OSDN Git Service

gfs2: Align read and write chunks to the page cache
authorAndreas Gruenbacher <agruenba@redhat.com>
Thu, 5 May 2022 11:32:23 +0000 (13:32 +0200)
committerAndreas Gruenbacher <agruenba@redhat.com>
Fri, 13 May 2022 20:00:22 +0000 (22:00 +0200)
Align the chunks that reads and writes are carried out in to the page
cache rather than the user buffers.  This will be more efficient in
general, especially for allocating writes.  Optimizing the case that the
user buffer is gfs2 backed isn't very useful; we only need to make sure
we won't deadlock.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
fs/gfs2/file.c

index ea87bef..11c4640 100644 (file)
@@ -771,6 +771,7 @@ static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 }
 
 static inline bool should_fault_in_pages(struct iov_iter *i,
+                                        struct kiocb *iocb,
                                         size_t *prev_count,
                                         size_t *window_size)
 {
@@ -783,15 +784,13 @@ static inline bool should_fault_in_pages(struct iov_iter *i,
                return false;
 
        size = PAGE_SIZE;
-       offs = offset_in_page(i->iov[0].iov_base + i->iov_offset);
+       offs = offset_in_page(iocb->ki_pos);
        if (*prev_count != count || !*window_size) {
                size_t nr_dirtied;
 
-               size = ALIGN(offs + count, PAGE_SIZE);
-               size = min_t(size_t, size, SZ_1M);
                nr_dirtied = max(current->nr_dirtied_pause -
                                 current->nr_dirtied, 8);
-               size = min(size, nr_dirtied << PAGE_SHIFT);
+               size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT);
        }
 
        *prev_count = count;
@@ -845,7 +844,7 @@ retry_under_glock:
        if (ret > 0)
                read = ret;
 
-       if (should_fault_in_pages(to, &prev_count, &window_size)) {
+       if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
                gfs2_holder_allow_demote(gh);
                window_size -= fault_in_iov_iter_writeable(to, window_size);
                gfs2_holder_disallow_demote(gh);
@@ -916,7 +915,7 @@ retry_under_glock:
        if (ret > 0)
                written = ret;
 
-       if (should_fault_in_pages(from, &prev_count, &window_size)) {
+       if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
                gfs2_holder_allow_demote(gh);
                window_size -= fault_in_iov_iter_readable(from, window_size);
                gfs2_holder_disallow_demote(gh);
@@ -984,7 +983,7 @@ retry_under_glock:
        if (ret > 0)
                read += ret;
 
-       if (should_fault_in_pages(to, &prev_count, &window_size)) {
+       if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
                gfs2_holder_allow_demote(&gh);
                window_size -= fault_in_iov_iter_writeable(to, window_size);
                gfs2_holder_disallow_demote(&gh);
@@ -1061,7 +1060,7 @@ retry_under_glock:
                goto out_unlock;
 
        from->count = orig_count - written;
-       if (should_fault_in_pages(from, &prev_count, &window_size)) {
+       if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
                gfs2_holder_allow_demote(gh);
                window_size -= fault_in_iov_iter_readable(from, window_size);
                gfs2_holder_disallow_demote(gh);