daxctl: Replace basename() usage with strrchr()
Alison Schofield
maintainerPATCH 1/1 daxctl: Replace basename() usage with strrchr()
A user reports that ndctl fails to compile on MUSL systems:
daxctl/device.c: In function 'parse_device_options':daxctl/device.c:377:26: error: implicit declaration of function 'basename' [-Wimplicit-function-declaration]
377 | device = basename(argv[0]);
| ^~~~~~~~There are two versions of basename() with different behaviors:
GNU basename() from <string.h>: doesn't modify its argument POSIX basename() from <libgen.h>: may modify its argument
glibc provides both versions, while MUSL libc only provides the POSIX version.
In daxctl/device.c, basename() is called without any header, relying on the GNU extension being implicitly available. And in daxctl/lib/ libdaxctl.c, libgen.h is included and the POSIX basename() used, which works but is needlessly complex as POSIX basename() can modify its argument.
Rather than conditionally including headers or dealing with platform differences, replace both basename() usages with a new implementation using strrchar to find the last '/' in the path and return everything after it, or the whole string if no '/' is found.
--- daxctl/device.c | 8 +++++--- daxctl/lib/libdaxctl.c | 4 ++-- 2 files changed, 7 insertions(+), 5 deletions(-)
daxctl/device.c +5 -3
@@ -362,11 +362,13 @@ static const char *parse_device_options(int argc, const char **argv, };unsigned long long units =1 ;int i, rc =0 ; -char *device =NULL ; +const char *device =NULL ; argc =parse_options (argc, argv, options, u,0 ); -if (argc >0 ) - device =basename (argv[0 ]); +if (argc >0 ) { + device =strrchr (argv[0 ],'/' ); + device = device ? device +1 : argv[0 ]; + }/* Handle action-agnostic non-option arguments */ if (argc == 0 &&
daxctl/lib/libdaxctl.c +2 -2
@@ -3,7 +3,6 @@#include <stdio.h> #include <errno.h> #include <limits.h> -#include <libgen.h> #include <stdlib.h> #include <dirent.h> #include <unistd.h>
@@ -408,7 +407,8 @@ DAXCTL_EXPORT int daxctl_dev_is_system_ram_capable(struct daxctl_dev *dev)if (!mod_path)return false; - mod_base =basename (mod_path); + mod_base =strrchr (mod_path,'/' ); + mod_base = mod_base ? mod_base +1 : mod_path;if (strcmp (mod_base, dax_modules[DAXCTL_DEV_MODE_RAM ]) ==0 ) {free (mod_path);return true;
Signature · 1 line
2.37.3
Hi Alison,
daxctl/device.c static const char *parse_device_options(int argc, const char **argv, · line 371
Quoted 8 lines · Alison Schofield
daxctl/device.c +4
-2argc =parse_options (argc, argv, options, u,0 ); -if (argc >0 ) - device =basename (argv[0 ]); +if (argc >0 ) { + device =strrchr (argv[0 ],'/' ); + device = device ? device +1 : argv[0 ]; + }
1. I would add a one-line comment in both places, something like "This is like basename but without the bugs and portability issues" because:
1.a) It's much faster to read such a comment than understanding the code. 1.b) Not everyone knows how much of GNU/POSIX disaster is "basename".
You summarized it well in the commit message but it's unlikely anyone will fetch the commit message from git without such a comment.
To avoid duplicating the comment, a small "my_basename()" inline function would not hurt while at it.
2. I believe this (unlike basename) returns an empty string when the argument has a trailing slash. Now, an argument with a trailing slash would probably be garbage and I'm OK with the "Garbage IN, garbage OUT" principle. BUT I also believe in the "Proportionate Response" principle, which means a small amount of garbage IN should IMHO not be punished by some utterly baffling error message or (much worse) a crash. Did you/could you test what happens with a trailing slash? If the resulting failure makes some kind of sense then don't change anything.
Note even when rare in interactive use, "Garbage IN" becomes more frequent in automation. Then good luck making sense of a cryptic error when you can't even reproduce the elaborate test configuration and test bugs that trigger it.
Alison Schofield
maintainerQuoted 16 lines · Marc Herbert
Hi Alison,
daxctl/device.c static const char *parse_device_options(int argc, const char **argv, · line 371
Quoted 8 lines · Alison Schofield
daxctl/device.c +4
-2argc =parse_options (argc, argv, options, u,0 ); -if (argc >0 ) - device =basename (argv[0 ]); +if (argc >0 ) { + device =strrchr (argv[0 ],'/' ); + device = device ? device +1 : argv[0 ]; + }1. I would add a one-line comment in both places, something like "This is like basename but without the bugs and portability issues" because:
Quoted 7 lines
1.a) It's much faster to read such a comment than understanding the code. 1.b) Not everyone knows how much of GNU/POSIX disaster is "basename".
You summarized it well in the commit message but it's unlikely anyone will fetch the commit message from git without such a comment.To avoid duplicating the comment, a small "
my_basename()" inline function would not hurt while at it.
Thanks for the review.
I'm headed down the Dan suggested path of adding a helper.
Quoted 11 lines
2. I believe this (unlike basename) returns an empty string when the argument has a trailing slash. Now, an argument with a trailing slash would probably be garbage and I'm OK with the "Garbage IN, garbage OUT" principle. BUT I also believe in the "Proportionate Response" principle, which means a small amount of garbage IN should IMHO not be punished by some utterly baffling error message or (much worse) a crash. Did you/could you test what happens with a trailing slash? If the resulting failure makes some kind of sense then don't change anything.
With the implementation moved to a new helper, path_basename() I've tested these:
assert(strcmp(path_basename("/usr/bin/foo"), "foo") == 0); assert(strcmp(path_basename("foo"), "foo") == 0); assert(strcmp(path_basename("/usr/bin/foo/"), "") == 0); assert(strcmp(path_basename("/"), "") == 0); assert(strcmp(path_basename(""), "") == 0);
I think this sticks w gigo principle and 'proportionate response' too. It's safe for valid paths, safe for unusual but harmless paths, no crashes or undefined behavior, and the thing that inspired this, is that it behave the same across libc implementations.
Watch for the v2 and let me know if i missed any string tests above.
Quoted 6 lines
Note even when rare in interactive use, "Garbage IN" becomes more frequent in automation. Then good luck making sense of a cryptic error when you can't even reproduce the elaborate test configuration and test bugs that trigger it.
Alison Schofield
maintainerQuoted 22 lines · Alison Schofield
Quoted 21 lines · Marc Herbert
Hi Alison,
daxctl/device.c static const char *parse_device_options(int argc, const char **argv, · line 371
Quoted 8 lines · Alison Schofield
daxctl/device.c +4
-2argc =parse_options (argc, argv, options, u,0 ); -if (argc >0 ) - device =basename (argv[0 ]); +if (argc >0 ) { + device =strrchr (argv[0 ],'/' ); + device = device ? device +1 : argv[0 ]; + }1. I would add a one-line comment in both places, something like "This is like basename but without the bugs and portability issues" because:
1.a) It's much faster to read such a comment than understanding the code. 1.b) Not everyone knows how much of GNU/POSIX disaster is "basename".
You summarized it well in the commit message but it's unlikely anyone will fetch the commit message from git without such a comment.
Quoted 6 lines
Quoted 2 lines
To avoid duplicating the comment, a small "
my_basename()" inline function would not hurt while at it.Thanks for the review.
I'm headed down the Dan suggested path of adding a helper.
meant to say - Marc and Dan suggested path :)
Alison Schofield wrote:
[..]
Quoted 4 lines
Rather than conditionally including headers or dealing with platform differences, replace both
basename()usages with a new implementation using strrchar to find the last '/' in the path and return everything after it, or the whole string if no '/' is found.
This feels like it wants a helper rather than a new open-coded thing. For example, devpath_to_devname() is an existing similar helper.
Alternatively, just create a local helper called basename when the C-library is missing the GNU version.
Alison Schofield
maintainerQuoted 9 lines · Dan Williams
Alison Schofield wrote:
[..]Quoted 4 lines
Rather than conditionally including headers or dealing with platform differences, replace both
basename()usages with a new implementation using strrchar to find the last '/' in the path and return everything after it, or the whole string if no '/' is found.This feels like it wants a helper rather than a new open-coded thing. For example,
devpath_to_devname()is an existing similar helper.
happy face
Quoted 3 lines
Alternatively, just create a local helper called basename when the C-library is missing the GNU version.
grimace
Prepping the happy face version next.