dax: fix ENOMEM handling in grab_mapping_entry()
authorJan Kara <jack@suse.cz>
Tue, 29 Jun 2021 02:35:04 +0000 (19:35 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 29 Jun 2021 17:53:47 +0000 (10:53 -0700)
grab_mapping_entry() has a bug in handling of ENOMEM condition.  Suppose
we have a PMD entry at index i which we are downgrading to a PTE entry.
grab_mapping_entry() will set pmd_downgrade to true, lock the entry, clear
the entry in xarray, and decrement mapping->nrpages.  The it will call:

entry = dax_make_entry(pfn_to_pfn_t(0), flags);
dax_lock_entry(xas, entry);

which inserts new PTE entry into xarray.  However this may fail allocating
the new node.  We handle this by:

if (xas_nomem(xas, mapping_gfp_mask(mapping) & ~__GFP_HIGHMEM))
goto retry;

however pmd_downgrade stays set to true even though 'entry' returned from
get_unlocked_entry() will be NULL now.  And we will go again through the
downgrade branch.  This is mostly harmless except that mapping->nrpages is
decremented again and we temporarily have an invalid entry stored in
xarray.  Fix the problem by setting pmd_downgrade to false each time we
lookup the entry we work with so that it matches the entry we found.

Link: https://lkml.kernel.org/r/20210622160015.18004-1-jack@suse.cz
Fixes: b15cd800682f ("dax: Convert page fault handlers to XArray")
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/dax.c

index 62352cb..da41f93 100644 (file)
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -488,10 +488,11 @@ static void *grab_mapping_entry(struct xa_state *xas,
                struct address_space *mapping, unsigned int order)
 {
        unsigned long index = xas->xa_index;
-       bool pmd_downgrade = false; /* splitting PMD entry into PTE entries? */
+       bool pmd_downgrade;     /* splitting PMD entry into PTE entries? */
        void *entry;
 
 retry:
+       pmd_downgrade = false;
        xas_lock_irq(xas);
        entry = get_unlocked_entry(xas, order);