1afc51aaaSDarrick J. Wong // SPDX-License-Identifier: GPL-2.0 2afc51aaaSDarrick J. Wong /* 3afc51aaaSDarrick J. Wong * Copyright (C) 2010 Red Hat, Inc. 4598ecfbaSChristoph Hellwig * Copyright (C) 2016-2019 Christoph Hellwig. 5afc51aaaSDarrick J. Wong */ 6afc51aaaSDarrick J. Wong #include <linux/module.h> 7afc51aaaSDarrick J. Wong #include <linux/compiler.h> 8afc51aaaSDarrick J. Wong #include <linux/fs.h> 9afc51aaaSDarrick J. Wong #include <linux/iomap.h> 10afc51aaaSDarrick J. Wong #include <linux/pagemap.h> 11afc51aaaSDarrick J. Wong #include <linux/uio.h> 12afc51aaaSDarrick J. Wong #include <linux/buffer_head.h> 13afc51aaaSDarrick J. Wong #include <linux/dax.h> 14afc51aaaSDarrick J. Wong #include <linux/writeback.h> 15598ecfbaSChristoph Hellwig #include <linux/list_sort.h> 16afc51aaaSDarrick J. Wong #include <linux/swap.h> 17afc51aaaSDarrick J. Wong #include <linux/bio.h> 18afc51aaaSDarrick J. Wong #include <linux/sched/signal.h> 19afc51aaaSDarrick J. Wong #include <linux/migrate.h> 209e91c572SChristoph Hellwig #include "trace.h" 21afc51aaaSDarrick J. Wong 22afc51aaaSDarrick J. Wong #include "../internal.h" 23afc51aaaSDarrick J. Wong 24ebb7fb15SDave Chinner #define IOEND_BATCH_SIZE 4096 25ebb7fb15SDave Chinner 26ab08b01eSChristoph Hellwig /* 2795c4cd05SMatthew Wilcox (Oracle) * Structure allocated for each folio when block size < folio size 2895c4cd05SMatthew Wilcox (Oracle) * to track sub-folio uptodate status and I/O completions. 29ab08b01eSChristoph Hellwig */ 30ab08b01eSChristoph Hellwig struct iomap_page { 317d636676SMatthew Wilcox (Oracle) atomic_t read_bytes_pending; 320fb2d720SMatthew Wilcox (Oracle) atomic_t write_bytes_pending; 331cea335dSChristoph Hellwig spinlock_t uptodate_lock; 340a195b91SMatthew Wilcox (Oracle) unsigned long uptodate[]; 35ab08b01eSChristoph Hellwig }; 36ab08b01eSChristoph Hellwig 3795c4cd05SMatthew Wilcox (Oracle) static inline struct iomap_page *to_iomap_page(struct folio *folio) 38ab08b01eSChristoph Hellwig { 3995c4cd05SMatthew Wilcox (Oracle) if (folio_test_private(folio)) 4095c4cd05SMatthew Wilcox (Oracle) return folio_get_private(folio); 41ab08b01eSChristoph Hellwig return NULL; 42ab08b01eSChristoph Hellwig } 43ab08b01eSChristoph Hellwig 44598ecfbaSChristoph Hellwig static struct bio_set iomap_ioend_bioset; 45598ecfbaSChristoph Hellwig 46afc51aaaSDarrick J. Wong static struct iomap_page * 479753b868SStefan Roesch iomap_page_create(struct inode *inode, struct folio *folio, unsigned int flags) 48afc51aaaSDarrick J. Wong { 4995c4cd05SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 50435d44b3SMatthew Wilcox (Oracle) unsigned int nr_blocks = i_blocks_per_folio(inode, folio); 519753b868SStefan Roesch gfp_t gfp; 52afc51aaaSDarrick J. Wong 530a195b91SMatthew Wilcox (Oracle) if (iop || nr_blocks <= 1) 54afc51aaaSDarrick J. Wong return iop; 55afc51aaaSDarrick J. Wong 569753b868SStefan Roesch if (flags & IOMAP_NOWAIT) 579753b868SStefan Roesch gfp = GFP_NOWAIT; 589753b868SStefan Roesch else 599753b868SStefan Roesch gfp = GFP_NOFS | __GFP_NOFAIL; 609753b868SStefan Roesch 610a195b91SMatthew Wilcox (Oracle) iop = kzalloc(struct_size(iop, uptodate, BITS_TO_LONGS(nr_blocks)), 629753b868SStefan Roesch gfp); 639753b868SStefan Roesch if (iop) { 641cea335dSChristoph Hellwig spin_lock_init(&iop->uptodate_lock); 65435d44b3SMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) 664595a298SMatthew Wilcox (Oracle) bitmap_fill(iop->uptodate, nr_blocks); 67435d44b3SMatthew Wilcox (Oracle) folio_attach_private(folio, iop); 689753b868SStefan Roesch } 69afc51aaaSDarrick J. Wong return iop; 70afc51aaaSDarrick J. Wong } 71afc51aaaSDarrick J. Wong 72c46e8324SMatthew Wilcox (Oracle) static void iomap_page_release(struct folio *folio) 73afc51aaaSDarrick J. Wong { 74c46e8324SMatthew Wilcox (Oracle) struct iomap_page *iop = folio_detach_private(folio); 75c46e8324SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 76c46e8324SMatthew Wilcox (Oracle) unsigned int nr_blocks = i_blocks_per_folio(inode, folio); 77afc51aaaSDarrick J. Wong 78afc51aaaSDarrick J. Wong if (!iop) 79afc51aaaSDarrick J. Wong return; 807d636676SMatthew Wilcox (Oracle) WARN_ON_ONCE(atomic_read(&iop->read_bytes_pending)); 810fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(atomic_read(&iop->write_bytes_pending)); 820a195b91SMatthew Wilcox (Oracle) WARN_ON_ONCE(bitmap_full(iop->uptodate, nr_blocks) != 83c46e8324SMatthew Wilcox (Oracle) folio_test_uptodate(folio)); 84afc51aaaSDarrick J. Wong kfree(iop); 85afc51aaaSDarrick J. Wong } 86afc51aaaSDarrick J. Wong 87afc51aaaSDarrick J. Wong /* 88431c0566SMatthew Wilcox (Oracle) * Calculate the range inside the folio that we actually need to read. 89afc51aaaSDarrick J. Wong */ 90431c0566SMatthew Wilcox (Oracle) static void iomap_adjust_read_range(struct inode *inode, struct folio *folio, 91431c0566SMatthew Wilcox (Oracle) loff_t *pos, loff_t length, size_t *offp, size_t *lenp) 92afc51aaaSDarrick J. Wong { 93431c0566SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 94afc51aaaSDarrick J. Wong loff_t orig_pos = *pos; 95afc51aaaSDarrick J. Wong loff_t isize = i_size_read(inode); 96afc51aaaSDarrick J. Wong unsigned block_bits = inode->i_blkbits; 97afc51aaaSDarrick J. Wong unsigned block_size = (1 << block_bits); 98431c0566SMatthew Wilcox (Oracle) size_t poff = offset_in_folio(folio, *pos); 99431c0566SMatthew Wilcox (Oracle) size_t plen = min_t(loff_t, folio_size(folio) - poff, length); 100afc51aaaSDarrick J. Wong unsigned first = poff >> block_bits; 101afc51aaaSDarrick J. Wong unsigned last = (poff + plen - 1) >> block_bits; 102afc51aaaSDarrick J. Wong 103afc51aaaSDarrick J. Wong /* 104f1f264b4SAndreas Gruenbacher * If the block size is smaller than the page size, we need to check the 105afc51aaaSDarrick J. Wong * per-block uptodate status and adjust the offset and length if needed 106afc51aaaSDarrick J. Wong * to avoid reading in already uptodate ranges. 107afc51aaaSDarrick J. Wong */ 108afc51aaaSDarrick J. Wong if (iop) { 109afc51aaaSDarrick J. Wong unsigned int i; 110afc51aaaSDarrick J. Wong 111afc51aaaSDarrick J. Wong /* move forward for each leading block marked uptodate */ 112afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) { 113afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 114afc51aaaSDarrick J. Wong break; 115afc51aaaSDarrick J. Wong *pos += block_size; 116afc51aaaSDarrick J. Wong poff += block_size; 117afc51aaaSDarrick J. Wong plen -= block_size; 118afc51aaaSDarrick J. Wong first++; 119afc51aaaSDarrick J. Wong } 120afc51aaaSDarrick J. Wong 121afc51aaaSDarrick J. Wong /* truncate len if we find any trailing uptodate block(s) */ 122afc51aaaSDarrick J. Wong for ( ; i <= last; i++) { 123afc51aaaSDarrick J. Wong if (test_bit(i, iop->uptodate)) { 124afc51aaaSDarrick J. Wong plen -= (last - i + 1) * block_size; 125afc51aaaSDarrick J. Wong last = i - 1; 126afc51aaaSDarrick J. Wong break; 127afc51aaaSDarrick J. Wong } 128afc51aaaSDarrick J. Wong } 129afc51aaaSDarrick J. Wong } 130afc51aaaSDarrick J. Wong 131afc51aaaSDarrick J. Wong /* 132f1f264b4SAndreas Gruenbacher * If the extent spans the block that contains the i_size, we need to 133afc51aaaSDarrick J. Wong * handle both halves separately so that we properly zero data in the 134afc51aaaSDarrick J. Wong * page cache for blocks that are entirely outside of i_size. 135afc51aaaSDarrick J. Wong */ 136afc51aaaSDarrick J. Wong if (orig_pos <= isize && orig_pos + length > isize) { 137431c0566SMatthew Wilcox (Oracle) unsigned end = offset_in_folio(folio, isize - 1) >> block_bits; 138afc51aaaSDarrick J. Wong 139afc51aaaSDarrick J. Wong if (first <= end && last > end) 140afc51aaaSDarrick J. Wong plen -= (last - end) * block_size; 141afc51aaaSDarrick J. Wong } 142afc51aaaSDarrick J. Wong 143afc51aaaSDarrick J. Wong *offp = poff; 144afc51aaaSDarrick J. Wong *lenp = plen; 145afc51aaaSDarrick J. Wong } 146afc51aaaSDarrick J. Wong 147431c0566SMatthew Wilcox (Oracle) static void iomap_iop_set_range_uptodate(struct folio *folio, 148431c0566SMatthew Wilcox (Oracle) struct iomap_page *iop, size_t off, size_t len) 149afc51aaaSDarrick J. Wong { 150431c0566SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 151afc51aaaSDarrick J. Wong unsigned first = off >> inode->i_blkbits; 152afc51aaaSDarrick J. Wong unsigned last = (off + len - 1) >> inode->i_blkbits; 1531cea335dSChristoph Hellwig unsigned long flags; 154afc51aaaSDarrick J. Wong 1551cea335dSChristoph Hellwig spin_lock_irqsave(&iop->uptodate_lock, flags); 156b21866f5SMatthew Wilcox (Oracle) bitmap_set(iop->uptodate, first, last - first + 1); 157431c0566SMatthew Wilcox (Oracle) if (bitmap_full(iop->uptodate, i_blocks_per_folio(inode, folio))) 158431c0566SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 1591cea335dSChristoph Hellwig spin_unlock_irqrestore(&iop->uptodate_lock, flags); 160afc51aaaSDarrick J. Wong } 161afc51aaaSDarrick J. Wong 162431c0566SMatthew Wilcox (Oracle) static void iomap_set_range_uptodate(struct folio *folio, 163431c0566SMatthew Wilcox (Oracle) struct iomap_page *iop, size_t off, size_t len) 1641cea335dSChristoph Hellwig { 165cd1e5afeSMatthew Wilcox (Oracle) if (iop) 166431c0566SMatthew Wilcox (Oracle) iomap_iop_set_range_uptodate(folio, iop, off, len); 1671cea335dSChristoph Hellwig else 168431c0566SMatthew Wilcox (Oracle) folio_mark_uptodate(folio); 169afc51aaaSDarrick J. Wong } 170afc51aaaSDarrick J. Wong 1718ffd74e9SMatthew Wilcox (Oracle) static void iomap_finish_folio_read(struct folio *folio, size_t offset, 1728ffd74e9SMatthew Wilcox (Oracle) size_t len, int error) 173afc51aaaSDarrick J. Wong { 17495c4cd05SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 175afc51aaaSDarrick J. Wong 176afc51aaaSDarrick J. Wong if (unlikely(error)) { 1778ffd74e9SMatthew Wilcox (Oracle) folio_clear_uptodate(folio); 1788ffd74e9SMatthew Wilcox (Oracle) folio_set_error(folio); 179afc51aaaSDarrick J. Wong } else { 180431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, offset, len); 181afc51aaaSDarrick J. Wong } 182afc51aaaSDarrick J. Wong 1838ffd74e9SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(len, &iop->read_bytes_pending)) 1848ffd74e9SMatthew Wilcox (Oracle) folio_unlock(folio); 185afc51aaaSDarrick J. Wong } 186afc51aaaSDarrick J. Wong 1878ffd74e9SMatthew Wilcox (Oracle) static void iomap_read_end_io(struct bio *bio) 188afc51aaaSDarrick J. Wong { 189afc51aaaSDarrick J. Wong int error = blk_status_to_errno(bio->bi_status); 1908ffd74e9SMatthew Wilcox (Oracle) struct folio_iter fi; 191afc51aaaSDarrick J. Wong 1928ffd74e9SMatthew Wilcox (Oracle) bio_for_each_folio_all(fi, bio) 1938ffd74e9SMatthew Wilcox (Oracle) iomap_finish_folio_read(fi.folio, fi.offset, fi.length, error); 194afc51aaaSDarrick J. Wong bio_put(bio); 195afc51aaaSDarrick J. Wong } 196afc51aaaSDarrick J. Wong 197afc51aaaSDarrick J. Wong struct iomap_readpage_ctx { 1983aa9c659SMatthew Wilcox (Oracle) struct folio *cur_folio; 1993aa9c659SMatthew Wilcox (Oracle) bool cur_folio_in_bio; 200afc51aaaSDarrick J. Wong struct bio *bio; 2019d24a13aSMatthew Wilcox (Oracle) struct readahead_control *rac; 202afc51aaaSDarrick J. Wong }; 203afc51aaaSDarrick J. Wong 2045ad448ceSAndreas Gruenbacher /** 2055ad448ceSAndreas Gruenbacher * iomap_read_inline_data - copy inline data into the page cache 2065ad448ceSAndreas Gruenbacher * @iter: iteration structure 207874628a2SMatthew Wilcox (Oracle) * @folio: folio to copy to 2085ad448ceSAndreas Gruenbacher * 209874628a2SMatthew Wilcox (Oracle) * Copy the inline data in @iter into @folio and zero out the rest of the folio. 2105ad448ceSAndreas Gruenbacher * Only a single IOMAP_INLINE extent is allowed at the end of each file. 2115ad448ceSAndreas Gruenbacher * Returns zero for success to complete the read, or the usual negative errno. 2125ad448ceSAndreas Gruenbacher */ 2135ad448ceSAndreas Gruenbacher static int iomap_read_inline_data(const struct iomap_iter *iter, 214874628a2SMatthew Wilcox (Oracle) struct folio *folio) 215afc51aaaSDarrick J. Wong { 216cd1e5afeSMatthew Wilcox (Oracle) struct iomap_page *iop; 217fad0a1abSChristoph Hellwig const struct iomap *iomap = iomap_iter_srcmap(iter); 2181b5c1e36SChristoph Hellwig size_t size = i_size_read(iter->inode) - iomap->offset; 219b405435bSMatthew Wilcox (Oracle) size_t poff = offset_in_page(iomap->offset); 220431c0566SMatthew Wilcox (Oracle) size_t offset = offset_in_folio(folio, iomap->offset); 221afc51aaaSDarrick J. Wong void *addr; 222afc51aaaSDarrick J. Wong 223874628a2SMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) 2245ad448ceSAndreas Gruenbacher return 0; 225afc51aaaSDarrick J. Wong 226ae44f9c2SMatthew Wilcox (Oracle) if (WARN_ON_ONCE(size > PAGE_SIZE - poff)) 227ae44f9c2SMatthew Wilcox (Oracle) return -EIO; 22869f4a26cSGao Xiang if (WARN_ON_ONCE(size > PAGE_SIZE - 22969f4a26cSGao Xiang offset_in_page(iomap->inline_data))) 23069f4a26cSGao Xiang return -EIO; 23169f4a26cSGao Xiang if (WARN_ON_ONCE(size > iomap->length)) 23269f4a26cSGao Xiang return -EIO; 233431c0566SMatthew Wilcox (Oracle) if (offset > 0) 2349753b868SStefan Roesch iop = iomap_page_create(iter->inode, folio, iter->flags); 235cd1e5afeSMatthew Wilcox (Oracle) else 236cd1e5afeSMatthew Wilcox (Oracle) iop = to_iomap_page(folio); 237afc51aaaSDarrick J. Wong 238874628a2SMatthew Wilcox (Oracle) addr = kmap_local_folio(folio, offset); 239afc51aaaSDarrick J. Wong memcpy(addr, iomap->inline_data, size); 240b405435bSMatthew Wilcox (Oracle) memset(addr + size, 0, PAGE_SIZE - poff - size); 241ab069d5fSMatthew Wilcox (Oracle) kunmap_local(addr); 242431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, offset, PAGE_SIZE - poff); 2435ad448ceSAndreas Gruenbacher return 0; 244afc51aaaSDarrick J. Wong } 245afc51aaaSDarrick J. Wong 246fad0a1abSChristoph Hellwig static inline bool iomap_block_needs_zeroing(const struct iomap_iter *iter, 2471b5c1e36SChristoph Hellwig loff_t pos) 248009d8d84SChristoph Hellwig { 249fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 2501b5c1e36SChristoph Hellwig 2511b5c1e36SChristoph Hellwig return srcmap->type != IOMAP_MAPPED || 2521b5c1e36SChristoph Hellwig (srcmap->flags & IOMAP_F_NEW) || 2531b5c1e36SChristoph Hellwig pos >= i_size_read(iter->inode); 254009d8d84SChristoph Hellwig } 255009d8d84SChristoph Hellwig 256fad0a1abSChristoph Hellwig static loff_t iomap_readpage_iter(const struct iomap_iter *iter, 257f6d48000SChristoph Hellwig struct iomap_readpage_ctx *ctx, loff_t offset) 258afc51aaaSDarrick J. Wong { 259fad0a1abSChristoph Hellwig const struct iomap *iomap = &iter->iomap; 260f6d48000SChristoph Hellwig loff_t pos = iter->pos + offset; 261f6d48000SChristoph Hellwig loff_t length = iomap_length(iter) - offset; 2623aa9c659SMatthew Wilcox (Oracle) struct folio *folio = ctx->cur_folio; 263637d3375SAndreas Gruenbacher struct iomap_page *iop; 264afc51aaaSDarrick J. Wong loff_t orig_pos = pos; 265431c0566SMatthew Wilcox (Oracle) size_t poff, plen; 266afc51aaaSDarrick J. Wong sector_t sector; 267afc51aaaSDarrick J. Wong 2685ad448ceSAndreas Gruenbacher if (iomap->type == IOMAP_INLINE) 269874628a2SMatthew Wilcox (Oracle) return iomap_read_inline_data(iter, folio); 270afc51aaaSDarrick J. Wong 271afc51aaaSDarrick J. Wong /* zero post-eof blocks as the page may be mapped */ 2729753b868SStefan Roesch iop = iomap_page_create(iter->inode, folio, iter->flags); 273431c0566SMatthew Wilcox (Oracle) iomap_adjust_read_range(iter->inode, folio, &pos, length, &poff, &plen); 274afc51aaaSDarrick J. Wong if (plen == 0) 275afc51aaaSDarrick J. Wong goto done; 276afc51aaaSDarrick J. Wong 2771b5c1e36SChristoph Hellwig if (iomap_block_needs_zeroing(iter, pos)) { 278431c0566SMatthew Wilcox (Oracle) folio_zero_range(folio, poff, plen); 279431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, poff, plen); 280afc51aaaSDarrick J. Wong goto done; 281afc51aaaSDarrick J. Wong } 282afc51aaaSDarrick J. Wong 2833aa9c659SMatthew Wilcox (Oracle) ctx->cur_folio_in_bio = true; 2847d636676SMatthew Wilcox (Oracle) if (iop) 2857d636676SMatthew Wilcox (Oracle) atomic_add(plen, &iop->read_bytes_pending); 286afc51aaaSDarrick J. Wong 287afc51aaaSDarrick J. Wong sector = iomap_sector(iomap, pos); 288d0364f94SChristoph Hellwig if (!ctx->bio || 289d0364f94SChristoph Hellwig bio_end_sector(ctx->bio) != sector || 290431c0566SMatthew Wilcox (Oracle) !bio_add_folio(ctx->bio, folio, plen, poff)) { 2913aa9c659SMatthew Wilcox (Oracle) gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL); 292457df33eSMatthew Wilcox (Oracle) gfp_t orig_gfp = gfp; 2935f7136dbSMatthew Wilcox (Oracle) unsigned int nr_vecs = DIV_ROUND_UP(length, PAGE_SIZE); 294afc51aaaSDarrick J. Wong 295afc51aaaSDarrick J. Wong if (ctx->bio) 296afc51aaaSDarrick J. Wong submit_bio(ctx->bio); 297afc51aaaSDarrick J. Wong 2989d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) /* same as readahead_gfp_mask */ 299afc51aaaSDarrick J. Wong gfp |= __GFP_NORETRY | __GFP_NOWARN; 30007888c66SChristoph Hellwig ctx->bio = bio_alloc(iomap->bdev, bio_max_segs(nr_vecs), 30107888c66SChristoph Hellwig REQ_OP_READ, gfp); 302457df33eSMatthew Wilcox (Oracle) /* 303457df33eSMatthew Wilcox (Oracle) * If the bio_alloc fails, try it again for a single page to 304457df33eSMatthew Wilcox (Oracle) * avoid having to deal with partial page reads. This emulates 305f132ab7dSMatthew Wilcox (Oracle) * what do_mpage_read_folio does. 306457df33eSMatthew Wilcox (Oracle) */ 30707888c66SChristoph Hellwig if (!ctx->bio) { 30807888c66SChristoph Hellwig ctx->bio = bio_alloc(iomap->bdev, 1, REQ_OP_READ, 30907888c66SChristoph Hellwig orig_gfp); 31007888c66SChristoph Hellwig } 3119d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) 312afc51aaaSDarrick J. Wong ctx->bio->bi_opf |= REQ_RAHEAD; 313afc51aaaSDarrick J. Wong ctx->bio->bi_iter.bi_sector = sector; 314afc51aaaSDarrick J. Wong ctx->bio->bi_end_io = iomap_read_end_io; 315*c2478469SJohannes Thumshirn bio_add_folio_nofail(ctx->bio, folio, plen, poff); 316afc51aaaSDarrick J. Wong } 317431c0566SMatthew Wilcox (Oracle) 318afc51aaaSDarrick J. Wong done: 319afc51aaaSDarrick J. Wong /* 320afc51aaaSDarrick J. Wong * Move the caller beyond our range so that it keeps making progress. 321f1f264b4SAndreas Gruenbacher * For that, we have to include any leading non-uptodate ranges, but 322afc51aaaSDarrick J. Wong * we can skip trailing ones as they will be handled in the next 323afc51aaaSDarrick J. Wong * iteration. 324afc51aaaSDarrick J. Wong */ 325afc51aaaSDarrick J. Wong return pos - orig_pos + plen; 326afc51aaaSDarrick J. Wong } 327afc51aaaSDarrick J. Wong 3287479c505SMatthew Wilcox (Oracle) int iomap_read_folio(struct folio *folio, const struct iomap_ops *ops) 329afc51aaaSDarrick J. Wong { 330f6d48000SChristoph Hellwig struct iomap_iter iter = { 3313aa9c659SMatthew Wilcox (Oracle) .inode = folio->mapping->host, 3323aa9c659SMatthew Wilcox (Oracle) .pos = folio_pos(folio), 3333aa9c659SMatthew Wilcox (Oracle) .len = folio_size(folio), 334f6d48000SChristoph Hellwig }; 335f6d48000SChristoph Hellwig struct iomap_readpage_ctx ctx = { 3363aa9c659SMatthew Wilcox (Oracle) .cur_folio = folio, 337f6d48000SChristoph Hellwig }; 338f6d48000SChristoph Hellwig int ret; 339afc51aaaSDarrick J. Wong 3403aa9c659SMatthew Wilcox (Oracle) trace_iomap_readpage(iter.inode, 1); 3419e91c572SChristoph Hellwig 342f6d48000SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 343f6d48000SChristoph Hellwig iter.processed = iomap_readpage_iter(&iter, &ctx, 0); 344f6d48000SChristoph Hellwig 345f6d48000SChristoph Hellwig if (ret < 0) 3463aa9c659SMatthew Wilcox (Oracle) folio_set_error(folio); 347afc51aaaSDarrick J. Wong 348afc51aaaSDarrick J. Wong if (ctx.bio) { 349afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 3503aa9c659SMatthew Wilcox (Oracle) WARN_ON_ONCE(!ctx.cur_folio_in_bio); 351afc51aaaSDarrick J. Wong } else { 3523aa9c659SMatthew Wilcox (Oracle) WARN_ON_ONCE(ctx.cur_folio_in_bio); 3533aa9c659SMatthew Wilcox (Oracle) folio_unlock(folio); 354afc51aaaSDarrick J. Wong } 355afc51aaaSDarrick J. Wong 356afc51aaaSDarrick J. Wong /* 3572c69e205SMatthew Wilcox (Oracle) * Just like mpage_readahead and block_read_full_folio, we always 3587479c505SMatthew Wilcox (Oracle) * return 0 and just set the folio error flag on errors. This 359f1f264b4SAndreas Gruenbacher * should be cleaned up throughout the stack eventually. 360afc51aaaSDarrick J. Wong */ 361afc51aaaSDarrick J. Wong return 0; 362afc51aaaSDarrick J. Wong } 3637479c505SMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(iomap_read_folio); 364afc51aaaSDarrick J. Wong 365fad0a1abSChristoph Hellwig static loff_t iomap_readahead_iter(const struct iomap_iter *iter, 366f6d48000SChristoph Hellwig struct iomap_readpage_ctx *ctx) 367afc51aaaSDarrick J. Wong { 368f6d48000SChristoph Hellwig loff_t length = iomap_length(iter); 369afc51aaaSDarrick J. Wong loff_t done, ret; 370afc51aaaSDarrick J. Wong 371afc51aaaSDarrick J. Wong for (done = 0; done < length; done += ret) { 3723aa9c659SMatthew Wilcox (Oracle) if (ctx->cur_folio && 3733aa9c659SMatthew Wilcox (Oracle) offset_in_folio(ctx->cur_folio, iter->pos + done) == 0) { 3743aa9c659SMatthew Wilcox (Oracle) if (!ctx->cur_folio_in_bio) 3753aa9c659SMatthew Wilcox (Oracle) folio_unlock(ctx->cur_folio); 3763aa9c659SMatthew Wilcox (Oracle) ctx->cur_folio = NULL; 377afc51aaaSDarrick J. Wong } 3783aa9c659SMatthew Wilcox (Oracle) if (!ctx->cur_folio) { 3793aa9c659SMatthew Wilcox (Oracle) ctx->cur_folio = readahead_folio(ctx->rac); 3803aa9c659SMatthew Wilcox (Oracle) ctx->cur_folio_in_bio = false; 381afc51aaaSDarrick J. Wong } 382f6d48000SChristoph Hellwig ret = iomap_readpage_iter(iter, ctx, done); 383d8af404fSAndreas Gruenbacher if (ret <= 0) 384d8af404fSAndreas Gruenbacher return ret; 385afc51aaaSDarrick J. Wong } 386afc51aaaSDarrick J. Wong 387afc51aaaSDarrick J. Wong return done; 388afc51aaaSDarrick J. Wong } 389afc51aaaSDarrick J. Wong 3909d24a13aSMatthew Wilcox (Oracle) /** 3919d24a13aSMatthew Wilcox (Oracle) * iomap_readahead - Attempt to read pages from a file. 3929d24a13aSMatthew Wilcox (Oracle) * @rac: Describes the pages to be read. 3939d24a13aSMatthew Wilcox (Oracle) * @ops: The operations vector for the filesystem. 3949d24a13aSMatthew Wilcox (Oracle) * 3959d24a13aSMatthew Wilcox (Oracle) * This function is for filesystems to call to implement their readahead 3969d24a13aSMatthew Wilcox (Oracle) * address_space operation. 3979d24a13aSMatthew Wilcox (Oracle) * 3989d24a13aSMatthew Wilcox (Oracle) * Context: The @ops callbacks may submit I/O (eg to read the addresses of 3999d24a13aSMatthew Wilcox (Oracle) * blocks from disc), and may wait for it. The caller may be trying to 4009d24a13aSMatthew Wilcox (Oracle) * access a different page, and so sleeping excessively should be avoided. 4019d24a13aSMatthew Wilcox (Oracle) * It may allocate memory, but should avoid costly allocations. This 4029d24a13aSMatthew Wilcox (Oracle) * function is called with memalloc_nofs set, so allocations will not cause 4039d24a13aSMatthew Wilcox (Oracle) * the filesystem to be reentered. 4049d24a13aSMatthew Wilcox (Oracle) */ 4059d24a13aSMatthew Wilcox (Oracle) void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops) 406afc51aaaSDarrick J. Wong { 407f6d48000SChristoph Hellwig struct iomap_iter iter = { 408f6d48000SChristoph Hellwig .inode = rac->mapping->host, 409f6d48000SChristoph Hellwig .pos = readahead_pos(rac), 410f6d48000SChristoph Hellwig .len = readahead_length(rac), 411f6d48000SChristoph Hellwig }; 412afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { 4139d24a13aSMatthew Wilcox (Oracle) .rac = rac, 414afc51aaaSDarrick J. Wong }; 415afc51aaaSDarrick J. Wong 416f6d48000SChristoph Hellwig trace_iomap_readahead(rac->mapping->host, readahead_count(rac)); 4179e91c572SChristoph Hellwig 418f6d48000SChristoph Hellwig while (iomap_iter(&iter, ops) > 0) 419f6d48000SChristoph Hellwig iter.processed = iomap_readahead_iter(&iter, &ctx); 4209d24a13aSMatthew Wilcox (Oracle) 421afc51aaaSDarrick J. Wong if (ctx.bio) 422afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 4233aa9c659SMatthew Wilcox (Oracle) if (ctx.cur_folio) { 4243aa9c659SMatthew Wilcox (Oracle) if (!ctx.cur_folio_in_bio) 4253aa9c659SMatthew Wilcox (Oracle) folio_unlock(ctx.cur_folio); 426afc51aaaSDarrick J. Wong } 427afc51aaaSDarrick J. Wong } 4289d24a13aSMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(iomap_readahead); 429afc51aaaSDarrick J. Wong 430afc51aaaSDarrick J. Wong /* 4312e7e80f7SMatthew Wilcox (Oracle) * iomap_is_partially_uptodate checks whether blocks within a folio are 432afc51aaaSDarrick J. Wong * uptodate or not. 433afc51aaaSDarrick J. Wong * 4342e7e80f7SMatthew Wilcox (Oracle) * Returns true if all blocks which correspond to the specified part 4352e7e80f7SMatthew Wilcox (Oracle) * of the folio are uptodate. 436afc51aaaSDarrick J. Wong */ 4372e7e80f7SMatthew Wilcox (Oracle) bool iomap_is_partially_uptodate(struct folio *folio, size_t from, size_t count) 438afc51aaaSDarrick J. Wong { 43995c4cd05SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 4402e7e80f7SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 4412e7e80f7SMatthew Wilcox (Oracle) unsigned first, last, i; 442afc51aaaSDarrick J. Wong 4432e7e80f7SMatthew Wilcox (Oracle) if (!iop) 4442e7e80f7SMatthew Wilcox (Oracle) return false; 4452e7e80f7SMatthew Wilcox (Oracle) 4462756c818SMatthew Wilcox (Oracle) /* Caller's range may extend past the end of this folio */ 4472756c818SMatthew Wilcox (Oracle) count = min(folio_size(folio) - from, count); 448afc51aaaSDarrick J. Wong 4492756c818SMatthew Wilcox (Oracle) /* First and last blocks in range within folio */ 450afc51aaaSDarrick J. Wong first = from >> inode->i_blkbits; 4512756c818SMatthew Wilcox (Oracle) last = (from + count - 1) >> inode->i_blkbits; 452afc51aaaSDarrick J. Wong 453afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) 454afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 4552e7e80f7SMatthew Wilcox (Oracle) return false; 4562e7e80f7SMatthew Wilcox (Oracle) return true; 457afc51aaaSDarrick J. Wong } 458afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate); 459afc51aaaSDarrick J. Wong 46098321b51SAndreas Gruenbacher /** 46198321b51SAndreas Gruenbacher * iomap_get_folio - get a folio reference for writing 46298321b51SAndreas Gruenbacher * @iter: iteration structure 46398321b51SAndreas Gruenbacher * @pos: start offset of write 46498321b51SAndreas Gruenbacher * 46598321b51SAndreas Gruenbacher * Returns a locked reference to the folio at @pos, or an error pointer if the 46698321b51SAndreas Gruenbacher * folio could not be obtained. 46798321b51SAndreas Gruenbacher */ 46898321b51SAndreas Gruenbacher struct folio *iomap_get_folio(struct iomap_iter *iter, loff_t pos) 46998321b51SAndreas Gruenbacher { 470e999a5c5SMatthew Wilcox unsigned fgp = FGP_WRITEBEGIN | FGP_NOFS; 47198321b51SAndreas Gruenbacher 47298321b51SAndreas Gruenbacher if (iter->flags & IOMAP_NOWAIT) 47398321b51SAndreas Gruenbacher fgp |= FGP_NOWAIT; 47498321b51SAndreas Gruenbacher 47566dabbb6SChristoph Hellwig return __filemap_get_folio(iter->inode->i_mapping, pos >> PAGE_SHIFT, 47698321b51SAndreas Gruenbacher fgp, mapping_gfp_mask(iter->inode->i_mapping)); 47798321b51SAndreas Gruenbacher } 47898321b51SAndreas Gruenbacher EXPORT_SYMBOL_GPL(iomap_get_folio); 47998321b51SAndreas Gruenbacher 4808597447dSMatthew Wilcox (Oracle) bool iomap_release_folio(struct folio *folio, gfp_t gfp_flags) 481afc51aaaSDarrick J. Wong { 4828597447dSMatthew Wilcox (Oracle) trace_iomap_release_folio(folio->mapping->host, folio_pos(folio), 48339f16c83SMatthew Wilcox (Oracle) folio_size(folio)); 4849e91c572SChristoph Hellwig 485afc51aaaSDarrick J. Wong /* 4868597447dSMatthew Wilcox (Oracle) * mm accommodates an old ext3 case where clean folios might 4878597447dSMatthew Wilcox (Oracle) * not have had the dirty bit cleared. Thus, it can send actual 4888597447dSMatthew Wilcox (Oracle) * dirty folios to ->release_folio() via shrink_active_list(); 4898597447dSMatthew Wilcox (Oracle) * skip those here. 490afc51aaaSDarrick J. Wong */ 49139f16c83SMatthew Wilcox (Oracle) if (folio_test_dirty(folio) || folio_test_writeback(folio)) 4928597447dSMatthew Wilcox (Oracle) return false; 493c46e8324SMatthew Wilcox (Oracle) iomap_page_release(folio); 4948597447dSMatthew Wilcox (Oracle) return true; 495afc51aaaSDarrick J. Wong } 4968597447dSMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(iomap_release_folio); 497afc51aaaSDarrick J. Wong 4988306a5f5SMatthew Wilcox (Oracle) void iomap_invalidate_folio(struct folio *folio, size_t offset, size_t len) 499afc51aaaSDarrick J. Wong { 500d82354f6SMatthew Wilcox (Oracle) trace_iomap_invalidate_folio(folio->mapping->host, 5011241ebecSMatthew Wilcox (Oracle) folio_pos(folio) + offset, len); 5029e91c572SChristoph Hellwig 503afc51aaaSDarrick J. Wong /* 50460d82310SMatthew Wilcox (Oracle) * If we're invalidating the entire folio, clear the dirty state 50560d82310SMatthew Wilcox (Oracle) * from it and release it to avoid unnecessary buildup of the LRU. 506afc51aaaSDarrick J. Wong */ 5078306a5f5SMatthew Wilcox (Oracle) if (offset == 0 && len == folio_size(folio)) { 5088306a5f5SMatthew Wilcox (Oracle) WARN_ON_ONCE(folio_test_writeback(folio)); 5098306a5f5SMatthew Wilcox (Oracle) folio_cancel_dirty(folio); 510c46e8324SMatthew Wilcox (Oracle) iomap_page_release(folio); 51160d82310SMatthew Wilcox (Oracle) } else if (folio_test_large(folio)) { 51260d82310SMatthew Wilcox (Oracle) /* Must release the iop so the page can be split */ 51360d82310SMatthew Wilcox (Oracle) WARN_ON_ONCE(!folio_test_uptodate(folio) && 51460d82310SMatthew Wilcox (Oracle) folio_test_dirty(folio)); 51560d82310SMatthew Wilcox (Oracle) iomap_page_release(folio); 516afc51aaaSDarrick J. Wong } 517afc51aaaSDarrick J. Wong } 5188306a5f5SMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(iomap_invalidate_folio); 5198306a5f5SMatthew Wilcox (Oracle) 520afc51aaaSDarrick J. Wong static void 521afc51aaaSDarrick J. Wong iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) 522afc51aaaSDarrick J. Wong { 523afc51aaaSDarrick J. Wong loff_t i_size = i_size_read(inode); 524afc51aaaSDarrick J. Wong 525afc51aaaSDarrick J. Wong /* 526afc51aaaSDarrick J. Wong * Only truncate newly allocated pages beyoned EOF, even if the 527afc51aaaSDarrick J. Wong * write started inside the existing inode size. 528afc51aaaSDarrick J. Wong */ 529afc51aaaSDarrick J. Wong if (pos + len > i_size) 530b71450e2SAndreas Gruenbacher truncate_pagecache_range(inode, max(pos, i_size), 531b71450e2SAndreas Gruenbacher pos + len - 1); 532afc51aaaSDarrick J. Wong } 533afc51aaaSDarrick J. Wong 534431c0566SMatthew Wilcox (Oracle) static int iomap_read_folio_sync(loff_t block_start, struct folio *folio, 535431c0566SMatthew Wilcox (Oracle) size_t poff, size_t plen, const struct iomap *iomap) 536afc51aaaSDarrick J. Wong { 537afc51aaaSDarrick J. Wong struct bio_vec bvec; 538afc51aaaSDarrick J. Wong struct bio bio; 539afc51aaaSDarrick J. Wong 54049add496SChristoph Hellwig bio_init(&bio, iomap->bdev, &bvec, 1, REQ_OP_READ); 541afc51aaaSDarrick J. Wong bio.bi_iter.bi_sector = iomap_sector(iomap, block_start); 542*c2478469SJohannes Thumshirn bio_add_folio_nofail(&bio, folio, plen, poff); 543afc51aaaSDarrick J. Wong return submit_bio_wait(&bio); 544afc51aaaSDarrick J. Wong } 545afc51aaaSDarrick J. Wong 546fad0a1abSChristoph Hellwig static int __iomap_write_begin(const struct iomap_iter *iter, loff_t pos, 547bc6123a8SMatthew Wilcox (Oracle) size_t len, struct folio *folio) 548afc51aaaSDarrick J. Wong { 549fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 5509753b868SStefan Roesch struct iomap_page *iop; 5511b5c1e36SChristoph Hellwig loff_t block_size = i_blocksize(iter->inode); 5526cc19c5fSNikolay Borisov loff_t block_start = round_down(pos, block_size); 5536cc19c5fSNikolay Borisov loff_t block_end = round_up(pos + len, block_size); 554cae2de69SStefan Roesch unsigned int nr_blocks = i_blocks_per_folio(iter->inode, folio); 555431c0566SMatthew Wilcox (Oracle) size_t from = offset_in_folio(folio, pos), to = from + len; 556431c0566SMatthew Wilcox (Oracle) size_t poff, plen; 557afc51aaaSDarrick J. Wong 558431c0566SMatthew Wilcox (Oracle) if (folio_test_uptodate(folio)) 559afc51aaaSDarrick J. Wong return 0; 560431c0566SMatthew Wilcox (Oracle) folio_clear_error(folio); 561afc51aaaSDarrick J. Wong 5629753b868SStefan Roesch iop = iomap_page_create(iter->inode, folio, iter->flags); 563cae2de69SStefan Roesch if ((iter->flags & IOMAP_NOWAIT) && !iop && nr_blocks > 1) 564cae2de69SStefan Roesch return -EAGAIN; 5659753b868SStefan Roesch 566afc51aaaSDarrick J. Wong do { 567431c0566SMatthew Wilcox (Oracle) iomap_adjust_read_range(iter->inode, folio, &block_start, 568afc51aaaSDarrick J. Wong block_end - block_start, &poff, &plen); 569afc51aaaSDarrick J. Wong if (plen == 0) 570afc51aaaSDarrick J. Wong break; 571afc51aaaSDarrick J. Wong 572b74b1293SChristoph Hellwig if (!(iter->flags & IOMAP_UNSHARE) && 57332a38a49SChristoph Hellwig (from <= poff || from >= poff + plen) && 574d3b40439SChristoph Hellwig (to <= poff || to >= poff + plen)) 575d3b40439SChristoph Hellwig continue; 576d3b40439SChristoph Hellwig 5771b5c1e36SChristoph Hellwig if (iomap_block_needs_zeroing(iter, block_start)) { 578b74b1293SChristoph Hellwig if (WARN_ON_ONCE(iter->flags & IOMAP_UNSHARE)) 57932a38a49SChristoph Hellwig return -EIO; 580431c0566SMatthew Wilcox (Oracle) folio_zero_segments(folio, poff, from, to, poff + plen); 58114284fedSMatthew Wilcox (Oracle) } else { 582cae2de69SStefan Roesch int status; 583cae2de69SStefan Roesch 584cae2de69SStefan Roesch if (iter->flags & IOMAP_NOWAIT) 585cae2de69SStefan Roesch return -EAGAIN; 586cae2de69SStefan Roesch 587cae2de69SStefan Roesch status = iomap_read_folio_sync(block_start, folio, 58814284fedSMatthew Wilcox (Oracle) poff, plen, srcmap); 589d3b40439SChristoph Hellwig if (status) 590d3b40439SChristoph Hellwig return status; 59114284fedSMatthew Wilcox (Oracle) } 592431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, poff, plen); 593afc51aaaSDarrick J. Wong } while ((block_start += plen) < block_end); 594afc51aaaSDarrick J. Wong 595d3b40439SChristoph Hellwig return 0; 596afc51aaaSDarrick J. Wong } 597afc51aaaSDarrick J. Wong 59807c22b56SAndreas Gruenbacher static struct folio *__iomap_get_folio(struct iomap_iter *iter, loff_t pos, 59907c22b56SAndreas Gruenbacher size_t len) 60007c22b56SAndreas Gruenbacher { 601471859f5SAndreas Gruenbacher const struct iomap_folio_ops *folio_ops = iter->iomap.folio_ops; 60207c22b56SAndreas Gruenbacher 603471859f5SAndreas Gruenbacher if (folio_ops && folio_ops->get_folio) 604471859f5SAndreas Gruenbacher return folio_ops->get_folio(iter, pos, len); 60507c22b56SAndreas Gruenbacher else 60607c22b56SAndreas Gruenbacher return iomap_get_folio(iter, pos); 60707c22b56SAndreas Gruenbacher } 60807c22b56SAndreas Gruenbacher 6097a70a508SAndreas Gruenbacher static void __iomap_put_folio(struct iomap_iter *iter, loff_t pos, size_t ret, 6107a70a508SAndreas Gruenbacher struct folio *folio) 6117a70a508SAndreas Gruenbacher { 612471859f5SAndreas Gruenbacher const struct iomap_folio_ops *folio_ops = iter->iomap.folio_ops; 6137a70a508SAndreas Gruenbacher 614471859f5SAndreas Gruenbacher if (folio_ops && folio_ops->put_folio) { 615471859f5SAndreas Gruenbacher folio_ops->put_folio(iter->inode, pos, ret, folio); 6169060bc4dSAndreas Gruenbacher } else { 6177a70a508SAndreas Gruenbacher folio_unlock(folio); 6187a70a508SAndreas Gruenbacher folio_put(folio); 6197a70a508SAndreas Gruenbacher } 62080baab88SAndreas Gruenbacher } 6217a70a508SAndreas Gruenbacher 622fad0a1abSChristoph Hellwig static int iomap_write_begin_inline(const struct iomap_iter *iter, 623bc6123a8SMatthew Wilcox (Oracle) struct folio *folio) 62469f4a26cSGao Xiang { 62569f4a26cSGao Xiang /* needs more work for the tailpacking case; disable for now */ 6261b5c1e36SChristoph Hellwig if (WARN_ON_ONCE(iomap_iter_srcmap(iter)->offset != 0)) 62769f4a26cSGao Xiang return -EIO; 628874628a2SMatthew Wilcox (Oracle) return iomap_read_inline_data(iter, folio); 62969f4a26cSGao Xiang } 63069f4a26cSGao Xiang 631d7b64041SDave Chinner static int iomap_write_begin(struct iomap_iter *iter, loff_t pos, 632bc6123a8SMatthew Wilcox (Oracle) size_t len, struct folio **foliop) 633afc51aaaSDarrick J. Wong { 634471859f5SAndreas Gruenbacher const struct iomap_folio_ops *folio_ops = iter->iomap.folio_ops; 635fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 636d1bd0b4eSMatthew Wilcox (Oracle) struct folio *folio; 637afc51aaaSDarrick J. Wong int status = 0; 638afc51aaaSDarrick J. Wong 6391b5c1e36SChristoph Hellwig BUG_ON(pos + len > iter->iomap.offset + iter->iomap.length); 6401b5c1e36SChristoph Hellwig if (srcmap != &iter->iomap) 641c039b997SGoldwyn Rodrigues BUG_ON(pos + len > srcmap->offset + srcmap->length); 642afc51aaaSDarrick J. Wong 643afc51aaaSDarrick J. Wong if (fatal_signal_pending(current)) 644afc51aaaSDarrick J. Wong return -EINTR; 645afc51aaaSDarrick J. Wong 646d454ab82SMatthew Wilcox (Oracle) if (!mapping_large_folio_support(iter->inode->i_mapping)) 647d454ab82SMatthew Wilcox (Oracle) len = min_t(size_t, len, PAGE_SIZE - offset_in_page(pos)); 648d454ab82SMatthew Wilcox (Oracle) 64907c22b56SAndreas Gruenbacher folio = __iomap_get_folio(iter, pos, len); 6509060bc4dSAndreas Gruenbacher if (IS_ERR(folio)) 65198321b51SAndreas Gruenbacher return PTR_ERR(folio); 652d7b64041SDave Chinner 653d7b64041SDave Chinner /* 654d7b64041SDave Chinner * Now we have a locked folio, before we do anything with it we need to 655d7b64041SDave Chinner * check that the iomap we have cached is not stale. The inode extent 656d7b64041SDave Chinner * mapping can change due to concurrent IO in flight (e.g. 657d7b64041SDave Chinner * IOMAP_UNWRITTEN state can change and memory reclaim could have 658d7b64041SDave Chinner * reclaimed a previously partially written page at this index after IO 659d7b64041SDave Chinner * completion before this write reaches this file offset) and hence we 660d7b64041SDave Chinner * could do the wrong thing here (zero a page range incorrectly or fail 661d7b64041SDave Chinner * to zero) and corrupt data. 662d7b64041SDave Chinner */ 663471859f5SAndreas Gruenbacher if (folio_ops && folio_ops->iomap_valid) { 664471859f5SAndreas Gruenbacher bool iomap_valid = folio_ops->iomap_valid(iter->inode, 665d7b64041SDave Chinner &iter->iomap); 666d7b64041SDave Chinner if (!iomap_valid) { 667d7b64041SDave Chinner iter->iomap.flags |= IOMAP_F_STALE; 668d7b64041SDave Chinner status = 0; 669d7b64041SDave Chinner goto out_unlock; 670d7b64041SDave Chinner } 671d7b64041SDave Chinner } 672d7b64041SDave Chinner 673d454ab82SMatthew Wilcox (Oracle) if (pos + len > folio_pos(folio) + folio_size(folio)) 674d454ab82SMatthew Wilcox (Oracle) len = folio_pos(folio) + folio_size(folio) - pos; 675afc51aaaSDarrick J. Wong 676c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) 677bc6123a8SMatthew Wilcox (Oracle) status = iomap_write_begin_inline(iter, folio); 6781b5c1e36SChristoph Hellwig else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) 679d1bd0b4eSMatthew Wilcox (Oracle) status = __block_write_begin_int(folio, pos, len, NULL, srcmap); 680afc51aaaSDarrick J. Wong else 681bc6123a8SMatthew Wilcox (Oracle) status = __iomap_write_begin(iter, pos, len, folio); 682afc51aaaSDarrick J. Wong 683afc51aaaSDarrick J. Wong if (unlikely(status)) 684afc51aaaSDarrick J. Wong goto out_unlock; 685afc51aaaSDarrick J. Wong 686bc6123a8SMatthew Wilcox (Oracle) *foliop = folio; 687afc51aaaSDarrick J. Wong return 0; 688afc51aaaSDarrick J. Wong 689afc51aaaSDarrick J. Wong out_unlock: 6907a70a508SAndreas Gruenbacher __iomap_put_folio(iter, pos, 0, folio); 6911b5c1e36SChristoph Hellwig iomap_write_failed(iter->inode, pos, len); 692afc51aaaSDarrick J. Wong 693afc51aaaSDarrick J. Wong return status; 694afc51aaaSDarrick J. Wong } 695afc51aaaSDarrick J. Wong 696e25ba8cbSMatthew Wilcox (Oracle) static size_t __iomap_write_end(struct inode *inode, loff_t pos, size_t len, 697bc6123a8SMatthew Wilcox (Oracle) size_t copied, struct folio *folio) 698afc51aaaSDarrick J. Wong { 699cd1e5afeSMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 700bc6123a8SMatthew Wilcox (Oracle) flush_dcache_folio(folio); 701afc51aaaSDarrick J. Wong 702afc51aaaSDarrick J. Wong /* 703afc51aaaSDarrick J. Wong * The blocks that were entirely written will now be uptodate, so we 7047479c505SMatthew Wilcox (Oracle) * don't have to worry about a read_folio reading them and overwriting a 705f1f264b4SAndreas Gruenbacher * partial write. However, if we've encountered a short write and only 706afc51aaaSDarrick J. Wong * partially written into a block, it will not be marked uptodate, so a 7077479c505SMatthew Wilcox (Oracle) * read_folio might come in and destroy our partial write. 708afc51aaaSDarrick J. Wong * 709f1f264b4SAndreas Gruenbacher * Do the simplest thing and just treat any short write to a 710f1f264b4SAndreas Gruenbacher * non-uptodate page as a zero-length write, and force the caller to 711f1f264b4SAndreas Gruenbacher * redo the whole thing. 712afc51aaaSDarrick J. Wong */ 713bc6123a8SMatthew Wilcox (Oracle) if (unlikely(copied < len && !folio_test_uptodate(folio))) 714afc51aaaSDarrick J. Wong return 0; 715431c0566SMatthew Wilcox (Oracle) iomap_set_range_uptodate(folio, iop, offset_in_folio(folio, pos), len); 716bc6123a8SMatthew Wilcox (Oracle) filemap_dirty_folio(inode->i_mapping, folio); 717afc51aaaSDarrick J. Wong return copied; 718afc51aaaSDarrick J. Wong } 719afc51aaaSDarrick J. Wong 720fad0a1abSChristoph Hellwig static size_t iomap_write_end_inline(const struct iomap_iter *iter, 7219c4ce08dSMatthew Wilcox (Oracle) struct folio *folio, loff_t pos, size_t copied) 722afc51aaaSDarrick J. Wong { 723fad0a1abSChristoph Hellwig const struct iomap *iomap = &iter->iomap; 724afc51aaaSDarrick J. Wong void *addr; 725afc51aaaSDarrick J. Wong 7269c4ce08dSMatthew Wilcox (Oracle) WARN_ON_ONCE(!folio_test_uptodate(folio)); 72769f4a26cSGao Xiang BUG_ON(!iomap_inline_data_valid(iomap)); 728afc51aaaSDarrick J. Wong 7299c4ce08dSMatthew Wilcox (Oracle) flush_dcache_folio(folio); 7309c4ce08dSMatthew Wilcox (Oracle) addr = kmap_local_folio(folio, pos); 731ab069d5fSMatthew Wilcox (Oracle) memcpy(iomap_inline_data(iomap, pos), addr, copied); 732ab069d5fSMatthew Wilcox (Oracle) kunmap_local(addr); 733afc51aaaSDarrick J. Wong 7341b5c1e36SChristoph Hellwig mark_inode_dirty(iter->inode); 735afc51aaaSDarrick J. Wong return copied; 736afc51aaaSDarrick J. Wong } 737afc51aaaSDarrick J. Wong 738e25ba8cbSMatthew Wilcox (Oracle) /* Returns the number of bytes copied. May be 0. Cannot be an errno. */ 7391b5c1e36SChristoph Hellwig static size_t iomap_write_end(struct iomap_iter *iter, loff_t pos, size_t len, 740bc6123a8SMatthew Wilcox (Oracle) size_t copied, struct folio *folio) 741afc51aaaSDarrick J. Wong { 742fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 7431b5c1e36SChristoph Hellwig loff_t old_size = iter->inode->i_size; 744e25ba8cbSMatthew Wilcox (Oracle) size_t ret; 745afc51aaaSDarrick J. Wong 746c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) { 7479c4ce08dSMatthew Wilcox (Oracle) ret = iomap_write_end_inline(iter, folio, pos, copied); 748c039b997SGoldwyn Rodrigues } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) { 7491b5c1e36SChristoph Hellwig ret = block_write_end(NULL, iter->inode->i_mapping, pos, len, 750bc6123a8SMatthew Wilcox (Oracle) copied, &folio->page, NULL); 751afc51aaaSDarrick J. Wong } else { 752bc6123a8SMatthew Wilcox (Oracle) ret = __iomap_write_end(iter->inode, pos, len, copied, folio); 753afc51aaaSDarrick J. Wong } 754afc51aaaSDarrick J. Wong 755afc51aaaSDarrick J. Wong /* 756afc51aaaSDarrick J. Wong * Update the in-memory inode size after copying the data into the page 757afc51aaaSDarrick J. Wong * cache. It's up to the file system to write the updated size to disk, 758afc51aaaSDarrick J. Wong * preferably after I/O completion so that no stale data is exposed. 759afc51aaaSDarrick J. Wong */ 760afc51aaaSDarrick J. Wong if (pos + ret > old_size) { 7611b5c1e36SChristoph Hellwig i_size_write(iter->inode, pos + ret); 7621b5c1e36SChristoph Hellwig iter->iomap.flags |= IOMAP_F_SIZE_CHANGED; 763afc51aaaSDarrick J. Wong } 7647a70a508SAndreas Gruenbacher __iomap_put_folio(iter, pos, ret, folio); 765afc51aaaSDarrick J. Wong 766afc51aaaSDarrick J. Wong if (old_size < pos) 7671b5c1e36SChristoph Hellwig pagecache_isize_extended(iter->inode, old_size, pos); 768afc51aaaSDarrick J. Wong if (ret < len) 769d74999c8SAndreas Gruenbacher iomap_write_failed(iter->inode, pos + ret, len - ret); 770afc51aaaSDarrick J. Wong return ret; 771afc51aaaSDarrick J. Wong } 772afc51aaaSDarrick J. Wong 773ce83a025SChristoph Hellwig static loff_t iomap_write_iter(struct iomap_iter *iter, struct iov_iter *i) 774afc51aaaSDarrick J. Wong { 775ce83a025SChristoph Hellwig loff_t length = iomap_length(iter); 776ce83a025SChristoph Hellwig loff_t pos = iter->pos; 777afc51aaaSDarrick J. Wong ssize_t written = 0; 778ce83a025SChristoph Hellwig long status = 0; 779cae2de69SStefan Roesch struct address_space *mapping = iter->inode->i_mapping; 780cae2de69SStefan Roesch unsigned int bdp_flags = (iter->flags & IOMAP_NOWAIT) ? BDP_ASYNC : 0; 781afc51aaaSDarrick J. Wong 782afc51aaaSDarrick J. Wong do { 783bc6123a8SMatthew Wilcox (Oracle) struct folio *folio; 784afc51aaaSDarrick J. Wong struct page *page; 785afc51aaaSDarrick J. Wong unsigned long offset; /* Offset into pagecache page */ 786afc51aaaSDarrick J. Wong unsigned long bytes; /* Bytes to write to page */ 787afc51aaaSDarrick J. Wong size_t copied; /* Bytes copied from user */ 788afc51aaaSDarrick J. Wong 789afc51aaaSDarrick J. Wong offset = offset_in_page(pos); 790afc51aaaSDarrick J. Wong bytes = min_t(unsigned long, PAGE_SIZE - offset, 791afc51aaaSDarrick J. Wong iov_iter_count(i)); 792afc51aaaSDarrick J. Wong again: 793cae2de69SStefan Roesch status = balance_dirty_pages_ratelimited_flags(mapping, 794cae2de69SStefan Roesch bdp_flags); 795cae2de69SStefan Roesch if (unlikely(status)) 796cae2de69SStefan Roesch break; 797cae2de69SStefan Roesch 798afc51aaaSDarrick J. Wong if (bytes > length) 799afc51aaaSDarrick J. Wong bytes = length; 800afc51aaaSDarrick J. Wong 801afc51aaaSDarrick J. Wong /* 802f1f264b4SAndreas Gruenbacher * Bring in the user page that we'll copy from _first_. 803afc51aaaSDarrick J. Wong * Otherwise there's a nasty deadlock on copying from the 804afc51aaaSDarrick J. Wong * same page as we're writing to, without it being marked 805afc51aaaSDarrick J. Wong * up-to-date. 806cae2de69SStefan Roesch * 807cae2de69SStefan Roesch * For async buffered writes the assumption is that the user 808cae2de69SStefan Roesch * page has already been faulted in. This can be optimized by 809cae2de69SStefan Roesch * faulting the user page. 810afc51aaaSDarrick J. Wong */ 811631f871fSAndreas Gruenbacher if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) { 812afc51aaaSDarrick J. Wong status = -EFAULT; 813afc51aaaSDarrick J. Wong break; 814afc51aaaSDarrick J. Wong } 815afc51aaaSDarrick J. Wong 816bc6123a8SMatthew Wilcox (Oracle) status = iomap_write_begin(iter, pos, bytes, &folio); 817afc51aaaSDarrick J. Wong if (unlikely(status)) 818afc51aaaSDarrick J. Wong break; 819d7b64041SDave Chinner if (iter->iomap.flags & IOMAP_F_STALE) 820d7b64041SDave Chinner break; 821afc51aaaSDarrick J. Wong 822bc6123a8SMatthew Wilcox (Oracle) page = folio_file_page(folio, pos >> PAGE_SHIFT); 823cae2de69SStefan Roesch if (mapping_writably_mapped(mapping)) 824afc51aaaSDarrick J. Wong flush_dcache_page(page); 825afc51aaaSDarrick J. Wong 826f0b65f39SAl Viro copied = copy_page_from_iter_atomic(page, offset, bytes, i); 827afc51aaaSDarrick J. Wong 828bc6123a8SMatthew Wilcox (Oracle) status = iomap_write_end(iter, pos, bytes, copied, folio); 829afc51aaaSDarrick J. Wong 830f0b65f39SAl Viro if (unlikely(copied != status)) 831f0b65f39SAl Viro iov_iter_revert(i, copied - status); 832afc51aaaSDarrick J. Wong 833f0b65f39SAl Viro cond_resched(); 834bc1bb416SAl Viro if (unlikely(status == 0)) { 835afc51aaaSDarrick J. Wong /* 836bc1bb416SAl Viro * A short copy made iomap_write_end() reject the 837bc1bb416SAl Viro * thing entirely. Might be memory poisoning 838bc1bb416SAl Viro * halfway through, might be a race with munmap, 839bc1bb416SAl Viro * might be severe memory pressure. 840afc51aaaSDarrick J. Wong */ 841bc1bb416SAl Viro if (copied) 842bc1bb416SAl Viro bytes = copied; 843afc51aaaSDarrick J. Wong goto again; 844afc51aaaSDarrick J. Wong } 845f0b65f39SAl Viro pos += status; 846f0b65f39SAl Viro written += status; 847f0b65f39SAl Viro length -= status; 848afc51aaaSDarrick J. Wong } while (iov_iter_count(i) && length); 849afc51aaaSDarrick J. Wong 85018e419f6SStefan Roesch if (status == -EAGAIN) { 85118e419f6SStefan Roesch iov_iter_revert(i, written); 85218e419f6SStefan Roesch return -EAGAIN; 85318e419f6SStefan Roesch } 854afc51aaaSDarrick J. Wong return written ? written : status; 855afc51aaaSDarrick J. Wong } 856afc51aaaSDarrick J. Wong 857afc51aaaSDarrick J. Wong ssize_t 858ce83a025SChristoph Hellwig iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *i, 859afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 860afc51aaaSDarrick J. Wong { 861ce83a025SChristoph Hellwig struct iomap_iter iter = { 862ce83a025SChristoph Hellwig .inode = iocb->ki_filp->f_mapping->host, 863ce83a025SChristoph Hellwig .pos = iocb->ki_pos, 864ce83a025SChristoph Hellwig .len = iov_iter_count(i), 865ce83a025SChristoph Hellwig .flags = IOMAP_WRITE, 866ce83a025SChristoph Hellwig }; 867ce83a025SChristoph Hellwig int ret; 868afc51aaaSDarrick J. Wong 869cae2de69SStefan Roesch if (iocb->ki_flags & IOCB_NOWAIT) 870cae2de69SStefan Roesch iter.flags |= IOMAP_NOWAIT; 871cae2de69SStefan Roesch 872ce83a025SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 873ce83a025SChristoph Hellwig iter.processed = iomap_write_iter(&iter, i); 874ce83a025SChristoph Hellwig if (iter.pos == iocb->ki_pos) 875ce83a025SChristoph Hellwig return ret; 876ce83a025SChristoph Hellwig return iter.pos - iocb->ki_pos; 877afc51aaaSDarrick J. Wong } 878afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_file_buffered_write); 879afc51aaaSDarrick J. Wong 8809c7babf9SDave Chinner /* 881f43dc4dcSDave Chinner * Scan the data range passed to us for dirty page cache folios. If we find a 882f43dc4dcSDave Chinner * dirty folio, punch out the preceeding range and update the offset from which 883f43dc4dcSDave Chinner * the next punch will start from. 884f43dc4dcSDave Chinner * 885f43dc4dcSDave Chinner * We can punch out storage reservations under clean pages because they either 886f43dc4dcSDave Chinner * contain data that has been written back - in which case the delalloc punch 887f43dc4dcSDave Chinner * over that range is a no-op - or they have been read faults in which case they 888f43dc4dcSDave Chinner * contain zeroes and we can remove the delalloc backing range and any new 889f43dc4dcSDave Chinner * writes to those pages will do the normal hole filling operation... 890f43dc4dcSDave Chinner * 891f43dc4dcSDave Chinner * This makes the logic simple: we only need to keep the delalloc extents only 892f43dc4dcSDave Chinner * over the dirty ranges of the page cache. 893f43dc4dcSDave Chinner * 894f43dc4dcSDave Chinner * This function uses [start_byte, end_byte) intervals (i.e. open ended) to 895f43dc4dcSDave Chinner * simplify range iterations. 896f43dc4dcSDave Chinner */ 897f43dc4dcSDave Chinner static int iomap_write_delalloc_scan(struct inode *inode, 898f43dc4dcSDave Chinner loff_t *punch_start_byte, loff_t start_byte, loff_t end_byte, 899f43dc4dcSDave Chinner int (*punch)(struct inode *inode, loff_t offset, loff_t length)) 900f43dc4dcSDave Chinner { 901f43dc4dcSDave Chinner while (start_byte < end_byte) { 902f43dc4dcSDave Chinner struct folio *folio; 903f43dc4dcSDave Chinner 904f43dc4dcSDave Chinner /* grab locked page */ 905f43dc4dcSDave Chinner folio = filemap_lock_folio(inode->i_mapping, 906f43dc4dcSDave Chinner start_byte >> PAGE_SHIFT); 90766dabbb6SChristoph Hellwig if (IS_ERR(folio)) { 908f43dc4dcSDave Chinner start_byte = ALIGN_DOWN(start_byte, PAGE_SIZE) + 909f43dc4dcSDave Chinner PAGE_SIZE; 910f43dc4dcSDave Chinner continue; 911f43dc4dcSDave Chinner } 912f43dc4dcSDave Chinner 913f43dc4dcSDave Chinner /* if dirty, punch up to offset */ 914f43dc4dcSDave Chinner if (folio_test_dirty(folio)) { 915f43dc4dcSDave Chinner if (start_byte > *punch_start_byte) { 916f43dc4dcSDave Chinner int error; 917f43dc4dcSDave Chinner 918f43dc4dcSDave Chinner error = punch(inode, *punch_start_byte, 919f43dc4dcSDave Chinner start_byte - *punch_start_byte); 920f43dc4dcSDave Chinner if (error) { 921f43dc4dcSDave Chinner folio_unlock(folio); 922f43dc4dcSDave Chinner folio_put(folio); 923f43dc4dcSDave Chinner return error; 924f43dc4dcSDave Chinner } 925f43dc4dcSDave Chinner } 926f43dc4dcSDave Chinner 927f43dc4dcSDave Chinner /* 928f43dc4dcSDave Chinner * Make sure the next punch start is correctly bound to 929f43dc4dcSDave Chinner * the end of this data range, not the end of the folio. 930f43dc4dcSDave Chinner */ 931f43dc4dcSDave Chinner *punch_start_byte = min_t(loff_t, end_byte, 932f43dc4dcSDave Chinner folio_next_index(folio) << PAGE_SHIFT); 933f43dc4dcSDave Chinner } 934f43dc4dcSDave Chinner 935f43dc4dcSDave Chinner /* move offset to start of next folio in range */ 936f43dc4dcSDave Chinner start_byte = folio_next_index(folio) << PAGE_SHIFT; 937f43dc4dcSDave Chinner folio_unlock(folio); 938f43dc4dcSDave Chinner folio_put(folio); 939f43dc4dcSDave Chinner } 940f43dc4dcSDave Chinner return 0; 941f43dc4dcSDave Chinner } 942f43dc4dcSDave Chinner 943f43dc4dcSDave Chinner /* 944f43dc4dcSDave Chinner * Punch out all the delalloc blocks in the range given except for those that 945f43dc4dcSDave Chinner * have dirty data still pending in the page cache - those are going to be 946f43dc4dcSDave Chinner * written and so must still retain the delalloc backing for writeback. 947f43dc4dcSDave Chinner * 948f43dc4dcSDave Chinner * As we are scanning the page cache for data, we don't need to reimplement the 949f43dc4dcSDave Chinner * wheel - mapping_seek_hole_data() does exactly what we need to identify the 950f43dc4dcSDave Chinner * start and end of data ranges correctly even for sub-folio block sizes. This 951f43dc4dcSDave Chinner * byte range based iteration is especially convenient because it means we 952f43dc4dcSDave Chinner * don't have to care about variable size folios, nor where the start or end of 953f43dc4dcSDave Chinner * the data range lies within a folio, if they lie within the same folio or even 954f43dc4dcSDave Chinner * if there are multiple discontiguous data ranges within the folio. 955f43dc4dcSDave Chinner * 956f43dc4dcSDave Chinner * It should be noted that mapping_seek_hole_data() is not aware of EOF, and so 957f43dc4dcSDave Chinner * can return data ranges that exist in the cache beyond EOF. e.g. a page fault 958f43dc4dcSDave Chinner * spanning EOF will initialise the post-EOF data to zeroes and mark it up to 959f43dc4dcSDave Chinner * date. A write page fault can then mark it dirty. If we then fail a write() 960f43dc4dcSDave Chinner * beyond EOF into that up to date cached range, we allocate a delalloc block 961f43dc4dcSDave Chinner * beyond EOF and then have to punch it out. Because the range is up to date, 962f43dc4dcSDave Chinner * mapping_seek_hole_data() will return it, and we will skip the punch because 963f43dc4dcSDave Chinner * the folio is dirty. THis is incorrect - we always need to punch out delalloc 964f43dc4dcSDave Chinner * beyond EOF in this case as writeback will never write back and covert that 965f43dc4dcSDave Chinner * delalloc block beyond EOF. Hence we limit the cached data scan range to EOF, 966f43dc4dcSDave Chinner * resulting in always punching out the range from the EOF to the end of the 967f43dc4dcSDave Chinner * range the iomap spans. 968f43dc4dcSDave Chinner * 969f43dc4dcSDave Chinner * Intervals are of the form [start_byte, end_byte) (i.e. open ended) because it 970f43dc4dcSDave Chinner * matches the intervals returned by mapping_seek_hole_data(). i.e. SEEK_DATA 971f43dc4dcSDave Chinner * returns the start of a data range (start_byte), and SEEK_HOLE(start_byte) 972f43dc4dcSDave Chinner * returns the end of the data range (data_end). Using closed intervals would 973f43dc4dcSDave Chinner * require sprinkling this code with magic "+ 1" and "- 1" arithmetic and expose 974f43dc4dcSDave Chinner * the code to subtle off-by-one bugs.... 975f43dc4dcSDave Chinner */ 976f43dc4dcSDave Chinner static int iomap_write_delalloc_release(struct inode *inode, 977f43dc4dcSDave Chinner loff_t start_byte, loff_t end_byte, 978f43dc4dcSDave Chinner int (*punch)(struct inode *inode, loff_t pos, loff_t length)) 979f43dc4dcSDave Chinner { 980f43dc4dcSDave Chinner loff_t punch_start_byte = start_byte; 981f43dc4dcSDave Chinner loff_t scan_end_byte = min(i_size_read(inode), end_byte); 982f43dc4dcSDave Chinner int error = 0; 983f43dc4dcSDave Chinner 984f43dc4dcSDave Chinner /* 985f43dc4dcSDave Chinner * Lock the mapping to avoid races with page faults re-instantiating 986f43dc4dcSDave Chinner * folios and dirtying them via ->page_mkwrite whilst we walk the 987f43dc4dcSDave Chinner * cache and perform delalloc extent removal. Failing to do this can 988f43dc4dcSDave Chinner * leave dirty pages with no space reservation in the cache. 989f43dc4dcSDave Chinner */ 990f43dc4dcSDave Chinner filemap_invalidate_lock(inode->i_mapping); 991f43dc4dcSDave Chinner while (start_byte < scan_end_byte) { 992f43dc4dcSDave Chinner loff_t data_end; 993f43dc4dcSDave Chinner 994f43dc4dcSDave Chinner start_byte = mapping_seek_hole_data(inode->i_mapping, 995f43dc4dcSDave Chinner start_byte, scan_end_byte, SEEK_DATA); 996f43dc4dcSDave Chinner /* 997f43dc4dcSDave Chinner * If there is no more data to scan, all that is left is to 998f43dc4dcSDave Chinner * punch out the remaining range. 999f43dc4dcSDave Chinner */ 1000f43dc4dcSDave Chinner if (start_byte == -ENXIO || start_byte == scan_end_byte) 1001f43dc4dcSDave Chinner break; 1002f43dc4dcSDave Chinner if (start_byte < 0) { 1003f43dc4dcSDave Chinner error = start_byte; 1004f43dc4dcSDave Chinner goto out_unlock; 1005f43dc4dcSDave Chinner } 1006f43dc4dcSDave Chinner WARN_ON_ONCE(start_byte < punch_start_byte); 1007f43dc4dcSDave Chinner WARN_ON_ONCE(start_byte > scan_end_byte); 1008f43dc4dcSDave Chinner 1009f43dc4dcSDave Chinner /* 1010f43dc4dcSDave Chinner * We find the end of this contiguous cached data range by 1011f43dc4dcSDave Chinner * seeking from start_byte to the beginning of the next hole. 1012f43dc4dcSDave Chinner */ 1013f43dc4dcSDave Chinner data_end = mapping_seek_hole_data(inode->i_mapping, start_byte, 1014f43dc4dcSDave Chinner scan_end_byte, SEEK_HOLE); 1015f43dc4dcSDave Chinner if (data_end < 0) { 1016f43dc4dcSDave Chinner error = data_end; 1017f43dc4dcSDave Chinner goto out_unlock; 1018f43dc4dcSDave Chinner } 1019f43dc4dcSDave Chinner WARN_ON_ONCE(data_end <= start_byte); 1020f43dc4dcSDave Chinner WARN_ON_ONCE(data_end > scan_end_byte); 1021f43dc4dcSDave Chinner 1022f43dc4dcSDave Chinner error = iomap_write_delalloc_scan(inode, &punch_start_byte, 1023f43dc4dcSDave Chinner start_byte, data_end, punch); 1024f43dc4dcSDave Chinner if (error) 1025f43dc4dcSDave Chinner goto out_unlock; 1026f43dc4dcSDave Chinner 1027f43dc4dcSDave Chinner /* The next data search starts at the end of this one. */ 1028f43dc4dcSDave Chinner start_byte = data_end; 1029f43dc4dcSDave Chinner } 1030f43dc4dcSDave Chinner 1031f43dc4dcSDave Chinner if (punch_start_byte < end_byte) 1032f43dc4dcSDave Chinner error = punch(inode, punch_start_byte, 1033f43dc4dcSDave Chinner end_byte - punch_start_byte); 1034f43dc4dcSDave Chinner out_unlock: 1035f43dc4dcSDave Chinner filemap_invalidate_unlock(inode->i_mapping); 1036f43dc4dcSDave Chinner return error; 1037f43dc4dcSDave Chinner } 1038f43dc4dcSDave Chinner 1039f43dc4dcSDave Chinner /* 10409c7babf9SDave Chinner * When a short write occurs, the filesystem may need to remove reserved space 10419c7babf9SDave Chinner * that was allocated in ->iomap_begin from it's ->iomap_end method. For 10429c7babf9SDave Chinner * filesystems that use delayed allocation, we need to punch out delalloc 10439c7babf9SDave Chinner * extents from the range that are not dirty in the page cache. As the write can 10449c7babf9SDave Chinner * race with page faults, there can be dirty pages over the delalloc extent 10459c7babf9SDave Chinner * outside the range of a short write but still within the delalloc extent 10469c7babf9SDave Chinner * allocated for this iomap. 10479c7babf9SDave Chinner * 10489c7babf9SDave Chinner * This function uses [start_byte, end_byte) intervals (i.e. open ended) to 1049f43dc4dcSDave Chinner * simplify range iterations. 1050f43dc4dcSDave Chinner * 1051f43dc4dcSDave Chinner * The punch() callback *must* only punch delalloc extents in the range passed 1052f43dc4dcSDave Chinner * to it. It must skip over all other types of extents in the range and leave 1053f43dc4dcSDave Chinner * them completely unchanged. It must do this punch atomically with respect to 1054f43dc4dcSDave Chinner * other extent modifications. 1055f43dc4dcSDave Chinner * 1056f43dc4dcSDave Chinner * The punch() callback may be called with a folio locked to prevent writeback 1057f43dc4dcSDave Chinner * extent allocation racing at the edge of the range we are currently punching. 1058f43dc4dcSDave Chinner * The locked folio may or may not cover the range being punched, so it is not 1059f43dc4dcSDave Chinner * safe for the punch() callback to lock folios itself. 1060f43dc4dcSDave Chinner * 1061f43dc4dcSDave Chinner * Lock order is: 1062f43dc4dcSDave Chinner * 1063f43dc4dcSDave Chinner * inode->i_rwsem (shared or exclusive) 1064f43dc4dcSDave Chinner * inode->i_mapping->invalidate_lock (exclusive) 1065f43dc4dcSDave Chinner * folio_lock() 1066f43dc4dcSDave Chinner * ->punch 1067f43dc4dcSDave Chinner * internal filesystem allocation lock 10689c7babf9SDave Chinner */ 10699c7babf9SDave Chinner int iomap_file_buffered_write_punch_delalloc(struct inode *inode, 10709c7babf9SDave Chinner struct iomap *iomap, loff_t pos, loff_t length, 10719c7babf9SDave Chinner ssize_t written, 10729c7babf9SDave Chinner int (*punch)(struct inode *inode, loff_t pos, loff_t length)) 10739c7babf9SDave Chinner { 10749c7babf9SDave Chinner loff_t start_byte; 10759c7babf9SDave Chinner loff_t end_byte; 10769c7babf9SDave Chinner int blocksize = i_blocksize(inode); 10779c7babf9SDave Chinner 10789c7babf9SDave Chinner if (iomap->type != IOMAP_DELALLOC) 10799c7babf9SDave Chinner return 0; 10809c7babf9SDave Chinner 10819c7babf9SDave Chinner /* If we didn't reserve the blocks, we're not allowed to punch them. */ 10829c7babf9SDave Chinner if (!(iomap->flags & IOMAP_F_NEW)) 10839c7babf9SDave Chinner return 0; 10849c7babf9SDave Chinner 10859c7babf9SDave Chinner /* 10869c7babf9SDave Chinner * start_byte refers to the first unused block after a short write. If 10879c7babf9SDave Chinner * nothing was written, round offset down to point at the first block in 10889c7babf9SDave Chinner * the range. 10899c7babf9SDave Chinner */ 10909c7babf9SDave Chinner if (unlikely(!written)) 10919c7babf9SDave Chinner start_byte = round_down(pos, blocksize); 10929c7babf9SDave Chinner else 10939c7babf9SDave Chinner start_byte = round_up(pos + written, blocksize); 10949c7babf9SDave Chinner end_byte = round_up(pos + length, blocksize); 10959c7babf9SDave Chinner 10969c7babf9SDave Chinner /* Nothing to do if we've written the entire delalloc extent */ 10979c7babf9SDave Chinner if (start_byte >= end_byte) 10989c7babf9SDave Chinner return 0; 10999c7babf9SDave Chinner 1100f43dc4dcSDave Chinner return iomap_write_delalloc_release(inode, start_byte, end_byte, 1101f43dc4dcSDave Chinner punch); 11029c7babf9SDave Chinner } 11039c7babf9SDave Chinner EXPORT_SYMBOL_GPL(iomap_file_buffered_write_punch_delalloc); 11049c7babf9SDave Chinner 11058fc274d1SChristoph Hellwig static loff_t iomap_unshare_iter(struct iomap_iter *iter) 1106afc51aaaSDarrick J. Wong { 11078fc274d1SChristoph Hellwig struct iomap *iomap = &iter->iomap; 1108fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 11098fc274d1SChristoph Hellwig loff_t pos = iter->pos; 11108fc274d1SChristoph Hellwig loff_t length = iomap_length(iter); 1111afc51aaaSDarrick J. Wong long status = 0; 1112d4ff3b2eSMatthew Wilcox (Oracle) loff_t written = 0; 1113afc51aaaSDarrick J. Wong 11143590c4d8SChristoph Hellwig /* don't bother with blocks that are not shared to start with */ 11153590c4d8SChristoph Hellwig if (!(iomap->flags & IOMAP_F_SHARED)) 11163590c4d8SChristoph Hellwig return length; 11173590c4d8SChristoph Hellwig /* don't bother with holes or unwritten extents */ 1118c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 11193590c4d8SChristoph Hellwig return length; 11203590c4d8SChristoph Hellwig 1121afc51aaaSDarrick J. Wong do { 112232a38a49SChristoph Hellwig unsigned long offset = offset_in_page(pos); 112332a38a49SChristoph Hellwig unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length); 1124bc6123a8SMatthew Wilcox (Oracle) struct folio *folio; 1125afc51aaaSDarrick J. Wong 1126bc6123a8SMatthew Wilcox (Oracle) status = iomap_write_begin(iter, pos, bytes, &folio); 1127afc51aaaSDarrick J. Wong if (unlikely(status)) 1128afc51aaaSDarrick J. Wong return status; 1129d7b64041SDave Chinner if (iter->iomap.flags & IOMAP_F_STALE) 1130d7b64041SDave Chinner break; 1131afc51aaaSDarrick J. Wong 1132bc6123a8SMatthew Wilcox (Oracle) status = iomap_write_end(iter, pos, bytes, bytes, folio); 1133afc51aaaSDarrick J. Wong if (WARN_ON_ONCE(status == 0)) 1134afc51aaaSDarrick J. Wong return -EIO; 1135afc51aaaSDarrick J. Wong 1136afc51aaaSDarrick J. Wong cond_resched(); 1137afc51aaaSDarrick J. Wong 1138afc51aaaSDarrick J. Wong pos += status; 1139afc51aaaSDarrick J. Wong written += status; 1140afc51aaaSDarrick J. Wong length -= status; 1141afc51aaaSDarrick J. Wong 11428fc274d1SChristoph Hellwig balance_dirty_pages_ratelimited(iter->inode->i_mapping); 1143afc51aaaSDarrick J. Wong } while (length); 1144afc51aaaSDarrick J. Wong 1145afc51aaaSDarrick J. Wong return written; 1146afc51aaaSDarrick J. Wong } 1147afc51aaaSDarrick J. Wong 1148afc51aaaSDarrick J. Wong int 11493590c4d8SChristoph Hellwig iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, 1150afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 1151afc51aaaSDarrick J. Wong { 11528fc274d1SChristoph Hellwig struct iomap_iter iter = { 11538fc274d1SChristoph Hellwig .inode = inode, 11548fc274d1SChristoph Hellwig .pos = pos, 11558fc274d1SChristoph Hellwig .len = len, 1156b74b1293SChristoph Hellwig .flags = IOMAP_WRITE | IOMAP_UNSHARE, 11578fc274d1SChristoph Hellwig }; 11588fc274d1SChristoph Hellwig int ret; 1159afc51aaaSDarrick J. Wong 11608fc274d1SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 11618fc274d1SChristoph Hellwig iter.processed = iomap_unshare_iter(&iter); 1162afc51aaaSDarrick J. Wong return ret; 1163afc51aaaSDarrick J. Wong } 11643590c4d8SChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_file_unshare); 1165afc51aaaSDarrick J. Wong 11662aa3048eSChristoph Hellwig static loff_t iomap_zero_iter(struct iomap_iter *iter, bool *did_zero) 1167afc51aaaSDarrick J. Wong { 1168fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 11692aa3048eSChristoph Hellwig loff_t pos = iter->pos; 11702aa3048eSChristoph Hellwig loff_t length = iomap_length(iter); 1171afc51aaaSDarrick J. Wong loff_t written = 0; 1172afc51aaaSDarrick J. Wong 1173afc51aaaSDarrick J. Wong /* already zeroed? we're done. */ 1174c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 117581ee8e52SMatthew Wilcox (Oracle) return length; 1176afc51aaaSDarrick J. Wong 1177afc51aaaSDarrick J. Wong do { 11784d7bd0ebSMatthew Wilcox (Oracle) struct folio *folio; 11794d7bd0ebSMatthew Wilcox (Oracle) int status; 11804d7bd0ebSMatthew Wilcox (Oracle) size_t offset; 11814d7bd0ebSMatthew Wilcox (Oracle) size_t bytes = min_t(u64, SIZE_MAX, length); 1182afc51aaaSDarrick J. Wong 11834d7bd0ebSMatthew Wilcox (Oracle) status = iomap_write_begin(iter, pos, bytes, &folio); 11844d7bd0ebSMatthew Wilcox (Oracle) if (status) 11854d7bd0ebSMatthew Wilcox (Oracle) return status; 1186d7b64041SDave Chinner if (iter->iomap.flags & IOMAP_F_STALE) 1187d7b64041SDave Chinner break; 11884d7bd0ebSMatthew Wilcox (Oracle) 11894d7bd0ebSMatthew Wilcox (Oracle) offset = offset_in_folio(folio, pos); 11904d7bd0ebSMatthew Wilcox (Oracle) if (bytes > folio_size(folio) - offset) 11914d7bd0ebSMatthew Wilcox (Oracle) bytes = folio_size(folio) - offset; 11924d7bd0ebSMatthew Wilcox (Oracle) 11934d7bd0ebSMatthew Wilcox (Oracle) folio_zero_range(folio, offset, bytes); 11944d7bd0ebSMatthew Wilcox (Oracle) folio_mark_accessed(folio); 11954d7bd0ebSMatthew Wilcox (Oracle) 11964d7bd0ebSMatthew Wilcox (Oracle) bytes = iomap_write_end(iter, pos, bytes, bytes, folio); 11974d7bd0ebSMatthew Wilcox (Oracle) if (WARN_ON_ONCE(bytes == 0)) 11984d7bd0ebSMatthew Wilcox (Oracle) return -EIO; 1199afc51aaaSDarrick J. Wong 1200afc51aaaSDarrick J. Wong pos += bytes; 120181ee8e52SMatthew Wilcox (Oracle) length -= bytes; 1202afc51aaaSDarrick J. Wong written += bytes; 120381ee8e52SMatthew Wilcox (Oracle) } while (length > 0); 1204afc51aaaSDarrick J. Wong 120598eb8d95SKaixu Xia if (did_zero) 120698eb8d95SKaixu Xia *did_zero = true; 1207afc51aaaSDarrick J. Wong return written; 1208afc51aaaSDarrick J. Wong } 1209afc51aaaSDarrick J. Wong 1210afc51aaaSDarrick J. Wong int 1211afc51aaaSDarrick J. Wong iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, 1212afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 1213afc51aaaSDarrick J. Wong { 12142aa3048eSChristoph Hellwig struct iomap_iter iter = { 12152aa3048eSChristoph Hellwig .inode = inode, 12162aa3048eSChristoph Hellwig .pos = pos, 12172aa3048eSChristoph Hellwig .len = len, 12182aa3048eSChristoph Hellwig .flags = IOMAP_ZERO, 12192aa3048eSChristoph Hellwig }; 12202aa3048eSChristoph Hellwig int ret; 1221afc51aaaSDarrick J. Wong 12222aa3048eSChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 12232aa3048eSChristoph Hellwig iter.processed = iomap_zero_iter(&iter, did_zero); 1224afc51aaaSDarrick J. Wong return ret; 1225afc51aaaSDarrick J. Wong } 1226afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_zero_range); 1227afc51aaaSDarrick J. Wong 1228afc51aaaSDarrick J. Wong int 1229afc51aaaSDarrick J. Wong iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, 1230afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 1231afc51aaaSDarrick J. Wong { 1232afc51aaaSDarrick J. Wong unsigned int blocksize = i_blocksize(inode); 1233afc51aaaSDarrick J. Wong unsigned int off = pos & (blocksize - 1); 1234afc51aaaSDarrick J. Wong 1235afc51aaaSDarrick J. Wong /* Block boundary? Nothing to do */ 1236afc51aaaSDarrick J. Wong if (!off) 1237afc51aaaSDarrick J. Wong return 0; 1238afc51aaaSDarrick J. Wong return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); 1239afc51aaaSDarrick J. Wong } 1240afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_truncate_page); 1241afc51aaaSDarrick J. Wong 1242ea0f843aSMatthew Wilcox (Oracle) static loff_t iomap_folio_mkwrite_iter(struct iomap_iter *iter, 1243ea0f843aSMatthew Wilcox (Oracle) struct folio *folio) 1244afc51aaaSDarrick J. Wong { 1245253564baSChristoph Hellwig loff_t length = iomap_length(iter); 1246afc51aaaSDarrick J. Wong int ret; 1247afc51aaaSDarrick J. Wong 1248253564baSChristoph Hellwig if (iter->iomap.flags & IOMAP_F_BUFFER_HEAD) { 1249d1bd0b4eSMatthew Wilcox (Oracle) ret = __block_write_begin_int(folio, iter->pos, length, NULL, 1250253564baSChristoph Hellwig &iter->iomap); 1251afc51aaaSDarrick J. Wong if (ret) 1252afc51aaaSDarrick J. Wong return ret; 1253ea0f843aSMatthew Wilcox (Oracle) block_commit_write(&folio->page, 0, length); 1254afc51aaaSDarrick J. Wong } else { 1255ea0f843aSMatthew Wilcox (Oracle) WARN_ON_ONCE(!folio_test_uptodate(folio)); 1256ea0f843aSMatthew Wilcox (Oracle) folio_mark_dirty(folio); 1257afc51aaaSDarrick J. Wong } 1258afc51aaaSDarrick J. Wong 1259afc51aaaSDarrick J. Wong return length; 1260afc51aaaSDarrick J. Wong } 1261afc51aaaSDarrick J. Wong 1262afc51aaaSDarrick J. Wong vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) 1263afc51aaaSDarrick J. Wong { 1264253564baSChristoph Hellwig struct iomap_iter iter = { 1265253564baSChristoph Hellwig .inode = file_inode(vmf->vma->vm_file), 1266253564baSChristoph Hellwig .flags = IOMAP_WRITE | IOMAP_FAULT, 1267253564baSChristoph Hellwig }; 1268ea0f843aSMatthew Wilcox (Oracle) struct folio *folio = page_folio(vmf->page); 1269afc51aaaSDarrick J. Wong ssize_t ret; 1270afc51aaaSDarrick J. Wong 1271ea0f843aSMatthew Wilcox (Oracle) folio_lock(folio); 1272ea0f843aSMatthew Wilcox (Oracle) ret = folio_mkwrite_check_truncate(folio, iter.inode); 1273243145bcSAndreas Gruenbacher if (ret < 0) 1274afc51aaaSDarrick J. Wong goto out_unlock; 1275ea0f843aSMatthew Wilcox (Oracle) iter.pos = folio_pos(folio); 1276253564baSChristoph Hellwig iter.len = ret; 1277253564baSChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 1278ea0f843aSMatthew Wilcox (Oracle) iter.processed = iomap_folio_mkwrite_iter(&iter, folio); 1279afc51aaaSDarrick J. Wong 1280253564baSChristoph Hellwig if (ret < 0) 1281afc51aaaSDarrick J. Wong goto out_unlock; 1282ea0f843aSMatthew Wilcox (Oracle) folio_wait_stable(folio); 1283afc51aaaSDarrick J. Wong return VM_FAULT_LOCKED; 1284afc51aaaSDarrick J. Wong out_unlock: 1285ea0f843aSMatthew Wilcox (Oracle) folio_unlock(folio); 1286afc51aaaSDarrick J. Wong return block_page_mkwrite_return(ret); 1287afc51aaaSDarrick J. Wong } 1288afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_page_mkwrite); 1289598ecfbaSChristoph Hellwig 12908ffd74e9SMatthew Wilcox (Oracle) static void iomap_finish_folio_write(struct inode *inode, struct folio *folio, 12918ffd74e9SMatthew Wilcox (Oracle) size_t len, int error) 1292598ecfbaSChristoph Hellwig { 129395c4cd05SMatthew Wilcox (Oracle) struct iomap_page *iop = to_iomap_page(folio); 1294598ecfbaSChristoph Hellwig 1295598ecfbaSChristoph Hellwig if (error) { 12968ffd74e9SMatthew Wilcox (Oracle) folio_set_error(folio); 1297b69eea82SDarrick J. Wong mapping_set_error(inode->i_mapping, error); 1298598ecfbaSChristoph Hellwig } 1299598ecfbaSChristoph Hellwig 13008ffd74e9SMatthew Wilcox (Oracle) WARN_ON_ONCE(i_blocks_per_folio(inode, folio) > 1 && !iop); 13010fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) <= 0); 1302598ecfbaSChristoph Hellwig 13030fb2d720SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(len, &iop->write_bytes_pending)) 13048ffd74e9SMatthew Wilcox (Oracle) folio_end_writeback(folio); 1305598ecfbaSChristoph Hellwig } 1306598ecfbaSChristoph Hellwig 1307598ecfbaSChristoph Hellwig /* 1308598ecfbaSChristoph Hellwig * We're now finished for good with this ioend structure. Update the page 1309598ecfbaSChristoph Hellwig * state, release holds on bios, and finally free up memory. Do not use the 1310598ecfbaSChristoph Hellwig * ioend after this. 1311598ecfbaSChristoph Hellwig */ 1312ebb7fb15SDave Chinner static u32 1313598ecfbaSChristoph Hellwig iomap_finish_ioend(struct iomap_ioend *ioend, int error) 1314598ecfbaSChristoph Hellwig { 1315598ecfbaSChristoph Hellwig struct inode *inode = ioend->io_inode; 1316598ecfbaSChristoph Hellwig struct bio *bio = &ioend->io_inline_bio; 1317598ecfbaSChristoph Hellwig struct bio *last = ioend->io_bio, *next; 1318598ecfbaSChristoph Hellwig u64 start = bio->bi_iter.bi_sector; 1319c275779fSZorro Lang loff_t offset = ioend->io_offset; 1320598ecfbaSChristoph Hellwig bool quiet = bio_flagged(bio, BIO_QUIET); 1321ebb7fb15SDave Chinner u32 folio_count = 0; 1322598ecfbaSChristoph Hellwig 1323598ecfbaSChristoph Hellwig for (bio = &ioend->io_inline_bio; bio; bio = next) { 13248ffd74e9SMatthew Wilcox (Oracle) struct folio_iter fi; 1325598ecfbaSChristoph Hellwig 1326598ecfbaSChristoph Hellwig /* 1327598ecfbaSChristoph Hellwig * For the last bio, bi_private points to the ioend, so we 1328598ecfbaSChristoph Hellwig * need to explicitly end the iteration here. 1329598ecfbaSChristoph Hellwig */ 1330598ecfbaSChristoph Hellwig if (bio == last) 1331598ecfbaSChristoph Hellwig next = NULL; 1332598ecfbaSChristoph Hellwig else 1333598ecfbaSChristoph Hellwig next = bio->bi_private; 1334598ecfbaSChristoph Hellwig 13358ffd74e9SMatthew Wilcox (Oracle) /* walk all folios in bio, ending page IO on them */ 1336ebb7fb15SDave Chinner bio_for_each_folio_all(fi, bio) { 13378ffd74e9SMatthew Wilcox (Oracle) iomap_finish_folio_write(inode, fi.folio, fi.length, 13388ffd74e9SMatthew Wilcox (Oracle) error); 1339ebb7fb15SDave Chinner folio_count++; 1340ebb7fb15SDave Chinner } 1341598ecfbaSChristoph Hellwig bio_put(bio); 1342598ecfbaSChristoph Hellwig } 1343c275779fSZorro Lang /* The ioend has been freed by bio_put() */ 1344598ecfbaSChristoph Hellwig 1345598ecfbaSChristoph Hellwig if (unlikely(error && !quiet)) { 1346598ecfbaSChristoph Hellwig printk_ratelimited(KERN_ERR 13479cd0ed63SDarrick J. Wong "%s: writeback error on inode %lu, offset %lld, sector %llu", 1348c275779fSZorro Lang inode->i_sb->s_id, inode->i_ino, offset, start); 1349598ecfbaSChristoph Hellwig } 1350ebb7fb15SDave Chinner return folio_count; 1351598ecfbaSChristoph Hellwig } 1352598ecfbaSChristoph Hellwig 1353ebb7fb15SDave Chinner /* 1354ebb7fb15SDave Chinner * Ioend completion routine for merged bios. This can only be called from task 1355ebb7fb15SDave Chinner * contexts as merged ioends can be of unbound length. Hence we have to break up 1356ebb7fb15SDave Chinner * the writeback completions into manageable chunks to avoid long scheduler 1357ebb7fb15SDave Chinner * holdoffs. We aim to keep scheduler holdoffs down below 10ms so that we get 1358ebb7fb15SDave Chinner * good batch processing throughput without creating adverse scheduler latency 1359ebb7fb15SDave Chinner * conditions. 1360ebb7fb15SDave Chinner */ 1361598ecfbaSChristoph Hellwig void 1362598ecfbaSChristoph Hellwig iomap_finish_ioends(struct iomap_ioend *ioend, int error) 1363598ecfbaSChristoph Hellwig { 1364598ecfbaSChristoph Hellwig struct list_head tmp; 1365ebb7fb15SDave Chinner u32 completions; 1366ebb7fb15SDave Chinner 1367ebb7fb15SDave Chinner might_sleep(); 1368598ecfbaSChristoph Hellwig 1369598ecfbaSChristoph Hellwig list_replace_init(&ioend->io_list, &tmp); 1370ebb7fb15SDave Chinner completions = iomap_finish_ioend(ioend, error); 1371598ecfbaSChristoph Hellwig 1372598ecfbaSChristoph Hellwig while (!list_empty(&tmp)) { 1373ebb7fb15SDave Chinner if (completions > IOEND_BATCH_SIZE * 8) { 1374ebb7fb15SDave Chinner cond_resched(); 1375ebb7fb15SDave Chinner completions = 0; 1376ebb7fb15SDave Chinner } 1377598ecfbaSChristoph Hellwig ioend = list_first_entry(&tmp, struct iomap_ioend, io_list); 1378598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1379ebb7fb15SDave Chinner completions += iomap_finish_ioend(ioend, error); 1380598ecfbaSChristoph Hellwig } 1381598ecfbaSChristoph Hellwig } 1382598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_finish_ioends); 1383598ecfbaSChristoph Hellwig 1384598ecfbaSChristoph Hellwig /* 1385598ecfbaSChristoph Hellwig * We can merge two adjacent ioends if they have the same set of work to do. 1386598ecfbaSChristoph Hellwig */ 1387598ecfbaSChristoph Hellwig static bool 1388598ecfbaSChristoph Hellwig iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next) 1389598ecfbaSChristoph Hellwig { 1390598ecfbaSChristoph Hellwig if (ioend->io_bio->bi_status != next->io_bio->bi_status) 1391598ecfbaSChristoph Hellwig return false; 1392598ecfbaSChristoph Hellwig if ((ioend->io_flags & IOMAP_F_SHARED) ^ 1393598ecfbaSChristoph Hellwig (next->io_flags & IOMAP_F_SHARED)) 1394598ecfbaSChristoph Hellwig return false; 1395598ecfbaSChristoph Hellwig if ((ioend->io_type == IOMAP_UNWRITTEN) ^ 1396598ecfbaSChristoph Hellwig (next->io_type == IOMAP_UNWRITTEN)) 1397598ecfbaSChristoph Hellwig return false; 1398598ecfbaSChristoph Hellwig if (ioend->io_offset + ioend->io_size != next->io_offset) 1399598ecfbaSChristoph Hellwig return false; 1400ebb7fb15SDave Chinner /* 1401ebb7fb15SDave Chinner * Do not merge physically discontiguous ioends. The filesystem 1402ebb7fb15SDave Chinner * completion functions will have to iterate the physical 1403ebb7fb15SDave Chinner * discontiguities even if we merge the ioends at a logical level, so 1404ebb7fb15SDave Chinner * we don't gain anything by merging physical discontiguities here. 1405ebb7fb15SDave Chinner * 1406ebb7fb15SDave Chinner * We cannot use bio->bi_iter.bi_sector here as it is modified during 1407ebb7fb15SDave Chinner * submission so does not point to the start sector of the bio at 1408ebb7fb15SDave Chinner * completion. 1409ebb7fb15SDave Chinner */ 1410ebb7fb15SDave Chinner if (ioend->io_sector + (ioend->io_size >> 9) != next->io_sector) 1411ebb7fb15SDave Chinner return false; 1412598ecfbaSChristoph Hellwig return true; 1413598ecfbaSChristoph Hellwig } 1414598ecfbaSChristoph Hellwig 1415598ecfbaSChristoph Hellwig void 14166e552494SBrian Foster iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends) 1417598ecfbaSChristoph Hellwig { 1418598ecfbaSChristoph Hellwig struct iomap_ioend *next; 1419598ecfbaSChristoph Hellwig 1420598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1421598ecfbaSChristoph Hellwig 1422598ecfbaSChristoph Hellwig while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend, 1423598ecfbaSChristoph Hellwig io_list))) { 1424598ecfbaSChristoph Hellwig if (!iomap_ioend_can_merge(ioend, next)) 1425598ecfbaSChristoph Hellwig break; 1426598ecfbaSChristoph Hellwig list_move_tail(&next->io_list, &ioend->io_list); 1427598ecfbaSChristoph Hellwig ioend->io_size += next->io_size; 1428598ecfbaSChristoph Hellwig } 1429598ecfbaSChristoph Hellwig } 1430598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_ioend_try_merge); 1431598ecfbaSChristoph Hellwig 1432598ecfbaSChristoph Hellwig static int 14334f0f586bSSami Tolvanen iomap_ioend_compare(void *priv, const struct list_head *a, 14344f0f586bSSami Tolvanen const struct list_head *b) 1435598ecfbaSChristoph Hellwig { 1436b3d423ecSChristoph Hellwig struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list); 1437b3d423ecSChristoph Hellwig struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list); 1438598ecfbaSChristoph Hellwig 1439598ecfbaSChristoph Hellwig if (ia->io_offset < ib->io_offset) 1440598ecfbaSChristoph Hellwig return -1; 1441b3d423ecSChristoph Hellwig if (ia->io_offset > ib->io_offset) 1442598ecfbaSChristoph Hellwig return 1; 1443598ecfbaSChristoph Hellwig return 0; 1444598ecfbaSChristoph Hellwig } 1445598ecfbaSChristoph Hellwig 1446598ecfbaSChristoph Hellwig void 1447598ecfbaSChristoph Hellwig iomap_sort_ioends(struct list_head *ioend_list) 1448598ecfbaSChristoph Hellwig { 1449598ecfbaSChristoph Hellwig list_sort(NULL, ioend_list, iomap_ioend_compare); 1450598ecfbaSChristoph Hellwig } 1451598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_sort_ioends); 1452598ecfbaSChristoph Hellwig 1453598ecfbaSChristoph Hellwig static void iomap_writepage_end_bio(struct bio *bio) 1454598ecfbaSChristoph Hellwig { 1455598ecfbaSChristoph Hellwig struct iomap_ioend *ioend = bio->bi_private; 1456598ecfbaSChristoph Hellwig 1457598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status)); 1458598ecfbaSChristoph Hellwig } 1459598ecfbaSChristoph Hellwig 1460598ecfbaSChristoph Hellwig /* 1461598ecfbaSChristoph Hellwig * Submit the final bio for an ioend. 1462598ecfbaSChristoph Hellwig * 1463598ecfbaSChristoph Hellwig * If @error is non-zero, it means that we have a situation where some part of 1464f1f264b4SAndreas Gruenbacher * the submission process has failed after we've marked pages for writeback 1465598ecfbaSChristoph Hellwig * and unlocked them. In this situation, we need to fail the bio instead of 1466598ecfbaSChristoph Hellwig * submitting it. This typically only happens on a filesystem shutdown. 1467598ecfbaSChristoph Hellwig */ 1468598ecfbaSChristoph Hellwig static int 1469598ecfbaSChristoph Hellwig iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend, 1470598ecfbaSChristoph Hellwig int error) 1471598ecfbaSChristoph Hellwig { 1472598ecfbaSChristoph Hellwig ioend->io_bio->bi_private = ioend; 1473598ecfbaSChristoph Hellwig ioend->io_bio->bi_end_io = iomap_writepage_end_bio; 1474598ecfbaSChristoph Hellwig 1475598ecfbaSChristoph Hellwig if (wpc->ops->prepare_ioend) 1476598ecfbaSChristoph Hellwig error = wpc->ops->prepare_ioend(ioend, error); 1477598ecfbaSChristoph Hellwig if (error) { 1478598ecfbaSChristoph Hellwig /* 1479f1f264b4SAndreas Gruenbacher * If we're failing the IO now, just mark the ioend with an 1480598ecfbaSChristoph Hellwig * error and finish it. This will run IO completion immediately 1481598ecfbaSChristoph Hellwig * as there is only one reference to the ioend at this point in 1482598ecfbaSChristoph Hellwig * time. 1483598ecfbaSChristoph Hellwig */ 1484598ecfbaSChristoph Hellwig ioend->io_bio->bi_status = errno_to_blk_status(error); 1485598ecfbaSChristoph Hellwig bio_endio(ioend->io_bio); 1486598ecfbaSChristoph Hellwig return error; 1487598ecfbaSChristoph Hellwig } 1488598ecfbaSChristoph Hellwig 1489598ecfbaSChristoph Hellwig submit_bio(ioend->io_bio); 1490598ecfbaSChristoph Hellwig return 0; 1491598ecfbaSChristoph Hellwig } 1492598ecfbaSChristoph Hellwig 1493598ecfbaSChristoph Hellwig static struct iomap_ioend * 1494598ecfbaSChristoph Hellwig iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc, 1495598ecfbaSChristoph Hellwig loff_t offset, sector_t sector, struct writeback_control *wbc) 1496598ecfbaSChristoph Hellwig { 1497598ecfbaSChristoph Hellwig struct iomap_ioend *ioend; 1498598ecfbaSChristoph Hellwig struct bio *bio; 1499598ecfbaSChristoph Hellwig 1500609be106SChristoph Hellwig bio = bio_alloc_bioset(wpc->iomap.bdev, BIO_MAX_VECS, 1501609be106SChristoph Hellwig REQ_OP_WRITE | wbc_to_write_flags(wbc), 1502609be106SChristoph Hellwig GFP_NOFS, &iomap_ioend_bioset); 1503598ecfbaSChristoph Hellwig bio->bi_iter.bi_sector = sector; 1504598ecfbaSChristoph Hellwig wbc_init_bio(wbc, bio); 1505598ecfbaSChristoph Hellwig 1506598ecfbaSChristoph Hellwig ioend = container_of(bio, struct iomap_ioend, io_inline_bio); 1507598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1508598ecfbaSChristoph Hellwig ioend->io_type = wpc->iomap.type; 1509598ecfbaSChristoph Hellwig ioend->io_flags = wpc->iomap.flags; 1510598ecfbaSChristoph Hellwig ioend->io_inode = inode; 1511598ecfbaSChristoph Hellwig ioend->io_size = 0; 1512ebb7fb15SDave Chinner ioend->io_folios = 0; 1513598ecfbaSChristoph Hellwig ioend->io_offset = offset; 1514598ecfbaSChristoph Hellwig ioend->io_bio = bio; 1515ebb7fb15SDave Chinner ioend->io_sector = sector; 1516598ecfbaSChristoph Hellwig return ioend; 1517598ecfbaSChristoph Hellwig } 1518598ecfbaSChristoph Hellwig 1519598ecfbaSChristoph Hellwig /* 1520598ecfbaSChristoph Hellwig * Allocate a new bio, and chain the old bio to the new one. 1521598ecfbaSChristoph Hellwig * 1522f1f264b4SAndreas Gruenbacher * Note that we have to perform the chaining in this unintuitive order 1523598ecfbaSChristoph Hellwig * so that the bi_private linkage is set up in the right direction for the 1524598ecfbaSChristoph Hellwig * traversal in iomap_finish_ioend(). 1525598ecfbaSChristoph Hellwig */ 1526598ecfbaSChristoph Hellwig static struct bio * 1527598ecfbaSChristoph Hellwig iomap_chain_bio(struct bio *prev) 1528598ecfbaSChristoph Hellwig { 1529598ecfbaSChristoph Hellwig struct bio *new; 1530598ecfbaSChristoph Hellwig 153107888c66SChristoph Hellwig new = bio_alloc(prev->bi_bdev, BIO_MAX_VECS, prev->bi_opf, GFP_NOFS); 153207888c66SChristoph Hellwig bio_clone_blkg_association(new, prev); 1533598ecfbaSChristoph Hellwig new->bi_iter.bi_sector = bio_end_sector(prev); 1534598ecfbaSChristoph Hellwig 1535598ecfbaSChristoph Hellwig bio_chain(prev, new); 1536598ecfbaSChristoph Hellwig bio_get(prev); /* for iomap_finish_ioend */ 1537598ecfbaSChristoph Hellwig submit_bio(prev); 1538598ecfbaSChristoph Hellwig return new; 1539598ecfbaSChristoph Hellwig } 1540598ecfbaSChristoph Hellwig 1541598ecfbaSChristoph Hellwig static bool 1542598ecfbaSChristoph Hellwig iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset, 1543598ecfbaSChristoph Hellwig sector_t sector) 1544598ecfbaSChristoph Hellwig { 1545598ecfbaSChristoph Hellwig if ((wpc->iomap.flags & IOMAP_F_SHARED) != 1546598ecfbaSChristoph Hellwig (wpc->ioend->io_flags & IOMAP_F_SHARED)) 1547598ecfbaSChristoph Hellwig return false; 1548598ecfbaSChristoph Hellwig if (wpc->iomap.type != wpc->ioend->io_type) 1549598ecfbaSChristoph Hellwig return false; 1550598ecfbaSChristoph Hellwig if (offset != wpc->ioend->io_offset + wpc->ioend->io_size) 1551598ecfbaSChristoph Hellwig return false; 1552598ecfbaSChristoph Hellwig if (sector != bio_end_sector(wpc->ioend->io_bio)) 1553598ecfbaSChristoph Hellwig return false; 1554ebb7fb15SDave Chinner /* 1555ebb7fb15SDave Chinner * Limit ioend bio chain lengths to minimise IO completion latency. This 1556ebb7fb15SDave Chinner * also prevents long tight loops ending page writeback on all the 1557ebb7fb15SDave Chinner * folios in the ioend. 1558ebb7fb15SDave Chinner */ 1559ebb7fb15SDave Chinner if (wpc->ioend->io_folios >= IOEND_BATCH_SIZE) 1560ebb7fb15SDave Chinner return false; 1561598ecfbaSChristoph Hellwig return true; 1562598ecfbaSChristoph Hellwig } 1563598ecfbaSChristoph Hellwig 1564598ecfbaSChristoph Hellwig /* 1565598ecfbaSChristoph Hellwig * Test to see if we have an existing ioend structure that we could append to 1566f1f264b4SAndreas Gruenbacher * first; otherwise finish off the current ioend and start another. 1567598ecfbaSChristoph Hellwig */ 1568598ecfbaSChristoph Hellwig static void 1569e735c007SMatthew Wilcox (Oracle) iomap_add_to_ioend(struct inode *inode, loff_t pos, struct folio *folio, 1570598ecfbaSChristoph Hellwig struct iomap_page *iop, struct iomap_writepage_ctx *wpc, 1571598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct list_head *iolist) 1572598ecfbaSChristoph Hellwig { 1573e735c007SMatthew Wilcox (Oracle) sector_t sector = iomap_sector(&wpc->iomap, pos); 1574598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1575e735c007SMatthew Wilcox (Oracle) size_t poff = offset_in_folio(folio, pos); 1576598ecfbaSChristoph Hellwig 1577e735c007SMatthew Wilcox (Oracle) if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, pos, sector)) { 1578598ecfbaSChristoph Hellwig if (wpc->ioend) 1579598ecfbaSChristoph Hellwig list_add(&wpc->ioend->io_list, iolist); 1580e735c007SMatthew Wilcox (Oracle) wpc->ioend = iomap_alloc_ioend(inode, wpc, pos, sector, wbc); 1581598ecfbaSChristoph Hellwig } 1582598ecfbaSChristoph Hellwig 1583e735c007SMatthew Wilcox (Oracle) if (!bio_add_folio(wpc->ioend->io_bio, folio, len, poff)) { 1584c1b79f11SChristoph Hellwig wpc->ioend->io_bio = iomap_chain_bio(wpc->ioend->io_bio); 1585*c2478469SJohannes Thumshirn bio_add_folio_nofail(wpc->ioend->io_bio, folio, len, poff); 1586c1b79f11SChristoph Hellwig } 1587c1b79f11SChristoph Hellwig 15880fb2d720SMatthew Wilcox (Oracle) if (iop) 15890fb2d720SMatthew Wilcox (Oracle) atomic_add(len, &iop->write_bytes_pending); 1590598ecfbaSChristoph Hellwig wpc->ioend->io_size += len; 1591e735c007SMatthew Wilcox (Oracle) wbc_account_cgroup_owner(wbc, &folio->page, len); 1592598ecfbaSChristoph Hellwig } 1593598ecfbaSChristoph Hellwig 1594598ecfbaSChristoph Hellwig /* 1595598ecfbaSChristoph Hellwig * We implement an immediate ioend submission policy here to avoid needing to 1596598ecfbaSChristoph Hellwig * chain multiple ioends and hence nest mempool allocations which can violate 1597f1f264b4SAndreas Gruenbacher * the forward progress guarantees we need to provide. The current ioend we're 1598f1f264b4SAndreas Gruenbacher * adding blocks to is cached in the writepage context, and if the new block 1599f1f264b4SAndreas Gruenbacher * doesn't append to the cached ioend, it will create a new ioend and cache that 1600598ecfbaSChristoph Hellwig * instead. 1601598ecfbaSChristoph Hellwig * 1602598ecfbaSChristoph Hellwig * If a new ioend is created and cached, the old ioend is returned and queued 1603598ecfbaSChristoph Hellwig * locally for submission once the entire page is processed or an error has been 1604598ecfbaSChristoph Hellwig * detected. While ioends are submitted immediately after they are completed, 1605598ecfbaSChristoph Hellwig * batching optimisations are provided by higher level block plugging. 1606598ecfbaSChristoph Hellwig * 1607598ecfbaSChristoph Hellwig * At the end of a writeback pass, there will be a cached ioend remaining on the 1608598ecfbaSChristoph Hellwig * writepage context that the caller will need to submit. 1609598ecfbaSChristoph Hellwig */ 1610598ecfbaSChristoph Hellwig static int 1611598ecfbaSChristoph Hellwig iomap_writepage_map(struct iomap_writepage_ctx *wpc, 1612598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct inode *inode, 1613e735c007SMatthew Wilcox (Oracle) struct folio *folio, u64 end_pos) 1614598ecfbaSChristoph Hellwig { 16159753b868SStefan Roesch struct iomap_page *iop = iomap_page_create(inode, folio, 0); 1616598ecfbaSChristoph Hellwig struct iomap_ioend *ioend, *next; 1617598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 161892655036SMatthew Wilcox (Oracle) unsigned nblocks = i_blocks_per_folio(inode, folio); 161992655036SMatthew Wilcox (Oracle) u64 pos = folio_pos(folio); 1620598ecfbaSChristoph Hellwig int error = 0, count = 0, i; 1621598ecfbaSChristoph Hellwig LIST_HEAD(submit_list); 1622598ecfbaSChristoph Hellwig 16230fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) != 0); 1624598ecfbaSChristoph Hellwig 1625598ecfbaSChristoph Hellwig /* 162692655036SMatthew Wilcox (Oracle) * Walk through the folio to find areas to write back. If we 162792655036SMatthew Wilcox (Oracle) * run off the end of the current map or find the current map 162892655036SMatthew Wilcox (Oracle) * invalid, grab a new one. 1629598ecfbaSChristoph Hellwig */ 163092655036SMatthew Wilcox (Oracle) for (i = 0; i < nblocks && pos < end_pos; i++, pos += len) { 1631598ecfbaSChristoph Hellwig if (iop && !test_bit(i, iop->uptodate)) 1632598ecfbaSChristoph Hellwig continue; 1633598ecfbaSChristoph Hellwig 163492655036SMatthew Wilcox (Oracle) error = wpc->ops->map_blocks(wpc, inode, pos); 1635598ecfbaSChristoph Hellwig if (error) 1636598ecfbaSChristoph Hellwig break; 1637adc9c2e5SDarrick J. Wong trace_iomap_writepage_map(inode, &wpc->iomap); 16383e19e6f3SChristoph Hellwig if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE)) 16393e19e6f3SChristoph Hellwig continue; 1640598ecfbaSChristoph Hellwig if (wpc->iomap.type == IOMAP_HOLE) 1641598ecfbaSChristoph Hellwig continue; 1642e735c007SMatthew Wilcox (Oracle) iomap_add_to_ioend(inode, pos, folio, iop, wpc, wbc, 1643598ecfbaSChristoph Hellwig &submit_list); 1644598ecfbaSChristoph Hellwig count++; 1645598ecfbaSChristoph Hellwig } 1646ebb7fb15SDave Chinner if (count) 1647ebb7fb15SDave Chinner wpc->ioend->io_folios++; 1648598ecfbaSChristoph Hellwig 1649598ecfbaSChristoph Hellwig WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list)); 1650e735c007SMatthew Wilcox (Oracle) WARN_ON_ONCE(!folio_test_locked(folio)); 1651e735c007SMatthew Wilcox (Oracle) WARN_ON_ONCE(folio_test_writeback(folio)); 1652e735c007SMatthew Wilcox (Oracle) WARN_ON_ONCE(folio_test_dirty(folio)); 1653598ecfbaSChristoph Hellwig 1654598ecfbaSChristoph Hellwig /* 1655598ecfbaSChristoph Hellwig * We cannot cancel the ioend directly here on error. We may have 1656598ecfbaSChristoph Hellwig * already set other pages under writeback and hence we have to run I/O 1657598ecfbaSChristoph Hellwig * completion to mark the error state of the pages under writeback 1658598ecfbaSChristoph Hellwig * appropriately. 1659598ecfbaSChristoph Hellwig */ 1660598ecfbaSChristoph Hellwig if (unlikely(error)) { 1661598ecfbaSChristoph Hellwig /* 1662763e4cdcSBrian Foster * Let the filesystem know what portion of the current page 1663f1f264b4SAndreas Gruenbacher * failed to map. If the page hasn't been added to ioend, it 1664763e4cdcSBrian Foster * won't be affected by I/O completion and we must unlock it 1665763e4cdcSBrian Foster * now. 1666598ecfbaSChristoph Hellwig */ 16676e478521SMatthew Wilcox (Oracle) if (wpc->ops->discard_folio) 166892655036SMatthew Wilcox (Oracle) wpc->ops->discard_folio(folio, pos); 1669763e4cdcSBrian Foster if (!count) { 1670e735c007SMatthew Wilcox (Oracle) folio_unlock(folio); 1671598ecfbaSChristoph Hellwig goto done; 1672598ecfbaSChristoph Hellwig } 1673598ecfbaSChristoph Hellwig } 1674598ecfbaSChristoph Hellwig 1675e735c007SMatthew Wilcox (Oracle) folio_start_writeback(folio); 1676e735c007SMatthew Wilcox (Oracle) folio_unlock(folio); 1677598ecfbaSChristoph Hellwig 1678598ecfbaSChristoph Hellwig /* 1679f1f264b4SAndreas Gruenbacher * Preserve the original error if there was one; catch 1680598ecfbaSChristoph Hellwig * submission errors here and propagate into subsequent ioend 1681598ecfbaSChristoph Hellwig * submissions. 1682598ecfbaSChristoph Hellwig */ 1683598ecfbaSChristoph Hellwig list_for_each_entry_safe(ioend, next, &submit_list, io_list) { 1684598ecfbaSChristoph Hellwig int error2; 1685598ecfbaSChristoph Hellwig 1686598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1687598ecfbaSChristoph Hellwig error2 = iomap_submit_ioend(wpc, ioend, error); 1688598ecfbaSChristoph Hellwig if (error2 && !error) 1689598ecfbaSChristoph Hellwig error = error2; 1690598ecfbaSChristoph Hellwig } 1691598ecfbaSChristoph Hellwig 1692598ecfbaSChristoph Hellwig /* 1693598ecfbaSChristoph Hellwig * We can end up here with no error and nothing to write only if we race 1694598ecfbaSChristoph Hellwig * with a partial page truncate on a sub-page block sized filesystem. 1695598ecfbaSChristoph Hellwig */ 1696598ecfbaSChristoph Hellwig if (!count) 1697e735c007SMatthew Wilcox (Oracle) folio_end_writeback(folio); 1698598ecfbaSChristoph Hellwig done: 16993d5f3ba1SDarrick J. Wong mapping_set_error(inode->i_mapping, error); 1700598ecfbaSChristoph Hellwig return error; 1701598ecfbaSChristoph Hellwig } 1702598ecfbaSChristoph Hellwig 1703598ecfbaSChristoph Hellwig /* 1704598ecfbaSChristoph Hellwig * Write out a dirty page. 1705598ecfbaSChristoph Hellwig * 1706f1f264b4SAndreas Gruenbacher * For delalloc space on the page, we need to allocate space and flush it. 1707f1f264b4SAndreas Gruenbacher * For unwritten space on the page, we need to start the conversion to 1708598ecfbaSChristoph Hellwig * regular allocated space. 1709598ecfbaSChristoph Hellwig */ 1710d585bdbeSMatthew Wilcox (Oracle) static int iomap_do_writepage(struct folio *folio, 1711d585bdbeSMatthew Wilcox (Oracle) struct writeback_control *wbc, void *data) 1712598ecfbaSChristoph Hellwig { 1713598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc = data; 1714e735c007SMatthew Wilcox (Oracle) struct inode *inode = folio->mapping->host; 171581d4782aSMatthew Wilcox (Oracle) u64 end_pos, isize; 1716598ecfbaSChristoph Hellwig 1717e735c007SMatthew Wilcox (Oracle) trace_iomap_writepage(inode, folio_pos(folio), folio_size(folio)); 1718598ecfbaSChristoph Hellwig 1719598ecfbaSChristoph Hellwig /* 1720e735c007SMatthew Wilcox (Oracle) * Refuse to write the folio out if we're called from reclaim context. 1721598ecfbaSChristoph Hellwig * 1722598ecfbaSChristoph Hellwig * This avoids stack overflows when called from deeply used stacks in 1723598ecfbaSChristoph Hellwig * random callers for direct reclaim or memcg reclaim. We explicitly 1724598ecfbaSChristoph Hellwig * allow reclaim from kswapd as the stack usage there is relatively low. 1725598ecfbaSChristoph Hellwig * 1726598ecfbaSChristoph Hellwig * This should never happen except in the case of a VM regression so 1727598ecfbaSChristoph Hellwig * warn about it. 1728598ecfbaSChristoph Hellwig */ 1729598ecfbaSChristoph Hellwig if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == 1730598ecfbaSChristoph Hellwig PF_MEMALLOC)) 1731598ecfbaSChristoph Hellwig goto redirty; 1732598ecfbaSChristoph Hellwig 1733598ecfbaSChristoph Hellwig /* 1734e735c007SMatthew Wilcox (Oracle) * Is this folio beyond the end of the file? 1735598ecfbaSChristoph Hellwig * 1736e735c007SMatthew Wilcox (Oracle) * The folio index is less than the end_index, adjust the end_pos 1737e735c007SMatthew Wilcox (Oracle) * to the highest offset that this folio should represent. 1738598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1739598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1740598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1741598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | | 1742598ecfbaSChristoph Hellwig * ^--------------------------------^----------|-------- 1743598ecfbaSChristoph Hellwig * | desired writeback range | see else | 1744598ecfbaSChristoph Hellwig * ---------------------------------^------------------| 1745598ecfbaSChristoph Hellwig */ 174681d4782aSMatthew Wilcox (Oracle) isize = i_size_read(inode); 1747e735c007SMatthew Wilcox (Oracle) end_pos = folio_pos(folio) + folio_size(folio); 174881d4782aSMatthew Wilcox (Oracle) if (end_pos > isize) { 1749598ecfbaSChristoph Hellwig /* 1750598ecfbaSChristoph Hellwig * Check whether the page to write out is beyond or straddles 1751598ecfbaSChristoph Hellwig * i_size or not. 1752598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1753598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1754598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1755598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | 1756598ecfbaSChristoph Hellwig * ^--------------------------------^-----------|--------- 1757598ecfbaSChristoph Hellwig * | | Straddles | 1758598ecfbaSChristoph Hellwig * ---------------------------------^-----------|--------| 1759598ecfbaSChristoph Hellwig */ 1760e735c007SMatthew Wilcox (Oracle) size_t poff = offset_in_folio(folio, isize); 176181d4782aSMatthew Wilcox (Oracle) pgoff_t end_index = isize >> PAGE_SHIFT; 1762598ecfbaSChristoph Hellwig 1763598ecfbaSChristoph Hellwig /* 1764d58562caSChris Mason * Skip the page if it's fully outside i_size, e.g. 1765d58562caSChris Mason * due to a truncate operation that's in progress. We've 1766d58562caSChris Mason * cleaned this page and truncate will finish things off for 1767d58562caSChris Mason * us. 1768598ecfbaSChristoph Hellwig * 1769f1f264b4SAndreas Gruenbacher * Note that the end_index is unsigned long. If the given 1770f1f264b4SAndreas Gruenbacher * offset is greater than 16TB on a 32-bit system then if we 1771f1f264b4SAndreas Gruenbacher * checked if the page is fully outside i_size with 1772f1f264b4SAndreas Gruenbacher * "if (page->index >= end_index + 1)", "end_index + 1" would 1773f1f264b4SAndreas Gruenbacher * overflow and evaluate to 0. Hence this page would be 1774f1f264b4SAndreas Gruenbacher * redirtied and written out repeatedly, which would result in 1775f1f264b4SAndreas Gruenbacher * an infinite loop; the user program performing this operation 1776f1f264b4SAndreas Gruenbacher * would hang. Instead, we can detect this situation by 1777f1f264b4SAndreas Gruenbacher * checking if the page is totally beyond i_size or if its 1778598ecfbaSChristoph Hellwig * offset is just equal to the EOF. 1779598ecfbaSChristoph Hellwig */ 1780e735c007SMatthew Wilcox (Oracle) if (folio->index > end_index || 1781e735c007SMatthew Wilcox (Oracle) (folio->index == end_index && poff == 0)) 1782d58562caSChris Mason goto unlock; 1783598ecfbaSChristoph Hellwig 1784598ecfbaSChristoph Hellwig /* 1785598ecfbaSChristoph Hellwig * The page straddles i_size. It must be zeroed out on each 1786598ecfbaSChristoph Hellwig * and every writepage invocation because it may be mmapped. 1787598ecfbaSChristoph Hellwig * "A file is mapped in multiples of the page size. For a file 1788598ecfbaSChristoph Hellwig * that is not a multiple of the page size, the remaining 1789598ecfbaSChristoph Hellwig * memory is zeroed when mapped, and writes to that region are 1790598ecfbaSChristoph Hellwig * not written out to the file." 1791598ecfbaSChristoph Hellwig */ 1792e735c007SMatthew Wilcox (Oracle) folio_zero_segment(folio, poff, folio_size(folio)); 179381d4782aSMatthew Wilcox (Oracle) end_pos = isize; 1794598ecfbaSChristoph Hellwig } 1795598ecfbaSChristoph Hellwig 1796e735c007SMatthew Wilcox (Oracle) return iomap_writepage_map(wpc, wbc, inode, folio, end_pos); 1797598ecfbaSChristoph Hellwig 1798598ecfbaSChristoph Hellwig redirty: 1799e735c007SMatthew Wilcox (Oracle) folio_redirty_for_writepage(wbc, folio); 1800d58562caSChris Mason unlock: 1801e735c007SMatthew Wilcox (Oracle) folio_unlock(folio); 1802598ecfbaSChristoph Hellwig return 0; 1803598ecfbaSChristoph Hellwig } 1804598ecfbaSChristoph Hellwig 1805598ecfbaSChristoph Hellwig int 1806598ecfbaSChristoph Hellwig iomap_writepages(struct address_space *mapping, struct writeback_control *wbc, 1807598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1808598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1809598ecfbaSChristoph Hellwig { 1810598ecfbaSChristoph Hellwig int ret; 1811598ecfbaSChristoph Hellwig 1812598ecfbaSChristoph Hellwig wpc->ops = ops; 1813598ecfbaSChristoph Hellwig ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc); 1814598ecfbaSChristoph Hellwig if (!wpc->ioend) 1815598ecfbaSChristoph Hellwig return ret; 1816598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1817598ecfbaSChristoph Hellwig } 1818598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepages); 1819598ecfbaSChristoph Hellwig 1820598ecfbaSChristoph Hellwig static int __init iomap_init(void) 1821598ecfbaSChristoph Hellwig { 1822598ecfbaSChristoph Hellwig return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE), 1823598ecfbaSChristoph Hellwig offsetof(struct iomap_ioend, io_inline_bio), 1824598ecfbaSChristoph Hellwig BIOSET_NEED_BVECS); 1825598ecfbaSChristoph Hellwig } 1826598ecfbaSChristoph Hellwig fs_initcall(iomap_init); 1827