]> err.no Git - linux-2.6/blobdiff - drivers/md/dm-crypt.c
[NETFILTER]: nfnetlink: parse attributes with nfattr_parse in nfnetlink_check_attribute
[linux-2.6] / drivers / md / dm-crypt.c
index 3783cf97885016d8509e69d1ee16dd8a9fe9837f..4c2471ee054aa099a8777de111e7b219fe6c6eb9 100644 (file)
 #include <linux/slab.h>
 #include <linux/crypto.h>
 #include <linux/workqueue.h>
+#include <linux/backing-dev.h>
 #include <asm/atomic.h>
 #include <linux/scatterlist.h>
 #include <asm/page.h>
+#include <asm/unaligned.h>
 
 #include "dm.h"
 
@@ -35,6 +37,7 @@ struct crypt_io {
        struct work_struct work;
        atomic_t pending;
        int error;
+       int post_process;
 };
 
 /*
@@ -76,13 +79,17 @@ struct crypt_config {
         */
        mempool_t *io_pool;
        mempool_t *page_pool;
+       struct bio_set *bs;
 
        /*
         * crypto related data
         */
        struct crypt_iv_operations *iv_gen_ops;
        char *iv_mode;
-       struct crypto_cipher *iv_gen_private;
+       union {
+               struct crypto_cipher *essiv_tfm;
+               int benbi_shift;
+       } iv_gen_private;
        sector_t iv_offset;
        unsigned int iv_size;
 
@@ -94,11 +101,11 @@ struct crypt_config {
        u8 key[0];
 };
 
-#define MIN_IOS        256
+#define MIN_IOS        16
 #define MIN_POOL_PAGES 32
 #define MIN_BIO_PAGES  8
 
-static kmem_cache_t *_crypt_io_pool;
+static struct kmem_cache *_crypt_io_pool;
 
 /*
  * Different IV generation algorithms:
@@ -110,6 +117,9 @@ static kmem_cache_t *_crypt_io_pool;
  *        encrypted with the bulk cipher using a salt as key. The salt
  *        should be derived from the bulk cipher's key via hashing.
  *
+ * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
+ *        (needed for LRW-32-AES and possible other narrow block modes)
+ *
  * plumb: unimplemented, see:
  * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454
  */
@@ -188,21 +198,61 @@ static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti,
        }
        kfree(salt);
 
-       cc->iv_gen_private = essiv_tfm;
+       cc->iv_gen_private.essiv_tfm = essiv_tfm;
        return 0;
 }
 
 static void crypt_iv_essiv_dtr(struct crypt_config *cc)
 {
-       crypto_free_cipher(cc->iv_gen_private);
-       cc->iv_gen_private = NULL;
+       crypto_free_cipher(cc->iv_gen_private.essiv_tfm);
+       cc->iv_gen_private.essiv_tfm = NULL;
 }
 
 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
 {
        memset(iv, 0, cc->iv_size);
        *(u64 *)iv = cpu_to_le64(sector);
-       crypto_cipher_encrypt_one(cc->iv_gen_private, iv, iv);
+       crypto_cipher_encrypt_one(cc->iv_gen_private.essiv_tfm, iv, iv);
+       return 0;
+}
+
+static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
+                             const char *opts)
+{
+       unsigned int bs = crypto_blkcipher_blocksize(cc->tfm);
+       int log = ilog2(bs);
+
+       /* we need to calculate how far we must shift the sector count
+        * to get the cipher block count, we use this shift in _gen */
+
+       if (1 << log != bs) {
+               ti->error = "cypher blocksize is not a power of 2";
+               return -EINVAL;
+       }
+
+       if (log > 9) {
+               ti->error = "cypher blocksize is > 512";
+               return -EINVAL;
+       }
+
+       cc->iv_gen_private.benbi_shift = 9 - log;
+
+       return 0;
+}
+
+static void crypt_iv_benbi_dtr(struct crypt_config *cc)
+{
+}
+
+static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv, sector_t sector)
+{
+       __be64 val;
+
+       memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
+
+       val = cpu_to_be64(((u64)sector << cc->iv_gen_private.benbi_shift) + 1);
+       put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
+
        return 0;
 }
 
@@ -216,13 +266,18 @@ static struct crypt_iv_operations crypt_iv_essiv_ops = {
        .generator = crypt_iv_essiv_gen
 };
 
+static struct crypt_iv_operations crypt_iv_benbi_ops = {
+       .ctr       = crypt_iv_benbi_ctr,
+       .dtr       = crypt_iv_benbi_dtr,
+       .generator = crypt_iv_benbi_gen
+};
 
 static int
 crypt_convert_scatterlist(struct crypt_config *cc, struct scatterlist *out,
                           struct scatterlist *in, unsigned int length,
                           int write, sector_t sector)
 {
-       u8 iv[cc->iv_size];
+       u8 iv[cc->iv_size] __attribute__ ((aligned(__alignof__(u64))));
        struct blkcipher_desc desc = {
                .tfm = cc->tfm,
                .info = iv,
@@ -310,6 +365,14 @@ static int crypt_convert(struct crypt_config *cc,
        return r;
 }
 
+ static void dm_crypt_bio_destructor(struct bio *bio)
+ {
+       struct crypt_io *io = bio->bi_private;
+       struct crypt_config *cc = io->target->private;
+
+       bio_free(bio, cc->bs);
+ }
+
 /*
  * Generate a new unfragmented bio with the given size
  * This should never violate the device limitations
@@ -324,18 +387,17 @@ crypt_alloc_buffer(struct crypt_config *cc, unsigned int size,
        gfp_t gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
        unsigned int i;
 
-       /*
-        * Use __GFP_NOMEMALLOC to tell the VM to act less aggressively and
-        * to fail earlier.  This is not necessary but increases throughput.
-        * FIXME: Is this really intelligent?
-        */
-       if (base_bio)
-               clone = bio_clone(base_bio, GFP_NOIO|__GFP_NOMEMALLOC);
-       else
-               clone = bio_alloc(GFP_NOIO|__GFP_NOMEMALLOC, nr_iovecs);
+       if (base_bio) {
+               clone = bio_alloc_bioset(GFP_NOIO, base_bio->bi_max_vecs, cc->bs);
+               __bio_clone(clone, base_bio);
+       } else
+               clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, cc->bs);
+
        if (!clone)
                return NULL;
 
+       clone->bi_destructor = dm_crypt_bio_destructor;
+
        /* if the last bio was not complete, continue where that one ended */
        clone->bi_idx = *bio_vec_idx;
        clone->bi_vcnt = *bio_vec_idx;
@@ -445,15 +507,14 @@ static void dec_pending(struct crypt_io *io, int error)
  * kcryptd:
  *
  * Needed because it would be very unwise to do decryption in an
- * interrupt context, so bios returning from read requests get
- * queued here.
+ * interrupt context.
  */
 static struct workqueue_struct *_kcryptd_workqueue;
-static void kcryptd_do_work(void *data);
+static void kcryptd_do_work(struct work_struct *work);
 
 static void kcryptd_queue_io(struct crypt_io *io)
 {
-       INIT_WORK(&io->work, kcryptd_do_work, io);
+       INIT_WORK(&io->work, kcryptd_do_work);
        queue_work(_kcryptd_workqueue, &io->work);
 }
 
@@ -470,12 +531,10 @@ static int crypt_endio(struct bio *clone, unsigned int done, int error)
        if (!read_io)
                crypt_free_buffer_pages(cc, clone, done);
 
+       /* keep going - not finished yet */
        if (unlikely(clone->bi_size))
                return 1;
 
-       /*
-        * successful reads are decrypted by the worker thread
-        */
        if (!read_io)
                goto out;
 
@@ -485,6 +544,7 @@ static int crypt_endio(struct bio *clone, unsigned int done, int error)
        }
 
        bio_put(clone);
+       io->post_process = 1;
        kcryptd_queue_io(io);
        return 0;
 
@@ -504,60 +564,99 @@ static void clone_init(struct crypt_io *io, struct bio *clone)
        clone->bi_rw      = io->base_bio->bi_rw;
 }
 
-static struct bio *clone_read(struct crypt_io *io,
-                             sector_t sector)
+static void process_read(struct crypt_io *io)
 {
        struct crypt_config *cc = io->target->private;
        struct bio *base_bio = io->base_bio;
        struct bio *clone;
+       sector_t sector = base_bio->bi_sector - io->target->begin;
+
+       atomic_inc(&io->pending);
 
        /*
         * The block layer might modify the bvec array, so always
         * copy the required bvecs because we need the original
         * one in order to decrypt the whole bio data *afterwards*.
         */
-       clone = bio_alloc(GFP_NOIO, bio_segments(base_bio));
-       if (unlikely(!clone))
-               return NULL;
+       clone = bio_alloc_bioset(GFP_NOIO, bio_segments(base_bio), cc->bs);
+       if (unlikely(!clone)) {
+               dec_pending(io, -ENOMEM);
+               return;
+       }
 
        clone_init(io, clone);
+       clone->bi_destructor = dm_crypt_bio_destructor;
        clone->bi_idx = 0;
        clone->bi_vcnt = bio_segments(base_bio);
        clone->bi_size = base_bio->bi_size;
+       clone->bi_sector = cc->start + sector;
        memcpy(clone->bi_io_vec, bio_iovec(base_bio),
               sizeof(struct bio_vec) * clone->bi_vcnt);
-       clone->bi_sector = cc->start + sector;
 
-       return clone;
+       generic_make_request(clone);
 }
 
-static struct bio *clone_write(struct crypt_io *io,
-                              sector_t sector,
-                              unsigned *bvec_idx,
-                              struct convert_context *ctx)
+static void process_write(struct crypt_io *io)
 {
        struct crypt_config *cc = io->target->private;
        struct bio *base_bio = io->base_bio;
        struct bio *clone;
+       struct convert_context ctx;
+       unsigned remaining = base_bio->bi_size;
+       sector_t sector = base_bio->bi_sector - io->target->begin;
+       unsigned bvec_idx = 0;
 
-       clone = crypt_alloc_buffer(cc, base_bio->bi_size,
-                                  io->first_clone, bvec_idx);
-       if (!clone)
-               return NULL;
+       atomic_inc(&io->pending);
 
-       ctx->bio_out = clone;
+       crypt_convert_init(cc, &ctx, NULL, base_bio, sector, 1);
 
-       if (unlikely(crypt_convert(cc, ctx) < 0)) {
-               crypt_free_buffer_pages(cc, clone,
-                                       clone->bi_size);
-               bio_put(clone);
-               return NULL;
-       }
+       /*
+        * The allocated buffers can be smaller than the whole bio,
+        * so repeat the whole process until all the data can be handled.
+        */
+       while (remaining) {
+               clone = crypt_alloc_buffer(cc, base_bio->bi_size,
+                                          io->first_clone, &bvec_idx);
+               if (unlikely(!clone)) {
+                       dec_pending(io, -ENOMEM);
+                       return;
+               }
 
-       clone_init(io, clone);
-       clone->bi_sector = cc->start + sector;
+               ctx.bio_out = clone;
 
-       return clone;
+               if (unlikely(crypt_convert(cc, &ctx) < 0)) {
+                       crypt_free_buffer_pages(cc, clone, clone->bi_size);
+                       bio_put(clone);
+                       dec_pending(io, -EIO);
+                       return;
+               }
+
+               clone_init(io, clone);
+               clone->bi_sector = cc->start + sector;
+
+               if (!io->first_clone) {
+                       /*
+                        * hold a reference to the first clone, because it
+                        * holds the bio_vec array and that can't be freed
+                        * before all other clones are released
+                        */
+                       bio_get(clone);
+                       io->first_clone = clone;
+               }
+
+               remaining -= clone->bi_size;
+               sector += bio_sectors(clone);
+
+               /* prevent bio_put of first_clone */
+               if (remaining)
+                       atomic_inc(&io->pending);
+
+               generic_make_request(clone);
+
+               /* out of memory -> run queues */
+               if (remaining)
+                       congestion_wait(bio_data_dir(clone), HZ/100);
+       }
 }
 
 static void process_read_endio(struct crypt_io *io)
@@ -571,11 +670,16 @@ static void process_read_endio(struct crypt_io *io)
        dec_pending(io, crypt_convert(cc, &ctx));
 }
 
-static void kcryptd_do_work(void *data)
+static void kcryptd_do_work(struct work_struct *work)
 {
-       struct crypt_io *io = data;
+       struct crypt_io *io = container_of(work, struct crypt_io, work);
 
-       process_read_endio(io);
+       if (io->post_process)
+               process_read_endio(io);
+       else if (bio_data_dir(io->base_bio) == READ)
+               process_read(io);
+       else
+               process_write(io);
 }
 
 /*
@@ -716,7 +820,7 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
        cc->tfm = tfm;
 
        /*
-        * Choose ivmode. Valid modes: "plain", "essiv:<esshash>".
+        * Choose ivmode. Valid modes: "plain", "essiv:<esshash>", "benbi".
         * See comments at iv code
         */
 
@@ -726,6 +830,8 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
                cc->iv_gen_ops = &crypt_iv_plain_ops;
        else if (strcmp(ivmode, "essiv") == 0)
                cc->iv_gen_ops = &crypt_iv_essiv_ops;
+       else if (strcmp(ivmode, "benbi") == 0)
+               cc->iv_gen_ops = &crypt_iv_benbi_ops;
        else {
                ti->error = "Invalid IV mode";
                goto bad2;
@@ -761,6 +867,12 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
                goto bad4;
        }
 
+       cc->bs = bioset_create(MIN_IOS, MIN_IOS, 4);
+       if (!cc->bs) {
+               ti->error = "Cannot allocate crypt bioset";
+               goto bad_bs;
+       }
+
        if (crypto_blkcipher_setkey(tfm, cc->key, key_size) < 0) {
                ti->error = "Error setting key";
                goto bad5;
@@ -800,6 +912,8 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
        return 0;
 
 bad5:
+       bioset_free(cc->bs);
+bad_bs:
        mempool_destroy(cc->page_pool);
 bad4:
        mempool_destroy(cc->io_pool);
@@ -819,6 +933,7 @@ static void crypt_dtr(struct dm_target *ti)
 {
        struct crypt_config *cc = (struct crypt_config *) ti->private;
 
+       bioset_free(cc->bs);
        mempool_destroy(cc->page_pool);
        mempool_destroy(cc->io_pool);
 
@@ -838,76 +953,22 @@ static int crypt_map(struct dm_target *ti, struct bio *bio,
 {
        struct crypt_config *cc = ti->private;
        struct crypt_io *io;
-       struct convert_context ctx;
-       struct bio *clone;
-       unsigned int remaining = bio->bi_size;
-       sector_t sector = bio->bi_sector - ti->begin;
-       unsigned int bvec_idx = 0;
 
        io = mempool_alloc(cc->io_pool, GFP_NOIO);
        io->target = ti;
        io->base_bio = bio;
        io->first_clone = NULL;
-       io->error = 0;
-       atomic_set(&io->pending, 1); /* hold a reference */
-
-       if (bio_data_dir(bio) == WRITE)
-               crypt_convert_init(cc, &ctx, NULL, bio, sector, 1);
-
-       /*
-        * The allocated buffers can be smaller than the whole bio,
-        * so repeat the whole process until all the data can be handled.
-        */
-       while (remaining) {
-               if (bio_data_dir(bio) == WRITE)
-                       clone = clone_write(io, sector, &bvec_idx, &ctx);
-               else
-                       clone = clone_read(io, sector);
-               if (!clone)
-                       goto cleanup;
-
-               if (!io->first_clone) {
-                       /*
-                        * hold a reference to the first clone, because it
-                        * holds the bio_vec array and that can't be freed
-                        * before all other clones are released
-                        */
-                       bio_get(clone);
-                       io->first_clone = clone;
-               }
-               atomic_inc(&io->pending);
-
-               remaining -= clone->bi_size;
-               sector += bio_sectors(clone);
-
-               generic_make_request(clone);
-
-               /* out of memory -> run queues */
-               if (remaining)
-                       blk_congestion_wait(bio_data_dir(clone), HZ/100);
-       }
-
-       /* drop reference, clones could have returned before we reach this */
-       dec_pending(io, 0);
-       return 0;
-
-cleanup:
-       if (io->first_clone) {
-               dec_pending(io, -ENOMEM);
-               return 0;
-       }
+       io->error = io->post_process = 0;
+       atomic_set(&io->pending, 0);
+       kcryptd_queue_io(io);
 
-       /* if no bio has been dispatched yet, we can directly return the error */
-       mempool_free(io, cc->io_pool);
-       return -ENOMEM;
+       return DM_MAPIO_SUBMITTED;
 }
 
 static int crypt_status(struct dm_target *ti, status_type_t type,
                        char *result, unsigned int maxlen)
 {
        struct crypt_config *cc = (struct crypt_config *) ti->private;
-       const char *cipher;
-       const char *chainmode = NULL;
        unsigned int sz = 0;
 
        switch (type) {
@@ -916,14 +977,11 @@ static int crypt_status(struct dm_target *ti, status_type_t type,
                break;
 
        case STATUSTYPE_TABLE:
-               cipher = crypto_blkcipher_name(cc->tfm);
-
-               chainmode = cc->chainmode;
-
                if (cc->iv_mode)
-                       DMEMIT("%s-%s-%s ", cipher, chainmode, cc->iv_mode);
+                       DMEMIT("%s-%s-%s ", cc->cipher, cc->chainmode,
+                              cc->iv_mode);
                else
-                       DMEMIT("%s-%s ", cipher, chainmode);
+                       DMEMIT("%s-%s ", cc->cipher, cc->chainmode);
 
                if (cc->key_size > 0) {
                        if ((maxlen - sz) < ((cc->key_size << 1) + 1))
@@ -999,7 +1057,7 @@ error:
 
 static struct target_type crypt_target = {
        .name   = "crypt",
-       .version= {1, 2, 0},
+       .version= {1, 3, 0},
        .module = THIS_MODULE,
        .ctr    = crypt_ctr,
        .dtr    = crypt_dtr,