mirror of
https://github.com/sheumann/hush.git
synced 2024-12-30 20:29:21 +00:00
8131eea3dc
function old new delta buffer_fill_and_print 179 196 +17 fflush_all - 9 +9 spawn 87 92 +5 rtcwake_main 455 453 -2 ... alarm_intr 93 84 -9 readcmd 1072 1062 -10 bb_ask 345 333 -12 more_main 845 832 -13 flush_stdout_stderr 42 23 -19 xfflush_stdout 27 - -27 flush_stderr 30 - -30 ------------------------------------------------------------------------------ (add/remove: 1/2 grow/shrink: 2/50 up/down: 31/-397) Total: -366 bytes Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
1302 lines
32 KiB
C
1302 lines
32 KiB
C
/* vi: set sw=4 ts=4: */
|
|
/*
|
|
* fsck.c - a file system consistency checker for Linux.
|
|
*
|
|
* (C) 1991, 1992 Linus Torvalds.
|
|
*
|
|
* Licensed under GPLv2, see file LICENSE in this tarball for details.
|
|
*/
|
|
|
|
/*
|
|
* 09.11.91 - made the first rudimentary functions
|
|
*
|
|
* 10.11.91 - updated, does checking, no repairs yet.
|
|
* Sent out to the mailing-list for testing.
|
|
*
|
|
* 14.11.91 - Testing seems to have gone well. Added some
|
|
* correction-code, and changed some functions.
|
|
*
|
|
* 15.11.91 - More correction code. Hopefully it notices most
|
|
* cases now, and tries to do something about them.
|
|
*
|
|
* 16.11.91 - More corrections (thanks to Mika Jalava). Most
|
|
* things seem to work now. Yeah, sure.
|
|
*
|
|
*
|
|
* 19.04.92 - Had to start over again from this old version, as a
|
|
* kernel bug ate my enhanced fsck in february.
|
|
*
|
|
* 28.02.93 - added support for different directory entry sizes..
|
|
*
|
|
* Sat Mar 6 18:59:42 1993, faith@cs.unc.edu: Output namelen with
|
|
* superblock information
|
|
*
|
|
* Sat Oct 9 11:17:11 1993, faith@cs.unc.edu: make exit status conform
|
|
* to that required by fsutil
|
|
*
|
|
* Mon Jan 3 11:06:52 1994 - Dr. Wettstein (greg%wind.uucp@plains.nodak.edu)
|
|
* Added support for file system valid flag. Also
|
|
* added program_version variable and output of
|
|
* program name and version number when program
|
|
* is executed.
|
|
*
|
|
* 30.10.94 - added support for v2 filesystem
|
|
* (Andreas Schwab, schwab@issan.informatik.uni-dortmund.de)
|
|
*
|
|
* 10.12.94 - added test to prevent checking of mounted fs adapted
|
|
* from Theodore Ts'o's (tytso@athena.mit.edu) e2fsck
|
|
* program. (Daniel Quinlan, quinlan@yggdrasil.com)
|
|
*
|
|
* 01.07.96 - Fixed the v2 fs stuff to use the right #defines and such
|
|
* for modern libcs (janl@math.uio.no, Nicolai Langfeldt)
|
|
*
|
|
* 02.07.96 - Added C bit fiddling routines from rmk@ecs.soton.ac.uk
|
|
* (Russell King). He made them for ARM. It would seem
|
|
* that the ARM is powerful enough to do this in C whereas
|
|
* i386 and m64k must use assembly to get it fast >:-)
|
|
* This should make minix fsck system-independent.
|
|
* (janl@math.uio.no, Nicolai Langfeldt)
|
|
*
|
|
* 04.11.96 - Added minor fixes from Andreas Schwab to avoid compiler
|
|
* warnings. Added mc68k bitops from
|
|
* Joerg Dorchain <dorchain@mpi-sb.mpg.de>.
|
|
*
|
|
* 06.11.96 - Added v2 code submitted by Joerg Dorchain, but written by
|
|
* Andreas Schwab.
|
|
*
|
|
* 1999-02-22 Arkadiusz Mickiewicz <misiek@misiek.eu.org>
|
|
* - added Native Language Support
|
|
*
|
|
*
|
|
* I've had no time to add comments - hopefully the function names
|
|
* are comments enough. As with all file system checkers, this assumes
|
|
* the file system is quiescent - don't use it on a mounted device
|
|
* unless you can be sure nobody is writing to it (and remember that the
|
|
* kernel can write to it when it searches for files).
|
|
*
|
|
* Usage: fsck [-larvsm] device
|
|
* -l for a listing of all the filenames
|
|
* -a for automatic repairs (not implemented)
|
|
* -r for repairs (interactive) (not implemented)
|
|
* -v for verbose (tells how many files)
|
|
* -s for superblock info
|
|
* -m for minix-like "mode not cleared" warnings
|
|
* -f force filesystem check even if filesystem marked as valid
|
|
*
|
|
* The device may be a block device or a image of one, but this isn't
|
|
* enforced (but it's not much fun on a character device :-).
|
|
*/
|
|
|
|
#include <mntent.h>
|
|
#include "libbb.h"
|
|
#include "minix.h"
|
|
|
|
#ifndef BLKGETSIZE
|
|
#define BLKGETSIZE _IO(0x12,96) /* return device size */
|
|
#endif
|
|
|
|
struct BUG_bad_inode_size {
|
|
char BUG_bad_inode1_size[(INODE_SIZE1 * MINIX1_INODES_PER_BLOCK != BLOCK_SIZE) ? -1 : 1];
|
|
#if ENABLE_FEATURE_MINIX2
|
|
char BUG_bad_inode2_size[(INODE_SIZE2 * MINIX2_INODES_PER_BLOCK != BLOCK_SIZE) ? -1 : 1];
|
|
#endif
|
|
};
|
|
|
|
enum {
|
|
#ifdef UNUSED
|
|
MINIX1_LINK_MAX = 250,
|
|
MINIX2_LINK_MAX = 65530,
|
|
MINIX_I_MAP_SLOTS = 8,
|
|
MINIX_Z_MAP_SLOTS = 64,
|
|
MINIX_V1 = 0x0001, /* original minix fs */
|
|
MINIX_V2 = 0x0002, /* minix V2 fs */
|
|
#endif
|
|
MINIX_NAME_MAX = 255, /* # chars in a file name */
|
|
};
|
|
|
|
|
|
#if !ENABLE_FEATURE_MINIX2
|
|
enum { version2 = 0 };
|
|
#endif
|
|
|
|
enum { MAX_DEPTH = 32 };
|
|
|
|
enum { dev_fd = 3 };
|
|
|
|
struct globals {
|
|
#if ENABLE_FEATURE_MINIX2
|
|
smallint version2;
|
|
#endif
|
|
smallint changed; /* is filesystem modified? */
|
|
smallint errors_uncorrected; /* flag if some error was not corrected */
|
|
smallint termios_set;
|
|
smallint dirsize;
|
|
smallint namelen;
|
|
const char *device_name;
|
|
int directory, regular, blockdev, chardev, links, symlinks, total;
|
|
char *inode_buffer;
|
|
|
|
char *inode_map;
|
|
char *zone_map;
|
|
|
|
unsigned char *inode_count;
|
|
unsigned char *zone_count;
|
|
|
|
/* File-name data */
|
|
int name_depth;
|
|
char *name_component[MAX_DEPTH+1];
|
|
|
|
/* Bigger stuff */
|
|
struct termios sv_termios;
|
|
char superblock_buffer[BLOCK_SIZE];
|
|
char add_zone_ind_blk[BLOCK_SIZE];
|
|
char add_zone_dind_blk[BLOCK_SIZE];
|
|
IF_FEATURE_MINIX2(char add_zone_tind_blk[BLOCK_SIZE];)
|
|
char check_file_blk[BLOCK_SIZE];
|
|
|
|
/* File-name data */
|
|
char current_name[MAX_DEPTH * MINIX_NAME_MAX];
|
|
};
|
|
|
|
#define G (*ptr_to_globals)
|
|
#if ENABLE_FEATURE_MINIX2
|
|
#define version2 (G.version2 )
|
|
#endif
|
|
#define changed (G.changed )
|
|
#define errors_uncorrected (G.errors_uncorrected )
|
|
#define termios_set (G.termios_set )
|
|
#define dirsize (G.dirsize )
|
|
#define namelen (G.namelen )
|
|
#define device_name (G.device_name )
|
|
#define directory (G.directory )
|
|
#define regular (G.regular )
|
|
#define blockdev (G.blockdev )
|
|
#define chardev (G.chardev )
|
|
#define links (G.links )
|
|
#define symlinks (G.symlinks )
|
|
#define total (G.total )
|
|
#define inode_buffer (G.inode_buffer )
|
|
#define inode_map (G.inode_map )
|
|
#define zone_map (G.zone_map )
|
|
#define inode_count (G.inode_count )
|
|
#define zone_count (G.zone_count )
|
|
#define name_depth (G.name_depth )
|
|
#define name_component (G.name_component )
|
|
#define sv_termios (G.sv_termios )
|
|
#define superblock_buffer (G.superblock_buffer )
|
|
#define add_zone_ind_blk (G.add_zone_ind_blk )
|
|
#define add_zone_dind_blk (G.add_zone_dind_blk )
|
|
#define add_zone_tind_blk (G.add_zone_tind_blk )
|
|
#define check_file_blk (G.check_file_blk )
|
|
#define current_name (G.current_name )
|
|
#define INIT_G() do { \
|
|
SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
|
|
dirsize = 16; \
|
|
namelen = 14; \
|
|
current_name[0] = '/'; \
|
|
/*current_name[1] = '\0';*/ \
|
|
name_component[0] = ¤t_name[0]; \
|
|
} while (0)
|
|
|
|
|
|
#define OPTION_STR "larvsmf"
|
|
enum {
|
|
OPT_l = (1 << 0),
|
|
OPT_a = (1 << 1),
|
|
OPT_r = (1 << 2),
|
|
OPT_v = (1 << 3),
|
|
OPT_s = (1 << 4),
|
|
OPT_w = (1 << 5),
|
|
OPT_f = (1 << 6),
|
|
};
|
|
#define OPT_list (option_mask32 & OPT_l)
|
|
#define OPT_automatic (option_mask32 & OPT_a)
|
|
#define OPT_repair (option_mask32 & OPT_r)
|
|
#define OPT_verbose (option_mask32 & OPT_v)
|
|
#define OPT_show (option_mask32 & OPT_s)
|
|
#define OPT_warn_mode (option_mask32 & OPT_w)
|
|
#define OPT_force (option_mask32 & OPT_f)
|
|
/* non-automatic repairs requested? */
|
|
#define OPT_manual ((option_mask32 & (OPT_a|OPT_r)) == OPT_r)
|
|
|
|
|
|
#define Inode1 (((struct minix1_inode *) inode_buffer)-1)
|
|
#define Inode2 (((struct minix2_inode *) inode_buffer)-1)
|
|
|
|
#define Super (*(struct minix_superblock *)(superblock_buffer))
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
# define ZONES ((unsigned)(version2 ? Super.s_zones : Super.s_nzones))
|
|
#else
|
|
# define ZONES ((unsigned)(Super.s_nzones))
|
|
#endif
|
|
#define INODES ((unsigned)Super.s_ninodes)
|
|
#define IMAPS ((unsigned)Super.s_imap_blocks)
|
|
#define ZMAPS ((unsigned)Super.s_zmap_blocks)
|
|
#define FIRSTZONE ((unsigned)Super.s_firstdatazone)
|
|
#define ZONESIZE ((unsigned)Super.s_log_zone_size)
|
|
#define MAXSIZE ((unsigned)Super.s_max_size)
|
|
#define MAGIC (Super.s_magic)
|
|
|
|
/* gcc likes this more (code is smaller) than macro variant */
|
|
static ALWAYS_INLINE unsigned div_roundup(unsigned size, unsigned n)
|
|
{
|
|
return (size + n-1) / n;
|
|
}
|
|
|
|
#if !ENABLE_FEATURE_MINIX2
|
|
#define INODE_BLOCKS div_roundup(INODES, MINIX1_INODES_PER_BLOCK)
|
|
#else
|
|
#define INODE_BLOCKS div_roundup(INODES, \
|
|
(version2 ? MINIX2_INODES_PER_BLOCK : MINIX1_INODES_PER_BLOCK))
|
|
#endif
|
|
|
|
#define INODE_BUFFER_SIZE (INODE_BLOCKS * BLOCK_SIZE)
|
|
#define NORM_FIRSTZONE (2 + IMAPS + ZMAPS + INODE_BLOCKS)
|
|
|
|
/* Before you ask "where they come from?": */
|
|
/* setbit/clrbit are supplied by sys/param.h */
|
|
|
|
static int minix_bit(const char *a, unsigned i)
|
|
{
|
|
return (a[i >> 3] & (1<<(i & 7)));
|
|
}
|
|
|
|
static void minix_setbit(char *a, unsigned i)
|
|
{
|
|
setbit(a, i);
|
|
changed = 1;
|
|
}
|
|
static void minix_clrbit(char *a, unsigned i)
|
|
{
|
|
clrbit(a, i);
|
|
changed = 1;
|
|
}
|
|
|
|
/* Note: do not assume 0/1, it is 0/nonzero */
|
|
#define zone_in_use(x) (minix_bit(zone_map,(x)-FIRSTZONE+1))
|
|
#define inode_in_use(x) (minix_bit(inode_map,(x)))
|
|
|
|
#define mark_inode(x) (minix_setbit(inode_map,(x)))
|
|
#define unmark_inode(x) (minix_clrbit(inode_map,(x)))
|
|
|
|
#define mark_zone(x) (minix_setbit(zone_map,(x)-FIRSTZONE+1))
|
|
#define unmark_zone(x) (minix_clrbit(zone_map,(x)-FIRSTZONE+1))
|
|
|
|
|
|
static void recursive_check(unsigned ino);
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void recursive_check2(unsigned ino);
|
|
#endif
|
|
|
|
static void die(const char *str) NORETURN;
|
|
static void die(const char *str)
|
|
{
|
|
if (termios_set)
|
|
tcsetattr_stdin_TCSANOW(&sv_termios);
|
|
bb_error_msg_and_die("%s", str);
|
|
}
|
|
|
|
static void push_filename(const char *name)
|
|
{
|
|
// /dir/dir/dir/file
|
|
// ^ ^ ^
|
|
// [0] [1] [2] <-name_component[i]
|
|
if (name_depth < MAX_DEPTH) {
|
|
int len;
|
|
char *p = name_component[name_depth];
|
|
*p++ = '/';
|
|
len = sprintf(p, "%.*s", namelen, name);
|
|
name_component[name_depth + 1] = p + len;
|
|
}
|
|
name_depth++;
|
|
}
|
|
|
|
static void pop_filename(void)
|
|
{
|
|
name_depth--;
|
|
if (name_depth < MAX_DEPTH) {
|
|
*name_component[name_depth] = '\0';
|
|
if (!name_depth) {
|
|
current_name[0] = '/';
|
|
current_name[1] = '\0';
|
|
}
|
|
}
|
|
}
|
|
|
|
static int ask(const char *string, int def)
|
|
{
|
|
int c;
|
|
|
|
if (!OPT_repair) {
|
|
bb_putchar('\n');
|
|
errors_uncorrected = 1;
|
|
return 0;
|
|
}
|
|
if (OPT_automatic) {
|
|
bb_putchar('\n');
|
|
if (!def)
|
|
errors_uncorrected = 1;
|
|
return def;
|
|
}
|
|
printf(def ? "%s (y/n)? " : "%s (n/y)? ", string);
|
|
for (;;) {
|
|
fflush_all();
|
|
c = getchar();
|
|
if (c == EOF) {
|
|
if (!def)
|
|
errors_uncorrected = 1;
|
|
return def;
|
|
}
|
|
if (c == '\n')
|
|
break;
|
|
c |= 0x20; /* tolower */
|
|
if (c == 'y') {
|
|
def = 1;
|
|
break;
|
|
}
|
|
if (c == 'n') {
|
|
def = 0;
|
|
break;
|
|
}
|
|
}
|
|
if (def)
|
|
printf("y\n");
|
|
else {
|
|
printf("n\n");
|
|
errors_uncorrected = 1;
|
|
}
|
|
return def;
|
|
}
|
|
|
|
/*
|
|
* Make certain that we aren't checking a filesystem that is on a
|
|
* mounted partition. Code adapted from e2fsck, Copyright (C) 1993,
|
|
* 1994 Theodore Ts'o. Also licensed under GPL.
|
|
*/
|
|
static void check_mount(void)
|
|
{
|
|
if (find_mount_point(device_name, 0)) {
|
|
int cont;
|
|
#if ENABLE_FEATURE_MTAB_SUPPORT
|
|
/*
|
|
* If the root is mounted read-only, then /etc/mtab is
|
|
* probably not correct; so we won't issue a warning based on
|
|
* it.
|
|
*/
|
|
int fd = open(bb_path_mtab_file, O_RDWR);
|
|
|
|
if (fd < 0 && errno == EROFS)
|
|
return;
|
|
close(fd);
|
|
#endif
|
|
printf("%s is mounted. ", device_name);
|
|
cont = 0;
|
|
if (isatty(0) && isatty(1))
|
|
cont = ask("Do you really want to continue", 0);
|
|
if (!cont) {
|
|
printf("Check aborted\n");
|
|
exit(EXIT_SUCCESS);
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
* check_zone_nr checks to see that *nr is a valid zone nr. If it
|
|
* isn't, it will possibly be repaired. Check_zone_nr sets *corrected
|
|
* if an error was corrected, and returns the zone (0 for no zone
|
|
* or a bad zone-number).
|
|
*/
|
|
static int check_zone_nr2(uint32_t *nr, smallint *corrected)
|
|
{
|
|
const char *msg;
|
|
if (!*nr)
|
|
return 0;
|
|
if (*nr < FIRSTZONE)
|
|
msg = "< FIRSTZONE";
|
|
else if (*nr >= ZONES)
|
|
msg = ">= ZONES";
|
|
else
|
|
return *nr;
|
|
printf("Zone nr %s in file '%s'. ", msg, current_name);
|
|
if (ask("Remove block", 1)) {
|
|
*nr = 0;
|
|
*corrected = 1;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
static int check_zone_nr(uint16_t *nr, smallint *corrected)
|
|
{
|
|
uint32_t nr32 = *nr;
|
|
int r = check_zone_nr2(&nr32, corrected);
|
|
*nr = (uint16_t)nr32;
|
|
return r;
|
|
}
|
|
|
|
/*
|
|
* read-block reads block nr into the buffer at addr.
|
|
*/
|
|
static void read_block(unsigned nr, void *addr)
|
|
{
|
|
if (!nr) {
|
|
memset(addr, 0, BLOCK_SIZE);
|
|
return;
|
|
}
|
|
xlseek(dev_fd, BLOCK_SIZE * nr, SEEK_SET);
|
|
if (BLOCK_SIZE != full_read(dev_fd, addr, BLOCK_SIZE)) {
|
|
printf("%s: bad block %u in file '%s'\n",
|
|
bb_msg_read_error, nr, current_name);
|
|
errors_uncorrected = 1;
|
|
memset(addr, 0, BLOCK_SIZE);
|
|
}
|
|
}
|
|
|
|
/*
|
|
* write_block writes block nr to disk.
|
|
*/
|
|
static void write_block(unsigned nr, void *addr)
|
|
{
|
|
if (!nr)
|
|
return;
|
|
if (nr < FIRSTZONE || nr >= ZONES) {
|
|
printf("Internal error: trying to write bad block\n"
|
|
"Write request ignored\n");
|
|
errors_uncorrected = 1;
|
|
return;
|
|
}
|
|
xlseek(dev_fd, BLOCK_SIZE * nr, SEEK_SET);
|
|
if (BLOCK_SIZE != full_write(dev_fd, addr, BLOCK_SIZE)) {
|
|
printf("%s: bad block %u in file '%s'\n",
|
|
bb_msg_write_error, nr, current_name);
|
|
errors_uncorrected = 1;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* map_block calculates the absolute block nr of a block in a file.
|
|
* It sets 'changed' if the inode has needed changing, and re-writes
|
|
* any indirect blocks with errors.
|
|
*/
|
|
static int map_block(struct minix1_inode *inode, unsigned blknr)
|
|
{
|
|
uint16_t ind[BLOCK_SIZE >> 1];
|
|
int block, result;
|
|
smallint blk_chg;
|
|
|
|
if (blknr < 7)
|
|
return check_zone_nr(inode->i_zone + blknr, &changed);
|
|
blknr -= 7;
|
|
if (blknr < 512) {
|
|
block = check_zone_nr(inode->i_zone + 7, &changed);
|
|
goto common;
|
|
}
|
|
blknr -= 512;
|
|
block = check_zone_nr(inode->i_zone + 8, &changed);
|
|
read_block(block, ind); /* double indirect */
|
|
blk_chg = 0;
|
|
result = check_zone_nr(&ind[blknr / 512], &blk_chg);
|
|
if (blk_chg)
|
|
write_block(block, ind);
|
|
block = result;
|
|
common:
|
|
read_block(block, ind);
|
|
blk_chg = 0;
|
|
result = check_zone_nr(&ind[blknr % 512], &blk_chg);
|
|
if (blk_chg)
|
|
write_block(block, ind);
|
|
return result;
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static int map_block2(struct minix2_inode *inode, unsigned blknr)
|
|
{
|
|
uint32_t ind[BLOCK_SIZE >> 2];
|
|
int block, result;
|
|
smallint blk_chg;
|
|
|
|
if (blknr < 7)
|
|
return check_zone_nr2(inode->i_zone + blknr, &changed);
|
|
blknr -= 7;
|
|
if (blknr < 256) {
|
|
block = check_zone_nr2(inode->i_zone + 7, &changed);
|
|
goto common2;
|
|
}
|
|
blknr -= 256;
|
|
if (blknr < 256 * 256) {
|
|
block = check_zone_nr2(inode->i_zone + 8, &changed);
|
|
goto common1;
|
|
}
|
|
blknr -= 256 * 256;
|
|
block = check_zone_nr2(inode->i_zone + 9, &changed);
|
|
read_block(block, ind); /* triple indirect */
|
|
blk_chg = 0;
|
|
result = check_zone_nr2(&ind[blknr / (256 * 256)], &blk_chg);
|
|
if (blk_chg)
|
|
write_block(block, ind);
|
|
block = result;
|
|
common1:
|
|
read_block(block, ind); /* double indirect */
|
|
blk_chg = 0;
|
|
result = check_zone_nr2(&ind[(blknr / 256) % 256], &blk_chg);
|
|
if (blk_chg)
|
|
write_block(block, ind);
|
|
block = result;
|
|
common2:
|
|
read_block(block, ind);
|
|
blk_chg = 0;
|
|
result = check_zone_nr2(&ind[blknr % 256], &blk_chg);
|
|
if (blk_chg)
|
|
write_block(block, ind);
|
|
return result;
|
|
}
|
|
#endif
|
|
|
|
static void write_superblock(void)
|
|
{
|
|
/*
|
|
* Set the state of the filesystem based on whether or not there
|
|
* are uncorrected errors. The filesystem valid flag is
|
|
* unconditionally set if we get this far.
|
|
*/
|
|
Super.s_state |= MINIX_VALID_FS | MINIX_ERROR_FS;
|
|
if (!errors_uncorrected)
|
|
Super.s_state &= ~MINIX_ERROR_FS;
|
|
|
|
xlseek(dev_fd, BLOCK_SIZE, SEEK_SET);
|
|
if (BLOCK_SIZE != full_write(dev_fd, superblock_buffer, BLOCK_SIZE))
|
|
die("cannot write superblock");
|
|
}
|
|
|
|
static void write_tables(void)
|
|
{
|
|
write_superblock();
|
|
|
|
if (IMAPS * BLOCK_SIZE != write(dev_fd, inode_map, IMAPS * BLOCK_SIZE))
|
|
die("cannot write inode map");
|
|
if (ZMAPS * BLOCK_SIZE != write(dev_fd, zone_map, ZMAPS * BLOCK_SIZE))
|
|
die("cannot write zone map");
|
|
if (INODE_BUFFER_SIZE != write(dev_fd, inode_buffer, INODE_BUFFER_SIZE))
|
|
die("cannot write inodes");
|
|
}
|
|
|
|
static void get_dirsize(void)
|
|
{
|
|
int block;
|
|
char blk[BLOCK_SIZE];
|
|
int size;
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
if (version2)
|
|
block = Inode2[MINIX_ROOT_INO].i_zone[0];
|
|
else
|
|
#endif
|
|
block = Inode1[MINIX_ROOT_INO].i_zone[0];
|
|
read_block(block, blk);
|
|
for (size = 16; size < BLOCK_SIZE; size <<= 1) {
|
|
if (strcmp(blk + size + 2, "..") == 0) {
|
|
dirsize = size;
|
|
namelen = size - 2;
|
|
return;
|
|
}
|
|
}
|
|
/* use defaults */
|
|
}
|
|
|
|
static void read_superblock(void)
|
|
{
|
|
xlseek(dev_fd, BLOCK_SIZE, SEEK_SET);
|
|
if (BLOCK_SIZE != full_read(dev_fd, superblock_buffer, BLOCK_SIZE))
|
|
die("cannot read superblock");
|
|
/* already initialized to:
|
|
namelen = 14;
|
|
dirsize = 16;
|
|
version2 = 0;
|
|
*/
|
|
if (MAGIC == MINIX1_SUPER_MAGIC) {
|
|
} else if (MAGIC == MINIX1_SUPER_MAGIC2) {
|
|
namelen = 30;
|
|
dirsize = 32;
|
|
#if ENABLE_FEATURE_MINIX2
|
|
} else if (MAGIC == MINIX2_SUPER_MAGIC) {
|
|
version2 = 1;
|
|
} else if (MAGIC == MINIX2_SUPER_MAGIC2) {
|
|
namelen = 30;
|
|
dirsize = 32;
|
|
version2 = 1;
|
|
#endif
|
|
} else
|
|
die("bad magic number in superblock");
|
|
if (ZONESIZE != 0 || BLOCK_SIZE != 1024)
|
|
die("only 1k blocks/zones supported");
|
|
if (IMAPS * BLOCK_SIZE * 8 < INODES + 1)
|
|
die("bad s_imap_blocks field in superblock");
|
|
if (ZMAPS * BLOCK_SIZE * 8 < ZONES - FIRSTZONE + 1)
|
|
die("bad s_zmap_blocks field in superblock");
|
|
}
|
|
|
|
static void read_tables(void)
|
|
{
|
|
inode_map = xzalloc(IMAPS * BLOCK_SIZE);
|
|
zone_map = xzalloc(ZMAPS * BLOCK_SIZE);
|
|
inode_buffer = xmalloc(INODE_BUFFER_SIZE);
|
|
inode_count = xmalloc(INODES + 1);
|
|
zone_count = xmalloc(ZONES);
|
|
if (IMAPS * BLOCK_SIZE != read(dev_fd, inode_map, IMAPS * BLOCK_SIZE))
|
|
die("cannot read inode map");
|
|
if (ZMAPS * BLOCK_SIZE != read(dev_fd, zone_map, ZMAPS * BLOCK_SIZE))
|
|
die("cannot read zone map");
|
|
if (INODE_BUFFER_SIZE != read(dev_fd, inode_buffer, INODE_BUFFER_SIZE))
|
|
die("cannot read inodes");
|
|
if (NORM_FIRSTZONE != FIRSTZONE) {
|
|
printf("warning: firstzone!=norm_firstzone\n");
|
|
errors_uncorrected = 1;
|
|
}
|
|
get_dirsize();
|
|
if (OPT_show) {
|
|
printf("%u inodes\n"
|
|
"%u blocks\n"
|
|
"Firstdatazone=%u (%u)\n"
|
|
"Zonesize=%u\n"
|
|
"Maxsize=%u\n"
|
|
"Filesystem state=%u\n"
|
|
"namelen=%u\n\n",
|
|
INODES,
|
|
ZONES,
|
|
FIRSTZONE, NORM_FIRSTZONE,
|
|
BLOCK_SIZE << ZONESIZE,
|
|
MAXSIZE,
|
|
Super.s_state,
|
|
namelen);
|
|
}
|
|
}
|
|
|
|
static void get_inode_common(unsigned nr, uint16_t i_mode)
|
|
{
|
|
total++;
|
|
if (!inode_count[nr]) {
|
|
if (!inode_in_use(nr)) {
|
|
printf("Inode %d is marked as 'unused', but it is used "
|
|
"for file '%s'\n", nr, current_name);
|
|
if (OPT_repair) {
|
|
if (ask("Mark as 'in use'", 1))
|
|
mark_inode(nr);
|
|
else
|
|
errors_uncorrected = 1;
|
|
}
|
|
}
|
|
if (S_ISDIR(i_mode))
|
|
directory++;
|
|
else if (S_ISREG(i_mode))
|
|
regular++;
|
|
else if (S_ISCHR(i_mode))
|
|
chardev++;
|
|
else if (S_ISBLK(i_mode))
|
|
blockdev++;
|
|
else if (S_ISLNK(i_mode))
|
|
symlinks++;
|
|
else if (S_ISSOCK(i_mode));
|
|
else if (S_ISFIFO(i_mode));
|
|
else {
|
|
printf("%s has mode %05o\n", current_name, i_mode);
|
|
}
|
|
} else
|
|
links++;
|
|
if (!++inode_count[nr]) {
|
|
printf("Warning: inode count too big\n");
|
|
inode_count[nr]--;
|
|
errors_uncorrected = 1;
|
|
}
|
|
}
|
|
|
|
static struct minix1_inode *get_inode(unsigned nr)
|
|
{
|
|
struct minix1_inode *inode;
|
|
|
|
if (!nr || nr > INODES)
|
|
return NULL;
|
|
inode = Inode1 + nr;
|
|
get_inode_common(nr, inode->i_mode);
|
|
return inode;
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static struct minix2_inode *get_inode2(unsigned nr)
|
|
{
|
|
struct minix2_inode *inode;
|
|
|
|
if (!nr || nr > INODES)
|
|
return NULL;
|
|
inode = Inode2 + nr;
|
|
get_inode_common(nr, inode->i_mode);
|
|
return inode;
|
|
}
|
|
#endif
|
|
|
|
static void check_root(void)
|
|
{
|
|
struct minix1_inode *inode = Inode1 + MINIX_ROOT_INO;
|
|
|
|
if (!inode || !S_ISDIR(inode->i_mode))
|
|
die("root inode isn't a directory");
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void check_root2(void)
|
|
{
|
|
struct minix2_inode *inode = Inode2 + MINIX_ROOT_INO;
|
|
|
|
if (!inode || !S_ISDIR(inode->i_mode))
|
|
die("root inode isn't a directory");
|
|
}
|
|
#else
|
|
void check_root2(void);
|
|
#endif
|
|
|
|
static int add_zone_common(int block, smallint *corrected)
|
|
{
|
|
if (!block)
|
|
return 0;
|
|
if (zone_count[block]) {
|
|
printf("Already used block is reused in file '%s'. ",
|
|
current_name);
|
|
if (ask("Clear", 1)) {
|
|
block = 0;
|
|
*corrected = 1;
|
|
return -1; /* "please zero out *znr" */
|
|
}
|
|
}
|
|
if (!zone_in_use(block)) {
|
|
printf("Block %d in file '%s' is marked as 'unused'. ",
|
|
block, current_name);
|
|
if (ask("Correct", 1))
|
|
mark_zone(block);
|
|
}
|
|
if (!++zone_count[block])
|
|
zone_count[block]--;
|
|
return block;
|
|
}
|
|
|
|
static int add_zone(uint16_t *znr, smallint *corrected)
|
|
{
|
|
int block;
|
|
|
|
block = check_zone_nr(znr, corrected);
|
|
block = add_zone_common(block, corrected);
|
|
if (block == -1) {
|
|
*znr = 0;
|
|
block = 0;
|
|
}
|
|
return block;
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static int add_zone2(uint32_t *znr, smallint *corrected)
|
|
{
|
|
int block;
|
|
|
|
block = check_zone_nr2(znr, corrected);
|
|
block = add_zone_common(block, corrected);
|
|
if (block == -1) {
|
|
*znr = 0;
|
|
block = 0;
|
|
}
|
|
return block;
|
|
}
|
|
#endif
|
|
|
|
static void add_zone_ind(uint16_t *znr, smallint *corrected)
|
|
{
|
|
int i;
|
|
int block;
|
|
smallint chg_blk = 0;
|
|
|
|
block = add_zone(znr, corrected);
|
|
if (!block)
|
|
return;
|
|
read_block(block, add_zone_ind_blk);
|
|
for (i = 0; i < (BLOCK_SIZE >> 1); i++)
|
|
add_zone(i + (uint16_t *) add_zone_ind_blk, &chg_blk);
|
|
if (chg_blk)
|
|
write_block(block, add_zone_ind_blk);
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void add_zone_ind2(uint32_t *znr, smallint *corrected)
|
|
{
|
|
int i;
|
|
int block;
|
|
smallint chg_blk = 0;
|
|
|
|
block = add_zone2(znr, corrected);
|
|
if (!block)
|
|
return;
|
|
read_block(block, add_zone_ind_blk);
|
|
for (i = 0; i < BLOCK_SIZE >> 2; i++)
|
|
add_zone2(i + (uint32_t *) add_zone_ind_blk, &chg_blk);
|
|
if (chg_blk)
|
|
write_block(block, add_zone_ind_blk);
|
|
}
|
|
#endif
|
|
|
|
static void add_zone_dind(uint16_t *znr, smallint *corrected)
|
|
{
|
|
int i;
|
|
int block;
|
|
smallint chg_blk = 0;
|
|
|
|
block = add_zone(znr, corrected);
|
|
if (!block)
|
|
return;
|
|
read_block(block, add_zone_dind_blk);
|
|
for (i = 0; i < (BLOCK_SIZE >> 1); i++)
|
|
add_zone_ind(i + (uint16_t *) add_zone_dind_blk, &chg_blk);
|
|
if (chg_blk)
|
|
write_block(block, add_zone_dind_blk);
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void add_zone_dind2(uint32_t *znr, smallint *corrected)
|
|
{
|
|
int i;
|
|
int block;
|
|
smallint chg_blk = 0;
|
|
|
|
block = add_zone2(znr, corrected);
|
|
if (!block)
|
|
return;
|
|
read_block(block, add_zone_dind_blk);
|
|
for (i = 0; i < BLOCK_SIZE >> 2; i++)
|
|
add_zone_ind2(i + (uint32_t *) add_zone_dind_blk, &chg_blk);
|
|
if (chg_blk)
|
|
write_block(block, add_zone_dind_blk);
|
|
}
|
|
|
|
static void add_zone_tind2(uint32_t *znr, smallint *corrected)
|
|
{
|
|
int i;
|
|
int block;
|
|
smallint chg_blk = 0;
|
|
|
|
block = add_zone2(znr, corrected);
|
|
if (!block)
|
|
return;
|
|
read_block(block, add_zone_tind_blk);
|
|
for (i = 0; i < BLOCK_SIZE >> 2; i++)
|
|
add_zone_dind2(i + (uint32_t *) add_zone_tind_blk, &chg_blk);
|
|
if (chg_blk)
|
|
write_block(block, add_zone_tind_blk);
|
|
}
|
|
#endif
|
|
|
|
static void check_zones(unsigned i)
|
|
{
|
|
struct minix1_inode *inode;
|
|
|
|
if (!i || i > INODES)
|
|
return;
|
|
if (inode_count[i] > 1) /* have we counted this file already? */
|
|
return;
|
|
inode = Inode1 + i;
|
|
if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
|
|
!S_ISLNK(inode->i_mode)) return;
|
|
for (i = 0; i < 7; i++)
|
|
add_zone(i + inode->i_zone, &changed);
|
|
add_zone_ind(7 + inode->i_zone, &changed);
|
|
add_zone_dind(8 + inode->i_zone, &changed);
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void check_zones2(unsigned i)
|
|
{
|
|
struct minix2_inode *inode;
|
|
|
|
if (!i || i > INODES)
|
|
return;
|
|
if (inode_count[i] > 1) /* have we counted this file already? */
|
|
return;
|
|
inode = Inode2 + i;
|
|
if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode)
|
|
&& !S_ISLNK(inode->i_mode))
|
|
return;
|
|
for (i = 0; i < 7; i++)
|
|
add_zone2(i + inode->i_zone, &changed);
|
|
add_zone_ind2(7 + inode->i_zone, &changed);
|
|
add_zone_dind2(8 + inode->i_zone, &changed);
|
|
add_zone_tind2(9 + inode->i_zone, &changed);
|
|
}
|
|
#endif
|
|
|
|
static void check_file(struct minix1_inode *dir, unsigned offset)
|
|
{
|
|
struct minix1_inode *inode;
|
|
int ino;
|
|
char *name;
|
|
int block;
|
|
|
|
block = map_block(dir, offset / BLOCK_SIZE);
|
|
read_block(block, check_file_blk);
|
|
name = check_file_blk + (offset % BLOCK_SIZE) + 2;
|
|
ino = *(uint16_t *) (name - 2);
|
|
if (ino > INODES) {
|
|
printf("%s contains a bad inode number for file '%.*s'. ",
|
|
current_name, namelen, name);
|
|
if (ask("Remove", 1)) {
|
|
*(uint16_t *) (name - 2) = 0;
|
|
write_block(block, check_file_blk);
|
|
}
|
|
ino = 0;
|
|
}
|
|
push_filename(name);
|
|
inode = get_inode(ino);
|
|
pop_filename();
|
|
if (!offset) {
|
|
if (inode && LONE_CHAR(name, '.'))
|
|
return;
|
|
printf("%s: bad directory: '.' isn't first\n", current_name);
|
|
errors_uncorrected = 1;
|
|
}
|
|
if (offset == dirsize) {
|
|
if (inode && strcmp("..", name) == 0)
|
|
return;
|
|
printf("%s: bad directory: '..' isn't second\n", current_name);
|
|
errors_uncorrected = 1;
|
|
}
|
|
if (!inode)
|
|
return;
|
|
push_filename(name);
|
|
if (OPT_list) {
|
|
if (OPT_verbose)
|
|
printf("%6d %07o %3d ", ino, inode->i_mode, inode->i_nlinks);
|
|
printf("%s%s\n", current_name, S_ISDIR(inode->i_mode) ? ":" : "");
|
|
}
|
|
check_zones(ino);
|
|
if (inode && S_ISDIR(inode->i_mode))
|
|
recursive_check(ino);
|
|
pop_filename();
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void check_file2(struct minix2_inode *dir, unsigned offset)
|
|
{
|
|
struct minix2_inode *inode;
|
|
int ino;
|
|
char *name;
|
|
int block;
|
|
|
|
block = map_block2(dir, offset / BLOCK_SIZE);
|
|
read_block(block, check_file_blk);
|
|
name = check_file_blk + (offset % BLOCK_SIZE) + 2;
|
|
ino = *(uint16_t *) (name - 2);
|
|
if (ino > INODES) {
|
|
printf("%s contains a bad inode number for file '%.*s'. ",
|
|
current_name, namelen, name);
|
|
if (ask("Remove", 1)) {
|
|
*(uint16_t *) (name - 2) = 0;
|
|
write_block(block, check_file_blk);
|
|
}
|
|
ino = 0;
|
|
}
|
|
push_filename(name);
|
|
inode = get_inode2(ino);
|
|
pop_filename();
|
|
if (!offset) {
|
|
if (inode && LONE_CHAR(name, '.'))
|
|
return;
|
|
printf("%s: bad directory: '.' isn't first\n", current_name);
|
|
errors_uncorrected = 1;
|
|
}
|
|
if (offset == dirsize) {
|
|
if (inode && strcmp("..", name) == 0)
|
|
return;
|
|
printf("%s: bad directory: '..' isn't second\n", current_name);
|
|
errors_uncorrected = 1;
|
|
}
|
|
if (!inode)
|
|
return;
|
|
push_filename(name);
|
|
if (OPT_list) {
|
|
if (OPT_verbose)
|
|
printf("%6d %07o %3d ", ino, inode->i_mode, inode->i_nlinks);
|
|
printf("%s%s\n", current_name, S_ISDIR(inode->i_mode) ? ":" : "");
|
|
}
|
|
check_zones2(ino);
|
|
if (inode && S_ISDIR(inode->i_mode))
|
|
recursive_check2(ino);
|
|
pop_filename();
|
|
}
|
|
#endif
|
|
|
|
static void recursive_check(unsigned ino)
|
|
{
|
|
struct minix1_inode *dir;
|
|
unsigned offset;
|
|
|
|
dir = Inode1 + ino;
|
|
if (!S_ISDIR(dir->i_mode))
|
|
die("internal error");
|
|
if (dir->i_size < 2 * dirsize) {
|
|
printf("%s: bad directory: size<32", current_name);
|
|
errors_uncorrected = 1;
|
|
}
|
|
for (offset = 0; offset < dir->i_size; offset += dirsize)
|
|
check_file(dir, offset);
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void recursive_check2(unsigned ino)
|
|
{
|
|
struct minix2_inode *dir;
|
|
unsigned offset;
|
|
|
|
dir = Inode2 + ino;
|
|
if (!S_ISDIR(dir->i_mode))
|
|
die("internal error");
|
|
if (dir->i_size < 2 * dirsize) {
|
|
printf("%s: bad directory: size<32", current_name);
|
|
errors_uncorrected = 1;
|
|
}
|
|
for (offset = 0; offset < dir->i_size; offset += dirsize)
|
|
check_file2(dir, offset);
|
|
}
|
|
#endif
|
|
|
|
static int bad_zone(int i)
|
|
{
|
|
char buffer[BLOCK_SIZE];
|
|
|
|
xlseek(dev_fd, BLOCK_SIZE * i, SEEK_SET);
|
|
return (BLOCK_SIZE != full_read(dev_fd, buffer, BLOCK_SIZE));
|
|
}
|
|
|
|
static void check_counts(void)
|
|
{
|
|
int i;
|
|
|
|
for (i = 1; i <= INODES; i++) {
|
|
if (OPT_warn_mode && Inode1[i].i_mode && !inode_in_use(i)) {
|
|
printf("Inode %d has non-zero mode. ", i);
|
|
if (ask("Clear", 1)) {
|
|
Inode1[i].i_mode = 0;
|
|
changed = 1;
|
|
}
|
|
}
|
|
if (!inode_count[i]) {
|
|
if (!inode_in_use(i))
|
|
continue;
|
|
printf("Unused inode %d is marked as 'used' in the bitmap. ", i);
|
|
if (ask("Clear", 1))
|
|
unmark_inode(i);
|
|
continue;
|
|
}
|
|
if (!inode_in_use(i)) {
|
|
printf("Inode %d is used, but marked as 'unused' in the bitmap. ", i);
|
|
if (ask("Set", 1))
|
|
mark_inode(i);
|
|
}
|
|
if (Inode1[i].i_nlinks != inode_count[i]) {
|
|
printf("Inode %d (mode=%07o), i_nlinks=%d, counted=%d. ",
|
|
i, Inode1[i].i_mode, Inode1[i].i_nlinks,
|
|
inode_count[i]);
|
|
if (ask("Set i_nlinks to count", 1)) {
|
|
Inode1[i].i_nlinks = inode_count[i];
|
|
changed = 1;
|
|
}
|
|
}
|
|
}
|
|
for (i = FIRSTZONE; i < ZONES; i++) {
|
|
if ((zone_in_use(i) != 0) == zone_count[i])
|
|
continue;
|
|
if (!zone_count[i]) {
|
|
if (bad_zone(i))
|
|
continue;
|
|
printf("Zone %d is marked 'in use', but no file uses it. ", i);
|
|
if (ask("Unmark", 1))
|
|
unmark_zone(i);
|
|
continue;
|
|
}
|
|
printf("Zone %d: %sin use, counted=%d\n",
|
|
i, zone_in_use(i) ? "" : "not ", zone_count[i]);
|
|
}
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void check_counts2(void)
|
|
{
|
|
int i;
|
|
|
|
for (i = 1; i <= INODES; i++) {
|
|
if (OPT_warn_mode && Inode2[i].i_mode && !inode_in_use(i)) {
|
|
printf("Inode %d has non-zero mode. ", i);
|
|
if (ask("Clear", 1)) {
|
|
Inode2[i].i_mode = 0;
|
|
changed = 1;
|
|
}
|
|
}
|
|
if (!inode_count[i]) {
|
|
if (!inode_in_use(i))
|
|
continue;
|
|
printf("Unused inode %d is marked as 'used' in the bitmap. ", i);
|
|
if (ask("Clear", 1))
|
|
unmark_inode(i);
|
|
continue;
|
|
}
|
|
if (!inode_in_use(i)) {
|
|
printf("Inode %d is used, but marked as 'unused' in the bitmap. ", i);
|
|
if (ask("Set", 1))
|
|
mark_inode(i);
|
|
}
|
|
if (Inode2[i].i_nlinks != inode_count[i]) {
|
|
printf("Inode %d (mode=%07o), i_nlinks=%d, counted=%d. ",
|
|
i, Inode2[i].i_mode, Inode2[i].i_nlinks,
|
|
inode_count[i]);
|
|
if (ask("Set i_nlinks to count", 1)) {
|
|
Inode2[i].i_nlinks = inode_count[i];
|
|
changed = 1;
|
|
}
|
|
}
|
|
}
|
|
for (i = FIRSTZONE; i < ZONES; i++) {
|
|
if ((zone_in_use(i) != 0) == zone_count[i])
|
|
continue;
|
|
if (!zone_count[i]) {
|
|
if (bad_zone(i))
|
|
continue;
|
|
printf("Zone %d is marked 'in use', but no file uses it. ", i);
|
|
if (ask("Unmark", 1))
|
|
unmark_zone(i);
|
|
continue;
|
|
}
|
|
printf("Zone %d: %sin use, counted=%d\n",
|
|
i, zone_in_use(i) ? "" : "not ", zone_count[i]);
|
|
}
|
|
}
|
|
#endif
|
|
|
|
static void check(void)
|
|
{
|
|
memset(inode_count, 0, (INODES + 1) * sizeof(*inode_count));
|
|
memset(zone_count, 0, ZONES * sizeof(*zone_count));
|
|
check_zones(MINIX_ROOT_INO);
|
|
recursive_check(MINIX_ROOT_INO);
|
|
check_counts();
|
|
}
|
|
|
|
#if ENABLE_FEATURE_MINIX2
|
|
static void check2(void)
|
|
{
|
|
memset(inode_count, 0, (INODES + 1) * sizeof(*inode_count));
|
|
memset(zone_count, 0, ZONES * sizeof(*zone_count));
|
|
check_zones2(MINIX_ROOT_INO);
|
|
recursive_check2(MINIX_ROOT_INO);
|
|
check_counts2();
|
|
}
|
|
#else
|
|
void check2(void);
|
|
#endif
|
|
|
|
int fsck_minix_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
|
|
int fsck_minix_main(int argc UNUSED_PARAM, char **argv)
|
|
{
|
|
struct termios tmp;
|
|
int retcode = 0;
|
|
|
|
xfunc_error_retval = 8;
|
|
|
|
INIT_G();
|
|
|
|
opt_complementary = "=1:ar"; /* one argument; -a assumes -r */
|
|
getopt32(argv, OPTION_STR);
|
|
argv += optind;
|
|
device_name = argv[0];
|
|
|
|
check_mount(); /* trying to check a mounted filesystem? */
|
|
if (OPT_manual) {
|
|
if (!isatty(0) || !isatty(1))
|
|
die("need terminal for interactive repairs");
|
|
}
|
|
xmove_fd(xopen(device_name, OPT_repair ? O_RDWR : O_RDONLY), dev_fd);
|
|
|
|
/*sync(); paranoia? */
|
|
read_superblock();
|
|
|
|
/*
|
|
* Determine whether or not we should continue with the checking.
|
|
* This is based on the status of the filesystem valid and error
|
|
* flags and whether or not the -f switch was specified on the
|
|
* command line.
|
|
*/
|
|
printf("%s: %s\n", applet_name, bb_banner);
|
|
|
|
if (!(Super.s_state & MINIX_ERROR_FS)
|
|
&& (Super.s_state & MINIX_VALID_FS) && !OPT_force
|
|
) {
|
|
if (OPT_repair)
|
|
printf("%s is clean, check is skipped\n", device_name);
|
|
return 0;
|
|
} else if (OPT_force)
|
|
printf("Forcing filesystem check on %s\n", device_name);
|
|
else if (OPT_repair)
|
|
printf("Filesystem on %s is dirty, needs checking\n",
|
|
device_name);
|
|
|
|
read_tables();
|
|
|
|
if (OPT_manual) {
|
|
tcgetattr(0, &sv_termios);
|
|
tmp = sv_termios;
|
|
tmp.c_lflag &= ~(ICANON | ECHO);
|
|
tcsetattr_stdin_TCSANOW(&tmp);
|
|
termios_set = 1;
|
|
}
|
|
|
|
if (version2) {
|
|
check_root2();
|
|
check2();
|
|
} else {
|
|
check_root();
|
|
check();
|
|
}
|
|
|
|
if (OPT_verbose) {
|
|
int i, free_cnt;
|
|
|
|
for (i = 1, free_cnt = 0; i <= INODES; i++)
|
|
if (!inode_in_use(i))
|
|
free_cnt++;
|
|
printf("\n%6u inodes used (%u%%)\n", (INODES - free_cnt),
|
|
100 * (INODES - free_cnt) / INODES);
|
|
for (i = FIRSTZONE, free_cnt = 0; i < ZONES; i++)
|
|
if (!zone_in_use(i))
|
|
free_cnt++;
|
|
printf("%6u zones used (%u%%)\n\n"
|
|
"%6u regular files\n"
|
|
"%6u directories\n"
|
|
"%6u character device files\n"
|
|
"%6u block device files\n"
|
|
"%6u links\n"
|
|
"%6u symbolic links\n"
|
|
"------\n"
|
|
"%6u files\n",
|
|
(ZONES - free_cnt), 100 * (ZONES - free_cnt) / ZONES,
|
|
regular, directory, chardev, blockdev,
|
|
links - 2 * directory + 1, symlinks,
|
|
total - 2 * directory + 1);
|
|
}
|
|
if (changed) {
|
|
write_tables();
|
|
printf("FILE SYSTEM HAS BEEN CHANGED\n");
|
|
sync();
|
|
} else if (OPT_repair)
|
|
write_superblock();
|
|
|
|
if (OPT_manual)
|
|
tcsetattr_stdin_TCSANOW(&sv_termios);
|
|
|
|
if (changed)
|
|
retcode += 3;
|
|
if (errors_uncorrected)
|
|
retcode += 4;
|
|
return retcode;
|
|
}
|