From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions When using eBPF to probe CS CALL instructions within a function, a crash can be triggered. The eBPF tool probes offset 257 of the __hrtimer_run_queues() function: <__hrtimer_run_queues+249>: nopl 0x0(%rax,%rax,1) <__hrtimer_run_queues+254>: mov %r14,%rdi <__hrtimer_run_queues+257>: cs call <__x86_indirect_thunk_r12> <__hrtimer_run_queues+263>: mov %eax,%r12d <__hrtimer_run_queues+266>: xchg %ax,%ax <__hrtimer_run_queues+268>: mov %r13,%rdi Which triggers this crash: BUG: unable to handle page fault for address: 00000000000f41c9 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P RIP: 0010:__hrtimer_run_queues+0x106/0x230 Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is at the 6th byte of the above CS CALL instruction. Since the CS CALL instruction occupies 6 bytes, the exception occurred in the middle of that call instruction. The root cause is that when using eBPF tools to probe in the middle of a function, a kprobe with INT3 is used as the underlying implementation. During single-step emulation of the original CALL instruction, int3_emulate_call() assumes that the probed CALL instruction is 5 bytes long. However, the actual CS-prefixed CALL instruction occupies 6 bytes, so it constructs an incorrect exception return address. When the CPU returns from the kprobe handler, the next instruction to be executed is at the address of the last byte of that CS CALL instruction. Coincidentally, starting from that address, the CPU fetches and decodes a completely different instruction, which ultimately triggers a kernel crash. Fix the issue by using the actual instruction length obtained from the instruction decoder when constructing the exception return address, rather than relying on the hardcoded CALL_INSN_SIZE macro. [ mingo: Refined the changelog ] Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step") Suggested-by: Masami Hiramatsu (Google) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- fs/ceph/ioctl.h | 101 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 101 insertions(+) create mode 100644 fs/ceph/ioctl.h (limited to 'fs/ceph/ioctl.h') diff --git a/fs/ceph/ioctl.h b/fs/ceph/ioctl.h new file mode 100644 index 000000000..51f7f1d39 --- /dev/null +++ b/fs/ceph/ioctl.h @@ -0,0 +1,101 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef FS_CEPH_IOCTL_H +#define FS_CEPH_IOCTL_H + +#include +#include + +#define CEPH_IOCTL_MAGIC 0x97 + +/* + * CEPH_IOC_GET_LAYOUT - get file layout or dir layout policy + * CEPH_IOC_SET_LAYOUT - set file layout + * CEPH_IOC_SET_LAYOUT_POLICY - set dir layout policy + * + * The file layout specifies how file data is striped over objects in + * the distributed object store, which object pool they belong to (if + * it differs from the default), and an optional 'preferred osd' to + * store them on. + * + * Files get a new layout based on the policy set on the containing + * directory or one of its ancestors. The GET_LAYOUT ioctl will let + * you examine the layout for a file or the policy on a directory. + * + * SET_LAYOUT will let you set a layout on a newly created file. This + * only works immediately after the file is created and before any + * data is written to it. + * + * SET_LAYOUT_POLICY will let you set a layout policy (default layout) + * on a directory that will apply to any new files created in that + * directory (or any child directory that doesn't specify a layout of + * its own). + */ + +/* use u64 to align sanely on all archs */ +struct ceph_ioctl_layout { + __u64 stripe_unit, stripe_count, object_size; + __u64 data_pool; + + /* obsolete. new values ignored, always return -1 */ + __s64 preferred_osd; +}; + +#define CEPH_IOC_GET_LAYOUT _IOR(CEPH_IOCTL_MAGIC, 1, \ + struct ceph_ioctl_layout) +#define CEPH_IOC_SET_LAYOUT _IOW(CEPH_IOCTL_MAGIC, 2, \ + struct ceph_ioctl_layout) +#define CEPH_IOC_SET_LAYOUT_POLICY _IOW(CEPH_IOCTL_MAGIC, 5, \ + struct ceph_ioctl_layout) + +/* + * CEPH_IOC_GET_DATALOC - get location of file data in the cluster + * + * Extract identity, address of the OSD and object storing a given + * file offset. + */ +struct ceph_ioctl_dataloc { + __u64 file_offset; /* in+out: file offset */ + __u64 object_offset; /* out: offset in object */ + __u64 object_no; /* out: object # */ + __u64 object_size; /* out: object size */ + char object_name[64]; /* out: object name */ + __u64 block_offset; /* out: offset in block */ + __u64 block_size; /* out: block length */ + __s64 osd; /* out: osd # */ + struct sockaddr_storage osd_addr; /* out: osd address */ +}; + +#define CEPH_IOC_GET_DATALOC _IOWR(CEPH_IOCTL_MAGIC, 3, \ + struct ceph_ioctl_dataloc) + +/* + * CEPH_IOC_LAZYIO - relax consistency + * + * Normally Ceph switches to synchronous IO when multiple clients have + * the file open (and or more for write). Reads and writes bypass the + * page cache and go directly to the OSD. Setting this flag on a file + * descriptor will allow buffered IO for this file in cases where the + * application knows it won't interfere with other nodes (or doesn't + * care). + */ +#define CEPH_IOC_LAZYIO _IO(CEPH_IOCTL_MAGIC, 4) + +/* + * CEPH_IOC_SYNCIO - force synchronous IO + * + * This ioctl sets a file flag that forces the synchronous IO that + * bypasses the page cache, even if it is not necessary. This is + * essentially the opposite behavior of IOC_LAZYIO. This forces the + * same read/write path as a file opened by multiple clients when one + * or more of those clients is opened for write. + * + * Note that this type of sync IO takes a different path than a file + * opened with O_SYNC/D_SYNC (writes hit the page cache and are + * immediately flushed on page boundaries). It is very similar to + * O_DIRECT (writes bypass the page cache) excep that O_DIRECT writes + * are not copied (user page must remain stable) and O_DIRECT writes + * have alignment restrictions (on the buffer and file offset). + */ +#define CEPH_IOC_SYNCIO _IO(CEPH_IOCTL_MAGIC, 5) + +#endif -- cgit v1.3.1