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 24ab08b01eSChristoph Hellwig /* 250a195b91SMatthew Wilcox (Oracle) * Structure allocated for each page or THP when block size < page size 260a195b91SMatthew Wilcox (Oracle) * to track sub-page uptodate status and I/O completions. 27ab08b01eSChristoph Hellwig */ 28ab08b01eSChristoph Hellwig struct iomap_page { 297d636676SMatthew Wilcox (Oracle) atomic_t read_bytes_pending; 300fb2d720SMatthew Wilcox (Oracle) atomic_t write_bytes_pending; 311cea335dSChristoph Hellwig spinlock_t uptodate_lock; 320a195b91SMatthew Wilcox (Oracle) unsigned long uptodate[]; 33ab08b01eSChristoph Hellwig }; 34ab08b01eSChristoph Hellwig 35ab08b01eSChristoph Hellwig static inline struct iomap_page *to_iomap_page(struct page *page) 36ab08b01eSChristoph Hellwig { 370a195b91SMatthew Wilcox (Oracle) /* 380a195b91SMatthew Wilcox (Oracle) * per-block data is stored in the head page. Callers should 39f1f264b4SAndreas Gruenbacher * not be dealing with tail pages, and if they are, they can 400a195b91SMatthew Wilcox (Oracle) * call thp_head() first. 410a195b91SMatthew Wilcox (Oracle) */ 420a195b91SMatthew Wilcox (Oracle) VM_BUG_ON_PGFLAGS(PageTail(page), page); 430a195b91SMatthew Wilcox (Oracle) 44ab08b01eSChristoph Hellwig if (page_has_private(page)) 45ab08b01eSChristoph Hellwig return (struct iomap_page *)page_private(page); 46ab08b01eSChristoph Hellwig return NULL; 47ab08b01eSChristoph Hellwig } 48ab08b01eSChristoph Hellwig 49598ecfbaSChristoph Hellwig static struct bio_set iomap_ioend_bioset; 50598ecfbaSChristoph Hellwig 51afc51aaaSDarrick J. Wong static struct iomap_page * 52afc51aaaSDarrick J. Wong iomap_page_create(struct inode *inode, struct page *page) 53afc51aaaSDarrick J. Wong { 54afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 550a195b91SMatthew Wilcox (Oracle) unsigned int nr_blocks = i_blocks_per_page(inode, page); 56afc51aaaSDarrick J. Wong 570a195b91SMatthew Wilcox (Oracle) if (iop || nr_blocks <= 1) 58afc51aaaSDarrick J. Wong return iop; 59afc51aaaSDarrick J. Wong 600a195b91SMatthew Wilcox (Oracle) iop = kzalloc(struct_size(iop, uptodate, BITS_TO_LONGS(nr_blocks)), 610a195b91SMatthew Wilcox (Oracle) GFP_NOFS | __GFP_NOFAIL); 621cea335dSChristoph Hellwig spin_lock_init(&iop->uptodate_lock); 634595a298SMatthew Wilcox (Oracle) if (PageUptodate(page)) 644595a298SMatthew Wilcox (Oracle) bitmap_fill(iop->uptodate, nr_blocks); 6558aeb731SGuoqing Jiang attach_page_private(page, iop); 66afc51aaaSDarrick J. Wong return iop; 67afc51aaaSDarrick J. Wong } 68afc51aaaSDarrick J. Wong 69afc51aaaSDarrick J. Wong static void 70afc51aaaSDarrick J. Wong iomap_page_release(struct page *page) 71afc51aaaSDarrick J. Wong { 7258aeb731SGuoqing Jiang struct iomap_page *iop = detach_page_private(page); 730a195b91SMatthew Wilcox (Oracle) unsigned int nr_blocks = i_blocks_per_page(page->mapping->host, page); 74afc51aaaSDarrick J. Wong 75afc51aaaSDarrick J. Wong if (!iop) 76afc51aaaSDarrick J. Wong return; 777d636676SMatthew Wilcox (Oracle) WARN_ON_ONCE(atomic_read(&iop->read_bytes_pending)); 780fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(atomic_read(&iop->write_bytes_pending)); 790a195b91SMatthew Wilcox (Oracle) WARN_ON_ONCE(bitmap_full(iop->uptodate, nr_blocks) != 800a195b91SMatthew Wilcox (Oracle) PageUptodate(page)); 81afc51aaaSDarrick J. Wong kfree(iop); 82afc51aaaSDarrick J. Wong } 83afc51aaaSDarrick J. Wong 84afc51aaaSDarrick J. Wong /* 85afc51aaaSDarrick J. Wong * Calculate the range inside the page that we actually need to read. 86afc51aaaSDarrick J. Wong */ 87afc51aaaSDarrick J. Wong static void 88afc51aaaSDarrick J. Wong iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop, 89afc51aaaSDarrick J. Wong loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp) 90afc51aaaSDarrick J. Wong { 91afc51aaaSDarrick J. Wong loff_t orig_pos = *pos; 92afc51aaaSDarrick J. Wong loff_t isize = i_size_read(inode); 93afc51aaaSDarrick J. Wong unsigned block_bits = inode->i_blkbits; 94afc51aaaSDarrick J. Wong unsigned block_size = (1 << block_bits); 95afc51aaaSDarrick J. Wong unsigned poff = offset_in_page(*pos); 96afc51aaaSDarrick J. Wong unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length); 97afc51aaaSDarrick J. Wong unsigned first = poff >> block_bits; 98afc51aaaSDarrick J. Wong unsigned last = (poff + plen - 1) >> block_bits; 99afc51aaaSDarrick J. Wong 100afc51aaaSDarrick J. Wong /* 101f1f264b4SAndreas Gruenbacher * If the block size is smaller than the page size, we need to check the 102afc51aaaSDarrick J. Wong * per-block uptodate status and adjust the offset and length if needed 103afc51aaaSDarrick J. Wong * to avoid reading in already uptodate ranges. 104afc51aaaSDarrick J. Wong */ 105afc51aaaSDarrick J. Wong if (iop) { 106afc51aaaSDarrick J. Wong unsigned int i; 107afc51aaaSDarrick J. Wong 108afc51aaaSDarrick J. Wong /* move forward for each leading block marked uptodate */ 109afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) { 110afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 111afc51aaaSDarrick J. Wong break; 112afc51aaaSDarrick J. Wong *pos += block_size; 113afc51aaaSDarrick J. Wong poff += block_size; 114afc51aaaSDarrick J. Wong plen -= block_size; 115afc51aaaSDarrick J. Wong first++; 116afc51aaaSDarrick J. Wong } 117afc51aaaSDarrick J. Wong 118afc51aaaSDarrick J. Wong /* truncate len if we find any trailing uptodate block(s) */ 119afc51aaaSDarrick J. Wong for ( ; i <= last; i++) { 120afc51aaaSDarrick J. Wong if (test_bit(i, iop->uptodate)) { 121afc51aaaSDarrick J. Wong plen -= (last - i + 1) * block_size; 122afc51aaaSDarrick J. Wong last = i - 1; 123afc51aaaSDarrick J. Wong break; 124afc51aaaSDarrick J. Wong } 125afc51aaaSDarrick J. Wong } 126afc51aaaSDarrick J. Wong } 127afc51aaaSDarrick J. Wong 128afc51aaaSDarrick J. Wong /* 129f1f264b4SAndreas Gruenbacher * If the extent spans the block that contains the i_size, we need to 130afc51aaaSDarrick J. Wong * handle both halves separately so that we properly zero data in the 131afc51aaaSDarrick J. Wong * page cache for blocks that are entirely outside of i_size. 132afc51aaaSDarrick J. Wong */ 133afc51aaaSDarrick J. Wong if (orig_pos <= isize && orig_pos + length > isize) { 134afc51aaaSDarrick J. Wong unsigned end = offset_in_page(isize - 1) >> block_bits; 135afc51aaaSDarrick J. Wong 136afc51aaaSDarrick J. Wong if (first <= end && last > end) 137afc51aaaSDarrick J. Wong plen -= (last - end) * block_size; 138afc51aaaSDarrick J. Wong } 139afc51aaaSDarrick J. Wong 140afc51aaaSDarrick J. Wong *offp = poff; 141afc51aaaSDarrick J. Wong *lenp = plen; 142afc51aaaSDarrick J. Wong } 143afc51aaaSDarrick J. Wong 144afc51aaaSDarrick J. Wong static void 1451cea335dSChristoph Hellwig iomap_iop_set_range_uptodate(struct page *page, unsigned off, unsigned len) 146afc51aaaSDarrick J. Wong { 147afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 148afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 149afc51aaaSDarrick J. Wong unsigned first = off >> inode->i_blkbits; 150afc51aaaSDarrick J. Wong unsigned last = (off + len - 1) >> inode->i_blkbits; 1511cea335dSChristoph Hellwig unsigned long flags; 152afc51aaaSDarrick J. Wong 1531cea335dSChristoph Hellwig spin_lock_irqsave(&iop->uptodate_lock, flags); 154b21866f5SMatthew Wilcox (Oracle) bitmap_set(iop->uptodate, first, last - first + 1); 155b21866f5SMatthew Wilcox (Oracle) if (bitmap_full(iop->uptodate, i_blocks_per_page(inode, page))) 1561cea335dSChristoph Hellwig SetPageUptodate(page); 1571cea335dSChristoph Hellwig spin_unlock_irqrestore(&iop->uptodate_lock, flags); 158afc51aaaSDarrick J. Wong } 159afc51aaaSDarrick J. Wong 1601cea335dSChristoph Hellwig static void 1611cea335dSChristoph Hellwig iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len) 1621cea335dSChristoph Hellwig { 1631cea335dSChristoph Hellwig if (PageError(page)) 1641cea335dSChristoph Hellwig return; 1651cea335dSChristoph Hellwig 1661cea335dSChristoph Hellwig if (page_has_private(page)) 1671cea335dSChristoph Hellwig iomap_iop_set_range_uptodate(page, off, len); 1681cea335dSChristoph Hellwig else 169afc51aaaSDarrick J. Wong SetPageUptodate(page); 170afc51aaaSDarrick J. Wong } 171afc51aaaSDarrick J. Wong 172afc51aaaSDarrick J. Wong static void 173afc51aaaSDarrick J. Wong iomap_read_page_end_io(struct bio_vec *bvec, int error) 174afc51aaaSDarrick J. Wong { 175afc51aaaSDarrick J. Wong struct page *page = bvec->bv_page; 176afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 177afc51aaaSDarrick J. Wong 178afc51aaaSDarrick J. Wong if (unlikely(error)) { 179afc51aaaSDarrick J. Wong ClearPageUptodate(page); 180afc51aaaSDarrick J. Wong SetPageError(page); 181afc51aaaSDarrick J. Wong } else { 182afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len); 183afc51aaaSDarrick J. Wong } 184afc51aaaSDarrick J. Wong 1857d636676SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(bvec->bv_len, &iop->read_bytes_pending)) 1867d636676SMatthew Wilcox (Oracle) unlock_page(page); 187afc51aaaSDarrick J. Wong } 188afc51aaaSDarrick J. Wong 189afc51aaaSDarrick J. Wong static void 190afc51aaaSDarrick J. Wong iomap_read_end_io(struct bio *bio) 191afc51aaaSDarrick J. Wong { 192afc51aaaSDarrick J. Wong int error = blk_status_to_errno(bio->bi_status); 193afc51aaaSDarrick J. Wong struct bio_vec *bvec; 194afc51aaaSDarrick J. Wong struct bvec_iter_all iter_all; 195afc51aaaSDarrick J. Wong 196afc51aaaSDarrick J. Wong bio_for_each_segment_all(bvec, bio, iter_all) 197afc51aaaSDarrick J. Wong iomap_read_page_end_io(bvec, error); 198afc51aaaSDarrick J. Wong bio_put(bio); 199afc51aaaSDarrick J. Wong } 200afc51aaaSDarrick J. Wong 201afc51aaaSDarrick J. Wong struct iomap_readpage_ctx { 202afc51aaaSDarrick J. Wong struct page *cur_page; 203afc51aaaSDarrick J. Wong bool cur_page_in_bio; 204afc51aaaSDarrick J. Wong struct bio *bio; 2059d24a13aSMatthew Wilcox (Oracle) struct readahead_control *rac; 206afc51aaaSDarrick J. Wong }; 207afc51aaaSDarrick J. Wong 208fad0a1abSChristoph Hellwig static loff_t iomap_read_inline_data(const struct iomap_iter *iter, 2091b5c1e36SChristoph Hellwig struct page *page) 210afc51aaaSDarrick J. Wong { 211fad0a1abSChristoph Hellwig const struct iomap *iomap = iomap_iter_srcmap(iter); 2121b5c1e36SChristoph Hellwig size_t size = i_size_read(iter->inode) - iomap->offset; 213b405435bSMatthew Wilcox (Oracle) size_t poff = offset_in_page(iomap->offset); 214afc51aaaSDarrick J. Wong void *addr; 215afc51aaaSDarrick J. Wong 216afc51aaaSDarrick J. Wong if (PageUptodate(page)) 217b405435bSMatthew Wilcox (Oracle) return PAGE_SIZE - poff; 218afc51aaaSDarrick J. Wong 219ae44f9c2SMatthew Wilcox (Oracle) if (WARN_ON_ONCE(size > PAGE_SIZE - poff)) 220ae44f9c2SMatthew Wilcox (Oracle) return -EIO; 22169f4a26cSGao Xiang if (WARN_ON_ONCE(size > PAGE_SIZE - 22269f4a26cSGao Xiang offset_in_page(iomap->inline_data))) 22369f4a26cSGao Xiang return -EIO; 22469f4a26cSGao Xiang if (WARN_ON_ONCE(size > iomap->length)) 22569f4a26cSGao Xiang return -EIO; 226b405435bSMatthew Wilcox (Oracle) if (poff > 0) 2271b5c1e36SChristoph Hellwig iomap_page_create(iter->inode, page); 228afc51aaaSDarrick J. Wong 229ab069d5fSMatthew Wilcox (Oracle) addr = kmap_local_page(page) + poff; 230afc51aaaSDarrick J. Wong memcpy(addr, iomap->inline_data, size); 231b405435bSMatthew Wilcox (Oracle) memset(addr + size, 0, PAGE_SIZE - poff - size); 232ab069d5fSMatthew Wilcox (Oracle) kunmap_local(addr); 233b405435bSMatthew Wilcox (Oracle) iomap_set_range_uptodate(page, poff, PAGE_SIZE - poff); 234b405435bSMatthew Wilcox (Oracle) return PAGE_SIZE - poff; 235afc51aaaSDarrick J. Wong } 236afc51aaaSDarrick J. Wong 237fad0a1abSChristoph Hellwig static inline bool iomap_block_needs_zeroing(const struct iomap_iter *iter, 2381b5c1e36SChristoph Hellwig loff_t pos) 239009d8d84SChristoph Hellwig { 240fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 2411b5c1e36SChristoph Hellwig 2421b5c1e36SChristoph Hellwig return srcmap->type != IOMAP_MAPPED || 2431b5c1e36SChristoph Hellwig (srcmap->flags & IOMAP_F_NEW) || 2441b5c1e36SChristoph Hellwig pos >= i_size_read(iter->inode); 245009d8d84SChristoph Hellwig } 246009d8d84SChristoph Hellwig 247fad0a1abSChristoph Hellwig static loff_t iomap_readpage_iter(const struct iomap_iter *iter, 248f6d48000SChristoph Hellwig struct iomap_readpage_ctx *ctx, loff_t offset) 249afc51aaaSDarrick J. Wong { 250fad0a1abSChristoph Hellwig const struct iomap *iomap = &iter->iomap; 251f6d48000SChristoph Hellwig loff_t pos = iter->pos + offset; 252f6d48000SChristoph Hellwig loff_t length = iomap_length(iter) - offset; 253afc51aaaSDarrick J. Wong struct page *page = ctx->cur_page; 254637d3375SAndreas Gruenbacher struct iomap_page *iop; 255afc51aaaSDarrick J. Wong loff_t orig_pos = pos; 256afc51aaaSDarrick J. Wong unsigned poff, plen; 257afc51aaaSDarrick J. Wong sector_t sector; 258afc51aaaSDarrick J. Wong 259b405435bSMatthew Wilcox (Oracle) if (iomap->type == IOMAP_INLINE) 2601b5c1e36SChristoph Hellwig return min(iomap_read_inline_data(iter, page), length); 261afc51aaaSDarrick J. Wong 262afc51aaaSDarrick J. Wong /* zero post-eof blocks as the page may be mapped */ 263f6d48000SChristoph Hellwig iop = iomap_page_create(iter->inode, page); 264f6d48000SChristoph Hellwig iomap_adjust_read_range(iter->inode, iop, &pos, length, &poff, &plen); 265afc51aaaSDarrick J. Wong if (plen == 0) 266afc51aaaSDarrick J. Wong goto done; 267afc51aaaSDarrick J. Wong 2681b5c1e36SChristoph Hellwig if (iomap_block_needs_zeroing(iter, pos)) { 269afc51aaaSDarrick J. Wong zero_user(page, poff, plen); 270afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, poff, plen); 271afc51aaaSDarrick J. Wong goto done; 272afc51aaaSDarrick J. Wong } 273afc51aaaSDarrick J. Wong 274afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = true; 2757d636676SMatthew Wilcox (Oracle) if (iop) 2767d636676SMatthew Wilcox (Oracle) atomic_add(plen, &iop->read_bytes_pending); 277afc51aaaSDarrick J. Wong 278afc51aaaSDarrick J. Wong sector = iomap_sector(iomap, pos); 279d0364f94SChristoph Hellwig if (!ctx->bio || 280d0364f94SChristoph Hellwig bio_end_sector(ctx->bio) != sector || 281d0364f94SChristoph Hellwig bio_add_page(ctx->bio, page, plen, poff) != plen) { 282afc51aaaSDarrick J. Wong gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL); 283457df33eSMatthew Wilcox (Oracle) gfp_t orig_gfp = gfp; 2845f7136dbSMatthew Wilcox (Oracle) unsigned int nr_vecs = DIV_ROUND_UP(length, PAGE_SIZE); 285afc51aaaSDarrick J. Wong 286afc51aaaSDarrick J. Wong if (ctx->bio) 287afc51aaaSDarrick J. Wong submit_bio(ctx->bio); 288afc51aaaSDarrick J. Wong 2899d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) /* same as readahead_gfp_mask */ 290afc51aaaSDarrick J. Wong gfp |= __GFP_NORETRY | __GFP_NOWARN; 2915f7136dbSMatthew Wilcox (Oracle) ctx->bio = bio_alloc(gfp, bio_max_segs(nr_vecs)); 292457df33eSMatthew Wilcox (Oracle) /* 293457df33eSMatthew Wilcox (Oracle) * If the bio_alloc fails, try it again for a single page to 294457df33eSMatthew Wilcox (Oracle) * avoid having to deal with partial page reads. This emulates 295457df33eSMatthew Wilcox (Oracle) * what do_mpage_readpage does. 296457df33eSMatthew Wilcox (Oracle) */ 297457df33eSMatthew Wilcox (Oracle) if (!ctx->bio) 298457df33eSMatthew Wilcox (Oracle) ctx->bio = bio_alloc(orig_gfp, 1); 299afc51aaaSDarrick J. Wong ctx->bio->bi_opf = REQ_OP_READ; 3009d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) 301afc51aaaSDarrick J. Wong ctx->bio->bi_opf |= REQ_RAHEAD; 302afc51aaaSDarrick J. Wong ctx->bio->bi_iter.bi_sector = sector; 303afc51aaaSDarrick J. Wong bio_set_dev(ctx->bio, iomap->bdev); 304afc51aaaSDarrick J. Wong ctx->bio->bi_end_io = iomap_read_end_io; 305d0364f94SChristoph Hellwig __bio_add_page(ctx->bio, page, plen, poff); 306afc51aaaSDarrick J. Wong } 307afc51aaaSDarrick J. Wong done: 308afc51aaaSDarrick J. Wong /* 309afc51aaaSDarrick J. Wong * Move the caller beyond our range so that it keeps making progress. 310f1f264b4SAndreas Gruenbacher * For that, we have to include any leading non-uptodate ranges, but 311afc51aaaSDarrick J. Wong * we can skip trailing ones as they will be handled in the next 312afc51aaaSDarrick J. Wong * iteration. 313afc51aaaSDarrick J. Wong */ 314afc51aaaSDarrick J. Wong return pos - orig_pos + plen; 315afc51aaaSDarrick J. Wong } 316afc51aaaSDarrick J. Wong 317afc51aaaSDarrick J. Wong int 318afc51aaaSDarrick J. Wong iomap_readpage(struct page *page, const struct iomap_ops *ops) 319afc51aaaSDarrick J. Wong { 320f6d48000SChristoph Hellwig struct iomap_iter iter = { 321f6d48000SChristoph Hellwig .inode = page->mapping->host, 322f6d48000SChristoph Hellwig .pos = page_offset(page), 323f6d48000SChristoph Hellwig .len = PAGE_SIZE, 324f6d48000SChristoph Hellwig }; 325f6d48000SChristoph Hellwig struct iomap_readpage_ctx ctx = { 326f6d48000SChristoph Hellwig .cur_page = page, 327f6d48000SChristoph Hellwig }; 328f6d48000SChristoph Hellwig int ret; 329afc51aaaSDarrick J. Wong 3309e91c572SChristoph Hellwig trace_iomap_readpage(page->mapping->host, 1); 3319e91c572SChristoph Hellwig 332f6d48000SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 333f6d48000SChristoph Hellwig iter.processed = iomap_readpage_iter(&iter, &ctx, 0); 334f6d48000SChristoph Hellwig 335f6d48000SChristoph Hellwig if (ret < 0) 336afc51aaaSDarrick J. Wong SetPageError(page); 337afc51aaaSDarrick J. Wong 338afc51aaaSDarrick J. Wong if (ctx.bio) { 339afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 340afc51aaaSDarrick J. Wong WARN_ON_ONCE(!ctx.cur_page_in_bio); 341afc51aaaSDarrick J. Wong } else { 342afc51aaaSDarrick J. Wong WARN_ON_ONCE(ctx.cur_page_in_bio); 343afc51aaaSDarrick J. Wong unlock_page(page); 344afc51aaaSDarrick J. Wong } 345afc51aaaSDarrick J. Wong 346afc51aaaSDarrick J. Wong /* 347f1f264b4SAndreas Gruenbacher * Just like mpage_readahead and block_read_full_page, we always 348afc51aaaSDarrick J. Wong * return 0 and just mark the page as PageError on errors. This 349f1f264b4SAndreas Gruenbacher * should be cleaned up throughout the stack eventually. 350afc51aaaSDarrick J. Wong */ 351afc51aaaSDarrick J. Wong return 0; 352afc51aaaSDarrick J. Wong } 353afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_readpage); 354afc51aaaSDarrick J. Wong 355fad0a1abSChristoph Hellwig static loff_t iomap_readahead_iter(const struct iomap_iter *iter, 356f6d48000SChristoph Hellwig struct iomap_readpage_ctx *ctx) 357afc51aaaSDarrick J. Wong { 358f6d48000SChristoph Hellwig loff_t length = iomap_length(iter); 359afc51aaaSDarrick J. Wong loff_t done, ret; 360afc51aaaSDarrick J. Wong 361afc51aaaSDarrick J. Wong for (done = 0; done < length; done += ret) { 362f6d48000SChristoph Hellwig if (ctx->cur_page && offset_in_page(iter->pos + done) == 0) { 363afc51aaaSDarrick J. Wong if (!ctx->cur_page_in_bio) 364afc51aaaSDarrick J. Wong unlock_page(ctx->cur_page); 365afc51aaaSDarrick J. Wong put_page(ctx->cur_page); 366afc51aaaSDarrick J. Wong ctx->cur_page = NULL; 367afc51aaaSDarrick J. Wong } 368afc51aaaSDarrick J. Wong if (!ctx->cur_page) { 3699d24a13aSMatthew Wilcox (Oracle) ctx->cur_page = readahead_page(ctx->rac); 370afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = false; 371afc51aaaSDarrick J. Wong } 372f6d48000SChristoph Hellwig ret = iomap_readpage_iter(iter, ctx, done); 373afc51aaaSDarrick J. Wong } 374afc51aaaSDarrick J. Wong 375afc51aaaSDarrick J. Wong return done; 376afc51aaaSDarrick J. Wong } 377afc51aaaSDarrick J. Wong 3789d24a13aSMatthew Wilcox (Oracle) /** 3799d24a13aSMatthew Wilcox (Oracle) * iomap_readahead - Attempt to read pages from a file. 3809d24a13aSMatthew Wilcox (Oracle) * @rac: Describes the pages to be read. 3819d24a13aSMatthew Wilcox (Oracle) * @ops: The operations vector for the filesystem. 3829d24a13aSMatthew Wilcox (Oracle) * 3839d24a13aSMatthew Wilcox (Oracle) * This function is for filesystems to call to implement their readahead 3849d24a13aSMatthew Wilcox (Oracle) * address_space operation. 3859d24a13aSMatthew Wilcox (Oracle) * 3869d24a13aSMatthew Wilcox (Oracle) * Context: The @ops callbacks may submit I/O (eg to read the addresses of 3879d24a13aSMatthew Wilcox (Oracle) * blocks from disc), and may wait for it. The caller may be trying to 3889d24a13aSMatthew Wilcox (Oracle) * access a different page, and so sleeping excessively should be avoided. 3899d24a13aSMatthew Wilcox (Oracle) * It may allocate memory, but should avoid costly allocations. This 3909d24a13aSMatthew Wilcox (Oracle) * function is called with memalloc_nofs set, so allocations will not cause 3919d24a13aSMatthew Wilcox (Oracle) * the filesystem to be reentered. 3929d24a13aSMatthew Wilcox (Oracle) */ 3939d24a13aSMatthew Wilcox (Oracle) void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops) 394afc51aaaSDarrick J. Wong { 395f6d48000SChristoph Hellwig struct iomap_iter iter = { 396f6d48000SChristoph Hellwig .inode = rac->mapping->host, 397f6d48000SChristoph Hellwig .pos = readahead_pos(rac), 398f6d48000SChristoph Hellwig .len = readahead_length(rac), 399f6d48000SChristoph Hellwig }; 400afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { 4019d24a13aSMatthew Wilcox (Oracle) .rac = rac, 402afc51aaaSDarrick J. Wong }; 403afc51aaaSDarrick J. Wong 404f6d48000SChristoph Hellwig trace_iomap_readahead(rac->mapping->host, readahead_count(rac)); 4059e91c572SChristoph Hellwig 406f6d48000SChristoph Hellwig while (iomap_iter(&iter, ops) > 0) 407f6d48000SChristoph Hellwig iter.processed = iomap_readahead_iter(&iter, &ctx); 4089d24a13aSMatthew Wilcox (Oracle) 409afc51aaaSDarrick J. Wong if (ctx.bio) 410afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 411afc51aaaSDarrick J. Wong if (ctx.cur_page) { 412afc51aaaSDarrick J. Wong if (!ctx.cur_page_in_bio) 413afc51aaaSDarrick J. Wong unlock_page(ctx.cur_page); 414afc51aaaSDarrick J. Wong put_page(ctx.cur_page); 415afc51aaaSDarrick J. Wong } 416afc51aaaSDarrick J. Wong } 4179d24a13aSMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(iomap_readahead); 418afc51aaaSDarrick J. Wong 419afc51aaaSDarrick J. Wong /* 420afc51aaaSDarrick J. Wong * iomap_is_partially_uptodate checks whether blocks within a page are 421afc51aaaSDarrick J. Wong * uptodate or not. 422afc51aaaSDarrick J. Wong * 423afc51aaaSDarrick J. Wong * Returns true if all blocks which correspond to a file portion 424afc51aaaSDarrick J. Wong * we want to read within the page are uptodate. 425afc51aaaSDarrick J. Wong */ 426afc51aaaSDarrick J. Wong int 427afc51aaaSDarrick J. Wong iomap_is_partially_uptodate(struct page *page, unsigned long from, 428afc51aaaSDarrick J. Wong unsigned long count) 429afc51aaaSDarrick J. Wong { 430afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 431afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 432afc51aaaSDarrick J. Wong unsigned len, first, last; 433afc51aaaSDarrick J. Wong unsigned i; 434afc51aaaSDarrick J. Wong 435afc51aaaSDarrick J. Wong /* Limit range to one page */ 436afc51aaaSDarrick J. Wong len = min_t(unsigned, PAGE_SIZE - from, count); 437afc51aaaSDarrick J. Wong 438afc51aaaSDarrick J. Wong /* First and last blocks in range within page */ 439afc51aaaSDarrick J. Wong first = from >> inode->i_blkbits; 440afc51aaaSDarrick J. Wong last = (from + len - 1) >> inode->i_blkbits; 441afc51aaaSDarrick J. Wong 442afc51aaaSDarrick J. Wong if (iop) { 443afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) 444afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 445afc51aaaSDarrick J. Wong return 0; 446afc51aaaSDarrick J. Wong return 1; 447afc51aaaSDarrick J. Wong } 448afc51aaaSDarrick J. Wong 449afc51aaaSDarrick J. Wong return 0; 450afc51aaaSDarrick J. Wong } 451afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate); 452afc51aaaSDarrick J. Wong 453afc51aaaSDarrick J. Wong int 454afc51aaaSDarrick J. Wong iomap_releasepage(struct page *page, gfp_t gfp_mask) 455afc51aaaSDarrick J. Wong { 4561ac99452SMatthew Wilcox (Oracle) trace_iomap_releasepage(page->mapping->host, page_offset(page), 4571ac99452SMatthew Wilcox (Oracle) PAGE_SIZE); 4589e91c572SChristoph Hellwig 459afc51aaaSDarrick J. Wong /* 460afc51aaaSDarrick J. Wong * mm accommodates an old ext3 case where clean pages might not have had 461afc51aaaSDarrick J. Wong * the dirty bit cleared. Thus, it can send actual dirty pages to 462f1f264b4SAndreas Gruenbacher * ->releasepage() via shrink_active_list(); skip those here. 463afc51aaaSDarrick J. Wong */ 464afc51aaaSDarrick J. Wong if (PageDirty(page) || PageWriteback(page)) 465afc51aaaSDarrick J. Wong return 0; 466afc51aaaSDarrick J. Wong iomap_page_release(page); 467afc51aaaSDarrick J. Wong return 1; 468afc51aaaSDarrick J. Wong } 469afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_releasepage); 470afc51aaaSDarrick J. Wong 471afc51aaaSDarrick J. Wong void 472afc51aaaSDarrick J. Wong iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len) 473afc51aaaSDarrick J. Wong { 4741ac99452SMatthew Wilcox (Oracle) trace_iomap_invalidatepage(page->mapping->host, offset, len); 4759e91c572SChristoph Hellwig 476afc51aaaSDarrick J. Wong /* 477f1f264b4SAndreas Gruenbacher * If we're invalidating the entire page, clear the dirty state from it 478afc51aaaSDarrick J. Wong * and release it to avoid unnecessary buildup of the LRU. 479afc51aaaSDarrick J. Wong */ 480afc51aaaSDarrick J. Wong if (offset == 0 && len == PAGE_SIZE) { 481afc51aaaSDarrick J. Wong WARN_ON_ONCE(PageWriteback(page)); 482afc51aaaSDarrick J. Wong cancel_dirty_page(page); 483afc51aaaSDarrick J. Wong iomap_page_release(page); 484afc51aaaSDarrick J. Wong } 485afc51aaaSDarrick J. Wong } 486afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_invalidatepage); 487afc51aaaSDarrick J. Wong 488afc51aaaSDarrick J. Wong #ifdef CONFIG_MIGRATION 489afc51aaaSDarrick J. Wong int 490afc51aaaSDarrick J. Wong iomap_migrate_page(struct address_space *mapping, struct page *newpage, 491afc51aaaSDarrick J. Wong struct page *page, enum migrate_mode mode) 492afc51aaaSDarrick J. Wong { 493afc51aaaSDarrick J. Wong int ret; 494afc51aaaSDarrick J. Wong 49526473f83SLinus Torvalds ret = migrate_page_move_mapping(mapping, newpage, page, 0); 496afc51aaaSDarrick J. Wong if (ret != MIGRATEPAGE_SUCCESS) 497afc51aaaSDarrick J. Wong return ret; 498afc51aaaSDarrick J. Wong 49958aeb731SGuoqing Jiang if (page_has_private(page)) 50058aeb731SGuoqing Jiang attach_page_private(newpage, detach_page_private(page)); 501afc51aaaSDarrick J. Wong 502afc51aaaSDarrick J. Wong if (mode != MIGRATE_SYNC_NO_COPY) 503afc51aaaSDarrick J. Wong migrate_page_copy(newpage, page); 504afc51aaaSDarrick J. Wong else 505afc51aaaSDarrick J. Wong migrate_page_states(newpage, page); 506afc51aaaSDarrick J. Wong return MIGRATEPAGE_SUCCESS; 507afc51aaaSDarrick J. Wong } 508afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_migrate_page); 509afc51aaaSDarrick J. Wong #endif /* CONFIG_MIGRATION */ 510afc51aaaSDarrick J. Wong 511afc51aaaSDarrick J. Wong static void 512afc51aaaSDarrick J. Wong iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) 513afc51aaaSDarrick J. Wong { 514afc51aaaSDarrick J. Wong loff_t i_size = i_size_read(inode); 515afc51aaaSDarrick J. Wong 516afc51aaaSDarrick J. Wong /* 517afc51aaaSDarrick J. Wong * Only truncate newly allocated pages beyoned EOF, even if the 518afc51aaaSDarrick J. Wong * write started inside the existing inode size. 519afc51aaaSDarrick J. Wong */ 520afc51aaaSDarrick J. Wong if (pos + len > i_size) 521afc51aaaSDarrick J. Wong truncate_pagecache_range(inode, max(pos, i_size), pos + len); 522afc51aaaSDarrick J. Wong } 523afc51aaaSDarrick J. Wong 524afc51aaaSDarrick J. Wong static int 525d3b40439SChristoph Hellwig iomap_read_page_sync(loff_t block_start, struct page *page, unsigned poff, 5261acd9e9cSChristoph Hellwig unsigned plen, const struct iomap *iomap) 527afc51aaaSDarrick J. Wong { 528afc51aaaSDarrick J. Wong struct bio_vec bvec; 529afc51aaaSDarrick J. Wong struct bio bio; 530afc51aaaSDarrick J. Wong 531afc51aaaSDarrick J. Wong bio_init(&bio, &bvec, 1); 532afc51aaaSDarrick J. Wong bio.bi_opf = REQ_OP_READ; 533afc51aaaSDarrick J. Wong bio.bi_iter.bi_sector = iomap_sector(iomap, block_start); 534afc51aaaSDarrick J. Wong bio_set_dev(&bio, iomap->bdev); 535afc51aaaSDarrick J. Wong __bio_add_page(&bio, page, plen, poff); 536afc51aaaSDarrick J. Wong return submit_bio_wait(&bio); 537afc51aaaSDarrick J. Wong } 538afc51aaaSDarrick J. Wong 539fad0a1abSChristoph Hellwig static int __iomap_write_begin(const struct iomap_iter *iter, loff_t pos, 540b74b1293SChristoph Hellwig unsigned len, struct page *page) 541afc51aaaSDarrick J. Wong { 542fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 5431b5c1e36SChristoph Hellwig struct iomap_page *iop = iomap_page_create(iter->inode, page); 5441b5c1e36SChristoph Hellwig loff_t block_size = i_blocksize(iter->inode); 5456cc19c5fSNikolay Borisov loff_t block_start = round_down(pos, block_size); 5466cc19c5fSNikolay Borisov loff_t block_end = round_up(pos + len, block_size); 547afc51aaaSDarrick J. Wong unsigned from = offset_in_page(pos), to = from + len, poff, plen; 548afc51aaaSDarrick J. Wong 549afc51aaaSDarrick J. Wong if (PageUptodate(page)) 550afc51aaaSDarrick J. Wong return 0; 551e6e7ca92SMatthew Wilcox (Oracle) ClearPageError(page); 552afc51aaaSDarrick J. Wong 553afc51aaaSDarrick J. Wong do { 5541b5c1e36SChristoph Hellwig iomap_adjust_read_range(iter->inode, iop, &block_start, 555afc51aaaSDarrick J. Wong block_end - block_start, &poff, &plen); 556afc51aaaSDarrick J. Wong if (plen == 0) 557afc51aaaSDarrick J. Wong break; 558afc51aaaSDarrick J. Wong 559b74b1293SChristoph Hellwig if (!(iter->flags & IOMAP_UNSHARE) && 56032a38a49SChristoph Hellwig (from <= poff || from >= poff + plen) && 561d3b40439SChristoph Hellwig (to <= poff || to >= poff + plen)) 562d3b40439SChristoph Hellwig continue; 563d3b40439SChristoph Hellwig 5641b5c1e36SChristoph Hellwig if (iomap_block_needs_zeroing(iter, block_start)) { 565b74b1293SChristoph Hellwig if (WARN_ON_ONCE(iter->flags & IOMAP_UNSHARE)) 56632a38a49SChristoph Hellwig return -EIO; 567d3b40439SChristoph Hellwig zero_user_segments(page, poff, from, to, poff + plen); 56814284fedSMatthew Wilcox (Oracle) } else { 56914284fedSMatthew Wilcox (Oracle) int status = iomap_read_page_sync(block_start, page, 57014284fedSMatthew Wilcox (Oracle) poff, plen, srcmap); 571d3b40439SChristoph Hellwig if (status) 572d3b40439SChristoph Hellwig return status; 57314284fedSMatthew Wilcox (Oracle) } 57414284fedSMatthew Wilcox (Oracle) iomap_set_range_uptodate(page, poff, plen); 575afc51aaaSDarrick J. Wong } while ((block_start += plen) < block_end); 576afc51aaaSDarrick J. Wong 577d3b40439SChristoph Hellwig return 0; 578afc51aaaSDarrick J. Wong } 579afc51aaaSDarrick J. Wong 580fad0a1abSChristoph Hellwig static int iomap_write_begin_inline(const struct iomap_iter *iter, 5811b5c1e36SChristoph Hellwig struct page *page) 58269f4a26cSGao Xiang { 583b405435bSMatthew Wilcox (Oracle) int ret; 584b405435bSMatthew Wilcox (Oracle) 58569f4a26cSGao Xiang /* needs more work for the tailpacking case; disable for now */ 5861b5c1e36SChristoph Hellwig if (WARN_ON_ONCE(iomap_iter_srcmap(iter)->offset != 0)) 58769f4a26cSGao Xiang return -EIO; 5881b5c1e36SChristoph Hellwig ret = iomap_read_inline_data(iter, page); 589b405435bSMatthew Wilcox (Oracle) if (ret < 0) 590b405435bSMatthew Wilcox (Oracle) return ret; 591b405435bSMatthew Wilcox (Oracle) return 0; 59269f4a26cSGao Xiang } 59369f4a26cSGao Xiang 594fad0a1abSChristoph Hellwig static int iomap_write_begin(const struct iomap_iter *iter, loff_t pos, 595fad0a1abSChristoph Hellwig unsigned len, struct page **pagep) 596afc51aaaSDarrick J. Wong { 5971b5c1e36SChristoph Hellwig const struct iomap_page_ops *page_ops = iter->iomap.page_ops; 598fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 599afc51aaaSDarrick J. Wong struct page *page; 600afc51aaaSDarrick J. Wong int status = 0; 601afc51aaaSDarrick J. Wong 6021b5c1e36SChristoph Hellwig BUG_ON(pos + len > iter->iomap.offset + iter->iomap.length); 6031b5c1e36SChristoph Hellwig if (srcmap != &iter->iomap) 604c039b997SGoldwyn Rodrigues BUG_ON(pos + len > srcmap->offset + srcmap->length); 605afc51aaaSDarrick J. Wong 606afc51aaaSDarrick J. Wong if (fatal_signal_pending(current)) 607afc51aaaSDarrick J. Wong return -EINTR; 608afc51aaaSDarrick J. Wong 609afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_prepare) { 6101b5c1e36SChristoph Hellwig status = page_ops->page_prepare(iter->inode, pos, len); 611afc51aaaSDarrick J. Wong if (status) 612afc51aaaSDarrick J. Wong return status; 613afc51aaaSDarrick J. Wong } 614afc51aaaSDarrick J. Wong 6151b5c1e36SChristoph Hellwig page = grab_cache_page_write_begin(iter->inode->i_mapping, 6161b5c1e36SChristoph Hellwig pos >> PAGE_SHIFT, AOP_FLAG_NOFS); 617afc51aaaSDarrick J. Wong if (!page) { 618afc51aaaSDarrick J. Wong status = -ENOMEM; 619afc51aaaSDarrick J. Wong goto out_no_page; 620afc51aaaSDarrick J. Wong } 621afc51aaaSDarrick J. Wong 622c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) 6231b5c1e36SChristoph Hellwig status = iomap_write_begin_inline(iter, page); 6241b5c1e36SChristoph Hellwig else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) 625c039b997SGoldwyn Rodrigues status = __block_write_begin_int(page, pos, len, NULL, srcmap); 626afc51aaaSDarrick J. Wong else 627b74b1293SChristoph Hellwig status = __iomap_write_begin(iter, pos, len, page); 628afc51aaaSDarrick J. Wong 629afc51aaaSDarrick J. Wong if (unlikely(status)) 630afc51aaaSDarrick J. Wong goto out_unlock; 631afc51aaaSDarrick J. Wong 632afc51aaaSDarrick J. Wong *pagep = page; 633afc51aaaSDarrick J. Wong return 0; 634afc51aaaSDarrick J. Wong 635afc51aaaSDarrick J. Wong out_unlock: 636afc51aaaSDarrick J. Wong unlock_page(page); 637afc51aaaSDarrick J. Wong put_page(page); 6381b5c1e36SChristoph Hellwig iomap_write_failed(iter->inode, pos, len); 639afc51aaaSDarrick J. Wong 640afc51aaaSDarrick J. Wong out_no_page: 641afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 6421b5c1e36SChristoph Hellwig page_ops->page_done(iter->inode, pos, 0, NULL); 643afc51aaaSDarrick J. Wong return status; 644afc51aaaSDarrick J. Wong } 645afc51aaaSDarrick J. Wong 646e25ba8cbSMatthew Wilcox (Oracle) static size_t __iomap_write_end(struct inode *inode, loff_t pos, size_t len, 647e25ba8cbSMatthew Wilcox (Oracle) size_t copied, struct page *page) 648afc51aaaSDarrick J. Wong { 649afc51aaaSDarrick J. Wong flush_dcache_page(page); 650afc51aaaSDarrick J. Wong 651afc51aaaSDarrick J. Wong /* 652afc51aaaSDarrick J. Wong * The blocks that were entirely written will now be uptodate, so we 653afc51aaaSDarrick J. Wong * don't have to worry about a readpage reading them and overwriting a 654f1f264b4SAndreas Gruenbacher * partial write. However, if we've encountered a short write and only 655afc51aaaSDarrick J. Wong * partially written into a block, it will not be marked uptodate, so a 656afc51aaaSDarrick J. Wong * readpage might come in and destroy our partial write. 657afc51aaaSDarrick J. Wong * 658f1f264b4SAndreas Gruenbacher * Do the simplest thing and just treat any short write to a 659f1f264b4SAndreas Gruenbacher * non-uptodate page as a zero-length write, and force the caller to 660f1f264b4SAndreas Gruenbacher * redo the whole thing. 661afc51aaaSDarrick J. Wong */ 662afc51aaaSDarrick J. Wong if (unlikely(copied < len && !PageUptodate(page))) 663afc51aaaSDarrick J. Wong return 0; 664afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, offset_in_page(pos), len); 665fd7353f8SMatthew Wilcox (Oracle) __set_page_dirty_nobuffers(page); 666afc51aaaSDarrick J. Wong return copied; 667afc51aaaSDarrick J. Wong } 668afc51aaaSDarrick J. Wong 669fad0a1abSChristoph Hellwig static size_t iomap_write_end_inline(const struct iomap_iter *iter, 670fad0a1abSChristoph Hellwig struct page *page, loff_t pos, size_t copied) 671afc51aaaSDarrick J. Wong { 672fad0a1abSChristoph Hellwig const struct iomap *iomap = &iter->iomap; 673afc51aaaSDarrick J. Wong void *addr; 674afc51aaaSDarrick J. Wong 675afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 67669f4a26cSGao Xiang BUG_ON(!iomap_inline_data_valid(iomap)); 677afc51aaaSDarrick J. Wong 6787ed3cd1aSMatthew Wilcox (Oracle) flush_dcache_page(page); 679ab069d5fSMatthew Wilcox (Oracle) addr = kmap_local_page(page) + pos; 680ab069d5fSMatthew Wilcox (Oracle) memcpy(iomap_inline_data(iomap, pos), addr, copied); 681ab069d5fSMatthew Wilcox (Oracle) kunmap_local(addr); 682afc51aaaSDarrick J. Wong 6831b5c1e36SChristoph Hellwig mark_inode_dirty(iter->inode); 684afc51aaaSDarrick J. Wong return copied; 685afc51aaaSDarrick J. Wong } 686afc51aaaSDarrick J. Wong 687e25ba8cbSMatthew Wilcox (Oracle) /* Returns the number of bytes copied. May be 0. Cannot be an errno. */ 6881b5c1e36SChristoph Hellwig static size_t iomap_write_end(struct iomap_iter *iter, loff_t pos, size_t len, 6891b5c1e36SChristoph Hellwig size_t copied, struct page *page) 690afc51aaaSDarrick J. Wong { 6911b5c1e36SChristoph Hellwig const struct iomap_page_ops *page_ops = iter->iomap.page_ops; 692fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 6931b5c1e36SChristoph Hellwig loff_t old_size = iter->inode->i_size; 694e25ba8cbSMatthew Wilcox (Oracle) size_t ret; 695afc51aaaSDarrick J. Wong 696c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) { 6971b5c1e36SChristoph Hellwig ret = iomap_write_end_inline(iter, page, pos, copied); 698c039b997SGoldwyn Rodrigues } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) { 6991b5c1e36SChristoph Hellwig ret = block_write_end(NULL, iter->inode->i_mapping, pos, len, 7001b5c1e36SChristoph Hellwig copied, page, NULL); 701afc51aaaSDarrick J. Wong } else { 7021b5c1e36SChristoph Hellwig ret = __iomap_write_end(iter->inode, pos, len, copied, page); 703afc51aaaSDarrick J. Wong } 704afc51aaaSDarrick J. Wong 705afc51aaaSDarrick J. Wong /* 706afc51aaaSDarrick J. Wong * Update the in-memory inode size after copying the data into the page 707afc51aaaSDarrick J. Wong * cache. It's up to the file system to write the updated size to disk, 708afc51aaaSDarrick J. Wong * preferably after I/O completion so that no stale data is exposed. 709afc51aaaSDarrick J. Wong */ 710afc51aaaSDarrick J. Wong if (pos + ret > old_size) { 7111b5c1e36SChristoph Hellwig i_size_write(iter->inode, pos + ret); 7121b5c1e36SChristoph Hellwig iter->iomap.flags |= IOMAP_F_SIZE_CHANGED; 713afc51aaaSDarrick J. Wong } 714afc51aaaSDarrick J. Wong unlock_page(page); 715afc51aaaSDarrick J. Wong 716afc51aaaSDarrick J. Wong if (old_size < pos) 7171b5c1e36SChristoph Hellwig pagecache_isize_extended(iter->inode, old_size, pos); 718afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 7191b5c1e36SChristoph Hellwig page_ops->page_done(iter->inode, pos, ret, page); 720afc51aaaSDarrick J. Wong put_page(page); 721afc51aaaSDarrick J. Wong 722afc51aaaSDarrick J. Wong if (ret < len) 7231b5c1e36SChristoph Hellwig iomap_write_failed(iter->inode, pos, len); 724afc51aaaSDarrick J. Wong return ret; 725afc51aaaSDarrick J. Wong } 726afc51aaaSDarrick J. Wong 727ce83a025SChristoph Hellwig static loff_t iomap_write_iter(struct iomap_iter *iter, struct iov_iter *i) 728afc51aaaSDarrick J. Wong { 729ce83a025SChristoph Hellwig loff_t length = iomap_length(iter); 730ce83a025SChristoph Hellwig loff_t pos = iter->pos; 731afc51aaaSDarrick J. Wong ssize_t written = 0; 732ce83a025SChristoph Hellwig long status = 0; 733afc51aaaSDarrick J. Wong 734afc51aaaSDarrick J. Wong do { 735afc51aaaSDarrick J. Wong struct page *page; 736afc51aaaSDarrick J. Wong unsigned long offset; /* Offset into pagecache page */ 737afc51aaaSDarrick J. Wong unsigned long bytes; /* Bytes to write to page */ 738afc51aaaSDarrick J. Wong size_t copied; /* Bytes copied from user */ 739afc51aaaSDarrick J. Wong 740afc51aaaSDarrick J. Wong offset = offset_in_page(pos); 741afc51aaaSDarrick J. Wong bytes = min_t(unsigned long, PAGE_SIZE - offset, 742afc51aaaSDarrick J. Wong iov_iter_count(i)); 743afc51aaaSDarrick J. Wong again: 744afc51aaaSDarrick J. Wong if (bytes > length) 745afc51aaaSDarrick J. Wong bytes = length; 746afc51aaaSDarrick J. Wong 747afc51aaaSDarrick J. Wong /* 748f1f264b4SAndreas Gruenbacher * Bring in the user page that we'll copy from _first_. 749afc51aaaSDarrick J. Wong * Otherwise there's a nasty deadlock on copying from the 750afc51aaaSDarrick J. Wong * same page as we're writing to, without it being marked 751afc51aaaSDarrick J. Wong * up-to-date. 752afc51aaaSDarrick J. Wong */ 753*631f871fSAndreas Gruenbacher if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) { 754afc51aaaSDarrick J. Wong status = -EFAULT; 755afc51aaaSDarrick J. Wong break; 756afc51aaaSDarrick J. Wong } 757afc51aaaSDarrick J. Wong 758b74b1293SChristoph Hellwig status = iomap_write_begin(iter, pos, bytes, &page); 759afc51aaaSDarrick J. Wong if (unlikely(status)) 760afc51aaaSDarrick J. Wong break; 761afc51aaaSDarrick J. Wong 762ce83a025SChristoph Hellwig if (mapping_writably_mapped(iter->inode->i_mapping)) 763afc51aaaSDarrick J. Wong flush_dcache_page(page); 764afc51aaaSDarrick J. Wong 765f0b65f39SAl Viro copied = copy_page_from_iter_atomic(page, offset, bytes, i); 766afc51aaaSDarrick J. Wong 7671b5c1e36SChristoph Hellwig status = iomap_write_end(iter, pos, bytes, copied, page); 768afc51aaaSDarrick J. Wong 769f0b65f39SAl Viro if (unlikely(copied != status)) 770f0b65f39SAl Viro iov_iter_revert(i, copied - status); 771afc51aaaSDarrick J. Wong 772f0b65f39SAl Viro cond_resched(); 773bc1bb416SAl Viro if (unlikely(status == 0)) { 774afc51aaaSDarrick J. Wong /* 775bc1bb416SAl Viro * A short copy made iomap_write_end() reject the 776bc1bb416SAl Viro * thing entirely. Might be memory poisoning 777bc1bb416SAl Viro * halfway through, might be a race with munmap, 778bc1bb416SAl Viro * might be severe memory pressure. 779afc51aaaSDarrick J. Wong */ 780bc1bb416SAl Viro if (copied) 781bc1bb416SAl Viro bytes = copied; 782afc51aaaSDarrick J. Wong goto again; 783afc51aaaSDarrick J. Wong } 784f0b65f39SAl Viro pos += status; 785f0b65f39SAl Viro written += status; 786f0b65f39SAl Viro length -= status; 787afc51aaaSDarrick J. Wong 788ce83a025SChristoph Hellwig balance_dirty_pages_ratelimited(iter->inode->i_mapping); 789afc51aaaSDarrick J. Wong } while (iov_iter_count(i) && length); 790afc51aaaSDarrick J. Wong 791afc51aaaSDarrick J. Wong return written ? written : status; 792afc51aaaSDarrick J. Wong } 793afc51aaaSDarrick J. Wong 794afc51aaaSDarrick J. Wong ssize_t 795ce83a025SChristoph Hellwig iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *i, 796afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 797afc51aaaSDarrick J. Wong { 798ce83a025SChristoph Hellwig struct iomap_iter iter = { 799ce83a025SChristoph Hellwig .inode = iocb->ki_filp->f_mapping->host, 800ce83a025SChristoph Hellwig .pos = iocb->ki_pos, 801ce83a025SChristoph Hellwig .len = iov_iter_count(i), 802ce83a025SChristoph Hellwig .flags = IOMAP_WRITE, 803ce83a025SChristoph Hellwig }; 804ce83a025SChristoph Hellwig int ret; 805afc51aaaSDarrick J. Wong 806ce83a025SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 807ce83a025SChristoph Hellwig iter.processed = iomap_write_iter(&iter, i); 808ce83a025SChristoph Hellwig if (iter.pos == iocb->ki_pos) 809ce83a025SChristoph Hellwig return ret; 810ce83a025SChristoph Hellwig return iter.pos - iocb->ki_pos; 811afc51aaaSDarrick J. Wong } 812afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_file_buffered_write); 813afc51aaaSDarrick J. Wong 8148fc274d1SChristoph Hellwig static loff_t iomap_unshare_iter(struct iomap_iter *iter) 815afc51aaaSDarrick J. Wong { 8168fc274d1SChristoph Hellwig struct iomap *iomap = &iter->iomap; 817fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 8188fc274d1SChristoph Hellwig loff_t pos = iter->pos; 8198fc274d1SChristoph Hellwig loff_t length = iomap_length(iter); 820afc51aaaSDarrick J. Wong long status = 0; 821d4ff3b2eSMatthew Wilcox (Oracle) loff_t written = 0; 822afc51aaaSDarrick J. Wong 8233590c4d8SChristoph Hellwig /* don't bother with blocks that are not shared to start with */ 8243590c4d8SChristoph Hellwig if (!(iomap->flags & IOMAP_F_SHARED)) 8253590c4d8SChristoph Hellwig return length; 8263590c4d8SChristoph Hellwig /* don't bother with holes or unwritten extents */ 827c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 8283590c4d8SChristoph Hellwig return length; 8293590c4d8SChristoph Hellwig 830afc51aaaSDarrick J. Wong do { 83132a38a49SChristoph Hellwig unsigned long offset = offset_in_page(pos); 83232a38a49SChristoph Hellwig unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length); 83332a38a49SChristoph Hellwig struct page *page; 834afc51aaaSDarrick J. Wong 835b74b1293SChristoph Hellwig status = iomap_write_begin(iter, pos, bytes, &page); 836afc51aaaSDarrick J. Wong if (unlikely(status)) 837afc51aaaSDarrick J. Wong return status; 838afc51aaaSDarrick J. Wong 8391b5c1e36SChristoph Hellwig status = iomap_write_end(iter, pos, bytes, bytes, page); 840afc51aaaSDarrick J. Wong if (WARN_ON_ONCE(status == 0)) 841afc51aaaSDarrick J. Wong return -EIO; 842afc51aaaSDarrick J. Wong 843afc51aaaSDarrick J. Wong cond_resched(); 844afc51aaaSDarrick J. Wong 845afc51aaaSDarrick J. Wong pos += status; 846afc51aaaSDarrick J. Wong written += status; 847afc51aaaSDarrick J. Wong length -= status; 848afc51aaaSDarrick J. Wong 8498fc274d1SChristoph Hellwig balance_dirty_pages_ratelimited(iter->inode->i_mapping); 850afc51aaaSDarrick J. Wong } while (length); 851afc51aaaSDarrick J. Wong 852afc51aaaSDarrick J. Wong return written; 853afc51aaaSDarrick J. Wong } 854afc51aaaSDarrick J. Wong 855afc51aaaSDarrick J. Wong int 8563590c4d8SChristoph Hellwig iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, 857afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 858afc51aaaSDarrick J. Wong { 8598fc274d1SChristoph Hellwig struct iomap_iter iter = { 8608fc274d1SChristoph Hellwig .inode = inode, 8618fc274d1SChristoph Hellwig .pos = pos, 8628fc274d1SChristoph Hellwig .len = len, 863b74b1293SChristoph Hellwig .flags = IOMAP_WRITE | IOMAP_UNSHARE, 8648fc274d1SChristoph Hellwig }; 8658fc274d1SChristoph Hellwig int ret; 866afc51aaaSDarrick J. Wong 8678fc274d1SChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 8688fc274d1SChristoph Hellwig iter.processed = iomap_unshare_iter(&iter); 869afc51aaaSDarrick J. Wong return ret; 870afc51aaaSDarrick J. Wong } 8713590c4d8SChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_file_unshare); 872afc51aaaSDarrick J. Wong 8731b5c1e36SChristoph Hellwig static s64 __iomap_zero_iter(struct iomap_iter *iter, loff_t pos, u64 length) 874afc51aaaSDarrick J. Wong { 875afc51aaaSDarrick J. Wong struct page *page; 876afc51aaaSDarrick J. Wong int status; 87781ee8e52SMatthew Wilcox (Oracle) unsigned offset = offset_in_page(pos); 87881ee8e52SMatthew Wilcox (Oracle) unsigned bytes = min_t(u64, PAGE_SIZE - offset, length); 879afc51aaaSDarrick J. Wong 880b74b1293SChristoph Hellwig status = iomap_write_begin(iter, pos, bytes, &page); 881afc51aaaSDarrick J. Wong if (status) 882afc51aaaSDarrick J. Wong return status; 883afc51aaaSDarrick J. Wong 884afc51aaaSDarrick J. Wong zero_user(page, offset, bytes); 885afc51aaaSDarrick J. Wong mark_page_accessed(page); 886afc51aaaSDarrick J. Wong 8871b5c1e36SChristoph Hellwig return iomap_write_end(iter, pos, bytes, bytes, page); 888afc51aaaSDarrick J. Wong } 889afc51aaaSDarrick J. Wong 8902aa3048eSChristoph Hellwig static loff_t iomap_zero_iter(struct iomap_iter *iter, bool *did_zero) 891afc51aaaSDarrick J. Wong { 8922aa3048eSChristoph Hellwig struct iomap *iomap = &iter->iomap; 893fad0a1abSChristoph Hellwig const struct iomap *srcmap = iomap_iter_srcmap(iter); 8942aa3048eSChristoph Hellwig loff_t pos = iter->pos; 8952aa3048eSChristoph Hellwig loff_t length = iomap_length(iter); 896afc51aaaSDarrick J. Wong loff_t written = 0; 897afc51aaaSDarrick J. Wong 898afc51aaaSDarrick J. Wong /* already zeroed? we're done. */ 899c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 90081ee8e52SMatthew Wilcox (Oracle) return length; 901afc51aaaSDarrick J. Wong 902afc51aaaSDarrick J. Wong do { 90381ee8e52SMatthew Wilcox (Oracle) s64 bytes; 904afc51aaaSDarrick J. Wong 9052aa3048eSChristoph Hellwig if (IS_DAX(iter->inode)) 90681ee8e52SMatthew Wilcox (Oracle) bytes = dax_iomap_zero(pos, length, iomap); 907afc51aaaSDarrick J. Wong else 9081b5c1e36SChristoph Hellwig bytes = __iomap_zero_iter(iter, pos, length); 90981ee8e52SMatthew Wilcox (Oracle) if (bytes < 0) 91081ee8e52SMatthew Wilcox (Oracle) return bytes; 911afc51aaaSDarrick J. Wong 912afc51aaaSDarrick J. Wong pos += bytes; 91381ee8e52SMatthew Wilcox (Oracle) length -= bytes; 914afc51aaaSDarrick J. Wong written += bytes; 915afc51aaaSDarrick J. Wong if (did_zero) 916afc51aaaSDarrick J. Wong *did_zero = true; 91781ee8e52SMatthew Wilcox (Oracle) } while (length > 0); 918afc51aaaSDarrick J. Wong 919afc51aaaSDarrick J. Wong return written; 920afc51aaaSDarrick J. Wong } 921afc51aaaSDarrick J. Wong 922afc51aaaSDarrick J. Wong int 923afc51aaaSDarrick J. Wong iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, 924afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 925afc51aaaSDarrick J. Wong { 9262aa3048eSChristoph Hellwig struct iomap_iter iter = { 9272aa3048eSChristoph Hellwig .inode = inode, 9282aa3048eSChristoph Hellwig .pos = pos, 9292aa3048eSChristoph Hellwig .len = len, 9302aa3048eSChristoph Hellwig .flags = IOMAP_ZERO, 9312aa3048eSChristoph Hellwig }; 9322aa3048eSChristoph Hellwig int ret; 933afc51aaaSDarrick J. Wong 9342aa3048eSChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 9352aa3048eSChristoph Hellwig iter.processed = iomap_zero_iter(&iter, did_zero); 936afc51aaaSDarrick J. Wong return ret; 937afc51aaaSDarrick J. Wong } 938afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_zero_range); 939afc51aaaSDarrick J. Wong 940afc51aaaSDarrick J. Wong int 941afc51aaaSDarrick J. Wong iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, 942afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 943afc51aaaSDarrick J. Wong { 944afc51aaaSDarrick J. Wong unsigned int blocksize = i_blocksize(inode); 945afc51aaaSDarrick J. Wong unsigned int off = pos & (blocksize - 1); 946afc51aaaSDarrick J. Wong 947afc51aaaSDarrick J. Wong /* Block boundary? Nothing to do */ 948afc51aaaSDarrick J. Wong if (!off) 949afc51aaaSDarrick J. Wong return 0; 950afc51aaaSDarrick J. Wong return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); 951afc51aaaSDarrick J. Wong } 952afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_truncate_page); 953afc51aaaSDarrick J. Wong 954253564baSChristoph Hellwig static loff_t iomap_page_mkwrite_iter(struct iomap_iter *iter, 955253564baSChristoph Hellwig struct page *page) 956afc51aaaSDarrick J. Wong { 957253564baSChristoph Hellwig loff_t length = iomap_length(iter); 958afc51aaaSDarrick J. Wong int ret; 959afc51aaaSDarrick J. Wong 960253564baSChristoph Hellwig if (iter->iomap.flags & IOMAP_F_BUFFER_HEAD) { 961253564baSChristoph Hellwig ret = __block_write_begin_int(page, iter->pos, length, NULL, 962253564baSChristoph Hellwig &iter->iomap); 963afc51aaaSDarrick J. Wong if (ret) 964afc51aaaSDarrick J. Wong return ret; 965afc51aaaSDarrick J. Wong block_commit_write(page, 0, length); 966afc51aaaSDarrick J. Wong } else { 967afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 968afc51aaaSDarrick J. Wong set_page_dirty(page); 969afc51aaaSDarrick J. Wong } 970afc51aaaSDarrick J. Wong 971afc51aaaSDarrick J. Wong return length; 972afc51aaaSDarrick J. Wong } 973afc51aaaSDarrick J. Wong 974afc51aaaSDarrick J. Wong vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) 975afc51aaaSDarrick J. Wong { 976253564baSChristoph Hellwig struct iomap_iter iter = { 977253564baSChristoph Hellwig .inode = file_inode(vmf->vma->vm_file), 978253564baSChristoph Hellwig .flags = IOMAP_WRITE | IOMAP_FAULT, 979253564baSChristoph Hellwig }; 980afc51aaaSDarrick J. Wong struct page *page = vmf->page; 981afc51aaaSDarrick J. Wong ssize_t ret; 982afc51aaaSDarrick J. Wong 983afc51aaaSDarrick J. Wong lock_page(page); 984253564baSChristoph Hellwig ret = page_mkwrite_check_truncate(page, iter.inode); 985243145bcSAndreas Gruenbacher if (ret < 0) 986afc51aaaSDarrick J. Wong goto out_unlock; 987253564baSChristoph Hellwig iter.pos = page_offset(page); 988253564baSChristoph Hellwig iter.len = ret; 989253564baSChristoph Hellwig while ((ret = iomap_iter(&iter, ops)) > 0) 990253564baSChristoph Hellwig iter.processed = iomap_page_mkwrite_iter(&iter, page); 991afc51aaaSDarrick J. Wong 992253564baSChristoph Hellwig if (ret < 0) 993afc51aaaSDarrick J. Wong goto out_unlock; 994afc51aaaSDarrick J. Wong wait_for_stable_page(page); 995afc51aaaSDarrick J. Wong return VM_FAULT_LOCKED; 996afc51aaaSDarrick J. Wong out_unlock: 997afc51aaaSDarrick J. Wong unlock_page(page); 998afc51aaaSDarrick J. Wong return block_page_mkwrite_return(ret); 999afc51aaaSDarrick J. Wong } 1000afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_page_mkwrite); 1001598ecfbaSChristoph Hellwig 1002598ecfbaSChristoph Hellwig static void 100348d64cd1SChristoph Hellwig iomap_finish_page_writeback(struct inode *inode, struct page *page, 10040fb2d720SMatthew Wilcox (Oracle) int error, unsigned int len) 1005598ecfbaSChristoph Hellwig { 100648d64cd1SChristoph Hellwig struct iomap_page *iop = to_iomap_page(page); 1007598ecfbaSChristoph Hellwig 1008598ecfbaSChristoph Hellwig if (error) { 100948d64cd1SChristoph Hellwig SetPageError(page); 1010b69eea82SDarrick J. Wong mapping_set_error(inode->i_mapping, error); 1011598ecfbaSChristoph Hellwig } 1012598ecfbaSChristoph Hellwig 101324addd84SMatthew Wilcox (Oracle) WARN_ON_ONCE(i_blocks_per_page(inode, page) > 1 && !iop); 10140fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) <= 0); 1015598ecfbaSChristoph Hellwig 10160fb2d720SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(len, &iop->write_bytes_pending)) 101748d64cd1SChristoph Hellwig end_page_writeback(page); 1018598ecfbaSChristoph Hellwig } 1019598ecfbaSChristoph Hellwig 1020598ecfbaSChristoph Hellwig /* 1021598ecfbaSChristoph Hellwig * We're now finished for good with this ioend structure. Update the page 1022598ecfbaSChristoph Hellwig * state, release holds on bios, and finally free up memory. Do not use the 1023598ecfbaSChristoph Hellwig * ioend after this. 1024598ecfbaSChristoph Hellwig */ 1025598ecfbaSChristoph Hellwig static void 1026598ecfbaSChristoph Hellwig iomap_finish_ioend(struct iomap_ioend *ioend, int error) 1027598ecfbaSChristoph Hellwig { 1028598ecfbaSChristoph Hellwig struct inode *inode = ioend->io_inode; 1029598ecfbaSChristoph Hellwig struct bio *bio = &ioend->io_inline_bio; 1030598ecfbaSChristoph Hellwig struct bio *last = ioend->io_bio, *next; 1031598ecfbaSChristoph Hellwig u64 start = bio->bi_iter.bi_sector; 1032c275779fSZorro Lang loff_t offset = ioend->io_offset; 1033598ecfbaSChristoph Hellwig bool quiet = bio_flagged(bio, BIO_QUIET); 1034598ecfbaSChristoph Hellwig 1035598ecfbaSChristoph Hellwig for (bio = &ioend->io_inline_bio; bio; bio = next) { 1036598ecfbaSChristoph Hellwig struct bio_vec *bv; 1037598ecfbaSChristoph Hellwig struct bvec_iter_all iter_all; 1038598ecfbaSChristoph Hellwig 1039598ecfbaSChristoph Hellwig /* 1040598ecfbaSChristoph Hellwig * For the last bio, bi_private points to the ioend, so we 1041598ecfbaSChristoph Hellwig * need to explicitly end the iteration here. 1042598ecfbaSChristoph Hellwig */ 1043598ecfbaSChristoph Hellwig if (bio == last) 1044598ecfbaSChristoph Hellwig next = NULL; 1045598ecfbaSChristoph Hellwig else 1046598ecfbaSChristoph Hellwig next = bio->bi_private; 1047598ecfbaSChristoph Hellwig 1048598ecfbaSChristoph Hellwig /* walk each page on bio, ending page IO on them */ 1049598ecfbaSChristoph Hellwig bio_for_each_segment_all(bv, bio, iter_all) 10500fb2d720SMatthew Wilcox (Oracle) iomap_finish_page_writeback(inode, bv->bv_page, error, 10510fb2d720SMatthew Wilcox (Oracle) bv->bv_len); 1052598ecfbaSChristoph Hellwig bio_put(bio); 1053598ecfbaSChristoph Hellwig } 1054c275779fSZorro Lang /* The ioend has been freed by bio_put() */ 1055598ecfbaSChristoph Hellwig 1056598ecfbaSChristoph Hellwig if (unlikely(error && !quiet)) { 1057598ecfbaSChristoph Hellwig printk_ratelimited(KERN_ERR 10589cd0ed63SDarrick J. Wong "%s: writeback error on inode %lu, offset %lld, sector %llu", 1059c275779fSZorro Lang inode->i_sb->s_id, inode->i_ino, offset, start); 1060598ecfbaSChristoph Hellwig } 1061598ecfbaSChristoph Hellwig } 1062598ecfbaSChristoph Hellwig 1063598ecfbaSChristoph Hellwig void 1064598ecfbaSChristoph Hellwig iomap_finish_ioends(struct iomap_ioend *ioend, int error) 1065598ecfbaSChristoph Hellwig { 1066598ecfbaSChristoph Hellwig struct list_head tmp; 1067598ecfbaSChristoph Hellwig 1068598ecfbaSChristoph Hellwig list_replace_init(&ioend->io_list, &tmp); 1069598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1070598ecfbaSChristoph Hellwig 1071598ecfbaSChristoph Hellwig while (!list_empty(&tmp)) { 1072598ecfbaSChristoph Hellwig ioend = list_first_entry(&tmp, struct iomap_ioend, io_list); 1073598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1074598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1075598ecfbaSChristoph Hellwig } 1076598ecfbaSChristoph Hellwig } 1077598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_finish_ioends); 1078598ecfbaSChristoph Hellwig 1079598ecfbaSChristoph Hellwig /* 1080598ecfbaSChristoph Hellwig * We can merge two adjacent ioends if they have the same set of work to do. 1081598ecfbaSChristoph Hellwig */ 1082598ecfbaSChristoph Hellwig static bool 1083598ecfbaSChristoph Hellwig iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next) 1084598ecfbaSChristoph Hellwig { 1085598ecfbaSChristoph Hellwig if (ioend->io_bio->bi_status != next->io_bio->bi_status) 1086598ecfbaSChristoph Hellwig return false; 1087598ecfbaSChristoph Hellwig if ((ioend->io_flags & IOMAP_F_SHARED) ^ 1088598ecfbaSChristoph Hellwig (next->io_flags & IOMAP_F_SHARED)) 1089598ecfbaSChristoph Hellwig return false; 1090598ecfbaSChristoph Hellwig if ((ioend->io_type == IOMAP_UNWRITTEN) ^ 1091598ecfbaSChristoph Hellwig (next->io_type == IOMAP_UNWRITTEN)) 1092598ecfbaSChristoph Hellwig return false; 1093598ecfbaSChristoph Hellwig if (ioend->io_offset + ioend->io_size != next->io_offset) 1094598ecfbaSChristoph Hellwig return false; 1095598ecfbaSChristoph Hellwig return true; 1096598ecfbaSChristoph Hellwig } 1097598ecfbaSChristoph Hellwig 1098598ecfbaSChristoph Hellwig void 10996e552494SBrian Foster iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends) 1100598ecfbaSChristoph Hellwig { 1101598ecfbaSChristoph Hellwig struct iomap_ioend *next; 1102598ecfbaSChristoph Hellwig 1103598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1104598ecfbaSChristoph Hellwig 1105598ecfbaSChristoph Hellwig while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend, 1106598ecfbaSChristoph Hellwig io_list))) { 1107598ecfbaSChristoph Hellwig if (!iomap_ioend_can_merge(ioend, next)) 1108598ecfbaSChristoph Hellwig break; 1109598ecfbaSChristoph Hellwig list_move_tail(&next->io_list, &ioend->io_list); 1110598ecfbaSChristoph Hellwig ioend->io_size += next->io_size; 1111598ecfbaSChristoph Hellwig } 1112598ecfbaSChristoph Hellwig } 1113598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_ioend_try_merge); 1114598ecfbaSChristoph Hellwig 1115598ecfbaSChristoph Hellwig static int 11164f0f586bSSami Tolvanen iomap_ioend_compare(void *priv, const struct list_head *a, 11174f0f586bSSami Tolvanen const struct list_head *b) 1118598ecfbaSChristoph Hellwig { 1119b3d423ecSChristoph Hellwig struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list); 1120b3d423ecSChristoph Hellwig struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list); 1121598ecfbaSChristoph Hellwig 1122598ecfbaSChristoph Hellwig if (ia->io_offset < ib->io_offset) 1123598ecfbaSChristoph Hellwig return -1; 1124b3d423ecSChristoph Hellwig if (ia->io_offset > ib->io_offset) 1125598ecfbaSChristoph Hellwig return 1; 1126598ecfbaSChristoph Hellwig return 0; 1127598ecfbaSChristoph Hellwig } 1128598ecfbaSChristoph Hellwig 1129598ecfbaSChristoph Hellwig void 1130598ecfbaSChristoph Hellwig iomap_sort_ioends(struct list_head *ioend_list) 1131598ecfbaSChristoph Hellwig { 1132598ecfbaSChristoph Hellwig list_sort(NULL, ioend_list, iomap_ioend_compare); 1133598ecfbaSChristoph Hellwig } 1134598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_sort_ioends); 1135598ecfbaSChristoph Hellwig 1136598ecfbaSChristoph Hellwig static void iomap_writepage_end_bio(struct bio *bio) 1137598ecfbaSChristoph Hellwig { 1138598ecfbaSChristoph Hellwig struct iomap_ioend *ioend = bio->bi_private; 1139598ecfbaSChristoph Hellwig 1140598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status)); 1141598ecfbaSChristoph Hellwig } 1142598ecfbaSChristoph Hellwig 1143598ecfbaSChristoph Hellwig /* 1144598ecfbaSChristoph Hellwig * Submit the final bio for an ioend. 1145598ecfbaSChristoph Hellwig * 1146598ecfbaSChristoph Hellwig * If @error is non-zero, it means that we have a situation where some part of 1147f1f264b4SAndreas Gruenbacher * the submission process has failed after we've marked pages for writeback 1148598ecfbaSChristoph Hellwig * and unlocked them. In this situation, we need to fail the bio instead of 1149598ecfbaSChristoph Hellwig * submitting it. This typically only happens on a filesystem shutdown. 1150598ecfbaSChristoph Hellwig */ 1151598ecfbaSChristoph Hellwig static int 1152598ecfbaSChristoph Hellwig iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend, 1153598ecfbaSChristoph Hellwig int error) 1154598ecfbaSChristoph Hellwig { 1155598ecfbaSChristoph Hellwig ioend->io_bio->bi_private = ioend; 1156598ecfbaSChristoph Hellwig ioend->io_bio->bi_end_io = iomap_writepage_end_bio; 1157598ecfbaSChristoph Hellwig 1158598ecfbaSChristoph Hellwig if (wpc->ops->prepare_ioend) 1159598ecfbaSChristoph Hellwig error = wpc->ops->prepare_ioend(ioend, error); 1160598ecfbaSChristoph Hellwig if (error) { 1161598ecfbaSChristoph Hellwig /* 1162f1f264b4SAndreas Gruenbacher * If we're failing the IO now, just mark the ioend with an 1163598ecfbaSChristoph Hellwig * error and finish it. This will run IO completion immediately 1164598ecfbaSChristoph Hellwig * as there is only one reference to the ioend at this point in 1165598ecfbaSChristoph Hellwig * time. 1166598ecfbaSChristoph Hellwig */ 1167598ecfbaSChristoph Hellwig ioend->io_bio->bi_status = errno_to_blk_status(error); 1168598ecfbaSChristoph Hellwig bio_endio(ioend->io_bio); 1169598ecfbaSChristoph Hellwig return error; 1170598ecfbaSChristoph Hellwig } 1171598ecfbaSChristoph Hellwig 1172598ecfbaSChristoph Hellwig submit_bio(ioend->io_bio); 1173598ecfbaSChristoph Hellwig return 0; 1174598ecfbaSChristoph Hellwig } 1175598ecfbaSChristoph Hellwig 1176598ecfbaSChristoph Hellwig static struct iomap_ioend * 1177598ecfbaSChristoph Hellwig iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc, 1178598ecfbaSChristoph Hellwig loff_t offset, sector_t sector, struct writeback_control *wbc) 1179598ecfbaSChristoph Hellwig { 1180598ecfbaSChristoph Hellwig struct iomap_ioend *ioend; 1181598ecfbaSChristoph Hellwig struct bio *bio; 1182598ecfbaSChristoph Hellwig 1183a8affc03SChristoph Hellwig bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, &iomap_ioend_bioset); 1184598ecfbaSChristoph Hellwig bio_set_dev(bio, wpc->iomap.bdev); 1185598ecfbaSChristoph Hellwig bio->bi_iter.bi_sector = sector; 1186598ecfbaSChristoph Hellwig bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); 1187598ecfbaSChristoph Hellwig bio->bi_write_hint = inode->i_write_hint; 1188598ecfbaSChristoph Hellwig wbc_init_bio(wbc, bio); 1189598ecfbaSChristoph Hellwig 1190598ecfbaSChristoph Hellwig ioend = container_of(bio, struct iomap_ioend, io_inline_bio); 1191598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1192598ecfbaSChristoph Hellwig ioend->io_type = wpc->iomap.type; 1193598ecfbaSChristoph Hellwig ioend->io_flags = wpc->iomap.flags; 1194598ecfbaSChristoph Hellwig ioend->io_inode = inode; 1195598ecfbaSChristoph Hellwig ioend->io_size = 0; 1196598ecfbaSChristoph Hellwig ioend->io_offset = offset; 1197598ecfbaSChristoph Hellwig ioend->io_bio = bio; 1198598ecfbaSChristoph Hellwig return ioend; 1199598ecfbaSChristoph Hellwig } 1200598ecfbaSChristoph Hellwig 1201598ecfbaSChristoph Hellwig /* 1202598ecfbaSChristoph Hellwig * Allocate a new bio, and chain the old bio to the new one. 1203598ecfbaSChristoph Hellwig * 1204f1f264b4SAndreas Gruenbacher * Note that we have to perform the chaining in this unintuitive order 1205598ecfbaSChristoph Hellwig * so that the bi_private linkage is set up in the right direction for the 1206598ecfbaSChristoph Hellwig * traversal in iomap_finish_ioend(). 1207598ecfbaSChristoph Hellwig */ 1208598ecfbaSChristoph Hellwig static struct bio * 1209598ecfbaSChristoph Hellwig iomap_chain_bio(struct bio *prev) 1210598ecfbaSChristoph Hellwig { 1211598ecfbaSChristoph Hellwig struct bio *new; 1212598ecfbaSChristoph Hellwig 1213a8affc03SChristoph Hellwig new = bio_alloc(GFP_NOFS, BIO_MAX_VECS); 1214598ecfbaSChristoph Hellwig bio_copy_dev(new, prev);/* also copies over blkcg information */ 1215598ecfbaSChristoph Hellwig new->bi_iter.bi_sector = bio_end_sector(prev); 1216598ecfbaSChristoph Hellwig new->bi_opf = prev->bi_opf; 1217598ecfbaSChristoph Hellwig new->bi_write_hint = prev->bi_write_hint; 1218598ecfbaSChristoph Hellwig 1219598ecfbaSChristoph Hellwig bio_chain(prev, new); 1220598ecfbaSChristoph Hellwig bio_get(prev); /* for iomap_finish_ioend */ 1221598ecfbaSChristoph Hellwig submit_bio(prev); 1222598ecfbaSChristoph Hellwig return new; 1223598ecfbaSChristoph Hellwig } 1224598ecfbaSChristoph Hellwig 1225598ecfbaSChristoph Hellwig static bool 1226598ecfbaSChristoph Hellwig iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset, 1227598ecfbaSChristoph Hellwig sector_t sector) 1228598ecfbaSChristoph Hellwig { 1229598ecfbaSChristoph Hellwig if ((wpc->iomap.flags & IOMAP_F_SHARED) != 1230598ecfbaSChristoph Hellwig (wpc->ioend->io_flags & IOMAP_F_SHARED)) 1231598ecfbaSChristoph Hellwig return false; 1232598ecfbaSChristoph Hellwig if (wpc->iomap.type != wpc->ioend->io_type) 1233598ecfbaSChristoph Hellwig return false; 1234598ecfbaSChristoph Hellwig if (offset != wpc->ioend->io_offset + wpc->ioend->io_size) 1235598ecfbaSChristoph Hellwig return false; 1236598ecfbaSChristoph Hellwig if (sector != bio_end_sector(wpc->ioend->io_bio)) 1237598ecfbaSChristoph Hellwig return false; 1238598ecfbaSChristoph Hellwig return true; 1239598ecfbaSChristoph Hellwig } 1240598ecfbaSChristoph Hellwig 1241598ecfbaSChristoph Hellwig /* 1242598ecfbaSChristoph Hellwig * Test to see if we have an existing ioend structure that we could append to 1243f1f264b4SAndreas Gruenbacher * first; otherwise finish off the current ioend and start another. 1244598ecfbaSChristoph Hellwig */ 1245598ecfbaSChristoph Hellwig static void 1246598ecfbaSChristoph Hellwig iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page, 1247598ecfbaSChristoph Hellwig struct iomap_page *iop, struct iomap_writepage_ctx *wpc, 1248598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct list_head *iolist) 1249598ecfbaSChristoph Hellwig { 1250598ecfbaSChristoph Hellwig sector_t sector = iomap_sector(&wpc->iomap, offset); 1251598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1252598ecfbaSChristoph Hellwig unsigned poff = offset & (PAGE_SIZE - 1); 1253598ecfbaSChristoph Hellwig 1254598ecfbaSChristoph Hellwig if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) { 1255598ecfbaSChristoph Hellwig if (wpc->ioend) 1256598ecfbaSChristoph Hellwig list_add(&wpc->ioend->io_list, iolist); 1257598ecfbaSChristoph Hellwig wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc); 1258598ecfbaSChristoph Hellwig } 1259598ecfbaSChristoph Hellwig 1260c1b79f11SChristoph Hellwig if (bio_add_page(wpc->ioend->io_bio, page, len, poff) != len) { 1261c1b79f11SChristoph Hellwig wpc->ioend->io_bio = iomap_chain_bio(wpc->ioend->io_bio); 1262c1b79f11SChristoph Hellwig __bio_add_page(wpc->ioend->io_bio, page, len, poff); 1263c1b79f11SChristoph Hellwig } 1264c1b79f11SChristoph Hellwig 12650fb2d720SMatthew Wilcox (Oracle) if (iop) 12660fb2d720SMatthew Wilcox (Oracle) atomic_add(len, &iop->write_bytes_pending); 1267598ecfbaSChristoph Hellwig wpc->ioend->io_size += len; 1268598ecfbaSChristoph Hellwig wbc_account_cgroup_owner(wbc, page, len); 1269598ecfbaSChristoph Hellwig } 1270598ecfbaSChristoph Hellwig 1271598ecfbaSChristoph Hellwig /* 1272598ecfbaSChristoph Hellwig * We implement an immediate ioend submission policy here to avoid needing to 1273598ecfbaSChristoph Hellwig * chain multiple ioends and hence nest mempool allocations which can violate 1274f1f264b4SAndreas Gruenbacher * the forward progress guarantees we need to provide. The current ioend we're 1275f1f264b4SAndreas Gruenbacher * adding blocks to is cached in the writepage context, and if the new block 1276f1f264b4SAndreas Gruenbacher * doesn't append to the cached ioend, it will create a new ioend and cache that 1277598ecfbaSChristoph Hellwig * instead. 1278598ecfbaSChristoph Hellwig * 1279598ecfbaSChristoph Hellwig * If a new ioend is created and cached, the old ioend is returned and queued 1280598ecfbaSChristoph Hellwig * locally for submission once the entire page is processed or an error has been 1281598ecfbaSChristoph Hellwig * detected. While ioends are submitted immediately after they are completed, 1282598ecfbaSChristoph Hellwig * batching optimisations are provided by higher level block plugging. 1283598ecfbaSChristoph Hellwig * 1284598ecfbaSChristoph Hellwig * At the end of a writeback pass, there will be a cached ioend remaining on the 1285598ecfbaSChristoph Hellwig * writepage context that the caller will need to submit. 1286598ecfbaSChristoph Hellwig */ 1287598ecfbaSChristoph Hellwig static int 1288598ecfbaSChristoph Hellwig iomap_writepage_map(struct iomap_writepage_ctx *wpc, 1289598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct inode *inode, 1290598ecfbaSChristoph Hellwig struct page *page, u64 end_offset) 1291598ecfbaSChristoph Hellwig { 12928e1bcef8SAndreas Gruenbacher struct iomap_page *iop = iomap_page_create(inode, page); 1293598ecfbaSChristoph Hellwig struct iomap_ioend *ioend, *next; 1294598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1295598ecfbaSChristoph Hellwig u64 file_offset; /* file offset of page */ 1296598ecfbaSChristoph Hellwig int error = 0, count = 0, i; 1297598ecfbaSChristoph Hellwig LIST_HEAD(submit_list); 1298598ecfbaSChristoph Hellwig 12990fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) != 0); 1300598ecfbaSChristoph Hellwig 1301598ecfbaSChristoph Hellwig /* 1302598ecfbaSChristoph Hellwig * Walk through the page to find areas to write back. If we run off the 1303598ecfbaSChristoph Hellwig * end of the current map or find the current map invalid, grab a new 1304598ecfbaSChristoph Hellwig * one. 1305598ecfbaSChristoph Hellwig */ 1306598ecfbaSChristoph Hellwig for (i = 0, file_offset = page_offset(page); 1307598ecfbaSChristoph Hellwig i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset; 1308598ecfbaSChristoph Hellwig i++, file_offset += len) { 1309598ecfbaSChristoph Hellwig if (iop && !test_bit(i, iop->uptodate)) 1310598ecfbaSChristoph Hellwig continue; 1311598ecfbaSChristoph Hellwig 1312598ecfbaSChristoph Hellwig error = wpc->ops->map_blocks(wpc, inode, file_offset); 1313598ecfbaSChristoph Hellwig if (error) 1314598ecfbaSChristoph Hellwig break; 13153e19e6f3SChristoph Hellwig if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE)) 13163e19e6f3SChristoph Hellwig continue; 1317598ecfbaSChristoph Hellwig if (wpc->iomap.type == IOMAP_HOLE) 1318598ecfbaSChristoph Hellwig continue; 1319598ecfbaSChristoph Hellwig iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc, 1320598ecfbaSChristoph Hellwig &submit_list); 1321598ecfbaSChristoph Hellwig count++; 1322598ecfbaSChristoph Hellwig } 1323598ecfbaSChristoph Hellwig 1324598ecfbaSChristoph Hellwig WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list)); 1325598ecfbaSChristoph Hellwig WARN_ON_ONCE(!PageLocked(page)); 1326598ecfbaSChristoph Hellwig WARN_ON_ONCE(PageWriteback(page)); 132750e7d6c7SBrian Foster WARN_ON_ONCE(PageDirty(page)); 1328598ecfbaSChristoph Hellwig 1329598ecfbaSChristoph Hellwig /* 1330598ecfbaSChristoph Hellwig * We cannot cancel the ioend directly here on error. We may have 1331598ecfbaSChristoph Hellwig * already set other pages under writeback and hence we have to run I/O 1332598ecfbaSChristoph Hellwig * completion to mark the error state of the pages under writeback 1333598ecfbaSChristoph Hellwig * appropriately. 1334598ecfbaSChristoph Hellwig */ 1335598ecfbaSChristoph Hellwig if (unlikely(error)) { 1336598ecfbaSChristoph Hellwig /* 1337763e4cdcSBrian Foster * Let the filesystem know what portion of the current page 1338f1f264b4SAndreas Gruenbacher * failed to map. If the page hasn't been added to ioend, it 1339763e4cdcSBrian Foster * won't be affected by I/O completion and we must unlock it 1340763e4cdcSBrian Foster * now. 1341598ecfbaSChristoph Hellwig */ 1342598ecfbaSChristoph Hellwig if (wpc->ops->discard_page) 1343763e4cdcSBrian Foster wpc->ops->discard_page(page, file_offset); 1344763e4cdcSBrian Foster if (!count) { 1345598ecfbaSChristoph Hellwig ClearPageUptodate(page); 1346598ecfbaSChristoph Hellwig unlock_page(page); 1347598ecfbaSChristoph Hellwig goto done; 1348598ecfbaSChristoph Hellwig } 1349598ecfbaSChristoph Hellwig } 1350598ecfbaSChristoph Hellwig 135150e7d6c7SBrian Foster set_page_writeback(page); 1352598ecfbaSChristoph Hellwig unlock_page(page); 1353598ecfbaSChristoph Hellwig 1354598ecfbaSChristoph Hellwig /* 1355f1f264b4SAndreas Gruenbacher * Preserve the original error if there was one; catch 1356598ecfbaSChristoph Hellwig * submission errors here and propagate into subsequent ioend 1357598ecfbaSChristoph Hellwig * submissions. 1358598ecfbaSChristoph Hellwig */ 1359598ecfbaSChristoph Hellwig list_for_each_entry_safe(ioend, next, &submit_list, io_list) { 1360598ecfbaSChristoph Hellwig int error2; 1361598ecfbaSChristoph Hellwig 1362598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1363598ecfbaSChristoph Hellwig error2 = iomap_submit_ioend(wpc, ioend, error); 1364598ecfbaSChristoph Hellwig if (error2 && !error) 1365598ecfbaSChristoph Hellwig error = error2; 1366598ecfbaSChristoph Hellwig } 1367598ecfbaSChristoph Hellwig 1368598ecfbaSChristoph Hellwig /* 1369598ecfbaSChristoph Hellwig * We can end up here with no error and nothing to write only if we race 1370598ecfbaSChristoph Hellwig * with a partial page truncate on a sub-page block sized filesystem. 1371598ecfbaSChristoph Hellwig */ 1372598ecfbaSChristoph Hellwig if (!count) 1373598ecfbaSChristoph Hellwig end_page_writeback(page); 1374598ecfbaSChristoph Hellwig done: 1375598ecfbaSChristoph Hellwig mapping_set_error(page->mapping, error); 1376598ecfbaSChristoph Hellwig return error; 1377598ecfbaSChristoph Hellwig } 1378598ecfbaSChristoph Hellwig 1379598ecfbaSChristoph Hellwig /* 1380598ecfbaSChristoph Hellwig * Write out a dirty page. 1381598ecfbaSChristoph Hellwig * 1382f1f264b4SAndreas Gruenbacher * For delalloc space on the page, we need to allocate space and flush it. 1383f1f264b4SAndreas Gruenbacher * For unwritten space on the page, we need to start the conversion to 1384598ecfbaSChristoph Hellwig * regular allocated space. 1385598ecfbaSChristoph Hellwig */ 1386598ecfbaSChristoph Hellwig static int 1387598ecfbaSChristoph Hellwig iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data) 1388598ecfbaSChristoph Hellwig { 1389598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc = data; 1390598ecfbaSChristoph Hellwig struct inode *inode = page->mapping->host; 1391598ecfbaSChristoph Hellwig pgoff_t end_index; 1392598ecfbaSChristoph Hellwig u64 end_offset; 1393598ecfbaSChristoph Hellwig loff_t offset; 1394598ecfbaSChristoph Hellwig 13951ac99452SMatthew Wilcox (Oracle) trace_iomap_writepage(inode, page_offset(page), PAGE_SIZE); 1396598ecfbaSChristoph Hellwig 1397598ecfbaSChristoph Hellwig /* 1398f1f264b4SAndreas Gruenbacher * Refuse to write the page out if we're called from reclaim context. 1399598ecfbaSChristoph Hellwig * 1400598ecfbaSChristoph Hellwig * This avoids stack overflows when called from deeply used stacks in 1401598ecfbaSChristoph Hellwig * random callers for direct reclaim or memcg reclaim. We explicitly 1402598ecfbaSChristoph Hellwig * allow reclaim from kswapd as the stack usage there is relatively low. 1403598ecfbaSChristoph Hellwig * 1404598ecfbaSChristoph Hellwig * This should never happen except in the case of a VM regression so 1405598ecfbaSChristoph Hellwig * warn about it. 1406598ecfbaSChristoph Hellwig */ 1407598ecfbaSChristoph Hellwig if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == 1408598ecfbaSChristoph Hellwig PF_MEMALLOC)) 1409598ecfbaSChristoph Hellwig goto redirty; 1410598ecfbaSChristoph Hellwig 1411598ecfbaSChristoph Hellwig /* 1412598ecfbaSChristoph Hellwig * Is this page beyond the end of the file? 1413598ecfbaSChristoph Hellwig * 1414598ecfbaSChristoph Hellwig * The page index is less than the end_index, adjust the end_offset 1415598ecfbaSChristoph Hellwig * to the highest offset that this page should represent. 1416598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1417598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1418598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1419598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | | 1420598ecfbaSChristoph Hellwig * ^--------------------------------^----------|-------- 1421598ecfbaSChristoph Hellwig * | desired writeback range | see else | 1422598ecfbaSChristoph Hellwig * ---------------------------------^------------------| 1423598ecfbaSChristoph Hellwig */ 1424598ecfbaSChristoph Hellwig offset = i_size_read(inode); 1425598ecfbaSChristoph Hellwig end_index = offset >> PAGE_SHIFT; 1426598ecfbaSChristoph Hellwig if (page->index < end_index) 1427598ecfbaSChristoph Hellwig end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT; 1428598ecfbaSChristoph Hellwig else { 1429598ecfbaSChristoph Hellwig /* 1430598ecfbaSChristoph Hellwig * Check whether the page to write out is beyond or straddles 1431598ecfbaSChristoph Hellwig * i_size or not. 1432598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1433598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1434598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1435598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | 1436598ecfbaSChristoph Hellwig * ^--------------------------------^-----------|--------- 1437598ecfbaSChristoph Hellwig * | | Straddles | 1438598ecfbaSChristoph Hellwig * ---------------------------------^-----------|--------| 1439598ecfbaSChristoph Hellwig */ 1440598ecfbaSChristoph Hellwig unsigned offset_into_page = offset & (PAGE_SIZE - 1); 1441598ecfbaSChristoph Hellwig 1442598ecfbaSChristoph Hellwig /* 1443f1f264b4SAndreas Gruenbacher * Skip the page if it's fully outside i_size, e.g. due to a 1444f1f264b4SAndreas Gruenbacher * truncate operation that's in progress. We must redirty the 1445598ecfbaSChristoph Hellwig * page so that reclaim stops reclaiming it. Otherwise 1446598ecfbaSChristoph Hellwig * iomap_vm_releasepage() is called on it and gets confused. 1447598ecfbaSChristoph Hellwig * 1448f1f264b4SAndreas Gruenbacher * Note that the end_index is unsigned long. If the given 1449f1f264b4SAndreas Gruenbacher * offset is greater than 16TB on a 32-bit system then if we 1450f1f264b4SAndreas Gruenbacher * checked if the page is fully outside i_size with 1451f1f264b4SAndreas Gruenbacher * "if (page->index >= end_index + 1)", "end_index + 1" would 1452f1f264b4SAndreas Gruenbacher * overflow and evaluate to 0. Hence this page would be 1453f1f264b4SAndreas Gruenbacher * redirtied and written out repeatedly, which would result in 1454f1f264b4SAndreas Gruenbacher * an infinite loop; the user program performing this operation 1455f1f264b4SAndreas Gruenbacher * would hang. Instead, we can detect this situation by 1456f1f264b4SAndreas Gruenbacher * checking if the page is totally beyond i_size or if its 1457598ecfbaSChristoph Hellwig * offset is just equal to the EOF. 1458598ecfbaSChristoph Hellwig */ 1459598ecfbaSChristoph Hellwig if (page->index > end_index || 1460598ecfbaSChristoph Hellwig (page->index == end_index && offset_into_page == 0)) 1461598ecfbaSChristoph Hellwig goto redirty; 1462598ecfbaSChristoph Hellwig 1463598ecfbaSChristoph Hellwig /* 1464598ecfbaSChristoph Hellwig * The page straddles i_size. It must be zeroed out on each 1465598ecfbaSChristoph Hellwig * and every writepage invocation because it may be mmapped. 1466598ecfbaSChristoph Hellwig * "A file is mapped in multiples of the page size. For a file 1467598ecfbaSChristoph Hellwig * that is not a multiple of the page size, the remaining 1468598ecfbaSChristoph Hellwig * memory is zeroed when mapped, and writes to that region are 1469598ecfbaSChristoph Hellwig * not written out to the file." 1470598ecfbaSChristoph Hellwig */ 1471598ecfbaSChristoph Hellwig zero_user_segment(page, offset_into_page, PAGE_SIZE); 1472598ecfbaSChristoph Hellwig 1473598ecfbaSChristoph Hellwig /* Adjust the end_offset to the end of file */ 1474598ecfbaSChristoph Hellwig end_offset = offset; 1475598ecfbaSChristoph Hellwig } 1476598ecfbaSChristoph Hellwig 1477598ecfbaSChristoph Hellwig return iomap_writepage_map(wpc, wbc, inode, page, end_offset); 1478598ecfbaSChristoph Hellwig 1479598ecfbaSChristoph Hellwig redirty: 1480598ecfbaSChristoph Hellwig redirty_page_for_writepage(wbc, page); 1481598ecfbaSChristoph Hellwig unlock_page(page); 1482598ecfbaSChristoph Hellwig return 0; 1483598ecfbaSChristoph Hellwig } 1484598ecfbaSChristoph Hellwig 1485598ecfbaSChristoph Hellwig int 1486598ecfbaSChristoph Hellwig iomap_writepage(struct page *page, struct writeback_control *wbc, 1487598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1488598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1489598ecfbaSChristoph Hellwig { 1490598ecfbaSChristoph Hellwig int ret; 1491598ecfbaSChristoph Hellwig 1492598ecfbaSChristoph Hellwig wpc->ops = ops; 1493598ecfbaSChristoph Hellwig ret = iomap_do_writepage(page, wbc, wpc); 1494598ecfbaSChristoph Hellwig if (!wpc->ioend) 1495598ecfbaSChristoph Hellwig return ret; 1496598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1497598ecfbaSChristoph Hellwig } 1498598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepage); 1499598ecfbaSChristoph Hellwig 1500598ecfbaSChristoph Hellwig int 1501598ecfbaSChristoph Hellwig iomap_writepages(struct address_space *mapping, struct writeback_control *wbc, 1502598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1503598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1504598ecfbaSChristoph Hellwig { 1505598ecfbaSChristoph Hellwig int ret; 1506598ecfbaSChristoph Hellwig 1507598ecfbaSChristoph Hellwig wpc->ops = ops; 1508598ecfbaSChristoph Hellwig ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc); 1509598ecfbaSChristoph Hellwig if (!wpc->ioend) 1510598ecfbaSChristoph Hellwig return ret; 1511598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1512598ecfbaSChristoph Hellwig } 1513598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepages); 1514598ecfbaSChristoph Hellwig 1515598ecfbaSChristoph Hellwig static int __init iomap_init(void) 1516598ecfbaSChristoph Hellwig { 1517598ecfbaSChristoph Hellwig return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE), 1518598ecfbaSChristoph Hellwig offsetof(struct iomap_ioend, io_inline_bio), 1519598ecfbaSChristoph Hellwig BIOSET_NEED_BVECS); 1520598ecfbaSChristoph Hellwig } 1521598ecfbaSChristoph Hellwig fs_initcall(iomap_init); 1522