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; 30*0fb2d720SMatthew 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); 6358aeb731SGuoqing Jiang attach_page_private(page, iop); 64afc51aaaSDarrick J. Wong return iop; 65afc51aaaSDarrick J. Wong } 66afc51aaaSDarrick J. Wong 67afc51aaaSDarrick J. Wong static void 68afc51aaaSDarrick J. Wong iomap_page_release(struct page *page) 69afc51aaaSDarrick J. Wong { 7058aeb731SGuoqing Jiang struct iomap_page *iop = detach_page_private(page); 710a195b91SMatthew Wilcox (Oracle) unsigned int nr_blocks = i_blocks_per_page(page->mapping->host, page); 72afc51aaaSDarrick J. Wong 73afc51aaaSDarrick J. Wong if (!iop) 74afc51aaaSDarrick J. Wong return; 757d636676SMatthew Wilcox (Oracle) WARN_ON_ONCE(atomic_read(&iop->read_bytes_pending)); 76*0fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(atomic_read(&iop->write_bytes_pending)); 770a195b91SMatthew Wilcox (Oracle) WARN_ON_ONCE(bitmap_full(iop->uptodate, nr_blocks) != 780a195b91SMatthew Wilcox (Oracle) PageUptodate(page)); 79afc51aaaSDarrick J. Wong kfree(iop); 80afc51aaaSDarrick J. Wong } 81afc51aaaSDarrick J. Wong 82afc51aaaSDarrick J. Wong /* 83afc51aaaSDarrick J. Wong * Calculate the range inside the page that we actually need to read. 84afc51aaaSDarrick J. Wong */ 85afc51aaaSDarrick J. Wong static void 86afc51aaaSDarrick J. Wong iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop, 87afc51aaaSDarrick J. Wong loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp) 88afc51aaaSDarrick J. Wong { 89afc51aaaSDarrick J. Wong loff_t orig_pos = *pos; 90afc51aaaSDarrick J. Wong loff_t isize = i_size_read(inode); 91afc51aaaSDarrick J. Wong unsigned block_bits = inode->i_blkbits; 92afc51aaaSDarrick J. Wong unsigned block_size = (1 << block_bits); 93afc51aaaSDarrick J. Wong unsigned poff = offset_in_page(*pos); 94afc51aaaSDarrick J. Wong unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length); 95afc51aaaSDarrick J. Wong unsigned first = poff >> block_bits; 96afc51aaaSDarrick J. Wong unsigned last = (poff + plen - 1) >> block_bits; 97afc51aaaSDarrick J. Wong 98afc51aaaSDarrick J. Wong /* 99afc51aaaSDarrick J. Wong * If the block size is smaller than the page size we need to check the 100afc51aaaSDarrick J. Wong * per-block uptodate status and adjust the offset and length if needed 101afc51aaaSDarrick J. Wong * to avoid reading in already uptodate ranges. 102afc51aaaSDarrick J. Wong */ 103afc51aaaSDarrick J. Wong if (iop) { 104afc51aaaSDarrick J. Wong unsigned int i; 105afc51aaaSDarrick J. Wong 106afc51aaaSDarrick J. Wong /* move forward for each leading block marked uptodate */ 107afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) { 108afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 109afc51aaaSDarrick J. Wong break; 110afc51aaaSDarrick J. Wong *pos += block_size; 111afc51aaaSDarrick J. Wong poff += block_size; 112afc51aaaSDarrick J. Wong plen -= block_size; 113afc51aaaSDarrick J. Wong first++; 114afc51aaaSDarrick J. Wong } 115afc51aaaSDarrick J. Wong 116afc51aaaSDarrick J. Wong /* truncate len if we find any trailing uptodate block(s) */ 117afc51aaaSDarrick J. Wong for ( ; i <= last; i++) { 118afc51aaaSDarrick J. Wong if (test_bit(i, iop->uptodate)) { 119afc51aaaSDarrick J. Wong plen -= (last - i + 1) * block_size; 120afc51aaaSDarrick J. Wong last = i - 1; 121afc51aaaSDarrick J. Wong break; 122afc51aaaSDarrick J. Wong } 123afc51aaaSDarrick J. Wong } 124afc51aaaSDarrick J. Wong } 125afc51aaaSDarrick J. Wong 126afc51aaaSDarrick J. Wong /* 127afc51aaaSDarrick J. Wong * If the extent spans the block that contains the i_size we need to 128afc51aaaSDarrick J. Wong * handle both halves separately so that we properly zero data in the 129afc51aaaSDarrick J. Wong * page cache for blocks that are entirely outside of i_size. 130afc51aaaSDarrick J. Wong */ 131afc51aaaSDarrick J. Wong if (orig_pos <= isize && orig_pos + length > isize) { 132afc51aaaSDarrick J. Wong unsigned end = offset_in_page(isize - 1) >> block_bits; 133afc51aaaSDarrick J. Wong 134afc51aaaSDarrick J. Wong if (first <= end && last > end) 135afc51aaaSDarrick J. Wong plen -= (last - end) * block_size; 136afc51aaaSDarrick J. Wong } 137afc51aaaSDarrick J. Wong 138afc51aaaSDarrick J. Wong *offp = poff; 139afc51aaaSDarrick J. Wong *lenp = plen; 140afc51aaaSDarrick J. Wong } 141afc51aaaSDarrick J. Wong 142afc51aaaSDarrick J. Wong static void 1431cea335dSChristoph Hellwig iomap_iop_set_range_uptodate(struct page *page, unsigned off, unsigned len) 144afc51aaaSDarrick J. Wong { 145afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 146afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 147afc51aaaSDarrick J. Wong unsigned first = off >> inode->i_blkbits; 148afc51aaaSDarrick J. Wong unsigned last = (off + len - 1) >> inode->i_blkbits; 1491cea335dSChristoph Hellwig unsigned long flags; 150afc51aaaSDarrick J. Wong 1511cea335dSChristoph Hellwig spin_lock_irqsave(&iop->uptodate_lock, flags); 152b21866f5SMatthew Wilcox (Oracle) bitmap_set(iop->uptodate, first, last - first + 1); 153b21866f5SMatthew Wilcox (Oracle) if (bitmap_full(iop->uptodate, i_blocks_per_page(inode, page))) 1541cea335dSChristoph Hellwig SetPageUptodate(page); 1551cea335dSChristoph Hellwig spin_unlock_irqrestore(&iop->uptodate_lock, flags); 156afc51aaaSDarrick J. Wong } 157afc51aaaSDarrick J. Wong 1581cea335dSChristoph Hellwig static void 1591cea335dSChristoph Hellwig iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len) 1601cea335dSChristoph Hellwig { 1611cea335dSChristoph Hellwig if (PageError(page)) 1621cea335dSChristoph Hellwig return; 1631cea335dSChristoph Hellwig 1641cea335dSChristoph Hellwig if (page_has_private(page)) 1651cea335dSChristoph Hellwig iomap_iop_set_range_uptodate(page, off, len); 1661cea335dSChristoph Hellwig else 167afc51aaaSDarrick J. Wong SetPageUptodate(page); 168afc51aaaSDarrick J. Wong } 169afc51aaaSDarrick J. Wong 170afc51aaaSDarrick J. Wong static void 171afc51aaaSDarrick J. Wong iomap_read_page_end_io(struct bio_vec *bvec, int error) 172afc51aaaSDarrick J. Wong { 173afc51aaaSDarrick J. Wong struct page *page = bvec->bv_page; 174afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 175afc51aaaSDarrick J. Wong 176afc51aaaSDarrick J. Wong if (unlikely(error)) { 177afc51aaaSDarrick J. Wong ClearPageUptodate(page); 178afc51aaaSDarrick J. Wong SetPageError(page); 179afc51aaaSDarrick J. Wong } else { 180afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len); 181afc51aaaSDarrick J. Wong } 182afc51aaaSDarrick J. Wong 1837d636676SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(bvec->bv_len, &iop->read_bytes_pending)) 1847d636676SMatthew Wilcox (Oracle) unlock_page(page); 185afc51aaaSDarrick J. Wong } 186afc51aaaSDarrick J. Wong 187afc51aaaSDarrick J. Wong static void 188afc51aaaSDarrick J. Wong iomap_read_end_io(struct bio *bio) 189afc51aaaSDarrick J. Wong { 190afc51aaaSDarrick J. Wong int error = blk_status_to_errno(bio->bi_status); 191afc51aaaSDarrick J. Wong struct bio_vec *bvec; 192afc51aaaSDarrick J. Wong struct bvec_iter_all iter_all; 193afc51aaaSDarrick J. Wong 194afc51aaaSDarrick J. Wong bio_for_each_segment_all(bvec, bio, iter_all) 195afc51aaaSDarrick J. Wong iomap_read_page_end_io(bvec, error); 196afc51aaaSDarrick J. Wong bio_put(bio); 197afc51aaaSDarrick J. Wong } 198afc51aaaSDarrick J. Wong 199afc51aaaSDarrick J. Wong struct iomap_readpage_ctx { 200afc51aaaSDarrick J. Wong struct page *cur_page; 201afc51aaaSDarrick J. Wong bool cur_page_in_bio; 202afc51aaaSDarrick J. Wong struct bio *bio; 2039d24a13aSMatthew Wilcox (Oracle) struct readahead_control *rac; 204afc51aaaSDarrick J. Wong }; 205afc51aaaSDarrick J. Wong 206afc51aaaSDarrick J. Wong static void 207afc51aaaSDarrick J. Wong iomap_read_inline_data(struct inode *inode, struct page *page, 208afc51aaaSDarrick J. Wong struct iomap *iomap) 209afc51aaaSDarrick J. Wong { 210afc51aaaSDarrick J. Wong size_t size = i_size_read(inode); 211afc51aaaSDarrick J. Wong void *addr; 212afc51aaaSDarrick J. Wong 213afc51aaaSDarrick J. Wong if (PageUptodate(page)) 214afc51aaaSDarrick J. Wong return; 215afc51aaaSDarrick J. Wong 216afc51aaaSDarrick J. Wong BUG_ON(page->index); 217afc51aaaSDarrick J. Wong BUG_ON(size > PAGE_SIZE - offset_in_page(iomap->inline_data)); 218afc51aaaSDarrick J. Wong 219afc51aaaSDarrick J. Wong addr = kmap_atomic(page); 220afc51aaaSDarrick J. Wong memcpy(addr, iomap->inline_data, size); 221afc51aaaSDarrick J. Wong memset(addr + size, 0, PAGE_SIZE - size); 222afc51aaaSDarrick J. Wong kunmap_atomic(addr); 223afc51aaaSDarrick J. Wong SetPageUptodate(page); 224afc51aaaSDarrick J. Wong } 225afc51aaaSDarrick J. Wong 226009d8d84SChristoph Hellwig static inline bool iomap_block_needs_zeroing(struct inode *inode, 227009d8d84SChristoph Hellwig struct iomap *iomap, loff_t pos) 228009d8d84SChristoph Hellwig { 229009d8d84SChristoph Hellwig return iomap->type != IOMAP_MAPPED || 230009d8d84SChristoph Hellwig (iomap->flags & IOMAP_F_NEW) || 231009d8d84SChristoph Hellwig pos >= i_size_read(inode); 232009d8d84SChristoph Hellwig } 233009d8d84SChristoph Hellwig 234afc51aaaSDarrick J. Wong static loff_t 235afc51aaaSDarrick J. Wong iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 236c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 237afc51aaaSDarrick J. Wong { 238afc51aaaSDarrick J. Wong struct iomap_readpage_ctx *ctx = data; 239afc51aaaSDarrick J. Wong struct page *page = ctx->cur_page; 240afc51aaaSDarrick J. Wong struct iomap_page *iop = iomap_page_create(inode, page); 241afc51aaaSDarrick J. Wong bool same_page = false, is_contig = false; 242afc51aaaSDarrick J. Wong loff_t orig_pos = pos; 243afc51aaaSDarrick J. Wong unsigned poff, plen; 244afc51aaaSDarrick J. Wong sector_t sector; 245afc51aaaSDarrick J. Wong 246afc51aaaSDarrick J. Wong if (iomap->type == IOMAP_INLINE) { 247afc51aaaSDarrick J. Wong WARN_ON_ONCE(pos); 248afc51aaaSDarrick J. Wong iomap_read_inline_data(inode, page, iomap); 249afc51aaaSDarrick J. Wong return PAGE_SIZE; 250afc51aaaSDarrick J. Wong } 251afc51aaaSDarrick J. Wong 252afc51aaaSDarrick J. Wong /* zero post-eof blocks as the page may be mapped */ 253afc51aaaSDarrick J. Wong iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen); 254afc51aaaSDarrick J. Wong if (plen == 0) 255afc51aaaSDarrick J. Wong goto done; 256afc51aaaSDarrick J. Wong 257009d8d84SChristoph Hellwig if (iomap_block_needs_zeroing(inode, iomap, pos)) { 258afc51aaaSDarrick J. Wong zero_user(page, poff, plen); 259afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, poff, plen); 260afc51aaaSDarrick J. Wong goto done; 261afc51aaaSDarrick J. Wong } 262afc51aaaSDarrick J. Wong 263afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = true; 2647d636676SMatthew Wilcox (Oracle) if (iop) 2657d636676SMatthew Wilcox (Oracle) atomic_add(plen, &iop->read_bytes_pending); 266afc51aaaSDarrick J. Wong 2677d636676SMatthew Wilcox (Oracle) /* Try to merge into a previous segment if we can */ 268afc51aaaSDarrick J. Wong sector = iomap_sector(iomap, pos); 2697d636676SMatthew Wilcox (Oracle) if (ctx->bio && bio_end_sector(ctx->bio) == sector) { 2707d636676SMatthew Wilcox (Oracle) if (__bio_try_merge_page(ctx->bio, page, plen, poff, 2717d636676SMatthew Wilcox (Oracle) &same_page)) 272afc51aaaSDarrick J. Wong goto done; 2737d636676SMatthew Wilcox (Oracle) is_contig = true; 274afc51aaaSDarrick J. Wong } 275afc51aaaSDarrick J. Wong 2767d636676SMatthew Wilcox (Oracle) if (!is_contig || bio_full(ctx->bio, plen)) { 277afc51aaaSDarrick J. Wong gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL); 278457df33eSMatthew Wilcox (Oracle) gfp_t orig_gfp = gfp; 279afc51aaaSDarrick J. Wong int nr_vecs = (length + PAGE_SIZE - 1) >> PAGE_SHIFT; 280afc51aaaSDarrick J. Wong 281afc51aaaSDarrick J. Wong if (ctx->bio) 282afc51aaaSDarrick J. Wong submit_bio(ctx->bio); 283afc51aaaSDarrick J. Wong 2849d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) /* same as readahead_gfp_mask */ 285afc51aaaSDarrick J. Wong gfp |= __GFP_NORETRY | __GFP_NOWARN; 286afc51aaaSDarrick J. Wong ctx->bio = bio_alloc(gfp, min(BIO_MAX_PAGES, nr_vecs)); 287457df33eSMatthew Wilcox (Oracle) /* 288457df33eSMatthew Wilcox (Oracle) * If the bio_alloc fails, try it again for a single page to 289457df33eSMatthew Wilcox (Oracle) * avoid having to deal with partial page reads. This emulates 290457df33eSMatthew Wilcox (Oracle) * what do_mpage_readpage does. 291457df33eSMatthew Wilcox (Oracle) */ 292457df33eSMatthew Wilcox (Oracle) if (!ctx->bio) 293457df33eSMatthew Wilcox (Oracle) ctx->bio = bio_alloc(orig_gfp, 1); 294afc51aaaSDarrick J. Wong ctx->bio->bi_opf = REQ_OP_READ; 2959d24a13aSMatthew Wilcox (Oracle) if (ctx->rac) 296afc51aaaSDarrick J. Wong ctx->bio->bi_opf |= REQ_RAHEAD; 297afc51aaaSDarrick J. Wong ctx->bio->bi_iter.bi_sector = sector; 298afc51aaaSDarrick J. Wong bio_set_dev(ctx->bio, iomap->bdev); 299afc51aaaSDarrick J. Wong ctx->bio->bi_end_io = iomap_read_end_io; 300afc51aaaSDarrick J. Wong } 301afc51aaaSDarrick J. Wong 302afc51aaaSDarrick J. Wong bio_add_page(ctx->bio, page, plen, poff); 303afc51aaaSDarrick J. Wong done: 304afc51aaaSDarrick J. Wong /* 305afc51aaaSDarrick J. Wong * Move the caller beyond our range so that it keeps making progress. 306afc51aaaSDarrick J. Wong * For that we have to include any leading non-uptodate ranges, but 307afc51aaaSDarrick J. Wong * we can skip trailing ones as they will be handled in the next 308afc51aaaSDarrick J. Wong * iteration. 309afc51aaaSDarrick J. Wong */ 310afc51aaaSDarrick J. Wong return pos - orig_pos + plen; 311afc51aaaSDarrick J. Wong } 312afc51aaaSDarrick J. Wong 313afc51aaaSDarrick J. Wong int 314afc51aaaSDarrick J. Wong iomap_readpage(struct page *page, const struct iomap_ops *ops) 315afc51aaaSDarrick J. Wong { 316afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { .cur_page = page }; 317afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 318afc51aaaSDarrick J. Wong unsigned poff; 319afc51aaaSDarrick J. Wong loff_t ret; 320afc51aaaSDarrick J. Wong 3219e91c572SChristoph Hellwig trace_iomap_readpage(page->mapping->host, 1); 3229e91c572SChristoph Hellwig 323afc51aaaSDarrick J. Wong for (poff = 0; poff < PAGE_SIZE; poff += ret) { 324afc51aaaSDarrick J. Wong ret = iomap_apply(inode, page_offset(page) + poff, 325afc51aaaSDarrick J. Wong PAGE_SIZE - poff, 0, ops, &ctx, 326afc51aaaSDarrick J. Wong iomap_readpage_actor); 327afc51aaaSDarrick J. Wong if (ret <= 0) { 328afc51aaaSDarrick J. Wong WARN_ON_ONCE(ret == 0); 329afc51aaaSDarrick J. Wong SetPageError(page); 330afc51aaaSDarrick J. Wong break; 331afc51aaaSDarrick J. Wong } 332afc51aaaSDarrick J. Wong } 333afc51aaaSDarrick J. Wong 334afc51aaaSDarrick J. Wong if (ctx.bio) { 335afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 336afc51aaaSDarrick J. Wong WARN_ON_ONCE(!ctx.cur_page_in_bio); 337afc51aaaSDarrick J. Wong } else { 338afc51aaaSDarrick J. Wong WARN_ON_ONCE(ctx.cur_page_in_bio); 339afc51aaaSDarrick J. Wong unlock_page(page); 340afc51aaaSDarrick J. Wong } 341afc51aaaSDarrick J. Wong 342afc51aaaSDarrick J. Wong /* 343d4388340SMatthew Wilcox (Oracle) * Just like mpage_readahead and block_read_full_page we always 344afc51aaaSDarrick J. Wong * return 0 and just mark the page as PageError on errors. This 345afc51aaaSDarrick J. Wong * should be cleaned up all through the stack eventually. 346afc51aaaSDarrick J. Wong */ 347afc51aaaSDarrick J. Wong return 0; 348afc51aaaSDarrick J. Wong } 349afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_readpage); 350afc51aaaSDarrick J. Wong 351afc51aaaSDarrick J. Wong static loff_t 3529d24a13aSMatthew Wilcox (Oracle) iomap_readahead_actor(struct inode *inode, loff_t pos, loff_t length, 353c039b997SGoldwyn Rodrigues void *data, struct iomap *iomap, struct iomap *srcmap) 354afc51aaaSDarrick J. Wong { 355afc51aaaSDarrick J. Wong struct iomap_readpage_ctx *ctx = data; 356afc51aaaSDarrick J. Wong loff_t done, ret; 357afc51aaaSDarrick J. Wong 358afc51aaaSDarrick J. Wong for (done = 0; done < length; done += ret) { 359afc51aaaSDarrick J. Wong if (ctx->cur_page && offset_in_page(pos + done) == 0) { 360afc51aaaSDarrick J. Wong if (!ctx->cur_page_in_bio) 361afc51aaaSDarrick J. Wong unlock_page(ctx->cur_page); 362afc51aaaSDarrick J. Wong put_page(ctx->cur_page); 363afc51aaaSDarrick J. Wong ctx->cur_page = NULL; 364afc51aaaSDarrick J. Wong } 365afc51aaaSDarrick J. Wong if (!ctx->cur_page) { 3669d24a13aSMatthew Wilcox (Oracle) ctx->cur_page = readahead_page(ctx->rac); 367afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = false; 368afc51aaaSDarrick J. Wong } 369afc51aaaSDarrick J. Wong ret = iomap_readpage_actor(inode, pos + done, length - done, 370c039b997SGoldwyn Rodrigues ctx, iomap, srcmap); 371afc51aaaSDarrick J. Wong } 372afc51aaaSDarrick J. Wong 373afc51aaaSDarrick J. Wong return done; 374afc51aaaSDarrick J. Wong } 375afc51aaaSDarrick J. Wong 3769d24a13aSMatthew Wilcox (Oracle) /** 3779d24a13aSMatthew Wilcox (Oracle) * iomap_readahead - Attempt to read pages from a file. 3789d24a13aSMatthew Wilcox (Oracle) * @rac: Describes the pages to be read. 3799d24a13aSMatthew Wilcox (Oracle) * @ops: The operations vector for the filesystem. 3809d24a13aSMatthew Wilcox (Oracle) * 3819d24a13aSMatthew Wilcox (Oracle) * This function is for filesystems to call to implement their readahead 3829d24a13aSMatthew Wilcox (Oracle) * address_space operation. 3839d24a13aSMatthew Wilcox (Oracle) * 3849d24a13aSMatthew Wilcox (Oracle) * Context: The @ops callbacks may submit I/O (eg to read the addresses of 3859d24a13aSMatthew Wilcox (Oracle) * blocks from disc), and may wait for it. The caller may be trying to 3869d24a13aSMatthew Wilcox (Oracle) * access a different page, and so sleeping excessively should be avoided. 3879d24a13aSMatthew Wilcox (Oracle) * It may allocate memory, but should avoid costly allocations. This 3889d24a13aSMatthew Wilcox (Oracle) * function is called with memalloc_nofs set, so allocations will not cause 3899d24a13aSMatthew Wilcox (Oracle) * the filesystem to be reentered. 3909d24a13aSMatthew Wilcox (Oracle) */ 3919d24a13aSMatthew Wilcox (Oracle) void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops) 392afc51aaaSDarrick J. Wong { 3939d24a13aSMatthew Wilcox (Oracle) struct inode *inode = rac->mapping->host; 3949d24a13aSMatthew Wilcox (Oracle) loff_t pos = readahead_pos(rac); 3959d24a13aSMatthew Wilcox (Oracle) loff_t length = readahead_length(rac); 396afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { 3979d24a13aSMatthew Wilcox (Oracle) .rac = rac, 398afc51aaaSDarrick J. Wong }; 399afc51aaaSDarrick J. Wong 4009d24a13aSMatthew Wilcox (Oracle) trace_iomap_readahead(inode, readahead_count(rac)); 4019e91c572SChristoph Hellwig 402afc51aaaSDarrick J. Wong while (length > 0) { 4039d24a13aSMatthew Wilcox (Oracle) loff_t ret = iomap_apply(inode, pos, length, 0, ops, 4049d24a13aSMatthew Wilcox (Oracle) &ctx, iomap_readahead_actor); 405afc51aaaSDarrick J. Wong if (ret <= 0) { 406afc51aaaSDarrick J. Wong WARN_ON_ONCE(ret == 0); 4079d24a13aSMatthew Wilcox (Oracle) break; 408afc51aaaSDarrick J. Wong } 409afc51aaaSDarrick J. Wong pos += ret; 410afc51aaaSDarrick J. Wong length -= ret; 411afc51aaaSDarrick J. Wong } 4129d24a13aSMatthew Wilcox (Oracle) 413afc51aaaSDarrick J. Wong if (ctx.bio) 414afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 415afc51aaaSDarrick J. Wong if (ctx.cur_page) { 416afc51aaaSDarrick J. Wong if (!ctx.cur_page_in_bio) 417afc51aaaSDarrick J. Wong unlock_page(ctx.cur_page); 418afc51aaaSDarrick J. Wong put_page(ctx.cur_page); 419afc51aaaSDarrick J. Wong } 420afc51aaaSDarrick J. Wong } 4219d24a13aSMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(iomap_readahead); 422afc51aaaSDarrick J. Wong 423afc51aaaSDarrick J. Wong /* 424afc51aaaSDarrick J. Wong * iomap_is_partially_uptodate checks whether blocks within a page are 425afc51aaaSDarrick J. Wong * uptodate or not. 426afc51aaaSDarrick J. Wong * 427afc51aaaSDarrick J. Wong * Returns true if all blocks which correspond to a file portion 428afc51aaaSDarrick J. Wong * we want to read within the page are uptodate. 429afc51aaaSDarrick J. Wong */ 430afc51aaaSDarrick J. Wong int 431afc51aaaSDarrick J. Wong iomap_is_partially_uptodate(struct page *page, unsigned long from, 432afc51aaaSDarrick J. Wong unsigned long count) 433afc51aaaSDarrick J. Wong { 434afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 435afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 436afc51aaaSDarrick J. Wong unsigned len, first, last; 437afc51aaaSDarrick J. Wong unsigned i; 438afc51aaaSDarrick J. Wong 439afc51aaaSDarrick J. Wong /* Limit range to one page */ 440afc51aaaSDarrick J. Wong len = min_t(unsigned, PAGE_SIZE - from, count); 441afc51aaaSDarrick J. Wong 442afc51aaaSDarrick J. Wong /* First and last blocks in range within page */ 443afc51aaaSDarrick J. Wong first = from >> inode->i_blkbits; 444afc51aaaSDarrick J. Wong last = (from + len - 1) >> inode->i_blkbits; 445afc51aaaSDarrick J. Wong 446afc51aaaSDarrick J. Wong if (iop) { 447afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) 448afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 449afc51aaaSDarrick J. Wong return 0; 450afc51aaaSDarrick J. Wong return 1; 451afc51aaaSDarrick J. Wong } 452afc51aaaSDarrick J. Wong 453afc51aaaSDarrick J. Wong return 0; 454afc51aaaSDarrick J. Wong } 455afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate); 456afc51aaaSDarrick J. Wong 457afc51aaaSDarrick J. Wong int 458afc51aaaSDarrick J. Wong iomap_releasepage(struct page *page, gfp_t gfp_mask) 459afc51aaaSDarrick J. Wong { 4601ac99452SMatthew Wilcox (Oracle) trace_iomap_releasepage(page->mapping->host, page_offset(page), 4611ac99452SMatthew Wilcox (Oracle) PAGE_SIZE); 4629e91c572SChristoph Hellwig 463afc51aaaSDarrick J. Wong /* 464afc51aaaSDarrick J. Wong * mm accommodates an old ext3 case where clean pages might not have had 465afc51aaaSDarrick J. Wong * the dirty bit cleared. Thus, it can send actual dirty pages to 466afc51aaaSDarrick J. Wong * ->releasepage() via shrink_active_list(), skip those here. 467afc51aaaSDarrick J. Wong */ 468afc51aaaSDarrick J. Wong if (PageDirty(page) || PageWriteback(page)) 469afc51aaaSDarrick J. Wong return 0; 470afc51aaaSDarrick J. Wong iomap_page_release(page); 471afc51aaaSDarrick J. Wong return 1; 472afc51aaaSDarrick J. Wong } 473afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_releasepage); 474afc51aaaSDarrick J. Wong 475afc51aaaSDarrick J. Wong void 476afc51aaaSDarrick J. Wong iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len) 477afc51aaaSDarrick J. Wong { 4781ac99452SMatthew Wilcox (Oracle) trace_iomap_invalidatepage(page->mapping->host, offset, len); 4799e91c572SChristoph Hellwig 480afc51aaaSDarrick J. Wong /* 481afc51aaaSDarrick J. Wong * If we are invalidating the entire page, clear the dirty state from it 482afc51aaaSDarrick J. Wong * and release it to avoid unnecessary buildup of the LRU. 483afc51aaaSDarrick J. Wong */ 484afc51aaaSDarrick J. Wong if (offset == 0 && len == PAGE_SIZE) { 485afc51aaaSDarrick J. Wong WARN_ON_ONCE(PageWriteback(page)); 486afc51aaaSDarrick J. Wong cancel_dirty_page(page); 487afc51aaaSDarrick J. Wong iomap_page_release(page); 488afc51aaaSDarrick J. Wong } 489afc51aaaSDarrick J. Wong } 490afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_invalidatepage); 491afc51aaaSDarrick J. Wong 492afc51aaaSDarrick J. Wong #ifdef CONFIG_MIGRATION 493afc51aaaSDarrick J. Wong int 494afc51aaaSDarrick J. Wong iomap_migrate_page(struct address_space *mapping, struct page *newpage, 495afc51aaaSDarrick J. Wong struct page *page, enum migrate_mode mode) 496afc51aaaSDarrick J. Wong { 497afc51aaaSDarrick J. Wong int ret; 498afc51aaaSDarrick J. Wong 49926473f83SLinus Torvalds ret = migrate_page_move_mapping(mapping, newpage, page, 0); 500afc51aaaSDarrick J. Wong if (ret != MIGRATEPAGE_SUCCESS) 501afc51aaaSDarrick J. Wong return ret; 502afc51aaaSDarrick J. Wong 50358aeb731SGuoqing Jiang if (page_has_private(page)) 50458aeb731SGuoqing Jiang attach_page_private(newpage, detach_page_private(page)); 505afc51aaaSDarrick J. Wong 506afc51aaaSDarrick J. Wong if (mode != MIGRATE_SYNC_NO_COPY) 507afc51aaaSDarrick J. Wong migrate_page_copy(newpage, page); 508afc51aaaSDarrick J. Wong else 509afc51aaaSDarrick J. Wong migrate_page_states(newpage, page); 510afc51aaaSDarrick J. Wong return MIGRATEPAGE_SUCCESS; 511afc51aaaSDarrick J. Wong } 512afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_migrate_page); 513afc51aaaSDarrick J. Wong #endif /* CONFIG_MIGRATION */ 514afc51aaaSDarrick J. Wong 51532a38a49SChristoph Hellwig enum { 51632a38a49SChristoph Hellwig IOMAP_WRITE_F_UNSHARE = (1 << 0), 51732a38a49SChristoph Hellwig }; 51832a38a49SChristoph Hellwig 519afc51aaaSDarrick J. Wong static void 520afc51aaaSDarrick J. Wong iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) 521afc51aaaSDarrick J. Wong { 522afc51aaaSDarrick J. Wong loff_t i_size = i_size_read(inode); 523afc51aaaSDarrick J. Wong 524afc51aaaSDarrick J. Wong /* 525afc51aaaSDarrick J. Wong * Only truncate newly allocated pages beyoned EOF, even if the 526afc51aaaSDarrick J. Wong * write started inside the existing inode size. 527afc51aaaSDarrick J. Wong */ 528afc51aaaSDarrick J. Wong if (pos + len > i_size) 529afc51aaaSDarrick J. Wong truncate_pagecache_range(inode, max(pos, i_size), pos + len); 530afc51aaaSDarrick J. Wong } 531afc51aaaSDarrick J. Wong 532afc51aaaSDarrick J. Wong static int 533d3b40439SChristoph Hellwig iomap_read_page_sync(loff_t block_start, struct page *page, unsigned poff, 534d3b40439SChristoph Hellwig unsigned plen, struct iomap *iomap) 535afc51aaaSDarrick J. Wong { 536afc51aaaSDarrick J. Wong struct bio_vec bvec; 537afc51aaaSDarrick J. Wong struct bio bio; 538afc51aaaSDarrick J. Wong 539afc51aaaSDarrick J. Wong bio_init(&bio, &bvec, 1); 540afc51aaaSDarrick J. Wong bio.bi_opf = REQ_OP_READ; 541afc51aaaSDarrick J. Wong bio.bi_iter.bi_sector = iomap_sector(iomap, block_start); 542afc51aaaSDarrick J. Wong bio_set_dev(&bio, iomap->bdev); 543afc51aaaSDarrick J. Wong __bio_add_page(&bio, page, plen, poff); 544afc51aaaSDarrick J. Wong return submit_bio_wait(&bio); 545afc51aaaSDarrick J. Wong } 546afc51aaaSDarrick J. Wong 547afc51aaaSDarrick J. Wong static int 54832a38a49SChristoph Hellwig __iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, int flags, 549c039b997SGoldwyn Rodrigues struct page *page, struct iomap *srcmap) 550afc51aaaSDarrick J. Wong { 551afc51aaaSDarrick J. Wong struct iomap_page *iop = iomap_page_create(inode, page); 552afc51aaaSDarrick J. Wong loff_t block_size = i_blocksize(inode); 5536cc19c5fSNikolay Borisov loff_t block_start = round_down(pos, block_size); 5546cc19c5fSNikolay Borisov loff_t block_end = round_up(pos + len, block_size); 555afc51aaaSDarrick J. Wong unsigned from = offset_in_page(pos), to = from + len, poff, plen; 556afc51aaaSDarrick J. Wong 557afc51aaaSDarrick J. Wong if (PageUptodate(page)) 558afc51aaaSDarrick J. Wong return 0; 559e6e7ca92SMatthew Wilcox (Oracle) ClearPageError(page); 560afc51aaaSDarrick J. Wong 561afc51aaaSDarrick J. Wong do { 562afc51aaaSDarrick J. Wong iomap_adjust_read_range(inode, iop, &block_start, 563afc51aaaSDarrick J. Wong block_end - block_start, &poff, &plen); 564afc51aaaSDarrick J. Wong if (plen == 0) 565afc51aaaSDarrick J. Wong break; 566afc51aaaSDarrick J. Wong 56732a38a49SChristoph Hellwig if (!(flags & IOMAP_WRITE_F_UNSHARE) && 56832a38a49SChristoph Hellwig (from <= poff || from >= poff + plen) && 569d3b40439SChristoph Hellwig (to <= poff || to >= poff + plen)) 570d3b40439SChristoph Hellwig continue; 571d3b40439SChristoph Hellwig 572c039b997SGoldwyn Rodrigues if (iomap_block_needs_zeroing(inode, srcmap, block_start)) { 57332a38a49SChristoph Hellwig if (WARN_ON_ONCE(flags & IOMAP_WRITE_F_UNSHARE)) 57432a38a49SChristoph Hellwig return -EIO; 575d3b40439SChristoph Hellwig zero_user_segments(page, poff, from, to, poff + plen); 57614284fedSMatthew Wilcox (Oracle) } else { 57714284fedSMatthew Wilcox (Oracle) int status = iomap_read_page_sync(block_start, page, 57814284fedSMatthew Wilcox (Oracle) poff, plen, srcmap); 579d3b40439SChristoph Hellwig if (status) 580d3b40439SChristoph Hellwig return status; 58114284fedSMatthew Wilcox (Oracle) } 58214284fedSMatthew Wilcox (Oracle) iomap_set_range_uptodate(page, poff, plen); 583afc51aaaSDarrick J. Wong } while ((block_start += plen) < block_end); 584afc51aaaSDarrick J. Wong 585d3b40439SChristoph Hellwig return 0; 586afc51aaaSDarrick J. Wong } 587afc51aaaSDarrick J. Wong 588afc51aaaSDarrick J. Wong static int 589afc51aaaSDarrick J. Wong iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags, 590c039b997SGoldwyn Rodrigues struct page **pagep, struct iomap *iomap, struct iomap *srcmap) 591afc51aaaSDarrick J. Wong { 592afc51aaaSDarrick J. Wong const struct iomap_page_ops *page_ops = iomap->page_ops; 593afc51aaaSDarrick J. Wong struct page *page; 594afc51aaaSDarrick J. Wong int status = 0; 595afc51aaaSDarrick J. Wong 596afc51aaaSDarrick J. Wong BUG_ON(pos + len > iomap->offset + iomap->length); 597c039b997SGoldwyn Rodrigues if (srcmap != iomap) 598c039b997SGoldwyn Rodrigues BUG_ON(pos + len > srcmap->offset + srcmap->length); 599afc51aaaSDarrick J. Wong 600afc51aaaSDarrick J. Wong if (fatal_signal_pending(current)) 601afc51aaaSDarrick J. Wong return -EINTR; 602afc51aaaSDarrick J. Wong 603afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_prepare) { 604afc51aaaSDarrick J. Wong status = page_ops->page_prepare(inode, pos, len, iomap); 605afc51aaaSDarrick J. Wong if (status) 606afc51aaaSDarrick J. Wong return status; 607afc51aaaSDarrick J. Wong } 608afc51aaaSDarrick J. Wong 609dcd6158dSChristoph Hellwig page = grab_cache_page_write_begin(inode->i_mapping, pos >> PAGE_SHIFT, 610dcd6158dSChristoph Hellwig AOP_FLAG_NOFS); 611afc51aaaSDarrick J. Wong if (!page) { 612afc51aaaSDarrick J. Wong status = -ENOMEM; 613afc51aaaSDarrick J. Wong goto out_no_page; 614afc51aaaSDarrick J. Wong } 615afc51aaaSDarrick J. Wong 616c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) 617c039b997SGoldwyn Rodrigues iomap_read_inline_data(inode, page, srcmap); 618afc51aaaSDarrick J. Wong else if (iomap->flags & IOMAP_F_BUFFER_HEAD) 619c039b997SGoldwyn Rodrigues status = __block_write_begin_int(page, pos, len, NULL, srcmap); 620afc51aaaSDarrick J. Wong else 62132a38a49SChristoph Hellwig status = __iomap_write_begin(inode, pos, len, flags, page, 622c039b997SGoldwyn Rodrigues srcmap); 623afc51aaaSDarrick J. Wong 624afc51aaaSDarrick J. Wong if (unlikely(status)) 625afc51aaaSDarrick J. Wong goto out_unlock; 626afc51aaaSDarrick J. Wong 627afc51aaaSDarrick J. Wong *pagep = page; 628afc51aaaSDarrick J. Wong return 0; 629afc51aaaSDarrick J. Wong 630afc51aaaSDarrick J. Wong out_unlock: 631afc51aaaSDarrick J. Wong unlock_page(page); 632afc51aaaSDarrick J. Wong put_page(page); 633afc51aaaSDarrick J. Wong iomap_write_failed(inode, pos, len); 634afc51aaaSDarrick J. Wong 635afc51aaaSDarrick J. Wong out_no_page: 636afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 637afc51aaaSDarrick J. Wong page_ops->page_done(inode, pos, 0, NULL, iomap); 638afc51aaaSDarrick J. Wong return status; 639afc51aaaSDarrick J. Wong } 640afc51aaaSDarrick J. Wong 641afc51aaaSDarrick J. Wong int 642afc51aaaSDarrick J. Wong iomap_set_page_dirty(struct page *page) 643afc51aaaSDarrick J. Wong { 644afc51aaaSDarrick J. Wong struct address_space *mapping = page_mapping(page); 645afc51aaaSDarrick J. Wong int newly_dirty; 646afc51aaaSDarrick J. Wong 647afc51aaaSDarrick J. Wong if (unlikely(!mapping)) 648afc51aaaSDarrick J. Wong return !TestSetPageDirty(page); 649afc51aaaSDarrick J. Wong 650afc51aaaSDarrick J. Wong /* 651afc51aaaSDarrick J. Wong * Lock out page->mem_cgroup migration to keep PageDirty 652afc51aaaSDarrick J. Wong * synchronized with per-memcg dirty page counters. 653afc51aaaSDarrick J. Wong */ 654afc51aaaSDarrick J. Wong lock_page_memcg(page); 655afc51aaaSDarrick J. Wong newly_dirty = !TestSetPageDirty(page); 656afc51aaaSDarrick J. Wong if (newly_dirty) 657afc51aaaSDarrick J. Wong __set_page_dirty(page, mapping, 0); 658afc51aaaSDarrick J. Wong unlock_page_memcg(page); 659afc51aaaSDarrick J. Wong 660afc51aaaSDarrick J. Wong if (newly_dirty) 661afc51aaaSDarrick J. Wong __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 662afc51aaaSDarrick J. Wong return newly_dirty; 663afc51aaaSDarrick J. Wong } 664afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_set_page_dirty); 665afc51aaaSDarrick J. Wong 666afc51aaaSDarrick J. Wong static int 667afc51aaaSDarrick J. Wong __iomap_write_end(struct inode *inode, loff_t pos, unsigned len, 668c12d6fa8SChristoph Hellwig unsigned copied, struct page *page) 669afc51aaaSDarrick J. Wong { 670afc51aaaSDarrick J. Wong flush_dcache_page(page); 671afc51aaaSDarrick J. Wong 672afc51aaaSDarrick J. Wong /* 673afc51aaaSDarrick J. Wong * The blocks that were entirely written will now be uptodate, so we 674afc51aaaSDarrick J. Wong * don't have to worry about a readpage reading them and overwriting a 675afc51aaaSDarrick J. Wong * partial write. However if we have encountered a short write and only 676afc51aaaSDarrick J. Wong * partially written into a block, it will not be marked uptodate, so a 677afc51aaaSDarrick J. Wong * readpage might come in and destroy our partial write. 678afc51aaaSDarrick J. Wong * 679afc51aaaSDarrick J. Wong * Do the simplest thing, and just treat any short write to a non 680afc51aaaSDarrick J. Wong * uptodate page as a zero-length write, and force the caller to redo 681afc51aaaSDarrick J. Wong * the whole thing. 682afc51aaaSDarrick J. Wong */ 683afc51aaaSDarrick J. Wong if (unlikely(copied < len && !PageUptodate(page))) 684afc51aaaSDarrick J. Wong return 0; 685afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, offset_in_page(pos), len); 686afc51aaaSDarrick J. Wong iomap_set_page_dirty(page); 687afc51aaaSDarrick J. Wong return copied; 688afc51aaaSDarrick J. Wong } 689afc51aaaSDarrick J. Wong 690afc51aaaSDarrick J. Wong static int 691afc51aaaSDarrick J. Wong iomap_write_end_inline(struct inode *inode, struct page *page, 692afc51aaaSDarrick J. Wong struct iomap *iomap, loff_t pos, unsigned copied) 693afc51aaaSDarrick J. Wong { 694afc51aaaSDarrick J. Wong void *addr; 695afc51aaaSDarrick J. Wong 696afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 697afc51aaaSDarrick J. Wong BUG_ON(pos + copied > PAGE_SIZE - offset_in_page(iomap->inline_data)); 698afc51aaaSDarrick J. Wong 6997ed3cd1aSMatthew Wilcox (Oracle) flush_dcache_page(page); 700afc51aaaSDarrick J. Wong addr = kmap_atomic(page); 701afc51aaaSDarrick J. Wong memcpy(iomap->inline_data + pos, addr + pos, copied); 702afc51aaaSDarrick J. Wong kunmap_atomic(addr); 703afc51aaaSDarrick J. Wong 704afc51aaaSDarrick J. Wong mark_inode_dirty(inode); 705afc51aaaSDarrick J. Wong return copied; 706afc51aaaSDarrick J. Wong } 707afc51aaaSDarrick J. Wong 708afc51aaaSDarrick J. Wong static int 709c039b997SGoldwyn Rodrigues iomap_write_end(struct inode *inode, loff_t pos, unsigned len, unsigned copied, 710c039b997SGoldwyn Rodrigues struct page *page, struct iomap *iomap, struct iomap *srcmap) 711afc51aaaSDarrick J. Wong { 712afc51aaaSDarrick J. Wong const struct iomap_page_ops *page_ops = iomap->page_ops; 713afc51aaaSDarrick J. Wong loff_t old_size = inode->i_size; 714afc51aaaSDarrick J. Wong int ret; 715afc51aaaSDarrick J. Wong 716c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) { 717afc51aaaSDarrick J. Wong ret = iomap_write_end_inline(inode, page, iomap, pos, copied); 718c039b997SGoldwyn Rodrigues } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) { 719afc51aaaSDarrick J. Wong ret = block_write_end(NULL, inode->i_mapping, pos, len, copied, 720afc51aaaSDarrick J. Wong page, NULL); 721afc51aaaSDarrick J. Wong } else { 722c12d6fa8SChristoph Hellwig ret = __iomap_write_end(inode, pos, len, copied, page); 723afc51aaaSDarrick J. Wong } 724afc51aaaSDarrick J. Wong 725afc51aaaSDarrick J. Wong /* 726afc51aaaSDarrick J. Wong * Update the in-memory inode size after copying the data into the page 727afc51aaaSDarrick J. Wong * cache. It's up to the file system to write the updated size to disk, 728afc51aaaSDarrick J. Wong * preferably after I/O completion so that no stale data is exposed. 729afc51aaaSDarrick J. Wong */ 730afc51aaaSDarrick J. Wong if (pos + ret > old_size) { 731afc51aaaSDarrick J. Wong i_size_write(inode, pos + ret); 732afc51aaaSDarrick J. Wong iomap->flags |= IOMAP_F_SIZE_CHANGED; 733afc51aaaSDarrick J. Wong } 734afc51aaaSDarrick J. Wong unlock_page(page); 735afc51aaaSDarrick J. Wong 736afc51aaaSDarrick J. Wong if (old_size < pos) 737afc51aaaSDarrick J. Wong pagecache_isize_extended(inode, old_size, pos); 738afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 739afc51aaaSDarrick J. Wong page_ops->page_done(inode, pos, ret, page, iomap); 740afc51aaaSDarrick J. Wong put_page(page); 741afc51aaaSDarrick J. Wong 742afc51aaaSDarrick J. Wong if (ret < len) 743afc51aaaSDarrick J. Wong iomap_write_failed(inode, pos, len); 744afc51aaaSDarrick J. Wong return ret; 745afc51aaaSDarrick J. Wong } 746afc51aaaSDarrick J. Wong 747afc51aaaSDarrick J. Wong static loff_t 748afc51aaaSDarrick J. Wong iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 749c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 750afc51aaaSDarrick J. Wong { 751afc51aaaSDarrick J. Wong struct iov_iter *i = data; 752afc51aaaSDarrick J. Wong long status = 0; 753afc51aaaSDarrick J. Wong ssize_t written = 0; 754afc51aaaSDarrick J. Wong 755afc51aaaSDarrick J. Wong do { 756afc51aaaSDarrick J. Wong struct page *page; 757afc51aaaSDarrick J. Wong unsigned long offset; /* Offset into pagecache page */ 758afc51aaaSDarrick J. Wong unsigned long bytes; /* Bytes to write to page */ 759afc51aaaSDarrick J. Wong size_t copied; /* Bytes copied from user */ 760afc51aaaSDarrick J. Wong 761afc51aaaSDarrick J. Wong offset = offset_in_page(pos); 762afc51aaaSDarrick J. Wong bytes = min_t(unsigned long, PAGE_SIZE - offset, 763afc51aaaSDarrick J. Wong iov_iter_count(i)); 764afc51aaaSDarrick J. Wong again: 765afc51aaaSDarrick J. Wong if (bytes > length) 766afc51aaaSDarrick J. Wong bytes = length; 767afc51aaaSDarrick J. Wong 768afc51aaaSDarrick J. Wong /* 769afc51aaaSDarrick J. Wong * Bring in the user page that we will copy from _first_. 770afc51aaaSDarrick J. Wong * Otherwise there's a nasty deadlock on copying from the 771afc51aaaSDarrick J. Wong * same page as we're writing to, without it being marked 772afc51aaaSDarrick J. Wong * up-to-date. 773afc51aaaSDarrick J. Wong * 774afc51aaaSDarrick J. Wong * Not only is this an optimisation, but it is also required 775afc51aaaSDarrick J. Wong * to check that the address is actually valid, when atomic 776afc51aaaSDarrick J. Wong * usercopies are used, below. 777afc51aaaSDarrick J. Wong */ 778afc51aaaSDarrick J. Wong if (unlikely(iov_iter_fault_in_readable(i, bytes))) { 779afc51aaaSDarrick J. Wong status = -EFAULT; 780afc51aaaSDarrick J. Wong break; 781afc51aaaSDarrick J. Wong } 782afc51aaaSDarrick J. Wong 783c039b997SGoldwyn Rodrigues status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, 784c039b997SGoldwyn Rodrigues srcmap); 785afc51aaaSDarrick J. Wong if (unlikely(status)) 786afc51aaaSDarrick J. Wong break; 787afc51aaaSDarrick J. Wong 788afc51aaaSDarrick J. Wong if (mapping_writably_mapped(inode->i_mapping)) 789afc51aaaSDarrick J. Wong flush_dcache_page(page); 790afc51aaaSDarrick J. Wong 791afc51aaaSDarrick J. Wong copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); 792afc51aaaSDarrick J. Wong 793c039b997SGoldwyn Rodrigues status = iomap_write_end(inode, pos, bytes, copied, page, iomap, 794c039b997SGoldwyn Rodrigues srcmap); 795afc51aaaSDarrick J. Wong if (unlikely(status < 0)) 796afc51aaaSDarrick J. Wong break; 797afc51aaaSDarrick J. Wong copied = status; 798afc51aaaSDarrick J. Wong 799afc51aaaSDarrick J. Wong cond_resched(); 800afc51aaaSDarrick J. Wong 801afc51aaaSDarrick J. Wong iov_iter_advance(i, copied); 802afc51aaaSDarrick J. Wong if (unlikely(copied == 0)) { 803afc51aaaSDarrick J. Wong /* 804afc51aaaSDarrick J. Wong * If we were unable to copy any data at all, we must 805afc51aaaSDarrick J. Wong * fall back to a single segment length write. 806afc51aaaSDarrick J. Wong * 807afc51aaaSDarrick J. Wong * If we didn't fallback here, we could livelock 808afc51aaaSDarrick J. Wong * because not all segments in the iov can be copied at 809afc51aaaSDarrick J. Wong * once without a pagefault. 810afc51aaaSDarrick J. Wong */ 811afc51aaaSDarrick J. Wong bytes = min_t(unsigned long, PAGE_SIZE - offset, 812afc51aaaSDarrick J. Wong iov_iter_single_seg_count(i)); 813afc51aaaSDarrick J. Wong goto again; 814afc51aaaSDarrick J. Wong } 815afc51aaaSDarrick J. Wong pos += copied; 816afc51aaaSDarrick J. Wong written += copied; 817afc51aaaSDarrick J. Wong length -= copied; 818afc51aaaSDarrick J. Wong 819afc51aaaSDarrick J. Wong balance_dirty_pages_ratelimited(inode->i_mapping); 820afc51aaaSDarrick J. Wong } while (iov_iter_count(i) && length); 821afc51aaaSDarrick J. Wong 822afc51aaaSDarrick J. Wong return written ? written : status; 823afc51aaaSDarrick J. Wong } 824afc51aaaSDarrick J. Wong 825afc51aaaSDarrick J. Wong ssize_t 826afc51aaaSDarrick J. Wong iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter, 827afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 828afc51aaaSDarrick J. Wong { 829afc51aaaSDarrick J. Wong struct inode *inode = iocb->ki_filp->f_mapping->host; 830afc51aaaSDarrick J. Wong loff_t pos = iocb->ki_pos, ret = 0, written = 0; 831afc51aaaSDarrick J. Wong 832afc51aaaSDarrick J. Wong while (iov_iter_count(iter)) { 833afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, iov_iter_count(iter), 834afc51aaaSDarrick J. Wong IOMAP_WRITE, ops, iter, iomap_write_actor); 835afc51aaaSDarrick J. Wong if (ret <= 0) 836afc51aaaSDarrick J. Wong break; 837afc51aaaSDarrick J. Wong pos += ret; 838afc51aaaSDarrick J. Wong written += ret; 839afc51aaaSDarrick J. Wong } 840afc51aaaSDarrick J. Wong 841afc51aaaSDarrick J. Wong return written ? written : ret; 842afc51aaaSDarrick J. Wong } 843afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_file_buffered_write); 844afc51aaaSDarrick J. Wong 845afc51aaaSDarrick J. Wong static loff_t 8463590c4d8SChristoph Hellwig iomap_unshare_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 847c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 848afc51aaaSDarrick J. Wong { 849afc51aaaSDarrick J. Wong long status = 0; 850d4ff3b2eSMatthew Wilcox (Oracle) loff_t written = 0; 851afc51aaaSDarrick J. Wong 8523590c4d8SChristoph Hellwig /* don't bother with blocks that are not shared to start with */ 8533590c4d8SChristoph Hellwig if (!(iomap->flags & IOMAP_F_SHARED)) 8543590c4d8SChristoph Hellwig return length; 8553590c4d8SChristoph Hellwig /* don't bother with holes or unwritten extents */ 856c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 8573590c4d8SChristoph Hellwig return length; 8583590c4d8SChristoph Hellwig 859afc51aaaSDarrick J. Wong do { 86032a38a49SChristoph Hellwig unsigned long offset = offset_in_page(pos); 86132a38a49SChristoph Hellwig unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length); 86232a38a49SChristoph Hellwig struct page *page; 863afc51aaaSDarrick J. Wong 86432a38a49SChristoph Hellwig status = iomap_write_begin(inode, pos, bytes, 865c039b997SGoldwyn Rodrigues IOMAP_WRITE_F_UNSHARE, &page, iomap, srcmap); 866afc51aaaSDarrick J. Wong if (unlikely(status)) 867afc51aaaSDarrick J. Wong return status; 868afc51aaaSDarrick J. Wong 869c039b997SGoldwyn Rodrigues status = iomap_write_end(inode, pos, bytes, bytes, page, iomap, 870c039b997SGoldwyn Rodrigues srcmap); 871afc51aaaSDarrick J. Wong if (unlikely(status <= 0)) { 872afc51aaaSDarrick J. Wong if (WARN_ON_ONCE(status == 0)) 873afc51aaaSDarrick J. Wong return -EIO; 874afc51aaaSDarrick J. Wong return status; 875afc51aaaSDarrick J. Wong } 876afc51aaaSDarrick J. Wong 877afc51aaaSDarrick J. Wong cond_resched(); 878afc51aaaSDarrick J. Wong 879afc51aaaSDarrick J. Wong pos += status; 880afc51aaaSDarrick J. Wong written += status; 881afc51aaaSDarrick J. Wong length -= status; 882afc51aaaSDarrick J. Wong 883afc51aaaSDarrick J. Wong balance_dirty_pages_ratelimited(inode->i_mapping); 884afc51aaaSDarrick J. Wong } while (length); 885afc51aaaSDarrick J. Wong 886afc51aaaSDarrick J. Wong return written; 887afc51aaaSDarrick J. Wong } 888afc51aaaSDarrick J. Wong 889afc51aaaSDarrick J. Wong int 8903590c4d8SChristoph Hellwig iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, 891afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 892afc51aaaSDarrick J. Wong { 893afc51aaaSDarrick J. Wong loff_t ret; 894afc51aaaSDarrick J. Wong 895afc51aaaSDarrick J. Wong while (len) { 896afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL, 8973590c4d8SChristoph Hellwig iomap_unshare_actor); 898afc51aaaSDarrick J. Wong if (ret <= 0) 899afc51aaaSDarrick J. Wong return ret; 900afc51aaaSDarrick J. Wong pos += ret; 901afc51aaaSDarrick J. Wong len -= ret; 902afc51aaaSDarrick J. Wong } 903afc51aaaSDarrick J. Wong 904afc51aaaSDarrick J. Wong return 0; 905afc51aaaSDarrick J. Wong } 9063590c4d8SChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_file_unshare); 907afc51aaaSDarrick J. Wong 908afc51aaaSDarrick J. Wong static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset, 909c039b997SGoldwyn Rodrigues unsigned bytes, struct iomap *iomap, struct iomap *srcmap) 910afc51aaaSDarrick J. Wong { 911afc51aaaSDarrick J. Wong struct page *page; 912afc51aaaSDarrick J. Wong int status; 913afc51aaaSDarrick J. Wong 914c039b997SGoldwyn Rodrigues status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, srcmap); 915afc51aaaSDarrick J. Wong if (status) 916afc51aaaSDarrick J. Wong return status; 917afc51aaaSDarrick J. Wong 918afc51aaaSDarrick J. Wong zero_user(page, offset, bytes); 919afc51aaaSDarrick J. Wong mark_page_accessed(page); 920afc51aaaSDarrick J. Wong 921c039b997SGoldwyn Rodrigues return iomap_write_end(inode, pos, bytes, bytes, page, iomap, srcmap); 922afc51aaaSDarrick J. Wong } 923afc51aaaSDarrick J. Wong 924afc51aaaSDarrick J. Wong static loff_t 925afc51aaaSDarrick J. Wong iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count, 926c039b997SGoldwyn Rodrigues void *data, struct iomap *iomap, struct iomap *srcmap) 927afc51aaaSDarrick J. Wong { 928afc51aaaSDarrick J. Wong bool *did_zero = data; 929afc51aaaSDarrick J. Wong loff_t written = 0; 930afc51aaaSDarrick J. Wong int status; 931afc51aaaSDarrick J. Wong 932afc51aaaSDarrick J. Wong /* already zeroed? we're done. */ 933c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 934afc51aaaSDarrick J. Wong return count; 935afc51aaaSDarrick J. Wong 936afc51aaaSDarrick J. Wong do { 937afc51aaaSDarrick J. Wong unsigned offset, bytes; 938afc51aaaSDarrick J. Wong 939afc51aaaSDarrick J. Wong offset = offset_in_page(pos); 940afc51aaaSDarrick J. Wong bytes = min_t(loff_t, PAGE_SIZE - offset, count); 941afc51aaaSDarrick J. Wong 942afc51aaaSDarrick J. Wong if (IS_DAX(inode)) 9434f3b4f16SVivek Goyal status = dax_iomap_zero(pos, offset, bytes, iomap); 944afc51aaaSDarrick J. Wong else 945c039b997SGoldwyn Rodrigues status = iomap_zero(inode, pos, offset, bytes, iomap, 946c039b997SGoldwyn Rodrigues srcmap); 947afc51aaaSDarrick J. Wong if (status < 0) 948afc51aaaSDarrick J. Wong return status; 949afc51aaaSDarrick J. Wong 950afc51aaaSDarrick J. Wong pos += bytes; 951afc51aaaSDarrick J. Wong count -= bytes; 952afc51aaaSDarrick J. Wong written += bytes; 953afc51aaaSDarrick J. Wong if (did_zero) 954afc51aaaSDarrick J. Wong *did_zero = true; 955afc51aaaSDarrick J. Wong } while (count > 0); 956afc51aaaSDarrick J. Wong 957afc51aaaSDarrick J. Wong return written; 958afc51aaaSDarrick J. Wong } 959afc51aaaSDarrick J. Wong 960afc51aaaSDarrick J. Wong int 961afc51aaaSDarrick J. Wong iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, 962afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 963afc51aaaSDarrick J. Wong { 964afc51aaaSDarrick J. Wong loff_t ret; 965afc51aaaSDarrick J. Wong 966afc51aaaSDarrick J. Wong while (len > 0) { 967afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, len, IOMAP_ZERO, 968afc51aaaSDarrick J. Wong ops, did_zero, iomap_zero_range_actor); 969afc51aaaSDarrick J. Wong if (ret <= 0) 970afc51aaaSDarrick J. Wong return ret; 971afc51aaaSDarrick J. Wong 972afc51aaaSDarrick J. Wong pos += ret; 973afc51aaaSDarrick J. Wong len -= ret; 974afc51aaaSDarrick J. Wong } 975afc51aaaSDarrick J. Wong 976afc51aaaSDarrick J. Wong return 0; 977afc51aaaSDarrick J. Wong } 978afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_zero_range); 979afc51aaaSDarrick J. Wong 980afc51aaaSDarrick J. Wong int 981afc51aaaSDarrick J. Wong iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, 982afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 983afc51aaaSDarrick J. Wong { 984afc51aaaSDarrick J. Wong unsigned int blocksize = i_blocksize(inode); 985afc51aaaSDarrick J. Wong unsigned int off = pos & (blocksize - 1); 986afc51aaaSDarrick J. Wong 987afc51aaaSDarrick J. Wong /* Block boundary? Nothing to do */ 988afc51aaaSDarrick J. Wong if (!off) 989afc51aaaSDarrick J. Wong return 0; 990afc51aaaSDarrick J. Wong return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); 991afc51aaaSDarrick J. Wong } 992afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_truncate_page); 993afc51aaaSDarrick J. Wong 994afc51aaaSDarrick J. Wong static loff_t 995afc51aaaSDarrick J. Wong iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length, 996c039b997SGoldwyn Rodrigues void *data, struct iomap *iomap, struct iomap *srcmap) 997afc51aaaSDarrick J. Wong { 998afc51aaaSDarrick J. Wong struct page *page = data; 999afc51aaaSDarrick J. Wong int ret; 1000afc51aaaSDarrick J. Wong 1001afc51aaaSDarrick J. Wong if (iomap->flags & IOMAP_F_BUFFER_HEAD) { 1002afc51aaaSDarrick J. Wong ret = __block_write_begin_int(page, pos, length, NULL, iomap); 1003afc51aaaSDarrick J. Wong if (ret) 1004afc51aaaSDarrick J. Wong return ret; 1005afc51aaaSDarrick J. Wong block_commit_write(page, 0, length); 1006afc51aaaSDarrick J. Wong } else { 1007afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 1008afc51aaaSDarrick J. Wong iomap_page_create(inode, page); 1009afc51aaaSDarrick J. Wong set_page_dirty(page); 1010afc51aaaSDarrick J. Wong } 1011afc51aaaSDarrick J. Wong 1012afc51aaaSDarrick J. Wong return length; 1013afc51aaaSDarrick J. Wong } 1014afc51aaaSDarrick J. Wong 1015afc51aaaSDarrick J. Wong vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) 1016afc51aaaSDarrick J. Wong { 1017afc51aaaSDarrick J. Wong struct page *page = vmf->page; 1018afc51aaaSDarrick J. Wong struct inode *inode = file_inode(vmf->vma->vm_file); 1019afc51aaaSDarrick J. Wong unsigned long length; 1020243145bcSAndreas Gruenbacher loff_t offset; 1021afc51aaaSDarrick J. Wong ssize_t ret; 1022afc51aaaSDarrick J. Wong 1023afc51aaaSDarrick J. Wong lock_page(page); 1024243145bcSAndreas Gruenbacher ret = page_mkwrite_check_truncate(page, inode); 1025243145bcSAndreas Gruenbacher if (ret < 0) 1026afc51aaaSDarrick J. Wong goto out_unlock; 1027243145bcSAndreas Gruenbacher length = ret; 1028afc51aaaSDarrick J. Wong 1029243145bcSAndreas Gruenbacher offset = page_offset(page); 1030afc51aaaSDarrick J. Wong while (length > 0) { 1031afc51aaaSDarrick J. Wong ret = iomap_apply(inode, offset, length, 1032afc51aaaSDarrick J. Wong IOMAP_WRITE | IOMAP_FAULT, ops, page, 1033afc51aaaSDarrick J. Wong iomap_page_mkwrite_actor); 1034afc51aaaSDarrick J. Wong if (unlikely(ret <= 0)) 1035afc51aaaSDarrick J. Wong goto out_unlock; 1036afc51aaaSDarrick J. Wong offset += ret; 1037afc51aaaSDarrick J. Wong length -= ret; 1038afc51aaaSDarrick J. Wong } 1039afc51aaaSDarrick J. Wong 1040afc51aaaSDarrick J. Wong wait_for_stable_page(page); 1041afc51aaaSDarrick J. Wong return VM_FAULT_LOCKED; 1042afc51aaaSDarrick J. Wong out_unlock: 1043afc51aaaSDarrick J. Wong unlock_page(page); 1044afc51aaaSDarrick J. Wong return block_page_mkwrite_return(ret); 1045afc51aaaSDarrick J. Wong } 1046afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_page_mkwrite); 1047598ecfbaSChristoph Hellwig 1048598ecfbaSChristoph Hellwig static void 104948d64cd1SChristoph Hellwig iomap_finish_page_writeback(struct inode *inode, struct page *page, 1050*0fb2d720SMatthew Wilcox (Oracle) int error, unsigned int len) 1051598ecfbaSChristoph Hellwig { 105248d64cd1SChristoph Hellwig struct iomap_page *iop = to_iomap_page(page); 1053598ecfbaSChristoph Hellwig 1054598ecfbaSChristoph Hellwig if (error) { 105548d64cd1SChristoph Hellwig SetPageError(page); 1056598ecfbaSChristoph Hellwig mapping_set_error(inode->i_mapping, -EIO); 1057598ecfbaSChristoph Hellwig } 1058598ecfbaSChristoph Hellwig 105924addd84SMatthew Wilcox (Oracle) WARN_ON_ONCE(i_blocks_per_page(inode, page) > 1 && !iop); 1060*0fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) <= 0); 1061598ecfbaSChristoph Hellwig 1062*0fb2d720SMatthew Wilcox (Oracle) if (!iop || atomic_sub_and_test(len, &iop->write_bytes_pending)) 106348d64cd1SChristoph Hellwig end_page_writeback(page); 1064598ecfbaSChristoph Hellwig } 1065598ecfbaSChristoph Hellwig 1066598ecfbaSChristoph Hellwig /* 1067598ecfbaSChristoph Hellwig * We're now finished for good with this ioend structure. Update the page 1068598ecfbaSChristoph Hellwig * state, release holds on bios, and finally free up memory. Do not use the 1069598ecfbaSChristoph Hellwig * ioend after this. 1070598ecfbaSChristoph Hellwig */ 1071598ecfbaSChristoph Hellwig static void 1072598ecfbaSChristoph Hellwig iomap_finish_ioend(struct iomap_ioend *ioend, int error) 1073598ecfbaSChristoph Hellwig { 1074598ecfbaSChristoph Hellwig struct inode *inode = ioend->io_inode; 1075598ecfbaSChristoph Hellwig struct bio *bio = &ioend->io_inline_bio; 1076598ecfbaSChristoph Hellwig struct bio *last = ioend->io_bio, *next; 1077598ecfbaSChristoph Hellwig u64 start = bio->bi_iter.bi_sector; 1078c275779fSZorro Lang loff_t offset = ioend->io_offset; 1079598ecfbaSChristoph Hellwig bool quiet = bio_flagged(bio, BIO_QUIET); 1080598ecfbaSChristoph Hellwig 1081598ecfbaSChristoph Hellwig for (bio = &ioend->io_inline_bio; bio; bio = next) { 1082598ecfbaSChristoph Hellwig struct bio_vec *bv; 1083598ecfbaSChristoph Hellwig struct bvec_iter_all iter_all; 1084598ecfbaSChristoph Hellwig 1085598ecfbaSChristoph Hellwig /* 1086598ecfbaSChristoph Hellwig * For the last bio, bi_private points to the ioend, so we 1087598ecfbaSChristoph Hellwig * need to explicitly end the iteration here. 1088598ecfbaSChristoph Hellwig */ 1089598ecfbaSChristoph Hellwig if (bio == last) 1090598ecfbaSChristoph Hellwig next = NULL; 1091598ecfbaSChristoph Hellwig else 1092598ecfbaSChristoph Hellwig next = bio->bi_private; 1093598ecfbaSChristoph Hellwig 1094598ecfbaSChristoph Hellwig /* walk each page on bio, ending page IO on them */ 1095598ecfbaSChristoph Hellwig bio_for_each_segment_all(bv, bio, iter_all) 1096*0fb2d720SMatthew Wilcox (Oracle) iomap_finish_page_writeback(inode, bv->bv_page, error, 1097*0fb2d720SMatthew Wilcox (Oracle) bv->bv_len); 1098598ecfbaSChristoph Hellwig bio_put(bio); 1099598ecfbaSChristoph Hellwig } 1100c275779fSZorro Lang /* The ioend has been freed by bio_put() */ 1101598ecfbaSChristoph Hellwig 1102598ecfbaSChristoph Hellwig if (unlikely(error && !quiet)) { 1103598ecfbaSChristoph Hellwig printk_ratelimited(KERN_ERR 11049cd0ed63SDarrick J. Wong "%s: writeback error on inode %lu, offset %lld, sector %llu", 1105c275779fSZorro Lang inode->i_sb->s_id, inode->i_ino, offset, start); 1106598ecfbaSChristoph Hellwig } 1107598ecfbaSChristoph Hellwig } 1108598ecfbaSChristoph Hellwig 1109598ecfbaSChristoph Hellwig void 1110598ecfbaSChristoph Hellwig iomap_finish_ioends(struct iomap_ioend *ioend, int error) 1111598ecfbaSChristoph Hellwig { 1112598ecfbaSChristoph Hellwig struct list_head tmp; 1113598ecfbaSChristoph Hellwig 1114598ecfbaSChristoph Hellwig list_replace_init(&ioend->io_list, &tmp); 1115598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1116598ecfbaSChristoph Hellwig 1117598ecfbaSChristoph Hellwig while (!list_empty(&tmp)) { 1118598ecfbaSChristoph Hellwig ioend = list_first_entry(&tmp, struct iomap_ioend, io_list); 1119598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1120598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1121598ecfbaSChristoph Hellwig } 1122598ecfbaSChristoph Hellwig } 1123598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_finish_ioends); 1124598ecfbaSChristoph Hellwig 1125598ecfbaSChristoph Hellwig /* 1126598ecfbaSChristoph Hellwig * We can merge two adjacent ioends if they have the same set of work to do. 1127598ecfbaSChristoph Hellwig */ 1128598ecfbaSChristoph Hellwig static bool 1129598ecfbaSChristoph Hellwig iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next) 1130598ecfbaSChristoph Hellwig { 1131598ecfbaSChristoph Hellwig if (ioend->io_bio->bi_status != next->io_bio->bi_status) 1132598ecfbaSChristoph Hellwig return false; 1133598ecfbaSChristoph Hellwig if ((ioend->io_flags & IOMAP_F_SHARED) ^ 1134598ecfbaSChristoph Hellwig (next->io_flags & IOMAP_F_SHARED)) 1135598ecfbaSChristoph Hellwig return false; 1136598ecfbaSChristoph Hellwig if ((ioend->io_type == IOMAP_UNWRITTEN) ^ 1137598ecfbaSChristoph Hellwig (next->io_type == IOMAP_UNWRITTEN)) 1138598ecfbaSChristoph Hellwig return false; 1139598ecfbaSChristoph Hellwig if (ioend->io_offset + ioend->io_size != next->io_offset) 1140598ecfbaSChristoph Hellwig return false; 1141598ecfbaSChristoph Hellwig return true; 1142598ecfbaSChristoph Hellwig } 1143598ecfbaSChristoph Hellwig 1144598ecfbaSChristoph Hellwig void 1145598ecfbaSChristoph Hellwig iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends, 1146598ecfbaSChristoph Hellwig void (*merge_private)(struct iomap_ioend *ioend, 1147598ecfbaSChristoph Hellwig struct iomap_ioend *next)) 1148598ecfbaSChristoph Hellwig { 1149598ecfbaSChristoph Hellwig struct iomap_ioend *next; 1150598ecfbaSChristoph Hellwig 1151598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1152598ecfbaSChristoph Hellwig 1153598ecfbaSChristoph Hellwig while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend, 1154598ecfbaSChristoph Hellwig io_list))) { 1155598ecfbaSChristoph Hellwig if (!iomap_ioend_can_merge(ioend, next)) 1156598ecfbaSChristoph Hellwig break; 1157598ecfbaSChristoph Hellwig list_move_tail(&next->io_list, &ioend->io_list); 1158598ecfbaSChristoph Hellwig ioend->io_size += next->io_size; 1159598ecfbaSChristoph Hellwig if (next->io_private && merge_private) 1160598ecfbaSChristoph Hellwig merge_private(ioend, next); 1161598ecfbaSChristoph Hellwig } 1162598ecfbaSChristoph Hellwig } 1163598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_ioend_try_merge); 1164598ecfbaSChristoph Hellwig 1165598ecfbaSChristoph Hellwig static int 1166598ecfbaSChristoph Hellwig iomap_ioend_compare(void *priv, struct list_head *a, struct list_head *b) 1167598ecfbaSChristoph Hellwig { 1168b3d423ecSChristoph Hellwig struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list); 1169b3d423ecSChristoph Hellwig struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list); 1170598ecfbaSChristoph Hellwig 1171598ecfbaSChristoph Hellwig if (ia->io_offset < ib->io_offset) 1172598ecfbaSChristoph Hellwig return -1; 1173b3d423ecSChristoph Hellwig if (ia->io_offset > ib->io_offset) 1174598ecfbaSChristoph Hellwig return 1; 1175598ecfbaSChristoph Hellwig return 0; 1176598ecfbaSChristoph Hellwig } 1177598ecfbaSChristoph Hellwig 1178598ecfbaSChristoph Hellwig void 1179598ecfbaSChristoph Hellwig iomap_sort_ioends(struct list_head *ioend_list) 1180598ecfbaSChristoph Hellwig { 1181598ecfbaSChristoph Hellwig list_sort(NULL, ioend_list, iomap_ioend_compare); 1182598ecfbaSChristoph Hellwig } 1183598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_sort_ioends); 1184598ecfbaSChristoph Hellwig 1185598ecfbaSChristoph Hellwig static void iomap_writepage_end_bio(struct bio *bio) 1186598ecfbaSChristoph Hellwig { 1187598ecfbaSChristoph Hellwig struct iomap_ioend *ioend = bio->bi_private; 1188598ecfbaSChristoph Hellwig 1189598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status)); 1190598ecfbaSChristoph Hellwig } 1191598ecfbaSChristoph Hellwig 1192598ecfbaSChristoph Hellwig /* 1193598ecfbaSChristoph Hellwig * Submit the final bio for an ioend. 1194598ecfbaSChristoph Hellwig * 1195598ecfbaSChristoph Hellwig * If @error is non-zero, it means that we have a situation where some part of 1196598ecfbaSChristoph Hellwig * the submission process has failed after we have marked paged for writeback 1197598ecfbaSChristoph Hellwig * and unlocked them. In this situation, we need to fail the bio instead of 1198598ecfbaSChristoph Hellwig * submitting it. This typically only happens on a filesystem shutdown. 1199598ecfbaSChristoph Hellwig */ 1200598ecfbaSChristoph Hellwig static int 1201598ecfbaSChristoph Hellwig iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend, 1202598ecfbaSChristoph Hellwig int error) 1203598ecfbaSChristoph Hellwig { 1204598ecfbaSChristoph Hellwig ioend->io_bio->bi_private = ioend; 1205598ecfbaSChristoph Hellwig ioend->io_bio->bi_end_io = iomap_writepage_end_bio; 1206598ecfbaSChristoph Hellwig 1207598ecfbaSChristoph Hellwig if (wpc->ops->prepare_ioend) 1208598ecfbaSChristoph Hellwig error = wpc->ops->prepare_ioend(ioend, error); 1209598ecfbaSChristoph Hellwig if (error) { 1210598ecfbaSChristoph Hellwig /* 1211598ecfbaSChristoph Hellwig * If we are failing the IO now, just mark the ioend with an 1212598ecfbaSChristoph Hellwig * error and finish it. This will run IO completion immediately 1213598ecfbaSChristoph Hellwig * as there is only one reference to the ioend at this point in 1214598ecfbaSChristoph Hellwig * time. 1215598ecfbaSChristoph Hellwig */ 1216598ecfbaSChristoph Hellwig ioend->io_bio->bi_status = errno_to_blk_status(error); 1217598ecfbaSChristoph Hellwig bio_endio(ioend->io_bio); 1218598ecfbaSChristoph Hellwig return error; 1219598ecfbaSChristoph Hellwig } 1220598ecfbaSChristoph Hellwig 1221598ecfbaSChristoph Hellwig submit_bio(ioend->io_bio); 1222598ecfbaSChristoph Hellwig return 0; 1223598ecfbaSChristoph Hellwig } 1224598ecfbaSChristoph Hellwig 1225598ecfbaSChristoph Hellwig static struct iomap_ioend * 1226598ecfbaSChristoph Hellwig iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc, 1227598ecfbaSChristoph Hellwig loff_t offset, sector_t sector, struct writeback_control *wbc) 1228598ecfbaSChristoph Hellwig { 1229598ecfbaSChristoph Hellwig struct iomap_ioend *ioend; 1230598ecfbaSChristoph Hellwig struct bio *bio; 1231598ecfbaSChristoph Hellwig 1232598ecfbaSChristoph Hellwig bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &iomap_ioend_bioset); 1233598ecfbaSChristoph Hellwig bio_set_dev(bio, wpc->iomap.bdev); 1234598ecfbaSChristoph Hellwig bio->bi_iter.bi_sector = sector; 1235598ecfbaSChristoph Hellwig bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); 1236598ecfbaSChristoph Hellwig bio->bi_write_hint = inode->i_write_hint; 1237598ecfbaSChristoph Hellwig wbc_init_bio(wbc, bio); 1238598ecfbaSChristoph Hellwig 1239598ecfbaSChristoph Hellwig ioend = container_of(bio, struct iomap_ioend, io_inline_bio); 1240598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1241598ecfbaSChristoph Hellwig ioend->io_type = wpc->iomap.type; 1242598ecfbaSChristoph Hellwig ioend->io_flags = wpc->iomap.flags; 1243598ecfbaSChristoph Hellwig ioend->io_inode = inode; 1244598ecfbaSChristoph Hellwig ioend->io_size = 0; 1245598ecfbaSChristoph Hellwig ioend->io_offset = offset; 1246598ecfbaSChristoph Hellwig ioend->io_private = NULL; 1247598ecfbaSChristoph Hellwig ioend->io_bio = bio; 1248598ecfbaSChristoph Hellwig return ioend; 1249598ecfbaSChristoph Hellwig } 1250598ecfbaSChristoph Hellwig 1251598ecfbaSChristoph Hellwig /* 1252598ecfbaSChristoph Hellwig * Allocate a new bio, and chain the old bio to the new one. 1253598ecfbaSChristoph Hellwig * 1254598ecfbaSChristoph Hellwig * Note that we have to do perform the chaining in this unintuitive order 1255598ecfbaSChristoph Hellwig * so that the bi_private linkage is set up in the right direction for the 1256598ecfbaSChristoph Hellwig * traversal in iomap_finish_ioend(). 1257598ecfbaSChristoph Hellwig */ 1258598ecfbaSChristoph Hellwig static struct bio * 1259598ecfbaSChristoph Hellwig iomap_chain_bio(struct bio *prev) 1260598ecfbaSChristoph Hellwig { 1261598ecfbaSChristoph Hellwig struct bio *new; 1262598ecfbaSChristoph Hellwig 1263598ecfbaSChristoph Hellwig new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); 1264598ecfbaSChristoph Hellwig bio_copy_dev(new, prev);/* also copies over blkcg information */ 1265598ecfbaSChristoph Hellwig new->bi_iter.bi_sector = bio_end_sector(prev); 1266598ecfbaSChristoph Hellwig new->bi_opf = prev->bi_opf; 1267598ecfbaSChristoph Hellwig new->bi_write_hint = prev->bi_write_hint; 1268598ecfbaSChristoph Hellwig 1269598ecfbaSChristoph Hellwig bio_chain(prev, new); 1270598ecfbaSChristoph Hellwig bio_get(prev); /* for iomap_finish_ioend */ 1271598ecfbaSChristoph Hellwig submit_bio(prev); 1272598ecfbaSChristoph Hellwig return new; 1273598ecfbaSChristoph Hellwig } 1274598ecfbaSChristoph Hellwig 1275598ecfbaSChristoph Hellwig static bool 1276598ecfbaSChristoph Hellwig iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset, 1277598ecfbaSChristoph Hellwig sector_t sector) 1278598ecfbaSChristoph Hellwig { 1279598ecfbaSChristoph Hellwig if ((wpc->iomap.flags & IOMAP_F_SHARED) != 1280598ecfbaSChristoph Hellwig (wpc->ioend->io_flags & IOMAP_F_SHARED)) 1281598ecfbaSChristoph Hellwig return false; 1282598ecfbaSChristoph Hellwig if (wpc->iomap.type != wpc->ioend->io_type) 1283598ecfbaSChristoph Hellwig return false; 1284598ecfbaSChristoph Hellwig if (offset != wpc->ioend->io_offset + wpc->ioend->io_size) 1285598ecfbaSChristoph Hellwig return false; 1286598ecfbaSChristoph Hellwig if (sector != bio_end_sector(wpc->ioend->io_bio)) 1287598ecfbaSChristoph Hellwig return false; 1288598ecfbaSChristoph Hellwig return true; 1289598ecfbaSChristoph Hellwig } 1290598ecfbaSChristoph Hellwig 1291598ecfbaSChristoph Hellwig /* 1292598ecfbaSChristoph Hellwig * Test to see if we have an existing ioend structure that we could append to 1293598ecfbaSChristoph Hellwig * first, otherwise finish off the current ioend and start another. 1294598ecfbaSChristoph Hellwig */ 1295598ecfbaSChristoph Hellwig static void 1296598ecfbaSChristoph Hellwig iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page, 1297598ecfbaSChristoph Hellwig struct iomap_page *iop, struct iomap_writepage_ctx *wpc, 1298598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct list_head *iolist) 1299598ecfbaSChristoph Hellwig { 1300598ecfbaSChristoph Hellwig sector_t sector = iomap_sector(&wpc->iomap, offset); 1301598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1302598ecfbaSChristoph Hellwig unsigned poff = offset & (PAGE_SIZE - 1); 1303598ecfbaSChristoph Hellwig bool merged, same_page = false; 1304598ecfbaSChristoph Hellwig 1305598ecfbaSChristoph Hellwig if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) { 1306598ecfbaSChristoph Hellwig if (wpc->ioend) 1307598ecfbaSChristoph Hellwig list_add(&wpc->ioend->io_list, iolist); 1308598ecfbaSChristoph Hellwig wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc); 1309598ecfbaSChristoph Hellwig } 1310598ecfbaSChristoph Hellwig 1311598ecfbaSChristoph Hellwig merged = __bio_try_merge_page(wpc->ioend->io_bio, page, len, poff, 1312598ecfbaSChristoph Hellwig &same_page); 1313*0fb2d720SMatthew Wilcox (Oracle) if (iop) 1314*0fb2d720SMatthew Wilcox (Oracle) atomic_add(len, &iop->write_bytes_pending); 1315598ecfbaSChristoph Hellwig 1316598ecfbaSChristoph Hellwig if (!merged) { 1317598ecfbaSChristoph Hellwig if (bio_full(wpc->ioend->io_bio, len)) { 1318598ecfbaSChristoph Hellwig wpc->ioend->io_bio = 1319598ecfbaSChristoph Hellwig iomap_chain_bio(wpc->ioend->io_bio); 1320598ecfbaSChristoph Hellwig } 1321598ecfbaSChristoph Hellwig bio_add_page(wpc->ioend->io_bio, page, len, poff); 1322598ecfbaSChristoph Hellwig } 1323598ecfbaSChristoph Hellwig 1324598ecfbaSChristoph Hellwig wpc->ioend->io_size += len; 1325598ecfbaSChristoph Hellwig wbc_account_cgroup_owner(wbc, page, len); 1326598ecfbaSChristoph Hellwig } 1327598ecfbaSChristoph Hellwig 1328598ecfbaSChristoph Hellwig /* 1329598ecfbaSChristoph Hellwig * We implement an immediate ioend submission policy here to avoid needing to 1330598ecfbaSChristoph Hellwig * chain multiple ioends and hence nest mempool allocations which can violate 1331598ecfbaSChristoph Hellwig * forward progress guarantees we need to provide. The current ioend we are 1332598ecfbaSChristoph Hellwig * adding blocks to is cached on the writepage context, and if the new block 1333598ecfbaSChristoph Hellwig * does not append to the cached ioend it will create a new ioend and cache that 1334598ecfbaSChristoph Hellwig * instead. 1335598ecfbaSChristoph Hellwig * 1336598ecfbaSChristoph Hellwig * If a new ioend is created and cached, the old ioend is returned and queued 1337598ecfbaSChristoph Hellwig * locally for submission once the entire page is processed or an error has been 1338598ecfbaSChristoph Hellwig * detected. While ioends are submitted immediately after they are completed, 1339598ecfbaSChristoph Hellwig * batching optimisations are provided by higher level block plugging. 1340598ecfbaSChristoph Hellwig * 1341598ecfbaSChristoph Hellwig * At the end of a writeback pass, there will be a cached ioend remaining on the 1342598ecfbaSChristoph Hellwig * writepage context that the caller will need to submit. 1343598ecfbaSChristoph Hellwig */ 1344598ecfbaSChristoph Hellwig static int 1345598ecfbaSChristoph Hellwig iomap_writepage_map(struct iomap_writepage_ctx *wpc, 1346598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct inode *inode, 1347598ecfbaSChristoph Hellwig struct page *page, u64 end_offset) 1348598ecfbaSChristoph Hellwig { 1349598ecfbaSChristoph Hellwig struct iomap_page *iop = to_iomap_page(page); 1350598ecfbaSChristoph Hellwig struct iomap_ioend *ioend, *next; 1351598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1352598ecfbaSChristoph Hellwig u64 file_offset; /* file offset of page */ 1353598ecfbaSChristoph Hellwig int error = 0, count = 0, i; 1354598ecfbaSChristoph Hellwig LIST_HEAD(submit_list); 1355598ecfbaSChristoph Hellwig 135624addd84SMatthew Wilcox (Oracle) WARN_ON_ONCE(i_blocks_per_page(inode, page) > 1 && !iop); 1357*0fb2d720SMatthew Wilcox (Oracle) WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) != 0); 1358598ecfbaSChristoph Hellwig 1359598ecfbaSChristoph Hellwig /* 1360598ecfbaSChristoph Hellwig * Walk through the page to find areas to write back. If we run off the 1361598ecfbaSChristoph Hellwig * end of the current map or find the current map invalid, grab a new 1362598ecfbaSChristoph Hellwig * one. 1363598ecfbaSChristoph Hellwig */ 1364598ecfbaSChristoph Hellwig for (i = 0, file_offset = page_offset(page); 1365598ecfbaSChristoph Hellwig i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset; 1366598ecfbaSChristoph Hellwig i++, file_offset += len) { 1367598ecfbaSChristoph Hellwig if (iop && !test_bit(i, iop->uptodate)) 1368598ecfbaSChristoph Hellwig continue; 1369598ecfbaSChristoph Hellwig 1370598ecfbaSChristoph Hellwig error = wpc->ops->map_blocks(wpc, inode, file_offset); 1371598ecfbaSChristoph Hellwig if (error) 1372598ecfbaSChristoph Hellwig break; 13733e19e6f3SChristoph Hellwig if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE)) 13743e19e6f3SChristoph Hellwig continue; 1375598ecfbaSChristoph Hellwig if (wpc->iomap.type == IOMAP_HOLE) 1376598ecfbaSChristoph Hellwig continue; 1377598ecfbaSChristoph Hellwig iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc, 1378598ecfbaSChristoph Hellwig &submit_list); 1379598ecfbaSChristoph Hellwig count++; 1380598ecfbaSChristoph Hellwig } 1381598ecfbaSChristoph Hellwig 1382598ecfbaSChristoph Hellwig WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list)); 1383598ecfbaSChristoph Hellwig WARN_ON_ONCE(!PageLocked(page)); 1384598ecfbaSChristoph Hellwig WARN_ON_ONCE(PageWriteback(page)); 1385598ecfbaSChristoph Hellwig 1386598ecfbaSChristoph Hellwig /* 1387598ecfbaSChristoph Hellwig * We cannot cancel the ioend directly here on error. We may have 1388598ecfbaSChristoph Hellwig * already set other pages under writeback and hence we have to run I/O 1389598ecfbaSChristoph Hellwig * completion to mark the error state of the pages under writeback 1390598ecfbaSChristoph Hellwig * appropriately. 1391598ecfbaSChristoph Hellwig */ 1392598ecfbaSChristoph Hellwig if (unlikely(error)) { 1393598ecfbaSChristoph Hellwig if (!count) { 1394598ecfbaSChristoph Hellwig /* 1395598ecfbaSChristoph Hellwig * If the current page hasn't been added to ioend, it 1396598ecfbaSChristoph Hellwig * won't be affected by I/O completions and we must 1397598ecfbaSChristoph Hellwig * discard and unlock it right here. 1398598ecfbaSChristoph Hellwig */ 1399598ecfbaSChristoph Hellwig if (wpc->ops->discard_page) 1400598ecfbaSChristoph Hellwig wpc->ops->discard_page(page); 1401598ecfbaSChristoph Hellwig ClearPageUptodate(page); 1402598ecfbaSChristoph Hellwig unlock_page(page); 1403598ecfbaSChristoph Hellwig goto done; 1404598ecfbaSChristoph Hellwig } 1405598ecfbaSChristoph Hellwig 1406598ecfbaSChristoph Hellwig /* 1407598ecfbaSChristoph Hellwig * If the page was not fully cleaned, we need to ensure that the 1408598ecfbaSChristoph Hellwig * higher layers come back to it correctly. That means we need 1409598ecfbaSChristoph Hellwig * to keep the page dirty, and for WB_SYNC_ALL writeback we need 1410598ecfbaSChristoph Hellwig * to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed 1411598ecfbaSChristoph Hellwig * so another attempt to write this page in this writeback sweep 1412598ecfbaSChristoph Hellwig * will be made. 1413598ecfbaSChristoph Hellwig */ 1414598ecfbaSChristoph Hellwig set_page_writeback_keepwrite(page); 1415598ecfbaSChristoph Hellwig } else { 1416598ecfbaSChristoph Hellwig clear_page_dirty_for_io(page); 1417598ecfbaSChristoph Hellwig set_page_writeback(page); 1418598ecfbaSChristoph Hellwig } 1419598ecfbaSChristoph Hellwig 1420598ecfbaSChristoph Hellwig unlock_page(page); 1421598ecfbaSChristoph Hellwig 1422598ecfbaSChristoph Hellwig /* 1423598ecfbaSChristoph Hellwig * Preserve the original error if there was one, otherwise catch 1424598ecfbaSChristoph Hellwig * submission errors here and propagate into subsequent ioend 1425598ecfbaSChristoph Hellwig * submissions. 1426598ecfbaSChristoph Hellwig */ 1427598ecfbaSChristoph Hellwig list_for_each_entry_safe(ioend, next, &submit_list, io_list) { 1428598ecfbaSChristoph Hellwig int error2; 1429598ecfbaSChristoph Hellwig 1430598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1431598ecfbaSChristoph Hellwig error2 = iomap_submit_ioend(wpc, ioend, error); 1432598ecfbaSChristoph Hellwig if (error2 && !error) 1433598ecfbaSChristoph Hellwig error = error2; 1434598ecfbaSChristoph Hellwig } 1435598ecfbaSChristoph Hellwig 1436598ecfbaSChristoph Hellwig /* 1437598ecfbaSChristoph Hellwig * We can end up here with no error and nothing to write only if we race 1438598ecfbaSChristoph Hellwig * with a partial page truncate on a sub-page block sized filesystem. 1439598ecfbaSChristoph Hellwig */ 1440598ecfbaSChristoph Hellwig if (!count) 1441598ecfbaSChristoph Hellwig end_page_writeback(page); 1442598ecfbaSChristoph Hellwig done: 1443598ecfbaSChristoph Hellwig mapping_set_error(page->mapping, error); 1444598ecfbaSChristoph Hellwig return error; 1445598ecfbaSChristoph Hellwig } 1446598ecfbaSChristoph Hellwig 1447598ecfbaSChristoph Hellwig /* 1448598ecfbaSChristoph Hellwig * Write out a dirty page. 1449598ecfbaSChristoph Hellwig * 1450598ecfbaSChristoph Hellwig * For delalloc space on the page we need to allocate space and flush it. 1451598ecfbaSChristoph Hellwig * For unwritten space on the page we need to start the conversion to 1452598ecfbaSChristoph Hellwig * regular allocated space. 1453598ecfbaSChristoph Hellwig */ 1454598ecfbaSChristoph Hellwig static int 1455598ecfbaSChristoph Hellwig iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data) 1456598ecfbaSChristoph Hellwig { 1457598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc = data; 1458598ecfbaSChristoph Hellwig struct inode *inode = page->mapping->host; 1459598ecfbaSChristoph Hellwig pgoff_t end_index; 1460598ecfbaSChristoph Hellwig u64 end_offset; 1461598ecfbaSChristoph Hellwig loff_t offset; 1462598ecfbaSChristoph Hellwig 14631ac99452SMatthew Wilcox (Oracle) trace_iomap_writepage(inode, page_offset(page), PAGE_SIZE); 1464598ecfbaSChristoph Hellwig 1465598ecfbaSChristoph Hellwig /* 1466598ecfbaSChristoph Hellwig * Refuse to write the page out if we are called from reclaim context. 1467598ecfbaSChristoph Hellwig * 1468598ecfbaSChristoph Hellwig * This avoids stack overflows when called from deeply used stacks in 1469598ecfbaSChristoph Hellwig * random callers for direct reclaim or memcg reclaim. We explicitly 1470598ecfbaSChristoph Hellwig * allow reclaim from kswapd as the stack usage there is relatively low. 1471598ecfbaSChristoph Hellwig * 1472598ecfbaSChristoph Hellwig * This should never happen except in the case of a VM regression so 1473598ecfbaSChristoph Hellwig * warn about it. 1474598ecfbaSChristoph Hellwig */ 1475598ecfbaSChristoph Hellwig if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == 1476598ecfbaSChristoph Hellwig PF_MEMALLOC)) 1477598ecfbaSChristoph Hellwig goto redirty; 1478598ecfbaSChristoph Hellwig 1479598ecfbaSChristoph Hellwig /* 1480598ecfbaSChristoph Hellwig * Given that we do not allow direct reclaim to call us, we should 1481598ecfbaSChristoph Hellwig * never be called in a recursive filesystem reclaim context. 1482598ecfbaSChristoph Hellwig */ 1483598ecfbaSChristoph Hellwig if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS)) 1484598ecfbaSChristoph Hellwig goto redirty; 1485598ecfbaSChristoph Hellwig 1486598ecfbaSChristoph Hellwig /* 1487598ecfbaSChristoph Hellwig * Is this page beyond the end of the file? 1488598ecfbaSChristoph Hellwig * 1489598ecfbaSChristoph Hellwig * The page index is less than the end_index, adjust the end_offset 1490598ecfbaSChristoph Hellwig * to the highest offset that this page should represent. 1491598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1492598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1493598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1494598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | | 1495598ecfbaSChristoph Hellwig * ^--------------------------------^----------|-------- 1496598ecfbaSChristoph Hellwig * | desired writeback range | see else | 1497598ecfbaSChristoph Hellwig * ---------------------------------^------------------| 1498598ecfbaSChristoph Hellwig */ 1499598ecfbaSChristoph Hellwig offset = i_size_read(inode); 1500598ecfbaSChristoph Hellwig end_index = offset >> PAGE_SHIFT; 1501598ecfbaSChristoph Hellwig if (page->index < end_index) 1502598ecfbaSChristoph Hellwig end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT; 1503598ecfbaSChristoph Hellwig else { 1504598ecfbaSChristoph Hellwig /* 1505598ecfbaSChristoph Hellwig * Check whether the page to write out is beyond or straddles 1506598ecfbaSChristoph Hellwig * i_size or not. 1507598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1508598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1509598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1510598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | 1511598ecfbaSChristoph Hellwig * ^--------------------------------^-----------|--------- 1512598ecfbaSChristoph Hellwig * | | Straddles | 1513598ecfbaSChristoph Hellwig * ---------------------------------^-----------|--------| 1514598ecfbaSChristoph Hellwig */ 1515598ecfbaSChristoph Hellwig unsigned offset_into_page = offset & (PAGE_SIZE - 1); 1516598ecfbaSChristoph Hellwig 1517598ecfbaSChristoph Hellwig /* 1518598ecfbaSChristoph Hellwig * Skip the page if it is fully outside i_size, e.g. due to a 1519598ecfbaSChristoph Hellwig * truncate operation that is in progress. We must redirty the 1520598ecfbaSChristoph Hellwig * page so that reclaim stops reclaiming it. Otherwise 1521598ecfbaSChristoph Hellwig * iomap_vm_releasepage() is called on it and gets confused. 1522598ecfbaSChristoph Hellwig * 1523598ecfbaSChristoph Hellwig * Note that the end_index is unsigned long, it would overflow 1524598ecfbaSChristoph Hellwig * if the given offset is greater than 16TB on 32-bit system 1525598ecfbaSChristoph Hellwig * and if we do check the page is fully outside i_size or not 1526598ecfbaSChristoph Hellwig * via "if (page->index >= end_index + 1)" as "end_index + 1" 1527598ecfbaSChristoph Hellwig * will be evaluated to 0. Hence this page will be redirtied 1528598ecfbaSChristoph Hellwig * and be written out repeatedly which would result in an 1529598ecfbaSChristoph Hellwig * infinite loop, the user program that perform this operation 1530598ecfbaSChristoph Hellwig * will hang. Instead, we can verify this situation by checking 1531598ecfbaSChristoph Hellwig * if the page to write is totally beyond the i_size or if it's 1532598ecfbaSChristoph Hellwig * offset is just equal to the EOF. 1533598ecfbaSChristoph Hellwig */ 1534598ecfbaSChristoph Hellwig if (page->index > end_index || 1535598ecfbaSChristoph Hellwig (page->index == end_index && offset_into_page == 0)) 1536598ecfbaSChristoph Hellwig goto redirty; 1537598ecfbaSChristoph Hellwig 1538598ecfbaSChristoph Hellwig /* 1539598ecfbaSChristoph Hellwig * The page straddles i_size. It must be zeroed out on each 1540598ecfbaSChristoph Hellwig * and every writepage invocation because it may be mmapped. 1541598ecfbaSChristoph Hellwig * "A file is mapped in multiples of the page size. For a file 1542598ecfbaSChristoph Hellwig * that is not a multiple of the page size, the remaining 1543598ecfbaSChristoph Hellwig * memory is zeroed when mapped, and writes to that region are 1544598ecfbaSChristoph Hellwig * not written out to the file." 1545598ecfbaSChristoph Hellwig */ 1546598ecfbaSChristoph Hellwig zero_user_segment(page, offset_into_page, PAGE_SIZE); 1547598ecfbaSChristoph Hellwig 1548598ecfbaSChristoph Hellwig /* Adjust the end_offset to the end of file */ 1549598ecfbaSChristoph Hellwig end_offset = offset; 1550598ecfbaSChristoph Hellwig } 1551598ecfbaSChristoph Hellwig 1552598ecfbaSChristoph Hellwig return iomap_writepage_map(wpc, wbc, inode, page, end_offset); 1553598ecfbaSChristoph Hellwig 1554598ecfbaSChristoph Hellwig redirty: 1555598ecfbaSChristoph Hellwig redirty_page_for_writepage(wbc, page); 1556598ecfbaSChristoph Hellwig unlock_page(page); 1557598ecfbaSChristoph Hellwig return 0; 1558598ecfbaSChristoph Hellwig } 1559598ecfbaSChristoph Hellwig 1560598ecfbaSChristoph Hellwig int 1561598ecfbaSChristoph Hellwig iomap_writepage(struct page *page, struct writeback_control *wbc, 1562598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1563598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1564598ecfbaSChristoph Hellwig { 1565598ecfbaSChristoph Hellwig int ret; 1566598ecfbaSChristoph Hellwig 1567598ecfbaSChristoph Hellwig wpc->ops = ops; 1568598ecfbaSChristoph Hellwig ret = iomap_do_writepage(page, wbc, wpc); 1569598ecfbaSChristoph Hellwig if (!wpc->ioend) 1570598ecfbaSChristoph Hellwig return ret; 1571598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1572598ecfbaSChristoph Hellwig } 1573598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepage); 1574598ecfbaSChristoph Hellwig 1575598ecfbaSChristoph Hellwig int 1576598ecfbaSChristoph Hellwig iomap_writepages(struct address_space *mapping, struct writeback_control *wbc, 1577598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1578598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1579598ecfbaSChristoph Hellwig { 1580598ecfbaSChristoph Hellwig int ret; 1581598ecfbaSChristoph Hellwig 1582598ecfbaSChristoph Hellwig wpc->ops = ops; 1583598ecfbaSChristoph Hellwig ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc); 1584598ecfbaSChristoph Hellwig if (!wpc->ioend) 1585598ecfbaSChristoph Hellwig return ret; 1586598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1587598ecfbaSChristoph Hellwig } 1588598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepages); 1589598ecfbaSChristoph Hellwig 1590598ecfbaSChristoph Hellwig static int __init iomap_init(void) 1591598ecfbaSChristoph Hellwig { 1592598ecfbaSChristoph Hellwig return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE), 1593598ecfbaSChristoph Hellwig offsetof(struct iomap_ioend, io_inline_bio), 1594598ecfbaSChristoph Hellwig BIOSET_NEED_BVECS); 1595598ecfbaSChristoph Hellwig } 1596598ecfbaSChristoph Hellwig fs_initcall(iomap_init); 1597