Nothing prevents a dentry found by path lookup before a return of
__debugfs_remove() to actually get opened after that return. Now, after
the return of __debugfs_remove(), there are no guarantees whatsoever
regarding the memory the corresponding inode's file_operations object
had been kept in.
Since __debugfs_remove() is seldomly invoked, usually from module exit
handlers only, the race is hard to trigger and the impact is very low.
A discussion of the problem outlined above as well as a suggested
solution can be found in the (sub-)thread rooted at
http://lkml.kernel.org/g/20130401203445.GA20862@ZenIV.linux.org.uk
("Yet another pipe related oops.")
Basically, Greg KH suggests to introduce an intermediate fops and
Al Viro points out that a pointer to the original ones may be stored in
->d_fsdata.
Follow this line of reasoning:
- Add SRCU as a reverse dependency of DEBUG_FS.
- Introduce a srcu_struct object for the debugfs subsystem.
- In debugfs_create_file(), store a pointer to the original
file_operations object in ->d_fsdata.
- Make debugfs_remove() and debugfs_remove_recursive() wait for a
SRCU grace period after the dentry has been delete()'d and before they
return to their callers.
- Introduce an intermediate file_operations object named
"debugfs_open_proxy_file_operations". It's ->open() functions checks,
under the protection of a SRCU read lock, whether the dentry is still
alive, i.e. has not been d_delete()'d and if so, tries to acquire a
reference on the owning module.
On success, it sets the file object's ->f_op to the original
file_operations and forwards the ongoing open() call to the original
->open().
- For clarity, rename the former debugfs_file_operations to
debugfs_noop_file_operations -- they are in no way canonical.
The choice of SRCU over "normal" RCU is justified by the fact, that the
former may also be used to protect ->i_private data from going away
during the execution of a file's readers and writers which may (and do)
sleep.
Finally, introduce the fs/debugfs/internal.h header containing some
declarations internal to the debugfs implementation.
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Within the copying loop in mpi_read_raw_from_sgl(), the last input SGE's
byte count gets artificially extended as follows:
if (sg_is_last(sg) && (len % BYTES_PER_MPI_LIMB))
len += BYTES_PER_MPI_LIMB - (len % BYTES_PER_MPI_LIMB);
Within the following byte copying loop, this causes reads beyond that
SGE's allocated buffer:
BUG: KASAN: slab-out-of-bounds in mpi_read_raw_from_sgl+0x331/0x650
at addr ffff8801e168d4d8
Read of size 1 by task systemd-udevd/721
[...]
Call Trace:
[<ffffffff818c4d35>] dump_stack+0xbc/0x117
[<ffffffff818c4c79>] ? _atomic_dec_and_lock+0x169/0x169
[<ffffffff814af5d1>] ? print_section+0x61/0xb0
[<ffffffff814b1109>] print_trailer+0x179/0x2c0
[<ffffffff814bc524>] object_err+0x34/0x40
[<ffffffff814bfdc7>] kasan_report_error+0x307/0x8c0
[<ffffffff814bf315>] ? kasan_unpoison_shadow+0x35/0x50
[<ffffffff814bf38e>] ? kasan_kmalloc+0x5e/0x70
[<ffffffff814c0ad1>] kasan_report+0x71/0xa0
[<ffffffff81938171>] ? mpi_read_raw_from_sgl+0x331/0x650
[<ffffffff814bf1a6>] __asan_load1+0x46/0x50
[<ffffffff81938171>] mpi_read_raw_from_sgl+0x331/0x650
[<ffffffff817f41b6>] rsa_verify+0x106/0x260
[<ffffffff817f40b0>] ? rsa_set_pub_key+0xf0/0xf0
[<ffffffff818edc79>] ? sg_init_table+0x29/0x50
[<ffffffff817f4d22>] ? pkcs1pad_sg_set_buf+0xb2/0x2e0
[<ffffffff817f5b74>] pkcs1pad_verify+0x1f4/0x2b0
[<ffffffff81831057>] public_key_verify_signature+0x3a7/0x5e0
[<ffffffff81830cb0>] ? public_key_describe+0x80/0x80
[<ffffffff817830f0>] ? keyring_search_aux+0x150/0x150
[<ffffffff818334a4>] ? x509_request_asymmetric_key+0x114/0x370
[<ffffffff814b83f0>] ? kfree+0x220/0x370
[<ffffffff818312c2>] public_key_verify_signature_2+0x32/0x50
[<ffffffff81830b5c>] verify_signature+0x7c/0xb0
[<ffffffff81835d0c>] pkcs7_validate_trust+0x42c/0x5f0
[<ffffffff813c391a>] system_verify_data+0xca/0x170
[<ffffffff813c3850>] ? top_trace_array+0x9b/0x9b
[<ffffffff81510b29>] ? __vfs_read+0x279/0x3d0
[<ffffffff8129372f>] mod_verify_sig+0x1ff/0x290
[...]
The exact purpose of the len extension isn't clear to me, but due to
its form, I suspect that it's a leftover somehow accounting for leading
zero bytes within the most significant output limb.
Note however that without that len adjustement, the total number of bytes
ever processed by the inner loop equals nbytes and thus, the last output
limb gets written at this point. Thus the net effect of the len adjustement
cited above is just to keep the inner loop running for some more
iterations, namely < BYTES_PER_MPI_LIMB ones, reading some extra bytes from
beyond the last SGE's buffer and discarding them afterwards.
Fix this issue by purging the extension of len beyond the last input SGE's
buffer length.
Fixes: 2d4d1eea54 ("lib/mpi: Add mpi sgl helpers")
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Within the byte reading loop in mpi_read_raw_sgl(), there are two
housekeeping indices used, z and x.
At all times, the index z represents the number of output bytes covered
by the input SGEs for which processing has completed so far. This includes
any leading zero bytes within the most significant limb.
The index x changes its meaning after the first outer loop's iteration
though: while processing the first input SGE, it represents
"number of leading zero bytes in most significant output limb" +
"current position within current SGE"
For the remaining SGEs OTOH, x corresponds just to
"current position within current SGE"
After all, it is only the sum of z and x that has any meaning for the
output buffer and thus, the
"number of leading zero bytes in most significant output limb"
part can be moved away from x into z from the beginning, opening up the
opportunity for cleaner code.
Before the outer loop iterating over the SGEs, don't initialize z with
zero, but with the number of leading zero bytes in the most significant
output limb. For the inner loop iterating over a single SGE's bytes,
get rid of the buf_shift offset to x' bounds and let x run from zero to
sg->length - 1.
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The number of bits, nbits, is calculated in mpi_read_raw_from_sgl() as
follows:
nbits = nbytes * 8;
Afterwards, the number of leading zero bits of the first byte get
subtracted:
nbits -= count_leading_zeros(*(u8 *)(sg_virt(sgl) + lzeros));
However, count_leading_zeros() takes an unsigned long and thus,
the u8 gets promoted to an unsigned long.
Thus, the above doesn't subtract the number of leading zeros in the most
significant nonzero input byte from nbits, but the number of leading
zeros of the most significant nonzero input byte promoted to unsigned long,
i.e. BITS_PER_LONG - 8 too many.
Fix this by subtracting
count_leading_zeros(...) - (BITS_PER_LONG - 8)
from nbits only.
Fixes: 2d4d1eea54 ("lib/mpi: Add mpi sgl helpers")
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In mpi_read_raw_from_sgl(), unsigned nbits is calculated as follows:
nbits = nbytes * 8;
and redundantly cleared later on if nbytes == 0:
if (nbytes > 0)
...
else
nbits = 0;
Purge this redundant clearing for the sake of clarity.
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
At the very beginning of mpi_read_raw_from_sgl(), the leading zeros of
the input scatterlist are counted:
lzeros = 0;
for_each_sg(sgl, sg, ents, i) {
...
if (/* sg contains nonzero bytes */)
break;
/* sg contains nothing but zeros here */
ents--;
lzeros = 0;
}
Later on, the total number of trailing nonzero bytes is calculated by
subtracting the number of leading zero bytes from the total number of input
bytes:
nbytes -= lzeros;
However, since lzeros gets reset to zero for each completely zero leading
sg in the loop above, it doesn't include those.
Besides wasting resources by allocating a too large output buffer,
this mistake propagates into the calculation of x, the number of
leading zeros within the most significant output limb:
x = BYTES_PER_MPI_LIMB - nbytes % BYTES_PER_MPI_LIMB;
What's more, the low order bytes of the output, equal in number to the
extra bytes in nbytes, are left uninitialized.
Fix this by adjusting nbytes for each completely zero leading scatterlist
entry.
Fixes: 2d4d1eea54 ("lib/mpi: Add mpi sgl helpers")
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently, the nbytes local variable is calculated from the len argument
as follows:
... mpi_read_raw_from_sgl(..., unsigned int len)
{
unsigned nbytes;
...
if (!ents)
nbytes = 0;
else
nbytes = len - lzeros;
...
}
Given that nbytes is derived from len in a trivial way and that the len
argument is shadowed by a local len variable in several loops, this is just
confusing.
Rename the len argument to nbytes and get rid of the nbytes local variable.
Do the nbytes calculation in place.
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently, mpi_read_buffer() writes full limbs to the output buffer
and moves memory around to purge leading zero limbs afterwards.
However, with
commit 9cbe21d8f8 ("lib/mpi: only require buffers as big as needed for
the integer")
the caller is only required to provide a buffer large enough to hold the
result without the leading zeros.
This might result in a buffer overflow for small MP numbers with leading
zeros.
Fix this by coping the result to its final destination within the output
buffer and not copying the leading zeros at all.
Fixes: 9cbe21d8f8 ("lib/mpi: only require buffers as big as needed for
the integer")
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently, the endian conversion from CPU order to BE is open coded in
mpi_read_buffer().
Replace this by the centrally provided cpu_to_be*() macros.
Copy from the temporary storage on stack to the destination buffer
by means of memcpy().
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently, if the number of leading zeros is greater than fits into a
complete limb, mpi_read_buffer() skips them by iterating over them
limb-wise.
Instead of skipping the high order zero limbs within the loop as shown
above, adjust the copying loop's bounds.
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently, the endian conversion from CPU order to BE is open coded in
mpi_write_sgl().
Replace this by the centrally provided cpu_to_be*() macros.
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Within the copying loop in mpi_write_sgl(), we have
if (lzeros) {
mpi_limb_t *limb1 = (void *)p - sizeof(alimb);
mpi_limb_t *limb2 = (void *)p - sizeof(alimb)
+ lzeros;
*limb1 = *limb2;
...
}
where p points past the end of alimb2 which lives on the stack and contains
the current limb in BE order.
The purpose of the above is to shift the non-zero bytes of alimb2 to its
beginning in memory, i.e. to skip its leading zero bytes.
However, limb2 points somewhere into the middle of alimb2 and thus, reading
*limb2 pulls in lzero bytes from somewhere.
Indeed, KASAN splats:
BUG: KASAN: stack-out-of-bounds in mpi_write_to_sgl+0x4e3/0x6f0
at addr ffff8800cb04f601
Read of size 8 by task systemd-udevd/391
page:ffffea00032c13c0 count:0 mapcount:0 mapping: (null) index:0x0
flags: 0x3fff8000000000()
page dumped because: kasan: bad access detected
CPU: 3 PID: 391 Comm: systemd-udevd Tainted: G B L
4.5.0-next-20160316+ #12
[...]
Call Trace:
[<ffffffff8194889e>] dump_stack+0xdc/0x15e
[<ffffffff819487c2>] ? _atomic_dec_and_lock+0xa2/0xa2
[<ffffffff814892b5>] ? __dump_page+0x185/0x330
[<ffffffff8150ffd6>] kasan_report_error+0x5e6/0x8b0
[<ffffffff814724cd>] ? kzfree+0x2d/0x40
[<ffffffff819c5bce>] ? mpi_free_limb_space+0xe/0x20
[<ffffffff819c469e>] ? mpi_powm+0x37e/0x16f0
[<ffffffff815109f1>] kasan_report+0x71/0xa0
[<ffffffff819c0353>] ? mpi_write_to_sgl+0x4e3/0x6f0
[<ffffffff8150ed34>] __asan_load8+0x64/0x70
[<ffffffff819c0353>] mpi_write_to_sgl+0x4e3/0x6f0
[<ffffffff819bfe70>] ? mpi_set_buffer+0x620/0x620
[<ffffffff819c0e6f>] ? mpi_cmp+0xbf/0x180
[<ffffffff8186e282>] rsa_verify+0x202/0x260
What's more, since lzeros can be anything from 1 to sizeof(mpi_limb_t)-1,
the above will cause unaligned accesses which is bad on non-x86 archs.
Fix the issue, by preparing the starting point p for the upcoming copy
operation instead of shifting the source memory, i.e. alimb2.
Fixes: 2d4d1eea54 ("lib/mpi: Add mpi sgl helpers")
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Within the copying loop in mpi_write_sgl(), we have
if (lzeros) {
...
p -= lzeros;
y = lzeros;
}
p = p - (sizeof(alimb) - y);
If lzeros == 0, then y == 0, too. Thus, lzeros gets subtracted and added
back again to p.
Purge this redundancy.
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Within the copying loop in mpi_write_sgl(), we have
if (lzeros > 0) {
...
lzeros -= sizeof(alimb);
}
However, at this point, lzeros < sizeof(alimb) holds. Make this fact
explicit by rewriting the above to
if (lzeros) {
...
lzeros = 0;
}
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently, if the number of leading zeros is greater than fits into a
complete limb, mpi_write_sgl() skips them by iterating over them limb-wise.
However, it fails to adjust its internal leading zeros tracking variable,
lzeros, accordingly: it does a
p -= sizeof(alimb);
continue;
which should really have been a
lzeros -= sizeof(alimb);
continue;
Since lzeros never decreases if its initial value >= sizeof(alimb), nothing
gets copied by mpi_write_sgl() in that case.
Instead of skipping the high order zero limbs within the loop as shown
above, fix the issue by adjusting the copying loop's bounds.
Fixes: 2d4d1eea54 ("lib/mpi: Add mpi sgl helpers")
Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In certain cases, the 802.11 mesh pathtable code wants to
iterate over all of the entries in the forwarding table from
the receive path, which is inside an RCU read-side critical
section. Enable walks inside atomic sections by allowing
GFP_ATOMIC allocations for the walker state.
Change all existing callsites to pass in GFP_KERNEL.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Bob Copeland <me@bobcopeland.com>
[also adjust gfs2/glock.c and rhashtable tests]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Rui Salvaterra
James Morris
Nicolai Stange
David S. Miller
Al Viro
Naveen N. Rao
Jerome Marchand
Bob Copeland