From ede95a63b5e84ddeea6b0c473b36ab8bfd8c6ce3 Mon Sep 17 00:00:00 2001 From: Daniel Borkmann Date: Tue, 23 Oct 2018 01:11:04 +0200 Subject: bpf: add bpf_jit_limit knob to restrict unpriv allocations Rick reported that the BPF JIT could potentially fill the entire module space with BPF programs from unprivileged users which would prevent later attempts to load normal kernel modules or privileged BPF programs, for example. If JIT was enabled but unsuccessful to generate the image, then before commit 290af86629b2 ("bpf: introduce BPF_JIT_ALWAYS_ON config") we would always fall back to the BPF interpreter. Nowadays in the case where the CONFIG_BPF_JIT_ALWAYS_ON could be set, then the load will abort with a failure since the BPF interpreter was compiled out. Add a global limit and enforce it for unprivileged users such that in case of BPF interpreter compiled out we fail once the limit has been reached or we fall back to BPF interpreter earlier w/o using module mem if latter was compiled in. In a next step, fair share among unprivileged users can be resolved in particular for the case where we would fail hard once limit is reached. Fixes: 290af86629b2 ("bpf: introduce BPF_JIT_ALWAYS_ON config") Fixes: 0a14842f5a3c ("net: filter: Just In Time compiler for x86-64") Co-Developed-by: Rick Edgecombe Signed-off-by: Daniel Borkmann Acked-by: Alexei Starovoitov Cc: Eric Dumazet Cc: Jann Horn Cc: Kees Cook Cc: LKML Signed-off-by: Alexei Starovoitov --- kernel/bpf/core.c | 49 ++++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 46 insertions(+), 3 deletions(-) (limited to 'kernel/bpf/core.c') diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c index 7c7eeea8cffc..6377225b2082 100644 --- a/kernel/bpf/core.c +++ b/kernel/bpf/core.c @@ -365,10 +365,13 @@ void bpf_prog_kallsyms_del_all(struct bpf_prog *fp) } #ifdef CONFIG_BPF_JIT +# define BPF_JIT_LIMIT_DEFAULT (PAGE_SIZE * 40000) + /* All BPF JIT sysctl knobs here. */ int bpf_jit_enable __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_ALWAYS_ON); int bpf_jit_harden __read_mostly; int bpf_jit_kallsyms __read_mostly; +int bpf_jit_limit __read_mostly = BPF_JIT_LIMIT_DEFAULT; static __always_inline void bpf_get_prog_addr_region(const struct bpf_prog *prog, @@ -577,27 +580,64 @@ int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, return ret; } +static atomic_long_t bpf_jit_current; + +#if defined(MODULES_VADDR) +static int __init bpf_jit_charge_init(void) +{ + /* Only used as heuristic here to derive limit. */ + bpf_jit_limit = min_t(u64, round_up((MODULES_END - MODULES_VADDR) >> 2, + PAGE_SIZE), INT_MAX); + return 0; +} +pure_initcall(bpf_jit_charge_init); +#endif + +static int bpf_jit_charge_modmem(u32 pages) +{ + if (atomic_long_add_return(pages, &bpf_jit_current) > + (bpf_jit_limit >> PAGE_SHIFT)) { + if (!capable(CAP_SYS_ADMIN)) { + atomic_long_sub(pages, &bpf_jit_current); + return -EPERM; + } + } + + return 0; +} + +static void bpf_jit_uncharge_modmem(u32 pages) +{ + atomic_long_sub(pages, &bpf_jit_current); +} + struct bpf_binary_header * bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, unsigned int alignment, bpf_jit_fill_hole_t bpf_fill_ill_insns) { struct bpf_binary_header *hdr; - unsigned int size, hole, start; + u32 size, hole, start, pages; /* Most of BPF filters are really small, but if some of them * fill a page, allow at least 128 extra bytes to insert a * random section of illegal instructions. */ size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE); + pages = size / PAGE_SIZE; + + if (bpf_jit_charge_modmem(pages)) + return NULL; hdr = module_alloc(size); - if (hdr == NULL) + if (!hdr) { + bpf_jit_uncharge_modmem(pages); return NULL; + } /* Fill space with illegal/arch-dep instructions. */ bpf_fill_ill_insns(hdr, size); - hdr->pages = size / PAGE_SIZE; + hdr->pages = pages; hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)), PAGE_SIZE - sizeof(*hdr)); start = (get_random_int() % hole) & ~(alignment - 1); @@ -610,7 +650,10 @@ bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr, void bpf_jit_binary_free(struct bpf_binary_header *hdr) { + u32 pages = hdr->pages; + module_memfree(hdr); + bpf_jit_uncharge_modmem(pages); } /* This symbol is only overridden by archs that have different -- cgit v1.2.3 From df0734702a7cbba49d6765bd5ba069340bf9c5db Mon Sep 17 00:00:00 2001 From: Song Liu Date: Fri, 2 Nov 2018 10:16:15 -0700 Subject: bpf: show real jited prog address in /proc/kallsyms Currently, /proc/kallsyms shows page address of jited bpf program. The main reason here is to not expose randomized start address. However, This is not ideal for detailed profiling (find hot instructions from stack traces). This patch replaces the page address with real prog start address. This change is OK because these addresses are still protected by sysctl kptr_restrict (see kallsyms_show_value()), and only programs loaded by root are added to kallsyms (see bpf_prog_kallsyms_add()). Signed-off-by: Song Liu Signed-off-by: Daniel Borkmann --- kernel/bpf/core.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) (limited to 'kernel/bpf/core.c') diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c index 6377225b2082..1a796e0799ec 100644 --- a/kernel/bpf/core.c +++ b/kernel/bpf/core.c @@ -553,7 +553,6 @@ bool is_bpf_text_address(unsigned long addr) int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, char *sym) { - unsigned long symbol_start, symbol_end; struct bpf_prog_aux *aux; unsigned int it = 0; int ret = -ERANGE; @@ -566,10 +565,9 @@ int bpf_get_kallsym(unsigned int symnum, unsigned long *value, char *type, if (it++ != symnum) continue; - bpf_get_prog_addr_region(aux->prog, &symbol_start, &symbol_end); bpf_get_prog_name(aux->prog, sym); - *value = symbol_start; + *value = (unsigned long)aux->prog->bpf_func; *type = BPF_SYM_ELF_TYPE; ret = 0; -- cgit v1.2.3 From e2c95a61656d29ceaac97b6a975c8a1f26e26f15 Mon Sep 17 00:00:00 2001 From: Daniel Borkmann Date: Mon, 26 Nov 2018 14:05:38 +0100 Subject: bpf, ppc64: generalize fetching subprog into bpf_jit_get_func_addr Make fetching of the BPF call address from ppc64 JIT generic. ppc64 was using a slightly different variant rather than through the insns' imm field encoding as the target address would not fit into that space. Therefore, the target subprog number was encoded into the insns' offset and fetched through fp->aux->func[off]->bpf_func instead. Given there are other JITs with this issue and the mechanism of fetching the address is JIT-generic, move it into the core as a helper instead. On the JIT side, we get information on whether the retrieved address is a fixed one, that is, not changing through JIT passes, or a dynamic one. For the former, JITs can optimize their imm emission because this doesn't change jump offsets throughout JIT process. Signed-off-by: Daniel Borkmann Reviewed-by: Sandipan Das Tested-by: Sandipan Das Signed-off-by: Alexei Starovoitov --- kernel/bpf/core.c | 34 ++++++++++++++++++++++++++++++++++ 1 file changed, 34 insertions(+) (limited to 'kernel/bpf/core.c') diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c index 1a796e0799ec..b1a3545d0ec8 100644 --- a/kernel/bpf/core.c +++ b/kernel/bpf/core.c @@ -672,6 +672,40 @@ void __weak bpf_jit_free(struct bpf_prog *fp) bpf_prog_unlock_free(fp); } +int bpf_jit_get_func_addr(const struct bpf_prog *prog, + const struct bpf_insn *insn, bool extra_pass, + u64 *func_addr, bool *func_addr_fixed) +{ + s16 off = insn->off; + s32 imm = insn->imm; + u8 *addr; + + *func_addr_fixed = insn->src_reg != BPF_PSEUDO_CALL; + if (!*func_addr_fixed) { + /* Place-holder address till the last pass has collected + * all addresses for JITed subprograms in which case we + * can pick them up from prog->aux. + */ + if (!extra_pass) + addr = NULL; + else if (prog->aux->func && + off >= 0 && off < prog->aux->func_cnt) + addr = (u8 *)prog->aux->func[off]->bpf_func; + else + return -EINVAL; + } else { + /* Address of a BPF helper call. Since part of the core + * kernel, it's always at a fixed location. __bpf_call_base + * and the helper with imm relative to it are both in core + * kernel. + */ + addr = (u8 *)__bpf_call_base + imm; + } + + *func_addr = (unsigned long)addr; + return 0; +} + static int bpf_jit_blind_insn(const struct bpf_insn *from, const struct bpf_insn *aux, struct bpf_insn *to_buff) -- cgit v1.2.3