610 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 * bio-integrity.c - bio data integrity extensions4 *5 * Copyright (C) 2007, 2008, 2009 Oracle Corporation6 * Written by: Martin K. Petersen <martin.petersen@oracle.com>7 */8 9#include <linux/blk-integrity.h>10#include <linux/mempool.h>11#include <linux/export.h>12#include <linux/bio.h>13#include <linux/workqueue.h>14#include <linux/slab.h>15#include "blk.h"16 17static struct kmem_cache *bip_slab;18static struct workqueue_struct *kintegrityd_wq;19 20void blk_flush_integrity(void)21{22 flush_workqueue(kintegrityd_wq);23}24 25/**26 * bio_integrity_free - Free bio integrity payload27 * @bio: bio containing bip to be freed28 *29 * Description: Free the integrity portion of a bio.30 */31void bio_integrity_free(struct bio *bio)32{33 struct bio_integrity_payload *bip = bio_integrity(bio);34 struct bio_set *bs = bio->bi_pool;35 36 if (bs && mempool_initialized(&bs->bio_integrity_pool)) {37 if (bip->bip_vec)38 bvec_free(&bs->bvec_integrity_pool, bip->bip_vec,39 bip->bip_max_vcnt);40 mempool_free(bip, &bs->bio_integrity_pool);41 } else {42 kfree(bip);43 }44 bio->bi_integrity = NULL;45 bio->bi_opf &= ~REQ_INTEGRITY;46}47 48/**49 * bio_integrity_alloc - Allocate integrity payload and attach it to bio50 * @bio: bio to attach integrity metadata to51 * @gfp_mask: Memory allocation mask52 * @nr_vecs: Number of integrity metadata scatter-gather elements53 *54 * Description: This function prepares a bio for attaching integrity55 * metadata. nr_vecs specifies the maximum number of pages containing56 * integrity metadata that can be attached.57 */58struct bio_integrity_payload *bio_integrity_alloc(struct bio *bio,59 gfp_t gfp_mask,60 unsigned int nr_vecs)61{62 struct bio_integrity_payload *bip;63 struct bio_set *bs = bio->bi_pool;64 unsigned inline_vecs;65 66 if (WARN_ON_ONCE(bio_has_crypt_ctx(bio)))67 return ERR_PTR(-EOPNOTSUPP);68 69 if (!bs || !mempool_initialized(&bs->bio_integrity_pool)) {70 bip = kmalloc(struct_size(bip, bip_inline_vecs, nr_vecs), gfp_mask);71 inline_vecs = nr_vecs;72 } else {73 bip = mempool_alloc(&bs->bio_integrity_pool, gfp_mask);74 inline_vecs = BIO_INLINE_VECS;75 }76 77 if (unlikely(!bip))78 return ERR_PTR(-ENOMEM);79 80 memset(bip, 0, sizeof(*bip));81 82 /* always report as many vecs as asked explicitly, not inline vecs */83 bip->bip_max_vcnt = nr_vecs;84 if (nr_vecs > inline_vecs) {85 bip->bip_vec = bvec_alloc(&bs->bvec_integrity_pool,86 &bip->bip_max_vcnt, gfp_mask);87 if (!bip->bip_vec)88 goto err;89 } else if (nr_vecs) {90 bip->bip_vec = bip->bip_inline_vecs;91 }92 93 bip->bip_bio = bio;94 bio->bi_integrity = bip;95 bio->bi_opf |= REQ_INTEGRITY;96 97 return bip;98err:99 if (bs && mempool_initialized(&bs->bio_integrity_pool))100 mempool_free(bip, &bs->bio_integrity_pool);101 else102 kfree(bip);103 return ERR_PTR(-ENOMEM);104}105EXPORT_SYMBOL(bio_integrity_alloc);106 107static void bio_integrity_unpin_bvec(struct bio_vec *bv, int nr_vecs,108 bool dirty)109{110 int i;111 112 for (i = 0; i < nr_vecs; i++) {113 if (dirty && !PageCompound(bv[i].bv_page))114 set_page_dirty_lock(bv[i].bv_page);115 unpin_user_page(bv[i].bv_page);116 }117}118 119static void bio_integrity_uncopy_user(struct bio_integrity_payload *bip)120{121 unsigned short nr_vecs = bip->bip_max_vcnt - 1;122 struct bio_vec *copy = &bip->bip_vec[1];123 size_t bytes = bip->bip_iter.bi_size;124 struct iov_iter iter;125 int ret;126 127 iov_iter_bvec(&iter, ITER_DEST, copy, nr_vecs, bytes);128 ret = copy_to_iter(bvec_virt(bip->bip_vec), bytes, &iter);129 WARN_ON_ONCE(ret != bytes);130 131 bio_integrity_unpin_bvec(copy, nr_vecs, true);132}133 134/**135 * bio_integrity_unmap_user - Unmap user integrity payload136 * @bio: bio containing bip to be unmapped137 *138 * Unmap the user mapped integrity portion of a bio.139 */140void bio_integrity_unmap_user(struct bio *bio)141{142 struct bio_integrity_payload *bip = bio_integrity(bio);143 144 if (bip->bip_flags & BIP_COPY_USER) {145 if (bio_data_dir(bio) == READ)146 bio_integrity_uncopy_user(bip);147 kfree(bvec_virt(bip->bip_vec));148 return;149 }150 151 bio_integrity_unpin_bvec(bip->bip_vec, bip->bip_max_vcnt,152 bio_data_dir(bio) == READ);153}154 155/**156 * bio_integrity_add_page - Attach integrity metadata157 * @bio: bio to update158 * @page: page containing integrity metadata159 * @len: number of bytes of integrity metadata in page160 * @offset: start offset within page161 *162 * Description: Attach a page containing integrity metadata to bio.163 */164int bio_integrity_add_page(struct bio *bio, struct page *page,165 unsigned int len, unsigned int offset)166{167 struct request_queue *q = bdev_get_queue(bio->bi_bdev);168 struct bio_integrity_payload *bip = bio_integrity(bio);169 170 if (bip->bip_vcnt > 0) {171 struct bio_vec *bv = &bip->bip_vec[bip->bip_vcnt - 1];172 bool same_page = false;173 174 if (bvec_try_merge_hw_page(q, bv, page, len, offset,175 &same_page)) {176 bip->bip_iter.bi_size += len;177 return len;178 }179 180 if (bip->bip_vcnt >=181 min(bip->bip_max_vcnt, queue_max_integrity_segments(q)))182 return 0;183 184 /*185 * If the queue doesn't support SG gaps and adding this segment186 * would create a gap, disallow it.187 */188 if (bvec_gap_to_prev(&q->limits, bv, offset))189 return 0;190 }191 192 bvec_set_page(&bip->bip_vec[bip->bip_vcnt], page, len, offset);193 bip->bip_vcnt++;194 bip->bip_iter.bi_size += len;195 196 return len;197}198EXPORT_SYMBOL(bio_integrity_add_page);199 200static int bio_integrity_copy_user(struct bio *bio, struct bio_vec *bvec,201 int nr_vecs, unsigned int len,202 unsigned int direction, u32 seed)203{204 bool write = direction == ITER_SOURCE;205 struct bio_integrity_payload *bip;206 struct iov_iter iter;207 void *buf;208 int ret;209 210 buf = kmalloc(len, GFP_KERNEL);211 if (!buf)212 return -ENOMEM;213 214 if (write) {215 iov_iter_bvec(&iter, direction, bvec, nr_vecs, len);216 if (!copy_from_iter_full(buf, len, &iter)) {217 ret = -EFAULT;218 goto free_buf;219 }220 221 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);222 } else {223 memset(buf, 0, len);224 225 /*226 * We need to preserve the original bvec and the number of vecs227 * in it for completion handling228 */229 bip = bio_integrity_alloc(bio, GFP_KERNEL, nr_vecs + 1);230 }231 232 if (IS_ERR(bip)) {233 ret = PTR_ERR(bip);234 goto free_buf;235 }236 237 if (write)238 bio_integrity_unpin_bvec(bvec, nr_vecs, false);239 else240 memcpy(&bip->bip_vec[1], bvec, nr_vecs * sizeof(*bvec));241 242 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,243 offset_in_page(buf));244 if (ret != len) {245 ret = -ENOMEM;246 goto free_bip;247 }248 249 bip->bip_flags |= BIP_COPY_USER;250 bip->bip_iter.bi_sector = seed;251 bip->bip_vcnt = nr_vecs;252 return 0;253free_bip:254 bio_integrity_free(bio);255free_buf:256 kfree(buf);257 return ret;258}259 260static int bio_integrity_init_user(struct bio *bio, struct bio_vec *bvec,261 int nr_vecs, unsigned int len, u32 seed)262{263 struct bio_integrity_payload *bip;264 265 bip = bio_integrity_alloc(bio, GFP_KERNEL, nr_vecs);266 if (IS_ERR(bip))267 return PTR_ERR(bip);268 269 memcpy(bip->bip_vec, bvec, nr_vecs * sizeof(*bvec));270 bip->bip_iter.bi_sector = seed;271 bip->bip_iter.bi_size = len;272 bip->bip_vcnt = nr_vecs;273 return 0;274}275 276static unsigned int bvec_from_pages(struct bio_vec *bvec, struct page **pages,277 int nr_vecs, ssize_t bytes, ssize_t offset)278{279 unsigned int nr_bvecs = 0;280 int i, j;281 282 for (i = 0; i < nr_vecs; i = j) {283 size_t size = min_t(size_t, bytes, PAGE_SIZE - offset);284 struct folio *folio = page_folio(pages[i]);285 286 bytes -= size;287 for (j = i + 1; j < nr_vecs; j++) {288 size_t next = min_t(size_t, PAGE_SIZE, bytes);289 290 if (page_folio(pages[j]) != folio ||291 pages[j] != pages[j - 1] + 1)292 break;293 unpin_user_page(pages[j]);294 size += next;295 bytes -= next;296 }297 298 bvec_set_page(&bvec[nr_bvecs], pages[i], size, offset);299 offset = 0;300 nr_bvecs++;301 }302 303 return nr_bvecs;304}305 306int bio_integrity_map_user(struct bio *bio, void __user *ubuf, ssize_t bytes,307 u32 seed)308{309 struct request_queue *q = bdev_get_queue(bio->bi_bdev);310 unsigned int align = blk_lim_dma_alignment_and_pad(&q->limits);311 struct page *stack_pages[UIO_FASTIOV], **pages = stack_pages;312 struct bio_vec stack_vec[UIO_FASTIOV], *bvec = stack_vec;313 unsigned int direction, nr_bvecs;314 struct iov_iter iter;315 int ret, nr_vecs;316 size_t offset;317 bool copy;318 319 if (bio_integrity(bio))320 return -EINVAL;321 if (bytes >> SECTOR_SHIFT > queue_max_hw_sectors(q))322 return -E2BIG;323 324 if (bio_data_dir(bio) == READ)325 direction = ITER_DEST;326 else327 direction = ITER_SOURCE;328 329 iov_iter_ubuf(&iter, direction, ubuf, bytes);330 nr_vecs = iov_iter_npages(&iter, BIO_MAX_VECS + 1);331 if (nr_vecs > BIO_MAX_VECS)332 return -E2BIG;333 if (nr_vecs > UIO_FASTIOV) {334 bvec = kcalloc(nr_vecs, sizeof(*bvec), GFP_KERNEL);335 if (!bvec)336 return -ENOMEM;337 pages = NULL;338 }339 340 copy = !iov_iter_is_aligned(&iter, align, align);341 ret = iov_iter_extract_pages(&iter, &pages, bytes, nr_vecs, 0, &offset);342 if (unlikely(ret < 0))343 goto free_bvec;344 345 nr_bvecs = bvec_from_pages(bvec, pages, nr_vecs, bytes, offset);346 if (pages != stack_pages)347 kvfree(pages);348 if (nr_bvecs > queue_max_integrity_segments(q))349 copy = true;350 351 if (copy)352 ret = bio_integrity_copy_user(bio, bvec, nr_bvecs, bytes,353 direction, seed);354 else355 ret = bio_integrity_init_user(bio, bvec, nr_bvecs, bytes, seed);356 if (ret)357 goto release_pages;358 if (bvec != stack_vec)359 kfree(bvec);360 361 return 0;362 363release_pages:364 bio_integrity_unpin_bvec(bvec, nr_bvecs, false);365free_bvec:366 if (bvec != stack_vec)367 kfree(bvec);368 return ret;369}370 371/**372 * bio_integrity_prep - Prepare bio for integrity I/O373 * @bio: bio to prepare374 *375 * Description: Checks if the bio already has an integrity payload attached.376 * If it does, the payload has been generated by another kernel subsystem,377 * and we just pass it through. Otherwise allocates integrity payload.378 * The bio must have data direction, target device and start sector set priot379 * to calling. In the WRITE case, integrity metadata will be generated using380 * the block device's integrity function. In the READ case, the buffer381 * will be prepared for DMA and a suitable end_io handler set up.382 */383bool bio_integrity_prep(struct bio *bio)384{385 struct bio_integrity_payload *bip;386 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);387 unsigned int len;388 void *buf;389 gfp_t gfp = GFP_NOIO;390 391 if (!bi)392 return true;393 394 if (!bio_sectors(bio))395 return true;396 397 /* Already protected? */398 if (bio_integrity(bio))399 return true;400 401 switch (bio_op(bio)) {402 case REQ_OP_READ:403 if (bi->flags & BLK_INTEGRITY_NOVERIFY)404 return true;405 break;406 case REQ_OP_WRITE:407 if (bi->flags & BLK_INTEGRITY_NOGENERATE)408 return true;409 410 /*411 * Zero the memory allocated to not leak uninitialized kernel412 * memory to disk for non-integrity metadata where nothing else413 * initializes the memory.414 */415 if (bi->csum_type == BLK_INTEGRITY_CSUM_NONE)416 gfp |= __GFP_ZERO;417 break;418 default:419 return true;420 }421 422 /* Allocate kernel buffer for protection data */423 len = bio_integrity_bytes(bi, bio_sectors(bio));424 buf = kmalloc(len, gfp);425 if (unlikely(buf == NULL)) {426 goto err_end_io;427 }428 429 bip = bio_integrity_alloc(bio, GFP_NOIO, 1);430 if (IS_ERR(bip)) {431 kfree(buf);432 goto err_end_io;433 }434 435 bip->bip_flags |= BIP_BLOCK_INTEGRITY;436 bip_set_seed(bip, bio->bi_iter.bi_sector);437 438 if (bi->csum_type == BLK_INTEGRITY_CSUM_IP)439 bip->bip_flags |= BIP_IP_CHECKSUM;440 441 if (bio_integrity_add_page(bio, virt_to_page(buf), len,442 offset_in_page(buf)) < len) {443 printk(KERN_ERR "could not attach integrity payload\n");444 goto err_end_io;445 }446 447 /* Auto-generate integrity metadata if this is a write */448 if (bio_data_dir(bio) == WRITE)449 blk_integrity_generate(bio);450 else451 bip->bio_iter = bio->bi_iter;452 return true;453 454err_end_io:455 bio->bi_status = BLK_STS_RESOURCE;456 bio_endio(bio);457 return false;458}459EXPORT_SYMBOL(bio_integrity_prep);460 461/**462 * bio_integrity_verify_fn - Integrity I/O completion worker463 * @work: Work struct stored in bio to be verified464 *465 * Description: This workqueue function is called to complete a READ466 * request. The function verifies the transferred integrity metadata467 * and then calls the original bio end_io function.468 */469static void bio_integrity_verify_fn(struct work_struct *work)470{471 struct bio_integrity_payload *bip =472 container_of(work, struct bio_integrity_payload, bip_work);473 struct bio *bio = bip->bip_bio;474 475 blk_integrity_verify(bio);476 477 kfree(bvec_virt(bip->bip_vec));478 bio_integrity_free(bio);479 bio_endio(bio);480}481 482/**483 * __bio_integrity_endio - Integrity I/O completion function484 * @bio: Protected bio485 *486 * Description: Completion for integrity I/O487 *488 * Normally I/O completion is done in interrupt context. However,489 * verifying I/O integrity is a time-consuming task which must be run490 * in process context. This function postpones completion491 * accordingly.492 */493bool __bio_integrity_endio(struct bio *bio)494{495 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);496 struct bio_integrity_payload *bip = bio_integrity(bio);497 498 if (bio_op(bio) == REQ_OP_READ && !bio->bi_status && bi->csum_type) {499 INIT_WORK(&bip->bip_work, bio_integrity_verify_fn);500 queue_work(kintegrityd_wq, &bip->bip_work);501 return false;502 }503 504 kfree(bvec_virt(bip->bip_vec));505 bio_integrity_free(bio);506 return true;507}508 509/**510 * bio_integrity_advance - Advance integrity vector511 * @bio: bio whose integrity vector to update512 * @bytes_done: number of data bytes that have been completed513 *514 * Description: This function calculates how many integrity bytes the515 * number of completed data bytes correspond to and advances the516 * integrity vector accordingly.517 */518void bio_integrity_advance(struct bio *bio, unsigned int bytes_done)519{520 struct bio_integrity_payload *bip = bio_integrity(bio);521 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);522 unsigned bytes = bio_integrity_bytes(bi, bytes_done >> 9);523 524 bip->bip_iter.bi_sector += bio_integrity_intervals(bi, bytes_done >> 9);525 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, bytes);526}527 528/**529 * bio_integrity_trim - Trim integrity vector530 * @bio: bio whose integrity vector to update531 *532 * Description: Used to trim the integrity vector in a cloned bio.533 */534void bio_integrity_trim(struct bio *bio)535{536 struct bio_integrity_payload *bip = bio_integrity(bio);537 struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk);538 539 bip->bip_iter.bi_size = bio_integrity_bytes(bi, bio_sectors(bio));540}541EXPORT_SYMBOL(bio_integrity_trim);542 543/**544 * bio_integrity_clone - Callback for cloning bios with integrity metadata545 * @bio: New bio546 * @bio_src: Original bio547 * @gfp_mask: Memory allocation mask548 *549 * Description: Called to allocate a bip when cloning a bio550 */551int bio_integrity_clone(struct bio *bio, struct bio *bio_src,552 gfp_t gfp_mask)553{554 struct bio_integrity_payload *bip_src = bio_integrity(bio_src);555 struct bio_integrity_payload *bip;556 557 BUG_ON(bip_src == NULL);558 559 bip = bio_integrity_alloc(bio, gfp_mask, 0);560 if (IS_ERR(bip))561 return PTR_ERR(bip);562 563 bip->bip_vec = bip_src->bip_vec;564 bip->bip_iter = bip_src->bip_iter;565 bip->bip_flags = bip_src->bip_flags & ~BIP_BLOCK_INTEGRITY;566 567 return 0;568}569 570int bioset_integrity_create(struct bio_set *bs, int pool_size)571{572 if (mempool_initialized(&bs->bio_integrity_pool))573 return 0;574 575 if (mempool_init_slab_pool(&bs->bio_integrity_pool,576 pool_size, bip_slab))577 return -1;578 579 if (biovec_init_pool(&bs->bvec_integrity_pool, pool_size)) {580 mempool_exit(&bs->bio_integrity_pool);581 return -1;582 }583 584 return 0;585}586EXPORT_SYMBOL(bioset_integrity_create);587 588void bioset_integrity_free(struct bio_set *bs)589{590 mempool_exit(&bs->bio_integrity_pool);591 mempool_exit(&bs->bvec_integrity_pool);592}593 594void __init bio_integrity_init(void)595{596 /*597 * kintegrityd won't block much but may burn a lot of CPU cycles.598 * Make it highpri CPU intensive wq with max concurrency of 1.599 */600 kintegrityd_wq = alloc_workqueue("kintegrityd", WQ_MEM_RECLAIM |601 WQ_HIGHPRI | WQ_CPU_INTENSIVE, 1);602 if (!kintegrityd_wq)603 panic("Failed to create kintegrityd\n");604 605 bip_slab = kmem_cache_create("bio_integrity_payload",606 sizeof(struct bio_integrity_payload) +607 sizeof(struct bio_vec) * BIO_INLINE_VECS,608 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);609}610