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 /* 25ab08b01eSChristoph Hellwig * Structure allocated for each page when block size < PAGE_SIZE to track 26ab08b01eSChristoph Hellwig * sub-page uptodate status and I/O completions. 27ab08b01eSChristoph Hellwig */ 28ab08b01eSChristoph Hellwig struct iomap_page { 29ab08b01eSChristoph Hellwig atomic_t read_count; 30ab08b01eSChristoph Hellwig atomic_t write_count; 311cea335dSChristoph Hellwig spinlock_t uptodate_lock; 32ab08b01eSChristoph Hellwig DECLARE_BITMAP(uptodate, PAGE_SIZE / 512); 33ab08b01eSChristoph Hellwig }; 34ab08b01eSChristoph Hellwig 35ab08b01eSChristoph Hellwig static inline struct iomap_page *to_iomap_page(struct page *page) 36ab08b01eSChristoph Hellwig { 37ab08b01eSChristoph Hellwig if (page_has_private(page)) 38ab08b01eSChristoph Hellwig return (struct iomap_page *)page_private(page); 39ab08b01eSChristoph Hellwig return NULL; 40ab08b01eSChristoph Hellwig } 41ab08b01eSChristoph Hellwig 42598ecfbaSChristoph Hellwig static struct bio_set iomap_ioend_bioset; 43598ecfbaSChristoph Hellwig 44afc51aaaSDarrick J. Wong static struct iomap_page * 45afc51aaaSDarrick J. Wong iomap_page_create(struct inode *inode, struct page *page) 46afc51aaaSDarrick J. Wong { 47afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 48afc51aaaSDarrick J. Wong 49afc51aaaSDarrick J. Wong if (iop || i_blocksize(inode) == PAGE_SIZE) 50afc51aaaSDarrick J. Wong return iop; 51afc51aaaSDarrick J. Wong 52afc51aaaSDarrick J. Wong iop = kmalloc(sizeof(*iop), GFP_NOFS | __GFP_NOFAIL); 53afc51aaaSDarrick J. Wong atomic_set(&iop->read_count, 0); 54afc51aaaSDarrick J. Wong atomic_set(&iop->write_count, 0); 551cea335dSChristoph Hellwig spin_lock_init(&iop->uptodate_lock); 56afc51aaaSDarrick J. Wong bitmap_zero(iop->uptodate, PAGE_SIZE / SECTOR_SIZE); 57afc51aaaSDarrick J. Wong 58afc51aaaSDarrick J. Wong /* 59afc51aaaSDarrick J. Wong * migrate_page_move_mapping() assumes that pages with private data have 60afc51aaaSDarrick J. Wong * their count elevated by 1. 61afc51aaaSDarrick J. Wong */ 62afc51aaaSDarrick J. Wong get_page(page); 63afc51aaaSDarrick J. Wong set_page_private(page, (unsigned long)iop); 64afc51aaaSDarrick J. Wong SetPagePrivate(page); 65afc51aaaSDarrick J. Wong return iop; 66afc51aaaSDarrick J. Wong } 67afc51aaaSDarrick J. Wong 68afc51aaaSDarrick J. Wong static void 69afc51aaaSDarrick J. Wong iomap_page_release(struct page *page) 70afc51aaaSDarrick J. Wong { 71afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 72afc51aaaSDarrick J. Wong 73afc51aaaSDarrick J. Wong if (!iop) 74afc51aaaSDarrick J. Wong return; 75afc51aaaSDarrick J. Wong WARN_ON_ONCE(atomic_read(&iop->read_count)); 76afc51aaaSDarrick J. Wong WARN_ON_ONCE(atomic_read(&iop->write_count)); 77afc51aaaSDarrick J. Wong ClearPagePrivate(page); 78afc51aaaSDarrick J. Wong set_page_private(page, 0); 79afc51aaaSDarrick J. Wong put_page(page); 80afc51aaaSDarrick J. Wong kfree(iop); 81afc51aaaSDarrick J. Wong } 82afc51aaaSDarrick J. Wong 83afc51aaaSDarrick J. Wong /* 84afc51aaaSDarrick J. Wong * Calculate the range inside the page that we actually need to read. 85afc51aaaSDarrick J. Wong */ 86afc51aaaSDarrick J. Wong static void 87afc51aaaSDarrick J. Wong iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop, 88afc51aaaSDarrick J. Wong loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp) 89afc51aaaSDarrick J. Wong { 90afc51aaaSDarrick J. Wong loff_t orig_pos = *pos; 91afc51aaaSDarrick J. Wong loff_t isize = i_size_read(inode); 92afc51aaaSDarrick J. Wong unsigned block_bits = inode->i_blkbits; 93afc51aaaSDarrick J. Wong unsigned block_size = (1 << block_bits); 94afc51aaaSDarrick J. Wong unsigned poff = offset_in_page(*pos); 95afc51aaaSDarrick J. Wong unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length); 96afc51aaaSDarrick J. Wong unsigned first = poff >> block_bits; 97afc51aaaSDarrick J. Wong unsigned last = (poff + plen - 1) >> block_bits; 98afc51aaaSDarrick J. Wong 99afc51aaaSDarrick J. Wong /* 100afc51aaaSDarrick J. Wong * If the block size is smaller than the page size we need to check the 101afc51aaaSDarrick J. Wong * per-block uptodate status and adjust the offset and length if needed 102afc51aaaSDarrick J. Wong * to avoid reading in already uptodate ranges. 103afc51aaaSDarrick J. Wong */ 104afc51aaaSDarrick J. Wong if (iop) { 105afc51aaaSDarrick J. Wong unsigned int i; 106afc51aaaSDarrick J. Wong 107afc51aaaSDarrick J. Wong /* move forward for each leading block marked uptodate */ 108afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) { 109afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 110afc51aaaSDarrick J. Wong break; 111afc51aaaSDarrick J. Wong *pos += block_size; 112afc51aaaSDarrick J. Wong poff += block_size; 113afc51aaaSDarrick J. Wong plen -= block_size; 114afc51aaaSDarrick J. Wong first++; 115afc51aaaSDarrick J. Wong } 116afc51aaaSDarrick J. Wong 117afc51aaaSDarrick J. Wong /* truncate len if we find any trailing uptodate block(s) */ 118afc51aaaSDarrick J. Wong for ( ; i <= last; i++) { 119afc51aaaSDarrick J. Wong if (test_bit(i, iop->uptodate)) { 120afc51aaaSDarrick J. Wong plen -= (last - i + 1) * block_size; 121afc51aaaSDarrick J. Wong last = i - 1; 122afc51aaaSDarrick J. Wong break; 123afc51aaaSDarrick J. Wong } 124afc51aaaSDarrick J. Wong } 125afc51aaaSDarrick J. Wong } 126afc51aaaSDarrick J. Wong 127afc51aaaSDarrick J. Wong /* 128afc51aaaSDarrick J. Wong * If the extent spans the block that contains the i_size we need to 129afc51aaaSDarrick J. Wong * handle both halves separately so that we properly zero data in the 130afc51aaaSDarrick J. Wong * page cache for blocks that are entirely outside of i_size. 131afc51aaaSDarrick J. Wong */ 132afc51aaaSDarrick J. Wong if (orig_pos <= isize && orig_pos + length > isize) { 133afc51aaaSDarrick J. Wong unsigned end = offset_in_page(isize - 1) >> block_bits; 134afc51aaaSDarrick J. Wong 135afc51aaaSDarrick J. Wong if (first <= end && last > end) 136afc51aaaSDarrick J. Wong plen -= (last - end) * block_size; 137afc51aaaSDarrick J. Wong } 138afc51aaaSDarrick J. Wong 139afc51aaaSDarrick J. Wong *offp = poff; 140afc51aaaSDarrick J. Wong *lenp = plen; 141afc51aaaSDarrick J. Wong } 142afc51aaaSDarrick J. Wong 143afc51aaaSDarrick J. Wong static void 1441cea335dSChristoph Hellwig iomap_iop_set_range_uptodate(struct page *page, unsigned off, unsigned len) 145afc51aaaSDarrick J. Wong { 146afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 147afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 148afc51aaaSDarrick J. Wong unsigned first = off >> inode->i_blkbits; 149afc51aaaSDarrick J. Wong unsigned last = (off + len - 1) >> inode->i_blkbits; 150afc51aaaSDarrick J. Wong bool uptodate = true; 1511cea335dSChristoph Hellwig unsigned long flags; 1521cea335dSChristoph Hellwig unsigned int i; 153afc51aaaSDarrick J. Wong 1541cea335dSChristoph Hellwig spin_lock_irqsave(&iop->uptodate_lock, flags); 155afc51aaaSDarrick J. Wong for (i = 0; i < PAGE_SIZE / i_blocksize(inode); i++) { 156afc51aaaSDarrick J. Wong if (i >= first && i <= last) 157afc51aaaSDarrick J. Wong set_bit(i, iop->uptodate); 158afc51aaaSDarrick J. Wong else if (!test_bit(i, iop->uptodate)) 159afc51aaaSDarrick J. Wong uptodate = false; 160afc51aaaSDarrick J. Wong } 1611cea335dSChristoph Hellwig 1621cea335dSChristoph Hellwig if (uptodate) 1631cea335dSChristoph Hellwig SetPageUptodate(page); 1641cea335dSChristoph Hellwig spin_unlock_irqrestore(&iop->uptodate_lock, flags); 165afc51aaaSDarrick J. Wong } 166afc51aaaSDarrick J. Wong 1671cea335dSChristoph Hellwig static void 1681cea335dSChristoph Hellwig iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len) 1691cea335dSChristoph Hellwig { 1701cea335dSChristoph Hellwig if (PageError(page)) 1711cea335dSChristoph Hellwig return; 1721cea335dSChristoph Hellwig 1731cea335dSChristoph Hellwig if (page_has_private(page)) 1741cea335dSChristoph Hellwig iomap_iop_set_range_uptodate(page, off, len); 1751cea335dSChristoph Hellwig else 176afc51aaaSDarrick J. Wong SetPageUptodate(page); 177afc51aaaSDarrick J. Wong } 178afc51aaaSDarrick J. Wong 179afc51aaaSDarrick J. Wong static void 180afc51aaaSDarrick J. Wong iomap_read_finish(struct iomap_page *iop, struct page *page) 181afc51aaaSDarrick J. Wong { 182afc51aaaSDarrick J. Wong if (!iop || atomic_dec_and_test(&iop->read_count)) 183afc51aaaSDarrick J. Wong unlock_page(page); 184afc51aaaSDarrick J. Wong } 185afc51aaaSDarrick J. Wong 186afc51aaaSDarrick J. Wong static void 187afc51aaaSDarrick J. Wong iomap_read_page_end_io(struct bio_vec *bvec, int error) 188afc51aaaSDarrick J. Wong { 189afc51aaaSDarrick J. Wong struct page *page = bvec->bv_page; 190afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 191afc51aaaSDarrick J. Wong 192afc51aaaSDarrick J. Wong if (unlikely(error)) { 193afc51aaaSDarrick J. Wong ClearPageUptodate(page); 194afc51aaaSDarrick J. Wong SetPageError(page); 195afc51aaaSDarrick J. Wong } else { 196afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len); 197afc51aaaSDarrick J. Wong } 198afc51aaaSDarrick J. Wong 199afc51aaaSDarrick J. Wong iomap_read_finish(iop, page); 200afc51aaaSDarrick J. Wong } 201afc51aaaSDarrick J. Wong 202afc51aaaSDarrick J. Wong static void 203afc51aaaSDarrick J. Wong iomap_read_end_io(struct bio *bio) 204afc51aaaSDarrick J. Wong { 205afc51aaaSDarrick J. Wong int error = blk_status_to_errno(bio->bi_status); 206afc51aaaSDarrick J. Wong struct bio_vec *bvec; 207afc51aaaSDarrick J. Wong struct bvec_iter_all iter_all; 208afc51aaaSDarrick J. Wong 209afc51aaaSDarrick J. Wong bio_for_each_segment_all(bvec, bio, iter_all) 210afc51aaaSDarrick J. Wong iomap_read_page_end_io(bvec, error); 211afc51aaaSDarrick J. Wong bio_put(bio); 212afc51aaaSDarrick J. Wong } 213afc51aaaSDarrick J. Wong 214afc51aaaSDarrick J. Wong struct iomap_readpage_ctx { 215afc51aaaSDarrick J. Wong struct page *cur_page; 216afc51aaaSDarrick J. Wong bool cur_page_in_bio; 217afc51aaaSDarrick J. Wong bool is_readahead; 218afc51aaaSDarrick J. Wong struct bio *bio; 219afc51aaaSDarrick J. Wong struct list_head *pages; 220afc51aaaSDarrick J. Wong }; 221afc51aaaSDarrick J. Wong 222afc51aaaSDarrick J. Wong static void 223afc51aaaSDarrick J. Wong iomap_read_inline_data(struct inode *inode, struct page *page, 224afc51aaaSDarrick J. Wong struct iomap *iomap) 225afc51aaaSDarrick J. Wong { 226afc51aaaSDarrick J. Wong size_t size = i_size_read(inode); 227afc51aaaSDarrick J. Wong void *addr; 228afc51aaaSDarrick J. Wong 229afc51aaaSDarrick J. Wong if (PageUptodate(page)) 230afc51aaaSDarrick J. Wong return; 231afc51aaaSDarrick J. Wong 232afc51aaaSDarrick J. Wong BUG_ON(page->index); 233afc51aaaSDarrick J. Wong BUG_ON(size > PAGE_SIZE - offset_in_page(iomap->inline_data)); 234afc51aaaSDarrick J. Wong 235afc51aaaSDarrick J. Wong addr = kmap_atomic(page); 236afc51aaaSDarrick J. Wong memcpy(addr, iomap->inline_data, size); 237afc51aaaSDarrick J. Wong memset(addr + size, 0, PAGE_SIZE - size); 238afc51aaaSDarrick J. Wong kunmap_atomic(addr); 239afc51aaaSDarrick J. Wong SetPageUptodate(page); 240afc51aaaSDarrick J. Wong } 241afc51aaaSDarrick J. Wong 242009d8d84SChristoph Hellwig static inline bool iomap_block_needs_zeroing(struct inode *inode, 243009d8d84SChristoph Hellwig struct iomap *iomap, loff_t pos) 244009d8d84SChristoph Hellwig { 245009d8d84SChristoph Hellwig return iomap->type != IOMAP_MAPPED || 246009d8d84SChristoph Hellwig (iomap->flags & IOMAP_F_NEW) || 247009d8d84SChristoph Hellwig pos >= i_size_read(inode); 248009d8d84SChristoph Hellwig } 249009d8d84SChristoph Hellwig 250afc51aaaSDarrick J. Wong static loff_t 251afc51aaaSDarrick J. Wong iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 252c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 253afc51aaaSDarrick J. Wong { 254afc51aaaSDarrick J. Wong struct iomap_readpage_ctx *ctx = data; 255afc51aaaSDarrick J. Wong struct page *page = ctx->cur_page; 256afc51aaaSDarrick J. Wong struct iomap_page *iop = iomap_page_create(inode, page); 257afc51aaaSDarrick J. Wong bool same_page = false, is_contig = false; 258afc51aaaSDarrick J. Wong loff_t orig_pos = pos; 259afc51aaaSDarrick J. Wong unsigned poff, plen; 260afc51aaaSDarrick J. Wong sector_t sector; 261afc51aaaSDarrick J. Wong 262afc51aaaSDarrick J. Wong if (iomap->type == IOMAP_INLINE) { 263afc51aaaSDarrick J. Wong WARN_ON_ONCE(pos); 264afc51aaaSDarrick J. Wong iomap_read_inline_data(inode, page, iomap); 265afc51aaaSDarrick J. Wong return PAGE_SIZE; 266afc51aaaSDarrick J. Wong } 267afc51aaaSDarrick J. Wong 268afc51aaaSDarrick J. Wong /* zero post-eof blocks as the page may be mapped */ 269afc51aaaSDarrick J. Wong iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen); 270afc51aaaSDarrick J. Wong if (plen == 0) 271afc51aaaSDarrick J. Wong goto done; 272afc51aaaSDarrick J. Wong 273009d8d84SChristoph Hellwig if (iomap_block_needs_zeroing(inode, iomap, pos)) { 274afc51aaaSDarrick J. Wong zero_user(page, poff, plen); 275afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, poff, plen); 276afc51aaaSDarrick J. Wong goto done; 277afc51aaaSDarrick J. Wong } 278afc51aaaSDarrick J. Wong 279afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = true; 280afc51aaaSDarrick J. Wong 281afc51aaaSDarrick J. Wong /* 282afc51aaaSDarrick J. Wong * Try to merge into a previous segment if we can. 283afc51aaaSDarrick J. Wong */ 284afc51aaaSDarrick J. Wong sector = iomap_sector(iomap, pos); 285afc51aaaSDarrick J. Wong if (ctx->bio && bio_end_sector(ctx->bio) == sector) 286afc51aaaSDarrick J. Wong is_contig = true; 287afc51aaaSDarrick J. Wong 288afc51aaaSDarrick J. Wong if (is_contig && 289afc51aaaSDarrick J. Wong __bio_try_merge_page(ctx->bio, page, plen, poff, &same_page)) { 290afc51aaaSDarrick J. Wong if (!same_page && iop) 291afc51aaaSDarrick J. Wong atomic_inc(&iop->read_count); 292afc51aaaSDarrick J. Wong goto done; 293afc51aaaSDarrick J. Wong } 294afc51aaaSDarrick J. Wong 295afc51aaaSDarrick J. Wong /* 296afc51aaaSDarrick J. Wong * If we start a new segment we need to increase the read count, and we 297afc51aaaSDarrick J. Wong * need to do so before submitting any previous full bio to make sure 298afc51aaaSDarrick J. Wong * that we don't prematurely unlock the page. 299afc51aaaSDarrick J. Wong */ 300afc51aaaSDarrick J. Wong if (iop) 301afc51aaaSDarrick J. Wong atomic_inc(&iop->read_count); 302afc51aaaSDarrick J. Wong 303afc51aaaSDarrick J. Wong if (!ctx->bio || !is_contig || bio_full(ctx->bio, plen)) { 304afc51aaaSDarrick J. Wong gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL); 305afc51aaaSDarrick J. Wong int nr_vecs = (length + PAGE_SIZE - 1) >> PAGE_SHIFT; 306afc51aaaSDarrick J. Wong 307afc51aaaSDarrick J. Wong if (ctx->bio) 308afc51aaaSDarrick J. Wong submit_bio(ctx->bio); 309afc51aaaSDarrick J. Wong 310afc51aaaSDarrick J. Wong if (ctx->is_readahead) /* same as readahead_gfp_mask */ 311afc51aaaSDarrick J. Wong gfp |= __GFP_NORETRY | __GFP_NOWARN; 312afc51aaaSDarrick J. Wong ctx->bio = bio_alloc(gfp, min(BIO_MAX_PAGES, nr_vecs)); 313afc51aaaSDarrick J. Wong ctx->bio->bi_opf = REQ_OP_READ; 314afc51aaaSDarrick J. Wong if (ctx->is_readahead) 315afc51aaaSDarrick J. Wong ctx->bio->bi_opf |= REQ_RAHEAD; 316afc51aaaSDarrick J. Wong ctx->bio->bi_iter.bi_sector = sector; 317afc51aaaSDarrick J. Wong bio_set_dev(ctx->bio, iomap->bdev); 318afc51aaaSDarrick J. Wong ctx->bio->bi_end_io = iomap_read_end_io; 319afc51aaaSDarrick J. Wong } 320afc51aaaSDarrick J. Wong 321afc51aaaSDarrick J. Wong bio_add_page(ctx->bio, page, plen, poff); 322afc51aaaSDarrick J. Wong done: 323afc51aaaSDarrick J. Wong /* 324afc51aaaSDarrick J. Wong * Move the caller beyond our range so that it keeps making progress. 325afc51aaaSDarrick J. Wong * For that we have to include any leading non-uptodate ranges, but 326afc51aaaSDarrick J. Wong * we can skip trailing ones as they will be handled in the next 327afc51aaaSDarrick J. Wong * iteration. 328afc51aaaSDarrick J. Wong */ 329afc51aaaSDarrick J. Wong return pos - orig_pos + plen; 330afc51aaaSDarrick J. Wong } 331afc51aaaSDarrick J. Wong 332afc51aaaSDarrick J. Wong int 333afc51aaaSDarrick J. Wong iomap_readpage(struct page *page, const struct iomap_ops *ops) 334afc51aaaSDarrick J. Wong { 335afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { .cur_page = page }; 336afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 337afc51aaaSDarrick J. Wong unsigned poff; 338afc51aaaSDarrick J. Wong loff_t ret; 339afc51aaaSDarrick J. Wong 3409e91c572SChristoph Hellwig trace_iomap_readpage(page->mapping->host, 1); 3419e91c572SChristoph Hellwig 342afc51aaaSDarrick J. Wong for (poff = 0; poff < PAGE_SIZE; poff += ret) { 343afc51aaaSDarrick J. Wong ret = iomap_apply(inode, page_offset(page) + poff, 344afc51aaaSDarrick J. Wong PAGE_SIZE - poff, 0, ops, &ctx, 345afc51aaaSDarrick J. Wong iomap_readpage_actor); 346afc51aaaSDarrick J. Wong if (ret <= 0) { 347afc51aaaSDarrick J. Wong WARN_ON_ONCE(ret == 0); 348afc51aaaSDarrick J. Wong SetPageError(page); 349afc51aaaSDarrick J. Wong break; 350afc51aaaSDarrick J. Wong } 351afc51aaaSDarrick J. Wong } 352afc51aaaSDarrick J. Wong 353afc51aaaSDarrick J. Wong if (ctx.bio) { 354afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 355afc51aaaSDarrick J. Wong WARN_ON_ONCE(!ctx.cur_page_in_bio); 356afc51aaaSDarrick J. Wong } else { 357afc51aaaSDarrick J. Wong WARN_ON_ONCE(ctx.cur_page_in_bio); 358afc51aaaSDarrick J. Wong unlock_page(page); 359afc51aaaSDarrick J. Wong } 360afc51aaaSDarrick J. Wong 361afc51aaaSDarrick J. Wong /* 362afc51aaaSDarrick J. Wong * Just like mpage_readpages and block_read_full_page we always 363afc51aaaSDarrick J. Wong * return 0 and just mark the page as PageError on errors. This 364afc51aaaSDarrick J. Wong * should be cleaned up all through the stack eventually. 365afc51aaaSDarrick J. Wong */ 366afc51aaaSDarrick J. Wong return 0; 367afc51aaaSDarrick J. Wong } 368afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_readpage); 369afc51aaaSDarrick J. Wong 370afc51aaaSDarrick J. Wong static struct page * 371afc51aaaSDarrick J. Wong iomap_next_page(struct inode *inode, struct list_head *pages, loff_t pos, 372afc51aaaSDarrick J. Wong loff_t length, loff_t *done) 373afc51aaaSDarrick J. Wong { 374afc51aaaSDarrick J. Wong while (!list_empty(pages)) { 375afc51aaaSDarrick J. Wong struct page *page = lru_to_page(pages); 376afc51aaaSDarrick J. Wong 377afc51aaaSDarrick J. Wong if (page_offset(page) >= (u64)pos + length) 378afc51aaaSDarrick J. Wong break; 379afc51aaaSDarrick J. Wong 380afc51aaaSDarrick J. Wong list_del(&page->lru); 381afc51aaaSDarrick J. Wong if (!add_to_page_cache_lru(page, inode->i_mapping, page->index, 382afc51aaaSDarrick J. Wong GFP_NOFS)) 383afc51aaaSDarrick J. Wong return page; 384afc51aaaSDarrick J. Wong 385afc51aaaSDarrick J. Wong /* 386afc51aaaSDarrick J. Wong * If we already have a page in the page cache at index we are 387afc51aaaSDarrick J. Wong * done. Upper layers don't care if it is uptodate after the 388afc51aaaSDarrick J. Wong * readpages call itself as every page gets checked again once 389afc51aaaSDarrick J. Wong * actually needed. 390afc51aaaSDarrick J. Wong */ 391afc51aaaSDarrick J. Wong *done += PAGE_SIZE; 392afc51aaaSDarrick J. Wong put_page(page); 393afc51aaaSDarrick J. Wong } 394afc51aaaSDarrick J. Wong 395afc51aaaSDarrick J. Wong return NULL; 396afc51aaaSDarrick J. Wong } 397afc51aaaSDarrick J. Wong 398afc51aaaSDarrick J. Wong static loff_t 399afc51aaaSDarrick J. Wong iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, 400c039b997SGoldwyn Rodrigues void *data, struct iomap *iomap, struct iomap *srcmap) 401afc51aaaSDarrick J. Wong { 402afc51aaaSDarrick J. Wong struct iomap_readpage_ctx *ctx = data; 403afc51aaaSDarrick J. Wong loff_t done, ret; 404afc51aaaSDarrick J. Wong 405afc51aaaSDarrick J. Wong for (done = 0; done < length; done += ret) { 406afc51aaaSDarrick J. Wong if (ctx->cur_page && offset_in_page(pos + done) == 0) { 407afc51aaaSDarrick J. Wong if (!ctx->cur_page_in_bio) 408afc51aaaSDarrick J. Wong unlock_page(ctx->cur_page); 409afc51aaaSDarrick J. Wong put_page(ctx->cur_page); 410afc51aaaSDarrick J. Wong ctx->cur_page = NULL; 411afc51aaaSDarrick J. Wong } 412afc51aaaSDarrick J. Wong if (!ctx->cur_page) { 413afc51aaaSDarrick J. Wong ctx->cur_page = iomap_next_page(inode, ctx->pages, 414afc51aaaSDarrick J. Wong pos, length, &done); 415afc51aaaSDarrick J. Wong if (!ctx->cur_page) 416afc51aaaSDarrick J. Wong break; 417afc51aaaSDarrick J. Wong ctx->cur_page_in_bio = false; 418afc51aaaSDarrick J. Wong } 419afc51aaaSDarrick J. Wong ret = iomap_readpage_actor(inode, pos + done, length - done, 420c039b997SGoldwyn Rodrigues ctx, iomap, srcmap); 421afc51aaaSDarrick J. Wong } 422afc51aaaSDarrick J. Wong 423afc51aaaSDarrick J. Wong return done; 424afc51aaaSDarrick J. Wong } 425afc51aaaSDarrick J. Wong 426afc51aaaSDarrick J. Wong int 427afc51aaaSDarrick J. Wong iomap_readpages(struct address_space *mapping, struct list_head *pages, 428afc51aaaSDarrick J. Wong unsigned nr_pages, const struct iomap_ops *ops) 429afc51aaaSDarrick J. Wong { 430afc51aaaSDarrick J. Wong struct iomap_readpage_ctx ctx = { 431afc51aaaSDarrick J. Wong .pages = pages, 432afc51aaaSDarrick J. Wong .is_readahead = true, 433afc51aaaSDarrick J. Wong }; 434afc51aaaSDarrick J. Wong loff_t pos = page_offset(list_entry(pages->prev, struct page, lru)); 435afc51aaaSDarrick J. Wong loff_t last = page_offset(list_entry(pages->next, struct page, lru)); 436afc51aaaSDarrick J. Wong loff_t length = last - pos + PAGE_SIZE, ret = 0; 437afc51aaaSDarrick J. Wong 4389e91c572SChristoph Hellwig trace_iomap_readpages(mapping->host, nr_pages); 4399e91c572SChristoph Hellwig 440afc51aaaSDarrick J. Wong while (length > 0) { 441afc51aaaSDarrick J. Wong ret = iomap_apply(mapping->host, pos, length, 0, ops, 442afc51aaaSDarrick J. Wong &ctx, iomap_readpages_actor); 443afc51aaaSDarrick J. Wong if (ret <= 0) { 444afc51aaaSDarrick J. Wong WARN_ON_ONCE(ret == 0); 445afc51aaaSDarrick J. Wong goto done; 446afc51aaaSDarrick J. Wong } 447afc51aaaSDarrick J. Wong pos += ret; 448afc51aaaSDarrick J. Wong length -= ret; 449afc51aaaSDarrick J. Wong } 450afc51aaaSDarrick J. Wong ret = 0; 451afc51aaaSDarrick J. Wong done: 452afc51aaaSDarrick J. Wong if (ctx.bio) 453afc51aaaSDarrick J. Wong submit_bio(ctx.bio); 454afc51aaaSDarrick J. Wong if (ctx.cur_page) { 455afc51aaaSDarrick J. Wong if (!ctx.cur_page_in_bio) 456afc51aaaSDarrick J. Wong unlock_page(ctx.cur_page); 457afc51aaaSDarrick J. Wong put_page(ctx.cur_page); 458afc51aaaSDarrick J. Wong } 459afc51aaaSDarrick J. Wong 460afc51aaaSDarrick J. Wong /* 461afc51aaaSDarrick J. Wong * Check that we didn't lose a page due to the arcance calling 462afc51aaaSDarrick J. Wong * conventions.. 463afc51aaaSDarrick J. Wong */ 464afc51aaaSDarrick J. Wong WARN_ON_ONCE(!ret && !list_empty(ctx.pages)); 465afc51aaaSDarrick J. Wong return ret; 466afc51aaaSDarrick J. Wong } 467afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_readpages); 468afc51aaaSDarrick J. Wong 469afc51aaaSDarrick J. Wong /* 470afc51aaaSDarrick J. Wong * iomap_is_partially_uptodate checks whether blocks within a page are 471afc51aaaSDarrick J. Wong * uptodate or not. 472afc51aaaSDarrick J. Wong * 473afc51aaaSDarrick J. Wong * Returns true if all blocks which correspond to a file portion 474afc51aaaSDarrick J. Wong * we want to read within the page are uptodate. 475afc51aaaSDarrick J. Wong */ 476afc51aaaSDarrick J. Wong int 477afc51aaaSDarrick J. Wong iomap_is_partially_uptodate(struct page *page, unsigned long from, 478afc51aaaSDarrick J. Wong unsigned long count) 479afc51aaaSDarrick J. Wong { 480afc51aaaSDarrick J. Wong struct iomap_page *iop = to_iomap_page(page); 481afc51aaaSDarrick J. Wong struct inode *inode = page->mapping->host; 482afc51aaaSDarrick J. Wong unsigned len, first, last; 483afc51aaaSDarrick J. Wong unsigned i; 484afc51aaaSDarrick J. Wong 485afc51aaaSDarrick J. Wong /* Limit range to one page */ 486afc51aaaSDarrick J. Wong len = min_t(unsigned, PAGE_SIZE - from, count); 487afc51aaaSDarrick J. Wong 488afc51aaaSDarrick J. Wong /* First and last blocks in range within page */ 489afc51aaaSDarrick J. Wong first = from >> inode->i_blkbits; 490afc51aaaSDarrick J. Wong last = (from + len - 1) >> inode->i_blkbits; 491afc51aaaSDarrick J. Wong 492afc51aaaSDarrick J. Wong if (iop) { 493afc51aaaSDarrick J. Wong for (i = first; i <= last; i++) 494afc51aaaSDarrick J. Wong if (!test_bit(i, iop->uptodate)) 495afc51aaaSDarrick J. Wong return 0; 496afc51aaaSDarrick J. Wong return 1; 497afc51aaaSDarrick J. Wong } 498afc51aaaSDarrick J. Wong 499afc51aaaSDarrick J. Wong return 0; 500afc51aaaSDarrick J. Wong } 501afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate); 502afc51aaaSDarrick J. Wong 503afc51aaaSDarrick J. Wong int 504afc51aaaSDarrick J. Wong iomap_releasepage(struct page *page, gfp_t gfp_mask) 505afc51aaaSDarrick J. Wong { 5069e91c572SChristoph Hellwig trace_iomap_releasepage(page->mapping->host, page, 0, 0); 5079e91c572SChristoph Hellwig 508afc51aaaSDarrick J. Wong /* 509afc51aaaSDarrick J. Wong * mm accommodates an old ext3 case where clean pages might not have had 510afc51aaaSDarrick J. Wong * the dirty bit cleared. Thus, it can send actual dirty pages to 511afc51aaaSDarrick J. Wong * ->releasepage() via shrink_active_list(), skip those here. 512afc51aaaSDarrick J. Wong */ 513afc51aaaSDarrick J. Wong if (PageDirty(page) || PageWriteback(page)) 514afc51aaaSDarrick J. Wong return 0; 515afc51aaaSDarrick J. Wong iomap_page_release(page); 516afc51aaaSDarrick J. Wong return 1; 517afc51aaaSDarrick J. Wong } 518afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_releasepage); 519afc51aaaSDarrick J. Wong 520afc51aaaSDarrick J. Wong void 521afc51aaaSDarrick J. Wong iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len) 522afc51aaaSDarrick J. Wong { 5239e91c572SChristoph Hellwig trace_iomap_invalidatepage(page->mapping->host, page, offset, len); 5249e91c572SChristoph Hellwig 525afc51aaaSDarrick J. Wong /* 526afc51aaaSDarrick J. Wong * If we are invalidating the entire page, clear the dirty state from it 527afc51aaaSDarrick J. Wong * and release it to avoid unnecessary buildup of the LRU. 528afc51aaaSDarrick J. Wong */ 529afc51aaaSDarrick J. Wong if (offset == 0 && len == PAGE_SIZE) { 530afc51aaaSDarrick J. Wong WARN_ON_ONCE(PageWriteback(page)); 531afc51aaaSDarrick J. Wong cancel_dirty_page(page); 532afc51aaaSDarrick J. Wong iomap_page_release(page); 533afc51aaaSDarrick J. Wong } 534afc51aaaSDarrick J. Wong } 535afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_invalidatepage); 536afc51aaaSDarrick J. Wong 537afc51aaaSDarrick J. Wong #ifdef CONFIG_MIGRATION 538afc51aaaSDarrick J. Wong int 539afc51aaaSDarrick J. Wong iomap_migrate_page(struct address_space *mapping, struct page *newpage, 540afc51aaaSDarrick J. Wong struct page *page, enum migrate_mode mode) 541afc51aaaSDarrick J. Wong { 542afc51aaaSDarrick J. Wong int ret; 543afc51aaaSDarrick J. Wong 54426473f83SLinus Torvalds ret = migrate_page_move_mapping(mapping, newpage, page, 0); 545afc51aaaSDarrick J. Wong if (ret != MIGRATEPAGE_SUCCESS) 546afc51aaaSDarrick J. Wong return ret; 547afc51aaaSDarrick J. Wong 548afc51aaaSDarrick J. Wong if (page_has_private(page)) { 549afc51aaaSDarrick J. Wong ClearPagePrivate(page); 550afc51aaaSDarrick J. Wong get_page(newpage); 551afc51aaaSDarrick J. Wong set_page_private(newpage, page_private(page)); 552afc51aaaSDarrick J. Wong set_page_private(page, 0); 553afc51aaaSDarrick J. Wong put_page(page); 554afc51aaaSDarrick J. Wong SetPagePrivate(newpage); 555afc51aaaSDarrick J. Wong } 556afc51aaaSDarrick J. Wong 557afc51aaaSDarrick J. Wong if (mode != MIGRATE_SYNC_NO_COPY) 558afc51aaaSDarrick J. Wong migrate_page_copy(newpage, page); 559afc51aaaSDarrick J. Wong else 560afc51aaaSDarrick J. Wong migrate_page_states(newpage, page); 561afc51aaaSDarrick J. Wong return MIGRATEPAGE_SUCCESS; 562afc51aaaSDarrick J. Wong } 563afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_migrate_page); 564afc51aaaSDarrick J. Wong #endif /* CONFIG_MIGRATION */ 565afc51aaaSDarrick J. Wong 56632a38a49SChristoph Hellwig enum { 56732a38a49SChristoph Hellwig IOMAP_WRITE_F_UNSHARE = (1 << 0), 56832a38a49SChristoph Hellwig }; 56932a38a49SChristoph Hellwig 570afc51aaaSDarrick J. Wong static void 571afc51aaaSDarrick J. Wong iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) 572afc51aaaSDarrick J. Wong { 573afc51aaaSDarrick J. Wong loff_t i_size = i_size_read(inode); 574afc51aaaSDarrick J. Wong 575afc51aaaSDarrick J. Wong /* 576afc51aaaSDarrick J. Wong * Only truncate newly allocated pages beyoned EOF, even if the 577afc51aaaSDarrick J. Wong * write started inside the existing inode size. 578afc51aaaSDarrick J. Wong */ 579afc51aaaSDarrick J. Wong if (pos + len > i_size) 580afc51aaaSDarrick J. Wong truncate_pagecache_range(inode, max(pos, i_size), pos + len); 581afc51aaaSDarrick J. Wong } 582afc51aaaSDarrick J. Wong 583afc51aaaSDarrick J. Wong static int 584d3b40439SChristoph Hellwig iomap_read_page_sync(loff_t block_start, struct page *page, unsigned poff, 585d3b40439SChristoph Hellwig unsigned plen, struct iomap *iomap) 586afc51aaaSDarrick J. Wong { 587afc51aaaSDarrick J. Wong struct bio_vec bvec; 588afc51aaaSDarrick J. Wong struct bio bio; 589afc51aaaSDarrick J. Wong 590afc51aaaSDarrick J. Wong bio_init(&bio, &bvec, 1); 591afc51aaaSDarrick J. Wong bio.bi_opf = REQ_OP_READ; 592afc51aaaSDarrick J. Wong bio.bi_iter.bi_sector = iomap_sector(iomap, block_start); 593afc51aaaSDarrick J. Wong bio_set_dev(&bio, iomap->bdev); 594afc51aaaSDarrick J. Wong __bio_add_page(&bio, page, plen, poff); 595afc51aaaSDarrick J. Wong return submit_bio_wait(&bio); 596afc51aaaSDarrick J. Wong } 597afc51aaaSDarrick J. Wong 598afc51aaaSDarrick J. Wong static int 59932a38a49SChristoph Hellwig __iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, int flags, 600c039b997SGoldwyn Rodrigues struct page *page, struct iomap *srcmap) 601afc51aaaSDarrick J. Wong { 602afc51aaaSDarrick J. Wong struct iomap_page *iop = iomap_page_create(inode, page); 603afc51aaaSDarrick J. Wong loff_t block_size = i_blocksize(inode); 604afc51aaaSDarrick J. Wong loff_t block_start = pos & ~(block_size - 1); 605afc51aaaSDarrick J. Wong loff_t block_end = (pos + len + block_size - 1) & ~(block_size - 1); 606afc51aaaSDarrick J. Wong unsigned from = offset_in_page(pos), to = from + len, poff, plen; 607d3b40439SChristoph Hellwig int status; 608afc51aaaSDarrick J. Wong 609afc51aaaSDarrick J. Wong if (PageUptodate(page)) 610afc51aaaSDarrick J. Wong return 0; 611afc51aaaSDarrick J. Wong 612afc51aaaSDarrick J. Wong do { 613afc51aaaSDarrick J. Wong iomap_adjust_read_range(inode, iop, &block_start, 614afc51aaaSDarrick J. Wong block_end - block_start, &poff, &plen); 615afc51aaaSDarrick J. Wong if (plen == 0) 616afc51aaaSDarrick J. Wong break; 617afc51aaaSDarrick J. Wong 61832a38a49SChristoph Hellwig if (!(flags & IOMAP_WRITE_F_UNSHARE) && 61932a38a49SChristoph Hellwig (from <= poff || from >= poff + plen) && 620d3b40439SChristoph Hellwig (to <= poff || to >= poff + plen)) 621d3b40439SChristoph Hellwig continue; 622d3b40439SChristoph Hellwig 623c039b997SGoldwyn Rodrigues if (iomap_block_needs_zeroing(inode, srcmap, block_start)) { 62432a38a49SChristoph Hellwig if (WARN_ON_ONCE(flags & IOMAP_WRITE_F_UNSHARE)) 62532a38a49SChristoph Hellwig return -EIO; 626d3b40439SChristoph Hellwig zero_user_segments(page, poff, from, to, poff + plen); 627d3b40439SChristoph Hellwig iomap_set_range_uptodate(page, poff, plen); 628d3b40439SChristoph Hellwig continue; 629afc51aaaSDarrick J. Wong } 630afc51aaaSDarrick J. Wong 631d3b40439SChristoph Hellwig status = iomap_read_page_sync(block_start, page, poff, plen, 632c039b997SGoldwyn Rodrigues srcmap); 633d3b40439SChristoph Hellwig if (status) 634d3b40439SChristoph Hellwig return status; 635afc51aaaSDarrick J. Wong } while ((block_start += plen) < block_end); 636afc51aaaSDarrick J. Wong 637d3b40439SChristoph Hellwig return 0; 638afc51aaaSDarrick J. Wong } 639afc51aaaSDarrick J. Wong 640afc51aaaSDarrick J. Wong static int 641afc51aaaSDarrick J. Wong iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags, 642c039b997SGoldwyn Rodrigues struct page **pagep, struct iomap *iomap, struct iomap *srcmap) 643afc51aaaSDarrick J. Wong { 644afc51aaaSDarrick J. Wong const struct iomap_page_ops *page_ops = iomap->page_ops; 645afc51aaaSDarrick J. Wong struct page *page; 646afc51aaaSDarrick J. Wong int status = 0; 647afc51aaaSDarrick J. Wong 648afc51aaaSDarrick J. Wong BUG_ON(pos + len > iomap->offset + iomap->length); 649c039b997SGoldwyn Rodrigues if (srcmap != iomap) 650c039b997SGoldwyn Rodrigues BUG_ON(pos + len > srcmap->offset + srcmap->length); 651afc51aaaSDarrick J. Wong 652afc51aaaSDarrick J. Wong if (fatal_signal_pending(current)) 653afc51aaaSDarrick J. Wong return -EINTR; 654afc51aaaSDarrick J. Wong 655afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_prepare) { 656afc51aaaSDarrick J. Wong status = page_ops->page_prepare(inode, pos, len, iomap); 657afc51aaaSDarrick J. Wong if (status) 658afc51aaaSDarrick J. Wong return status; 659afc51aaaSDarrick J. Wong } 660afc51aaaSDarrick J. Wong 661dcd6158dSChristoph Hellwig page = grab_cache_page_write_begin(inode->i_mapping, pos >> PAGE_SHIFT, 662dcd6158dSChristoph Hellwig AOP_FLAG_NOFS); 663afc51aaaSDarrick J. Wong if (!page) { 664afc51aaaSDarrick J. Wong status = -ENOMEM; 665afc51aaaSDarrick J. Wong goto out_no_page; 666afc51aaaSDarrick J. Wong } 667afc51aaaSDarrick J. Wong 668c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) 669c039b997SGoldwyn Rodrigues iomap_read_inline_data(inode, page, srcmap); 670afc51aaaSDarrick J. Wong else if (iomap->flags & IOMAP_F_BUFFER_HEAD) 671c039b997SGoldwyn Rodrigues status = __block_write_begin_int(page, pos, len, NULL, srcmap); 672afc51aaaSDarrick J. Wong else 67332a38a49SChristoph Hellwig status = __iomap_write_begin(inode, pos, len, flags, page, 674c039b997SGoldwyn Rodrigues srcmap); 675afc51aaaSDarrick J. Wong 676afc51aaaSDarrick J. Wong if (unlikely(status)) 677afc51aaaSDarrick J. Wong goto out_unlock; 678afc51aaaSDarrick J. Wong 679afc51aaaSDarrick J. Wong *pagep = page; 680afc51aaaSDarrick J. Wong return 0; 681afc51aaaSDarrick J. Wong 682afc51aaaSDarrick J. Wong out_unlock: 683afc51aaaSDarrick J. Wong unlock_page(page); 684afc51aaaSDarrick J. Wong put_page(page); 685afc51aaaSDarrick J. Wong iomap_write_failed(inode, pos, len); 686afc51aaaSDarrick J. Wong 687afc51aaaSDarrick J. Wong out_no_page: 688afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 689afc51aaaSDarrick J. Wong page_ops->page_done(inode, pos, 0, NULL, iomap); 690afc51aaaSDarrick J. Wong return status; 691afc51aaaSDarrick J. Wong } 692afc51aaaSDarrick J. Wong 693afc51aaaSDarrick J. Wong int 694afc51aaaSDarrick J. Wong iomap_set_page_dirty(struct page *page) 695afc51aaaSDarrick J. Wong { 696afc51aaaSDarrick J. Wong struct address_space *mapping = page_mapping(page); 697afc51aaaSDarrick J. Wong int newly_dirty; 698afc51aaaSDarrick J. Wong 699afc51aaaSDarrick J. Wong if (unlikely(!mapping)) 700afc51aaaSDarrick J. Wong return !TestSetPageDirty(page); 701afc51aaaSDarrick J. Wong 702afc51aaaSDarrick J. Wong /* 703afc51aaaSDarrick J. Wong * Lock out page->mem_cgroup migration to keep PageDirty 704afc51aaaSDarrick J. Wong * synchronized with per-memcg dirty page counters. 705afc51aaaSDarrick J. Wong */ 706afc51aaaSDarrick J. Wong lock_page_memcg(page); 707afc51aaaSDarrick J. Wong newly_dirty = !TestSetPageDirty(page); 708afc51aaaSDarrick J. Wong if (newly_dirty) 709afc51aaaSDarrick J. Wong __set_page_dirty(page, mapping, 0); 710afc51aaaSDarrick J. Wong unlock_page_memcg(page); 711afc51aaaSDarrick J. Wong 712afc51aaaSDarrick J. Wong if (newly_dirty) 713afc51aaaSDarrick J. Wong __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 714afc51aaaSDarrick J. Wong return newly_dirty; 715afc51aaaSDarrick J. Wong } 716afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_set_page_dirty); 717afc51aaaSDarrick J. Wong 718afc51aaaSDarrick J. Wong static int 719afc51aaaSDarrick J. Wong __iomap_write_end(struct inode *inode, loff_t pos, unsigned len, 720c12d6fa8SChristoph Hellwig unsigned copied, struct page *page) 721afc51aaaSDarrick J. Wong { 722afc51aaaSDarrick J. Wong flush_dcache_page(page); 723afc51aaaSDarrick J. Wong 724afc51aaaSDarrick J. Wong /* 725afc51aaaSDarrick J. Wong * The blocks that were entirely written will now be uptodate, so we 726afc51aaaSDarrick J. Wong * don't have to worry about a readpage reading them and overwriting a 727afc51aaaSDarrick J. Wong * partial write. However if we have encountered a short write and only 728afc51aaaSDarrick J. Wong * partially written into a block, it will not be marked uptodate, so a 729afc51aaaSDarrick J. Wong * readpage might come in and destroy our partial write. 730afc51aaaSDarrick J. Wong * 731afc51aaaSDarrick J. Wong * Do the simplest thing, and just treat any short write to a non 732afc51aaaSDarrick J. Wong * uptodate page as a zero-length write, and force the caller to redo 733afc51aaaSDarrick J. Wong * the whole thing. 734afc51aaaSDarrick J. Wong */ 735afc51aaaSDarrick J. Wong if (unlikely(copied < len && !PageUptodate(page))) 736afc51aaaSDarrick J. Wong return 0; 737afc51aaaSDarrick J. Wong iomap_set_range_uptodate(page, offset_in_page(pos), len); 738afc51aaaSDarrick J. Wong iomap_set_page_dirty(page); 739afc51aaaSDarrick J. Wong return copied; 740afc51aaaSDarrick J. Wong } 741afc51aaaSDarrick J. Wong 742afc51aaaSDarrick J. Wong static int 743afc51aaaSDarrick J. Wong iomap_write_end_inline(struct inode *inode, struct page *page, 744afc51aaaSDarrick J. Wong struct iomap *iomap, loff_t pos, unsigned copied) 745afc51aaaSDarrick J. Wong { 746afc51aaaSDarrick J. Wong void *addr; 747afc51aaaSDarrick J. Wong 748afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 749afc51aaaSDarrick J. Wong BUG_ON(pos + copied > PAGE_SIZE - offset_in_page(iomap->inline_data)); 750afc51aaaSDarrick J. Wong 751afc51aaaSDarrick J. Wong addr = kmap_atomic(page); 752afc51aaaSDarrick J. Wong memcpy(iomap->inline_data + pos, addr + pos, copied); 753afc51aaaSDarrick J. Wong kunmap_atomic(addr); 754afc51aaaSDarrick J. Wong 755afc51aaaSDarrick J. Wong mark_inode_dirty(inode); 756afc51aaaSDarrick J. Wong return copied; 757afc51aaaSDarrick J. Wong } 758afc51aaaSDarrick J. Wong 759afc51aaaSDarrick J. Wong static int 760c039b997SGoldwyn Rodrigues iomap_write_end(struct inode *inode, loff_t pos, unsigned len, unsigned copied, 761c039b997SGoldwyn Rodrigues struct page *page, struct iomap *iomap, struct iomap *srcmap) 762afc51aaaSDarrick J. Wong { 763afc51aaaSDarrick J. Wong const struct iomap_page_ops *page_ops = iomap->page_ops; 764afc51aaaSDarrick J. Wong loff_t old_size = inode->i_size; 765afc51aaaSDarrick J. Wong int ret; 766afc51aaaSDarrick J. Wong 767c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_INLINE) { 768afc51aaaSDarrick J. Wong ret = iomap_write_end_inline(inode, page, iomap, pos, copied); 769c039b997SGoldwyn Rodrigues } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) { 770afc51aaaSDarrick J. Wong ret = block_write_end(NULL, inode->i_mapping, pos, len, copied, 771afc51aaaSDarrick J. Wong page, NULL); 772afc51aaaSDarrick J. Wong } else { 773c12d6fa8SChristoph Hellwig ret = __iomap_write_end(inode, pos, len, copied, page); 774afc51aaaSDarrick J. Wong } 775afc51aaaSDarrick J. Wong 776afc51aaaSDarrick J. Wong /* 777afc51aaaSDarrick J. Wong * Update the in-memory inode size after copying the data into the page 778afc51aaaSDarrick J. Wong * cache. It's up to the file system to write the updated size to disk, 779afc51aaaSDarrick J. Wong * preferably after I/O completion so that no stale data is exposed. 780afc51aaaSDarrick J. Wong */ 781afc51aaaSDarrick J. Wong if (pos + ret > old_size) { 782afc51aaaSDarrick J. Wong i_size_write(inode, pos + ret); 783afc51aaaSDarrick J. Wong iomap->flags |= IOMAP_F_SIZE_CHANGED; 784afc51aaaSDarrick J. Wong } 785afc51aaaSDarrick J. Wong unlock_page(page); 786afc51aaaSDarrick J. Wong 787afc51aaaSDarrick J. Wong if (old_size < pos) 788afc51aaaSDarrick J. Wong pagecache_isize_extended(inode, old_size, pos); 789afc51aaaSDarrick J. Wong if (page_ops && page_ops->page_done) 790afc51aaaSDarrick J. Wong page_ops->page_done(inode, pos, ret, page, iomap); 791afc51aaaSDarrick J. Wong put_page(page); 792afc51aaaSDarrick J. Wong 793afc51aaaSDarrick J. Wong if (ret < len) 794afc51aaaSDarrick J. Wong iomap_write_failed(inode, pos, len); 795afc51aaaSDarrick J. Wong return ret; 796afc51aaaSDarrick J. Wong } 797afc51aaaSDarrick J. Wong 798afc51aaaSDarrick J. Wong static loff_t 799afc51aaaSDarrick J. Wong iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 800c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 801afc51aaaSDarrick J. Wong { 802afc51aaaSDarrick J. Wong struct iov_iter *i = data; 803afc51aaaSDarrick J. Wong long status = 0; 804afc51aaaSDarrick J. Wong ssize_t written = 0; 805afc51aaaSDarrick J. Wong 806afc51aaaSDarrick J. Wong do { 807afc51aaaSDarrick J. Wong struct page *page; 808afc51aaaSDarrick J. Wong unsigned long offset; /* Offset into pagecache page */ 809afc51aaaSDarrick J. Wong unsigned long bytes; /* Bytes to write to page */ 810afc51aaaSDarrick J. Wong size_t copied; /* Bytes copied from user */ 811afc51aaaSDarrick J. Wong 812afc51aaaSDarrick J. Wong offset = offset_in_page(pos); 813afc51aaaSDarrick J. Wong bytes = min_t(unsigned long, PAGE_SIZE - offset, 814afc51aaaSDarrick J. Wong iov_iter_count(i)); 815afc51aaaSDarrick J. Wong again: 816afc51aaaSDarrick J. Wong if (bytes > length) 817afc51aaaSDarrick J. Wong bytes = length; 818afc51aaaSDarrick J. Wong 819afc51aaaSDarrick J. Wong /* 820afc51aaaSDarrick J. Wong * Bring in the user page that we will copy from _first_. 821afc51aaaSDarrick J. Wong * Otherwise there's a nasty deadlock on copying from the 822afc51aaaSDarrick J. Wong * same page as we're writing to, without it being marked 823afc51aaaSDarrick J. Wong * up-to-date. 824afc51aaaSDarrick J. Wong * 825afc51aaaSDarrick J. Wong * Not only is this an optimisation, but it is also required 826afc51aaaSDarrick J. Wong * to check that the address is actually valid, when atomic 827afc51aaaSDarrick J. Wong * usercopies are used, below. 828afc51aaaSDarrick J. Wong */ 829afc51aaaSDarrick J. Wong if (unlikely(iov_iter_fault_in_readable(i, bytes))) { 830afc51aaaSDarrick J. Wong status = -EFAULT; 831afc51aaaSDarrick J. Wong break; 832afc51aaaSDarrick J. Wong } 833afc51aaaSDarrick J. Wong 834c039b997SGoldwyn Rodrigues status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, 835c039b997SGoldwyn Rodrigues srcmap); 836afc51aaaSDarrick J. Wong if (unlikely(status)) 837afc51aaaSDarrick J. Wong break; 838afc51aaaSDarrick J. Wong 839afc51aaaSDarrick J. Wong if (mapping_writably_mapped(inode->i_mapping)) 840afc51aaaSDarrick J. Wong flush_dcache_page(page); 841afc51aaaSDarrick J. Wong 842afc51aaaSDarrick J. Wong copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); 843afc51aaaSDarrick J. Wong 844afc51aaaSDarrick J. Wong flush_dcache_page(page); 845afc51aaaSDarrick J. Wong 846c039b997SGoldwyn Rodrigues status = iomap_write_end(inode, pos, bytes, copied, page, iomap, 847c039b997SGoldwyn Rodrigues srcmap); 848afc51aaaSDarrick J. Wong if (unlikely(status < 0)) 849afc51aaaSDarrick J. Wong break; 850afc51aaaSDarrick J. Wong copied = status; 851afc51aaaSDarrick J. Wong 852afc51aaaSDarrick J. Wong cond_resched(); 853afc51aaaSDarrick J. Wong 854afc51aaaSDarrick J. Wong iov_iter_advance(i, copied); 855afc51aaaSDarrick J. Wong if (unlikely(copied == 0)) { 856afc51aaaSDarrick J. Wong /* 857afc51aaaSDarrick J. Wong * If we were unable to copy any data at all, we must 858afc51aaaSDarrick J. Wong * fall back to a single segment length write. 859afc51aaaSDarrick J. Wong * 860afc51aaaSDarrick J. Wong * If we didn't fallback here, we could livelock 861afc51aaaSDarrick J. Wong * because not all segments in the iov can be copied at 862afc51aaaSDarrick J. Wong * once without a pagefault. 863afc51aaaSDarrick J. Wong */ 864afc51aaaSDarrick J. Wong bytes = min_t(unsigned long, PAGE_SIZE - offset, 865afc51aaaSDarrick J. Wong iov_iter_single_seg_count(i)); 866afc51aaaSDarrick J. Wong goto again; 867afc51aaaSDarrick J. Wong } 868afc51aaaSDarrick J. Wong pos += copied; 869afc51aaaSDarrick J. Wong written += copied; 870afc51aaaSDarrick J. Wong length -= copied; 871afc51aaaSDarrick J. Wong 872afc51aaaSDarrick J. Wong balance_dirty_pages_ratelimited(inode->i_mapping); 873afc51aaaSDarrick J. Wong } while (iov_iter_count(i) && length); 874afc51aaaSDarrick J. Wong 875afc51aaaSDarrick J. Wong return written ? written : status; 876afc51aaaSDarrick J. Wong } 877afc51aaaSDarrick J. Wong 878afc51aaaSDarrick J. Wong ssize_t 879afc51aaaSDarrick J. Wong iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter, 880afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 881afc51aaaSDarrick J. Wong { 882afc51aaaSDarrick J. Wong struct inode *inode = iocb->ki_filp->f_mapping->host; 883afc51aaaSDarrick J. Wong loff_t pos = iocb->ki_pos, ret = 0, written = 0; 884afc51aaaSDarrick J. Wong 885afc51aaaSDarrick J. Wong while (iov_iter_count(iter)) { 886afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, iov_iter_count(iter), 887afc51aaaSDarrick J. Wong IOMAP_WRITE, ops, iter, iomap_write_actor); 888afc51aaaSDarrick J. Wong if (ret <= 0) 889afc51aaaSDarrick J. Wong break; 890afc51aaaSDarrick J. Wong pos += ret; 891afc51aaaSDarrick J. Wong written += ret; 892afc51aaaSDarrick J. Wong } 893afc51aaaSDarrick J. Wong 894afc51aaaSDarrick J. Wong return written ? written : ret; 895afc51aaaSDarrick J. Wong } 896afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_file_buffered_write); 897afc51aaaSDarrick J. Wong 898afc51aaaSDarrick J. Wong static loff_t 8993590c4d8SChristoph Hellwig iomap_unshare_actor(struct inode *inode, loff_t pos, loff_t length, void *data, 900c039b997SGoldwyn Rodrigues struct iomap *iomap, struct iomap *srcmap) 901afc51aaaSDarrick J. Wong { 902afc51aaaSDarrick J. Wong long status = 0; 903afc51aaaSDarrick J. Wong ssize_t written = 0; 904afc51aaaSDarrick J. Wong 9053590c4d8SChristoph Hellwig /* don't bother with blocks that are not shared to start with */ 9063590c4d8SChristoph Hellwig if (!(iomap->flags & IOMAP_F_SHARED)) 9073590c4d8SChristoph Hellwig return length; 9083590c4d8SChristoph Hellwig /* don't bother with holes or unwritten extents */ 909c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 9103590c4d8SChristoph Hellwig return length; 9113590c4d8SChristoph Hellwig 912afc51aaaSDarrick J. Wong do { 91332a38a49SChristoph Hellwig unsigned long offset = offset_in_page(pos); 91432a38a49SChristoph Hellwig unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length); 91532a38a49SChristoph Hellwig struct page *page; 916afc51aaaSDarrick J. Wong 91732a38a49SChristoph Hellwig status = iomap_write_begin(inode, pos, bytes, 918c039b997SGoldwyn Rodrigues IOMAP_WRITE_F_UNSHARE, &page, iomap, srcmap); 919afc51aaaSDarrick J. Wong if (unlikely(status)) 920afc51aaaSDarrick J. Wong return status; 921afc51aaaSDarrick J. Wong 922c039b997SGoldwyn Rodrigues status = iomap_write_end(inode, pos, bytes, bytes, page, iomap, 923c039b997SGoldwyn Rodrigues srcmap); 924afc51aaaSDarrick J. Wong if (unlikely(status <= 0)) { 925afc51aaaSDarrick J. Wong if (WARN_ON_ONCE(status == 0)) 926afc51aaaSDarrick J. Wong return -EIO; 927afc51aaaSDarrick J. Wong return status; 928afc51aaaSDarrick J. Wong } 929afc51aaaSDarrick J. Wong 930afc51aaaSDarrick J. Wong cond_resched(); 931afc51aaaSDarrick J. Wong 932afc51aaaSDarrick J. Wong pos += status; 933afc51aaaSDarrick J. Wong written += status; 934afc51aaaSDarrick J. Wong length -= status; 935afc51aaaSDarrick J. Wong 936afc51aaaSDarrick J. Wong balance_dirty_pages_ratelimited(inode->i_mapping); 937afc51aaaSDarrick J. Wong } while (length); 938afc51aaaSDarrick J. Wong 939afc51aaaSDarrick J. Wong return written; 940afc51aaaSDarrick J. Wong } 941afc51aaaSDarrick J. Wong 942afc51aaaSDarrick J. Wong int 9433590c4d8SChristoph Hellwig iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, 944afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 945afc51aaaSDarrick J. Wong { 946afc51aaaSDarrick J. Wong loff_t ret; 947afc51aaaSDarrick J. Wong 948afc51aaaSDarrick J. Wong while (len) { 949afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL, 9503590c4d8SChristoph Hellwig iomap_unshare_actor); 951afc51aaaSDarrick J. Wong if (ret <= 0) 952afc51aaaSDarrick J. Wong return ret; 953afc51aaaSDarrick J. Wong pos += ret; 954afc51aaaSDarrick J. Wong len -= ret; 955afc51aaaSDarrick J. Wong } 956afc51aaaSDarrick J. Wong 957afc51aaaSDarrick J. Wong return 0; 958afc51aaaSDarrick J. Wong } 9593590c4d8SChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_file_unshare); 960afc51aaaSDarrick J. Wong 961afc51aaaSDarrick J. Wong static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset, 962c039b997SGoldwyn Rodrigues unsigned bytes, struct iomap *iomap, struct iomap *srcmap) 963afc51aaaSDarrick J. Wong { 964afc51aaaSDarrick J. Wong struct page *page; 965afc51aaaSDarrick J. Wong int status; 966afc51aaaSDarrick J. Wong 967c039b997SGoldwyn Rodrigues status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, srcmap); 968afc51aaaSDarrick J. Wong if (status) 969afc51aaaSDarrick J. Wong return status; 970afc51aaaSDarrick J. Wong 971afc51aaaSDarrick J. Wong zero_user(page, offset, bytes); 972afc51aaaSDarrick J. Wong mark_page_accessed(page); 973afc51aaaSDarrick J. Wong 974c039b997SGoldwyn Rodrigues return iomap_write_end(inode, pos, bytes, bytes, page, iomap, srcmap); 975afc51aaaSDarrick J. Wong } 976afc51aaaSDarrick J. Wong 977afc51aaaSDarrick J. Wong static int iomap_dax_zero(loff_t pos, unsigned offset, unsigned bytes, 978afc51aaaSDarrick J. Wong struct iomap *iomap) 979afc51aaaSDarrick J. Wong { 980afc51aaaSDarrick J. Wong return __dax_zero_page_range(iomap->bdev, iomap->dax_dev, 981afc51aaaSDarrick J. Wong iomap_sector(iomap, pos & PAGE_MASK), offset, bytes); 982afc51aaaSDarrick J. Wong } 983afc51aaaSDarrick J. Wong 984afc51aaaSDarrick J. Wong static loff_t 985afc51aaaSDarrick J. Wong iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count, 986c039b997SGoldwyn Rodrigues void *data, struct iomap *iomap, struct iomap *srcmap) 987afc51aaaSDarrick J. Wong { 988afc51aaaSDarrick J. Wong bool *did_zero = data; 989afc51aaaSDarrick J. Wong loff_t written = 0; 990afc51aaaSDarrick J. Wong int status; 991afc51aaaSDarrick J. Wong 992afc51aaaSDarrick J. Wong /* already zeroed? we're done. */ 993c039b997SGoldwyn Rodrigues if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) 994afc51aaaSDarrick J. Wong return count; 995afc51aaaSDarrick J. Wong 996afc51aaaSDarrick J. Wong do { 997afc51aaaSDarrick J. Wong unsigned offset, bytes; 998afc51aaaSDarrick J. Wong 999afc51aaaSDarrick J. Wong offset = offset_in_page(pos); 1000afc51aaaSDarrick J. Wong bytes = min_t(loff_t, PAGE_SIZE - offset, count); 1001afc51aaaSDarrick J. Wong 1002afc51aaaSDarrick J. Wong if (IS_DAX(inode)) 1003afc51aaaSDarrick J. Wong status = iomap_dax_zero(pos, offset, bytes, iomap); 1004afc51aaaSDarrick J. Wong else 1005c039b997SGoldwyn Rodrigues status = iomap_zero(inode, pos, offset, bytes, iomap, 1006c039b997SGoldwyn Rodrigues srcmap); 1007afc51aaaSDarrick J. Wong if (status < 0) 1008afc51aaaSDarrick J. Wong return status; 1009afc51aaaSDarrick J. Wong 1010afc51aaaSDarrick J. Wong pos += bytes; 1011afc51aaaSDarrick J. Wong count -= bytes; 1012afc51aaaSDarrick J. Wong written += bytes; 1013afc51aaaSDarrick J. Wong if (did_zero) 1014afc51aaaSDarrick J. Wong *did_zero = true; 1015afc51aaaSDarrick J. Wong } while (count > 0); 1016afc51aaaSDarrick J. Wong 1017afc51aaaSDarrick J. Wong return written; 1018afc51aaaSDarrick J. Wong } 1019afc51aaaSDarrick J. Wong 1020afc51aaaSDarrick J. Wong int 1021afc51aaaSDarrick J. Wong iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, 1022afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 1023afc51aaaSDarrick J. Wong { 1024afc51aaaSDarrick J. Wong loff_t ret; 1025afc51aaaSDarrick J. Wong 1026afc51aaaSDarrick J. Wong while (len > 0) { 1027afc51aaaSDarrick J. Wong ret = iomap_apply(inode, pos, len, IOMAP_ZERO, 1028afc51aaaSDarrick J. Wong ops, did_zero, iomap_zero_range_actor); 1029afc51aaaSDarrick J. Wong if (ret <= 0) 1030afc51aaaSDarrick J. Wong return ret; 1031afc51aaaSDarrick J. Wong 1032afc51aaaSDarrick J. Wong pos += ret; 1033afc51aaaSDarrick J. Wong len -= ret; 1034afc51aaaSDarrick J. Wong } 1035afc51aaaSDarrick J. Wong 1036afc51aaaSDarrick J. Wong return 0; 1037afc51aaaSDarrick J. Wong } 1038afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_zero_range); 1039afc51aaaSDarrick J. Wong 1040afc51aaaSDarrick J. Wong int 1041afc51aaaSDarrick J. Wong iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, 1042afc51aaaSDarrick J. Wong const struct iomap_ops *ops) 1043afc51aaaSDarrick J. Wong { 1044afc51aaaSDarrick J. Wong unsigned int blocksize = i_blocksize(inode); 1045afc51aaaSDarrick J. Wong unsigned int off = pos & (blocksize - 1); 1046afc51aaaSDarrick J. Wong 1047afc51aaaSDarrick J. Wong /* Block boundary? Nothing to do */ 1048afc51aaaSDarrick J. Wong if (!off) 1049afc51aaaSDarrick J. Wong return 0; 1050afc51aaaSDarrick J. Wong return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); 1051afc51aaaSDarrick J. Wong } 1052afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_truncate_page); 1053afc51aaaSDarrick J. Wong 1054afc51aaaSDarrick J. Wong static loff_t 1055afc51aaaSDarrick J. Wong iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length, 1056c039b997SGoldwyn Rodrigues void *data, struct iomap *iomap, struct iomap *srcmap) 1057afc51aaaSDarrick J. Wong { 1058afc51aaaSDarrick J. Wong struct page *page = data; 1059afc51aaaSDarrick J. Wong int ret; 1060afc51aaaSDarrick J. Wong 1061afc51aaaSDarrick J. Wong if (iomap->flags & IOMAP_F_BUFFER_HEAD) { 1062afc51aaaSDarrick J. Wong ret = __block_write_begin_int(page, pos, length, NULL, iomap); 1063afc51aaaSDarrick J. Wong if (ret) 1064afc51aaaSDarrick J. Wong return ret; 1065afc51aaaSDarrick J. Wong block_commit_write(page, 0, length); 1066afc51aaaSDarrick J. Wong } else { 1067afc51aaaSDarrick J. Wong WARN_ON_ONCE(!PageUptodate(page)); 1068afc51aaaSDarrick J. Wong iomap_page_create(inode, page); 1069afc51aaaSDarrick J. Wong set_page_dirty(page); 1070afc51aaaSDarrick J. Wong } 1071afc51aaaSDarrick J. Wong 1072afc51aaaSDarrick J. Wong return length; 1073afc51aaaSDarrick J. Wong } 1074afc51aaaSDarrick J. Wong 1075afc51aaaSDarrick J. Wong vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) 1076afc51aaaSDarrick J. Wong { 1077afc51aaaSDarrick J. Wong struct page *page = vmf->page; 1078afc51aaaSDarrick J. Wong struct inode *inode = file_inode(vmf->vma->vm_file); 1079afc51aaaSDarrick J. Wong unsigned long length; 1080*243145bcSAndreas Gruenbacher loff_t offset; 1081afc51aaaSDarrick J. Wong ssize_t ret; 1082afc51aaaSDarrick J. Wong 1083afc51aaaSDarrick J. Wong lock_page(page); 1084*243145bcSAndreas Gruenbacher ret = page_mkwrite_check_truncate(page, inode); 1085*243145bcSAndreas Gruenbacher if (ret < 0) 1086afc51aaaSDarrick J. Wong goto out_unlock; 1087*243145bcSAndreas Gruenbacher length = ret; 1088afc51aaaSDarrick J. Wong 1089*243145bcSAndreas Gruenbacher offset = page_offset(page); 1090afc51aaaSDarrick J. Wong while (length > 0) { 1091afc51aaaSDarrick J. Wong ret = iomap_apply(inode, offset, length, 1092afc51aaaSDarrick J. Wong IOMAP_WRITE | IOMAP_FAULT, ops, page, 1093afc51aaaSDarrick J. Wong iomap_page_mkwrite_actor); 1094afc51aaaSDarrick J. Wong if (unlikely(ret <= 0)) 1095afc51aaaSDarrick J. Wong goto out_unlock; 1096afc51aaaSDarrick J. Wong offset += ret; 1097afc51aaaSDarrick J. Wong length -= ret; 1098afc51aaaSDarrick J. Wong } 1099afc51aaaSDarrick J. Wong 1100afc51aaaSDarrick J. Wong wait_for_stable_page(page); 1101afc51aaaSDarrick J. Wong return VM_FAULT_LOCKED; 1102afc51aaaSDarrick J. Wong out_unlock: 1103afc51aaaSDarrick J. Wong unlock_page(page); 1104afc51aaaSDarrick J. Wong return block_page_mkwrite_return(ret); 1105afc51aaaSDarrick J. Wong } 1106afc51aaaSDarrick J. Wong EXPORT_SYMBOL_GPL(iomap_page_mkwrite); 1107598ecfbaSChristoph Hellwig 1108598ecfbaSChristoph Hellwig static void 110948d64cd1SChristoph Hellwig iomap_finish_page_writeback(struct inode *inode, struct page *page, 1110598ecfbaSChristoph Hellwig int error) 1111598ecfbaSChristoph Hellwig { 111248d64cd1SChristoph Hellwig struct iomap_page *iop = to_iomap_page(page); 1113598ecfbaSChristoph Hellwig 1114598ecfbaSChristoph Hellwig if (error) { 111548d64cd1SChristoph Hellwig SetPageError(page); 1116598ecfbaSChristoph Hellwig mapping_set_error(inode->i_mapping, -EIO); 1117598ecfbaSChristoph Hellwig } 1118598ecfbaSChristoph Hellwig 1119598ecfbaSChristoph Hellwig WARN_ON_ONCE(i_blocksize(inode) < PAGE_SIZE && !iop); 1120598ecfbaSChristoph Hellwig WARN_ON_ONCE(iop && atomic_read(&iop->write_count) <= 0); 1121598ecfbaSChristoph Hellwig 1122598ecfbaSChristoph Hellwig if (!iop || atomic_dec_and_test(&iop->write_count)) 112348d64cd1SChristoph Hellwig end_page_writeback(page); 1124598ecfbaSChristoph Hellwig } 1125598ecfbaSChristoph Hellwig 1126598ecfbaSChristoph Hellwig /* 1127598ecfbaSChristoph Hellwig * We're now finished for good with this ioend structure. Update the page 1128598ecfbaSChristoph Hellwig * state, release holds on bios, and finally free up memory. Do not use the 1129598ecfbaSChristoph Hellwig * ioend after this. 1130598ecfbaSChristoph Hellwig */ 1131598ecfbaSChristoph Hellwig static void 1132598ecfbaSChristoph Hellwig iomap_finish_ioend(struct iomap_ioend *ioend, int error) 1133598ecfbaSChristoph Hellwig { 1134598ecfbaSChristoph Hellwig struct inode *inode = ioend->io_inode; 1135598ecfbaSChristoph Hellwig struct bio *bio = &ioend->io_inline_bio; 1136598ecfbaSChristoph Hellwig struct bio *last = ioend->io_bio, *next; 1137598ecfbaSChristoph Hellwig u64 start = bio->bi_iter.bi_sector; 1138c275779fSZorro Lang loff_t offset = ioend->io_offset; 1139598ecfbaSChristoph Hellwig bool quiet = bio_flagged(bio, BIO_QUIET); 1140598ecfbaSChristoph Hellwig 1141598ecfbaSChristoph Hellwig for (bio = &ioend->io_inline_bio; bio; bio = next) { 1142598ecfbaSChristoph Hellwig struct bio_vec *bv; 1143598ecfbaSChristoph Hellwig struct bvec_iter_all iter_all; 1144598ecfbaSChristoph Hellwig 1145598ecfbaSChristoph Hellwig /* 1146598ecfbaSChristoph Hellwig * For the last bio, bi_private points to the ioend, so we 1147598ecfbaSChristoph Hellwig * need to explicitly end the iteration here. 1148598ecfbaSChristoph Hellwig */ 1149598ecfbaSChristoph Hellwig if (bio == last) 1150598ecfbaSChristoph Hellwig next = NULL; 1151598ecfbaSChristoph Hellwig else 1152598ecfbaSChristoph Hellwig next = bio->bi_private; 1153598ecfbaSChristoph Hellwig 1154598ecfbaSChristoph Hellwig /* walk each page on bio, ending page IO on them */ 1155598ecfbaSChristoph Hellwig bio_for_each_segment_all(bv, bio, iter_all) 115648d64cd1SChristoph Hellwig iomap_finish_page_writeback(inode, bv->bv_page, error); 1157598ecfbaSChristoph Hellwig bio_put(bio); 1158598ecfbaSChristoph Hellwig } 1159c275779fSZorro Lang /* The ioend has been freed by bio_put() */ 1160598ecfbaSChristoph Hellwig 1161598ecfbaSChristoph Hellwig if (unlikely(error && !quiet)) { 1162598ecfbaSChristoph Hellwig printk_ratelimited(KERN_ERR 11639cd0ed63SDarrick J. Wong "%s: writeback error on inode %lu, offset %lld, sector %llu", 1164c275779fSZorro Lang inode->i_sb->s_id, inode->i_ino, offset, start); 1165598ecfbaSChristoph Hellwig } 1166598ecfbaSChristoph Hellwig } 1167598ecfbaSChristoph Hellwig 1168598ecfbaSChristoph Hellwig void 1169598ecfbaSChristoph Hellwig iomap_finish_ioends(struct iomap_ioend *ioend, int error) 1170598ecfbaSChristoph Hellwig { 1171598ecfbaSChristoph Hellwig struct list_head tmp; 1172598ecfbaSChristoph Hellwig 1173598ecfbaSChristoph Hellwig list_replace_init(&ioend->io_list, &tmp); 1174598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1175598ecfbaSChristoph Hellwig 1176598ecfbaSChristoph Hellwig while (!list_empty(&tmp)) { 1177598ecfbaSChristoph Hellwig ioend = list_first_entry(&tmp, struct iomap_ioend, io_list); 1178598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1179598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, error); 1180598ecfbaSChristoph Hellwig } 1181598ecfbaSChristoph Hellwig } 1182598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_finish_ioends); 1183598ecfbaSChristoph Hellwig 1184598ecfbaSChristoph Hellwig /* 1185598ecfbaSChristoph Hellwig * We can merge two adjacent ioends if they have the same set of work to do. 1186598ecfbaSChristoph Hellwig */ 1187598ecfbaSChristoph Hellwig static bool 1188598ecfbaSChristoph Hellwig iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next) 1189598ecfbaSChristoph Hellwig { 1190598ecfbaSChristoph Hellwig if (ioend->io_bio->bi_status != next->io_bio->bi_status) 1191598ecfbaSChristoph Hellwig return false; 1192598ecfbaSChristoph Hellwig if ((ioend->io_flags & IOMAP_F_SHARED) ^ 1193598ecfbaSChristoph Hellwig (next->io_flags & IOMAP_F_SHARED)) 1194598ecfbaSChristoph Hellwig return false; 1195598ecfbaSChristoph Hellwig if ((ioend->io_type == IOMAP_UNWRITTEN) ^ 1196598ecfbaSChristoph Hellwig (next->io_type == IOMAP_UNWRITTEN)) 1197598ecfbaSChristoph Hellwig return false; 1198598ecfbaSChristoph Hellwig if (ioend->io_offset + ioend->io_size != next->io_offset) 1199598ecfbaSChristoph Hellwig return false; 1200598ecfbaSChristoph Hellwig return true; 1201598ecfbaSChristoph Hellwig } 1202598ecfbaSChristoph Hellwig 1203598ecfbaSChristoph Hellwig void 1204598ecfbaSChristoph Hellwig iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends, 1205598ecfbaSChristoph Hellwig void (*merge_private)(struct iomap_ioend *ioend, 1206598ecfbaSChristoph Hellwig struct iomap_ioend *next)) 1207598ecfbaSChristoph Hellwig { 1208598ecfbaSChristoph Hellwig struct iomap_ioend *next; 1209598ecfbaSChristoph Hellwig 1210598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1211598ecfbaSChristoph Hellwig 1212598ecfbaSChristoph Hellwig while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend, 1213598ecfbaSChristoph Hellwig io_list))) { 1214598ecfbaSChristoph Hellwig if (!iomap_ioend_can_merge(ioend, next)) 1215598ecfbaSChristoph Hellwig break; 1216598ecfbaSChristoph Hellwig list_move_tail(&next->io_list, &ioend->io_list); 1217598ecfbaSChristoph Hellwig ioend->io_size += next->io_size; 1218598ecfbaSChristoph Hellwig if (next->io_private && merge_private) 1219598ecfbaSChristoph Hellwig merge_private(ioend, next); 1220598ecfbaSChristoph Hellwig } 1221598ecfbaSChristoph Hellwig } 1222598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_ioend_try_merge); 1223598ecfbaSChristoph Hellwig 1224598ecfbaSChristoph Hellwig static int 1225598ecfbaSChristoph Hellwig iomap_ioend_compare(void *priv, struct list_head *a, struct list_head *b) 1226598ecfbaSChristoph Hellwig { 1227b3d423ecSChristoph Hellwig struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list); 1228b3d423ecSChristoph Hellwig struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list); 1229598ecfbaSChristoph Hellwig 1230598ecfbaSChristoph Hellwig if (ia->io_offset < ib->io_offset) 1231598ecfbaSChristoph Hellwig return -1; 1232b3d423ecSChristoph Hellwig if (ia->io_offset > ib->io_offset) 1233598ecfbaSChristoph Hellwig return 1; 1234598ecfbaSChristoph Hellwig return 0; 1235598ecfbaSChristoph Hellwig } 1236598ecfbaSChristoph Hellwig 1237598ecfbaSChristoph Hellwig void 1238598ecfbaSChristoph Hellwig iomap_sort_ioends(struct list_head *ioend_list) 1239598ecfbaSChristoph Hellwig { 1240598ecfbaSChristoph Hellwig list_sort(NULL, ioend_list, iomap_ioend_compare); 1241598ecfbaSChristoph Hellwig } 1242598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_sort_ioends); 1243598ecfbaSChristoph Hellwig 1244598ecfbaSChristoph Hellwig static void iomap_writepage_end_bio(struct bio *bio) 1245598ecfbaSChristoph Hellwig { 1246598ecfbaSChristoph Hellwig struct iomap_ioend *ioend = bio->bi_private; 1247598ecfbaSChristoph Hellwig 1248598ecfbaSChristoph Hellwig iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status)); 1249598ecfbaSChristoph Hellwig } 1250598ecfbaSChristoph Hellwig 1251598ecfbaSChristoph Hellwig /* 1252598ecfbaSChristoph Hellwig * Submit the final bio for an ioend. 1253598ecfbaSChristoph Hellwig * 1254598ecfbaSChristoph Hellwig * If @error is non-zero, it means that we have a situation where some part of 1255598ecfbaSChristoph Hellwig * the submission process has failed after we have marked paged for writeback 1256598ecfbaSChristoph Hellwig * and unlocked them. In this situation, we need to fail the bio instead of 1257598ecfbaSChristoph Hellwig * submitting it. This typically only happens on a filesystem shutdown. 1258598ecfbaSChristoph Hellwig */ 1259598ecfbaSChristoph Hellwig static int 1260598ecfbaSChristoph Hellwig iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend, 1261598ecfbaSChristoph Hellwig int error) 1262598ecfbaSChristoph Hellwig { 1263598ecfbaSChristoph Hellwig ioend->io_bio->bi_private = ioend; 1264598ecfbaSChristoph Hellwig ioend->io_bio->bi_end_io = iomap_writepage_end_bio; 1265598ecfbaSChristoph Hellwig 1266598ecfbaSChristoph Hellwig if (wpc->ops->prepare_ioend) 1267598ecfbaSChristoph Hellwig error = wpc->ops->prepare_ioend(ioend, error); 1268598ecfbaSChristoph Hellwig if (error) { 1269598ecfbaSChristoph Hellwig /* 1270598ecfbaSChristoph Hellwig * If we are failing the IO now, just mark the ioend with an 1271598ecfbaSChristoph Hellwig * error and finish it. This will run IO completion immediately 1272598ecfbaSChristoph Hellwig * as there is only one reference to the ioend at this point in 1273598ecfbaSChristoph Hellwig * time. 1274598ecfbaSChristoph Hellwig */ 1275598ecfbaSChristoph Hellwig ioend->io_bio->bi_status = errno_to_blk_status(error); 1276598ecfbaSChristoph Hellwig bio_endio(ioend->io_bio); 1277598ecfbaSChristoph Hellwig return error; 1278598ecfbaSChristoph Hellwig } 1279598ecfbaSChristoph Hellwig 1280598ecfbaSChristoph Hellwig submit_bio(ioend->io_bio); 1281598ecfbaSChristoph Hellwig return 0; 1282598ecfbaSChristoph Hellwig } 1283598ecfbaSChristoph Hellwig 1284598ecfbaSChristoph Hellwig static struct iomap_ioend * 1285598ecfbaSChristoph Hellwig iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc, 1286598ecfbaSChristoph Hellwig loff_t offset, sector_t sector, struct writeback_control *wbc) 1287598ecfbaSChristoph Hellwig { 1288598ecfbaSChristoph Hellwig struct iomap_ioend *ioend; 1289598ecfbaSChristoph Hellwig struct bio *bio; 1290598ecfbaSChristoph Hellwig 1291598ecfbaSChristoph Hellwig bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &iomap_ioend_bioset); 1292598ecfbaSChristoph Hellwig bio_set_dev(bio, wpc->iomap.bdev); 1293598ecfbaSChristoph Hellwig bio->bi_iter.bi_sector = sector; 1294598ecfbaSChristoph Hellwig bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); 1295598ecfbaSChristoph Hellwig bio->bi_write_hint = inode->i_write_hint; 1296598ecfbaSChristoph Hellwig wbc_init_bio(wbc, bio); 1297598ecfbaSChristoph Hellwig 1298598ecfbaSChristoph Hellwig ioend = container_of(bio, struct iomap_ioend, io_inline_bio); 1299598ecfbaSChristoph Hellwig INIT_LIST_HEAD(&ioend->io_list); 1300598ecfbaSChristoph Hellwig ioend->io_type = wpc->iomap.type; 1301598ecfbaSChristoph Hellwig ioend->io_flags = wpc->iomap.flags; 1302598ecfbaSChristoph Hellwig ioend->io_inode = inode; 1303598ecfbaSChristoph Hellwig ioend->io_size = 0; 1304598ecfbaSChristoph Hellwig ioend->io_offset = offset; 1305598ecfbaSChristoph Hellwig ioend->io_private = NULL; 1306598ecfbaSChristoph Hellwig ioend->io_bio = bio; 1307598ecfbaSChristoph Hellwig return ioend; 1308598ecfbaSChristoph Hellwig } 1309598ecfbaSChristoph Hellwig 1310598ecfbaSChristoph Hellwig /* 1311598ecfbaSChristoph Hellwig * Allocate a new bio, and chain the old bio to the new one. 1312598ecfbaSChristoph Hellwig * 1313598ecfbaSChristoph Hellwig * Note that we have to do perform the chaining in this unintuitive order 1314598ecfbaSChristoph Hellwig * so that the bi_private linkage is set up in the right direction for the 1315598ecfbaSChristoph Hellwig * traversal in iomap_finish_ioend(). 1316598ecfbaSChristoph Hellwig */ 1317598ecfbaSChristoph Hellwig static struct bio * 1318598ecfbaSChristoph Hellwig iomap_chain_bio(struct bio *prev) 1319598ecfbaSChristoph Hellwig { 1320598ecfbaSChristoph Hellwig struct bio *new; 1321598ecfbaSChristoph Hellwig 1322598ecfbaSChristoph Hellwig new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); 1323598ecfbaSChristoph Hellwig bio_copy_dev(new, prev);/* also copies over blkcg information */ 1324598ecfbaSChristoph Hellwig new->bi_iter.bi_sector = bio_end_sector(prev); 1325598ecfbaSChristoph Hellwig new->bi_opf = prev->bi_opf; 1326598ecfbaSChristoph Hellwig new->bi_write_hint = prev->bi_write_hint; 1327598ecfbaSChristoph Hellwig 1328598ecfbaSChristoph Hellwig bio_chain(prev, new); 1329598ecfbaSChristoph Hellwig bio_get(prev); /* for iomap_finish_ioend */ 1330598ecfbaSChristoph Hellwig submit_bio(prev); 1331598ecfbaSChristoph Hellwig return new; 1332598ecfbaSChristoph Hellwig } 1333598ecfbaSChristoph Hellwig 1334598ecfbaSChristoph Hellwig static bool 1335598ecfbaSChristoph Hellwig iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset, 1336598ecfbaSChristoph Hellwig sector_t sector) 1337598ecfbaSChristoph Hellwig { 1338598ecfbaSChristoph Hellwig if ((wpc->iomap.flags & IOMAP_F_SHARED) != 1339598ecfbaSChristoph Hellwig (wpc->ioend->io_flags & IOMAP_F_SHARED)) 1340598ecfbaSChristoph Hellwig return false; 1341598ecfbaSChristoph Hellwig if (wpc->iomap.type != wpc->ioend->io_type) 1342598ecfbaSChristoph Hellwig return false; 1343598ecfbaSChristoph Hellwig if (offset != wpc->ioend->io_offset + wpc->ioend->io_size) 1344598ecfbaSChristoph Hellwig return false; 1345598ecfbaSChristoph Hellwig if (sector != bio_end_sector(wpc->ioend->io_bio)) 1346598ecfbaSChristoph Hellwig return false; 1347598ecfbaSChristoph Hellwig return true; 1348598ecfbaSChristoph Hellwig } 1349598ecfbaSChristoph Hellwig 1350598ecfbaSChristoph Hellwig /* 1351598ecfbaSChristoph Hellwig * Test to see if we have an existing ioend structure that we could append to 1352598ecfbaSChristoph Hellwig * first, otherwise finish off the current ioend and start another. 1353598ecfbaSChristoph Hellwig */ 1354598ecfbaSChristoph Hellwig static void 1355598ecfbaSChristoph Hellwig iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page, 1356598ecfbaSChristoph Hellwig struct iomap_page *iop, struct iomap_writepage_ctx *wpc, 1357598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct list_head *iolist) 1358598ecfbaSChristoph Hellwig { 1359598ecfbaSChristoph Hellwig sector_t sector = iomap_sector(&wpc->iomap, offset); 1360598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1361598ecfbaSChristoph Hellwig unsigned poff = offset & (PAGE_SIZE - 1); 1362598ecfbaSChristoph Hellwig bool merged, same_page = false; 1363598ecfbaSChristoph Hellwig 1364598ecfbaSChristoph Hellwig if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) { 1365598ecfbaSChristoph Hellwig if (wpc->ioend) 1366598ecfbaSChristoph Hellwig list_add(&wpc->ioend->io_list, iolist); 1367598ecfbaSChristoph Hellwig wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc); 1368598ecfbaSChristoph Hellwig } 1369598ecfbaSChristoph Hellwig 1370598ecfbaSChristoph Hellwig merged = __bio_try_merge_page(wpc->ioend->io_bio, page, len, poff, 1371598ecfbaSChristoph Hellwig &same_page); 1372598ecfbaSChristoph Hellwig if (iop && !same_page) 1373598ecfbaSChristoph Hellwig atomic_inc(&iop->write_count); 1374598ecfbaSChristoph Hellwig 1375598ecfbaSChristoph Hellwig if (!merged) { 1376598ecfbaSChristoph Hellwig if (bio_full(wpc->ioend->io_bio, len)) { 1377598ecfbaSChristoph Hellwig wpc->ioend->io_bio = 1378598ecfbaSChristoph Hellwig iomap_chain_bio(wpc->ioend->io_bio); 1379598ecfbaSChristoph Hellwig } 1380598ecfbaSChristoph Hellwig bio_add_page(wpc->ioend->io_bio, page, len, poff); 1381598ecfbaSChristoph Hellwig } 1382598ecfbaSChristoph Hellwig 1383598ecfbaSChristoph Hellwig wpc->ioend->io_size += len; 1384598ecfbaSChristoph Hellwig wbc_account_cgroup_owner(wbc, page, len); 1385598ecfbaSChristoph Hellwig } 1386598ecfbaSChristoph Hellwig 1387598ecfbaSChristoph Hellwig /* 1388598ecfbaSChristoph Hellwig * We implement an immediate ioend submission policy here to avoid needing to 1389598ecfbaSChristoph Hellwig * chain multiple ioends and hence nest mempool allocations which can violate 1390598ecfbaSChristoph Hellwig * forward progress guarantees we need to provide. The current ioend we are 1391598ecfbaSChristoph Hellwig * adding blocks to is cached on the writepage context, and if the new block 1392598ecfbaSChristoph Hellwig * does not append to the cached ioend it will create a new ioend and cache that 1393598ecfbaSChristoph Hellwig * instead. 1394598ecfbaSChristoph Hellwig * 1395598ecfbaSChristoph Hellwig * If a new ioend is created and cached, the old ioend is returned and queued 1396598ecfbaSChristoph Hellwig * locally for submission once the entire page is processed or an error has been 1397598ecfbaSChristoph Hellwig * detected. While ioends are submitted immediately after they are completed, 1398598ecfbaSChristoph Hellwig * batching optimisations are provided by higher level block plugging. 1399598ecfbaSChristoph Hellwig * 1400598ecfbaSChristoph Hellwig * At the end of a writeback pass, there will be a cached ioend remaining on the 1401598ecfbaSChristoph Hellwig * writepage context that the caller will need to submit. 1402598ecfbaSChristoph Hellwig */ 1403598ecfbaSChristoph Hellwig static int 1404598ecfbaSChristoph Hellwig iomap_writepage_map(struct iomap_writepage_ctx *wpc, 1405598ecfbaSChristoph Hellwig struct writeback_control *wbc, struct inode *inode, 1406598ecfbaSChristoph Hellwig struct page *page, u64 end_offset) 1407598ecfbaSChristoph Hellwig { 1408598ecfbaSChristoph Hellwig struct iomap_page *iop = to_iomap_page(page); 1409598ecfbaSChristoph Hellwig struct iomap_ioend *ioend, *next; 1410598ecfbaSChristoph Hellwig unsigned len = i_blocksize(inode); 1411598ecfbaSChristoph Hellwig u64 file_offset; /* file offset of page */ 1412598ecfbaSChristoph Hellwig int error = 0, count = 0, i; 1413598ecfbaSChristoph Hellwig LIST_HEAD(submit_list); 1414598ecfbaSChristoph Hellwig 1415598ecfbaSChristoph Hellwig WARN_ON_ONCE(i_blocksize(inode) < PAGE_SIZE && !iop); 1416598ecfbaSChristoph Hellwig WARN_ON_ONCE(iop && atomic_read(&iop->write_count) != 0); 1417598ecfbaSChristoph Hellwig 1418598ecfbaSChristoph Hellwig /* 1419598ecfbaSChristoph Hellwig * Walk through the page to find areas to write back. If we run off the 1420598ecfbaSChristoph Hellwig * end of the current map or find the current map invalid, grab a new 1421598ecfbaSChristoph Hellwig * one. 1422598ecfbaSChristoph Hellwig */ 1423598ecfbaSChristoph Hellwig for (i = 0, file_offset = page_offset(page); 1424598ecfbaSChristoph Hellwig i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset; 1425598ecfbaSChristoph Hellwig i++, file_offset += len) { 1426598ecfbaSChristoph Hellwig if (iop && !test_bit(i, iop->uptodate)) 1427598ecfbaSChristoph Hellwig continue; 1428598ecfbaSChristoph Hellwig 1429598ecfbaSChristoph Hellwig error = wpc->ops->map_blocks(wpc, inode, file_offset); 1430598ecfbaSChristoph Hellwig if (error) 1431598ecfbaSChristoph Hellwig break; 14323e19e6f3SChristoph Hellwig if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE)) 14333e19e6f3SChristoph Hellwig continue; 1434598ecfbaSChristoph Hellwig if (wpc->iomap.type == IOMAP_HOLE) 1435598ecfbaSChristoph Hellwig continue; 1436598ecfbaSChristoph Hellwig iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc, 1437598ecfbaSChristoph Hellwig &submit_list); 1438598ecfbaSChristoph Hellwig count++; 1439598ecfbaSChristoph Hellwig } 1440598ecfbaSChristoph Hellwig 1441598ecfbaSChristoph Hellwig WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list)); 1442598ecfbaSChristoph Hellwig WARN_ON_ONCE(!PageLocked(page)); 1443598ecfbaSChristoph Hellwig WARN_ON_ONCE(PageWriteback(page)); 1444598ecfbaSChristoph Hellwig 1445598ecfbaSChristoph Hellwig /* 1446598ecfbaSChristoph Hellwig * We cannot cancel the ioend directly here on error. We may have 1447598ecfbaSChristoph Hellwig * already set other pages under writeback and hence we have to run I/O 1448598ecfbaSChristoph Hellwig * completion to mark the error state of the pages under writeback 1449598ecfbaSChristoph Hellwig * appropriately. 1450598ecfbaSChristoph Hellwig */ 1451598ecfbaSChristoph Hellwig if (unlikely(error)) { 1452598ecfbaSChristoph Hellwig if (!count) { 1453598ecfbaSChristoph Hellwig /* 1454598ecfbaSChristoph Hellwig * If the current page hasn't been added to ioend, it 1455598ecfbaSChristoph Hellwig * won't be affected by I/O completions and we must 1456598ecfbaSChristoph Hellwig * discard and unlock it right here. 1457598ecfbaSChristoph Hellwig */ 1458598ecfbaSChristoph Hellwig if (wpc->ops->discard_page) 1459598ecfbaSChristoph Hellwig wpc->ops->discard_page(page); 1460598ecfbaSChristoph Hellwig ClearPageUptodate(page); 1461598ecfbaSChristoph Hellwig unlock_page(page); 1462598ecfbaSChristoph Hellwig goto done; 1463598ecfbaSChristoph Hellwig } 1464598ecfbaSChristoph Hellwig 1465598ecfbaSChristoph Hellwig /* 1466598ecfbaSChristoph Hellwig * If the page was not fully cleaned, we need to ensure that the 1467598ecfbaSChristoph Hellwig * higher layers come back to it correctly. That means we need 1468598ecfbaSChristoph Hellwig * to keep the page dirty, and for WB_SYNC_ALL writeback we need 1469598ecfbaSChristoph Hellwig * to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed 1470598ecfbaSChristoph Hellwig * so another attempt to write this page in this writeback sweep 1471598ecfbaSChristoph Hellwig * will be made. 1472598ecfbaSChristoph Hellwig */ 1473598ecfbaSChristoph Hellwig set_page_writeback_keepwrite(page); 1474598ecfbaSChristoph Hellwig } else { 1475598ecfbaSChristoph Hellwig clear_page_dirty_for_io(page); 1476598ecfbaSChristoph Hellwig set_page_writeback(page); 1477598ecfbaSChristoph Hellwig } 1478598ecfbaSChristoph Hellwig 1479598ecfbaSChristoph Hellwig unlock_page(page); 1480598ecfbaSChristoph Hellwig 1481598ecfbaSChristoph Hellwig /* 1482598ecfbaSChristoph Hellwig * Preserve the original error if there was one, otherwise catch 1483598ecfbaSChristoph Hellwig * submission errors here and propagate into subsequent ioend 1484598ecfbaSChristoph Hellwig * submissions. 1485598ecfbaSChristoph Hellwig */ 1486598ecfbaSChristoph Hellwig list_for_each_entry_safe(ioend, next, &submit_list, io_list) { 1487598ecfbaSChristoph Hellwig int error2; 1488598ecfbaSChristoph Hellwig 1489598ecfbaSChristoph Hellwig list_del_init(&ioend->io_list); 1490598ecfbaSChristoph Hellwig error2 = iomap_submit_ioend(wpc, ioend, error); 1491598ecfbaSChristoph Hellwig if (error2 && !error) 1492598ecfbaSChristoph Hellwig error = error2; 1493598ecfbaSChristoph Hellwig } 1494598ecfbaSChristoph Hellwig 1495598ecfbaSChristoph Hellwig /* 1496598ecfbaSChristoph Hellwig * We can end up here with no error and nothing to write only if we race 1497598ecfbaSChristoph Hellwig * with a partial page truncate on a sub-page block sized filesystem. 1498598ecfbaSChristoph Hellwig */ 1499598ecfbaSChristoph Hellwig if (!count) 1500598ecfbaSChristoph Hellwig end_page_writeback(page); 1501598ecfbaSChristoph Hellwig done: 1502598ecfbaSChristoph Hellwig mapping_set_error(page->mapping, error); 1503598ecfbaSChristoph Hellwig return error; 1504598ecfbaSChristoph Hellwig } 1505598ecfbaSChristoph Hellwig 1506598ecfbaSChristoph Hellwig /* 1507598ecfbaSChristoph Hellwig * Write out a dirty page. 1508598ecfbaSChristoph Hellwig * 1509598ecfbaSChristoph Hellwig * For delalloc space on the page we need to allocate space and flush it. 1510598ecfbaSChristoph Hellwig * For unwritten space on the page we need to start the conversion to 1511598ecfbaSChristoph Hellwig * regular allocated space. 1512598ecfbaSChristoph Hellwig */ 1513598ecfbaSChristoph Hellwig static int 1514598ecfbaSChristoph Hellwig iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data) 1515598ecfbaSChristoph Hellwig { 1516598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc = data; 1517598ecfbaSChristoph Hellwig struct inode *inode = page->mapping->host; 1518598ecfbaSChristoph Hellwig pgoff_t end_index; 1519598ecfbaSChristoph Hellwig u64 end_offset; 1520598ecfbaSChristoph Hellwig loff_t offset; 1521598ecfbaSChristoph Hellwig 1522598ecfbaSChristoph Hellwig trace_iomap_writepage(inode, page, 0, 0); 1523598ecfbaSChristoph Hellwig 1524598ecfbaSChristoph Hellwig /* 1525598ecfbaSChristoph Hellwig * Refuse to write the page out if we are called from reclaim context. 1526598ecfbaSChristoph Hellwig * 1527598ecfbaSChristoph Hellwig * This avoids stack overflows when called from deeply used stacks in 1528598ecfbaSChristoph Hellwig * random callers for direct reclaim or memcg reclaim. We explicitly 1529598ecfbaSChristoph Hellwig * allow reclaim from kswapd as the stack usage there is relatively low. 1530598ecfbaSChristoph Hellwig * 1531598ecfbaSChristoph Hellwig * This should never happen except in the case of a VM regression so 1532598ecfbaSChristoph Hellwig * warn about it. 1533598ecfbaSChristoph Hellwig */ 1534598ecfbaSChristoph Hellwig if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == 1535598ecfbaSChristoph Hellwig PF_MEMALLOC)) 1536598ecfbaSChristoph Hellwig goto redirty; 1537598ecfbaSChristoph Hellwig 1538598ecfbaSChristoph Hellwig /* 1539598ecfbaSChristoph Hellwig * Given that we do not allow direct reclaim to call us, we should 1540598ecfbaSChristoph Hellwig * never be called in a recursive filesystem reclaim context. 1541598ecfbaSChristoph Hellwig */ 1542598ecfbaSChristoph Hellwig if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS)) 1543598ecfbaSChristoph Hellwig goto redirty; 1544598ecfbaSChristoph Hellwig 1545598ecfbaSChristoph Hellwig /* 1546598ecfbaSChristoph Hellwig * Is this page beyond the end of the file? 1547598ecfbaSChristoph Hellwig * 1548598ecfbaSChristoph Hellwig * The page index is less than the end_index, adjust the end_offset 1549598ecfbaSChristoph Hellwig * to the highest offset that this page should represent. 1550598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1551598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1552598ecfbaSChristoph Hellwig * ----------------------------------------------------- 1553598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | | 1554598ecfbaSChristoph Hellwig * ^--------------------------------^----------|-------- 1555598ecfbaSChristoph Hellwig * | desired writeback range | see else | 1556598ecfbaSChristoph Hellwig * ---------------------------------^------------------| 1557598ecfbaSChristoph Hellwig */ 1558598ecfbaSChristoph Hellwig offset = i_size_read(inode); 1559598ecfbaSChristoph Hellwig end_index = offset >> PAGE_SHIFT; 1560598ecfbaSChristoph Hellwig if (page->index < end_index) 1561598ecfbaSChristoph Hellwig end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT; 1562598ecfbaSChristoph Hellwig else { 1563598ecfbaSChristoph Hellwig /* 1564598ecfbaSChristoph Hellwig * Check whether the page to write out is beyond or straddles 1565598ecfbaSChristoph Hellwig * i_size or not. 1566598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1567598ecfbaSChristoph Hellwig * | file mapping | <EOF> | 1568598ecfbaSChristoph Hellwig * ------------------------------------------------------- 1569598ecfbaSChristoph Hellwig * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | 1570598ecfbaSChristoph Hellwig * ^--------------------------------^-----------|--------- 1571598ecfbaSChristoph Hellwig * | | Straddles | 1572598ecfbaSChristoph Hellwig * ---------------------------------^-----------|--------| 1573598ecfbaSChristoph Hellwig */ 1574598ecfbaSChristoph Hellwig unsigned offset_into_page = offset & (PAGE_SIZE - 1); 1575598ecfbaSChristoph Hellwig 1576598ecfbaSChristoph Hellwig /* 1577598ecfbaSChristoph Hellwig * Skip the page if it is fully outside i_size, e.g. due to a 1578598ecfbaSChristoph Hellwig * truncate operation that is in progress. We must redirty the 1579598ecfbaSChristoph Hellwig * page so that reclaim stops reclaiming it. Otherwise 1580598ecfbaSChristoph Hellwig * iomap_vm_releasepage() is called on it and gets confused. 1581598ecfbaSChristoph Hellwig * 1582598ecfbaSChristoph Hellwig * Note that the end_index is unsigned long, it would overflow 1583598ecfbaSChristoph Hellwig * if the given offset is greater than 16TB on 32-bit system 1584598ecfbaSChristoph Hellwig * and if we do check the page is fully outside i_size or not 1585598ecfbaSChristoph Hellwig * via "if (page->index >= end_index + 1)" as "end_index + 1" 1586598ecfbaSChristoph Hellwig * will be evaluated to 0. Hence this page will be redirtied 1587598ecfbaSChristoph Hellwig * and be written out repeatedly which would result in an 1588598ecfbaSChristoph Hellwig * infinite loop, the user program that perform this operation 1589598ecfbaSChristoph Hellwig * will hang. Instead, we can verify this situation by checking 1590598ecfbaSChristoph Hellwig * if the page to write is totally beyond the i_size or if it's 1591598ecfbaSChristoph Hellwig * offset is just equal to the EOF. 1592598ecfbaSChristoph Hellwig */ 1593598ecfbaSChristoph Hellwig if (page->index > end_index || 1594598ecfbaSChristoph Hellwig (page->index == end_index && offset_into_page == 0)) 1595598ecfbaSChristoph Hellwig goto redirty; 1596598ecfbaSChristoph Hellwig 1597598ecfbaSChristoph Hellwig /* 1598598ecfbaSChristoph Hellwig * The page straddles i_size. It must be zeroed out on each 1599598ecfbaSChristoph Hellwig * and every writepage invocation because it may be mmapped. 1600598ecfbaSChristoph Hellwig * "A file is mapped in multiples of the page size. For a file 1601598ecfbaSChristoph Hellwig * that is not a multiple of the page size, the remaining 1602598ecfbaSChristoph Hellwig * memory is zeroed when mapped, and writes to that region are 1603598ecfbaSChristoph Hellwig * not written out to the file." 1604598ecfbaSChristoph Hellwig */ 1605598ecfbaSChristoph Hellwig zero_user_segment(page, offset_into_page, PAGE_SIZE); 1606598ecfbaSChristoph Hellwig 1607598ecfbaSChristoph Hellwig /* Adjust the end_offset to the end of file */ 1608598ecfbaSChristoph Hellwig end_offset = offset; 1609598ecfbaSChristoph Hellwig } 1610598ecfbaSChristoph Hellwig 1611598ecfbaSChristoph Hellwig return iomap_writepage_map(wpc, wbc, inode, page, end_offset); 1612598ecfbaSChristoph Hellwig 1613598ecfbaSChristoph Hellwig redirty: 1614598ecfbaSChristoph Hellwig redirty_page_for_writepage(wbc, page); 1615598ecfbaSChristoph Hellwig unlock_page(page); 1616598ecfbaSChristoph Hellwig return 0; 1617598ecfbaSChristoph Hellwig } 1618598ecfbaSChristoph Hellwig 1619598ecfbaSChristoph Hellwig int 1620598ecfbaSChristoph Hellwig iomap_writepage(struct page *page, struct writeback_control *wbc, 1621598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1622598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1623598ecfbaSChristoph Hellwig { 1624598ecfbaSChristoph Hellwig int ret; 1625598ecfbaSChristoph Hellwig 1626598ecfbaSChristoph Hellwig wpc->ops = ops; 1627598ecfbaSChristoph Hellwig ret = iomap_do_writepage(page, wbc, wpc); 1628598ecfbaSChristoph Hellwig if (!wpc->ioend) 1629598ecfbaSChristoph Hellwig return ret; 1630598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1631598ecfbaSChristoph Hellwig } 1632598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepage); 1633598ecfbaSChristoph Hellwig 1634598ecfbaSChristoph Hellwig int 1635598ecfbaSChristoph Hellwig iomap_writepages(struct address_space *mapping, struct writeback_control *wbc, 1636598ecfbaSChristoph Hellwig struct iomap_writepage_ctx *wpc, 1637598ecfbaSChristoph Hellwig const struct iomap_writeback_ops *ops) 1638598ecfbaSChristoph Hellwig { 1639598ecfbaSChristoph Hellwig int ret; 1640598ecfbaSChristoph Hellwig 1641598ecfbaSChristoph Hellwig wpc->ops = ops; 1642598ecfbaSChristoph Hellwig ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc); 1643598ecfbaSChristoph Hellwig if (!wpc->ioend) 1644598ecfbaSChristoph Hellwig return ret; 1645598ecfbaSChristoph Hellwig return iomap_submit_ioend(wpc, wpc->ioend, ret); 1646598ecfbaSChristoph Hellwig } 1647598ecfbaSChristoph Hellwig EXPORT_SYMBOL_GPL(iomap_writepages); 1648598ecfbaSChristoph Hellwig 1649598ecfbaSChristoph Hellwig static int __init iomap_init(void) 1650598ecfbaSChristoph Hellwig { 1651598ecfbaSChristoph Hellwig return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE), 1652598ecfbaSChristoph Hellwig offsetof(struct iomap_ioend, io_inline_bio), 1653598ecfbaSChristoph Hellwig BIOSET_NEED_BVECS); 1654598ecfbaSChristoph Hellwig } 1655598ecfbaSChristoph Hellwig fs_initcall(iomap_init); 1656