From patchwork Fri Nov 12 18:51:50 2010 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Milan Broz X-Patchwork-Id: 320702 Received: from mx4-phx2.redhat.com (mx4-phx2.redhat.com [209.132.183.25]) by demeter1.kernel.org (8.14.4/8.14.3) with ESMTP id oACIs48L027882 for ; Fri, 12 Nov 2010 18:54:27 GMT Received: from lists01.pubmisc.prod.ext.phx2.redhat.com (lists01.pubmisc.prod.ext.phx2.redhat.com [10.5.19.33]) by mx4-phx2.redhat.com (8.13.8/8.13.8) with ESMTP id oACIq1EJ019262; Fri, 12 Nov 2010 13:52:02 -0500 Received: from int-mx09.intmail.prod.int.phx2.redhat.com (int-mx09.intmail.prod.int.phx2.redhat.com [10.5.11.22]) by lists01.pubmisc.prod.ext.phx2.redhat.com (8.13.8/8.13.8) with ESMTP id oACIpxN9009541 for ; Fri, 12 Nov 2010 13:51:59 -0500 Received: from [10.34.26.53] (tawny.brq.redhat.com [10.34.26.53]) by int-mx09.intmail.prod.int.phx2.redhat.com (8.14.4/8.14.4) with ESMTP id oACIpokj028569; Fri, 12 Nov 2010 13:51:51 -0500 Message-ID: <4CDD8CC6.3070108@redhat.com> Date: Fri, 12 Nov 2010 19:51:50 +0100 From: Milan Broz User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.9) Gecko/20100918 Thunderbird/3.1.4 MIME-Version: 1.0 To: device-mapper development X-Enigmail-Version: 1.1.2 X-Scanned-By: MIMEDefang 2.68 on 10.5.11.22 X-loop: dm-devel@redhat.com Cc: Mike Snitzer , Andi Kleen , Linux Kernel Mailing List , Alasdair G Kergon Subject: [dm-devel] [PATCH v5] dm crypt: scale to multiple CPUs X-BeenThere: dm-devel@redhat.com X-Mailman-Version: 2.1.12 Precedence: junk Reply-To: device-mapper development List-Id: device-mapper development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: dm-devel-bounces@redhat.com Errors-To: dm-devel-bounces@redhat.com X-Greylist: IP, sender and recipient auto-whitelisted, not delayed by milter-greylist-4.2.3 (demeter1.kernel.org [140.211.167.41]); Fri, 12 Nov 2010 18:54:27 +0000 (UTC) diff --git a/drivers/md/dm-crypt.c b/drivers/md/dm-crypt.c index d5b0e4c..b4dac68 100644 --- a/drivers/md/dm-crypt.c +++ b/drivers/md/dm-crypt.c @@ -18,6 +18,7 @@ #include #include #include +#include #include #include #include @@ -77,7 +78,6 @@ struct crypt_iv_operations { }; struct iv_essiv_private { - struct crypto_cipher *tfm; struct crypto_hash *hash_tfm; u8 *salt; }; @@ -91,6 +91,20 @@ struct iv_benbi_private { * and encrypts / decrypts at the same time. */ enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID }; + +/* Duplicated per CPU state for cipher */ +struct crypt_cpu { + struct ablkcipher_request *req; + struct crypto_ablkcipher *tfm; + + /* ESSIV: struct crypto_cipher *essiv_tfm */ + void *iv_private; +}; + +/* + * The fields in here must be read only after initialization, + * changing state should be in crypt_cpu. + */ struct crypt_config { struct dm_dev *dev; sector_t start; @@ -119,6 +133,12 @@ struct crypt_config { unsigned int iv_size; /* + * Duplicated per cpu state. Access through + * per_cpu_ptr() only. + */ + struct crypt_cpu __percpu *cpu; + + /* * Layout of each crypto request: * * struct ablkcipher_request @@ -132,9 +152,7 @@ struct crypt_config { * correctly aligned. */ unsigned int dmreq_start; - struct ablkcipher_request *req; - struct crypto_ablkcipher *tfm; unsigned long flags; unsigned int key_size; u8 key[0]; @@ -149,6 +167,20 @@ static struct kmem_cache *_crypt_io_pool; static void clone_init(struct dm_crypt_io *, struct bio *); static void kcryptd_queue_crypt(struct dm_crypt_io *io); +static struct crypt_cpu *crypt_me(struct crypt_config *cc) +{ + return this_cpu_ptr(cc->cpu); +} + +/* Use this for cipher attributes that are the same for all cpus */ +static struct crypto_ablkcipher *any_tfm(struct crypt_config *cc) +{ + struct crypto_ablkcipher *tfm; + /* cpu doesn't matter, output is always the same */ + tfm = __this_cpu_ptr(cc->cpu)->tfm; + return tfm; +} + /* * Different IV generation algorithms: * @@ -195,7 +227,9 @@ static int crypt_iv_essiv_init(struct crypt_config *cc) struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv; struct hash_desc desc; struct scatterlist sg; - int err; + struct crypt_cpu *cs; + struct crypto_cipher *essiv_tfm; + int err, n, cpu; sg_init_one(&sg, cc->key, cc->key_size); desc.tfm = essiv->hash_tfm; @@ -205,8 +239,19 @@ static int crypt_iv_essiv_init(struct crypt_config *cc) if (err) return err; - return crypto_cipher_setkey(essiv->tfm, essiv->salt, + for_each_possible_cpu (cpu) { + cs = per_cpu_ptr(cc->cpu, cpu); + essiv_tfm = cs->iv_private, + + n = crypto_cipher_setkey(essiv_tfm, essiv->salt, crypto_hash_digestsize(essiv->hash_tfm)); + if (n) { + err = n; + break; + } + } + + return err; } /* Wipe salt and reset key derived from volume key */ @@ -214,24 +259,75 @@ static int crypt_iv_essiv_wipe(struct crypt_config *cc) { struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv; unsigned salt_size = crypto_hash_digestsize(essiv->hash_tfm); + struct crypt_cpu *cs; + struct crypto_cipher *essiv_tfm; + int cpu, err, n; memset(essiv->salt, 0, salt_size); - return crypto_cipher_setkey(essiv->tfm, essiv->salt, salt_size); + err = 0; + for_each_possible_cpu (cpu) { + cs = per_cpu_ptr(cc->cpu, cpu); + essiv_tfm = cs->iv_private; + n = crypto_cipher_setkey(essiv_tfm, essiv->salt, salt_size); + if (n) + err = n; + } + return err; +} + +/* Set up per cpu cipher state */ +static struct crypto_cipher *setup_essiv_cpu(struct crypt_config *cc, + struct dm_target *ti, + u8 *salt, unsigned saltsize) +{ + struct crypto_cipher *essiv_tfm; + int err; + + /* Setup the essiv_tfm with the given salt */ + essiv_tfm = crypto_alloc_cipher(cc->cipher, 0, CRYPTO_ALG_ASYNC); + if (IS_ERR(essiv_tfm)) { + ti->error = "Error allocating crypto tfm for ESSIV"; + return essiv_tfm; + } + + if (crypto_cipher_blocksize(essiv_tfm) != + crypto_ablkcipher_ivsize(any_tfm(cc))) { + ti->error = "Block size of ESSIV cipher does " + "not match IV size of block cipher"; + crypto_free_cipher(essiv_tfm); + return ERR_PTR(-EINVAL); + } + err = crypto_cipher_setkey(essiv_tfm, salt, saltsize); + if (err) { + ti->error = "Failed to set key for ESSIV cipher"; + crypto_free_cipher(essiv_tfm); + return ERR_PTR(err); + } + + return essiv_tfm; } static void crypt_iv_essiv_dtr(struct crypt_config *cc) { + int cpu; + struct crypt_cpu *cs; + struct crypto_cipher *essiv_tfm; struct iv_essiv_private *essiv = &cc->iv_gen_private.essiv; - crypto_free_cipher(essiv->tfm); - essiv->tfm = NULL; - crypto_free_hash(essiv->hash_tfm); essiv->hash_tfm = NULL; kzfree(essiv->salt); essiv->salt = NULL; + + for_each_possible_cpu (cpu) { + cs = per_cpu_ptr(cc->cpu, cpu); + essiv_tfm = cs->iv_private; + if (essiv_tfm) + crypto_free_cipher(essiv_tfm); + cs->iv_private = NULL; + } } static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti, @@ -240,7 +336,7 @@ static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti, struct crypto_cipher *essiv_tfm = NULL; struct crypto_hash *hash_tfm = NULL; u8 *salt = NULL; - int err; + int err, cpu; if (!opts) { ti->error = "Digest algorithm missing for ESSIV mode"; @@ -262,30 +358,21 @@ static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti, goto bad; } - /* Allocate essiv_tfm */ - essiv_tfm = crypto_alloc_cipher(cc->cipher, 0, CRYPTO_ALG_ASYNC); - if (IS_ERR(essiv_tfm)) { - ti->error = "Error allocating crypto tfm for ESSIV"; - err = PTR_ERR(essiv_tfm); - goto bad; - } - if (crypto_cipher_blocksize(essiv_tfm) != - crypto_ablkcipher_ivsize(cc->tfm)) { - ti->error = "Block size of ESSIV cipher does " - "not match IV size of block cipher"; - err = -EINVAL; - goto bad; - } - cc->iv_gen_private.essiv.salt = salt; - cc->iv_gen_private.essiv.tfm = essiv_tfm; cc->iv_gen_private.essiv.hash_tfm = hash_tfm; + for_each_possible_cpu (cpu) { + essiv_tfm = setup_essiv_cpu(cc, ti, salt, + crypto_hash_digestsize(hash_tfm)); + if (IS_ERR(essiv_tfm)) { + crypt_iv_essiv_dtr(cc); + return PTR_ERR(essiv_tfm); + } + per_cpu_ptr(cc->cpu, cpu)->iv_private = essiv_tfm; + } return 0; bad: - if (essiv_tfm && !IS_ERR(essiv_tfm)) - crypto_free_cipher(essiv_tfm); if (hash_tfm && !IS_ERR(hash_tfm)) crypto_free_hash(hash_tfm); kfree(salt); @@ -294,16 +381,18 @@ bad: static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector) { + struct crypto_cipher *essiv_tfm = crypt_me(cc)->iv_private; + memset(iv, 0, cc->iv_size); *(u64 *)iv = cpu_to_le64(sector); - crypto_cipher_encrypt_one(cc->iv_gen_private.essiv.tfm, iv, iv); + crypto_cipher_encrypt_one(essiv_tfm, iv, iv); return 0; } static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti, const char *opts) { - unsigned bs = crypto_ablkcipher_blocksize(cc->tfm); + unsigned bs = crypto_ablkcipher_blocksize(any_tfm(cc)); int log = ilog2(bs); /* we need to calculate how far we must shift the sector count @@ -412,7 +501,7 @@ static int crypt_convert_block(struct crypt_config *cc, dmreq = dmreq_of_req(cc, req); iv = (u8 *)ALIGN((unsigned long)(dmreq + 1), - crypto_ablkcipher_alignmask(cc->tfm) + 1); + crypto_ablkcipher_alignmask(any_tfm(cc)) + 1); dmreq->ctx = ctx; sg_init_table(&dmreq->sg_in, 1); @@ -457,13 +546,14 @@ static void kcryptd_async_done(struct crypto_async_request *async_req, static void crypt_alloc_req(struct crypt_config *cc, struct convert_context *ctx) { - if (!cc->req) - cc->req = mempool_alloc(cc->req_pool, GFP_NOIO); - ablkcipher_request_set_tfm(cc->req, cc->tfm); - ablkcipher_request_set_callback(cc->req, CRYPTO_TFM_REQ_MAY_BACKLOG | + struct crypt_cpu *cs = crypt_me(cc); + if (!cs->req) + cs->req = mempool_alloc(cc->req_pool, GFP_NOIO); + ablkcipher_request_set_tfm(cs->req, cs->tfm); + ablkcipher_request_set_callback(cs->req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP, kcryptd_async_done, - dmreq_of_req(cc, cc->req)); + dmreq_of_req(cc, cs->req)); } /* @@ -472,6 +562,7 @@ static void crypt_alloc_req(struct crypt_config *cc, static int crypt_convert(struct crypt_config *cc, struct convert_context *ctx) { + struct crypt_cpu *cs = crypt_me(cc); int r; atomic_set(&ctx->pending, 1); @@ -483,7 +574,7 @@ static int crypt_convert(struct crypt_config *cc, atomic_inc(&ctx->pending); - r = crypt_convert_block(cc, ctx, cc->req); + r = crypt_convert_block(cc, ctx, cs->req); switch (r) { /* async */ @@ -492,7 +583,7 @@ static int crypt_convert(struct crypt_config *cc, INIT_COMPLETION(ctx->restart); /* fall through*/ case -EINPROGRESS: - cc->req = NULL; + cs->req = NULL; ctx->sector++; continue; @@ -651,6 +742,9 @@ static void crypt_dec_pending(struct dm_crypt_io *io) * They must be separated as otherwise the final stages could be * starved by new requests which can block in the first stages due * to memory allocation. + * + * The work is done per CPU global for all dmcrypt instances. + * They should not depend on each other and do not block. */ static void crypt_endio(struct bio *clone, int error) { @@ -971,6 +1065,22 @@ static void crypt_encode_key(char *hex, u8 *key, unsigned int size) } } +static int crypt_setkey_allcpus(struct crypt_config *cc) +{ + struct crypt_cpu *cs; + int cpu, n, err; + + err = 0; + for_each_possible_cpu(cpu) { + cs = per_cpu_ptr(cc->cpu, cpu); + n = crypto_ablkcipher_setkey(cs->tfm, cc->key, cc->key_size); + if (n) + err = n; + } + return err; +} + + static int crypt_set_key(struct crypt_config *cc, char *key) { unsigned key_size = strlen(key) >> 1; @@ -986,19 +1096,35 @@ static int crypt_set_key(struct crypt_config *cc, char *key) set_bit(DM_CRYPT_KEY_VALID, &cc->flags); - return crypto_ablkcipher_setkey(cc->tfm, cc->key, cc->key_size); + return crypt_setkey_allcpus(cc); } static int crypt_wipe_key(struct crypt_config *cc) { clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); memset(&cc->key, 0, cc->key_size * sizeof(u8)); - return crypto_ablkcipher_setkey(cc->tfm, cc->key, cc->key_size); + return crypt_setkey_allcpus(cc); +} + +static void crypt_dtr_cpus(struct crypt_config *cc) +{ + struct crypt_cpu *cs; + int cpu; + + for_each_possible_cpu(cpu) { + cs = per_cpu_ptr(cc->cpu, cpu); + if (cs->tfm) { + crypto_free_ablkcipher(cs->tfm); + cs->tfm = NULL; + } + } } static void crypt_dtr(struct dm_target *ti) { struct crypt_config *cc = ti->private; + struct crypt_cpu *cs; + int cpu; ti->private = NULL; @@ -1010,6 +1136,12 @@ static void crypt_dtr(struct dm_target *ti) if (cc->crypt_queue) destroy_workqueue(cc->crypt_queue); + for_each_possible_cpu(cpu) { + cs = per_cpu_ptr(cc->cpu, cpu); + if (cs->req) + mempool_free(cs->req, cc->req_pool); + } + if (cc->bs) bioset_free(cc->bs); @@ -1023,12 +1155,14 @@ static void crypt_dtr(struct dm_target *ti) if (cc->iv_gen_ops && cc->iv_gen_ops->dtr) cc->iv_gen_ops->dtr(cc); - if (cc->tfm && !IS_ERR(cc->tfm)) - crypto_free_ablkcipher(cc->tfm); - if (cc->dev) dm_put_device(ti, cc->dev); + crypt_dtr_cpus(cc); + + if (cc->cpu) + free_percpu(cc->cpu); + kzfree(cc->cipher); kzfree(cc->cipher_mode); @@ -1040,9 +1174,10 @@ static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key) { struct crypt_config *cc = ti->private; + struct crypto_ablkcipher *tfm; char *tmp, *cipher, *chainmode, *ivmode, *ivopts; char *cipher_api = NULL; - int ret = -EINVAL; + int cpu, ret = -EINVAL; /* Convert to crypto api definition? */ if (strchr(cipher_in, '(')) { @@ -1074,6 +1209,12 @@ static int crypt_ctr_cipher(struct dm_target *ti, if (tmp) DMWARN("Ignoring unexpected additional cipher options"); + cc->cpu = alloc_percpu(struct crypt_cpu); + if (!cc->cpu) { + ti->error = "Cannot allocate per cpu state"; + goto bad_mem; + } + /* Compatibility mode for old dm-crypt mappings */ if (!chainmode || (!strcmp(chainmode, "plain") && !ivmode)) { kfree(cc->cipher_mode); @@ -1099,12 +1240,16 @@ static int crypt_ctr_cipher(struct dm_target *ti, } /* Allocate cipher */ - cc->tfm = crypto_alloc_ablkcipher(cipher_api, 0, 0); - if (IS_ERR(cc->tfm)) { - ret = PTR_ERR(cc->tfm); - ti->error = "Error allocating crypto tfm"; - goto bad; - } + for_each_possible_cpu (cpu) { + tfm = crypto_alloc_ablkcipher(cipher_api, 0, 0); + if (IS_ERR(tfm)) { + ret = PTR_ERR(tfm); + ti->error = "Error allocating crypto tfm"; + goto bad; + } + per_cpu_ptr(cc->cpu, cpu)->tfm = tfm; + } + tfm = any_tfm(cc); /* Initialize and set key */ ret = crypt_set_key(cc, key); @@ -1114,7 +1259,7 @@ static int crypt_ctr_cipher(struct dm_target *ti, } /* Initialize IV */ - cc->iv_size = crypto_ablkcipher_ivsize(cc->tfm); + cc->iv_size = crypto_ablkcipher_ivsize(tfm); if (cc->iv_size) /* at least a 64 bit sector number should fit in our buffer */ cc->iv_size = max(cc->iv_size, @@ -1178,9 +1323,10 @@ bad_mem: static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) { struct crypt_config *cc; + struct crypt_cpu *cs; unsigned int key_size; unsigned long long tmpll; - int ret; + int ret, cpu; if (argc != 5) { ti->error = "Not enough arguments"; @@ -1208,9 +1354,9 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) } cc->dmreq_start = sizeof(struct ablkcipher_request); - cc->dmreq_start += crypto_ablkcipher_reqsize(cc->tfm); + cc->dmreq_start += crypto_ablkcipher_reqsize(any_tfm(cc)); cc->dmreq_start = ALIGN(cc->dmreq_start, crypto_tfm_ctx_alignment()); - cc->dmreq_start += crypto_ablkcipher_alignmask(cc->tfm) & + cc->dmreq_start += crypto_ablkcipher_alignmask(any_tfm(cc)) & ~(crypto_tfm_ctx_alignment() - 1); cc->req_pool = mempool_create_kmalloc_pool(MIN_IOS, cc->dmreq_start + @@ -1219,7 +1365,6 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) ti->error = "Cannot allocate crypt request mempool"; goto bad; } - cc->req = NULL; cc->page_pool = mempool_create_page_pool(MIN_POOL_PAGES, 0); if (!cc->page_pool) { @@ -1233,6 +1378,14 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) goto bad; } + for_each_possible_cpu(cpu) { + cs = per_cpu_ptr(cc->cpu, cpu); + if (crypto_ablkcipher_setkey(cs->tfm, cc->key, key_size) < 0) { + ti->error = "Error setting key"; + goto bad; + } + } + ret = -EINVAL; if (sscanf(argv[2], "%llu", &tmpll) != 1) { ti->error = "Invalid iv_offset sector"; @@ -1252,13 +1405,20 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) cc->start = tmpll; ret = -ENOMEM; - cc->io_queue = create_singlethread_workqueue("kcryptd_io"); + cc->io_queue = alloc_workqueue("kcryptd_io", + WQ_NON_REENTRANT| + WQ_MEM_RECLAIM, + 1); if (!cc->io_queue) { ti->error = "Couldn't create kcryptd io queue"; goto bad; } - cc->crypt_queue = create_singlethread_workqueue("kcryptd"); + cc->crypt_queue = alloc_workqueue("kcryptd", + WQ_NON_REENTRANT| + WQ_CPU_INTENSIVE| + WQ_MEM_RECLAIM, + 1); if (!cc->crypt_queue) { ti->error = "Couldn't create kcryptd queue"; goto bad;