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/verity/enable.c | 413 +++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 413 insertions(+) create mode 100644 fs/verity/enable.c (limited to 'fs/verity/enable.c') diff --git a/fs/verity/enable.c b/fs/verity/enable.c new file mode 100644 index 000000000..42dfed1ce --- /dev/null +++ b/fs/verity/enable.c @@ -0,0 +1,413 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Ioctl to enable verity on a file + * + * Copyright 2019 Google LLC + */ + +#include "fsverity_private.h" + +#include +#include +#include +#include + +struct block_buffer { + u32 filled; + bool is_root_hash; + u8 *data; +}; + +/* Hash a block, writing the result to the next level's pending block buffer. */ +static int hash_one_block(const struct merkle_tree_params *params, + struct block_buffer *cur) +{ + struct block_buffer *next = cur + 1; + + /* + * Safety check to prevent a buffer overflow in case of a filesystem bug + * that allows the file size to change despite deny_write_access(), or a + * bug in the Merkle tree logic itself + */ + if (WARN_ON_ONCE(next->is_root_hash && next->filled != 0)) + return -EINVAL; + + /* Zero-pad the block if it's shorter than the block size. */ + memset(&cur->data[cur->filled], 0, params->block_size - cur->filled); + + fsverity_hash_block(params, cur->data, &next->data[next->filled]); + next->filled += params->digest_size; + cur->filled = 0; + return 0; +} + +static int write_merkle_tree_block(struct file *file, const u8 *buf, + unsigned long index, + const struct merkle_tree_params *params) +{ + struct inode *inode = file_inode(file); + u64 pos = (u64)index << params->log_blocksize; + int err; + + err = inode->i_sb->s_vop->write_merkle_tree_block(file, buf, pos, + params->block_size); + if (err) + fsverity_err(inode, "Error %d writing Merkle tree block %lu", + err, index); + return err; +} + +/* + * Build the Merkle tree for the given file using the given parameters, and + * return the root hash in @root_hash. + * + * The tree is written to a filesystem-specific location as determined by the + * ->write_merkle_tree_block() method. However, the blocks that comprise the + * tree are the same for all filesystems. + */ +static int build_merkle_tree(struct file *filp, + const struct merkle_tree_params *params, + u8 *root_hash) +{ + struct inode *inode = file_inode(filp); + const u64 data_size = inode->i_size; + const int num_levels = params->num_levels; + struct block_buffer _buffers[1 + FS_VERITY_MAX_LEVELS + 1] = {}; + struct block_buffer *buffers = &_buffers[1]; + unsigned long level_offset[FS_VERITY_MAX_LEVELS]; + int level; + u64 offset; + int err; + + if (data_size == 0) { + /* Empty file is a special case; root hash is all 0's */ + memset(root_hash, 0, params->digest_size); + return 0; + } + + /* + * Allocate the block buffers. Buffer "-1" is for data blocks. + * Buffers 0 <= level < num_levels are for the actual tree levels. + * Buffer 'num_levels' is for the root hash. + */ + for (level = -1; level < num_levels; level++) { + buffers[level].data = kzalloc(params->block_size, GFP_KERNEL); + if (!buffers[level].data) { + err = -ENOMEM; + goto out; + } + } + buffers[num_levels].data = root_hash; + buffers[num_levels].is_root_hash = true; + + BUILD_BUG_ON(sizeof(level_offset) != sizeof(params->level_start)); + memcpy(level_offset, params->level_start, sizeof(level_offset)); + + /* Hash each data block, also hashing the tree blocks as they fill up */ + for (offset = 0; offset < data_size; offset += params->block_size) { + ssize_t bytes_read; + loff_t pos = offset; + + buffers[-1].filled = min_t(u64, params->block_size, + data_size - offset); + bytes_read = __kernel_read(filp, buffers[-1].data, + buffers[-1].filled, &pos); + if (bytes_read < 0) { + err = bytes_read; + fsverity_err(inode, "Error %d reading file data", err); + goto out; + } + if (bytes_read != buffers[-1].filled) { + err = -EINVAL; + fsverity_err(inode, "Short read of file data"); + goto out; + } + err = hash_one_block(params, &buffers[-1]); + if (err) + goto out; + for (level = 0; level < num_levels; level++) { + if (buffers[level].filled + params->digest_size <= + params->block_size) { + /* Next block at @level isn't full yet */ + break; + } + /* Next block at @level is full */ + + err = hash_one_block(params, &buffers[level]); + if (err) + goto out; + err = write_merkle_tree_block(filp, + buffers[level].data, + level_offset[level], + params); + if (err) + goto out; + level_offset[level]++; + } + if (fatal_signal_pending(current)) { + err = -EINTR; + goto out; + } + cond_resched(); + } + /* Finish all nonempty pending tree blocks. */ + for (level = 0; level < num_levels; level++) { + if (buffers[level].filled != 0) { + err = hash_one_block(params, &buffers[level]); + if (err) + goto out; + err = write_merkle_tree_block(filp, + buffers[level].data, + level_offset[level], + params); + if (err) + goto out; + } + } + /* The root hash was filled by the last call to hash_one_block(). */ + if (WARN_ON_ONCE(buffers[num_levels].filled != params->digest_size)) { + err = -EINVAL; + goto out; + } + err = 0; +out: + for (level = -1; level < num_levels; level++) + kfree(buffers[level].data); + return err; +} + +static int enable_verity(struct file *filp, + const struct fsverity_enable_arg *arg) +{ + struct inode *inode = file_inode(filp); + const struct fsverity_operations *vops = inode->i_sb->s_vop; + struct merkle_tree_params params = { }; + struct fsverity_descriptor *desc; + size_t desc_size = struct_size(desc, signature, arg->sig_size); + struct fsverity_info *vi; + int err; + + /* Start initializing the fsverity_descriptor */ + desc = kzalloc(desc_size, GFP_KERNEL); + if (!desc) + return -ENOMEM; + desc->version = 1; + desc->hash_algorithm = arg->hash_algorithm; + desc->log_blocksize = ilog2(arg->block_size); + + /* Get the salt if the user provided one */ + if (arg->salt_size && + copy_from_user(desc->salt, u64_to_user_ptr(arg->salt_ptr), + arg->salt_size)) { + err = -EFAULT; + goto out; + } + desc->salt_size = arg->salt_size; + + /* Get the builtin signature if the user provided one */ + if (arg->sig_size && + copy_from_user(desc->signature, u64_to_user_ptr(arg->sig_ptr), + arg->sig_size)) { + err = -EFAULT; + goto out; + } + desc->sig_size = cpu_to_le32(arg->sig_size); + + desc->data_size = cpu_to_le64(inode->i_size); + + /* Prepare the Merkle tree parameters */ + err = fsverity_init_merkle_tree_params(¶ms, inode, + arg->hash_algorithm, + desc->log_blocksize, + desc->salt, desc->salt_size); + if (err) + goto out; + + trace_fsverity_enable(inode, ¶ms); + + /* + * Start enabling verity on this file, serialized by the inode lock. + * Fail if verity is already enabled or is already being enabled. + */ + inode_lock(inode); + if (IS_VERITY(inode)) + err = -EEXIST; + else + err = vops->begin_enable_verity(filp); + inode_unlock(inode); + if (err) + goto out; + + /* + * Build the Merkle tree. Don't hold the inode lock during this, since + * on huge files this may take a very long time and we don't want to + * force unrelated syscalls like chown() to block forever. We don't + * need the inode lock here because deny_write_access() already prevents + * the file from being written to or truncated, and we still serialize + * ->begin_enable_verity() and ->end_enable_verity() using the inode + * lock and only allow one process to be here at a time on a given file. + */ + BUILD_BUG_ON(sizeof(desc->root_hash) < FS_VERITY_MAX_DIGEST_SIZE); + err = build_merkle_tree(filp, ¶ms, desc->root_hash); + if (err) { + fsverity_err(inode, "Error %d building Merkle tree", err); + goto rollback; + } + + /* + * Create the fsverity_info. Don't bother trying to save work by + * reusing the merkle_tree_params from above. Instead, just create the + * fsverity_info from the fsverity_descriptor as if it were just loaded + * from disk. This is simpler, and it serves as an extra check that the + * metadata we're writing is valid before actually enabling verity. + */ + vi = fsverity_create_info(inode, desc); + if (IS_ERR(vi)) { + err = PTR_ERR(vi); + goto rollback; + } + + trace_fsverity_tree_done(inode, vi, ¶ms); + + /* + * Add the fsverity_info into the hash table before finishing the + * initialization so that we don't have to undo the enabling when memory + * allocation for the hash table fails. This is safe because looking up + * the fsverity_info always first checks the S_VERITY flag on the inode, + * which will only be set at the very end of the ->end_enable_verity + * method. + */ + err = fsverity_set_info(vi); + if (err) { + fsverity_free_info(vi); + goto rollback; + } + + /* + * Tell the filesystem to finish enabling verity on the file. + * Serialized with ->begin_enable_verity() by the inode lock. The file + * system needs to set the S_VERITY flag on the inode at the very end of + * the method, at which point the fsverity information can be accessed + * by other threads. + */ + inode_lock(inode); + err = vops->end_enable_verity(filp, desc, desc_size, params.tree_size); + inode_unlock(inode); + if (err) { + fsverity_err(inode, "%ps() failed with err %d", + vops->end_enable_verity, err); + fsverity_remove_info(vi); + } else if (WARN_ON_ONCE(!IS_VERITY(inode))) { + fsverity_remove_info(vi); + err = -EINVAL; + } +out: + kfree(params.hashstate); + kfree(desc); + return err; + +rollback: + inode_lock(inode); + (void)vops->end_enable_verity(filp, NULL, 0, params.tree_size); + inode_unlock(inode); + goto out; +} + +/** + * fsverity_ioctl_enable() - enable verity on a file + * @filp: file to enable verity on + * @uarg: user pointer to fsverity_enable_arg + * + * Enable fs-verity on a file. See the "FS_IOC_ENABLE_VERITY" section of + * Documentation/filesystems/fsverity.rst for the documentation. + * + * Return: 0 on success, -errno on failure + */ +int fsverity_ioctl_enable(struct file *filp, const void __user *uarg) +{ + struct inode *inode = file_inode(filp); + struct fsverity_enable_arg arg; + int err; + + if (copy_from_user(&arg, uarg, sizeof(arg))) + return -EFAULT; + + if (arg.version != 1) + return -EINVAL; + + if (arg.__reserved1 || + memchr_inv(arg.__reserved2, 0, sizeof(arg.__reserved2))) + return -EINVAL; + + if (!is_power_of_2(arg.block_size)) + return -EINVAL; + + if (arg.salt_size > sizeof_field(struct fsverity_descriptor, salt)) + return -EMSGSIZE; + + if (arg.sig_size > FS_VERITY_MAX_SIGNATURE_SIZE) + return -EMSGSIZE; + + /* + * Require a regular file with write access. But the actual fd must + * still be readonly so that we can lock out all writers. This is + * needed to guarantee that no writable fds exist to the file once it + * has verity enabled, and to stabilize the data being hashed. + */ + + err = file_permission(filp, MAY_WRITE); + if (err) + return err; + /* + * __kernel_read() is used while building the Merkle tree. So, we can't + * allow file descriptors that were opened for ioctl access only, using + * the special nonstandard access mode 3. O_RDONLY only, please! + */ + if (!(filp->f_mode & FMODE_READ)) + return -EBADF; + + if (IS_APPEND(inode)) + return -EPERM; + + if (S_ISDIR(inode->i_mode)) + return -EISDIR; + + if (!S_ISREG(inode->i_mode)) + return -EINVAL; + + err = mnt_want_write_file(filp); + if (err) /* -EROFS */ + return err; + + err = deny_write_access(filp); + if (err) /* -ETXTBSY */ + goto out_drop_write; + + err = enable_verity(filp, &arg); + + /* + * We no longer drop the inode's pagecache after enabling verity. This + * used to be done to try to avoid a race condition where pages could be + * evicted after being used in the Merkle tree construction, then + * re-instantiated by a concurrent read. Such pages are unverified, and + * the backing storage could have filled them with different content, so + * they shouldn't be used to fulfill reads once verity is enabled. + * + * But, dropping the pagecache has a big performance impact, and it + * doesn't fully solve the race condition anyway. So for those reasons, + * and also because this race condition isn't very important relatively + * speaking (especially for small-ish files, where the chance of a page + * being used, evicted, *and* re-instantiated all while enabling verity + * is quite small), we no longer drop the inode's pagecache. + */ + + /* + * allow_write_access() is needed to pair with deny_write_access(). + * Regardless, the filesystem won't allow writing to verity files. + */ + allow_write_access(filp); +out_drop_write: + mnt_drop_write_file(filp); + return err; +} +EXPORT_SYMBOL_GPL(fsverity_ioctl_enable); -- cgit v1.3.1