Re: [PATCH] iomap: fix iomap_dio_zero() for fs bs > system page size

[Pankaj Raghav <p.raghav@xxxxxxxxxxx>]

On 27/10/2023 12:47, Matthew Wilcox wrote:
> On Fri, Oct 27, 2023 at 10:03:15AM +0200, Pankaj Raghav wrote:
>> I also noticed this pattern in fscrypt_zeroout_range_inline_crypt().
>> Probably there are more places which could use a ZERO_FOLIO directly
>> instead of iterating with ZERO_PAGE.
>>
>> Chinner also had a similar comment. It would be nice if we can reserve
>> a zero huge page that is the size of MAX_PAGECACHE_ORDER and add it as
>> one folio to the bio.
> 
> i'm on holiday atm.  start looking at mm_get_huge_zero_page()

Thanks for the pointer. I made a rough version of how it might
look like if I use that API:

diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
index bcd3f8cf5ea4..6ae21bd16dbe 100644
--- a/fs/iomap/direct-io.c
+++ b/fs/iomap/direct-io.c
@@ -236,17 +236,43 @@ static void iomap_dio_zero(const struct iomap_iter *iter, struct iomap_dio *dio,
                loff_t pos, unsigned len)
 {
        struct inode *inode = file_inode(dio->iocb->ki_filp);
-       struct page *page = ZERO_PAGE(0);
+       struct page *page = NULL;
+       bool huge_page = false;
+       bool fallback = false;
        struct bio *bio;

-       bio = iomap_dio_alloc_bio(iter, dio, 1, REQ_OP_WRITE | REQ_SYNC | REQ_IDLE);
+       if (len > PAGE_SIZE) {
+               page = mm_get_huge_zero_page(current->mm);
+               if (likely(page))
+                       huge_page = true;
+       }
+
+       if (!huge_page)
+               page = ZERO_PAGE(0);
+
+       fallback = ((len > PAGE_SIZE) && !huge_page);
+
+       bio = iomap_dio_alloc_bio(iter, dio, BIO_MAX_VECS,
+                                 REQ_OP_WRITE | REQ_SYNC | REQ_IDLE);
        fscrypt_set_bio_crypt_ctx(bio, inode, pos >> inode->i_blkbits,
                                  GFP_KERNEL);
+
        bio->bi_iter.bi_sector = iomap_sector(&iter->iomap, pos);
        bio->bi_private = dio;
        bio->bi_end_io = iomap_dio_bio_end_io;

-       __bio_add_page(bio, page, len, 0);
+       if (!fallback) {
+               bio_add_folio_nofail(bio, page_folio(page), len, 0);
+       } else {
+               while (len) {
+                       unsigned int io_len =
+                               min_t(unsigned int, len, PAGE_SIZE);
+
+                       __bio_add_page(bio, page, io_len, 0);
+                       len -= io_len;
+               }
+       }
+
        iomap_dio_submit_bio(iter, dio, bio, pos);
 }

The only issue with mm_get_huge_zero_page() is that it can fail, and we need
to fallback to iteration if it cannot allocate a huge folio.

PS: Enjoy your holidays :)