X-Git-Url: https://git.dogcows.com/gitweb?a=blobdiff_plain;f=src%2Fcreate.c;h=6381f9fe6ff71d14e01b87484e8f51b277c190f9;hb=d659cbaccdc1f3279c49107cf15f15a639738529;hp=6f38836e18b4674588ffd776b2abf6790f37ad88;hpb=ba2ef4d0752d3440fa449599cdd6def5c96bfbaf;p=chaz%2Ftar diff --git a/src/create.c b/src/create.c index 6f38836..6381f9f 100644 --- a/src/create.c +++ b/src/create.c @@ -1,5 +1,8 @@ /* Create a tar archive. - Copyright 1985, 92, 93, 94, 96, 97, 99, 2000 Free Software Foundation, Inc. + + Copyright (C) 1985, 1992, 1993, 1994, 1996, 1997, 1999, 2000, 2001, + 2003, 2004, 2005 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 @@ -14,47 +17,23 @@ 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. */ - -#include "system.h" - -#if !MSDOS -# include -# include -#endif + 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */ -#if HAVE_UTIME_H -# include -#else -struct utimbuf - { - long actime; - long modtime; - }; -#endif +#include #include +#include #include "common.h" - -#ifndef MSDOS -extern dev_t ar_dev; -extern ino_t ar_ino; -#endif - -extern struct name *gnu_list_name; - -/* This module is the only one that cares about `struct link's. */ +#include struct link { - struct link *next; dev_t dev; ino_t ino; + size_t nlink; char name[1]; }; - -static struct link *linklist; /* points to first link in list */ /* The maximum uintmax_t value that can be represented with DIGITS digits, assuming that each digit is BITS_PER_DIGIT wide. */ @@ -63,6 +42,10 @@ static struct link *linklist; /* points to first link in list */ ? ((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. */ @@ -81,6 +64,29 @@ 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. @@ -119,7 +125,7 @@ to_base256 (int negative, uintmax_t value, char *where, size_t size) static void to_chars (int negative, uintmax_t value, size_t valsize, - uintmax_t (*substitute) PARAMS ((int *)), + uintmax_t (*substitute) (int *), char *where, size_t size, const char *type) { int base256_allowed = (archive_format == GNU_FORMAT @@ -198,6 +204,8 @@ to_chars (int negative, uintmax_t value, size_t valsize, { int negsub; uintmax_t sub = substitute (&negsub) & maxval; + /* FIXME: This is the only place where GNU_FORMAT differs from + OLDGNU_FORMAT. 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); @@ -263,6 +271,7 @@ 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; @@ -332,26 +341,33 @@ uintmax_to_chars (uintmax_t v, char *p, size_t s) { to_chars (0, v, sizeof v, 0, p, s, "uintmax_t"); } - -/* Writing routines. */ -/*-----------------------------------------------------------------------. -| Just zeroes out the buffer so we don't confuse ourselves with leftover | -| data. | -`-----------------------------------------------------------------------*/ +void +string_to_chars (char const *str, char *p, size_t s) +{ + tar_copy_str (p, str, s); + p[s - 1] = '\0'; +} + + +/* A file is not dumpable if + a) it is empty *and* world-readable, or + b) current archive is /dev/null */ -static void -clear_buffer (char *buffer) +bool +file_dumpable_p (struct tar_stat_info *st) { - memset (buffer, 0, BLOCKSIZE); + return !(dev_null_output + || (st->archive_file_size == 0 + && (st->stat.st_mode & MODE_R) == MODE_R)); } -/*-------------------------------------------------------------------------. -| Write the EOT block(s). We zero at least two blocks, through | -| the end of the record. Old tar, as previous versions of GNU tar, writes | -| garbage after two zeroed blocks. | -`-------------------------------------------------------------------------*/ + +/* 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) { @@ -363,28 +379,73 @@ 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 (&t); + 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; +} + +/* Create a new header and store there at most NAME_FIELD_SIZE bytes of + the file name */ -/* FIXME: Cross recursion between start_header and write_long! */ +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; +} -static union block *start_header PARAMS ((const char *, struct stat *)); +#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 stat foo; - - memset (&foo, 0, sizeof foo); - foo.st_size = size; - - header = start_header ("././@LongLink", &foo); + char *tmpname; + + header = start_private_header ("././@LongLink", size); + 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); + finish_header (st, header, -1); header = find_next_block (); @@ -403,73 +464,173 @@ 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 whose stat info is st, - and return its address. */ +static size_t +split_long_name (const char *name, size_t length) +{ + size_t i; + + if (length > PREFIX_FIELD_SIZE) + length = PREFIX_FIELD_SIZE+2; + 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; union block *header; - if (!absolute_names_option) + if (length > PREFIX_FIELD_SIZE + NAME_FIELD_SIZE + 1) { - size_t prefix_len = FILESYSTEM_PREFIX_LEN (name); + 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 (prefix_len) - { - static int warned_once; - if (!warned_once) - { - warned_once = 1; - WARN ((0, 0, _("Removing `%.*s' prefix from member names"), - (int) prefix_len, name)); - } - name += prefix_len; - } + i = split_long_name (name, length); + if (i == 0 || length - i - 1 > NAME_FIELD_SIZE) + { + 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; + + 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; - while (*name == '/') + case V7_FORMAT: + if (strlen (st->file_name) > NAME_FIELD_SIZE-1) { - static int warned_once; - if (!warned_once) - { - warned_once = 1; - WARN ((0, 0, _("Removing leading `/' from member names"))); - } - name++; + 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; - { - static int warned_once; - if (! warned_once && contains_dot_dot (name)) - { - warned_once = 1; - WARN ((0, 0, _("Member names contain `..'"))); - } - } + 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); +} + +static union block * +write_extended (struct tar_stat_info *st, union block *old_header) +{ + union block *header, hp; + char *p; + + if (extended_header.buffer || extended_header.stk == NULL) + return old_header; - if (sizeof header->header.name <= strlen (name)) - write_long (name, GNUTYPE_LONGNAME); + xheader_finish (&extended_header); + memcpy (hp.buffer, old_header, sizeof (hp)); + p = xheader_xhdr_name (st); + xheader_write (XHDTYPE, p, &extended_header); + 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); +} - assign_string (¤t_file_name, name); + +/* Header handling. */ + +/* Make a header block for the file whose stat info is st, + and return its address. */ + +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 & ~MODE_ALL) - | mode_adjust (st->st_mode, mode_option)); + st->stat.st_mode = + ((st->stat.st_mode & ~MODE_ALL) + | mode_adjust (st->stat.st_mode, mode_option, initial_umask)); /* Paul Eggert tried the trivial test ($WRITER cf a b; $READER tvf a) for a few tars and came up with the following interoperability @@ -494,21 +655,96 @@ start_header (const char *name, struct stat *st) above, thus making GNU tar both a universal donor and a universal acceptor for Paul's test. */ - if (archive_format == V7_FORMAT) - MODE_TO_CHARS (st->st_mode & MODE_ALL, 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_CHARS (st->st_mode, header->header.mode); + MODE_TO_CHARS (st->stat.st_mode, header->header.mode); - UID_TO_CHARS (st->st_uid, header->header.uid); - GID_TO_CHARS (st->st_gid, header->header.gid); - OFF_TO_CHARS (st->st_size, header->header.size); - TIME_TO_CHARS (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; + } + UID_TO_CHARS (uid, header->header.uid); + } - 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; + } + GID_TO_CHARS (gid, header->header.gid); + } + + { + 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; + } + OFF_TO_CHARS (size, header->header.size); + } + + { + struct timespec mtime = get_stat_mtime (&st->stat); + if (archive_format == POSIX_FORMAT) + { + if (MAX_OCTAL_VAL (header->header.mtime) < mtime.tv_sec + || mtime.tv_nsec != 0) + xheader_store ("mtime", st, NULL); + if (MAX_OCTAL_VAL (header->header.mtime) < mtime.tv_sec) + mtime.tv_sec = 0; + } + TIME_TO_CHARS (mtime.tv_sec, header->header.mtime); + } + + /* 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; + } + MAJOR_TO_CHARS (devmajor, header->header.devmajor); + + if (archive_format == POSIX_FORMAT + && MAX_OCTAL_VAL (header->header.devminor) < devminor) + { + xheader_store ("devminor", st, NULL); + devminor = 0; + } + MINOR_TO_CHARS (devminor, header->header.devminor); + } + 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 (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_CHARS (st->st_atime, header->oldgnu_header.atime); - TIME_TO_CHARS (st->st_ctime, header->oldgnu_header.ctime); + TIME_TO_CHARS (st->stat.st_atime, header->oldgnu_header.atime); + TIME_TO_CHARS (st->stat.st_ctime, header->oldgnu_header.ctime); } header->header.typeflag = archive_format == V7_FORMAT ? AREGTYPE : REGTYPE; @@ -519,12 +755,13 @@ start_header (const char *name, struct stat *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; @@ -539,20 +776,27 @@ start_header (const char *name, struct stat *st) } else { - uid_to_uname (st->st_uid, header->header.uname); - gid_to_gname (st->st_gid, header->header.gname); + 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 + || !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); } 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; @@ -578,1005 +822,802 @@ finish_header (union block *header) 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. */ current_format = archive_format; - print_header (); + print_header (st, block_ordinal); } - set_next_block_after (header); + header = write_extended (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) -{ - int counter; - - for (counter = 0; counter < BLOCKSIZE; counter++) - if (buffer[counter] != '\0') - return 0; - return 1; -} - -/*---. -| ? | -`---*/ -static void -init_sparsearray (void) +void +pad_archive (off_t size_left) { - int counter; - - sp_array_size = 10; - - /* Make room for our scratch space -- initially is 10 elts long. */ - - sparsearray = xmalloc (sp_array_size * sizeof (struct sp_array)); - for (counter = 0; counter < sp_array_size; counter++) + union block *blk; + while (size_left > 0) { - sparsearray[counter].offset = 0; - sparsearray[counter].numbytes = 0; + save_sizeleft = size_left; + 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) -{ - int i; - off_t s = 0; - for (i = 0; i < sparses; i++) - s += sparsearray[i].numbytes; - return s; -} - -/*-----------------------------------------------------------------------. -| Make one pass over the file NAME, studying where any non-zero data is, | -| that is, how far into the file each instance of data is, and how many | -| bytes are there. Save this information in the sparsearray, which will | -| later be translated into header information. | -`-----------------------------------------------------------------------*/ - -/* There is little point in trimming small amounts of null data at the head - and tail of blocks, only avoid dumping full null blocks. */ - -/* FIXME: this routine might accept bits of algorithmic cleanup, it is - too kludgey for my taste... */ - -static int -deal_with_sparse (char *name, union block *header) +static enum dump_status +dump_regular_file (int fd, struct tar_stat_info *st) { - size_t numbytes = 0; - off_t offset = 0; - int file; - int sparses = 0; - ssize_t count; - char buffer[BLOCKSIZE]; + off_t size_left = st->stat.st_size; + off_t block_ordinal; + union block *blk; - if (archive_format == OLDGNU_FORMAT) - header->oldgnu_header.isextended = 0; + block_ordinal = current_block_ordinal (); + blk = start_header (st); + if (!blk) + return dump_status_fail; - if (file = open (name, O_RDONLY), file < 0) - /* This problem will be caught later on, so just return. */ - return 0; + /* Mark contiguous files, if we support them. */ + if (archive_format != V7_FORMAT && S_ISCTG (st->stat.st_mode)) + blk->header.typeflag = CONTTYPE; - init_sparsearray (); - clear_buffer (buffer); + finish_header (st, blk, block_ordinal); - while (0 < (count = safe_read (file, buffer, sizeof buffer))) + while (size_left > 0) { - /* Realloc the scratch area as necessary. FIXME: should reallocate - only at beginning of a new instance of non-zero data. */ + size_t bufsize, count; - if (sp_array_size <= sparses) + if (multi_volume_option) { - sparsearray = - xrealloc (sparsearray, - 2 * sp_array_size * sizeof (struct sp_array)); - sp_array_size *= 2; + assign_string (&save_name, st->orig_file_name); + save_sizeleft = size_left; + save_totsize = st->stat.st_size; } + blk = find_next_block (); - /* Process one block. */ - - if (count == sizeof buffer) - - if (zero_block_p (buffer)) - { - if (numbytes) - { - sparsearray[sparses++].numbytes = numbytes; - numbytes = 0; - } - } - else - { - if (!numbytes) - sparsearray[sparses].offset = offset; - numbytes += count; - } - - else + bufsize = available_space_after (blk); - /* Since count < sizeof buffer, we have the last bit of the file. */ + 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); + } - 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); - } + 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; + if (count) + set_next_block_after (blk + (bufsize - 1) / BLOCKSIZE); - if (numbytes) - sparsearray[sparses++].numbytes = numbytes; - else - { - sparsearray[sparses].offset = offset - 1; - sparsearray[sparses++].numbytes = 1; + if (count != bufsize) + { + char buf[UINTMAX_STRSIZE_BOUND]; + memset (blk->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", + size_left), + quotearg_colon (st->orig_file_name), + STRINGIFY_BIGINT (size_left, buf))); + if (! ignore_failed_read_option) + exit_status = TAREXIT_FAILURE; + pad_archive (size_left - (bufsize-count)); + return dump_status_short; + } } - - close (file); - return count < 0 ? 0 : sparses; + return dump_status_ok; } -/*---. -| ? | -`---*/ - -static int -finish_sparse_file (int file, off_t *sizeleft, off_t fullsize, char *name) +static void +dump_regular_finish (int fd, struct tar_stat_info *st, + struct timespec original_ctime) { - union block *start; - size_t bufsize; - int sparses = 0; - ssize_t count; - - while (*sizeleft > 0) + if (fd >= 0) { - 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) + struct stat final_stat; + if (fstat (fd, &final_stat) != 0) { - (ignore_failed_read_option ? seek_warn_details : seek_error_details) - (name, sparsearray[sparses - 1].offset); - break; + stat_diag (st->orig_file_name); } - - /* If the number of bytes to be written here exceeds the size of - the temporary buffer, do it in steps. */ - - while (bufsize > BLOCKSIZE) + else if (final_stat.st_ctime != original_ctime.tv_sec + || (get_stat_ctime (&final_stat).tv_nsec + != original_ctime.tv_nsec)) { - count = safe_read (file, start->buffer, BLOCKSIZE); - if (count < 0) - { - (ignore_failed_read_option - ? read_warn_details - : read_error_details) - (name, fullsize - *sizeleft, bufsize); - return 1; - } - bufsize -= count; - *sizeleft -= count; - set_next_block_after (start); - start = find_next_block (); - memset (start->buffer, 0, BLOCKSIZE); + WARN ((0, 0, _("%s: file changed as we read it"), + quotearg_colon (st->orig_file_name))); } - - { - char buffer[BLOCKSIZE]; - - clear_buffer (buffer); - count = safe_read (file, buffer, bufsize); - memcpy (start->buffer, buffer, BLOCKSIZE); - } - - if (count < 0) + if (close (fd) != 0) { - (ignore_failed_read_option - ? read_warn_details - : read_error_details) - (name, fullsize - *sizeleft, bufsize); - return 1; + close_diag (st->orig_file_name); } - - *sizeleft -= count; - set_next_block_after (start); } - free (sparsearray); -#if 0 - set_next_block_after (start + (count - 1) / BLOCKSIZE); -#endif - return 0; + if (remove_files_option) + { + if (unlink (st->orig_file_name) == -1) + unlink_error (st->orig_file_name); + } } - -/* Main functions of this module. */ -/*---. -| ? | -`---*/ +/* Look in directory DIRNAME for a cache directory tag file + with the magic name "CACHEDIR.TAG" and a standard 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. */ -void -create_archive (void) +#define CACHEDIR_SIGNATURE "Signature: 8a477f597d28d172789f06886806bc55" +#define CACHEDIR_SIGNATURE_SIZE (sizeof CACHEDIR_SIGNATURE - 1) + +static bool +check_cache_directory (char *dirname) { - char *p; + static char tagname[] = "CACHEDIR.TAG"; + char *tagpath; + int fd; + int tag_present = false; - open_archive (ACCESS_WRITE); + tagpath = xmalloc (strlen (dirname) + strlen (tagname) + 1); + strcpy (tagpath, dirname); + strcat (tagpath, tagname); - if (incremental_option) + fd = open (tagpath, O_RDONLY); + if (fd >= 0) { - char *buffer = xmalloc (PATH_MAX); - const char *q; - char *bufp; + static char tagbuf[CACHEDIR_SIGNATURE_SIZE]; - collect_and_sort_names (); - - while (p = name_from_list (), p) - if (!excluded_name (p)) - dump_file (p, -1, (dev_t) 0); + if (read (fd, tagbuf, CACHEDIR_SIGNATURE_SIZE) + == CACHEDIR_SIGNATURE_SIZE + && memcmp (tagbuf, CACHEDIR_SIGNATURE, CACHEDIR_SIGNATURE_SIZE) == 0) + tag_present = true; - blank_name_list (); - while (p = name_from_list (), p) - if (!excluded_name (p)) - { - strcpy (buffer, p); - if (p[strlen (p) - 1] != '/') - strcat (buffer, "/"); - bufp = buffer + strlen (buffer); - q = gnu_list_name->dir_contents; - if (q) - for (; *q; q += strlen (q) + 1) - if (*q == 'Y') - { - strcpy (bufp, q + 1); - dump_file (buffer, -1, (dev_t) 0); - } - } - free (buffer); - } - else - { - while (p = name_next (1), p) - if (!excluded_name (p)) - dump_file (p, 1, (dev_t) 0); + close (fd); } - write_eot (); - close_archive (); + free (tagpath); - if (listed_incremental_option) - write_directory_file (); + return tag_present; } -/* Dump a single file, recursing on directories. P is the file name - to dump. TOP_LEVEL tells whether this is a top-level call; zero - means no, positive means yes, and negative means an incremental - dump. PARENT_DEVICE is the device of P's - parent directory; it is examined only if TOP_LEVEL is zero. - - Set global CURRENT_STAT to stat output for this file. */ - -/* FIXME: One should make sure that for *every* path leading to setting - exit_status to failure, a clear diagnostic has been issued. */ - -void -dump_file (char *p, int top_level, dev_t parent_device) +static void +dump_dir0 (char *directory, + struct tar_stat_info *st, int top_level, dev_t parent_device) { - union block *header; - char type; - union block *exhdr; - char save_typeflag; - time_t original_ctime; - struct utimbuf restore_times; + dev_t our_device = st->stat.st_dev; - /* FIXME: `header' and `sparses' 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) != 0) + if (!is_avoided_name (st->orig_file_name)) { - if (ignore_failed_read_option) - stat_warn (p); - else - stat_error (p); - return; - } + union block *blk = NULL; + off_t block_ordinal = current_block_ordinal (); + st->stat.st_size = 0; /* force 0 size on dir */ - original_ctime = current_stat.st_ctime; - restore_times.actime = current_stat.st_atime; - restore_times.modtime = current_stat.st_mtime; + blk = start_header (st); + if (!blk) + return; -#ifdef S_ISHIDDEN - if (S_ISHIDDEN (current_stat.st_mode)) - { - char *new = (char *) alloca (strlen (p) + 2); - if (new) - { - strcpy (new, p); - strcat (new, "@"); - p = new; - } - } -#endif + if (incremental_option) + blk->header.typeflag = GNUTYPE_DUMPDIR; + else /* if (standard_option) */ + blk->header.typeflag = DIRTYPE; - /* See if we want only new files, and check if this one is too old to - put in the archive. */ + /* If we're gnudumping, we aren't done yet so don't close it. */ - if ((0 < top_level || !incremental_option) - && !S_ISDIR (current_stat.st_mode) - && current_stat.st_mtime < newer_mtime_option - && (!after_date_option || current_stat.st_ctime < newer_ctime_option)) - { - if (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.st_dev == ar_dev && current_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.st_mode)) - { - DIR *directory; - struct dirent *entry; - char *namebuf; - size_t buflen; - size_t len; - dev_t our_device = current_stat.st_dev; - - errno = 0; - - directory = opendir (p); - if (! directory) + if (!incremental_option) + finish_header (st, blk, block_ordinal); + else if (gnu_list_name->dir_contents) { - if (ignore_failed_read_option) - opendir_warn (p); - else - opendir_error (p); - return; - } - - /* Build new prototype name. Ensure exactly one trailing slash. */ - - 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'; - - 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. */ - - current_stat.st_size = 0; /* force 0 size on dir */ - - /* FIXME: If people could really read standard archives, this - should be: - - header - = start_header (standard_option ? p : namebuf, ¤t_stat); - - but since they'd interpret DIRTYPE blocks as regular - files, we'd better put the / on the name. */ - - header = start_header (namebuf, ¤t_stat); - - if (incremental_option) - header->header.typeflag = GNUTYPE_DUMPDIR; - else /* if (standard_option) */ - header->header.typeflag = DIRTYPE; - - /* If we're gnudumping, we aren't done yet so don't close it. */ - - if (!incremental_option) - finish_header (header); /* done with directory header */ - } - - if (incremental_option && gnu_list_name->dir_contents) - { - off_t sizeleft; + off_t size_left; off_t totsize; size_t bufsize; - union block *start; ssize_t count; const char *buffer, *p_buffer; + block_ordinal = current_block_ordinal (); buffer = gnu_list_name->dir_contents; /* FOO */ totsize = 0; - for (p_buffer = buffer; p_buffer && *p_buffer;) - { - size_t tmp; - - tmp = strlen (p_buffer) + 1; - totsize += tmp; - p_buffer += tmp; - } + 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); + OFF_TO_CHARS (totsize, blk->header.size); + finish_header (st, blk, block_ordinal); p_buffer = buffer; - sizeleft = totsize; - while (sizeleft > 0) + size_left = totsize; + while (size_left > 0) { if (multi_volume_option) { - assign_string (&save_name, p); - save_sizeleft = sizeleft; + assign_string (&save_name, st->orig_file_name); + save_sizeleft = size_left; save_totsize = totsize; } - start = find_next_block (); - bufsize = available_space_after (start); - if (sizeleft < bufsize) + blk = find_next_block (); + bufsize = available_space_after (blk); + if (size_left < bufsize) { - bufsize = sizeleft; + bufsize = size_left; count = bufsize % BLOCKSIZE; if (count) - memset (start->buffer + sizeleft, 0, BLOCKSIZE - count); + memset (blk->buffer + size_left, 0, BLOCKSIZE - count); } - memcpy (start->buffer, p_buffer, bufsize); - sizeleft -= bufsize; + memcpy (blk->buffer, p_buffer, bufsize); + size_left -= bufsize; p_buffer += bufsize; - set_next_block_after (start + (bufsize - 1) / BLOCKSIZE); + set_next_block_after (blk + (bufsize - 1) / BLOCKSIZE); } if (multi_volume_option) assign_string (&save_name, 0); - goto finish_dir; + 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. */ + if (!recursion_option) + return; - if (no_recurse_option) - goto finish_dir; + if (one_file_system_option + && !top_level + && parent_device != st->stat.st_dev) + { + if (verbose_option) + WARN ((0, 0, + _("%s: file is on a different filesystem; not dumped"), + quotearg_colon (st->orig_file_name))); + 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 (exclude_caches_option + && check_cache_directory(st->orig_file_name)) + { + if (verbose_option) + WARN ((0, 0, + _("%s: contains a cache directory tag; not dumped"), + quotearg_colon (st->orig_file_name))); + return; + } - if (one_file_system_option && !top_level - && parent_device != current_stat.st_dev) - { - if (verbose_option) - WARN ((0, 0, - _("%s: file is on a different filesystem; not dumped"), - quotearg_colon (p))); - goto finish_dir; - } + { + char const *entry; + size_t entry_len; + char *name_buf = xstrdup (st->orig_file_name); + size_t name_size = strlen (name_buf); + size_t name_len = name_size; - /* Now output all the files in the directory. */ + /* Now output all the files in the directory. */ + /* FIXME: Should speed this up by cd-ing into the dir. */ - /* FIXME: Should speed this up by cd-ing into the dir. */ + 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)) + dump_file (name_buf, 0, our_device); + } - while (errno = 0, (entry = readdir (directory))) - { - /* Skip `.', `..', and excluded file names. */ + free (name_buf); + } +} - if (is_dot_or_dotdot (entry->d_name)) - continue; +/* 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 ((int) NAMLEN (entry) + len >= buflen) - { - buflen = len + NAMLEN (entry); - namebuf = xrealloc (namebuf, buflen + 1); - } - strcpy (namebuf + len, entry->d_name); - if (!excluded_name (namebuf)) - dump_file (namebuf, 0, our_device); - } +static bool +dump_dir (struct tar_stat_info *st, int top_level, dev_t parent_device) +{ + char *directory; - if (errno) - { - if (ignore_failed_read_option) - readdir_warn (p); - else - readdir_error (p); - } + directory = savedir (st->orig_file_name); + if (!directory) + { + savedir_diag (st->orig_file_name); + return false; + } - finish_dir: - if (closedir (directory) != 0) - { - if (ignore_failed_read_option) - closedir_warn (p); - else - closedir_error (p); - } + ensure_slash (&st->orig_file_name); + ensure_slash (&st->file_name); - free (namebuf); - if (atime_preserve_option) - utime (p, &restore_times); - return; - } - else if (is_avoided_name (p)) - return; - else - { - /* Check for multiple links. + dump_dir0 (directory, st, top_level, parent_device); - 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. */ + free (directory); + return true; +} - if (1 < current_stat.st_nlink) - { - struct link *lp; + +/* Main functions of this module. */ - /* FIXME: First quick and dirty. Hashing, etc later. */ +void +create_archive (void) +{ + char *p; - 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; + open_archive (ACCESS_WRITE); + xheader_write_global (); + + if (incremental_option) + { + size_t buffer_size = 1000; + char *buffer = xmalloc (buffer_size); + const char *q; - /* We found a link. */ + collect_and_sort_names (); - while (!absolute_names_option && *link_name == '/') - { - static int warned_once; - if (!warned_once) - { - warned_once = 1; - WARN ((0, 0, - _("Removing leading `/' from link names"))); - } - link_name++; - } - if (strlen (link_name) >= NAME_FIELD_SIZE) - write_long (link_name, GNUTYPE_LONGLINK); - assign_string (¤t_link_name, link_name); + while ((p = name_from_list ()) != NULL) + if (!excluded_name (p)) + dump_file (p, -1, (dev_t) 0); + + blank_name_list (); + while ((p = name_from_list ()) != NULL) + if (!excluded_name (p)) + { + size_t plen = strlen (p); + if (buffer_size <= plen) + { + while ((buffer_size *= 2) <= plen) + continue; + buffer = xrealloc (buffer, buffer_size); + } + memcpy (buffer, p, plen); + if (! ISSLASH (buffer[plen - 1])) + buffer[plen++] = '/'; + q = gnu_list_name->dir_contents; + if (q) + while (*q) + { + size_t qlen = strlen (q); + if (*q == 'Y') + { + if (buffer_size < plen + qlen) + { + while ((buffer_size *=2 ) < plen + qlen) + continue; + buffer = xrealloc (buffer, buffer_size); + } + strcpy (buffer + plen, q + 1); + dump_file (buffer, -1, (dev_t) 0); + } + q += qlen + 1; + } + } + free (buffer); + } + else + { + while ((p = name_next (1)) != NULL) + if (!excluded_name (p)) + dump_file (p, 1, (dev_t) 0); + } - current_stat.st_size = 0; - header = start_header (p, ¤t_stat); - strncpy (header->header.linkname, - link_name, NAME_FIELD_SIZE); + write_eot (); + close_archive (); - /* Force null truncated. */ + if (listed_incremental_option) + write_directory_file (); +} - header->header.linkname[NAME_FIELD_SIZE - 1] = 0; - header->header.typeflag = LNKTYPE; - finish_header (header); +/* 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; +} - /* FIXME: Maybe remove from list after all links found? */ +/* Compare two links for equality. */ +static bool +compare_links (void const *entry1, void const *entry2) +{ + struct link const *link1 = entry1; + struct link const *link2 = entry2; + return ((link1->dev ^ link2->dev) | (link1->ino ^ link2->ino)) == 0; +} - if (remove_files_option) - if (unlink (p) == -1) - unlink_error (p); +static void +unknown_file_error (char *p) +{ + WARN ((0, 0, _("%s: Unknown file type; file ignored"), + quotearg_colon (p))); + if (!ignore_failed_read_option) + exit_status = TAREXIT_FAILURE; +} - /* We dumped it. */ - return; - } + +/* Handling of hard links */ - /* Not found. Add it to the list of possible 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; - lp = xmalloc (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; - } +/* Try to dump stat as a hard link to another file in the archive. If + succeeded returns true */ +static bool +dump_hard_link (struct tar_stat_info *st) +{ + if (link_table && st->stat.st_nlink > 1) + { + struct link lp; + struct link *duplicate; + off_t block_ordinal; + union block *blk; - /* This is not a link to a previously dumped file, so dump it. */ + lp.ino = st->stat.st_ino; + lp.dev = st->stat.st_dev; - if (S_ISREG (current_stat.st_mode) - || S_ISCTG (current_stat.st_mode)) + if ((duplicate = hash_lookup (link_table, &lp))) { - int f; /* file descriptor */ - size_t bufsize; - ssize_t count; - off_t sizeleft; - union block *start; - int header_moved; - char isextended = 0; - int sparses; + /* We found a link. */ + char const *link_name = safer_name_suffix (duplicate->name, true, + absolute_names_option); - header_moved = 0; + duplicate->nlink--; - if (sparse_option) - { - /* Check the size of the file against the number of blocks - allocated for it, counting both data and indirect blocks. - If there is a smaller number of blocks that would be - necessary to accommodate a file of this size, this is safe - to say that we have a sparse file: at least one of those - blocks in the file is just a useless hole. For sparse - files not having more hole blocks than indirect blocks, the - sparseness will go undetected. */ - - /* Bruno Haible sent me these statistics for Linux. It seems - that some filesystems count indirect blocks in st_blocks, - while others do not seem to: - - minix-fs tar: size=7205, st_blocks=18 and ST_NBLOCKS=18 - extfs tar: size=7205, st_blocks=18 and ST_NBLOCKS=18 - ext2fs tar: size=7205, st_blocks=16 and ST_NBLOCKS=16 - msdos-fs tar: size=7205, st_blocks=16 and ST_NBLOCKS=16 - - Dick Streefland reports the previous numbers as misleading, - because ext2fs use 12 direct blocks, while minix-fs uses only - 6 direct blocks. Dick gets: - - ext2 size=20480 ls listed blocks=21 - minix size=20480 ls listed blocks=21 - msdos size=20480 ls listed blocks=20 - - It seems that indirect blocks *are* included in st_blocks. - The minix filesystem does not account for phantom blocks in - st_blocks, so `du' and `ls -s' give wrong results. So, the - --sparse option would not work on a minix filesystem. */ - - if (ST_NBLOCKS (current_stat) - < (current_stat.st_size / ST_NBLOCKSIZE - + (current_stat.st_size % ST_NBLOCKSIZE != 0))) - { - int counter; + 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); - header = start_header (p, ¤t_stat); - header->header.typeflag = GNUTYPE_SPARSE; - header_moved = 1; + st->stat.st_size = 0; + blk = start_header (st); + if (!blk) + return true; + tar_copy_str (blk->header.linkname, link_name, NAME_FIELD_SIZE); - /* 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. */ + blk->header.typeflag = LNKTYPE; + finish_header (st, blk, block_ordinal); - sparses = deal_with_sparse (p, header); + if (remove_files_option && unlink (st->orig_file_name) != 0) + unlink_error (st->orig_file_name); - /* See if we'll need an extended header later. */ + return true; + } + } + return false; +} - if (SPARSES_IN_OLDGNU_HEADER < sparses) - header->oldgnu_header.isextended = 1; +static void +file_count_links (struct tar_stat_info *st) +{ + if (st->stat.st_nlink > 1) + { + struct link *duplicate; + struct link *lp = xmalloc (offsetof (struct link, name) + + strlen (st->orig_file_name) + 1); + lp->ino = st->stat.st_ino; + lp->dev = st->stat.st_dev; + lp->nlink = st->stat.st_nlink; + strcpy (lp->name, st->orig_file_name); + + 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--; + } +} - /* 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. */ +/* 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; - OFF_TO_CHARS (current_stat.st_size, - header->oldgnu_header.realsize); + if (!link_table) + 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. */ + 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"), quote (lp->name))); + } + } +} - current_stat.st_size = find_new_file_size (sparses); - OFF_TO_CHARS (current_stat.st_size, header->header.size); - for (counter = 0; - counter < sparses && counter < SPARSES_IN_OLDGNU_HEADER; - counter++) - { - OFF_TO_CHARS (sparsearray[counter].offset, - header->oldgnu_header.sp[counter].offset); - SIZE_TO_CHARS (sparsearray[counter].numbytes, - header->oldgnu_header.sp[counter].numbytes); - } - } - } - else - sparses = SPARSES_IN_OLDGNU_HEADER; +/* 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. */ - sizeleft = current_stat.st_size; +/* FIXME: One should make sure that for *every* path leading to setting + exit_status to failure, a clear diagnostic has been issued. */ - /* Don't bother opening empty, world readable files. Also do not open - files when archive is meant for /dev/null. */ +static void +dump_file0 (struct tar_stat_info *st, char *p, + int top_level, dev_t parent_device) +{ + union block *header; + char type; + struct timespec original_ctime; + struct timespec restore_times[2]; + off_t block_ordinal = -1; - if (dev_null_output - || (sizeleft == 0 - && MODE_R == (MODE_R & current_stat.st_mode))) - f = -1; - else - { - f = open (p, O_RDONLY | O_BINARY); - if (f < 0) - { - if (! top_level && errno == ENOENT) - WARN ((0, 0, _("%s: File removed before we read it"), - quotearg_colon (p))); - else - (ignore_failed_read_option ? open_warn : open_error) (p); - return; - } - } + if (interactive_option && !confirm ("add", p)) + return; - /* If the file is sparse, we've already taken care of this. */ + assign_string (&st->orig_file_name, p); + assign_string (&st->file_name, + safer_name_suffix (p, false, absolute_names_option)); - if (!header_moved) - header = start_header (p, ¤t_stat); + if (deref_stat (dereference_option, p, &st->stat) != 0) + { + stat_diag (p); + return; + } + st->archive_file_size = st->stat.st_size; + original_ctime = get_stat_ctime (&st->stat); + restore_times[0] = get_stat_atime (&st->stat); + restore_times[1] = get_stat_mtime (&st->stat); - /* Mark contiguous files, if we support them. */ +#ifdef S_ISHIDDEN + if (S_ISHIDDEN (st->stat.st_mode)) + { + char *new = (char *) alloca (strlen (p) + 2); + if (new) + { + strcpy (new, p); + strcat (new, "@"); + p = new; + } + } +#endif - if (archive_format != V7_FORMAT && S_ISCTG (current_stat.st_mode)) - header->header.typeflag = CONTTYPE; + /* See if we want only new files, and check if this one is too old to + put in the archive. - isextended = header->oldgnu_header.isextended; - save_typeflag = header->header.typeflag; - finish_header (header); - if (isextended) - { - int counter; - /* static */ int index_offset = SPARSES_IN_OLDGNU_HEADER; + This check is omitted if incremental_option is set *and* the + requested file is not explicitely listed in the command line. */ - extend: - exhdr = find_next_block (); - memset (exhdr->buffer, 0, BLOCKSIZE); - for (counter = 0; counter < SPARSES_IN_SPARSE_HEADER; counter++) - { - if (sparses <= counter + index_offset) - break; + if (!(incremental_option && !is_individual_file (p)) + && !S_ISDIR (st->stat.st_mode) + && OLDER_STAT_TIME (st->stat, m) + && (!after_date_option || OLDER_STAT_TIME (st->stat, c))) + { + if (!incremental_option && verbose_option) + WARN ((0, 0, _("%s: file is unchanged; not dumped"), + quotearg_colon (p))); + return; + } - SIZE_TO_CHARS (sparsearray[counter + index_offset].numbytes, - exhdr->sparse_header.sp[counter].numbytes); - OFF_TO_CHARS (sparsearray[counter + index_offset].offset, - exhdr->sparse_header.sp[counter].offset); - } - set_next_block_after (exhdr); - if (counter + index_offset < sparses) - { - index_offset += counter; - exhdr->sparse_header.isextended = 1; - goto extend; - } + /* See if we are trying to dump the archive. */ + if (sys_file_is_archive (st)) + { + WARN ((0, 0, _("%s: file is the archive; not dumped"), + quotearg_colon (p))); + return; + } - } - if (save_typeflag == GNUTYPE_SPARSE) - { - if (f < 0 - || finish_sparse_file (f, &sizeleft, - current_stat.st_size, p)) - goto padit; - } - else - while (sizeleft > 0) - { - if (multi_volume_option) - { - assign_string (&save_name, p); - save_sizeleft = sizeleft; - save_totsize = current_stat.st_size; - } - start = find_next_block (); - - bufsize = available_space_after (start); - - if (sizeleft < bufsize) - { - /* Last read -- zero out area beyond. */ - - bufsize = sizeleft; - count = bufsize % BLOCKSIZE; - if (count) - memset (start->buffer + sizeleft, 0, BLOCKSIZE - count); - } - if (f < 0) - count = bufsize; - else - count = safe_read (f, start->buffer, bufsize); - if (count < 0) - { - (ignore_failed_read_option - ? read_warn_details - : read_error_details) - (p, current_stat.st_size - sizeleft, bufsize); - goto padit; - } - sizeleft -= count; - - /* This is nonportable (the type of set_next_block_after's arg). */ - - set_next_block_after (start + (count - 1) / BLOCKSIZE); - - if (count == bufsize) - continue; - else - { - char buf[UINTMAX_STRSIZE_BOUND]; - WARN ((0, 0, - _("%s: File shrank by %s bytes; padding with zeros"), - quotearg_colon (p), - STRINGIFY_BIGINT (sizeleft, buf))); - if (! ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; - goto padit; /* short read */ - } - } + if (is_avoided_name (p)) + return; + if (S_ISDIR (st->stat.st_mode)) + { + dump_dir (st, top_level, parent_device); + if (atime_preserve_option) + utimens (p, restore_times); + return; + } + else + { + /* Check for multiple links. */ + if (dump_hard_link (st)) + return; - if (multi_volume_option) - assign_string (&save_name, 0); + /* This is not a link to a previously dumped file, so dump it. */ - if (f >= 0) + if (S_ISREG (st->stat.st_mode) + || S_ISCTG (st->stat.st_mode)) + { + int fd; + enum dump_status status; + + if (file_dumpable_p (st)) { - 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) + fd = open (st->orig_file_name, + O_RDONLY | O_BINARY); + if (fd < 0) { - char const *qp = quotearg_colon (p); - WARN ((0, 0, _("%s: file changed as we read it"), qp)); - if (! ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; - } - if (close (f) != 0) - { - if (ignore_failed_read_option) - close_warn (p); + if (!top_level && errno == ENOENT) + WARN ((0, 0, _("%s: File removed before we read it"), + quotearg_colon (st->orig_file_name))); else - close_error (p); + open_diag (st->orig_file_name); + return; } - if (atime_preserve_option) - utime (p, &restore_times); - } - if (remove_files_option) - { - if (unlink (p) == -1) - unlink_error (p); } - return; - - /* File shrunk or gave error, pad out tape to match the size we - specified in the header. */ + else + fd = -1; - padit: - while (sizeleft > 0) + if (sparse_option && sparse_file_p (st)) { - save_sizeleft = sizeleft; - start = find_next_block (); - memset (start->buffer, 0, BLOCKSIZE); - set_next_block_after (start); - sizeleft -= BLOCKSIZE; + status = sparse_dump_file (fd, st); + if (status == dump_status_not_implemented) + status = dump_regular_file (fd, st); } - if (multi_volume_option) - assign_string (&save_name, 0); - if (f >= 0) + else + status = dump_regular_file (fd, st); + + switch (status) { - close (f); - if (atime_preserve_option) - utime (p, &restore_times); + case dump_status_ok: + if (multi_volume_option) + assign_string (&save_name, 0); + dump_regular_finish (fd, st, original_ctime); + break; + + case dump_status_short: + if (multi_volume_option) + assign_string (&save_name, 0); + close (fd); + break; + + case dump_status_fail: + close (fd); + return; + + case dump_status_not_implemented: + abort (); } + + if (atime_preserve_option) + utimens (st->orig_file_name, restore_times); + file_count_links (st); return; } #ifdef HAVE_READLINK - else if (S_ISLNK (current_stat.st_mode)) + else if (S_ISLNK (st->stat.st_mode)) { + char *buffer; int size; - char *buffer = (char *) alloca (PATH_MAX + 1); - - size = readlink (p, buffer, PATH_MAX + 1); + size_t linklen = st->stat.st_size; + if (linklen != st->stat.st_size || linklen + 1 == 0) + xalloc_die (); + buffer = (char *) alloca (linklen + 1); + size = readlink (p, buffer, linklen + 1); if (size < 0) { - if (ignore_failed_read_option) - readlink_warn (p); - else - readlink_error (p); + readlink_diag (p); return; } buffer[size] = '\0'; - if (size >= NAME_FIELD_SIZE) - write_long (buffer, GNUTYPE_LONGLINK); - assign_string (¤t_link_name, buffer); - - current_stat.st_size = 0; /* force 0 size on symlink */ - header = start_header (p, ¤t_stat); - strncpy (header->header.linkname, buffer, NAME_FIELD_SIZE); - header->header.linkname[NAME_FIELD_SIZE - 1] = '\0'; + assign_string (&st->link_name, buffer); + 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, buffer, NAME_FIELD_SIZE); header->header.typeflag = SYMTYPE; - finish_header (header); /* nothing more to do to it */ + finish_header (st, header, block_ordinal); + /* nothing more to do to it */ + if (remove_files_option) { if (unlink (p) == -1) unlink_error (p); } + file_count_links (st); return; } #endif - else if (S_ISCHR (current_stat.st_mode)) + else if (S_ISCHR (st->stat.st_mode)) type = CHRTYPE; - else if (S_ISBLK (current_stat.st_mode)) + else if (S_ISBLK (st->stat.st_mode)) type = BLKTYPE; - else if (S_ISFIFO (current_stat.st_mode)) + else if (S_ISFIFO (st->stat.st_mode)) type = FIFOTYPE; - else if (S_ISSOCK (current_stat.st_mode)) + else if (S_ISSOCK (st->stat.st_mode)) { WARN ((0, 0, _("%s: socket ignored"), quotearg_colon (p))); return; } - else if (S_ISDOOR (current_stat.st_mode)) + else if (S_ISDOOR (st->stat.st_mode)) { WARN ((0, 0, _("%s: door ignored"), quotearg_colon (p))); return; } else - goto unknown; + { + unknown_file_error (p); + return; + } } if (archive_format == V7_FORMAT) - goto unknown; + { + unknown_file_error (p); + return; + } - current_stat.st_size = 0; /* force 0 size */ - header = start_header (p, ¤t_stat); + 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 (type != FIFOTYPE) { - MAJOR_TO_CHARS (major (current_stat.st_rdev), header->header.devmajor); - MINOR_TO_CHARS (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); } - finish_header (header); + finish_header (st, header, block_ordinal); if (remove_files_option) { if (unlink (p) == -1) unlink_error (p); } - return; +} -unknown: - WARN ((0, 0, _("%s: Unknown file type; file ignored"), - quotearg_colon (p))); - if (! ignore_failed_read_option) - exit_status = TAREXIT_FAILURE; +void +dump_file (char *p, int top_level, dev_t parent_device) +{ + struct tar_stat_info st; + tar_stat_init (&st); + dump_file0 (&st, p, top_level, parent_device); + tar_stat_destroy (&st); }