2 * Copyright (C) 2010 Red Hat, Inc.
3 * Copyright (c) 2016 Christoph Hellwig.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 #include <linux/module.h>
15 #include <linux/compiler.h>
17 #include <linux/iomap.h>
18 #include <linux/uaccess.h>
19 #include <linux/gfp.h>
21 #include <linux/swap.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/backing-dev.h>
26 #include <linux/buffer_head.h>
27 #include <linux/dax.h>
31 * Execute a iomap write on a segment of the mapping that spans a
32 * contiguous range of pages that have identical block mapping state.
34 * This avoids the need to map pages individually, do individual allocations
35 * for each page and most importantly avoid the need for filesystem specific
36 * locking per page. Instead, all the operations are amortised over the entire
37 * range of pages. It is assumed that the filesystems will lock whatever
38 * resources they require in the iomap_begin call, and release them in the
42 iomap_apply(struct inode *inode, loff_t pos, loff_t length, unsigned flags,
43 struct iomap_ops *ops, void *data, iomap_actor_t actor)
45 struct iomap iomap = { 0 };
46 loff_t written = 0, ret;
49 * Need to map a range from start position for length bytes. This can
50 * span multiple pages - it is only guaranteed to return a range of a
51 * single type of pages (e.g. all into a hole, all mapped or all
52 * unwritten). Failure at this point has nothing to undo.
54 * If allocation is required for this range, reserve the space now so
55 * that the allocation is guaranteed to succeed later on. Once we copy
56 * the data into the page cache pages, then we cannot fail otherwise we
57 * expose transient stale data. If the reserve fails, we can safely
58 * back out at this point as there is nothing to undo.
60 ret = ops->iomap_begin(inode, pos, length, flags, &iomap);
63 if (WARN_ON(iomap.offset > pos))
67 * Cut down the length to the one actually provided by the filesystem,
68 * as it might not be able to give us the whole size that we requested.
70 if (iomap.offset + iomap.length < pos + length)
71 length = iomap.offset + iomap.length - pos;
74 * Now that we have guaranteed that the space allocation will succeed.
75 * we can do the copy-in page by page without having to worry about
76 * failures exposing transient data.
78 written = actor(inode, pos, length, data, &iomap);
81 * Now the data has been copied, commit the range we've copied. This
82 * should not fail unless the filesystem has had a fatal error.
85 ret = ops->iomap_end(inode, pos, length,
86 written > 0 ? written : 0,
90 return written ? written : ret;
94 iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
96 loff_t i_size = i_size_read(inode);
99 * Only truncate newly allocated pages beyoned EOF, even if the
100 * write started inside the existing inode size.
102 if (pos + len > i_size)
103 truncate_pagecache_range(inode, max(pos, i_size), pos + len);
107 iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags,
108 struct page **pagep, struct iomap *iomap)
110 pgoff_t index = pos >> PAGE_SHIFT;
114 BUG_ON(pos + len > iomap->offset + iomap->length);
116 if (fatal_signal_pending(current))
119 page = grab_cache_page_write_begin(inode->i_mapping, index, flags);
123 status = __block_write_begin_int(page, pos, len, NULL, iomap);
124 if (unlikely(status)) {
129 iomap_write_failed(inode, pos, len);
137 iomap_write_end(struct inode *inode, loff_t pos, unsigned len,
138 unsigned copied, struct page *page)
142 ret = generic_write_end(NULL, inode->i_mapping, pos, len,
145 iomap_write_failed(inode, pos, len);
150 iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
153 struct iov_iter *i = data;
156 unsigned int flags = AOP_FLAG_NOFS;
159 * Copies from kernel address space cannot fail (NFSD is a big user).
161 if (!iter_is_iovec(i))
162 flags |= AOP_FLAG_UNINTERRUPTIBLE;
166 unsigned long offset; /* Offset into pagecache page */
167 unsigned long bytes; /* Bytes to write to page */
168 size_t copied; /* Bytes copied from user */
170 offset = (pos & (PAGE_SIZE - 1));
171 bytes = min_t(unsigned long, PAGE_SIZE - offset,
178 * Bring in the user page that we will copy from _first_.
179 * Otherwise there's a nasty deadlock on copying from the
180 * same page as we're writing to, without it being marked
183 * Not only is this an optimisation, but it is also required
184 * to check that the address is actually valid, when atomic
185 * usercopies are used, below.
187 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
192 status = iomap_write_begin(inode, pos, bytes, flags, &page,
194 if (unlikely(status))
197 if (mapping_writably_mapped(inode->i_mapping))
198 flush_dcache_page(page);
200 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
202 flush_dcache_page(page);
204 status = iomap_write_end(inode, pos, bytes, copied, page);
205 if (unlikely(status < 0))
211 iov_iter_advance(i, copied);
212 if (unlikely(copied == 0)) {
214 * If we were unable to copy any data at all, we must
215 * fall back to a single segment length write.
217 * If we didn't fallback here, we could livelock
218 * because not all segments in the iov can be copied at
219 * once without a pagefault.
221 bytes = min_t(unsigned long, PAGE_SIZE - offset,
222 iov_iter_single_seg_count(i));
229 balance_dirty_pages_ratelimited(inode->i_mapping);
230 } while (iov_iter_count(i) && length);
232 return written ? written : status;
236 iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter,
237 struct iomap_ops *ops)
239 struct inode *inode = iocb->ki_filp->f_mapping->host;
240 loff_t pos = iocb->ki_pos, ret = 0, written = 0;
242 while (iov_iter_count(iter)) {
243 ret = iomap_apply(inode, pos, iov_iter_count(iter),
244 IOMAP_WRITE, ops, iter, iomap_write_actor);
251 return written ? written : ret;
253 EXPORT_SYMBOL_GPL(iomap_file_buffered_write);
256 __iomap_read_page(struct inode *inode, loff_t offset)
258 struct address_space *mapping = inode->i_mapping;
261 page = read_mapping_page(mapping, offset >> PAGE_SHIFT, NULL);
264 if (!PageUptodate(page)) {
266 return ERR_PTR(-EIO);
272 iomap_dirty_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
279 struct page *page, *rpage;
280 unsigned long offset; /* Offset into pagecache page */
281 unsigned long bytes; /* Bytes to write to page */
283 offset = (pos & (PAGE_SIZE - 1));
284 bytes = min_t(loff_t, PAGE_SIZE - offset, length);
286 rpage = __iomap_read_page(inode, pos);
288 return PTR_ERR(rpage);
290 status = iomap_write_begin(inode, pos, bytes,
291 AOP_FLAG_NOFS | AOP_FLAG_UNINTERRUPTIBLE,
294 if (unlikely(status))
297 WARN_ON_ONCE(!PageUptodate(page));
299 status = iomap_write_end(inode, pos, bytes, bytes, page);
300 if (unlikely(status <= 0)) {
301 if (WARN_ON_ONCE(status == 0))
312 balance_dirty_pages_ratelimited(inode->i_mapping);
319 iomap_file_dirty(struct inode *inode, loff_t pos, loff_t len,
320 struct iomap_ops *ops)
325 ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL,
335 EXPORT_SYMBOL_GPL(iomap_file_dirty);
337 static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset,
338 unsigned bytes, struct iomap *iomap)
343 status = iomap_write_begin(inode, pos, bytes,
344 AOP_FLAG_UNINTERRUPTIBLE | AOP_FLAG_NOFS, &page, iomap);
348 zero_user(page, offset, bytes);
349 mark_page_accessed(page);
351 return iomap_write_end(inode, pos, bytes, bytes, page);
354 static int iomap_dax_zero(loff_t pos, unsigned offset, unsigned bytes,
357 sector_t sector = iomap->blkno +
358 (((pos & ~(PAGE_SIZE - 1)) - iomap->offset) >> 9);
360 return __dax_zero_page_range(iomap->bdev, sector, offset, bytes);
364 iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count,
365 void *data, struct iomap *iomap)
367 bool *did_zero = data;
371 /* already zeroed? we're done. */
372 if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN)
376 unsigned offset, bytes;
378 offset = pos & (PAGE_SIZE - 1); /* Within page */
379 bytes = min_t(loff_t, PAGE_SIZE - offset, count);
382 status = iomap_dax_zero(pos, offset, bytes, iomap);
384 status = iomap_zero(inode, pos, offset, bytes, iomap);
399 iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
400 struct iomap_ops *ops)
405 ret = iomap_apply(inode, pos, len, IOMAP_ZERO,
406 ops, did_zero, iomap_zero_range_actor);
416 EXPORT_SYMBOL_GPL(iomap_zero_range);
419 iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
420 struct iomap_ops *ops)
422 unsigned int blocksize = i_blocksize(inode);
423 unsigned int off = pos & (blocksize - 1);
425 /* Block boundary? Nothing to do */
428 return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
430 EXPORT_SYMBOL_GPL(iomap_truncate_page);
433 iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length,
434 void *data, struct iomap *iomap)
436 struct page *page = data;
439 ret = __block_write_begin_int(page, pos, length, NULL, iomap);
443 block_commit_write(page, 0, length);
447 int iomap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
448 struct iomap_ops *ops)
450 struct page *page = vmf->page;
451 struct inode *inode = file_inode(vma->vm_file);
452 unsigned long length;
457 size = i_size_read(inode);
458 if ((page->mapping != inode->i_mapping) ||
459 (page_offset(page) > size)) {
460 /* We overload EFAULT to mean page got truncated */
465 /* page is wholly or partially inside EOF */
466 if (((page->index + 1) << PAGE_SHIFT) > size)
467 length = size & ~PAGE_MASK;
471 offset = page_offset(page);
473 ret = iomap_apply(inode, offset, length, IOMAP_WRITE,
474 ops, page, iomap_page_mkwrite_actor);
475 if (unlikely(ret <= 0))
481 set_page_dirty(page);
482 wait_for_stable_page(page);
488 EXPORT_SYMBOL_GPL(iomap_page_mkwrite);
491 struct fiemap_extent_info *fi;
495 static int iomap_to_fiemap(struct fiemap_extent_info *fi,
496 struct iomap *iomap, u32 flags)
498 switch (iomap->type) {
503 flags |= FIEMAP_EXTENT_DELALLOC | FIEMAP_EXTENT_UNKNOWN;
505 case IOMAP_UNWRITTEN:
506 flags |= FIEMAP_EXTENT_UNWRITTEN;
512 if (iomap->flags & IOMAP_F_MERGED)
513 flags |= FIEMAP_EXTENT_MERGED;
514 if (iomap->flags & IOMAP_F_SHARED)
515 flags |= FIEMAP_EXTENT_SHARED;
517 return fiemap_fill_next_extent(fi, iomap->offset,
518 iomap->blkno != IOMAP_NULL_BLOCK ? iomap->blkno << 9: 0,
519 iomap->length, flags);
524 iomap_fiemap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
527 struct fiemap_ctx *ctx = data;
530 if (iomap->type == IOMAP_HOLE)
533 ret = iomap_to_fiemap(ctx->fi, &ctx->prev, 0);
536 case 0: /* success */
538 case 1: /* extent array full */
545 int iomap_fiemap(struct inode *inode, struct fiemap_extent_info *fi,
546 loff_t start, loff_t len, struct iomap_ops *ops)
548 struct fiemap_ctx ctx;
551 memset(&ctx, 0, sizeof(ctx));
553 ctx.prev.type = IOMAP_HOLE;
555 ret = fiemap_check_flags(fi, FIEMAP_FLAG_SYNC);
559 if (fi->fi_flags & FIEMAP_FLAG_SYNC) {
560 ret = filemap_write_and_wait(inode->i_mapping);
566 ret = iomap_apply(inode, start, len, IOMAP_REPORT, ops, &ctx,
568 /* inode with no (attribute) mapping will give ENOENT */
580 if (ctx.prev.type != IOMAP_HOLE) {
581 ret = iomap_to_fiemap(fi, &ctx.prev, FIEMAP_EXTENT_LAST);
588 EXPORT_SYMBOL_GPL(iomap_fiemap);