| Message ID | 20241210-bpf-fix-uprobe-uaf-v3-1-ce50ae2a2f0f@google.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [bpf,v3] bpf: Fix theoretical prog_array UAF in __uprobe_perf_func() | expand |
On Tue, Dec 10, 2024 at 7:34 AM Jann Horn <jannh@google.com> wrote: > > Currently, the pointer stored in call->prog_array is loaded in > __uprobe_perf_func(), with no RCU annotation and no immediately visible > RCU protection, so it looks as if the loaded pointer can immediately be > dangling. > Later, bpf_prog_run_array_uprobe() starts a RCU-trace read-side critical > section, but this is too late. It then uses rcu_dereference_check(), but > this use of rcu_dereference_check() does not actually dereference anything. > > Fix it by aligning the semantics to bpf_prog_run_array(): Let the caller > provide rcu_read_lock_trace() protection and then load call->prog_array > with rcu_dereference_check(). > > This issue seems to be theoretical: I don't know of any way to reach this > code without having handle_swbp() further up the stack, which is already > holding a rcu_read_lock_trace() lock, so where we take > rcu_read_lock_trace() in __uprobe_perf_func()/bpf_prog_run_array_uprobe() > doesn't actually have any effect. > > Fixes: 8c7dcb84e3b7 ("bpf: implement sleepable uprobes by chaining gps") > Suggested-by: Andrii Nakryiko <andrii@kernel.org> > Signed-off-by: Jann Horn <jannh@google.com> > --- > Changes in v3: > - align semantics with bpf_prog_run_array() > - correct commit message: the issue is theoretical > - remove stable CC > - Link to v2: https://lore.kernel.org/r/20241206-bpf-fix-uprobe-uaf-v2-1-4c75c54fe424@google.com > > Changes in v2: > - remove diff chunk in patch notes that confuses git > - Link to v1: https://lore.kernel.org/r/20241206-bpf-fix-uprobe-uaf-v1-1-6869c8a17258@google.com > --- > include/linux/bpf.h | 11 +++-------- > kernel/trace/trace_uprobe.c | 6 +++++- > 2 files changed, 8 insertions(+), 9 deletions(-) > > diff --git a/include/linux/bpf.h b/include/linux/bpf.h > index eaee2a819f4c150a34a7b1075584711609682e4c..7fe5cf181511d543b1b100028db94ebb2a44da5d 100644 > --- a/include/linux/bpf.h > +++ b/include/linux/bpf.h > @@ -2193,26 +2193,22 @@ bpf_prog_run_array(const struct bpf_prog_array *array, > * rcu-protected dynamically sized maps. > */ > static __always_inline u32 > -bpf_prog_run_array_uprobe(const struct bpf_prog_array __rcu *array_rcu, > +bpf_prog_run_array_uprobe(const struct bpf_prog_array *array, > const void *ctx, bpf_prog_run_fn run_prog) > { > const struct bpf_prog_array_item *item; > const struct bpf_prog *prog; > - const struct bpf_prog_array *array; > struct bpf_run_ctx *old_run_ctx; > struct bpf_trace_run_ctx run_ctx; > u32 ret = 1; > > might_fault(); > + RCU_LOCKDEP_WARN(!rcu_read_lock_trace_held(), "no rcu lock held"); > > - rcu_read_lock_trace(); > migrate_disable(); > > run_ctx.is_uprobe = true; > > - array = rcu_dereference_check(array_rcu, rcu_read_lock_trace_held()); > - if (unlikely(!array)) > - goto out; I think we should keep this unlikely(NULL) check, bpf_prog_run_array() has it and see bpf_prog_array_valid() comment below pw-bot: cr > old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); > item = &array->items[0]; > while ((prog = READ_ONCE(item->prog))) { > @@ -2227,9 +2223,8 @@ bpf_prog_run_array_uprobe(const struct bpf_prog_array __rcu *array_rcu, > rcu_read_unlock(); > } > bpf_reset_run_ctx(old_run_ctx); > -out: > + > migrate_enable(); > - rcu_read_unlock_trace(); > return ret; > } > > diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c > index fed382b7881b82ee3c334ea77860cce77581a74d..4875e7f5de3db249af34c539c079fbedd38f4107 100644 > --- a/kernel/trace/trace_uprobe.c > +++ b/kernel/trace/trace_uprobe.c > @@ -1402,9 +1402,13 @@ static void __uprobe_perf_func(struct trace_uprobe *tu, > > #ifdef CONFIG_BPF_EVENTS > if (bpf_prog_array_valid(call)) { bpf_prog_array_valid() explicitly calls out that it's just an opportunistic check and bpf_prog_run_array*() should double check for NULL > + const struct bpf_prog_array *array; > u32 ret; > > - ret = bpf_prog_run_array_uprobe(call->prog_array, regs, bpf_prog_run); > + rcu_read_lock_trace(); > + array = rcu_dereference_check(call->prog_array, rcu_read_lock_trace_held()); > + ret = bpf_prog_run_array_uprobe(array, regs, bpf_prog_run); > + rcu_read_unlock_trace(); > if (!ret) > return; > } > > --- > base-commit: 509df676c2d79c985ec2eaa3e3a3bbe557645861 > change-id: 20241206-bpf-fix-uprobe-uaf-53d928bab3d0 > > -- > Jann Horn <jannh@google.com> >
On Tue, Dec 10, 2024 at 6:52 PM Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote: > On Tue, Dec 10, 2024 at 7:34 AM Jann Horn <jannh@google.com> wrote: > > Currently, the pointer stored in call->prog_array is loaded in > > __uprobe_perf_func(), with no RCU annotation and no immediately visible > > RCU protection, so it looks as if the loaded pointer can immediately be > > dangling. > > Later, bpf_prog_run_array_uprobe() starts a RCU-trace read-side critical > > section, but this is too late. It then uses rcu_dereference_check(), but > > this use of rcu_dereference_check() does not actually dereference anything. > > > > Fix it by aligning the semantics to bpf_prog_run_array(): Let the caller > > provide rcu_read_lock_trace() protection and then load call->prog_array > > with rcu_dereference_check(). > > > > This issue seems to be theoretical: I don't know of any way to reach this > > code without having handle_swbp() further up the stack, which is already > > holding a rcu_read_lock_trace() lock, so where we take > > rcu_read_lock_trace() in __uprobe_perf_func()/bpf_prog_run_array_uprobe() > > doesn't actually have any effect. > > > > Fixes: 8c7dcb84e3b7 ("bpf: implement sleepable uprobes by chaining gps") > > Suggested-by: Andrii Nakryiko <andrii@kernel.org> > > Signed-off-by: Jann Horn <jannh@google.com> > > --- > > Changes in v3: > > - align semantics with bpf_prog_run_array() > > - correct commit message: the issue is theoretical > > - remove stable CC > > - Link to v2: https://lore.kernel.org/r/20241206-bpf-fix-uprobe-uaf-v2-1-4c75c54fe424@google.com > > > > Changes in v2: > > - remove diff chunk in patch notes that confuses git > > - Link to v1: https://lore.kernel.org/r/20241206-bpf-fix-uprobe-uaf-v1-1-6869c8a17258@google.com > > --- > > include/linux/bpf.h | 11 +++-------- > > kernel/trace/trace_uprobe.c | 6 +++++- > > 2 files changed, 8 insertions(+), 9 deletions(-) > > > > diff --git a/include/linux/bpf.h b/include/linux/bpf.h > > index eaee2a819f4c150a34a7b1075584711609682e4c..7fe5cf181511d543b1b100028db94ebb2a44da5d 100644 > > --- a/include/linux/bpf.h > > +++ b/include/linux/bpf.h > > @@ -2193,26 +2193,22 @@ bpf_prog_run_array(const struct bpf_prog_array *array, > > * rcu-protected dynamically sized maps. > > */ > > static __always_inline u32 > > -bpf_prog_run_array_uprobe(const struct bpf_prog_array __rcu *array_rcu, > > +bpf_prog_run_array_uprobe(const struct bpf_prog_array *array, > > const void *ctx, bpf_prog_run_fn run_prog) > > { > > const struct bpf_prog_array_item *item; > > const struct bpf_prog *prog; > > - const struct bpf_prog_array *array; > > struct bpf_run_ctx *old_run_ctx; > > struct bpf_trace_run_ctx run_ctx; > > u32 ret = 1; > > > > might_fault(); > > + RCU_LOCKDEP_WARN(!rcu_read_lock_trace_held(), "no rcu lock held"); > > > > - rcu_read_lock_trace(); > > migrate_disable(); > > > > run_ctx.is_uprobe = true; > > > > - array = rcu_dereference_check(array_rcu, rcu_read_lock_trace_held()); > > - if (unlikely(!array)) > > - goto out; > > I think we should keep this unlikely(NULL) check, bpf_prog_run_array() > has it and see bpf_prog_array_valid() comment below Whoops, yeah, I removed it here at some point while moving the dereference around and then forgot to re-add it; will fix.
diff --git a/include/linux/bpf.h b/include/linux/bpf.h index eaee2a819f4c150a34a7b1075584711609682e4c..7fe5cf181511d543b1b100028db94ebb2a44da5d 100644 --- a/include/linux/bpf.h +++ b/include/linux/bpf.h @@ -2193,26 +2193,22 @@ bpf_prog_run_array(const struct bpf_prog_array *array, * rcu-protected dynamically sized maps. */ static __always_inline u32 -bpf_prog_run_array_uprobe(const struct bpf_prog_array __rcu *array_rcu, +bpf_prog_run_array_uprobe(const struct bpf_prog_array *array, const void *ctx, bpf_prog_run_fn run_prog) { const struct bpf_prog_array_item *item; const struct bpf_prog *prog; - const struct bpf_prog_array *array; struct bpf_run_ctx *old_run_ctx; struct bpf_trace_run_ctx run_ctx; u32 ret = 1; might_fault(); + RCU_LOCKDEP_WARN(!rcu_read_lock_trace_held(), "no rcu lock held"); - rcu_read_lock_trace(); migrate_disable(); run_ctx.is_uprobe = true; - array = rcu_dereference_check(array_rcu, rcu_read_lock_trace_held()); - if (unlikely(!array)) - goto out; old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); item = &array->items[0]; while ((prog = READ_ONCE(item->prog))) { @@ -2227,9 +2223,8 @@ bpf_prog_run_array_uprobe(const struct bpf_prog_array __rcu *array_rcu, rcu_read_unlock(); } bpf_reset_run_ctx(old_run_ctx); -out: + migrate_enable(); - rcu_read_unlock_trace(); return ret; } diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c index fed382b7881b82ee3c334ea77860cce77581a74d..4875e7f5de3db249af34c539c079fbedd38f4107 100644 --- a/kernel/trace/trace_uprobe.c +++ b/kernel/trace/trace_uprobe.c @@ -1402,9 +1402,13 @@ static void __uprobe_perf_func(struct trace_uprobe *tu, #ifdef CONFIG_BPF_EVENTS if (bpf_prog_array_valid(call)) { + const struct bpf_prog_array *array; u32 ret; - ret = bpf_prog_run_array_uprobe(call->prog_array, regs, bpf_prog_run); + rcu_read_lock_trace(); + array = rcu_dereference_check(call->prog_array, rcu_read_lock_trace_held()); + ret = bpf_prog_run_array_uprobe(array, regs, bpf_prog_run); + rcu_read_unlock_trace(); if (!ret) return; }
Currently, the pointer stored in call->prog_array is loaded in __uprobe_perf_func(), with no RCU annotation and no immediately visible RCU protection, so it looks as if the loaded pointer can immediately be dangling. Later, bpf_prog_run_array_uprobe() starts a RCU-trace read-side critical section, but this is too late. It then uses rcu_dereference_check(), but this use of rcu_dereference_check() does not actually dereference anything. Fix it by aligning the semantics to bpf_prog_run_array(): Let the caller provide rcu_read_lock_trace() protection and then load call->prog_array with rcu_dereference_check(). This issue seems to be theoretical: I don't know of any way to reach this code without having handle_swbp() further up the stack, which is already holding a rcu_read_lock_trace() lock, so where we take rcu_read_lock_trace() in __uprobe_perf_func()/bpf_prog_run_array_uprobe() doesn't actually have any effect. Fixes: 8c7dcb84e3b7 ("bpf: implement sleepable uprobes by chaining gps") Suggested-by: Andrii Nakryiko <andrii@kernel.org> Signed-off-by: Jann Horn <jannh@google.com> --- Changes in v3: - align semantics with bpf_prog_run_array() - correct commit message: the issue is theoretical - remove stable CC - Link to v2: https://lore.kernel.org/r/20241206-bpf-fix-uprobe-uaf-v2-1-4c75c54fe424@google.com Changes in v2: - remove diff chunk in patch notes that confuses git - Link to v1: https://lore.kernel.org/r/20241206-bpf-fix-uprobe-uaf-v1-1-6869c8a17258@google.com --- include/linux/bpf.h | 11 +++-------- kernel/trace/trace_uprobe.c | 6 +++++- 2 files changed, 8 insertions(+), 9 deletions(-) --- base-commit: 509df676c2d79c985ec2eaa3e3a3bbe557645861 change-id: 20241206-bpf-fix-uprobe-uaf-53d928bab3d0