]> git.baikalelectronics.ru Git - kernel.git/commitdiff
mm: handle THP in swap_*page_fs()
authorNeilBrown <neilb@suse.de>
Tue, 10 May 2022 01:20:49 +0000 (18:20 -0700)
committerakpm <akpm@linux-foundation.org>
Tue, 10 May 2022 01:20:49 +0000 (18:20 -0700)
Pages passed to swap_readpage()/swap_writepage() are not necessarily all
the same size - there may be transparent-huge-pages involves.

The BIO paths of swap_*page() handle this correctly, but the SWP_FS_OPS
path does not.

So we need to use thp_size() to find the size, not just assume PAGE_SIZE,
and we need to track the total length of the request, not just assume it
is "page * PAGE_SIZE".

Link: https://lkml.kernel.org/r/165119301488.15698.9457662928942765453.stgit@noble.brown
Signed-off-by: NeilBrown <neilb@suse.de>
Reported-by: Miaohe Lin <linmiaohe@huawei.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Howells <dhowells@redhat.com>
Cc: Geert Uytterhoeven <geert+renesas@glider.be>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/page_io.c

index c132511f521c5255bf25cf49efb64af530f6f109..d636a3531cad1382f36598fbd6ff760faff4f93e 100644 (file)
@@ -239,6 +239,7 @@ struct swap_iocb {
        struct kiocb            iocb;
        struct bio_vec          bvec[SWAP_CLUSTER_MAX];
        int                     pages;
+       int                     len;
 };
 static mempool_t *sio_pool;
 
@@ -261,7 +262,7 @@ static void sio_write_complete(struct kiocb *iocb, long ret)
        struct page *page = sio->bvec[0].bv_page;
        int p;
 
-       if (ret != PAGE_SIZE * sio->pages) {
+       if (ret != sio->len) {
                /*
                 * In the case of swap-over-nfs, this can be a
                 * temporary failure if the system has limited
@@ -301,7 +302,7 @@ static int swap_writepage_fs(struct page *page, struct writeback_control *wbc)
                sio = *wbc->swap_plug;
        if (sio) {
                if (sio->iocb.ki_filp != swap_file ||
-                   sio->iocb.ki_pos + sio->pages * PAGE_SIZE != pos) {
+                   sio->iocb.ki_pos + sio->len != pos) {
                        swap_write_unplug(sio);
                        sio = NULL;
                }
@@ -312,10 +313,12 @@ static int swap_writepage_fs(struct page *page, struct writeback_control *wbc)
                sio->iocb.ki_complete = sio_write_complete;
                sio->iocb.ki_pos = pos;
                sio->pages = 0;
+               sio->len = 0;
        }
        sio->bvec[sio->pages].bv_page = page;
-       sio->bvec[sio->pages].bv_len = PAGE_SIZE;
+       sio->bvec[sio->pages].bv_len = thp_size(page);
        sio->bvec[sio->pages].bv_offset = 0;
+       sio->len += thp_size(page);
        sio->pages += 1;
        if (sio->pages == ARRAY_SIZE(sio->bvec) || !wbc->swap_plug) {
                swap_write_unplug(sio);
@@ -371,8 +374,7 @@ void swap_write_unplug(struct swap_iocb *sio)
        struct address_space *mapping = sio->iocb.ki_filp->f_mapping;
        int ret;
 
-       iov_iter_bvec(&from, WRITE, sio->bvec, sio->pages,
-                     PAGE_SIZE * sio->pages);
+       iov_iter_bvec(&from, WRITE, sio->bvec, sio->pages, sio->len);
        ret = mapping->a_ops->swap_rw(&sio->iocb, &from);
        if (ret != -EIOCBQUEUED)
                sio_write_complete(&sio->iocb, ret);
@@ -383,7 +385,7 @@ static void sio_read_complete(struct kiocb *iocb, long ret)
        struct swap_iocb *sio = container_of(iocb, struct swap_iocb, iocb);
        int p;
 
-       if (ret == PAGE_SIZE * sio->pages) {
+       if (ret == sio->len) {
                for (p = 0; p < sio->pages; p++) {
                        struct page *page = sio->bvec[p].bv_page;
 
@@ -415,7 +417,7 @@ static void swap_readpage_fs(struct page *page,
                sio = *plug;
        if (sio) {
                if (sio->iocb.ki_filp != sis->swap_file ||
-                   sio->iocb.ki_pos + sio->pages * PAGE_SIZE != pos) {
+                   sio->iocb.ki_pos + sio->len != pos) {
                        swap_read_unplug(sio);
                        sio = NULL;
                }
@@ -426,10 +428,12 @@ static void swap_readpage_fs(struct page *page,
                sio->iocb.ki_pos = pos;
                sio->iocb.ki_complete = sio_read_complete;
                sio->pages = 0;
+               sio->len = 0;
        }
        sio->bvec[sio->pages].bv_page = page;
-       sio->bvec[sio->pages].bv_len = PAGE_SIZE;
+       sio->bvec[sio->pages].bv_len = thp_size(page);
        sio->bvec[sio->pages].bv_offset = 0;
+       sio->len += thp_size(page);
        sio->pages += 1;
        if (sio->pages == ARRAY_SIZE(sio->bvec) || !plug) {
                swap_read_unplug(sio);
@@ -521,8 +525,7 @@ void __swap_read_unplug(struct swap_iocb *sio)
        struct address_space *mapping = sio->iocb.ki_filp->f_mapping;
        int ret;
 
-       iov_iter_bvec(&from, READ, sio->bvec, sio->pages,
-                     PAGE_SIZE * sio->pages);
+       iov_iter_bvec(&from, READ, sio->bvec, sio->pages, sio->len);
        ret = mapping->a_ops->swap_rw(&sio->iocb, &from);
        if (ret != -EIOCBQUEUED)
                sio_read_complete(&sio->iocb, ret);