kexec-tools: Fix option/argument parsing
Subsystems (2)
RFC PATCH 1/1 kexec-tools: Fix option/argument parsing
Hi,
In playing with kexec-tools I've noticed various problems with the argument passing, meaning one has to be careful to use certain forms of arguments and present them in a certain order.
The arguments end up being parsed three times, each getting more specific than the last. main() looks for general args, arch_process_options() looks for options common to all loaders for the arch, and then finally many file_type load() methods check for options specific to that filetype.
As GNU getopt works, it re-orders the argv[] pushing any args it doesn't recognise to the end. This includes arguments to valid options which are simply not recognised the first time around.
For example, this does not work:
# ./kexec -l --append "init=/bin/foo" /boot/vmlinux Cannot open `init=/bin/foo': No such file or directory
but this does:
# ./kexec -l --append="init=/bin/foo" /boot/vmlinux <joy>
Using the --opt<space>arg variant results in the first non-option argument in main() being "init=/bin/foo" which is then used as the kernel filename, failing. Also, the order affects things in unintuitive ways:
# ./kexec -l /boot/vmlinux --append "init=/bin/foo" <appears to load correctly, but command line appended with "/boot/vmlinux"!>
This behaviour is avoided by using the --opt= forms, but getopt does allow both and hence the user can have a fairly frustrating experience. (Doing something unexpected (ex. 3) is more annoying than an error exit (ex. 1) in many cases.)
The fix presented here is to create a new #define to encapsulate all possible options for an architecture build. The intention is that this set includes all possible loaders' all possible options.
main() walks through the options and silently ignores any non-general options (BUT respects that "--arg foo" should remain together, as getopt_long() does now recognise it). arch_process_options() walks through them again and responds to any arch-specific options, again ignoring but respecting any non-arch options. Finally the loader may look for its options, and find them in-order and present. Any outside of this combined set are complained-about as usual.
So, comments please. Is this a reasonable way to do it or is there an obvious better way I've missed? :-) So far I have been able to test on x86(32,64) and ppc(32,64) but none of the others. :(
Cheers,
Matt
---
The argument parsing is currently a bit broken as main()'s getopt_long() knows nothing about either the architecture-specific options or, even more specifically, the architecture-and-loader-specific options.
This patch introduces new #defines for all architectures, KEXEC_ALL_OPTIONS and KEXEC_ALL_OPT_STR. These contain all possible options for a given build, and the getopt_long() passes in main() and arch_process_options() will now recognise arch- and loader-specific options; these will not be re-ordered in argv[], there is no confusion over which argv[] entry is the kernel filename, and using '--opt=foo' and '--opt foo' both work.
All architectures have command line options (and #define OPT_BLAHs) consolidated into their include/arch/option.h files. x86_64 builds parts of i386/ as well, so now both share a single option.h file (with a symlink).
--- kexec/arch/alpha/include/arch/options.h | 9 +++++ kexec/arch/arm/include/arch/options.h | 29 +++++++++++++++++ kexec/arch/arm/kexec-zImage-arm.c | 3 +- kexec/arch/cris/include/arch/options.h | 32 ++++++++++++++++-- kexec/arch/cris/kexec-elf-cris.c | 4 +- kexec/arch/i386/include/arch/options.h | 51 ++++++++++++++++++++++++++++++ kexec/arch/i386/kexec-beoboot-x86.c | 3 +- kexec/arch/i386/kexec-bzImage.c | 6 +-- kexec/arch/i386/kexec-elf-x86.c | 7 +--- kexec/arch/i386/kexec-multiboot-x86.c | 5 +-- kexec/arch/i386/kexec-x86.c | 4 +- kexec/arch/ia64/include/arch/options.h | 33 ++++++++++++++++++- kexec/arch/ia64/kexec-elf-ia64.c | 7 +--- kexec/arch/mips/include/arch/options.h | 27 +++++++++++++++- kexec/arch/mips/kexec-elf-mips.c | 3 +- kexec/arch/ppc/include/arch/options.h | 34 ++++++++++++++++++++ kexec/arch/ppc/kexec-dol-ppc.c | 4 +- kexec/arch/ppc/kexec-elf-ppc.c | 7 +--- kexec/arch/ppc/kexec-uImage-ppc.c | 6 +-- kexec/arch/ppc64/include/arch/options.h | 37 ++++++++++++++++++++- kexec/arch/ppc64/kexec-elf-ppc64.c | 12 ++++--- kexec/arch/ppc64/kexec-ppc64.c | 12 +++---- kexec/arch/s390/include/arch/options.h | 28 ++++++++++++++++- kexec/arch/s390/kexec-image.c | 4 +-- kexec/arch/sh/include/arch/options.h | 26 +++++++++++++-- kexec/arch/sh/kexec-sh.c | 7 +++- kexec/arch/x86_64/include/arch/options.h | 23 +------------- kexec/arch/x86_64/kexec-elf-x86_64.c | 7 +--- kexec/arch/x86_64/kexec-x86_64.c | 4 +- kexec/kexec.c | 4 +- 30 files changed, 340 insertions(+), 98 deletions(-) mode change 100644 => 120000 kexec/arch/x86_64/include/arch/options.h
kexec/arch/alpha/include/arch/options.h +9 -0
@@ -3,9 +3,18 @@#define OPT_ARCH_MAX (OPT_MAX+0) +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* See the other architectures for details of these; Alpha has no +* loader-specific options yet. +*/ +#define KEXEC_ALL_OPTIONS KEXEC_ARCH_OPTIONS +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR +#endif /* KEXEC_ARCH_ALPHA_OPTIONS_H */
kexec/arch/arm/include/arch/options.h +29 -0
@@ -3,9 +3,38 @@ #define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_APPEND 'a' +#define OPT_RAMDISK 'r' + +/* Options relevant to the architecture (excluding loader-specific ones), + * in this case none: + */ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \ #define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the + * architecture's loaders. + * o main() uses this complete list to scan for its options, ignoring + * arch-specific/loader-specific ones. + * o Then, arch_process_options() uses this complete list to scan for its + * options, ignoring general/loader-specific ones. + * o Then, the file_type[n].load re-scans for options, using + * KEXEC_ARCH_OPTIONS plus its loader-specific options subset. + * Any unrecognised options cause an error here. + * + * This is done so that main()'s/arch_process_options()'s getopt_long() calls + * don't choose a kernel filename from random arguments to options they don't + * recognise -- as they now recognise (if not act upon) all possible options. + */ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS + { "command-line", 1, 0, OPT_APPEND }, + { "append", 1, 0, OPT_APPEND }, + { "initrd", 1, 0, OPT_RAMDISK }, + { "ramdisk", 1, 0, OPT_RAMDISK }, + +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR "a:r:" + #endif /* KEXEC_ARCH_ARM_OPTIONS_H */
kexec/arch/arm/kexec-zImage-arm.c +1 -2
@@ -219,8 +219,7 @@ int zImage_arm_load(int argc, char **argv, const char *buf, off_t len,off_t ramdisk_length;off_t ramdisk_offset;int opt; -#define OPT_APPEND 'a' -#define OPT_RAMDISK 'r' +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS { "command-line", 1, 0, OPT_APPEND },
kexec/arch/cris/include/arch/options.h +28 -4
@@ -1,11 +1,35 @@ -#ifndef KEXEC_ARCH_MIPS_OPTIONS_H -#define KEXEC_ARCH_MIPS_OPTIONS_H +#ifndef KEXEC_ARCH_CRIS_OPTIONS_H +#define KEXEC_ARCH_CRIS_OPTIONS_H -#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_APPEND (OPT_ARCH_MAX+0) +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" -#endif /* KEXEC_ARCH_MIPS_OPTIONS_H */ +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. +* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS \ + {"append", 1, 0, OPT_APPEND}, + +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR + +#endif /* KEXEC_ARCH_CRIS_OPTIONS_H */
kexec/arch/cris/kexec-elf-cris.c +2 -2
@@ -40,8 +40,6 @@#include <arch/options.h> #include "kexec-cris.h" -#define OPT_APPEND (OPT_ARCH_MAX+0) -int elf_cris_probe (const char *buf,off_t len) {struct mem_ehdr ehdr;
@@ -89,8 +87,10 @@ int elf_cris_load(int argc, char **argv, const char *buf, off_t len,unsigned int regs[16 ]; } cris_regframe; +/* See options.h -- add any more there, too. */ static const struct option options[] = { KEXEC_ARCH_OPTIONS + {"append" ,1 ,0 ,OPT_APPEND }, {0 ,0 ,0 ,0 }, };
kexec/arch/i386/include/arch/options.h +51 -0
68 lines not on this page · raw message
kexec/arch/i386/kexec-beoboot-x86.c +2 -1
@@ -84,7 +84,8 @@ int beoboot_load(int argc, char **argv, const char *buf, off_t UNUSED(len),int debug, real_mode_entry;int opt;int result; -#define OPT_REAL_MODE (OPT_ARCH_MAX+0) + +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS { "debug", 0, 0, OPT_DEBUG },
kexec/arch/i386/kexec-bzImage.c +2 -4
@@ -341,10 +341,8 @@ int bzImage_load(int argc, char **argv, const char *buf, off_t len,int debug, real_mode_entry;int opt;int result; -#define OPT_APPEND (OPT_ARCH_MAX+0) -#define OPT_REUSE_CMDLINE (OPT_ARCH_MAX+1) -#define OPT_RAMDISK (OPT_ARCH_MAX+2) -#define OPT_REAL_MODE (OPT_ARCH_MAX+3) + +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS { "debug", 0, 0, OPT_DEBUG },
kexec/arch/i386/kexec-elf-x86.c +1 -6
@@ -99,13 +99,8 @@ int elf_x86_load(int argc, char **argv, const char *buf, off_t len,#define ARG_STYLE_LINUX 1#define ARG_STYLE_NONE 2int opt; -#define OPT_APPEND (OPT_ARCH_MAX+0) -#define OPT_REUSE_CMDLINE (OPT_ARCH_MAX+1) -#define OPT_RAMDISK (OPT_ARCH_MAX+2) -#define OPT_ARGS_ELF (OPT_ARCH_MAX+3) -#define OPT_ARGS_LINUX (OPT_ARCH_MAX+4) -#define OPT_ARGS_NONE (OPT_ARCH_MAX+5) +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS { "command-line", 1, NULL, OPT_APPEND },
kexec/arch/i386/kexec-multiboot-x86.c +1 -4
@@ -157,10 +157,7 @@ int multiboot_x86_load(int argc, char **argv, const char *buf, off_t len,uint32_t u;int opt;int modules, mod_command_line_space; -#define OPT_CL (OPT_ARCH_MAX+0) -#define OPT_REUSE_CMDLINE (OPT_ARCH_MAX+1) -#define OPT_MOD (OPT_ARCH_MAX+2) -#define OPT_VGA (OPT_ARCH_MAX+3) +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS { "command-line", 1, 0, OPT_CL },
kexec/arch/i386/kexec-x86.c +2 -2
@@ -69,10 +69,10 @@ struct arch_options_t arch_options = {int arch_process_options (int argc,char **argv) {static const struct option options[] = { - KEXEC_ARCH_OPTIONS + KEXEC_ALL_OPTIONS {0 ,0 ,NULL ,0 }, }; -static const char short_options[] =KEXEC_ARCH_OPT_STR ; +static const char short_options[] =KEXEC_ALL_OPT_STR ;int opt;unsigned long value;char *end;
kexec/arch/ia64/include/arch/options.h +32 -1
@@ -1,11 +1,42 @@#ifndef KEXEC_ARCH_IA64_OPTIONS_H #define KEXEC_ARCH_IA64_OPTIONS_H -#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_APPEND (OPT_ARCH_MAX+0) +#define OPT_RAMDISK (OPT_ARCH_MAX+1) +#define OPT_NOIO (OPT_ARCH_MAX+2) +#define OPT_VMM (OPT_ARCH_MAX+3) +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. +* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS \ + {"command-line", 1, 0, OPT_APPEND}, \ + {"append", 1, 0, OPT_APPEND}, \ + {"initrd", 1, 0, OPT_RAMDISK}, \ + {"noio", 0, 0, OPT_NOIO}, \ + {"vmm", 1, 0, OPT_VMM}, \ + +#define KEXEC_ALL_OPT_STR KEXEC_OPT_STR +#endif /* KEXEC_ARCH_IA64_OPTIONS_H */
kexec/arch/ia64/kexec-elf-ia64.c +2 -5
@@ -45,11 +45,6 @@#include "crashdump-ia64.h" #include <arch/options.h> -#define OPT_APPEND (OPT_ARCH_MAX+0) -#define OPT_RAMDISK (OPT_ARCH_MAX+1) -#define OPT_NOIO (OPT_ARCH_MAX+2) -#define OPT_VMM (OPT_ARCH_MAX+3) -static const int probe_debug =0 ;extern unsigned long saved_efi_memmap_size;
@@ -136,6 +131,8 @@ int elf_ia64_load(int argc, char **argv, const char *buf, off_t len,int result;int opt;char *efi_memmap_buf, *boot_param; + +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS {"command-line", 1, 0, OPT_APPEND},
kexec/arch/mips/include/arch/options.h +26 -1
@@ -1,11 +1,36 @@#ifndef KEXEC_ARCH_MIPS_OPTIONS_H #define KEXEC_ARCH_MIPS_OPTIONS_H -#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_APPEND (OPT_ARCH_MAX+0) +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. +* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS \ + {"command-line", 1, 0, OPT_APPEND}, \ + {"append", 1, 0, OPT_APPEND}, + +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR +#endif /* KEXEC_ARCH_MIPS_OPTIONS_H */
kexec/arch/mips/kexec-elf-mips.c +2 -1
@@ -35,7 +35,6 @@ static const int probe_debug = 0;#define BOOTLOADER "kexec"#define MAX_COMMAND_LINE 256#define UPSZ (X ) ((sizeof(X) + 3) & ~3) -#define OPT_APPEND (OPT_ARCH_MAX+0)static char cmdline_buf[256 ] ="kexec " ;int elf_mips_probe (const char *buf,off_t len)
@@ -82,6 +81,8 @@ int elf_mips_load(int argc, char **argv, const char *buf, off_t len,unsigned long cmdline_addr;size_t i;unsigned long bss_start =0 , bss_size =0 ; + +/* See options.h if adding any more options. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS {"command-line", 1, 0, OPT_APPEND},
kexec/arch/ppc/include/arch/options.h +34 -0
@@ -3,9 +3,43 @@#define OPT_ARCH_MAX (OPT_MAX+0) +/* All 'local' loader options: */ +#define OPT_APPEND (OPT_ARCH_MAX+0) +#define OPT_GAMECUBE (OPT_ARCH_MAX+1) +#define OPT_DTB (OPT_ARCH_MAX+2) +#define OPT_NODES (OPT_ARCH_MAX+3) + +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. +* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS \ + {"command-line", 1, 0, OPT_APPEND},\ + {"append", 1, 0, OPT_APPEND},\ + {"gamecube", 1, 0, OPT_GAMECUBE},\ + {"dtb", 1, 0, OPT_DTB},\ + {"reuse-node", 1, 0, OPT_NODES},\ + {"debug", 0, 0, OPT_DEBUG}, + +#define KEXEC_ALL_OPT_STR KEXEC_OPT_STR +#endif /* KEXEC_ARCH_PPC_OPTIONS_H */
kexec/arch/ppc/kexec-dol-ppc.c +2 -2
@@ -335,8 +335,8 @@ int dol_ppc_load(int argc, char **argv, const char *buf, off_t UNUSED(len),unsigned long lowest_start;int i, j, k;int opt; -#define OPT_APPEND (OPT_ARCH_MAX+0) +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS {"debug", 0, 0, OPT_DEBUG},
@@ -344,7 +344,7 @@ int dol_ppc_load(int argc, char **argv, const char *buf, off_t UNUSED(len), {"append", 1, 0, OPT_APPEND}, {0, 0, 0, 0}, }; -static const char short_options[] =KEXEC_ARCH_OPT_STR "d" ; +static const char short_options[] =KEXEC_ARCH_OPT_STR ;/* * Parse the command line arguments
kexec/arch/ppc/kexec-elf-ppc.c +2 -5
@@ -113,10 +113,7 @@ static void gamecube_hack_addresses(struct mem_ehdr *ehdr) } } -#define OPT_APPEND (OPT_ARCH_MAX+0) -#define OPT_GAMECUBE (OPT_ARCH_MAX+1) -#define OPT_DTB (OPT_ARCH_MAX+2) -#define OPT_NODES (OPT_ARCH_MAX+3) +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS {"command-line", 1, 0, OPT_APPEND},
@@ -126,7 +123,7 @@ static const struct option options[] = { {"reuse-node", 1, 0, OPT_NODES}, {0, 0, 0, 0}, }; -static const char short_options[] =KEXEC_ARCH_OPT_STR "d" ; +static const char short_options[] =KEXEC_ARCH_OPT_STR ;void elf_ppc_usage (void ) {
kexec/arch/ppc/kexec-uImage-ppc.c +2 -4
@@ -14,9 +14,7 @@#include "fixup_dtb.h" #include <kexec-uImage.h> -#define OPT_APPEND (OPT_ARCH_MAX+0) -#define OPT_DTB (OPT_ARCH_MAX+1) -#define OPT_NODES (OPT_ARCH_MAX+2) +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS {"command-line", 1, 0, OPT_APPEND},
@@ -25,7 +23,7 @@ static const struct option options[] = { {"reuse-node", 1, 0, OPT_NODES}, {0, 0, 0, 0}, }; -static const char short_options[] =KEXEC_ARCH_OPT_STR "d" ; +static const char short_options[] =KEXEC_ARCH_OPT_STR ;void uImage_ppc_usage (void ) {
kexec/arch/ppc64/include/arch/options.h +35 -2
@@ -1,13 +1,46 @@#ifndef KEXEC_ARCH_PPC64_OPTIONS_H #define KEXEC_ARCH_PPC64_OPTIONS_H -#define OPT_ARCH_MAX (OPT_MAX+0) -#define OPT_ELF64_CORE (OPT_MAX+1) +#define OPT_ELF64_CORE (OPT_MAX+0) +#define OPT_ARCH_MAX (OPT_MAX+1) +/* All 'local' loader options: */ +#define OPT_APPEND (OPT_ARCH_MAX+0) +#define OPT_RAMDISK (OPT_ARCH_MAX+1) +#define OPT_DEVICETREEBLOB (OPT_ARCH_MAX+2) +#define OPT_ARGS_IGNORE (OPT_ARCH_MAX+3) + +/* Options relevant to the architecture (excluding loader-specific ones): */ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \ { "elf64-core-headers", 0, 0, OPT_ELF64_CORE }, \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. +* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS \ + { "command-line", 1, NULL, OPT_APPEND }, \ + { "append", 1, NULL, OPT_APPEND }, \ + { "ramdisk", 1, NULL, OPT_RAMDISK }, \ + { "initrd", 1, NULL, OPT_RAMDISK }, \ + { "devicetreeblob", 1, NULL, OPT_DEVICETREEBLOB }, \ + { "args-linux", 0, NULL, OPT_ARGS_IGNORE }, + +#define KEXEC_ALL_OPT_STR KEXEC_OPT_STR + +#endif /* KEXEC_ARCH_PPC64_OPTIONS_H */
kexec/arch/ppc64/kexec-elf-ppc64.c +7 -5
@@ -93,11 +93,7 @@ int elf_ppc64_load(int argc, char **argv, const char *buf, off_t len,uint32_t my_run_at_load;unsigned int slave_code[256 /sizeof (unsigned int )], master_entry; -#define OPT_APPEND (OPT_ARCH_MAX+0) -#define OPT_RAMDISK (OPT_ARCH_MAX+1) -#define OPT_DEVICETREEBLOB (OPT_ARCH_MAX+2) -#define OPT_ARGS_IGNORE (OPT_ARCH_MAX+3) - +/* See options.h -- add any more there, too. */ static const struct option options[] = {KEXEC_ARCH_OPTIONS { "command-line", 1, NULL, OPT_APPEND },
@@ -347,5 +343,11 @@ int elf_ppc64_load(int argc, char **argv, const char *buf, off_t len,void elf_ppc64_usage (void ) { +fprintf (stderr," --command-line=<Command line> command line to append.\n" ); +fprintf (stderr," --append=<Command line> same as --command-line.\n" ); +fprintf (stderr," --ramdisk=<filename> Initial RAM disk.\n" ); +fprintf (stderr," --initrd=<filename> same as --ramdisk.\n" ); +fprintf (stderr," --devicetreeblob=<filename> Specify device tree blob file.\n" ); +fprintf (stderr,"elf support is still broken\n" ); }
kexec/arch/ppc64/kexec-ppc64.c +5 -7
@@ -735,11 +735,6 @@ int file_types = sizeof(file_type) / sizeof(file_type[0]);void arch_usage (void ) { -fprintf (stderr," --command-line=<Command line> command line to append.\n" ); -fprintf (stderr," --append=<Command line> same as --command-line.\n" ); -fprintf (stderr," --ramdisk=<filename> Initial RAM disk.\n" ); -fprintf (stderr," --initrd=<filename> same as --ramdisk.\n" ); -fprintf (stderr," --devicetreeblob=<filename> Specify device tree blob file.\n" );fprintf (stderr," --elf64-core-headers Prepare core headers in ELF64 format\n" ); }
@@ -749,11 +744,14 @@ struct arch_options_t arch_options = {int arch_process_options (int argc,char **argv) { +/* We look for all options so getopt_long doesn't start reordering +* argv[] before file_type[n].load() gets a look in. +*/ static const struct option options[] = { - KEXEC_ARCH_OPTIONS + KEXEC_ALL_OPTIONS {0 ,0 ,NULL ,0 }, }; -static const char short_options[] =KEXEC_ARCH_OPT_STR ; +static const char short_options[] =KEXEC_ALL_OPT_STR ;int opt; opterr =0 ;/* Don't complain about unrecognized options here */
kexec/arch/s390/include/arch/options.h +27 -1
@@ -1,11 +1,37 @@#ifndef KEXEC_ARCH_S390_OPTIONS_H #define KEXEC_ARCH_S390_OPTIONS_H -#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_APPEND OPT_MAX+0 +#define OPT_RAMDISK OPT_MAX+1 +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. +* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS \ + {"command-line", 1, 0, OPT_APPEND}, \ + {"initrd", 1, 0, OPT_RAMDISK}, + +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR +#endif /* KEXEC_ARCH_S390_OPTIONS_H */
kexec/arch/s390/kexec-image.c +1 -3
@@ -19,9 +19,7 @@#include <getopt.h> #include "../../kexec.h" #include "kexec-s390.h" - -#define OPT_APPEND OPT_MAX+0 -#define OPT_RAMDISK OPT_MAX+1 +#include <arch/options.h> int image_s390_load(int argc, char **argv, const char *kernel_buf,
kexec/arch/sh/include/arch/options.h +23 -3
@@ -7,16 +7,36 @@#define OPT_NBSD_HOWTO (OPT_ARCH_MAX+3)#define OPT_NBSD_MROOT (OPT_ARCH_MAX+4) - +/* Options relevant to the architecture (excluding loader-specific ones): */ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \ {"command-line", 1, 0, OPT_APPEND}, \ {"append", 1, 0, OPT_APPEND}, \ {"empty-zero", 1, 0, OPT_APPEND}, \ {"howto", 1, 0, OPT_NBSD_HOWTO}, \ - {"miniroot", 1, 0, OPT_NBSD_MROOT}, \ - + {"miniroot", 1, 0, OPT_NBSD_MROOT}, +/* These options seem to be loader-specific rather than cris-specific, so +* ought to be moved to KEXEC_ALL_OPTIONS below and parsed in the relevant +* loader, e.g. kexec-netbsd-sh.c +*/ #define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. +* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS KEXEC_ARCH_OPTIONS +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR +#endif /* KEXEC_ARCH_SH_OPTIONS_H */
kexec/arch/sh/kexec-sh.c +6 -1
@@ -97,8 +97,13 @@ void arch_usage(void)int arch_process_options (int argc,char **argv) { +/* The common options amongst loaders (e.g. --append) should be read +* here, and the loader-specific options (e.g. NetBSD stuff) should +* then be re-parsed in the loader. +* (e.g. in kexec-netbsd-sh.c, for example.) +*/ static const struct option options[] = { - KEXEC_ARCH_OPTIONS + KEXEC_ALL_OPTIONS {0 ,0 ,NULL ,0 }, };static const char short_options[] =KEXEC_ARCH_OPT_STR ;
kexec/arch/x86_64/include/arch/options.h deleted +0 -22
@@ -1,22 +0,0 @@ -#ifndef KEXEC_ARCH_X86_64_OPTIONS_H -#define KEXEC_ARCH_X86_64_OPTIONS_H - -#define OPT_RESET_VGA (OPT_MAX+0) -#define OPT_SERIAL (OPT_MAX+1) -#define OPT_SERIAL_BAUD (OPT_MAX+2) -#define OPT_CONSOLE_VGA (OPT_MAX+3) -#define OPT_CONSOLE_SERIAL (OPT_MAX+4) -#define OPT_ARCH_MAX (OPT_MAX+5) - -#define KEXEC_ARCH_OPTIONS \ - KEXEC_OPTIONS \ - { "reset-vga", 0, 0, OPT_RESET_VGA }, \ - { "serial", 1, 0, OPT_SERIAL }, \ - { "serial-baud", 1, 0, OPT_SERIAL_BAUD }, \ - { "console-vga", 0, 0, OPT_CONSOLE_VGA }, \ - { "console-serial", 0, 0, OPT_CONSOLE_SERIAL }, \ - -#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" - -#endif /* KEXEC_ARCH_X86_64_OPTIONS_H */ -
kexec/arch/x86_64/include/arch/options.h new file +1 -0
@@ -0,0 +1 @@ +../../../i386/include/arch/options.h
kexec/arch/x86_64/kexec-elf-x86_64.c +1 -6
@@ -98,13 +98,8 @@ int elf_x86_64_load(int argc, char **argv, const char *buf, off_t len,#define ARG_STYLE_LINUX 1#define ARG_STYLE_NONE 2int opt; -#define OPT_APPEND (OPT_ARCH_MAX+0) -#define OPT_REUSE_CMDLINE (OPT_ARCH_MAX+1) -#define OPT_RAMDISK (OPT_ARCH_MAX+2) -#define OPT_ARGS_ELF (OPT_ARCH_MAX+3) -#define OPT_ARGS_LINUX (OPT_ARCH_MAX+4) -#define OPT_ARGS_NONE (OPT_ARCH_MAX+5) +/* See options.h and add any new options there too! */ static const struct option options[] = {KEXEC_ARCH_OPTIONS { "command-line", 1, NULL, OPT_APPEND },
kexec/arch/x86_64/kexec-x86_64.c +2 -2
@@ -74,10 +74,10 @@ static struct {int arch_process_options (int argc,char **argv) {static const struct option options[] = { - KEXEC_ARCH_OPTIONS + KEXEC_ALL_OPTIONS {0 ,0 ,NULL ,0 }, }; -static const char short_options[] =KEXEC_ARCH_OPT_STR ; +static const char short_options[] =KEXEC_ALL_OPT_STR ;int opt;unsigned long value;char *end;
kexec/kexec.c +2 -2
@@ -1062,10 +1062,10 @@ int main(int argc, char *argv[])int result =0 ;int fileind;static const struct option options[] = { - KEXEC_ARCH_OPTIONS + KEXEC_ALL_OPTIONS {0 ,0 ,0 ,0 }, }; -static const char short_options[] =KEXEC_OPT_STR ; +static const char short_options[] =KEXEC_ALL_OPT_STR ;arch_init ();
Signature · 7 lines
1.6.3.3 _______________________________________________ kexec mailing list [email protected] http://lists.infradead.org/mailman/listinfo/kexec
Quoted 7 lines · Matt Evans
Hi,
In playing with kexec-tools I've noticed various problems with the argument passing, meaning one has to be careful to use certain forms of arguments and present them in a certain order.
Quoted 8 lines
The arguments end up being parsed three times, each getting more specific than the last.
main()looks for general args,arch_process_options()looks for options common to all loaders for the arch, and then finally manyfile_typeload()methods check for options specific to that filetype.As GNU getopt works, it re-orders the argv[] pushing any args it doesn't recognise to the end. This includes arguments to valid options which are simply not recognised the first time around.
This is a long standing problem; I found it years ago. My patch saved a copy of the original argc and then passed that original version to each phase instead of the mangled vector.
I hope your suggestion is accepted or prompts a solution, as the problem can be quite vexing...
Signature · 7 lines
David N. Lombard, Intel, Irvine, CA I do not speak for Intel Corporation; all comments are strictly my own. _______________________________________________ kexec mailing list [email protected] http://lists.infradead.org/mailman/listinfo/kexec
Quoted 48 lines · Matt Evans
Hi,
In playing with kexec-tools I've noticed various problems with the argument passing, meaning one has to be careful to use certain forms of arguments and present them in a certain order.
The arguments end up being parsed three times, each getting more specific than the last.
main()looks for general args,arch_process_options()looks for options common to all loaders for the arch, and then finally manyfile_typeload()methods check for options specific to that filetype.As GNU getopt works, it re-orders the argv[] pushing any args it doesn't recognise to the end. This includes arguments to valid options which are simply not recognised the first time around.
For example, this does not work:
# ./kexec -l --append "init=/bin/foo" /boot/vmlinux Cannot open `init=/bin/foo': No such file or directorybut this does:
# ./kexec -l --append="init=/bin/foo" /boot/vmlinux <joy>Using the --opt<space>arg variant results in the first non-option argument in
main()being "init=/bin/foo" which is then used as the kernel filename, failing. Also, the order affects things in unintuitive ways:# ./kexec -l /boot/vmlinux --append "init=/bin/foo" <appears to load correctly, but command line appended with "/boot/vmlinux"!>This behaviour is avoided by using the --opt= forms, but getopt does allow both and hence the user can have a fairly frustrating experience. (Doing something unexpected (ex. 3) is more annoying than an error exit (ex. 1) in many cases.)
The fix presented here is to create a new #define to encapsulate all possible options for an architecture build. The intention is that this set includes all possible loaders' all possible options.
main()walks through the options and silently ignores any non-general options (BUT respects that "--arg foo" should remain together, asgetopt_long()does now recognise it).arch_process_options()walks through them again and responds to any arch-specific options, again ignoring but respecting any non-arch options. Finally the loader may look for its options, and find them in-order and present. Any outside of this combined set are complained-about as usual.
Quoted 3 lines
So, comments please. Is this a reasonable way to do it or is there an obvious better way I've missed? :-) So far I have been able to test on
x86(32,64)andppc(32,64)but none of the others. :(
This seems reasonable to me.
I've compiled tested the code on all architectures except cris (I don't have my build environment at the moment). I found a minor problem on arm which I have noted below.
[snip]
Quoted 28 lines
kexec/arch/arm/include/arch/options.h +15
-0@@ -3,9 +3,38 @@#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_APPEND 'a' +#define OPT_RAMDISK 'r' + +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here.
kexec/arch/arm/include/arch/options.h line 35
Quoted 10 lines
kexec/arch/arm/include/arch/options.h +10
-0+* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS + { "command-line", 1, 0, OPT_APPEND }, + { "append", 1, 0, OPT_APPEND }, + { "initrd", 1, 0, OPT_RAMDISK },
The above 4 lines seem to be missing a trailing '\'
kexec/arch/arm/include/arch/options.h line 40
Quoted 5 lines
kexec/arch/arm/include/arch/options.h +4
-0+ { "ramdisk", 1, 0, OPT_RAMDISK }, + +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR "a:r:" +#endif /* KEXEC_ARCH_ARM_OPTIONS_H */
[snip]
_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
Quoted 33 lines · In message <[email protected]> you wrote:
Quoted 32 lines · Matt Evans
Hi,
In playing with kexec-tools I've noticed various problems with the argument passing, meaning one has to be careful to use certain forms of arguments and present them in a certain order.
The arguments end up being parsed three times, each getting more specific than the last.
main()looks for general args,arch_process_options()looks for options common to all loaders for the arch, and then finally manyfile_typeload()methods check for options specific to that filetype.As GNU getopt works, it re-orders the argv[] pushing any args it doesn't recognise to the end. This includes arguments to valid options which are simply not recognised the first time around.
For example, this does not work:
# ./kexec -l --append "init=/bin/foo" /boot/vmlinux Cannot open `init=/bin/foo': No such file or directorybut this does:
# ./kexec -l --append="init=/bin/foo" /boot/vmlinux <joy>Using the --opt<space>arg variant results in the first non-option argument in
main()being "init=/bin/foo" which is then used as the kernel filename, failing. Also, the order affects things in unintuitive ways:# ./kexec -l /boot/vmlinux --append "init=/bin/foo" <appears to load correctly, but command line appended with "/boot/vmlinux"!
Quoted 6 lines
Quoted 6 lines
This behaviour is avoided by using the --opt= forms, but getopt does allow both and hence the user can have a fairly frustrating experience. (Doing something unexpected (ex. 3) is more annoying than an error exit (ex. 1) in many cases.)
The fix presented here is to create a new #define to encapsulate all possib
le
Quoted 15 lines
Quoted 11 lines
options for an architecture build. The intention is that this set includes all possible loaders' all possible options.
main()walks through the options and silently ignores any non-general options (BUT respects that "--arg foo" should remain together, asgetopt_long()does now recognise it).arch_process_options()walks through them again and responds to any arch-specific options, again ignoring but respecting any non-arch options. Finally the loader may look for its options, and find them in-order and present. Any outside of this combined set are complained-about as usual.Quoted 3 lines
So, comments please. Is this a reasonable way to do it or is there an obvious better way I've missed? :-) So far I have been able to test on
x86(32,64)andppc(32,64)but none of the others. :(
Quoted 5 lines
This seems reasonable to me.
I've compiled tested the code on all architectures except cris (I don't have my build environment at the moment). I found a minor problem on arm which I have noted below.
I suspect it'll break someones kexec scripts, so maybe we take this patch (or something like it) but bump up the release revision to 2.1?
Mikey
Quoted 4 lines
[snip]
kexec/arch/arm/include
Quoted 1 line
kexec/arch/arm/include renamed from kexec/arch/arm/include/arch/options.h +0
-0
/arch/options.h
Quoted 16 lines
Quoted 16 lines
index 248230b..a76539e 100644
kexec/arch/arm/include/arch/options.h +6
-0@@ -3,9 +3,38 @@#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_APPEND 'a' +#define OPT_RAMDISK 'r' + +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \
Quoted 5 lines
kexec/arch/arm/include/arch/options.h
Quoted 5 lines
kexec/arch/arm/include/arch/options.h +3
-0#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignorin
g
Quoted 2 lines
kexec/arch/arm/include/arch/options.h line 21
Quoted 2 lines
kexec/arch/arm/include/arch/options.h +2
-0+* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan fo
r its
Quoted 6 lines
kexec/arch/arm/include/arch/options.h
Quoted 6 lines
kexec/arch/arm/include/arch/options.h +6
-0+* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. +* +* This is done so that main()'s/arch_process_options()'s getopt_long() ca
lls
Quoted 1 line
kexec/arch/arm/include/arch/options.h line 28
Quoted 1 line
kexec/arch/arm/include/arch/options.h +1
-0+* don't choose a kernel filename from random arguments to options they do
n't
Quoted 1 line
kexec/arch/arm/include/arch/options.h line 29
Quoted 1 line
kexec/arch/arm/include/arch/options.h +1
-0+* recognise -- as they now recognise (if not act upon) all possible optio
ns.
Quoted 9 lines
kexec/arch/arm/include/arch/options.h line 35
Quoted 6 lines
kexec/arch/arm/include/arch/options.h +6
-0+*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS + { "command-line", 1, 0, OPT_APPEND }, + { "append", 1, 0, OPT_APPEND }, + { "initrd", 1, 0, OPT_RAMDISK },The above 4 lines seem to be missing a trailing '\'
Quoted 8 lines
Quoted 5 lines
kexec/arch/arm/include/arch/options.h +4
-0+ { "ramdisk", 1, 0, OPT_RAMDISK }, + +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR "a:r:" +#endif /* KEXEC_ARCH_ARM_OPTIONS_H */ [snip]
_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
Quoted 53 lines · Michael Neuling
Quoted 29 lines · In message <[email protected]> you wrote:
Quoted 28 lines · Matt Evans
Hi,
In playing with kexec-tools I've noticed various problems with the argument passing, meaning one has to be careful to use certain forms of arguments and present them in a certain order.
The arguments end up being parsed three times, each getting more specific than the last.
main()looks for general args,arch_process_options()looks for options common to all loaders for the arch, and then finally manyfile_typeload()methods check for options specific to that filetype.As GNU getopt works, it re-orders the argv[] pushing any args it doesn't recognise to the end. This includes arguments to valid options which are simply not recognised the first time around.
For example, this does not work:
# ./kexec -l --append "init=/bin/foo" /boot/vmlinux Cannot open `init=/bin/foo': No such file or directorybut this does:
# ./kexec -l --append="init=/bin/foo" /boot/vmlinux <joy>Using the --opt<space>arg variant results in the first non-option argument in
main()being "init=/bin/foo" which is then used as the kernel filename, failing. Also, the order affects things in unintuitive ways:Quoted 8 lines
Quoted 8 lines
# ./kexec -l /boot/vmlinux --append "init=/bin/foo" <appears to load correctly, but command line appended with "/boot/vmlinux"! This behaviour is avoided by using the --opt= forms, but getopt does allow both and hence the user can have a fairly frustrating experience. (Doing something unexpected (ex. 3) is more annoying than an error exit (ex. 1) in many cases.)The fix presented here is to create a new #define to encapsulate all possib
le
Quoted 14 lines
Quoted 14 lines
options for an architecture build. The intention is that this set includes all possible loaders' all possible options.
main()walks through the options and silently ignores any non-general options (BUT respects that "--arg foo" should remain together, asgetopt_long()does now recognise it).arch_process_options()walks through them again and responds to any arch-specific options, again ignoring but respecting any non-arch options. Finally the loader may look for its options, and find them in-order and present. Any outside of this combined set are complained-about as usual.So, comments please. Is this a reasonable way to do it or is there an obvious better way I've missed? :-) So far I have been able to test on
x86(32,64)andppc(32,64)but none of the others. :(
Quoted 8 lines
Quoted 5 lines
This seems reasonable to me.
I've compiled tested the code on all architectures except cris (I don't have my build environment at the moment). I found a minor problem on arm which I have noted below.
I suspect it'll break someones kexec scripts, so maybe we take this patch (or something like it) but bump up the release revision to 2.1?
It /should/ not, as I haven't changed existing working behaviour -- unless someone with a script is relying on existing broken behaviour? I guess we've all been burned at least once by saying, "but SURELY nobody would be doing that", but hey. :)
Cheers,
Matt
_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
Quoted 7 lines · In message <[email protected]> you wrote:
Quoted 6 lines · Michael Neuling
Quoted 5 lines · In message <[email protected]> you wrote:
Quoted 4 lines · Matt Evans
Hi,
In playing with kexec-tools I've noticed various problems with the argume
nt
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
passing, meaning one has to be careful to use certain forms of arguments and present them in a certain order.
The arguments end up being parsed three times, each getting more specific than the last.
main()looks for general args,arch_process_options()loo
ks
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
for options common to all loaders for the arch, and then finally many
file_typeload()methods check for options specific to that filetype.As GNU getopt works, it re-orders the argv[] pushing any args it doesn't recognise to the end. This includes arguments to valid options which are simply not recognised the first time around.
For example, this does not work:
# ./kexec -l --append "init=/bin/foo" /boot/vmlinux Cannot open `init=/bin/foo': No such file or directory
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
but this does:
# ./kexec -l --append="init=/bin/foo" /boot/vmlinux <joy>Using the --opt<space>arg variant results in the first non-option argumen
t
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
in
main()being "init=/bin/foo" which is then used as the kernel filename
,
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
failing. Also, the order affects things in unintuitive ways:
# ./kexec -l /boot/vmlinux --append "init=/bin/foo" <appears to load correctly, but command line appended with "/boot/vmlinux
"!
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
This behaviour is avoided by using the --opt= forms, but getopt does allo
w
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
both and hence the user can have a fairly frustrating experience. (Doing something unexpected (ex. 3) is more annoying than an error exit (ex. 1) in many cases.)
The fix presented here is to create a new #define to encapsulate all poss
ib
Quoted 2 lines
Quoted 2 lines
le
Quoted 1 line
Quoted 1 line
options for an architecture build. The intention is that this set includ
es
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
all possible loaders' all possible options.
main()walks through the options and silently ignores any non-general options (BUT respects that "--arg foo" should remain together, asgetopt_long()does now recognise it).arch_process_options()walks throu
gh
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
them again and responds to any arch-specific options, again ignoring but respecting any non-arch options. Finally the loader may look for its options, and find them in-order and present. Any outside of this combine
d
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
Quoted 5 lines
set are complained-about as usual.
So, comments please. Is this a reasonable way to do it or is there an obvious better way I've missed? :-) So far I have been able to test on
x86(32,64)andppc(32,64)but none of the others. :(This seems reasonable to me.
I've compiled tested the code on all architectures except cris (I don't have my build environment at the moment). I found a minor problem on arm which I have noted below.
Quoted 7 lines
Quoted 2 lines
I suspect it'll break someones kexec scripts, so maybe we take this patch (or something like it) but bump up the release revision to 2.1?
It /should/ not, as I haven't changed existing working behaviour -- unless someone with a script is relying on existing broken behaviour? I guess we've all been burned at least once by saying, "but SURELY nobody would be doing that", but hey. :)
Yeah, that's what I'm concerned about. Users are pretty dumb ;-)
Mikey
_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
Quoted 7 lines · Michael Neuling
Quoted 6 lines · In message <[email protected]> you wrote:
Quoted 5 lines · Matt Evans
In playing with kexec-tools I've noticed various problems with the argument passing, meaning one has to be careful to use certain forms of arguments and present them in a certain order.
[deletia]
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
This behaviour is avoided by using the --opt= forms, but getopt does allow both and hence the user can have a fairly frustrating experience. (Doing something unexpected (ex. 3) is more annoying than an error exit (ex. 1) in many cases.)
Quoted 9 lines
Quoted 5 lines
This seems reasonable to me.
I've compiled tested the code on all architectures except cris (I don't have my build environment at the moment). I found a minor problem on arm which I have noted below.
I suspect it'll break someones kexec scripts, so maybe we take this patch (or something like it) but bump up the release revision to 2.1?
How?
Command lines that previously worked will continue to work. Command lines that should have worked, but didn't, will now work. Command lines that shouldn't have worked will still not work.
The only scripts that may fail are those doing negative testing to check for a form that *should* have been allowed--quite clearly, any such negative testing was incorrect.
Signature · 7 lines
David N. Lombard, Intel, Irvine, CA I do not speak for Intel Corporation; all comments are strictly my own. _______________________________________________ kexec mailing list [email protected] http://lists.infradead.org/mailman/listinfo/kexec
Quoted 5 lines · In message <[email protected]> you wrote:
Quoted 4 lines · Michael Neuling
Quoted 3 lines · In message <[email protected]> you wrote:
Quoted 2 lines · Matt Evans
In playing with kexec-tools I've noticed various problems with the argu
ment
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
passing, meaning one has to be careful to use certain forms of argument
s
Quoted 5 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
and present them in a certain order.
[deletia]
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
This behaviour is avoided by using the --opt= forms, but getopt does al
low
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
both and hence the user can have a fairly frustrating experience. (Doi
ng
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
something unexpected (ex. 3) is more annoying than an error exit (ex. 1
)
Quoted 14 lines
Quoted 12 lines
Quoted 8 lines
Quoted 2 lines
in many cases.)
This seems reasonable to me.
I've compiled tested the code on all architectures except cris (I don't have my build environment at the moment). I found a minor problem on arm which I have noted below.
I suspect it'll break someones kexec scripts, so maybe we take this patch (or something like it) but bump up the release revision to 2.1?
How?
Quoted 7 lines
Command lines that previously worked will continue to work. Command lines that should have worked, but didn't, will now work. Command lines that shouldn't have worked will still not work.
The only scripts that may fail are those doing negative testing to check for a form that *should* have been allowed--quite clearly, any such negative testing was incorrect.
Yeah, and like I said, users are pretty dumb, so put 1 and 1 together... :-)
That being said, I don't have a strong opinion. If others think it's unlikely enough that anyone will hit it, then let's keep the release numbering as it is.
... and I think we all agree we need the patch.
Mikey
_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
Quoted 53 lines · Simon Horman
Quoted 48 lines · Matt Evans
Hi,
In playing with kexec-tools I've noticed various problems with the argument passing, meaning one has to be careful to use certain forms of arguments and present them in a certain order.
The arguments end up being parsed three times, each getting more specific than the last.
main()looks for general args,arch_process_options()looks for options common to all loaders for the arch, and then finally manyfile_typeload()methods check for options specific to that filetype.As GNU getopt works, it re-orders the argv[] pushing any args it doesn't recognise to the end. This includes arguments to valid options which are simply not recognised the first time around.
For example, this does not work:
# ./kexec -l --append "init=/bin/foo" /boot/vmlinux Cannot open `init=/bin/foo': No such file or directorybut this does:
# ./kexec -l --append="init=/bin/foo" /boot/vmlinux <joy>Using the --opt<space>arg variant results in the first non-option argument in
main()being "init=/bin/foo" which is then used as the kernel filename, failing. Also, the order affects things in unintuitive ways:# ./kexec -l /boot/vmlinux --append "init=/bin/foo" <appears to load correctly, but command line appended with "/boot/vmlinux"!>This behaviour is avoided by using the --opt= forms, but getopt does allow both and hence the user can have a fairly frustrating experience. (Doing something unexpected (ex. 3) is more annoying than an error exit (ex. 1) in many cases.)
The fix presented here is to create a new #define to encapsulate all possible options for an architecture build. The intention is that this set includes all possible loaders' all possible options.
main()walks through the options and silently ignores any non-general options (BUT respects that "--arg foo" should remain together, asgetopt_long()does now recognise it).arch_process_options()walks through them again and responds to any arch-specific options, again ignoring but respecting any non-arch options. Finally the loader may look for its options, and find them in-order and present. Any outside of this combined set are complained-about as usual.Quoted 3 lines
So, comments please. Is this a reasonable way to do it or is there an obvious better way I've missed? :-) So far I have been able to test on
x86(32,64)andppc(32,64)but none of the others. :(
Quoted 5 lines
This seems reasonable to me.
I've compiled tested the code on all architectures except cris (I don't have my build environment at the moment). I found a minor problem on arm which I have noted below.
That's superb, thanks very much for compiling them! And thanks also for catching the cut-n-paste-no-brain-involved ARM build problem below. I'll repost.
Cheers,
Matt
Quoted 44 lines
[snip]
Quoted 28 lines
kexec/arch/arm/include/arch/options.h +15
-0@@ -3,9 +3,38 @@#define OPT_ARCH_MAX (OPT_MAX+0) +#define OPT_APPEND 'a' +#define OPT_RAMDISK 'r' + +/* Options relevant to the architecture (excluding loader-specific ones), +* in this case none: +*/ #define KEXEC_ARCH_OPTIONS \ KEXEC_OPTIONS \#define KEXEC_ARCH_OPT_STR KEXEC_OPT_STR "" +/* The following two #defines list ALL of the options added by all of the +* architecture's loaders. +* o main() uses this complete list to scan for its options, ignoring +* arch-specific/loader-specific ones. +* o Then, arch_process_options() uses this complete list to scan for its +* options, ignoring general/loader-specific ones. +* o Then, the file_type[n].load re-scans for options, using +* KEXEC_ARCH_OPTIONS plus its loader-specific options subset. +* Any unrecognised options cause an error here. kexec/arch/arm/include/arch/options.h line 35
Quoted 10 lines
kexec/arch/arm/include/arch/options.h +10
-0+* +* This is done so that main()'s/arch_process_options()'s getopt_long() calls +* don't choose a kernel filename from random arguments to options they don't +* recognise -- as they now recognise (if not act upon) all possible options. +*/ +#define KEXEC_ALL_OPTIONS \ + KEXEC_ARCH_OPTIONS + { "command-line", 1, 0, OPT_APPEND }, + { "append", 1, 0, OPT_APPEND }, + { "initrd", 1, 0, OPT_RAMDISK },The above 4 lines seem to be missing a trailing '\'
Quoted 7 lines
Quoted 5 lines
kexec/arch/arm/include/arch/options.h +4
-0+ { "ramdisk", 1, 0, OPT_RAMDISK }, + +#define KEXEC_ALL_OPT_STR KEXEC_ARCH_OPT_STR "a:r:" +#endif /* KEXEC_ARCH_ARM_OPTIONS_H */ [snip]
_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec