diff mbox

[2/2] Btrfs: fix invalid attempt to free reserved space on failure to cow range

Message ID 1488843186-8831-1-git-send-email-fdmanana@kernel.org (mailing list archive)
State Superseded, archived
Headers show

Commit Message

Filipe Manana March 6, 2017, 11:33 p.m. UTC
From: Filipe Manana <fdmanana@suse.com>

When attempting to COW a file range (we are starting writeback and doing
COW), if we manage to reserve an extent for the range we will write into
but fail after reserving it and before creating the respective ordered
extent, we end up in an error path where we attempt to decrement the
data space's reserved_bytes counter after we already did it while
reserving the extent, leading to a warning/trace like the following:

[  847.621524] ------------[ cut here ]------------
[  847.625441] WARNING: CPU: 5 PID: 4905 at fs/btrfs/extent-tree.c:4316 btrfs_free_reserved_data_space_noquota+0x60/0x9f [btrfs]
[  847.633704] Modules linked in: btrfs crc32c_generic xor raid6_pq acpi_cpufreq i2c_piix4 ppdev psmouse tpm_tis serio_raw pcspkr parport_pc tpm_tis_core i2c_core sg
[  847.644616] CPU: 5 PID: 4905 Comm: xfs_io Not tainted 4.10.0-rc8-btrfs-next-37+ #2
[  847.648601] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.1-0-gb3ef39f-prebuilt.qemu-project.org 04/01/2014
[  847.648601] Call Trace:
[  847.648601]  dump_stack+0x67/0x90
[  847.648601]  __warn+0xc2/0xdd
[  847.648601]  warn_slowpath_null+0x1d/0x1f
[  847.648601]  btrfs_free_reserved_data_space_noquota+0x60/0x9f [btrfs]
[  847.648601]  btrfs_clear_bit_hook+0x140/0x258 [btrfs]
[  847.648601]  clear_state_bit+0x87/0x128 [btrfs]
[  847.648601]  __clear_extent_bit+0x222/0x2b7 [btrfs]
[  847.648601]  clear_extent_bit+0x17/0x19 [btrfs]
[  847.648601]  extent_clear_unlock_delalloc+0x3b/0x6b [btrfs]
[  847.648601]  cow_file_range.isra.39+0x387/0x39a [btrfs]
[  847.648601]  run_delalloc_nocow+0x4d7/0x70e [btrfs]
[  847.648601]  ? arch_local_irq_save+0x9/0xc
[  847.648601]  run_delalloc_range+0xa7/0x2b5 [btrfs]
[  847.648601]  writepage_delalloc.isra.31+0xb9/0x15c [btrfs]
[  847.648601]  __extent_writepage+0x249/0x2e8 [btrfs]
[  847.648601]  extent_write_cache_pages.constprop.33+0x28b/0x36c [btrfs]
[  847.648601]  ? arch_local_irq_save+0x9/0xc
[  847.648601]  ? mark_lock+0x24/0x201
[  847.648601]  extent_writepages+0x4b/0x5c [btrfs]
[  847.648601]  ? btrfs_writepage_start_hook+0xed/0xed [btrfs]
[  847.648601]  btrfs_writepages+0x28/0x2a [btrfs]
[  847.648601]  do_writepages+0x23/0x2c
[  847.648601]  __filemap_fdatawrite_range+0x5a/0x61
[  847.648601]  filemap_fdatawrite_range+0x13/0x15
[  847.648601]  btrfs_fdatawrite_range+0x20/0x46 [btrfs]
[  847.648601]  start_ordered_ops+0x19/0x23 [btrfs]
[  847.648601]  btrfs_sync_file+0x136/0x42c [btrfs]
[  847.648601]  vfs_fsync_range+0x8c/0x9e
[  847.648601]  vfs_fsync+0x1c/0x1e
[  847.648601]  do_fsync+0x31/0x4a
[  847.648601]  SyS_fsync+0x10/0x14
[  847.648601]  entry_SYSCALL_64_fastpath+0x18/0xad
[  847.648601] RIP: 0033:0x7f5b05200800
[  847.648601] RSP: 002b:00007ffe204f71c8 EFLAGS: 00000246 ORIG_RAX: 000000000000004a
[  847.648601] RAX: ffffffffffffffda RBX: ffffffff8109637b RCX: 00007f5b05200800
[  847.648601] RDX: 00000000008bd0a0 RSI: 00000000008bd2e0 RDI: 0000000000000003
[  847.648601] RBP: ffffc90001d67f98 R08: 000000000000ffff R09: 000000000000001f
[  847.648601] R10: 00000000000001f6 R11: 0000000000000246 R12: 0000000000000046
[  847.648601] R13: ffffc90001d67f78 R14: 00007f5b054be740 R15: 00007f5b054be740
[  847.648601]  ? trace_hardirqs_off_caller+0x3f/0xaa
[  847.685787] ---[ end trace 2a4a3e15382508e8 ]---

So fix this by not attempting to decrement the data space info's reserved
bytes counter if we already reserved the extent and an error happened before
creating the ordered extent. We are already correctly freeing the reserved
extent if an error happens, so there's no additional measure needed.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
---
 fs/btrfs/extent_io.h |  4 +++-
 fs/btrfs/inode.c     | 45 ++++++++++++++++++++++++++++++---------------
 2 files changed, 33 insertions(+), 16 deletions(-)
diff mbox

Patch

diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index b12f4db..ba7b410 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -21,8 +21,10 @@ 
 #define EXTENT_NORESERVE	(1U << 15)
 #define EXTENT_QGROUP_RESERVED	(1U << 16)
 #define EXTENT_CLEAR_DATA_RESV	(1U << 17)
+#define EXTENT_SPACE_FREED	(1U << 18)
 #define EXTENT_IOBITS		(EXTENT_LOCKED | EXTENT_WRITEBACK)
-#define EXTENT_CTLBITS		(EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC)
+#define EXTENT_CTLBITS		(EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC |\
+				 EXTENT_SPACE_FREED)
 
 /*
  * flags for bio submission. The high bits indicate the compression
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 07bee7d..0d09ef2 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -912,10 +912,13 @@  static noinline int cow_file_range(struct inode *inode,
 	u64 num_bytes;
 	unsigned long ram_size;
 	u64 disk_num_bytes;
-	u64 cur_alloc_size;
+	u64 cur_alloc_size = 0;
 	u64 blocksize = fs_info->sectorsize;
 	struct btrfs_key ins;
 	struct extent_map *em;
+	unsigned clear_bits;
+	unsigned long page_ops;
+	bool extent_reserved = false;
 	int ret = 0;
 
 	if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
@@ -960,14 +963,15 @@  static noinline int cow_file_range(struct inode *inode,
 			start + num_bytes - 1, 0);
 
 	while (disk_num_bytes > 0) {
-		unsigned long op;
-
+		extent_reserved = false;
 		cur_alloc_size = disk_num_bytes;
 		ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
 					   fs_info->sectorsize, 0, alloc_hint,
 					   &ins, 1, 1);
 		if (ret < 0)
 			goto out_unlock;
+		cur_alloc_size = ins.offset;
+		extent_reserved = true;
 
 		ram_size = ins.offset;
 		em = create_io_em(inode, start, ins.offset, /* len */
@@ -982,7 +986,6 @@  static noinline int cow_file_range(struct inode *inode,
 			goto out_reserve;
 		free_extent_map(em);
 
-		cur_alloc_size = ins.offset;
 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 					       ram_size, cur_alloc_size, 0);
 		if (ret)
@@ -1008,14 +1011,14 @@  static noinline int cow_file_range(struct inode *inode,
 		 * Do set the Private2 bit so we know this page was properly
 		 * setup for writepage
 		 */
-		op = unlock ? PAGE_UNLOCK : 0;
-		op |= PAGE_SET_PRIVATE2;
+		page_ops = unlock ? PAGE_UNLOCK : 0;
+		page_ops |= PAGE_SET_PRIVATE2;
 
 		extent_clear_unlock_delalloc(inode, start,
 					     start + ram_size - 1,
 					     locked_page,
 					     EXTENT_LOCKED | EXTENT_DELALLOC,
-					     op);
+					     page_ops);
 		disk_num_bytes -= cur_alloc_size;
 		num_bytes -= cur_alloc_size;
 		alloc_hint = ins.objectid + ins.offset;
@@ -1030,12 +1033,24 @@  static noinline int cow_file_range(struct inode *inode,
 	btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
 out_unlock:
+	clear_bits = EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | EXTENT_DELALLOC |
+		EXTENT_DEFRAG;
+	page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
+		PAGE_END_WRITEBACK;
+	if (extent_reserved) {
+		extent_clear_unlock_delalloc(inode, start,
+					     start + cur_alloc_size,
+					     locked_page,
+					     clear_bits | EXTENT_SPACE_FREED,
+					     page_ops);
+		start += cur_alloc_size;
+		if (start >= end)
+			goto out;
+	}
 	extent_clear_unlock_delalloc(inode, start, end,
 				     locked_page,
-				     EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
-				     EXTENT_DELALLOC | EXTENT_DEFRAG,
-				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
-				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
+				     clear_bits,
+				     page_ops);
 	goto out;
 }
 
@@ -1735,10 +1750,10 @@  static void btrfs_clear_bit_hook(struct btrfs_inode *inode,
 		if (btrfs_is_testing(fs_info))
 			return;
 
-		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
-		    && do_list && !(state->state & EXTENT_NORESERVE)
-		    && (*bits & (EXTENT_DO_ACCOUNTING |
-		    EXTENT_CLEAR_DATA_RESV)))
+		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
+		    do_list && !(state->state & EXTENT_NORESERVE) &&
+		    (*bits & (EXTENT_DO_ACCOUNTING | EXTENT_CLEAR_DATA_RESV)) &&
+		    !(*bits & EXTENT_SPACE_FREED))
 			btrfs_free_reserved_data_space_noquota(
 					&inode->vfs_inode,
 					state->start, len);