@@ -21,6 +21,7 @@
#include <linux/filelock.h>
#include <linux/splice.h>
#include <linux/task_io_accounting_ops.h>
+#include <linux/iomap.h>
static int fuse_send_open(struct fuse_mount *fm, u64 nodeid,
unsigned int open_flags, int opcode,
@@ -1419,6 +1420,63 @@ static void fuse_dio_unlock(struct kiocb *iocb, bool exclusive)
}
}
+static struct folio *fuse_iomap_get_folio(struct iomap_iter *iter,
+ loff_t pos, unsigned len)
+{
+ struct file *file = (struct file *)iter->private;
+ struct inode *inode = iter->inode;
+ struct folio *folio;
+ loff_t fsize;
+
+ folio = iomap_get_folio(iter, pos, len);
+ if (IS_ERR(folio))
+ return folio;
+
+ fuse_wait_on_folio_writeback(inode, folio);
+
+ if (folio_test_uptodate(folio))
+ return folio;
+
+ /*
+ * If we're going to write past EOF then avoid the read, but zero the
+ * whole thing and mark it uptodate so that if we get a short write we
+ * don't try to re-read this page, we just carry on.
+ */
+ fsize = i_size_read(inode);
+ if (fsize <= folio_pos(folio)) {
+ folio_zero_range(folio, 0, folio_size(folio));
+ } else {
+ int err = fuse_do_readpage(file, &folio->page);
+ if (err) {
+ folio_unlock(folio);
+ folio_put(folio);
+ return ERR_PTR(err);
+ }
+ }
+
+ return folio;
+}
+
+static const struct iomap_folio_ops fuse_iomap_folio_ops = {
+ .get_folio = fuse_iomap_get_folio,
+};
+
+static int fuse_iomap_begin_write(struct inode *inode, loff_t pos, loff_t length,
+ unsigned flags, struct iomap *iomap,
+ struct iomap *srcmap)
+{
+ iomap->type = IOMAP_DELALLOC;
+ iomap->addr = IOMAP_NULL_ADDR;
+ iomap->offset = pos;
+ iomap->length = length;
+ iomap->folio_ops = &fuse_iomap_folio_ops;
+ return 0;
+}
+
+static const struct iomap_ops fuse_iomap_write_ops = {
+ .iomap_begin = fuse_iomap_begin_write,
+};
+
static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from)
{
struct file *file = iocb->ki_filp;
@@ -1427,6 +1485,7 @@ static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from)
struct inode *inode = mapping->host;
ssize_t err, count;
struct fuse_conn *fc = get_fuse_conn(inode);
+ bool writethrough = (fc->writeback_cache == 0);
if (fc->writeback_cache) {
/* Update size (EOF optimization) and mode (SUID clearing) */
@@ -1437,14 +1496,10 @@ static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from)
if (fc->handle_killpriv_v2 &&
setattr_should_drop_suidgid(&nop_mnt_idmap,
- file_inode(file))) {
- goto writethrough;
- }
-
- return generic_file_write_iter(iocb, from);
+ file_inode(file)))
+ writethrough = true;
}
-writethrough:
inode_lock(inode);
err = count = generic_write_checks(iocb, from);
@@ -1463,8 +1518,12 @@ static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from)
goto out;
written = direct_write_fallback(iocb, from, written,
fuse_perform_write(iocb, from));
- } else {
+ } else if (writethrough) {
written = fuse_perform_write(iocb, from);
+ } else {
+ written = iomap_file_buffered_write(iocb, from,
+ &fuse_iomap_write_ops,
+ file);
}
out:
inode_unlock(inode);
@@ -2407,85 +2466,6 @@ static int fuse_writepages(struct address_space *mapping,
return err;
}
-/*
- * It's worthy to make sure that space is reserved on disk for the write,
- * but how to implement it without killing performance need more thinking.
- */
-static int fuse_write_begin(struct file *file, struct address_space *mapping,
- loff_t pos, unsigned len, struct page **pagep, void **fsdata)
-{
- pgoff_t index = pos >> PAGE_SHIFT;
- struct fuse_conn *fc = get_fuse_conn(file_inode(file));
- struct page *page;
- loff_t fsize;
- int err = -ENOMEM;
-
- WARN_ON(!fc->writeback_cache);
-
- page = grab_cache_page_write_begin(mapping, index);
- if (!page)
- goto error;
-
- fuse_wait_on_page_writeback(mapping->host, page->index);
-
- if (PageUptodate(page) || len == PAGE_SIZE)
- goto success;
- /*
- * Check if the start this page comes after the end of file, in which
- * case the readpage can be optimized away.
- */
- fsize = i_size_read(mapping->host);
- if (fsize <= (pos & PAGE_MASK)) {
- size_t off = pos & ~PAGE_MASK;
- if (off)
- zero_user_segment(page, 0, off);
- goto success;
- }
- err = fuse_do_readpage(file, page);
- if (err)
- goto cleanup;
-success:
- *pagep = page;
- return 0;
-
-cleanup:
- unlock_page(page);
- put_page(page);
-error:
- return err;
-}
-
-static int fuse_write_end(struct file *file, struct address_space *mapping,
- loff_t pos, unsigned len, unsigned copied,
- struct page *page, void *fsdata)
-{
- struct inode *inode = page->mapping->host;
-
- /* Haven't copied anything? Skip zeroing, size extending, dirtying. */
- if (!copied)
- goto unlock;
-
- pos += copied;
- if (!PageUptodate(page)) {
- /* Zero any unwritten bytes at the end of the page */
- size_t endoff = pos & ~PAGE_MASK;
- if (endoff)
- zero_user_segment(page, endoff, PAGE_SIZE);
- SetPageUptodate(page);
- }
-
- if (pos > inode->i_size)
- i_size_write(inode, pos);
-
- set_page_dirty(page);
-
-unlock:
- unlock_page(page);
- put_page(page);
-
- return copied;
-}
-
static int fuse_launder_folio(struct folio *folio)
{
int err = 0;
@@ -3351,8 +3331,6 @@ static const struct address_space_operations fuse_file_aops = {
.migrate_folio = filemap_migrate_folio,
.bmap = fuse_bmap,
.direct_IO = fuse_direct_IO,
- .write_begin = fuse_write_begin,
- .write_end = fuse_write_end,
};
void fuse_init_file_inode(struct inode *inode, unsigned int flags)
We're currently using the old ->write_begin()/->write_end() method of doing buffered writes. This isn't a huge deal for fuse since we basically just want to copy the pages and move on, but the iomap infrastructure gives us access to having huge folios. Rework the buffered write path when we have writeback cache to use the iomap buffered write code, the ->get_folio() callback now handles the work that we did in ->write_begin(), the rest of the work is handled inside of iomap so we don't need a replacement for ->write_end. Signed-off-by: Josef Bacik <josef@toxicpanda.com> --- fs/fuse/file.c | 154 +++++++++++++++++++++---------------------------- 1 file changed, 66 insertions(+), 88 deletions(-)