dm: add integrity target
authorMikulas Patocka <mpatocka@redhat.com>
Wed, 4 Jan 2017 19:23:53 +0000 (20:23 +0100)
committerMike Snitzer <snitzer@redhat.com>
Fri, 24 Mar 2017 19:49:07 +0000 (15:49 -0400)
The dm-integrity target emulates a block device that has additional
per-sector tags that can be used for storing integrity information.

A general problem with storing integrity tags with every sector is that
writing the sector and the integrity tag must be atomic - i.e. in case of
crash, either both sector and integrity tag or none of them is written.

To guarantee write atomicity the dm-integrity target uses a journal. It
writes sector data and integrity tags into a journal, commits the journal
and then copies the data and integrity tags to their respective location.

The dm-integrity target can be used with the dm-crypt target - in this
situation the dm-crypt target creates the integrity data and passes them
to the dm-integrity target via bio_integrity_payload attached to the bio.
In this mode, the dm-crypt and dm-integrity targets provide authenticated
disk encryption - if the attacker modifies the encrypted device, an I/O
error is returned instead of random data.

The dm-integrity target can also be used as a standalone target, in this
mode it calculates and verifies the integrity tag internally. In this
mode, the dm-integrity target can be used to detect silent data
corruption on the disk or in the I/O path.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Milan Broz <gmazyland@gmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Documentation/device-mapper/dm-integrity.txt [new file with mode: 0644]
drivers/md/Kconfig
drivers/md/Makefile
drivers/md/dm-integrity.c [new file with mode: 0644]

diff --git a/Documentation/device-mapper/dm-integrity.txt b/Documentation/device-mapper/dm-integrity.txt
new file mode 100644 (file)
index 0000000..2406f56
--- /dev/null
@@ -0,0 +1,189 @@
+The dm-integrity target emulates a block device that has additional
+per-sector tags that can be used for storing integrity information.
+
+A general problem with storing integrity tags with every sector is that
+writing the sector and the integrity tag must be atomic - i.e. in case of
+crash, either both sector and integrity tag or none of them is written.
+
+To guarantee write atomicity, the dm-integrity target uses journal, it
+writes sector data and integrity tags into a journal, commits the journal
+and then copies the data and integrity tags to their respective location.
+
+The dm-integrity target can be used with the dm-crypt target - in this
+situation the dm-crypt target creates the integrity data and passes them
+to the dm-integrity target via bio_integrity_payload attached to the bio.
+In this mode, the dm-crypt and dm-integrity targets provide authenticated
+disk encryption - if the attacker modifies the encrypted device, an I/O
+error is returned instead of random data.
+
+The dm-integrity target can also be used as a standalone target, in this
+mode it calculates and verifies the integrity tag internally. In this
+mode, the dm-integrity target can be used to detect silent data
+corruption on the disk or in the I/O path.
+
+
+When loading the target for the first time, the kernel driver will format
+the device. But it will only format the device if the superblock contains
+zeroes. If the superblock is neither valid nor zeroed, the dm-integrity
+target can't be loaded.
+
+To use the target for the first time:
+1. overwrite the superblock with zeroes
+2. load the dm-integrity target with one-sector size, the kernel driver
+       will format the device
+3. unload the dm-integrity target
+4. read the "provided_data_sectors" value from the superblock
+5. load the dm-integrity target with the the target size
+       "provided_data_sectors"
+6. if you want to use dm-integrity with dm-crypt, load the dm-crypt target
+       with the size "provided_data_sectors"
+
+
+Target arguments:
+
+1. the underlying block device
+
+2. the number of reserved sector at the beginning of the device - the
+       dm-integrity won't read of write these sectors
+
+3. the size of the integrity tag (if "-" is used, the size is taken from
+       the internal-hash algorithm)
+
+4. mode:
+       D - direct writes (without journal) - in this mode, journaling is
+               not used and data sectors and integrity tags are written
+               separately. In case of crash, it is possible that the data
+               and integrity tag doesn't match.
+       J - journaled writes - data and integrity tags are written to the
+               journal and atomicity is guaranteed. In case of crash,
+               either both data and tag or none of them are written. The
+               journaled mode degrades write throughput twice because the
+               data have to be written twice.
+
+5. the number of additional arguments
+
+Additional arguments:
+
+journal-sectors:number
+       The size of journal, this argument is used only if formatting the
+       device. If the device is already formatted, the value from the
+       superblock is used.
+
+interleave-sectors:number
+       The number of interleaved sectors. This values is rounded down to
+       a power of two. If the device is already formatted, the value from
+       the superblock is used.
+
+buffer-sectors:number
+       The number of sectors in one buffer. The value is rounded down to
+       a power of two.
+
+       The tag area is accessed using buffers, the buffer size is
+       configurable. The large buffer size means that the I/O size will
+       be larger, but there could be less I/Os issued.
+
+journal-watermark:number
+       The journal watermark in percents. When the size of the journal
+       exceeds this watermark, the thread that flushes the journal will
+       be started.
+
+commit-time:number
+       Commit time in milliseconds. When this time passes, the journal is
+       written. The journal is also written immediatelly if the FLUSH
+       request is received.
+
+internal-hash:algorithm(:key)  (the key is optional)
+       Use internal hash or crc.
+       When this argument is used, the dm-integrity target won't accept
+       integrity tags from the upper target, but it will automatically
+       generate and verify the integrity tags.
+
+       You can use a crc algorithm (such as crc32), then integrity target
+       will protect the data against accidental corruption.
+       You can also use a hmac algorithm (for example
+       "hmac(sha256):0123456789abcdef"), in this mode it will provide
+       cryptographic authentication of the data without encryption.
+
+       When this argument is not used, the integrity tags are accepted
+       from an upper layer target, such as dm-crypt. The upper layer
+       target should check the validity of the integrity tags.
+
+journal-crypt:algorithm(:key)  (the key is optional)
+       Encrypt the journal using given algorithm to make sure that the
+       attacker can't read the journal. You can use a block cipher here
+       (such as "cbc(aes)") or a stream cipher (for example "chacha20",
+       "salsa20", "ctr(aes)" or "ecb(arc4)").
+
+       The journal contains history of last writes to the block device,
+       an attacker reading the journal could see the last sector nubmers
+       that were written. From the sector numbers, the attacker can infer
+       the size of files that were written. To protect against this
+       situation, you can encrypt the journal.
+
+journal-mac:algorithm(:key)    (the key is optional)
+       Protect sector numbers in the journal from accidental or malicious
+       modification. To protect against accidental modification, use a
+       crc algorithm, to protect against malicious modification, use a
+       hmac algorithm with a key.
+
+       This option is not needed when using internal-hash because in this
+       mode, the integrity of journal entries is checked when replaying
+       the journal. Thus, modified sector number would be detected at
+       this stage.
+
+
+The journal mode (D/J), buffer-sectors, journal-watermark, commit-time can
+be changed when reloading the target (load an inactive table and swap the
+tables with suspend and resume). The other arguments should not be changed
+when reloading the target because the layout of disk data depend on them
+and the reloaded target would be non-functional.
+
+
+The layout of the formatted block device:
+* reserved sectors (they are not used by this target, they can be used for
+  storing LUKS metadata or for other purpose), the size of the reserved
+  area is specified in the target arguments
+* superblock (4kiB)
+       * magic string - identifies that the device was formatted
+       * version
+       * log2(interleave sectors)
+       * integrity tag size
+       * the number of journal sections
+       * provided data sectors - the number of sectors that this target
+         provides (i.e. the size of the device minus the size of all
+         metadata and padding). The user of this target should not send
+         bios that access data beyond the "provided data sectors" limit.
+       * flags - a flag is set if journal-mac is used
+* journal
+       The journal is divided into sections, each section contains:
+       * metadata area (4kiB), it contains journal entries
+         every journal entry contains:
+               * logical sector (specifies where the data and tag should
+                 be written)
+               * last 8 bytes of data
+               * integrity tag (the size is specified in the superblock)
+           every metadata sector ends with
+               * mac (8-bytes), all the macs in 8 metadata sectors form a
+                 64-byte value. It is used to store hmac of sector
+                 numbers in the journal section, to protect against a
+                 possibility that the attacker tampers with sector
+                 numbers in the journal.
+               * commit id
+       * data area (the size is variable; it depends on how many journal
+         entries fit into the metadata area)
+           every sector in the data area contains:
+               * data (504 bytes of data, the last 8 bytes are stored in
+                 the journal entry)
+               * commit id
+       To test if the whole journal section was written correctly, every
+       512-byte sector of the journal ends with 8-byte commit id. If the
+       commit id matches on all sectors in a journal section, then it is
+       assumed that the section was written correctly. If the commit id
+       doesn't match, the section was written partially and it should not
+       be replayed.
+* one or more runs of interleaved tags and data. Each run contains:
+       * tag area - it contains integrity tags. There is one tag for each
+         sector in the data area
+       * data area - it contains data sectors. The number of data sectors
+         in one run must be a power of two. log2 of this value is stored
+         in the superblock.
index 982cd0626bc76bfa87cf36911037cc86b5d92833..5c5ed97c9fda699a629ce170e426ea2856d2ad6f 100644 (file)
@@ -500,4 +500,14 @@ config DM_LOG_WRITES
 
          If unsure, say N.
 
+config DM_INTEGRITY
+       tristate "Integrity target"
+       depends on BLK_DEV_DM
+       select BLK_DEV_INTEGRITY
+       select DM_BUFIO
+       select CRYPTO
+       select ASYNC_XOR
+       ---help---
+          This is the integrity target.
+
 endif # MD
index 2801b2fb452d88b6979da721e9f4108498e40d32..39cf2a1b5f90156572ee0383fff51903cf7edbde 100644 (file)
@@ -59,6 +59,7 @@ obj-$(CONFIG_DM_CACHE)                += dm-cache.o
 obj-$(CONFIG_DM_CACHE_SMQ)     += dm-cache-smq.o
 obj-$(CONFIG_DM_ERA)           += dm-era.o
 obj-$(CONFIG_DM_LOG_WRITES)    += dm-log-writes.o
+obj-$(CONFIG_DM_INTEGRITY)     += dm-integrity.o
 
 ifeq ($(CONFIG_DM_UEVENT),y)
 dm-mod-objs                    += dm-uevent.o
diff --git a/drivers/md/dm-integrity.c b/drivers/md/dm-integrity.c
new file mode 100644 (file)
index 0000000..ea779cc
--- /dev/null
@@ -0,0 +1,3085 @@
+/*
+ * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
+ * Copyright (C) 2016-2017 Milan Broz
+ * Copyright (C) 2016-2017 Mikulas Patocka
+ *
+ * This file is released under the GPL.
+ */
+
+#include <linux/module.h>
+#include <linux/device-mapper.h>
+#include <linux/dm-io.h>
+#include <linux/vmalloc.h>
+#include <linux/sort.h>
+#include <linux/rbtree.h>
+#include <linux/delay.h>
+#include <linux/random.h>
+#include <crypto/hash.h>
+#include <crypto/skcipher.h>
+#include <linux/async_tx.h>
+#include "dm-bufio.h"
+
+#define DM_MSG_PREFIX "integrity"
+
+#define DEFAULT_INTERLEAVE_SECTORS     32768
+#define DEFAULT_JOURNAL_SIZE_FACTOR    7
+#define DEFAULT_BUFFER_SECTORS         128
+#define DEFAULT_JOURNAL_WATERMARK      50
+#define DEFAULT_SYNC_MSEC              10000
+#define DEFAULT_MAX_JOURNAL_SECTORS    131072
+#define MIN_INTERLEAVE_SECTORS         3
+#define MAX_INTERLEAVE_SECTORS         31
+#define METADATA_WORKQUEUE_MAX_ACTIVE  16
+
+/*
+ * Warning - DEBUG_PRINT prints security-sensitive data to the log,
+ * so it should not be enabled in the official kernel
+ */
+//#define DEBUG_PRINT
+//#define INTERNAL_VERIFY
+
+/*
+ * On disk structures
+ */
+
+#define SB_MAGIC                       "integrt"
+#define SB_VERSION                     1
+#define SB_SECTORS                     8
+
+struct superblock {
+       __u8 magic[8];
+       __u8 version;
+       __u8 log2_interleave_sectors;
+       __u16 integrity_tag_size;
+       __u32 journal_sections;
+       __u64 provided_data_sectors;    /* userspace uses this value */
+       __u32 flags;
+};
+
+#define SB_FLAG_HAVE_JOURNAL_MAC       0x1
+
+#define        JOURNAL_ENTRY_ROUNDUP           8
+
+typedef __u64 commit_id_t;
+#define JOURNAL_MAC_PER_SECTOR         8
+
+struct journal_entry {
+       union {
+               struct {
+                       __u32 sector_lo;
+                       __u32 sector_hi;
+               } s;
+               __u64 sector;
+       } u;
+       commit_id_t last_bytes;
+       __u8 tag[0];
+};
+
+#if BITS_PER_LONG == 64
+#define journal_entry_set_sector(je, x)                do { smp_wmb(); ACCESS_ONCE((je)->u.sector) = cpu_to_le64(x); } while (0)
+#define journal_entry_get_sector(je)           le64_to_cpu((je)->u.sector)
+#elif defined(CONFIG_LBDAF)
+#define journal_entry_set_sector(je, x)                do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); ACCESS_ONCE((je)->u.s.sector_hi) = cpu_to_le32((x) >> 32); } while (0)
+#define journal_entry_get_sector(je)           le64_to_cpu((je)->u.sector)
+#else
+#define journal_entry_set_sector(je, x)                do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); ACCESS_ONCE((je)->u.s.sector_hi) = cpu_to_le32(0); } while (0)
+#define journal_entry_get_sector(je)           le32_to_cpu((je)->u.s.sector_lo)
+#endif
+#define journal_entry_is_unused(je)            ((je)->u.s.sector_hi == cpu_to_le32(-1))
+#define journal_entry_set_unused(je)           do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
+#define journal_entry_is_inprogress(je)                ((je)->u.s.sector_hi == cpu_to_le32(-2))
+#define journal_entry_set_inprogress(je)       do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
+
+#define JOURNAL_BLOCK_SECTORS          8
+#define JOURNAL_SECTOR_DATA            ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
+#define JOURNAL_MAC_SIZE               (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
+
+struct journal_sector {
+       __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
+       __u8 mac[JOURNAL_MAC_PER_SECTOR];
+       commit_id_t commit_id;
+};
+
+#define MAX_TAG_SIZE                   (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, tag))
+
+#define METADATA_PADDING_SECTORS       8
+
+#define N_COMMIT_IDS                   4
+
+static unsigned char prev_commit_seq(unsigned char seq)
+{
+       return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
+}
+
+static unsigned char next_commit_seq(unsigned char seq)
+{
+       return (seq + 1) % N_COMMIT_IDS;
+}
+
+/*
+ * In-memory structures
+ */
+
+struct journal_node {
+       struct rb_node node;
+       sector_t sector;
+};
+
+struct alg_spec {
+       char *alg_string;
+       char *key_string;
+       __u8 *key;
+       unsigned key_size;
+};
+
+struct dm_integrity_c {
+       struct dm_dev *dev;
+       unsigned tag_size;
+       __s8 log2_tag_size;
+       sector_t start;
+       mempool_t *journal_io_mempool;
+       struct dm_io_client *io;
+       struct dm_bufio_client *bufio;
+       struct workqueue_struct *metadata_wq;
+       struct superblock *sb;
+       unsigned journal_pages;
+       struct page_list *journal;
+       struct page_list *journal_io;
+       struct page_list *journal_xor;
+
+       struct crypto_skcipher *journal_crypt;
+       struct scatterlist **journal_scatterlist;
+       struct scatterlist **journal_io_scatterlist;
+       struct skcipher_request **sk_requests;
+
+       struct crypto_shash *journal_mac;
+
+       struct journal_node *journal_tree;
+       struct rb_root journal_tree_root;
+
+       sector_t provided_data_sectors;
+
+       unsigned short journal_entry_size;
+       unsigned char journal_entries_per_sector;
+       unsigned char journal_section_entries;
+       unsigned char journal_section_sectors;
+       unsigned journal_sections;
+       unsigned journal_entries;
+       sector_t device_sectors;
+       unsigned initial_sectors;
+       unsigned metadata_run;
+       __s8 log2_metadata_run;
+       __u8 log2_buffer_sectors;
+
+       unsigned char mode;
+       bool suspending;
+
+       int failed;
+
+       struct crypto_shash *internal_hash;
+
+       /* these variables are locked with endio_wait.lock */
+       struct rb_root in_progress;
+       wait_queue_head_t endio_wait;
+       struct workqueue_struct *wait_wq;
+
+       unsigned char commit_seq;
+       commit_id_t commit_ids[N_COMMIT_IDS];
+
+       unsigned committed_section;
+       unsigned n_committed_sections;
+
+       unsigned uncommitted_section;
+       unsigned n_uncommitted_sections;
+
+       unsigned free_section;
+       unsigned char free_section_entry;
+       unsigned free_sectors;
+
+       unsigned free_sectors_threshold;
+
+       struct workqueue_struct *commit_wq;
+       struct work_struct commit_work;
+
+       struct workqueue_struct *writer_wq;
+       struct work_struct writer_work;
+
+       struct bio_list flush_bio_list;
+
+       unsigned long autocommit_jiffies;
+       struct timer_list autocommit_timer;
+       unsigned autocommit_msec;
+
+       wait_queue_head_t copy_to_journal_wait;
+
+       struct completion crypto_backoff;
+
+       bool journal_uptodate;
+       bool just_formatted;
+
+       struct alg_spec internal_hash_alg;
+       struct alg_spec journal_crypt_alg;
+       struct alg_spec journal_mac_alg;
+};
+
+struct dm_integrity_range {
+       sector_t logical_sector;
+       unsigned n_sectors;
+       struct rb_node node;
+};
+
+struct dm_integrity_io {
+       struct work_struct work;
+
+       struct dm_integrity_c *ic;
+       bool write;
+       bool fua;
+
+       struct dm_integrity_range range;
+
+       sector_t metadata_block;
+       unsigned metadata_offset;
+
+       atomic_t in_flight;
+       int bi_error;
+
+       struct completion *completion;
+
+       struct block_device *orig_bi_bdev;
+       bio_end_io_t *orig_bi_end_io;
+       struct bio_integrity_payload *orig_bi_integrity;
+       struct bvec_iter orig_bi_iter;
+};
+
+struct journal_completion {
+       struct dm_integrity_c *ic;
+       atomic_t in_flight;
+       struct completion comp;
+};
+
+struct journal_io {
+       struct dm_integrity_range range;
+       struct journal_completion *comp;
+};
+
+static struct kmem_cache *journal_io_cache;
+
+#define JOURNAL_IO_MEMPOOL     32
+
+#ifdef DEBUG_PRINT
+#define DEBUG_print(x, ...)    printk(KERN_DEBUG x, ##__VA_ARGS__)
+static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
+{
+       va_list args;
+       va_start(args, msg);
+       vprintk(msg, args);
+       va_end(args);
+       if (len)
+               pr_cont(":");
+       while (len) {
+               pr_cont(" %02x", *bytes);
+               bytes++;
+               len--;
+       }
+       pr_cont("\n");
+}
+#define DEBUG_bytes(bytes, len, msg, ...)      __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
+#else
+#define DEBUG_print(x, ...)                    do { } while (0)
+#define DEBUG_bytes(bytes, len, msg, ...)      do { } while (0)
+#endif
+
+/*
+ * DM Integrity profile, protection is performed layer above (dm-crypt)
+ */
+static struct blk_integrity_profile dm_integrity_profile = {
+       .name                   = "DM-DIF-EXT-TAG",
+       .generate_fn            = NULL,
+       .verify_fn              = NULL,
+};
+
+static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
+static void integrity_bio_wait(struct work_struct *w);
+static void dm_integrity_dtr(struct dm_target *ti);
+
+static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
+{
+       if (!cmpxchg(&ic->failed, 0, err))
+               DMERR("Error on %s: %d", msg, err);
+}
+
+static int dm_integrity_failed(struct dm_integrity_c *ic)
+{
+       return ACCESS_ONCE(ic->failed);
+}
+
+static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
+                                         unsigned j, unsigned char seq)
+{
+       /*
+        * Xor the number with section and sector, so that if a piece of
+        * journal is written at wrong place, it is detected.
+        */
+       return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
+}
+
+static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
+                               sector_t *area, sector_t *offset)
+{
+       __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
+
+       *area = data_sector >> log2_interleave_sectors;
+       *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
+}
+
+static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
+                                           sector_t offset, unsigned *metadata_offset)
+{
+       __u64 ms;
+       unsigned mo;
+
+       ms = area << ic->sb->log2_interleave_sectors;
+       if (likely(ic->log2_metadata_run >= 0))
+               ms += area << ic->log2_metadata_run;
+       else
+               ms += area * ic->metadata_run;
+       ms >>= ic->log2_buffer_sectors;
+
+       if (likely(ic->log2_tag_size >= 0)) {
+               ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
+               mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
+       } else {
+               ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
+               mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
+       }
+       *metadata_offset = mo;
+       return ms;
+}
+
+static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
+{
+       sector_t result;
+
+       result = area << ic->sb->log2_interleave_sectors;
+       if (likely(ic->log2_metadata_run >= 0))
+               result += (area + 1) << ic->log2_metadata_run;
+       else
+               result += (area + 1) * ic->metadata_run;
+
+       result += (sector_t)ic->initial_sectors + offset;
+       return result;
+}
+
+static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
+{
+       if (unlikely(*sec_ptr >= ic->journal_sections))
+               *sec_ptr -= ic->journal_sections;
+}
+
+static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
+{
+       struct dm_io_request io_req;
+       struct dm_io_region io_loc;
+
+       io_req.bi_op = op;
+       io_req.bi_op_flags = op_flags;
+       io_req.mem.type = DM_IO_KMEM;
+       io_req.mem.ptr.addr = ic->sb;
+       io_req.notify.fn = NULL;
+       io_req.client = ic->io;
+       io_loc.bdev = ic->dev->bdev;
+       io_loc.sector = ic->start;
+       io_loc.count = SB_SECTORS;
+
+       return dm_io(&io_req, 1, &io_loc, NULL);
+}
+
+static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
+                                bool e, const char *function)
+{
+#if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
+       unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
+
+       if (unlikely(section >= ic->journal_sections) ||
+           unlikely(offset >= limit)) {
+               printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
+                       function, section, offset, ic->journal_sections, limit);
+               BUG();
+       }
+#endif
+}
+
+static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
+                              unsigned *pl_index, unsigned *pl_offset)
+{
+       unsigned sector;
+
+       access_journal_check(ic, section, offset, false, "access_journal");
+
+       sector = section * ic->journal_section_sectors + offset;
+
+       *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
+       *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
+}
+
+static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
+                                              unsigned section, unsigned offset, unsigned *n_sectors)
+{
+       unsigned pl_index, pl_offset;
+       char *va;
+
+       page_list_location(ic, section, offset, &pl_index, &pl_offset);
+
+       if (n_sectors)
+               *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
+
+       va = lowmem_page_address(pl[pl_index].page);
+
+       return (struct journal_sector *)(va + pl_offset);
+}
+
+static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
+{
+       return access_page_list(ic, ic->journal, section, offset, NULL);
+}
+
+static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
+{
+       unsigned rel_sector, offset;
+       struct journal_sector *js;
+
+       access_journal_check(ic, section, n, true, "access_journal_entry");
+
+       rel_sector = n % JOURNAL_BLOCK_SECTORS;
+       offset = n / JOURNAL_BLOCK_SECTORS;
+
+       js = access_journal(ic, section, rel_sector);
+       return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
+}
+
+static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
+{
+       access_journal_check(ic, section, n, true, "access_journal_data");
+
+       return access_journal(ic, section, n + JOURNAL_BLOCK_SECTORS);
+}
+
+static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
+{
+       SHASH_DESC_ON_STACK(desc, ic->journal_mac);
+       int r;
+       unsigned j, size;
+
+       desc->tfm = ic->journal_mac;
+       desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
+
+       r = crypto_shash_init(desc);
+       if (unlikely(r)) {
+               dm_integrity_io_error(ic, "crypto_shash_init", r);
+               goto err;
+       }
+
+       for (j = 0; j < ic->journal_section_entries; j++) {
+               struct journal_entry *je = access_journal_entry(ic, section, j);
+               r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
+               if (unlikely(r)) {
+                       dm_integrity_io_error(ic, "crypto_shash_update", r);
+                       goto err;
+               }
+       }
+
+       size = crypto_shash_digestsize(ic->journal_mac);
+
+       if (likely(size <= JOURNAL_MAC_SIZE)) {
+               r = crypto_shash_final(desc, result);
+               if (unlikely(r)) {
+                       dm_integrity_io_error(ic, "crypto_shash_final", r);
+                       goto err;
+               }
+               memset(result + size, 0, JOURNAL_MAC_SIZE - size);
+       } else {
+               __u8 digest[size];
+               r = crypto_shash_final(desc, digest);
+               if (unlikely(r)) {
+                       dm_integrity_io_error(ic, "crypto_shash_final", r);
+                       goto err;
+               }
+               memcpy(result, digest, JOURNAL_MAC_SIZE);
+       }
+
+       return;
+err:
+       memset(result, 0, JOURNAL_MAC_SIZE);
+}
+
+static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
+{
+       __u8 result[JOURNAL_MAC_SIZE];
+       unsigned j;
+
+       if (!ic->journal_mac)
+               return;
+
+       section_mac(ic, section, result);
+
+       for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
+               struct journal_sector *js = access_journal(ic, section, j);
+
+               if (likely(wr))
+                       memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
+               else {
+                       if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
+                               dm_integrity_io_error(ic, "journal mac", -EILSEQ);
+               }
+       }
+}
+
+static void complete_journal_op(void *context)
+{
+       struct journal_completion *comp = context;
+       BUG_ON(!atomic_read(&comp->in_flight));
+       if (likely(atomic_dec_and_test(&comp->in_flight)))
+               complete(&comp->comp);
+}
+
+static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
+                       unsigned n_sections, struct journal_completion *comp)
+{
+       struct async_submit_ctl submit;
+       size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
+       unsigned pl_index, pl_offset, section_index;
+       struct page_list *source_pl, *target_pl;
+
+       if (likely(encrypt)) {
+               source_pl = ic->journal;
+               target_pl = ic->journal_io;
+       } else {
+               source_pl = ic->journal_io;
+               target_pl = ic->journal;
+       }
+
+       page_list_location(ic, section, 0, &pl_index, &pl_offset);
+
+       atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
+
+       init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
+
+       section_index = pl_index;
+
+       do {
+               size_t this_step;
+               struct page *src_pages[2];
+               struct page *dst_page;
+
+               while (unlikely(pl_index == section_index)) {
+                       unsigned dummy;
+                       if (likely(encrypt))
+                               rw_section_mac(ic, section, true);
+                       section++;
+                       n_sections--;
+                       if (!n_sections)
+                               break;
+                       page_list_location(ic, section, 0, &section_index, &dummy);
+               }
+
+               this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
+               dst_page = target_pl[pl_index].page;
+               src_pages[0] = source_pl[pl_index].page;
+               src_pages[1] = ic->journal_xor[pl_index].page;
+
+               async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
+
+               pl_index++;
+               pl_offset = 0;
+               n_bytes -= this_step;
+       } while (n_bytes);
+
+       BUG_ON(n_sections);
+
+       async_tx_issue_pending_all();
+}
+
+static void complete_journal_encrypt(struct crypto_async_request *req, int err)
+{
+       struct journal_completion *comp = req->data;
+       if (unlikely(err)) {
+               if (likely(err == -EINPROGRESS)) {
+                       complete(&comp->ic->crypto_backoff);
+                       return;
+               }
+               dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
+       }
+       complete_journal_op(comp);
+}
+
+static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
+{
+       int r;
+       skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
+                                     complete_journal_encrypt, comp);
+       if (likely(encrypt))
+               r = crypto_skcipher_encrypt(req);
+       else
+               r = crypto_skcipher_decrypt(req);
+       if (likely(!r))
+               return false;
+       if (likely(r == -EINPROGRESS))
+               return true;
+       if (likely(r == -EBUSY)) {
+               wait_for_completion(&comp->ic->crypto_backoff);
+               reinit_completion(&comp->ic->crypto_backoff);
+               return true;
+       }
+       dm_integrity_io_error(comp->ic, "encrypt", r);
+       return false;
+}
+
+static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
+                         unsigned n_sections, struct journal_completion *comp)
+{
+       struct scatterlist **source_sg;
+       struct scatterlist **target_sg;
+
+       atomic_add(2, &comp->in_flight);
+
+       if (likely(encrypt)) {
+               source_sg = ic->journal_scatterlist;
+               target_sg = ic->journal_io_scatterlist;
+       } else {
+               source_sg = ic->journal_io_scatterlist;
+               target_sg = ic->journal_scatterlist;
+       }
+
+       do {
+               struct skcipher_request *req;
+               unsigned ivsize;
+               char *iv;
+
+               if (likely(encrypt))
+                       rw_section_mac(ic, section, true);
+
+               req = ic->sk_requests[section];
+               ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
+               iv = req->iv;
+
+               memcpy(iv, iv + ivsize, ivsize);
+
+               req->src = source_sg[section];
+               req->dst = target_sg[section];
+
+               if (unlikely(do_crypt(encrypt, req, comp)))
+                       atomic_inc(&comp->in_flight);
+
+               section++;
+               n_sections--;
+       } while (n_sections);
+
+       atomic_dec(&comp->in_flight);
+       complete_journal_op(comp);
+}
+
+static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
+                           unsigned n_sections, struct journal_completion *comp)
+{
+       if (ic->journal_xor)
+               return xor_journal(ic, encrypt, section, n_sections, comp);
+       else
+               return crypt_journal(ic, encrypt, section, n_sections, comp);
+}
+
+static void complete_journal_io(unsigned long error, void *context)
+{
+       struct journal_completion *comp = context;
+       if (unlikely(error != 0))
+               dm_integrity_io_error(comp->ic, "writing journal", -EIO);
+       complete_journal_op(comp);
+}
+
+static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
+                      unsigned n_sections, struct journal_completion *comp)
+{
+       struct dm_io_request io_req;
+       struct dm_io_region io_loc;
+       unsigned sector, n_sectors, pl_index, pl_offset;
+       int r;
+
+       if (unlikely(dm_integrity_failed(ic))) {
+               if (comp)
+                       complete_journal_io(-1UL, comp);
+               return;
+       }
+
+       sector = section * ic->journal_section_sectors;
+       n_sectors = n_sections * ic->journal_section_sectors;
+
+       pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
+       pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
+
+       io_req.bi_op = op;
+       io_req.bi_op_flags = op_flags;
+       io_req.mem.type = DM_IO_PAGE_LIST;
+       if (ic->journal_io)
+               io_req.mem.ptr.pl = &ic->journal_io[pl_index];
+       else
+               io_req.mem.ptr.pl = &ic->journal[pl_index];
+       io_req.mem.offset = pl_offset;
+       if (likely(comp != NULL)) {
+               io_req.notify.fn = complete_journal_io;
+               io_req.notify.context = comp;
+       } else {
+               io_req.notify.fn = NULL;
+       }
+       io_req.client = ic->io;
+       io_loc.bdev = ic->dev->bdev;
+       io_loc.sector = ic->start + SB_SECTORS + sector;
+       io_loc.count = n_sectors;
+
+       r = dm_io(&io_req, 1, &io_loc, NULL);
+       if (unlikely(r)) {
+               dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
+               if (comp) {
+                       WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
+                       complete_journal_io(-1UL, comp);
+               }
+       }
+}
+
+static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
+{
+       struct journal_completion io_comp;
+       struct journal_completion crypt_comp_1;
+       struct journal_completion crypt_comp_2;
+       unsigned i;
+
+       io_comp.ic = ic;
+       io_comp.comp = COMPLETION_INITIALIZER_ONSTACK(io_comp.comp);
+
+       if (commit_start + commit_sections <= ic->journal_sections) {
+               io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
+               if (ic->journal_io) {
+                       crypt_comp_1.ic = ic;
+                       crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
+                       crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
+                       encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
+                       wait_for_completion_io(&crypt_comp_1.comp);
+               } else {
+                       for (i = 0; i < commit_sections; i++)
+                               rw_section_mac(ic, commit_start + i, true);
+               }
+               rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, commit_sections, &io_comp);
+       } else {
+               unsigned to_end;
+               io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
+               to_end = ic->journal_sections - commit_start;
+               if (ic->journal_io) {
+                       crypt_comp_1.ic = ic;
+                       crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
+                       crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
+                       encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
+                       if (try_wait_for_completion(&crypt_comp_1.comp)) {
+                               rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
+                               crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
+                               crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
+                               encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
+                               wait_for_completion_io(&crypt_comp_1.comp);
+                       } else {
+                               crypt_comp_2.ic = ic;
+                               crypt_comp_2.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_2.comp);
+                               crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
+                               encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
+                               wait_for_completion_io(&crypt_comp_1.comp);
+                               rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
+                               wait_for_completion_io(&crypt_comp_2.comp);
+                       }
+               } else {
+                       for (i = 0; i < to_end; i++)
+                               rw_section_mac(ic, commit_start + i, true);
+                       rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
+                       for (i = 0; i < commit_sections - to_end; i++)
+                               rw_section_mac(ic, i, true);
+               }
+               rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
+       }
+
+       wait_for_completion_io(&io_comp.comp);
+}
+
+static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
+                             unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
+{
+       struct dm_io_request io_req;
+       struct dm_io_region io_loc;
+       int r;
+       unsigned sector, pl_index, pl_offset;
+
+       if (unlikely(dm_integrity_failed(ic))) {
+               fn(-1UL, data);
+               return;
+       }
+
+       sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
+
+       pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
+       pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
+
+       io_req.bi_op = REQ_OP_WRITE;
+       io_req.bi_op_flags = 0;
+       io_req.mem.type = DM_IO_PAGE_LIST;
+       io_req.mem.ptr.pl = &ic->journal[pl_index];
+       io_req.mem.offset = pl_offset;
+       io_req.notify.fn = fn;
+       io_req.notify.context = data;
+       io_req.client = ic->io;
+       io_loc.bdev = ic->dev->bdev;
+       io_loc.sector = ic->start + target;
+       io_loc.count = n_sectors;
+
+       r = dm_io(&io_req, 1, &io_loc, NULL);
+       if (unlikely(r)) {
+               WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
+               fn(-1UL, data);
+       }
+}
+
+static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
+{
+       struct rb_node **n = &ic->in_progress.rb_node;
+       struct rb_node *parent;
+
+       parent = NULL;
+
+       while (*n) {
+               struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
+
+               parent = *n;
+               if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
+                       n = &range->node.rb_left;
+               } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
+                       n = &range->node.rb_right;
+               } else {
+                       return false;
+               }
+       }
+
+       rb_link_node(&new_range->node, parent, n);
+       rb_insert_color(&new_range->node, &ic->in_progress);
+
+       return true;
+}
+
+static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
+{
+       rb_erase(&range->node, &ic->in_progress);
+       wake_up_locked(&ic->endio_wait);
+}
+
+static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&ic->endio_wait.lock, flags);
+       remove_range_unlocked(ic, range);
+       spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
+}
+
+static void init_journal_node(struct journal_node *node)
+{
+       RB_CLEAR_NODE(&node->node);
+       node->sector = (sector_t)-1;
+}
+
+static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
+{
+       struct rb_node **link;
+       struct rb_node *parent;
+
+       node->sector = sector;
+       BUG_ON(!RB_EMPTY_NODE(&node->node));
+
+       link = &ic->journal_tree_root.rb_node;
+       parent = NULL;
+
+       while (*link) {
+               struct journal_node *j;
+               parent = *link;
+               j = container_of(parent, struct journal_node, node);
+               if (sector < j->sector)
+                       link = &j->node.rb_left;
+               else
+                       link = &j->node.rb_right;
+       }
+
+       rb_link_node(&node->node, parent, link);
+       rb_insert_color(&node->node, &ic->journal_tree_root);
+}
+
+static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
+{
+       BUG_ON(RB_EMPTY_NODE(&node->node));
+       rb_erase(&node->node, &ic->journal_tree_root);
+       init_journal_node(node);
+}
+
+#define NOT_FOUND      (-1U)
+
+static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
+{
+       struct rb_node *n = ic->journal_tree_root.rb_node;
+       unsigned found = NOT_FOUND;
+       *next_sector = (sector_t)-1;
+       while (n) {
+               struct journal_node *j = container_of(n, struct journal_node, node);
+               if (sector == j->sector) {
+                       found = j - ic->journal_tree;
+               }
+               if (sector < j->sector) {
+                       *next_sector = j->sector;
+                       n = j->node.rb_left;
+               } else {
+                       n = j->node.rb_right;
+               }
+       }
+
+       return found;
+}
+
+static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
+{
+       struct journal_node *node, *next_node;
+       struct rb_node *next;
+
+       if (unlikely(pos >= ic->journal_entries))
+               return false;
+       node = &ic->journal_tree[pos];
+       if (unlikely(RB_EMPTY_NODE(&node->node)))
+               return false;
+       if (unlikely(node->sector != sector))
+               return false;
+
+       next = rb_next(&node->node);
+       if (unlikely(!next))
+               return true;
+
+       next_node = container_of(next, struct journal_node, node);
+       return next_node->sector != sector;
+}
+
+static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
+{
+       struct rb_node *next;
+       struct journal_node *next_node;
+       unsigned next_section;
+
+       BUG_ON(RB_EMPTY_NODE(&node->node));
+
+       next = rb_next(&node->node);
+       if (unlikely(!next))
+               return false;
+
+       next_node = container_of(next, struct journal_node, node);
+
+       if (next_node->sector != node->sector)
+               return false;
+
+       next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
+       if (next_section >= ic->committed_section &&
+           next_section < ic->committed_section + ic->n_committed_sections)
+               return true;
+       if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
+               return true;
+
+       return false;
+}
+
+#define TAG_READ       0
+#define TAG_WRITE      1
+#define TAG_CMP                2
+
+static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
+                              unsigned *metadata_offset, unsigned total_size, int op)
+{
+       do {
+               unsigned char *data, *dp;
+               struct dm_buffer *b;
+               unsigned to_copy;
+               int r;
+
+               r = dm_integrity_failed(ic);
+               if (unlikely(r))
+                       return r;
+
+               data = dm_bufio_read(ic->bufio, *metadata_block, &b);
+               if (unlikely(IS_ERR(data)))
+                       return PTR_ERR(data);
+
+               to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
+               dp = data + *metadata_offset;
+               if (op == TAG_READ) {
+                       memcpy(tag, dp, to_copy);
+               } else if (op == TAG_WRITE) {
+                       memcpy(dp, tag, to_copy);
+                       dm_bufio_mark_buffer_dirty(b);
+               } else  {
+                       /* e.g.: op == TAG_CMP */
+                       if (unlikely(memcmp(dp, tag, to_copy))) {
+                               unsigned i;
+
+                               for (i = 0; i < to_copy; i++) {
+                                       if (dp[i] != tag[i])
+                                               break;
+                                       total_size--;
+                               }
+                               dm_bufio_release(b);
+                               return total_size;
+                       }
+               }
+               dm_bufio_release(b);
+
+               tag += to_copy;
+               *metadata_offset += to_copy;
+               if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
+                       (*metadata_block)++;
+                       *metadata_offset = 0;
+               }
+               total_size -= to_copy;
+       } while (unlikely(total_size));
+
+       return 0;
+}
+
+static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
+{
+       int r;
+       r = dm_bufio_write_dirty_buffers(ic->bufio);
+       if (unlikely(r))
+               dm_integrity_io_error(ic, "writing tags", r);
+}
+
+static void sleep_on_endio_wait(struct dm_integrity_c *ic)
+{
+       DECLARE_WAITQUEUE(wait, current);
+       __add_wait_queue(&ic->endio_wait, &wait);
+       __set_current_state(TASK_UNINTERRUPTIBLE);
+       spin_unlock_irq(&ic->endio_wait.lock);
+       io_schedule();
+       spin_lock_irq(&ic->endio_wait.lock);
+       __remove_wait_queue(&ic->endio_wait, &wait);
+}
+
+static void autocommit_fn(unsigned long data)
+{
+       struct dm_integrity_c *ic = (struct dm_integrity_c *)data;
+
+       if (likely(!dm_integrity_failed(ic)))
+               queue_work(ic->commit_wq, &ic->commit_work);
+}
+
+static void schedule_autocommit(struct dm_integrity_c *ic)
+{
+       if (!timer_pending(&ic->autocommit_timer))
+               mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
+}
+
+static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
+{
+       struct bio *bio;
+       spin_lock_irq(&ic->endio_wait.lock);
+       bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
+       bio_list_add(&ic->flush_bio_list, bio);
+       spin_unlock_irq(&ic->endio_wait.lock);
+       queue_work(ic->commit_wq, &ic->commit_work);
+}
+
+static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
+{
+       int r = dm_integrity_failed(ic);
+       if (unlikely(r) && !bio->bi_error)
+               bio->bi_error = r;
+       bio_endio(bio);
+}
+
+static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
+{
+       struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
+
+       if (unlikely(dio->fua) && likely(!bio->bi_error) && likely(!dm_integrity_failed(ic)))
+               submit_flush_bio(ic, dio);
+       else
+               do_endio(ic, bio);
+}
+
+static void dec_in_flight(struct dm_integrity_io *dio)
+{
+       if (atomic_dec_and_test(&dio->in_flight)) {
+               struct dm_integrity_c *ic = dio->ic;
+               struct bio *bio;
+
+               remove_range(ic, &dio->range);
+
+               if (unlikely(dio->write))
+                       schedule_autocommit(ic);
+
+               bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
+
+               if (unlikely(dio->bi_error) && !bio->bi_error)
+                       bio->bi_error = dio->bi_error;
+               if (likely(!bio->bi_error) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
+                       dio->range.logical_sector += dio->range.n_sectors;
+                       bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
+                       INIT_WORK(&dio->work, integrity_bio_wait);
+                       queue_work(ic->wait_wq, &dio->work);
+                       return;
+               }
+               do_endio_flush(ic, dio);
+       }
+}
+
+static void integrity_end_io(struct bio *bio)
+{
+       struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
+
+       bio->bi_iter = dio->orig_bi_iter;
+       bio->bi_bdev = dio->orig_bi_bdev;
+       if (dio->orig_bi_integrity) {
+               bio->bi_integrity = dio->orig_bi_integrity;
+               bio->bi_opf |= REQ_INTEGRITY;
+       }
+       bio->bi_end_io = dio->orig_bi_end_io;
+
+       if (dio->completion)
+               complete(dio->completion);
+
+       dec_in_flight(dio);
+}
+
+static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
+                                     const char *data, char *result)
+{
+       __u64 sector_le = cpu_to_le64(sector);
+       SHASH_DESC_ON_STACK(req, ic->internal_hash);
+       int r;
+       unsigned digest_size;
+
+       req->tfm = ic->internal_hash;
+       req->flags = 0;
+
+       r = crypto_shash_init(req);
+       if (unlikely(r < 0)) {
+               dm_integrity_io_error(ic, "crypto_shash_init", r);
+               goto failed;
+       }
+
+       r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
+       if (unlikely(r < 0)) {
+               dm_integrity_io_error(ic, "crypto_shash_update", r);
+               goto failed;
+       }
+
+       r = crypto_shash_update(req, data, 1 << SECTOR_SHIFT);
+       if (unlikely(r < 0)) {
+               dm_integrity_io_error(ic, "crypto_shash_update", r);
+               goto failed;
+       }
+
+       r = crypto_shash_final(req, result);
+       if (unlikely(r < 0)) {
+               dm_integrity_io_error(ic, "crypto_shash_final", r);
+               goto failed;
+       }
+
+       digest_size = crypto_shash_digestsize(ic->internal_hash);
+       if (unlikely(digest_size < ic->tag_size))
+               memset(result + digest_size, 0, ic->tag_size - digest_size);
+
+       return;
+
+failed:
+       /* this shouldn't happen anyway, the hash functions have no reason to fail */
+       get_random_bytes(result, ic->tag_size);
+}
+
+static void integrity_metadata(struct work_struct *w)
+{
+       struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
+       struct dm_integrity_c *ic = dio->ic;
+
+       int r;
+
+       if (ic->internal_hash) {
+               struct bvec_iter iter;
+               struct bio_vec bv;
+               unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
+               struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
+               char *checksums;
+               unsigned extra_space = digest_size > ic->tag_size ? digest_size - ic->tag_size : 0;
+               char checksums_onstack[ic->tag_size + extra_space];
+               unsigned sectors_to_process = dio->range.n_sectors;
+               sector_t sector = dio->range.logical_sector;
+
+               checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT) * ic->tag_size + extra_space,
+                                   GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
+               if (!checksums)
+                       checksums = checksums_onstack;
+
+               __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
+                       unsigned pos;
+                       char *mem, *checksums_ptr;
+
+again:
+                       mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
+                       pos = 0;
+                       checksums_ptr = checksums;
+                       do {
+                               integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
+                               checksums_ptr += ic->tag_size;
+                               sectors_to_process--;
+                               pos += 1 << SECTOR_SHIFT;
+                               sector++;
+                       } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
+                       kunmap_atomic(mem);
+
+                       r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
+                                               checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
+                       if (unlikely(r)) {
+                               if (r > 0) {
+                                       DMERR("Checksum failed at sector 0x%llx",
+                                             (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
+                                       r = -EILSEQ;
+                               }
+                               if (likely(checksums != checksums_onstack))
+                                       kfree(checksums);
+                               goto error;
+                       }
+
+                       if (!sectors_to_process)
+                               break;
+
+                       if (unlikely(pos < bv.bv_len)) {
+                               bv.bv_offset += pos;
+                               bv.bv_len -= pos;
+                               goto again;
+                       }
+               }
+
+               if (likely(checksums != checksums_onstack))
+                       kfree(checksums);
+       } else {
+               struct bio_integrity_payload *bip = dio->orig_bi_integrity;
+
+               if (bip) {
+                       struct bio_vec biv;
+                       struct bvec_iter iter;
+                       unsigned data_to_process = dio->range.n_sectors * ic->tag_size;
+
+                       bip_for_each_vec(biv, bip, iter) {
+                               unsigned char *tag;
+                               unsigned this_len;
+
+                               BUG_ON(PageHighMem(biv.bv_page));
+                               tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
+                               this_len = min(biv.bv_len, data_to_process);
+                               r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
+                                                       this_len, !dio->write ? TAG_READ : TAG_WRITE);
+                               if (unlikely(r))
+                                       goto error;
+                               data_to_process -= this_len;
+                               if (!data_to_process)
+                                       break;
+                       }
+               }
+       }
+       dec_in_flight(dio);
+       return;
+error:
+       dio->bi_error = r;
+       dec_in_flight(dio);
+}
+
+static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
+{
+       struct dm_integrity_c *ic = ti->private;
+       struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
+
+       sector_t area, offset;
+
+       dio->ic = ic;
+       dio->bi_error = 0;
+
+       if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
+               submit_flush_bio(ic, dio);
+               return DM_MAPIO_SUBMITTED;
+       }
+
+       dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
+       dio->write = bio_op(bio) == REQ_OP_WRITE;
+       dio->fua = dio->write && bio->bi_opf & REQ_FUA;
+       if (unlikely(dio->fua)) {
+               /*
+                * Don't pass down the FUA flag because we have to flush
+                * disk cache anyway.
+                */
+               bio->bi_opf &= ~REQ_FUA;
+       }
+       if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
+               DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
+                     (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
+                     (unsigned long long)ic->provided_data_sectors);
+               return -EIO;
+       }
+
+       get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
+       dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
+       bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
+
+       dm_integrity_map_continue(dio, true);
+       return DM_MAPIO_SUBMITTED;
+}
+
+static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
+                                unsigned journal_section, unsigned journal_entry)
+{
+       struct dm_integrity_c *ic = dio->ic;
+       sector_t logical_sector;
+       unsigned n_sectors;
+
+       logical_sector = dio->range.logical_sector;
+       n_sectors = dio->range.n_sectors;
+       do {
+               struct bio_vec bv = bio_iovec(bio);
+               char *mem;
+
+               if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
+                       bv.bv_len = n_sectors << SECTOR_SHIFT;
+               n_sectors -= bv.bv_len >> SECTOR_SHIFT;
+               bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
+retry_kmap:
+               mem = kmap_atomic(bv.bv_page);
+               if (likely(dio->write))
+                       flush_dcache_page(bv.bv_page);
+
+               do {
+                       struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
+
+                       if (unlikely(!dio->write)) {
+                               struct journal_sector *js;
+
+                               if (unlikely(journal_entry_is_inprogress(je))) {
+                                       flush_dcache_page(bv.bv_page);
+                                       kunmap_atomic(mem);
+
+                                       __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
+                                       goto retry_kmap;
+                               }
+                               smp_rmb();
+                               BUG_ON(journal_entry_get_sector(je) != logical_sector);
+                               js = access_journal_data(ic, journal_section, journal_entry);
+                               memcpy(mem + bv.bv_offset, js, JOURNAL_SECTOR_DATA);
+                               memcpy(mem + bv.bv_offset + JOURNAL_SECTOR_DATA, &je->last_bytes, sizeof je->last_bytes);
+#ifdef INTERNAL_VERIFY
+                               if (ic->internal_hash) {
+                                       char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
+
+                                       integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
+                                       if (unlikely(memcmp(checksums_onstack, je->tag, ic->tag_size))) {
+                                               DMERR("Checksum failed when reading from journal, at sector 0x%llx",
+                                                     (unsigned long long)logical_sector);
+                                       }
+                               }
+#endif
+                       }
+
+                       if (!ic->internal_hash) {
+                               struct bio_integrity_payload *bip = bio_integrity(bio);
+                               unsigned tag_todo = ic->tag_size;
+                               char *tag_ptr = je->tag;
+
+                               if (bip) do {
+                                       struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
+                                       unsigned tag_now = min(biv.bv_len, tag_todo);
+                                       char *tag_addr;
+                                       BUG_ON(PageHighMem(biv.bv_page));
+                                       tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
+                                       if (likely(dio->write))
+                                               memcpy(tag_ptr, tag_addr, tag_now);
+                                       else
+                                               memcpy(tag_addr, tag_ptr, tag_now);
+                                       bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
+                                       tag_ptr += tag_now;
+                                       tag_todo -= tag_now;
+                               } while (unlikely(tag_todo)); else {
+                                       if (likely(dio->write))
+                                               memset(tag_ptr, 0, tag_todo);
+                               }
+                       }
+
+                       if (likely(dio->write)) {
+                               struct journal_sector *js;
+
+                               js = access_journal_data(ic, journal_section, journal_entry);
+                               memcpy(js, mem + bv.bv_offset, 1 << SECTOR_SHIFT);
+                               je->last_bytes = js->commit_id;
+
+                               if (ic->internal_hash) {
+                                       unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
+                                       if (unlikely(digest_size > ic->tag_size)) {
+                                               char checksums_onstack[digest_size];
+                                               integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
+                                               memcpy(je->tag, checksums_onstack, ic->tag_size);
+                                       } else
+                                               integrity_sector_checksum(ic, logical_sector, (char *)js, je->tag);
+                               }
+
+                               journal_entry_set_sector(je, logical_sector);
+                       }
+                       logical_sector++;
+
+                       journal_entry++;
+                       if (unlikely(journal_entry == ic->journal_section_entries)) {
+                               journal_entry = 0;
+                               journal_section++;
+                               wraparound_section(ic, &journal_section);
+                       }
+
+                       bv.bv_offset += 1 << SECTOR_SHIFT;
+               } while (bv.bv_len -= 1 << SECTOR_SHIFT);
+
+               if (unlikely(!dio->write))
+                       flush_dcache_page(bv.bv_page);
+               kunmap_atomic(mem);
+       } while (n_sectors);
+
+       if (likely(dio->write)) {
+               smp_mb();
+               if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
+                       wake_up(&ic->copy_to_journal_wait);
+               if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
+                       queue_work(ic->commit_wq, &ic->commit_work);
+               } else {
+                       schedule_autocommit(ic);
+               }
+       } else {
+               remove_range(ic, &dio->range);
+       }
+
+       if (unlikely(bio->bi_iter.bi_size)) {
+               sector_t area, offset;
+
+               dio->range.logical_sector = logical_sector;
+               get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
+               dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
+               return true;
+       }
+
+       return false;
+}
+
+static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
+{
+       struct dm_integrity_c *ic = dio->ic;
+       struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
+       unsigned journal_section, journal_entry;
+       unsigned journal_read_pos;
+       struct completion read_comp;
+       bool need_sync_io = ic->internal_hash && !dio->write;
+
+       if (need_sync_io && from_map) {
+               INIT_WORK(&dio->work, integrity_bio_wait);
+               queue_work(ic->metadata_wq, &dio->work);
+               return;
+       }
+
+lock_retry:
+       spin_lock_irq(&ic->endio_wait.lock);
+retry:
+       if (unlikely(dm_integrity_failed(ic))) {
+               spin_unlock_irq(&ic->endio_wait.lock);
+               do_endio(ic, bio);
+               return;
+       }
+       dio->range.n_sectors = bio_sectors(bio);
+       journal_read_pos = NOT_FOUND;
+       if (likely(ic->mode == 'J')) {
+               if (dio->write) {
+                       unsigned next_entry, i, pos;
+                       unsigned ws, we;
+
+                       dio->range.n_sectors = min(dio->range.n_sectors, ic->free_sectors);
+                       if (unlikely(!dio->range.n_sectors))
+                               goto sleep;
+                       ic->free_sectors -= dio->range.n_sectors;
+                       journal_section = ic->free_section;
+                       journal_entry = ic->free_section_entry;
+
+                       next_entry = ic->free_section_entry + dio->range.n_sectors;
+                       ic->free_section_entry = next_entry % ic->journal_section_entries;
+                       ic->free_section += next_entry / ic->journal_section_entries;
+                       ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
+                       wraparound_section(ic, &ic->free_section);
+
+                       pos = journal_section * ic->journal_section_entries + journal_entry;
+                       ws = journal_section;
+                       we = journal_entry;
+                       for (i = 0; i < dio->range.n_sectors; i++) {
+                               struct journal_entry *je;
+
+                               add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
+                               pos++;
+                               if (unlikely(pos >= ic->journal_entries))
+                                       pos = 0;
+
+                               je = access_journal_entry(ic, ws, we);
+                               BUG_ON(!journal_entry_is_unused(je));
+                               journal_entry_set_inprogress(je);
+                               we++;
+                               if (unlikely(we == ic->journal_section_entries)) {
+                                       we = 0;
+                                       ws++;
+                                       wraparound_section(ic, &ws);
+                               }
+                       }
+
+                       spin_unlock_irq(&ic->endio_wait.lock);
+                       goto journal_read_write;
+               } else {
+                       sector_t next_sector;
+                       journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
+                       if (likely(journal_read_pos == NOT_FOUND)) {
+                               if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
+                                       dio->range.n_sectors = next_sector - dio->range.logical_sector;
+                       } else {
+                               unsigned i;
+                               for (i = 1; i < dio->range.n_sectors; i++) {
+                                       if (!test_journal_node(ic, journal_read_pos + i, dio->range.logical_sector + i))
+                                               break;
+                               }
+                               dio->range.n_sectors = i;
+                       }
+               }
+       }
+       if (unlikely(!add_new_range(ic, &dio->range))) {
+               /*
+                * We must not sleep in the request routine because it could
+                * stall bios on current->bio_list.
+                * So, we offload the bio to a workqueue if we have to sleep.
+                */
+sleep:
+               if (from_map) {
+                       spin_unlock_irq(&ic->endio_wait.lock);
+                       INIT_WORK(&dio->work, integrity_bio_wait);
+                       queue_work(ic->wait_wq, &dio->work);
+                       return;
+               } else {
+                       sleep_on_endio_wait(ic);
+                       goto retry;
+               }
+       }
+       spin_unlock_irq(&ic->endio_wait.lock);
+
+       if (unlikely(journal_read_pos != NOT_FOUND)) {
+               journal_section = journal_read_pos / ic->journal_section_entries;
+               journal_entry = journal_read_pos % ic->journal_section_entries;
+               goto journal_read_write;
+       }
+
+       dio->in_flight = (atomic_t)ATOMIC_INIT(2);
+
+       if (need_sync_io) {
+               read_comp = COMPLETION_INITIALIZER_ONSTACK(read_comp);
+               dio->completion = &read_comp;
+       } else
+               dio->completion = NULL;
+
+       dio->orig_bi_iter = bio->bi_iter;
+
+       dio->orig_bi_bdev = bio->bi_bdev;
+       bio->bi_bdev = ic->dev->bdev;
+
+       dio->orig_bi_integrity = bio_integrity(bio);
+       bio->bi_integrity = NULL;
+       bio->bi_opf &= ~REQ_INTEGRITY;
+
+       dio->orig_bi_end_io = bio->bi_end_io;
+       bio->bi_end_io = integrity_end_io;
+
+       bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
+       bio->bi_iter.bi_sector += ic->start;
+       generic_make_request(bio);
+
+       if (need_sync_io) {
+               wait_for_completion_io(&read_comp);
+               integrity_metadata(&dio->work);
+       } else {
+               INIT_WORK(&dio->work, integrity_metadata);
+               queue_work(ic->metadata_wq, &dio->work);
+       }
+
+       return;
+
+journal_read_write:
+       if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
+               goto lock_retry;
+
+       do_endio_flush(ic, dio);
+}
+
+
+static void integrity_bio_wait(struct work_struct *w)
+{
+       struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
+
+       dm_integrity_map_continue(dio, false);
+}
+
+static void pad_uncommitted(struct dm_integrity_c *ic)
+{
+       if (ic->free_section_entry) {
+               ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
+               ic->free_section_entry = 0;
+               ic->free_section++;
+               wraparound_section(ic, &ic->free_section);
+               ic->n_uncommitted_sections++;
+       }
+}
+
+static void integrity_commit(struct work_struct *w)
+{
+       struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
+       unsigned commit_start, commit_sections;
+       unsigned i, j, n;
+       struct bio *flushes;
+
+       del_timer(&ic->autocommit_timer);
+
+       spin_lock_irq(&ic->endio_wait.lock);
+       flushes = bio_list_get(&ic->flush_bio_list);
+       if (unlikely(ic->mode != 'J')) {
+               spin_unlock_irq(&ic->endio_wait.lock);
+               dm_integrity_flush_buffers(ic);
+               goto release_flush_bios;
+       }
+
+       pad_uncommitted(ic);
+       commit_start = ic->uncommitted_section;
+       commit_sections = ic->n_uncommitted_sections;
+       spin_unlock_irq(&ic->endio_wait.lock);
+
+       if (!commit_sections)
+               goto release_flush_bios;
+
+       i = commit_start;
+       for (n = 0; n < commit_sections; n++) {
+               for (j = 0; j < ic->journal_section_entries; j++) {
+                       struct journal_entry *je;
+                       je = access_journal_entry(ic, i, j);
+                       io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
+               }
+               for (j = 0; j < ic->journal_section_sectors; j++) {
+                       struct journal_sector *js;
+                       js = access_journal(ic, i, j);
+                       js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
+               }
+               i++;
+               if (unlikely(i >= ic->journal_sections))
+                       ic->commit_seq = next_commit_seq(ic->commit_seq);
+               wraparound_section(ic, &i);
+       }
+       smp_rmb();
+
+       write_journal(ic, commit_start, commit_sections);
+
+       spin_lock_irq(&ic->endio_wait.lock);
+       ic->uncommitted_section += commit_sections;
+       wraparound_section(ic, &ic->uncommitted_section);
+       ic->n_uncommitted_sections -= commit_sections;
+       ic->n_committed_sections += commit_sections;
+       spin_unlock_irq(&ic->endio_wait.lock);
+
+       if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
+               queue_work(ic->writer_wq, &ic->writer_work);
+
+release_flush_bios:
+       while (flushes) {
+               struct bio *next = flushes->bi_next;
+               flushes->bi_next = NULL;
+               do_endio(ic, flushes);
+               flushes = next;
+       }
+}
+
+static void complete_copy_from_journal(unsigned long error, void *context)
+{
+       struct journal_io *io = context;
+       struct journal_completion *comp = io->comp;
+       struct dm_integrity_c *ic = comp->ic;
+       remove_range(ic, &io->range);
+       mempool_free(io, ic->journal_io_mempool);
+       if (unlikely(error != 0))
+               dm_integrity_io_error(ic, "copying from journal", -EIO);
+       complete_journal_op(comp);
+}
+
+static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
+                            unsigned write_sections, bool from_replay)
+{
+       unsigned i, j, n;
+       struct journal_completion comp;
+
+       comp.ic = ic;
+       comp.in_flight = (atomic_t)ATOMIC_INIT(1);
+       comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
+
+       i = write_start;
+       for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
+#ifndef INTERNAL_VERIFY
+               if (unlikely(from_replay))
+#endif
+                       rw_section_mac(ic, i, false);
+               for (j = 0; j < ic->journal_section_entries; j++) {
+                       struct journal_entry *je = access_journal_entry(ic, i, j);
+                       sector_t sec, area, offset;
+                       unsigned k, l, next_loop;
+                       sector_t metadata_block;
+                       unsigned metadata_offset;
+                       struct journal_io *io;
+
+                       if (journal_entry_is_unused(je))
+                               continue;
+                       BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
+                       sec = journal_entry_get_sector(je);
+                       get_area_and_offset(ic, sec, &area, &offset);
+                       access_journal_data(ic, i, j)->commit_id = je->last_bytes;
+                       for (k = j + 1; k < ic->journal_section_entries; k++) {
+                               struct journal_entry *je2 = access_journal_entry(ic, i, k);
+                               sector_t sec2, area2, offset2;
+                               if (journal_entry_is_unused(je2))
+                                       break;
+                               BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
+                               sec2 = journal_entry_get_sector(je2);
+                               get_area_and_offset(ic, sec2, &area2, &offset2);
+                               if (area2 != area || offset2 != offset + (k - j))
+                                       break;
+                               access_journal_data(ic, i, k)->commit_id = je2->last_bytes;
+                       }
+                       next_loop = k - 1;
+
+                       io = mempool_alloc(ic->journal_io_mempool, GFP_NOIO);
+                       io->comp = &comp;
+                       io->range.logical_sector = sec;
+                       io->range.n_sectors = k - j;
+
+                       spin_lock_irq(&ic->endio_wait.lock);
+                       while (unlikely(!add_new_range(ic, &io->range)))
+                               sleep_on_endio_wait(ic);
+
+                       if (likely(!from_replay)) {
+                               struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
+
+                               /* don't write if there is newer committed sector */
+                               while (j < k && find_newer_committed_node(ic, &section_node[j])) {
+                                       struct journal_entry *je2 = access_journal_entry(ic, i, j);
+
+                                       journal_entry_set_unused(je2);
+                                       remove_journal_node(ic, &section_node[j]);
+                                       j++;
+                                       sec++;
+                                       offset++;
+                               }
+                               while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
+                                       struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
+
+                                       journal_entry_set_unused(je2);
+                                       remove_journal_node(ic, &section_node[k - 1]);
+                                       k--;
+                               }
+                               if (j == k) {
+                                       remove_range_unlocked(ic, &io->range);
+                                       spin_unlock_irq(&ic->endio_wait.lock);
+                                       mempool_free(io, ic->journal_io_mempool);
+                                       goto skip_io;
+                               }
+                               for (l = j; l < k; l++) {
+                                       remove_journal_node(ic, &section_node[l]);
+                               }
+                       }
+                       spin_unlock_irq(&ic->endio_wait.lock);
+
+                       metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
+                       for (l = j; l < k; l++) {
+                               int r;
+                               struct journal_entry *je2 = access_journal_entry(ic, i, l);
+
+                               if (
+#ifndef INTERNAL_VERIFY
+                                   unlikely(from_replay) &&
+#endif
+                                   ic->internal_hash) {
+                                       unsigned char test_tag[ic->tag_size];
+
+                                       integrity_sector_checksum(ic, sec + (l - j),
+                                                                 (char *)access_journal_data(ic, i, l), test_tag);
+                                       if (unlikely(memcmp(test_tag, je2->tag, ic->tag_size)))
+                                               dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
+                               }
+
+                               journal_entry_set_unused(je2);
+                               r = dm_integrity_rw_tag(ic, je2->tag, &metadata_block, &metadata_offset,
+                                                       ic->tag_size, TAG_WRITE);
+                               if (unlikely(r)) {
+                                       dm_integrity_io_error(ic, "reading tags", r);
+                               }
+                       }
+
+                       atomic_inc(&comp.in_flight);
+                       copy_from_journal(ic, i, j, k - j, get_data_sector(ic, area, offset),
+                                         complete_copy_from_journal, io);
+skip_io:
+                       j = next_loop;
+               }
+       }
+
+       dm_bufio_write_dirty_buffers_async(ic->bufio);
+
+       complete_journal_op(&comp);
+       wait_for_completion_io(&comp.comp);
+
+       dm_integrity_flush_buffers(ic);
+}
+
+static void integrity_writer(struct work_struct *w)
+{
+       struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
+       unsigned write_start, write_sections;
+
+       unsigned prev_free_sectors;
+
+       /* the following test is not needed, but it tests the replay code */
+       if (ACCESS_ONCE(ic->suspending))
+               return;
+
+       spin_lock_irq(&ic->endio_wait.lock);
+       write_start = ic->committed_section;
+       write_sections = ic->n_committed_sections;
+       spin_unlock_irq(&ic->endio_wait.lock);
+
+       if (!write_sections)
+               return;
+
+       do_journal_write(ic, write_start, write_sections, false);
+
+       spin_lock_irq(&ic->endio_wait.lock);
+
+       ic->committed_section += write_sections;
+       wraparound_section(ic, &ic->committed_section);
+       ic->n_committed_sections -= write_sections;
+
+       prev_free_sectors = ic->free_sectors;
+       ic->free_sectors += write_sections * ic->journal_section_entries;
+       if (unlikely(!prev_free_sectors))
+               wake_up_locked(&ic->endio_wait);
+
+       spin_unlock_irq(&ic->endio_wait.lock);
+}
+
+static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
+                        unsigned n_sections, unsigned char commit_seq)
+{
+       unsigned i, j, n;
+
+       if (!n_sections)
+               return;
+
+       for (n = 0; n < n_sections; n++) {
+               i = start_section + n;
+               wraparound_section(ic, &i);
+               for (j = 0; j < ic->journal_section_sectors; j++) {
+                       struct journal_sector *js = access_journal(ic, i, j);
+                       memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
+                       js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
+               }
+               for (j = 0; j < ic->journal_section_entries; j++) {
+                       struct journal_entry *je = access_journal_entry(ic, i, j);
+                       journal_entry_set_unused(je);
+               }
+       }
+
+       write_journal(ic, start_section, n_sections);
+}
+
+static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
+{
+       unsigned char k;
+       for (k = 0; k < N_COMMIT_IDS; k++) {
+               if (dm_integrity_commit_id(ic, i, j, k) == id)
+                       return k;
+       }
+       dm_integrity_io_error(ic, "journal commit id", -EIO);
+       return -EIO;
+}
+
+static void replay_journal(struct dm_integrity_c *ic)
+{
+       unsigned i, j;
+       bool used_commit_ids[N_COMMIT_IDS];
+       unsigned max_commit_id_sections[N_COMMIT_IDS];
+       unsigned write_start, write_sections;
+       unsigned continue_section;
+       bool journal_empty;
+       unsigned char unused, last_used, want_commit_seq;
+
+       if (ic->journal_uptodate)
+               return;
+
+       last_used = 0;
+       write_start = 0;
+
+       if (!ic->just_formatted) {
+               DEBUG_print("reading journal\n");
+               rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
+               if (ic->journal_io)
+                       DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
+               if (ic->journal_io) {
+                       struct journal_completion crypt_comp;
+                       crypt_comp.ic = ic;
+                       crypt_comp.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp.comp);
+                       crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
+                       encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
+                       wait_for_completion(&crypt_comp.comp);
+               }
+               DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
+       }
+
+       if (dm_integrity_failed(ic))
+               goto clear_journal;
+
+       journal_empty = true;
+       memset(used_commit_ids, 0, sizeof used_commit_ids);
+       memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
+       for (i = 0; i < ic->journal_sections; i++) {
+               for (j = 0; j < ic->journal_section_sectors; j++) {
+                       int k;
+                       struct journal_sector *js = access_journal(ic, i, j);
+                       k = find_commit_seq(ic, i, j, js->commit_id);
+                       if (k < 0)
+                               goto clear_journal;
+                       used_commit_ids[k] = true;
+                       max_commit_id_sections[k] = i;
+               }
+               if (journal_empty) {
+                       for (j = 0; j < ic->journal_section_entries; j++) {
+                               struct journal_entry *je = access_journal_entry(ic, i, j);
+                               if (!journal_entry_is_unused(je)) {
+                                       journal_empty = false;
+                                       break;
+                               }
+                       }
+               }
+       }
+
+       if (!used_commit_ids[N_COMMIT_IDS - 1]) {
+               unused = N_COMMIT_IDS - 1;
+               while (unused && !used_commit_ids[unused - 1])
+                       unused--;
+       } else {
+               for (unused = 0; unused < N_COMMIT_IDS; unused++)
+                       if (!used_commit_ids[unused])
+                               break;
+               if (unused == N_COMMIT_IDS) {
+                       dm_integrity_io_error(ic, "journal commit ids", -EIO);
+                       goto clear_journal;
+               }
+       }
+       DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
+                   unused, used_commit_ids[0], used_commit_ids[1],
+                   used_commit_ids[2], used_commit_ids[3]);
+
+       last_used = prev_commit_seq(unused);
+       want_commit_seq = prev_commit_seq(last_used);
+
+       if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
+               journal_empty = true;
+
+       write_start = max_commit_id_sections[last_used] + 1;
+       if (unlikely(write_start >= ic->journal_sections))
+               want_commit_seq = next_commit_seq(want_commit_seq);
+       wraparound_section(ic, &write_start);
+
+       i = write_start;
+       for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
+               for (j = 0; j < ic->journal_section_sectors; j++) {
+                       struct journal_sector *js = access_journal(ic, i, j);
+
+                       if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
+                               /*
+                                * This could be caused by crash during writing.
+                                * We won't replay the inconsistent part of the
+                                * journal.
+                                */
+                               DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
+                                           i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
+                               goto brk;
+                       }
+               }
+               i++;
+               if (unlikely(i >= ic->journal_sections))
+                       want_commit_seq = next_commit_seq(want_commit_seq);
+               wraparound_section(ic, &i);
+       }
+brk:
+
+       if (!journal_empty) {
+               DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
+                           write_sections, write_start, want_commit_seq);
+               do_journal_write(ic, write_start, write_sections, true);
+       }
+
+       if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
+               continue_section = write_start;
+               ic->commit_seq = want_commit_seq;
+               DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
+       } else {
+               unsigned s;
+               unsigned char erase_seq;
+clear_journal:
+               DEBUG_print("clearing journal\n");
+
+               erase_seq = prev_commit_seq(prev_commit_seq(last_used));
+               s = write_start;
+               init_journal(ic, s, 1, erase_seq);
+               s++;
+               wraparound_section(ic, &s);
+               if (ic->journal_sections >= 2) {
+                       init_journal(ic, s, ic->journal_sections - 2, erase_seq);
+                       s += ic->journal_sections - 2;
+                       wraparound_section(ic, &s);
+                       init_journal(ic, s, 1, erase_seq);
+               }
+
+               continue_section = 0;
+               ic->commit_seq = next_commit_seq(erase_seq);
+       }
+
+       ic->committed_section = continue_section;
+       ic->n_committed_sections = 0;
+
+       ic->uncommitted_section = continue_section;
+       ic->n_uncommitted_sections = 0;
+
+       ic->free_section = continue_section;
+       ic->free_section_entry = 0;
+       ic->free_sectors = ic->journal_entries;
+
+       ic->journal_tree_root = RB_ROOT;
+       for (i = 0; i < ic->journal_entries; i++)
+               init_journal_node(&ic->journal_tree[i]);
+}
+
+static void dm_integrity_postsuspend(struct dm_target *ti)
+{
+       struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
+
+       del_timer_sync(&ic->autocommit_timer);
+
+       ic->suspending = true;
+
+       queue_work(ic->commit_wq, &ic->commit_work);
+       drain_workqueue(ic->commit_wq);
+
+       if (ic->mode == 'J') {
+               drain_workqueue(ic->writer_wq);
+               dm_integrity_flush_buffers(ic);
+       }
+
+       ic->suspending = false;
+
+       BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
+
+       ic->journal_uptodate = true;
+}
+
+static void dm_integrity_resume(struct dm_target *ti)
+{
+       struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
+
+       replay_journal(ic);
+}
+
+static void dm_integrity_status(struct dm_target *ti, status_type_t type,
+                               unsigned status_flags, char *result, unsigned maxlen)
+{
+       struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
+       unsigned arg_count;
+       size_t sz = 0;
+
+       switch (type) {
+       case STATUSTYPE_INFO:
+               result[0] = '\0';
+               break;
+
+       case STATUSTYPE_TABLE: {
+               __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
+               watermark_percentage += ic->journal_entries / 2;
+               do_div(watermark_percentage, ic->journal_entries);
+               arg_count = 5;
+               arg_count += !!ic->internal_hash_alg.alg_string;
+               arg_count += !!ic->journal_crypt_alg.alg_string;
+               arg_count += !!ic->journal_mac_alg.alg_string;
+               DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
+                      ic->tag_size, ic->mode, arg_count);
+               DMEMIT(" journal-sectors:%u", ic->initial_sectors - SB_SECTORS);
+               DMEMIT(" interleave-sectors:%u", 1U << ic->sb->log2_interleave_sectors);
+               DMEMIT(" buffer-sectors:%u", 1U << ic->log2_buffer_sectors);
+               DMEMIT(" journal-watermark:%u", (unsigned)watermark_percentage);
+               DMEMIT(" commit-time:%u", ic->autocommit_msec);
+
+#define EMIT_ALG(a, n)                                                 \
+               do {                                                    \
+                       if (ic->a.alg_string) {                         \
+                               DMEMIT(" %s:%s", n, ic->a.alg_string);  \
+                               if (ic->a.key_string)                   \
+                                       DMEMIT(":%s", ic->a.key_string);\
+                       }                                               \
+               } while (0)
+               EMIT_ALG(internal_hash_alg, "internal-hash");
+               EMIT_ALG(journal_crypt_alg, "journal-crypt");
+               EMIT_ALG(journal_mac_alg, "journal-mac");
+               break;
+       }
+       }
+}
+
+static int dm_integrity_iterate_devices(struct dm_target *ti,
+                                       iterate_devices_callout_fn fn, void *data)
+{
+       struct dm_integrity_c *ic = ti->private;
+
+       return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
+}
+
+static void calculate_journal_section_size(struct dm_integrity_c *ic)
+{
+       unsigned sector_space = JOURNAL_SECTOR_DATA;
+
+       ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
+       ic->journal_entry_size = roundup(offsetof(struct journal_entry, tag) + ic->tag_size,
+                                        JOURNAL_ENTRY_ROUNDUP);
+
+       if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
+               sector_space -= JOURNAL_MAC_PER_SECTOR;
+       ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
+       ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
+       ic->journal_section_sectors = ic->journal_section_entries + JOURNAL_BLOCK_SECTORS;
+       ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
+}
+
+static int calculate_device_limits(struct dm_integrity_c *ic)
+{
+       __u64 initial_sectors;
+       sector_t last_sector, last_area, last_offset;
+
+       calculate_journal_section_size(ic);
+       initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
+       if (initial_sectors + METADATA_PADDING_SECTORS >= ic->device_sectors || initial_sectors > UINT_MAX)
+               return -EINVAL;
+       ic->initial_sectors = initial_sectors;
+
+       ic->metadata_run = roundup((__u64)ic->tag_size << ic->sb->log2_interleave_sectors,
+                                  (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
+       if (!(ic->metadata_run & (ic->metadata_run - 1)))
+               ic->log2_metadata_run = __ffs(ic->metadata_run);
+       else
+               ic->log2_metadata_run = -1;
+
+       get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
+       last_sector = get_data_sector(ic, last_area, last_offset);
+
+       if (ic->start + last_sector < last_sector || ic->start + last_sector >= ic->device_sectors)
+               return -EINVAL;
+
+       return 0;
+}
+
+static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
+{
+       unsigned journal_sections;
+       int test_bit;
+
+       memcpy(ic->sb->magic, SB_MAGIC, 8);
+       ic->sb->version = SB_VERSION;
+       ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
+       if (ic->journal_mac_alg.alg_string)
+               ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
+
+       calculate_journal_section_size(ic);
+       journal_sections = journal_sectors / ic->journal_section_sectors;
+       if (!journal_sections)
+               journal_sections = 1;
+       ic->sb->journal_sections = cpu_to_le32(journal_sections);
+
+       ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
+       ic->sb->log2_interleave_sectors = max((__u8)MIN_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
+       ic->sb->log2_interleave_sectors = min((__u8)MAX_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
+
+       ic->provided_data_sectors = 0;
+       for (test_bit = fls64(ic->device_sectors) - 1; test_bit >= 3; test_bit--) {
+               __u64 prev_data_sectors = ic->provided_data_sectors;
+
+               ic->provided_data_sectors |= (sector_t)1 << test_bit;
+               if (calculate_device_limits(ic))
+                       ic->provided_data_sectors = prev_data_sectors;
+       }
+
+       if (!le64_to_cpu(ic->provided_data_sectors))
+               return -EINVAL;
+
+       ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
+
+       return 0;
+}
+
+static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
+{
+       struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
+       struct blk_integrity bi;
+
+       memset(&bi, 0, sizeof(bi));
+       bi.profile = &dm_integrity_profile;
+       bi.tuple_size = ic->tag_size * (queue_logical_block_size(disk->queue) >> SECTOR_SHIFT);
+       bi.tag_size = ic->tag_size;
+
+       blk_integrity_register(disk, &bi);
+       blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
+}
+
+/* FIXME: use new kvmalloc */
+static void *dm_integrity_kvmalloc(size_t size, gfp_t gfp)
+{
+       void *ptr = NULL;
+
+       if (size <= PAGE_SIZE)
+               ptr = kmalloc(size, GFP_KERNEL | gfp);
+       if (!ptr && size <= KMALLOC_MAX_SIZE)
+               ptr = kmalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY | gfp);
+       if (!ptr)
+               ptr = __vmalloc(size, GFP_KERNEL | gfp, PAGE_KERNEL);
+
+       return ptr;
+}
+
+static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
+{
+       unsigned i;
+
+       if (!pl)
+               return;
+       for (i = 0; i < ic->journal_pages; i++)
+               if (pl[i].page)
+                       __free_page(pl[i].page);
+       kvfree(pl);
+}
+
+static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
+{
+       size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
+       struct page_list *pl;
+       unsigned i;
+
+       pl = dm_integrity_kvmalloc(page_list_desc_size, __GFP_ZERO);
+       if (!pl)
+               return NULL;
+
+       for (i = 0; i < ic->journal_pages; i++) {
+               pl[i].page = alloc_page(GFP_KERNEL);
+               if (!pl[i].page) {
+                       dm_integrity_free_page_list(ic, pl);
+                       return NULL;
+               }
+               if (i)
+                       pl[i - 1].next = &pl[i];
+       }
+
+       return pl;
+}
+
+static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
+{
+       unsigned i;
+       for (i = 0; i < ic->journal_sections; i++)
+               kvfree(sl[i]);
+       kfree(sl);
+}
+
+static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
+{
+       struct scatterlist **sl;
+       unsigned i;
+
+       sl = dm_integrity_kvmalloc(ic->journal_sections * sizeof(struct scatterlist *), __GFP_ZERO);
+       if (!sl)
+               return NULL;
+
+       for (i = 0; i < ic->journal_sections; i++) {
+               struct scatterlist *s;
+               unsigned start_index, start_offset;
+               unsigned end_index, end_offset;
+               unsigned n_pages;
+               unsigned idx;
+
+               page_list_location(ic, i, 0, &start_index, &start_offset);
+               page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
+
+               n_pages = (end_index - start_index + 1);
+
+               s = dm_integrity_kvmalloc(n_pages * sizeof(struct scatterlist), 0);
+               if (!s) {
+                       dm_integrity_free_journal_scatterlist(ic, sl);
+                       return NULL;
+               }
+
+               sg_init_table(s, n_pages);
+               for (idx = start_index; idx <= end_index; idx++) {
+                       char *va = lowmem_page_address(pl[idx].page);
+                       unsigned start = 0, end = PAGE_SIZE;
+                       if (idx == start_index)
+                               start = start_offset;
+                       if (idx == end_index)
+                               end = end_offset + (1 << SECTOR_SHIFT);
+                       sg_set_buf(&s[idx - start_index], va + start, end - start);
+               }
+
+               sl[i] = s;
+       }
+
+       return sl;
+}
+
+static void free_alg(struct alg_spec *a)
+{
+       kzfree(a->alg_string);
+       kzfree(a->key);
+       memset(a, 0, sizeof *a);
+}
+
+static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
+{
+       char *k;
+
+       free_alg(a);
+
+       a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
+       if (!a->alg_string)
+               goto nomem;
+
+       k = strchr(a->alg_string, ':');
+       if (k) {
+               unsigned i;
+
+               *k = 0;
+               a->key_string = k + 1;
+               if (strlen(a->key_string) & 1)
+                       goto inval;
+
+               a->key_size = strlen(a->key_string) / 2;
+               a->key = kmalloc(a->key_size, GFP_KERNEL);
+               if (!a->key)
+                       goto nomem;
+               for (i = 0; i < a->key_size; i++) {
+                       char digit[3];
+                       digit[0] = a->key_string[i * 2];
+                       digit[1] = a->key_string[i * 2 + 1];
+                       digit[2] = 0;
+                       if (strspn(digit, "0123456789abcdefABCDEF") != 2)
+                               goto inval;
+                       if (kstrtou8(digit, 16, &a->key[i]))
+                               goto inval;
+               }
+       }
+
+       return 0;
+inval:
+       *error = error_inval;
+       return -EINVAL;
+nomem:
+       *error = "Out of memory for an argument";
+       return -ENOMEM;
+}
+
+static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
+                  char *error_alg, char *error_key)
+{
+       int r;
+
+       if (a->alg_string) {
+               *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
+               if (IS_ERR(*hash)) {
+                       *error = error_alg;
+                       r = PTR_ERR(*hash);
+                       *hash = NULL;
+                       return r;
+               }
+
+               if (a->key) {
+                       r = crypto_shash_setkey(*hash, a->key, a->key_size);
+                       if (r) {
+                               *error = error_key;
+                               return r;
+                       }
+               }
+       }
+
+       return 0;
+}
+
+/*
+ * Construct a integrity mapping: <dev_path> <offset> <tag_size>
+ *
+ * Arguments:
+ *     device
+ *     offset from the start of the device
+ *     tag size
+ *     D - direct writes, J - journal writes
+ *     number of optional arguments
+ *     optional arguments:
+ *             journal-sectors
+ *             interleave-sectors
+ *             buffer-sectors
+ *             journal-watermark
+ *             commit-time
+ *             internal-hash
+ *             journal-crypt
+ *             journal-mac
+ */
+static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
+{
+       struct dm_integrity_c *ic;
+       char dummy;
+       int r;
+       unsigned i;
+       unsigned extra_args;
+       struct dm_arg_set as;
+       static struct dm_arg _args[] = {
+               {0, 7, "Invalid number of feature args"},
+       };
+       unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
+       bool should_write_sb;
+       __u64 journal_pages, journal_desc_size, journal_tree_size;
+       __u64 threshold;
+       unsigned long long start;
+
+#define DIRECT_ARGUMENTS       4
+
+       if (argc <= DIRECT_ARGUMENTS) {
+               ti->error = "Invalid argument count";
+               return -EINVAL;
+       }
+
+       ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
+       if (!ic) {
+               ti->error = "Cannot allocate integrity context";
+               return -ENOMEM;
+       }
+       ti->private = ic;
+       ti->per_io_data_size = sizeof(struct dm_integrity_io);
+
+       ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
+       ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
+       ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
+       ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
+
+       ic->in_progress = RB_ROOT;
+       init_waitqueue_head(&ic->endio_wait);
+       bio_list_init(&ic->flush_bio_list);
+       init_waitqueue_head(&ic->copy_to_journal_wait);
+       init_completion(&ic->crypto_backoff);
+
+       r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
+       if (r) {
+               ti->error = "Device lookup failed";
+               goto bad;
+       }
+
+       if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
+               ti->error = "Invalid starting offset";
+               r = -EINVAL;
+               goto bad;
+       }
+       ic->start = start;
+
+       if (strcmp(argv[2], "-")) {
+               if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
+                       ti->error = "Invalid tag size";
+                       r = -EINVAL;
+                       goto bad;
+               }
+       }
+
+       if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D"))
+               ic->mode = argv[3][0];
+       else {
+               ti->error = "Invalid mode (expecting J or D)";
+               r = -EINVAL;
+               goto bad;
+       }
+
+       ic->device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
+       journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
+                       ic->device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
+       interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
+       buffer_sectors = DEFAULT_BUFFER_SECTORS;
+       journal_watermark = DEFAULT_JOURNAL_WATERMARK;
+       sync_msec = DEFAULT_SYNC_MSEC;
+
+       as.argc = argc - DIRECT_ARGUMENTS;
+       as.argv = argv + DIRECT_ARGUMENTS;
+       r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
+       if (r)
+               goto bad;
+
+       while (extra_args--) {
+               const char *opt_string;
+               unsigned val;
+               opt_string = dm_shift_arg(&as);
+               if (!opt_string) {
+                       r = -EINVAL;
+                       ti->error = "Not enough feature arguments";
+                       goto bad;
+               }
+               if (sscanf(opt_string, "journal-sectors:%u%c", &val, &dummy) == 1)
+                       journal_sectors = val;
+               else if (sscanf(opt_string, "interleave-sectors:%u%c", &val, &dummy) == 1)
+                       interleave_sectors = val;
+               else if (sscanf(opt_string, "buffer-sectors:%u%c", &val, &dummy) == 1)
+                       buffer_sectors = val;
+               else if (sscanf(opt_string, "journal-watermark:%u%c", &val, &dummy) == 1 && val <= 100)
+                       journal_watermark = val;
+               else if (sscanf(opt_string, "commit-time:%u%c", &val, &dummy) == 1)
+                       sync_msec = val;
+               else if (!memcmp(opt_string, "internal-hash:", strlen("internal-hash:"))) {
+                       r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
+                                           "Invalid internal-hash argument");
+                       if (r)
+                               goto bad;
+               } else if (!memcmp(opt_string, "journal-crypt:", strlen("journal-crypt:"))) {
+                       r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
+                                           "Invalid journal-crypt argument");
+                       if (r)
+                               goto bad;
+               } else if (!memcmp(opt_string, "journal-mac:", strlen("journal-mac:"))) {
+                       r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
+                                           "Invalid journal-mac argument");
+                       if (r)
+                               goto bad;
+               } else {
+                       r = -EINVAL;
+                       ti->error = "Invalid argument";
+                       goto bad;
+               }
+       }
+
+       r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
+                   "Invalid internal hash", "Error setting internal hash key");
+       if (r)
+               goto bad;
+
+       r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
+                   "Invalid journal mac", "Error setting journal mac key");
+       if (r)
+               goto bad;
+
+       if (!ic->tag_size) {
+               if (!ic->internal_hash) {
+                       ti->error = "Unknown tag size";
+                       r = -EINVAL;
+                       goto bad;
+               }
+               ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
+       }
+       if (ic->tag_size > MAX_TAG_SIZE) {
+               ti->error = "Too big tag size";
+               r = -EINVAL;
+               goto bad;
+       }
+       if (!(ic->tag_size & (ic->tag_size - 1)))
+               ic->log2_tag_size = __ffs(ic->tag_size);
+       else
+               ic->log2_tag_size = -1;
+
+       ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
+       ic->autocommit_msec = sync_msec;
+       setup_timer(&ic->autocommit_timer, autocommit_fn, (unsigned long)ic);
+
+       ic->io = dm_io_client_create();
+       if (IS_ERR(ic->io)) {
+               r = PTR_ERR(ic->io);
+               ic->io = NULL;
+               ti->error = "Cannot allocate dm io";
+               goto bad;
+       }
+
+       ic->journal_io_mempool = mempool_create_slab_pool(JOURNAL_IO_MEMPOOL, journal_io_cache);
+       if (!ic->journal_io_mempool) {
+               r = -ENOMEM;
+               ti->error = "Cannot allocate mempool";
+               goto bad;
+       }
+
+       ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
+                                         WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
+       if (!ic->metadata_wq) {
+               ti->error = "Cannot allocate workqueue";
+               r = -ENOMEM;
+               goto bad;
+       }
+
+       /*
+        * If this workqueue were percpu, it would cause bio reordering
+        * and reduced performance.
+        */
+       ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
+       if (!ic->wait_wq) {
+               ti->error = "Cannot allocate workqueue";
+               r = -ENOMEM;
+               goto bad;
+       }
+
+       ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
+       if (!ic->commit_wq) {
+               ti->error = "Cannot allocate workqueue";
+               r = -ENOMEM;
+               goto bad;
+       }
+       INIT_WORK(&ic->commit_work, integrity_commit);
+
+       if (ic->mode == 'J') {
+               ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
+               if (!ic->writer_wq) {
+                       ti->error = "Cannot allocate workqueue";
+                       r = -ENOMEM;
+                       goto bad;
+               }
+               INIT_WORK(&ic->writer_work, integrity_writer);
+       }
+
+       ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
+       if (!ic->sb) {
+               r = -ENOMEM;
+               ti->error = "Cannot allocate superblock area";
+               goto bad;
+       }
+
+       r = sync_rw_sb(ic, REQ_OP_READ, 0);
+       if (r) {
+               ti->error = "Error reading superblock";
+               goto bad;
+       }
+       if (!memcmp(ic->sb->magic, SB_MAGIC, 8)) {
+               should_write_sb = false;
+       } else {
+               for (i = 0; i < 512; i += 8) {
+                       if (*(__u64 *)((__u8 *)ic->sb + i)) {
+                               r = -EINVAL;
+                               ti->error = "The device is not initialized";
+                               goto bad;
+                       }
+               }
+
+               r = initialize_superblock(ic, journal_sectors, interleave_sectors);
+               if (r) {
+                       ti->error = "Could not initialize superblock";
+                       goto bad;
+               }
+               should_write_sb = true;
+       }
+
+       if (ic->sb->version != SB_VERSION) {
+               r = -EINVAL;
+               ti->error = "Unknown version";
+               goto bad;
+       }
+       if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
+               r = -EINVAL;
+               ti->error = "Invalid tag size";
+               goto bad;
+       }
+       /* make sure that ti->max_io_len doesn't overflow */
+       if (ic->sb->log2_interleave_sectors < MIN_INTERLEAVE_SECTORS ||
+           ic->sb->log2_interleave_sectors > MAX_INTERLEAVE_SECTORS) {
+               r = -EINVAL;
+               ti->error = "Invalid interleave_sectors in the superblock";
+               goto bad;
+       }
+       ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
+       if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
+               /* test for overflow */
+               r = -EINVAL;
+               ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
+               goto bad;
+       }
+       if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
+               r = -EINVAL;
+               ti->error = "Journal mac mismatch";
+               goto bad;
+       }
+       r = calculate_device_limits(ic);
+       if (r) {
+               ti->error = "The device is too small";
+               goto bad;
+       }
+
+       if (!buffer_sectors)
+               buffer_sectors = 1;
+       ic->log2_buffer_sectors = min3((int)__fls(buffer_sectors), (int)__ffs(ic->metadata_run), 31 - SECTOR_SHIFT);
+
+       threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
+       threshold += 50;
+       do_div(threshold, 100);
+       ic->free_sectors_threshold = threshold;
+
+       DEBUG_print("initialized:\n");
+       DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
+       DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
+       DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
+       DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
+       DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
+       DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
+       DEBUG_print("   journal_entries %u\n", ic->journal_entries);
+       DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
+       DEBUG_print("   device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
+       DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
+       DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
+       DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
+       DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
+                   (unsigned long long)ic->provided_data_sectors);
+       DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
+
+       ic->bufio = dm_bufio_client_create(ic->dev->bdev, 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors),
+                                          1, 0, NULL, NULL);
+       if (IS_ERR(ic->bufio)) {
+               r = PTR_ERR(ic->bufio);
+               ti->error = "Cannot initialize dm-bufio";
+               ic->bufio = NULL;
+               goto bad;
+       }
+       dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
+
+       journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
+                               PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
+       journal_desc_size = journal_pages * sizeof(struct page_list);
+       if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
+               ti->error = "Journal doesn't fit into memory";
+               r = -ENOMEM;
+               goto bad;
+       }
+       ic->journal_pages = journal_pages;
+
+       ic->journal = dm_integrity_alloc_page_list(ic);
+       if (!ic->journal) {
+               ti->error = "Could not allocate memory for journal";
+               r = -ENOMEM;
+               goto bad;
+       }
+       if (ic->journal_crypt_alg.alg_string) {
+               unsigned ivsize, blocksize;
+               struct journal_completion comp;
+               comp.ic = ic;
+
+               ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
+               if (IS_ERR(ic->journal_crypt)) {
+                       ti->error = "Invalid journal cipher";
+                       r = PTR_ERR(ic->journal_crypt);
+                       ic->journal_crypt = NULL;
+                       goto bad;
+               }
+               ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
+               blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
+
+               if (ic->journal_crypt_alg.key) {
+                       r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
+                                                  ic->journal_crypt_alg.key_size);
+                       if (r) {
+                               ti->error = "Error setting encryption key";
+                               goto bad;
+                       }
+               }
+               DEBUG_print("cipher %s, block size %u iv size %u\n",
+                           ic->journal_crypt_alg.alg_string, blocksize, ivsize);
+
+               ic->journal_io = dm_integrity_alloc_page_list(ic);
+               if (!ic->journal_io) {
+                       ti->error = "Could not allocate memory for journal io";
+                       r = -ENOMEM;
+                       goto bad;
+               }
+
+               if (blocksize == 1) {
+                       struct scatterlist *sg;
+                       SKCIPHER_REQUEST_ON_STACK(req, ic->journal_crypt);
+                       unsigned char iv[ivsize];
+                       skcipher_request_set_tfm(req, ic->journal_crypt);
+
+                       ic->journal_xor = dm_integrity_alloc_page_list(ic);
+                       if (!ic->journal_xor) {
+                               ti->error = "Could not allocate memory for journal xor";
+                               r = -ENOMEM;
+                               goto bad;
+                       }
+
+                       sg = dm_integrity_kvmalloc((ic->journal_pages + 1) * sizeof(struct scatterlist), 0);
+                       if (!sg) {
+                               ti->error = "Unable to allocate sg list";
+                               r = -ENOMEM;
+                               goto bad;
+                       }
+                       sg_init_table(sg, ic->journal_pages + 1);
+                       for (i = 0; i < ic->journal_pages; i++) {
+                               char *va = lowmem_page_address(ic->journal_xor[i].page);
+                               clear_page(va);
+                               sg_set_buf(&sg[i], va, PAGE_SIZE);
+                       }
+                       sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
+                       memset(iv, 0x00, ivsize);
+
+                       skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, iv);
+                       comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
+                       comp.in_flight = (atomic_t)ATOMIC_INIT(1);
+                       if (do_crypt(true, req, &comp))
+                               wait_for_completion(&comp.comp);
+                       kvfree(sg);
+                       if ((r = dm_integrity_failed(ic))) {
+                               ti->error = "Unable to encrypt journal";
+                               goto bad;
+                       }
+                       DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
+
+                       crypto_free_skcipher(ic->journal_crypt);
+                       ic->journal_crypt = NULL;
+               } else {
+                       SKCIPHER_REQUEST_ON_STACK(req, ic->journal_crypt);
+                       unsigned char iv[ivsize];
+                       unsigned crypt_len = roundup(ivsize, blocksize);
+                       unsigned char crypt_data[crypt_len];
+
+                       skcipher_request_set_tfm(req, ic->journal_crypt);
+
+                       ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
+                       if (!ic->journal_scatterlist) {
+                               ti->error = "Unable to allocate sg list";
+                               r = -ENOMEM;
+                               goto bad;
+                       }
+                       ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
+                       if (!ic->journal_io_scatterlist) {
+                               ti->error = "Unable to allocate sg list";
+                               r = -ENOMEM;
+                               goto bad;
+                       }
+                       ic->sk_requests = dm_integrity_kvmalloc(ic->journal_sections * sizeof(struct skcipher_request *), __GFP_ZERO);
+                       if (!ic->sk_requests) {
+                               ti->error = "Unable to allocate sk requests";
+                               r = -ENOMEM;
+                               goto bad;
+                       }
+                       for (i = 0; i < ic->journal_sections; i++) {
+                               struct scatterlist sg;
+                               struct skcipher_request *section_req;
+                               __u32 section_le = cpu_to_le32(i);
+
+                               memset(iv, 0x00, ivsize);
+                               memset(crypt_data, 0x00, crypt_len);
+                               memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
+
+                               sg_init_one(&sg, crypt_data, crypt_len);
+                               skcipher_request_set_crypt(req, &sg, &sg, crypt_len, iv);
+                               comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
+                               comp.in_flight = (atomic_t)ATOMIC_INIT(1);
+                               if (do_crypt(true, req, &comp))
+                                       wait_for_completion(&comp.comp);
+
+                               if ((r = dm_integrity_failed(ic))) {
+                                       ti->error = "Unable to generate iv";
+                                       goto bad;
+                               }
+
+                               section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
+                               if (!section_req) {
+                                       ti->error = "Unable to allocate crypt request";
+                                       r = -ENOMEM;
+                                       goto bad;
+                               }
+                               section_req->iv = kmalloc(ivsize * 2, GFP_KERNEL);
+                               if (!section_req->iv) {
+                                       skcipher_request_free(section_req);
+                                       ti->error = "Unable to allocate iv";
+                                       r = -ENOMEM;
+                                       goto bad;
+                               }
+                               memcpy(section_req->iv + ivsize, crypt_data, ivsize);
+                               section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
+                               ic->sk_requests[i] = section_req;
+                               DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
+                       }
+               }
+       }
+
+       for (i = 0; i < N_COMMIT_IDS; i++) {
+               unsigned j;
+retest_commit_id:
+               for (j = 0; j < i; j++) {
+                       if (ic->commit_ids[j] == ic->commit_ids[i]) {
+                               ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
+                               goto retest_commit_id;
+                       }
+               }
+               DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
+       }
+
+       journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
+       if (journal_tree_size > ULONG_MAX) {
+               ti->error = "Journal doesn't fit into memory";
+               r = -ENOMEM;
+               goto bad;
+       }
+       ic->journal_tree = dm_integrity_kvmalloc(journal_tree_size, 0);
+       if (!ic->journal_tree) {
+               ti->error = "Could not allocate memory for journal tree";
+               r = -ENOMEM;
+               goto bad;
+       }
+
+       if (should_write_sb) {
+               int r;
+
+               init_journal(ic, 0, ic->journal_sections, 0);
+               r = dm_integrity_failed(ic);
+               if (unlikely(r)) {
+                       ti->error = "Error initializing journal";
+                       goto bad;
+               }
+               r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
+               if (r) {
+                       ti->error = "Error initializing superblock";
+                       goto bad;
+               }
+               ic->just_formatted = true;
+       }
+
+       r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
+       if (r)
+               goto bad;
+
+       if (!ic->internal_hash)
+               dm_integrity_set(ti, ic);
+
+       ti->num_flush_bios = 1;
+       ti->flush_supported = true;
+
+       return 0;
+bad:
+       dm_integrity_dtr(ti);
+       return r;
+}
+
+static void dm_integrity_dtr(struct dm_target *ti)
+{
+       struct dm_integrity_c *ic = ti->private;
+
+       BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
+
+       if (ic->metadata_wq)
+               destroy_workqueue(ic->metadata_wq);
+       if (ic->wait_wq)
+               destroy_workqueue(ic->wait_wq);
+       if (ic->commit_wq)
+               destroy_workqueue(ic->commit_wq);
+       if (ic->writer_wq)
+               destroy_workqueue(ic->writer_wq);
+       if (ic->bufio)
+               dm_bufio_client_destroy(ic->bufio);
+       mempool_destroy(ic->journal_io_mempool);
+       if (ic->io)
+               dm_io_client_destroy(ic->io);
+       if (ic->dev)
+               dm_put_device(ti, ic->dev);
+       dm_integrity_free_page_list(ic, ic->journal);
+       dm_integrity_free_page_list(ic, ic->journal_io);
+       dm_integrity_free_page_list(ic, ic->journal_xor);
+       if (ic->journal_scatterlist)
+               dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
+       if (ic->journal_io_scatterlist)
+               dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
+       if (ic->sk_requests) {
+               unsigned i;
+
+               for (i = 0; i < ic->journal_sections; i++) {
+                       struct skcipher_request *req = ic->sk_requests[i];
+                       if (req) {
+                               kzfree(req->iv);
+                               skcipher_request_free(req);
+                       }
+               }
+               kvfree(ic->sk_requests);
+       }
+       kvfree(ic->journal_tree);
+       if (ic->sb)
+               free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
+
+       if (ic->internal_hash)
+               crypto_free_shash(ic->internal_hash);
+       free_alg(&ic->internal_hash_alg);
+
+       if (ic->journal_crypt)
+               crypto_free_skcipher(ic->journal_crypt);
+       free_alg(&ic->journal_crypt_alg);
+
+       if (ic->journal_mac)
+               crypto_free_shash(ic->journal_mac);
+       free_alg(&ic->journal_mac_alg);
+
+       kfree(ic);
+}
+
+static struct target_type integrity_target = {
+       .name                   = "integrity",
+       .version                = {1, 0, 0},
+       .module                 = THIS_MODULE,
+       .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
+       .ctr                    = dm_integrity_ctr,
+       .dtr                    = dm_integrity_dtr,
+       .map                    = dm_integrity_map,
+       .postsuspend            = dm_integrity_postsuspend,
+       .resume                 = dm_integrity_resume,
+       .status                 = dm_integrity_status,
+       .iterate_devices        = dm_integrity_iterate_devices,
+};
+
+int __init dm_integrity_init(void)
+{
+       int r;
+
+       journal_io_cache = kmem_cache_create("integrity_journal_io",
+                                            sizeof(struct journal_io), 0, 0, NULL);
+       if (!journal_io_cache) {
+               DMERR("can't allocate journal io cache");
+               return -ENOMEM;
+       }
+
+       r = dm_register_target(&integrity_target);
+
+       if (r < 0)
+               DMERR("register failed %d", r);
+
+       return r;
+}
+
+void dm_integrity_exit(void)
+{
+       dm_unregister_target(&integrity_target);
+       kmem_cache_destroy(journal_io_cache);
+}
+
+module_init(dm_integrity_init);
+module_exit(dm_integrity_exit);
+
+MODULE_AUTHOR("Milan Broz");
+MODULE_AUTHOR("Mikulas Patocka");
+MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
+MODULE_LICENSE("GPL");