[RFC PATCH 1/4] mm/shmem: add SGP_GET to get unlocked folio

[Chi Zhiling]

From: Chi Zhiling <chizhiling@xxxxxxxxxx>

Add a new sgp_type SGP_GET which is similar to SGP_READ but returns
the folio unlocked with an increased refcount. This eliminates the
lock/unlock overhead for read-only operations.

SGP_GET skips folio lock and mapping check, suitable only for
short-lived access. Caller must not rely on folio->mapping validity
as it can become invalid due to concurrent truncate. Safety relies
on refcount and uptodate flag (truncate doesn't clear content).

Signed-off-by: Chi Zhiling <chizhiling@xxxxxxxxxx>
---
 include/linux/shmem_fs.h |  3 ++-
 mm/shmem.c               | 15 ++++++++++++---
 2 files changed, 14 insertions(+), 4 deletions(-)

diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
index 93a0ba872ebe..24698faea5a4 100644
--- a/include/linux/shmem_fs.h
+++ b/include/linux/shmem_fs.h
@@ -164,7 +164,8 @@ extern unsigned long shmem_partial_swap_usage(struct address_space *mapping,
 
 /* Flag allocation requirements to shmem_get_folio */
 enum sgp_type {
-	SGP_READ,	/* don't exceed i_size, don't allocate page */
+	SGP_GET,	/* don't exceed i_size, don't allocate page, don't lock */
+	SGP_READ,	/* don't exceed i_size, don't allocate page, lock folio */
 	SGP_NOALLOC,	/* similar, but fail on hole or use fallocated page */
 	SGP_CACHE,	/* don't exceed i_size, may allocate page */
 	SGP_WRITE,	/* may exceed i_size, may allocate !Uptodate page */
diff --git a/mm/shmem.c b/mm/shmem.c
index 3b5dc21b323c..ef19968cc51c 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -2504,6 +2504,13 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
 	}
 
 	if (folio) {
+		if (sgp == SGP_GET) {
+			if (!folio_test_uptodate(folio)) {
+				folio_put(folio);
+				folio = NULL;
+			}
+			goto out;
+		}
 		folio_lock(folio);
 
 		/* Has the folio been truncated or swapped out? */
@@ -2524,11 +2531,11 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
 	}
 
 	/*
-	 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
+	 * SGP_READ/SGP_GET: succeed on hole, with NULL folio, letting caller zero.
 	 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
 	 */
 	*foliop = NULL;
-	if (sgp == SGP_READ)
+	if (sgp == SGP_READ || sgp == SGP_GET)
 		return 0;
 	if (sgp == SGP_NOALLOC)
 		return -ENOENT;
@@ -2649,13 +2656,15 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
  * @sgp:	SGP_* flags to control behavior
  *
  * Looks up the page cache entry at @inode & @index.  If a folio is
- * present, it is returned locked with an increased refcount.
+ * present, it is returned locked with an increased refcount, except
+ * for SGP_GET which returns the folio unlocked with an increased refcount.
  *
  * If the caller modifies data in the folio, it must call folio_mark_dirty()
  * before unlocking the folio to ensure that the folio is not reclaimed.
  * There is no need to reserve space before calling folio_mark_dirty().
  *
  * When no folio is found, the behavior depends on @sgp:
+ *  - for SGP_GET, *@foliop is %NULL and 0 is returned
  *  - for SGP_READ, *@foliop is %NULL and 0 is returned
  *  - for SGP_NOALLOC, *@foliop is %NULL and -ENOENT is returned
  *  - for all other flags a new folio is allocated, inserted into the
-- 
2.43.0