su - user Hang!!!! Bug in reading history
Hi, All
I met a bug in bash when dealing with one case about bash history.
~~
Setps to reproduce the problem:
create a new user like jack.
setup the 4 factors to reproduce the problem, user with the following
condition would encount this problem.
1.HISTFILESIZE is larger than 0, for example 1000
2.HISTSIZE=0
3..bash_history is not a empty file
4.time stamp is enabled in .bash_history file.
Then we use root user to run su - jack, bash would hang.
~
I did some investigation and I think it should be a bug of bash.
1. if HISTSIZE=0, then the length of array to store history command in
memory would be zero.
The global variable the_history[history_length] is the array to store
history command, history_length would be set 0 if HISTSIZE=0,
2. if HISTFILESIZE is set to none zero value, for example 1000, bash
would truncate .bash_history size to 1000.
history_truncate_file (".bash_hisotry, 1000);
3. then read_history_range function would be call to put item read from
.bash_history file to array the_history[history_length].
add_history would have this action done,
void
add_history (string)
const char *string;
{
HIST_ENTRY *temp;
if (history_stifled && (history_length == history_max_entries))
{
register int i;
/* If the history is stifled, and history_length is zero,
and it equals history_max_entries, we don't save items. */
if (history_length == 0)
return; <--- we can see that if history_length=0,
add_history would return directly, would not add any item.
4. Then it would add try to add timestamp if they are enable in
.bash_history.
if (HIST_TIMESTAMP_START(line_start) == 0)
{
add_history (line_start);
if (last_ts)
{
add_history_time (last_ts);
last_ts = NULL;
}
}
5. Because add_history did not any thing if history_length = 0. and
add_history_time would meet a wrong array index exception at
hs = the_history[history_length - 1];
it try to read the_history[-1] which does not exist.
void
add_history_time (string)
const char *string;
{
HIST_ENTRY *hs;
hs = the_history[history_length - 1];
FREE (hs->timestamp);
hs->timestamp = savestring (string);
}
6. So the problem happens and shell stops there.
(gdb) where
#0 0x00482027 in add_history_time (string=0x18779b55
"#1357531487") at history.c:322
#1 0x00484d26 in read_history_range (filename=out>, from=0, to=4535) at histfile.c:272
#2 0x0044de3e in load_history () at bashhist.c:284
#3 0x0041b445 in main (argc=,
argv=0x7fff6eee2458, env=0x7fff6eee2468) at shell.c:710
(gdb) f
#0 0x00482027 in add_history_time (string=0x18779b55
"#1357531487") at history.c:322
322 hs = the_history[history_length - 1];
~~
Fix:
We should exit add_history_time if history_length=0.
/* Change the time stamp of the most recent history entry to STRING. */
void
add_history_time (string)
const char *string;
{
HIST_ENTRY *hs;
if ( history_length ==0 ) <---change needs to be
done here to avoid invalid array index if history_length=0.
return;
hs = the_history[history_length - 1];
FREE (hs->timestamp);
hs->timestamp = savestring (string);
}
Gengsheng Liu
RHCE
Redhat GSS Support
Re: su - user Hang!!!! Bug in reading history
On 7/8/13 1:50 AM, Geng Sheng Liu wrote: > Hi, All > > I met a bug in bash when dealing with one case about bash history. Thanks for the report. This will be fixed in the next bash/readline release. Chet -- ``The lyf so short, the craft so long to lerne.'' - Chaucer ``Ars longa, vita brevis'' - Hippocrates Chet Ramey, ITS, CWRUc...@case.eduhttp://cnswww.cns.cwru.edu/~chet/
Re: typeset -p & manpage on it are confusing w/rt funcs
On 6/6/13 6:48 PM, Linda Walsh wrote: > > I wanted to test to see if a function was defined and looking at > typeset in the bash man page, I see > typeset ... The -p option will > display the attributes and values of each name. When -p is used > with name arguments, additional options are ignored. When -p is > supplied without name arguments, it will display the attributes > and values of all variables having the attributes specified by > the additional options. If no other options are supplied with > -p, declare will display the attributes and values of all shell > variables. The -f option will restrict the display to shell > functions. The -F option inhibits the display of function defi- > nitions; > > > ok ... so reading the above, how does "-f" and -F" tie in with > "-p" ?? If I use -f with -p does that limit it to functions only? I think the original intent of the -p option was to have it interact with -f and -F to limit each name argument to the function namespace, and to display definitions and attributes. I will change the code and revise the documentation to reflect that. Chet -- ``The lyf so short, the craft so long to lerne.'' - Chaucer ``Ars longa, vita brevis'' - Hippocrates Chet Ramey, ITS, CWRUc...@case.eduhttp://cnswww.cns.cwru.edu/~chet/
Re: typeset -p & manpage on it are confusing w/rt funcs
Chet Ramey wrote: On 6/6/13 6:48 PM, Linda Walsh wrote: I wanted to test to see if a function was defined and looking at typeset in the bash man page, I see typeset ... The -p option will display the attributes and values of each name. When -p is used with name arguments, additional options are ignored. When -p is supplied without name arguments, it will display the attributes and values of all variables having the attributes specified by the additional options. If no other options are supplied with -p, declare will display the attributes and values of all shell variables. The -f option will restrict the display to shell functions. The -F option inhibits the display of function defi- nitions; ok ... so reading the above, how does "-f" and -F" tie in with "-p" ?? If I use -f with -p does that limit it to functions only? I think the original intent of the -p option was to have it interact with -f and -F to limit each name argument to the function namespace, and to display definitions and attributes. I will change the code and revise the documentation to reflect that. --- Yeah, right now, if you specify -f or -F with -p, it's as though -p is ignored, whereas typeset -p by itself seems to give the attributes of -F (non-funcs). I don't know how much it's worth, but it seems that w/o -f or -F, -p might display attributes&values of funcs+vars, with -f, funcs only, and -F as it acts now by itself (i.e. attribs of func+vars). Probably meant to do something orthogonal like that but maybe got interrupted in the middle of implementation and the idea fell off "the stack" so it never got completedor at least I can see that happening to me...;-)
Re: PS1 multiline with colors
Greg Wooledge wrote: normal=$(tput sgr0) red=$(tput setaf 1) green=$(tput setaf 2) ... --- BTW If you ever trace your code with "-x", tracing through the above will change your terminal text color. You can get around that by using read: read _CRST < <(tput sgr0) #Reset read _CRed < <(tput setaf 1) #Red read _CBLD < <(tput bold) #Bold
