Use spl_fstrans_mark instead of memalloc_noio_save

For earlier versions of the kernel with memalloc_noio_save, it only turns
off __GFP_IO but leaves __GFP_FS untouched during direct reclaim. This
would cause threads to direct reclaim into ZFS and cause deadlock.

Instead, we should stick to using spl_fstrans_mark. Since we would
explicitly turn off both __GFP_IO and __GFP_FS before allocation, it
will work on every version of the kernel.

This impacts kernel versions 3.9-3.17, see upstream kernel commit
torvalds/linux@934f307 for reference.

Signed-off-by: Chunwei Chen <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Signed-off-by: Tim Chase <[email protected]>
Closes #515
Issue zfsonlinux/zfs#4111

4 files changed, 5 insertions, 37 deletions

diff --git a/module/spl/spl-kmem-cache.c b/module/spl/spl-kmem-cache.c
index 846d24300..e3edca5a0 100644
--- a/module/spl/spl-kmem-cache.c
+++ b/module/spl/spl-kmem-cache.c
@@ -201,7 +201,7 @@ kv_alloc(spl_kmem_cache_t *skc, int size, int flags)
 		ASSERT(ISP2(size));
 		ptr = (void *)__get_free_pages(lflags, get_order(size));
 	} else {
-		ptr = spl_vmalloc(size, lflags | __GFP_HIGHMEM, PAGE_KERNEL);
+		ptr = __vmalloc(size, lflags | __GFP_HIGHMEM, PAGE_KERNEL);
 	}
 
 	/* Resulting allocated memory will be page aligned */
@@ -1156,15 +1156,10 @@ spl_cache_grow_work(void *data)
 	spl_kmem_cache_t *skc = ska->ska_cache;
 	spl_kmem_slab_t *sks;
 
-#if defined(PF_MEMALLOC_NOIO)
-	unsigned noio_flag = memalloc_noio_save();
-	sks = spl_slab_alloc(skc, ska->ska_flags);
-	memalloc_noio_restore(noio_flag);
-#else
 	fstrans_cookie_t cookie = spl_fstrans_mark();
 	sks = spl_slab_alloc(skc, ska->ska_flags);
 	spl_fstrans_unmark(cookie);
-#endif
+
 	spin_lock(&skc->skc_lock);
 	if (sks) {
 		skc->skc_slab_total++;
diff --git a/module/spl/spl-kmem.c b/module/spl/spl-kmem.c
index 92263f8f4..0931fdfcf 100644
--- a/module/spl/spl-kmem.c
+++ b/module/spl/spl-kmem.c
@@ -185,7 +185,7 @@ spl_kmem_alloc_impl(size_t size, int flags, int node)
 		 */
 		if ((size > spl_kmem_alloc_max) || use_vmem) {
 			if (flags & KM_VMEM) {
-				ptr = spl_vmalloc(size, lflags, PAGE_KERNEL);
+				ptr = __vmalloc(size, lflags, PAGE_KERNEL);
 			} else {
 				return (NULL);
 			}
@@ -198,7 +198,7 @@ spl_kmem_alloc_impl(size_t size, int flags, int node)
 
 		/*
 		 * For vmem_alloc() and vmem_zalloc() callers retry immediately
-		 * using spl_vmalloc() which is unlikely to fail.
+		 * using __vmalloc() which is unlikely to fail.
 		 */
 		if ((flags & KM_VMEM) && (use_vmem == 0))  {
 			use_vmem = 1;
diff --git a/module/spl/spl-taskq.c b/module/spl/spl-taskq.c
index db5d8fba1..5f2b725c9 100644
--- a/module/spl/spl-taskq.c
+++ b/module/spl/spl-taskq.c
@@ -849,9 +849,7 @@ taskq_thread(void *args)
 	tq = tqt->tqt_tq;
 	current->flags |= PF_NOFREEZE;
 
-#if defined(PF_MEMALLOC_NOIO)
-	(void) memalloc_noio_save();
-#endif
+	(void) spl_fstrans_mark();
 
 	sigfillset(&blocked);
 	sigprocmask(SIG_BLOCK, &blocked, NULL);
diff --git a/module/spl/spl-vmem.c b/module/spl/spl-vmem.c
index bca27f263..e177988a7 100644
--- a/module/spl/spl-vmem.c
+++ b/module/spl/spl-vmem.c
@@ -97,31 +97,6 @@ spl_vmem_free(const void *buf, size_t size)
 }
 EXPORT_SYMBOL(spl_vmem_free);
 
-/*
- * Public vmalloc() interface designed to be safe to be called during I/O.
- */
-void *
-spl_vmalloc(unsigned long size, gfp_t lflags, pgprot_t prot)
-{
-#if defined(PF_MEMALLOC_NOIO)
-	void *ptr;
-	unsigned noio_flag = 0;
-
-	if (spl_fstrans_check())
-		noio_flag = memalloc_noio_save();
-
-	ptr =  __vmalloc(size, lflags, prot);
-
-	if (spl_fstrans_check())
-		memalloc_noio_restore(noio_flag);
-
-	return (ptr);
-#else
-	return (__vmalloc(size, lflags, prot));
-#endif
-}
-EXPORT_SYMBOL(spl_vmalloc);
-
 int
 spl_vmem_init(void)
 {