Skip to contentSkip to conversation

ntfs: trim the sizes after a short write

PATCH In review Zhan Xusheng 2 participants 2 messages Nov 20

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v8 1/1 ntfs: trim the sizes after a short write

From: Zhan Xusheng <[email protected]>

ntfs_file_write_iter() moves the data size and, for a non-resident uncompressed file, the initialized size to pos + count before the copy. __ntfs_extend_initialized_size() only zeroes [old initialized size, pos), so the range [pos, pos + count) is initialized only by the copy itself. When iomap_file_buffered_write() copies less than count, it returns a positive short count and the error rollback at out: does not run. The file keeps the size pos + count, and [pos + ret, pos + count) reads back whatever the freed clusters held before, also after a remount. A local user can read the old contents of freed clusters this way.

Run the existing rollback for a short write as well, and roll the data size back to the end of what was written, keeping the clusters that were allocated on disk before the write, such as space reserved with FALLOC_FL_KEEP_SIZE. Keep the initialized size, as a page fault through a shared mapping may rely on it while the write is stalled on its source, and zero what the write did not reach, up to the end of the block holding pos + count. Only if that fails, take the initialized size back to where the zeroing started, if it is above that.

Fixes b041ca562526 ("ntfs: update iomap and address space operations")

---
v8: when taking the data size back without truncating, mark the file name dirty as ntfs_non_resident_attr_shrink() does. If the inode was written while the write waited on its source, the size in the directory index otherwise stays at the extended value (Baolin).
v7: when taking the data size back, keep the clusters allocated on disk before the write instead of truncating, which freed space reserved with FALLOC_FL_KEEP_SIZE; a failed write already did so before this patch (Hyunchul). Holes and delayed allocations past the data size, such as the preallocation of an earlier extending write, are freed as before. Baolin's Reviewed-by on v6 is not carried over, as this changes the data size rollback he reviewed.
v6: cap that fallback at the current initialized size, which a failed extension before the write can leave below pos (Baolin).
v5 (Baolin):
- start the zeroing at pos, not at the old initialized size, which overwrote data stored through a shared mapping below pos
- do the same for a direct write, instead of truncating the page cache and taking the initialized size back
- if the zeroing fails, take the initialized size back to where it started, instead of skipping both rollbacks
v4: for a buffered write, zero the unwritten range instead of truncating the page cache and taking the initialized size back, which dropped data stored through a shared mapping during the copy (Baolin).
v3: clear the cache up to EOF, not up to the extended initialized size, which left the rest of the block holding pos + count (Baolin).
v2: also clear the cached tail of the last block, which v1 left in place when the write stayed inside the old data size (Baolin).

v7: https://lore.kernel.org/all/[email protected]/
v6: https://lore.kernel.org/all/[email protected]/
v5: https://lore.kernel.org/all/[email protected]/
v4: https://lore.kernel.org/all/[email protected]/
v3: https://lore.kernel.org/all/[email protected]/
v2: https://lore.kernel.org/all/[email protected]/
v1: https://lore.kernel.org/all/[email protected]/

Tested in a VM on ntfs-next 4862a81576e3, with all free space filled with a known pattern first. Short buffered writes of 4M, old data readable after drop_caches and after remount, 3 runs each:

                                    unfixed    v8
  fallocated, ends block-aligned    3137536    0
  fallocated, ends inside a block   3141332    0
  fallocated, unaligned start       3137536    0
  same, pos + count in that page    -          0
  append                            3137536    0

Marker stored through a shared mapping at 64K while pwritev() is stalled on a userfaultfd source page, state after remount, with and without fallocate():

                                          unfixed  v8
  buffered 4K at 128K, EFAULT             lost     kept
  O_DIRECT 12K at 0, EFAULT               lost     kept
  buffered 12K at 0, short 8192           kept     kept

1M reserved with FALLOC_FL_KEEP_SIZE, then a 12K write at 0, 512-byte blocks after remount:

                                          unfixed  v8
  short, 8192 written                     2048     2048
  failed, EFAULT                          0        2048

A short write past 8K reserved this way frees what it allocated beyond the 8K and keeps the 8K. A store into [pos + ret, pos + count) itself during the write is zeroed, as that range belongs to the write. With iomap_zero_range() failing (injected -ENOMEM), a 32K write to a 16K file that returns 12288 leaves the size at 28K with no old data, also after remount.

In the 8192 case above, with syncfs() while the write waits on its source, the data size in the directory index after remount is 12288 with v7 and 8192 with v8.

fsx (buffered, no-mmap, O_DIRECT and mixed), compressed files, fallocate, insert/collapse range and punch hole give the same results with v8 as with v7.

---
 fs/ntfs/attrib.c |  37 +++++++++++++++
 fs/ntfs/attrib.h |   1 +
 fs/ntfs/file.c   | 121 +++++++++++++++++++++++++++++++++++++++++++----
 3 files changed, 151 insertions(+), 8 deletions(-)
fs/ntfs/attrib.c +37 -0
index 29eae1eb3d02..ccaeb484aaa0 100644
@@ -2456,6 +2456,43 @@ int ntfs_attr_set_initialized_size(struct ntfs_inode *ni, loff_t new_size)
 	return err;
 }
 
+/*
+ * ntfs_attr_set_data_size - set the data size of a non-resident attribute
+ * without changing its allocation
+ *
+ * The caller holds mrec_lock.  @new_size must not be below the initialized
+ * size or above the allocated size.
+ */
+int ntfs_attr_set_data_size(struct ntfs_inode *ni, loff_t new_size)
+{
+	struct ntfs_attr_search_ctx *ctx;
+	struct ntfs_inode *base_ni = ntfs_base_inode(ni);
+	int err = 0;
+
+	if (!NInoNonResident(ni) || new_size < ni->initialized_size ||
+	    new_size > ni->allocated_size)
+		return -EINVAL;
+
+	ctx = ntfs_attr_get_search_ctx(base_ni, NULL);
+	if (!ctx)
+		return -ENOMEM;
+
+	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
+			       CASE_SENSITIVE, 0, NULL, 0, ctx);
+	if (err)
+		goto out_ctx;
+
+	ctx->attr->data.non_resident.data_size = cpu_to_le64(new_size);
+	ni->data_size = new_size;
+	/* Update data size in the index. */
+	if (ni->type == AT_DATA && ni->name == AT_UNNAMED)
+		NInoSetFileNameDirty(ni);
+	mark_mft_record_dirty(ctx->ntfs_ino);
+out_ctx:
+	ntfs_attr_put_search_ctx(ctx);
+	return err;
+}
+
 /*
  * ntfs_make_room_for_attr - make room for an attribute inside an mft record
  * @m:		mft record
fs/ntfs/attrib.h +1 -0
index bee91c9f6f81..edb62826a6ab 100644
@@ -102,6 +102,7 @@ int ntfs_attr_make_non_resident(struct ntfs_inode *ni, const u32 data_size);
 int ntfs_attr_set(struct ntfs_inode *ni, const s64 ofs, const s64 cnt,
 		const u8 val);
 int ntfs_attr_set_initialized_size(struct ntfs_inode *ni, loff_t new_size);
+int ntfs_attr_set_data_size(struct ntfs_inode *ni, loff_t new_size);
 int ntfs_attr_open(struct ntfs_inode *ni, const __le32 type,
 		__le16 *name, u32 name_len);
 void ntfs_attr_close(struct ntfs_inode *n);
fs/ntfs/file.c +113 -8

169 lines not on this page · raw message

base-commit: 4862a81576e3b20ba40cb94a38c5df0450047ffe

Signature · 1 line
2.43.0
Reviewed-by Baolin Liu on v8 1/1
#1 lore raw
Quoted 348 lines · Zhan Xusheng

From: Zhan Xusheng <[email protected]>

ntfs_file_write_iter() moves the data size and, for a non-resident uncompressed file, the initialized size to pos + count before the copy. __ntfs_extend_initialized_size() only zeroes [old initialized size, pos), so the range [pos, pos + count) is initialized only by the copy itself. When iomap_file_buffered_write() copies less than count, it returns a positive short count and the error rollback at out: does not run. The file keeps the size pos + count, and [pos + ret, pos + count) reads back whatever the freed clusters held before, also after a remount. A local user can read the old contents of freed clusters this way.

Run the existing rollback for a short write as well, and roll the data size back to the end of what was written, keeping the clusters that were allocated on disk before the write, such as space reserved with FALLOC_FL_KEEP_SIZE. Keep the initialized size, as a page fault through a shared mapping may rely on it while the write is stalled on its source, and zero what the write did not reach, up to the end of the block holding pos + count. Only if that fails, take the initialized size back to where the zeroing started, if it is above that.

Fixes b041ca562526 ("ntfs: update iomap and address space operations")

---
v8: when taking the data size back without truncating, mark the file name dirty as ntfs_non_resident_attr_shrink() does. If the inode was written while the write waited on its source, the size in the directory index otherwise stays at the extended value (Baolin).
v7: when taking the data size back, keep the clusters allocated on disk before the write instead of truncating, which freed space reserved with FALLOC_FL_KEEP_SIZE; a failed write already did so before this patch (Hyunchul). Holes and delayed allocations past the data size, such as the preallocation of an earlier extending write, are freed as before. Baolin's Reviewed-by on v6 is not carried over, as this changes the data size rollback he reviewed.
v6: cap that fallback at the current initialized size, which a failed extension before the write can leave below pos (Baolin).
v5 (Baolin):
- start the zeroing at pos, not at the old initialized size, which overwrote data stored through a shared mapping below pos
- do the same for a direct write, instead of truncating the page cache and taking the initialized size back
- if the zeroing fails, take the initialized size back to where it started, instead of skipping both rollbacks
v4: for a buffered write, zero the unwritten range instead of truncating the page cache and taking the initialized size back, which dropped data stored through a shared mapping during the copy (Baolin).
v3: clear the cache up to EOF, not up to the extended initialized size, which left the rest of the block holding pos + count (Baolin).
v2: also clear the cached tail of the last block, which v1 left in place when the write stayed inside the old data size (Baolin).

v7: https://lore.kernel.org/all/[email protected]/
v6: https://lore.kernel.org/all/[email protected]/
v5: https://lore.kernel.org/all/[email protected]/
v4: https://lore.kernel.org/all/[email protected]/
v3: https://lore.kernel.org/all/[email protected]/
v2: https://lore.kernel.org/all/[email protected]/
v1: https://lore.kernel.org/all/[email protected]/

Tested in a VM on ntfs-next 4862a81576e3, with all free space filled with a known pattern first. Short buffered writes of 4M, old data readable after drop_caches and after remount, 3 runs each:

                                     unfixed    v8
   fallocated, ends block-aligned    3137536    0
   fallocated, ends inside a block   3141332    0
   fallocated, unaligned start       3137536    0
   same, pos + count in that page    -          0
   append                            3137536    0

Marker stored through a shared mapping at 64K while pwritev() is stalled on a userfaultfd source page, state after remount, with and without fallocate():

                                           unfixed  v8
   buffered 4K at 128K, EFAULT             lost     kept
   O_DIRECT 12K at 0, EFAULT               lost     kept
   buffered 12K at 0, short 8192           kept     kept

1M reserved with FALLOC_FL_KEEP_SIZE, then a 12K write at 0, 512-byte blocks after remount:

                                           unfixed  v8
   short, 8192 written                     2048     2048
   failed, EFAULT                          0        2048

A short write past 8K reserved this way frees what it allocated beyond the 8K and keeps the 8K. A store into [pos + ret, pos + count) itself during the write is zeroed, as that range belongs to the write. With iomap_zero_range() failing (injected -ENOMEM), a 32K write to a 16K file that returns 12288 leaves the size at 28K with no old data, also after remount.

In the 8192 case above, with syncfs() while the write waits on its source, the data size in the directory index after remount is 12288 with v7 and 8192 with v8.

fsx (buffered, no-mmap, O_DIRECT and mixed), compressed files, fallocate, insert/collapse range and punch hole give the same results with v8 as with v7.

---
  fs/ntfs/attrib.c |  37 +++++++++++++++
  fs/ntfs/attrib.h |   1 +
  fs/ntfs/file.c   | 121 +++++++++++++++++++++++++++++++++++++++++++----
  3 files changed, 151 insertions(+), 8 deletions(-)
fs/ntfs/attrib.c +37 -0
index 29eae1eb3d02..ccaeb484aaa0 100644
@@ -2456,6 +2456,43 @@ int ntfs_attr_set_initialized_size(struct ntfs_inode *ni, loff_t new_size)
  	return err;
  }
 
+/*
+ * ntfs_attr_set_data_size - set the data size of a non-resident attribute
+ * without changing its allocation
+ *
+ * The caller holds mrec_lock.  @new_size must not be below the initialized
+ * size or above the allocated size.
+ */
+int ntfs_attr_set_data_size(struct ntfs_inode *ni, loff_t new_size)
+{
+	struct ntfs_attr_search_ctx *ctx;
+	struct ntfs_inode *base_ni = ntfs_base_inode(ni);
+	int err = 0;
+
+	if (!NInoNonResident(ni) || new_size < ni->initialized_size ||
+	    new_size > ni->allocated_size)
+		return -EINVAL;
+
+	ctx = ntfs_attr_get_search_ctx(base_ni, NULL);
+	if (!ctx)
+		return -ENOMEM;
+
+	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
+			       CASE_SENSITIVE, 0, NULL, 0, ctx);
+	if (err)
+		goto out_ctx;
+
+	ctx->attr->data.non_resident.data_size = cpu_to_le64(new_size);
+	ni->data_size = new_size;
+	/* Update data size in the index. */
+	if (ni->type == AT_DATA && ni->name == AT_UNNAMED)
+		NInoSetFileNameDirty(ni);
+	mark_mft_record_dirty(ctx->ntfs_ino);
+out_ctx:
+	ntfs_attr_put_search_ctx(ctx);
+	return err;
+}
+
  /*
   * ntfs_make_room_for_attr - make room for an attribute inside an mft record
   * @m:		mft record
fs/ntfs/attrib.h +1 -0
index bee91c9f6f81..edb62826a6ab 100644
@@ -102,6 +102,7 @@ int ntfs_attr_make_non_resident(struct ntfs_inode *ni, const u32 data_size);
  int ntfs_attr_set(struct ntfs_inode *ni, const s64 ofs, const s64 cnt,
  		const u8 val);
  int ntfs_attr_set_initialized_size(struct ntfs_inode *ni, loff_t new_size);
+int ntfs_attr_set_data_size(struct ntfs_inode *ni, loff_t new_size);
  int ntfs_attr_open(struct ntfs_inode *ni, const __le32 type,
  		__le16 *name, u32 name_len);
  void ntfs_attr_close(struct ntfs_inode *n);
fs/ntfs/file.c +113 -8
index a52483d16c6e..25034735d249 100644
@@ -843,6 +843,34 @@ static int ntfs_expand_for_write(struct ntfs_inode *ni, loff_t end)
  	return err;
  }
 
+/*
+ * End of the last run of clusters allocated on disk.  Holes and delayed
+ * allocations past it, such as the preallocation of an extending write,
+ * do not count.  Falls back to the allocated size if the runlist is not
+ * fully mapped.
+ */
+static loff_t ntfs_allocated_end_on_disk(struct ntfs_inode *ni)
+{
+	loff_t end = ni->allocated_size;
+	ssize_t i;
+
+	down_read(&ni->runlist.lock);
+	if (NInoFullyMapped(ni) && ni->runlist.rl) {
+		end = 0;
+		for (i = ni->runlist.count - 1; i >= 0; i--) {
+			struct runlist_element *rl = &ni->runlist.rl[i];
+
+			if (rl->length && rl->lcn >= 0) {
+				end = ntfs_cluster_to_bytes(ni->vol,
+						rl->vcn + rl->length);
+				break;
+			}
+		}
+	}
+	up_read(&ni->runlist.lock);
+	return end;
+}
+
  ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  {
  	struct file *file = iocb->ki_filp;
@@ -853,7 +881,7 @@ ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  	ssize_t count;
  	loff_t pos, end;
  	int err;
-	loff_t old_data_size, old_init_size;
+	loff_t old_data_size, old_init_size, old_alloc_size;
 
  	if (NVolShutdown(vol))
  		return -EIO;
@@ -901,6 +929,10 @@ ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 
  	old_data_size = ni->data_size;
  	old_init_size = ni->initialized_size;
+	old_alloc_size = ni->allocated_size;
+	if (NInoNonResident(ni) && !NInoCompressed(ni) &&
+	    old_alloc_size > round_up(old_data_size, vol->cluster_size))
+		old_alloc_size = ntfs_allocated_end_on_disk(ni);
 
  	if (end > old_data_size) {
  		ret = ntfs_expand_for_write(ni, end);
@@ -928,10 +960,21 @@ ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  		ret = iomap_file_buffered_write(iocb, from, ntfs_write_iomap_next,
  				&ntfs_iomap_folio_ops, NULL);
  out:
-	if (ret < 0 && ret != -EIOCBQUEUED) {
+	/*
+	 * Both sizes were moved to end before the copy, so a short write,
+	 * like a failed one, must take back what it did not write.
+	 */
+	if (ret < count && ret != -EIOCBQUEUED) {
+		loff_t new_init_size = old_init_size;
+		loff_t new_data_size = old_data_size;
  		int init_err = 0;
  		int data_err = 0;
 
+		if (ret > 0) {
+			new_init_size = max(old_init_size, pos + ret);
+			new_data_size = max(old_data_size, pos + ret);
+		}
+
  		/*
  		 * Like ntfs_setattr_size(): __ntfs_attr_truncate_vfs()
  		 * expects the invalidate lock held across the size
@@ -939,14 +982,59 @@ ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  		 */
  		filemap_invalidate_lock(vi->i_mapping);
 
+		/*
+		 * A page fault through a shared mapping may have used the
+		 * extended initialized size while the write was stalled on
+		 * its source, so keep that size and zero what the write did
+		 * not reach instead.  [old_init_size, pos) was zeroed before
+		 * the write.  Past end, zero the rest of the block that a
+		 * buffered copy may have read from disk, unless such a fault
+		 * already moved the initialized size past end and zeroed it.
+		 */
+		if (NInoNonResident(ni) && !NInoCompressed(ni) &&
+		    ni->initialized_size > new_init_size) {
+			loff_t zstart = max(pos, new_init_size);
+			loff_t zend = ni->initialized_size > end ? end :
+				      round_up(end, i_blocksize(vi));
+			int err = 0;
+
+			zend = min(zend, new_data_size);
+			if (zend > zstart) {
+				err = iomap_zero_range(vi, zstart,
+						zend - zstart, NULL,
+						ntfs_seek_iomap_next,
+						&ntfs_iomap_folio_ops, NULL);
+				/* A direct read must not see the old blocks. */
+				if (!err && iocb->ki_flags & IOCB_DIRECT)
+					err = filemap_write_and_wait_range(
+						vi->i_mapping, zstart, zend - 1);
+			}
+			if (!err)
+				new_init_size = min(ni->initialized_size,
+						    new_data_size);
+			else
+				new_init_size = min3(ni->initialized_size,
+						     zstart, new_data_size);
+		}
+
  		/* ntfs_attr_set_initialized_size() rejects a resident one. */
-		if (ni->initialized_size != old_init_size &&
+		if (ni->initialized_size != new_init_size &&
  		    NInoNonResident(ni)) {
  			struct ntfs_inode *mrec_ni = ntfs_base_inode(ni);
 
+			/*
+			 * Only reached when the zeroing above failed or did
+			 * not apply.  Drop what was cached past the new
+			 * initialized size, as __ntfs_extend_initialized_size()
+			 * does.
+			 */
+			if (ni->initialized_size > new_init_size)
+				truncate_pagecache_range(vi, new_init_size,
+						max_t(loff_t, i_size_read(vi),
+						      ni->initialized_size) - 1);
  			mutex_lock(&mrec_ni->mrec_lock);
  			init_err = ntfs_attr_set_initialized_size(ni,
-					old_init_size);
+					new_init_size);
  			mutex_unlock(&mrec_ni->mrec_lock);
  		}
  		/*
@@ -955,13 +1043,30 @@ ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
  		 * the uninitialized tail, so skip it unless the rollback
  		 * above restored that size.
  		 */
-		if (ni->data_size != old_data_size && !init_err) {
+		if (ni->data_size != new_data_size && !init_err) {
  			struct ntfs_inode *mrec_ni = ntfs_base_inode(ni);
 
-			truncate_setsize(vi, old_data_size);
+			truncate_setsize(vi, new_data_size);
  			mutex_lock(&mrec_ni->mrec_lock);
-			data_err = __ntfs_attr_truncate_vfs(ni, old_data_size,
-					old_data_size);
+			if (NInoNonResident(ni) && !NInoCompressed(ni)) {
+				/*
+				 * Only free what the write allocated, not
+				 * clusters that were there before, such as
+				 * space reserved with FALLOC_FL_KEEP_SIZE.
+				 */
+				loff_t keep = max(new_data_size,
+						  old_alloc_size);
+
+				if (ni->allocated_size > keep)
+					data_err = __ntfs_attr_truncate_vfs(ni,
+							keep, keep);
+				if (!data_err && ni->data_size != new_data_size)
+					data_err = ntfs_attr_set_data_size(ni,
+							new_data_size);
+			} else {
+				data_err = __ntfs_attr_truncate_vfs(ni,
+						new_data_size, new_data_size);
+			}
  			mutex_unlock(&mrec_ni->mrec_lock);
  			/* Like ntfs_setattr_size(): stay on the durable size. */
  			if (data_err)
Quoted 1 line

base-commit: 4862a81576e3b20ba40cb94a38c5df0450047ffe

Looks good to me.