Fix invalid memory write in case of failing mmap(...) in PosixLockedPageAllocator::AllocateLocked#15117
Conversation
…geAllocator::AllocateLocked
| return static_cast<T*>(LockedPoolManager::Instance().alloc(sizeof(T) * n)); | ||
| T* allocation = static_cast<T*>(LockedPoolManager::Instance().alloc(sizeof(T) * n)); | ||
| if (!allocation) { | ||
| throw std::bad_alloc(); |
There was a problem hiding this comment.
I don't think bad_alloc have any parameterized ctor:s?
There was a problem hiding this comment.
Indeed - I guess the other option would be to log it in the allocate method. (nit only)
| addr = mmap(nullptr, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); | ||
| if (addr == MAP_FAILED) { | ||
| return nullptr; | ||
| } |
There was a problem hiding this comment.
Judging from leveldb's handling, and the mmap docs, the null value case is negligible:
bitcoin/src/leveldb/util/env_posix.cc
Lines 367 to 372 in 9c71998
There was a problem hiding this comment.
I'm not sure how to parse TBH: are you suggesting a change here? :-)
There was a problem hiding this comment.
Just that the if (addr) is strictly unnecessary.
|
Concept ACK |
|
TIL. Always thought it returned utACK ca126d4 |
|
Concept ACK. I too assumed mmap returned NULL on failure. |
|
Concept ACK - good spot! |
…n PosixLockedPageAllocator::AllocateLocked ca126d4 Fix out-of-bounds write in case of failing mmap(...) in PosixLockedPageAllocator::AllocateLocked (practicalswift) Pull request description: `mmap(...)` returns `MAP_FAILED` (`(void *) -1`) in case of allocation failure. `PosixLockedPageAllocator::AllocateLocked(...)` did not check for allocation failures prior to this PR. Instead the invalid memory address `(void *) -1` (`0xffffffffffffffff`) was passed to the caller as if it was a valid address. After some operations the address is wrapped around from `0xffffffffffffffff` to `0x00000003ffdf` (`0xffffffffffffffff + 262112 == 0x00000003ffdf`); The resulting address `0x00000003ffdf` is then written to. Before this patch (with failing `mmap` call): ``` $ src/bitcoind … 2019-01-06T16:28:14Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation 2019-01-06T16:28:14Z Using RdRand as an additional entropy source Segmentation fault (core dumped) ``` Before this patch (under `valgrind` with failing `mmap` call): ``` $ valgrind src/bitcoind … 2019-01-06T16:28:51Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation ==17812== Invalid write of size 1 ==17812== at 0x500B7E: void __gnu_cxx::new_allocator<unsigned char>::construct<unsigned char>(unsigned char*) (new_allocator.h:136) ==17812== by 0x500B52: _ZNSt16allocator_traitsI16secure_allocatorIhEE12_S_constructIhJEEENSt9enable_ifIXsr6__and_INS2_18__construct_helperIT_JDpT0_EE4typeEEE5valueEvE4typeERS1_PS6_DpOS7_ (alloc_traits.h:243) ==17812== by 0x500B22: _ZNSt16allocator_traitsI16secure_allocatorIhEE9constructIhJEEEDTcl12_S_constructfp_fp0_spclsr3stdE7forwardIT0_Efp1_EEERS1_PT_DpOS4_ (alloc_traits.h:344) ==17812== by 0x500982: unsigned char* std::__uninitialized_default_n_a<unsigned char*, unsigned long, secure_allocator<unsigned char> >(unsigned char*, unsigned long, secure_allocator<unsigned char>&) (stl_uninitialized.h:631) ==17812== by 0x60BFC2: std::vector<unsigned char, secure_allocator<unsigned char> >::_M_default_initialize(unsigned long) (stl_vector.h:1347) ==17812== by 0x60BD86: std::vector<unsigned char, secure_allocator<unsigned char> >::vector(unsigned long, secure_allocator<unsigned char> const&) (stl_vector.h:285) ==17812== by 0x60BB55: ECC_Start() (key.cpp:351) ==17812== by 0x16AC90: AppInitSanityChecks() (init.cpp:1162) ==17812== by 0x15BAC9: AppInit(int, char**) (bitcoind.cpp:138) ==17812== by 0x15B6C8: main (bitcoind.cpp:201) ==17812== Address 0x3ffdf is not stack'd, malloc'd or (recently) free'd … Segmentation fault (core dumped) ``` After this patch (with failing `mmap` call): ``` $ src/bitcoind … 2019-01-06T15:50:18Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation 2019-01-06T15:50:18Z Using RdRand as an additional entropy source 2019-01-06T15:50:18Z ************************ EXCEPTION: St9bad_alloc std::bad_alloc bitcoin in AppInit() ************************ EXCEPTION: St9bad_alloc std::bad_alloc bitcoin in AppInit() 2019-01-06T15:50:18Z Shutdown: In progress... 2019-01-06T15:50:18Z Shutdown: done ``` To simulate the failing `mmap` call apply the following to `master`: ```diff diff --git a/src/support/lockedpool.cpp b/src/support/lockedpool.cpp index 8d577cf..ce79e569b 100644 --- a/src/support/lockedpool.cpp +++ b/src/support/lockedpool.cpp @@ -247,7 +247,8 @@ void *PosixLockedPageAllocator::AllocateLocked(size_t len, bool *lockingSuccess) { void *addr; len = align_up(len, page_size); - addr = mmap(nullptr, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); + // addr = mmap(nullptr, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); + addr = MAP_FAILED; if (addr) { *lockingSuccess = mlock(addr, len) == 0; } ``` Tree-SHA512: 66947f5fc0fbb19afb3e1edbd51df07df9d16b77018cff3d48d30f378a53d6a0dc62bc36622b3966b7e374e61edbcca114ef4ac8ae8d725022c1a597edcbf7c7
Locked memory manager Add a pool for locked memory chunks, replacing `LockedPageManager`. Cherry-picked from the following upstream PRs: - bitcoin/bitcoin#8321 - bitcoin/bitcoin#8753 - bitcoin/bitcoin#9063 - bitcoin/bitcoin#9070 - bitcoin/bitcoin#11385 - bitcoin/bitcoin#12048 - Excludes change to benchmark. - bitcoin/bitcoin#15117 - bitcoin/bitcoin#16161 - Excludes Travis CI changes. - Includes change from bitcoin/bitcoin#13163 - bitcoin/bitcoin#15600 - bitcoin/bitcoin#18443 - Assorted small changes from: - bitcoin/bitcoin#9233 - bitcoin/bitcoin#10483 - bitcoin/bitcoin#10645 - bitcoin/bitcoin#10969 - bitcoin/bitcoin#11351 - bitcoin/bitcoin#19111 - Excludes change to `src/rpc/server.cpp` - bitcoin/bitcoin#9804 - Only the commit for `src/key.cpp` - bitcoin/bitcoin#9598
…geAllocator::AllocateLocked Summary: PR description: > mmap(...) returns MAP_FAILED ((void *) -1) in case of allocation failure. > > PosixLockedPageAllocator::AllocateLocked(...) did not check for allocation failures prior to this PR. > > Instead the invalid memory address (void *) -1 (0xffffffffffffffff) was passed to the caller as if it was a valid address. > > After some operations the address is wrapped around from 0xffffffffffffffff to 0x00000003ffdf (0xffffffffffffffff + 262112 == 0x00000003ffdf); > > The resulting address 0x00000003ffdf is then written to. > > Before this patch (with failing mmap call): > > ``` > $ src/bitcoind > … > 2019-01-06T16:28:14Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation > 2019-01-06T16:28:14Z Using RdRand as an additional entropy source > Segmentation fault (core dumped) > ``` > > After this patch (with failing mmap call): > ``` > $ src/bitcoind > … > 2019-01-06T15:50:18Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation > 2019-01-06T15:50:18Z Using RdRand as an additional entropy source > 2019-01-06T15:50:18Z > > ************************ > EXCEPTION: St9bad_alloc > std::bad_alloc > bitcoin in AppInit() > > > > ************************ > EXCEPTION: St9bad_alloc > std::bad_alloc > bitcoin in AppInit() > > 2019-01-06T15:50:18Z Shutdown: In progress... > 2019-01-06T15:50:18Z Shutdown: done > ``` This is a backport of Core [[bitcoin/bitcoin#15117 | PR15117]] Test Plan: `ninja all check-all` Reviewers: #bitcoin_abc, majcosta Reviewed By: #bitcoin_abc, majcosta Differential Revision: https://reviews.bitcoinabc.org/D8354
…(...) in PosixLockedPageAllocator::AllocateLocked ca126d4 Fix out-of-bounds write in case of failing mmap(...) in PosixLockedPageAllocator::AllocateLocked (practicalswift) Pull request description: `mmap(...)` returns `MAP_FAILED` (`(void *) -1`) in case of allocation failure. `PosixLockedPageAllocator::AllocateLocked(...)` did not check for allocation failures prior to this PR. Instead the invalid memory address `(void *) -1` (`0xffffffffffffffff`) was passed to the caller as if it was a valid address. After some operations the address is wrapped around from `0xffffffffffffffff` to `0x00000003ffdf` (`0xffffffffffffffff + 262112 == 0x00000003ffdf`); The resulting address `0x00000003ffdf` is then written to. Before this patch (with failing `mmap` call): ``` $ src/bitcoind … 2019-01-06T16:28:14Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation 2019-01-06T16:28:14Z Using RdRand as an additional entropy source Segmentation fault (core dumped) ``` Before this patch (under `valgrind` with failing `mmap` call): ``` $ valgrind src/bitcoind … 2019-01-06T16:28:51Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation ==17812== Invalid write of size 1 ==17812== at 0x500B7E: void __gnu_cxx::new_allocator<unsigned char>::construct<unsigned char>(unsigned char*) (new_allocator.h:136) ==17812== by 0x500B52: _ZNSt16allocator_traitsI16secure_allocatorIhEE12_S_constructIhJEEENSt9enable_ifIXsr6__and_INS2_18__construct_helperIT_JDpT0_EE4typeEEE5valueEvE4typeERS1_PS6_DpOS7_ (alloc_traits.h:243) ==17812== by 0x500B22: _ZNSt16allocator_traitsI16secure_allocatorIhEE9constructIhJEEEDTcl12_S_constructfp_fp0_spclsr3stdE7forwardIT0_Efp1_EEERS1_PT_DpOS4_ (alloc_traits.h:344) ==17812== by 0x500982: unsigned char* std::__uninitialized_default_n_a<unsigned char*, unsigned long, secure_allocator<unsigned char> >(unsigned char*, unsigned long, secure_allocator<unsigned char>&) (stl_uninitialized.h:631) ==17812== by 0x60BFC2: std::vector<unsigned char, secure_allocator<unsigned char> >::_M_default_initialize(unsigned long) (stl_vector.h:1347) ==17812== by 0x60BD86: std::vector<unsigned char, secure_allocator<unsigned char> >::vector(unsigned long, secure_allocator<unsigned char> const&) (stl_vector.h:285) ==17812== by 0x60BB55: ECC_Start() (key.cpp:351) ==17812== by 0x16AC90: AppInitSanityChecks() (init.cpp:1162) ==17812== by 0x15BAC9: AppInit(int, char**) (bitcoind.cpp:138) ==17812== by 0x15B6C8: main (bitcoind.cpp:201) ==17812== Address 0x3ffdf is not stack'd, malloc'd or (recently) free'd … Segmentation fault (core dumped) ``` After this patch (with failing `mmap` call): ``` $ src/bitcoind … 2019-01-06T15:50:18Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation 2019-01-06T15:50:18Z Using RdRand as an additional entropy source 2019-01-06T15:50:18Z ************************ EXCEPTION: St9bad_alloc std::bad_alloc bitcoin in AppInit() ************************ EXCEPTION: St9bad_alloc std::bad_alloc bitcoin in AppInit() 2019-01-06T15:50:18Z Shutdown: In progress... 2019-01-06T15:50:18Z Shutdown: done ``` To simulate the failing `mmap` call apply the following to `master`: ```diff diff --git a/src/support/lockedpool.cpp b/src/support/lockedpool.cpp index 8d577cf..ce79e569b 100644 --- a/src/support/lockedpool.cpp +++ b/src/support/lockedpool.cpp @@ -247,7 +247,8 @@ void *PosixLockedPageAllocator::AllocateLocked(size_t len, bool *lockingSuccess) { void *addr; len = align_up(len, page_size); - addr = mmap(nullptr, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); + // addr = mmap(nullptr, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); + addr = MAP_FAILED; if (addr) { *lockingSuccess = mlock(addr, len) == 0; } ``` Tree-SHA512: 66947f5fc0fbb19afb3e1edbd51df07df9d16b77018cff3d48d30f378a53d6a0dc62bc36622b3966b7e374e61edbcca114ef4ac8ae8d725022c1a597edcbf7c7
9295568 Fix segfault in allocator_tests/arena_tests (Jeffrey Czyz) f7d88f5 Define ARENA_DEBUG in GA test runs (furszy) cae2e13 Fix compilation errors in support/lockedpool.cpp (Jeffrey Czyz) d19359f fix nits: variable naming, typos (Martin Ankerl) 5cbd967 Use best-fit strategy in Arena, now O(log(n)) instead O(n) (Martin Ankerl) 1ccd62e Fix out-of-bounds write in case of failing mmap(...) in PosixLockedPageAllocator::AllocateLocked (practicalswift) 7ad7157 LockedPool: avoid quadratic-time allocation (Kaz Wesley) 39e1b3b LockedPool: fix explosion for illegal-sized alloc (Kaz Wesley) adb1f1c LockedPool: test handling of invalid allocations (Kaz Wesley) 626b865 Do not shadow variable, use deprecated MAP_ANON if MAP_ANONYMOUS is not defined. (Pavel Janík) Pull request description: Follow up to #2276. Completing the locked memory manager todo list. Back porting: * bitcoin#9063. * bitcoin#9070. * bitcoin#12048. * bitcoin#15117. * bitcoin#16161. ACKs for top commit: random-zebra: ACK 9295568 Fuzzbawls: ACK 9295568 Tree-SHA512: 0d14c7e8231386ee32f1fefac08de9ee278c02d6b0e5172c055851d33c803b0be3398fc5173457e539179712a13d4736033a926f318e576789b4d57b67eb0596
…(...) in PosixLockedPageAllocator::AllocateLocked ca126d4 Fix out-of-bounds write in case of failing mmap(...) in PosixLockedPageAllocator::AllocateLocked (practicalswift) Pull request description: `mmap(...)` returns `MAP_FAILED` (`(void *) -1`) in case of allocation failure. `PosixLockedPageAllocator::AllocateLocked(...)` did not check for allocation failures prior to this PR. Instead the invalid memory address `(void *) -1` (`0xffffffffffffffff`) was passed to the caller as if it was a valid address. After some operations the address is wrapped around from `0xffffffffffffffff` to `0x00000003ffdf` (`0xffffffffffffffff + 262112 == 0x00000003ffdf`); The resulting address `0x00000003ffdf` is then written to. Before this patch (with failing `mmap` call): ``` $ src/bitcoind … 2019-01-06T16:28:14Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation 2019-01-06T16:28:14Z Using RdRand as an additional entropy source Segmentation fault (core dumped) ``` Before this patch (under `valgrind` with failing `mmap` call): ``` $ valgrind src/bitcoind … 2019-01-06T16:28:51Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation ==17812== Invalid write of size 1 ==17812== at 0x500B7E: void __gnu_cxx::new_allocator<unsigned char>::construct<unsigned char>(unsigned char*) (new_allocator.h:136) ==17812== by 0x500B52: _ZNSt16allocator_traitsI16secure_allocatorIhEE12_S_constructIhJEEENSt9enable_ifIXsr6__and_INS2_18__construct_helperIT_JDpT0_EE4typeEEE5valueEvE4typeERS1_PS6_DpOS7_ (alloc_traits.h:243) ==17812== by 0x500B22: _ZNSt16allocator_traitsI16secure_allocatorIhEE9constructIhJEEEDTcl12_S_constructfp_fp0_spclsr3stdE7forwardIT0_Efp1_EEERS1_PT_DpOS4_ (alloc_traits.h:344) ==17812== by 0x500982: unsigned char* std::__uninitialized_default_n_a<unsigned char*, unsigned long, secure_allocator<unsigned char> >(unsigned char*, unsigned long, secure_allocator<unsigned char>&) (stl_uninitialized.h:631) ==17812== by 0x60BFC2: std::vector<unsigned char, secure_allocator<unsigned char> >::_M_default_initialize(unsigned long) (stl_vector.h:1347) ==17812== by 0x60BD86: std::vector<unsigned char, secure_allocator<unsigned char> >::vector(unsigned long, secure_allocator<unsigned char> const&) (stl_vector.h:285) ==17812== by 0x60BB55: ECC_Start() (key.cpp:351) ==17812== by 0x16AC90: AppInitSanityChecks() (init.cpp:1162) ==17812== by 0x15BAC9: AppInit(int, char**) (bitcoind.cpp:138) ==17812== by 0x15B6C8: main (bitcoind.cpp:201) ==17812== Address 0x3ffdf is not stack'd, malloc'd or (recently) free'd … Segmentation fault (core dumped) ``` After this patch (with failing `mmap` call): ``` $ src/bitcoind … 2019-01-06T15:50:18Z Using the 'sse4(1way),sse41(4way)' SHA256 implementation 2019-01-06T15:50:18Z Using RdRand as an additional entropy source 2019-01-06T15:50:18Z ************************ EXCEPTION: St9bad_alloc std::bad_alloc bitcoin in AppInit() ************************ EXCEPTION: St9bad_alloc std::bad_alloc bitcoin in AppInit() 2019-01-06T15:50:18Z Shutdown: In progress... 2019-01-06T15:50:18Z Shutdown: done ``` To simulate the failing `mmap` call apply the following to `master`: ```diff diff --git a/src/support/lockedpool.cpp b/src/support/lockedpool.cpp index 8d577cf..ce79e569b 100644 --- a/src/support/lockedpool.cpp +++ b/src/support/lockedpool.cpp @@ -247,7 +247,8 @@ void *PosixLockedPageAllocator::AllocateLocked(size_t len, bool *lockingSuccess) { void *addr; len = align_up(len, page_size); - addr = mmap(nullptr, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); + // addr = mmap(nullptr, len, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); + addr = MAP_FAILED; if (addr) { *lockingSuccess = mlock(addr, len) == 0; } ``` Tree-SHA512: 66947f5fc0fbb19afb3e1edbd51df07df9d16b77018cff3d48d30f378a53d6a0dc62bc36622b3966b7e374e61edbcca114ef4ac8ae8d725022c1a597edcbf7c7
mmap(...)returnsMAP_FAILED((void *) -1) in case of allocation failure.PosixLockedPageAllocator::AllocateLocked(...)did not check for allocation failures prior to this PR.Instead the invalid memory address
(void *) -1(0xffffffffffffffff) was passed to the caller as if it was a valid address.After some operations the address is wrapped around from
0xffffffffffffffffto0x00000003ffdf(0xffffffffffffffff + 262112 == 0x00000003ffdf);The resulting address
0x00000003ffdfis then written to.Before this patch (with failing
mmapcall):Before this patch (under
valgrindwith failingmmapcall):After this patch (with failing
mmapcall):To simulate the failing
mmapcall apply the following tomaster: