bcachefs: Wrap write path in memalloc_nofs_save()

This fixes a lockdep splat where we're allocating memory with vmalloc in
the compression bounce path, which doesn't always obey GFP_NOFS.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

diff --git a/fs/bcachefs/compress.c b/fs/bcachefs/compress.c
index 920460a182b4efdb5c7155e8bd8f4be914207431..595d76aa3956bae67711f0fda5db687e37582cc0 100644 (file)
--- a/fs/bcachefs/compress.c
+++ b/fs/bcachefs/compress.c
@@ -7,7 +7,6 @@
 #include "super-io.h"
 
 #include <linux/lz4.h>
-#include <linux/sched/mm.h>
 #include <linux/zlib.h>
 #include <linux/zstd.h>
 
@@ -64,7 +63,7 @@ static struct bbuf __bio_map_or_bounce(struct bch_fs *c, struct bio *bio,
        struct bbuf ret;
        struct bio_vec bv;
        struct bvec_iter iter;
-       unsigned nr_pages = 0, flags;
+       unsigned nr_pages = 0;
        struct page *stack_pages[16];
        struct page **pages = NULL;
        void *data;
@@ -104,10 +103,7 @@ static struct bbuf __bio_map_or_bounce(struct bch_fs *c, struct bio *bio,
        __bio_for_each_segment(bv, bio, iter, start)
                pages[nr_pages++] = bv.bv_page;
 
-       flags = memalloc_nofs_save();
        data = vmap(pages, nr_pages, VM_MAP, PAGE_KERNEL);
-       memalloc_nofs_restore(flags);
-
        if (pages != stack_pages)
                kfree(pages);
 

diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c
index be59b615b2db89ee56630132813e86be82afa59e..ae2688d3aee60f4e24695cd5bc2793ed0a2f4313 100644 (file)
--- a/fs/bcachefs/io.c
+++ b/fs/bcachefs/io.c
@@ -32,6 +32,7 @@
 
 #include <linux/blkdev.h>
 #include <linux/random.h>
+#include <linux/sched/mm.h>
 
 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
 
@@ -1052,7 +1053,10 @@ static void __bch2_write(struct closure *cl)
        struct write_point *wp;
        struct bio *bio;
        bool skip_put = true;
+       unsigned nofs_flags;
        int ret;
+
+       nofs_flags = memalloc_nofs_save();
 again:
        memset(&op->failed, 0, sizeof(op->failed));
 
@@ -1134,13 +1138,15 @@ again:
 
        if (!skip_put)
                continue_at(cl, bch2_write_index, index_update_wq(op));
+out:
+       memalloc_nofs_restore(nofs_flags);
        return;
 err:
        op->error = ret;
        op->flags |= BCH_WRITE_DONE;
 
        continue_at(cl, bch2_write_index, index_update_wq(op));
-       return;
+       goto out;
 flush_io:
        /*
         * If the write can't all be submitted at once, we generally want to
@@ -1151,7 +1157,7 @@ flush_io:
         */
        if (current->flags & PF_WQ_WORKER) {
                continue_at(cl, bch2_write_index, index_update_wq(op));
-               return;
+               goto out;
        }
 
        closure_sync(cl);
@@ -1162,7 +1168,7 @@ flush_io:
                if (op->error) {
                        op->flags |= BCH_WRITE_DONE;
                        continue_at_nobarrier(cl, bch2_write_done, NULL);
-                       return;
+                       goto out;
                }
        }