Bug 651120

+++ This bug was initially created as a clone of Bug #587516 +++

Description of problem:
I was trying to set some user.DOSATTRIB values on read-only samba share with setfattr. Samba seems to require the values for that attribute to be hexadecimal, starting with C style 0x prefix. Unfortunately such a value is rejected by setfattr:

gklab-63-001:/var/tmp> setfattr -n user.DOSATTRIB -v '0x2' aqq
bad input encoding
gklab-63-001:/var/tmp> setfattr -n DOSATTRIB -v '0x2' aqq
bad input encoding
gklab-63-001:/var/tmp> getfattr -d aqq
gklab-63-001:/var/tmp> attr -s DOSATTRIB -V 0x2 aqq
Atrybut "DOSATTRIB" dla aqq ustawiono na wartość 3-bajtową:
0x2
gklab-63-001:/var/tmp> getfattr -d aqq
# file: aqq
user.DOSATTRIB="0x2"

Curiously (as seen above) attr -s seems to have no problem with such value.

Version-Release number of selected component (if applicable):
attr-2.4.44-1.fc12.x86_64

How reproducible:
Always

Actual results:
Cannot set the value 0x2 for the attribute.

Expected results:
Can set the value 0x2 for the attribute.


Additional info:
I've also seen this bug on CentOS 5.4 so you should consider cloning the bug there as well.

--- Additional comment from kdudka on 2010-04-30 09:22:39 CEST ---

Thank you for filling the bug!  I'll check if it is a real bug or feature.

FWIW, here is a trace of the failure:

$ gdb -q --args setfattr -n user.DOSATTRIB -v 0x2 /tmp/x
(gdb) break decode
Breakpoint 1 at 0x401839: file setfattr.c, line 291.
(gdb) run

Breakpoint 1, decode (value=0x7fffffffe6cb "0x2", size=0x7fffffffe250) at setfattr.c:291
291             if (value[0] == '0' && (value[1] == 'x' || value[1] == 'X')) {
(gdb) next
292                     const char *v = value+2, *end = value + *size;
(gdb)
296                                          *size / 2)) {
(gdb)
295                     if (high_water_alloc((void **)&decoded, &decoded_size,
(gdb)
302                     d = decoded;
(gdb)
303                     while (v < end) {
(gdb)
306                             while (v < end && isspace(*v))
(gdb)
308                             if (v == end)
(gdb)
310                             d1 = hex_digit(*v++);
(gdb)
311                             while (v < end && isspace(*v))
(gdb)
313                             if (v == end) {
(gdb)
315                                     fprintf(stderr, "bad input encoding\n");
(gdb) info locals
d1 = 2
d0 = 32767
v = 0x7fffffffe6ce ""
end = 0x7fffffffe6ce ""
d = 0x605000 ""
decoded = 0x605000 ""
decoded_size = 256

--- Additional comment from kdudka on 2010-05-11 13:13:55 CEST ---

That's not a bug.  setfattr(1) takes also values in either hexadecimal or base64 format, given the 0[xX] respectively 0[sS] prefix.  You should be able to pass 0x2 as string by surrounding it to quotes.  Please confirm.

I agree there is a dearth of documentation for this feature.  If you had some suggestion what to add to man page or whatever, it would be also appreciated.

--- Additional comment from tomek on 2010-05-11 19:57:58 CEST ---

I would say: it works weirdly:

vesemir:/var/tmp> setfattr -n user.DOSATTRIB -v '0x2' rpm-tmp.P2sHiq
bad input encoding
vesemir:/var/tmp> setfattr -n user.DOSATTRIB -v "0x2" rpm-tmp.P2sHiq
bad input encoding
vesemir:/var/tmp> setfattr -n user.DOSATTRIB -v '"0x2"' rpm-tmp.P2sHiq
vesemir:/var/tmp> getfa
getfacl  getfattr 
vesemir:/var/tmp> getfattr -d rpm-tmp.P2sHiq
# file: rpm-tmp.P2sHiq
user.DOSATTRIB="0x2"

So, to make it work - you need single quotes in doubles quotes, the other way around or plenty of backslashes....

I love hearing: it is not a bug, it is a feature ;).

--- Additional comment from kdudka on 2010-05-11 20:20:11 CEST ---

Sure, that's a feature of the _shell_ you're using ;-)  It drops the outer single or double quotes.  You need to either double quote, or escape the quotes to disable their special meaning:

$ setfattr -n user.DOSATTRIB -v \"0x2\"

You will observe basically the same with e.g. echo(1):

$ echo "0x2"
0x2

$ echo \"0x2\"
"0x2"

--- Additional comment from tomek on 2010-05-11 22:28:00 CEST ---

That doesn't explain why I have to put quotes around 0x2 which is hexadecimal number and not when passing straight string like zzz:

vesemir:/var/tmp> setfattr -n user.DOSATTRIB -v '"0x2"' kdecache-tomek
vesemir:/var/tmp> getfattr -d kdecache-tomek
# file: kdecache-tomek
user.DOSATTRIB="0x2"

vesemir:/var/tmp> setfattr -n user.DOSATTRIB -v zzz kdecache-tomek
vesemir:/var/tmp> getfattr -d kdecache-tomek
# file: kdecache-tomek
user.DOSATTRIB="zzz"

My understanding is that all attributes are just strings and I consider the requirement to quote hexadecimal number in this case unreasonable.

But that's another story, I think you can close the bug (unless you agree with me and want to fix it ;) ).

--- Additional comment from kdudka on 2010-05-11 23:17:30 CEST ---

Well, my mistake.  I should have explained this properly.

The mentioned feature is obstruction from your point of view since you pass only text values through.  Nevertheless there should be a way to pass some binary values as well.  That's why the hexadecimal and base64 format has been introduced, at least from my understanding of that.

Each enhancement has its cost.  In this case, it means that all strings starting with '0x', '0X', '0s', or '0S' are treated specially.  To get the old behavior back, just surround your value by double quotes.  It will work well for "0x2", as well as "zzz".

If you have any idea how to implement the feature in a better way (without breaking the stuff), patches are welcome.  I left the bug open because I hadn't found the feature mentioned in documentation.  If the information is indeed missing, we should perhaps a paragraph there about the feature.

FWIW, I've CC'd the upstream maintainer of attr.

--- Additional comment from kdudka on 2010-11-08 20:25:06 CET ---

Created attachment 458858 [details]
proposed improvement of setfattr.1 man page