f2fs: clean up post-read processing

Rework the post-read processing logic to be much easier to understand.

At least one bug is fixed by this: if an I/O error occurred when reading
from disk, decryption and verity would be performed on the uninitialized
data, causing misleading messages in the kernel log.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

fs/f2fs/compress.c

@@ -756,38 +756,27 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
 	return ret;
 }
 
-void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
+static void f2fs_decompress_cluster(struct decompress_io_ctx *dic)
 {
-	struct decompress_io_ctx *dic =
-			(struct decompress_io_ctx *)page_private(page);
 	struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
-	struct f2fs_inode_info *fi= F2FS_I(dic->inode);
+	struct f2fs_inode_info *fi = F2FS_I(dic->inode);
 	const struct f2fs_compress_ops *cops =
 			f2fs_cops[fi->i_compress_algorithm];
 	int ret;
 	int i;
 
-	dec_page_count(sbi, F2FS_RD_DATA);
-
-	if (bio->bi_status || PageError(page))
-		dic->failed = true;
-
-	if (atomic_dec_return(&dic->pending_pages))
-		return;
-
 	trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
 				dic->cluster_size, fi->i_compress_algorithm);
 
-	/* submit partial compressed pages */
 	if (dic->failed) {
 		ret = -EIO;
-		goto out_free_dic;
+		goto out_end_io;
 	}
 
 	dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
 	if (!dic->tpages) {
 		ret = -ENOMEM;
-		goto out_free_dic;
+		goto out_end_io;
 	}
 
 	for (i = 0; i < dic->cluster_size; i++) {
@@ -799,20 +788,20 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
 		dic->tpages[i] = f2fs_compress_alloc_page();
 		if (!dic->tpages[i]) {
 			ret = -ENOMEM;
-			goto out_free_dic;
+			goto out_end_io;
 		}
 	}
 
 	if (cops->init_decompress_ctx) {
 		ret = cops->init_decompress_ctx(dic);
 		if (ret)
-			goto out_free_dic;
+			goto out_end_io;
 	}
 
 	dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
 	if (!dic->rbuf) {
 		ret = -ENOMEM;
-		goto destroy_decompress_ctx;
+		goto out_destroy_decompress_ctx;
 	}
 
 	dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
@@ -851,18 +840,34 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
 	vm_unmap_ram(dic->cbuf, dic->nr_cpages);
 out_vunmap_rbuf:
 	vm_unmap_ram(dic->rbuf, dic->cluster_size);
-destroy_decompress_ctx:
+out_destroy_decompress_ctx:
 	if (cops->destroy_decompress_ctx)
 		cops->destroy_decompress_ctx(dic);
-out_free_dic:
-	if (!verity)
-		f2fs_decompress_end_io(dic->rpages, dic->cluster_size,
-								ret, false);
-
+out_end_io:
 	trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
 							dic->clen, ret);
-	if (!verity)
-		f2fs_free_dic(dic);
+	f2fs_decompress_end_io(dic, ret);
+}
+
+/*
+ * This is called when a page of a compressed cluster has been read from disk
+ * (or failed to be read from disk).  It checks whether this page was the last
+ * page being waited on in the cluster, and if so, it decompresses the cluster
+ * (or in the case of a failure, cleans up without actually decompressing).
+ */
+void f2fs_end_read_compressed_page(struct page *page, bool failed)
+{
+	struct decompress_io_ctx *dic =
+			(struct decompress_io_ctx *)page_private(page);
+	struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
+
+	dec_page_count(sbi, F2FS_RD_DATA);
+
+	if (failed)
+		WRITE_ONCE(dic->failed, true);
+
+	if (atomic_dec_and_test(&dic->remaining_pages))
+		f2fs_decompress_cluster(dic);
 }
 
 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
@@ -1529,6 +1534,8 @@ int f2fs_write_multi_pages(struct compress_ctx *cc,
 	return err;
 }
 
+static void f2fs_free_dic(struct decompress_io_ctx *dic);
+
 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
 {
 	struct decompress_io_ctx *dic;
@@ -1547,12 +1554,14 @@ struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
 
 	dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
 	dic->inode = cc->inode;
-	atomic_set(&dic->pending_pages, cc->nr_cpages);
+	atomic_set(&dic->remaining_pages, cc->nr_cpages);
 	dic->cluster_idx = cc->cluster_idx;
 	dic->cluster_size = cc->cluster_size;
 	dic->log_cluster_size = cc->log_cluster_size;
 	dic->nr_cpages = cc->nr_cpages;
+	refcount_set(&dic->refcnt, 1);
 	dic->failed = false;
+	dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
 
 	for (i = 0; i < dic->cluster_size; i++)
 		dic->rpages[i] = cc->rpages[i];
@@ -1581,7 +1590,7 @@ struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
 	return ERR_PTR(-ENOMEM);
 }
 
-void f2fs_free_dic(struct decompress_io_ctx *dic)
+static void f2fs_free_dic(struct decompress_io_ctx *dic)
 {
 	int i;
 
@@ -1609,30 +1618,88 @@ void f2fs_free_dic(struct decompress_io_ctx *dic)
 	kmem_cache_free(dic_entry_slab, dic);
 }
 
-void f2fs_decompress_end_io(struct page **rpages,
-			unsigned int cluster_size, bool err, bool verity)
+static void f2fs_put_dic(struct decompress_io_ctx *dic)
+{
+	if (refcount_dec_and_test(&dic->refcnt))
+		f2fs_free_dic(dic);
+}
+
+/*
+ * Update and unlock the cluster's pagecache pages, and release the reference to
+ * the decompress_io_ctx that was being held for I/O completion.
+ */
+static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
 {
 	int i;
 
-	for (i = 0; i < cluster_size; i++) {
-		struct page *rpage = rpages[i];
+	for (i = 0; i < dic->cluster_size; i++) {
+		struct page *rpage = dic->rpages[i];
 
 		if (!rpage)
 			continue;
 
-		if (err || PageError(rpage))
-			goto clear_uptodate;
-
-		if (!verity || fsverity_verify_page(rpage)) {
-			SetPageUptodate(rpage);
-			goto unlock;
-		}
-clear_uptodate:
+		/* PG_error was set if verity failed. */
+		if (failed || PageError(rpage)) {
 			ClearPageUptodate(rpage);
+			/* will re-read again later */
 			ClearPageError(rpage);
-unlock:
+		} else {
+			SetPageUptodate(rpage);
+		}
 		unlock_page(rpage);
 	}
+
+	f2fs_put_dic(dic);
+}
+
+static void f2fs_verify_cluster(struct work_struct *work)
+{
+	struct decompress_io_ctx *dic =
+		container_of(work, struct decompress_io_ctx, verity_work);
+	int i;
+
+	/* Verify the cluster's decompressed pages with fs-verity. */
+	for (i = 0; i < dic->cluster_size; i++) {
+		struct page *rpage = dic->rpages[i];
+
+		if (rpage && !fsverity_verify_page(rpage))
+			SetPageError(rpage);
+	}
+
+	__f2fs_decompress_end_io(dic, false);
+}
+
+/*
+ * This is called when a compressed cluster has been decompressed
+ * (or failed to be read and/or decompressed).
+ */
+void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
+{
+	if (!failed && dic->need_verity) {
+		/*
+		 * Note that to avoid deadlocks, the verity work can't be done
+		 * on the decompression workqueue.  This is because verifying
+		 * the data pages can involve reading metadata pages from the
+		 * file, and these metadata pages may be compressed.
+		 */
+		INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
+		fsverity_enqueue_verify_work(&dic->verity_work);
+	} else {
+		__f2fs_decompress_end_io(dic, failed);
+	}
+}
+
+/*
+ * Put a reference to a compressed page's decompress_io_ctx.
+ *
+ * This is called when the page is no longer needed and can be freed.
+ */
+void f2fs_put_page_dic(struct page *page)
+{
+	struct decompress_io_ctx *dic =
+			(struct decompress_io_ctx *)page_private(page);
+
+	f2fs_put_dic(dic);
 }
 
 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)

fs/f2fs/f2fs.h

@@ -1341,7 +1341,7 @@ struct compress_io_ctx {
 	atomic_t pending_pages;		/* in-flight compressed page count */
 };
 
-/* decompress io context for read IO path */
+/* Context for decompressing one cluster on the read IO path */
 struct decompress_io_ctx {
 	u32 magic;			/* magic number to indicate page is compressed */
 	struct inode *inode;		/* inode the context belong to */
@@ -1357,11 +1357,37 @@ struct decompress_io_ctx {
 	struct compress_data *cbuf;	/* virtual mapped address on cpages */
 	size_t rlen;			/* valid data length in rbuf */
 	size_t clen;			/* valid data length in cbuf */
-	atomic_t pending_pages;		/* in-flight compressed page count */
-	atomic_t verity_pages;		/* in-flight page count for verity */
-	bool failed;			/* indicate IO error during decompression */
+
+	/*
+	 * The number of compressed pages remaining to be read in this cluster.
+	 * This is initially nr_cpages.  It is decremented by 1 each time a page
+	 * has been read (or failed to be read).  When it reaches 0, the cluster
+	 * is decompressed (or an error is reported).
+	 *
+	 * If an error occurs before all the pages have been submitted for I/O,
+	 * then this will never reach 0.  In this case the I/O submitter is
+	 * responsible for calling f2fs_decompress_end_io() instead.
+	 */
+	atomic_t remaining_pages;
+
+	/*
+	 * Number of references to this decompress_io_ctx.
+	 *
+	 * One reference is held for I/O completion.  This reference is dropped
+	 * after the pagecache pages are updated and unlocked -- either after
+	 * decompression (and verity if enabled), or after an error.
+	 *
+	 * In addition, each compressed page holds a reference while it is in a
+	 * bio.  These references are necessary prevent compressed pages from
+	 * being freed while they are still in a bio.
+	 */
+	refcount_t refcnt;
+
+	bool failed;			/* IO error occurred before decompression? */
+	bool need_verity;		/* need fs-verity verification after decompression? */
 	void *private;			/* payload buffer for specified decompression algorithm */
 	void *private2;			/* extra payload buffer */
+	struct work_struct verity_work;	/* work to verify the decompressed pages */
 };
 
 #define NULL_CLUSTER			((unsigned int)(~0))
@@ -3883,7 +3909,7 @@ void f2fs_compress_write_end_io(struct bio *bio, struct page *page);
 bool f2fs_is_compress_backend_ready(struct inode *inode);
 int f2fs_init_compress_mempool(void);
 void f2fs_destroy_compress_mempool(void);
-void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity);
+void f2fs_end_read_compressed_page(struct page *page, bool failed);
 bool f2fs_cluster_is_empty(struct compress_ctx *cc);
 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index);
 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page);
@@ -3896,9 +3922,8 @@ int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
 				unsigned nr_pages, sector_t *last_block_in_bio,
 				bool is_readahead, bool for_write);
 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc);
-void f2fs_free_dic(struct decompress_io_ctx *dic);
-void f2fs_decompress_end_io(struct page **rpages,
-			unsigned int cluster_size, bool err, bool verity);
+void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed);
+void f2fs_put_page_dic(struct page *page);
 int f2fs_init_compress_ctx(struct compress_ctx *cc);
 void f2fs_destroy_compress_ctx(struct compress_ctx *cc);
 void f2fs_init_compress_info(struct f2fs_sb_info *sbi);
@@ -3922,6 +3947,14 @@ static inline struct page *f2fs_compress_control_page(struct page *page)
 }
 static inline int f2fs_init_compress_mempool(void) { return 0; }
 static inline void f2fs_destroy_compress_mempool(void) { }
+static inline void f2fs_end_read_compressed_page(struct page *page, bool failed)
+{
+	WARN_ON_ONCE(1);
+}
+static inline void f2fs_put_page_dic(struct page *page)
+{
+	WARN_ON_ONCE(1);
+}
 static inline int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi) { return 0; }
 static inline void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi) { }
 static inline int __init f2fs_init_compress_cache(void) { return 0; }
@@ -4126,6 +4159,12 @@ static inline bool f2fs_force_buffered_io(struct inode *inode,
 	return false;
 }
 
+static inline bool f2fs_need_verity(const struct inode *inode, pgoff_t idx)
+{
+	return fsverity_active(inode) &&
+	       idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
+}
+
 #ifdef CONFIG_F2FS_FAULT_INJECTION
 extern void f2fs_build_fault_attr(struct f2fs_sb_info *sbi, unsigned int rate,
 							unsigned int type);