[PATCH 0/12 v4] fs: Hole punch vs page cache filling races

[PATCH 0/12 v4] fs: Hole punch vs page cache filling races

[Jan Kara]

Hello,

here is another version of my patches to address races between hole punching
and page cache filling functions for ext4 and other filesystems. I think
we are coming close to a complete solution so I've removed the RFC tag from
the subject. I went through all filesystems supporting hole punching and
converted them from their private locks to a generic one (usually fixing the
race ext4 had as a side effect). I also found out ceph & cifs didn't have
any protection from the hole punch vs page fault race either so I've added
appropriate protections there. Open are still GFS2 and OCFS2 filesystems.
GFS2 actually avoids the race but is prone to deadlocks (acquires the same lock
both above and below mmap_sem), OCFS2 locking seems kind of hosed and some
read, write, and hole punch paths are not properly serialized possibly leading
to fs corruption. Both issues are non-trivial so respective fs maintainers
have to deal with those (I've informed them and problems were generally
confirmed). Anyway, for all the other filesystem this kind of race should
be closed.

As a next step, I'd like to actually make sure all calls to
truncate_inode_pages() happen under mapping->invalidate_lock, add the assert
and then we can also get rid of i_size checks in some places (truncate can
use the same serialization scheme as hole punch). But that step is mostly
a cleanup so I'd like to get these functional fixes in first.

Changes since v3:
* Renamed and moved lock to struct address_space
* Added conversions of tmpfs, ceph, cifs, fuse, f2fs
* Fixed error handling path in filemap_read()
* Removed .page_mkwrite() cleanup from the series for now

Changes since v2:
* Added documentation and comments regarding lock ordering and how the lock is
  supposed to be used
* Added conversions of ext2, xfs, zonefs
* Added patch removing i_mapping_sem protection from .page_mkwrite handlers

Changes since v1:
* Moved to using inode->i_mapping_sem instead of aops handler to acquire
  appropriate lock

---
Motivation:

Amir has reported [1] a that ext4 has a potential issues when reads can race
with hole punching possibly exposing stale data from freed blocks or even
corrupting filesystem when stale mapping data gets used for writeout. The
problem is that during hole punching, new page cache pages can get instantiated
and block mapping from the looked up in a punched range after
truncate_inode_pages() has run but before the filesystem removes blocks from
the file. In principle any filesystem implementing hole punching thus needs to
implement a mechanism to block instantiating page cache pages during hole
punching to avoid this race. This is further complicated by the fact that there
are multiple places that can instantiate pages in page cache.  We can have
regular read(2) or page fault doing this but fadvise(2) or madvise(2) can also
result in reading in page cache pages through force_page_cache_readahead().

There are couple of ways how to fix this. First way (currently implemented by
XFS) is to protect read(2) and *advise(2) calls with i_rwsem so that they are
serialized with hole punching. This is easy to do but as a result all reads
would then be serialized with writes and thus mixed read-write workloads suffer
heavily on ext4. Thus this series introduces inode->i_mapping_sem and uses it
when creating new pages in the page cache and looking up their corresponding
block mapping. We also replace EXT4_I(inode)->i_mmap_sem with this new rwsem
which provides necessary serialization with hole punching for ext4.

								Honza

[1] https://lore.kernel.org/linux-fsdevel/CAOQ4uxjQNmxqmtA_VbYW0Su9rKRk2zobJmahcyeaEVOFKVQ5dw@mail.gmail.com/

Previous versions:
Link: https://lore.kernel.org/linux-fsdevel/20210208163918.7871-1-jack@suse.cz/
Link: http://lore.kernel.org/r/20210413105205.3093-1-jack@suse.cz

CC: ceph-devel@vger.kernel.org
CC: Chao Yu <yuchao0@huawei.com>
CC: Damien Le Moal <damien.lemoal@wdc.com>
CC: "Darrick J. Wong" <darrick.wong@oracle.com>
CC: Hugh Dickins <hughd@google.com>
CC: Jaegeuk Kim <jaegeuk@kernel.org>
CC: Jeff Layton <jlayton@kernel.org>
CC: Johannes Thumshirn <jth@kernel.org>
CC: linux-cifs@vger.kernel.org
CC: <linux-ext4@vger.kernel.org>
CC: linux-f2fs-devel@lists.sourceforge.net
CC: <linux-fsdevel@vger.kernel.org>
CC: <linux-mm@kvack.org>
CC: <linux-xfs@vger.kernel.org>
CC: Miklos Szeredi <miklos@szeredi.hu>
CC: Steve French <sfrench@samba.org>
CC: Ted Tso <tytso@mit.edu>

[PATCH 02/12] mm: Protect operations adding pages to page cache with invalidate_lock

[Jan Kara]

Currently, serializing operations such as page fault, read, or readahead
against hole punching is rather difficult. The basic race scheme is
like:

fallocate(FALLOC_FL_PUNCH_HOLE)			read / fault / ..
  truncate_inode_pages_range()
						  <create pages in page
						   cache here>
  <update fs block mapping and free blocks>

Now the problem is in this way read / page fault / readahead can
instantiate pages in page cache with potentially stale data (if blocks
get quickly reused). Avoiding this race is not simple - page locks do
not work because we want to make sure there are *no* pages in given
range. inode->i_rwsem does not work because page fault happens under
mmap_sem which ranks below inode->i_rwsem. Also using it for reads makes
the performance for mixed read-write workloads suffer.

So create a new rw_semaphore in the address_space - invalidate_lock -
that protects adding of pages to page cache for page faults / reads /
readahead.

Signed-off-by: Jan Kara <jack@suse.cz>
CC: ceph-devel@vger.kernel.org
CC: Chao Yu <yuchao0@huawei.com>
CC: Damien Le Moal <damien.lemoal@wdc.com>
CC: "Darrick J. Wong" <darrick.wong@oracle.com>
CC: Hugh Dickins <hughd@google.com>
CC: Jaegeuk Kim <jaegeuk@kernel.org>
CC: Jeff Layton <jlayton@kernel.org>
CC: Johannes Thumshirn <jth@kernel.org>
CC: linux-cifs@vger.kernel.org
CC: <linux-ext4@vger.kernel.org>
CC: linux-f2fs-devel@lists.sourceforge.net
CC: <linux-fsdevel@vger.kernel.org>
CC: <linux-mm@kvack.org>
CC: <linux-xfs@vger.kernel.org>
CC: Miklos Szeredi <miklos@szeredi.hu>
CC: Steve French <sfrench@samba.org>
CC: Ted Tso <tytso@mit.edu>
---
 Documentation/filesystems/locking.rst | 39 ++++++++++++-----
 fs/inode.c                            |  3 ++
 include/linux/fs.h                    |  4 ++
 mm/filemap.c                          | 61 +++++++++++++++++++++------
 mm/readahead.c                        |  2 +
 mm/rmap.c                             | 37 ++++++++--------
 mm/truncate.c                         |  2 +-
 7 files changed, 106 insertions(+), 42 deletions(-)

diff --git a/Documentation/filesystems/locking.rst b/Documentation/filesystems/locking.rst
index b7dcc86c92a4..7cbf72862832 100644
--- a/Documentation/filesystems/locking.rst
+++ b/Documentation/filesystems/locking.rst
@@ -266,19 +266,19 @@ prototypes::
 locking rules:
 	All except set_page_dirty and freepage may block
 
-======================	======================== =========
-ops			PageLocked(page)	 i_rwsem
-======================	======================== =========
+======================	======================== =========	===============
+ops			PageLocked(page)	 i_rwsem	invalidate_lock
+======================	======================== =========	===============
 writepage:		yes, unlocks (see below)
-readpage:		yes, unlocks
+readpage:		yes, unlocks				shared
 writepages:
 set_page_dirty		no
-readahead:		yes, unlocks
-readpages:		no
+readahead:		yes, unlocks				shared
+readpages:		no					shared
 write_begin:		locks the page		 exclusive
 write_end:		yes, unlocks		 exclusive
 bmap:
-invalidatepage:		yes
+invalidatepage:		yes					exclusive
 releasepage:		yes
 freepage:		yes
 direct_IO:
@@ -373,7 +373,10 @@ keep it that way and don't breed new callers.
 ->invalidatepage() is called when the filesystem must attempt to drop
 some or all of the buffers from the page when it is being truncated. It
 returns zero on success. If ->invalidatepage is zero, the kernel uses
-block_invalidatepage() instead.
+block_invalidatepage() instead. The filesystem should exclusively acquire
+invalidate_lock before invalidating page cache in truncate / hole punch path (and
+thus calling into ->invalidatepage) to block races between page cache
+invalidation and page cache filling functions (fault, read, ...).
 
 ->releasepage() is called when the kernel is about to try to drop the
 buffers from the page in preparation for freeing it.  It returns zero to
@@ -567,6 +570,20 @@ in sys_read() and friends.
 the lease within the individual filesystem to record the result of the
 operation
 
+->fallocate implementation must be really careful to maintain page cache
+consistency when punching holes or performing other operations that invalidate
+page cache contents. Usually the filesystem needs to call
+truncate_inode_pages_range() to invalidate relevant range of the page cache.
+However the filesystem usually also needs to update its internal (and on disk)
+view of file offset -> disk block mapping. Until this update is finished, the
+filesystem needs to block page faults and reads from reloading now-stale page
+cache contents from the disk. VFS provides mapping->invalidate_lock for this
+and acquires it in shared mode in paths loading pages from disk
+(filemap_fault(), filemap_read(), readahead paths). The filesystem is
+responsible for taking this lock in its fallocate implementation and generally
+whenever the page cache contents needs to be invalidated because a block is
+moving from under a page.
+
 dquot_operations
 ================
 
@@ -628,9 +645,9 @@ access:		yes
 to be faulted in. The filesystem must find and return the page associated
 with the passed in "pgoff" in the vm_fault structure. If it is possible that
 the page may be truncated and/or invalidated, then the filesystem must lock
-the page, then ensure it is not already truncated (the page lock will block
-subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
-locked. The VM will unlock the page.
+invalidate_lock, then ensure the page is not already truncated (invalidate_lock
+will block subsequent truncate), and then return with VM_FAULT_LOCKED, and the
+page locked. The VM will unlock the page.
 
 ->map_pages() is called when VM asks to map easy accessible pages.
 Filesystem should find and map pages associated with offsets from "start_pgoff"
diff --git a/fs/inode.c b/fs/inode.c
index a047ab306f9a..43596dd8b61e 100644
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -191,6 +191,9 @@ int inode_init_always(struct super_block *sb, struct inode *inode)
 	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 	mapping->private_data = NULL;
 	mapping->writeback_index = 0;
+	init_rwsem(&mapping->invalidate_lock);
+	lockdep_set_class(&mapping->invalidate_lock,
+			  &sb->s_type->invalidate_lock_key);
 	inode->i_private = NULL;
 	inode->i_mapping = mapping;
 	INIT_HLIST_HEAD(&inode->i_dentry);	/* buggered by rcu freeing */
diff --git a/include/linux/fs.h b/include/linux/fs.h
index ec8f3ddf4a6a..3fca7bf2d0fb 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -435,6 +435,8 @@ int pagecache_write_end(struct file *, struct address_space *mapping,
  * struct address_space - Contents of a cacheable, mappable object.
  * @host: Owner, either the inode or the block_device.
  * @i_pages: Cached pages.
+ * @invalidate_lock: Guards coherency between page cache contents and
+ *   file offset->disk block mappings in the filesystem during invalidates
  * @gfp_mask: Memory allocation flags to use for allocating pages.
  * @i_mmap_writable: Number of VM_SHARED mappings.
  * @nr_thps: Number of THPs in the pagecache (non-shmem only).
@@ -453,6 +455,7 @@ int pagecache_write_end(struct file *, struct address_space *mapping,
 struct address_space {
 	struct inode		*host;
 	struct xarray		i_pages;
+	struct rw_semaphore	invalidate_lock;
 	gfp_t			gfp_mask;
 	atomic_t		i_mmap_writable;
 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
@@ -2351,6 +2354,7 @@ struct file_system_type {
 
 	struct lock_class_key i_lock_key;
 	struct lock_class_key i_mutex_key;
+	struct lock_class_key invalidate_lock_key;
 	struct lock_class_key i_mutex_dir_key;
 };
 
diff --git a/mm/filemap.c b/mm/filemap.c
index bd7c50e060a9..9ea8dfb0609c 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -77,7 +77,8 @@
  *        ->i_pages lock
  *
  *  ->i_rwsem
- *    ->i_mmap_rwsem		(truncate->unmap_mapping_range)
+ *    ->invalidate_lock		(acquired by fs in truncate path)
+ *      ->i_mmap_rwsem		(truncate->unmap_mapping_range)
  *
  *  ->mmap_lock
  *    ->i_mmap_rwsem
@@ -85,7 +86,8 @@
  *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
  *
  *  ->mmap_lock
- *    ->lock_page		(access_process_vm)
+ *    ->invalidate_lock		(filemap_fault)
+ *      ->lock_page		(filemap_fault, access_process_vm)
  *
  *  ->i_rwsem			(generic_perform_write)
  *    ->mmap_lock		(fault_in_pages_readable->do_page_fault)
@@ -2276,20 +2278,30 @@ static int filemap_update_page(struct kiocb *iocb,
 {
 	int error;
 
+	if (iocb->ki_flags & IOCB_NOWAIT) {
+		if (!down_read_trylock(&mapping->invalidate_lock))
+			return -EAGAIN;
+	} else {
+		down_read(&mapping->invalidate_lock);
+	}
+
 	if (!trylock_page(page)) {
+		error = -EAGAIN;
 		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
-			return -EAGAIN;
+			goto unlock_mapping;
 		if (!(iocb->ki_flags & IOCB_WAITQ)) {
+			up_read(&mapping->invalidate_lock);
 			put_and_wait_on_page_locked(page, TASK_KILLABLE);
 			return AOP_TRUNCATED_PAGE;
 		}
 		error = __lock_page_async(page, iocb->ki_waitq);
 		if (error)
-			return error;
+			goto unlock_mapping;
 	}
 
+	error = AOP_TRUNCATED_PAGE;
 	if (!page->mapping)
-		goto truncated;
+		goto unlock;
 
 	error = 0;
 	if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page))
@@ -2300,15 +2312,13 @@ static int filemap_update_page(struct kiocb *iocb,
 		goto unlock;
 
 	error = filemap_read_page(iocb->ki_filp, mapping, page);
-	if (error == AOP_TRUNCATED_PAGE)
-		put_page(page);
-	return error;
-truncated:
-	unlock_page(page);
-	put_page(page);
-	return AOP_TRUNCATED_PAGE;
+	goto unlock_mapping;
 unlock:
 	unlock_page(page);
+unlock_mapping:
+	up_read(&mapping->invalidate_lock);
+	if (error == AOP_TRUNCATED_PAGE)
+		put_page(page);
 	return error;
 }
 
@@ -2323,6 +2333,19 @@ static int filemap_create_page(struct file *file,
 	if (!page)
 		return -ENOMEM;
 
+	/*
+	 * Protect against truncate / hole punch. Grabbing invalidate_lock here
+	 * assures we cannot instantiate and bring uptodate new pagecache pages
+	 * after evicting page cache during truncate and before actually
+	 * freeing blocks.  Note that we could release invalidate_lock after
+	 * inserting the page into page cache as the locked page would then be
+	 * enough to synchronize with hole punching. But there are code paths
+	 * such as filemap_update_page() filling in partially uptodate pages or
+	 * ->readpages() that need to hold invalidate_lock while mapping blocks
+	 * for IO so let's hold the lock here as well to keep locking rules
+	 * simple.
+	 */
+	down_read(&mapping->invalidate_lock);
 	error = add_to_page_cache_lru(page, mapping, index,
 			mapping_gfp_constraint(mapping, GFP_KERNEL));
 	if (error == -EEXIST)
@@ -2334,9 +2357,11 @@ static int filemap_create_page(struct file *file,
 	if (error)
 		goto error;
 
+	up_read(&mapping->invalidate_lock);
 	pagevec_add(pvec, page);
 	return 0;
 error:
+	up_read(&mapping->invalidate_lock);
 	put_page(page);
 	return error;
 }
@@ -2896,6 +2921,13 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
 		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
 		ret = VM_FAULT_MAJOR;
 		fpin = do_sync_mmap_readahead(vmf);
+	}
+
+	/*
+	 * See comment in filemap_create_page() why we need invalidate_lock
+	 */
+	down_read(&mapping->invalidate_lock);
+	if (!page) {
 retry_find:
 		page = pagecache_get_page(mapping, offset,
 					  FGP_CREAT|FGP_FOR_MMAP,
@@ -2903,6 +2935,7 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
 		if (!page) {
 			if (fpin)
 				goto out_retry;
+			up_read(&mapping->invalidate_lock);
 			return VM_FAULT_OOM;
 		}
 	}
@@ -2943,9 +2976,11 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
 	if (unlikely(offset >= max_off)) {
 		unlock_page(page);
 		put_page(page);
+		up_read(&mapping->invalidate_lock);
 		return VM_FAULT_SIGBUS;
 	}
 
+	up_read(&mapping->invalidate_lock);
 	vmf->page = page;
 	return ret | VM_FAULT_LOCKED;
 
@@ -2971,6 +3006,7 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
 	if (!error || error == AOP_TRUNCATED_PAGE)
 		goto retry_find;
 
+	up_read(&mapping->invalidate_lock);
 	shrink_readahead_size_eio(ra);
 	return VM_FAULT_SIGBUS;
 
@@ -2982,6 +3018,7 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
 	 */
 	if (page)
 		put_page(page);
+	up_read(&mapping->invalidate_lock);
 	if (fpin)
 		fput(fpin);
 	return ret | VM_FAULT_RETRY;
diff --git a/mm/readahead.c b/mm/readahead.c
index c5b0457415be..37dd07b32c67 100644
--- a/mm/readahead.c
+++ b/mm/readahead.c
@@ -192,6 +192,7 @@ void page_cache_ra_unbounded(struct readahead_control *ractl,
 	 */
 	unsigned int nofs = memalloc_nofs_save();
 
+	down_read(&mapping->invalidate_lock);
 	/*
 	 * Preallocate as many pages as we will need.
 	 */
@@ -236,6 +237,7 @@ void page_cache_ra_unbounded(struct readahead_control *ractl,
 	 * will then handle the error.
 	 */
 	read_pages(ractl, &page_pool, false);
+	up_read(&mapping->invalidate_lock);
 	memalloc_nofs_restore(nofs);
 }
 EXPORT_SYMBOL_GPL(page_cache_ra_unbounded);
diff --git a/mm/rmap.c b/mm/rmap.c
index dba8cb8a5578..e4f769a4dcc8 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -22,24 +22,25 @@
  *
  * inode->i_rwsem	(while writing or truncating, not reading or faulting)
  *   mm->mmap_lock
- *     page->flags PG_locked (lock_page)   * (see hugetlbfs below)
- *       hugetlbfs_i_mmap_rwsem_key (in huge_pmd_share)
- *         mapping->i_mmap_rwsem
- *           hugetlb_fault_mutex (hugetlbfs specific page fault mutex)
- *           anon_vma->rwsem
- *             mm->page_table_lock or pte_lock
- *               swap_lock (in swap_duplicate, swap_info_get)
- *                 mmlist_lock (in mmput, drain_mmlist and others)
- *                 mapping->private_lock (in __set_page_dirty_buffers)
- *                   lock_page_memcg move_lock (in __set_page_dirty_buffers)
- *                     i_pages lock (widely used)
- *                       lruvec->lru_lock (in lock_page_lruvec_irq)
- *                 inode->i_lock (in set_page_dirty's __mark_inode_dirty)
- *                 bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
- *                   sb_lock (within inode_lock in fs/fs-writeback.c)
- *                   i_pages lock (widely used, in set_page_dirty,
- *                             in arch-dependent flush_dcache_mmap_lock,
- *                             within bdi.wb->list_lock in __sync_single_inode)
+ *     mapping->invalidate_lock (in filemap_fault)
+ *       page->flags PG_locked (lock_page)   * (see hugetlbfs below)
+ *         hugetlbfs_i_mmap_rwsem_key (in huge_pmd_share)
+ *           mapping->i_mmap_rwsem
+ *             hugetlb_fault_mutex (hugetlbfs specific page fault mutex)
+ *             anon_vma->rwsem
+ *               mm->page_table_lock or pte_lock
+ *                 swap_lock (in swap_duplicate, swap_info_get)
+ *                   mmlist_lock (in mmput, drain_mmlist and others)
+ *                   mapping->private_lock (in __set_page_dirty_buffers)
+ *                     lock_page_memcg move_lock (in __set_page_dirty_buffers)
+ *                       i_pages lock (widely used)
+ *                         lruvec->lru_lock (in lock_page_lruvec_irq)
+ *                   inode->i_lock (in set_page_dirty's __mark_inode_dirty)
+ *                   bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
+ *                     sb_lock (within inode_lock in fs/fs-writeback.c)
+ *                     i_pages lock (widely used, in set_page_dirty,
+ *                               in arch-dependent flush_dcache_mmap_lock,
+ *                               within bdi.wb->list_lock in __sync_single_inode)
  *
  * anon_vma->rwsem,mapping->i_mmap_rwsem   (memory_failure, collect_procs_anon)
  *   ->tasklist_lock
diff --git a/mm/truncate.c b/mm/truncate.c
index 2cf71d8c3c62..464ad70a081f 100644
--- a/mm/truncate.c
+++ b/mm/truncate.c
@@ -416,7 +416,7 @@ EXPORT_SYMBOL(truncate_inode_pages_range);
  * @mapping: mapping to truncate
  * @lstart: offset from which to truncate
  *
- * Called under (and serialised by) inode->i_rwsem.
+ * Called under (and serialised by) inode->i_rwsem and inode->i_mapping_rwsem.
  *
  * Note: When this function returns, there can be a page in the process of
  * deletion (inside __delete_from_page_cache()) in the specified range.  Thus
-- 
2.26.2

[PATCH 10/12] shmem: Use invalidate_lock to protect fallocate

[Jan Kara]

We have to handle pages added by currently running shmem_fallocate()
specially in shmem_writepage(). For this we use serialization mechanism
using structure attached to inode->i_private field. If we protect
allocation of pages in shmem_fallocate() with invalidate_lock instead,
we are sure added pages cannot be dirtied until shmem_fallocate() is done
(invalidate_lock blocks faults, i_rwsem blocks writes) and thus we
cannot see those pages in shmem_writepage() and there's no need for the
serialization mechanism.

CC: Hugh Dickins <hughd@google.com>
CC: <linux-mm@kvack.org>
Signed-off-by: Jan Kara <jack@suse.cz>
---
 mm/shmem.c | 61 ++++++------------------------------------------------
 1 file changed, 6 insertions(+), 55 deletions(-)

diff --git a/mm/shmem.c b/mm/shmem.c
index f34162ac46de..7a2b0744031e 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -94,18 +94,6 @@ static struct vfsmount *shm_mnt;
 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
 #define SHORT_SYMLINK_LEN 128
 
-/*
- * shmem_fallocate communicates with shmem_writepage via inode->i_private (with
- * i_rwsem making sure that it has only one user at a time): we would prefer
- * not to enlarge the shmem inode just for that.
- */
-struct shmem_falloc {
-	pgoff_t start;		/* start of range currently being fallocated */
-	pgoff_t next;		/* the next page offset to be fallocated */
-	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
-	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
-};
-
 struct shmem_options {
 	unsigned long long blocks;
 	unsigned long long inodes;
@@ -1364,28 +1352,11 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
 	 * value into swapfile.c, the only way we can correctly account for a
 	 * fallocated page arriving here is now to initialize it and write it.
-	 *
-	 * That's okay for a page already fallocated earlier, but if we have
-	 * not yet completed the fallocation, then (a) we want to keep track
-	 * of this page in case we have to undo it, and (b) it may not be a
-	 * good idea to continue anyway, once we're pushing into swap.  So
-	 * reactivate the page, and let shmem_fallocate() quit when too many.
+	 * Since a page added by currently running fallocate call cannot be
+	 * dirtied and thus arrive here we know the fallocate has already
+	 * completed and we are fine writing it out.
 	 */
 	if (!PageUptodate(page)) {
-		if (inode->i_private) {
-			struct shmem_falloc *shmem_falloc;
-			spin_lock(&inode->i_lock);
-			shmem_falloc = inode->i_private;
-			if (shmem_falloc &&
-			    index >= shmem_falloc->start &&
-			    index < shmem_falloc->next)
-				shmem_falloc->nr_unswapped++;
-			else
-				shmem_falloc = NULL;
-			spin_unlock(&inode->i_lock);
-			if (shmem_falloc)
-				goto redirty;
-		}
 		clear_highpage(page);
 		flush_dcache_page(page);
 		SetPageUptodate(page);
@@ -2629,9 +2600,9 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 							 loff_t len)
 {
 	struct inode *inode = file_inode(file);
+	struct address_space *mapping = file->f_mapping;
 	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;
 	int error;
 
@@ -2641,7 +2612,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 	inode_lock(inode);
 
 	if (mode & FALLOC_FL_PUNCH_HOLE) {
-		struct address_space *mapping = file->f_mapping;
 		loff_t unmap_start = round_up(offset, PAGE_SIZE);
 		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
 
@@ -2680,14 +2650,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 		goto out;
 	}
 
-	shmem_falloc.start = start;
-	shmem_falloc.next  = start;
-	shmem_falloc.nr_falloced = 0;
-	shmem_falloc.nr_unswapped = 0;
-	spin_lock(&inode->i_lock);
-	inode->i_private = &shmem_falloc;
-	spin_unlock(&inode->i_lock);
-
+	down_write(&mapping->invalidate_lock);
 	for (index = start; index < end; index++) {
 		struct page *page;
 
@@ -2697,8 +2660,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 		 */
 		if (signal_pending(current))
 			error = -EINTR;
-		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
-			error = -ENOMEM;
 		else
 			error = shmem_getpage(inode, index, &page, SGP_FALLOC);
 		if (error) {
@@ -2711,14 +2672,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 			goto undone;
 		}
 
-		/*
-		 * Inform shmem_writepage() how far we have reached.
-		 * No need for lock or barrier: we have the page lock.
-		 */
-		shmem_falloc.next++;
-		if (!PageUptodate(page))
-			shmem_falloc.nr_falloced++;
-
 		/*
 		 * If !PageUptodate, leave it that way so that freeable pages
 		 * can be recognized if we need to rollback on error later.
@@ -2736,9 +2689,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 		i_size_write(inode, offset + len);
 	inode->i_ctime = current_time(inode);
 undone:
-	spin_lock(&inode->i_lock);
-	inode->i_private = NULL;
-	spin_unlock(&inode->i_lock);
+	up_write(&mapping->invalidate_lock);
 out:
 	inode_unlock(inode);
 	return error;
-- 
2.26.2

Re: [PATCH 02/12] mm: Protect operations adding pages to page cache with invalidate_lock

[Matthew Wilcox]

On Fri, Apr 23, 2021 at 07:29:31PM +0200, Jan Kara wrote:
> Currently, serializing operations such as page fault, read, or readahead
> against hole punching is rather difficult. The basic race scheme is
> like:
> 
> fallocate(FALLOC_FL_PUNCH_HOLE)			read / fault / ..
>   truncate_inode_pages_range()
> 						  <create pages in page
> 						   cache here>
>   <update fs block mapping and free blocks>
> 
> Now the problem is in this way read / page fault / readahead can
> instantiate pages in page cache with potentially stale data (if blocks
> get quickly reused). Avoiding this race is not simple - page locks do
> not work because we want to make sure there are *no* pages in given
> range.

One of the things I've had in mind for a while is moving the DAX locked
entry concept into the page cache proper.  It would avoid creating the
new semaphore, at the cost of taking the i_pages lock twice (once to
insert the entries that cover the range, and once to delete the entries).

It'd have pretty much the same effect, though -- read/fault/... would
block until the entry was deleted from the page cache.

Re: [PATCH 10/12] shmem: Use invalidate_lock to protect fallocate

[kernel test robot]

[-- Attachment #1: Type: text/plain, Size: 9742 bytes --]

Hi Jan,

I love your patch! Perhaps something to improve:

[auto build test WARNING on ext4/dev]
[also build test WARNING on fuse/for-next linus/master v5.12-rc8]
[cannot apply to hnaz-linux-mm/master next-20210423]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/Jan-Kara/fs-Hole-punch-vs-page-cache-filling-races/20210424-013114
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4.git dev
config: s390-randconfig-s031-20210424 (attached as .config)
compiler: s390-linux-gcc (GCC) 9.3.0
reproduce:
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # apt-get install sparse
        # sparse version: v0.6.3-341-g8af24329-dirty
        # https://github.com/0day-ci/linux/commit/800cf89f11d437415eead2da969c3b07908fd406
        git remote add linux-review https://github.com/0day-ci/linux
        git fetch --no-tags linux-review Jan-Kara/fs-Hole-punch-vs-page-cache-filling-races/20210424-013114
        git checkout 800cf89f11d437415eead2da969c3b07908fd406
        # save the attached .config to linux build tree
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-9.3.0 make.cross C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' W=1 ARCH=s390 

If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>

All warnings (new ones prefixed by >>):

   mm/shmem.c: In function 'shmem_writepage':
>> mm/shmem.c:1326:10: warning: variable 'index' set but not used [-Wunused-but-set-variable]
    1326 |  pgoff_t index;
         |          ^~~~~


vim +/index +1326 mm/shmem.c

^1da177e4c3f41 Linus Torvalds         2005-04-16  1316  
^1da177e4c3f41 Linus Torvalds         2005-04-16  1317  /*
^1da177e4c3f41 Linus Torvalds         2005-04-16  1318   * Move the page from the page cache to the swap cache.
^1da177e4c3f41 Linus Torvalds         2005-04-16  1319   */
^1da177e4c3f41 Linus Torvalds         2005-04-16  1320  static int shmem_writepage(struct page *page, struct writeback_control *wbc)
^1da177e4c3f41 Linus Torvalds         2005-04-16  1321  {
^1da177e4c3f41 Linus Torvalds         2005-04-16  1322  	struct shmem_inode_info *info;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1323  	struct address_space *mapping;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1324  	struct inode *inode;
6922c0c7abd387 Hugh Dickins           2011-08-03  1325  	swp_entry_t swap;
6922c0c7abd387 Hugh Dickins           2011-08-03 @1326  	pgoff_t index;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1327  
800d8c63b2e989 Kirill A. Shutemov     2016-07-26  1328  	VM_BUG_ON_PAGE(PageCompound(page), page);
^1da177e4c3f41 Linus Torvalds         2005-04-16  1329  	BUG_ON(!PageLocked(page));
^1da177e4c3f41 Linus Torvalds         2005-04-16  1330  	mapping = page->mapping;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1331  	index = page->index;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1332  	inode = mapping->host;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1333  	info = SHMEM_I(inode);
^1da177e4c3f41 Linus Torvalds         2005-04-16  1334  	if (info->flags & VM_LOCKED)
^1da177e4c3f41 Linus Torvalds         2005-04-16  1335  		goto redirty;
d9fe526a83b84e Hugh Dickins           2008-02-04  1336  	if (!total_swap_pages)
^1da177e4c3f41 Linus Torvalds         2005-04-16  1337  		goto redirty;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1338  
d9fe526a83b84e Hugh Dickins           2008-02-04  1339  	/*
97b713ba3ebaa6 Christoph Hellwig      2015-01-14  1340  	 * Our capabilities prevent regular writeback or sync from ever calling
97b713ba3ebaa6 Christoph Hellwig      2015-01-14  1341  	 * shmem_writepage; but a stacking filesystem might use ->writepage of
97b713ba3ebaa6 Christoph Hellwig      2015-01-14  1342  	 * its underlying filesystem, in which case tmpfs should write out to
97b713ba3ebaa6 Christoph Hellwig      2015-01-14  1343  	 * swap only in response to memory pressure, and not for the writeback
97b713ba3ebaa6 Christoph Hellwig      2015-01-14  1344  	 * threads or sync.
d9fe526a83b84e Hugh Dickins           2008-02-04  1345  	 */
48f170fb7d7db8 Hugh Dickins           2011-07-25  1346  	if (!wbc->for_reclaim) {
48f170fb7d7db8 Hugh Dickins           2011-07-25  1347  		WARN_ON_ONCE(1);	/* Still happens? Tell us about it! */
48f170fb7d7db8 Hugh Dickins           2011-07-25  1348  		goto redirty;
48f170fb7d7db8 Hugh Dickins           2011-07-25  1349  	}
1635f6a74152f1 Hugh Dickins           2012-05-29  1350  
1635f6a74152f1 Hugh Dickins           2012-05-29  1351  	/*
1635f6a74152f1 Hugh Dickins           2012-05-29  1352  	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1635f6a74152f1 Hugh Dickins           2012-05-29  1353  	 * value into swapfile.c, the only way we can correctly account for a
1635f6a74152f1 Hugh Dickins           2012-05-29  1354  	 * fallocated page arriving here is now to initialize it and write it.
800cf89f11d437 Jan Kara               2021-04-23  1355  	 * Since a page added by currently running fallocate call cannot be
800cf89f11d437 Jan Kara               2021-04-23  1356  	 * dirtied and thus arrive here we know the fallocate has already
800cf89f11d437 Jan Kara               2021-04-23  1357  	 * completed and we are fine writing it out.
1635f6a74152f1 Hugh Dickins           2012-05-29  1358  	 */
1635f6a74152f1 Hugh Dickins           2012-05-29  1359  	if (!PageUptodate(page)) {
1635f6a74152f1 Hugh Dickins           2012-05-29  1360  		clear_highpage(page);
1635f6a74152f1 Hugh Dickins           2012-05-29  1361  		flush_dcache_page(page);
1635f6a74152f1 Hugh Dickins           2012-05-29  1362  		SetPageUptodate(page);
1635f6a74152f1 Hugh Dickins           2012-05-29  1363  	}
1635f6a74152f1 Hugh Dickins           2012-05-29  1364  
38d8b4e6bdc872 Huang Ying             2017-07-06  1365  	swap = get_swap_page(page);
48f170fb7d7db8 Hugh Dickins           2011-07-25  1366  	if (!swap.val)
48f170fb7d7db8 Hugh Dickins           2011-07-25  1367  		goto redirty;
d9fe526a83b84e Hugh Dickins           2008-02-04  1368  
b1dea800ac3959 Hugh Dickins           2011-05-11  1369  	/*
b1dea800ac3959 Hugh Dickins           2011-05-11  1370  	 * Add inode to shmem_unuse()'s list of swapped-out inodes,
6922c0c7abd387 Hugh Dickins           2011-08-03  1371  	 * if it's not already there.  Do it now before the page is
6922c0c7abd387 Hugh Dickins           2011-08-03  1372  	 * moved to swap cache, when its pagelock no longer protects
b1dea800ac3959 Hugh Dickins           2011-05-11  1373  	 * the inode from eviction.  But don't unlock the mutex until
6922c0c7abd387 Hugh Dickins           2011-08-03  1374  	 * we've incremented swapped, because shmem_unuse_inode() will
6922c0c7abd387 Hugh Dickins           2011-08-03  1375  	 * prune a !swapped inode from the swaplist under this mutex.
b1dea800ac3959 Hugh Dickins           2011-05-11  1376  	 */
b1dea800ac3959 Hugh Dickins           2011-05-11  1377  	mutex_lock(&shmem_swaplist_mutex);
05bf86b4ccfd0f Hugh Dickins           2011-05-14  1378  	if (list_empty(&info->swaplist))
b56a2d8af9147a Vineeth Remanan Pillai 2019-03-05  1379  		list_add(&info->swaplist, &shmem_swaplist);
b1dea800ac3959 Hugh Dickins           2011-05-11  1380  
4afab1cd256e42 Yang Shi               2019-11-30  1381  	if (add_to_swap_cache(page, swap,
3852f6768ede54 Joonsoo Kim            2020-08-11  1382  			__GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
3852f6768ede54 Joonsoo Kim            2020-08-11  1383  			NULL) == 0) {
4595ef88d13613 Kirill A. Shutemov     2016-07-26  1384  		spin_lock_irq(&info->lock);
6922c0c7abd387 Hugh Dickins           2011-08-03  1385  		shmem_recalc_inode(inode);
267a4c76bbdb95 Hugh Dickins           2015-12-11  1386  		info->swapped++;
4595ef88d13613 Kirill A. Shutemov     2016-07-26  1387  		spin_unlock_irq(&info->lock);
6922c0c7abd387 Hugh Dickins           2011-08-03  1388  
267a4c76bbdb95 Hugh Dickins           2015-12-11  1389  		swap_shmem_alloc(swap);
267a4c76bbdb95 Hugh Dickins           2015-12-11  1390  		shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
267a4c76bbdb95 Hugh Dickins           2015-12-11  1391  
6922c0c7abd387 Hugh Dickins           2011-08-03  1392  		mutex_unlock(&shmem_swaplist_mutex);
d9fe526a83b84e Hugh Dickins           2008-02-04  1393  		BUG_ON(page_mapped(page));
9fab5619bdd7f8 Hugh Dickins           2009-03-31  1394  		swap_writepage(page, wbc);
^1da177e4c3f41 Linus Torvalds         2005-04-16  1395  		return 0;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1396  	}
^1da177e4c3f41 Linus Torvalds         2005-04-16  1397  
6922c0c7abd387 Hugh Dickins           2011-08-03  1398  	mutex_unlock(&shmem_swaplist_mutex);
75f6d6d29a40b5 Minchan Kim            2017-07-06  1399  	put_swap_page(page, swap);
^1da177e4c3f41 Linus Torvalds         2005-04-16  1400  redirty:
^1da177e4c3f41 Linus Torvalds         2005-04-16  1401  	set_page_dirty(page);
d9fe526a83b84e Hugh Dickins           2008-02-04  1402  	if (wbc->for_reclaim)
d9fe526a83b84e Hugh Dickins           2008-02-04  1403  		return AOP_WRITEPAGE_ACTIVATE;	/* Return with page locked */
d9fe526a83b84e Hugh Dickins           2008-02-04  1404  	unlock_page(page);
d9fe526a83b84e Hugh Dickins           2008-02-04  1405  	return 0;
^1da177e4c3f41 Linus Torvalds         2005-04-16  1406  }
^1da177e4c3f41 Linus Torvalds         2005-04-16  1407  

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org

[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 14079 bytes --]

Re: [PATCH 0/12 v4] fs: Hole punch vs page cache filling races

[Dave Chinner]

Hi Jan,

In future, can you please use the same cc-list for the entire
patchset?

The stuff that has hit the XFS list (where I'm replying from)
doesn't give me any context as to what the core changes are that
allow XFS to be changed, so I can't review them in isolation.

I've got to spend time now reconstructing the patchset into a single
series because the delivery has been spread across three different
mailing lists and so hit 3 different procmail filters.  I'll comment
on the patches once I've reconstructed the series and read through
it as a whole...

/me considers the way people use "cc" tags in git commits for
including mailing lists on individual patches actively harmful.
Unless the recipient is subscribed to all the mailing lists the
patchset was CC'd to, they can't easily find the bits of the
patchset that didn't arrive in their mail box. Individual mailing
lists should receive entire patchsets for review, not random,
individual, context free patches. 

And, FWIW, cc'ing the cover letter to all the mailing lists is not
good enough. Being able to see the code change as a whole is what
matters for review, not the cover letter...

Cheers,

Dave.

On Fri, Apr 23, 2021 at 07:29:29PM +0200, Jan Kara wrote:
> Hello,
> 
> here is another version of my patches to address races between hole punching
> and page cache filling functions for ext4 and other filesystems. I think
> we are coming close to a complete solution so I've removed the RFC tag from
> the subject. I went through all filesystems supporting hole punching and
> converted them from their private locks to a generic one (usually fixing the
> race ext4 had as a side effect). I also found out ceph & cifs didn't have
> any protection from the hole punch vs page fault race either so I've added
> appropriate protections there. Open are still GFS2 and OCFS2 filesystems.
> GFS2 actually avoids the race but is prone to deadlocks (acquires the same lock
> both above and below mmap_sem), OCFS2 locking seems kind of hosed and some
> read, write, and hole punch paths are not properly serialized possibly leading
> to fs corruption. Both issues are non-trivial so respective fs maintainers
> have to deal with those (I've informed them and problems were generally
> confirmed). Anyway, for all the other filesystem this kind of race should
> be closed.
> 
> As a next step, I'd like to actually make sure all calls to
> truncate_inode_pages() happen under mapping->invalidate_lock, add the assert
> and then we can also get rid of i_size checks in some places (truncate can
> use the same serialization scheme as hole punch). But that step is mostly
> a cleanup so I'd like to get these functional fixes in first.
> 
> Changes since v3:
> * Renamed and moved lock to struct address_space
> * Added conversions of tmpfs, ceph, cifs, fuse, f2fs
> * Fixed error handling path in filemap_read()
> * Removed .page_mkwrite() cleanup from the series for now
> 
> Changes since v2:
> * Added documentation and comments regarding lock ordering and how the lock is
>   supposed to be used
> * Added conversions of ext2, xfs, zonefs
> * Added patch removing i_mapping_sem protection from .page_mkwrite handlers
> 
> Changes since v1:
> * Moved to using inode->i_mapping_sem instead of aops handler to acquire
>   appropriate lock
> 
> ---
> Motivation:
> 
> Amir has reported [1] a that ext4 has a potential issues when reads can race
> with hole punching possibly exposing stale data from freed blocks or even
> corrupting filesystem when stale mapping data gets used for writeout. The
> problem is that during hole punching, new page cache pages can get instantiated
> and block mapping from the looked up in a punched range after
> truncate_inode_pages() has run but before the filesystem removes blocks from
> the file. In principle any filesystem implementing hole punching thus needs to
> implement a mechanism to block instantiating page cache pages during hole
> punching to avoid this race. This is further complicated by the fact that there
> are multiple places that can instantiate pages in page cache.  We can have
> regular read(2) or page fault doing this but fadvise(2) or madvise(2) can also
> result in reading in page cache pages through force_page_cache_readahead().
> 
> There are couple of ways how to fix this. First way (currently implemented by
> XFS) is to protect read(2) and *advise(2) calls with i_rwsem so that they are
> serialized with hole punching. This is easy to do but as a result all reads
> would then be serialized with writes and thus mixed read-write workloads suffer
> heavily on ext4. Thus this series introduces inode->i_mapping_sem and uses it
> when creating new pages in the page cache and looking up their corresponding
> block mapping. We also replace EXT4_I(inode)->i_mmap_sem with this new rwsem
> which provides necessary serialization with hole punching for ext4.
> 
> 								Honza
> 
> [1] https://lore.kernel.org/linux-fsdevel/CAOQ4uxjQNmxqmtA_VbYW0Su9rKRk2zobJmahcyeaEVOFKVQ5dw@mail.gmail.com/
> 
> Previous versions:
> Link: https://lore.kernel.org/linux-fsdevel/20210208163918.7871-1-jack@suse.cz/
> Link: http://lore.kernel.org/r/20210413105205.3093-1-jack@suse.cz
> 
> CC: ceph-devel@vger.kernel.org
> CC: Chao Yu <yuchao0@huawei.com>
> CC: Damien Le Moal <damien.lemoal@wdc.com>
> CC: "Darrick J. Wong" <darrick.wong@oracle.com>
> CC: Hugh Dickins <hughd@google.com>
> CC: Jaegeuk Kim <jaegeuk@kernel.org>
> CC: Jeff Layton <jlayton@kernel.org>
> CC: Johannes Thumshirn <jth@kernel.org>
> CC: linux-cifs@vger.kernel.org
> CC: <linux-ext4@vger.kernel.org>
> CC: linux-f2fs-devel@lists.sourceforge.net
> CC: <linux-fsdevel@vger.kernel.org>
> CC: <linux-mm@kvack.org>
> CC: <linux-xfs@vger.kernel.org>
> CC: Miklos Szeredi <miklos@szeredi.hu>
> CC: Steve French <sfrench@samba.org>
> CC: Ted Tso <tytso@mit.edu>
> 

-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 02/12] mm: Protect operations adding pages to page cache with invalidate_lock

[Dave Chinner]

On Fri, Apr 23, 2021 at 07:29:31PM +0200, Jan Kara wrote:
> Currently, serializing operations such as page fault, read, or readahead
> against hole punching is rather difficult. The basic race scheme is
> like:
> 
> fallocate(FALLOC_FL_PUNCH_HOLE)			read / fault / ..
>   truncate_inode_pages_range()
> 						  <create pages in page
> 						   cache here>
>   <update fs block mapping and free blocks>
> 
> Now the problem is in this way read / page fault / readahead can
> instantiate pages in page cache with potentially stale data (if blocks
> get quickly reused). Avoiding this race is not simple - page locks do
> not work because we want to make sure there are *no* pages in given
> range. inode->i_rwsem does not work because page fault happens under
> mmap_sem which ranks below inode->i_rwsem. Also using it for reads makes
> the performance for mixed read-write workloads suffer.
> 
> So create a new rw_semaphore in the address_space - invalidate_lock -
> that protects adding of pages to page cache for page faults / reads /
> readahead.
.....
> diff --git a/fs/inode.c b/fs/inode.c
> index a047ab306f9a..43596dd8b61e 100644
> --- a/fs/inode.c
> +++ b/fs/inode.c
> @@ -191,6 +191,9 @@ int inode_init_always(struct super_block *sb, struct inode *inode)
>  	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
>  	mapping->private_data = NULL;
>  	mapping->writeback_index = 0;
> +	init_rwsem(&mapping->invalidate_lock);
> +	lockdep_set_class(&mapping->invalidate_lock,
> +			  &sb->s_type->invalidate_lock_key);
>  	inode->i_private = NULL;
>  	inode->i_mapping = mapping;
>  	INIT_HLIST_HEAD(&inode->i_dentry);	/* buggered by rcu freeing */

Oh, lockdep. That might be a problem here.

The XFS_MMAPLOCK has non-trivial lockdep annotations so that it is
tracked as nesting properly against the IOLOCK and the ILOCK. When
you end up using xfs_ilock(XFS_MMAPLOCK..) to lock this, XFS will
add subclass annotations to the lock and they are going to be
different to the locking that the VFS does.

We'll see this from xfs_lock_two_inodes() (e.g. in
xfs_swap_extents()) and xfs_ilock2_io_mmap() during reflink
oper.....

Oooooh. The page cache copy done when breaking a shared extent needs
to lock out page faults on both the source and destination, but it
still needs to be able to populate the page cache of both the source
and destination file.....

.... and vfs_dedupe_file_range_compare() has to be able to read
pages from both the source and destination file to determine that
the contents are identical and that's done while we hold the
XFS_MMAPLOCK exclusively so the compare is atomic w.r.t. all other
user data modification operations being run....

I now have many doubts that this "serialise page faults by locking
out page cache instantiation" method actually works as a generic
mechanism. It's not just page cache invalidation that relies on
being able to lock out page faults: copy-on-write and deduplication
both require the ability to populate the page cache with source data
while page faults are locked out so the data can be compared/copied
atomically with the extent level manipulations and so user data
modifications cannot occur until the physical extent manipulation
operation has completed.

Having only just realised this is a problem, no solution has
immediately popped into my mind. I'll chew on it over the weekend,
but I'm not hopeful at this point...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 0/12 v4] fs: Hole punch vs page cache filling races

[Matthew Wilcox]

On Sat, Apr 24, 2021 at 08:07:51AM +1000, Dave Chinner wrote:
> I've got to spend time now reconstructing the patchset into a single
> series because the delivery has been spread across three different
> mailing lists and so hit 3 different procmail filters.  I'll comment
> on the patches once I've reconstructed the series and read through
> it as a whole...

$ b4 mbox 20210423171010.12-1-jack@suse.cz
Looking up https://lore.kernel.org/r/20210423171010.12-1-jack%40suse.cz
Grabbing thread from lore.kernel.org/ceph-devel
6 messages in the thread
Saved ./20210423171010.12-1-jack@suse.cz.mbx

Re: [PATCH 0/12 v4] fs: Hole punch vs page cache filling races

[Christoph Hellwig]

On Sat, Apr 24, 2021 at 12:51:49AM +0100, Matthew Wilcox wrote:
> On Sat, Apr 24, 2021 at 08:07:51AM +1000, Dave Chinner wrote:
> > I've got to spend time now reconstructing the patchset into a single
> > series because the delivery has been spread across three different
> > mailing lists and so hit 3 different procmail filters.  I'll comment
> > on the patches once I've reconstructed the series and read through
> > it as a whole...
> 
> $ b4 mbox 20210423171010.12-1-jack@suse.cz
> Looking up https://lore.kernel.org/r/20210423171010.12-1-jack%40suse.cz
> Grabbing thread from lore.kernel.org/ceph-devel
> 6 messages in the thread
> Saved ./20210423171010.12-1-jack@suse.cz.mbx

Yikes.  Just send them damn mails.  Or switch the lists to NNTP, but
don't let the people who are reviewing your patches do stupid work
with weird tools.

Re: [PATCH 02/12] mm: Protect operations adding pages to page cache with invalidate_lock

[Jan Kara]

On Sat 24-04-21 09:04:49, Dave Chinner wrote:
> On Fri, Apr 23, 2021 at 07:29:31PM +0200, Jan Kara wrote:
> > Currently, serializing operations such as page fault, read, or readahead
> > against hole punching is rather difficult. The basic race scheme is
> > like:
> > 
> > fallocate(FALLOC_FL_PUNCH_HOLE)			read / fault / ..
> >   truncate_inode_pages_range()
> > 						  <create pages in page
> > 						   cache here>
> >   <update fs block mapping and free blocks>
> > 
> > Now the problem is in this way read / page fault / readahead can
> > instantiate pages in page cache with potentially stale data (if blocks
> > get quickly reused). Avoiding this race is not simple - page locks do
> > not work because we want to make sure there are *no* pages in given
> > range. inode->i_rwsem does not work because page fault happens under
> > mmap_sem which ranks below inode->i_rwsem. Also using it for reads makes
> > the performance for mixed read-write workloads suffer.
> > 
> > So create a new rw_semaphore in the address_space - invalidate_lock -
> > that protects adding of pages to page cache for page faults / reads /
> > readahead.
> .....
> > diff --git a/fs/inode.c b/fs/inode.c
> > index a047ab306f9a..43596dd8b61e 100644
> > --- a/fs/inode.c
> > +++ b/fs/inode.c
> > @@ -191,6 +191,9 @@ int inode_init_always(struct super_block *sb, struct inode *inode)
> >  	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
> >  	mapping->private_data = NULL;
> >  	mapping->writeback_index = 0;
> > +	init_rwsem(&mapping->invalidate_lock);
> > +	lockdep_set_class(&mapping->invalidate_lock,
> > +			  &sb->s_type->invalidate_lock_key);
> >  	inode->i_private = NULL;
> >  	inode->i_mapping = mapping;
> >  	INIT_HLIST_HEAD(&inode->i_dentry);	/* buggered by rcu freeing */
> 
> Oh, lockdep. That might be a problem here.
> 
> The XFS_MMAPLOCK has non-trivial lockdep annotations so that it is
> tracked as nesting properly against the IOLOCK and the ILOCK. When
> you end up using xfs_ilock(XFS_MMAPLOCK..) to lock this, XFS will
> add subclass annotations to the lock and they are going to be
> different to the locking that the VFS does.
> 
> We'll see this from xfs_lock_two_inodes() (e.g. in
> xfs_swap_extents()) and xfs_ilock2_io_mmap() during reflink
> oper.....

Thanks for the pointer. I was kind of wondering what lockdep nesting games
XFS plays but then forgot to look into details. Anyway, I've preserved the
nesting annotations in XFS and fstests run on XFS passed without lockdep
complaining so there isn't at least an obvious breakage. Also as far as I'm
checking the code XFS usage in and lock nesting of MMAPLOCK should be
compatible with the nesting VFS enforces (also see below)...
 
> Oooooh. The page cache copy done when breaking a shared extent needs
> to lock out page faults on both the source and destination, but it
> still needs to be able to populate the page cache of both the source
> and destination file.....
> 
> .... and vfs_dedupe_file_range_compare() has to be able to read
> pages from both the source and destination file to determine that
> the contents are identical and that's done while we hold the
> XFS_MMAPLOCK exclusively so the compare is atomic w.r.t. all other
> user data modification operations being run....

So I started wondering why fstests passed when reading this :) The reason
is that vfs_dedupe_get_page() does not use standard page cache filling path
(neither readahead API nor filemap_read()), instead it uses
read_mapping_page() and so gets into page cache filling path below the
level at which we get invalidate_lock and thus everything works as it
should. So read_mapping_page() is similar to places like e.g.
block_truncate_page() or block_write_begin() which may end up filling in
page cache contents but they rely on upper layers to already hold
appropriate locks. I'll add a comment to read_mapping_page() about this.
Once all filesystems are converted to use invalidate_lock, I also want to
add WARN_ON_ONCE() to various places verifying that invalidate_lock is held
as it should...
 
> I now have many doubts that this "serialise page faults by locking
> out page cache instantiation" method actually works as a generic
> mechanism. It's not just page cache invalidation that relies on
> being able to lock out page faults: copy-on-write and deduplication
> both require the ability to populate the page cache with source data
> while page faults are locked out so the data can be compared/copied
> atomically with the extent level manipulations and so user data
> modifications cannot occur until the physical extent manipulation
> operation has completed.

Hum, that is a good point. So there are actually two different things you
want to block at different places:

1) You really want to block page cache instantiation for operations such as
hole punch as that operation mutates data and thus contents would become
stale.

2) You want to block page cache *modification* for operations such as
dedupe while keeping page cache in place. This is somewhat different
requirement but invalidate_lock can, in principle, cover it as well.
Basically we just need to keep invalidate_lock usage in .page_mkwrite
helpers. The question remains whether invalidate_lock is still a good name
with this usage in mind and I probably need to update a documentation to
reflect this usage.

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH 10/12] shmem: Use invalidate_lock to protect fallocate

[Hugh Dickins]

On Fri, 23 Apr 2021, Jan Kara wrote:

> We have to handle pages added by currently running shmem_fallocate()
> specially in shmem_writepage(). For this we use serialization mechanism
> using structure attached to inode->i_private field. If we protect
> allocation of pages in shmem_fallocate() with invalidate_lock instead,
> we are sure added pages cannot be dirtied until shmem_fallocate() is done
> (invalidate_lock blocks faults, i_rwsem blocks writes) and thus we
> cannot see those pages in shmem_writepage() and there's no need for the
> serialization mechanism.

Appealing diffstat, but NAK, this patch is based on a false premise.

All those pages that are added by shmem_fallocate() are marked dirty:
see the set_page_dirty(page) and comment above it in shmem_fallocate().

It's intentional, that they should percolate through to shmem_writepage()
when memory is tight, and give feedback to fallocate that it should stop
(instead of getting into that horrid OOM-kill flurry that never even
frees any tmpfs anyway).

Hugh

> 
> CC: Hugh Dickins <hughd@google.com>
> CC: <linux-mm@kvack.org>
> Signed-off-by: Jan Kara <jack@suse.cz>
> ---
>  mm/shmem.c | 61 ++++++------------------------------------------------
>  1 file changed, 6 insertions(+), 55 deletions(-)
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index f34162ac46de..7a2b0744031e 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -94,18 +94,6 @@ static struct vfsmount *shm_mnt;
>  /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
>  #define SHORT_SYMLINK_LEN 128
>  
> -/*
> - * shmem_fallocate communicates with shmem_writepage via inode->i_private (with
> - * i_rwsem making sure that it has only one user at a time): we would prefer
> - * not to enlarge the shmem inode just for that.
> - */
> -struct shmem_falloc {
> -	pgoff_t start;		/* start of range currently being fallocated */
> -	pgoff_t next;		/* the next page offset to be fallocated */
> -	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
> -	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
> -};
> -
>  struct shmem_options {
>  	unsigned long long blocks;
>  	unsigned long long inodes;
> @@ -1364,28 +1352,11 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc)
>  	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
>  	 * value into swapfile.c, the only way we can correctly account for a
>  	 * fallocated page arriving here is now to initialize it and write it.
> -	 *
> -	 * That's okay for a page already fallocated earlier, but if we have
> -	 * not yet completed the fallocation, then (a) we want to keep track
> -	 * of this page in case we have to undo it, and (b) it may not be a
> -	 * good idea to continue anyway, once we're pushing into swap.  So
> -	 * reactivate the page, and let shmem_fallocate() quit when too many.

(b) there commenting on communicating back to fallocate by nr_unswapped.

> +	 * Since a page added by currently running fallocate call cannot be
> +	 * dirtied and thus arrive here we know the fallocate has already
> +	 * completed and we are fine writing it out.
>  	 */
>  	if (!PageUptodate(page)) {
> -		if (inode->i_private) {
> -			struct shmem_falloc *shmem_falloc;
> -			spin_lock(&inode->i_lock);
> -			shmem_falloc = inode->i_private;
> -			if (shmem_falloc &&
> -			    index >= shmem_falloc->start &&
> -			    index < shmem_falloc->next)
> -				shmem_falloc->nr_unswapped++;
> -			else
> -				shmem_falloc = NULL;
> -			spin_unlock(&inode->i_lock);
> -			if (shmem_falloc)
> -				goto redirty;
> -		}
>  		clear_highpage(page);
>  		flush_dcache_page(page);
>  		SetPageUptodate(page);
> @@ -2629,9 +2600,9 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  							 loff_t len)
>  {
>  	struct inode *inode = file_inode(file);
> +	struct address_space *mapping = file->f_mapping;
>  	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;
>  	int error;
>  
> @@ -2641,7 +2612,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  	inode_lock(inode);
>  
>  	if (mode & FALLOC_FL_PUNCH_HOLE) {
> -		struct address_space *mapping = file->f_mapping;
>  		loff_t unmap_start = round_up(offset, PAGE_SIZE);
>  		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
>  
> @@ -2680,14 +2650,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  		goto out;
>  	}
>  
> -	shmem_falloc.start = start;
> -	shmem_falloc.next  = start;
> -	shmem_falloc.nr_falloced = 0;
> -	shmem_falloc.nr_unswapped = 0;
> -	spin_lock(&inode->i_lock);
> -	inode->i_private = &shmem_falloc;
> -	spin_unlock(&inode->i_lock);
> -
> +	down_write(&mapping->invalidate_lock);
>  	for (index = start; index < end; index++) {
>  		struct page *page;
>  
> @@ -2697,8 +2660,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  		 */
>  		if (signal_pending(current))
>  			error = -EINTR;
> -		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
> -			error = -ENOMEM;
>  		else
>  			error = shmem_getpage(inode, index, &page, SGP_FALLOC);
>  		if (error) {
> @@ -2711,14 +2672,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  			goto undone;
>  		}
>  
> -		/*
> -		 * Inform shmem_writepage() how far we have reached.
> -		 * No need for lock or barrier: we have the page lock.
> -		 */
> -		shmem_falloc.next++;
> -		if (!PageUptodate(page))
> -			shmem_falloc.nr_falloced++;
> -
>  		/*
>  		 * If !PageUptodate, leave it that way so that freeable pages
>  		 * can be recognized if we need to rollback on error later.

Which goes on to say:

		 * But set_page_dirty so that memory pressure will swap rather
		 * than free the pages we are allocating (and SGP_CACHE pages
		 * might still be clean: we now need to mark those dirty too).
		 */
		set_page_dirty(page);
		unlock_page(page);
		put_page(page);
		cond_resched();

> @@ -2736,9 +2689,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  		i_size_write(inode, offset + len);
>  	inode->i_ctime = current_time(inode);
>  undone:
> -	spin_lock(&inode->i_lock);
> -	inode->i_private = NULL;
> -	spin_unlock(&inode->i_lock);
> +	up_write(&mapping->invalidate_lock);
>  out:
>  	inode_unlock(inode);
>  	return error;
> -- 
> 2.26.2

Re: [PATCH 09/12] shmem: Convert to using invalidate_lock

[Hugh Dickins]

On Fri, 23 Apr 2021, Jan Kara wrote:

> Shmem uses a home-grown mechanism for serializing hole punch with page
> fault. Use mapping->invalidate_lock for it instead. Admittedly the
> home-grown mechanism locks out only the range being actually punched out
> while invalidate_lock locks the whole mapping so it is serializing more.
> But hole punch doesn't seem to be that critical operation and the
> simplification is noticeable.

Home-grown indeed (and went through several different bugginesses,
Linus fixing issues in its waitq handling found years later).

I'd love to remove it all (rather than replace it by a new rwsem),
but never enough courage+time to do so: on optimistic days (that is,
rarely) I like to think that none of it would be needed nowadays;
but its gestation was difficult, and I cannot easily reproduce the
testing that demanded it (Sasha and Vlastimil helped a lot).

If you're interested in the history, I cannot point to one thread,
but "shmem: fix faulting into a hole while it's punched" finds
some of them, June/July 2014.  You've pushed me into re-reading
there, but I've not yet found the crucial evidence that stopped us
from reverting this mechanism, once we had abandoned the hole-punch
"pincer" in shmem_undo_range().

tmpfs's problem with faulting versus hole-punch was not the data
integrity issue you are attacking with invalidate_lock, but a
starvation issue triggered in Trinity fuzzing.

If invalidate_lock had existed at the time, I might have reused it
for this purpose too - I certainly wanted to avoid enlarging the
inode with another rwsem just for this; but also reluctant to add
another layer of locking to the common path (maybe I'm just silly
to try to avoid an rwsem which is so rarely taken for writing?).

But the code as it stands is working satisfactorily with minimal
overhead: so I'm not in a rush to remove or replace it yet. Thank
you for including tmpfs in your reach, but I think for the moment
I'd prefer you to leave this change out of the series. Maybe later
when it's settled in the fs/ filesystems (perhaps making guarantees
that we might want to extend to tmpfs) we could make this change -
but I'd still rather let hole-punch and fault race freely without it.

But your 01/12, fixing mm comments mentioning i_mutex, looked good:
Acked-by: Hugh Dickins <hughd@google.com>
to that one.  But I think it would be better extracted from this
invalidate_lock series, and just sent to akpm cc linux-mm on its own.

Thanks,
Hugh

> 
> CC: Hugh Dickins <hughd@google.com>
> CC: <linux-mm@kvack.org>
> Signed-off-by: Jan Kara <jack@suse.cz>
> ---
>  mm/shmem.c | 98 ++++--------------------------------------------------
>  1 file changed, 7 insertions(+), 91 deletions(-)
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 55b2888db542..f34162ac46de 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -95,12 +95,11 @@ static struct vfsmount *shm_mnt;
>  #define SHORT_SYMLINK_LEN 128
>  
>  /*
> - * shmem_fallocate communicates with shmem_fault or shmem_writepage via
> - * inode->i_private (with i_rwsem making sure that it has only one user at
> - * a time): we would prefer not to enlarge the shmem inode just for that.
> + * shmem_fallocate communicates with shmem_writepage via inode->i_private (with
> + * i_rwsem making sure that it has only one user at a time): we would prefer
> + * not to enlarge the shmem inode just for that.
>   */
>  struct shmem_falloc {
> -	wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
>  	pgoff_t start;		/* start of range currently being fallocated */
>  	pgoff_t next;		/* the next page offset to be fallocated */
>  	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
> @@ -1378,7 +1377,6 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc)
>  			spin_lock(&inode->i_lock);
>  			shmem_falloc = inode->i_private;
>  			if (shmem_falloc &&
> -			    !shmem_falloc->waitq &&
>  			    index >= shmem_falloc->start &&
>  			    index < shmem_falloc->next)
>  				shmem_falloc->nr_unswapped++;
> @@ -2025,18 +2023,6 @@ static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
>  	return error;
>  }
>  
> -/*
> - * This is like autoremove_wake_function, but it removes the wait queue
> - * entry unconditionally - even if something else had already woken the
> - * target.
> - */
> -static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
> -{
> -	int ret = default_wake_function(wait, mode, sync, key);
> -	list_del_init(&wait->entry);
> -	return ret;
> -}
> -
>  static vm_fault_t shmem_fault(struct vm_fault *vmf)
>  {
>  	struct vm_area_struct *vma = vmf->vma;
> @@ -2046,65 +2032,6 @@ static vm_fault_t shmem_fault(struct vm_fault *vmf)
>  	int err;
>  	vm_fault_t ret = VM_FAULT_LOCKED;
>  
> -	/*
> -	 * Trinity finds that probing a hole which tmpfs is punching can
> -	 * prevent the hole-punch from ever completing: which in turn
> -	 * locks writers out with its hold on i_rwsem.  So refrain from
> -	 * faulting pages into the hole while it's being punched.  Although
> -	 * shmem_undo_range() does remove the additions, it may be unable to
> -	 * keep up, as each new page needs its own unmap_mapping_range() call,
> -	 * and the i_mmap tree grows ever slower to scan if new vmas are added.
> -	 *
> -	 * It does not matter if we sometimes reach this check just before the
> -	 * hole-punch begins, so that one fault then races with the punch:
> -	 * we just need to make racing faults a rare case.
> -	 *
> -	 * The implementation below would be much simpler if we just used a
> -	 * standard mutex or completion: but we cannot take i_rwsem in fault,
> -	 * and bloating every shmem inode for this unlikely case would be sad.
> -	 */
> -	if (unlikely(inode->i_private)) {
> -		struct shmem_falloc *shmem_falloc;
> -
> -		spin_lock(&inode->i_lock);
> -		shmem_falloc = inode->i_private;
> -		if (shmem_falloc &&
> -		    shmem_falloc->waitq &&
> -		    vmf->pgoff >= shmem_falloc->start &&
> -		    vmf->pgoff < shmem_falloc->next) {
> -			struct file *fpin;
> -			wait_queue_head_t *shmem_falloc_waitq;
> -			DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
> -
> -			ret = VM_FAULT_NOPAGE;
> -			fpin = maybe_unlock_mmap_for_io(vmf, NULL);
> -			if (fpin)
> -				ret = VM_FAULT_RETRY;
> -
> -			shmem_falloc_waitq = shmem_falloc->waitq;
> -			prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
> -					TASK_UNINTERRUPTIBLE);
> -			spin_unlock(&inode->i_lock);
> -			schedule();
> -
> -			/*
> -			 * shmem_falloc_waitq points into the shmem_fallocate()
> -			 * stack of the hole-punching task: shmem_falloc_waitq
> -			 * is usually invalid by the time we reach here, but
> -			 * finish_wait() does not dereference it in that case;
> -			 * though i_lock needed lest racing with wake_up_all().
> -			 */
> -			spin_lock(&inode->i_lock);
> -			finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
> -			spin_unlock(&inode->i_lock);
> -
> -			if (fpin)
> -				fput(fpin);
> -			return ret;
> -		}
> -		spin_unlock(&inode->i_lock);
> -	}
> -
>  	sgp = SGP_CACHE;
>  
>  	if ((vma->vm_flags & VM_NOHUGEPAGE) ||
> @@ -2113,8 +2040,10 @@ static vm_fault_t shmem_fault(struct vm_fault *vmf)
>  	else if (vma->vm_flags & VM_HUGEPAGE)
>  		sgp = SGP_HUGE;
>  
> +	down_read(&inode->i_mapping->invalidate_lock);
>  	err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
>  				  gfp, vma, vmf, &ret);
> +	up_read(&inode->i_mapping->invalidate_lock);
>  	if (err)
>  		return vmf_error(err);
>  	return ret;
> @@ -2715,7 +2644,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  		struct address_space *mapping = file->f_mapping;
>  		loff_t unmap_start = round_up(offset, PAGE_SIZE);
>  		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
> -		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
>  
>  		/* protected by i_rwsem */
>  		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
> @@ -2723,24 +2651,13 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  			goto out;
>  		}
>  
> -		shmem_falloc.waitq = &shmem_falloc_waitq;
> -		shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
> -		shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
> -		spin_lock(&inode->i_lock);
> -		inode->i_private = &shmem_falloc;
> -		spin_unlock(&inode->i_lock);
> -
> +		down_write(&mapping->invalidate_lock);
>  		if ((u64)unmap_end > (u64)unmap_start)
>  			unmap_mapping_range(mapping, unmap_start,
>  					    1 + unmap_end - unmap_start, 0);
>  		shmem_truncate_range(inode, offset, offset + len - 1);
>  		/* No need to unmap again: hole-punching leaves COWed pages */
> -
> -		spin_lock(&inode->i_lock);
> -		inode->i_private = NULL;
> -		wake_up_all(&shmem_falloc_waitq);
> -		WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
> -		spin_unlock(&inode->i_lock);
> +		up_write(&mapping->invalidate_lock);
>  		error = 0;
>  		goto out;
>  	}
> @@ -2763,7 +2680,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
>  		goto out;
>  	}
>  
> -	shmem_falloc.waitq = NULL;
>  	shmem_falloc.start = start;
>  	shmem_falloc.next  = start;
>  	shmem_falloc.nr_falloced = 0;
> -- 
> 2.26.2

Re: [PATCH 10/12] shmem: Use invalidate_lock to protect fallocate

[Jan Kara]

On Wed 28-04-21 20:24:59, Hugh Dickins wrote:
> On Fri, 23 Apr 2021, Jan Kara wrote:
> 
> > We have to handle pages added by currently running shmem_fallocate()
> > specially in shmem_writepage(). For this we use serialization mechanism
> > using structure attached to inode->i_private field. If we protect
> > allocation of pages in shmem_fallocate() with invalidate_lock instead,
> > we are sure added pages cannot be dirtied until shmem_fallocate() is done
> > (invalidate_lock blocks faults, i_rwsem blocks writes) and thus we
> > cannot see those pages in shmem_writepage() and there's no need for the
> > serialization mechanism.
> 
> Appealing diffstat, but NAK, this patch is based on a false premise.
> 
> All those pages that are added by shmem_fallocate() are marked dirty:
> see the set_page_dirty(page) and comment above it in shmem_fallocate().

Aha, I missed that set_page_dirty(). Thanks for correcting me.

> It's intentional, that they should percolate through to shmem_writepage()
> when memory is tight, and give feedback to fallocate that it should stop
> (instead of getting into that horrid OOM-kill flurry that never even
> frees any tmpfs anyway).

I understand the reason, I just feel there should be a better way of
communicating memory pressure than ->writepage talking to ->fallocate
through structure attached to inode->i_private. But now I see it isn't as
simple as I thought ;) Anyway for now I'll just remove this patch. Thanks
for review!

								Honza

> > CC: Hugh Dickins <hughd@google.com>
> > CC: <linux-mm@kvack.org>
> > Signed-off-by: Jan Kara <jack@suse.cz>
> > ---
> >  mm/shmem.c | 61 ++++++------------------------------------------------
> >  1 file changed, 6 insertions(+), 55 deletions(-)
> > 
> > diff --git a/mm/shmem.c b/mm/shmem.c
> > index f34162ac46de..7a2b0744031e 100644
> > --- a/mm/shmem.c
> > +++ b/mm/shmem.c
> > @@ -94,18 +94,6 @@ static struct vfsmount *shm_mnt;
> >  /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
> >  #define SHORT_SYMLINK_LEN 128
> >  
> > -/*
> > - * shmem_fallocate communicates with shmem_writepage via inode->i_private (with
> > - * i_rwsem making sure that it has only one user at a time): we would prefer
> > - * not to enlarge the shmem inode just for that.
> > - */
> > -struct shmem_falloc {
> > -	pgoff_t start;		/* start of range currently being fallocated */
> > -	pgoff_t next;		/* the next page offset to be fallocated */
> > -	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
> > -	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
> > -};
> > -
> >  struct shmem_options {
> >  	unsigned long long blocks;
> >  	unsigned long long inodes;
> > @@ -1364,28 +1352,11 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc)
> >  	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
> >  	 * value into swapfile.c, the only way we can correctly account for a
> >  	 * fallocated page arriving here is now to initialize it and write it.
> > -	 *
> > -	 * That's okay for a page already fallocated earlier, but if we have
> > -	 * not yet completed the fallocation, then (a) we want to keep track
> > -	 * of this page in case we have to undo it, and (b) it may not be a
> > -	 * good idea to continue anyway, once we're pushing into swap.  So
> > -	 * reactivate the page, and let shmem_fallocate() quit when too many.
> 
> (b) there commenting on communicating back to fallocate by nr_unswapped.
> 
> > +	 * Since a page added by currently running fallocate call cannot be
> > +	 * dirtied and thus arrive here we know the fallocate has already
> > +	 * completed and we are fine writing it out.
> >  	 */
> >  	if (!PageUptodate(page)) {
> > -		if (inode->i_private) {
> > -			struct shmem_falloc *shmem_falloc;
> > -			spin_lock(&inode->i_lock);
> > -			shmem_falloc = inode->i_private;
> > -			if (shmem_falloc &&
> > -			    index >= shmem_falloc->start &&
> > -			    index < shmem_falloc->next)
> > -				shmem_falloc->nr_unswapped++;
> > -			else
> > -				shmem_falloc = NULL;
> > -			spin_unlock(&inode->i_lock);
> > -			if (shmem_falloc)
> > -				goto redirty;
> > -		}
> >  		clear_highpage(page);
> >  		flush_dcache_page(page);
> >  		SetPageUptodate(page);
> > @@ -2629,9 +2600,9 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  							 loff_t len)
> >  {
> >  	struct inode *inode = file_inode(file);
> > +	struct address_space *mapping = file->f_mapping;
> >  	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;
> >  	int error;
> >  
> > @@ -2641,7 +2612,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  	inode_lock(inode);
> >  
> >  	if (mode & FALLOC_FL_PUNCH_HOLE) {
> > -		struct address_space *mapping = file->f_mapping;
> >  		loff_t unmap_start = round_up(offset, PAGE_SIZE);
> >  		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
> >  
> > @@ -2680,14 +2650,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  		goto out;
> >  	}
> >  
> > -	shmem_falloc.start = start;
> > -	shmem_falloc.next  = start;
> > -	shmem_falloc.nr_falloced = 0;
> > -	shmem_falloc.nr_unswapped = 0;
> > -	spin_lock(&inode->i_lock);
> > -	inode->i_private = &shmem_falloc;
> > -	spin_unlock(&inode->i_lock);
> > -
> > +	down_write(&mapping->invalidate_lock);
> >  	for (index = start; index < end; index++) {
> >  		struct page *page;
> >  
> > @@ -2697,8 +2660,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  		 */
> >  		if (signal_pending(current))
> >  			error = -EINTR;
> > -		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
> > -			error = -ENOMEM;
> >  		else
> >  			error = shmem_getpage(inode, index, &page, SGP_FALLOC);
> >  		if (error) {
> > @@ -2711,14 +2672,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  			goto undone;
> >  		}
> >  
> > -		/*
> > -		 * Inform shmem_writepage() how far we have reached.
> > -		 * No need for lock or barrier: we have the page lock.
> > -		 */
> > -		shmem_falloc.next++;
> > -		if (!PageUptodate(page))
> > -			shmem_falloc.nr_falloced++;
> > -
> >  		/*
> >  		 * If !PageUptodate, leave it that way so that freeable pages
> >  		 * can be recognized if we need to rollback on error later.
> 
> Which goes on to say:
> 
> 		 * But set_page_dirty so that memory pressure will swap rather
> 		 * than free the pages we are allocating (and SGP_CACHE pages
> 		 * might still be clean: we now need to mark those dirty too).
> 		 */
> 		set_page_dirty(page);
> 		unlock_page(page);
> 		put_page(page);
> 		cond_resched();
> 
> > @@ -2736,9 +2689,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  		i_size_write(inode, offset + len);
> >  	inode->i_ctime = current_time(inode);
> >  undone:
> > -	spin_lock(&inode->i_lock);
> > -	inode->i_private = NULL;
> > -	spin_unlock(&inode->i_lock);
> > +	up_write(&mapping->invalidate_lock);
> >  out:
> >  	inode_unlock(inode);
> >  	return error;
> > -- 
> > 2.26.2
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH 09/12] shmem: Convert to using invalidate_lock

[Jan Kara]

On Wed 28-04-21 21:12:36, Hugh Dickins wrote:
> On Fri, 23 Apr 2021, Jan Kara wrote:
> 
> > Shmem uses a home-grown mechanism for serializing hole punch with page
> > fault. Use mapping->invalidate_lock for it instead. Admittedly the
> > home-grown mechanism locks out only the range being actually punched out
> > while invalidate_lock locks the whole mapping so it is serializing more.
> > But hole punch doesn't seem to be that critical operation and the
> > simplification is noticeable.
> 
> Home-grown indeed (and went through several different bugginesses,
> Linus fixing issues in its waitq handling found years later).
> 
> I'd love to remove it all (rather than replace it by a new rwsem),
> but never enough courage+time to do so: on optimistic days (that is,
> rarely) I like to think that none of it would be needed nowadays;
> but its gestation was difficult, and I cannot easily reproduce the
> testing that demanded it (Sasha and Vlastimil helped a lot).
> 
> If you're interested in the history, I cannot point to one thread,
> but "shmem: fix faulting into a hole while it's punched" finds
> some of them, June/July 2014.  You've pushed me into re-reading
> there, but I've not yet found the crucial evidence that stopped us
> from reverting this mechanism, once we had abandoned the hole-punch
> "pincer" in shmem_undo_range().
> 
> tmpfs's problem with faulting versus hole-punch was not the data
> integrity issue you are attacking with invalidate_lock, but a
> starvation issue triggered in Trinity fuzzing.
> 
> If invalidate_lock had existed at the time, I might have reused it
> for this purpose too - I certainly wanted to avoid enlarging the
> inode with another rwsem just for this; but also reluctant to add
> another layer of locking to the common path (maybe I'm just silly
> to try to avoid an rwsem which is so rarely taken for writing?).
> 
> But the code as it stands is working satisfactorily with minimal
> overhead: so I'm not in a rush to remove or replace it yet. Thank
> you for including tmpfs in your reach, but I think for the moment
> I'd prefer you to leave this change out of the series. Maybe later
> when it's settled in the fs/ filesystems (perhaps making guarantees
> that we might want to extend to tmpfs) we could make this change -
> but I'd still rather let hole-punch and fault race freely without it.

OK, I'll remove the patch from the series for now. As you say, tmpfs is not
buggy so we can postpone the cleanup for later.

> But your 01/12, fixing mm comments mentioning i_mutex, looked good:
> Acked-by: Hugh Dickins <hughd@google.com>
> to that one.  But I think it would be better extracted from this
> invalidate_lock series, and just sent to akpm cc linux-mm on its own.

Thanks for review and yes, I guess I can send that patch to Andrew earlier.

								Honza

> > 
> > CC: Hugh Dickins <hughd@google.com>
> > CC: <linux-mm@kvack.org>
> > Signed-off-by: Jan Kara <jack@suse.cz>
> > ---
> >  mm/shmem.c | 98 ++++--------------------------------------------------
> >  1 file changed, 7 insertions(+), 91 deletions(-)
> > 
> > diff --git a/mm/shmem.c b/mm/shmem.c
> > index 55b2888db542..f34162ac46de 100644
> > --- a/mm/shmem.c
> > +++ b/mm/shmem.c
> > @@ -95,12 +95,11 @@ static struct vfsmount *shm_mnt;
> >  #define SHORT_SYMLINK_LEN 128
> >  
> >  /*
> > - * shmem_fallocate communicates with shmem_fault or shmem_writepage via
> > - * inode->i_private (with i_rwsem making sure that it has only one user at
> > - * a time): we would prefer not to enlarge the shmem inode just for that.
> > + * shmem_fallocate communicates with shmem_writepage via inode->i_private (with
> > + * i_rwsem making sure that it has only one user at a time): we would prefer
> > + * not to enlarge the shmem inode just for that.
> >   */
> >  struct shmem_falloc {
> > -	wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
> >  	pgoff_t start;		/* start of range currently being fallocated */
> >  	pgoff_t next;		/* the next page offset to be fallocated */
> >  	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
> > @@ -1378,7 +1377,6 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc)
> >  			spin_lock(&inode->i_lock);
> >  			shmem_falloc = inode->i_private;
> >  			if (shmem_falloc &&
> > -			    !shmem_falloc->waitq &&
> >  			    index >= shmem_falloc->start &&
> >  			    index < shmem_falloc->next)
> >  				shmem_falloc->nr_unswapped++;
> > @@ -2025,18 +2023,6 @@ static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
> >  	return error;
> >  }
> >  
> > -/*
> > - * This is like autoremove_wake_function, but it removes the wait queue
> > - * entry unconditionally - even if something else had already woken the
> > - * target.
> > - */
> > -static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
> > -{
> > -	int ret = default_wake_function(wait, mode, sync, key);
> > -	list_del_init(&wait->entry);
> > -	return ret;
> > -}
> > -
> >  static vm_fault_t shmem_fault(struct vm_fault *vmf)
> >  {
> >  	struct vm_area_struct *vma = vmf->vma;
> > @@ -2046,65 +2032,6 @@ static vm_fault_t shmem_fault(struct vm_fault *vmf)
> >  	int err;
> >  	vm_fault_t ret = VM_FAULT_LOCKED;
> >  
> > -	/*
> > -	 * Trinity finds that probing a hole which tmpfs is punching can
> > -	 * prevent the hole-punch from ever completing: which in turn
> > -	 * locks writers out with its hold on i_rwsem.  So refrain from
> > -	 * faulting pages into the hole while it's being punched.  Although
> > -	 * shmem_undo_range() does remove the additions, it may be unable to
> > -	 * keep up, as each new page needs its own unmap_mapping_range() call,
> > -	 * and the i_mmap tree grows ever slower to scan if new vmas are added.
> > -	 *
> > -	 * It does not matter if we sometimes reach this check just before the
> > -	 * hole-punch begins, so that one fault then races with the punch:
> > -	 * we just need to make racing faults a rare case.
> > -	 *
> > -	 * The implementation below would be much simpler if we just used a
> > -	 * standard mutex or completion: but we cannot take i_rwsem in fault,
> > -	 * and bloating every shmem inode for this unlikely case would be sad.
> > -	 */
> > -	if (unlikely(inode->i_private)) {
> > -		struct shmem_falloc *shmem_falloc;
> > -
> > -		spin_lock(&inode->i_lock);
> > -		shmem_falloc = inode->i_private;
> > -		if (shmem_falloc &&
> > -		    shmem_falloc->waitq &&
> > -		    vmf->pgoff >= shmem_falloc->start &&
> > -		    vmf->pgoff < shmem_falloc->next) {
> > -			struct file *fpin;
> > -			wait_queue_head_t *shmem_falloc_waitq;
> > -			DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
> > -
> > -			ret = VM_FAULT_NOPAGE;
> > -			fpin = maybe_unlock_mmap_for_io(vmf, NULL);
> > -			if (fpin)
> > -				ret = VM_FAULT_RETRY;
> > -
> > -			shmem_falloc_waitq = shmem_falloc->waitq;
> > -			prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
> > -					TASK_UNINTERRUPTIBLE);
> > -			spin_unlock(&inode->i_lock);
> > -			schedule();
> > -
> > -			/*
> > -			 * shmem_falloc_waitq points into the shmem_fallocate()
> > -			 * stack of the hole-punching task: shmem_falloc_waitq
> > -			 * is usually invalid by the time we reach here, but
> > -			 * finish_wait() does not dereference it in that case;
> > -			 * though i_lock needed lest racing with wake_up_all().
> > -			 */
> > -			spin_lock(&inode->i_lock);
> > -			finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
> > -			spin_unlock(&inode->i_lock);
> > -
> > -			if (fpin)
> > -				fput(fpin);
> > -			return ret;
> > -		}
> > -		spin_unlock(&inode->i_lock);
> > -	}
> > -
> >  	sgp = SGP_CACHE;
> >  
> >  	if ((vma->vm_flags & VM_NOHUGEPAGE) ||
> > @@ -2113,8 +2040,10 @@ static vm_fault_t shmem_fault(struct vm_fault *vmf)
> >  	else if (vma->vm_flags & VM_HUGEPAGE)
> >  		sgp = SGP_HUGE;
> >  
> > +	down_read(&inode->i_mapping->invalidate_lock);
> >  	err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
> >  				  gfp, vma, vmf, &ret);
> > +	up_read(&inode->i_mapping->invalidate_lock);
> >  	if (err)
> >  		return vmf_error(err);
> >  	return ret;
> > @@ -2715,7 +2644,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  		struct address_space *mapping = file->f_mapping;
> >  		loff_t unmap_start = round_up(offset, PAGE_SIZE);
> >  		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
> > -		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
> >  
> >  		/* protected by i_rwsem */
> >  		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
> > @@ -2723,24 +2651,13 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  			goto out;
> >  		}
> >  
> > -		shmem_falloc.waitq = &shmem_falloc_waitq;
> > -		shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
> > -		shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
> > -		spin_lock(&inode->i_lock);
> > -		inode->i_private = &shmem_falloc;
> > -		spin_unlock(&inode->i_lock);
> > -
> > +		down_write(&mapping->invalidate_lock);
> >  		if ((u64)unmap_end > (u64)unmap_start)
> >  			unmap_mapping_range(mapping, unmap_start,
> >  					    1 + unmap_end - unmap_start, 0);
> >  		shmem_truncate_range(inode, offset, offset + len - 1);
> >  		/* No need to unmap again: hole-punching leaves COWed pages */
> > -
> > -		spin_lock(&inode->i_lock);
> > -		inode->i_private = NULL;
> > -		wake_up_all(&shmem_falloc_waitq);
> > -		WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
> > -		spin_unlock(&inode->i_lock);
> > +		up_write(&mapping->invalidate_lock);
> >  		error = 0;
> >  		goto out;
> >  	}
> > @@ -2763,7 +2680,6 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
> >  		goto out;
> >  	}
> >  
> > -	shmem_falloc.waitq = NULL;
> >  	shmem_falloc.start = start;
> >  	shmem_falloc.next  = start;
> >  	shmem_falloc.nr_falloced = 0;
> > -- 
> > 2.26.2
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR