The verity target provides transparent integrity checking of block devices
using a cryptographic digest.
dm-verity is meant to be setup as part of a verified boot path. This
may be anything ranging from a boot using tboot or trustedgrub to just
booting from a known-good device (like a USB drive or CD).
dm-verity is part of ChromeOS's verified boot path. It is used to verify
the integrity of the root filesystem on boot. The root filesystem is
mounted on a dm-verity partition which transparently verifies each block
with a bootloader verified hash passed into the kernel at boot.
Changes in V4:
* Discussion over phone (Alasdair G Kergon)
* copy _ioctl fix from dm-linear
* verity_status format fixes to match dm conventions
* s/dm-bht/verity_tree
* put everything into dm-verity.c
* ctr changed to dm conventions
* use hex2bin
* use conventional dm names for function
* s/dm_//
* for example: verity_ctr versus dm_verity_ctr
* use per_cpu API
Changes in V3:
* Discussion over irc (Alasdair G Kergon)
* Implement ioctl hook
Changes in V2:
* https://lkml.org/lkml/2011/11/10/85 (Steffen Klassert)
* Use shash API instead of older hash API
Signed-off-by: Will Drewry <[email protected]>
Signed-off-by: Elly Jones <[email protected]>
Signed-off-by: Mandeep Singh Baines <[email protected]>
Cc: Alasdair G Kergon <[email protected]>
Cc: Milan Broz <[email protected]>
Cc: Olof Johansson <[email protected]>
Cc: Steffen Klassert <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Mikulas Patocka <[email protected]>
Cc: [email protected]
---
Documentation/device-mapper/verity.txt | 149 ++++
drivers/md/Kconfig | 16 +
drivers/md/Makefile | 1 +
drivers/md/dm-verity.c | 1411 ++++++++++++++++++++++++++++++++
4 files changed, 1577 insertions(+), 0 deletions(-)
create mode 100644 Documentation/device-mapper/verity.txt
create mode 100644 drivers/md/dm-verity.c
diff --git a/Documentation/device-mapper/verity.txt b/Documentation/device-mapper/verity.txt
new file mode 100644
index 0000000..b631f12
--- /dev/null
+++ b/Documentation/device-mapper/verity.txt
@@ -0,0 +1,149 @@
+dm-verity
+==========
+
+Device-Mapper's "verity" target provides transparent integrity checking of
+block devices using a cryptographic digest provided by the kernel crypto API.
+This target is read-only.
+
+Parameters:
+ <version> <dev> <hash_dev> <hash_start> <block_size> <alg> <digest> <salt>
+
+<version>
+ This is the version number of the on-disk format. Currently, there is
+ only version 0.
+
+<dev>
+ This is the device that is going to be integrity checked. It may be
+ a subset of the full device as specified to dmsetup (start sector and count)
+ It may be specified as a path, like /dev/sdaX, or a device number,
+ <major>:<minor>.
+
+<hash_dev>
+ This is the device that that supplies the hash tree data. It may be
+ specified similarly to the device path and may be the same device. If the
+ same device is used, the hash offset should be outside of the dm-verity
+ configured device size.
+
+<hash_start>
+ This is the offset, in 512-byte sectors, from the start of hash_dev to
+ the root block of the hash tree.
+
+<block_size>
+ The size of a hash block. Also, the size of a block to be hashed.
+
+<alg>
+ The cryptographic hash algorithm used for this device. This should
+ be the name of the algorithm, like "sha1".
+
+<digest>
+ The hexadecimal encoding of the cryptographic hash of all of the
+ neighboring nodes at the first level of the tree. This hash should be
+ trusted as there is no other authenticity beyond this point.
+
+<salt>
+ The hexadecimal encoding of the salt value.
+
+Theory of operation
+===================
+
+dm-verity is meant to be setup as part of a verified boot path. This
+may be anything ranging from a boot using tboot or trustedgrub to just
+booting from a known-good device (like a USB drive or CD).
+
+When a dm-verity device is configured, it is expected that the caller
+has been authenticated in some way (cryptographic signatures, etc).
+After instantiation, all hashes will be verified on-demand during
+disk access. If they cannot be verified up to the root node of the
+tree, the root hash, then the I/O will fail. This should identify
+tampering with any data on the device and the hash data.
+
+Cryptographic hashes are used to assert the integrity of the device on a
+per-block basis. This allows for a lightweight hash computation on first read
+into the page cache. Block hashes are stored linearly aligned to the nearest
+block the size of a page.
+
+Hash Tree
+---------
+
+Each node in the tree is a cryptographic hash. If it is a leaf node, the hash
+is of some block data on disk. If it is an intermediary node, then the hash is
+of a number of child nodes.
+
+Each entry in the tree is a collection of neighboring nodes that fit in one
+block. The number is determined based on block_size and the size of the
+selected cryptographic digest algorithm. The hashes are linearly ordered in
+this entry and any unaligned trailing space is ignored but included when
+calculating the parent node.
+
+The tree looks something like:
+
+alg = sha256, num_blocks = 32768, block_size = 4096
+
+ [ root ]
+ / . . . \
+ [entry_0] [entry_1]
+ / . . . \ . . . \
+ [entry_0_0] . . . [entry_0_127] . . . . [entry_1_127]
+ / ... \ / . . . \ / \
+ blk_0 ... blk_127 blk_16256 blk_16383 blk_32640 . . . blk_32767
+
+On-disk format
+==============
+
+Below is the recommended on-disk format. The verity kernel code does not
+read the on-disk header. It only reads the hash blocks which directly
+follow the header. It is expected that a user-space tool will verify the
+integrity of the verity_header and then call dm_setup with the correct
+parameters. Alternatively, the header can be omitted and the dm_setup
+parameters can be passed via the kernel command-line in a rooted chain
+of trust where the command-line is verified.
+
+The on-disk format is especially useful in cases where the hash blocks
+are on a separate partition. The magic number allows easy identification
+of the partition contents. Alternatively, the hash blocks can be stored
+in the same partition as the data to be verified. In such a configuration
+the filesystem on the partition would be sized a little smaller than
+the full-partition, leaving room for the hash blocks.
+
+struct verity_header {
+ uint64_t magic = 0x7665726974790a00;
+ uint32_t version;
+ uint32_t block_size;
+ char digest[128]; /* in hex-ascii, null-terminated or 128-bytes */
+ char salt[128]; /* in hex-ascii, null-terminated or 128-bytes */
+}
+
+struct verity_header_block {
+ struct verity_header;
+ char unused[block_size - sizeof(struct verity_header) - sizeof(sig)];
+ char sig[128]; /* in hex-ascii, null-terminated or 128-bytes */
+}
+
+Directly following the header are the hash blocks which are stored a depth
+at a time (starting from the root), sorted in order of increasing index.
+
+Usage
+=====
+
+The API provides mechanisms for reading and verifying a tree. When reading, all
+required data for the hash tree should be populated for a block before
+attempting a verify. This can be done by calling dm_bht_populate(). When all
+data is ready, a call to dm_bht_verify_block() with the expected hash value will
+perform both the direct block hash check and the hashes of the parent and
+neighboring nodes where needed to ensure validity up to the root hash. Note,
+dm_bht_set_root_hexdigest() should be called before any verification attempts
+occur.
+
+Example
+=======
+
+Setup a device;
+[[
+ dmsetup create vroot --table \
+ "0 204800 verity /dev/sda1 /dev/sda2 alg=sha1 "\
+ "root_hexdigest=9f74809a2ee7607b16fcc70d9399a4de9725a727"
+]]
+
+A command line tool is available to compute the hash tree and return the
+root hash value.
+ http://git.chromium.org/cgi-bin/gitweb.cgi?p=dm-verity.git;a=tree
diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
index faa4741..b8bb690 100644
--- a/drivers/md/Kconfig
+++ b/drivers/md/Kconfig
@@ -370,4 +370,20 @@ config DM_FLAKEY
---help---
A target that intermittently fails I/O for debugging purposes.
+config DM_VERITY
+ tristate "Verity target support"
+ depends on BLK_DEV_DM
+ select CRYPTO
+ select CRYPTO_HASH
+ ---help---
+ This device-mapper target allows you to create a device that
+ transparently integrity checks the data on it. You'll need to
+ activate the digests you're going to use in the cryptoapi
+ configuration.
+
+ To compile this code as a module, choose M here: the module will
+ be called dm-verity.
+
+ If unsure, say N.
+
endif # MD
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index 046860c..70a29af 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -39,6 +39,7 @@ obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o
obj-$(CONFIG_DM_PERSISTENT_DATA) += persistent-data/
obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o dm-region-hash.o
obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
+obj-$(CONFIG_DM_VERITY) += dm-verity.o
obj-$(CONFIG_DM_ZERO) += dm-zero.o
obj-$(CONFIG_DM_RAID) += dm-raid.o
obj-$(CONFIG_DM_THIN_PROVISIONING) += dm-thin-pool.o
diff --git a/drivers/md/dm-verity.c b/drivers/md/dm-verity.c
new file mode 100644
index 0000000..87b7958
--- /dev/null
+++ b/drivers/md/dm-verity.c
@@ -0,0 +1,1411 @@
+/*
+ * Originally based on dm-crypt.c,
+ * Copyright (C) 2003 Christophe Saout <[email protected]>
+ * Copyright (C) 2004 Clemens Fruhwirth <[email protected]>
+ * Copyright (C) 2006-2008 Red Hat, Inc. All rights reserved.
+ * Copyright (C) 2012 The Chromium OS Authors <[email protected]>
+ * All Rights Reserved.
+ *
+ * This file is released under the GPLv2.
+ *
+ * Implements a verifying transparent block device.
+ * See Documentation/device-mapper/dm-verity.txt
+ */
+#include <crypto/hash.h>
+#include <linux/atomic.h>
+#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <linux/genhd.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/mempool.h>
+#include <linux/module.h>
+#include <linux/workqueue.h>
+#include <linux/device-mapper.h>
+
+
+#define DM_MSG_PREFIX "verity"
+
+
+/* Helper for printing sector_t */
+#define ULL(x) ((unsigned long long)(x))
+
+#define MIN_IOS 32
+#define MIN_BIOS (MIN_IOS * 2)
+
+/* To avoid allocating memory for digest tests, we just setup a
+ * max to use for now.
+ */
+#define VERITY_MAX_DIGEST_SIZE 64 /* Supports up to 512-bit digests */
+#define VERITY_SALT_SIZE 32 /* 256 bits of salt is a lot */
+
+/* UNALLOCATED, PENDING, READY, and VERIFIED are valid states. All other
+ * values are entry-related return codes.
+ */
+#define VERITY_TREE_ENTRY_VERIFIED 8 /* 'nodes' checked against parent */
+#define VERITY_TREE_ENTRY_READY 4 /* 'nodes' is loaded and available */
+#define VERITY_TREE_ENTRY_PENDING 2 /* 'nodes' is being loaded */
+#define VERITY_TREE_ENTRY_UNALLOCATED 0 /* untouched */
+#define VERITY_TREE_ENTRY_ERROR -1 /* entry is unsuitable for use */
+#define VERITY_TREE_ENTRY_ERROR_IO -2 /* I/O error on load */
+
+/* Additional possible return codes */
+#define VERITY_TREE_ENTRY_ERROR_MISMATCH -3 /* Digest mismatch */
+
+
+/* verity_tree_entry
+ * Contains verity_tree->node_count tree nodes at a given tree depth.
+ * state is used to transactionally assure that data is paged in
+ * from disk. Unless verity_tree kept running crypto contexts for each
+ * level, we need to load in the data for on-demand verification.
+ */
+struct verity_tree_entry {
+ atomic_t state; /* see defines */
+ /* Keeping an extra pointer per entry wastes up to ~33k of
+ * memory if a 1m blocks are used (or 66 on 64-bit arch)
+ */
+ void *io_context; /* Reserve a pointer for use during io */
+ /* data should only be non-NULL if fully populated. */
+ void *nodes; /* The hash data used to verify the children.
+ * Guaranteed to be page-aligned.
+ */
+};
+
+/* verity_tree_level
+ * Contains an array of entries which represent a page of hashes where
+ * each hash is a node in the tree at the given tree depth/level.
+ */
+struct verity_tree_level {
+ struct verity_tree_entry *entries; /* array of entries of tree nodes */
+ unsigned int count; /* number of entries at this level */
+ sector_t sector; /* starting sector for this level */
+};
+
+/* opaque context, start, databuf, sector_count */
+typedef int(*verity_tree_callback)(void *, /* external context */
+ sector_t, /* start sector */
+ u8 *, /* destination page */
+ sector_t, /* num sectors */
+ struct verity_tree_entry *);
+/* verity_tree - Device mapper block hash tree
+ * verity_tree provides a fixed interface for comparing data blocks
+ * against a cryptographic hashes stored in a hash tree. It
+ * optimizes the tree structure for storage on disk.
+ *
+ * The tree is built from the bottom up. A collection of data,
+ * external to the tree, is hashed and these hashes are stored
+ * as the blocks in the tree. For some number of these hashes,
+ * a parent node is created by hashing them. These steps are
+ * repeated.
+ */
+struct verity_tree {
+ /* Configured values */
+ int depth; /* Depth of the tree including the root */
+ unsigned int block_count; /* Number of blocks hashed */
+ unsigned int block_size; /* Size of a hash block */
+ char hash_alg[CRYPTO_MAX_ALG_NAME];
+ u8 salt[VERITY_SALT_SIZE];
+
+ /* Computed values */
+ unsigned int node_count; /* Data size (in hashes) for each entry */
+ unsigned int node_count_shift; /* first bit set - 1 */
+ /*
+ * There is one per CPU so that verified can be simultaneous.
+ * Access through per_cpu_ptr() only
+ */
+ struct shash_desc * __percpu *hash_desc; /* Container for hash alg */
+ unsigned int digest_size;
+ sector_t sectors; /* Number of disk sectors used */
+
+ /* bool verified; Full tree is verified */
+ u8 digest[VERITY_MAX_DIGEST_SIZE];
+ struct verity_tree_level *levels; /* in reverse order */
+ /* Callback for reading from the hash device */
+ verity_tree_callback read_cb;
+};
+
+/* per-requested-bio private data */
+enum verity_io_flags {
+ VERITY_IOFLAGS_CLONED = 0x1, /* original bio has been cloned */
+};
+
+struct verity_io {
+ struct dm_target *target;
+ struct bio *bio;
+ struct delayed_work work;
+ unsigned int flags;
+
+ int error;
+ atomic_t pending;
+
+ u64 block; /* aligned block index */
+ u64 count; /* aligned count in blocks */
+};
+
+struct verity_config {
+ struct dm_dev *dev;
+ sector_t start;
+ sector_t size;
+
+ struct dm_dev *hash_dev;
+ sector_t hash_start;
+
+ struct verity_tree bht;
+
+ /* Pool required for io contexts */
+ mempool_t *io_pool;
+ /* Pool and bios required for making sure that backing device reads are
+ * in PAGE_SIZE increments.
+ */
+ struct bio_set *bs;
+
+ char hash_alg[CRYPTO_MAX_ALG_NAME];
+};
+
+
+static struct kmem_cache *_verity_io_pool;
+static struct workqueue_struct *kveritydq, *kverityd_ioq;
+
+
+static void kverityd_verify(struct work_struct *work);
+static void kverityd_io(struct work_struct *work);
+static void kverityd_io_bht_populate(struct verity_io *io);
+static void kverityd_io_bht_populate_end(struct bio *, int error);
+
+
+/*
+ * Utilities
+ */
+
+static void bin2hex(char *dst, const u8 *src, size_t count)
+{
+ while (count-- > 0) {
+ sprintf(dst, "%02hhx", (int)*src);
+ dst += 2;
+ src++;
+ }
+}
+
+/*
+ * Verity Tree
+ */
+
+/* Functions for converting indices to nodes. */
+
+static inline unsigned int verity_tree_get_level_shift(struct verity_tree *bht,
+ int depth)
+{
+ return (bht->depth - depth) * bht->node_count_shift;
+}
+
+/* For the given depth, this is the entry index. At depth+1 it is the node
+ * index for depth.
+ */
+static inline unsigned int verity_tree_index_at_level(struct verity_tree *bht,
+ int depth,
+ unsigned int leaf)
+{
+ return leaf >> verity_tree_get_level_shift(bht, depth);
+}
+
+static inline struct verity_tree_entry *verity_tree_get_entry(
+ struct verity_tree *bht,
+ int depth,
+ unsigned int block)
+{
+ unsigned int index = verity_tree_index_at_level(bht, depth, block);
+ struct verity_tree_level *level = &bht->levels[depth];
+
+ return &level->entries[index];
+}
+
+static inline void *verity_tree_get_node(struct verity_tree *bht,
+ struct verity_tree_entry *entry,
+ int depth, unsigned int block)
+{
+ unsigned int index = verity_tree_index_at_level(bht, depth, block);
+ unsigned int node_index = index % bht->node_count;
+
+ return entry->nodes + (node_index * bht->digest_size);
+}
+/**
+ * verity_tree_compute_hash: hashes a page of data
+ */
+static int verity_tree_compute_hash(struct verity_tree *vt, struct page *pg,
+ unsigned int offset, u8 *digest)
+{
+ struct shash_desc *hash_desc;
+ void *data;
+ int err;
+
+ hash_desc = *per_cpu_ptr(vt->hash_desc, smp_processor_id());
+
+ if (crypto_shash_init(hash_desc)) {
+ DMCRIT("failed to reinitialize crypto hash (proc:%d)",
+ smp_processor_id());
+ return -EINVAL;
+ }
+ data = kmap_atomic(pg);
+ err = crypto_shash_update(hash_desc, data + offset, PAGE_SIZE);
+ kunmap_atomic(data);
+ if (err) {
+ DMCRIT("crypto_hash_update failed");
+ return -EINVAL;
+ }
+ if (crypto_shash_update(hash_desc, vt->salt, sizeof(vt->salt))) {
+ DMCRIT("crypto_hash_update failed");
+ return -EINVAL;
+ }
+ if (crypto_shash_final(hash_desc, digest)) {
+ DMCRIT("crypto_hash_final failed");
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int verity_tree_initialize_entries(struct verity_tree *vt)
+{
+ /* last represents the index of the last digest store in the tree.
+ * By walking the tree with that index, it is possible to compute the
+ * total number of entries at each level.
+ *
+ * Since each entry will contain up to |node_count| nodes of the tree,
+ * it is possible that the last index may not be at the end of a given
+ * entry->nodes. In that case, it is assumed the value is padded.
+ *
+ * Note, we treat both the tree root (1 hash) and the tree leaves
+ * independently from the vt data structures. Logically, the root is
+ * depth=-1 and the block layer level is depth=vt->depth
+ */
+ unsigned int last = vt->block_count;
+ int depth;
+
+ /* check that the largest level->count can't result in an int overflow
+ * on allocation or sector calculation.
+ */
+ if (((last >> vt->node_count_shift) + 1) >
+ UINT_MAX / max((unsigned int)sizeof(struct verity_tree_entry),
+ (unsigned int)to_sector(vt->block_size))) {
+ DMCRIT("required entries %u is too large", last + 1);
+ return -EINVAL;
+ }
+
+ /* Track the current sector location for each level so we don't have to
+ * compute it during traversals.
+ */
+ vt->sectors = 0;
+ for (depth = 0; depth < vt->depth; ++depth) {
+ struct verity_tree_level *level = &vt->levels[depth];
+
+ level->count = verity_tree_index_at_level(vt, depth, last) + 1;
+ level->entries = (struct verity_tree_entry *)
+ kcalloc(level->count,
+ sizeof(struct verity_tree_entry),
+ GFP_KERNEL);
+ if (!level->entries) {
+ DMERR("failed to allocate entries for depth %d", depth);
+ return -ENOMEM;
+ }
+ level->sector = vt->sectors;
+ vt->sectors += level->count * to_sector(vt->block_size);
+ }
+
+ return 0;
+}
+
+/**
+ * verity_tree_create - prepares @vt for us
+ * @vt: pointer to a verity_tree_create()d vt
+ * @depth: tree depth without the root; including block hashes
+ * @block_count: the number of block hashes / tree leaves
+ * @alg_name: crypto hash algorithm name
+ *
+ * Returns 0 on success.
+ *
+ * Callers can offset into devices by storing the data in the io callbacks.
+ */
+static int verity_tree_create(struct verity_tree *vt, unsigned int block_count,
+ unsigned int block_size, const char *alg_name)
+{
+ struct crypto_shash *tfm;
+ int size, cpu, status = 0;
+
+ vt->block_size = block_size;
+ /* Verify that PAGE_SIZE >= block_size >= SECTOR_SIZE. */
+ if ((block_size > PAGE_SIZE) ||
+ (PAGE_SIZE % block_size) ||
+ (to_sector(block_size) == 0))
+ return -EINVAL;
+
+ tfm = crypto_alloc_shash(alg_name, 0, 0);
+ if (IS_ERR(tfm)) {
+ DMERR("failed to allocate crypto hash '%s'", alg_name);
+ return -ENOMEM;
+ }
+ size = sizeof(struct shash_desc) + crypto_shash_descsize(tfm);
+
+ vt->hash_desc = alloc_percpu(struct shash_desc *);
+ if (!vt->hash_desc) {
+ DMERR("Failed to allocate per cpu hash_desc");
+ status = -ENOMEM;
+ goto bad_per_cpu;
+ }
+
+ /* Pre-allocate per-cpu crypto contexts to avoid having to
+ * kmalloc/kfree a context for every hash operation.
+ */
+ for_each_possible_cpu(cpu) {
+ struct shash_desc *hash_desc = kmalloc(size, GFP_KERNEL);
+
+ *per_cpu_ptr(vt->hash_desc, cpu) = hash_desc;
+ if (!hash_desc) {
+ DMERR("failed to allocate crypto hash contexts");
+ status = -ENOMEM;
+ goto bad_hash_alloc;
+ }
+ hash_desc->tfm = tfm;
+ hash_desc->flags = 0x0;
+ }
+ vt->digest_size = crypto_shash_digestsize(tfm);
+ /* We expect to be able to pack >=2 hashes into a block */
+ if (block_size / vt->digest_size < 2) {
+ DMERR("too few hashes fit in a block");
+ status = -EINVAL;
+ goto bad_arg;
+ }
+
+ if (vt->digest_size > VERITY_MAX_DIGEST_SIZE) {
+ DMERR("VERITY_MAX_DIGEST_SIZE too small for digest");
+ status = -EINVAL;
+ goto bad_arg;
+ }
+
+ /* Configure the tree */
+ vt->block_count = block_count;
+ if (block_count == 0) {
+ DMERR("block_count must be non-zero");
+ status = -EINVAL;
+ goto bad_arg;
+ }
+
+ /* Each verity_tree_entry->nodes is one block. The node code tracks
+ * how many nodes fit into one entry where a node is a single
+ * hash (message digest).
+ */
+ vt->node_count_shift = fls(block_size / vt->digest_size) - 1;
+ /* Round down to the nearest power of two. This makes indexing
+ * into the tree much less painful.
+ */
+ vt->node_count = 1 << vt->node_count_shift;
+
+ /* This is unlikely to happen, but with 64k pages, who knows. */
+ if (vt->node_count > UINT_MAX / vt->digest_size) {
+ DMERR("node_count * hash_len exceeds UINT_MAX!");
+ status = -EINVAL;
+ goto bad_arg;
+ }
+
+ vt->depth = DIV_ROUND_UP(fls(block_count - 1), vt->node_count_shift);
+
+ /* Ensure that we can safely shift by this value. */
+ if (vt->depth * vt->node_count_shift >= sizeof(unsigned int) * 8) {
+ DMERR("specified depth and node_count_shift is too large");
+ status = -EINVAL;
+ goto bad_arg;
+ }
+
+ /* Allocate levels. Each level of the tree may have an arbitrary number
+ * of verity_tree_entry structs. Each entry contains node_count nodes.
+ * Each node in the tree is a cryptographic digest of either node_count
+ * nodes on the subsequent level or of a specific block on disk.
+ */
+ vt->levels = (struct verity_tree_level *)
+ kcalloc(vt->depth,
+ sizeof(struct verity_tree_level), GFP_KERNEL);
+ if (!vt->levels) {
+ DMERR("failed to allocate tree levels");
+ status = -ENOMEM;
+ goto bad_level_alloc;
+ }
+
+ vt->read_cb = NULL;
+
+ status = verity_tree_initialize_entries(vt);
+ if (status)
+ goto bad_entries_alloc;
+
+ /* We compute depth such that there is only be 1 block at level 0. */
+ BUG_ON(vt->levels[0].count != 1);
+
+ return 0;
+
+bad_entries_alloc:
+ while (vt->depth-- > 0)
+ kfree(vt->levels[vt->depth].entries);
+ kfree(vt->levels);
+bad_level_alloc:
+bad_arg:
+bad_hash_alloc:
+ for_each_possible_cpu(cpu)
+ if (*per_cpu_ptr(vt->hash_desc, cpu))
+ kfree(*per_cpu_ptr(vt->hash_desc, cpu));
+ free_percpu(vt->hash_desc);
+bad_per_cpu:
+ crypto_free_shash(tfm);
+ return status;
+}
+
+/**
+ * verity_tree_read_completed
+ * @entry: pointer to the entry that's been loaded
+ * @status: I/O status. Non-zero is failure.
+ * MUST always be called after a read_cb completes.
+ */
+static void verity_tree_read_completed(struct verity_tree_entry *entry,
+ int status)
+{
+ if (status) {
+ DMCRIT("an I/O error occurred while reading entry");
+ atomic_set(&entry->state, VERITY_TREE_ENTRY_ERROR_IO);
+ return;
+ }
+ BUG_ON(atomic_read(&entry->state) != VERITY_TREE_ENTRY_PENDING);
+ atomic_set(&entry->state, VERITY_TREE_ENTRY_READY);
+}
+
+/**
+ * verity_tree_verify_block - checks that all path nodes for @block are valid
+ * @vt: pointer to a verity_tree_create()d vt
+ * @block: specific block data is expected from
+ * @pg: page holding the block data
+ * @offset: offset into the page
+ *
+ * Returns 0 on success, VERITY_TREE_ENTRY_ERROR_MISMATCH on error.
+ */
+static int verity_tree_verify_block(struct verity_tree *vt, unsigned int block,
+ struct page *pg, unsigned int offset)
+{
+ int state, depth = vt->depth;
+ u8 digest[VERITY_MAX_DIGEST_SIZE];
+ struct verity_tree_entry *entry;
+ void *node;
+
+ do {
+ /* Need to check that the hash of the current block is accurate
+ * in its parent.
+ */
+ entry = verity_tree_get_entry(vt, depth - 1, block);
+ state = atomic_read(&entry->state);
+ /* This call is only safe if all nodes along the path
+ * are already populated (i.e. READY) via verity_tree_populate.
+ */
+ BUG_ON(state < VERITY_TREE_ENTRY_READY);
+ node = verity_tree_get_node(vt, entry, depth, block);
+
+ if (verity_tree_compute_hash(vt, pg, offset, digest) ||
+ memcmp(digest, node, vt->digest_size))
+ goto mismatch;
+
+ /* Keep the containing block of hashes to be verified in the
+ * next pass.
+ */
+ pg = virt_to_page(entry->nodes);
+ offset = offset_in_page(entry->nodes);
+ } while (--depth > 0 && state != VERITY_TREE_ENTRY_VERIFIED);
+
+ if (depth == 0 && state != VERITY_TREE_ENTRY_VERIFIED) {
+ if (verity_tree_compute_hash(vt, pg, offset, digest) ||
+ memcmp(digest, vt->digest, vt->digest_size))
+ goto mismatch;
+ atomic_set(&entry->state, VERITY_TREE_ENTRY_VERIFIED);
+ }
+
+ /* Mark path to leaf as verified. */
+ for (depth++; depth < vt->depth; depth++) {
+ entry = verity_tree_get_entry(vt, depth, block);
+ /* At this point, entry can only be in VERIFIED or READY state.
+ * So it is safe to use atomic_set instead of atomic_cmpxchg.
+ */
+ atomic_set(&entry->state, VERITY_TREE_ENTRY_VERIFIED);
+ }
+
+ return 0;
+
+mismatch:
+ DMERR_LIMIT("verify_path: failed to verify hash (d=%d,bi=%u)",
+ depth, block);
+ return VERITY_TREE_ENTRY_ERROR_MISMATCH;
+}
+
+/**
+ * verity_tree_is_populated - check that nodes needed to verify a given
+ * block are all ready
+ * @vt: pointer to a verity_tree_create()d vt
+ * @block: specific block data is expected from
+ *
+ * Callers may wish to call verity_tree_is_populated() when checking an io
+ * for which entries were already pending.
+ */
+static bool verity_tree_is_populated(struct verity_tree *vt, unsigned int block)
+{
+ int depth;
+
+ for (depth = vt->depth - 1; depth >= 0; depth--) {
+ struct verity_tree_entry *entry;
+ entry = verity_tree_get_entry(vt, depth, block);
+ if (atomic_read(&entry->state) < VERITY_TREE_ENTRY_READY)
+ return false;
+ }
+
+ return true;
+}
+
+/**
+ * verity_tree_populate - reads entries from disk needed to verify a given block
+ * @vt: pointer to a verity_tree_create()d vt
+ * @ctx: context used for all read_cb calls on this request
+ * @block: specific block data is expected from
+ *
+ * Returns negative value on error. Returns 0 on success.
+ */
+static int verity_tree_populate(struct verity_tree *vt, void *ctx,
+ unsigned int block)
+{
+ int depth, state;
+
+ BUG_ON(block >= vt->block_count);
+
+ for (depth = vt->depth - 1; depth >= 0; --depth) {
+ unsigned int index;
+ struct verity_tree_level *level;
+ struct verity_tree_entry *entry;
+
+ index = verity_tree_index_at_level(vt, depth, block);
+ level = &vt->levels[depth];
+ entry = verity_tree_get_entry(vt, depth, block);
+ state = atomic_cmpxchg(&entry->state,
+ VERITY_TREE_ENTRY_UNALLOCATED,
+ VERITY_TREE_ENTRY_PENDING);
+ if (state == VERITY_TREE_ENTRY_VERIFIED)
+ break;
+ if (state <= VERITY_TREE_ENTRY_ERROR)
+ goto error_state;
+ if (state != VERITY_TREE_ENTRY_UNALLOCATED)
+ continue;
+
+ /* Current entry is claimed for allocation and loading */
+ entry->nodes = kmalloc(vt->block_size, GFP_NOIO);
+ if (!entry->nodes)
+ goto nomem;
+
+ vt->read_cb(ctx,
+ level->sector + to_sector(index * vt->block_size),
+ entry->nodes, to_sector(vt->block_size), entry);
+ }
+
+ return 0;
+
+error_state:
+ DMCRIT("block %u at depth %d is in an error state", block, depth);
+ return -EPERM;
+
+nomem:
+ DMCRIT("failed to allocate memory for entry->nodes");
+ return -ENOMEM;
+}
+
+/**
+ * verity_tree_destroy - cleans up all memory used by @vt
+ * @vt: pointer to a verity_tree_create()d vt
+ */
+static void verity_tree_destroy(struct verity_tree *vt)
+{
+ int depth, cpu;
+
+ for (depth = 0; depth < vt->depth; depth++) {
+ struct verity_tree_entry *entry = vt->levels[depth].entries;
+ struct verity_tree_entry *entry_end = entry +
+ vt->levels[depth].count;
+ for (; entry < entry_end; ++entry)
+ kfree(entry->nodes);
+ kfree(vt->levels[depth].entries);
+ }
+ kfree(vt->levels);
+ crypto_free_shash((*per_cpu_ptr(vt->hash_desc, 0))->tfm);
+ for_each_possible_cpu(cpu)
+ kfree(*per_cpu_ptr(vt->hash_desc, cpu));
+}
+
+/*
+ * Verity Tree Accessors
+ */
+
+/**
+ * verity_tree_set_digest - sets an unverified root digest hash from hex
+ * @vt: pointer to a verity_tree_create()d vt
+ * @digest: string containing the digest in hex
+ * Returns non-zero on error.
+ */
+static int verity_tree_set_digest(struct verity_tree *vt, const char *digest)
+{
+ /* Make sure we have at least the bytes expected */
+ if (strnlen((char *)digest, vt->digest_size * 2) !=
+ vt->digest_size * 2) {
+ DMERR("root digest length does not match hash algorithm");
+ return -1;
+ }
+ return hex2bin(vt->digest, digest, vt->digest_size);
+}
+
+/**
+ * verity_tree_digest - returns root digest in hex
+ * @vt: pointer to a verity_tree_create()d vt
+ * @digest: buffer to put into, must be of length VERITY_SALT_SIZE * 2 + 1.
+ */
+int verity_tree_digest(struct verity_tree *vt, char *digest)
+{
+ bin2hex(digest, vt->digest, vt->digest_size);
+ return 0;
+}
+
+/**
+ * verity_tree_set_salt - sets the salt
+ * @vt: pointer to a verity_tree_create()d vt
+ * @salt: string containing the salt in hex
+ * Returns non-zero on error.
+ */
+int verity_tree_set_salt(struct verity_tree *vt, const char *salt)
+{
+ size_t saltlen = min(strlen(salt) / 2, sizeof(vt->salt));
+ memset(vt->salt, 0, sizeof(vt->salt));
+ return hex2bin(vt->salt, salt, saltlen);
+}
+
+
+/**
+ * verity_tree_salt - returns the salt in hex
+ * @vt: pointer to a verity_tree_create()d vt
+ * @salt: buffer to put salt into, of length VERITY_SALT_SIZE * 2 + 1.
+ */
+int verity_tree_salt(struct verity_tree *vt, char *salt)
+{
+ bin2hex(salt, vt->salt, sizeof(vt->salt));
+ return 0;
+}
+
+/*
+ * Allocation and utility functions
+ */
+
+static void kverityd_src_io_read_end(struct bio *clone, int error);
+
+/* Shared destructor for all internal bios */
+static void verity_bio_destructor(struct bio *bio)
+{
+ struct verity_io *io = bio->bi_private;
+ struct verity_config *vc = io->target->private;
+ bio_free(bio, vc->bs);
+}
+
+static struct bio *verity_alloc_bioset(struct verity_config *vc, gfp_t gfp_mask,
+ int nr_iovecs)
+{
+ return bio_alloc_bioset(gfp_mask, nr_iovecs, vc->bs);
+}
+
+static struct verity_io *verity_io_alloc(struct dm_target *ti,
+ struct bio *bio)
+{
+ struct verity_config *vc = ti->private;
+ sector_t sector = bio->bi_sector - ti->begin;
+ struct verity_io *io;
+
+ io = mempool_alloc(vc->io_pool, GFP_NOIO);
+ if (unlikely(!io))
+ return NULL;
+ io->flags = 0;
+ io->target = ti;
+ io->bio = bio;
+ io->error = 0;
+
+ /* Adjust the sector by the virtual starting sector */
+ io->block = to_bytes(sector) / vc->bht.block_size;
+ io->count = bio->bi_size / vc->bht.block_size;
+
+ atomic_set(&io->pending, 0);
+
+ return io;
+}
+
+static struct bio *verity_bio_clone(struct verity_io *io)
+{
+ struct verity_config *vc = io->target->private;
+ struct bio *bio = io->bio;
+ struct bio *clone = verity_alloc_bioset(vc, GFP_NOIO, bio->bi_max_vecs);
+
+ if (!clone)
+ return NULL;
+
+ __bio_clone(clone, bio);
+ clone->bi_private = io;
+ clone->bi_end_io = kverityd_src_io_read_end;
+ clone->bi_bdev = vc->dev->bdev;
+ clone->bi_sector += vc->start - io->target->begin;
+ clone->bi_destructor = verity_bio_destructor;
+
+ return clone;
+}
+
+/*
+ * Reverse flow of requests into the device.
+ *
+ * (Start at the bottom with verity_map and work your way upward).
+ */
+
+static void verity_inc_pending(struct verity_io *io);
+
+static void verity_return_bio_to_caller(struct verity_io *io)
+{
+ struct verity_config *vc = io->target->private;
+
+ if (io->error)
+ io->error = -EIO;
+
+ bio_endio(io->bio, io->error);
+ mempool_free(io, vc->io_pool);
+}
+
+/* Check for any missing bht hashes. */
+static bool verity_is_bht_populated(struct verity_io *io)
+{
+ struct verity_config *vc = io->target->private;
+ u64 block;
+
+ for (block = io->block; block < io->block + io->count; ++block)
+ if (!verity_tree_is_populated(&vc->bht, block))
+ return false;
+
+ return true;
+}
+
+/* verity_dec_pending manages the lifetime of all verity_io structs.
+ * Non-bug error handling is centralized through this interface and
+ * all passage from workqueue to workqueue.
+ */
+static void verity_dec_pending(struct verity_io *io)
+{
+ if (!atomic_dec_and_test(&io->pending))
+ goto done;
+
+ if (unlikely(io->error))
+ goto io_error;
+
+ /* I/Os that were pending may now be ready */
+ if (verity_is_bht_populated(io)) {
+ INIT_DELAYED_WORK(&io->work, kverityd_verify);
+ queue_delayed_work(kveritydq, &io->work, 0);
+ } else {
+ INIT_DELAYED_WORK(&io->work, kverityd_io);
+ queue_delayed_work(kverityd_ioq, &io->work, HZ/10);
+ }
+
+done:
+ return;
+
+io_error:
+ verity_return_bio_to_caller(io);
+}
+
+/* Walks the data set and computes the hash of the data read from the
+ * untrusted source device. The computed hash is then passed to verity-tree
+ * for verification.
+ */
+static int verity_verify(struct verity_config *vc,
+ struct verity_io *io)
+{
+ unsigned int block_size = vc->bht.block_size;
+ struct bio *bio = io->bio;
+ u64 block = io->block;
+ unsigned int idx;
+ int r;
+
+ for (idx = bio->bi_idx; idx < bio->bi_vcnt; idx++) {
+ struct bio_vec *bv = bio_iovec_idx(bio, idx);
+ unsigned int offset = bv->bv_offset;
+ unsigned int len = bv->bv_len;
+
+ BUG_ON(offset % block_size);
+ BUG_ON(len % block_size);
+
+ while (len) {
+ r = verity_tree_verify_block(&vc->bht, block,
+ bv->bv_page, offset);
+ if (r)
+ goto bad_return;
+
+ offset += block_size;
+ len -= block_size;
+ block++;
+ cond_resched();
+ }
+ }
+
+ return 0;
+
+bad_return:
+ /* verity_tree functions aren't expected to return errno friendly
+ * values. They are converted here for uniformity.
+ */
+ if (r > 0) {
+ DMERR("Pending data for block %llu seen at verify", ULL(block));
+ r = -EBUSY;
+ } else {
+ DMERR_LIMIT("Block hash does not match!");
+ r = -EACCES;
+ }
+ return r;
+}
+
+/* Services the verify workqueue */
+static void kverityd_verify(struct work_struct *work)
+{
+ struct delayed_work *dwork = container_of(work, struct delayed_work,
+ work);
+ struct verity_io *io = container_of(dwork, struct verity_io,
+ work);
+ struct verity_config *vc = io->target->private;
+
+ io->error = verity_verify(vc, io);
+
+ /* Free up the bio and tag with the return value */
+ verity_return_bio_to_caller(io);
+}
+
+/* Asynchronously called upon the completion of verity-tree I/O. The status
+ * of the operation is passed back to verity-tree and the next steps are
+ * decided by verity_dec_pending.
+ */
+static void kverityd_io_bht_populate_end(struct bio *bio, int error)
+{
+ struct verity_tree_entry *entry;
+ struct verity_io *io;
+
+ entry = (struct verity_tree_entry *) bio->bi_private;
+ io = (struct verity_io *) entry->io_context;
+
+ /* Tell the tree to atomically update now that we've populated
+ * the given entry.
+ */
+ verity_tree_read_completed(entry, error);
+
+ /* Clean up for reuse when reading data to be checked */
+ bio->bi_vcnt = 0;
+ bio->bi_io_vec->bv_offset = 0;
+ bio->bi_io_vec->bv_len = 0;
+ bio->bi_io_vec->bv_page = NULL;
+ /* Restore the private data to I/O so the destructor can be shared. */
+ bio->bi_private = (void *) io;
+ bio_put(bio);
+
+ /* We bail but assume the tree has been marked bad. */
+ if (unlikely(error)) {
+ DMERR("Failed to read for sector %llu (%u)",
+ ULL(io->bio->bi_sector), io->bio->bi_size);
+ io->error = error;
+ /* Pass through the error to verity_dec_pending below */
+ }
+ /* When pending = 0, it will transition to reading real data */
+ verity_dec_pending(io);
+}
+
+/* Called by verity-tree (via verity_tree_populate), this function provides
+ * the message digests to verity-tree that are stored on disk.
+ */
+static int kverityd_bht_read_callback(void *ctx, sector_t start, u8 *dst,
+ sector_t count,
+ struct verity_tree_entry *entry)
+{
+ struct verity_io *io = ctx; /* I/O for this batch */
+ struct verity_config *vc;
+ struct bio *bio;
+
+ vc = io->target->private;
+
+ /* The I/O context is nested inside the entry so that we don't need one
+ * io context per page read.
+ */
+ entry->io_context = ctx;
+
+ /* We should only get page size requests at present. */
+ verity_inc_pending(io);
+ bio = verity_alloc_bioset(vc, GFP_NOIO, 1);
+ if (unlikely(!bio)) {
+ DMCRIT("Out of memory at bio_alloc_bioset");
+ verity_tree_read_completed(entry, -ENOMEM);
+ return -ENOMEM;
+ }
+ bio->bi_private = (void *) entry;
+ bio->bi_idx = 0;
+ bio->bi_size = vc->bht.block_size;
+ bio->bi_sector = vc->hash_start + start;
+ bio->bi_bdev = vc->hash_dev->bdev;
+ bio->bi_end_io = kverityd_io_bht_populate_end;
+ bio->bi_rw = REQ_META;
+ /* Only need to free the bio since the page is managed by bht */
+ bio->bi_destructor = verity_bio_destructor;
+ bio->bi_vcnt = 1;
+ bio->bi_io_vec->bv_offset = offset_in_page(dst);
+ bio->bi_io_vec->bv_len = to_bytes(count);
+ /* dst is guaranteed to be a page_pool allocation */
+ bio->bi_io_vec->bv_page = virt_to_page(dst);
+ /* Track that this I/O is in use. There should be no risk of the io
+ * being removed prior since this is called synchronously.
+ */
+ generic_make_request(bio);
+ return 0;
+}
+
+/* Submits an io request for each missing block of block hashes.
+ * The last one to return will then enqueue this on the io workqueue.
+ */
+static void kverityd_io_bht_populate(struct verity_io *io)
+{
+ struct verity_config *vc = io->target->private;
+ u64 block;
+
+ for (block = io->block; block < io->block + io->count; ++block) {
+ int ret = verity_tree_populate(&vc->bht, io, block);
+
+ if (ret < 0) {
+ /* verity_dec_pending will handle the error case. */
+ io->error = ret;
+ break;
+ }
+ }
+}
+
+/* Asynchronously called upon the completion of I/O issued
+ * from kverityd_src_io_read. verity_dec_pending() acts as
+ * the scheduler/flow manager.
+ */
+static void kverityd_src_io_read_end(struct bio *clone, int error)
+{
+ struct verity_io *io = clone->bi_private;
+
+ if (unlikely(!bio_flagged(clone, BIO_UPTODATE) && !error))
+ error = -EIO;
+
+ if (unlikely(error)) {
+ DMERR("Error occurred: %d (%llu, %u)",
+ error, ULL(clone->bi_sector), clone->bi_size);
+ io->error = error;
+ }
+
+ /* Release the clone which just avoids the block layer from
+ * leaving offsets, etc in unexpected states.
+ */
+ bio_put(clone);
+
+ verity_dec_pending(io);
+}
+
+/* If not yet underway, an I/O request will be issued to the vc->dev
+ * device for the data needed. It is cloned to avoid unexpected changes
+ * to the original bio struct.
+ */
+static void kverityd_src_io_read(struct verity_io *io)
+{
+ struct bio *clone;
+
+ /* Check if the read is already issued. */
+ if (io->flags & VERITY_IOFLAGS_CLONED)
+ return;
+
+ io->flags |= VERITY_IOFLAGS_CLONED;
+
+ /* Clone the bio. The block layer may modify the bvec array. */
+ clone = verity_bio_clone(io);
+ if (unlikely(!clone)) {
+ io->error = -ENOMEM;
+ return;
+ }
+
+ verity_inc_pending(io);
+
+ generic_make_request(clone);
+}
+
+/* kverityd_io services the I/O workqueue. For each pass through
+ * the I/O workqueue, a call to populate both the origin drive
+ * data and the hash tree data is made.
+ */
+static void kverityd_io(struct work_struct *work)
+{
+ struct delayed_work *dwork = container_of(work, struct delayed_work,
+ work);
+ struct verity_io *io = container_of(dwork, struct verity_io,
+ work);
+
+ /* Issue requests asynchronously. */
+ verity_inc_pending(io);
+ kverityd_src_io_read(io);
+ kverityd_io_bht_populate(io);
+ verity_dec_pending(io);
+}
+
+/* Paired with verity_dec_pending, the pending value in the io dictate the
+ * lifetime of a request and when it is ready to be processed on the
+ * workqueues.
+ */
+static void verity_inc_pending(struct verity_io *io)
+{
+ atomic_inc(&io->pending);
+}
+
+/* Block-level requests start here. */
+static int verity_map(struct dm_target *ti, struct bio *bio,
+ union map_info *map_context)
+{
+ struct verity_io *io;
+ struct verity_config *vc;
+ struct request_queue *r_queue;
+
+ if (unlikely(!ti)) {
+ DMERR("dm_target was NULL");
+ return -EIO;
+ }
+
+ vc = ti->private;
+ r_queue = bdev_get_queue(vc->dev->bdev);
+
+ if (bio_data_dir(bio) == WRITE) {
+ /* If we silently drop writes, then the VFS layer will cache
+ * the write and persist it in memory. While it doesn't change
+ * the underlying storage, it still may be contrary to the
+ * behavior expected by a verified, read-only device.
+ */
+ DMWARN_LIMIT("write request received. rejecting with -EIO.");
+ return -EIO;
+ } else {
+ /* Queue up the request to be verified */
+ io = verity_io_alloc(ti, bio);
+ if (!io) {
+ DMERR_LIMIT("Failed to allocate and init IO data");
+ return DM_MAPIO_REQUEUE;
+ }
+ INIT_DELAYED_WORK(&io->work, kverityd_io);
+ queue_delayed_work(kverityd_ioq, &io->work, 0);
+ }
+
+ return DM_MAPIO_SUBMITTED;
+}
+
+/*
+ * Non-block interfaces and device-mapper specific code
+ */
+
+/*
+ * Verity target parameters:
+ *
+ * <dev> <hash_dev> <hash_start> <block_size> <alg> <digest> <salt>
+ *
+ * version: version of the hash tree on-disk format
+ * dev: device to verify
+ * hash_dev: device hashtree is stored on
+ * hash_start: start address of hashes
+ * block_size: size of a hash block
+ * alg: hash algorithm
+ * digest: toplevel hash of the tree
+ * salt: salt
+ */
+static int verity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
+{
+ struct verity_config *vc = NULL;
+ const char *dev, *hash_dev, *alg, *digest, *salt;
+ unsigned long hash_start, block_size, version;
+ sector_t blocks;
+ int ret;
+
+ if (argc != 8) {
+ ti->error = "Invalid argument count";
+ return -EINVAL;
+ }
+
+ if (strict_strtoul(argv[0], 10, &version) ||
+ (version != 0)) {
+ ti->error = "Invalid version";
+ return -EINVAL;
+ }
+ dev = argv[1];
+ hash_dev = argv[2];
+ if (strict_strtoul(argv[3], 10, &hash_start)) {
+ ti->error = "Invalid hash_start";
+ return -EINVAL;
+ }
+ if (strict_strtoul(argv[4], 10, &block_size) ||
+ (block_size > UINT_MAX)) {
+ ti->error = "Invalid block_size";
+ return -EINVAL;
+ }
+ alg = argv[5];
+ digest = argv[6];
+ salt = argv[7];
+
+ /* The device mapper device should be setup read-only */
+ if ((dm_table_get_mode(ti->table) & ~FMODE_READ) != 0) {
+ ti->error = "Must be created readonly.";
+ return -EINVAL;
+ }
+
+ vc = kzalloc(sizeof(*vc), GFP_KERNEL);
+ if (!vc) {
+ return -EINVAL;
+ }
+
+ /* Calculate the blocks from the given device size */
+ vc->size = ti->len;
+ blocks = to_bytes(vc->size) / block_size;
+ if (verity_tree_create(&vc->bht, blocks, block_size, alg)) {
+ DMERR("failed to create required bht");
+ goto bad_bht;
+ }
+ if (verity_tree_set_digest(&vc->bht, digest)) {
+ DMERR("digest error");
+ goto bad_digest;
+ }
+ verity_tree_set_salt(&vc->bht, salt);
+ vc->bht.read_cb = kverityd_bht_read_callback;
+
+ vc->start = 0;
+ /* We only ever grab the device in read-only mode. */
+ ret = dm_get_device(ti, dev, dm_table_get_mode(ti->table), &vc->dev);
+ if (ret) {
+ DMERR("Failed to acquire device '%s': %d", dev, ret);
+ ti->error = "Device lookup failed";
+ goto bad_verity_dev;
+ }
+
+ if ((to_bytes(vc->start) % block_size) ||
+ (to_bytes(vc->size) % block_size)) {
+ ti->error = "Device must be block_size divisble/aligned";
+ goto bad_hash_start;
+ }
+
+ vc->hash_start = (sector_t)hash_start;
+
+ /*
+ * Note, dev == hash_dev is okay as long as the size of
+ * ti->len passed to device mapper does not include
+ * the hashes.
+ */
+ if (dm_get_device(ti, hash_dev,
+ dm_table_get_mode(ti->table), &vc->hash_dev)) {
+ ti->error = "Hash device lookup failed";
+ goto bad_hash_dev;
+ }
+
+ if (snprintf(vc->hash_alg, CRYPTO_MAX_ALG_NAME, "%s", alg) >=
+ CRYPTO_MAX_ALG_NAME) {
+ ti->error = "Hash algorithm name is too long";
+ goto bad_hash;
+ }
+
+ vc->io_pool = mempool_create_slab_pool(MIN_IOS, _verity_io_pool);
+ if (!vc->io_pool) {
+ ti->error = "Cannot allocate verity io mempool";
+ goto bad_slab_pool;
+ }
+
+ vc->bs = bioset_create(MIN_BIOS, 0);
+ if (!vc->bs) {
+ ti->error = "Cannot allocate verity bioset";
+ goto bad_bs;
+ }
+
+ ti->private = vc;
+
+ return 0;
+
+bad_bs:
+ mempool_destroy(vc->io_pool);
+bad_slab_pool:
+bad_hash:
+ dm_put_device(ti, vc->hash_dev);
+bad_hash_dev:
+bad_hash_start:
+ dm_put_device(ti, vc->dev);
+bad_bht:
+bad_digest:
+bad_verity_dev:
+ kfree(vc); /* hash is not secret so no need to zero */
+ return -EINVAL;
+}
+
+static void verity_dtr(struct dm_target *ti)
+{
+ struct verity_config *vc = (struct verity_config *) ti->private;
+
+ bioset_free(vc->bs);
+ mempool_destroy(vc->io_pool);
+ verity_tree_destroy(&vc->bht);
+ dm_put_device(ti, vc->hash_dev);
+ dm_put_device(ti, vc->dev);
+ kfree(vc);
+}
+
+static int verity_ioctl(struct dm_target *ti, unsigned int cmd,
+ unsigned long arg)
+{
+ struct verity_config *vc = (struct verity_config *) ti->private;
+ struct dm_dev *dev = vc->dev;
+ int r = 0;
+
+ /*
+ * Only pass ioctls through if the device sizes match exactly.
+ */
+ if (vc->start ||
+ ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT)
+ r = scsi_verify_blk_ioctl(NULL, cmd);
+
+ return r ? : __blkdev_driver_ioctl(dev->bdev, dev->mode, cmd, arg);
+}
+
+static int verity_status(struct dm_target *ti, status_type_t type,
+ char *result, unsigned int maxlen)
+{
+ struct verity_config *vc = (struct verity_config *) ti->private;
+ unsigned int sz = 0;
+ char digest[VERITY_MAX_DIGEST_SIZE * 2 + 1] = { 0 };
+ char salt[VERITY_SALT_SIZE * 2 + 1] = { 0 };
+
+ verity_tree_digest(&vc->bht, digest);
+ verity_tree_salt(&vc->bht, salt);
+
+ switch (type) {
+ case STATUSTYPE_INFO:
+ result[0] = '\0';
+ break;
+ case STATUSTYPE_TABLE:
+ DMEMIT("%s %s %llu %llu %s %s %s",
+ vc->dev->name,
+ vc->hash_dev->name,
+ ULL(vc->hash_start),
+ ULL(vc->bht.block_size),
+ vc->hash_alg,
+ digest,
+ salt);
+ break;
+ }
+ return 0;
+}
+
+static int verity_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
+ struct bio_vec *biovec, int max_size)
+{
+ struct verity_config *vc = ti->private;
+ struct request_queue *q = bdev_get_queue(vc->dev->bdev);
+
+ if (!q->merge_bvec_fn)
+ return max_size;
+
+ bvm->bi_bdev = vc->dev->bdev;
+ bvm->bi_sector = vc->start + bvm->bi_sector - ti->begin;
+
+ /* Optionally, this could just return 0 to stick to single pages. */
+ return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
+}
+
+static int verity_iterate_devices(struct dm_target *ti,
+ iterate_devices_callout_fn fn, void *data)
+{
+ struct verity_config *vc = ti->private;
+
+ return fn(ti, vc->dev, vc->start, ti->len, data);
+}
+
+static void verity_io_hints(struct dm_target *ti,
+ struct queue_limits *limits)
+{
+ struct verity_config *vc = ti->private;
+ unsigned int block_size = vc->bht.block_size;
+
+ limits->logical_block_size = block_size;
+ limits->physical_block_size = block_size;
+ blk_limits_io_min(limits, block_size);
+}
+
+static struct target_type verity_target = {
+ .name = "verity",
+ .version = {0, 1, 0},
+ .module = THIS_MODULE,
+ .ctr = verity_ctr,
+ .dtr = verity_dtr,
+ .ioctl = verity_ioctl,
+ .map = verity_map,
+ .merge = verity_merge,
+ .status = verity_status,
+ .iterate_devices = verity_iterate_devices,
+ .io_hints = verity_io_hints,
+};
+
+#define VERITY_WQ_FLAGS (WQ_CPU_INTENSIVE|WQ_HIGHPRI)
+
+static int __init verity_init(void)
+{
+ int r = -ENOMEM;
+
+ _verity_io_pool = KMEM_CACHE(verity_io, 0);
+ if (!_verity_io_pool) {
+ DMERR("failed to allocate pool verity_io");
+ goto bad_io_pool;
+ }
+
+ kverityd_ioq = alloc_workqueue("kverityd_io", VERITY_WQ_FLAGS, 1);
+ if (!kverityd_ioq) {
+ DMERR("failed to create workqueue kverityd_ioq");
+ goto bad_io_queue;
+ }
+
+ kveritydq = alloc_workqueue("kverityd", VERITY_WQ_FLAGS, 1);
+ if (!kveritydq) {
+ DMERR("failed to create workqueue kveritydq");
+ goto bad_verify_queue;
+ }
+
+ r = dm_register_target(&verity_target);
+ if (r < 0) {
+ DMERR("register failed %d", r);
+ goto register_failed;
+ }
+
+ DMINFO("version %u.%u.%u loaded", verity_target.version[0],
+ verity_target.version[1], verity_target.version[2]);
+
+ return r;
+
+register_failed:
+ destroy_workqueue(kveritydq);
+bad_verify_queue:
+ destroy_workqueue(kverityd_ioq);
+bad_io_queue:
+ kmem_cache_destroy(_verity_io_pool);
+bad_io_pool:
+ return r;
+}
+
+static void __exit verity_exit(void)
+{
+ destroy_workqueue(kveritydq);
+ destroy_workqueue(kverityd_ioq);
+
+ dm_unregister_target(&verity_target);
+ kmem_cache_destroy(_verity_io_pool);
+}
+
+module_init(verity_init);
+module_exit(verity_exit);
+
+MODULE_AUTHOR("The Chromium OS Authors <[email protected]>");
+MODULE_DESCRIPTION(DM_NAME " target for transparent disk integrity checking");
+MODULE_LICENSE("GPL");
--
1.7.7.3
Hi
This crashes if the device size is 64MiB (and sha256 hash is used).
I tested it with the userspace utility and it doesn't work with device
>= 128MiB, it fails to verify the output of the utility.
I run this (/dev/vg1/verity_long_data has 128MiB size):
./verity mode=create alg=sha256 payload=/dev/vg1/verity_long_data
hashtree=/dev/vg1/verity_long_hash
salt=1234000000000000000000000000000000000000000000000000000000000000
dm:dm bht[DEBUG] Setting block_count 32768
dm:dm bht[DEBUG] Setting depth to 3.
dm:dm bht[DEBUG] depth: 0 entries: 1
dm:dm bht[DEBUG] depth: 1 entries: 2
dm:dm bht[DEBUG] depth: 2 entries: 256
0 262144 verity payload=ROOT_DEV hashtree=HASH_DEV hashstart=262144
alg=sha256
root_hexdigest=6e46e106b288812a881a9da3f11180433f90ce264c4f1e8fa191fb40409846fb
salt=1234000000000000000000000000000000000000000000000000000000000000
dmsetup -r create verity --table "0 `blockdev --getsize
/dev/vg1/verity_long_data` verity 0 /dev/vg1/verity_long_data
/dev/vg1/verity_long_hash 0 4096 sha256
6e46e106b288812a881a9da3f11180433f90ce264c4f1e8fa191fb40409846fb
1234000000000000000000000000000000000000000000000000000000000000"
and get a lot of log messages "failed to verify hash (d=3,bi=0)"
If the device is smaller than 128MiB, it works (except for 64MiB device
where it crashes).
Mikulas
dmsetup -r create verity --table "0 131072 verity 0
/dev/vg1/verity_long_data /dev/vg1/verity_long_hash 0 4096 sha256
d821fec17e151a6e7b91c4a7a71487760185c25af49a74955a3e7a718c1f97dd
1234000000000000000000000000000000000000000000000000000000000000"
[14356.819694] ------------[ cut here ]------------
[14356.819747] kernel BUG at drivers/md/dm-verity2.c:439!
[14356.819796] invalid opcode: 0000 [#1] PREEMPT SMP
[14356.819879] CPU 5
[14356.819897] Modules linked in: dm_verity2 md5 sha1_generic cryptomgr
aead sha256_generic dm_zero dm_bufio crypto_hash crypto_algapi crypto
dm_loop dm_mod parport_pc parport powernow_k8 mperf cpufreq_stats
cpufreq_powersave cpufreq_userspace cpufreq_conservative cpufreq_ondemand
freq_table snd_usb_audio snd_pcm_oss snd_mixer_oss snd_pcm snd_timer
snd_page_alloc snd_hwdep snd_usbmidi_lib snd_rawmidi snd soundcore fuse
raid0 md_mod lm85 hwmon_vid ide_cd_mod cdrom ohci_hcd ehci_hcd sata_svw
libata serverworks ide_core usbcore floppy usb_common rtc_cmos e100 tg3
mii libphy k10temp skge hwmon button i2c_piix4 processor unix [last
unloaded: dm_verity2]
[14356.820536]
[14356.820580] Pid: 10909, comm: dmsetup Not tainted 3.2.0 #19 empty
empty/S3992-E
[14356.820680] RIP: 0010:[<ffffffffa0195908>] [<ffffffffa0195908>]
verity_ctr+0x808/0x860 [dm_verity2]
[14356.820772] RSP: 0018:ffff880141b7bcc8 EFLAGS: 00010202
[14356.820821] RAX: 0000000000000408 RBX: ffff8802afedcc28 RCX:
0000000000000002[14356.820875] RDX: 0000000000000081 RSI: ffff88044685a540
RDI: ffff8802afc4b000[14356.820929] RBP: ffff8802afedcc00 R08:
0000000000000000 R09: ffff8802afc4a000[14356.820987] R10: 0000000000000001
R11: ffffff9000000018 R12: ffffffff8147a630[14356.821041] R13:
ffff88044685a558 R14: 0000000000004000 R15: 0000000000000002[14356.821114]
FS: 00007f1c9ea567a0(0000) GS:ffff880447c80000(0000)
knlGS:0000000000000000
[14356.821199] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[14356.821257] CR2: 00007f1c9e199f80 CR3: 00000003fe98e000 CR4:
00000000000006e0[14356.821330] DR0: 0000000000000000 DR1: 0000000000000000
DR2: 0000000000000000[14356.821394] DR3: 0000000000000000 DR6:
00000000ffff0ff0 DR7: 0000000000000400[14356.821464] Process dmsetup (pid:
10909, threadinfo ffff880141b7a000, task ffff88023d8dec90)
[14356.821557] Stack:
[14356.821606] ffff880141b7bd34 0000000000000007 ffffc90012ac9040
ffff880429987340
[14356.821700] ffffc90012ac41a4 0000000000020000 ffffc90012ac419d
ffffc90012ac4162
[14356.821787] ffffc90012ac417c ffffc90012ac41e5 0000000000020000
0000000000000000
[14356.821905] Call Trace:
[14356.821965] [<ffffffffa01d2cab>] ? dm_table_add_target+0x19b/0x450
[dm_mod] [14356.822034] [<ffffffffa01d5690>] ? table_clear+0x80/0x80
[dm_mod]
[14356.822090] [<ffffffffa01d5762>] ? table_load+0xd2/0x330 [dm_mod]
[14356.822150] [<ffffffffa01d5690>] ? table_clear+0x80/0x80 [dm_mod]
[14356.822213] [<ffffffffa01d6bf9>] ? ctl_ioctl+0x159/0x2a0 [dm_mod]
[14356.822286] [<ffffffff8117351d>] ? ipc_addid+0x4d/0xd0
[14356.822338] [<ffffffffa01d6d4e>] ? dm_ctl_ioctl+0xe/0x20 [dm_mod]
[14356.822411] [<ffffffff81105a9e>] ? do_vfs_ioctl+0x8e/0x4f0
[14356.822474] [<ffffffff8110a980>] ? dput+0x20/0x230
[14356.822533] [<ffffffff810f6282>] ? fput+0x162/0x220
[14356.822590] [<ffffffff81105f49>] ? sys_ioctl+0x49/0x90
[14356.822652] [<ffffffff81313abb>] ? system_call_fastpath+0x16/0x1b
[14356.822702] Code: 38 ce 65 19 a0 e9 61 fb ff ff 48 c7 c7 d8 62 19 a0 31
c0 e8 d9 80 17 e1 48 c7 c7 28 63 19 a0 31 c0 e8 cb 80 17 e1 e9 40 fb ff ff
<0f> 0b 48 c7 c7 b8 60 19 a0 31 c0 e8 b6 80 17 e1 e9 9b fa ff ff
[14356.823081] RIP [<ffffffffa0195908>] verity_ctr+0x808/0x860
[dm_verity2]
[14356.823146] RSP <ffff880141b7bcc8>
[14356.823530] ---[ end trace 773c24b9dbd5cfff ]---
On Tue, 28 Feb 2012, Mandeep Singh Baines wrote:
> The verity target provides transparent integrity checking of block devices
> using a cryptographic digest.
>
> dm-verity is meant to be setup as part of a verified boot path. This
> may be anything ranging from a boot using tboot or trustedgrub to just
> booting from a known-good device (like a USB drive or CD).
>
> dm-verity is part of ChromeOS's verified boot path. It is used to verify
> the integrity of the root filesystem on boot. The root filesystem is
> mounted on a dm-verity partition which transparently verifies each block
> with a bootloader verified hash passed into the kernel at boot.
>
> Changes in V4:
> * Discussion over phone (Alasdair G Kergon)
> * copy _ioctl fix from dm-linear
> * verity_status format fixes to match dm conventions
> * s/dm-bht/verity_tree
> * put everything into dm-verity.c
> * ctr changed to dm conventions
> * use hex2bin
> * use conventional dm names for function
> * s/dm_//
> * for example: verity_ctr versus dm_verity_ctr
> * use per_cpu API
> Changes in V3:
> * Discussion over irc (Alasdair G Kergon)
> * Implement ioctl hook
> Changes in V2:
> * https://lkml.org/lkml/2011/11/10/85 (Steffen Klassert)
> * Use shash API instead of older hash API
>
> Signed-off-by: Will Drewry <[email protected]>
> Signed-off-by: Elly Jones <[email protected]>
> Signed-off-by: Mandeep Singh Baines <[email protected]>
> Cc: Alasdair G Kergon <[email protected]>
> Cc: Milan Broz <[email protected]>
> Cc: Olof Johansson <[email protected]>
> Cc: Steffen Klassert <[email protected]>
> Cc: Andrew Morton <[email protected]>
> Cc: Mikulas Patocka <[email protected]>
> Cc: [email protected]
> ---
> Documentation/device-mapper/verity.txt | 149 ++++
> drivers/md/Kconfig | 16 +
> drivers/md/Makefile | 1 +
> drivers/md/dm-verity.c | 1411 ++++++++++++++++++++++++++++++++
> 4 files changed, 1577 insertions(+), 0 deletions(-)
> create mode 100644 Documentation/device-mapper/verity.txt
> create mode 100644 drivers/md/dm-verity.c
>
> diff --git a/Documentation/device-mapper/verity.txt b/Documentation/device-mapper/verity.txt
> new file mode 100644
> index 0000000..b631f12
> --- /dev/null
> +++ b/Documentation/device-mapper/verity.txt
> @@ -0,0 +1,149 @@
> +dm-verity
> +==========
> +
> +Device-Mapper's "verity" target provides transparent integrity checking of
> +block devices using a cryptographic digest provided by the kernel crypto API.
> +This target is read-only.
> +
> +Parameters:
> + <version> <dev> <hash_dev> <hash_start> <block_size> <alg> <digest> <salt>
> +
> +<version>
> + This is the version number of the on-disk format. Currently, there is
> + only version 0.
> +
> +<dev>
> + This is the device that is going to be integrity checked. It may be
> + a subset of the full device as specified to dmsetup (start sector and count)
> + It may be specified as a path, like /dev/sdaX, or a device number,
> + <major>:<minor>.
> +
> +<hash_dev>
> + This is the device that that supplies the hash tree data. It may be
> + specified similarly to the device path and may be the same device. If the
> + same device is used, the hash offset should be outside of the dm-verity
> + configured device size.
> +
> +<hash_start>
> + This is the offset, in 512-byte sectors, from the start of hash_dev to
> + the root block of the hash tree.
> +
> +<block_size>
> + The size of a hash block. Also, the size of a block to be hashed.
> +
> +<alg>
> + The cryptographic hash algorithm used for this device. This should
> + be the name of the algorithm, like "sha1".
> +
> +<digest>
> + The hexadecimal encoding of the cryptographic hash of all of the
> + neighboring nodes at the first level of the tree. This hash should be
> + trusted as there is no other authenticity beyond this point.
> +
> +<salt>
> + The hexadecimal encoding of the salt value.
> +
> +Theory of operation
> +===================
> +
> +dm-verity is meant to be setup as part of a verified boot path. This
> +may be anything ranging from a boot using tboot or trustedgrub to just
> +booting from a known-good device (like a USB drive or CD).
> +
> +When a dm-verity device is configured, it is expected that the caller
> +has been authenticated in some way (cryptographic signatures, etc).
> +After instantiation, all hashes will be verified on-demand during
> +disk access. If they cannot be verified up to the root node of the
> +tree, the root hash, then the I/O will fail. This should identify
> +tampering with any data on the device and the hash data.
> +
> +Cryptographic hashes are used to assert the integrity of the device on a
> +per-block basis. This allows for a lightweight hash computation on first read
> +into the page cache. Block hashes are stored linearly aligned to the nearest
> +block the size of a page.
> +
> +Hash Tree
> +---------
> +
> +Each node in the tree is a cryptographic hash. If it is a leaf node, the hash
> +is of some block data on disk. If it is an intermediary node, then the hash is
> +of a number of child nodes.
> +
> +Each entry in the tree is a collection of neighboring nodes that fit in one
> +block. The number is determined based on block_size and the size of the
> +selected cryptographic digest algorithm. The hashes are linearly ordered in
> +this entry and any unaligned trailing space is ignored but included when
> +calculating the parent node.
> +
> +The tree looks something like:
> +
> +alg = sha256, num_blocks = 32768, block_size = 4096
> +
> + [ root ]
> + / . . . \
> + [entry_0] [entry_1]
> + / . . . \ . . . \
> + [entry_0_0] . . . [entry_0_127] . . . . [entry_1_127]
> + / ... \ / . . . \ / \
> + blk_0 ... blk_127 blk_16256 blk_16383 blk_32640 . . . blk_32767
> +
> +On-disk format
> +==============
> +
> +Below is the recommended on-disk format. The verity kernel code does not
> +read the on-disk header. It only reads the hash blocks which directly
> +follow the header. It is expected that a user-space tool will verify the
> +integrity of the verity_header and then call dm_setup with the correct
> +parameters. Alternatively, the header can be omitted and the dm_setup
> +parameters can be passed via the kernel command-line in a rooted chain
> +of trust where the command-line is verified.
> +
> +The on-disk format is especially useful in cases where the hash blocks
> +are on a separate partition. The magic number allows easy identification
> +of the partition contents. Alternatively, the hash blocks can be stored
> +in the same partition as the data to be verified. In such a configuration
> +the filesystem on the partition would be sized a little smaller than
> +the full-partition, leaving room for the hash blocks.
> +
> +struct verity_header {
> + uint64_t magic = 0x7665726974790a00;
> + uint32_t version;
> + uint32_t block_size;
> + char digest[128]; /* in hex-ascii, null-terminated or 128-bytes */
> + char salt[128]; /* in hex-ascii, null-terminated or 128-bytes */
> +}
> +
> +struct verity_header_block {
> + struct verity_header;
> + char unused[block_size - sizeof(struct verity_header) - sizeof(sig)];
> + char sig[128]; /* in hex-ascii, null-terminated or 128-bytes */
> +}
> +
> +Directly following the header are the hash blocks which are stored a depth
> +at a time (starting from the root), sorted in order of increasing index.
> +
> +Usage
> +=====
> +
> +The API provides mechanisms for reading and verifying a tree. When reading, all
> +required data for the hash tree should be populated for a block before
> +attempting a verify. This can be done by calling dm_bht_populate(). When all
> +data is ready, a call to dm_bht_verify_block() with the expected hash value will
> +perform both the direct block hash check and the hashes of the parent and
> +neighboring nodes where needed to ensure validity up to the root hash. Note,
> +dm_bht_set_root_hexdigest() should be called before any verification attempts
> +occur.
> +
> +Example
> +=======
> +
> +Setup a device;
> +[[
> + dmsetup create vroot --table \
> + "0 204800 verity /dev/sda1 /dev/sda2 alg=sha1 "\
> + "root_hexdigest=9f74809a2ee7607b16fcc70d9399a4de9725a727"
> +]]
> +
> +A command line tool is available to compute the hash tree and return the
> +root hash value.
> + http://git.chromium.org/cgi-bin/gitweb.cgi?p=dm-verity.git;a=tree
> diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
> index faa4741..b8bb690 100644
> --- a/drivers/md/Kconfig
> +++ b/drivers/md/Kconfig
> @@ -370,4 +370,20 @@ config DM_FLAKEY
> ---help---
> A target that intermittently fails I/O for debugging purposes.
>
> +config DM_VERITY
> + tristate "Verity target support"
> + depends on BLK_DEV_DM
> + select CRYPTO
> + select CRYPTO_HASH
> + ---help---
> + This device-mapper target allows you to create a device that
> + transparently integrity checks the data on it. You'll need to
> + activate the digests you're going to use in the cryptoapi
> + configuration.
> +
> + To compile this code as a module, choose M here: the module will
> + be called dm-verity.
> +
> + If unsure, say N.
> +
> endif # MD
> diff --git a/drivers/md/Makefile b/drivers/md/Makefile
> index 046860c..70a29af 100644
> --- a/drivers/md/Makefile
> +++ b/drivers/md/Makefile
> @@ -39,6 +39,7 @@ obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o
> obj-$(CONFIG_DM_PERSISTENT_DATA) += persistent-data/
> obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o dm-region-hash.o
> obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
> +obj-$(CONFIG_DM_VERITY) += dm-verity.o
> obj-$(CONFIG_DM_ZERO) += dm-zero.o
> obj-$(CONFIG_DM_RAID) += dm-raid.o
> obj-$(CONFIG_DM_THIN_PROVISIONING) += dm-thin-pool.o
> diff --git a/drivers/md/dm-verity.c b/drivers/md/dm-verity.c
> new file mode 100644
> index 0000000..87b7958
> --- /dev/null
> +++ b/drivers/md/dm-verity.c
> @@ -0,0 +1,1411 @@
> +/*
> + * Originally based on dm-crypt.c,
> + * Copyright (C) 2003 Christophe Saout <[email protected]>
> + * Copyright (C) 2004 Clemens Fruhwirth <[email protected]>
> + * Copyright (C) 2006-2008 Red Hat, Inc. All rights reserved.
> + * Copyright (C) 2012 The Chromium OS Authors <[email protected]>
> + * All Rights Reserved.
> + *
> + * This file is released under the GPLv2.
> + *
> + * Implements a verifying transparent block device.
> + * See Documentation/device-mapper/dm-verity.txt
> + */
> +#include <crypto/hash.h>
> +#include <linux/atomic.h>
> +#include <linux/bio.h>
> +#include <linux/blkdev.h>
> +#include <linux/genhd.h>
> +#include <linux/init.h>
> +#include <linux/kernel.h>
> +#include <linux/mempool.h>
> +#include <linux/module.h>
> +#include <linux/workqueue.h>
> +#include <linux/device-mapper.h>
> +
> +
> +#define DM_MSG_PREFIX "verity"
> +
> +
> +/* Helper for printing sector_t */
> +#define ULL(x) ((unsigned long long)(x))
> +
> +#define MIN_IOS 32
> +#define MIN_BIOS (MIN_IOS * 2)
> +
> +/* To avoid allocating memory for digest tests, we just setup a
> + * max to use for now.
> + */
> +#define VERITY_MAX_DIGEST_SIZE 64 /* Supports up to 512-bit digests */
> +#define VERITY_SALT_SIZE 32 /* 256 bits of salt is a lot */
> +
> +/* UNALLOCATED, PENDING, READY, and VERIFIED are valid states. All other
> + * values are entry-related return codes.
> + */
> +#define VERITY_TREE_ENTRY_VERIFIED 8 /* 'nodes' checked against parent */
> +#define VERITY_TREE_ENTRY_READY 4 /* 'nodes' is loaded and available */
> +#define VERITY_TREE_ENTRY_PENDING 2 /* 'nodes' is being loaded */
> +#define VERITY_TREE_ENTRY_UNALLOCATED 0 /* untouched */
> +#define VERITY_TREE_ENTRY_ERROR -1 /* entry is unsuitable for use */
> +#define VERITY_TREE_ENTRY_ERROR_IO -2 /* I/O error on load */
> +
> +/* Additional possible return codes */
> +#define VERITY_TREE_ENTRY_ERROR_MISMATCH -3 /* Digest mismatch */
> +
> +
> +/* verity_tree_entry
> + * Contains verity_tree->node_count tree nodes at a given tree depth.
> + * state is used to transactionally assure that data is paged in
> + * from disk. Unless verity_tree kept running crypto contexts for each
> + * level, we need to load in the data for on-demand verification.
> + */
> +struct verity_tree_entry {
> + atomic_t state; /* see defines */
> + /* Keeping an extra pointer per entry wastes up to ~33k of
> + * memory if a 1m blocks are used (or 66 on 64-bit arch)
> + */
> + void *io_context; /* Reserve a pointer for use during io */
> + /* data should only be non-NULL if fully populated. */
> + void *nodes; /* The hash data used to verify the children.
> + * Guaranteed to be page-aligned.
> + */
> +};
> +
> +/* verity_tree_level
> + * Contains an array of entries which represent a page of hashes where
> + * each hash is a node in the tree at the given tree depth/level.
> + */
> +struct verity_tree_level {
> + struct verity_tree_entry *entries; /* array of entries of tree nodes */
> + unsigned int count; /* number of entries at this level */
> + sector_t sector; /* starting sector for this level */
> +};
> +
> +/* opaque context, start, databuf, sector_count */
> +typedef int(*verity_tree_callback)(void *, /* external context */
> + sector_t, /* start sector */
> + u8 *, /* destination page */
> + sector_t, /* num sectors */
> + struct verity_tree_entry *);
> +/* verity_tree - Device mapper block hash tree
> + * verity_tree provides a fixed interface for comparing data blocks
> + * against a cryptographic hashes stored in a hash tree. It
> + * optimizes the tree structure for storage on disk.
> + *
> + * The tree is built from the bottom up. A collection of data,
> + * external to the tree, is hashed and these hashes are stored
> + * as the blocks in the tree. For some number of these hashes,
> + * a parent node is created by hashing them. These steps are
> + * repeated.
> + */
> +struct verity_tree {
> + /* Configured values */
> + int depth; /* Depth of the tree including the root */
> + unsigned int block_count; /* Number of blocks hashed */
> + unsigned int block_size; /* Size of a hash block */
> + char hash_alg[CRYPTO_MAX_ALG_NAME];
> + u8 salt[VERITY_SALT_SIZE];
> +
> + /* Computed values */
> + unsigned int node_count; /* Data size (in hashes) for each entry */
> + unsigned int node_count_shift; /* first bit set - 1 */
> + /*
> + * There is one per CPU so that verified can be simultaneous.
> + * Access through per_cpu_ptr() only
> + */
> + struct shash_desc * __percpu *hash_desc; /* Container for hash alg */
> + unsigned int digest_size;
> + sector_t sectors; /* Number of disk sectors used */
> +
> + /* bool verified; Full tree is verified */
> + u8 digest[VERITY_MAX_DIGEST_SIZE];
> + struct verity_tree_level *levels; /* in reverse order */
> + /* Callback for reading from the hash device */
> + verity_tree_callback read_cb;
> +};
> +
> +/* per-requested-bio private data */
> +enum verity_io_flags {
> + VERITY_IOFLAGS_CLONED = 0x1, /* original bio has been cloned */
> +};
> +
> +struct verity_io {
> + struct dm_target *target;
> + struct bio *bio;
> + struct delayed_work work;
> + unsigned int flags;
> +
> + int error;
> + atomic_t pending;
> +
> + u64 block; /* aligned block index */
> + u64 count; /* aligned count in blocks */
> +};
> +
> +struct verity_config {
> + struct dm_dev *dev;
> + sector_t start;
> + sector_t size;
> +
> + struct dm_dev *hash_dev;
> + sector_t hash_start;
> +
> + struct verity_tree bht;
> +
> + /* Pool required for io contexts */
> + mempool_t *io_pool;
> + /* Pool and bios required for making sure that backing device reads are
> + * in PAGE_SIZE increments.
> + */
> + struct bio_set *bs;
> +
> + char hash_alg[CRYPTO_MAX_ALG_NAME];
> +};
> +
> +
> +static struct kmem_cache *_verity_io_pool;
> +static struct workqueue_struct *kveritydq, *kverityd_ioq;
> +
> +
> +static void kverityd_verify(struct work_struct *work);
> +static void kverityd_io(struct work_struct *work);
> +static void kverityd_io_bht_populate(struct verity_io *io);
> +static void kverityd_io_bht_populate_end(struct bio *, int error);
> +
> +
> +/*
> + * Utilities
> + */
> +
> +static void bin2hex(char *dst, const u8 *src, size_t count)
> +{
> + while (count-- > 0) {
> + sprintf(dst, "%02hhx", (int)*src);
> + dst += 2;
> + src++;
> + }
> +}
> +
> +/*
> + * Verity Tree
> + */
> +
> +/* Functions for converting indices to nodes. */
> +
> +static inline unsigned int verity_tree_get_level_shift(struct verity_tree *bht,
> + int depth)
> +{
> + return (bht->depth - depth) * bht->node_count_shift;
> +}
> +
> +/* For the given depth, this is the entry index. At depth+1 it is the node
> + * index for depth.
> + */
> +static inline unsigned int verity_tree_index_at_level(struct verity_tree *bht,
> + int depth,
> + unsigned int leaf)
> +{
> + return leaf >> verity_tree_get_level_shift(bht, depth);
> +}
> +
> +static inline struct verity_tree_entry *verity_tree_get_entry(
> + struct verity_tree *bht,
> + int depth,
> + unsigned int block)
> +{
> + unsigned int index = verity_tree_index_at_level(bht, depth, block);
> + struct verity_tree_level *level = &bht->levels[depth];
> +
> + return &level->entries[index];
> +}
> +
> +static inline void *verity_tree_get_node(struct verity_tree *bht,
> + struct verity_tree_entry *entry,
> + int depth, unsigned int block)
> +{
> + unsigned int index = verity_tree_index_at_level(bht, depth, block);
> + unsigned int node_index = index % bht->node_count;
> +
> + return entry->nodes + (node_index * bht->digest_size);
> +}
> +/**
> + * verity_tree_compute_hash: hashes a page of data
> + */
> +static int verity_tree_compute_hash(struct verity_tree *vt, struct page *pg,
> + unsigned int offset, u8 *digest)
> +{
> + struct shash_desc *hash_desc;
> + void *data;
> + int err;
> +
> + hash_desc = *per_cpu_ptr(vt->hash_desc, smp_processor_id());
> +
> + if (crypto_shash_init(hash_desc)) {
> + DMCRIT("failed to reinitialize crypto hash (proc:%d)",
> + smp_processor_id());
> + return -EINVAL;
> + }
> + data = kmap_atomic(pg);
> + err = crypto_shash_update(hash_desc, data + offset, PAGE_SIZE);
> + kunmap_atomic(data);
> + if (err) {
> + DMCRIT("crypto_hash_update failed");
> + return -EINVAL;
> + }
> + if (crypto_shash_update(hash_desc, vt->salt, sizeof(vt->salt))) {
> + DMCRIT("crypto_hash_update failed");
> + return -EINVAL;
> + }
> + if (crypto_shash_final(hash_desc, digest)) {
> + DMCRIT("crypto_hash_final failed");
> + return -EINVAL;
> + }
> +
> + return 0;
> +}
> +
> +static int verity_tree_initialize_entries(struct verity_tree *vt)
> +{
> + /* last represents the index of the last digest store in the tree.
> + * By walking the tree with that index, it is possible to compute the
> + * total number of entries at each level.
> + *
> + * Since each entry will contain up to |node_count| nodes of the tree,
> + * it is possible that the last index may not be at the end of a given
> + * entry->nodes. In that case, it is assumed the value is padded.
> + *
> + * Note, we treat both the tree root (1 hash) and the tree leaves
> + * independently from the vt data structures. Logically, the root is
> + * depth=-1 and the block layer level is depth=vt->depth
> + */
> + unsigned int last = vt->block_count;
> + int depth;
> +
> + /* check that the largest level->count can't result in an int overflow
> + * on allocation or sector calculation.
> + */
> + if (((last >> vt->node_count_shift) + 1) >
> + UINT_MAX / max((unsigned int)sizeof(struct verity_tree_entry),
> + (unsigned int)to_sector(vt->block_size))) {
> + DMCRIT("required entries %u is too large", last + 1);
> + return -EINVAL;
> + }
> +
> + /* Track the current sector location for each level so we don't have to
> + * compute it during traversals.
> + */
> + vt->sectors = 0;
> + for (depth = 0; depth < vt->depth; ++depth) {
> + struct verity_tree_level *level = &vt->levels[depth];
> +
> + level->count = verity_tree_index_at_level(vt, depth, last) + 1;
> + level->entries = (struct verity_tree_entry *)
> + kcalloc(level->count,
> + sizeof(struct verity_tree_entry),
> + GFP_KERNEL);
> + if (!level->entries) {
> + DMERR("failed to allocate entries for depth %d", depth);
> + return -ENOMEM;
> + }
> + level->sector = vt->sectors;
> + vt->sectors += level->count * to_sector(vt->block_size);
> + }
> +
> + return 0;
> +}
> +
> +/**
> + * verity_tree_create - prepares @vt for us
> + * @vt: pointer to a verity_tree_create()d vt
> + * @depth: tree depth without the root; including block hashes
> + * @block_count: the number of block hashes / tree leaves
> + * @alg_name: crypto hash algorithm name
> + *
> + * Returns 0 on success.
> + *
> + * Callers can offset into devices by storing the data in the io callbacks.
> + */
> +static int verity_tree_create(struct verity_tree *vt, unsigned int block_count,
> + unsigned int block_size, const char *alg_name)
> +{
> + struct crypto_shash *tfm;
> + int size, cpu, status = 0;
> +
> + vt->block_size = block_size;
> + /* Verify that PAGE_SIZE >= block_size >= SECTOR_SIZE. */
> + if ((block_size > PAGE_SIZE) ||
> + (PAGE_SIZE % block_size) ||
> + (to_sector(block_size) == 0))
> + return -EINVAL;
> +
> + tfm = crypto_alloc_shash(alg_name, 0, 0);
> + if (IS_ERR(tfm)) {
> + DMERR("failed to allocate crypto hash '%s'", alg_name);
> + return -ENOMEM;
> + }
> + size = sizeof(struct shash_desc) + crypto_shash_descsize(tfm);
> +
> + vt->hash_desc = alloc_percpu(struct shash_desc *);
> + if (!vt->hash_desc) {
> + DMERR("Failed to allocate per cpu hash_desc");
> + status = -ENOMEM;
> + goto bad_per_cpu;
> + }
> +
> + /* Pre-allocate per-cpu crypto contexts to avoid having to
> + * kmalloc/kfree a context for every hash operation.
> + */
> + for_each_possible_cpu(cpu) {
> + struct shash_desc *hash_desc = kmalloc(size, GFP_KERNEL);
> +
> + *per_cpu_ptr(vt->hash_desc, cpu) = hash_desc;
> + if (!hash_desc) {
> + DMERR("failed to allocate crypto hash contexts");
> + status = -ENOMEM;
> + goto bad_hash_alloc;
> + }
> + hash_desc->tfm = tfm;
> + hash_desc->flags = 0x0;
> + }
> + vt->digest_size = crypto_shash_digestsize(tfm);
> + /* We expect to be able to pack >=2 hashes into a block */
> + if (block_size / vt->digest_size < 2) {
> + DMERR("too few hashes fit in a block");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + if (vt->digest_size > VERITY_MAX_DIGEST_SIZE) {
> + DMERR("VERITY_MAX_DIGEST_SIZE too small for digest");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + /* Configure the tree */
> + vt->block_count = block_count;
> + if (block_count == 0) {
> + DMERR("block_count must be non-zero");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + /* Each verity_tree_entry->nodes is one block. The node code tracks
> + * how many nodes fit into one entry where a node is a single
> + * hash (message digest).
> + */
> + vt->node_count_shift = fls(block_size / vt->digest_size) - 1;
> + /* Round down to the nearest power of two. This makes indexing
> + * into the tree much less painful.
> + */
> + vt->node_count = 1 << vt->node_count_shift;
> +
> + /* This is unlikely to happen, but with 64k pages, who knows. */
> + if (vt->node_count > UINT_MAX / vt->digest_size) {
> + DMERR("node_count * hash_len exceeds UINT_MAX!");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + vt->depth = DIV_ROUND_UP(fls(block_count - 1), vt->node_count_shift);
> +
> + /* Ensure that we can safely shift by this value. */
> + if (vt->depth * vt->node_count_shift >= sizeof(unsigned int) * 8) {
> + DMERR("specified depth and node_count_shift is too large");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + /* Allocate levels. Each level of the tree may have an arbitrary number
> + * of verity_tree_entry structs. Each entry contains node_count nodes.
> + * Each node in the tree is a cryptographic digest of either node_count
> + * nodes on the subsequent level or of a specific block on disk.
> + */
> + vt->levels = (struct verity_tree_level *)
> + kcalloc(vt->depth,
> + sizeof(struct verity_tree_level), GFP_KERNEL);
> + if (!vt->levels) {
> + DMERR("failed to allocate tree levels");
> + status = -ENOMEM;
> + goto bad_level_alloc;
> + }
> +
> + vt->read_cb = NULL;
> +
> + status = verity_tree_initialize_entries(vt);
> + if (status)
> + goto bad_entries_alloc;
> +
> + /* We compute depth such that there is only be 1 block at level 0. */
> + BUG_ON(vt->levels[0].count != 1);
> +
> + return 0;
> +
> +bad_entries_alloc:
> + while (vt->depth-- > 0)
> + kfree(vt->levels[vt->depth].entries);
> + kfree(vt->levels);
> +bad_level_alloc:
> +bad_arg:
> +bad_hash_alloc:
> + for_each_possible_cpu(cpu)
> + if (*per_cpu_ptr(vt->hash_desc, cpu))
> + kfree(*per_cpu_ptr(vt->hash_desc, cpu));
> + free_percpu(vt->hash_desc);
> +bad_per_cpu:
> + crypto_free_shash(tfm);
> + return status;
> +}
> +
> +/**
> + * verity_tree_read_completed
> + * @entry: pointer to the entry that's been loaded
> + * @status: I/O status. Non-zero is failure.
> + * MUST always be called after a read_cb completes.
> + */
> +static void verity_tree_read_completed(struct verity_tree_entry *entry,
> + int status)
> +{
> + if (status) {
> + DMCRIT("an I/O error occurred while reading entry");
> + atomic_set(&entry->state, VERITY_TREE_ENTRY_ERROR_IO);
> + return;
> + }
> + BUG_ON(atomic_read(&entry->state) != VERITY_TREE_ENTRY_PENDING);
> + atomic_set(&entry->state, VERITY_TREE_ENTRY_READY);
> +}
> +
> +/**
> + * verity_tree_verify_block - checks that all path nodes for @block are valid
> + * @vt: pointer to a verity_tree_create()d vt
> + * @block: specific block data is expected from
> + * @pg: page holding the block data
> + * @offset: offset into the page
> + *
> + * Returns 0 on success, VERITY_TREE_ENTRY_ERROR_MISMATCH on error.
> + */
> +static int verity_tree_verify_block(struct verity_tree *vt, unsigned int block,
> + struct page *pg, unsigned int offset)
> +{
> + int state, depth = vt->depth;
> + u8 digest[VERITY_MAX_DIGEST_SIZE];
> + struct verity_tree_entry *entry;
> + void *node;
> +
> + do {
> + /* Need to check that the hash of the current block is accurate
> + * in its parent.
> + */
> + entry = verity_tree_get_entry(vt, depth - 1, block);
> + state = atomic_read(&entry->state);
> + /* This call is only safe if all nodes along the path
> + * are already populated (i.e. READY) via verity_tree_populate.
> + */
> + BUG_ON(state < VERITY_TREE_ENTRY_READY);
> + node = verity_tree_get_node(vt, entry, depth, block);
> +
> + if (verity_tree_compute_hash(vt, pg, offset, digest) ||
> + memcmp(digest, node, vt->digest_size))
> + goto mismatch;
> +
> + /* Keep the containing block of hashes to be verified in the
> + * next pass.
> + */
> + pg = virt_to_page(entry->nodes);
> + offset = offset_in_page(entry->nodes);
> + } while (--depth > 0 && state != VERITY_TREE_ENTRY_VERIFIED);
> +
> + if (depth == 0 && state != VERITY_TREE_ENTRY_VERIFIED) {
> + if (verity_tree_compute_hash(vt, pg, offset, digest) ||
> + memcmp(digest, vt->digest, vt->digest_size))
> + goto mismatch;
> + atomic_set(&entry->state, VERITY_TREE_ENTRY_VERIFIED);
> + }
> +
> + /* Mark path to leaf as verified. */
> + for (depth++; depth < vt->depth; depth++) {
> + entry = verity_tree_get_entry(vt, depth, block);
> + /* At this point, entry can only be in VERIFIED or READY state.
> + * So it is safe to use atomic_set instead of atomic_cmpxchg.
> + */
> + atomic_set(&entry->state, VERITY_TREE_ENTRY_VERIFIED);
> + }
> +
> + return 0;
> +
> +mismatch:
> + DMERR_LIMIT("verify_path: failed to verify hash (d=%d,bi=%u)",
> + depth, block);
> + return VERITY_TREE_ENTRY_ERROR_MISMATCH;
> +}
> +
> +/**
> + * verity_tree_is_populated - check that nodes needed to verify a given
> + * block are all ready
> + * @vt: pointer to a verity_tree_create()d vt
> + * @block: specific block data is expected from
> + *
> + * Callers may wish to call verity_tree_is_populated() when checking an io
> + * for which entries were already pending.
> + */
> +static bool verity_tree_is_populated(struct verity_tree *vt, unsigned int block)
> +{
> + int depth;
> +
> + for (depth = vt->depth - 1; depth >= 0; depth--) {
> + struct verity_tree_entry *entry;
> + entry = verity_tree_get_entry(vt, depth, block);
> + if (atomic_read(&entry->state) < VERITY_TREE_ENTRY_READY)
> + return false;
> + }
> +
> + return true;
> +}
> +
> +/**
> + * verity_tree_populate - reads entries from disk needed to verify a given block
> + * @vt: pointer to a verity_tree_create()d vt
> + * @ctx: context used for all read_cb calls on this request
> + * @block: specific block data is expected from
> + *
> + * Returns negative value on error. Returns 0 on success.
> + */
> +static int verity_tree_populate(struct verity_tree *vt, void *ctx,
> + unsigned int block)
> +{
> + int depth, state;
> +
> + BUG_ON(block >= vt->block_count);
> +
> + for (depth = vt->depth - 1; depth >= 0; --depth) {
> + unsigned int index;
> + struct verity_tree_level *level;
> + struct verity_tree_entry *entry;
> +
> + index = verity_tree_index_at_level(vt, depth, block);
> + level = &vt->levels[depth];
> + entry = verity_tree_get_entry(vt, depth, block);
> + state = atomic_cmpxchg(&entry->state,
> + VERITY_TREE_ENTRY_UNALLOCATED,
> + VERITY_TREE_ENTRY_PENDING);
> + if (state == VERITY_TREE_ENTRY_VERIFIED)
> + break;
> + if (state <= VERITY_TREE_ENTRY_ERROR)
> + goto error_state;
> + if (state != VERITY_TREE_ENTRY_UNALLOCATED)
> + continue;
> +
> + /* Current entry is claimed for allocation and loading */
> + entry->nodes = kmalloc(vt->block_size, GFP_NOIO);
> + if (!entry->nodes)
> + goto nomem;
> +
> + vt->read_cb(ctx,
> + level->sector + to_sector(index * vt->block_size),
> + entry->nodes, to_sector(vt->block_size), entry);
> + }
> +
> + return 0;
> +
> +error_state:
> + DMCRIT("block %u at depth %d is in an error state", block, depth);
> + return -EPERM;
> +
> +nomem:
> + DMCRIT("failed to allocate memory for entry->nodes");
> + return -ENOMEM;
> +}
> +
> +/**
> + * verity_tree_destroy - cleans up all memory used by @vt
> + * @vt: pointer to a verity_tree_create()d vt
> + */
> +static void verity_tree_destroy(struct verity_tree *vt)
> +{
> + int depth, cpu;
> +
> + for (depth = 0; depth < vt->depth; depth++) {
> + struct verity_tree_entry *entry = vt->levels[depth].entries;
> + struct verity_tree_entry *entry_end = entry +
> + vt->levels[depth].count;
> + for (; entry < entry_end; ++entry)
> + kfree(entry->nodes);
> + kfree(vt->levels[depth].entries);
> + }
> + kfree(vt->levels);
> + crypto_free_shash((*per_cpu_ptr(vt->hash_desc, 0))->tfm);
> + for_each_possible_cpu(cpu)
> + kfree(*per_cpu_ptr(vt->hash_desc, cpu));
> +}
> +
> +/*
> + * Verity Tree Accessors
> + */
> +
> +/**
> + * verity_tree_set_digest - sets an unverified root digest hash from hex
> + * @vt: pointer to a verity_tree_create()d vt
> + * @digest: string containing the digest in hex
> + * Returns non-zero on error.
> + */
> +static int verity_tree_set_digest(struct verity_tree *vt, const char *digest)
> +{
> + /* Make sure we have at least the bytes expected */
> + if (strnlen((char *)digest, vt->digest_size * 2) !=
> + vt->digest_size * 2) {
> + DMERR("root digest length does not match hash algorithm");
> + return -1;
> + }
> + return hex2bin(vt->digest, digest, vt->digest_size);
> +}
> +
> +/**
> + * verity_tree_digest - returns root digest in hex
> + * @vt: pointer to a verity_tree_create()d vt
> + * @digest: buffer to put into, must be of length VERITY_SALT_SIZE * 2 + 1.
> + */
> +int verity_tree_digest(struct verity_tree *vt, char *digest)
> +{
> + bin2hex(digest, vt->digest, vt->digest_size);
> + return 0;
> +}
> +
> +/**
> + * verity_tree_set_salt - sets the salt
> + * @vt: pointer to a verity_tree_create()d vt
> + * @salt: string containing the salt in hex
> + * Returns non-zero on error.
> + */
> +int verity_tree_set_salt(struct verity_tree *vt, const char *salt)
> +{
> + size_t saltlen = min(strlen(salt) / 2, sizeof(vt->salt));
> + memset(vt->salt, 0, sizeof(vt->salt));
> + return hex2bin(vt->salt, salt, saltlen);
> +}
> +
> +
> +/**
> + * verity_tree_salt - returns the salt in hex
> + * @vt: pointer to a verity_tree_create()d vt
> + * @salt: buffer to put salt into, of length VERITY_SALT_SIZE * 2 + 1.
> + */
> +int verity_tree_salt(struct verity_tree *vt, char *salt)
> +{
> + bin2hex(salt, vt->salt, sizeof(vt->salt));
> + return 0;
> +}
> +
> +/*
> + * Allocation and utility functions
> + */
> +
> +static void kverityd_src_io_read_end(struct bio *clone, int error);
> +
> +/* Shared destructor for all internal bios */
> +static void verity_bio_destructor(struct bio *bio)
> +{
> + struct verity_io *io = bio->bi_private;
> + struct verity_config *vc = io->target->private;
> + bio_free(bio, vc->bs);
> +}
> +
> +static struct bio *verity_alloc_bioset(struct verity_config *vc, gfp_t gfp_mask,
> + int nr_iovecs)
> +{
> + return bio_alloc_bioset(gfp_mask, nr_iovecs, vc->bs);
> +}
> +
> +static struct verity_io *verity_io_alloc(struct dm_target *ti,
> + struct bio *bio)
> +{
> + struct verity_config *vc = ti->private;
> + sector_t sector = bio->bi_sector - ti->begin;
> + struct verity_io *io;
> +
> + io = mempool_alloc(vc->io_pool, GFP_NOIO);
> + if (unlikely(!io))
> + return NULL;
> + io->flags = 0;
> + io->target = ti;
> + io->bio = bio;
> + io->error = 0;
> +
> + /* Adjust the sector by the virtual starting sector */
> + io->block = to_bytes(sector) / vc->bht.block_size;
> + io->count = bio->bi_size / vc->bht.block_size;
> +
> + atomic_set(&io->pending, 0);
> +
> + return io;
> +}
> +
> +static struct bio *verity_bio_clone(struct verity_io *io)
> +{
> + struct verity_config *vc = io->target->private;
> + struct bio *bio = io->bio;
> + struct bio *clone = verity_alloc_bioset(vc, GFP_NOIO, bio->bi_max_vecs);
> +
> + if (!clone)
> + return NULL;
> +
> + __bio_clone(clone, bio);
> + clone->bi_private = io;
> + clone->bi_end_io = kverityd_src_io_read_end;
> + clone->bi_bdev = vc->dev->bdev;
> + clone->bi_sector += vc->start - io->target->begin;
> + clone->bi_destructor = verity_bio_destructor;
> +
> + return clone;
> +}
> +
> +/*
> + * Reverse flow of requests into the device.
> + *
> + * (Start at the bottom with verity_map and work your way upward).
> + */
> +
> +static void verity_inc_pending(struct verity_io *io);
> +
> +static void verity_return_bio_to_caller(struct verity_io *io)
> +{
> + struct verity_config *vc = io->target->private;
> +
> + if (io->error)
> + io->error = -EIO;
> +
> + bio_endio(io->bio, io->error);
> + mempool_free(io, vc->io_pool);
> +}
> +
> +/* Check for any missing bht hashes. */
> +static bool verity_is_bht_populated(struct verity_io *io)
> +{
> + struct verity_config *vc = io->target->private;
> + u64 block;
> +
> + for (block = io->block; block < io->block + io->count; ++block)
> + if (!verity_tree_is_populated(&vc->bht, block))
> + return false;
> +
> + return true;
> +}
> +
> +/* verity_dec_pending manages the lifetime of all verity_io structs.
> + * Non-bug error handling is centralized through this interface and
> + * all passage from workqueue to workqueue.
> + */
> +static void verity_dec_pending(struct verity_io *io)
> +{
> + if (!atomic_dec_and_test(&io->pending))
> + goto done;
> +
> + if (unlikely(io->error))
> + goto io_error;
> +
> + /* I/Os that were pending may now be ready */
> + if (verity_is_bht_populated(io)) {
> + INIT_DELAYED_WORK(&io->work, kverityd_verify);
> + queue_delayed_work(kveritydq, &io->work, 0);
> + } else {
> + INIT_DELAYED_WORK(&io->work, kverityd_io);
> + queue_delayed_work(kverityd_ioq, &io->work, HZ/10);
> + }
> +
> +done:
> + return;
> +
> +io_error:
> + verity_return_bio_to_caller(io);
> +}
> +
> +/* Walks the data set and computes the hash of the data read from the
> + * untrusted source device. The computed hash is then passed to verity-tree
> + * for verification.
> + */
> +static int verity_verify(struct verity_config *vc,
> + struct verity_io *io)
> +{
> + unsigned int block_size = vc->bht.block_size;
> + struct bio *bio = io->bio;
> + u64 block = io->block;
> + unsigned int idx;
> + int r;
> +
> + for (idx = bio->bi_idx; idx < bio->bi_vcnt; idx++) {
> + struct bio_vec *bv = bio_iovec_idx(bio, idx);
> + unsigned int offset = bv->bv_offset;
> + unsigned int len = bv->bv_len;
> +
> + BUG_ON(offset % block_size);
> + BUG_ON(len % block_size);
> +
> + while (len) {
> + r = verity_tree_verify_block(&vc->bht, block,
> + bv->bv_page, offset);
> + if (r)
> + goto bad_return;
> +
> + offset += block_size;
> + len -= block_size;
> + block++;
> + cond_resched();
> + }
> + }
> +
> + return 0;
> +
> +bad_return:
> + /* verity_tree functions aren't expected to return errno friendly
> + * values. They are converted here for uniformity.
> + */
> + if (r > 0) {
> + DMERR("Pending data for block %llu seen at verify", ULL(block));
> + r = -EBUSY;
> + } else {
> + DMERR_LIMIT("Block hash does not match!");
> + r = -EACCES;
> + }
> + return r;
> +}
> +
> +/* Services the verify workqueue */
> +static void kverityd_verify(struct work_struct *work)
> +{
> + struct delayed_work *dwork = container_of(work, struct delayed_work,
> + work);
> + struct verity_io *io = container_of(dwork, struct verity_io,
> + work);
> + struct verity_config *vc = io->target->private;
> +
> + io->error = verity_verify(vc, io);
> +
> + /* Free up the bio and tag with the return value */
> + verity_return_bio_to_caller(io);
> +}
> +
> +/* Asynchronously called upon the completion of verity-tree I/O. The status
> + * of the operation is passed back to verity-tree and the next steps are
> + * decided by verity_dec_pending.
> + */
> +static void kverityd_io_bht_populate_end(struct bio *bio, int error)
> +{
> + struct verity_tree_entry *entry;
> + struct verity_io *io;
> +
> + entry = (struct verity_tree_entry *) bio->bi_private;
> + io = (struct verity_io *) entry->io_context;
> +
> + /* Tell the tree to atomically update now that we've populated
> + * the given entry.
> + */
> + verity_tree_read_completed(entry, error);
> +
> + /* Clean up for reuse when reading data to be checked */
> + bio->bi_vcnt = 0;
> + bio->bi_io_vec->bv_offset = 0;
> + bio->bi_io_vec->bv_len = 0;
> + bio->bi_io_vec->bv_page = NULL;
> + /* Restore the private data to I/O so the destructor can be shared. */
> + bio->bi_private = (void *) io;
> + bio_put(bio);
> +
> + /* We bail but assume the tree has been marked bad. */
> + if (unlikely(error)) {
> + DMERR("Failed to read for sector %llu (%u)",
> + ULL(io->bio->bi_sector), io->bio->bi_size);
> + io->error = error;
> + /* Pass through the error to verity_dec_pending below */
> + }
> + /* When pending = 0, it will transition to reading real data */
> + verity_dec_pending(io);
> +}
> +
> +/* Called by verity-tree (via verity_tree_populate), this function provides
> + * the message digests to verity-tree that are stored on disk.
> + */
> +static int kverityd_bht_read_callback(void *ctx, sector_t start, u8 *dst,
> + sector_t count,
> + struct verity_tree_entry *entry)
> +{
> + struct verity_io *io = ctx; /* I/O for this batch */
> + struct verity_config *vc;
> + struct bio *bio;
> +
> + vc = io->target->private;
> +
> + /* The I/O context is nested inside the entry so that we don't need one
> + * io context per page read.
> + */
> + entry->io_context = ctx;
> +
> + /* We should only get page size requests at present. */
> + verity_inc_pending(io);
> + bio = verity_alloc_bioset(vc, GFP_NOIO, 1);
> + if (unlikely(!bio)) {
> + DMCRIT("Out of memory at bio_alloc_bioset");
> + verity_tree_read_completed(entry, -ENOMEM);
> + return -ENOMEM;
> + }
> + bio->bi_private = (void *) entry;
> + bio->bi_idx = 0;
> + bio->bi_size = vc->bht.block_size;
> + bio->bi_sector = vc->hash_start + start;
> + bio->bi_bdev = vc->hash_dev->bdev;
> + bio->bi_end_io = kverityd_io_bht_populate_end;
> + bio->bi_rw = REQ_META;
> + /* Only need to free the bio since the page is managed by bht */
> + bio->bi_destructor = verity_bio_destructor;
> + bio->bi_vcnt = 1;
> + bio->bi_io_vec->bv_offset = offset_in_page(dst);
> + bio->bi_io_vec->bv_len = to_bytes(count);
> + /* dst is guaranteed to be a page_pool allocation */
> + bio->bi_io_vec->bv_page = virt_to_page(dst);
> + /* Track that this I/O is in use. There should be no risk of the io
> + * being removed prior since this is called synchronously.
> + */
> + generic_make_request(bio);
> + return 0;
> +}
> +
> +/* Submits an io request for each missing block of block hashes.
> + * The last one to return will then enqueue this on the io workqueue.
> + */
> +static void kverityd_io_bht_populate(struct verity_io *io)
> +{
> + struct verity_config *vc = io->target->private;
> + u64 block;
> +
> + for (block = io->block; block < io->block + io->count; ++block) {
> + int ret = verity_tree_populate(&vc->bht, io, block);
> +
> + if (ret < 0) {
> + /* verity_dec_pending will handle the error case. */
> + io->error = ret;
> + break;
> + }
> + }
> +}
> +
> +/* Asynchronously called upon the completion of I/O issued
> + * from kverityd_src_io_read. verity_dec_pending() acts as
> + * the scheduler/flow manager.
> + */
> +static void kverityd_src_io_read_end(struct bio *clone, int error)
> +{
> + struct verity_io *io = clone->bi_private;
> +
> + if (unlikely(!bio_flagged(clone, BIO_UPTODATE) && !error))
> + error = -EIO;
> +
> + if (unlikely(error)) {
> + DMERR("Error occurred: %d (%llu, %u)",
> + error, ULL(clone->bi_sector), clone->bi_size);
> + io->error = error;
> + }
> +
> + /* Release the clone which just avoids the block layer from
> + * leaving offsets, etc in unexpected states.
> + */
> + bio_put(clone);
> +
> + verity_dec_pending(io);
> +}
> +
> +/* If not yet underway, an I/O request will be issued to the vc->dev
> + * device for the data needed. It is cloned to avoid unexpected changes
> + * to the original bio struct.
> + */
> +static void kverityd_src_io_read(struct verity_io *io)
> +{
> + struct bio *clone;
> +
> + /* Check if the read is already issued. */
> + if (io->flags & VERITY_IOFLAGS_CLONED)
> + return;
> +
> + io->flags |= VERITY_IOFLAGS_CLONED;
> +
> + /* Clone the bio. The block layer may modify the bvec array. */
> + clone = verity_bio_clone(io);
> + if (unlikely(!clone)) {
> + io->error = -ENOMEM;
> + return;
> + }
> +
> + verity_inc_pending(io);
> +
> + generic_make_request(clone);
> +}
> +
> +/* kverityd_io services the I/O workqueue. For each pass through
> + * the I/O workqueue, a call to populate both the origin drive
> + * data and the hash tree data is made.
> + */
> +static void kverityd_io(struct work_struct *work)
> +{
> + struct delayed_work *dwork = container_of(work, struct delayed_work,
> + work);
> + struct verity_io *io = container_of(dwork, struct verity_io,
> + work);
> +
> + /* Issue requests asynchronously. */
> + verity_inc_pending(io);
> + kverityd_src_io_read(io);
> + kverityd_io_bht_populate(io);
> + verity_dec_pending(io);
> +}
> +
> +/* Paired with verity_dec_pending, the pending value in the io dictate the
> + * lifetime of a request and when it is ready to be processed on the
> + * workqueues.
> + */
> +static void verity_inc_pending(struct verity_io *io)
> +{
> + atomic_inc(&io->pending);
> +}
> +
> +/* Block-level requests start here. */
> +static int verity_map(struct dm_target *ti, struct bio *bio,
> + union map_info *map_context)
> +{
> + struct verity_io *io;
> + struct verity_config *vc;
> + struct request_queue *r_queue;
> +
> + if (unlikely(!ti)) {
> + DMERR("dm_target was NULL");
> + return -EIO;
> + }
> +
> + vc = ti->private;
> + r_queue = bdev_get_queue(vc->dev->bdev);
> +
> + if (bio_data_dir(bio) == WRITE) {
> + /* If we silently drop writes, then the VFS layer will cache
> + * the write and persist it in memory. While it doesn't change
> + * the underlying storage, it still may be contrary to the
> + * behavior expected by a verified, read-only device.
> + */
> + DMWARN_LIMIT("write request received. rejecting with -EIO.");
> + return -EIO;
> + } else {
> + /* Queue up the request to be verified */
> + io = verity_io_alloc(ti, bio);
> + if (!io) {
> + DMERR_LIMIT("Failed to allocate and init IO data");
> + return DM_MAPIO_REQUEUE;
> + }
> + INIT_DELAYED_WORK(&io->work, kverityd_io);
> + queue_delayed_work(kverityd_ioq, &io->work, 0);
> + }
> +
> + return DM_MAPIO_SUBMITTED;
> +}
> +
> +/*
> + * Non-block interfaces and device-mapper specific code
> + */
> +
> +/*
> + * Verity target parameters:
> + *
> + * <dev> <hash_dev> <hash_start> <block_size> <alg> <digest> <salt>
> + *
> + * version: version of the hash tree on-disk format
> + * dev: device to verify
> + * hash_dev: device hashtree is stored on
> + * hash_start: start address of hashes
> + * block_size: size of a hash block
> + * alg: hash algorithm
> + * digest: toplevel hash of the tree
> + * salt: salt
> + */
> +static int verity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
> +{
> + struct verity_config *vc = NULL;
> + const char *dev, *hash_dev, *alg, *digest, *salt;
> + unsigned long hash_start, block_size, version;
> + sector_t blocks;
> + int ret;
> +
> + if (argc != 8) {
> + ti->error = "Invalid argument count";
> + return -EINVAL;
> + }
> +
> + if (strict_strtoul(argv[0], 10, &version) ||
> + (version != 0)) {
> + ti->error = "Invalid version";
> + return -EINVAL;
> + }
> + dev = argv[1];
> + hash_dev = argv[2];
> + if (strict_strtoul(argv[3], 10, &hash_start)) {
> + ti->error = "Invalid hash_start";
> + return -EINVAL;
> + }
> + if (strict_strtoul(argv[4], 10, &block_size) ||
> + (block_size > UINT_MAX)) {
> + ti->error = "Invalid block_size";
> + return -EINVAL;
> + }
> + alg = argv[5];
> + digest = argv[6];
> + salt = argv[7];
> +
> + /* The device mapper device should be setup read-only */
> + if ((dm_table_get_mode(ti->table) & ~FMODE_READ) != 0) {
> + ti->error = "Must be created readonly.";
> + return -EINVAL;
> + }
> +
> + vc = kzalloc(sizeof(*vc), GFP_KERNEL);
> + if (!vc) {
> + return -EINVAL;
> + }
> +
> + /* Calculate the blocks from the given device size */
> + vc->size = ti->len;
> + blocks = to_bytes(vc->size) / block_size;
> + if (verity_tree_create(&vc->bht, blocks, block_size, alg)) {
> + DMERR("failed to create required bht");
> + goto bad_bht;
> + }
> + if (verity_tree_set_digest(&vc->bht, digest)) {
> + DMERR("digest error");
> + goto bad_digest;
> + }
> + verity_tree_set_salt(&vc->bht, salt);
> + vc->bht.read_cb = kverityd_bht_read_callback;
> +
> + vc->start = 0;
> + /* We only ever grab the device in read-only mode. */
> + ret = dm_get_device(ti, dev, dm_table_get_mode(ti->table), &vc->dev);
> + if (ret) {
> + DMERR("Failed to acquire device '%s': %d", dev, ret);
> + ti->error = "Device lookup failed";
> + goto bad_verity_dev;
> + }
> +
> + if ((to_bytes(vc->start) % block_size) ||
> + (to_bytes(vc->size) % block_size)) {
> + ti->error = "Device must be block_size divisble/aligned";
> + goto bad_hash_start;
> + }
> +
> + vc->hash_start = (sector_t)hash_start;
> +
> + /*
> + * Note, dev == hash_dev is okay as long as the size of
> + * ti->len passed to device mapper does not include
> + * the hashes.
> + */
> + if (dm_get_device(ti, hash_dev,
> + dm_table_get_mode(ti->table), &vc->hash_dev)) {
> + ti->error = "Hash device lookup failed";
> + goto bad_hash_dev;
> + }
> +
> + if (snprintf(vc->hash_alg, CRYPTO_MAX_ALG_NAME, "%s", alg) >=
> + CRYPTO_MAX_ALG_NAME) {
> + ti->error = "Hash algorithm name is too long";
> + goto bad_hash;
> + }
> +
> + vc->io_pool = mempool_create_slab_pool(MIN_IOS, _verity_io_pool);
> + if (!vc->io_pool) {
> + ti->error = "Cannot allocate verity io mempool";
> + goto bad_slab_pool;
> + }
> +
> + vc->bs = bioset_create(MIN_BIOS, 0);
> + if (!vc->bs) {
> + ti->error = "Cannot allocate verity bioset";
> + goto bad_bs;
> + }
> +
> + ti->private = vc;
> +
> + return 0;
> +
> +bad_bs:
> + mempool_destroy(vc->io_pool);
> +bad_slab_pool:
> +bad_hash:
> + dm_put_device(ti, vc->hash_dev);
> +bad_hash_dev:
> +bad_hash_start:
> + dm_put_device(ti, vc->dev);
> +bad_bht:
> +bad_digest:
> +bad_verity_dev:
> + kfree(vc); /* hash is not secret so no need to zero */
> + return -EINVAL;
> +}
> +
> +static void verity_dtr(struct dm_target *ti)
> +{
> + struct verity_config *vc = (struct verity_config *) ti->private;
> +
> + bioset_free(vc->bs);
> + mempool_destroy(vc->io_pool);
> + verity_tree_destroy(&vc->bht);
> + dm_put_device(ti, vc->hash_dev);
> + dm_put_device(ti, vc->dev);
> + kfree(vc);
> +}
> +
> +static int verity_ioctl(struct dm_target *ti, unsigned int cmd,
> + unsigned long arg)
> +{
> + struct verity_config *vc = (struct verity_config *) ti->private;
> + struct dm_dev *dev = vc->dev;
> + int r = 0;
> +
> + /*
> + * Only pass ioctls through if the device sizes match exactly.
> + */
> + if (vc->start ||
> + ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT)
> + r = scsi_verify_blk_ioctl(NULL, cmd);
> +
> + return r ? : __blkdev_driver_ioctl(dev->bdev, dev->mode, cmd, arg);
> +}
> +
> +static int verity_status(struct dm_target *ti, status_type_t type,
> + char *result, unsigned int maxlen)
> +{
> + struct verity_config *vc = (struct verity_config *) ti->private;
> + unsigned int sz = 0;
> + char digest[VERITY_MAX_DIGEST_SIZE * 2 + 1] = { 0 };
> + char salt[VERITY_SALT_SIZE * 2 + 1] = { 0 };
> +
> + verity_tree_digest(&vc->bht, digest);
> + verity_tree_salt(&vc->bht, salt);
> +
> + switch (type) {
> + case STATUSTYPE_INFO:
> + result[0] = '\0';
> + break;
> + case STATUSTYPE_TABLE:
> + DMEMIT("%s %s %llu %llu %s %s %s",
> + vc->dev->name,
> + vc->hash_dev->name,
> + ULL(vc->hash_start),
> + ULL(vc->bht.block_size),
> + vc->hash_alg,
> + digest,
> + salt);
> + break;
> + }
> + return 0;
> +}
> +
> +static int verity_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
> + struct bio_vec *biovec, int max_size)
> +{
> + struct verity_config *vc = ti->private;
> + struct request_queue *q = bdev_get_queue(vc->dev->bdev);
> +
> + if (!q->merge_bvec_fn)
> + return max_size;
> +
> + bvm->bi_bdev = vc->dev->bdev;
> + bvm->bi_sector = vc->start + bvm->bi_sector - ti->begin;
> +
> + /* Optionally, this could just return 0 to stick to single pages. */
> + return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
> +}
> +
> +static int verity_iterate_devices(struct dm_target *ti,
> + iterate_devices_callout_fn fn, void *data)
> +{
> + struct verity_config *vc = ti->private;
> +
> + return fn(ti, vc->dev, vc->start, ti->len, data);
> +}
> +
> +static void verity_io_hints(struct dm_target *ti,
> + struct queue_limits *limits)
> +{
> + struct verity_config *vc = ti->private;
> + unsigned int block_size = vc->bht.block_size;
> +
> + limits->logical_block_size = block_size;
> + limits->physical_block_size = block_size;
> + blk_limits_io_min(limits, block_size);
> +}
> +
> +static struct target_type verity_target = {
> + .name = "verity",
> + .version = {0, 1, 0},
> + .module = THIS_MODULE,
> + .ctr = verity_ctr,
> + .dtr = verity_dtr,
> + .ioctl = verity_ioctl,
> + .map = verity_map,
> + .merge = verity_merge,
> + .status = verity_status,
> + .iterate_devices = verity_iterate_devices,
> + .io_hints = verity_io_hints,
> +};
> +
> +#define VERITY_WQ_FLAGS (WQ_CPU_INTENSIVE|WQ_HIGHPRI)
> +
> +static int __init verity_init(void)
> +{
> + int r = -ENOMEM;
> +
> + _verity_io_pool = KMEM_CACHE(verity_io, 0);
> + if (!_verity_io_pool) {
> + DMERR("failed to allocate pool verity_io");
> + goto bad_io_pool;
> + }
> +
> + kverityd_ioq = alloc_workqueue("kverityd_io", VERITY_WQ_FLAGS, 1);
> + if (!kverityd_ioq) {
> + DMERR("failed to create workqueue kverityd_ioq");
> + goto bad_io_queue;
> + }
> +
> + kveritydq = alloc_workqueue("kverityd", VERITY_WQ_FLAGS, 1);
> + if (!kveritydq) {
> + DMERR("failed to create workqueue kveritydq");
> + goto bad_verify_queue;
> + }
> +
> + r = dm_register_target(&verity_target);
> + if (r < 0) {
> + DMERR("register failed %d", r);
> + goto register_failed;
> + }
> +
> + DMINFO("version %u.%u.%u loaded", verity_target.version[0],
> + verity_target.version[1], verity_target.version[2]);
> +
> + return r;
> +
> +register_failed:
> + destroy_workqueue(kveritydq);
> +bad_verify_queue:
> + destroy_workqueue(kverityd_ioq);
> +bad_io_queue:
> + kmem_cache_destroy(_verity_io_pool);
> +bad_io_pool:
> + return r;
> +}
> +
> +static void __exit verity_exit(void)
> +{
> + destroy_workqueue(kveritydq);
> + destroy_workqueue(kverityd_ioq);
> +
> + dm_unregister_target(&verity_target);
> + kmem_cache_destroy(_verity_io_pool);
> +}
> +
> +module_init(verity_init);
> +module_exit(verity_exit);
> +
> +MODULE_AUTHOR("The Chromium OS Authors <[email protected]>");
> +MODULE_DESCRIPTION(DM_NAME " target for transparent disk integrity checking");
> +MODULE_LICENSE("GPL");
> --
> 1.7.7.3
>
On Tue, 28 Feb 2012 14:57:52 -0800
Mandeep Singh Baines <[email protected]> wrote:
> The verity target provides transparent integrity checking of block devices
> using a cryptographic digest.
>
> dm-verity is meant to be setup as part of a verified boot path. This
> may be anything ranging from a boot using tboot or trustedgrub to just
> booting from a known-good device (like a USB drive or CD).
>
> dm-verity is part of ChromeOS's verified boot path. It is used to verify
> the integrity of the root filesystem on boot. The root filesystem is
> mounted on a dm-verity partition which transparently verifies each block
> with a bootloader verified hash passed into the kernel at boot.
I brought my towering knowledge of DM drivers to bear upon your patch!
The documentation in this patch is brilliant. You usefully documented
the data structures! Never seen that before.
>
> ...
>
> +static int verity_tree_create(struct verity_tree *vt, unsigned int block_count,
> + unsigned int block_size, const char *alg_name)
> +{
> + struct crypto_shash *tfm;
> + int size, cpu, status = 0;
> +
> + vt->block_size = block_size;
> + /* Verify that PAGE_SIZE >= block_size >= SECTOR_SIZE. */
> + if ((block_size > PAGE_SIZE) ||
> + (PAGE_SIZE % block_size) ||
> + (to_sector(block_size) == 0))
> + return -EINVAL;
> +
> + tfm = crypto_alloc_shash(alg_name, 0, 0);
> + if (IS_ERR(tfm)) {
> + DMERR("failed to allocate crypto hash '%s'", alg_name);
> + return -ENOMEM;
> + }
> + size = sizeof(struct shash_desc) + crypto_shash_descsize(tfm);
> +
> + vt->hash_desc = alloc_percpu(struct shash_desc *);
> + if (!vt->hash_desc) {
> + DMERR("Failed to allocate per cpu hash_desc");
> + status = -ENOMEM;
> + goto bad_per_cpu;
> + }
> +
> + /* Pre-allocate per-cpu crypto contexts to avoid having to
> + * kmalloc/kfree a context for every hash operation.
> + */
> + for_each_possible_cpu(cpu) {
Is lame. Can/should it be made cpu-hotplug-aware, so we use
for_each_online_cpu()?
> + struct shash_desc *hash_desc = kmalloc(size, GFP_KERNEL);
> +
> + *per_cpu_ptr(vt->hash_desc, cpu) = hash_desc;
> + if (!hash_desc) {
> + DMERR("failed to allocate crypto hash contexts");
> + status = -ENOMEM;
> + goto bad_hash_alloc;
> + }
> + hash_desc->tfm = tfm;
> + hash_desc->flags = 0x0;
> + }
> + vt->digest_size = crypto_shash_digestsize(tfm);
> + /* We expect to be able to pack >=2 hashes into a block */
> + if (block_size / vt->digest_size < 2) {
> + DMERR("too few hashes fit in a block");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + if (vt->digest_size > VERITY_MAX_DIGEST_SIZE) {
> + DMERR("VERITY_MAX_DIGEST_SIZE too small for digest");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + /* Configure the tree */
> + vt->block_count = block_count;
> + if (block_count == 0) {
> + DMERR("block_count must be non-zero");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + /* Each verity_tree_entry->nodes is one block. The node code tracks
> + * how many nodes fit into one entry where a node is a single
> + * hash (message digest).
> + */
> + vt->node_count_shift = fls(block_size / vt->digest_size) - 1;
> + /* Round down to the nearest power of two. This makes indexing
> + * into the tree much less painful.
> + */
> + vt->node_count = 1 << vt->node_count_shift;
> +
> + /* This is unlikely to happen, but with 64k pages, who knows. */
> + if (vt->node_count > UINT_MAX / vt->digest_size) {
> + DMERR("node_count * hash_len exceeds UINT_MAX!");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + vt->depth = DIV_ROUND_UP(fls(block_count - 1), vt->node_count_shift);
> +
> + /* Ensure that we can safely shift by this value. */
> + if (vt->depth * vt->node_count_shift >= sizeof(unsigned int) * 8) {
> + DMERR("specified depth and node_count_shift is too large");
> + status = -EINVAL;
> + goto bad_arg;
> + }
> +
> + /* Allocate levels. Each level of the tree may have an arbitrary number
> + * of verity_tree_entry structs. Each entry contains node_count nodes.
> + * Each node in the tree is a cryptographic digest of either node_count
> + * nodes on the subsequent level or of a specific block on disk.
> + */
> + vt->levels = (struct verity_tree_level *)
> + kcalloc(vt->depth,
> + sizeof(struct verity_tree_level), GFP_KERNEL);
> + if (!vt->levels) {
> + DMERR("failed to allocate tree levels");
> + status = -ENOMEM;
> + goto bad_level_alloc;
> + }
> +
> + vt->read_cb = NULL;
> +
> + status = verity_tree_initialize_entries(vt);
> + if (status)
> + goto bad_entries_alloc;
> +
> + /* We compute depth such that there is only be 1 block at level 0. */
> + BUG_ON(vt->levels[0].count != 1);
> +
> + return 0;
> +
> +bad_entries_alloc:
> + while (vt->depth-- > 0)
> + kfree(vt->levels[vt->depth].entries);
> + kfree(vt->levels);
> +bad_level_alloc:
> +bad_arg:
> +bad_hash_alloc:
> + for_each_possible_cpu(cpu)
> + if (*per_cpu_ptr(vt->hash_desc, cpu))
This test assumes that alloc_percpu() zeroed out the target memory.
Not true, is it?
> + kfree(*per_cpu_ptr(vt->hash_desc, cpu));
Also, kfree(NULL) is OK, so the test was unneeded. But it will crash
the kernel either way ;)
> + free_percpu(vt->hash_desc);
> +bad_per_cpu:
> + crypto_free_shash(tfm);
> + return status;
> +}
> +
>
> ...
>
> +static struct verity_io *verity_io_alloc(struct dm_target *ti,
> + struct bio *bio)
> +{
> + struct verity_config *vc = ti->private;
> + sector_t sector = bio->bi_sector - ti->begin;
> + struct verity_io *io;
> +
> + io = mempool_alloc(vc->io_pool, GFP_NOIO);
> + if (unlikely(!io))
> + return NULL;
Actually, mempool_alloc(..., __GFP_WAIT) cannot fail. But I wouldn't
trust it either ;)
> + io->flags = 0;
> + io->target = ti;
> + io->bio = bio;
> + io->error = 0;
> +
> + /* Adjust the sector by the virtual starting sector */
> + io->block = to_bytes(sector) / vc->bht.block_size;
> + io->count = bio->bi_size / vc->bht.block_size;
> +
> + atomic_set(&io->pending, 0);
> +
> + return io;
> +}
> +
>
> ...
>
> +static void verity_return_bio_to_caller(struct verity_io *io)
> +{
> + struct verity_config *vc = io->target->private;
> +
> + if (io->error)
> + io->error = -EIO;
That's odd. Why overwrite a potentially useful errno?
> + bio_endio(io->bio, io->error);
> + mempool_free(io, vc->io_pool);
> +}
> +
>
> ...
>
> +static void kverityd_io_bht_populate_end(struct bio *bio, int error)
> +{
> + struct verity_tree_entry *entry;
> + struct verity_io *io;
> +
> + entry = (struct verity_tree_entry *) bio->bi_private;
Unneeded and undesirable cast of void*.
> + io = (struct verity_io *) entry->io_context;
Ditto.
> + /* Tell the tree to atomically update now that we've populated
> + * the given entry.
> + */
> + verity_tree_read_completed(entry, error);
> +
> + /* Clean up for reuse when reading data to be checked */
> + bio->bi_vcnt = 0;
> + bio->bi_io_vec->bv_offset = 0;
> + bio->bi_io_vec->bv_len = 0;
> + bio->bi_io_vec->bv_page = NULL;
> + /* Restore the private data to I/O so the destructor can be shared. */
> + bio->bi_private = (void *) io;
> + bio_put(bio);
> +
> + /* We bail but assume the tree has been marked bad. */
> + if (unlikely(error)) {
> + DMERR("Failed to read for sector %llu (%u)",
> + ULL(io->bio->bi_sector), io->bio->bi_size);
> + io->error = error;
> + /* Pass through the error to verity_dec_pending below */
> + }
> + /* When pending = 0, it will transition to reading real data */
> + verity_dec_pending(io);
> +}
> +
> +/* Called by verity-tree (via verity_tree_populate), this function provides
> + * the message digests to verity-tree that are stored on disk.
> + */
> +static int kverityd_bht_read_callback(void *ctx, sector_t start, u8 *dst,
> + sector_t count,
> + struct verity_tree_entry *entry)
> +{
> + struct verity_io *io = ctx; /* I/O for this batch */
> + struct verity_config *vc;
> + struct bio *bio;
> +
> + vc = io->target->private;
> +
> + /* The I/O context is nested inside the entry so that we don't need one
> + * io context per page read.
> + */
> + entry->io_context = ctx;
> +
> + /* We should only get page size requests at present. */
> + verity_inc_pending(io);
> + bio = verity_alloc_bioset(vc, GFP_NOIO, 1);
> + if (unlikely(!bio)) {
> + DMCRIT("Out of memory at bio_alloc_bioset");
> + verity_tree_read_completed(entry, -ENOMEM);
> + return -ENOMEM;
> + }
> + bio->bi_private = (void *) entry;
Another unneeded cast. And it's "undesirable" because the cast defeats
typechecking. Suppose someone were to change "entry"'s type to "long".
> + bio->bi_idx = 0;
> + bio->bi_size = vc->bht.block_size;
> + bio->bi_sector = vc->hash_start + start;
> + bio->bi_bdev = vc->hash_dev->bdev;
> + bio->bi_end_io = kverityd_io_bht_populate_end;
> + bio->bi_rw = REQ_META;
> + /* Only need to free the bio since the page is managed by bht */
> + bio->bi_destructor = verity_bio_destructor;
> + bio->bi_vcnt = 1;
> + bio->bi_io_vec->bv_offset = offset_in_page(dst);
> + bio->bi_io_vec->bv_len = to_bytes(count);
> + /* dst is guaranteed to be a page_pool allocation */
> + bio->bi_io_vec->bv_page = virt_to_page(dst);
> + /* Track that this I/O is in use. There should be no risk of the io
> + * being removed prior since this is called synchronously.
> + */
> + generic_make_request(bio);
> + return 0;
> +}
> +
>
> ...
>
> +static void kverityd_src_io_read_end(struct bio *clone, int error)
> +{
> + struct verity_io *io = clone->bi_private;
> +
> + if (unlikely(!bio_flagged(clone, BIO_UPTODATE) && !error))
> + error = -EIO;
> +
> + if (unlikely(error)) {
> + DMERR("Error occurred: %d (%llu, %u)",
> + error, ULL(clone->bi_sector), clone->bi_size);
Doing a printk() on each I/O error is often a bad idea - if a device
dies it can cause enormous uncontrollable log storms.
printk_ratelimited(), perhaps?
> + io->error = error;
> + }
> +
> + /* Release the clone which just avoids the block layer from
> + * leaving offsets, etc in unexpected states.
> + */
> + bio_put(clone);
> +
> + verity_dec_pending(io);
> +}
> +
>
> ...
>
> +static int verity_map(struct dm_target *ti, struct bio *bio,
> + union map_info *map_context)
> +{
> + struct verity_io *io;
> + struct verity_config *vc;
> + struct request_queue *r_queue;
> +
> + if (unlikely(!ti)) {
> + DMERR("dm_target was NULL");
> + return -EIO;
> + }
> +
> + vc = ti->private;
> + r_queue = bdev_get_queue(vc->dev->bdev);
> +
> + if (bio_data_dir(bio) == WRITE) {
> + /* If we silently drop writes, then the VFS layer will cache
> + * the write and persist it in memory. While it doesn't change
> + * the underlying storage, it still may be contrary to the
> + * behavior expected by a verified, read-only device.
> + */
> + DMWARN_LIMIT("write request received. rejecting with -EIO.");
> + return -EIO;
> + } else {
> + /* Queue up the request to be verified */
> + io = verity_io_alloc(ti, bio);
> + if (!io) {
> + DMERR_LIMIT("Failed to allocate and init IO data");
> + return DM_MAPIO_REQUEUE;
> + }
> + INIT_DELAYED_WORK(&io->work, kverityd_io);
> + queue_delayed_work(kverityd_ioq, &io->work, 0);
hm, I'm seeing delayed works being queued but I'm not seeing anywhere
which explicitly flushes them all out on the shutdown/rmmod paths. Are
you sure we can't accidentally leave works in flight?
> + }
> +
> + return DM_MAPIO_SUBMITTED;
> +}
> +
>
> ...
>
> +static int verity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
> +{
> + struct verity_config *vc = NULL;
> + const char *dev, *hash_dev, *alg, *digest, *salt;
> + unsigned long hash_start, block_size, version;
> + sector_t blocks;
> + int ret;
> +
> + if (argc != 8) {
> + ti->error = "Invalid argument count";
> + return -EINVAL;
> + }
> +
> + if (strict_strtoul(argv[0], 10, &version) ||
There's no point in me telling you things which checkpatch knows about ;)
> + (version != 0)) {
> + ti->error = "Invalid version";
> + return -EINVAL;
> + }
>
> ...
>
> +static int verity_ioctl(struct dm_target *ti, unsigned int cmd,
> + unsigned long arg)
> +{
> + struct verity_config *vc = (struct verity_config *) ti->private;
Another unneeded/undesirable cast. Multiple of instances of this one.
> + struct dm_dev *dev = vc->dev;
> + int r = 0;
> +
> + /*
> + * Only pass ioctls through if the device sizes match exactly.
> + */
> + if (vc->start ||
> + ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT)
> + r = scsi_verify_blk_ioctl(NULL, cmd);
> +
> + return r ? : __blkdev_driver_ioctl(dev->bdev, dev->mode, cmd, arg);
> +}
> +
> +static int verity_status(struct dm_target *ti, status_type_t type,
> + char *result, unsigned int maxlen)
> +{
> + struct verity_config *vc = (struct verity_config *) ti->private;
> + unsigned int sz = 0;
unused
> + char digest[VERITY_MAX_DIGEST_SIZE * 2 + 1] = { 0 };
> + char salt[VERITY_SALT_SIZE * 2 + 1] = { 0 };
> +
> + verity_tree_digest(&vc->bht, digest);
> + verity_tree_salt(&vc->bht, salt);
> +
> + switch (type) {
> + case STATUSTYPE_INFO:
> + result[0] = '\0';
> + break;
> + case STATUSTYPE_TABLE:
> + DMEMIT("%s %s %llu %llu %s %s %s",
> + vc->dev->name,
> + vc->hash_dev->name,
> + ULL(vc->hash_start),
> + ULL(vc->bht.block_size),
> + vc->hash_alg,
> + digest,
> + salt);
> + break;
> + }
> + return 0;
> +}
> +
>
> ...
>
> +static void verity_io_hints(struct dm_target *ti,
> + struct queue_limits *limits)
> +{
> + struct verity_config *vc = ti->private;
Did it right that time!
> + unsigned int block_size = vc->bht.block_size;
> +
> + limits->logical_block_size = block_size;
> + limits->physical_block_size = block_size;
> + blk_limits_io_min(limits, block_size);
> +}
> +
>
> ...
>