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 390a195b91SMatthew Wilcox (Oracle) * 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 /* 101afc51aaaSDarrick J. Wong * 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 /* 129afc51aaaSDarrick J. Wong * 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 208afc51aaaSDarrick J. Wong static void 209afc51aaaSDarrick J. Wong iomap_read_inline_data(struct inode *inode, struct page *page, 210afc51aaaSDarrick J. Wong struct iomap *iomap) 211afc51aaaSDarrick J. Wong { 212afc51aaaSDarrick J. Wong size_t size = i_size_read(inode); 213afc51aaaSDarrick J. Wong void *addr; 214afc51aaaSDarrick J. Wong 215afc51aaaSDarrick J. Wong if (PageUptodate(page)) 216afc51aaaSDarrick J. Wong return; 217afc51aaaSDarrick J. Wong 218637d3375SAndreas Gruenbacher BUG_ON(page_has_private(page)); 219afc51aaaSDarrick J. Wong BUG_ON(page->index); 220afc51aaaSDarrick J. Wong BUG_ON(size > PAGE_SIZE - offset_in_page(iomap->inline_data)); 221afc51aaaSDarrick J. Wong 222afc51aaaSDarrick J. Wong addr = kmap_atomic(page); 223afc51aaaSDarrick J. Wong memcpy(addr, iomap->inline_data, size); 224afc51aaaSDarrick J. Wong memset(addr + size, 0, PAGE_SIZE - size); 225afc51aaaSDarrick J. Wong kunmap_atomic(addr); 226afc51aaaSDarrick J. Wong SetPageUptodate(page); 227afc51aaaSDarrick J. Wong } 228afc51aaaSDarrick J. Wong 229009d8d84SChristoph Hellwig static inline bool iomap_block_needs_zeroing(struct inode *inode, 230009d8d84SChristoph Hellwig struct iomap *iomap, loff_t pos) 231009d8d84SChristoph Hellwig { 232009d8d84SChristoph Hellwig return iomap->type != IOMAP_MAPPED || 233009d8d84SChristoph Hellwig (iomap->flags & IOMAP_F_NEW) || 234009d8d84SChristoph Hellwig pos >= i_size_read(inode); 235009d8d84SChristoph Hellwig } 236009d8d84SChristoph Hellwig 237afc51aaaSDarrick J. Wong static loff_t 238afc51aaaSDarrick J. Wong iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 239c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 240afc51aaaSDarrick J. Wong { 241afc51aaaSDarrick J. Wong struct iomap_readpage_ctx *ctx = data; 242afc51aaaSDarrick J. Wong struct page *page = ctx->cur_page; 243637d3375SAndreas Gruenbacher struct iomap_page *iop; 244afc51aaaSDarrick J. Wong loff_t orig_pos = pos; 245afc51aaaSDarrick J. Wong unsigned poff, plen; 246afc51aaaSDarrick J. Wong sector_t sector; 247afc51aaaSDarrick J. Wong 248afc51aaaSDarrick J. Wong if (iomap->type == IOMAP_INLINE) { 249afc51aaaSDarrick J. Wong WARN_ON_ONCE(pos); 250afc51aaaSDarrick J. Wong iomap_read_inline_data(inode, page, iomap); 251afc51aaaSDarrick J. Wong return PAGE_SIZE; 252afc51aaaSDarrick J. Wong } 253afc51aaaSDarrick J. Wong 254afc51aaaSDarrick J. Wong /* zero post-eof blocks as the page may be mapped */ 255637d3375SAndreas Gruenbacher iop = iomap_page_create(inode, page); 256afc51aaaSDarrick J. Wong iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen); 257afc51aaaSDarrick J. Wong if (plen == 0) 258afc51aaaSDarrick J. Wong goto done; 259afc51aaaSDarrick J. Wong 260009d8d84SChristoph Hellwig if (iomap_block_needs_zeroing(inode, iomap, pos)) { 261afc51aaaSDarrick J. Wong zero_user(page, poff, plen); 262afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, poff, plen); 263afc51aaaSDarrick J. Wong goto done; 264afc51aaaSDarrick J. Wong } 265afc51aaaSDarrick J. Wong 266afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = true; 2677d636676SMatthew Wilcox (Oracle) if (iop) 2687d636676SMatthew Wilcox (Oracle) atomic_add(plen, &iop->read_bytes_pending); 269afc51aaaSDarrick J. Wong 270afc51aaaSDarrick J. Wong sector = iomap_sector(iomap, pos); 271d0364f94SChristoph Hellwig if (!ctx->bio || 272d0364f94SChristoph Hellwig bio_end_sector(ctx->bio) != sector || 273d0364f94SChristoph Hellwig bio_add_page(ctx->bio, page, plen, poff) != plen) { 274afc51aaaSDarrick J. Wong gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL); 275457df33eSMatthew Wilcox (Oracle) gfp_t orig_gfp = gfp; 2765f7136dbSMatthew Wilcox (Oracle) unsigned int nr_vecs = DIV_ROUND_UP(length, PAGE_SIZE); 277afc51aaaSDarrick J. Wong 278afc51aaaSDarrick J. Wong if (ctx->bio) 279afc51aaaSDarrick J. Wong submit_bio(ctx->bio); 280afc51aaaSDarrick J. Wong 2819d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) /* same as readahead_gfp_mask */ 282afc51aaaSDarrick J. Wong gfp |= __GFP_NORETRY | __GFP_NOWARN; 2835f7136dbSMatthew Wilcox (Oracle) ctx->bio = bio_alloc(gfp, bio_max_segs(nr_vecs)); 284457df33eSMatthew Wilcox (Oracle) /* 285457df33eSMatthew Wilcox (Oracle) * If the bio_alloc fails, try it again for a single page to 286457df33eSMatthew Wilcox (Oracle) * avoid having to deal with partial page reads. This emulates 287457df33eSMatthew Wilcox (Oracle) * what do_mpage_readpage does. 288457df33eSMatthew Wilcox (Oracle) */ 289457df33eSMatthew Wilcox (Oracle) if (!ctx->bio) 290457df33eSMatthew Wilcox (Oracle) ctx->bio = bio_alloc(orig_gfp, 1); 291afc51aaaSDarrick J. Wong ctx->bio->bi_opf = REQ_OP_READ; 2929d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) 293afc51aaaSDarrick J. Wong ctx->bio->bi_opf |= REQ_RAHEAD; 294afc51aaaSDarrick J. Wong ctx->bio->bi_iter.bi_sector = sector; 295afc51aaaSDarrick J. Wong bio_set_dev(ctx->bio, iomap->bdev); 296afc51aaaSDarrick J. Wong ctx->bio->bi_end_io = iomap_read_end_io; 297d0364f94SChristoph Hellwig __bio_add_page(ctx->bio, page, plen, poff); 298afc51aaaSDarrick J. Wong } 299afc51aaaSDarrick J. Wong done: 300afc51aaaSDarrick J. Wong /* 301afc51aaaSDarrick J. Wong * Move the caller beyond our range so that it keeps making progress. 302afc51aaaSDarrick J. Wong * For that we have to include any leading non-uptodate ranges, but 303afc51aaaSDarrick J. Wong * we can skip trailing ones as they will be handled in the next 304afc51aaaSDarrick J. Wong * iteration. 305afc51aaaSDarrick J. Wong */ 306afc51aaaSDarrick J. Wong return pos - orig_pos + plen; 307afc51aaaSDarrick J. Wong } 308afc51aaaSDarrick J. Wong 309afc51aaaSDarrick J. Wong int 310afc51aaaSDarrick J. Wong iomap_readpage(struct page *page, const struct iomap_ops *ops) 311afc51aaaSDarrick J. Wong { 312afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { .cur_page = page }; 313afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 314afc51aaaSDarrick J. Wong unsigned poff; 315afc51aaaSDarrick J. Wong loff_t ret; 316afc51aaaSDarrick J. Wong 3179e91c572SChristoph Hellwig trace_iomap_readpage(page->mapping->host, 1); 3189e91c572SChristoph Hellwig 319afc51aaaSDarrick J. Wong for (poff = 0; poff < PAGE_SIZE; poff += ret) { 320afc51aaaSDarrick J. Wong ret = iomap_apply(inode, page_offset(page) + poff, 321afc51aaaSDarrick J. Wong PAGE_SIZE - poff, 0, ops, &ctx, 322afc51aaaSDarrick J. Wong iomap_readpage_actor); 323afc51aaaSDarrick J. Wong if (ret <= 0) { 324afc51aaaSDarrick J. Wong WARN_ON_ONCE(ret == 0); 325afc51aaaSDarrick J. Wong SetPageError(page); 326afc51aaaSDarrick J. Wong break; 327afc51aaaSDarrick J. Wong } 328afc51aaaSDarrick J. Wong } 329afc51aaaSDarrick J. Wong 330afc51aaaSDarrick J. Wong if (ctx.bio) { 331afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 332afc51aaaSDarrick J. Wong WARN_ON_ONCE(!ctx.cur_page_in_bio); 333afc51aaaSDarrick J. Wong } else { 334afc51aaaSDarrick J. Wong WARN_ON_ONCE(ctx.cur_page_in_bio); 335afc51aaaSDarrick J. Wong unlock_page(page); 336afc51aaaSDarrick J. Wong } 337afc51aaaSDarrick J. Wong 338afc51aaaSDarrick J. Wong /* 339d4388340SMatthew Wilcox (Oracle) * Just like mpage_readahead and block_read_full_page we always 340afc51aaaSDarrick J. Wong * return 0 and just mark the page as PageError on errors. This 341afc51aaaSDarrick J. Wong * should be cleaned up all through the stack eventually. 342afc51aaaSDarrick J. Wong */ 343afc51aaaSDarrick J. Wong return 0; 344afc51aaaSDarrick J. Wong } 345afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_readpage); 346afc51aaaSDarrick J. Wong 347afc51aaaSDarrick J. Wong static loff_t 3489d24a13aSMatthew Wilcox (Oracle) iomap_readahead_actor(struct inode *inode, loff_t pos, loff_t length, 349c039b997SGoldwyn Rodrigues void *data, struct iomap *iomap, struct iomap *srcmap) 350afc51aaaSDarrick J. Wong { 351afc51aaaSDarrick J. Wong struct iomap_readpage_ctx *ctx = data; 352afc51aaaSDarrick J. Wong loff_t done, ret; 353afc51aaaSDarrick J. Wong 354afc51aaaSDarrick J. Wong for (done = 0; done < length; done += ret) { 355afc51aaaSDarrick J. Wong if (ctx->cur_page && offset_in_page(pos + done) == 0) { 356afc51aaaSDarrick J. Wong if (!ctx->cur_page_in_bio) 357afc51aaaSDarrick J. Wong unlock_page(ctx->cur_page); 358afc51aaaSDarrick J. Wong put_page(ctx->cur_page); 359afc51aaaSDarrick J. Wong ctx->cur_page = NULL; 360afc51aaaSDarrick J. Wong } 361afc51aaaSDarrick J. Wong if (!ctx->cur_page) { 3629d24a13aSMatthew Wilcox (Oracle) ctx->cur_page = readahead_page(ctx->rac); 363afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = false; 364afc51aaaSDarrick J. Wong } 365afc51aaaSDarrick J. Wong ret = iomap_readpage_actor(inode, pos + done, length - done, 366c039b997SGoldwyn Rodrigues ctx, iomap, srcmap); 367afc51aaaSDarrick J. Wong } 368afc51aaaSDarrick J. Wong 369afc51aaaSDarrick J. Wong return done; 370afc51aaaSDarrick J. Wong } 371afc51aaaSDarrick J. Wong 3729d24a13aSMatthew Wilcox (Oracle) /** 3739d24a13aSMatthew Wilcox (Oracle) * iomap_readahead - Attempt to read pages from a file. 3749d24a13aSMatthew Wilcox (Oracle) * @rac: Describes the pages to be read. 3759d24a13aSMatthew Wilcox (Oracle) * @ops: The operations vector for the filesystem. 3769d24a13aSMatthew Wilcox (Oracle) * 3779d24a13aSMatthew Wilcox (Oracle) * This function is for filesystems to call to implement their readahead 3789d24a13aSMatthew Wilcox (Oracle) * address_space operation. 3799d24a13aSMatthew Wilcox (Oracle) * 3809d24a13aSMatthew Wilcox (Oracle) * Context: The @ops callbacks may submit I/O (eg to read the addresses of 3819d24a13aSMatthew Wilcox (Oracle) * blocks from disc), and may wait for it. The caller may be trying to 3829d24a13aSMatthew Wilcox (Oracle) * access a different page, and so sleeping excessively should be avoided. 3839d24a13aSMatthew Wilcox (Oracle) * It may allocate memory, but should avoid costly allocations. This 3849d24a13aSMatthew Wilcox (Oracle) * function is called with memalloc_nofs set, so allocations will not cause 3859d24a13aSMatthew Wilcox (Oracle) * the filesystem to be reentered. 3869d24a13aSMatthew Wilcox (Oracle) */ 3879d24a13aSMatthew Wilcox (Oracle) void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops) 388afc51aaaSDarrick J. Wong { 3899d24a13aSMatthew Wilcox (Oracle) struct inode *inode = rac->mapping->host; 3909d24a13aSMatthew Wilcox (Oracle) loff_t pos = readahead_pos(rac); 391076171a6SMatthew Wilcox (Oracle) size_t length = readahead_length(rac); 392afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { 3939d24a13aSMatthew Wilcox (Oracle) .rac = rac, 394afc51aaaSDarrick J. Wong }; 395afc51aaaSDarrick J. Wong 3969d24a13aSMatthew Wilcox (Oracle) trace_iomap_readahead(inode, readahead_count(rac)); 3979e91c572SChristoph Hellwig 398afc51aaaSDarrick J. Wong while (length > 0) { 399076171a6SMatthew Wilcox (Oracle) ssize_t ret = iomap_apply(inode, pos, length, 0, ops, 4009d24a13aSMatthew Wilcox (Oracle) &ctx, iomap_readahead_actor); 401afc51aaaSDarrick J. Wong if (ret <= 0) { 402afc51aaaSDarrick J. Wong WARN_ON_ONCE(ret == 0); 4039d24a13aSMatthew Wilcox (Oracle) break; 404afc51aaaSDarrick J. Wong } 405afc51aaaSDarrick J. Wong pos += ret; 406afc51aaaSDarrick J. Wong length -= ret; 407afc51aaaSDarrick J. Wong } 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 462afc51aaaSDarrick J. Wong * ->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 /* 477afc51aaaSDarrick J. Wong * If we are 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 51132a38a49SChristoph Hellwig enum { 51232a38a49SChristoph Hellwig IOMAP_WRITE_F_UNSHARE = (1 << 0), 51332a38a49SChristoph Hellwig }; 51432a38a49SChristoph Hellwig 515afc51aaaSDarrick J. Wong static void 516afc51aaaSDarrick J. Wong iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) 517afc51aaaSDarrick J. Wong { 518afc51aaaSDarrick J. Wong loff_t i_size = i_size_read(inode); 519afc51aaaSDarrick J. Wong 520afc51aaaSDarrick J. Wong /* 521afc51aaaSDarrick J. Wong * Only truncate newly allocated pages beyoned EOF, even if the 522afc51aaaSDarrick J. Wong * write started inside the existing inode size. 523afc51aaaSDarrick J. Wong */ 524afc51aaaSDarrick J. Wong if (pos + len > i_size) 525afc51aaaSDarrick J. Wong truncate_pagecache_range(inode, max(pos, i_size), pos + len); 526afc51aaaSDarrick J. Wong } 527afc51aaaSDarrick J. Wong 528afc51aaaSDarrick J. Wong static int 529d3b40439SChristoph Hellwig iomap_read_page_sync(loff_t block_start, struct page *page, unsigned poff, 530d3b40439SChristoph Hellwig unsigned plen, struct iomap *iomap) 531afc51aaaSDarrick J. Wong { 532afc51aaaSDarrick J. Wong struct bio_vec bvec; 533afc51aaaSDarrick J. Wong struct bio bio; 534afc51aaaSDarrick J. Wong 535afc51aaaSDarrick J. Wong bio_init(&bio, &bvec, 1); 536afc51aaaSDarrick J. Wong bio.bi_opf = REQ_OP_READ; 537afc51aaaSDarrick J. Wong bio.bi_iter.bi_sector = iomap_sector(iomap, block_start); 538afc51aaaSDarrick J. Wong bio_set_dev(&bio, iomap->bdev); 539afc51aaaSDarrick J. Wong __bio_add_page(&bio, page, plen, poff); 540afc51aaaSDarrick J. Wong return submit_bio_wait(&bio); 541afc51aaaSDarrick J. Wong } 542afc51aaaSDarrick J. Wong 543afc51aaaSDarrick J. Wong static int 54432a38a49SChristoph Hellwig __iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, int flags, 545c039b997SGoldwyn Rodrigues struct page *page, struct iomap *srcmap) 546afc51aaaSDarrick J. Wong { 547afc51aaaSDarrick J. Wong struct iomap_page *iop = iomap_page_create(inode, page); 548afc51aaaSDarrick J. Wong loff_t block_size = i_blocksize(inode); 5496cc19c5fSNikolay Borisov loff_t block_start = round_down(pos, block_size); 5506cc19c5fSNikolay Borisov loff_t block_end = round_up(pos + len, block_size); 551afc51aaaSDarrick J. Wong unsigned from = offset_in_page(pos), to = from + len, poff, plen; 552afc51aaaSDarrick J. Wong 553afc51aaaSDarrick J. Wong if (PageUptodate(page)) 554afc51aaaSDarrick J. Wong return 0; 555e6e7ca92SMatthew Wilcox (Oracle) ClearPageError(page); 556afc51aaaSDarrick J. Wong 557afc51aaaSDarrick J. Wong do { 558afc51aaaSDarrick J. Wong iomap_adjust_read_range(inode, iop, &block_start, 559afc51aaaSDarrick J. Wong block_end - block_start, &poff, &plen); 560afc51aaaSDarrick J. Wong if (plen == 0) 561afc51aaaSDarrick J. Wong break; 562afc51aaaSDarrick J. Wong 56332a38a49SChristoph Hellwig if (!(flags & IOMAP_WRITE_F_UNSHARE) && 56432a38a49SChristoph Hellwig (from <= poff || from >= poff + plen) && 565d3b40439SChristoph Hellwig (to <= poff || to >= poff + plen)) 566d3b40439SChristoph Hellwig continue; 567d3b40439SChristoph Hellwig 568c039b997SGoldwyn Rodrigues if (iomap_block_needs_zeroing(inode, srcmap, block_start)) { 56932a38a49SChristoph Hellwig if (WARN_ON_ONCE(flags & IOMAP_WRITE_F_UNSHARE)) 57032a38a49SChristoph Hellwig return -EIO; 571d3b40439SChristoph Hellwig zero_user_segments(page, poff, from, to, poff + plen); 57214284fedSMatthew Wilcox (Oracle) } else { 57314284fedSMatthew Wilcox (Oracle) int status = iomap_read_page_sync(block_start, page, 57414284fedSMatthew Wilcox (Oracle) poff, plen, srcmap); 575d3b40439SChristoph Hellwig if (status) 576d3b40439SChristoph Hellwig return status; 57714284fedSMatthew Wilcox (Oracle) } 57814284fedSMatthew Wilcox (Oracle) iomap_set_range_uptodate(page, poff, plen); 579afc51aaaSDarrick J. Wong } while ((block_start += plen) < block_end); 580afc51aaaSDarrick J. Wong 581d3b40439SChristoph Hellwig return 0; 582afc51aaaSDarrick J. Wong } 583afc51aaaSDarrick J. Wong 584afc51aaaSDarrick J. Wong static int 585afc51aaaSDarrick J. Wong iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags, 586c039b997SGoldwyn Rodrigues struct page **pagep, struct iomap *iomap, struct iomap *srcmap) 587afc51aaaSDarrick J. Wong { 588afc51aaaSDarrick J. Wong const struct iomap_page_ops *page_ops = iomap->page_ops; 589afc51aaaSDarrick J. Wong struct page *page; 590afc51aaaSDarrick J. Wong int status = 0; 591afc51aaaSDarrick J. Wong 592afc51aaaSDarrick J. Wong BUG_ON(pos + len > iomap->offset + iomap->length); 593c039b997SGoldwyn Rodrigues if (srcmap != iomap) 594c039b997SGoldwyn Rodrigues BUG_ON(pos + len > srcmap->offset + srcmap->length); 595afc51aaaSDarrick J. Wong 596afc51aaaSDarrick J. Wong if (fatal_signal_pending(current)) 597afc51aaaSDarrick J. Wong return -EINTR; 598afc51aaaSDarrick J. Wong 599afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_prepare) { 600afc51aaaSDarrick J. Wong status = page_ops->page_prepare(inode, pos, len, iomap); 601afc51aaaSDarrick J. Wong if (status) 602afc51aaaSDarrick J. Wong return status; 603afc51aaaSDarrick J. Wong } 604afc51aaaSDarrick J. Wong 605dcd6158dSChristoph Hellwig page = grab_cache_page_write_begin(inode->i_mapping, pos >> PAGE_SHIFT, 606dcd6158dSChristoph Hellwig AOP_FLAG_NOFS); 607afc51aaaSDarrick J. Wong if (!page) { 608afc51aaaSDarrick J. Wong status = -ENOMEM; 609afc51aaaSDarrick J. Wong goto out_no_page; 610afc51aaaSDarrick J. Wong } 611afc51aaaSDarrick J. Wong 612c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) 613c039b997SGoldwyn Rodrigues iomap_read_inline_data(inode, page, srcmap); 614afc51aaaSDarrick J. Wong else if (iomap->flags & IOMAP_F_BUFFER_HEAD) 615c039b997SGoldwyn Rodrigues status = __block_write_begin_int(page, pos, len, NULL, srcmap); 616afc51aaaSDarrick J. Wong else 61732a38a49SChristoph Hellwig status = __iomap_write_begin(inode, pos, len, flags, page, 618c039b997SGoldwyn Rodrigues srcmap); 619afc51aaaSDarrick J. Wong 620afc51aaaSDarrick J. Wong if (unlikely(status)) 621afc51aaaSDarrick J. Wong goto out_unlock; 622afc51aaaSDarrick J. Wong 623afc51aaaSDarrick J. Wong *pagep = page; 624afc51aaaSDarrick J. Wong return 0; 625afc51aaaSDarrick J. Wong 626afc51aaaSDarrick J. Wong out_unlock: 627afc51aaaSDarrick J. Wong unlock_page(page); 628afc51aaaSDarrick J. Wong put_page(page); 629afc51aaaSDarrick J. Wong iomap_write_failed(inode, pos, len); 630afc51aaaSDarrick J. Wong 631afc51aaaSDarrick J. Wong out_no_page: 632afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 633afc51aaaSDarrick J. Wong page_ops->page_done(inode, pos, 0, NULL, iomap); 634afc51aaaSDarrick J. Wong return status; 635afc51aaaSDarrick J. Wong } 636afc51aaaSDarrick J. Wong 637e25ba8cbSMatthew Wilcox (Oracle) static size_t __iomap_write_end(struct inode *inode, loff_t pos, size_t len, 638e25ba8cbSMatthew Wilcox (Oracle) size_t copied, struct page *page) 639afc51aaaSDarrick J. Wong { 640afc51aaaSDarrick J. Wong flush_dcache_page(page); 641afc51aaaSDarrick J. Wong 642afc51aaaSDarrick J. Wong /* 643afc51aaaSDarrick J. Wong * The blocks that were entirely written will now be uptodate, so we 644afc51aaaSDarrick J. Wong * don't have to worry about a readpage reading them and overwriting a 645afc51aaaSDarrick J. Wong * partial write. However if we have encountered a short write and only 646afc51aaaSDarrick J. Wong * partially written into a block, it will not be marked uptodate, so a 647afc51aaaSDarrick J. Wong * readpage might come in and destroy our partial write. 648afc51aaaSDarrick J. Wong * 649afc51aaaSDarrick J. Wong * Do the simplest thing, and just treat any short write to a non 650afc51aaaSDarrick J. Wong * uptodate page as a zero-length write, and force the caller to redo 651afc51aaaSDarrick J. Wong * the whole thing. 652afc51aaaSDarrick J. Wong */ 653afc51aaaSDarrick J. Wong if (unlikely(copied < len && !PageUptodate(page))) 654afc51aaaSDarrick J. Wong return 0; 655afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, offset_in_page(pos), len); 656fd7353f8SMatthew Wilcox (Oracle) __set_page_dirty_nobuffers(page); 657afc51aaaSDarrick J. Wong return copied; 658afc51aaaSDarrick J. Wong } 659afc51aaaSDarrick J. Wong 660e25ba8cbSMatthew Wilcox (Oracle) static size_t iomap_write_end_inline(struct inode *inode, struct page *page, 661e25ba8cbSMatthew Wilcox (Oracle) struct iomap *iomap, loff_t pos, size_t copied) 662afc51aaaSDarrick J. Wong { 663afc51aaaSDarrick J. Wong void *addr; 664afc51aaaSDarrick J. Wong 665afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 666afc51aaaSDarrick J. Wong BUG_ON(pos + copied > PAGE_SIZE - offset_in_page(iomap->inline_data)); 667afc51aaaSDarrick J. Wong 6687ed3cd1aSMatthew Wilcox (Oracle) flush_dcache_page(page); 669afc51aaaSDarrick J. Wong addr = kmap_atomic(page); 670afc51aaaSDarrick J. Wong memcpy(iomap->inline_data + pos, addr + pos, copied); 671afc51aaaSDarrick J. Wong kunmap_atomic(addr); 672afc51aaaSDarrick J. Wong 673afc51aaaSDarrick J. Wong mark_inode_dirty(inode); 674afc51aaaSDarrick J. Wong return copied; 675afc51aaaSDarrick J. Wong } 676afc51aaaSDarrick J. Wong 677e25ba8cbSMatthew Wilcox (Oracle) /* Returns the number of bytes copied. May be 0. Cannot be an errno. */ 678e25ba8cbSMatthew Wilcox (Oracle) static size_t iomap_write_end(struct inode *inode, loff_t pos, size_t len, 679e25ba8cbSMatthew Wilcox (Oracle) size_t copied, struct page *page, struct iomap *iomap, 680e25ba8cbSMatthew Wilcox (Oracle) struct iomap *srcmap) 681afc51aaaSDarrick J. Wong { 682afc51aaaSDarrick J. Wong const struct iomap_page_ops *page_ops = iomap->page_ops; 683afc51aaaSDarrick J. Wong loff_t old_size = inode->i_size; 684e25ba8cbSMatthew Wilcox (Oracle) size_t ret; 685afc51aaaSDarrick J. Wong 686c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) { 687afc51aaaSDarrick J. Wong ret = iomap_write_end_inline(inode, page, iomap, pos, copied); 688c039b997SGoldwyn Rodrigues } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) { 689afc51aaaSDarrick J. Wong ret = block_write_end(NULL, inode->i_mapping, pos, len, copied, 690afc51aaaSDarrick J. Wong page, NULL); 691afc51aaaSDarrick J. Wong } else { 692c12d6fa8SChristoph Hellwig ret = __iomap_write_end(inode, pos, len, copied, page); 693afc51aaaSDarrick J. Wong } 694afc51aaaSDarrick J. Wong 695afc51aaaSDarrick J. Wong /* 696afc51aaaSDarrick J. Wong * Update the in-memory inode size after copying the data into the page 697afc51aaaSDarrick J. Wong * cache. It's up to the file system to write the updated size to disk, 698afc51aaaSDarrick J. Wong * preferably after I/O completion so that no stale data is exposed. 699afc51aaaSDarrick J. Wong */ 700afc51aaaSDarrick J. Wong if (pos + ret > old_size) { 701afc51aaaSDarrick J. Wong i_size_write(inode, pos + ret); 702afc51aaaSDarrick J. Wong iomap->flags |= IOMAP_F_SIZE_CHANGED; 703afc51aaaSDarrick J. Wong } 704afc51aaaSDarrick J. Wong unlock_page(page); 705afc51aaaSDarrick J. Wong 706afc51aaaSDarrick J. Wong if (old_size < pos) 707afc51aaaSDarrick J. Wong pagecache_isize_extended(inode, old_size, pos); 708afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 709afc51aaaSDarrick J. Wong page_ops->page_done(inode, pos, ret, page, iomap); 710afc51aaaSDarrick J. Wong put_page(page); 711afc51aaaSDarrick J. Wong 712afc51aaaSDarrick J. Wong if (ret < len) 713afc51aaaSDarrick J. Wong iomap_write_failed(inode, pos, len); 714afc51aaaSDarrick J. Wong return ret; 715afc51aaaSDarrick J. Wong } 716afc51aaaSDarrick J. Wong 717afc51aaaSDarrick J. Wong static loff_t 718afc51aaaSDarrick J. Wong iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 719c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 720afc51aaaSDarrick J. Wong { 721afc51aaaSDarrick J. Wong struct iov_iter *i = data; 722afc51aaaSDarrick J. Wong long status = 0; 723afc51aaaSDarrick J. Wong ssize_t written = 0; 724afc51aaaSDarrick J. Wong 725afc51aaaSDarrick J. Wong do { 726afc51aaaSDarrick J. Wong struct page *page; 727afc51aaaSDarrick J. Wong unsigned long offset; /* Offset into pagecache page */ 728afc51aaaSDarrick J. Wong unsigned long bytes; /* Bytes to write to page */ 729afc51aaaSDarrick J. Wong size_t copied; /* Bytes copied from user */ 730afc51aaaSDarrick J. Wong 731afc51aaaSDarrick J. Wong offset = offset_in_page(pos); 732afc51aaaSDarrick J. Wong bytes = min_t(unsigned long, PAGE_SIZE - offset, 733afc51aaaSDarrick J. Wong iov_iter_count(i)); 734afc51aaaSDarrick J. Wong again: 735afc51aaaSDarrick J. Wong if (bytes > length) 736afc51aaaSDarrick J. Wong bytes = length; 737afc51aaaSDarrick J. Wong 738afc51aaaSDarrick J. Wong /* 739afc51aaaSDarrick J. Wong * Bring in the user page that we will copy from _first_. 740afc51aaaSDarrick J. Wong * Otherwise there's a nasty deadlock on copying from the 741afc51aaaSDarrick J. Wong * same page as we're writing to, without it being marked 742afc51aaaSDarrick J. Wong * up-to-date. 743afc51aaaSDarrick J. Wong */ 744afc51aaaSDarrick J. Wong if (unlikely(iov_iter_fault_in_readable(i, bytes))) { 745afc51aaaSDarrick J. Wong status = -EFAULT; 746afc51aaaSDarrick J. Wong break; 747afc51aaaSDarrick J. Wong } 748afc51aaaSDarrick J. Wong 749c039b997SGoldwyn Rodrigues status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, 750c039b997SGoldwyn Rodrigues srcmap); 751afc51aaaSDarrick J. Wong if (unlikely(status)) 752afc51aaaSDarrick J. Wong break; 753afc51aaaSDarrick J. Wong 754afc51aaaSDarrick J. Wong if (mapping_writably_mapped(inode->i_mapping)) 755afc51aaaSDarrick J. Wong flush_dcache_page(page); 756afc51aaaSDarrick J. Wong 757f0b65f39SAl Viro copied = copy_page_from_iter_atomic(page, offset, bytes, i); 758afc51aaaSDarrick J. Wong 759bc1bb416SAl Viro status = iomap_write_end(inode, pos, bytes, copied, page, iomap, 760c039b997SGoldwyn Rodrigues srcmap); 761afc51aaaSDarrick J. Wong 762f0b65f39SAl Viro if (unlikely(copied != status)) 763f0b65f39SAl Viro iov_iter_revert(i, copied - status); 764afc51aaaSDarrick J. Wong 765f0b65f39SAl Viro cond_resched(); 766bc1bb416SAl Viro if (unlikely(status == 0)) { 767afc51aaaSDarrick J. Wong /* 768bc1bb416SAl Viro * A short copy made iomap_write_end() reject the 769bc1bb416SAl Viro * thing entirely. Might be memory poisoning 770bc1bb416SAl Viro * halfway through, might be a race with munmap, 771bc1bb416SAl Viro * might be severe memory pressure. 772afc51aaaSDarrick J. Wong */ 773bc1bb416SAl Viro if (copied) 774bc1bb416SAl Viro bytes = copied; 775afc51aaaSDarrick J. Wong goto again; 776afc51aaaSDarrick J. Wong } 777f0b65f39SAl Viro pos += status; 778f0b65f39SAl Viro written += status; 779f0b65f39SAl Viro length -= status; 780afc51aaaSDarrick J. Wong 781afc51aaaSDarrick J. Wong balance_dirty_pages_ratelimited(inode->i_mapping); 782afc51aaaSDarrick J. Wong } while (iov_iter_count(i) && length); 783afc51aaaSDarrick J. Wong 784afc51aaaSDarrick J. Wong return written ? written : status; 785afc51aaaSDarrick J. Wong } 786afc51aaaSDarrick J. Wong 787afc51aaaSDarrick J. Wong ssize_t 788afc51aaaSDarrick J. Wong iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter, 789afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 790afc51aaaSDarrick J. Wong { 791afc51aaaSDarrick J. Wong struct inode *inode = iocb->ki_filp->f_mapping->host; 792afc51aaaSDarrick J. Wong loff_t pos = iocb->ki_pos, ret = 0, written = 0; 793afc51aaaSDarrick J. Wong 794afc51aaaSDarrick J. Wong while (iov_iter_count(iter)) { 795afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, iov_iter_count(iter), 796afc51aaaSDarrick J. Wong IOMAP_WRITE, ops, iter, iomap_write_actor); 797afc51aaaSDarrick J. Wong if (ret <= 0) 798afc51aaaSDarrick J. Wong break; 799afc51aaaSDarrick J. Wong pos += ret; 800afc51aaaSDarrick J. Wong written += ret; 801afc51aaaSDarrick J. Wong } 802afc51aaaSDarrick J. Wong 803afc51aaaSDarrick J. Wong return written ? written : ret; 804afc51aaaSDarrick J. Wong } 805afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_file_buffered_write); 806afc51aaaSDarrick J. Wong 807afc51aaaSDarrick J. Wong static loff_t 8083590c4d8SChristoph Hellwig iomap_unshare_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 809c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 810afc51aaaSDarrick J. Wong { 811afc51aaaSDarrick J. Wong long status = 0; 812d4ff3b2eSMatthew Wilcox (Oracle) loff_t written = 0; 813afc51aaaSDarrick J. Wong 8143590c4d8SChristoph Hellwig /* don't bother with blocks that are not shared to start with */ 8153590c4d8SChristoph Hellwig if (!(iomap->flags & IOMAP_F_SHARED)) 8163590c4d8SChristoph Hellwig return length; 8173590c4d8SChristoph Hellwig /* don't bother with holes or unwritten extents */ 818c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 8193590c4d8SChristoph Hellwig return length; 8203590c4d8SChristoph Hellwig 821afc51aaaSDarrick J. Wong do { 82232a38a49SChristoph Hellwig unsigned long offset = offset_in_page(pos); 82332a38a49SChristoph Hellwig unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length); 82432a38a49SChristoph Hellwig struct page *page; 825afc51aaaSDarrick J. Wong 82632a38a49SChristoph Hellwig status = iomap_write_begin(inode, pos, bytes, 827c039b997SGoldwyn Rodrigues IOMAP_WRITE_F_UNSHARE, &page, iomap, srcmap); 828afc51aaaSDarrick J. Wong if (unlikely(status)) 829afc51aaaSDarrick J. Wong return status; 830afc51aaaSDarrick J. Wong 831c039b997SGoldwyn Rodrigues status = iomap_write_end(inode, pos, bytes, bytes, page, iomap, 832c039b997SGoldwyn Rodrigues srcmap); 833afc51aaaSDarrick J. Wong if (WARN_ON_ONCE(status == 0)) 834afc51aaaSDarrick J. Wong return -EIO; 835afc51aaaSDarrick J. Wong 836afc51aaaSDarrick J. Wong cond_resched(); 837afc51aaaSDarrick J. Wong 838afc51aaaSDarrick J. Wong pos += status; 839afc51aaaSDarrick J. Wong written += status; 840afc51aaaSDarrick J. Wong length -= status; 841afc51aaaSDarrick J. Wong 842afc51aaaSDarrick J. Wong balance_dirty_pages_ratelimited(inode->i_mapping); 843afc51aaaSDarrick J. Wong } while (length); 844afc51aaaSDarrick J. Wong 845afc51aaaSDarrick J. Wong return written; 846afc51aaaSDarrick J. Wong } 847afc51aaaSDarrick J. Wong 848afc51aaaSDarrick J. Wong int 8493590c4d8SChristoph Hellwig iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, 850afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 851afc51aaaSDarrick J. Wong { 852afc51aaaSDarrick J. Wong loff_t ret; 853afc51aaaSDarrick J. Wong 854afc51aaaSDarrick J. Wong while (len) { 855afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL, 8563590c4d8SChristoph Hellwig iomap_unshare_actor); 857afc51aaaSDarrick J. Wong if (ret <= 0) 858afc51aaaSDarrick J. Wong return ret; 859afc51aaaSDarrick J. Wong pos += ret; 860afc51aaaSDarrick J. Wong len -= ret; 861afc51aaaSDarrick J. Wong } 862afc51aaaSDarrick J. Wong 863afc51aaaSDarrick J. Wong return 0; 864afc51aaaSDarrick J. Wong } 8653590c4d8SChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_file_unshare); 866afc51aaaSDarrick J. Wong 86781ee8e52SMatthew Wilcox (Oracle) static s64 iomap_zero(struct inode *inode, loff_t pos, u64 length, 86881ee8e52SMatthew Wilcox (Oracle) struct iomap *iomap, struct iomap *srcmap) 869afc51aaaSDarrick J. Wong { 870afc51aaaSDarrick J. Wong struct page *page; 871afc51aaaSDarrick J. Wong int status; 87281ee8e52SMatthew Wilcox (Oracle) unsigned offset = offset_in_page(pos); 87381ee8e52SMatthew Wilcox (Oracle) unsigned bytes = min_t(u64, PAGE_SIZE - offset, length); 874afc51aaaSDarrick J. Wong 875c039b997SGoldwyn Rodrigues status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, srcmap); 876afc51aaaSDarrick J. Wong if (status) 877afc51aaaSDarrick J. Wong return status; 878afc51aaaSDarrick J. Wong 879afc51aaaSDarrick J. Wong zero_user(page, offset, bytes); 880afc51aaaSDarrick J. Wong mark_page_accessed(page); 881afc51aaaSDarrick J. Wong 882c039b997SGoldwyn Rodrigues return iomap_write_end(inode, pos, bytes, bytes, page, iomap, srcmap); 883afc51aaaSDarrick J. Wong } 884afc51aaaSDarrick J. Wong 88581ee8e52SMatthew Wilcox (Oracle) static loff_t iomap_zero_range_actor(struct inode *inode, loff_t pos, 88681ee8e52SMatthew Wilcox (Oracle) loff_t length, void *data, struct iomap *iomap, 88781ee8e52SMatthew Wilcox (Oracle) struct iomap *srcmap) 888afc51aaaSDarrick J. Wong { 889afc51aaaSDarrick J. Wong bool *did_zero = data; 890afc51aaaSDarrick J. Wong loff_t written = 0; 891afc51aaaSDarrick J. Wong 892afc51aaaSDarrick J. Wong /* already zeroed? we're done. */ 893c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 89481ee8e52SMatthew Wilcox (Oracle) return length; 895afc51aaaSDarrick J. Wong 896afc51aaaSDarrick J. Wong do { 89781ee8e52SMatthew Wilcox (Oracle) s64 bytes; 898afc51aaaSDarrick J. Wong 899afc51aaaSDarrick J. Wong if (IS_DAX(inode)) 90081ee8e52SMatthew Wilcox (Oracle) bytes = dax_iomap_zero(pos, length, iomap); 901afc51aaaSDarrick J. Wong else 90281ee8e52SMatthew Wilcox (Oracle) bytes = iomap_zero(inode, pos, length, iomap, srcmap); 90381ee8e52SMatthew Wilcox (Oracle) if (bytes < 0) 90481ee8e52SMatthew Wilcox (Oracle) return bytes; 905afc51aaaSDarrick J. Wong 906afc51aaaSDarrick J. Wong pos += bytes; 90781ee8e52SMatthew Wilcox (Oracle) length -= bytes; 908afc51aaaSDarrick J. Wong written += bytes; 909afc51aaaSDarrick J. Wong if (did_zero) 910afc51aaaSDarrick J. Wong *did_zero = true; 91181ee8e52SMatthew Wilcox (Oracle) } while (length > 0); 912afc51aaaSDarrick J. Wong 913afc51aaaSDarrick J. Wong return written; 914afc51aaaSDarrick J. Wong } 915afc51aaaSDarrick J. Wong 916afc51aaaSDarrick J. Wong int 917afc51aaaSDarrick J. Wong iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, 918afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 919afc51aaaSDarrick J. Wong { 920afc51aaaSDarrick J. Wong loff_t ret; 921afc51aaaSDarrick J. Wong 922afc51aaaSDarrick J. Wong while (len > 0) { 923afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, len, IOMAP_ZERO, 924afc51aaaSDarrick J. Wong ops, did_zero, iomap_zero_range_actor); 925afc51aaaSDarrick J. Wong if (ret <= 0) 926afc51aaaSDarrick J. Wong return ret; 927afc51aaaSDarrick J. Wong 928afc51aaaSDarrick J. Wong pos += ret; 929afc51aaaSDarrick J. Wong len -= ret; 930afc51aaaSDarrick J. Wong } 931afc51aaaSDarrick J. Wong 932afc51aaaSDarrick J. Wong return 0; 933afc51aaaSDarrick J. Wong } 934afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_zero_range); 935afc51aaaSDarrick J. Wong 936afc51aaaSDarrick J. Wong int 937afc51aaaSDarrick J. Wong iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, 938afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 939afc51aaaSDarrick J. Wong { 940afc51aaaSDarrick J. Wong unsigned int blocksize = i_blocksize(inode); 941afc51aaaSDarrick J. Wong unsigned int off = pos & (blocksize - 1); 942afc51aaaSDarrick J. Wong 943afc51aaaSDarrick J. Wong /* Block boundary? Nothing to do */ 944afc51aaaSDarrick J. Wong if (!off) 945afc51aaaSDarrick J. Wong return 0; 946afc51aaaSDarrick J. Wong return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); 947afc51aaaSDarrick J. Wong } 948afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_truncate_page); 949afc51aaaSDarrick J. Wong 950afc51aaaSDarrick J. Wong static loff_t 951afc51aaaSDarrick J. Wong iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length, 952c039b997SGoldwyn Rodrigues void *data, struct iomap *iomap, struct iomap *srcmap) 953afc51aaaSDarrick J. Wong { 954afc51aaaSDarrick J. Wong struct page *page = data; 955afc51aaaSDarrick J. Wong int ret; 956afc51aaaSDarrick J. Wong 957afc51aaaSDarrick J. Wong if (iomap->flags & IOMAP_F_BUFFER_HEAD) { 958afc51aaaSDarrick J. Wong ret = __block_write_begin_int(page, pos, length, NULL, iomap); 959afc51aaaSDarrick J. Wong if (ret) 960afc51aaaSDarrick J. Wong return ret; 961afc51aaaSDarrick J. Wong block_commit_write(page, 0, length); 962afc51aaaSDarrick J. Wong } else { 963afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 964afc51aaaSDarrick J. Wong set_page_dirty(page); 965afc51aaaSDarrick J. Wong } 966afc51aaaSDarrick J. Wong 967afc51aaaSDarrick J. Wong return length; 968afc51aaaSDarrick J. Wong } 969afc51aaaSDarrick J. Wong 970afc51aaaSDarrick J. Wong vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) 971afc51aaaSDarrick J. Wong { 972afc51aaaSDarrick J. Wong struct page *page = vmf->page; 973afc51aaaSDarrick J. Wong struct inode *inode = file_inode(vmf->vma->vm_file); 974afc51aaaSDarrick J. Wong unsigned long length; 975243145bcSAndreas Gruenbacher loff_t offset; 976afc51aaaSDarrick J. Wong ssize_t ret; 977afc51aaaSDarrick J. Wong 978afc51aaaSDarrick J. Wong lock_page(page); 979243145bcSAndreas Gruenbacher ret = page_mkwrite_check_truncate(page, inode); 980243145bcSAndreas Gruenbacher if (ret < 0) 981afc51aaaSDarrick J. Wong goto out_unlock; 982243145bcSAndreas Gruenbacher length = ret; 983afc51aaaSDarrick J. Wong 984243145bcSAndreas Gruenbacher offset = page_offset(page); 985afc51aaaSDarrick J. Wong while (length > 0) { 986afc51aaaSDarrick J. Wong ret = iomap_apply(inode, offset, length, 987afc51aaaSDarrick J. Wong IOMAP_WRITE | IOMAP_FAULT, ops, page, 988afc51aaaSDarrick J. Wong iomap_page_mkwrite_actor); 989afc51aaaSDarrick J. Wong if (unlikely(ret <= 0)) 990afc51aaaSDarrick J. Wong goto out_unlock; 991afc51aaaSDarrick J. Wong offset += ret; 992afc51aaaSDarrick J. Wong length -= ret; 993afc51aaaSDarrick J. Wong } 994afc51aaaSDarrick J. Wong 995afc51aaaSDarrick J. Wong wait_for_stable_page(page); 996afc51aaaSDarrick J. Wong return VM_FAULT_LOCKED; 997afc51aaaSDarrick J. Wong out_unlock: 998afc51aaaSDarrick J. Wong unlock_page(page); 999afc51aaaSDarrick J. Wong return block_page_mkwrite_return(ret); 1000afc51aaaSDarrick J. Wong } 1001afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_page_mkwrite); 1002598ecfbaSChristoph Hellwig 1003598ecfbaSChristoph Hellwig static void 100448d64cd1SChristoph Hellwig iomap_finish_page_writeback(struct inode *inode, struct page *page, 10050fb2d720SMatthew Wilcox (Oracle) int error, unsigned int len) 1006598ecfbaSChristoph Hellwig { 100748d64cd1SChristoph Hellwig struct iomap_page *iop = to_iomap_page(page); 1008598ecfbaSChristoph Hellwig 1009598ecfbaSChristoph Hellwig if (error) { 101048d64cd1SChristoph Hellwig SetPageError(page); 1011598ecfbaSChristoph Hellwig mapping_set_error(inode->i_mapping, -EIO); 1012598ecfbaSChristoph Hellwig } 1013598ecfbaSChristoph Hellwig 101424addd84SMatthew Wilcox (Oracle) WARN_ON_ONCE(i_blocks_per_page(inode, page) > 1 && !iop); 10150fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) <= 0); 1016598ecfbaSChristoph Hellwig 10170fb2d720SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(len, &iop->write_bytes_pending)) 101848d64cd1SChristoph Hellwig end_page_writeback(page); 1019598ecfbaSChristoph Hellwig } 1020598ecfbaSChristoph Hellwig 1021598ecfbaSChristoph Hellwig /* 1022598ecfbaSChristoph Hellwig * We're now finished for good with this ioend structure. Update the page 1023598ecfbaSChristoph Hellwig * state, release holds on bios, and finally free up memory. Do not use the 1024598ecfbaSChristoph Hellwig * ioend after this. 1025598ecfbaSChristoph Hellwig */ 1026598ecfbaSChristoph Hellwig static void 1027598ecfbaSChristoph Hellwig iomap_finish_ioend(struct iomap_ioend *ioend, int error) 1028598ecfbaSChristoph Hellwig { 1029598ecfbaSChristoph Hellwig struct inode *inode = ioend->io_inode; 1030598ecfbaSChristoph Hellwig struct bio *bio = &ioend->io_inline_bio; 1031598ecfbaSChristoph Hellwig struct bio *last = ioend->io_bio, *next; 1032598ecfbaSChristoph Hellwig u64 start = bio->bi_iter.bi_sector; 1033c275779fSZorro Lang loff_t offset = ioend->io_offset; 1034598ecfbaSChristoph Hellwig bool quiet = bio_flagged(bio, BIO_QUIET); 1035598ecfbaSChristoph Hellwig 1036598ecfbaSChristoph Hellwig for (bio = &ioend->io_inline_bio; bio; bio = next) { 1037598ecfbaSChristoph Hellwig struct bio_vec *bv; 1038598ecfbaSChristoph Hellwig struct bvec_iter_all iter_all; 1039598ecfbaSChristoph Hellwig 1040598ecfbaSChristoph Hellwig /* 1041598ecfbaSChristoph Hellwig * For the last bio, bi_private points to the ioend, so we 1042598ecfbaSChristoph Hellwig * need to explicitly end the iteration here. 1043598ecfbaSChristoph Hellwig */ 1044598ecfbaSChristoph Hellwig if (bio == last) 1045598ecfbaSChristoph Hellwig next = NULL; 1046598ecfbaSChristoph Hellwig else 1047598ecfbaSChristoph Hellwig next = bio->bi_private; 1048598ecfbaSChristoph Hellwig 1049598ecfbaSChristoph Hellwig /* walk each page on bio, ending page IO on them */ 1050598ecfbaSChristoph Hellwig bio_for_each_segment_all(bv, bio, iter_all) 10510fb2d720SMatthew Wilcox (Oracle) iomap_finish_page_writeback(inode, bv->bv_page, error, 10520fb2d720SMatthew Wilcox (Oracle) bv->bv_len); 1053598ecfbaSChristoph Hellwig bio_put(bio); 1054598ecfbaSChristoph Hellwig } 1055c275779fSZorro Lang /* The ioend has been freed by bio_put() */ 1056598ecfbaSChristoph Hellwig 1057598ecfbaSChristoph Hellwig if (unlikely(error && !quiet)) { 1058598ecfbaSChristoph Hellwig printk_ratelimited(KERN_ERR 10599cd0ed63SDarrick J. Wong "%s: writeback error on inode %lu, offset %lld, sector %llu", 1060c275779fSZorro Lang inode->i_sb->s_id, inode->i_ino, offset, start); 1061598ecfbaSChristoph Hellwig } 1062598ecfbaSChristoph Hellwig } 1063598ecfbaSChristoph Hellwig 1064598ecfbaSChristoph Hellwig void 1065598ecfbaSChristoph Hellwig iomap_finish_ioends(struct iomap_ioend *ioend, int error) 1066598ecfbaSChristoph Hellwig { 1067598ecfbaSChristoph Hellwig struct list_head tmp; 1068598ecfbaSChristoph Hellwig 1069598ecfbaSChristoph Hellwig list_replace_init(&ioend->io_list, &tmp); 1070598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1071598ecfbaSChristoph Hellwig 1072598ecfbaSChristoph Hellwig while (!list_empty(&tmp)) { 1073598ecfbaSChristoph Hellwig ioend = list_first_entry(&tmp, struct iomap_ioend, io_list); 1074598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1075598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1076598ecfbaSChristoph Hellwig } 1077598ecfbaSChristoph Hellwig } 1078598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_finish_ioends); 1079598ecfbaSChristoph Hellwig 1080598ecfbaSChristoph Hellwig /* 1081598ecfbaSChristoph Hellwig * We can merge two adjacent ioends if they have the same set of work to do. 1082598ecfbaSChristoph Hellwig */ 1083598ecfbaSChristoph Hellwig static bool 1084598ecfbaSChristoph Hellwig iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next) 1085598ecfbaSChristoph Hellwig { 1086598ecfbaSChristoph Hellwig if (ioend->io_bio->bi_status != next->io_bio->bi_status) 1087598ecfbaSChristoph Hellwig return false; 1088598ecfbaSChristoph Hellwig if ((ioend->io_flags & IOMAP_F_SHARED) ^ 1089598ecfbaSChristoph Hellwig (next->io_flags & IOMAP_F_SHARED)) 1090598ecfbaSChristoph Hellwig return false; 1091598ecfbaSChristoph Hellwig if ((ioend->io_type == IOMAP_UNWRITTEN) ^ 1092598ecfbaSChristoph Hellwig (next->io_type == IOMAP_UNWRITTEN)) 1093598ecfbaSChristoph Hellwig return false; 1094598ecfbaSChristoph Hellwig if (ioend->io_offset + ioend->io_size != next->io_offset) 1095598ecfbaSChristoph Hellwig return false; 1096598ecfbaSChristoph Hellwig return true; 1097598ecfbaSChristoph Hellwig } 1098598ecfbaSChristoph Hellwig 1099598ecfbaSChristoph Hellwig void 11006e552494SBrian Foster iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends) 1101598ecfbaSChristoph Hellwig { 1102598ecfbaSChristoph Hellwig struct iomap_ioend *next; 1103598ecfbaSChristoph Hellwig 1104598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1105598ecfbaSChristoph Hellwig 1106598ecfbaSChristoph Hellwig while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend, 1107598ecfbaSChristoph Hellwig io_list))) { 1108598ecfbaSChristoph Hellwig if (!iomap_ioend_can_merge(ioend, next)) 1109598ecfbaSChristoph Hellwig break; 1110598ecfbaSChristoph Hellwig list_move_tail(&next->io_list, &ioend->io_list); 1111598ecfbaSChristoph Hellwig ioend->io_size += next->io_size; 1112598ecfbaSChristoph Hellwig } 1113598ecfbaSChristoph Hellwig } 1114598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_ioend_try_merge); 1115598ecfbaSChristoph Hellwig 1116598ecfbaSChristoph Hellwig static int 11174f0f586bSSami Tolvanen iomap_ioend_compare(void *priv, const struct list_head *a, 11184f0f586bSSami Tolvanen const struct list_head *b) 1119598ecfbaSChristoph Hellwig { 1120b3d423ecSChristoph Hellwig struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list); 1121b3d423ecSChristoph Hellwig struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list); 1122598ecfbaSChristoph Hellwig 1123598ecfbaSChristoph Hellwig if (ia->io_offset < ib->io_offset) 1124598ecfbaSChristoph Hellwig return -1; 1125b3d423ecSChristoph Hellwig if (ia->io_offset > ib->io_offset) 1126598ecfbaSChristoph Hellwig return 1; 1127598ecfbaSChristoph Hellwig return 0; 1128598ecfbaSChristoph Hellwig } 1129598ecfbaSChristoph Hellwig 1130598ecfbaSChristoph Hellwig void 1131598ecfbaSChristoph Hellwig iomap_sort_ioends(struct list_head *ioend_list) 1132598ecfbaSChristoph Hellwig { 1133598ecfbaSChristoph Hellwig list_sort(NULL, ioend_list, iomap_ioend_compare); 1134598ecfbaSChristoph Hellwig } 1135598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_sort_ioends); 1136598ecfbaSChristoph Hellwig 1137598ecfbaSChristoph Hellwig static void iomap_writepage_end_bio(struct bio *bio) 1138598ecfbaSChristoph Hellwig { 1139598ecfbaSChristoph Hellwig struct iomap_ioend *ioend = bio->bi_private; 1140598ecfbaSChristoph Hellwig 1141598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status)); 1142598ecfbaSChristoph Hellwig } 1143598ecfbaSChristoph Hellwig 1144598ecfbaSChristoph Hellwig /* 1145598ecfbaSChristoph Hellwig * Submit the final bio for an ioend. 1146598ecfbaSChristoph Hellwig * 1147598ecfbaSChristoph Hellwig * If @error is non-zero, it means that we have a situation where some part of 1148598ecfbaSChristoph Hellwig * the submission process has failed after we have marked paged for writeback 1149598ecfbaSChristoph Hellwig * and unlocked them. In this situation, we need to fail the bio instead of 1150598ecfbaSChristoph Hellwig * submitting it. This typically only happens on a filesystem shutdown. 1151598ecfbaSChristoph Hellwig */ 1152598ecfbaSChristoph Hellwig static int 1153598ecfbaSChristoph Hellwig iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend, 1154598ecfbaSChristoph Hellwig int error) 1155598ecfbaSChristoph Hellwig { 1156598ecfbaSChristoph Hellwig ioend->io_bio->bi_private = ioend; 1157598ecfbaSChristoph Hellwig ioend->io_bio->bi_end_io = iomap_writepage_end_bio; 1158598ecfbaSChristoph Hellwig 1159598ecfbaSChristoph Hellwig if (wpc->ops->prepare_ioend) 1160598ecfbaSChristoph Hellwig error = wpc->ops->prepare_ioend(ioend, error); 1161598ecfbaSChristoph Hellwig if (error) { 1162598ecfbaSChristoph Hellwig /* 1163598ecfbaSChristoph Hellwig * If we are failing the IO now, just mark the ioend with an 1164598ecfbaSChristoph Hellwig * error and finish it. This will run IO completion immediately 1165598ecfbaSChristoph Hellwig * as there is only one reference to the ioend at this point in 1166598ecfbaSChristoph Hellwig * time. 1167598ecfbaSChristoph Hellwig */ 1168598ecfbaSChristoph Hellwig ioend->io_bio->bi_status = errno_to_blk_status(error); 1169598ecfbaSChristoph Hellwig bio_endio(ioend->io_bio); 1170598ecfbaSChristoph Hellwig return error; 1171598ecfbaSChristoph Hellwig } 1172598ecfbaSChristoph Hellwig 1173598ecfbaSChristoph Hellwig submit_bio(ioend->io_bio); 1174598ecfbaSChristoph Hellwig return 0; 1175598ecfbaSChristoph Hellwig } 1176598ecfbaSChristoph Hellwig 1177598ecfbaSChristoph Hellwig static struct iomap_ioend * 1178598ecfbaSChristoph Hellwig iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc, 1179598ecfbaSChristoph Hellwig loff_t offset, sector_t sector, struct writeback_control *wbc) 1180598ecfbaSChristoph Hellwig { 1181598ecfbaSChristoph Hellwig struct iomap_ioend *ioend; 1182598ecfbaSChristoph Hellwig struct bio *bio; 1183598ecfbaSChristoph Hellwig 1184a8affc03SChristoph Hellwig bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, &iomap_ioend_bioset); 1185598ecfbaSChristoph Hellwig bio_set_dev(bio, wpc->iomap.bdev); 1186598ecfbaSChristoph Hellwig bio->bi_iter.bi_sector = sector; 1187598ecfbaSChristoph Hellwig bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); 1188598ecfbaSChristoph Hellwig bio->bi_write_hint = inode->i_write_hint; 1189598ecfbaSChristoph Hellwig wbc_init_bio(wbc, bio); 1190598ecfbaSChristoph Hellwig 1191598ecfbaSChristoph Hellwig ioend = container_of(bio, struct iomap_ioend, io_inline_bio); 1192598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1193598ecfbaSChristoph Hellwig ioend->io_type = wpc->iomap.type; 1194598ecfbaSChristoph Hellwig ioend->io_flags = wpc->iomap.flags; 1195598ecfbaSChristoph Hellwig ioend->io_inode = inode; 1196598ecfbaSChristoph Hellwig ioend->io_size = 0; 1197598ecfbaSChristoph Hellwig ioend->io_offset = offset; 1198598ecfbaSChristoph Hellwig ioend->io_bio = bio; 1199598ecfbaSChristoph Hellwig return ioend; 1200598ecfbaSChristoph Hellwig } 1201598ecfbaSChristoph Hellwig 1202598ecfbaSChristoph Hellwig /* 1203598ecfbaSChristoph Hellwig * Allocate a new bio, and chain the old bio to the new one. 1204598ecfbaSChristoph Hellwig * 1205598ecfbaSChristoph Hellwig * Note that we have to do perform the chaining in this unintuitive order 1206598ecfbaSChristoph Hellwig * so that the bi_private linkage is set up in the right direction for the 1207598ecfbaSChristoph Hellwig * traversal in iomap_finish_ioend(). 1208598ecfbaSChristoph Hellwig */ 1209598ecfbaSChristoph Hellwig static struct bio * 1210598ecfbaSChristoph Hellwig iomap_chain_bio(struct bio *prev) 1211598ecfbaSChristoph Hellwig { 1212598ecfbaSChristoph Hellwig struct bio *new; 1213598ecfbaSChristoph Hellwig 1214a8affc03SChristoph Hellwig new = bio_alloc(GFP_NOFS, BIO_MAX_VECS); 1215598ecfbaSChristoph Hellwig bio_copy_dev(new, prev);/* also copies over blkcg information */ 1216598ecfbaSChristoph Hellwig new->bi_iter.bi_sector = bio_end_sector(prev); 1217598ecfbaSChristoph Hellwig new->bi_opf = prev->bi_opf; 1218598ecfbaSChristoph Hellwig new->bi_write_hint = prev->bi_write_hint; 1219598ecfbaSChristoph Hellwig 1220598ecfbaSChristoph Hellwig bio_chain(prev, new); 1221598ecfbaSChristoph Hellwig bio_get(prev); /* for iomap_finish_ioend */ 1222598ecfbaSChristoph Hellwig submit_bio(prev); 1223598ecfbaSChristoph Hellwig return new; 1224598ecfbaSChristoph Hellwig } 1225598ecfbaSChristoph Hellwig 1226598ecfbaSChristoph Hellwig static bool 1227598ecfbaSChristoph Hellwig iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset, 1228598ecfbaSChristoph Hellwig sector_t sector) 1229598ecfbaSChristoph Hellwig { 1230598ecfbaSChristoph Hellwig if ((wpc->iomap.flags & IOMAP_F_SHARED) != 1231598ecfbaSChristoph Hellwig (wpc->ioend->io_flags & IOMAP_F_SHARED)) 1232598ecfbaSChristoph Hellwig return false; 1233598ecfbaSChristoph Hellwig if (wpc->iomap.type != wpc->ioend->io_type) 1234598ecfbaSChristoph Hellwig return false; 1235598ecfbaSChristoph Hellwig if (offset != wpc->ioend->io_offset + wpc->ioend->io_size) 1236598ecfbaSChristoph Hellwig return false; 1237598ecfbaSChristoph Hellwig if (sector != bio_end_sector(wpc->ioend->io_bio)) 1238598ecfbaSChristoph Hellwig return false; 1239598ecfbaSChristoph Hellwig return true; 1240598ecfbaSChristoph Hellwig } 1241598ecfbaSChristoph Hellwig 1242598ecfbaSChristoph Hellwig /* 1243598ecfbaSChristoph Hellwig * Test to see if we have an existing ioend structure that we could append to 1244598ecfbaSChristoph Hellwig * first, otherwise finish off the current ioend and start another. 1245598ecfbaSChristoph Hellwig */ 1246598ecfbaSChristoph Hellwig static void 1247598ecfbaSChristoph Hellwig iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page, 1248598ecfbaSChristoph Hellwig struct iomap_page *iop, struct iomap_writepage_ctx *wpc, 1249598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct list_head *iolist) 1250598ecfbaSChristoph Hellwig { 1251598ecfbaSChristoph Hellwig sector_t sector = iomap_sector(&wpc->iomap, offset); 1252598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1253598ecfbaSChristoph Hellwig unsigned poff = offset & (PAGE_SIZE - 1); 1254598ecfbaSChristoph Hellwig 1255598ecfbaSChristoph Hellwig if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) { 1256598ecfbaSChristoph Hellwig if (wpc->ioend) 1257598ecfbaSChristoph Hellwig list_add(&wpc->ioend->io_list, iolist); 1258598ecfbaSChristoph Hellwig wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc); 1259598ecfbaSChristoph Hellwig } 1260598ecfbaSChristoph Hellwig 1261*c1b79f11SChristoph Hellwig if (bio_add_page(wpc->ioend->io_bio, page, len, poff) != len) { 1262*c1b79f11SChristoph Hellwig wpc->ioend->io_bio = iomap_chain_bio(wpc->ioend->io_bio); 1263*c1b79f11SChristoph Hellwig __bio_add_page(wpc->ioend->io_bio, page, len, poff); 1264*c1b79f11SChristoph Hellwig } 1265*c1b79f11SChristoph Hellwig 12660fb2d720SMatthew Wilcox (Oracle) if (iop) 12670fb2d720SMatthew Wilcox (Oracle) atomic_add(len, &iop->write_bytes_pending); 1268598ecfbaSChristoph Hellwig wpc->ioend->io_size += len; 1269598ecfbaSChristoph Hellwig wbc_account_cgroup_owner(wbc, page, len); 1270598ecfbaSChristoph Hellwig } 1271598ecfbaSChristoph Hellwig 1272598ecfbaSChristoph Hellwig /* 1273598ecfbaSChristoph Hellwig * We implement an immediate ioend submission policy here to avoid needing to 1274598ecfbaSChristoph Hellwig * chain multiple ioends and hence nest mempool allocations which can violate 1275598ecfbaSChristoph Hellwig * forward progress guarantees we need to provide. The current ioend we are 1276598ecfbaSChristoph Hellwig * adding blocks to is cached on the writepage context, and if the new block 1277598ecfbaSChristoph Hellwig * does not append to the cached ioend it will create a new ioend and cache that 1278598ecfbaSChristoph Hellwig * instead. 1279598ecfbaSChristoph Hellwig * 1280598ecfbaSChristoph Hellwig * If a new ioend is created and cached, the old ioend is returned and queued 1281598ecfbaSChristoph Hellwig * locally for submission once the entire page is processed or an error has been 1282598ecfbaSChristoph Hellwig * detected. While ioends are submitted immediately after they are completed, 1283598ecfbaSChristoph Hellwig * batching optimisations are provided by higher level block plugging. 1284598ecfbaSChristoph Hellwig * 1285598ecfbaSChristoph Hellwig * At the end of a writeback pass, there will be a cached ioend remaining on the 1286598ecfbaSChristoph Hellwig * writepage context that the caller will need to submit. 1287598ecfbaSChristoph Hellwig */ 1288598ecfbaSChristoph Hellwig static int 1289598ecfbaSChristoph Hellwig iomap_writepage_map(struct iomap_writepage_ctx *wpc, 1290598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct inode *inode, 1291598ecfbaSChristoph Hellwig struct page *page, u64 end_offset) 1292598ecfbaSChristoph Hellwig { 12938e1bcef8SAndreas Gruenbacher struct iomap_page *iop = iomap_page_create(inode, page); 1294598ecfbaSChristoph Hellwig struct iomap_ioend *ioend, *next; 1295598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1296598ecfbaSChristoph Hellwig u64 file_offset; /* file offset of page */ 1297598ecfbaSChristoph Hellwig int error = 0, count = 0, i; 1298598ecfbaSChristoph Hellwig LIST_HEAD(submit_list); 1299598ecfbaSChristoph Hellwig 13000fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) != 0); 1301598ecfbaSChristoph Hellwig 1302598ecfbaSChristoph Hellwig /* 1303598ecfbaSChristoph Hellwig * Walk through the page to find areas to write back. If we run off the 1304598ecfbaSChristoph Hellwig * end of the current map or find the current map invalid, grab a new 1305598ecfbaSChristoph Hellwig * one. 1306598ecfbaSChristoph Hellwig */ 1307598ecfbaSChristoph Hellwig for (i = 0, file_offset = page_offset(page); 1308598ecfbaSChristoph Hellwig i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset; 1309598ecfbaSChristoph Hellwig i++, file_offset += len) { 1310598ecfbaSChristoph Hellwig if (iop && !test_bit(i, iop->uptodate)) 1311598ecfbaSChristoph Hellwig continue; 1312598ecfbaSChristoph Hellwig 1313598ecfbaSChristoph Hellwig error = wpc->ops->map_blocks(wpc, inode, file_offset); 1314598ecfbaSChristoph Hellwig if (error) 1315598ecfbaSChristoph Hellwig break; 13163e19e6f3SChristoph Hellwig if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE)) 13173e19e6f3SChristoph Hellwig continue; 1318598ecfbaSChristoph Hellwig if (wpc->iomap.type == IOMAP_HOLE) 1319598ecfbaSChristoph Hellwig continue; 1320598ecfbaSChristoph Hellwig iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc, 1321598ecfbaSChristoph Hellwig &submit_list); 1322598ecfbaSChristoph Hellwig count++; 1323598ecfbaSChristoph Hellwig } 1324598ecfbaSChristoph Hellwig 1325598ecfbaSChristoph Hellwig WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list)); 1326598ecfbaSChristoph Hellwig WARN_ON_ONCE(!PageLocked(page)); 1327598ecfbaSChristoph Hellwig WARN_ON_ONCE(PageWriteback(page)); 132850e7d6c7SBrian Foster WARN_ON_ONCE(PageDirty(page)); 1329598ecfbaSChristoph Hellwig 1330598ecfbaSChristoph Hellwig /* 1331598ecfbaSChristoph Hellwig * We cannot cancel the ioend directly here on error. We may have 1332598ecfbaSChristoph Hellwig * already set other pages under writeback and hence we have to run I/O 1333598ecfbaSChristoph Hellwig * completion to mark the error state of the pages under writeback 1334598ecfbaSChristoph Hellwig * appropriately. 1335598ecfbaSChristoph Hellwig */ 1336598ecfbaSChristoph Hellwig if (unlikely(error)) { 1337598ecfbaSChristoph Hellwig /* 1338763e4cdcSBrian Foster * Let the filesystem know what portion of the current page 1339763e4cdcSBrian Foster * failed to map. If the page wasn't been added to ioend, it 1340763e4cdcSBrian Foster * won't be affected by I/O completion and we must unlock it 1341763e4cdcSBrian Foster * now. 1342598ecfbaSChristoph Hellwig */ 1343598ecfbaSChristoph Hellwig if (wpc->ops->discard_page) 1344763e4cdcSBrian Foster wpc->ops->discard_page(page, file_offset); 1345763e4cdcSBrian Foster if (!count) { 1346598ecfbaSChristoph Hellwig ClearPageUptodate(page); 1347598ecfbaSChristoph Hellwig unlock_page(page); 1348598ecfbaSChristoph Hellwig goto done; 1349598ecfbaSChristoph Hellwig } 1350598ecfbaSChristoph Hellwig } 1351598ecfbaSChristoph Hellwig 135250e7d6c7SBrian Foster set_page_writeback(page); 1353598ecfbaSChristoph Hellwig unlock_page(page); 1354598ecfbaSChristoph Hellwig 1355598ecfbaSChristoph Hellwig /* 1356598ecfbaSChristoph Hellwig * Preserve the original error if there was one, otherwise catch 1357598ecfbaSChristoph Hellwig * submission errors here and propagate into subsequent ioend 1358598ecfbaSChristoph Hellwig * submissions. 1359598ecfbaSChristoph Hellwig */ 1360598ecfbaSChristoph Hellwig list_for_each_entry_safe(ioend, next, &submit_list, io_list) { 1361598ecfbaSChristoph Hellwig int error2; 1362598ecfbaSChristoph Hellwig 1363598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1364598ecfbaSChristoph Hellwig error2 = iomap_submit_ioend(wpc, ioend, error); 1365598ecfbaSChristoph Hellwig if (error2 && !error) 1366598ecfbaSChristoph Hellwig error = error2; 1367598ecfbaSChristoph Hellwig } 1368598ecfbaSChristoph Hellwig 1369598ecfbaSChristoph Hellwig /* 1370598ecfbaSChristoph Hellwig * We can end up here with no error and nothing to write only if we race 1371598ecfbaSChristoph Hellwig * with a partial page truncate on a sub-page block sized filesystem. 1372598ecfbaSChristoph Hellwig */ 1373598ecfbaSChristoph Hellwig if (!count) 1374598ecfbaSChristoph Hellwig end_page_writeback(page); 1375598ecfbaSChristoph Hellwig done: 1376598ecfbaSChristoph Hellwig mapping_set_error(page->mapping, error); 1377598ecfbaSChristoph Hellwig return error; 1378598ecfbaSChristoph Hellwig } 1379598ecfbaSChristoph Hellwig 1380598ecfbaSChristoph Hellwig /* 1381598ecfbaSChristoph Hellwig * Write out a dirty page. 1382598ecfbaSChristoph Hellwig * 1383598ecfbaSChristoph Hellwig * For delalloc space on the page we need to allocate space and flush it. 1384598ecfbaSChristoph Hellwig * For unwritten space on the page we need to start the conversion to 1385598ecfbaSChristoph Hellwig * regular allocated space. 1386598ecfbaSChristoph Hellwig */ 1387598ecfbaSChristoph Hellwig static int 1388598ecfbaSChristoph Hellwig iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data) 1389598ecfbaSChristoph Hellwig { 1390598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc = data; 1391598ecfbaSChristoph Hellwig struct inode *inode = page->mapping->host; 1392598ecfbaSChristoph Hellwig pgoff_t end_index; 1393598ecfbaSChristoph Hellwig u64 end_offset; 1394598ecfbaSChristoph Hellwig loff_t offset; 1395598ecfbaSChristoph Hellwig 13961ac99452SMatthew Wilcox (Oracle) trace_iomap_writepage(inode, page_offset(page), PAGE_SIZE); 1397598ecfbaSChristoph Hellwig 1398598ecfbaSChristoph Hellwig /* 1399598ecfbaSChristoph Hellwig * Refuse to write the page out if we are called from reclaim context. 1400598ecfbaSChristoph Hellwig * 1401598ecfbaSChristoph Hellwig * This avoids stack overflows when called from deeply used stacks in 1402598ecfbaSChristoph Hellwig * random callers for direct reclaim or memcg reclaim. We explicitly 1403598ecfbaSChristoph Hellwig * allow reclaim from kswapd as the stack usage there is relatively low. 1404598ecfbaSChristoph Hellwig * 1405598ecfbaSChristoph Hellwig * This should never happen except in the case of a VM regression so 1406598ecfbaSChristoph Hellwig * warn about it. 1407598ecfbaSChristoph Hellwig */ 1408598ecfbaSChristoph Hellwig if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == 1409598ecfbaSChristoph Hellwig PF_MEMALLOC)) 1410598ecfbaSChristoph Hellwig goto redirty; 1411598ecfbaSChristoph Hellwig 1412598ecfbaSChristoph Hellwig /* 1413598ecfbaSChristoph Hellwig * Is this page beyond the end of the file? 1414598ecfbaSChristoph Hellwig * 1415598ecfbaSChristoph Hellwig * The page index is less than the end_index, adjust the end_offset 1416598ecfbaSChristoph Hellwig * to the highest offset that this page should represent. 1417598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1418598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1419598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1420598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | | 1421598ecfbaSChristoph Hellwig * ^--------------------------------^----------|-------- 1422598ecfbaSChristoph Hellwig * | desired writeback range | see else | 1423598ecfbaSChristoph Hellwig * ---------------------------------^------------------| 1424598ecfbaSChristoph Hellwig */ 1425598ecfbaSChristoph Hellwig offset = i_size_read(inode); 1426598ecfbaSChristoph Hellwig end_index = offset >> PAGE_SHIFT; 1427598ecfbaSChristoph Hellwig if (page->index < end_index) 1428598ecfbaSChristoph Hellwig end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT; 1429598ecfbaSChristoph Hellwig else { 1430598ecfbaSChristoph Hellwig /* 1431598ecfbaSChristoph Hellwig * Check whether the page to write out is beyond or straddles 1432598ecfbaSChristoph Hellwig * i_size or not. 1433598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1434598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1435598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1436598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | 1437598ecfbaSChristoph Hellwig * ^--------------------------------^-----------|--------- 1438598ecfbaSChristoph Hellwig * | | Straddles | 1439598ecfbaSChristoph Hellwig * ---------------------------------^-----------|--------| 1440598ecfbaSChristoph Hellwig */ 1441598ecfbaSChristoph Hellwig unsigned offset_into_page = offset & (PAGE_SIZE - 1); 1442598ecfbaSChristoph Hellwig 1443598ecfbaSChristoph Hellwig /* 1444598ecfbaSChristoph Hellwig * Skip the page if it is fully outside i_size, e.g. due to a 1445598ecfbaSChristoph Hellwig * truncate operation that is in progress. We must redirty the 1446598ecfbaSChristoph Hellwig * page so that reclaim stops reclaiming it. Otherwise 1447598ecfbaSChristoph Hellwig * iomap_vm_releasepage() is called on it and gets confused. 1448598ecfbaSChristoph Hellwig * 1449598ecfbaSChristoph Hellwig * Note that the end_index is unsigned long, it would overflow 1450598ecfbaSChristoph Hellwig * if the given offset is greater than 16TB on 32-bit system 1451598ecfbaSChristoph Hellwig * and if we do check the page is fully outside i_size or not 1452598ecfbaSChristoph Hellwig * via "if (page->index >= end_index + 1)" as "end_index + 1" 1453598ecfbaSChristoph Hellwig * will be evaluated to 0. Hence this page will be redirtied 1454598ecfbaSChristoph Hellwig * and be written out repeatedly which would result in an 1455598ecfbaSChristoph Hellwig * infinite loop, the user program that perform this operation 1456598ecfbaSChristoph Hellwig * will hang. Instead, we can verify this situation by checking 1457598ecfbaSChristoph Hellwig * if the page to write is totally beyond the i_size or if it's 1458598ecfbaSChristoph Hellwig * offset is just equal to the EOF. 1459598ecfbaSChristoph Hellwig */ 1460598ecfbaSChristoph Hellwig if (page->index > end_index || 1461598ecfbaSChristoph Hellwig (page->index == end_index && offset_into_page == 0)) 1462598ecfbaSChristoph Hellwig goto redirty; 1463598ecfbaSChristoph Hellwig 1464598ecfbaSChristoph Hellwig /* 1465598ecfbaSChristoph Hellwig * The page straddles i_size. It must be zeroed out on each 1466598ecfbaSChristoph Hellwig * and every writepage invocation because it may be mmapped. 1467598ecfbaSChristoph Hellwig * "A file is mapped in multiples of the page size. For a file 1468598ecfbaSChristoph Hellwig * that is not a multiple of the page size, the remaining 1469598ecfbaSChristoph Hellwig * memory is zeroed when mapped, and writes to that region are 1470598ecfbaSChristoph Hellwig * not written out to the file." 1471598ecfbaSChristoph Hellwig */ 1472598ecfbaSChristoph Hellwig zero_user_segment(page, offset_into_page, PAGE_SIZE); 1473598ecfbaSChristoph Hellwig 1474598ecfbaSChristoph Hellwig /* Adjust the end_offset to the end of file */ 1475598ecfbaSChristoph Hellwig end_offset = offset; 1476598ecfbaSChristoph Hellwig } 1477598ecfbaSChristoph Hellwig 1478598ecfbaSChristoph Hellwig return iomap_writepage_map(wpc, wbc, inode, page, end_offset); 1479598ecfbaSChristoph Hellwig 1480598ecfbaSChristoph Hellwig redirty: 1481598ecfbaSChristoph Hellwig redirty_page_for_writepage(wbc, page); 1482598ecfbaSChristoph Hellwig unlock_page(page); 1483598ecfbaSChristoph Hellwig return 0; 1484598ecfbaSChristoph Hellwig } 1485598ecfbaSChristoph Hellwig 1486598ecfbaSChristoph Hellwig int 1487598ecfbaSChristoph Hellwig iomap_writepage(struct page *page, struct writeback_control *wbc, 1488598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1489598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1490598ecfbaSChristoph Hellwig { 1491598ecfbaSChristoph Hellwig int ret; 1492598ecfbaSChristoph Hellwig 1493598ecfbaSChristoph Hellwig wpc->ops = ops; 1494598ecfbaSChristoph Hellwig ret = iomap_do_writepage(page, wbc, wpc); 1495598ecfbaSChristoph Hellwig if (!wpc->ioend) 1496598ecfbaSChristoph Hellwig return ret; 1497598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1498598ecfbaSChristoph Hellwig } 1499598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepage); 1500598ecfbaSChristoph Hellwig 1501598ecfbaSChristoph Hellwig int 1502598ecfbaSChristoph Hellwig iomap_writepages(struct address_space *mapping, struct writeback_control *wbc, 1503598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1504598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1505598ecfbaSChristoph Hellwig { 1506598ecfbaSChristoph Hellwig int ret; 1507598ecfbaSChristoph Hellwig 1508598ecfbaSChristoph Hellwig wpc->ops = ops; 1509598ecfbaSChristoph Hellwig ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc); 1510598ecfbaSChristoph Hellwig if (!wpc->ioend) 1511598ecfbaSChristoph Hellwig return ret; 1512598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1513598ecfbaSChristoph Hellwig } 1514598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepages); 1515598ecfbaSChristoph Hellwig 1516598ecfbaSChristoph Hellwig static int __init iomap_init(void) 1517598ecfbaSChristoph Hellwig { 1518598ecfbaSChristoph Hellwig return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE), 1519598ecfbaSChristoph Hellwig offsetof(struct iomap_ioend, io_inline_bio), 1520598ecfbaSChristoph Hellwig BIOSET_NEED_BVECS); 1521598ecfbaSChristoph Hellwig } 1522598ecfbaSChristoph Hellwig fs_initcall(iomap_init); 1523