]> git.baikalelectronics.ru Git - kernel.git/commitdiff
dm integrity: support larger block sizes
authorMikulas Patocka <mpatocka@redhat.com>
Tue, 18 Apr 2017 20:51:52 +0000 (16:51 -0400)
committerMike Snitzer <snitzer@redhat.com>
Mon, 24 Apr 2017 16:04:33 +0000 (12:04 -0400)
The DM integrity block size can now be 512, 1k, 2k or 4k.  Using larger
blocks reduces metadata handling overhead.  The block size can be
configured at table load time using the "block_size:<value>" option;
where <value> is expressed in bytes (defult is still 512 bytes).

It is safe to use larger block sizes with DM integrity, because the
DM integrity journal makes sure that the whole block is updated
atomically even if the underlying device doesn't support atomic writes
of that size (e.g. 4k block ontop of a 512b device).

Depends-on: 2859323e ("block: fix blk_integrity_register to use template's interval_exp if not 0")
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Documentation/device-mapper/dm-integrity.txt
drivers/md/dm-integrity.c

index ced34cd915efb7a01b5a4110affb0b0236ca595a..f33e3ade7a09bdaf9f6b338d5142acc512a94cb2 100644 (file)
@@ -136,6 +136,11 @@ journal_mac:algorithm(:key)        (the key is optional)
        the journal. Thus, modified sector number would be detected at
        this stage.
 
+block_size:number
+       The size of a data block in bytes.  The larger the block size the
+       less overhead there is for per-block integrity metadata.
+       Supported values are 512, 1024, 2048 and 4096 bytes.  If not
+       specified the default block size is 512 bytes.
 
 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
index 95cdffbb206ccd8266712b7d472ecdc1161592f5..0354af4cd7139d1d7c07bdd2966c83838b2609b1 100644 (file)
@@ -14,6 +14,7 @@
 #include <linux/rbtree.h>
 #include <linux/delay.h>
 #include <linux/random.h>
+#include <linux/log2.h>
 #include <crypto/hash.h>
 #include <crypto/skcipher.h>
 #include <linux/async_tx.h>
@@ -45,6 +46,7 @@
 #define SB_MAGIC                       "integrt"
 #define SB_VERSION                     1
 #define SB_SECTORS                     8
+#define MAX_SECTORS_PER_BLOCK          8
 
 struct superblock {
        __u8 magic[8];
@@ -54,6 +56,7 @@ struct superblock {
        __u32 journal_sections;
        __u64 provided_data_sectors;    /* userspace uses this value */
        __u32 flags;
+       __u8 log2_sectors_per_block;
 };
 
 #define SB_FLAG_HAVE_JOURNAL_MAC       0x1
@@ -71,10 +74,12 @@ struct journal_entry {
                } s;
                __u64 sector;
        } u;
-       commit_id_t last_bytes;
-       __u8 tag[0];
+       commit_id_t last_bytes[0];
+       /* __u8 tag[0]; */
 };
 
+#define journal_entry_tag(ic, je)              ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
+
 #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)
@@ -100,7 +105,7 @@ struct journal_sector {
        commit_id_t commit_id;
 };
 
-#define MAX_TAG_SIZE                   (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, tag))
+#define MAX_TAG_SIZE                   (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 
 #define METADATA_PADDING_SECTORS       8
 
@@ -162,7 +167,7 @@ struct dm_integrity_c {
        unsigned short journal_entry_size;
        unsigned char journal_entries_per_sector;
        unsigned char journal_section_entries;
-       unsigned char journal_section_sectors;
+       unsigned short journal_section_sectors;
        unsigned journal_sections;
        unsigned journal_entries;
        sector_t device_sectors;
@@ -170,6 +175,7 @@ struct dm_integrity_c {
        unsigned metadata_run;
        __s8 log2_metadata_run;
        __u8 log2_buffer_sectors;
+       __u8 sectors_per_block;
 
        unsigned char mode;
        bool suspending;
@@ -332,6 +338,12 @@ static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
        *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 }
 
+#define sector_to_block(ic, n)                                         \
+do {                                                                   \
+       BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
+       (n) >>= (ic)->sb->log2_sectors_per_block;                       \
+} while (0)
+
 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
                                            sector_t offset, unsigned *metadata_offset)
 {
@@ -345,6 +357,8 @@ static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t
                ms += area * ic->metadata_run;
        ms >>= ic->log2_buffer_sectors;
 
+       sector_to_block(ic, offset);
+
        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);
@@ -459,9 +473,13 @@ static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, uns
 
 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");
+       n <<= ic->sb->log2_sectors_per_block;
 
-       return access_journal(ic, section, n + JOURNAL_BLOCK_SECTORS);
+       n += JOURNAL_BLOCK_SECTORS;
+
+       access_journal_check(ic, section, n, false, "access_journal_data");
+
+       return access_journal(ic, section, n);
 }
 
 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
@@ -812,6 +830,8 @@ static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsig
        int r;
        unsigned sector, pl_index, pl_offset;
 
+       BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
+
        if (unlikely(dm_integrity_failed(ic))) {
                fn(-1UL, data);
                return;
@@ -846,6 +866,8 @@ static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *
        struct rb_node **n = &ic->in_progress.rb_node;
        struct rb_node *parent;
 
+       BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
+
        parent = NULL;
 
        while (*n) {
@@ -1175,7 +1197,7 @@ static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector
                goto failed;
        }
 
-       r = crypto_shash_update(req, data, 1 << SECTOR_SHIFT);
+       r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
        if (unlikely(r < 0)) {
                dm_integrity_io_error(ic, "crypto_shash_update", r);
                goto failed;
@@ -1219,7 +1241,7 @@ static void integrity_metadata(struct work_struct *w)
                if (unlikely(ic->mode == 'R'))
                        goto skip_io;
 
-               checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT) * ic->tag_size + extra_space,
+               checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
                                    GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
                if (!checksums)
                        checksums = checksums_onstack;
@@ -1235,9 +1257,9 @@ again:
                        do {
                                integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
                                checksums_ptr += ic->tag_size;
-                               sectors_to_process--;
-                               pos += 1 << SECTOR_SHIFT;
-                               sector++;
+                               sectors_to_process -= ic->sectors_per_block;
+                               pos += ic->sectors_per_block << SECTOR_SHIFT;
+                               sector += ic->sectors_per_block;
                        } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
                        kunmap_atomic(mem);
 
@@ -1272,7 +1294,9 @@ again:
                if (bip) {
                        struct bio_vec biv;
                        struct bvec_iter iter;
-                       unsigned data_to_process = dio->range.n_sectors * ic->tag_size;
+                       unsigned data_to_process = dio->range.n_sectors;
+                       sector_to_block(ic, data_to_process);
+                       data_to_process *= ic->tag_size;
 
                        bip_for_each_vec(biv, bip, iter) {
                                unsigned char *tag;
@@ -1303,6 +1327,7 @@ 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));
+       struct bio_integrity_payload *bip;
 
        sector_t area, offset;
 
@@ -1330,6 +1355,44 @@ static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
                      (unsigned long long)ic->provided_data_sectors);
                return -EIO;
        }
+       if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
+               DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
+                     ic->sectors_per_block,
+                     (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
+               return -EIO;
+       }
+
+       if (ic->sectors_per_block > 1) {
+               struct bvec_iter iter;
+               struct bio_vec bv;
+               bio_for_each_segment(bv, bio, iter) {
+                       if (unlikely((bv.bv_offset | bv.bv_len) & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
+                               DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
+                                       bv.bv_offset, bv.bv_len, ic->sectors_per_block);
+                               return -EIO;
+                       }
+               }
+       }
+
+       bip = bio_integrity(bio);
+       if (!ic->internal_hash) {
+               if (bip) {
+                       unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
+                       if (ic->log2_tag_size >= 0)
+                               wanted_tag_size <<= ic->log2_tag_size;
+                       else
+                               wanted_tag_size *= ic->tag_size;
+                       if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
+                               DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
+                               return -EIO;
+                       }
+               }
+       } else {
+               if (unlikely(bip != NULL)) {
+                       DMERR("Unexpected integrity data when using internal hash");
+                       return -EIO;
+               }
+       }
 
        if (unlikely(ic->mode == 'R') && unlikely(dio->write))
                return -EIO;
@@ -1369,6 +1432,8 @@ retry_kmap:
 
                        if (unlikely(!dio->write)) {
                                struct journal_sector *js;
+                               char *mem_ptr;
+                               unsigned s;
 
                                if (unlikely(journal_entry_is_inprogress(je))) {
                                        flush_dcache_page(bv.bv_page);
@@ -1380,14 +1445,20 @@ 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);
+                               mem_ptr = mem + bv.bv_offset;
+                               s = 0;
+                               do {
+                                       memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
+                                       *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
+                                       js++;
+                                       mem_ptr += 1 << SECTOR_SHIFT;
+                               } while (++s < ic->sectors_per_block);
 #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))) {
+                                       if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
                                                DMERR("Checksum failed when reading from journal, at sector 0x%llx",
                                                      (unsigned long long)logical_sector);
                                        }
@@ -1398,7 +1469,7 @@ retry_kmap:
                        if (!ic->internal_hash) {
                                struct bio_integrity_payload *bip = bio_integrity(bio);
                                unsigned tag_todo = ic->tag_size;
-                               char *tag_ptr = je->tag;
+                               char *tag_ptr = journal_entry_tag(ic, je);
 
                                if (bip) do {
                                        struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
@@ -1421,24 +1492,29 @@ retry_kmap:
 
                        if (likely(dio->write)) {
                                struct journal_sector *js;
+                               unsigned s;
 
                                js = access_journal_data(ic, journal_section, journal_entry);
-                               memcpy(js, mem + bv.bv_offset, 1 << SECTOR_SHIFT);
-                               je->last_bytes = js->commit_id;
+                               memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
+
+                               s = 0;
+                               do {
+                                       je->last_bytes[s] = js[s].commit_id;
+                               } while (++s < ic->sectors_per_block);
 
                                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);
+                                               memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
                                        } else
-                                               integrity_sector_checksum(ic, logical_sector, (char *)js, je->tag);
+                                               integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
                                }
 
                                journal_entry_set_sector(je, logical_sector);
                        }
-                       logical_sector++;
+                       logical_sector += ic->sectors_per_block;
 
                        journal_entry++;
                        if (unlikely(journal_entry == ic->journal_section_entries)) {
@@ -1447,8 +1523,8 @@ retry_kmap:
                                wraparound_section(ic, &journal_section);
                        }
 
-                       bv.bv_offset += 1 << SECTOR_SHIFT;
-               } while (bv.bv_len -= 1 << SECTOR_SHIFT);
+                       bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
+               } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
 
                if (unlikely(!dio->write))
                        flush_dcache_page(bv.bv_page);
@@ -1526,7 +1602,8 @@ retry:
                        pos = journal_section * ic->journal_section_entries + journal_entry;
                        ws = journal_section;
                        we = journal_entry;
-                       for (i = 0; i < dio->range.n_sectors; i++) {
+                       i = 0;
+                       do {
                                struct journal_entry *je;
 
                                add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
@@ -1543,7 +1620,7 @@ retry:
                                        ws++;
                                        wraparound_section(ic, &ws);
                                }
-                       }
+                       } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
 
                        spin_unlock_irq(&ic->endio_wait.lock);
                        goto journal_read_write;
@@ -1555,8 +1632,9 @@ retry:
                                        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))
+                               unsigned jp = journal_read_pos + 1;
+                               for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
+                                       if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
                                                break;
                                }
                                dio->range.n_sectors = i;
@@ -1725,6 +1803,16 @@ static void complete_copy_from_journal(unsigned long error, void *context)
        complete_journal_op(comp);
 }
 
+static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
+                              struct journal_entry *je)
+{
+       unsigned s = 0;
+       do {
+               js->commit_id = je->last_bytes[s];
+               js++;
+       } while (++s < ic->sectors_per_block);
+}
+
 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
                             unsigned write_sections, bool from_replay)
 {
@@ -1753,8 +1841,14 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
                                continue;
                        BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
                        sec = journal_entry_get_sector(je);
+                       if (unlikely(from_replay)) {
+                               if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
+                                       dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
+                                       sec &= ~(sector_t)(ic->sectors_per_block - 1);
+                               }
+                       }
                        get_area_and_offset(ic, sec, &area, &offset);
-                       access_journal_data(ic, i, j)->commit_id = je->last_bytes;
+                       restore_last_bytes(ic, access_journal_data(ic, i, j), je);
                        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;
@@ -1763,16 +1857,16 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
                                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))
+                               if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
                                        break;
-                               access_journal_data(ic, i, k)->commit_id = je2->last_bytes;
+                               restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
                        }
                        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;
+                       io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
 
                        spin_lock_irq(&ic->endio_wait.lock);
                        while (unlikely(!add_new_range(ic, &io->range)))
@@ -1788,8 +1882,8 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
                                        journal_entry_set_unused(je2);
                                        remove_journal_node(ic, &section_node[j]);
                                        j++;
-                                       sec++;
-                                       offset++;
+                                       sec += ic->sectors_per_block;
+                                       offset += ic->sectors_per_block;
                                }
                                while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
                                        struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
@@ -1822,14 +1916,14 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
                                    ic->internal_hash) {
                                        char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
 
-                                       integrity_sector_checksum(ic, sec + (l - j),
+                                       integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
                                                                  (char *)access_journal_data(ic, i, l), test_tag);
-                                       if (unlikely(memcmp(test_tag, je2->tag, ic->tag_size)))
+                                       if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), 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,
+                               r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
                                                        ic->tag_size, TAG_WRITE);
                                if (unlikely(r)) {
                                        dm_integrity_io_error(ic, "reading tags", r);
@@ -1837,7 +1931,9 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
                        }
 
                        atomic_inc(&comp.in_flight);
-                       copy_from_journal(ic, i, j, k - j, get_data_sector(ic, area, offset),
+                       copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
+                                         (k - j) << ic->sb->log2_sectors_per_block,
+                                         get_data_sector(ic, area, offset),
                                          complete_copy_from_journal, io);
 skip_io:
                        j = next_loop;
@@ -2130,6 +2226,7 @@ static void dm_integrity_status(struct dm_target *ti, status_type_t type,
                watermark_percentage += ic->journal_entries / 2;
                do_div(watermark_percentage, ic->journal_entries);
                arg_count = 5;
+               arg_count += ic->sectors_per_block != 1;
                arg_count += !!ic->internal_hash_alg.alg_string;
                arg_count += !!ic->journal_crypt_alg.alg_string;
                arg_count += !!ic->journal_mac_alg.alg_string;
@@ -2140,6 +2237,8 @@ static void dm_integrity_status(struct dm_target *ti, status_type_t type,
                DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
                DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
                DMEMIT(" commit_time:%u", ic->autocommit_msec);
+               if (ic->sectors_per_block != 1)
+                       DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
 
 #define EMIT_ALG(a, n)                                                 \
                do {                                                    \
@@ -2165,19 +2264,30 @@ static int dm_integrity_iterate_devices(struct dm_target *ti,
        return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
 }
 
+static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
+{
+       struct dm_integrity_c *ic = ti->private;
+
+       if (ic->sectors_per_block > 1) {
+               limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
+               limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
+               blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
+       }
+}
+
 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,
+       ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + 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_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
        ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
 }
 
@@ -2192,7 +2302,7 @@ static int calculate_device_limits(struct dm_integrity_c *ic)
                return -EINVAL;
        ic->initial_sectors = initial_sectors;
 
-       ic->metadata_run = roundup((__u64)ic->tag_size << ic->sb->log2_interleave_sectors,
+       ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
                                   (__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);
@@ -2217,6 +2327,7 @@ static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sec
        memcpy(ic->sb->magic, SB_MAGIC, 8);
        ic->sb->version = SB_VERSION;
        ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
+       ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
        if (ic->journal_mac_alg.alg_string)
                ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
 
@@ -2256,8 +2367,9 @@ static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
 
        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;
+       bi.tuple_size = ic->tag_size;
+       bi.tag_size = bi.tuple_size;
+       bi.interval_exp = ilog2(ic->sectors_per_block << SECTOR_SHIFT);
 
        blk_integrity_register(disk, &bi);
        blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
@@ -2667,6 +2779,7 @@ bad:
  *             internal_hash
  *             journal_crypt
  *             journal_mac
+ *             block_size
  */
 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
 {
@@ -2676,7 +2789,7 @@ static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
        unsigned extra_args;
        struct dm_arg_set as;
        static struct dm_arg _args[] = {
-               {0, 8, "Invalid number of feature args"},
+               {0, 9, "Invalid number of feature args"},
        };
        unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
        bool should_write_sb;
@@ -2740,6 +2853,7 @@ static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
        buffer_sectors = DEFAULT_BUFFER_SECTORS;
        journal_watermark = DEFAULT_JOURNAL_WATERMARK;
        sync_msec = DEFAULT_SYNC_MSEC;
+       ic->sectors_per_block = 1;
 
        as.argc = argc - DIRECT_ARGUMENTS;
        as.argv = argv + DIRECT_ARGUMENTS;
@@ -2766,7 +2880,16 @@ static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
                        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:"))) {
+               else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
+                       if (val < 1 << SECTOR_SHIFT ||
+                           val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
+                           (val & (val -1))) {
+                               r = -EINVAL;
+                               ti->error = "Invalid block_size argument";
+                               goto bad;
+                       }
+                       ic->sectors_per_block = val >> SECTOR_SHIFT;
+               } 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)
@@ -2910,7 +3033,12 @@ static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
        }
        if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
                r = -EINVAL;
-               ti->error = "Invalid tag size";
+               ti->error = "Tag size doesn't match the information in superblock";
+               goto bad;
+       }
+       if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
+               r = -EINVAL;
+               ti->error = "Block size doesn't match the information in superblock";
                goto bad;
        }
        /* make sure that ti->max_io_len doesn't overflow */
@@ -3084,6 +3212,7 @@ static struct target_type integrity_target = {
        .resume                 = dm_integrity_resume,
        .status                 = dm_integrity_status,
        .iterate_devices        = dm_integrity_iterate_devices,
+       .io_hints               = dm_integrity_io_hints,
 };
 
 int __init dm_integrity_init(void)