X-Git-Url: https://git.dogcows.com/gitweb?a=blobdiff_plain;f=src%2Fcreate.c;h=1b08e0b21fd3e5f34675c01b8ad1f43b8477ca9c;hb=4eb1484dced64b74cafda40918c0d96e5084846a;hp=d5c8c9204987f09259b0b939ec4a1dafe459fd63;hpb=48be206fbfd81aa56056d7327fa8295e27011e5f;p=chaz%2Ftar diff --git a/src/create.c b/src/create.c index d5c8c92..1b08e0b 100644 --- a/src/create.c +++ b/src/create.c @@ -1,202 +1,560 @@ /* Create a tar archive. - Copyright (C) 1985, 92, 93, 94, 96, 97 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 - version. + Copyright 1985, 1992-1994, 1996-1997, 1999-2001, 2003-2007, + 2009-2010, 2012-2014 Free Software Foundation, Inc. - This program is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General - Public License for more details. + This file is part of GNU tar. - 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 Place - Suite 330, Boston, MA 02111-1307, USA. */ + GNU tar 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 3 of the License, or + (at your option) any later version. -#include "system.h" + GNU tar is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. -#if !MSDOS -# include -# include -#endif + You should have received a copy of the GNU General Public License + along with this program. If not, see . -#if HAVE_UTIME_H -# include -#else -struct utimbuf - { - long actime; - long modtime; - }; -#endif + Written by John Gilmore, on 1985-08-25. */ -#include "common.h" +#include -#ifndef MSDOS -extern dev_t ar_dev; -extern ino_t ar_ino; -#endif +#include -extern struct name *gnu_list_name; +#include "common.h" +#include -/* This module is the only one that cares about `struct link's. */ +/* Error number to use when an impostor is discovered. + Pretend the impostor isn't there. */ +enum { IMPOSTOR_ERRNO = ENOENT }; struct link { - struct link *next; dev_t dev; ino_t ino; - short linkcount; + nlink_t nlink; char name[1]; }; -struct link *linklist = NULL; /* points to first link in list */ +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. */ +#define MAX_VAL_WITH_DIGITS(digits, bits_per_digit) \ + ((digits) * (bits_per_digit) < sizeof (uintmax_t) * CHAR_BIT \ + ? ((uintmax_t) 1 << ((digits) * (bits_per_digit))) - 1 \ + : (uintmax_t) -1) -/*------------------------------------------------------------------------. -| Convert VALUE into a size-SIZE field at WHERE, including a | -| trailing space. For example, 3 for SIZE means two digits and a space. | -| | -| We assume the trailing NUL is already there and don't fill it in. This | -| fact is used by start_header and finish_header, so don't change it! | -`------------------------------------------------------------------------*/ +/* 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) -/* This should be equivalent to: sprintf (WHERE, "%*lo ", SIZE - 1, VALUE); - except that we don't assume VALUE fits in an unsigned long, and - except that sprintf fills in the trailing NUL and we don't. */ +/* 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_oct (uintmax_t value, char *where, size_t size, const char *type) +to_octal (uintmax_t value, char *where, size_t size) { uintmax_t v = value; size_t i = size; - where[--i] = ' '; /* put in the space, though */ - /* Produce the digits -- at least one. */ + do + { + where[--i] = '0' + (v & ((1 << LG_8) - 1)); + v >>= LG_8; + } + 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. + 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] = '0' + (int) (v & 7); /* one octal digit */ - v >>= 3; + where[--i] = v & ((1 << LG_256) - 1); + v = propagated_sign_bits | (v >> LG_256); + } + 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; } - while (i != 0 && v != 0); + else + minval_string = "0"; - /* Leading spaces, if necessary. */ - while (i != 0) - where[--i] = ' '; + if (negative) + { + char *p = STRINGIFY_BIGINT (- value, valbuf + 1); + *--p = '-'; + value_string = p; + } + else + value_string = STRINGIFY_BIGINT (value, valbuf); - if (v != 0) + if (substitute) { - char buf[UINTMAX_STRSIZE_BOUND]; - ERROR ((0, 0, _("%s value %s is too large to fit in a %u-bit field"), - type, STRINGIFY_BIGINT (value, buf), - (unsigned) ((size - 1) * 3))); + 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; } -void -gid_to_oct (gid_t v, char *p, size_t s) + +/* 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 bool +to_chars (int negative, uintmax_t value, size_t valsize, + uintmax_t (*substitute) (int *), + char *where, size_t size, const char *type) { - to_oct (v, p, s, "gid_t"); + 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; + } + else if (gnu_format) + { + /* Try to cope with the number by using traditional GNU format + methods */ + + /* Generate the base-256 representation if the number fits. */ + if (((negative ? -1 - value : value) + <= MAX_VAL_WITH_DIGITS (size - 1, LG_256))) + { + where[0] = negative ? -1 : 1 << (LG_256 - 1); + to_base256 (negative, value, where + 1, size - 1); + return true; + } + + /* 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); + return true; + } + /* 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); } -void -major_to_oct (major_t v, char *p, size_t s) + +static uintmax_t +gid_substitute (int *negative) { - to_oct (v, p, s, "major_t"); + gid_t r; +#ifdef GID_NOBODY + r = GID_NOBODY; +#else + static gid_t gid_nobody; + if (!gid_nobody && !gname_to_gid ("nobody", &gid_nobody)) + gid_nobody = -2; + r = gid_nobody; +#endif + *negative = r < 0; + return r; } -void -minor_to_oct (minor_t v, char *p, size_t s) + +static bool +gid_to_chars (gid_t v, char *p, size_t s) { - to_oct (v, p, s, "minor_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, gid_substitute, p, s, "gid_t"); } -void -mode_to_oct (mode_t v, char *p, size_t s) + +static bool +major_to_chars (major_t v, char *p, size_t s) { - to_oct (v, p, s, "mode_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "major_t"); } -void -off_to_oct (off_t v, char *p, size_t s) + +static bool +minor_to_chars (minor_t v, char *p, size_t s) { - to_oct (v, p, s, "off_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "minor_t"); } -void -size_to_oct (size_t v, char *p, size_t s) + +static bool +mode_to_chars (mode_t v, char *p, size_t s) { - to_oct (v, p, s, "size_t"); + /* 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. */ + int negative; + uintmax_t u; + 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 + && archive_format != POSIX_FORMAT + && archive_format != USTAR_FORMAT + && archive_format != GNU_FORMAT) + { + negative = v < 0; + u = v; + } + else + { + negative = 0; + u = ((v & S_ISUID ? TSUID : 0) + | (v & S_ISGID ? TSGID : 0) + | (v & S_ISVTX ? TSVTX : 0) + | (v & S_IRUSR ? TUREAD : 0) + | (v & S_IWUSR ? TUWRITE : 0) + | (v & S_IXUSR ? TUEXEC : 0) + | (v & S_IRGRP ? TGREAD : 0) + | (v & S_IWGRP ? TGWRITE : 0) + | (v & S_IXGRP ? TGEXEC : 0) + | (v & S_IROTH ? TOREAD : 0) + | (v & S_IWOTH ? TOWRITE : 0) + | (v & S_IXOTH ? TOEXEC : 0)); + } + return to_chars (negative, u, sizeof v, 0, p, s, "mode_t"); } -void -time_to_oct (time_t v, char *p, size_t s) + +bool +off_to_chars (off_t v, char *p, size_t s) { - to_oct (v, p, s, "time_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "off_t"); } -void -uid_to_oct (uid_t v, char *p, size_t s) + +bool +time_to_chars (time_t v, char *p, size_t s) { - to_oct (v, p, s, "uid_t"); + return to_chars (v < 0, (uintmax_t) v, sizeof v, 0, p, s, "time_t"); } -void -uintmax_to_oct (uintmax_t v, char *p, size_t s) + +static uintmax_t +uid_substitute (int *negative) { - to_oct (v, p, s, "uintmax_t"); + uid_t r; +#ifdef UID_NOBODY + r = UID_NOBODY; +#else + static uid_t uid_nobody; + if (!uid_nobody && !uname_to_uid ("nobody", &uid_nobody)) + uid_nobody = -2; + r = uid_nobody; +#endif + *negative = r < 0; + return r; +} + +static bool +uid_to_chars (uid_t v, char *p, size_t s) +{ + return to_chars (v < 0, (uintmax_t) v, sizeof v, uid_substitute, p, s, "uid_t"); } - -/* Writing routines. */ -/*-----------------------------------------------------------------------. -| Just zeroes out the buffer so we don't confuse ourselves with leftover | -| data. | -`-----------------------------------------------------------------------*/ +static bool +uintmax_to_chars (uintmax_t v, char *p, size_t s) +{ + return to_chars (0, v, sizeof v, 0, p, s, "uintmax_t"); +} static void -clear_buffer (char *buffer) +string_to_chars (char const *str, char *p, size_t s) +{ + tar_copy_str (p, str, s); + p[s - 1] = '\0'; +} + + +/* 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: + + 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); } -/*-------------------------------------------------------------------------. -| Write the EOT block(s). We actually zero at least one block, through | -| the end of the record. Old tar, as previous versions of GNU tar, writes | -| garbage after two zeroed blocks. | -`-------------------------------------------------------------------------*/ + +/* 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. */ void write_eot (void) { union block *pointer = find_next_block (); + memset (pointer->buffer, 0, BLOCKSIZE); + set_next_block_after (pointer); + pointer = find_next_block (); + memset (pointer->buffer, 0, available_space_after (pointer)); + set_next_block_after (pointer); +} - if (pointer) - { - size_t space = available_space_after (pointer); +/* 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 (pointer->buffer, 0, space); - set_next_block_after (pointer); - } -} + memset (header->buffer, 0, sizeof (union block)); -/*-----------------------------------------------------. -| Write a GNUTYPE_LONGLINK or GNUTYPE_LONGNAME block. | -`-----------------------------------------------------*/ + tar_name_copy_str (header->header.name, name, NAME_FIELD_SIZE); + OFF_TO_CHARS (size, header->header.size); -/* FIXME: Cross recursion between start_header and write_long! */ + TIME_TO_CHARS (t < 0 ? 0 : min (t, MAX_OCTAL_VAL (header->header.mtime)), + header->header.mtime); + MODE_TO_CHARS (S_IFREG|S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH, header->header.mode); + UID_TO_CHARS (0, header->header.uid); + GID_TO_CHARS (0, header->header.gid); + strncpy (header->header.magic, TMAGIC, TMAGLEN); + strncpy (header->header.version, TVERSION, TVERSLEN); + return header; +} -static union block *start_header PARAMS ((const char *, struct stat *)); +/* Create a new header and store there at most NAME_FIELD_SIZE bytes of + the file name */ + +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; +} +/* 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 stat foo; - - memset (&foo, 0, sizeof foo); - foo.st_size = size; - - header = start_header ("././@LongLink", &foo); + char *tmpname; + + header = start_private_header ("././@LongLink", size, 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->buffer + offsetof (struct posix_header, magic), + OLDGNU_MAGIC); header->header.typeflag = type; - finish_header (header); + finish_header (st, header, -1); header = find_next_block (); @@ -215,68 +573,189 @@ write_long (const char *p, char type) memset (header->buffer + size, 0, bufsize - size); set_next_block_after (header + (size - 1) / BLOCKSIZE); } - -/* 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. | -`---------------------------------------------------------------------*/ +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 * -start_header (const char *name, struct stat *st) +write_ustar_long_name (const char *name) { + size_t length = strlen (name); + size_t i, nlen; union block *header; - if (!absolute_names_option) + if (length > PREFIX_FIELD_SIZE + NAME_FIELD_SIZE + 1) { - static int warned_once = 0; + 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; + } -#if MSDOS - if (name[1] == ':') - { - name += 2; - if (!warned_once) - { - warned_once = 1; - WARN ((0, 0, _("Removing drive spec from names in the archive"))); - } - } -#endif + 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) +{ + 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; - while (*name == '/') + case OLDGNU_FORMAT: + case GNU_FORMAT: + write_gnu_long_link (st, st->link_name, GNUTYPE_LONGLINK); + break; + + default: + abort(); /*FIXME*/ + } +} + +static union block * +write_long_name (struct tar_stat_info *st) +{ + switch (archive_format) + { + case POSIX_FORMAT: + xheader_store ("path", st, NULL); + break; + + case V7_FORMAT: + if (strlen (st->file_name) > NAME_FIELD_SIZE-1) { - name++; /* force relative path */ - if (!warned_once) - { - warned_once = 1; - WARN ((0, 0, _("\ -Removing leading `/' from absolute path names in the archive"))); - } + 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; + + case USTAR_FORMAT: + case STAR_FORMAT: + return write_ustar_long_name (st->file_name); + + 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); +} + +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; - /* Check the file name and put it in the block. */ + if (st->xhdr.buffer || st->xhdr.stk == NULL) + return old_header; - if (strlen (name) >= (size_t) NAME_FIELD_SIZE) - write_long (name, GNUTYPE_LONGNAME); + xheader_finish (&st->xhdr); + memcpy (hp.buffer, old_header, sizeof (hp)); + if (global) + { + type = XGLTYPE; + p = xheader_ghdr_name (); + t = start_time.tv_sec; + } + else + { + type = XHDTYPE; + p = xheader_xhdr_name (st); + t = set_mtime_option ? mtime_option.tv_sec : st->stat.st_mtime; + } + xheader_write (type, p, t, &st->xhdr); + free (p); header = find_next_block (); - memset (header->buffer, 0, sizeof (union 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. */ - assign_string (¤t_file_name, name); +union block * +start_header (struct tar_stat_info *st) +{ + union block *header; - strncpy (header->header.name, name, NAME_FIELD_SIZE); - header->header.name[NAME_FIELD_SIZE - 1] = '\0'; + header = write_header_name (st); + if (!header) + return NULL; /* Override some stat fields, if requested to do so. */ if (owner_option != (uid_t) -1) - st->st_uid = owner_option; + st->stat.st_uid = owner_option; if (group_option != (gid_t) -1) - st->st_gid = group_option; + st->stat.st_gid = group_option; if (mode_option) - st->st_mode = ((st->st_mode & S_IFMT) - | mode_adjust (st->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 @@ -295,47 +774,133 @@ Removing leading `/' from absolute path names in the archive"))); . . . . . . . . . 9 = Omron UNIOS-B 4.3BSD 1.60Beta . = works - # = ``impossible file type'' + # = "impossible file type" - The following mask for old archive removes the `#'s in column 4 + The following mask for old archive removes the '#'s in column 4 above, thus making GNU tar both a universal donor and a universal acceptor for Paul's test. */ - if (archive_format == V7_FORMAT) - MODE_TO_OCT (st->st_mode & 07777, header->header.mode); + if (archive_format == V7_FORMAT || archive_format == USTAR_FORMAT) + MODE_TO_CHARS (st->stat.st_mode & MODE_ALL, header->header.mode); else - MODE_TO_OCT (st->st_mode, header->header.mode); + MODE_TO_CHARS (st->stat.st_mode, header->header.mode); - UID_TO_OCT (st->st_uid, header->header.uid); - GID_TO_OCT (st->st_gid, header->header.gid); - OFF_TO_OCT (st->st_size, header->header.size); - TIME_TO_OCT (st->st_mtime, header->header.mtime); + { + 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 (incremental_option) - if (archive_format == OLDGNU_FORMAT) + { + 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; + } + + { + 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)) + { + major_t devmajor = major (st->stat.st_rdev); + minor_t devminor = minor (st->stat.st_rdev); + + 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 (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 if (archive_format != GNU_FORMAT && archive_format != OLDGNU_FORMAT) + { + 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, NULL); + xheader_store ("ctime", st, NULL); + } + else if (incremental_option) + if (archive_format == OLDGNU_FORMAT || archive_format == GNU_FORMAT) { - TIME_TO_OCT (st->st_atime, header->oldgnu_header.atime); - TIME_TO_OCT (st->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; switch (archive_format) { - case DEFAULT_FORMAT: case V7_FORMAT: break; case OLDGNU_FORMAT: + case GNU_FORMAT: /*FIXME?*/ /* Overwrite header->header.magic and header.version in one blow. */ - strcpy (header->header.magic, OLDGNU_MAGIC); + strcpy (header->buffer + offsetof (struct posix_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; + + default: + abort (); } if (archive_format == V7_FORMAT || numeric_owner_option) @@ -344,1161 +909,1051 @@ Removing leading `/' from absolute path names in the archive"))); } else { - uid_to_uname (st->st_uid, header->header.uname); - gid_to_gname (st->st_gid, header->header.gname); + if (owner_name_option) + st->uname = xstrdup (owner_name_option); + else + uid_to_uname (st->stat.st_uid, &st->uname); + + if (group_name_option) + st->gname = xstrdup (group_name_option); + else + gid_to_gname (st->stat.st_gid, &st->gname); + + if (archive_format == POSIX_FORMAT + && (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 + || !string_ascii_p (st->gname))) + xheader_store ("gname", st, NULL); + GNAME_TO_CHARS (st->gname, header->header.gname); + } + + if (archive_format == POSIX_FORMAT) + { + if (acls_option > 0) + { + if (st->acls_a_ptr) + xheader_store ("SCHILY.acl.access", st, NULL); + if (st->acls_d_ptr) + xheader_store ("SCHILY.acl.default", st, NULL); + } + if ((selinux_context_option > 0) && st->cntx_name) + xheader_store ("RHT.security.selinux", st, NULL); + if (xattrs_option > 0) + { + size_t scan_xattr = 0; + struct xattr_array *xattr_map = st->xattr_map; + + while (scan_xattr < st->xattr_map_size) + { + xheader_store (xattr_map[scan_xattr].xkey, st, &scan_xattr); + ++scan_xattr; + } + } } 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. | -`-------------------------------------------------------------------------*/ - void -finish_header (union block *header) +simple_finish_header (union block *header) { size_t i; 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++; /* Fill in the checksum field. It's formatted differently from the other fields: it has [6] digits, a null, then a space -- rather than - digits, a space, then a null. We use to_oct then write the null in - over to_oct's space. The final space is already there, from - checksumming, and to_oct doesn't modify it. + digits, then a null. We use to_chars. + The final space is already there, from + checksumming, and to_chars doesn't modify it. This is a fast way to do: sprintf(header->header.chksum, "%6o", sum); */ - UINTMAX_TO_OCT ((uintmax_t) sum, header->header.chksum); - header->header.chksum[6] = '\0'; /* zap the space */ + uintmax_to_chars ((uintmax_t) sum, header->header.chksum, 7); set_next_block_after (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 (struct tar_stat_info *st, + union block *header, off_t block_ordinal) +{ + /* 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 != GNUTYPE_LONGNAME + && header->header.typeflag != XHDTYPE + && header->header.typeflag != XGLTYPE) { - /* These globals are parameters to print_header, sigh. */ - - current_header = header; - /* current_stat is already set up. */ + /* FIXME: This global is used in print_header, sigh. */ current_format = archive_format; - print_header (); + print_header (st, header, block_ordinal); } + + header = write_extended (false, st, header); + simple_finish_header (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) +void +pad_archive (off_t size_left) { - int counter; - - for (counter = 0; counter < BLOCKSIZE; counter++) - if (buffer[counter] != '\0') - return 0; - return 1; + 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 void -init_sparsearray (void) +static enum dump_status +dump_regular_file (int fd, struct tar_stat_info *st) { - int counter; + 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; - sp_array_size = 10; + /* Mark contiguous files, if we support them. */ + if (archive_format != V7_FORMAT && S_ISCTG (st->stat.st_mode)) + blk->header.typeflag = CONTTYPE; - /* Make room for our scratch space -- initially is 10 elts long. */ + finish_header (st, blk, block_ordinal); - sparsearray = (struct sp_array *) - xmalloc (sp_array_size * sizeof (struct sp_array)); - for (counter = 0; counter < sp_array_size; counter++) + mv_begin_write (st->file_name, st->stat.st_size, st->stat.st_size); + while (size_left > 0) { - sparsearray[counter].offset = 0; - sparsearray[counter].numbytes = 0; - } -} + size_t bufsize, count; -/*---. -| ? | -`---*/ + blk = find_next_block (); -static void -find_new_file_size (off_t *filesize, int highest_index) -{ - int counter; + bufsize = available_space_after (blk); - *filesize = 0; - for (counter = 0; - sparsearray[counter].numbytes && counter <= highest_index; - counter++) - *filesize += sparsearray[counter].numbytes; -} + 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); + } -/*-----------------------------------------------------------------------. -| 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. | -`-----------------------------------------------------------------------*/ + count = (fd <= 0) ? bufsize : blocking_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); -/* There is little point in trimming small amounts of null data at the head - and tail of blocks, only avoid dumping full null blocks. */ + 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; +} -/* FIXME: this routine might accept bits of algorithmic cleanup, it is - too kludgey for my taste... */ + +/* Copy info from the directory identified by ST into the archive. + DIRECTORY contains the directory's entries. */ -static int -deal_with_sparse (char *name, union block *header) +static void +dump_dir0 (struct tar_stat_info *st, char const *directory) { - size_t numbytes = 0; - off_t offset = 0; - int file; - int sparse_index = 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); + info_attach_exclist (st); + + if (incremental_option && archive_format != POSIX_FORMAT) + blk->header.typeflag = GNUTYPE_DUMPDIR; + else /* if (standard_option) */ + blk->header.typeflag = DIRTYPE; - while (count = read (file, buffer, sizeof buffer), count != 0) - { - /* 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 (sparse_index > sp_array_size - 1) + if (!incremental_option) + finish_header (st, blk, block_ordinal); + else if (gnu_list_name->directory) + { + if (archive_format == POSIX_FORMAT) { - - sparsearray = (struct sp_array *) - 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); } - - /* Process one block. */ - - if (count == sizeof buffer) - - if (zero_block_p (buffer)) - { - if (numbytes) - { - sparsearray[sparse_index++].numbytes = numbytes; - numbytes = 0; - } - } - else - { - if (!numbytes) - sparsearray[sparse_index].offset = offset; - numbytes += count; - } - else + { + off_t size_left; + off_t totsize; + size_t bufsize; + ssize_t count; + const char *buffer, *p_buffer; - /* Since count < sizeof buffer, we have the last bit of the file. */ + 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 (!zero_block_p (buffer)) - { - if (!numbytes) - sparsearray[sparse_index].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); + 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 (numbytes) - sparsearray[sparse_index++].numbytes = numbytes; - else + if (!recursion_option) + return; + + if (one_file_system_option + && !top_level + && st->parent->stat.st_dev != st->stat.st_dev) { - sparsearray[sparse_index].offset = offset - 1; - sparsearray[sparse_index++].numbytes = 1; + 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; - close (file); - return sparse_index - 1; -} + switch (check_exclusion_tags (st, &tag_file_name)) + { + case exclusion_tag_all: + /* Handled in dump_file0 */ + break; -/*---. -| ? | -`---*/ + case exclusion_tag_none: + { + 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) + { + 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, st)) + dump_file (st, entry, name_buf); + } -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; - ssize_t count; + free (name_buf); + } + break; - 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)); + 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 (lseek (file, sparsearray[sparse_index++].offset, 0) < 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)); + case exclusion_tag_under: + exclusion_tag_warning (st->orig_file_name, tag_file_name, + _("contents not dumped")); break; } + } +} - /* If the number of bytes to be written here exceeds the size of - the temporary buffer, do it in steps. */ +/* 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'; +} - while (bufsize > BLOCKSIZE) - { -#if 0 - if (amount_read) - { - count = 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. */ +/* 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. - count = 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)); - return 1; - } - bufsize -= count; - *sizeleft -= count; - set_next_block_after (start); - start = find_next_block (); - memset (start->buffer, 0, BLOCKSIZE); - } + 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. */ - { - char buffer[BLOCKSIZE]; +static bool +open_failure_recover (struct tar_stat_info const *dir) +{ + if (errno == EMFILE && dir && dir->parent) + { + 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; + } - clear_buffer (buffer); - count = read (file, buffer, bufsize); - memcpy (start->buffer, buffer, BLOCKSIZE); - } + return false; +} - 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)); - 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); +/* 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) +{ + while (! (st->dirstream = fdopendir (st->fd))) + if (! open_failure_recover (st)) + return 0; + return streamsavedir (st->dirstream, savedir_sort_order); +} +/* 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; } - free (sparsearray); -#if 0 - set_next_block_after (start + (count - 1) / BLOCKSIZE); -#endif - return 0; + + dump_dir0 (st, directory); + + restore_parent_fd (st); + free (directory); + return true; } + -/* Main functions of this module. */ +/* 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) { - 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) - dump_file (p, (dev_t) -1, 1); + while ((p = name_from_list ()) != NULL) + if (!excluded_name (p->name, NULL)) + dump_file (0, p->name, p->name); blank_name_list (); - while (p = name_from_list (), p) - { - strcpy (buffer, p); - if (p[strlen (p) - 1] != '/') - strcat (buffer, "/"); - bufp = buffer + strlen (buffer); - for (q = gnu_list_name->dir_contents; - q && *q; - q += strlen (q) + 1) - { - if (*q == 'Y') + while ((p = name_from_list ()) != NULL) + if (!excluded_name (p->name, NULL)) + { + 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->name, plen); + if (! ISSLASH (buffer[plen - 1])) + buffer[plen++] = DIRECTORY_SEPARATOR; + tar_stat_init (&st); + q = directory_contents (p->directory); + if (q) + while (*q) { - strcpy (bufp, q + 1); - dump_file (buffer, (dev_t) -1, 1); + size_t qlen = strlen (q); + if (*q == 'Y') + { + if (! st.orig_file_name) + { + int fd = openat (chdir_fd, 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 (&st, q + 1, buffer); + } + q += qlen + 1; } - } - } + tar_stat_destroy (&st); + } free (buffer); } else { - while (p = name_next (1), p) - dump_file (p, (dev_t) -1, 1); + const char *name; + while ((name = name_next (1)) != NULL) + if (!excluded_name (name, NULL)) + dump_file (0, name, name); } write_eot (); close_archive (); - + finish_deferred_unlinks (); if (listed_incremental_option) - write_dir_file (); + write_directory_file (); } -/*----------------------------------------------------------------------. -| 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. | -`----------------------------------------------------------------------*/ -/* FIXME: One should make sure that for *every* path leading to setting - exit_status to failure, a clear diagnostic has been issued. */ +/* Calculate the hash of a link. */ +static size_t +hash_link (void const *entry, size_t n_buckets) +{ + struct link const *l = entry; + uintmax_t num = l->dev ^ l->ino; + return num % n_buckets; +} -void -dump_file (char *p, dev_t parent_device, int top_level) +/* Compare two links for equality. */ +static bool +compare_links (void const *entry1, void const *entry2) { - union block *header; - char type; - union block *exhdr; - char save_typeflag; - struct utimbuf restore_times; + struct link const *link1 = entry1; + struct link const *link2 = entry2; + return ((link1->dev ^ link2->dev) | (link1->ino ^ link2->ino)) == 0; +} - /* FIXME: `header' and `upperbound' might be used uninitialized in this - function. Reported by Bruno Haible. */ +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); +} - if (interactive_option && !confirm ("add", p)) - return; + +/* Handling of hard links */ - /* Use stat if following (rather than dumping) 4.2BSD's symbolic links. - Otherwise, use lstat (which falls back to stat if no symbolic 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; - if (dereference_option != 0 -#ifdef STX_HIDDEN /* 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 - ) +/* 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) +{ + if (link_table + && (trivial_link_count < st->stat.st_nlink || remove_files_option)) { - WARN ((0, errno, _("Cannot add file %s"), p)); - if (!ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; - return; - } + struct link lp; + struct link *duplicate; + off_t block_ordinal; + union block *blk; - restore_times.actime = current_stat.st_atime; - restore_times.modtime = current_stat.st_mtime; + lp.ino = st->stat.st_ino; + lp.dev = st->stat.st_dev; -#ifdef S_ISHIDDEN - if (S_ISHIDDEN (current_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; + /* We found a link. */ + char const *link_name = safer_name_suffix (duplicate->name, true, + absolute_names_option); + if (duplicate->nlink) + duplicate->nlink--; + + 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); + + st->stat.st_size = 0; + blk = start_header (st); + if (!blk) + return false; + tar_copy_str (blk->header.linkname, link_name, NAME_FIELD_SIZE); + + blk->header.typeflag = LNKTYPE; + finish_header (st, blk, block_ordinal); + + if (remove_files_option) + queue_deferred_unlink (st->orig_file_name, false); + + return true; } } -#endif - - /* See if we only want new files, and check if this one is too old to - put in the archive. */ - - if (!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 (parent_device == (dev_t) -1) - WARN ((0, 0, _("%s: is unchanged; not dumped"), p)); - /* FIXME: recheck this return. */ - return; - } - -#if !MSDOS - /* See if we are trying to dump the archive. */ - - 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. */ + return false; +} - if (current_stat.st_nlink > 1 - && (S_ISREG (current_stat.st_mode) -#ifdef S_ISCTG - || S_ISCTG (current_stat.st_mode) -#endif -#ifdef S_ISCHR - || S_ISCHR (current_stat.st_mode) -#endif -#ifdef S_ISBLK - || S_ISBLK (current_stat.st_mode) -#endif -#ifdef S_ISFIFO - || S_ISFIFO (current_stat.st_mode) -#endif - )) +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; - /* 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 = 0; - - if (!warned_once) - { - warned_once = 1; - WARN ((0, 0, _("\ -Removing leading `/' from absolute links"))); - } - 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; + assign_string (&linkname, st->orig_file_name); + transform_name (&linkname, XFORM_LINK); + + 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); + + if (! ((link_table + || (link_table = hash_initialize (0, 0, hash_link, + compare_links, 0))) + && (duplicate = hash_insert (link_table, lp)))) + xalloc_die (); + + if (duplicate != lp) + abort (); + lp->nlink--; } +} - /* This is not a link to a previously dumped file, so dump it. */ - - if (S_ISREG (current_stat.st_mode) -#ifdef S_ISCTG - || S_ISCTG (current_stat.st_mode) -#endif - ) - { - 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 +/* 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; - header_moved = 0; + if (!link_table) + return; - if (sparse_option) + for (lp = hash_get_first (link_table); lp; + lp = hash_get_next (link_table, lp)) + { + if (lp->nlink) { - /* 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_OCT (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_OCT (filesize, header->header.size); - - for (counter = 0; counter < SPARSES_IN_OLDGNU_HEADER; counter++) - { - if (!sparsearray[counter].numbytes) - break; - - OFF_TO_OCT (sparsearray[counter].offset, - header->oldgnu_header.sp[counter].offset); - SIZE_TO_OCT (sparsearray[counter].numbytes, - header->oldgnu_header.sp[counter].numbytes); - } - - } + WARN ((0, 0, _("Missing links to %s."), quote (lp->name))); } - 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. */ +/* 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 (dev_null_output - || (sizeleft == 0 && 0444 == (0444 & 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; - } - } + 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 (""); + } - /* If the file is sparse, we've already taken care of this. */ + while ((fd = openat (dir ? dir->fd : chdir_fd, file, flags)) < 0 + && open_failure_recover (dir)) + continue; + return fd; +} - if (!header_moved) +/* 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)) { - header = start_header (p, ¤t_stat); - if (header == NULL) - { - if (f >= 0) - close (f); - exit_status = TAREXIT_FAILURE; - return; - } + close (parentfd); + parentfd = IMPOSTOR_ERRNO; } -#ifdef S_ISCTG - /* Mark contiguous files, if we support them. */ - if (archive_format != V7_FORMAT && S_ISCTG (current_stat.st_mode)) - header->header.typeflag = CONTTYPE; -#endif - isextended = header->oldgnu_header.isextended; - save_typeflag = header->header.typeflag; - finish_header (header); - if (isextended) + if (parentfd < 0) { -#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) + int origfd = openat (chdir_fd, parent->orig_file_name, + open_searchdir_flags); + if (0 <= origfd) { - exit_status = TAREXIT_FAILURE; - return; + if (fstat (parentfd, &parentstat) == 0 + && parent->stat.st_ino == parentstat.st_ino + && parent->stat.st_dev == parentstat.st_dev) + parentfd = origfd; + else + close (origfd); } - memset (exhdr->buffer, 0, BLOCKSIZE); - for (counter = 0; counter < SPARSES_IN_SPARSE_HEADER; counter++) - { - if (counter + index_offset > upperbound) - break; - - SIZE_TO_OCT (sparsearray[counter + index_offset].numbytes, - exhdr->sparse_header.sp[counter].numbytes); - OFF_TO_OCT (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 (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); + parent->fd = parentfd; + } +} - if (sizeleft < bufsize) - { - /* Last read -- zero out area beyond. */ +/* 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). */ - bufsize = sizeleft; - count = bufsize % BLOCKSIZE; - if (count) - memset (start->buffer + sizeleft, 0, - (size_t) (BLOCKSIZE - count)); - } - if (f < 0) - count = bufsize; - else - count = 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; +/* FIXME: One should make sure that for *every* path leading to setting + exit_status to failure, a clear diagnostic has been issued. */ - /* This is nonportable (the type of set_next_block_after's arg). */ +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; + 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 ? chdir_fd : parent->fd; + void (*diag) (char const *) = 0; - set_next_block_after (start + (count - 1) / BLOCKSIZE); + if (interactive_option && !confirm ("add", p)) + return; - 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 */ - } - } + assign_string (&st->orig_file_name, p); + assign_string (&st->file_name, + safer_name_suffix (p, false, absolute_names_option)); - if (multi_volume_option) - assign_string (&save_name, NULL); + transform_name (&st->file_name, XFORM_REGFILE); - if (f >= 0) - { - close (f); - if (atime_preserve_option) - utime (p, &restore_times); - } - if (remove_files_option) + 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 { - if (unlink (p) == -1) - ERROR ((0, errno, _("Cannot remove %s"), p)); + st->fd = fd; + if (fstat (fd, &st->stat) != 0) + diag = stat_diag; } + } + if (diag) + { + file_removed_diag (p, top_level, diag); return; + } - /* File shrunk or gave error, pad out tape to match the size we - specified in the header. */ + st->archive_file_size = original_size = st->stat.st_size; + st->atime = get_stat_atime (&st->stat); + st->mtime = get_stat_mtime (&st->stat); + st->ctime = original_ctime = get_stat_ctime (&st->stat); - 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) +#ifdef S_ISHIDDEN + if (S_ISHIDDEN (st->stat.st_mode)) + { + char *new = (char *) alloca (strlen (p) + 2); + if (new) { - close (f); - if (atime_preserve_option) - utime (p, &restore_times); + strcpy (new, p); + strcat (new, "@"); + p = new; } - return; } +#endif -#ifdef S_ISLNK - else if (S_ISLNK (current_stat.st_mode)) - { - int size; - char *buffer = (char *) alloca (PATH_MAX + 1); + /* See if we want only new files, and check if this one is too old to + put in the archive. - 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); + This check is omitted if incremental_option is set *and* the + requested file is not explicitly listed in the command line. */ - 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)); - } + if (! (incremental_option && ! top_level) + && !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; } -#endif /* S_ISLNK */ - else if (S_ISDIR (current_stat.st_mode)) + /* See if we are trying to dump the archive. */ + if (sys_file_is_archive (st)) { - DIR *directory; - struct dirent *entry; - 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. + WARNOPT (WARN_IGNORE_ARCHIVE, + (0, 0, _("%s: file is the archive; not dumped"), + quotearg_colon (p))); + return; + } - 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. */ + is_dir = S_ISDIR (st->stat.st_mode) != 0; - if (access (p, R_OK) == -1 && geteuid () != 0) - { - WARN ((0, errno, _("Cannot add directory %s"), p)); - if (!ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; - return; - } + if (!is_dir && dump_hard_link (st)) + return; - /* Build new prototype name. Ensure exactly one trailing slash. */ + if (is_dir || S_ISREG (st->stat.st_mode) || S_ISCTG (st->stat.st_mode)) + { + bool ok; + struct stat final_stat; - len = strlen (p); - buflen = len + NAME_FIELD_SIZE; - namebuf = xmalloc (buflen + 1); - strncpy (namebuf, p, buflen); - while (len >= 1 && namebuf[len - 1] == '/') - len--; - namebuf[len++] = '/'; - namebuf[len] = '\0'; + xattrs_acls_get (parentfd, name, st, 0, !is_dir); + xattrs_selinux_get (parentfd, name, st, fd); + xattrs_xattrs_get (parentfd, name, st, fd); - if (1) + if (is_dir) { - /* The "1" above used to be "archive_format != V7_FORMAT", GNU tar - was just 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. */ - - current_stat.st_size = 0; /* force 0 size on dir */ + const char *tag_file_name; + ensure_slash (&st->orig_file_name); + ensure_slash (&st->file_name); - /* FIXME: If people could really read standard archives, this - should be: + 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; + } - header - = start_header (standard_option ? p : namebuf, ¤t_stat); + ok = dump_dir (st); - but since they'd interpret DIRTYPE blocks as regular - files, we'd better put the / on the name. */ + fd = st->fd; + parentfd = top_level ? chdir_fd : parent->fd; + } + else + { + enum dump_status status; - header = start_header (namebuf, ¤t_stat); - if (header == NULL) + if (fd && sparse_option && ST_IS_SPARSE (st->stat)) { - exit_status = TAREXIT_FAILURE; - return; /* eg name too long */ + 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 (incremental_option) - header->header.typeflag = GNUTYPE_DUMPDIR; - else /* if (standard_option) */ - header->header.typeflag = DIRTYPE; + switch (status) + { + case dump_status_ok: + case dump_status_short: + file_count_links (st); + break; - /* If we're gnudumping, we aren't done yet so don't close it. */ + case dump_status_fail: + break; - if (!incremental_option) - finish_header (header); /* done with directory header */ + case dump_status_not_implemented: + abort (); + } + + ok = status == dump_status_ok; } - if (incremental_option && gnu_list_name->dir_contents) + if (ok) { - 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; - for (p_buffer = buffer; p_buffer && *p_buffer;) + if (fd < 0) { - size_t tmp; - - tmp = strlen (p_buffer) + 1; - totsize += tmp; - p_buffer += tmp; + errno = - fd; + ok = false; } - totsize++; - OFF_TO_OCT (totsize, header->header.size); - finish_header (header); - p_buffer = buffer; - sizeleft = totsize; - while (sizeleft > 0) + else if (fd == 0) { - if (multi_volume_option) + if (parentfd < 0 && ! top_level) { - assign_string (&save_name, p); - save_sizeleft = sizeleft; - save_totsize = totsize; + errno = - parentfd; + ok = false; } - start = find_next_block (); - bufsize = available_space_after (start); - if (sizeleft < bufsize) - { - bufsize = sizeleft; - count = bufsize % BLOCKSIZE; - if (count) - memset (start->buffer + sizeleft, 0, - (size_t) (BLOCKSIZE - count)); - } - memcpy (start->buffer, p_buffer, bufsize); - sizeleft -= bufsize; - p_buffer += bufsize; - set_next_block_after (start + (bufsize - 1) / BLOCKSIZE); + else + ok = fstatat (parentfd, name, &final_stat, fstatat_flags) == 0; } - if (multi_volume_option) - assign_string (&save_name, NULL); - if (atime_preserve_option) - utime (p, &restore_times); - return; - } - - /* 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. */ + else + ok = fstat (fd, &final_stat) == 0; - if (no_recurse_option) - return; - - /* 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. */ + if (! ok) + file_removed_diag (p, top_level, stat_diag); + } - if (one_file_system_option && !top_level - && parent_device != current_stat.st_dev) + if (ok) { - if (verbose_option) - WARN ((0, 0, _("%s: On a different filesystem; not dumped"), p)); - return; + 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) + { + WARNOPT (WARN_FILE_CHANGED, + (0, 0, _("%s: file changed as we read it"), + quotearg_colon (p))); + set_exit_status (TAREXIT_DIFFERS); + } + else if (atime_preserve_option == replace_atime_preserve + && fd && (is_dir || original_size != 0) + && set_file_atime (fd, parentfd, name, st->atime) != 0) + utime_error (p); } - /* Now output all the files in the directory. */ - - errno = 0; /* FIXME: errno should be read-only */ + ok &= tar_stat_close (st); + if (ok && remove_files_option) + queue_deferred_unlink (p, is_dir); - directory = opendir (p); - if (!directory) + 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) { - ERROR ((0, errno, _("Cannot open directory %s"), p)); + file_removed_diag (p, top_level, readlink_diag); return; } + 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); + + xattrs_selinux_get (parentfd, name, st, 0); + xattrs_xattrs_get (parentfd, name, st, 0); + + 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 */ - /* Hack to remove "./" from the front of all the file names. */ - - if (len == 2 && namebuf[0] == '.' && namebuf[1] == '/') - len = 0; - - /* FIXME: Should speed this up by cd-ing into the dir. */ - - while (entry = readdir (directory), entry) - { - /* Skip `.' and `..'. */ - - if (is_dot_or_dotdot (entry->d_name)) - continue; - - if ((int) NAMLEN (entry) + len >= buflen) - { - 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 - } - strcpy (namebuf + len, entry->d_name); - if (exclude_option && check_exclude (namebuf)) - continue; - dump_file (namebuf, our_device, 0); - } + if (remove_files_option) + queue_deferred_unlink (p, false); - closedir (directory); - free (namebuf); - if (atime_preserve_option) - utime (p, &restore_times); + file_count_links (st); return; } - -#ifdef S_ISCHR - else if (S_ISCHR (current_stat.st_mode)) - type = CHRTYPE; -#endif - -#ifdef S_ISBLK - else if (S_ISBLK (current_stat.st_mode)) - type = BLKTYPE; -#endif - - /* Avoid screwy apollo lossage where S_IFIFO == S_IFSOCK. */ - -#if (_ISP__M68K == 0) && (_ISP__A88K == 0) && defined(S_ISFIFO) - else if (S_ISFIFO (current_stat.st_mode)) - type = FIFOTYPE; -#endif - -#ifdef S_ISSOCK - else if (S_ISSOCK (current_stat.st_mode)) - type = FIFOTYPE; #endif - + else if (S_ISCHR (st->stat.st_mode)) + { + type = CHRTYPE; + xattrs_acls_get (parentfd, name, st, 0, true); + xattrs_selinux_get (parentfd, name, st, 0); + xattrs_xattrs_get (parentfd, name, st, 0); + } + else if (S_ISBLK (st->stat.st_mode)) + { + type = BLKTYPE; + xattrs_acls_get (parentfd, name, st, 0, true); + xattrs_selinux_get (parentfd, name, st, 0); + xattrs_xattrs_get (parentfd, name, st, 0); + } + else if (S_ISFIFO (st->stat.st_mode)) + { + type = FIFOTYPE; + xattrs_acls_get (parentfd, name, st, 0, true); + xattrs_selinux_get (parentfd, name, st, 0); + xattrs_xattrs_get (parentfd, name, st, 0); + } + 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 - goto unknown; + { + unknown_file_error (p); + return; + } 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 */ + unknown_file_error (p); + return; } + block_ordinal = current_block_ordinal (); + st->stat.st_size = 0; /* force 0 size */ + header = start_header (st); + if (!header) + return; header->header.typeflag = type; -#if defined(S_IFBLK) || defined(S_IFCHR) if (type != FIFOTYPE) { - MAJOR_TO_OCT (major (current_stat.st_rdev), header->header.devmajor); - MINOR_TO_OCT (minor (current_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); } -#endif - finish_header (header); + finish_header (st, header, block_ordinal); if (remove_files_option) - { - if (unlink (p) == -1) - ERROR ((0, errno, _("Cannot remove %s"), p)); - } - return; + queue_deferred_unlink (p, false); +} -unknown: - ERROR ((0, 0, _("%s: Unknown file type; file ignored"), p)); +/* 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 +dump_file (struct tar_stat_info *parent, char const *name, + char const *fullname) +{ + 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); }