Files
lbmk/util/nvmutil/nvmutil.c
Leah Rowe 5d95d4dfe9 util/nvmutil: stricter return in pread
we were returning if verified is not off, but we
were not doing the check soon enough.

now it's clearer: just after either a reset,
or we found out offset doesn't match, we
return sooner.

otherwise, we read, and we verify again right
after. in the old code, we verified twice in
a row.

this is just more optimal, for error handling.

Signed-off-by: Leah Rowe <leah@libreboot.org>
2026-03-15 02:01:20 +00:00

2378 lines
52 KiB
C

/* SPDX-License-Identifier: MIT
*
* Copyright (c) 2022-2026 Leah Rowe <leah@libreboot.org>
* Copyright (c) 2023 Riku Viitanen <riku.viitanen@protonmail.com>
*
* This tool lets you modify Intel GbE NVM (Gigabit Ethernet
* Non-Volatile Memory) images, e.g. change the MAC address.
* These images configure your Intel Gigabit Ethernet adapter.
*
* This code is designed to be portable, running on as many
* Unix and Unix-like systems as possible (mainly BSD/Linux).
*
* Recommended CFLAGS for Clang/GCC:
*
* -Os -Wall -Wextra -Werror -pedantic -std=c90
*/
#define OFF_ERR 0
#ifndef OFF_RESET
#define OFF_RESET 1
#endif
/*
* NOTE: older Linux lacked arc4random.
* added in glibc 2.36. Just pass HAVE_ARC4RANDOM_BUF=0
* at build time if you need old Linux / other libc.
*/
#if defined(__OpenBSD__) || defined(__FreeBSD__) || \
defined(__NetBSD__) || defined(__APPLE__) || \
defined(__linux__)
#ifndef HAVE_ARC4RANDOM_BUF
#define HAVE_ARC4RANDOM_BUF 1
#endif
#endif
/*
* I/O config (build-time)
*
* Regarding:
* Retries on zero-return.
*
* 5 retries is generous,
* but also conservative.
* This is enough for e.g.
* slow USB flash drives,
* busy NFS servers, etc.
* Any more is too much
* and not of much benefit.
*
* 3-5 will tolerate buggy
* USB drives for example,
* but won't spin as long
* on really buggy and slow
* networks e.g. slow NFS.
*
* At least 3-5 recommended.
* Pass this at build time.
*/
#ifndef MAX_ZERO_RW_RETRY
#define MAX_ZERO_RW_RETRY 5
#endif
/*
* 0: portable pread/pwrite
* 1: real pread/pwrite (thread-safe)
* Pass this at build-time
*/
#ifndef HAVE_REAL_PREAD_PWRITE
#define HAVE_REAL_PREAD_PWRITE 0
#endif
/*
* Configure whether to wait on
* EINTR on files, or EAGAIN on
* cmd cat (stdout).
*
* Pass these at build time.
*/
#ifndef LOOP_EAGAIN
#define LOOP_EAGAIN 1
#endif
#ifndef LOOP_EINTR
#define LOOP_EINTR 1
#endif
/*
* Major TODO: split this into multiple files.
* This program has become quite large now, mostly
* due to all the extra sanity checks / portability.
* Make most of nvmutil a *library* for re-use
*
* TODO: gettimeofday not posible - use portable functions.
* TODO: ux fallback: modify the program instead
* to run on 16-bit systems: smaller buffers, and do
* operations byte-based instead of word-based.
*
* TODO: _XOPEN_SOURCE 500 probably not needed anymore.
* the portable fallbacks alone are likely enough.
* e.g. i don't need stdint, and i don't use pwrite/pread
* anymore.
*
* TODO: version detection of various BSDs to detect
* arc4random, use that if available. but also work on
* older versions of those BSDs (also MacOS) that lack it.
*
* TODO: portability/testing on non-Unix systems:
* old DOS. all windows versions (probably irrelevant
* because you can use cygwin/wsl, whatever), classic MacOS,
* also test really old unix e.g. sunos and irix. Be/Haiku too!
*
* TODO: reliance on global variables for status. make
* functions use structs passed as args instead, make
* functions re-useable (including libraries), etc.
*
* TODO: bound checks for files per-command, e.g. only
* first 6 bytes for CMD_SETMAC
*
* TODO: in command sanitizer: verify that each given
* entry corresponds to the correct function, in the
* pointer (this check is currently missing)
*
* TODO: general modularisierung of the entire codebase.
* TODO: better explain copy/swap read inversion trick
* by improving existing comments
* TODO: lots of overwritten comments in code. tidy it up.
*
* TODO: use getopt for nvmutil args, so that multiple
* operations can be performed, and also on many
* files at once (noting limitations with cat)
* BONUS: implement own getopt(), for portability
*
* TODO: document fuzzing / static analysis methods
* for the code, and:
* TODO: implement rigorous unit tests (separate util)
* NOTE: this would *include* known good test files
* in various configurations, also invalid files.
* the tests would likely be portable posix shell
* scripts rather than a new C program, but a modularisiert
* codebase would allow me to write a separate C
* program to test some finer intricacies
* TODO: the unit tests would basically test regressions
* TODO: after writing back a gbe to file, close() and
* open() it again, read it again, and check that
* the contents were written correctly, providing
* a warning if they were. do this in the main
* program.
* TODO: the unit tests would include an aggressive set
* of fuzz tests, under controlled conditions
*
* TODO: also document the layout of Intel GbE files, so
* that wily individuals can easily expand the
* featureset of nvmutil.
* TODO: write a manpage
* TODO: simplify the command sanitization, implement more
* of it as build time checks, e.g. static asserts.
* generally remove cleverness from the code, instead
* prefyerring readibility
* TODO: also document nvmutil's coding style, which is
* its own style at this point!
* TODO: when all the above (and possibly more) is done,
* submit this tool to coreboot with a further change
* to their build system that lets users modify
* GbE images, especially set MAC addresses, when
* including GbE files in coreboot configs.
*/
/*
BONUS TODO:
CI/CD. woodpecker is good enough, sourcehut also has one.
tie this in with other things mentioned here,
e.g. fuzzer / unit tests
*/
/* Major TODO: reproducible builds
Test with and without these:
CFLAGS += -fno-record-gcc-switches
CFLAGS += -ffile-prefix-map=$(PWD)=.
CFLAGS += -fdebug-prefix-map=$(PWD)=.
I already avoid unique timestamps per-build,
by not using them, e.g. not reporting build
time in the program.
When splitting the nvmutil.c file later, do e.g.:
SRC = main.c io.c nvm.c cmd.c
OBJ = $(SRC:.c=.o)
^ explicitly declare the order in which to build
*/
/*
TODO:
further note when fuzzing is implemented:
use deterministic randomisation, with a
guaranteed seed - so e.g. don't use /dev/urandom
in test builds. e.g. just use normal rand()
but with a static seed e.g. 1234
*/
/*
TODO: stricter build flags, e.g.
CFLAGS += -fstack-protector-strong
CFLAGS += -fno-common
CFLAGS += -D_FORTIFY_SOURCE=2
CFLAGS += -fPIE
also consider:
-fstack-clash-protection
-Wl,-z,relro
-Wl,-z,now
*/
#ifndef _FILE_OFFSET_BITS
#define _FILE_OFFSET_BITS 64
#endif
#ifdef __OpenBSD__
#include <sys/param.h>
#endif
#include <sys/types.h>
#include <sys/time.h>
#include <sys/stat.h>
#include <errno.h>
#include <fcntl.h>
#include <limits.h>
#include <stdarg.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <unistd.h>
typedef unsigned char u8;
typedef unsigned short ushort;
typedef unsigned int uint;
typedef unsigned long ulong;
/* type asserts */
typedef char static_assert_char_is_8_bits[(CHAR_BIT == 8) ? 1 : -1];
typedef char static_assert_char_is_1[(sizeof(char) == 1) ? 1 : -1];
typedef char static_assert_u8_is_1[
(sizeof(u8) == 1) ? 1 : -1];
typedef char static_assert_ushort_is_2[
(sizeof(ushort) >= 2) ? 1 : -1];
typedef char static_assert_short_is_2[(sizeof(short) >= 2) ? 1 : -1];
typedef char static_assert_uint_is_4[
(sizeof(uint) >= 4) ? 1 : -1];
typedef char static_assert_ulong_is_4[
(sizeof(ulong) >= 4) ? 1 : -1];
typedef char static_assert_int_ge_32[(sizeof(int) >= 4) ? 1 : -1];
typedef char static_assert_twos_complement[
((-1 & 3) == 3) ? 1 : -1
];
typedef char assert_ulong_ptr[
(sizeof(ulong) >= sizeof(void *)) ? 1 : -1
];
typedef char assert_size_t_ptr[
(sizeof(size_t) >= sizeof(void *)) ? 1 : -1
];
/*
* We set _FILE_OFFSET_BITS 64, but we only handle
* files that are 128KB in size at a maximum, so we
* realistically only need 32-bit at a minimum.
*
* We set 64 anyway, because there's no reason not
* to, but some systems may ignore _FILE_OFFSET_BITS
*/
typedef char static_assert_off_t_is_32[(sizeof(off_t) >= 4) ? 1 : -1];
/*
* Older versions of BSD to the early 2000s
* could compile nvmutil, but pledge was
* added in the 2010s. Therefore, for extra
* portability, we will only pledge/unveil
* on OpenBSD versions that have it.
*/
#if defined(__OpenBSD__) && defined(OpenBSD)
#if OpenBSD >= 604
#ifndef NVMUTIL_UNVEIL
#define NVMUTIL_UNVEIL 1
#endif
#endif
#if OpenBSD >= 509
#ifndef NVMUTIL_PLEDGE
#define NVMUTIL_PLEDGE 1
#endif
#endif
#endif
#ifndef EXIT_FAILURE
#define EXIT_FAILURE 1
#endif
#ifndef EXIT_SUCCESS
#define EXIT_SUCCESS 0
#endif
#ifndef O_BINARY
#define O_BINARY 0
#endif
#ifndef O_NOFOLLOW
#define O_NOFOLLOW 0
#endif
/*
* Sanitize command tables.
*/
static void sanitize_command_list(void);
static void sanitize_command_index(size_t c);
/*
* Argument handling (user input)
*/
static void set_cmd(int argc, char *argv[]);
static void set_cmd_args(int argc, char *argv[]);
static size_t conv_argv_part_num(const char *part_str);
static int xstrxcmp(const char *a, const char *b, size_t maxlen);
/*
* Prep files for reading
*/
static void open_gbe_file(void);
static void lock_gbe_file(void);
static void xopen(int *fd, const char *path, int flags, struct stat *st);
/*
* Read GbE file and verify
* checksums.
*
* After this, we can run commands.
*/
static void read_gbe_file(void);
static void read_checksums(void);
static int good_checksum(size_t partnum);
/*
* Execute user command on GbE data.
* These are stubs that call helpers.
*/
static void run_cmd(size_t c);
static void check_command_num(size_t c);
static u8 valid_command(size_t c);
/*
* Helper functions for command: setmac
*/
static void cmd_helper_setmac(void);
static void parse_mac_string(void);
static size_t xstrxlen(const char *scmp, size_t maxlen);
static void set_mac_byte(size_t mac_byte_pos);
static void set_mac_nib(size_t mac_str_pos,
size_t mac_byte_pos, size_t mac_nib_pos);
static ushort hextonum(char ch_s);
static ushort rhex(void);
#if !defined(HAVE_ARC4RANDOM_BUF) || \
(HAVE_ARC4RANDOM_BUF) < 1
static ulong entropy_jitter(void);
#endif
static void write_mac_part(size_t partnum);
/*
* Helper functions for command: dump
*/
static void cmd_helper_dump(void);
static void print_mac_from_nvm(size_t partnum);
static void hexdump(size_t partnum);
/*
* Helper functions for commands:
* cat, cat16 and cat128
*/
static void cmd_helper_cat(void);
static void gbe_cat_buf(u8 *b);
/*
* After command processing, write
* the modified GbE file back.
*
* These are stub functions: check
* below for the actual functions.
*/
static void write_gbe_file(void);
static void override_part_modified(void);
static void set_checksum(size_t part);
static ushort calculated_checksum(size_t p);
/*
* Helper functions for accessing
* the NVM area during operation.
*/
static ushort nvm_word(size_t pos16, size_t part);
static void set_nvm_word(size_t pos16, size_t part, ushort val16);
static void set_part_modified(size_t p);
static void check_nvm_bound(size_t pos16, size_t part);
static void check_bin(size_t a, const char *a_name);
/*
* Helper functions for stub functions
* that handle GbE file reads/writes.
*/
static void rw_gbe_file_part(size_t p, int rw_type,
const char *rw_type_str);
static void check_written_part(size_t p);
static void report_io_err_rw(void);
static u8 *gbe_mem_offset(size_t part, const char *f_op);
static off_t gbe_file_offset(size_t part, const char *f_op);
static off_t gbe_x_offset(size_t part, const char *f_op,
const char *d_type, off_t nsize, off_t ncmp);
static ssize_t rw_gbe_file_exact(int fd, u8 *mem, size_t nrw,
off_t off, int rw_type);
static ssize_t rw_file_exact(int fd, u8 *mem, size_t len,
off_t off, int rw_type, int loop_eagain, int loop_eintr,
size_t max_retries, int off_reset);
static ssize_t prw(int fd, void *mem, size_t nrw,
off_t off, int rw_type, int loop_eagain, int loop_eintr,
int off_reset);
static int io_args(int fd, void *mem, size_t nrw,
off_t off, int rw_type);
static int check_file(int fd, struct stat *st);
static ssize_t rw_over_nrw(ssize_t r, size_t nrw);
#if !defined(HAVE_REAL_PREAD_PWRITE) || \
HAVE_REAL_PREAD_PWRITE < 1
static off_t lseek_loop(int fd, off_t off,
int whence, int loop_eagain, int loop_eintr);
#endif
static int try_err(int loop_err, int errval);
/*
* Error handling and cleanup
*/
static void err(int nvm_errval, const char *msg, ...);
static int close_files(void);
static const char *getnvmprogname(void);
static void usage(int usage_exit);
/*
* Sizes in bytes:
*/
#define SIZE_1KB 1024
#define SIZE_4KB (4 * SIZE_1KB)
#define SIZE_8KB (8 * SIZE_1KB)
#define SIZE_16KB (16 * SIZE_1KB)
#define SIZE_128KB (128 * SIZE_1KB)
/*
* First 128 bytes of a GbE part contains
* the regular NVM (Non-Volatile-Memory)
* area. All of these bytes must add up,
* truncated to 0xBABA.
*
* The full GbE region is 4KB, but only
* the first 128 bytes are used here.
*
* There is a second 4KB part with the same
* rules, and it *should* be identical.
*/
#define GBE_FILE_SIZE SIZE_8KB /* for buf */
#define GBE_PART_SIZE (GBE_FILE_SIZE >> 1)
#define NVM_CHECKSUM 0xBABA
#define NVM_SIZE 128
#define NVM_WORDS (NVM_SIZE >> 1)
#define NVM_CHECKSUM_WORD (NVM_WORDS - 1)
/*
* Portable macro based on BSD nitems.
* Used to count the number of commands (see below).
*/
#define items(x) (sizeof((x)) / sizeof((x)[0]))
/*
* GbE files can be 8KB, 16KB or 128KB,
* but we only need the two 4KB parts
* from offset zero and offset 64KB in
* a 128KB file, or zero and 8KB in a 16KB
* file, or zero and 4KB in an 8KB file.
*
* The code will handle this properly.
*/
static u8 real_buf[GBE_FILE_SIZE];
static u8 pad[GBE_FILE_SIZE]; /* the file that wouldn't die */
static u8 *buf = real_buf;
static ushort mac_buf[3];
static off_t gbe_file_size;
static int gbe_fd = -1;
static size_t part;
static u8 part_modified[2];
static u8 part_valid[2];
static const char rmac[] = "xx:xx:xx:xx:xx:xx";
static const char *mac_str;
static const char *fname;
static const char *argv0;
#ifndef SSIZE_MAX
#define SSIZE_MAX ((ssize_t)(~((size_t)1 << (sizeof(ssize_t)*CHAR_BIT-1))))
#endif
/*
* Use these for .invert in command[]:
* If set to 1: read/write inverter (p0->p1, p1->p0)
*/
#define PART_INVERT 1
#define NO_INVERT 0
/*
* Use these for .argc in command[]:
*/
#define ARGC_3 3
#define ARGC_4 4
#define NO_LOOP_EAGAIN 0
#define NO_LOOP_EINTR 0
enum {
IO_READ,
IO_WRITE,
IO_PREAD,
IO_PWRITE
};
/*
* Used as indices for command[]
* MUST be in the same order as entries in command[]
*/
enum {
CMD_DUMP,
CMD_SETMAC,
CMD_SWAP,
CMD_COPY,
CMD_CAT,
CMD_CAT16,
CMD_CAT128
};
/*
* If set, a given part will always be written.
*/
enum {
SET_MOD_OFF, /* don't manually set part modified */
SET_MOD_0, /* set part 0 modified */
SET_MOD_1, /* set part 1 modified */
SET_MOD_N, /* set user-specified part modified */
/* affected by command[].invert */
SET_MOD_BOTH /* set both parts modified */
};
enum {
ARG_NOPART,
ARG_PART
};
enum {
SKIP_CHECKSUM_READ,
CHECKSUM_READ
};
enum {
SKIP_CHECKSUM_WRITE,
CHECKSUM_WRITE
};
struct commands {
size_t chk;
const char *str;
void (*run)(void);
int argc;
u8 invert;
u8 set_modified;
u8 arg_part;
u8 chksum_read;
u8 chksum_write;
size_t rw_size; /* within the 4KB GbE part */
int flags; /* e.g. O_RDWR or O_RDONLY */
};
/*
* Command table, for nvmutil commands
*/
static const struct commands command[] = {
{ CMD_DUMP, "dump", cmd_helper_dump, ARGC_3,
NO_INVERT, SET_MOD_OFF,
ARG_NOPART,
SKIP_CHECKSUM_READ, SKIP_CHECKSUM_WRITE,
NVM_SIZE, O_RDONLY },
{ CMD_SETMAC, "setmac", cmd_helper_setmac, ARGC_3,
NO_INVERT, SET_MOD_OFF,
ARG_NOPART,
CHECKSUM_READ, CHECKSUM_WRITE,
NVM_SIZE, O_RDWR },
/*
* OPTIMISATION: Read inverted, so no copying is needed.
*/
{ CMD_SWAP, "swap", NULL, ARGC_3,
PART_INVERT, SET_MOD_BOTH,
ARG_NOPART,
CHECKSUM_READ, SKIP_CHECKSUM_WRITE,
GBE_PART_SIZE, O_RDWR },
/*
* OPTIMISATION: Read inverted, so no copying is needed.
* The non-target part will not be read.
*/
{ CMD_COPY, "copy", NULL, ARGC_4,
PART_INVERT, SET_MOD_N,
ARG_PART,
CHECKSUM_READ, SKIP_CHECKSUM_WRITE,
GBE_PART_SIZE, O_RDWR },
{ CMD_CAT, "cat", cmd_helper_cat, ARGC_3,
NO_INVERT, SET_MOD_OFF,
ARG_NOPART,
CHECKSUM_READ, SKIP_CHECKSUM_WRITE,
GBE_PART_SIZE, O_RDONLY },
{ CMD_CAT16, "cat16", cmd_helper_cat, ARGC_3,
NO_INVERT, SET_MOD_OFF,
ARG_NOPART,
CHECKSUM_READ, SKIP_CHECKSUM_WRITE,
GBE_PART_SIZE, O_RDONLY },
{ CMD_CAT128, "cat128", cmd_helper_cat, ARGC_3,
NO_INVERT, SET_MOD_OFF,
ARG_NOPART,
CHECKSUM_READ, SKIP_CHECKSUM_WRITE,
GBE_PART_SIZE, O_RDONLY },
};
#define MAX_CMD_LEN 50
#define N_COMMANDS items(command)
#define CMD_NULL N_COMMANDS
/*
* Index in command[], will be set later
*/
static size_t cmd_index = CMD_NULL;
/*
* asserts (variables/defines sanity check)
*/
typedef char assert_argc3[(ARGC_3==3)?1:-1];
typedef char assert_argc4[(ARGC_4==4)?1:-1];
typedef char assert_read[(IO_READ==0)?1:-1];
typedef char assert_write[(IO_WRITE==1)?1:-1];
typedef char assert_pread[(IO_PREAD==2)?1:-1];
typedef char assert_pwrite[(IO_PWRITE==3)?1:-1];
/* commands */
typedef char assert_cmd_dump[(CMD_DUMP==0)?1:-1];
typedef char assert_cmd_setmac[(CMD_SETMAC==1)?1:-1];
typedef char assert_cmd_swap[(CMD_SWAP==2)?1:-1];
typedef char assert_cmd_copy[(CMD_COPY==3)?1:-1];
typedef char assert_cmd_cat[(CMD_CAT==4)?1:-1];
typedef char assert_cmd_cat16[(CMD_CAT16==5)?1:-1];
typedef char assert_cmd_cat128[(CMD_CAT128==6)?1:-1];
/* mod_type */
typedef char assert_mod_off[(SET_MOD_OFF==0)?1:-1];
typedef char assert_mod_0[(SET_MOD_0==1)?1:-1];
typedef char assert_mod_1[(SET_MOD_1==2)?1:-1];
typedef char assert_mod_n[(SET_MOD_N==3)?1:-1];
typedef char assert_mod_both[(SET_MOD_BOTH==4)?1:-1];
/* bool */
typedef char bool_arg_nopart[(ARG_NOPART==0)?1:-1];
typedef char bool_arg_part[(ARG_PART==1)?1:-1];
typedef char bool_skip_checksum_read[(SKIP_CHECKSUM_READ==0)?1:-1];
typedef char bool_checksum_read[(CHECKSUM_READ==1)?1:-1];
typedef char bool_skip_checksum_write[(SKIP_CHECKSUM_WRITE==0)?1:-1];
typedef char bool_checksum_write[(CHECKSUM_WRITE==1)?1:-1];
typedef char bool_no_invert[(NO_INVERT==0)?1:-1];
typedef char bool_part_invert[(PART_INVERT==1)?1:-1];
typedef char bool_loop_eintr[(LOOP_EINTR==1||LOOP_EINTR==0)?1:-1];
typedef char bool_loop_eagain[(LOOP_EAGAIN==1||LOOP_EAGAIN==0)?1:-1];
typedef char bool_no_loop_eintr[(NO_LOOP_EINTR==0)?1:-1];
typedef char bool_no_loop_eagain[(NO_LOOP_EAGAIN==0)?1:-1];
typedef char bool_off_err[(OFF_ERR==0)?1:-1];
typedef char bool_off_reset[(OFF_RESET==0||OFF_RESET==1)?1:-1];
static int io_err_gbe = 0;
static int rw_check_err_read[] = {0, 0};
static int rw_check_partial_read[] = {0, 0};
static int rw_check_bad_part[] = {0, 0};
static int post_rw_checksum[] = {0, 0};
static dev_t gbe_dev;
static ino_t gbe_ino;
#if defined(HAVE_ARC4RANDOM_BUF) && \
(HAVE_ARC4RANDOM_BUF) > 0
void arc4random_buf(void *buf, size_t n);
#endif
int
main(int argc, char *argv[])
{
argv0 = argv[0];
if (argc < 3)
usage(1);
fname = argv[1];
#ifdef NVMUTIL_PLEDGE
#ifdef NVMUTIL_UNVEIL
if (pledge("stdio flock rpath wpath unveil", NULL) == -1)
err(errno, "pledge");
if (unveil("/dev/null", "r") == -1)
err(errno, "unveil /dev/null");
#else
if (pledge("stdio flock rpath wpath", NULL) == -1)
err(errno, "pledge");
#endif
#endif
sanitize_command_list();
set_cmd(argc, argv);
set_cmd_args(argc, argv);
#ifdef NVMUTIL_PLEDGE
#ifdef NVMUTIL_UNVEIL
if (command[cmd_index].flags == O_RDONLY) {
if (unveil(fname, "r") == -1)
err(errno, "%s: unveil ro", fname);
if (unveil(NULL, NULL) == -1)
err(errno, "unveil block (ro)");
if (pledge("stdio flock rpath", NULL) == -1)
err(errno, "pledge ro (kill unveil)");
} else {
if (unveil(fname, "rw") == -1)
err(errno, "%s: unveil rw", fname);
if (unveil(NULL, NULL) == -1)
err(errno, "unveil block (rw)");
if (pledge("stdio flock rpath wpath", NULL) == -1)
err(errno, "pledge rw (kill unveil)");
}
#else
if (command[cmd_index].flags == O_RDONLY) {
if (pledge("stdio flock rpath", NULL) == -1)
err(errno, "pledge ro");
}
#endif
#endif
#if !defined(HAVE_ARC4RANDOM_BUF) || \
(HAVE_ARC4RANDOM_BUF) < 1
srand((uint)(time(NULL) ^ getpid()));
#endif
open_gbe_file();
lock_gbe_file();
#ifdef NVMUTIL_PLEDGE
if (pledge("stdio", NULL) == -1)
err(errno, "pledge stdio (main)");
#endif
/*
* Used by CMD_CAT, for padding
*/
memset(pad, 0xff, sizeof(pad));
read_gbe_file();
read_checksums();
run_cmd(cmd_index);
if (command[cmd_index].flags == O_RDWR) {
write_gbe_file();
/*
* We may otherwise read from
* cache, so we must sync.
*/
if (fsync(gbe_fd) == -1)
err(errno, "%s: fsync (pre-verification)",
fname);
check_written_part(0);
check_written_part(1);
report_io_err_rw();
if (io_err_gbe)
err(EIO, "%s: bad write", fname);
}
if (close_files() == -1)
err(EIO, "%s: close", fname);
return EXIT_SUCCESS;
}
/*
* Guard against regressions by maintainers (command table)
*/
static void
sanitize_command_list(void)
{
size_t c;
for (c = 0; c < N_COMMANDS; c++)
sanitize_command_index(c);
}
/*
* TODO: specific config checks per command
*/
static void
sanitize_command_index(size_t c)
{
u8 mod_type;
size_t gbe_rw_size;
check_command_num(c);
if (command[c].argc < 3)
err(EINVAL, "cmd index %lu: argc below 3, %d",
(ulong)c, command[c].argc);
if (command[c].str == NULL)
err(EINVAL, "cmd index %lu: NULL str",
(ulong)c);
if (*command[c].str == '\0')
err(EINVAL, "cmd index %lu: empty str",
(ulong)c);
if (xstrxlen(command[c].str, MAX_CMD_LEN + 1) >
MAX_CMD_LEN) {
err(EINVAL, "cmd index %lu: str too long: %s",
(ulong)c, command[c].str);
}
mod_type = command[c].set_modified;
switch (mod_type) {
case SET_MOD_0:
case SET_MOD_1:
case SET_MOD_N:
case SET_MOD_BOTH:
case SET_MOD_OFF:
break;
default:
err(EINVAL, "Unsupported set_mod type: %u", mod_type);
}
check_bin(command[c].invert, "cmd.invert");
check_bin(command[c].arg_part, "cmd.arg_part");
check_bin(command[c].chksum_read, "cmd.chksum_read");
check_bin(command[c].chksum_write, "cmd.chksum_write");
gbe_rw_size = command[c].rw_size;
switch (gbe_rw_size) {
case GBE_PART_SIZE:
case NVM_SIZE:
break;
default:
err(EINVAL, "Unsupported rw_size: %lu",
(ulong)gbe_rw_size);
}
if (gbe_rw_size > GBE_PART_SIZE)
err(EINVAL, "rw_size larger than GbE part: %lu",
(ulong)gbe_rw_size);
if (command[c].flags != O_RDONLY &&
command[c].flags != O_RDWR)
err(EINVAL, "invalid cmd.flags setting");
}
static void
set_cmd(int argc, char *argv[])
{
const char *cmd_str;
for (cmd_index = 0; valid_command(cmd_index); cmd_index++) {
cmd_str = command[cmd_index].str;
if (xstrxcmp(argv[2], cmd_str, MAX_CMD_LEN) != 0)
continue;
else if (argc >= command[cmd_index].argc)
return;
err(EINVAL, "Too few args on command '%s'", cmd_str);
}
cmd_index = CMD_NULL;
}
static void
set_cmd_args(int argc, char *argv[])
{
u8 arg_part;
if (!valid_command(cmd_index) || argc < 3)
usage(1);
arg_part = command[cmd_index].arg_part;
/* Maintainer bugs */
if (arg_part && argc < 4)
err(EINVAL,
"arg_part set for command that needs argc4");
if (arg_part && cmd_index == CMD_SETMAC)
err(EINVAL,
"arg_part set on CMD_SETMAC");
if (cmd_index == CMD_SETMAC)
mac_str = argc >= 4 ? argv[3] : rmac;
else if (arg_part)
part = conv_argv_part_num(argv[3]);
}
static size_t
conv_argv_part_num(const char *part_str)
{
u8 ch;
if (part_str[0] == '\0' || part_str[1] != '\0')
err(EINVAL, "Partnum string '%s' wrong length", part_str);
/* char signedness is implementation-defined */
ch = (u8)part_str[0];
if (ch < '0' || ch > '1')
err(EINVAL, "Bad part number (%c)", ch);
return (size_t)(ch - '0');
}
/*
* Portable strcmp() but blocks NULL/empty/unterminated
* strings. Even stricter than strncmp().
*/
static int
xstrxcmp(const char *a, const char *b, size_t maxlen)
{
size_t i;
if (a == NULL || b == NULL)
err(EINVAL, "NULL input to xstrxcmp");
if (*a == '\0' || *b == '\0')
err(EINVAL, "Empty string in xstrxcmp");
for (i = 0; i < maxlen; i++) {
u8 ac = (u8)a[i];
u8 bc = (u8)b[i];
if (ac == '\0' || bc == '\0') {
if (ac == bc)
return 0;
return ac - bc;
}
if (ac != bc)
return ac - bc;
}
/*
* We reached maxlen, so assume unterminated string.
*/
err(EINVAL, "Unterminated string in xstrxcmp");
/*
* Should never reach here. This keeps compilers happy.
*/
errno = EINVAL;
return -1;
}
static void
open_gbe_file(void)
{
struct stat gbe_st;
int flags;
xopen(&gbe_fd, fname,
command[cmd_index].flags | O_BINARY | O_NOFOLLOW, &gbe_st);
/* inode will be checked later on write */
gbe_dev = gbe_st.st_dev;
gbe_ino = gbe_st.st_ino;
if (gbe_st.st_nlink > 1)
fprintf(stderr,
"%s: warning: file has %lu hard links\n",
fname, (ulong)gbe_st.st_nlink);
if (gbe_st.st_nlink == 0)
err(EIO, "%s: file unlinked while open", fname);
flags = fcntl(gbe_fd, F_GETFL);
if (flags == -1)
err(errno, "%s: fcntl(F_GETFL)", fname);
/*
* O_APPEND must not be used, because this
* allows POSIX write() to ignore the
* current write offset and write at EOF,
* which would therefore break pread/pwrite
*/
if (flags & O_APPEND)
err(EIO, "%s: O_APPEND flag");
gbe_file_size = gbe_st.st_size;
switch (gbe_file_size) {
case SIZE_8KB:
case SIZE_16KB:
case SIZE_128KB:
break;
default:
err(EINVAL, "File size must be 8KB, 16KB or 128KB");
}
}
static void
lock_gbe_file(void)
{
struct flock fl;
memset(&fl, 0, sizeof(fl));
if (command[cmd_index].flags == O_RDONLY)
fl.l_type = F_RDLCK;
else
fl.l_type = F_WRLCK;
fl.l_whence = SEEK_SET;
if (fcntl(gbe_fd, F_SETLK, &fl) == -1)
err(errno, "file is locked by another process");
}
static void
xopen(int *fd_ptr, const char *path, int flags, struct stat *st)
{
if ((*fd_ptr = open(path, flags)) == -1)
err(errno, "%s", path);
if (fstat(*fd_ptr, st) == -1)
err(errno, "%s", path);
if (!S_ISREG(st->st_mode))
err(errno, "%s: not a regular file", path);
if (lseek(*fd_ptr, 0, SEEK_CUR) == (off_t)-1)
err(errno, "%s: file not seekable", path);
}
static void
read_gbe_file(void)
{
size_t p;
u8 do_read[2] = {1, 1};
/*
* Commands specifying a partnum only
* need the given GbE part to be read.
*/
if (command[cmd_index].arg_part)
do_read[part ^ 1] = 0;
for (p = 0; p < 2; p++) {
if (do_read[p])
rw_gbe_file_part(p, IO_PREAD, "pread");
}
}
static void
read_checksums(void)
{
size_t p;
size_t skip_part;
u8 invert;
u8 arg_part;
u8 num_invalid;
u8 max_invalid;
part_valid[0] = 0;
part_valid[1] = 0;
if (!command[cmd_index].chksum_read)
return;
num_invalid = 0;
max_invalid = 2;
invert = command[cmd_index].invert;
arg_part = command[cmd_index].arg_part;
if (arg_part)
max_invalid = 1;
/*
* Skip verification on this part,
* but only when arg_part is set.
*/
skip_part = part ^ 1 ^ invert;
for (p = 0; p < 2; p++) {
/*
* Only verify a part if it was *read*
*/
if (arg_part && (p == skip_part))
continue;
part_valid[p] = good_checksum(p);
if (!part_valid[p])
++num_invalid;
}
if (num_invalid >= max_invalid) {
if (max_invalid == 1)
err(ECANCELED, "%s: part %lu has a bad checksum",
fname, (ulong)part);
err(ECANCELED, "%s: No valid checksum found in file",
fname);
}
}
static int
good_checksum(size_t partnum)
{
ushort expected_checksum = calculated_checksum(partnum);
ushort current_checksum = nvm_word(NVM_CHECKSUM_WORD, partnum);
if (current_checksum == expected_checksum)
return 1;
return 0;
}
static void
run_cmd(size_t c)
{
check_command_num(c);
if (command[c].run != NULL)
command[c].run();
}
static void
check_command_num(size_t c)
{
if (!valid_command(c))
err(EINVAL, "Invalid run_cmd arg: %lu",
(ulong)c);
}
static u8
valid_command(size_t c)
{
if (c >= N_COMMANDS)
return 0;
if (c != command[c].chk)
err(EINVAL, "Invalid cmd chk value (%lu) vs arg: %lu",
(ulong)command[c].chk, (ulong)c);
return 1;
}
static void
cmd_helper_setmac(void)
{
size_t partnum;
printf("MAC address to be written: %s\n", mac_str);
parse_mac_string();
for (partnum = 0; partnum < 2; partnum++)
write_mac_part(partnum);
}
static void
parse_mac_string(void)
{
size_t mac_byte;
if (xstrxlen(mac_str, 18) != 17)
err(EINVAL, "MAC address is the wrong length");
memset(mac_buf, 0, sizeof(mac_buf));
for (mac_byte = 0; mac_byte < 6; mac_byte++)
set_mac_byte(mac_byte);
if ((mac_buf[0] | mac_buf[1] | mac_buf[2]) == 0)
err(EINVAL, "Must not specify all-zeroes MAC address");
if (mac_buf[0] & 1)
err(EINVAL, "Must not specify multicast MAC address");
}
/*
* strnlen() but aborts on NULL input, and empty strings.
* Our version also prohibits unterminated strings.
* strnlen() was standardized in POSIX.1-2008 and is not
* available on some older systems, so we provide our own.
*/
static size_t
xstrxlen(const char *scmp, size_t maxlen)
{
size_t xstr_index;
if (scmp == NULL)
err(EINVAL, "NULL input to xstrxlen");
if (*scmp == '\0')
err(EINVAL, "Empty string in xstrxlen");
for (xstr_index = 0;
xstr_index < maxlen && scmp[xstr_index] != '\0';
xstr_index++);
if (xstr_index == maxlen)
err(EINVAL, "Unterminated string in xstrxlen");
return xstr_index;
}
static void
set_mac_byte(size_t mac_byte_pos)
{
size_t mac_str_pos = mac_byte_pos * 3;
size_t mac_nib_pos;
char separator;
if (mac_str_pos < 15) {
if ((separator = mac_str[mac_str_pos + 2]) != ':')
err(EINVAL, "Invalid MAC address separator '%c'",
separator);
}
for (mac_nib_pos = 0; mac_nib_pos < 2; mac_nib_pos++)
set_mac_nib(mac_str_pos, mac_byte_pos, mac_nib_pos);
}
static void
set_mac_nib(size_t mac_str_pos,
size_t mac_byte_pos, size_t mac_nib_pos)
{
char mac_ch;
ushort hex_num;
mac_ch = mac_str[mac_str_pos + mac_nib_pos];
if ((hex_num = hextonum(mac_ch)) > 15)
err(EINVAL, "Invalid character '%c'",
mac_str[mac_str_pos + mac_nib_pos]);
/*
* If random, ensure that local/unicast bits are set.
*/
if ((mac_byte_pos == 0) && (mac_nib_pos == 1) &&
((mac_ch | 0x20) == 'x' ||
(mac_ch == '?')))
hex_num = (hex_num & 0xE) | 2; /* local, unicast */
/*
* MAC words stored big endian in-file, little-endian
* logically, so we reverse the order.
*/
mac_buf[mac_byte_pos >> 1] |= hex_num <<
(((mac_byte_pos & 1) << 3) /* left or right byte? */
| ((mac_nib_pos ^ 1) << 2)); /* left or right nib? */
}
static ushort
hextonum(char ch_s)
{
u8 ch = (u8)ch_s;
if ((uint)(ch - '0') <= 9)
return ch - '0';
ch |= 0x20;
if ((uint)(ch - 'a') <= 5)
return ch - 'a' + 10;
if (ch == '?' || ch == 'x')
return rhex(); /* random character */
return 16; /* invalid character */
}
#if defined(HAVE_ARC4RANDOM_BUF) && \
(HAVE_ARC4RANDOM_BUF) > 0
static ushort
rhex(void)
{
static u8 num[12];
static size_t n = 0;
if (!n) {
n = 12;
arc4random_buf(num, 12);
}
return num[--n] & 0xf;
}
#else
static ushort
rhex(void)
{
struct timeval tv;
ulong mix;
static ulong counter = 0;
gettimeofday(&tv, NULL);
mix = (ulong)tv.tv_sec
^ (ulong)tv.tv_usec
^ (ulong)getpid()
^ (ulong)&mix
^ counter++
^ entropy_jitter();
/*
* Stack addresses can vary between
* calls, thus increasing entropy.
*/
mix ^= (ulong)&mix;
mix ^= (ulong)&tv;
mix ^= (ulong)&counter;
return (ushort)(mix & 0xf);
}
static ulong
entropy_jitter(void)
{
struct timeval a, b;
ulong mix = 0;
long mix_diff;
int i;
for (i = 0; i < 8; i++) {
gettimeofday(&a, NULL);
getpid();
gettimeofday(&b, NULL);
/*
* prevent negative numbers to prevent overflow,
* which would bias rand to large numbers
*/
mix_diff = (long)(b.tv_usec - a.tv_usec);
if (mix_diff < 0)
mix_diff = -mix_diff;
mix ^= (ulong)(mix_diff);
mix ^= (ulong)&mix;
}
return mix;
}
#endif
static void
write_mac_part(size_t partnum)
{
size_t w;
check_bin(partnum, "part number");
if (!part_valid[partnum])
return;
for (w = 0; w < 3; w++)
set_nvm_word(w, partnum, mac_buf[w]);
printf("Wrote MAC address to part %lu: ",
(ulong)partnum);
print_mac_from_nvm(partnum);
}
static void
cmd_helper_dump(void)
{
size_t partnum;
part_valid[0] = good_checksum(0);
part_valid[1] = good_checksum(1);
for (partnum = 0; partnum < 2; partnum++) {
if (!part_valid[partnum])
fprintf(stderr,
"BAD checksum %04x in part %lu (expected %04x)\n",
nvm_word(NVM_CHECKSUM_WORD, partnum),
(ulong)partnum,
calculated_checksum(partnum));
printf("MAC (part %lu): ",
(ulong)partnum);
print_mac_from_nvm(partnum);
hexdump(partnum);
}
}
static void
print_mac_from_nvm(size_t partnum)
{
size_t c;
ushort val16;
for (c = 0; c < 3; c++) {
val16 = nvm_word(c, partnum);
printf("%02x:%02x",
(uint)(val16 & 0xff),
(uint)(val16 >> 8));
if (c == 2)
printf("\n");
else
printf(":");
}
}
static void
hexdump(size_t partnum)
{
size_t c;
size_t row;
ushort val16;
for (row = 0; row < 8; row++) {
printf("%08lx ", (ulong)((size_t)row << 4));
for (c = 0; c < 8; c++) {
val16 = nvm_word((row << 3) + c, partnum);
if (c == 4)
printf(" ");
printf(" %02x %02x",
(uint)(val16 & 0xff),
(uint)(val16 >> 8));
}
printf("\n");
}
}
static void
cmd_helper_cat(void)
{
size_t p;
size_t ff;
size_t n = 0;
if (cmd_index == CMD_CAT16)
n = 1;
else if (cmd_index == CMD_CAT128)
n = 15;
else if (cmd_index != CMD_CAT)
err(EINVAL, "cmd_helper_cat called erroneously");
fflush(NULL);
for (p = 0; p < 2; p++) {
gbe_cat_buf(buf + (size_t)(p * GBE_PART_SIZE));
for (ff = 0; ff < n; ff++)
gbe_cat_buf(pad);
}
}
static void
gbe_cat_buf(u8 *b)
{
if (rw_file_exact(STDOUT_FILENO, b,
GBE_PART_SIZE, 0, IO_WRITE, LOOP_EAGAIN, LOOP_EINTR,
MAX_ZERO_RW_RETRY, OFF_ERR) < 0)
err(errno, "stdout: cat");
}
static void
write_gbe_file(void)
{
struct stat gbe_st;
size_t p;
size_t partnum;
u8 update_checksum;
if (command[cmd_index].flags == O_RDONLY)
return;
update_checksum = command[cmd_index].chksum_write;
override_part_modified();
if (fstat(gbe_fd, &gbe_st) == -1)
err(errno, "%s: re-check", fname);
if (gbe_st.st_dev != gbe_dev || gbe_st.st_ino != gbe_ino)
err(EIO, "%s: file replaced while open", fname);
if (gbe_st.st_size != gbe_file_size)
err(errno, "%s: file size changed before write", fname);
if (!S_ISREG(gbe_st.st_mode))
err(errno, "%s: file type changed before write", fname);
for (p = 0; p < 2; p++) {
partnum = p ^ command[cmd_index].invert;
if (!part_modified[partnum])
continue;
if (update_checksum)
set_checksum(partnum);
rw_gbe_file_part(partnum, IO_PWRITE, "pwrite");
}
}
static void
override_part_modified(void)
{
u8 mod_type = command[cmd_index].set_modified;
switch (mod_type) {
case SET_MOD_0:
set_part_modified(0);
break;
case SET_MOD_1:
set_part_modified(1);
break;
case SET_MOD_N:
set_part_modified(part ^ command[cmd_index].invert);
break;
case SET_MOD_BOTH:
set_part_modified(0);
set_part_modified(1);
break;
case SET_MOD_OFF:
break;
default:
err(EINVAL, "Unsupported set_mod type: %u",
mod_type);
}
}
static void
set_checksum(size_t p)
{
check_bin(p, "part number");
set_nvm_word(NVM_CHECKSUM_WORD, p, calculated_checksum(p));
}
static ushort
calculated_checksum(size_t p)
{
size_t c;
uint val16 = 0;
for (c = 0; c < NVM_CHECKSUM_WORD; c++)
val16 += (uint)nvm_word(c, p);
return (ushort)((NVM_CHECKSUM - val16) & 0xffff);
}
/*
* GbE NVM files store 16-bit (2-byte) little-endian words.
* We must therefore swap the order when reading or writing.
*
* NOTE: The MAC address words are stored big-endian in the
* file, but we assume otherwise and adapt accordingly.
*/
static ushort
nvm_word(size_t pos16, size_t p)
{
size_t pos;
check_nvm_bound(pos16, p);
pos = (pos16 << 1) + (p * GBE_PART_SIZE);
return (ushort)buf[pos] |
((ushort)buf[pos + 1] << 8);
}
static void
set_nvm_word(size_t pos16, size_t p, ushort val16)
{
size_t pos;
check_nvm_bound(pos16, p);
pos = (pos16 << 1) + (p * GBE_PART_SIZE);
buf[pos] = (u8)(val16 & 0xff);
buf[pos + 1] = (u8)(val16 >> 8);
set_part_modified(p);
}
static void
set_part_modified(size_t p)
{
check_bin(p, "part number");
part_modified[p] = 1;
}
static void
check_nvm_bound(size_t c, size_t p)
{
/*
* NVM_SIZE assumed as the limit, because this
* current design assumes that we will only
* ever modified the NVM area.
*/
check_bin(p, "part number");
if (c >= NVM_WORDS)
err(ECANCELED, "check_nvm_bound: out of bounds %lu",
(ulong)c);
}
static void
check_bin(size_t a, const char *a_name)
{
if (a > 1)
err(EINVAL, "%s must be 0 or 1, but is %lu",
a_name, (ulong)a);
}
static void
rw_gbe_file_part(size_t p, int rw_type,
const char *rw_type_str)
{
ssize_t r;
size_t gbe_rw_size = command[cmd_index].rw_size;
u8 invert = command[cmd_index].invert;
u8 *mem_offset;
off_t file_offset;
if (rw_type < IO_PREAD || rw_type > IO_PWRITE)
err(errno, "%s: %s: part %lu: invalid rw_type, %d",
fname, rw_type_str, (ulong)p, rw_type);
if (rw_type == IO_PWRITE)
invert = 0;
/*
* Inverted reads are used by copy/swap.
* E.g. read from p0 (file) to p1 (mem).
*/
mem_offset = gbe_mem_offset(p ^ invert, rw_type_str);
file_offset = (off_t)gbe_file_offset(p, rw_type_str);
r = rw_gbe_file_exact(gbe_fd, mem_offset,
gbe_rw_size, file_offset, rw_type);
if (r == -1)
err(errno, "%s: %s: part %lu",
fname, rw_type_str, (ulong)p);
if ((size_t)r != gbe_rw_size)
err(EIO, "%s: partial %s: part %lu",
fname, rw_type_str, (ulong)p);
}
static void
check_written_part(size_t p)
{
ssize_t r;
size_t gbe_rw_size;
u8 *mem_offset;
off_t file_offset;
u8 *buf_restore;
struct stat st;
if (!part_modified[p])
return;
gbe_rw_size = command[cmd_index].rw_size;
/* invert not needed for pwrite */
mem_offset = gbe_mem_offset(p, "pwrite");
file_offset = (off_t)gbe_file_offset(p, "pwrite");
memset(pad, 0xff, sizeof(pad));
if (fstat(gbe_fd, &st) == -1)
err(errno, "%s: fstat (post-write)", fname);
if (st.st_dev != gbe_dev || st.st_ino != gbe_ino)
err(EIO, "%s: file changed during write", fname);
r = rw_gbe_file_exact(gbe_fd, pad,
gbe_rw_size, file_offset, IO_PREAD);
if (r == -1)
rw_check_err_read[p] = io_err_gbe = 1;
else if ((size_t)r != gbe_rw_size)
rw_check_partial_read[p] = io_err_gbe = 1;
else if (memcmp(mem_offset, pad, gbe_rw_size) != 0)
rw_check_bad_part[p] = io_err_gbe = 1;
if (rw_check_err_read[p] ||
rw_check_partial_read[p])
return;
/*
* We only load one part on-file, into memory but
* always at offset zero, for post-write checks.
* That's why we hardcode good_checksum(0).
*/
buf_restore = buf;
buf = pad;
post_rw_checksum[p] = good_checksum(0);
buf = buf_restore;
}
static void
report_io_err_rw(void)
{
size_t p;
if (!io_err_gbe)
return;
for (p = 0; p < 2; p++) {
if (!part_modified[p])
continue;
if (rw_check_err_read[p])
fprintf(stderr,
"%s: pread: p%lu (post-verification)\n",
fname, (ulong)p);
if (rw_check_partial_read[p])
fprintf(stderr,
"%s: partial pread: p%lu (post-verification)\n",
fname, (ulong)p);
if (rw_check_bad_part[p])
fprintf(stderr,
"%s: pwrite: corrupt write on p%lu\n",
fname, (ulong)p);
if (rw_check_err_read[p] ||
rw_check_partial_read[p]) {
fprintf(stderr,
"%s: p%lu: skipped checksum verification "
"(because read failed)\n",
fname, (ulong)p);
continue;
}
fprintf(stderr, "%s: ", fname);
if (post_rw_checksum[p])
fprintf(stderr, "GOOD");
else
fprintf(stderr, "BAD");
fprintf(stderr, " checksum in p%lu on-disk.\n",
(ulong)p);
if (post_rw_checksum[p]) {
fprintf(stderr,
" This does NOT mean it's safe. it may be\n"
" salvageable if you use the cat feature.\n");
}
}
}
/*
* This one is similar to gbe_file_offset,
* but used to check Gbe bounds in memory,
* and it is *also* used during file I/O.
*/
static u8 *
gbe_mem_offset(size_t p, const char *f_op)
{
off_t gbe_off = gbe_x_offset(p, f_op, "mem",
GBE_PART_SIZE, GBE_FILE_SIZE);
return (u8 *)(buf + (size_t)gbe_off);
}
/*
* I/O operations filtered here. These operations must
* only write from the 0th position or the half position
* within the GbE file, and write 4KB of data.
*
* This check is called, to ensure just that.
*/
static off_t
gbe_file_offset(size_t p, const char *f_op)
{
off_t gbe_file_half_size = gbe_file_size >> 1;
return gbe_x_offset(p, f_op, "file",
gbe_file_half_size, gbe_file_size);
}
static off_t
gbe_x_offset(size_t p, const char *f_op, const char *d_type,
off_t nsize, off_t ncmp)
{
off_t off;
check_bin(p, "part number");
off = ((off_t)p) * (off_t)nsize;
if (off > ncmp - GBE_PART_SIZE)
err(ECANCELED, "%s: GbE %s %s out of bounds",
fname, d_type, f_op);
if (off != 0 && off != ncmp >> 1)
err(ECANCELED, "%s: GbE %s %s at bad offset",
fname, d_type, f_op);
return off;
}
static ssize_t
rw_gbe_file_exact(int fd, u8 *mem, size_t nrw,
off_t off, int rw_type)
{
size_t mem_addr;
size_t buf_addr;
ssize_t r;
if (io_args(fd, mem, nrw, off, rw_type) == -1)
return -1;
mem_addr = (size_t)(void *)mem;
buf_addr = (size_t)(void *)buf;
if (mem != (void *)pad) {
if (mem_addr < buf_addr)
goto err_rw_gbe_file_exact;
if ((mem_addr - buf_addr) >= (size_t)GBE_FILE_SIZE)
goto err_rw_gbe_file_exact;
}
if (off < 0 || off >= gbe_file_size)
goto err_rw_gbe_file_exact;
if (nrw > (size_t)(gbe_file_size - off))
goto err_rw_gbe_file_exact;
if (nrw > (size_t)GBE_PART_SIZE)
goto err_rw_gbe_file_exact;
r = rw_file_exact(fd, mem, nrw, off, rw_type,
NO_LOOP_EAGAIN, LOOP_EINTR, MAX_ZERO_RW_RETRY,
OFF_ERR);
return rw_over_nrw(r, nrw);
err_rw_gbe_file_exact:
errno = EIO;
return -1;
}
/*
* Safe I/O functions wrapping around
* read(), write() and providing a portable
* analog of both pread() and pwrite().
* These functions are designed for maximum
* robustness, checking NULL inputs, overflowed
* outputs, and all kinds of errors that the
* standard libc functions don't.
*
* Looping on EINTR and EAGAIN is supported.
* EINTR/EAGAIN looping is done indefinitely.
*/
/*
* rw_file_exact() - Read perfectly or die
*
* Read/write, and absolutely insist on an
* absolute read; e.g. if 100 bytes are
* requested, this MUST return 100.
*
* This function will never return zero.
* It will only return below (error),
* or above (success). On error, -1 is
* returned and errno is set accordingly.
*
* Zero-byte returns are not allowed.
* It will re-spin a finite number of
* times upon zero-return, to recover,
* otherwise it will return an error.
*/
static ssize_t
rw_file_exact(int fd, u8 *mem, size_t nrw,
off_t off, int rw_type, int loop_eagain,
int loop_eintr, size_t max_retries,
int off_reset)
{
ssize_t rv = 0;
ssize_t rc = 0;
size_t retries_on_zero = 0;
off_t off_cur;
size_t nrw_cur;
void *mem_cur;
if (io_args(fd, mem, nrw, off, rw_type) == -1)
return -1;
while (1) {
/* Prevent theoretical overflow */
if (rv >= 0 && (size_t)rv > (nrw - rc))
goto err_rw_file_exact;
rc += rv;
if ((size_t)rc >= nrw)
break;
mem_cur = (void *)(mem + (size_t)rc);
nrw_cur = (size_t)(nrw - (size_t)rc);
if (off < 0)
goto err_rw_file_exact;
off_cur = (off_t)((size_t)off + (size_t)rc);
rv = prw(fd, mem_cur, nrw_cur, off_cur,
rw_type, loop_eagain, loop_eintr,
off_reset);
if (rv < 0)
return -1;
if (rv == 0) {
if (retries_on_zero++ < max_retries)
continue;
goto err_rw_file_exact;
}
retries_on_zero = 0;
}
if ((size_t)rc != nrw)
goto err_rw_file_exact;
return rw_over_nrw(rc, nrw);
err_rw_file_exact:
errno = EIO;
return -1;
}
/*
* prw() - portable read-write
*
* This implements a portable analog of pwrite()
* and pread() - note that this version is not
* thread-safe (race conditions are possible on
* shared file descriptors).
*
* This limitation is acceptable, since nvmutil is
* single-threaded. Portability is the main goal.
*
* If you need real pwrite/pread, just compile
* with flag: HAVE_REAL_PREAD_PWRITE=1
*
* A fallback is provided for regular read/write.
* rw_type can be IO_READ, IO_WRITE, IO_PREAD
* or IO_PWRITE
*
* loop_eagain does a retry loop on EAGAIN if set
* loop_eintr does a retry loop on EINTR if set
*
* Unlike the bare syscalls, prw() does security
* checks e.g. checks NULL strings, checks bounds,
* also mitigates a few theoretical libc bugs.
* It is designed for extremely safe single-threaded
* I/O on applications that need it.
*
* NOTE: If you use loop_eagain (1), you enable wait
* loop on EAGAIN. Beware if using this on a non-blocking
* pipe (it could spin indefinitely).
*
* off_reset: if zero, and using fallback pwrite/pread
* analogs, we check if a file offset changed,
* which would indicate another thread changed
* it, and return error, without resetting the
* file - this would allow that thread to keep
* running, but we could then cause a whole
* program exit if we wanted to.
* if not zero:
* we reset and continue, and pray for the worst.
*/
static ssize_t
prw(int fd, void *mem, size_t nrw,
off_t off, int rw_type,
int loop_eagain, int loop_eintr,
int off_reset)
{
ssize_t r;
int positional_rw;
struct stat st;
#if !defined(HAVE_REAL_PREAD_PWRITE) || \
HAVE_REAL_PREAD_PWRITE < 1
int saved_errno;
off_t verified;
off_t off_orig;
off_t off_last;
#endif
if (io_args(fd, mem, nrw, off, rw_type) == -1)
return -1;
r = -1;
/* Programs like cat can use this,
so we only check if it's a normal
file if not looping EAGAIN */
if (!loop_eagain) {
/*
* Checking on every run of prw()
* is expensive if called many
* times, but is defensive in
* case the status changes.
*/
if (check_file(fd, &st) == -1)
return -1;
}
if (rw_type >= IO_PREAD)
positional_rw = 1; /* pread/pwrite */
else
positional_rw = 0; /* read/write */
try_rw_again:
if (!positional_rw) {
#if defined(HAVE_REAL_PREAD_PWRITE) && \
HAVE_REAL_PREAD_PWRITE > 0
real_pread_pwrite:
#endif
if (rw_type == IO_WRITE)
r = write(fd, mem, nrw);
else if (rw_type == IO_READ)
r = read(fd, mem, nrw);
#if defined(HAVE_REAL_PREAD_PWRITE) && \
HAVE_REAL_PREAD_PWRITE > 0
else if (rw_type == IO_PWRITE)
r = pwrite(fd, mem, nrw, off);
else if (rw_type == IO_PREAD)
r = pread(fd, mem, nrw, off);
#endif
if (r == -1 && (errno == try_err(loop_eintr, EINTR)
|| errno == try_err(loop_eagain, EAGAIN)))
goto try_rw_again;
return rw_over_nrw(r, nrw);
}
#if defined(HAVE_REAL_PREAD_PWRITE) && \
HAVE_REAL_PREAD_PWRITE > 0
goto real_pread_pwrite;
#else
if ((off_orig = lseek_loop(fd, (off_t)0, SEEK_CUR,
loop_eagain, loop_eintr)) == (off_t)-1) {
r = -1;
} else if (lseek_loop(fd, off, SEEK_SET,
loop_eagain, loop_eintr) == (off_t)-1) {
r = -1;
} else {
verified = lseek_loop(fd, (off_t)0, SEEK_CUR,
loop_eagain, loop_eintr);
/*
* Partial thread-safety: detect
* if the offset changed to what
* we previously got. If it did,
* then another thread may have
* changed it. Enabled if
* off_reset is OFF_RESET.
*
* We do this *once*, on the theory
* that nothing is touching it now.
*/
if (off_reset && off != verified)
lseek_loop(fd, off, SEEK_SET,
loop_eagain, loop_eintr);
do {
if (off != verified)
return -1;
if (rw_type == IO_PREAD)
r = read(fd, mem, nrw);
else if (rw_type == IO_PWRITE)
r = write(fd, mem, nrw);
if (rw_over_nrw(r, nrw) == -1) {
errno = EIO;
break;
}
/*
* Verify again before I/O
* (even with OFF_ERR)
*
* This implements the first check
* even with OFF_ERR, but without
* the recovery. On ERR_RESET, if
* the check fails again, then we
* know something else is touching
* the file, so it's best that we
* probably leave it alone and err.
*
* In other words, ERR_RESET only
* tolerates one change. Any more
* will cause an exit, including
* per EINTR/EAGAIN re-spin.
*/
verified = lseek_loop(fd, (off_t)0, SEEK_CUR,
loop_eagain, loop_eintr);
} while (r == -1 &&
(errno == try_err(loop_eintr, EINTR)
|| errno == try_err(loop_eagain, EAGAIN)));
}
saved_errno = errno;
off_last = lseek_loop(fd, off_orig, SEEK_SET,
loop_eagain, loop_eintr);
if (off_last != off_orig) {
errno = saved_errno;
goto err_prw;
}
errno = saved_errno;
return rw_over_nrw(r, nrw);
#endif
err_prw:
errno = EIO;
return -1;
}
static int
io_args(int fd, void *mem, size_t nrw,
off_t off, int rw_type)
{
/* obviously */
if (mem == NULL)
goto err_io_args;
/* uninitialised fd */
if (fd < 0)
goto err_io_args;
/* negative offset */
if (off < 0)
goto err_io_args;
/* prevent zero-byte rw */
if (!nrw)
goto err_io_args;
/* prevent overflow */
if (nrw > (size_t)SSIZE_MAX)
goto err_io_args;
/* prevent overflow */
if (((size_t)off + nrw) < (size_t)off)
goto err_io_args;
if (rw_type > IO_PWRITE)
goto err_io_args;
return 0;
err_io_args:
errno = EIO;
return -1;
}
static int
check_file(int fd, struct stat *st)
{
if (fstat(fd, st) == -1)
goto err_is_file;
if (!S_ISREG(st->st_mode))
goto err_is_file;
return 0;
err_is_file:
errno = EIO;
return -1;
}
/*
* Check overflows caused by buggy libc.
*
* POSIX can say whatever it wants.
* specification != implementation
*/
static ssize_t
rw_over_nrw(ssize_t r, size_t nrw)
{
/*
* If a byte length of zero
* was requested, that is
* clearly a bug. No way.
*/
if (!nrw)
goto err_rw_over_nrw;
if (r == -1)
return r;
if ((size_t)r > SSIZE_MAX) {
/*
* Theoretical buggy libc
* check. Extremely academic.
*
* Specifications never
* allow this return value
* to exceed SSIZE_MAX, but
* spec != implementation
*
* Check this after using
* [p]read() or [p]write()
*/
goto err_rw_over_nrw;
}
/*
* Theoretical buggy libc:
* Should never return a number of
* bytes above the requested length.
*/
if ((size_t)r > nrw)
goto err_rw_over_nrw;
return r;
err_rw_over_nrw:
errno = EIO;
return -1;
}
#if !defined(HAVE_REAL_PREAD_PWRITE) || \
HAVE_REAL_PREAD_PWRITE < 1
/*
* lseek_loop() does lseek() but optionally
* on an EINTR/EAGAIN wait loop. Used by prw()
* for setting offsets for positional I/O.
*/
static off_t
lseek_loop(int fd, off_t off, int whence,
int loop_eagain, int loop_eintr)
{
off_t old = -1;
do {
old = lseek(fd, off, whence);
} while (old == (off_t)-1 && (
errno == try_err(loop_eintr, EINTR) ||
errno == try_err(loop_eagain, EAGAIN)));
return old;
}
#endif
/*
* If a given error loop is enabled,
* e.g. EINTR or EAGAIN, an I/O operation
* will loop until errno isn't -1 and one
* of these, e.g. -1 and EINTR
*/
static int
try_err(int loop_err, int errval)
{
if (loop_err)
return errval;
/* errno is never negative,
so functions checking it
can use it accordingly */
return -1;
}
static void
err(int nvm_errval, const char *msg, ...)
{
va_list args;
if (errno == 0)
errno = nvm_errval;
(void)close_files();
fprintf(stderr, "%s: ", getnvmprogname());
va_start(args, msg);
vfprintf(stderr, msg, args);
va_end(args);
fprintf(stderr, ": %s", strerror(errno));
fprintf(stderr, "\n");
exit(EXIT_FAILURE);
}
static int
close_files(void)
{
int close_err_gbe = 0;
int saved_errno = errno;
if (gbe_fd > -1) {
if (close(gbe_fd) == -1)
close_err_gbe = errno;
gbe_fd = -1;
}
if (saved_errno)
errno = saved_errno;
if (close_err_gbe)
return -1;
return 0;
}
static const char *
getnvmprogname(void)
{
const char *p;
if (argv0 == NULL || *argv0 == '\0')
return "";
p = strrchr(argv0, '/');
if (p)
return p + 1;
else
return argv0;
}
static void
usage(int usage_exit)
{
const char *util = getnvmprogname();
#ifdef NVMUTIL_PLEDGE
if (pledge("stdio", NULL) == -1)
err(errno, "pledge");
#endif
fprintf(stderr,
"Modify Intel GbE NVM images e.g. set MAC\n"
"USAGE:\n"
"\t%s FILE dump\n"
"\t%s FILE setmac [MAC]\n"
"\t%s FILE swap\n"
"\t%s FILE copy 0|1\n"
"\t%s FILE cat\n"
"\t%s FILE cat16\n"
"\t%s FILE cat128\n",
util, util, util, util,
util, util, util);
if (usage_exit)
err(EINVAL, "Too few arguments");
}