f2fs: fix race of pending_pages in decompression

I found out f2fs_free_dic() is invoked in a wrong timing, but
f2fs_verify_bio() still needed the dic info and it triggered the
below kernel panic. It has been caused by the race condition of
pending_pages value between decompression and verity logic, when
the same compression cluster had been split in different bios.
By split bios, f2fs_verify_bio() ended up with decreasing
pending_pages value before it is reset to nr_cpages by
f2fs_decompress_pages() and caused the kernel panic.

[ 4416.564763] Unable to handle kernel NULL pointer dereference
               at virtual address 0000000000000000
...
[ 4416.896016] Workqueue: fsverity_read_queue f2fs_verity_work
[ 4416.908515] pc : fsverity_verify_page+0x20/0x78
[ 4416.913721] lr : f2fs_verify_bio+0x11c/0x29c
[ 4416.913722] sp : ffffffc019533cd0
[ 4416.913723] x29: ffffffc019533cd0 x28: 0000000000000402
[ 4416.913724] x27: 0000000000000001 x26: 0000000000000100
[ 4416.913726] x25: 0000000000000001 x24: 0000000000000004
[ 4416.913727] x23: 0000000000001000 x22: 0000000000000000
[ 4416.913728] x21: 0000000000000000 x20: ffffffff2076f9c0
[ 4416.913729] x19: ffffffff2076f9c0 x18: ffffff8a32380c30
[ 4416.913731] x17: ffffffc01f966d97 x16: 0000000000000298
[ 4416.913732] x15: 0000000000000000 x14: 0000000000000000
[ 4416.913733] x13: f074faec89ffffff x12: 0000000000000000
[ 4416.913734] x11: 0000000000001000 x10: 0000000000001000
[ 4416.929176] x9 : ffffffff20d1f5c7 x8 : 0000000000000000
[ 4416.929178] x7 : 626d7464ff286b6b x6 : ffffffc019533ade
[ 4416.929179] x5 : 000000008049000e x4 : ffffffff2793e9e0
[ 4416.929180] x3 : 000000008049000e x2 : ffffff89ecfa74d0
[ 4416.929181] x1 : 0000000000000c40 x0 : ffffffff2076f9c0
[ 4416.929184] Call trace:
[ 4416.929187]  fsverity_verify_page+0x20/0x78
[ 4416.929189]  f2fs_verify_bio+0x11c/0x29c
[ 4416.929192]  f2fs_verity_work+0x58/0x84
[ 4417.050667]  process_one_work+0x270/0x47c
[ 4417.055354]  worker_thread+0x27c/0x4d8
[ 4417.059784]  kthread+0x13c/0x320
[ 4417.063693]  ret_from_fork+0x10/0x18

Chao pointed this can happen by the below race condition.

Thread A        f2fs_post_read_wq          fsverity_wq
- f2fs_read_multi_pages()
  - f2fs_alloc_dic
   - dic->pending_pages = 2
   - submit_bio()
   - submit_bio()
               - f2fs_post_read_work() handle first bio
                - f2fs_decompress_work()
                 - __read_end_io()
                  - f2fs_decompress_pages()
                   - dic->pending_pages--
                - enqueue f2fs_verity_work()
                                           - f2fs_verity_work() handle first bio
                                            - f2fs_verify_bio()
                                             - dic->pending_pages--
               - f2fs_post_read_work() handle second bio
                - f2fs_decompress_work()
                - enqueue f2fs_verity_work()
                                            - f2fs_verify_pages()
                                            - f2fs_free_dic()

                                          - f2fs_verity_work() handle second bio
                                           - f2fs_verfy_bio()
                                                 - use-after-free on dic

Signed-off-by: Daeho Jeong <daehojeong@google.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

1 files changed, 47 insertions, 11 deletions

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index cb28089e1eff..aa34d620bec9 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -202,7 +202,7 @@ static void f2fs_verify_bio(struct bio *bio)
 		dic = (struct decompress_io_ctx *)page_private(page);
 
 		if (dic) {
-			if (atomic_dec_return(&dic->pending_pages))
+			if (atomic_dec_return(&dic->verity_pages))
 				continue;
 			f2fs_verify_pages(dic->rpages,
 						dic->cluster_size);
@@ -1030,7 +1030,8 @@ static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx)
 
 static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 				      unsigned nr_pages, unsigned op_flag,
-				      pgoff_t first_idx, bool for_write)
+				      pgoff_t first_idx, bool for_write,
+				      bool for_verity)
 {
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 	struct bio *bio;
@@ -1052,7 +1053,7 @@ static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
 		post_read_steps |= 1 << STEP_DECRYPT;
 	if (f2fs_compressed_file(inode))
 		post_read_steps |= 1 << STEP_DECOMPRESS_NOWQ;
-	if (f2fs_need_verity(inode, first_idx))
+	if (for_verity && f2fs_need_verity(inode, first_idx))
 		post_read_steps |= 1 << STEP_VERITY;
 
 	if (post_read_steps) {
@@ -1082,7 +1083,7 @@ static int f2fs_submit_page_read(struct inode *inode, struct page *page,
 	struct bio *bio;
 
 	bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
-					page->index, for_write);
+					page->index, for_write, true);
 	if (IS_ERR(bio))
 		return PTR_ERR(bio);
 
@@ -2120,7 +2121,7 @@ submit_and_realloc:
 	if (bio == NULL) {
 		bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
 				is_readahead ? REQ_RAHEAD : 0, page->index,
-				false);
+				false, true);
 		if (IS_ERR(bio)) {
 			ret = PTR_ERR(bio);
 			bio = NULL;
@@ -2166,6 +2167,8 @@ int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
 	sector_t last_block_in_file;
 	const unsigned blocksize = blks_to_bytes(inode, 1);
 	struct decompress_io_ctx *dic = NULL;
+	struct bio_post_read_ctx *ctx;
+	bool for_verity = false;
 	int i;
 	int ret = 0;
 
@@ -2231,10 +2234,29 @@ int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
 		goto out_put_dnode;
 	}
 
+	/*
+	 * It's possible to enable fsverity on the fly when handling a cluster,
+	 * which requires complicated error handling. Instead of adding more
+	 * complexity, let's give a rule where end_io post-processes fsverity
+	 * per cluster. In order to do that, we need to submit bio, if previous
+	 * bio sets a different post-process policy.
+	 */
+	if (fsverity_active(cc->inode)) {
+		atomic_set(&dic->verity_pages, cc->nr_cpages);
+		for_verity = true;
+
+		if (bio) {
+			ctx = bio->bi_private;
+			if (!(ctx->enabled_steps & (1 << STEP_VERITY))) {
+				__submit_bio(sbi, bio, DATA);
+				bio = NULL;
+			}
+		}
+	}
+
 	for (i = 0; i < dic->nr_cpages; i++) {
 		struct page *page = dic->cpages[i];
 		block_t blkaddr;
-		struct bio_post_read_ctx *ctx;
 
 		blkaddr = data_blkaddr(dn.inode, dn.node_page,
 						dn.ofs_in_node + i + 1);
@@ -2250,17 +2272,31 @@ submit_and_realloc:
 		if (!bio) {
 			bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
 					is_readahead ? REQ_RAHEAD : 0,
-					page->index, for_write);
+					page->index, for_write, for_verity);
 			if (IS_ERR(bio)) {
+				unsigned int remained = dic->nr_cpages - i;
+				bool release = false;
+
 				ret = PTR_ERR(bio);
 				dic->failed = true;
-				if (!atomic_sub_return(dic->nr_cpages - i,
-							&dic->pending_pages)) {
+
+				if (for_verity) {
+					if (!atomic_sub_return(remained,
+						&dic->verity_pages))
+						release = true;
+				} else {
+					if (!atomic_sub_return(remained,
+						&dic->pending_pages))
+						release = true;
+				}
+
+				if (release) {
 					f2fs_decompress_end_io(dic->rpages,
-							cc->cluster_size, true,
-							false);
+						cc->cluster_size, true,
+						false);
 					f2fs_free_dic(dic);
 				}
+
 				f2fs_put_dnode(&dn);
 				*bio_ret = NULL;
 				return ret;