multipath config fixes
Benjamin Marzinski
maintainerv2 cover multipath config fixes
The first two patches are a cleanup and a fix for a memory leak in the config code. The third patch improves multipath's validation of the strings it passes directly into the table, features, path_selector, and hardware_handler. These three strings all have argument counts, and getting them wrong causes the kernel to parse the table incorrectly. When this happens the table load fails, but the error messages from the kernel are often completely unhelpful. A bad argument count will cause the rest of the table to be parsed incorrectly, and the kernel might not hit an unworkable token till later in the parsing. Multipath now makes sure that the count matches the actual number of arguments that it is passing.
V2 Changes (based on suggestions from Martin Wilck):
1/3: unrolled loop in validate_config_strvec() to explicitly check the
possible quote strings
3/3: spaces is now a "const char * const" and doesn't include '\n'
Benjamin Marzinski (3):
libmpathutil: simplify set_value and validate_config_strvec
libmultipath: don't leak memory on invalid strings
libmutipath: validate the argument count of config strings
libmpathutil/parser.c | 109 +++++++++++++++++++--------------------- libmultipath/dict.c | 112 ++++++++++++++++++++++++++++++++++++++---- 2 files changed, 154 insertions(+), 67 deletions(-)
Signature · 6 lines
2.17.2 -- dm-devel mailing list [email protected] https://listman.redhat.com/mailman/listinfo/dm-devel
Benjamin Marzinski
maintainerPATCH v2 1/3 libmpathutil: simplify set_value and validate_config_strvec
alloc_strvec() will never create a strvec with multiple tokens between the quote tokens. Verify this in validate_config_strvec(), and simplify set_value() by only reading one value after a quote token.
--- libmpathutil/parser.c | 109 ++++++++++++++++++++---------------------- 1 file changed, 51 insertions(+), 58 deletions(-)
libmpathutil/parser.c +51 -58
@@ -333,59 +333,33 @@ void * set_value(vector strvec) {char *str =VECTOR_SLOT (strvec,1 ); -size_t size; -int i =0 ; -int len =0 ;char *alloc =NULL ; -char *tmp;if (!str) {condlog (0 ,"option '%s' missing value" , (char *)VECTOR_SLOT (strvec,0 ));return NULL ; } -if (!is_quote (str)) { - size =strlen (str); -if (size ==0 ) { -condlog (0 ,"option '%s' has empty value" , - (char *)VECTOR_SLOT (strvec,0 )); -return NULL ; - } - alloc =calloc (1 ,sizeof (char ) * (size +1 )); -if (alloc) -memcpy (alloc, str, size); -else - goto oom; -return alloc; - } -/* Even empty quotes counts as a value (An empty string) */ - alloc = (char *)calloc (1 ,sizeof (char )); -if (!alloc) - goto oom; -for (i =2 ; i <VECTOR_SIZE (strvec); i++) { - str =VECTOR_SLOT (strvec, i); -if (!str) { -free (alloc); -condlog (0 ,"parse error for option '%s'" , - (char *)VECTOR_SLOT (strvec,0 )); -return NULL ; +if (is_quote (str)) { +if (VECTOR_SIZE (strvec) >2 ) { + str =VECTOR_SLOT (strvec,2 ); +if (!str) { +condlog (0 ,"parse error for option '%s'" , + (char *)VECTOR_SLOT (strvec,0 )); +return NULL ; + } } -if (is_quote (str)) -break ; - tmp = alloc; -/* The first +1 is for the NULL byte. The rest are for the -* spaces between words */ - len +=strlen (str) +1 ; - alloc =realloc (alloc,sizeof (char ) * len); -if (!alloc) { -free (tmp); - goto oom; +/* Even empty quotes counts as a value (An empty string) */ +if (is_quote (str)) { + alloc = (char *)calloc (1 ,sizeof (char )); +if (!alloc) + goto oom; +return alloc; } -if (*alloc !='\0' ) -strncat (alloc," " , len -strlen (alloc)); -strncat (alloc, str, len -strlen (alloc) -1 ); } -return alloc; + alloc =strdup (str); +if (alloc) +return alloc; oom:condlog (0 ,"can't allocate memory for option '%s'" , (char *)VECTOR_SLOT (strvec,0 ));
@@ -442,7 +416,6 @@ int validate_config_strvec(vector strvec, const char *file) { char *str = NULL; -int i;if (strvec &&VECTOR_SIZE (strvec) >0 ) str =VECTOR_SLOT (strvec,0 );
@@ -485,21 +458,41 @@ validate_config_strvec(vector strvec, const char *file)condlog (0 ,"ignoring extra data starting with '%s' on line %d of %s" , (char *)VECTOR_SLOT (strvec,2 ), line_nr, file);return 0 ; } -for (i =2 ; i <VECTOR_SIZE (strvec); i++) { - str =VECTOR_SLOT (strvec, i); -if (str ==NULL ) { -condlog (0 ,"can't parse value on line %d of %s" , - line_nr, file); -return -1 ; - } -if (is_quote (str)) { -if (VECTOR_SIZE (strvec) > i +1 ) -condlog (0 ,"ignoring extra data starting with '%s' on line %d of %s" , (char *)VECTOR_SLOT (strvec, (i +1 )), line_nr, file); -return 0 ; - } +if (VECTOR_SIZE (strvec) ==2 ) { +condlog (0 ,"missing closing quotes on line %d of %s" , + line_nr, file); +return 0 ; + } + str =VECTOR_SLOT (strvec,2 ); +if (str ==NULL ) { +condlog (0 ,"can't parse value on line %d of %s" , + line_nr, file); +return -1 ; + } +if (is_quote (str)) { +if (VECTOR_SIZE (strvec) >3 ) +condlog (0 ,"ignoring extra data starting with '%s' on line %d of %s" , (char *)VECTOR_SLOT (strvec,3 ), line_nr, file); +return 0 ; + } +if (VECTOR_SIZE (strvec) ==3 ) { +condlog (0 ,"missing closing quotes on line %d of %s" , + line_nr, file); +return 0 ; + } + str =VECTOR_SLOT (strvec,3 ); +if (str ==NULL ) { +condlog (0 ,"can't parse value on line %d of %s" , + line_nr, file); +return -1 ; + } +if (!is_quote (str)) { +/* There should only ever be one token between quotes */ +condlog (0 ,"parsing error starting with '%s' on line %d of %s" , + str, line_nr, file); +return -1 ; } -condlog (0 ,"missing closing quotes on line %d of %s" , - line_nr, file); +if (VECTOR_SIZE (strvec) >4 ) +condlog (0 ,"ignoring extra data starting with '%s' on line %d of %s" , (char *)VECTOR_SLOT (strvec,4 ), line_nr, file);return 0 ; }
Signature · 6 lines
2.17.2 -- dm-devel mailing list [email protected] https://listman.redhat.com/mailman/listinfo/dm-devel
Benjamin Marzinski
maintainerPATCH v2 2/3 libmultipath: don't leak memory on invalid strings
In reply to Benjamin MarzinskiIf set_path() or set_str_noslash() are called with a bad value, they ignore it and continue to use the old value. But they weren't freeing the bad value, causing a memory leak.
--- libmultipath/dict.c | 2 ++ 1 file changed, 2 insertions(+)
libmultipath/dict.c +2 -0
@@ -130,6 +130,7 @@ set_path(vector strvec, void *ptr, const char *file, int line_nr)if ((*str_ptr)[0 ] !='/' ){condlog (1 ,"%s line %d, %s is not an absolute path. Ignoring" , file, line_nr, *str_ptr); +free (*str_ptr); *str_ptr = old_str; }else free (old_str);
@@ -150,6 +151,7 @@ set_str_noslash(vector strvec, void *ptr, const char *file, int line_nr)if (strchr (*str_ptr,'/' )) {condlog (1 ,"%s line %d, %s cannot contain a slash. Ignoring" , file, line_nr, *str_ptr); +free (*str_ptr); *str_ptr = old_str; }else free (old_str);
Signature · 6 lines
2.17.2 -- dm-devel mailing list [email protected] https://listman.redhat.com/mailman/listinfo/dm-devel
Benjamin Marzinski
maintainerPATCH v2 3/3 libmutipath: validate the argument count of config strings
In reply to Benjamin MarzinskiThe features, path_selector, and hardware_handler config options pass their strings directly into the kernel. If users omit the argument counts from these strings, or use the wrong value, the kernel's table parsing gets completely messed up, and the error messages it prints don't reflect what actully went wrong. To avoid messing up the kernel table parsing, verify that these strings correctly set the argument count to the number of arguments they have.
--- libmultipath/dict.c | 110 ++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 101 insertions(+), 9 deletions(-)
libmultipath/dict.c +101 -9
@@ -116,6 +116,58 @@ set_str(vector strvec, void *ptr, const char *file, int line_nr)return 0 ; } +static int +set_arg_str (vector strvec,void *ptr,int count_idx,const char *file, +int line_nr) +{ +char **str_ptr = (char **)ptr; +char *old_str = *str_ptr; +const char *const spaces =" \f\r\t\v" ; +char *p, *end; +int idx =-1 ; +long int count =-1 ; + + *str_ptr =set_value (strvec); +if (!*str_ptr) { +free (old_str); +return 1 ; + } + p = *str_ptr; +while (*p !='\0' ) { + p +=strspn (p, spaces); +if (*p =='\0' ) +break ; + idx +=1 ; +if (idx == count_idx) { + errno =0 ; + count =strtol (p, &end,10 ); +if (errno ==ERANGE || end == p || + !(isspace (*end) || *end =='\0' )) { + count =-1 ; +break ; + } + } + p +=strcspn (p, spaces); + } +if (count <0 ) { +condlog (1 ,"%s line %d, missing argument count for %s" , + file, line_nr, (char *)VECTOR_SLOT (strvec,0 )); + goto fail; + } +if (count != idx - count_idx) { +condlog (1 ,"%s line %d, invalid argument count for %s:, got '%ld' expected '%d'" , + file, line_nr, (char *)VECTOR_SLOT (strvec,0 ), count, + idx - count_idx); + goto fail; + } +free (old_str); +return 0 ; +fail: +free (*str_ptr); + *str_ptr = old_str; +return 0 ; +} +static int set_path (vector strvec,void *ptr,const char *file,int line_nr) {
@@ -288,6 +340,14 @@ def_ ## option ## _handler (struct config *conf, vector strvec, \return set_int (strvec, &conf->option, minval, maxval, file, line_nr); \ } +#define declare_def_arg_str_handler (option, count_idx) \ +static int \ +def_ ## option ## _handler (struct config *conf, vector strvec, \ + const char *file, int line_nr) \ +{ \ + return set_arg_str(strvec, &conf->option, count_idx, file, line_nr); \ +} +#define declare_def_snprint (option, function) \ static int \ snprint_def_ ## option (struct config *conf, struct strbuf *buff, \
@@ -340,6 +400,17 @@ hw_ ## option ## _handler (struct config *conf, vector strvec, \return set_int (strvec, &hwe->option, minval, maxval, file, line_nr); \ } +#define declare_hw_arg_str_handler (option, count_idx) \ +static int \ +hw_ ## option ## _handler (struct config *conf, vector strvec, \ + const char *file, int line_nr) \ +{ \ + struct hwentry * hwe = VECTOR_LAST_SLOT(conf->hwtable); \ + if (!hwe) \ + return 1; \ + return set_arg_str(strvec, &hwe->option, count_idx, file, line_nr); \ +} +#define declare_hw_snprint (option, function) \ static int \
@@ -371,6 +442,16 @@ ovr_ ## option ## _handler (struct config *conf, vector strvec, \ file, line_nr); \ } +#define declare_ovr_arg_str_handler (option, count_idx) \ +static int \ +ovr_ ## option ## _handler (struct config *conf, vector strvec, \ + const char *file, int line_nr) \ +{ \ + if (!conf->overrides) \ + return 1; \ + return set_arg_str(strvec, &conf->overrides->option, count_idx, file, line_nr); \ +} +#define declare_ovr_snprint (option, function) \ static int \ snprint_ovr_ ## option (struct config *conf, struct strbuf *buff, \
@@ -401,6 +482,17 @@ mp_ ## option ## _handler (struct config *conf, vector strvec, \return set_int (strvec, &mpe->option, minval, maxval, file, line_nr); \ } +#define declare_mp_arg_str_handler (option, count_idx) \ +static int \ +mp_ ## option ## _handler (struct config *conf, vector strvec, \ + const char *file, int line_nr) \ +{ \ + struct mpentry * mpe = VECTOR_LAST_SLOT(conf->mptable); \ + if (!mpe) \ + return 1; \ + return set_arg_str(strvec, &mpe->option, count_idx, file, line_nr); \ +} +#define declare_mp_snprint (option, function) \ static int \ snprint_mp_ ## option (struct config *conf, struct strbuf *buff, \
@@ -584,13 +676,13 @@ snprint_def_marginal_pathgroups(struct config *conf, struct strbuf *buff, } -declare_def_handler(selector, set_str) +declare_def_arg_str_handler(selector, 1) declare_def_snprint_defstr(selector, print_str, DEFAULT_SELECTOR) -declare_hw_handler(selector, set_str) +declare_hw_arg_str_handler(selector, 1) declare_hw_snprint(selector, print_str) -declare_ovr_handler(selector, set_str) +declare_ovr_arg_str_handler(selector, 1) declare_ovr_snprint(selector, print_str) -declare_mp_handler(selector, set_str) +declare_mp_arg_str_handler(selector, 1) declare_mp_snprint(selector, print_str) static int snprint_uid_attrs(struct config *conf, struct strbuf *buff,
@@ -663,13 +755,13 @@ declare_hw_snprint(prio_args, print_str) declare_mp_handler(prio_args, set_str) declare_mp_snprint(prio_args, print_str) -declare_def_handler(features, set_str) +declare_def_arg_str_handler(features, 0) declare_def_snprint_defstr(features, print_str, DEFAULT_FEATURES) -declare_ovr_handler(features, set_str) +declare_ovr_arg_str_handler(features, 0) declare_ovr_snprint(features, print_str) -declare_hw_handler(features, set_str) +declare_hw_arg_str_handler(features, 0) declare_hw_snprint(features, print_str) -declare_mp_handler(features, set_str) +declare_mp_arg_str_handler(features, 0) declare_mp_snprint(features, print_str)declare_def_handler (checker_name, set_str)
@@ -1821,7 +1913,7 @@ declare_hw_snprint(revision, print_str) declare_hw_handler(bl_product, set_str) declare_hw_snprint(bl_product, print_str) -declare_hw_handler(hwhandler, set_str) +declare_hw_arg_str_handler(hwhandler, 0)declare_hw_snprint (hwhandler, print_str)/*
Signature · 6 lines
2.17.2 -- dm-devel mailing list [email protected] https://listman.redhat.com/mailman/listinfo/dm-devel
Quoted 18 lines · Benjamin Marzinski
The first two patches are a cleanup and a fix for a memory leak in the
config code. The third patch improves multipath's validation of the strings it passes directly into the table, features,path_selector, and
hardware_handler. These three strings all have argument counts, and getting them wrong causes the kernel to parse the table incorrectly. When this happens the table load fails, but the error messages from the
kernel are often completely unhelpful. A bad argument count will cause
the rest of the table to be parsed incorrectly, and the kernel might not
hit an unworkable token till later in the parsing. Multipath now makes
sure that the count matches the actual number of arguments that it is passing.
Quoted 5 lines
V2 Changes (based on suggestions from Martin Wilck):
1/3: unrolled loop invalidate_config_strvec()to explicitly check the
possible quote strings
3/3: spaces is now a "const char * const" and doesn't include '\n'
Hm, my suggestion was wrong. It shouldn't be a pointer at all but an array:
static const char spaces[] = " \f\r\t\v";
"static" makes sure it's only initialized once, and ends up in the .rodata section.
In practice, there's no significant difference between either version. So, for the set:
Martin
--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel
Benjamin Marzinski
maintainerQuoted 25 lines · Martin Wilck
Quoted 18 lines · Benjamin Marzinski
The first two patches are a cleanup and a fix for a memory leak in the
config code. The third patch improves multipath's validation of the strings it passes directly into the table, features,path_selector, and
hardware_handler. These three strings all have argument counts, and getting them wrong causes the kernel to parse the table incorrectly. When this happens the table load fails, but the error messages from the
kernel are often completely unhelpful. A bad argument count will cause
the rest of the table to be parsed incorrectly, and the kernel might not
hit an unworkable token till later in the parsing. Multipath now makes
sure that the count matches the actual number of arguments that it is passing.Quoted 5 lines
V2 Changes (based on suggestions from Martin Wilck):
1/3: unrolled loop invalidate_config_strvec()to explicitly check the
possible quote strings
3/3: spaces is now a "const char * const" and doesn't include '\n'
Quoted 7 lines
Hm, my suggestion was wrong. It shouldn't be a pointer at all but an array:
static const char spaces[] = " \f\r\t\v";"static" makes sure it's only initialized once, and ends up in the .rodata section.
Dumb question. If you explicitly make it an array, does that mean the compiler will always allocate separate memory for it, even if there is another identical const array? With multiple pointers to the same const string, the compiler will often just have one string in memory, which all the pointers refer to. Not sure if the same thing happens when they're defined as arrays.
-Ben
Quoted 7 lines
In practice, there's no significant difference between either version. So, for the set:
Reviewed-by Martin Wilck <[email protected]>Martin
--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel
Benjamin Marzinski
maintainerQuoted 34 lines · Benjamin Marzinski
Quoted 25 lines · Martin Wilck
Quoted 18 lines · Benjamin Marzinski
The first two patches are a cleanup and a fix for a memory leak in the
config code. The third patch improves multipath's validation of the strings it passes directly into the table, features,path_selector, and
hardware_handler. These three strings all have argument counts, and getting them wrong causes the kernel to parse the table incorrectly. When this happens the table load fails, but the error messages from the
kernel are often completely unhelpful. A bad argument count will cause
the rest of the table to be parsed incorrectly, and the kernel might not
hit an unworkable token till later in the parsing. Multipath now makes
sure that the count matches the actual number of arguments that it is passing.Quoted 5 lines
V2 Changes (based on suggestions from Martin Wilck):
1/3: unrolled loop invalidate_config_strvec()to explicitly check the
possible quote strings
3/3: spaces is now a "const char * const" and doesn't include '\n'Quoted 7 lines
Hm, my suggestion was wrong. It shouldn't be a pointer at all but an array:
static const char spaces[] = " \f\r\t\v";"static" makes sure it's only initialized once, and ends up in the .rodata section.
Quoted 8 lines
Dumb question. If you explicitly make it an array, does that mean the compiler will always allocate separate memory for it, even if there is another identical const array? With multiple pointers to the same const string, the compiler will often just have one string in memory, which all the pointers refer to. Not sure if the same thing happens when they're defined as arrays.
-Ben
FWIW, I just checked with gcc version 11.3.1. Using:
static const char * const ptr1 = "test";
static const char * const ptr2 = "test";
static const char arr1[] = "test";
static const char arr2[] = "test";
printf("pointers are %s\n", (ptr1 == ptr2) ? "equal" : "not equal"); printf("arrays are %s\n", (arr1 == arr2) ? "equal" : "not equal");
I get:
pointers are equal
arrays are not equal
So depending on whether or not we use this multiple times, we can save ourselves.. a whopping 6 bytes. At any rate, now I know.
-Ben
Quoted 8 lines
Quoted 8 lines
In practice, there's no significant difference between either version. So, for the set:
Reviewed-by Martin Wilck <[email protected]>Martin
--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel
Quoted 11 lines · Benjamin Marzinski
Quoted 9 lines · Martin Wilck
Hm, my suggestion was wrong. It shouldn't be a pointer at all but an
array:static const char spaces[] = " \f\r\t\v";
"static" makes sure it's only initialized once, and ends up in the .rodata section.
Quoted 10 lines
Dumb question. If you explicitly make it an array, does that mean the compiler will always allocate separate memory for it, even if there is
another identical const array? With multiple pointers to the same const
string, the compiler will often just have one string in memory, which all
the pointers refer to. Not sure if the same thing happens when they're
defined as arrays.
I am unsure about it, either. A quick experiment shows that the compiler does not merge multiple const char arrays with the same value into one memory location, while it does for string constants. (I suppose this works only for constants defined in the same source file?).
Also, if you use a pointer (rather than an array), the compiler is smart enough not to allocate memory for the pointer variable. At least in simple situations, it's probably the best idea to just use the string constant without declaring a variable. So you could have written
#define SPACES " \f\r\t\v"
...
p += strspn(p, SPACES);
This would benefit from gcc's constant merging. In the case at hand, the compiler output would be nearly the same between all 3 approaches. But this would change if you started using SPACES in some other function.
What definitely should not be used is
const char spaces[] = " \f\r\t\v";
(without "static") because it causes the array to be allocated on the stack and re-initialized at every function call.
The moral: we shouldn't try to be smarter than the compiler (this refers to my review :-/)
Martin
--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel