Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
blk-crypto: optimize bio splitting in blk_crypto_fallback_encrypt_bio
The current code in blk_crypto_fallback_encrypt_bio is inefficient and
prone to deadlocks under memory pressure: It first walks the passed in
plaintext bio to see how much of it can fit into a single encrypted
bio using up to BIO_MAX_VEC PAGE_SIZE segments, and then allocates a
plaintext clone that fits the size, only to allocate another bio for
the ciphertext later. While the plaintext clone uses a bioset to avoid
deadlocks when allocations could fail, the ciphertex one uses bio_kmalloc
which is a no-go in the file system I/O path.
Switch blk_crypto_fallback_encrypt_bio to walk the source plaintext bio
while consuming bi_iter without cloning it, and instead allocate a
ciphertext bio at the beginning and whenever we fille up the previous
one. The existing bio_set for the plaintext clones is reused for the
ciphertext bios to remove the deadlock risk.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
authored by
Christoph Hellwig
and committed by
Jens Axboe
b37fbce4
aefc2a1f
+76
-105
+76
-105
block/blk-crypto-fallback.c
+76
-105
block/blk-crypto-fallback.c
···
81
81
static struct blk_crypto_profile *blk_crypto_fallback_profile;
82
82
static struct workqueue_struct *blk_crypto_wq;
83
83
static mempool_t *blk_crypto_bounce_page_pool;
84
-
static struct bio_set crypto_bio_split;
84
+
static struct bio_set enc_bio_set;
85
85
86
86
/*
87
87
* This is the key we set when evicting a keyslot. This *should* be the all 0's
···
150
150
mempool_free(enc_bio->bi_io_vec[i].bv_page,
151
151
blk_crypto_bounce_page_pool);
152
152
153
-
src_bio->bi_status = enc_bio->bi_status;
153
+
if (enc_bio->bi_status)
154
+
cmpxchg(&src_bio->bi_status, 0, enc_bio->bi_status);
154
155
155
-
bio_uninit(enc_bio);
156
-
kfree(enc_bio);
156
+
bio_put(enc_bio);
157
157
bio_endio(src_bio);
158
158
}
159
159
160
-
static struct bio *blk_crypto_fallback_clone_bio(struct bio *bio_src)
160
+
static struct bio *blk_crypto_alloc_enc_bio(struct bio *bio_src,
161
+
unsigned int nr_segs)
161
162
{
162
-
unsigned int nr_segs = bio_segments(bio_src);
163
-
struct bvec_iter iter;
164
-
struct bio_vec bv;
165
163
struct bio *bio;
166
164
167
-
bio = bio_kmalloc(nr_segs, GFP_NOIO);
168
-
if (!bio)
169
-
return NULL;
170
-
bio_init_inline(bio, bio_src->bi_bdev, nr_segs, bio_src->bi_opf);
165
+
bio = bio_alloc_bioset(bio_src->bi_bdev, nr_segs, bio_src->bi_opf,
166
+
GFP_NOIO, &enc_bio_set);
171
167
if (bio_flagged(bio_src, BIO_REMAPPED))
172
168
bio_set_flag(bio, BIO_REMAPPED);
169
+
bio->bi_private = bio_src;
170
+
bio->bi_end_io = blk_crypto_fallback_encrypt_endio;
173
171
bio->bi_ioprio = bio_src->bi_ioprio;
174
172
bio->bi_write_hint = bio_src->bi_write_hint;
175
173
bio->bi_write_stream = bio_src->bi_write_stream;
176
174
bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
177
-
bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
178
-
179
-
bio_for_each_segment(bv, bio_src, iter)
180
-
bio->bi_io_vec[bio->bi_vcnt++] = bv;
181
-
182
175
bio_clone_blkg_association(bio, bio_src);
183
-
184
176
return bio;
185
177
}
186
178
···
200
208
return true;
201
209
}
202
210
203
-
static bool blk_crypto_fallback_split_bio_if_needed(struct bio **bio_ptr)
204
-
{
205
-
struct bio *bio = *bio_ptr;
206
-
unsigned int i = 0;
207
-
unsigned int num_sectors = 0;
208
-
struct bio_vec bv;
209
-
struct bvec_iter iter;
210
-
211
-
bio_for_each_segment(bv, bio, iter) {
212
-
num_sectors += bv.bv_len >> SECTOR_SHIFT;
213
-
if (++i == BIO_MAX_VECS)
214
-
break;
215
-
}
216
-
217
-
if (num_sectors < bio_sectors(bio)) {
218
-
bio = bio_submit_split_bioset(bio, num_sectors,
219
-
&crypto_bio_split);
220
-
if (!bio)
221
-
return false;
222
-
223
-
*bio_ptr = bio;
224
-
}
225
-
226
-
return true;
227
-
}
228
-
229
211
union blk_crypto_iv {
230
212
__le64 dun[BLK_CRYPTO_DUN_ARRAY_SIZE];
231
213
u8 bytes[BLK_CRYPTO_MAX_IV_SIZE];
···
223
257
*/
224
258
static void blk_crypto_fallback_encrypt_bio(struct bio *src_bio)
225
259
{
226
-
struct bio *enc_bio;
227
-
struct bio_crypt_ctx *bc;
228
-
struct blk_crypto_keyslot *slot;
229
-
int data_unit_size;
260
+
struct bio_crypt_ctx *bc = src_bio->bi_crypt_context;
261
+
int data_unit_size = bc->bc_key->crypto_cfg.data_unit_size;
230
262
struct skcipher_request *ciph_req = NULL;
263
+
struct blk_crypto_keyslot *slot;
231
264
DECLARE_CRYPTO_WAIT(wait);
232
265
u64 curr_dun[BLK_CRYPTO_DUN_ARRAY_SIZE];
233
266
struct scatterlist src, dst;
234
267
union blk_crypto_iv iv;
235
-
unsigned int i, j;
236
-
blk_status_t blk_st;
237
-
238
-
/* Split the bio if it's too big for single page bvec */
239
-
if (!blk_crypto_fallback_split_bio_if_needed(&src_bio))
240
-
goto out_endio;
241
-
242
-
bc = src_bio->bi_crypt_context;
243
-
data_unit_size = bc->bc_key->crypto_cfg.data_unit_size;
244
-
245
-
/* Allocate bounce bio for encryption */
246
-
enc_bio = blk_crypto_fallback_clone_bio(src_bio);
247
-
if (!enc_bio) {
248
-
src_bio->bi_status = BLK_STS_RESOURCE;
249
-
goto out_endio;
250
-
}
268
+
unsigned int nr_enc_pages, enc_idx;
269
+
struct bio *enc_bio;
270
+
blk_status_t status;
271
+
unsigned int i;
251
272
252
273
/*
253
274
* Get a blk-crypto-fallback keyslot that contains a crypto_skcipher for
254
275
* this bio's algorithm and key.
255
276
*/
256
-
blk_st = blk_crypto_get_keyslot(blk_crypto_fallback_profile,
277
+
status = blk_crypto_get_keyslot(blk_crypto_fallback_profile,
257
278
bc->bc_key, &slot);
258
-
if (blk_st != BLK_STS_OK) {
259
-
src_bio->bi_status = blk_st;
260
-
goto out_put_enc_bio;
279
+
if (status != BLK_STS_OK) {
280
+
src_bio->bi_status = status;
281
+
bio_endio(src_bio);
282
+
return;
261
283
}
262
284
263
285
/* and then allocate an skcipher_request for it */
264
286
if (!blk_crypto_fallback_alloc_cipher_req(slot, &ciph_req, &wait)) {
265
287
src_bio->bi_status = BLK_STS_RESOURCE;
288
+
bio_endio(src_bio);
266
289
goto out_release_keyslot;
267
290
}
268
291
···
262
307
skcipher_request_set_crypt(ciph_req, &src, &dst, data_unit_size,
263
308
iv.bytes);
264
309
265
-
/* Encrypt each page in the bounce bio */
266
-
for (i = 0; i < enc_bio->bi_vcnt; i++) {
267
-
struct bio_vec *enc_bvec = &enc_bio->bi_io_vec[i];
268
-
struct page *plaintext_page = enc_bvec->bv_page;
269
-
struct page *ciphertext_page =
270
-
mempool_alloc(blk_crypto_bounce_page_pool, GFP_NOIO);
310
+
/*
311
+
* Encrypt each page in the source bio. Because the source bio could
312
+
* have bio_vecs that span more than a single page, but the encrypted
313
+
* bios are limited to a single page per bio_vec, this can generate
314
+
* more than a single encrypted bio per source bio.
315
+
*/
316
+
new_bio:
317
+
nr_enc_pages = min(bio_segments(src_bio), BIO_MAX_VECS);
318
+
enc_bio = blk_crypto_alloc_enc_bio(src_bio, nr_enc_pages);
319
+
enc_idx = 0;
320
+
for (;;) {
321
+
struct bio_vec src_bv =
322
+
bio_iter_iovec(src_bio, src_bio->bi_iter);
323
+
struct page *enc_page;
271
324
272
-
enc_bvec->bv_page = ciphertext_page;
325
+
enc_page = mempool_alloc(blk_crypto_bounce_page_pool,
326
+
GFP_NOIO);
327
+
__bio_add_page(enc_bio, enc_page, src_bv.bv_len,
328
+
src_bv.bv_offset);
273
329
274
-
if (!ciphertext_page) {
275
-
src_bio->bi_status = BLK_STS_RESOURCE;
276
-
goto out_free_bounce_pages;
277
-
}
330
+
sg_set_page(&src, src_bv.bv_page, data_unit_size,
331
+
src_bv.bv_offset);
332
+
sg_set_page(&dst, enc_page, data_unit_size, src_bv.bv_offset);
278
333
279
-
sg_set_page(&src, plaintext_page, data_unit_size,
280
-
enc_bvec->bv_offset);
281
-
sg_set_page(&dst, ciphertext_page, data_unit_size,
282
-
enc_bvec->bv_offset);
334
+
/*
335
+
* Increment the index now that the encrypted page is added to
336
+
* the bio. This is important for the error unwind path.
337
+
*/
338
+
enc_idx++;
283
339
284
-
/* Encrypt each data unit in this page */
285
-
for (j = 0; j < enc_bvec->bv_len; j += data_unit_size) {
340
+
/*
341
+
* Encrypt each data unit in this page.
342
+
*/
343
+
for (i = 0; i < src_bv.bv_len; i += data_unit_size) {
286
344
blk_crypto_dun_to_iv(curr_dun, &iv);
287
345
if (crypto_wait_req(crypto_skcipher_encrypt(ciph_req),
288
346
&wait)) {
289
-
i++;
290
-
src_bio->bi_status = BLK_STS_IOERR;
291
-
goto out_free_bounce_pages;
347
+
bio_io_error(enc_bio);
348
+
goto out_free_request;
292
349
}
293
350
bio_crypt_dun_increment(curr_dun, 1);
294
351
src.offset += data_unit_size;
295
352
dst.offset += data_unit_size;
296
353
}
354
+
355
+
bio_advance_iter_single(src_bio, &src_bio->bi_iter,
356
+
src_bv.bv_len);
357
+
if (!src_bio->bi_iter.bi_size)
358
+
break;
359
+
360
+
if (enc_idx == nr_enc_pages) {
361
+
/*
362
+
* For each additional encrypted bio submitted,
363
+
* increment the source bio's remaining count. Each
364
+
* encrypted bio's completion handler calls bio_endio on
365
+
* the source bio, so this keeps the source bio from
366
+
* completing until the last encrypted bio does.
367
+
*/
368
+
bio_inc_remaining(src_bio);
369
+
submit_bio(enc_bio);
370
+
goto new_bio;
371
+
}
297
372
}
298
373
299
-
enc_bio->bi_private = src_bio;
300
-
enc_bio->bi_end_io = blk_crypto_fallback_encrypt_endio;
301
-
skcipher_request_free(ciph_req);
302
-
blk_crypto_put_keyslot(slot);
303
374
submit_bio(enc_bio);
304
-
return;
305
-
306
-
out_free_bounce_pages:
307
-
while (i > 0)
308
-
mempool_free(enc_bio->bi_io_vec[--i].bv_page,
309
-
blk_crypto_bounce_page_pool);
375
+
out_free_request:
310
376
skcipher_request_free(ciph_req);
311
377
out_release_keyslot:
312
378
blk_crypto_put_keyslot(slot);
313
-
out_put_enc_bio:
314
-
bio_uninit(enc_bio);
315
-
kfree(enc_bio);
316
-
out_endio:
317
-
bio_endio(src_bio);
318
379
}
319
380
320
381
/*
···
504
533
505
534
get_random_bytes(blank_key, sizeof(blank_key));
506
535
507
-
err = bioset_init(&crypto_bio_split, 64, 0, 0);
536
+
err = bioset_init(&enc_bio_set, 64, 0, BIOSET_NEED_BVECS);
508
537
if (err)
509
538
goto out;
510
539
···
574
603
fail_free_profile:
575
604
kfree(blk_crypto_fallback_profile);
576
605
fail_free_bioset:
577
-
bioset_exit(&crypto_bio_split);
606
+
bioset_exit(&enc_bio_set);
578
607
out:
579
608
return err;
580
609
}