Revert "btrfs: turn name_cache radix tree into XArray in send_ctx"

This reverts commit 4076942021fe14efecae33bf98566df6dd5ae6f7.

Revert the xarray conversion, there's a problem with potential
sleep-inside-spinlock [1] when calling xa_insert that triggers GFP_NOFS
allocation. The radix tree used the preloading mechanism to avoid
sleeping but this is not available in xarray.

Conversion from spin lock to mutex is possible but at time of rc6 is
riskier than a clean revert.

[1] https://lore.kernel.org/linux-btrfs/cover.1657097693.git.fdmanana@suse.com/

Reported-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c
index 5a05bea..d6a2428 100644 (file)
--- a/fs/btrfs/send.c
+++ b/fs/btrfs/send.c
@@ -10,6 +10,7 @@
 #include <linux/mount.h>
 #include <linux/xattr.h>
 #include <linux/posix_acl_xattr.h>
+#include <linux/radix-tree.h>
 #include <linux/vmalloc.h>
 #include <linux/string.h>
 #include <linux/compat.h>
@@ -127,7 +128,7 @@ struct send_ctx {
        struct list_head new_refs;
        struct list_head deleted_refs;
 
-       struct xarray name_cache;
+       struct radix_tree_root name_cache;
        struct list_head name_cache_list;
        int name_cache_size;
 
@@ -268,13 +269,14 @@ struct orphan_dir_info {
 struct name_cache_entry {
        struct list_head list;
        /*
-        * On 32bit kernels, xarray has only 32bit indices, but we need to
-        * handle 64bit inums. We use the lower 32bit of the 64bit inum to store
-        * it in the tree. If more than one inum would fall into the same entry,
-        * we use inum_aliases to store the additional entries. inum_aliases is
-        * also used to store entries with the same inum but different generations.
+        * radix_tree has only 32bit entries but we need to handle 64bit inums.
+        * We use the lower 32bit of the 64bit inum to store it in the tree. If
+        * more then one inum would fall into the same entry, we use radix_list
+        * to store the additional entries. radix_list is also used to store
+        * entries where two entries have the same inum but different
+        * generations.
         */
-       struct list_head inum_aliases;
+       struct list_head radix_list;
        u64 ino;
        u64 gen;
        u64 parent_ino;
@@ -2024,9 +2026,9 @@ out:
 }
 
 /*
- * Insert a name cache entry. On 32bit kernels the xarray index is 32bit,
+ * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
  * so we need to do some special handling in case we have clashes. This function
- * takes care of this with the help of name_cache_entry::inum_aliases.
+ * takes care of this with the help of name_cache_entry::radix_list.
  * In case of error, nce is kfreed.
  */
 static int name_cache_insert(struct send_ctx *sctx,
@@ -2035,7 +2037,8 @@ static int name_cache_insert(struct send_ctx *sctx,
        int ret = 0;
        struct list_head *nce_head;
 
-       nce_head = xa_load(&sctx->name_cache, (unsigned long)nce->ino);
+       nce_head = radix_tree_lookup(&sctx->name_cache,
+                       (unsigned long)nce->ino);
        if (!nce_head) {
                nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
                if (!nce_head) {
@@ -2044,14 +2047,14 @@ static int name_cache_insert(struct send_ctx *sctx,
                }
                INIT_LIST_HEAD(nce_head);
 
-               ret = xa_insert(&sctx->name_cache, nce->ino, nce_head, GFP_KERNEL);
+               ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
                if (ret < 0) {
                        kfree(nce_head);
                        kfree(nce);
                        return ret;
                }
        }
-       list_add_tail(&nce->inum_aliases, nce_head);
+       list_add_tail(&nce->radix_list, nce_head);
        list_add_tail(&nce->list, &sctx->name_cache_list);
        sctx->name_cache_size++;
 
@@ -2063,14 +2066,15 @@ static void name_cache_delete(struct send_ctx *sctx,
 {
        struct list_head *nce_head;
 
-       nce_head = xa_load(&sctx->name_cache, (unsigned long)nce->ino);
+       nce_head = radix_tree_lookup(&sctx->name_cache,
+                       (unsigned long)nce->ino);
        if (!nce_head) {
                btrfs_err(sctx->send_root->fs_info,
              "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
                        nce->ino, sctx->name_cache_size);
        }
 
-       list_del(&nce->inum_aliases);
+       list_del(&nce->radix_list);
        list_del(&nce->list);
        sctx->name_cache_size--;
 
@@ -2078,7 +2082,7 @@ static void name_cache_delete(struct send_ctx *sctx,
         * We may not get to the final release of nce_head if the lookup fails
         */
        if (nce_head && list_empty(nce_head)) {
-               xa_erase(&sctx->name_cache, (unsigned long)nce->ino);
+               radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
                kfree(nce_head);
        }
 }
@@ -2089,11 +2093,11 @@ static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
        struct list_head *nce_head;
        struct name_cache_entry *cur;
 
-       nce_head = xa_load(&sctx->name_cache, (unsigned long)ino);
+       nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
        if (!nce_head)
                return NULL;
 
-       list_for_each_entry(cur, nce_head, inum_aliases) {
+       list_for_each_entry(cur, nce_head, radix_list) {
                if (cur->ino == ino && cur->gen == gen)
                        return cur;
        }
@@ -7518,7 +7522,7 @@ long btrfs_ioctl_send(struct inode *inode, struct btrfs_ioctl_send_args *arg)
 
        INIT_LIST_HEAD(&sctx->new_refs);
        INIT_LIST_HEAD(&sctx->deleted_refs);
-       xa_init_flags(&sctx->name_cache, GFP_KERNEL);
+       INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
        INIT_LIST_HEAD(&sctx->name_cache_list);
 
        sctx->flags = arg->flags;