EX-8421 llite: disable kernel readahead for pcc mmap

Set ra_pages to 0 for PCC files when mmaped, because
otherwise this setting carries through to Lustre and will
cause crashes and possible inconsistencies.  This happens
because the PCC file and Lustre file share a mapping, which
is a weird trick required to have mmap work on PCC.

Add a set of asserts which confirm kernel readahead is
disabled and wasn't used for mmap.

Signed-off-by: Patrick Farrell <pfarrell@whamcloud.com>
Change-Id: I117042d68fac25158e8141c243acba698cf1930f
Reviewed-on: https://review.whamcloud.com/c/ex/lustre-release/+/52732
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Qian Yingjin <qian@ddn.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>

diff --git a/lustre/llite/pcc.c b/lustre/llite/pcc.c
index 7997c09..0f8562c 100644 (file)
--- a/lustre/llite/pcc.c
+++ b/lustre/llite/pcc.c
@@ -3194,6 +3194,14 @@ int pcc_file_mmap(struct file *file, struct vm_area_struct *vma,
        if (!pcc_file || !file_inode(pcc_file)->i_fop->mmap)
                RETURN(0);
 
+       /* with PCC, the files are created in an unusual way, then we do some
+        * special magic with mmap to allow Lustre and PCC to share the mmap,
+        * so we must manually set the ra_pages or we can get kernel readahead
+        * occurring (which Lustre does not support)
+        */
+       pcc_file->f_ra.ra_pages = 0;
+       file->f_ra.ra_pages = 0;
+
        pcc_inode_lock(inode);
        pcci = ll_i2pcci(inode);
        if (pcci && pcc_inode_has_layout(pcci)) {

diff --git a/lustre/llite/rw.c b/lustre/llite/rw.c
index 44d2ef6..f628905 100644 (file)
--- a/lustre/llite/rw.c
+++ b/lustre/llite/rw.c
@@ -1885,10 +1885,12 @@ int ll_readpage(struct file *file, struct page *vmpage)
        struct inode *inode = file_inode(file);
        struct cl_object *clob = ll_i2info(inode)->lli_clob;
        struct ll_sb_info *sbi = ll_i2sbi(inode);
+       struct super_block *sb = inode->i_sb;
        const struct lu_env *env = NULL;
        struct cl_read_ahead ra = { 0 };
        struct ll_cl_context *lcc;
        struct cl_io *io = NULL;
+       bool ra_assert = false;
        struct cl_page *page;
        int result;
        ENTRY;
@@ -2033,6 +2035,32 @@ int ll_readpage(struct file *file, struct page *vmpage)
                }
        }
 
+       /* this is a sequence of checks verifying that kernel readahead is
+        * truly disabled
+        */
+       if (lcc && lcc->lcc_type == LCC_MMAP) {
+               if (io->u.ci_fault.ft_index != vmpage->index) {
+                       CERROR("ft_index %lu, vmpage index %lu\n",
+                          io->u.ci_fault.ft_index, vmpage->index);
+                       ra_assert = true;
+               }
+       }
+
+       if (ra_assert || sb->s_bdi->ra_pages != 0 || file->f_ra.ra_pages != 0) {
+               CERROR("sbi ra pages %lu, file ra pages %d\n",
+                      sb->s_bdi->ra_pages, file->f_ra.ra_pages);
+               ra_assert = true;
+       }
+
+
+#ifdef HAVE_BDI_IO_PAGES
+       if (ra_assert || sb->s_bdi->io_pages != 0) {
+               CERROR("bdi io_pages %lu\n", sb->s_bdi->io_pages);
+               ra_assert = true;
+       }
+#endif
+       if (ra_assert)
+               LBUG();
        /**
         * Direct read can fall back to buffered read, but DIO is done
         * with lockless i/o, and buffered requires LDLM locking, so in