| Message ID | dda23def071a8d087cca3538289de1e0@bspu.by (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | modinfo must show real module info, not context if filename set | expand |
On Wed, Feb 26, 2020 at 04:53:45AM +0200, Denis Kaganovich wrote: > After commit e7e2cb61fa9f1db3429d91ef6accff549500d268, even if real filename > passed - modinfo show info from context (so, I got built-in info from > running > kernel, but asking for new kernel's external module). This behaviour > unobvious > and incompatible with pre-v27. Simple use fake context for filename - IMHO > much less ugly then current results. Can you give an example of this misbehavior ? I have a two kernel directories (generic and current) with modules.builtin.modinfo: $ ls -1 /lib/modules/{5.2.0-generic,`uname -r`}/modules.builtin.modinfo /lib/modules/5.2.0-generic/modules.builtin.modinfo /lib/modules/5.2.0-current/modules.builtin.modinfo The ext4 module is built into both: $ tr '\0' '\n' < /lib/modules/`uname -r`/modules.builtin.modinfo |grep ^ext4.description ext4.description=Fourth Extended Filesystem $ tr '\0' '\n' < /lib/modules/5.2.0-generic/modules.builtin.modinfo |grep ^ext4.description ext4.description=Fourth Extended Filesystem Now I have build this module separately and put it into the tree: $ find /lib/modules/{5.2.0-generic,`uname -r`} -name 'ext4.*' /lib/modules/5.2.0-generic/kernel/fs/ext4/ext4.ko.gz For current kernel: $ tools/modinfo ext4 |grep filename filename: (builtin) For the other kernel: $ tools/modinfo -k 5.2.0-generic ext4 |grep filename filename: /lib/modules/5.2.0-generic/kernel/fs/ext4/ext4.ko.gz Only the external module is displayed, even if there is a module with the same name in the modules.builtin.modinfo. I also have a very old kernel without modules.builtin.modinfo: $ tools/modinfo -k 4.14.96-generic ext4 |grep filename filename: /lib/modules/4.14.96-generic/kernel/fs/ext4/ext4.ko.gz > Signed-off-by: Dzianis Kahanovich <mahatma@eu.by> > > --- a/tools/modinfo.c 2020-02-25 13:46:38.181693570 +0300 > +++ b/tools/modinfo.c 2020-02-26 05:18:39.393790919 +0300 > @@ -359,7 +359,7 @@ static bool is_module_filename(const cha > > static int do_modinfo(int argc, char *argv[]) > { > - struct kmod_ctx *ctx; > + struct kmod_ctx *ctx, *ctx0; > char dirname_buf[PATH_MAX]; > const char *dirname = NULL; > const char *kversion = NULL; > @@ -437,7 +437,8 @@ static int do_modinfo(int argc, char *ar > } > > ctx = kmod_new(dirname, &null_config); > - if (!ctx) { > + ctx0 = kmod_new("/dev/null", &null_config); > + if (!ctx || !ctx0) { > ERR("kmod_new() failed!\n"); > return EXIT_FAILURE; > } > @@ -448,7 +449,7 @@ static int do_modinfo(int argc, char *ar > int r; > > if (is_module_filename(name)) > - r = modinfo_path_do(ctx, name); > + r = modinfo_path_do(ctx0, name); > else > r = modinfo_alias_do(ctx, name); > > @@ -456,6 +457,7 @@ static int do_modinfo(int argc, char *ar > err = r; > } > > + kmod_unref(ctx0); > kmod_unref(ctx); > return err >= 0 ? EXIT_SUCCESS : EXIT_FAILURE; > } >
On Wed, Feb 26, 2020 at 10:33 AM Alexey Gladkov <gladkov.alexey@gmail.com> wrote: > > On Wed, Feb 26, 2020 at 04:53:45AM +0200, Denis Kaganovich wrote: > > After commit e7e2cb61fa9f1db3429d91ef6accff549500d268, even if real filename > > passed - modinfo show info from context (so, I got built-in info from > > running > > kernel, but asking for new kernel's external module). This behaviour > > unobvious > > and incompatible with pre-v27. Simple use fake context for filename - IMHO > > much less ugly then current results. > > Can you give an example of this misbehavior ? > > I have a two kernel directories (generic and current) with > modules.builtin.modinfo: > > $ ls -1 /lib/modules/{5.2.0-generic,`uname -r`}/modules.builtin.modinfo > /lib/modules/5.2.0-generic/modules.builtin.modinfo > /lib/modules/5.2.0-current/modules.builtin.modinfo > > The ext4 module is built into both: > > $ tr '\0' '\n' < /lib/modules/`uname -r`/modules.builtin.modinfo |grep ^ext4.description > ext4.description=Fourth Extended Filesystem > > $ tr '\0' '\n' < /lib/modules/5.2.0-generic/modules.builtin.modinfo |grep ^ext4.description > ext4.description=Fourth Extended Filesystem > > Now I have build this module separately and put it into the tree: that's why you don't have this problem. If you just build a new ext4.ko module and give it in the command line without putting it into the tree, then this problem would be triggered. $ ~/p/kmod/tools/modinfo fs/ext4/ext4.ko filename: /home/ldmartin/p/gfx-internal/drm-intel-internal/fs/ext4/ext4.ko modinfo: ERROR: could not get modinfo from 'ext4': No such file or directory $ modinfo fs/ext4/ext4.ko filename: /home/ldmartin/p/gfx-internal/drm-intel-internal/fs/ext4/ext4.ko softdep: pre: crc32c license: GPL description: Fourth Extended Filesystem author: Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others alias: fs-ext4 alias: ext3 alias: fs-ext3 alias: ext2 alias: fs-ext2 depends: mbcache,jbd2 retpoline: Y intree: Y name: ext4 vermagic: 5.4.17+ SMP preempt mod_unload Lucas De Marchi > > $ find /lib/modules/{5.2.0-generic,`uname -r`} -name 'ext4.*' > /lib/modules/5.2.0-generic/kernel/fs/ext4/ext4.ko.gz > > For current kernel: > > $ tools/modinfo ext4 |grep filename > filename: (builtin) > > For the other kernel: > > $ tools/modinfo -k 5.2.0-generic ext4 |grep filename > filename: /lib/modules/5.2.0-generic/kernel/fs/ext4/ext4.ko.gz > > Only the external module is displayed, even if there is a module with the > same name in the modules.builtin.modinfo. > > I also have a very old kernel without modules.builtin.modinfo: > > $ tools/modinfo -k 4.14.96-generic ext4 |grep filename > filename: /lib/modules/4.14.96-generic/kernel/fs/ext4/ext4.ko.gz > > > Signed-off-by: Dzianis Kahanovich <mahatma@eu.by> > > > > --- a/tools/modinfo.c 2020-02-25 13:46:38.181693570 +0300 > > +++ b/tools/modinfo.c 2020-02-26 05:18:39.393790919 +0300 > > @@ -359,7 +359,7 @@ static bool is_module_filename(const cha > > > > static int do_modinfo(int argc, char *argv[]) > > { > > - struct kmod_ctx *ctx; > > + struct kmod_ctx *ctx, *ctx0; > > char dirname_buf[PATH_MAX]; > > const char *dirname = NULL; > > const char *kversion = NULL; > > @@ -437,7 +437,8 @@ static int do_modinfo(int argc, char *ar > > } > > > > ctx = kmod_new(dirname, &null_config); > > - if (!ctx) { > > + ctx0 = kmod_new("/dev/null", &null_config); > > + if (!ctx || !ctx0) { > > ERR("kmod_new() failed!\n"); > > return EXIT_FAILURE; > > } > > @@ -448,7 +449,7 @@ static int do_modinfo(int argc, char *ar > > int r; > > > > if (is_module_filename(name)) > > - r = modinfo_path_do(ctx, name); > > + r = modinfo_path_do(ctx0, name); > > else > > r = modinfo_alias_do(ctx, name); > > > > @@ -456,6 +457,7 @@ static int do_modinfo(int argc, char *ar > > err = r; > > } > > > > + kmod_unref(ctx0); > > kmod_unref(ctx); > > return err >= 0 ? EXIT_SUCCESS : EXIT_FAILURE; > > } > > > > -- > Rgrds, legion >
On Wed, Feb 26, 2020 at 11:36:00AM -0800, Lucas De Marchi wrote: > On Wed, Feb 26, 2020 at 10:33 AM Alexey Gladkov > <gladkov.alexey@gmail.com> wrote: > > > > On Wed, Feb 26, 2020 at 04:53:45AM +0200, Denis Kaganovich wrote: > > > After commit e7e2cb61fa9f1db3429d91ef6accff549500d268, even if real filename > > > passed - modinfo show info from context (so, I got built-in info from > > > running > > > kernel, but asking for new kernel's external module). This behaviour > > > unobvious > > > and incompatible with pre-v27. Simple use fake context for filename - IMHO > > > much less ugly then current results. > > > > Can you give an example of this misbehavior ? > > > > I have a two kernel directories (generic and current) with > > modules.builtin.modinfo: > > > > $ ls -1 /lib/modules/{5.2.0-generic,`uname -r`}/modules.builtin.modinfo > > /lib/modules/5.2.0-generic/modules.builtin.modinfo > > /lib/modules/5.2.0-current/modules.builtin.modinfo > > > > The ext4 module is built into both: > > > > $ tr '\0' '\n' < /lib/modules/`uname -r`/modules.builtin.modinfo |grep ^ext4.description > > ext4.description=Fourth Extended Filesystem > > > > $ tr '\0' '\n' < /lib/modules/5.2.0-generic/modules.builtin.modinfo |grep ^ext4.description > > ext4.description=Fourth Extended Filesystem > > > > Now I have build this module separately and put it into the tree: > > that's why you don't have this problem. If you just build a new > ext4.ko module and give it > in the command line without putting it into the tree, then this > problem would be triggered. Ouch, I see. The phrase "new kernel's external module" confused me because usually the modules for the new kernel are in the tree. My bad. > $ ~/p/kmod/tools/modinfo fs/ext4/ext4.ko > filename: /home/ldmartin/p/gfx-internal/drm-intel-internal/fs/ext4/ext4.ko > modinfo: ERROR: could not get modinfo from 'ext4': No such file or directory > > $ modinfo fs/ext4/ext4.ko > filename: /home/ldmartin/p/gfx-internal/drm-intel-internal/fs/ext4/ext4.ko > softdep: pre: crc32c > license: GPL > description: Fourth Extended Filesystem > author: Remy Card, Stephen Tweedie, Andrew Morton, Andreas > Dilger, Theodore Ts'o and others > alias: fs-ext4 > alias: ext3 > alias: fs-ext3 > alias: ext2 > alias: fs-ext2 > depends: mbcache,jbd2 > retpoline: Y > intree: Y > name: ext4 > vermagic: 5.4.17+ SMP preempt mod_unload > > Lucas De Marchi
--- a/tools/modinfo.c 2020-02-25 13:46:38.181693570 +0300 +++ b/tools/modinfo.c 2020-02-26 05:18:39.393790919 +0300 @@ -359,7 +359,7 @@ static bool is_module_filename(const cha static int do_modinfo(int argc, char *argv[]) { - struct kmod_ctx *ctx; + struct kmod_ctx *ctx, *ctx0; char dirname_buf[PATH_MAX]; const char *dirname = NULL; const char *kversion = NULL; @@ -437,7 +437,8 @@ static int do_modinfo(int argc, char *ar } ctx = kmod_new(dirname, &null_config); - if (!ctx) { + ctx0 = kmod_new("/dev/null", &null_config); + if (!ctx || !ctx0) { ERR("kmod_new() failed!\n"); return EXIT_FAILURE; } @@ -448,7 +449,7 @@ static int do_modinfo(int argc, char *ar int r; if (is_module_filename(name)) - r = modinfo_path_do(ctx, name); + r = modinfo_path_do(ctx0, name); else r = modinfo_alias_do(ctx, name); @@ -456,6 +457,7 @@ static int do_modinfo(int argc, char *ar err = r; } + kmod_unref(ctx0); kmod_unref(ctx); return err >= 0 ? EXIT_SUCCESS : EXIT_FAILURE; }
After commit e7e2cb61fa9f1db3429d91ef6accff549500d268, even if real filename passed - modinfo show info from context (so, I got built-in info from running kernel, but asking for new kernel's external module). This behaviour unobvious and incompatible with pre-v27. Simple use fake context for filename - IMHO much less ugly then current results. Signed-off-by: Dzianis Kahanovich <mahatma@eu.by>