[PATCH 2/9] huge tmpfs: fix split_huge_page() after FALLOC_FL_KEEP_SIZE

[Hugh Dickins <hughd@google.com>]

A successful shmem_fallocate() guarantees that the extent has been
reserved, even beyond i_size when the FALLOC_FL_KEEP_SIZE flag was used.
But that guarantee is broken by shmem_unused_huge_shrink()'s attempts to
split huge pages and free their excess beyond i_size; and by other uses
of split_huge_page() near i_size.

It's sad to add a shmem inode field just for this, but I did not find a
better way to keep the guarantee.  A flag to say KEEP_SIZE has been used
would be cheaper, but I'm averse to unclearable flags.  The fallocend
field is not perfect either (many disjoint ranges might be fallocated),
but good enough; and gains another use later on.

Fixes: 779750d20b93 ("shmem: split huge pages beyond i_size under memory pressure")
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
---
 include/linux/shmem_fs.h | 13 +++++++++++++
 mm/huge_memory.c         |  6 ++++--
 mm/shmem.c               | 15 ++++++++++++++-
 3 files changed, 31 insertions(+), 3 deletions(-)

diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
index 8e775ce517bb..9b7f7ac52351 100644
--- a/include/linux/shmem_fs.h
+++ b/include/linux/shmem_fs.h
@@ -18,6 +18,7 @@ struct shmem_inode_info {
 	unsigned long		flags;
 	unsigned long		alloced;	/* data pages alloced to file */
 	unsigned long		swapped;	/* subtotal assigned to swap */
+	pgoff_t			fallocend;	/* highest fallocate endindex */
 	struct list_head        shrinklist;     /* shrinkable hpage inodes */
 	struct list_head	swaplist;	/* chain of maybes on swap */
 	struct shared_policy	policy;		/* NUMA memory alloc policy */
@@ -119,6 +120,18 @@ static inline bool shmem_file(struct file *file)
 	return shmem_mapping(file->f_mapping);
 }
 
+/*
+ * If fallocate(FALLOC_FL_KEEP_SIZE) has been used, there may be pages
+ * beyond i_size's notion of EOF, which fallocate has committed to reserving:
+ * which split_huge_page() must therefore not delete.  This use of a single
+ * "fallocend" per inode errs on the side of not deleting a reservation when
+ * in doubt: there are plenty of cases when it preserves unreserved pages.
+ */
+static inline pgoff_t shmem_fallocend(struct inode *inode, pgoff_t eof)
+{
+	return max(eof, SHMEM_I(inode)->fallocend);
+}
+
 extern bool shmem_charge(struct inode *inode, long pages);
 extern void shmem_uncharge(struct inode *inode, long pages);
 
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index afff3ac87067..890fb73ac89b 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2454,11 +2454,11 @@ static void __split_huge_page(struct page *page, struct list_head *list,
 
 	for (i = nr - 1; i >= 1; i--) {
 		__split_huge_page_tail(head, i, lruvec, list);
-		/* Some pages can be beyond i_size: drop them from page cache */
+		/* Some pages can be beyond EOF: drop them from page cache */
 		if (head[i].index >= end) {
 			ClearPageDirty(head + i);
 			__delete_from_page_cache(head + i, NULL);
-			if (IS_ENABLED(CONFIG_SHMEM) && PageSwapBacked(head))
+			if (shmem_mapping(head->mapping))
 				shmem_uncharge(head->mapping->host, 1);
 			put_page(head + i);
 		} else if (!PageAnon(page)) {
@@ -2686,6 +2686,8 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
 		 * head page lock is good enough to serialize the trimming.
 		 */
 		end = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
+		if (shmem_mapping(mapping))
+			end = shmem_fallocend(mapping->host, end);
 	}
 
 	/*
diff --git a/mm/shmem.c b/mm/shmem.c
index 0cd5c9156457..24c9da6b41c2 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -905,6 +905,9 @@ static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 	if (lend == -1)
 		end = -1;	/* unsigned, so actually very big */
 
+	if (info->fallocend > start && info->fallocend <= end && !unfalloc)
+		info->fallocend = start;
+
 	pagevec_init(&pvec);
 	index = start;
 	while (index < end && find_lock_entries(mapping, index, end - 1,
@@ -2667,7 +2670,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 	struct shmem_inode_info *info = SHMEM_I(inode);
 	struct shmem_falloc shmem_falloc;
-	pgoff_t start, index, end;
+	pgoff_t start, index, end, undo_fallocend;
 	int error;
 
 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
@@ -2736,6 +2739,15 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 	inode->i_private = &shmem_falloc;
 	spin_unlock(&inode->i_lock);
 
+	/*
+	 * info->fallocend is only relevant when huge pages might be
+	 * involved: to prevent split_huge_page() freeing fallocated
+	 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
+	 */
+	undo_fallocend = info->fallocend;
+	if (info->fallocend < end)
+		info->fallocend = end;
+
 	for (index = start; index < end; ) {
 		struct page *page;
 
@@ -2750,6 +2762,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 		else
 			error = shmem_getpage(inode, index, &page, SGP_FALLOC);
 		if (error) {
+			info->fallocend = undo_fallocend;
 			/* Remove the !PageUptodate pages we added */
 			if (index > start) {
 				shmem_undo_range(inode,
-- 
2.26.2