1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 2d6d48196SJens Axboe /* 3d6d48196SJens Axboe * Functions related to segment and merge handling 4d6d48196SJens Axboe */ 5d6d48196SJens Axboe #include <linux/kernel.h> 6d6d48196SJens Axboe #include <linux/module.h> 7d6d48196SJens Axboe #include <linux/bio.h> 8d6d48196SJens Axboe #include <linux/blkdev.h> 9fe45e630SChristoph Hellwig #include <linux/blk-integrity.h> 10d6d48196SJens Axboe #include <linux/scatterlist.h> 1182d981d4SChristoph Hellwig #include <linux/part_stat.h> 126b2b0459STejun Heo #include <linux/blk-cgroup.h> 13d6d48196SJens Axboe 14cda22646SMike Krinkin #include <trace/events/block.h> 15cda22646SMike Krinkin 16d6d48196SJens Axboe #include "blk.h" 172aa7745bSChristoph Hellwig #include "blk-mq-sched.h" 188e756373SBaolin Wang #include "blk-rq-qos.h" 19a7b36ee6SJens Axboe #include "blk-throttle.h" 20d6d48196SJens Axboe 21ff18d77bSChristoph Hellwig static inline void bio_get_first_bvec(struct bio *bio, struct bio_vec *bv) 22ff18d77bSChristoph Hellwig { 23ff18d77bSChristoph Hellwig *bv = mp_bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 24ff18d77bSChristoph Hellwig } 25ff18d77bSChristoph Hellwig 26ff18d77bSChristoph Hellwig static inline void bio_get_last_bvec(struct bio *bio, struct bio_vec *bv) 27ff18d77bSChristoph Hellwig { 28ff18d77bSChristoph Hellwig struct bvec_iter iter = bio->bi_iter; 29ff18d77bSChristoph Hellwig int idx; 30ff18d77bSChristoph Hellwig 31ff18d77bSChristoph Hellwig bio_get_first_bvec(bio, bv); 32ff18d77bSChristoph Hellwig if (bv->bv_len == bio->bi_iter.bi_size) 33ff18d77bSChristoph Hellwig return; /* this bio only has a single bvec */ 34ff18d77bSChristoph Hellwig 35ff18d77bSChristoph Hellwig bio_advance_iter(bio, &iter, iter.bi_size); 36ff18d77bSChristoph Hellwig 37ff18d77bSChristoph Hellwig if (!iter.bi_bvec_done) 38ff18d77bSChristoph Hellwig idx = iter.bi_idx - 1; 39ff18d77bSChristoph Hellwig else /* in the middle of bvec */ 40ff18d77bSChristoph Hellwig idx = iter.bi_idx; 41ff18d77bSChristoph Hellwig 42ff18d77bSChristoph Hellwig *bv = bio->bi_io_vec[idx]; 43ff18d77bSChristoph Hellwig 44ff18d77bSChristoph Hellwig /* 45ff18d77bSChristoph Hellwig * iter.bi_bvec_done records actual length of the last bvec 46ff18d77bSChristoph Hellwig * if this bio ends in the middle of one io vector 47ff18d77bSChristoph Hellwig */ 48ff18d77bSChristoph Hellwig if (iter.bi_bvec_done) 49ff18d77bSChristoph Hellwig bv->bv_len = iter.bi_bvec_done; 50ff18d77bSChristoph Hellwig } 51ff18d77bSChristoph Hellwig 52e9907009SChristoph Hellwig static inline bool bio_will_gap(struct request_queue *q, 53e9907009SChristoph Hellwig struct request *prev_rq, struct bio *prev, struct bio *next) 54e9907009SChristoph Hellwig { 55e9907009SChristoph Hellwig struct bio_vec pb, nb; 56e9907009SChristoph Hellwig 57e9907009SChristoph Hellwig if (!bio_has_data(prev) || !queue_virt_boundary(q)) 58e9907009SChristoph Hellwig return false; 59e9907009SChristoph Hellwig 60e9907009SChristoph Hellwig /* 61e9907009SChristoph Hellwig * Don't merge if the 1st bio starts with non-zero offset, otherwise it 62e9907009SChristoph Hellwig * is quite difficult to respect the sg gap limit. We work hard to 63e9907009SChristoph Hellwig * merge a huge number of small single bios in case of mkfs. 64e9907009SChristoph Hellwig */ 65e9907009SChristoph Hellwig if (prev_rq) 66e9907009SChristoph Hellwig bio_get_first_bvec(prev_rq->bio, &pb); 67e9907009SChristoph Hellwig else 68e9907009SChristoph Hellwig bio_get_first_bvec(prev, &pb); 69df376b2eSJohannes Thumshirn if (pb.bv_offset & queue_virt_boundary(q)) 70e9907009SChristoph Hellwig return true; 71e9907009SChristoph Hellwig 72e9907009SChristoph Hellwig /* 73e9907009SChristoph Hellwig * We don't need to worry about the situation that the merged segment 74e9907009SChristoph Hellwig * ends in unaligned virt boundary: 75e9907009SChristoph Hellwig * 76e9907009SChristoph Hellwig * - if 'pb' ends aligned, the merged segment ends aligned 77e9907009SChristoph Hellwig * - if 'pb' ends unaligned, the next bio must include 78e9907009SChristoph Hellwig * one single bvec of 'nb', otherwise the 'nb' can't 79e9907009SChristoph Hellwig * merge with 'pb' 80e9907009SChristoph Hellwig */ 81e9907009SChristoph Hellwig bio_get_last_bvec(prev, &pb); 82e9907009SChristoph Hellwig bio_get_first_bvec(next, &nb); 83200a9affSChristoph Hellwig if (biovec_phys_mergeable(q, &pb, &nb)) 84e9907009SChristoph Hellwig return false; 85e9907009SChristoph Hellwig return __bvec_gap_to_prev(q, &pb, nb.bv_offset); 86e9907009SChristoph Hellwig } 87e9907009SChristoph Hellwig 88e9907009SChristoph Hellwig static inline bool req_gap_back_merge(struct request *req, struct bio *bio) 89e9907009SChristoph Hellwig { 90e9907009SChristoph Hellwig return bio_will_gap(req->q, req, req->biotail, bio); 91e9907009SChristoph Hellwig } 92e9907009SChristoph Hellwig 93e9907009SChristoph Hellwig static inline bool req_gap_front_merge(struct request *req, struct bio *bio) 94e9907009SChristoph Hellwig { 95e9907009SChristoph Hellwig return bio_will_gap(req->q, NULL, bio, req->bio); 96e9907009SChristoph Hellwig } 97e9907009SChristoph Hellwig 9854efd50bSKent Overstreet static struct bio *blk_bio_discard_split(struct request_queue *q, 9954efd50bSKent Overstreet struct bio *bio, 100bdced438SMing Lei struct bio_set *bs, 101bdced438SMing Lei unsigned *nsegs) 10254efd50bSKent Overstreet { 10354efd50bSKent Overstreet unsigned int max_discard_sectors, granularity; 10454efd50bSKent Overstreet int alignment; 10554efd50bSKent Overstreet sector_t tmp; 10654efd50bSKent Overstreet unsigned split_sectors; 10754efd50bSKent Overstreet 108bdced438SMing Lei *nsegs = 1; 109bdced438SMing Lei 11054efd50bSKent Overstreet /* Zero-sector (unknown) and one-sector granularities are the same. */ 11154efd50bSKent Overstreet granularity = max(q->limits.discard_granularity >> 9, 1U); 11254efd50bSKent Overstreet 1131adfc5e4SMing Lei max_discard_sectors = min(q->limits.max_discard_sectors, 1141adfc5e4SMing Lei bio_allowed_max_sectors(q)); 11554efd50bSKent Overstreet max_discard_sectors -= max_discard_sectors % granularity; 11654efd50bSKent Overstreet 11754efd50bSKent Overstreet if (unlikely(!max_discard_sectors)) { 11854efd50bSKent Overstreet /* XXX: warn */ 11954efd50bSKent Overstreet return NULL; 12054efd50bSKent Overstreet } 12154efd50bSKent Overstreet 12254efd50bSKent Overstreet if (bio_sectors(bio) <= max_discard_sectors) 12354efd50bSKent Overstreet return NULL; 12454efd50bSKent Overstreet 12554efd50bSKent Overstreet split_sectors = max_discard_sectors; 12654efd50bSKent Overstreet 12754efd50bSKent Overstreet /* 12854efd50bSKent Overstreet * If the next starting sector would be misaligned, stop the discard at 12954efd50bSKent Overstreet * the previous aligned sector. 13054efd50bSKent Overstreet */ 13154efd50bSKent Overstreet alignment = (q->limits.discard_alignment >> 9) % granularity; 13254efd50bSKent Overstreet 13354efd50bSKent Overstreet tmp = bio->bi_iter.bi_sector + split_sectors - alignment; 13454efd50bSKent Overstreet tmp = sector_div(tmp, granularity); 13554efd50bSKent Overstreet 13654efd50bSKent Overstreet if (split_sectors > tmp) 13754efd50bSKent Overstreet split_sectors -= tmp; 13854efd50bSKent Overstreet 13954efd50bSKent Overstreet return bio_split(bio, split_sectors, GFP_NOIO, bs); 14054efd50bSKent Overstreet } 14154efd50bSKent Overstreet 142885fa13fSChristoph Hellwig static struct bio *blk_bio_write_zeroes_split(struct request_queue *q, 143885fa13fSChristoph Hellwig struct bio *bio, struct bio_set *bs, unsigned *nsegs) 144885fa13fSChristoph Hellwig { 145d665e12aSChristoph Hellwig *nsegs = 0; 146885fa13fSChristoph Hellwig 147885fa13fSChristoph Hellwig if (!q->limits.max_write_zeroes_sectors) 148885fa13fSChristoph Hellwig return NULL; 149885fa13fSChristoph Hellwig 150885fa13fSChristoph Hellwig if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors) 151885fa13fSChristoph Hellwig return NULL; 152885fa13fSChristoph Hellwig 153885fa13fSChristoph Hellwig return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs); 154885fa13fSChristoph Hellwig } 155885fa13fSChristoph Hellwig 1569cc5169cSBart Van Assche /* 1579cc5169cSBart Van Assche * Return the maximum number of sectors from the start of a bio that may be 1589cc5169cSBart Van Assche * submitted as a single request to a block device. If enough sectors remain, 1599cc5169cSBart Van Assche * align the end to the physical block size. Otherwise align the end to the 1609cc5169cSBart Van Assche * logical block size. This approach minimizes the number of non-aligned 1619cc5169cSBart Van Assche * requests that are submitted to a block device if the start of a bio is not 1629cc5169cSBart Van Assche * aligned to a physical block boundary. 1639cc5169cSBart Van Assche */ 164d0e5fbb0SMing Lei static inline unsigned get_max_io_size(struct request_queue *q, 165d0e5fbb0SMing Lei struct bio *bio) 166d0e5fbb0SMing Lei { 1679cc5169cSBart Van Assche unsigned pbs = queue_physical_block_size(q) >> SECTOR_SHIFT; 1689cc5169cSBart Van Assche unsigned lbs = queue_logical_block_size(q) >> SECTOR_SHIFT; 169efef739dSChristoph Hellwig unsigned max_sectors = queue_max_sectors(q), start, end; 170d0e5fbb0SMing Lei 171efef739dSChristoph Hellwig if (q->limits.chunk_sectors) { 172efef739dSChristoph Hellwig max_sectors = min(max_sectors, 173efef739dSChristoph Hellwig blk_chunk_sectors_left(bio->bi_iter.bi_sector, 174efef739dSChristoph Hellwig q->limits.chunk_sectors)); 175efef739dSChristoph Hellwig } 176efef739dSChristoph Hellwig 17784613bedSChristoph Hellwig start = bio->bi_iter.bi_sector & (pbs - 1); 17884613bedSChristoph Hellwig end = (start + max_sectors) & ~(pbs - 1); 17984613bedSChristoph Hellwig if (end > start) 18084613bedSChristoph Hellwig return end - start; 18184613bedSChristoph Hellwig return max_sectors & ~(lbs - 1); 182d0e5fbb0SMing Lei } 183d0e5fbb0SMing Lei 184429120f3SMing Lei static inline unsigned get_max_segment_size(const struct request_queue *q, 185429120f3SMing Lei struct page *start_page, 186429120f3SMing Lei unsigned long offset) 187dcebd755SMing Lei { 188dcebd755SMing Lei unsigned long mask = queue_segment_boundary(q); 189dcebd755SMing Lei 190429120f3SMing Lei offset = mask & (page_to_phys(start_page) + offset); 1914a2f704eSMing Lei 1924a2f704eSMing Lei /* 1934a2f704eSMing Lei * overflow may be triggered in case of zero page physical address 1944a2f704eSMing Lei * on 32bit arch, use queue's max segment size when that happens. 1954a2f704eSMing Lei */ 1964a2f704eSMing Lei return min_not_zero(mask - offset + 1, 1974a2f704eSMing Lei (unsigned long)queue_max_segment_size(q)); 198dcebd755SMing Lei } 199dcebd755SMing Lei 200708b25b3SBart Van Assche /** 201708b25b3SBart Van Assche * bvec_split_segs - verify whether or not a bvec should be split in the middle 202708b25b3SBart Van Assche * @q: [in] request queue associated with the bio associated with @bv 203708b25b3SBart Van Assche * @bv: [in] bvec to examine 204708b25b3SBart Van Assche * @nsegs: [in,out] Number of segments in the bio being built. Incremented 205708b25b3SBart Van Assche * by the number of segments from @bv that may be appended to that 206708b25b3SBart Van Assche * bio without exceeding @max_segs 20767927d22SKeith Busch * @bytes: [in,out] Number of bytes in the bio being built. Incremented 20867927d22SKeith Busch * by the number of bytes from @bv that may be appended to that 20967927d22SKeith Busch * bio without exceeding @max_bytes 210708b25b3SBart Van Assche * @max_segs: [in] upper bound for *@nsegs 21167927d22SKeith Busch * @max_bytes: [in] upper bound for *@bytes 212708b25b3SBart Van Assche * 213708b25b3SBart Van Assche * When splitting a bio, it can happen that a bvec is encountered that is too 214708b25b3SBart Van Assche * big to fit in a single segment and hence that it has to be split in the 215708b25b3SBart Van Assche * middle. This function verifies whether or not that should happen. The value 216708b25b3SBart Van Assche * %true is returned if and only if appending the entire @bv to a bio with 217708b25b3SBart Van Assche * *@nsegs segments and *@sectors sectors would make that bio unacceptable for 218708b25b3SBart Van Assche * the block driver. 219dcebd755SMing Lei */ 220af2c68feSBart Van Assche static bool bvec_split_segs(const struct request_queue *q, 221af2c68feSBart Van Assche const struct bio_vec *bv, unsigned *nsegs, 22267927d22SKeith Busch unsigned *bytes, unsigned max_segs, 22367927d22SKeith Busch unsigned max_bytes) 224dcebd755SMing Lei { 22567927d22SKeith Busch unsigned max_len = min(max_bytes, UINT_MAX) - *bytes; 226708b25b3SBart Van Assche unsigned len = min(bv->bv_len, max_len); 227dcebd755SMing Lei unsigned total_len = 0; 228ff9811b3SBart Van Assche unsigned seg_size = 0; 229dcebd755SMing Lei 230ff9811b3SBart Van Assche while (len && *nsegs < max_segs) { 231429120f3SMing Lei seg_size = get_max_segment_size(q, bv->bv_page, 232429120f3SMing Lei bv->bv_offset + total_len); 233dcebd755SMing Lei seg_size = min(seg_size, len); 234dcebd755SMing Lei 235ff9811b3SBart Van Assche (*nsegs)++; 236dcebd755SMing Lei total_len += seg_size; 237dcebd755SMing Lei len -= seg_size; 238dcebd755SMing Lei 239dcebd755SMing Lei if ((bv->bv_offset + total_len) & queue_virt_boundary(q)) 240dcebd755SMing Lei break; 241dcebd755SMing Lei } 242dcebd755SMing Lei 24367927d22SKeith Busch *bytes += total_len; 244dcebd755SMing Lei 245708b25b3SBart Van Assche /* tell the caller to split the bvec if it is too big to fit */ 246708b25b3SBart Van Assche return len > 0 || bv->bv_len > max_len; 247dcebd755SMing Lei } 248dcebd755SMing Lei 249dad77584SBart Van Assche /** 250dad77584SBart Van Assche * blk_bio_segment_split - split a bio in two bios 251dad77584SBart Van Assche * @q: [in] request queue pointer 252dad77584SBart Van Assche * @bio: [in] bio to be split 253dad77584SBart Van Assche * @bs: [in] bio set to allocate the clone from 254dad77584SBart Van Assche * @segs: [out] number of segments in the bio with the first half of the sectors 255dad77584SBart Van Assche * 256dad77584SBart Van Assche * Clone @bio, update the bi_iter of the clone to represent the first sectors 257dad77584SBart Van Assche * of @bio and update @bio->bi_iter to represent the remaining sectors. The 258dad77584SBart Van Assche * following is guaranteed for the cloned bio: 259dad77584SBart Van Assche * - That it has at most get_max_io_size(@q, @bio) sectors. 260dad77584SBart Van Assche * - That it has at most queue_max_segments(@q) segments. 261dad77584SBart Van Assche * 262dad77584SBart Van Assche * Except for discard requests the cloned bio will point at the bi_io_vec of 263dad77584SBart Van Assche * the original bio. It is the responsibility of the caller to ensure that the 264dad77584SBart Van Assche * original bio is not freed before the cloned bio. The caller is also 265dad77584SBart Van Assche * responsible for ensuring that @bs is only destroyed after processing of the 266dad77584SBart Van Assche * split bio has finished. 267dad77584SBart Van Assche */ 26854efd50bSKent Overstreet static struct bio *blk_bio_segment_split(struct request_queue *q, 26954efd50bSKent Overstreet struct bio *bio, 270bdced438SMing Lei struct bio_set *bs, 271bdced438SMing Lei unsigned *segs) 27254efd50bSKent Overstreet { 2735014c311SJens Axboe struct bio_vec bv, bvprv, *bvprvp = NULL; 27454efd50bSKent Overstreet struct bvec_iter iter; 27567927d22SKeith Busch unsigned nsegs = 0, bytes = 0; 27667927d22SKeith Busch const unsigned max_bytes = get_max_io_size(q, bio) << 9; 27705b700baSMing Lei const unsigned max_segs = queue_max_segments(q); 27854efd50bSKent Overstreet 279dcebd755SMing Lei bio_for_each_bvec(bv, bio, iter) { 28054efd50bSKent Overstreet /* 28154efd50bSKent Overstreet * If the queue doesn't support SG gaps and adding this 28254efd50bSKent Overstreet * offset would create a gap, disallow it. 28354efd50bSKent Overstreet */ 2845014c311SJens Axboe if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset)) 28554efd50bSKent Overstreet goto split; 28654efd50bSKent Overstreet 28705b700baSMing Lei if (nsegs < max_segs && 28867927d22SKeith Busch bytes + bv.bv_len <= max_bytes && 289708b25b3SBart Van Assche bv.bv_offset + bv.bv_len <= PAGE_SIZE) { 290708b25b3SBart Van Assche nsegs++; 29167927d22SKeith Busch bytes += bv.bv_len; 29267927d22SKeith Busch } else if (bvec_split_segs(q, &bv, &nsegs, &bytes, max_segs, 29367927d22SKeith Busch max_bytes)) { 294e36f6204SKeith Busch goto split; 295e36f6204SKeith Busch } 296e36f6204SKeith Busch 29754efd50bSKent Overstreet bvprv = bv; 298578270bfSMing Lei bvprvp = &bvprv; 29954efd50bSKent Overstreet } 30054efd50bSKent Overstreet 301d627065dSChristoph Hellwig *segs = nsegs; 302d627065dSChristoph Hellwig return NULL; 30354efd50bSKent Overstreet split: 304bdced438SMing Lei *segs = nsegs; 305cc29e1bfSJeffle Xu 306cc29e1bfSJeffle Xu /* 30767927d22SKeith Busch * Individual bvecs might not be logical block aligned. Round down the 30867927d22SKeith Busch * split size so that each bio is properly block size aligned, even if 30967927d22SKeith Busch * we do not use the full hardware limits. 31067927d22SKeith Busch */ 31167927d22SKeith Busch bytes = ALIGN_DOWN(bytes, queue_logical_block_size(q)); 31267927d22SKeith Busch 31367927d22SKeith Busch /* 314cc29e1bfSJeffle Xu * Bio splitting may cause subtle trouble such as hang when doing sync 315cc29e1bfSJeffle Xu * iopoll in direct IO routine. Given performance gain of iopoll for 316cc29e1bfSJeffle Xu * big IO can be trival, disable iopoll when split needed. 317cc29e1bfSJeffle Xu */ 3186ce913feSChristoph Hellwig bio_clear_polled(bio); 31967927d22SKeith Busch return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs); 32054efd50bSKent Overstreet } 32154efd50bSKent Overstreet 322dad77584SBart Van Assche /** 323dad77584SBart Van Assche * __blk_queue_split - split a bio and submit the second half 324abd45c15SJens Axboe * @q: [in] request_queue new bio is being queued at 325dad77584SBart Van Assche * @bio: [in, out] bio to be split 326dad77584SBart Van Assche * @nr_segs: [out] number of segments in the first bio 327dad77584SBart Van Assche * 328dad77584SBart Van Assche * Split a bio into two bios, chain the two bios, submit the second half and 329dad77584SBart Van Assche * store a pointer to the first half in *@bio. If the second bio is still too 330dad77584SBart Van Assche * big it will be split by a recursive call to this function. Since this 331309dca30SChristoph Hellwig * function may allocate a new bio from q->bio_split, it is the responsibility 332309dca30SChristoph Hellwig * of the caller to ensure that q->bio_split is only released after processing 333309dca30SChristoph Hellwig * of the split bio has finished. 334dad77584SBart Van Assche */ 335abd45c15SJens Axboe void __blk_queue_split(struct request_queue *q, struct bio **bio, 336abd45c15SJens Axboe unsigned int *nr_segs) 33754efd50bSKent Overstreet { 338fa532287SChristoph Hellwig struct bio *split = NULL; 33954efd50bSKent Overstreet 3407afafc8aSAdrian Hunter switch (bio_op(*bio)) { 3417afafc8aSAdrian Hunter case REQ_OP_DISCARD: 3427afafc8aSAdrian Hunter case REQ_OP_SECURE_ERASE: 34314ccb66bSChristoph Hellwig split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs); 3447afafc8aSAdrian Hunter break; 345a6f0788eSChaitanya Kulkarni case REQ_OP_WRITE_ZEROES: 34614ccb66bSChristoph Hellwig split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split, 34714ccb66bSChristoph Hellwig nr_segs); 348a6f0788eSChaitanya Kulkarni break; 3497afafc8aSAdrian Hunter default: 35014ccb66bSChristoph Hellwig split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs); 3517afafc8aSAdrian Hunter break; 3527afafc8aSAdrian Hunter } 353bdced438SMing Lei 35454efd50bSKent Overstreet if (split) { 3556ac45aebSMing Lei /* there isn't chance to merge the splitted bio */ 3561eff9d32SJens Axboe split->bi_opf |= REQ_NOMERGE; 3576ac45aebSMing Lei 35854efd50bSKent Overstreet bio_chain(split, *bio); 359eb6f7f7cSChristoph Hellwig trace_block_split(split, (*bio)->bi_iter.bi_sector); 360ed00aabdSChristoph Hellwig submit_bio_noacct(*bio); 36154efd50bSKent Overstreet *bio = split; 36254efd50bSKent Overstreet } 36354efd50bSKent Overstreet } 36414ccb66bSChristoph Hellwig 365dad77584SBart Van Assche /** 366dad77584SBart Van Assche * blk_queue_split - split a bio and submit the second half 367dad77584SBart Van Assche * @bio: [in, out] bio to be split 368dad77584SBart Van Assche * 369dad77584SBart Van Assche * Split a bio into two bios, chains the two bios, submit the second half and 370dad77584SBart Van Assche * store a pointer to the first half in *@bio. Since this function may allocate 371309dca30SChristoph Hellwig * a new bio from q->bio_split, it is the responsibility of the caller to ensure 372309dca30SChristoph Hellwig * that q->bio_split is only released after processing of the split bio has 373309dca30SChristoph Hellwig * finished. 374dad77584SBart Van Assche */ 375f695ca38SChristoph Hellwig void blk_queue_split(struct bio **bio) 37614ccb66bSChristoph Hellwig { 377859897c3SPavel Begunkov struct request_queue *q = bdev_get_queue((*bio)->bi_bdev); 37814ccb66bSChristoph Hellwig unsigned int nr_segs; 37914ccb66bSChristoph Hellwig 380abd45c15SJens Axboe if (blk_may_split(q, *bio)) 381abd45c15SJens Axboe __blk_queue_split(q, bio, &nr_segs); 38214ccb66bSChristoph Hellwig } 38354efd50bSKent Overstreet EXPORT_SYMBOL(blk_queue_split); 38454efd50bSKent Overstreet 385e9cd19c0SChristoph Hellwig unsigned int blk_recalc_rq_segments(struct request *rq) 386d6d48196SJens Axboe { 3876869875fSChristoph Hellwig unsigned int nr_phys_segs = 0; 38867927d22SKeith Busch unsigned int bytes = 0; 389e9cd19c0SChristoph Hellwig struct req_iterator iter; 3906869875fSChristoph Hellwig struct bio_vec bv; 391d6d48196SJens Axboe 392e9cd19c0SChristoph Hellwig if (!rq->bio) 3931e428079SJens Axboe return 0; 394d6d48196SJens Axboe 395e9cd19c0SChristoph Hellwig switch (bio_op(rq->bio)) { 396a6f0788eSChaitanya Kulkarni case REQ_OP_DISCARD: 397a6f0788eSChaitanya Kulkarni case REQ_OP_SECURE_ERASE: 398a958937fSDavid Jeffery if (queue_max_discard_segments(rq->q) > 1) { 399a958937fSDavid Jeffery struct bio *bio = rq->bio; 400a958937fSDavid Jeffery 401a958937fSDavid Jeffery for_each_bio(bio) 402a958937fSDavid Jeffery nr_phys_segs++; 403a958937fSDavid Jeffery return nr_phys_segs; 404a958937fSDavid Jeffery } 405a958937fSDavid Jeffery return 1; 406a6f0788eSChaitanya Kulkarni case REQ_OP_WRITE_ZEROES: 407f9d03f96SChristoph Hellwig return 0; 4082d9b02beSBart Van Assche default: 4092d9b02beSBart Van Assche break; 410a6f0788eSChaitanya Kulkarni } 4115cb8850cSKent Overstreet 412e9cd19c0SChristoph Hellwig rq_for_each_bvec(bv, rq, iter) 41367927d22SKeith Busch bvec_split_segs(rq->q, &bv, &nr_phys_segs, &bytes, 414708b25b3SBart Van Assche UINT_MAX, UINT_MAX); 4151e428079SJens Axboe return nr_phys_segs; 4161e428079SJens Axboe } 4171e428079SJens Axboe 41848d7727cSMing Lei static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, 419862e5a5eSMing Lei struct scatterlist *sglist) 420862e5a5eSMing Lei { 421862e5a5eSMing Lei if (!*sg) 422862e5a5eSMing Lei return sglist; 423862e5a5eSMing Lei 424862e5a5eSMing Lei /* 425862e5a5eSMing Lei * If the driver previously mapped a shorter list, we could see a 426862e5a5eSMing Lei * termination bit prematurely unless it fully inits the sg table 427862e5a5eSMing Lei * on each mapping. We KNOW that there must be more entries here 428862e5a5eSMing Lei * or the driver would be buggy, so force clear the termination bit 429862e5a5eSMing Lei * to avoid doing a full sg_init_table() in drivers for each command. 430862e5a5eSMing Lei */ 431862e5a5eSMing Lei sg_unmark_end(*sg); 432862e5a5eSMing Lei return sg_next(*sg); 433862e5a5eSMing Lei } 434862e5a5eSMing Lei 435862e5a5eSMing Lei static unsigned blk_bvec_map_sg(struct request_queue *q, 436862e5a5eSMing Lei struct bio_vec *bvec, struct scatterlist *sglist, 437862e5a5eSMing Lei struct scatterlist **sg) 438862e5a5eSMing Lei { 439862e5a5eSMing Lei unsigned nbytes = bvec->bv_len; 4408a96a0e4SChristoph Hellwig unsigned nsegs = 0, total = 0; 441862e5a5eSMing Lei 442862e5a5eSMing Lei while (nbytes > 0) { 4438a96a0e4SChristoph Hellwig unsigned offset = bvec->bv_offset + total; 444429120f3SMing Lei unsigned len = min(get_max_segment_size(q, bvec->bv_page, 445429120f3SMing Lei offset), nbytes); 446f9f76879SChristoph Hellwig struct page *page = bvec->bv_page; 447f9f76879SChristoph Hellwig 448f9f76879SChristoph Hellwig /* 449f9f76879SChristoph Hellwig * Unfortunately a fair number of drivers barf on scatterlists 450f9f76879SChristoph Hellwig * that have an offset larger than PAGE_SIZE, despite other 451f9f76879SChristoph Hellwig * subsystems dealing with that invariant just fine. For now 452f9f76879SChristoph Hellwig * stick to the legacy format where we never present those from 453f9f76879SChristoph Hellwig * the block layer, but the code below should be removed once 454f9f76879SChristoph Hellwig * these offenders (mostly MMC/SD drivers) are fixed. 455f9f76879SChristoph Hellwig */ 456f9f76879SChristoph Hellwig page += (offset >> PAGE_SHIFT); 457f9f76879SChristoph Hellwig offset &= ~PAGE_MASK; 458862e5a5eSMing Lei 459862e5a5eSMing Lei *sg = blk_next_sg(sg, sglist); 460f9f76879SChristoph Hellwig sg_set_page(*sg, page, len, offset); 461862e5a5eSMing Lei 4628a96a0e4SChristoph Hellwig total += len; 4638a96a0e4SChristoph Hellwig nbytes -= len; 464862e5a5eSMing Lei nsegs++; 465862e5a5eSMing Lei } 466862e5a5eSMing Lei 467862e5a5eSMing Lei return nsegs; 468862e5a5eSMing Lei } 469862e5a5eSMing Lei 47016e3e418SMing Lei static inline int __blk_bvec_map_sg(struct bio_vec bv, 47116e3e418SMing Lei struct scatterlist *sglist, struct scatterlist **sg) 47216e3e418SMing Lei { 47316e3e418SMing Lei *sg = blk_next_sg(sg, sglist); 47416e3e418SMing Lei sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); 47516e3e418SMing Lei return 1; 47616e3e418SMing Lei } 47716e3e418SMing Lei 478f6970f83SMing Lei /* only try to merge bvecs into one sg if they are from two bios */ 479f6970f83SMing Lei static inline bool 480f6970f83SMing Lei __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, 481f6970f83SMing Lei struct bio_vec *bvprv, struct scatterlist **sg) 482963ab9e5SAsias He { 483963ab9e5SAsias He 484963ab9e5SAsias He int nbytes = bvec->bv_len; 485963ab9e5SAsias He 486f6970f83SMing Lei if (!*sg) 487f6970f83SMing Lei return false; 488f6970f83SMing Lei 489b4b6cb61SMing Lei if ((*sg)->length + nbytes > queue_max_segment_size(q)) 490f6970f83SMing Lei return false; 491f6970f83SMing Lei 4923dccdae5SChristoph Hellwig if (!biovec_phys_mergeable(q, bvprv, bvec)) 493f6970f83SMing Lei return false; 494963ab9e5SAsias He 495963ab9e5SAsias He (*sg)->length += nbytes; 496f6970f83SMing Lei 497f6970f83SMing Lei return true; 498963ab9e5SAsias He } 499963ab9e5SAsias He 5005cb8850cSKent Overstreet static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, 5015cb8850cSKent Overstreet struct scatterlist *sglist, 5025cb8850cSKent Overstreet struct scatterlist **sg) 5035cb8850cSKent Overstreet { 5043f649ab7SKees Cook struct bio_vec bvec, bvprv = { NULL }; 5055cb8850cSKent Overstreet struct bvec_iter iter; 50638417468SChristoph Hellwig int nsegs = 0; 507f6970f83SMing Lei bool new_bio = false; 5085cb8850cSKent Overstreet 509f6970f83SMing Lei for_each_bio(bio) { 510f6970f83SMing Lei bio_for_each_bvec(bvec, bio, iter) { 511f6970f83SMing Lei /* 512f6970f83SMing Lei * Only try to merge bvecs from two bios given we 513f6970f83SMing Lei * have done bio internal merge when adding pages 514f6970f83SMing Lei * to bio 515f6970f83SMing Lei */ 516f6970f83SMing Lei if (new_bio && 517f6970f83SMing Lei __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) 518f6970f83SMing Lei goto next_bvec; 519f6970f83SMing Lei 520f6970f83SMing Lei if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) 521f6970f83SMing Lei nsegs += __blk_bvec_map_sg(bvec, sglist, sg); 522f6970f83SMing Lei else 523f6970f83SMing Lei nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); 524f6970f83SMing Lei next_bvec: 525f6970f83SMing Lei new_bio = false; 526f6970f83SMing Lei } 527b21e11c5SMing Lei if (likely(bio->bi_iter.bi_size)) { 528f6970f83SMing Lei bvprv = bvec; 529f6970f83SMing Lei new_bio = true; 530f6970f83SMing Lei } 531b21e11c5SMing Lei } 5325cb8850cSKent Overstreet 5335cb8850cSKent Overstreet return nsegs; 5345cb8850cSKent Overstreet } 5355cb8850cSKent Overstreet 536d6d48196SJens Axboe /* 537d6d48196SJens Axboe * map a request to scatterlist, return number of sg entries setup. Caller 538d6d48196SJens Axboe * must make sure sg can hold rq->nr_phys_segments entries 539d6d48196SJens Axboe */ 54089de1504SChristoph Hellwig int __blk_rq_map_sg(struct request_queue *q, struct request *rq, 54189de1504SChristoph Hellwig struct scatterlist *sglist, struct scatterlist **last_sg) 542d6d48196SJens Axboe { 5435cb8850cSKent Overstreet int nsegs = 0; 544d6d48196SJens Axboe 545f9d03f96SChristoph Hellwig if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 54689de1504SChristoph Hellwig nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg); 547f9d03f96SChristoph Hellwig else if (rq->bio) 54889de1504SChristoph Hellwig nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg); 549f18573abSFUJITA Tomonori 55089de1504SChristoph Hellwig if (*last_sg) 55189de1504SChristoph Hellwig sg_mark_end(*last_sg); 552d6d48196SJens Axboe 55312e57f59SMing Lei /* 55412e57f59SMing Lei * Something must have been wrong if the figured number of 55512e57f59SMing Lei * segment is bigger than number of req's physical segments 55612e57f59SMing Lei */ 557f9d03f96SChristoph Hellwig WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); 55812e57f59SMing Lei 559d6d48196SJens Axboe return nsegs; 560d6d48196SJens Axboe } 56189de1504SChristoph Hellwig EXPORT_SYMBOL(__blk_rq_map_sg); 562d6d48196SJens Axboe 563943b40c8SMing Lei static inline unsigned int blk_rq_get_max_segments(struct request *rq) 564943b40c8SMing Lei { 565943b40c8SMing Lei if (req_op(rq) == REQ_OP_DISCARD) 566943b40c8SMing Lei return queue_max_discard_segments(rq->q); 567943b40c8SMing Lei return queue_max_segments(rq->q); 568943b40c8SMing Lei } 569943b40c8SMing Lei 570badf7f64SChristoph Hellwig static inline unsigned int blk_rq_get_max_sectors(struct request *rq, 571badf7f64SChristoph Hellwig sector_t offset) 572badf7f64SChristoph Hellwig { 573badf7f64SChristoph Hellwig struct request_queue *q = rq->q; 574c8875190SChristoph Hellwig unsigned int max_sectors; 575badf7f64SChristoph Hellwig 576badf7f64SChristoph Hellwig if (blk_rq_is_passthrough(rq)) 577badf7f64SChristoph Hellwig return q->limits.max_hw_sectors; 578badf7f64SChristoph Hellwig 579c8875190SChristoph Hellwig max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); 580badf7f64SChristoph Hellwig if (!q->limits.chunk_sectors || 581badf7f64SChristoph Hellwig req_op(rq) == REQ_OP_DISCARD || 582badf7f64SChristoph Hellwig req_op(rq) == REQ_OP_SECURE_ERASE) 583c8875190SChristoph Hellwig return max_sectors; 584c8875190SChristoph Hellwig return min(max_sectors, 585c8875190SChristoph Hellwig blk_chunk_sectors_left(offset, q->limits.chunk_sectors)); 586badf7f64SChristoph Hellwig } 587badf7f64SChristoph Hellwig 58814ccb66bSChristoph Hellwig static inline int ll_new_hw_segment(struct request *req, struct bio *bio, 58914ccb66bSChristoph Hellwig unsigned int nr_phys_segs) 590d6d48196SJens Axboe { 5916b2b0459STejun Heo if (!blk_cgroup_mergeable(req, bio)) 5926b2b0459STejun Heo goto no_merge; 5936b2b0459STejun Heo 5942705dfb2SMing Lei if (blk_integrity_merge_bio(req->q, req, bio) == false) 59513f05c8dSMartin K. Petersen goto no_merge; 59613f05c8dSMartin K. Petersen 5972705dfb2SMing Lei /* discard request merge won't add new segment */ 5982705dfb2SMing Lei if (req_op(req) == REQ_OP_DISCARD) 5992705dfb2SMing Lei return 1; 6002705dfb2SMing Lei 6012705dfb2SMing Lei if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req)) 60213f05c8dSMartin K. Petersen goto no_merge; 603d6d48196SJens Axboe 604d6d48196SJens Axboe /* 605d6d48196SJens Axboe * This will form the start of a new hw segment. Bump both 606d6d48196SJens Axboe * counters. 607d6d48196SJens Axboe */ 608d6d48196SJens Axboe req->nr_phys_segments += nr_phys_segs; 609d6d48196SJens Axboe return 1; 61013f05c8dSMartin K. Petersen 61113f05c8dSMartin K. Petersen no_merge: 61214ccb66bSChristoph Hellwig req_set_nomerge(req->q, req); 61313f05c8dSMartin K. Petersen return 0; 614d6d48196SJens Axboe } 615d6d48196SJens Axboe 61614ccb66bSChristoph Hellwig int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) 617d6d48196SJens Axboe { 6185e7c4274SJens Axboe if (req_gap_back_merge(req, bio)) 6195e7c4274SJens Axboe return 0; 6207f39add3SSagi Grimberg if (blk_integrity_rq(req) && 6217f39add3SSagi Grimberg integrity_req_gap_back_merge(req, bio)) 6227f39add3SSagi Grimberg return 0; 623a892c8d5SSatya Tangirala if (!bio_crypt_ctx_back_mergeable(req, bio)) 624a892c8d5SSatya Tangirala return 0; 625f31dc1cdSMartin K. Petersen if (blk_rq_sectors(req) + bio_sectors(bio) > 62617007f39SDamien Le Moal blk_rq_get_max_sectors(req, blk_rq_pos(req))) { 62714ccb66bSChristoph Hellwig req_set_nomerge(req->q, req); 628d6d48196SJens Axboe return 0; 629d6d48196SJens Axboe } 630d6d48196SJens Axboe 63114ccb66bSChristoph Hellwig return ll_new_hw_segment(req, bio, nr_segs); 632d6d48196SJens Axboe } 633d6d48196SJens Axboe 634eda5cc99SChristoph Hellwig static int ll_front_merge_fn(struct request *req, struct bio *bio, 635eda5cc99SChristoph Hellwig unsigned int nr_segs) 636d6d48196SJens Axboe { 6375e7c4274SJens Axboe if (req_gap_front_merge(req, bio)) 6385e7c4274SJens Axboe return 0; 6397f39add3SSagi Grimberg if (blk_integrity_rq(req) && 6407f39add3SSagi Grimberg integrity_req_gap_front_merge(req, bio)) 6417f39add3SSagi Grimberg return 0; 642a892c8d5SSatya Tangirala if (!bio_crypt_ctx_front_mergeable(req, bio)) 643a892c8d5SSatya Tangirala return 0; 644f31dc1cdSMartin K. Petersen if (blk_rq_sectors(req) + bio_sectors(bio) > 64517007f39SDamien Le Moal blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { 64614ccb66bSChristoph Hellwig req_set_nomerge(req->q, req); 647d6d48196SJens Axboe return 0; 648d6d48196SJens Axboe } 649d6d48196SJens Axboe 65014ccb66bSChristoph Hellwig return ll_new_hw_segment(req, bio, nr_segs); 651d6d48196SJens Axboe } 652d6d48196SJens Axboe 653445251d0SJens Axboe static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, 654445251d0SJens Axboe struct request *next) 655445251d0SJens Axboe { 656445251d0SJens Axboe unsigned short segments = blk_rq_nr_discard_segments(req); 657445251d0SJens Axboe 658445251d0SJens Axboe if (segments >= queue_max_discard_segments(q)) 659445251d0SJens Axboe goto no_merge; 660445251d0SJens Axboe if (blk_rq_sectors(req) + bio_sectors(next->bio) > 661445251d0SJens Axboe blk_rq_get_max_sectors(req, blk_rq_pos(req))) 662445251d0SJens Axboe goto no_merge; 663445251d0SJens Axboe 664445251d0SJens Axboe req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); 665445251d0SJens Axboe return true; 666445251d0SJens Axboe no_merge: 667445251d0SJens Axboe req_set_nomerge(q, req); 668445251d0SJens Axboe return false; 669445251d0SJens Axboe } 670445251d0SJens Axboe 671d6d48196SJens Axboe static int ll_merge_requests_fn(struct request_queue *q, struct request *req, 672d6d48196SJens Axboe struct request *next) 673d6d48196SJens Axboe { 674d6d48196SJens Axboe int total_phys_segments; 675d6d48196SJens Axboe 6765e7c4274SJens Axboe if (req_gap_back_merge(req, next->bio)) 677854fbb9cSKeith Busch return 0; 678854fbb9cSKeith Busch 679d6d48196SJens Axboe /* 680d6d48196SJens Axboe * Will it become too large? 681d6d48196SJens Axboe */ 682f31dc1cdSMartin K. Petersen if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > 68317007f39SDamien Le Moal blk_rq_get_max_sectors(req, blk_rq_pos(req))) 684d6d48196SJens Axboe return 0; 685d6d48196SJens Axboe 686d6d48196SJens Axboe total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; 687943b40c8SMing Lei if (total_phys_segments > blk_rq_get_max_segments(req)) 688d6d48196SJens Axboe return 0; 689d6d48196SJens Axboe 6906b2b0459STejun Heo if (!blk_cgroup_mergeable(req, next->bio)) 6916b2b0459STejun Heo return 0; 6926b2b0459STejun Heo 6934eaf99beSMartin K. Petersen if (blk_integrity_merge_rq(q, req, next) == false) 69413f05c8dSMartin K. Petersen return 0; 69513f05c8dSMartin K. Petersen 696a892c8d5SSatya Tangirala if (!bio_crypt_ctx_merge_rq(req, next)) 697a892c8d5SSatya Tangirala return 0; 698a892c8d5SSatya Tangirala 699d6d48196SJens Axboe /* Merge is OK... */ 700d6d48196SJens Axboe req->nr_phys_segments = total_phys_segments; 701d6d48196SJens Axboe return 1; 702d6d48196SJens Axboe } 703d6d48196SJens Axboe 70480a761fdSTejun Heo /** 70580a761fdSTejun Heo * blk_rq_set_mixed_merge - mark a request as mixed merge 70680a761fdSTejun Heo * @rq: request to mark as mixed merge 70780a761fdSTejun Heo * 70880a761fdSTejun Heo * Description: 70980a761fdSTejun Heo * @rq is about to be mixed merged. Make sure the attributes 71080a761fdSTejun Heo * which can be mixed are set in each bio and mark @rq as mixed 71180a761fdSTejun Heo * merged. 71280a761fdSTejun Heo */ 71380a761fdSTejun Heo void blk_rq_set_mixed_merge(struct request *rq) 71480a761fdSTejun Heo { 715*16458cf3SBart Van Assche blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK; 71680a761fdSTejun Heo struct bio *bio; 71780a761fdSTejun Heo 718e8064021SChristoph Hellwig if (rq->rq_flags & RQF_MIXED_MERGE) 71980a761fdSTejun Heo return; 72080a761fdSTejun Heo 72180a761fdSTejun Heo /* 72280a761fdSTejun Heo * @rq will no longer represent mixable attributes for all the 72380a761fdSTejun Heo * contained bios. It will just track those of the first one. 72480a761fdSTejun Heo * Distributes the attributs to each bio. 72580a761fdSTejun Heo */ 72680a761fdSTejun Heo for (bio = rq->bio; bio; bio = bio->bi_next) { 7271eff9d32SJens Axboe WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && 7281eff9d32SJens Axboe (bio->bi_opf & REQ_FAILFAST_MASK) != ff); 7291eff9d32SJens Axboe bio->bi_opf |= ff; 73080a761fdSTejun Heo } 731e8064021SChristoph Hellwig rq->rq_flags |= RQF_MIXED_MERGE; 73280a761fdSTejun Heo } 73380a761fdSTejun Heo 734b9c54f56SKonstantin Khlebnikov static void blk_account_io_merge_request(struct request *req) 73526308eabSJerome Marchand { 73626308eabSJerome Marchand if (blk_do_io_stat(req)) { 737112f158fSMike Snitzer part_stat_lock(); 738b9c54f56SKonstantin Khlebnikov part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); 73926308eabSJerome Marchand part_stat_unlock(); 74026308eabSJerome Marchand } 74126308eabSJerome Marchand } 742b9c54f56SKonstantin Khlebnikov 743e96c0d83SEric Biggers static enum elv_merge blk_try_req_merge(struct request *req, 744e96c0d83SEric Biggers struct request *next) 74569840466SJianchao Wang { 74669840466SJianchao Wang if (blk_discard_mergable(req)) 74769840466SJianchao Wang return ELEVATOR_DISCARD_MERGE; 74869840466SJianchao Wang else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) 74969840466SJianchao Wang return ELEVATOR_BACK_MERGE; 75069840466SJianchao Wang 75169840466SJianchao Wang return ELEVATOR_NO_MERGE; 75269840466SJianchao Wang } 75326308eabSJerome Marchand 754d6d48196SJens Axboe /* 755b973cb7eSJens Axboe * For non-mq, this has to be called with the request spinlock acquired. 756b973cb7eSJens Axboe * For mq with scheduling, the appropriate queue wide lock should be held. 757d6d48196SJens Axboe */ 758b973cb7eSJens Axboe static struct request *attempt_merge(struct request_queue *q, 759b973cb7eSJens Axboe struct request *req, struct request *next) 760d6d48196SJens Axboe { 761d6d48196SJens Axboe if (!rq_mergeable(req) || !rq_mergeable(next)) 762b973cb7eSJens Axboe return NULL; 763d6d48196SJens Axboe 764288dab8aSChristoph Hellwig if (req_op(req) != req_op(next)) 765b973cb7eSJens Axboe return NULL; 766f31dc1cdSMartin K. Petersen 76779bb1dbdSChristoph Hellwig if (rq_data_dir(req) != rq_data_dir(next)) 768b973cb7eSJens Axboe return NULL; 769d6d48196SJens Axboe 770668ffc03SDamien Le Moal if (req->ioprio != next->ioprio) 771668ffc03SDamien Le Moal return NULL; 772668ffc03SDamien Le Moal 773cb6934f8SJens Axboe /* 774d6d48196SJens Axboe * If we are allowed to merge, then append bio list 775d6d48196SJens Axboe * from next to rq and release next. merge_requests_fn 776d6d48196SJens Axboe * will have updated segment counts, update sector 777445251d0SJens Axboe * counts here. Handle DISCARDs separately, as they 778445251d0SJens Axboe * have separate settings. 779d6d48196SJens Axboe */ 78069840466SJianchao Wang 78169840466SJianchao Wang switch (blk_try_req_merge(req, next)) { 78269840466SJianchao Wang case ELEVATOR_DISCARD_MERGE: 783445251d0SJens Axboe if (!req_attempt_discard_merge(q, req, next)) 784445251d0SJens Axboe return NULL; 78569840466SJianchao Wang break; 78669840466SJianchao Wang case ELEVATOR_BACK_MERGE: 78769840466SJianchao Wang if (!ll_merge_requests_fn(q, req, next)) 788b973cb7eSJens Axboe return NULL; 78969840466SJianchao Wang break; 79069840466SJianchao Wang default: 79169840466SJianchao Wang return NULL; 79269840466SJianchao Wang } 793d6d48196SJens Axboe 794d6d48196SJens Axboe /* 79580a761fdSTejun Heo * If failfast settings disagree or any of the two is already 79680a761fdSTejun Heo * a mixed merge, mark both as mixed before proceeding. This 79780a761fdSTejun Heo * makes sure that all involved bios have mixable attributes 79880a761fdSTejun Heo * set properly. 79980a761fdSTejun Heo */ 800e8064021SChristoph Hellwig if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || 80180a761fdSTejun Heo (req->cmd_flags & REQ_FAILFAST_MASK) != 80280a761fdSTejun Heo (next->cmd_flags & REQ_FAILFAST_MASK)) { 80380a761fdSTejun Heo blk_rq_set_mixed_merge(req); 80480a761fdSTejun Heo blk_rq_set_mixed_merge(next); 80580a761fdSTejun Heo } 80680a761fdSTejun Heo 80780a761fdSTejun Heo /* 808522a7775SOmar Sandoval * At this point we have either done a back merge or front merge. We 809522a7775SOmar Sandoval * need the smaller start_time_ns of the merged requests to be the 810522a7775SOmar Sandoval * current request for accounting purposes. 811d6d48196SJens Axboe */ 812522a7775SOmar Sandoval if (next->start_time_ns < req->start_time_ns) 813522a7775SOmar Sandoval req->start_time_ns = next->start_time_ns; 814d6d48196SJens Axboe 815d6d48196SJens Axboe req->biotail->bi_next = next->bio; 816d6d48196SJens Axboe req->biotail = next->biotail; 817d6d48196SJens Axboe 818a2dec7b3STejun Heo req->__data_len += blk_rq_bytes(next); 819d6d48196SJens Axboe 8202a5cf35cSMing Lei if (!blk_discard_mergable(req)) 821d6d48196SJens Axboe elv_merge_requests(q, req, next); 822d6d48196SJens Axboe 82342dad764SJerome Marchand /* 82442dad764SJerome Marchand * 'next' is going away, so update stats accordingly 82542dad764SJerome Marchand */ 826b9c54f56SKonstantin Khlebnikov blk_account_io_merge_request(next); 827d6d48196SJens Axboe 828a54895faSChristoph Hellwig trace_block_rq_merge(next); 829f3bdc62fSJan Kara 830e4d750c9SJens Axboe /* 831e4d750c9SJens Axboe * ownership of bio passed from next to req, return 'next' for 832e4d750c9SJens Axboe * the caller to free 833e4d750c9SJens Axboe */ 8341cd96c24SBoaz Harrosh next->bio = NULL; 835b973cb7eSJens Axboe return next; 836d6d48196SJens Axboe } 837d6d48196SJens Axboe 838eda5cc99SChristoph Hellwig static struct request *attempt_back_merge(struct request_queue *q, 839eda5cc99SChristoph Hellwig struct request *rq) 840d6d48196SJens Axboe { 841d6d48196SJens Axboe struct request *next = elv_latter_request(q, rq); 842d6d48196SJens Axboe 843d6d48196SJens Axboe if (next) 844d6d48196SJens Axboe return attempt_merge(q, rq, next); 845d6d48196SJens Axboe 846b973cb7eSJens Axboe return NULL; 847d6d48196SJens Axboe } 848d6d48196SJens Axboe 849eda5cc99SChristoph Hellwig static struct request *attempt_front_merge(struct request_queue *q, 850eda5cc99SChristoph Hellwig struct request *rq) 851d6d48196SJens Axboe { 852d6d48196SJens Axboe struct request *prev = elv_former_request(q, rq); 853d6d48196SJens Axboe 854d6d48196SJens Axboe if (prev) 855d6d48196SJens Axboe return attempt_merge(q, prev, rq); 856d6d48196SJens Axboe 857b973cb7eSJens Axboe return NULL; 858d6d48196SJens Axboe } 8595e84ea3aSJens Axboe 860fd2ef39cSJan Kara /* 861fd2ef39cSJan Kara * Try to merge 'next' into 'rq'. Return true if the merge happened, false 862fd2ef39cSJan Kara * otherwise. The caller is responsible for freeing 'next' if the merge 863fd2ef39cSJan Kara * happened. 864fd2ef39cSJan Kara */ 865fd2ef39cSJan Kara bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, 8665e84ea3aSJens Axboe struct request *next) 8675e84ea3aSJens Axboe { 868fd2ef39cSJan Kara return attempt_merge(q, rq, next); 8695e84ea3aSJens Axboe } 870050c8ea8STejun Heo 871050c8ea8STejun Heo bool blk_rq_merge_ok(struct request *rq, struct bio *bio) 872050c8ea8STejun Heo { 873e2a60da7SMartin K. Petersen if (!rq_mergeable(rq) || !bio_mergeable(bio)) 874050c8ea8STejun Heo return false; 875050c8ea8STejun Heo 876288dab8aSChristoph Hellwig if (req_op(rq) != bio_op(bio)) 877f31dc1cdSMartin K. Petersen return false; 878f31dc1cdSMartin K. Petersen 879050c8ea8STejun Heo /* different data direction or already started, don't merge */ 880050c8ea8STejun Heo if (bio_data_dir(bio) != rq_data_dir(rq)) 881050c8ea8STejun Heo return false; 882050c8ea8STejun Heo 8836b2b0459STejun Heo /* don't merge across cgroup boundaries */ 8846b2b0459STejun Heo if (!blk_cgroup_mergeable(rq, bio)) 8856b2b0459STejun Heo return false; 8866b2b0459STejun Heo 887050c8ea8STejun Heo /* only merge integrity protected bio into ditto rq */ 8884eaf99beSMartin K. Petersen if (blk_integrity_merge_bio(rq->q, rq, bio) == false) 889050c8ea8STejun Heo return false; 890050c8ea8STejun Heo 891a892c8d5SSatya Tangirala /* Only merge if the crypt contexts are compatible */ 892a892c8d5SSatya Tangirala if (!bio_crypt_rq_ctx_compatible(rq, bio)) 893a892c8d5SSatya Tangirala return false; 894a892c8d5SSatya Tangirala 895668ffc03SDamien Le Moal if (rq->ioprio != bio_prio(bio)) 896668ffc03SDamien Le Moal return false; 897668ffc03SDamien Le Moal 898050c8ea8STejun Heo return true; 899050c8ea8STejun Heo } 900050c8ea8STejun Heo 90134fe7c05SChristoph Hellwig enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) 902050c8ea8STejun Heo { 90369840466SJianchao Wang if (blk_discard_mergable(rq)) 9041e739730SChristoph Hellwig return ELEVATOR_DISCARD_MERGE; 9051e739730SChristoph Hellwig else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) 906050c8ea8STejun Heo return ELEVATOR_BACK_MERGE; 9074f024f37SKent Overstreet else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) 908050c8ea8STejun Heo return ELEVATOR_FRONT_MERGE; 909050c8ea8STejun Heo return ELEVATOR_NO_MERGE; 910050c8ea8STejun Heo } 9118e756373SBaolin Wang 9128e756373SBaolin Wang static void blk_account_io_merge_bio(struct request *req) 9138e756373SBaolin Wang { 9148e756373SBaolin Wang if (!blk_do_io_stat(req)) 9158e756373SBaolin Wang return; 9168e756373SBaolin Wang 9178e756373SBaolin Wang part_stat_lock(); 9188e756373SBaolin Wang part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); 9198e756373SBaolin Wang part_stat_unlock(); 9208e756373SBaolin Wang } 9218e756373SBaolin Wang 922eda5cc99SChristoph Hellwig enum bio_merge_status { 923eda5cc99SChristoph Hellwig BIO_MERGE_OK, 924eda5cc99SChristoph Hellwig BIO_MERGE_NONE, 925eda5cc99SChristoph Hellwig BIO_MERGE_FAILED, 926eda5cc99SChristoph Hellwig }; 927eda5cc99SChristoph Hellwig 928eda5cc99SChristoph Hellwig static enum bio_merge_status bio_attempt_back_merge(struct request *req, 929eda5cc99SChristoph Hellwig struct bio *bio, unsigned int nr_segs) 9308e756373SBaolin Wang { 931*16458cf3SBart Van Assche const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK; 9328e756373SBaolin Wang 9338e756373SBaolin Wang if (!ll_back_merge_fn(req, bio, nr_segs)) 9347d7ca7c5SBaolin Wang return BIO_MERGE_FAILED; 9358e756373SBaolin Wang 936e8a676d6SChristoph Hellwig trace_block_bio_backmerge(bio); 9378e756373SBaolin Wang rq_qos_merge(req->q, req, bio); 9388e756373SBaolin Wang 9398e756373SBaolin Wang if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 9408e756373SBaolin Wang blk_rq_set_mixed_merge(req); 9418e756373SBaolin Wang 9428e756373SBaolin Wang req->biotail->bi_next = bio; 9438e756373SBaolin Wang req->biotail = bio; 9448e756373SBaolin Wang req->__data_len += bio->bi_iter.bi_size; 9458e756373SBaolin Wang 9468e756373SBaolin Wang bio_crypt_free_ctx(bio); 9478e756373SBaolin Wang 9488e756373SBaolin Wang blk_account_io_merge_bio(req); 9497d7ca7c5SBaolin Wang return BIO_MERGE_OK; 9508e756373SBaolin Wang } 9518e756373SBaolin Wang 952eda5cc99SChristoph Hellwig static enum bio_merge_status bio_attempt_front_merge(struct request *req, 953eda5cc99SChristoph Hellwig struct bio *bio, unsigned int nr_segs) 9548e756373SBaolin Wang { 955*16458cf3SBart Van Assche const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK; 9568e756373SBaolin Wang 9578e756373SBaolin Wang if (!ll_front_merge_fn(req, bio, nr_segs)) 9587d7ca7c5SBaolin Wang return BIO_MERGE_FAILED; 9598e756373SBaolin Wang 960e8a676d6SChristoph Hellwig trace_block_bio_frontmerge(bio); 9618e756373SBaolin Wang rq_qos_merge(req->q, req, bio); 9628e756373SBaolin Wang 9638e756373SBaolin Wang if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 9648e756373SBaolin Wang blk_rq_set_mixed_merge(req); 9658e756373SBaolin Wang 9668e756373SBaolin Wang bio->bi_next = req->bio; 9678e756373SBaolin Wang req->bio = bio; 9688e756373SBaolin Wang 9698e756373SBaolin Wang req->__sector = bio->bi_iter.bi_sector; 9708e756373SBaolin Wang req->__data_len += bio->bi_iter.bi_size; 9718e756373SBaolin Wang 9728e756373SBaolin Wang bio_crypt_do_front_merge(req, bio); 9738e756373SBaolin Wang 9748e756373SBaolin Wang blk_account_io_merge_bio(req); 9757d7ca7c5SBaolin Wang return BIO_MERGE_OK; 9768e756373SBaolin Wang } 9778e756373SBaolin Wang 978eda5cc99SChristoph Hellwig static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q, 979eda5cc99SChristoph Hellwig struct request *req, struct bio *bio) 9808e756373SBaolin Wang { 9818e756373SBaolin Wang unsigned short segments = blk_rq_nr_discard_segments(req); 9828e756373SBaolin Wang 9838e756373SBaolin Wang if (segments >= queue_max_discard_segments(q)) 9848e756373SBaolin Wang goto no_merge; 9858e756373SBaolin Wang if (blk_rq_sectors(req) + bio_sectors(bio) > 9868e756373SBaolin Wang blk_rq_get_max_sectors(req, blk_rq_pos(req))) 9878e756373SBaolin Wang goto no_merge; 9888e756373SBaolin Wang 9898e756373SBaolin Wang rq_qos_merge(q, req, bio); 9908e756373SBaolin Wang 9918e756373SBaolin Wang req->biotail->bi_next = bio; 9928e756373SBaolin Wang req->biotail = bio; 9938e756373SBaolin Wang req->__data_len += bio->bi_iter.bi_size; 9948e756373SBaolin Wang req->nr_phys_segments = segments + 1; 9958e756373SBaolin Wang 9968e756373SBaolin Wang blk_account_io_merge_bio(req); 9977d7ca7c5SBaolin Wang return BIO_MERGE_OK; 9988e756373SBaolin Wang no_merge: 9998e756373SBaolin Wang req_set_nomerge(q, req); 10007d7ca7c5SBaolin Wang return BIO_MERGE_FAILED; 10017d7ca7c5SBaolin Wang } 10027d7ca7c5SBaolin Wang 10037d7ca7c5SBaolin Wang static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q, 10047d7ca7c5SBaolin Wang struct request *rq, 10057d7ca7c5SBaolin Wang struct bio *bio, 10067d7ca7c5SBaolin Wang unsigned int nr_segs, 10077d7ca7c5SBaolin Wang bool sched_allow_merge) 10087d7ca7c5SBaolin Wang { 10097d7ca7c5SBaolin Wang if (!blk_rq_merge_ok(rq, bio)) 10107d7ca7c5SBaolin Wang return BIO_MERGE_NONE; 10117d7ca7c5SBaolin Wang 10127d7ca7c5SBaolin Wang switch (blk_try_merge(rq, bio)) { 10137d7ca7c5SBaolin Wang case ELEVATOR_BACK_MERGE: 1014265600b7SBaolin Wang if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) 10157d7ca7c5SBaolin Wang return bio_attempt_back_merge(rq, bio, nr_segs); 10167d7ca7c5SBaolin Wang break; 10177d7ca7c5SBaolin Wang case ELEVATOR_FRONT_MERGE: 1018265600b7SBaolin Wang if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) 10197d7ca7c5SBaolin Wang return bio_attempt_front_merge(rq, bio, nr_segs); 10207d7ca7c5SBaolin Wang break; 10217d7ca7c5SBaolin Wang case ELEVATOR_DISCARD_MERGE: 10227d7ca7c5SBaolin Wang return bio_attempt_discard_merge(q, rq, bio); 10237d7ca7c5SBaolin Wang default: 10247d7ca7c5SBaolin Wang return BIO_MERGE_NONE; 10257d7ca7c5SBaolin Wang } 10267d7ca7c5SBaolin Wang 10277d7ca7c5SBaolin Wang return BIO_MERGE_FAILED; 10288e756373SBaolin Wang } 10298e756373SBaolin Wang 10308e756373SBaolin Wang /** 10318e756373SBaolin Wang * blk_attempt_plug_merge - try to merge with %current's plugged list 10328e756373SBaolin Wang * @q: request_queue new bio is being queued at 10338e756373SBaolin Wang * @bio: new bio being queued 10348e756373SBaolin Wang * @nr_segs: number of segments in @bio 103587c037d1SJens Axboe * from the passed in @q already in the plug list 10368e756373SBaolin Wang * 1037d38a9c04SJens Axboe * Determine whether @bio being queued on @q can be merged with the previous 1038d38a9c04SJens Axboe * request on %current's plugged list. Returns %true if merge was successful, 10398e756373SBaolin Wang * otherwise %false. 10408e756373SBaolin Wang * 10418e756373SBaolin Wang * Plugging coalesces IOs from the same issuer for the same purpose without 10428e756373SBaolin Wang * going through @q->queue_lock. As such it's more of an issuing mechanism 10438e756373SBaolin Wang * than scheduling, and the request, while may have elvpriv data, is not 10448e756373SBaolin Wang * added on the elevator at this point. In addition, we don't have 10458e756373SBaolin Wang * reliable access to the elevator outside queue lock. Only check basic 10468e756373SBaolin Wang * merging parameters without querying the elevator. 10478e756373SBaolin Wang * 10488e756373SBaolin Wang * Caller must ensure !blk_queue_nomerges(q) beforehand. 10498e756373SBaolin Wang */ 10508e756373SBaolin Wang bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, 10510c5bcc92SChristoph Hellwig unsigned int nr_segs) 10528e756373SBaolin Wang { 10538e756373SBaolin Wang struct blk_plug *plug; 10548e756373SBaolin Wang struct request *rq; 10558e756373SBaolin Wang 10566deacb3bSChristoph Hellwig plug = blk_mq_plug(bio); 1057bc490f81SJens Axboe if (!plug || rq_list_empty(plug->mq_list)) 10588e756373SBaolin Wang return false; 10598e756373SBaolin Wang 10605b205071SJens Axboe rq_list_for_each(&plug->mq_list, rq) { 106187c037d1SJens Axboe if (rq->q == q) { 1062a1cb6537SMing Lei if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) == 1063a1cb6537SMing Lei BIO_MERGE_OK) 10648e756373SBaolin Wang return true; 10655b205071SJens Axboe break; 10665b205071SJens Axboe } 10675b205071SJens Axboe 10685b205071SJens Axboe /* 10695b205071SJens Axboe * Only keep iterating plug list for merges if we have multiple 10705b205071SJens Axboe * queues 10715b205071SJens Axboe */ 10725b205071SJens Axboe if (!plug->multiple_queues) 10735b205071SJens Axboe break; 1074a1cb6537SMing Lei } 10758e756373SBaolin Wang return false; 10768e756373SBaolin Wang } 1077bdc6a287SBaolin Wang 1078bdc6a287SBaolin Wang /* 1079bdc6a287SBaolin Wang * Iterate list of requests and see if we can merge this bio with any 1080bdc6a287SBaolin Wang * of them. 1081bdc6a287SBaolin Wang */ 1082bdc6a287SBaolin Wang bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, 1083bdc6a287SBaolin Wang struct bio *bio, unsigned int nr_segs) 1084bdc6a287SBaolin Wang { 1085bdc6a287SBaolin Wang struct request *rq; 1086bdc6a287SBaolin Wang int checked = 8; 1087bdc6a287SBaolin Wang 1088bdc6a287SBaolin Wang list_for_each_entry_reverse(rq, list, queuelist) { 1089bdc6a287SBaolin Wang if (!checked--) 1090bdc6a287SBaolin Wang break; 1091bdc6a287SBaolin Wang 10927d7ca7c5SBaolin Wang switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) { 10937d7ca7c5SBaolin Wang case BIO_MERGE_NONE: 1094bdc6a287SBaolin Wang continue; 10957d7ca7c5SBaolin Wang case BIO_MERGE_OK: 10967d7ca7c5SBaolin Wang return true; 10977d7ca7c5SBaolin Wang case BIO_MERGE_FAILED: 10987d7ca7c5SBaolin Wang return false; 1099bdc6a287SBaolin Wang } 1100bdc6a287SBaolin Wang 1101bdc6a287SBaolin Wang } 1102bdc6a287SBaolin Wang 1103bdc6a287SBaolin Wang return false; 1104bdc6a287SBaolin Wang } 1105bdc6a287SBaolin Wang EXPORT_SYMBOL_GPL(blk_bio_list_merge); 1106eda5cc99SChristoph Hellwig 1107eda5cc99SChristoph Hellwig bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, 1108eda5cc99SChristoph Hellwig unsigned int nr_segs, struct request **merged_request) 1109eda5cc99SChristoph Hellwig { 1110eda5cc99SChristoph Hellwig struct request *rq; 1111eda5cc99SChristoph Hellwig 1112eda5cc99SChristoph Hellwig switch (elv_merge(q, &rq, bio)) { 1113eda5cc99SChristoph Hellwig case ELEVATOR_BACK_MERGE: 1114eda5cc99SChristoph Hellwig if (!blk_mq_sched_allow_merge(q, rq, bio)) 1115eda5cc99SChristoph Hellwig return false; 1116eda5cc99SChristoph Hellwig if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK) 1117eda5cc99SChristoph Hellwig return false; 1118eda5cc99SChristoph Hellwig *merged_request = attempt_back_merge(q, rq); 1119eda5cc99SChristoph Hellwig if (!*merged_request) 1120eda5cc99SChristoph Hellwig elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); 1121eda5cc99SChristoph Hellwig return true; 1122eda5cc99SChristoph Hellwig case ELEVATOR_FRONT_MERGE: 1123eda5cc99SChristoph Hellwig if (!blk_mq_sched_allow_merge(q, rq, bio)) 1124eda5cc99SChristoph Hellwig return false; 1125eda5cc99SChristoph Hellwig if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK) 1126eda5cc99SChristoph Hellwig return false; 1127eda5cc99SChristoph Hellwig *merged_request = attempt_front_merge(q, rq); 1128eda5cc99SChristoph Hellwig if (!*merged_request) 1129eda5cc99SChristoph Hellwig elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); 1130eda5cc99SChristoph Hellwig return true; 1131eda5cc99SChristoph Hellwig case ELEVATOR_DISCARD_MERGE: 1132eda5cc99SChristoph Hellwig return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK; 1133eda5cc99SChristoph Hellwig default: 1134eda5cc99SChristoph Hellwig return false; 1135eda5cc99SChristoph Hellwig } 1136eda5cc99SChristoph Hellwig } 1137eda5cc99SChristoph Hellwig EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); 1138