readeither() caches (in new_p) a pointer to the proc_t of a task-group
leader, but readeither()'s callers can do pretty much anything with the
proc_t structure passed to and/or returned by this function. For
example, they can 1/ free it or 2/ recycle it (by passing it to
readeither() as x).
1/ leads to a use-after-free, and 2/ leads to unexpected behavior when
taskreader()/simple_readtask() is called with new_p equal to x (this is
not a theoretical flaw: 2/ happens in readproctab3() when want_task()
returns false and p is a group leader).
As a workaround, we keep a copy of new_p's first member (tid) in static
storage, and the next times we enter readeither() we check this "canary"
against the tid in new_p: if they differ, we reset new_p to NULL, which
forces the allocation of a new proc_t (the new "leader", or reference).
This always detects 2/ (because free_acquired(x,1) memsets x and hence
new_p); always detects 1/ if freed via free_acquired() and/or freeproc()
(very likely, otherwise memory may be leaked); probably detects 1/ even
if freed directly via free() (because the canary is the first member of
proc_t, likely to be overwritten by free()); but can not detect 1/ if
free() does not write to new_p's chunk at all.
Moreover, accessing new_p->tid to check the canary in case 1/ is itself
a use-after-free, so a better long-term solution should be implemented
at some point (we wanted to avoid intrusive and backward-incompatible
changes in this library function, hence this imperfect workaround).
If an integer overflow is about to be reached, call xalloc_err_handler()
(since it would have been caught by calloc() or reallocarray()) and then
exit(): these integer overflows are far from reachable, with the current
PID_MAX_LIMIT (2^22), so if they are there is something very wrong going
on. Note: we check the n_*alloc variables against INT_MAX even when they
are size_t because they are later stored as int in a struct proc_data_t.
If QUICK_THREADS is not defined (it is not by default, but most
distributions enable it) and task_dir_missing is true (only on very old
kernels), then readtask() forgets to reset some of the struct proc_t t's
members, which later results in double-free()s in free_acquired().
For now, we simply synchronized the list of members to be reset with the
list of members freed in free_acquired().
The memset() in the PROC_LOOSE_TASKS loop leaves a struct proc_t
uninitialized (the one at data+n_used), which leads to a use-after-free.
ps calls readproctab2(), but only if !TF_loose_tasks, and this U-A-F is
triggered only if PROC_LOOSE_TASKS, so there seems to be no vulnerable
call in the procps package itself (other users of the libprocps may be
vulnerable, though).
Check the return value of snprintf(), otherwise dst may point
out-of-bounds when it reaches the end of the dst_buffer (the snprintf()
always returns 1 in that case, even if there is not enough space left),
and vMAX becomes negative and is passed to snprintf() as a size_t.
This detects an integer overflow of "strlen + 1", prevents an integer
overflow of "tot + adj + (2 * pSZ)", and avoids calling snprintf with a
string longer than INT_MAX. Truncate rather than fail, since the callers
do not expect a failure of this function.
1/ Prevent an out-of-bounds write if sz is 0.
2/ Limit sz to INT_MAX, because the return value is an int, not an
unsigned int (and because if INT_MAX is equal to SSIZE_MAX, man 2 read
says "If count is greater than SSIZE_MAX, the result is unspecified.")
3/ Always null-terminate dst (unless sz is 0), because a return value of
0 because of an open() error (for example) is indistinguishable from a
return value of 0 because of an empty file.
4/ Use an unsigned int for i (just like n), not an int.
5/ Check for snprintf() truncation.
Note: this is by far the most important and complex patch of the whole
series, please review it carefully; thank you very much!
For this patch, we decided to keep the original function's design and
skeleton, to avoid regressions and behavior changes, while fixing the
various bugs and overflows. And like the "Harden file2str()" patch, this
patch does not fail when about to overflow, but truncates instead: there
is information available about this process, so return it to the caller;
also, we used INT_MAX as a limit, but a lower limit could be used.
The easy changes:
- Replace sprintf() with snprintf() (and check for truncation).
- Replace "if (n == 0 && rbuf == 0)" with "if (n <= 0 && tot <= 0)" and
do break instead of return: it simplifies the code (only one place to
handle errors), and also guarantees that in the while loop either n or
tot is > 0 (or both), even if n is reset to 0 when about to overflow.
- Remove the "if (n < 0)" block in the while loop: it is (and was) dead
code, since we enter the while loop only if n >= 0.
- Rewrite the missing-null-terminator detection: in the original
function, if the size of the file is a multiple of 2047, a null-
terminator is appended even if the file is already null-terminated.
- Replace "if (n <= 0 && !end_of_file)" with "if (n < 0 || tot <= 0)":
originally, it was equivalent to "if (n < 0)", but we added "tot <= 0"
to handle the first break of the while loop, and to guarantee that in
the rest of the function tot is > 0.
- Double-force ("belt and suspenders") the null-termination of rbuf:
this is (and was) essential to the correctness of the function.
- Replace the final "while" loop with a "for" loop that behaves just
like the preceding "for" loop: in the original function, this would
lead to unexpected results (for example, if rbuf is |\0|A|\0|, this
would return the array {"",NULL} but should return {"","A",NULL}; and
if rbuf is |A|\0|B| (should never happen because rbuf should be null-
terminated), this would make room for two pointers in ret, but would
write three pointers to ret).
The hard changes:
- Prevent the integer overflow of tot in the while loop, but unlike
file2str(), file2strvec() cannot let tot grow until it almost reaches
INT_MAX, because it needs more space for the pointers: this is why we
introduced ARG_LEN, which also guarantees that we can add "align" and
a few sizeof(char*)s to tot without overflowing.
- Prevent the integer overflow of "tot + c + align": when INT_MAX is
(almost) reached, we write the maximal safe amount of pointers to ret
(ARG_LEN guarantees that there is always space for *ret = rbuf and the
NULL terminator).
1/ Replace sprintf() with snprintf() (and check for truncation).
2/ Prevent an integer overflow of ub->siz. The "tot_read--" is needed to
avoid an off-by-one overflow in "ub->buf[tot_read] = '\0'". It is safe
to decrement tot_read here, because we know that tot_read is equal to
ub->siz (and ub->siz is very large).
We believe that truncation is a better option than failure (implementing
failure instead should be as easy as replacing the "tot_read--" with
"tot_read = 0").
1/ Use a "size_t num" instead of an "unsigned num" (also, do not store
the return value of sscanf() into num, it was unused anyway).
2/ Check the return value of strchr() and strrchr().
3/ Never jump over the terminating null byte with "S = tmp + 2".
1/ Prevent an integer overflow of t.
2/ Avoid an infinite loop if s contains characters other than comma,
spaces, +, -, and digits.
3/ Handle all possible return values of snprintf().
1/ Do not read past the terminating null byte when hashing the name.
2/ S[x] is used as an index, but S is "char *S" (signed) and hence may
index the array out-of-bounds. Bit-mask S[x] with 127 (the array has 128
entries).
3/ Use a size_t for j, not an int (strlen() returns a size_t).
Notes:
- These are (mostly) theoretical problems, because the contents of
/proc/PID/status are (mostly) trusted.
- The "name" member of the status_table_struct has 8 bytes, and
"RssShmem" occupies exactly 8 bytes, which means that "name" is not
null-terminated. This is fine right now, because status2proc() uses
memcmp(), not strcmp(), but it is worth mentioning.
This function is unused (SIGNAL_STRING is defined by default, and if it
is not, procps does not compile -- for example, there is no "outbuf" in
help_pr_sig()) but fix it anyway. There are two bugs:
- it accepts non-hexadecimal characters (anything >= 0x30);
- "(c - (c>0x57) ? 0x57 : 0x30)" is always equal to 0x57.
There was a "buff[BUFFSIZE-1] = 0;" but there may be garbage between
what is read() (less than BUFFSIZE-1 bytes) and this null byte. Reuse
the construct from the preceding getrunners().
To avoid an out-of-bounds access at checkoff[tmp]. The strspn() at the
beginning of the function protects against it already, but double-check
this in case of some future change.
Right now, "we _exit() anyway" is not always true: for example, the
default action for SIGURG is to ignore the signal, which means that
"kill(getpid(), signo);" does not terminate the process. Call _exit()
explicitly, in this case (rather than exit(), because the terminating
kill() calls do not call the functions registered with atexit() either).
Before "strlen(outbuf)", if one of the pr_*() functions forgot to do it.
This prevents an out-of-bounds read in strlen(), and an out-of-bounds
write in "outbuf[sz] = '\n'". Another solution would be to replace
strlen() with strnlen(), but this is not used anywhere else in the
code-base and may not exist in all libc's.
pr_bsdstart(): Replace "strcpy(outbuf," with "snprintf(outbuf, COLWID,"
(which is used in all surrounding functions). (side note: the fact that
many pr_*() functions simply return "snprintf(outbuf, COLWID," justifies
the "amount" checks added to show_one_proc() by the "ps/output.c:
Replace strcpy() with snprintf() in show_one_proc()." patch)
pr_stime(): Check the return value of strftime() (in case of an error,
"the contents of the array are undefined").
help_pr_sig(): Handle the "len < 8" case, otherwise "sig+len-8" may
point outside the sig string.
pr_context(): Handle the empty string case, or else "outbuf[len-1]"
points outside outbuf.
Enforce a maximum max_rightward of OUTBUF_SIZE-1, because it is used in
constructs such as "snprintf(outbuf, max_rightward+1," (we could remove
the extra check at the beginning of forest_helper() now, but we decided
to leave it, as a precaution and reminder).
The minimum max_rightward check is not strictly needed, because it is
unsigned. However, we decided to add it anyway:
- most of the other variables are signed;
- make it visually clear that this case is properly handled;
- ideally, the minimum max_rightward should be 1, not 0 (to prevent
integer overflows such as "max_rightward-1"), but this might change
the behavior/output of ps, so we decided against it, for now.
Instead, we fixed the only function that overflows if max_rightward is
0. Also, enforce the same safe range for max_leftward, although it is
never used throughout the code-base.
1/ Do not go deeper than the size of forest_prefix[], to prevent a
buffer overflow (sizeof(forest_prefix) is roughly 128K, but the maximum
/proc/sys/kernel/pid_max is 4M). (actually, we go deeper, but we stop
adding bytes to forest_prefix[])
2/ Always null-terminate forest_prefix[] at the current level.
This patch solves several problems:
1/ Limit the number of characters written (to outbuf) to OUTBUF_SIZE-1
(-1 for the null-terminator).
2/ Always null-terminate outbuf at q.
3/ Move the "rightward" checks *before* the strcpy() calls.
4/ Avoid an integer overflow in these checks (e.g., rightward-4).
May happen if strlen(src) > INT_MAX for example. This patch prevents
escaped_copy() from increasing maxroom and returning -1 (= number of
bytes consumed in dst).
This solves several problems:
1/ outbuf[1] was written to, but not outbuf[0], which was left
uninitialized (well, SECURE_ESCAPE_ARGS() already fixes this, but do it
explicitly as well); we know it is safe to write one byte to outbuf,
because SECURE_ESCAPE_ARGS() guarantees it.
2/ If bytes was 1, the write to outbuf[1] was an off-by-one overflow.
3/ Do not call escape_str() with a 0 bufsize if bytes == overhead.
4/ Prevent various buffer overflows if bytes <= overhead.
Simply rearrange the old comparisons. The new comparisons are safe,
because we know from previous checks that:
1/ wlen > 0
2/ my_cells < *maxcells (also: my_cells >= 0 and *maxcells > 0)
3/ len > 1
4/ my_bytes+1 < bufsize (also: my_bytes >= 0 and bufsize > 0)
This should never happen, because wcwidth() is called only if iswprint()
returns nonzero. But belt-and-suspenders, and make it visually clear
(very important for the next patch).
The SECURE_ESCAPE_ARGS() macro solves several potential problems
(although we found no problematic calls to the escape*() functions in
procps's code-base, but had to thoroughly review every call; and this is
library code):
1/ off-by-one overflows if the size of the destination buffer is 0;
2/ buffer overflows if this size (or "maxroom") is negative;
3/ integer overflows (for example, "*maxcells+1");
4/ always null-terminate the destination buffer (unless its size is 0).
In the human_readable case; otherwise the strcat() that follows may
append bytes to the previous contents of buf.
Also, slightly enlarge buf, as it was a bit too tight.
Could also replace all sprintf()s with snprintf()s, but all the calls
here output a limited number of characters, so they should be safe.
Especially its "next" member: this is what caused the crash in "slabtop:
Reset slab_list if get_slabinfo() fails." (if parse_slabinfo*() fails in
sscanf(), for example, then curr is set to NULL but it is already linked
into the "list" and its "next" member was never initialized).
In proc/slab.c, functions parse_slabinfo20() and parse_slabinfo11(),
sscanf() might overflow curr->name, because "String input conversions
store a terminating null byte ('\0') to mark the end of the input; the
maximum field width does not include this terminator."
Add one byte to name[] for this terminator.
