ocfs2: zero tail of sparse files on truncate

Since we don't zero on extend anymore, truncate needs to be fixed up to zero
the part of a file between i_size and and end of it's cluster. Otherwise a
subsequent extend could expose bad data.

This introduced a new helper, which can be used in ocfs2_write().

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

7 files changed, 328 insertions, 25 deletions

diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index 9a40603c4d4..98694a1add4 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -27,6 +27,7 @@
 #include <linux/types.h>
 #include <linux/slab.h>
 #include <linux/highmem.h>
+#include <linux/swap.h>
 
 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
 #include <cluster/masklog.h>
@@ -34,6 +35,7 @@
 #include "ocfs2.h"
 
 #include "alloc.h"
+#include "aops.h"
 #include "dlmglue.h"
 #include "extent_map.h"
 #include "inode.h"
@@ -3342,6 +3344,228 @@ bail:
 	return status;
 }
 
+static int ocfs2_writeback_zero_func(handle_t *handle, struct buffer_head *bh)
+{
+	set_buffer_uptodate(bh);
+	mark_buffer_dirty(bh);
+	return 0;
+}
+
+static int ocfs2_ordered_zero_func(handle_t *handle, struct buffer_head *bh)
+{
+	set_buffer_uptodate(bh);
+	mark_buffer_dirty(bh);
+	return ocfs2_journal_dirty_data(handle, bh);
+}
+
+static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t isize,
+				     struct page **pages, int numpages,
+				     u64 phys, handle_t *handle)
+{
+	int i, ret, partial = 0;
+	void *kaddr;
+	struct page *page;
+	unsigned int from, to = PAGE_CACHE_SIZE;
+	struct super_block *sb = inode->i_sb;
+
+	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
+
+	if (numpages == 0)
+		goto out;
+
+	from = isize & (PAGE_CACHE_SIZE - 1); /* 1st page offset */
+	if (PAGE_CACHE_SHIFT > OCFS2_SB(sb)->s_clustersize_bits) {
+		/*
+		 * Since 'from' has been capped to a value below page
+		 * size, this calculation won't be able to overflow
+		 * 'to'
+		 */
+		to = ocfs2_align_bytes_to_clusters(sb, from);
+
+		/*
+		 * The truncate tail in this case should never contain
+		 * more than one page at maximum. The loop below also
+		 * assumes this.
+		 */
+		BUG_ON(numpages != 1);
+	}
+
+	for(i = 0; i < numpages; i++) {
+		page = pages[i];
+
+		BUG_ON(from > PAGE_CACHE_SIZE);
+		BUG_ON(to > PAGE_CACHE_SIZE);
+
+		ret = ocfs2_map_page_blocks(page, &phys, inode, from, to, 0);
+		if (ret)
+			mlog_errno(ret);
+
+		kaddr = kmap_atomic(page, KM_USER0);
+		memset(kaddr + from, 0, to - from);
+		kunmap_atomic(kaddr, KM_USER0);
+
+		/*
+		 * Need to set the buffers we zero'd into uptodate
+		 * here if they aren't - ocfs2_map_page_blocks()
+		 * might've skipped some
+		 */
+		if (ocfs2_should_order_data(inode)) {
+			ret = walk_page_buffers(handle,
+						page_buffers(page),
+						from, to, &partial,
+						ocfs2_ordered_zero_func);
+			if (ret < 0)
+				mlog_errno(ret);
+		} else {
+			ret = walk_page_buffers(handle, page_buffers(page),
+						from, to, &partial,
+						ocfs2_writeback_zero_func);
+			if (ret < 0)
+				mlog_errno(ret);
+		}
+
+		if (!partial)
+			SetPageUptodate(page);
+
+		flush_dcache_page(page);
+
+		/*
+		 * Every page after the 1st one should be completely zero'd.
+		 */
+		from = 0;
+	}
+out:
+	if (pages) {
+		for (i = 0; i < numpages; i++) {
+			page = pages[i];
+			unlock_page(page);
+			mark_page_accessed(page);
+			page_cache_release(page);
+		}
+	}
+}
+
+static int ocfs2_grab_eof_pages(struct inode *inode, loff_t isize, struct page **pages,
+				int *num, u64 *phys)
+{
+	int i, numpages = 0, ret = 0;
+	unsigned int csize = OCFS2_SB(inode->i_sb)->s_clustersize;
+	struct super_block *sb = inode->i_sb;
+	struct address_space *mapping = inode->i_mapping;
+	unsigned long index;
+	u64 next_cluster_bytes;
+
+	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
+
+	/* Cluster boundary, so we don't need to grab any pages. */
+	if ((isize & (csize - 1)) == 0)
+		goto out;
+
+	ret = ocfs2_extent_map_get_blocks(inode, isize >> sb->s_blocksize_bits,
+					  phys, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Tail is a hole. */
+	if (*phys == 0)
+		goto out;
+
+	next_cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, isize);
+	index = isize >> PAGE_CACHE_SHIFT;
+	do {
+		pages[numpages] = grab_cache_page(mapping, index);
+		if (!pages[numpages]) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			goto out;
+		}
+
+		numpages++;
+		index++;
+	} while (index < (next_cluster_bytes >> PAGE_CACHE_SHIFT));
+
+out:
+	if (ret != 0) {
+		if (pages) {
+			for (i = 0; i < numpages; i++) {
+				if (pages[i]) {
+					unlock_page(pages[i]);
+					page_cache_release(pages[i]);
+				}
+			}
+		}
+		numpages = 0;
+	}
+
+	*num = numpages;
+
+	return ret;
+}
+
+/*
+ * Zero the area past i_size but still within an allocated
+ * cluster. This avoids exposing nonzero data on subsequent file
+ * extends.
+ *
+ * We need to call this before i_size is updated on the inode because
+ * otherwise block_write_full_page() will skip writeout of pages past
+ * i_size. The new_i_size parameter is passed for this reason.
+ */
+int ocfs2_zero_tail_for_truncate(struct inode *inode, handle_t *handle,
+				 u64 new_i_size)
+{
+	int ret, numpages;
+	struct page **pages = NULL;
+	u64 phys;
+
+	/*
+	 * File systems which don't support sparse files zero on every
+	 * extend.
+	 */
+	if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+		return 0;
+
+	pages = kcalloc(ocfs2_pages_per_cluster(inode->i_sb),
+			sizeof(struct page *), GFP_NOFS);
+	if (pages == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_grab_eof_pages(inode, new_i_size, pages, &numpages, &phys);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Truncate on an i_size boundary - nothing more to do.
+	 */
+	if (numpages == 0)
+		goto out;
+
+	ocfs2_zero_cluster_pages(inode, new_i_size, pages, numpages, phys,
+				 handle);
+
+	/*
+	 * Initiate writeout of the pages we zero'd here. We don't
+	 * wait on them - the truncate_inode_pages() call later will
+	 * do that for us.
+	 */
+	ret = filemap_fdatawrite(inode->i_mapping);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	if (pages)
+		kfree(pages);
+
+	return ret;
+}
+
 /*
  * It is expected, that by the time you call this function,
  * inode->i_size and fe->i_size have been adjusted.
diff --git a/fs/ocfs2/alloc.h b/fs/ocfs2/alloc.h
index bff2a162b03..3cb39cd5e47 100644
--- a/fs/ocfs2/alloc.h
+++ b/fs/ocfs2/alloc.h
@@ -71,6 +71,8 @@ struct ocfs2_truncate_context {
 	struct buffer_head *tc_last_eb_bh;
 };
 
+int ocfs2_zero_tail_for_truncate(struct inode *inode, handle_t *handle,
+				 u64 new_i_size);
 int ocfs2_prepare_truncate(struct ocfs2_super *osb,
 			   struct inode *inode,
 			   struct buffer_head *fe_bh,
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index acf8f000672..605c82a93f0 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -308,13 +308,13 @@ int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
  * functionality yet, but IMHO it's better to cut and paste the whole
  * thing so we can avoid introducing our own bugs (and easily pick up
  * their fixes when they happen) --Mark */
-static int walk_page_buffers(	handle_t *handle,
-				struct buffer_head *head,
-				unsigned from,
-				unsigned to,
-				int *partial,
-				int (*fn)(	handle_t *handle,
-						struct buffer_head *bh))
+int walk_page_buffers(	handle_t *handle,
+			struct buffer_head *head,
+			unsigned from,
+			unsigned to,
+			int *partial,
+			int (*fn)(	handle_t *handle,
+					struct buffer_head *bh))
 {
 	struct buffer_head *bh;
 	unsigned block_start, block_end;
@@ -654,9 +654,9 @@ static void ocfs2_clear_page_regions(struct page *page,
  *
  * This will also skip zeroing, which is handled externally.
  */
-static int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
-				 struct inode *inode, unsigned int from,
-				 unsigned int to, int new)
+int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
+			  struct inode *inode, unsigned int from,
+			  unsigned int to, int new)
 {
 	int ret = 0;
 	struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
@@ -675,8 +675,7 @@ static int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
 		 * Ignore blocks outside of our i/o range -
 		 * they may belong to unallocated clusters.
 		 */
-		if (block_start >= to ||
-		    (block_start + bsize) <= from) {
+		if (block_start >= to || block_end <= from) {
 			if (PageUptodate(page))
 				set_buffer_uptodate(bh);
 			continue;
@@ -971,7 +970,6 @@ static ssize_t ocfs2_write(struct file *file, u32 phys, handle_t *handle,
 	u64 v_blkno, p_blkno;
 	struct address_space *mapping = file->f_mapping;
 	struct inode *inode = mapping->host;
-	unsigned int cbits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
 	unsigned long index, start;
 	struct page **cpages;
 
@@ -979,13 +977,11 @@ static ssize_t ocfs2_write(struct file *file, u32 phys, handle_t *handle,
 
 	/*
 	 * Figure out how many pages we'll be manipulating here. For
-	 * non-allocating write, or any writes where cluster size is
-	 * less than page size, we only need one page. Otherwise,
-	 * allocating writes of cluster size larger than page size
-	 * need cluster size pages.
+	 * non allocating write, we just change the one
+	 * page. Otherwise, we'll need a whole clusters worth.
 	 */
-	if (new && !wc->w_large_pages)
-		numpages = (1 << cbits) / PAGE_SIZE;
+	if (new)
+		numpages = ocfs2_pages_per_cluster(inode->i_sb);
 
 	cpages = kzalloc(sizeof(*cpages) * numpages, GFP_NOFS);
 	if (!cpages) {
diff --git a/fs/ocfs2/aops.h b/fs/ocfs2/aops.h
index eeb2c42483e..7d94071f0ab 100644
--- a/fs/ocfs2/aops.h
+++ b/fs/ocfs2/aops.h
@@ -30,6 +30,18 @@ handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
 							 unsigned from,
 							 unsigned to);
 
+int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
+			  struct inode *inode, unsigned int from,
+			  unsigned int to, int new);
+
+int walk_page_buffers(	handle_t *handle,
+			struct buffer_head *head,
+			unsigned from,
+			unsigned to,
+			int *partial,
+			int (*fn)(	handle_t *handle,
+					struct buffer_head *bh));
+
 struct ocfs2_write_ctxt;
 typedef int (ocfs2_page_writer)(struct inode *, struct ocfs2_write_ctxt *,
 				u64 *, unsigned int *, unsigned int *);
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 667e5a869bf..5fd49ec169d 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -262,6 +262,7 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 {
 	int status;
 	handle_t *handle;
+	struct ocfs2_dinode *di;
 
 	mlog_entry_void();
 
@@ -275,12 +276,39 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 		goto out;
 	}
 
-	status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size);
+	status = ocfs2_journal_access(handle, inode, fe_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	/*
+	 * Do this before setting i_size.
+	 */
+	status = ocfs2_zero_tail_for_truncate(inode, handle, new_i_size);
+	if (status) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	i_size_write(inode, new_i_size);
+	inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
+	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+
+	di = (struct ocfs2_dinode *) fe_bh->b_data;
+	di->i_size = cpu_to_le64(new_i_size);
+	di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
+	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+
+	status = ocfs2_journal_dirty(handle, fe_bh);
 	if (status < 0)
 		mlog_errno(status);
 
+out_commit:
 	ocfs2_commit_trans(osb, handle);
 out:
+
 	mlog_exit(status);
 	return status;
 }
@@ -343,7 +371,6 @@ static int ocfs2_truncate_file(struct inode *inode,
 		mlog_errno(status);
 		goto bail;
 	}
-	ocfs2_data_unlock(inode, 1);
 
 	/* alright, we're going to need to do a full blown alloc size
 	 * change. Orphan the inode so that recovery can complete the
@@ -352,22 +379,25 @@ static int ocfs2_truncate_file(struct inode *inode,
 	status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
 	if (status < 0) {
 		mlog_errno(status);
-		goto bail;
+		goto bail_unlock_data;
 	}
 
 	status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
 	if (status < 0) {
 		mlog_errno(status);
-		goto bail;
+		goto bail_unlock_data;
 	}
 
 	status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
 	if (status < 0) {
 		mlog_errno(status);
-		goto bail;
+		goto bail_unlock_data;
 	}
 
 	/* TODO: orphan dir cleanup here. */
+bail_unlock_data:
+	ocfs2_data_unlock(inode, 1);
+
 bail:
 
 	mlog_exit(status);
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index 0bd86a13759..78c99b5050d 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -489,12 +489,38 @@ static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
 	int status = 0;
 	struct ocfs2_truncate_context *tc = NULL;
 	struct ocfs2_dinode *fe;
+	handle_t *handle = NULL;
 
 	mlog_entry_void();
 
 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
 
 	if (fe->i_clusters) {
+		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+		if (IS_ERR(handle)) {
+			status = PTR_ERR(handle);
+			mlog_errno(status);
+			goto out;
+		}
+
+		status = ocfs2_journal_access(handle, inode, fe_bh,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
+
+		i_size_write(inode, 0);
+
+		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
+
+		ocfs2_commit_trans(osb, handle);
+		handle = NULL;
+
 		status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
 		if (status < 0) {
 			mlog_errno(status);
@@ -507,8 +533,10 @@ static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
 			goto out;
 		}
 	}
-out:
 
+out:
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
 	mlog_exit(status);
 	return status;
 }
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
index 2699f7cac21..82cc92dcf8a 100644
--- a/fs/ocfs2/ocfs2.h
+++ b/fs/ocfs2/ocfs2.h
@@ -495,6 +495,17 @@ static inline unsigned long ocfs2_align_clusters_to_page_index(struct super_bloc
 	return index;
 }
 
+static inline unsigned int ocfs2_pages_per_cluster(struct super_block *sb)
+{
+	unsigned int cbits = OCFS2_SB(sb)->s_clustersize_bits;
+	unsigned int pages_per_cluster = 1;
+
+	if (PAGE_CACHE_SHIFT < cbits)
+		pages_per_cluster = 1 << (cbits - PAGE_CACHE_SHIFT);
+
+	return pages_per_cluster;
+}
+
 #define ocfs2_set_bit ext2_set_bit
 #define ocfs2_clear_bit ext2_clear_bit
 #define ocfs2_test_bit ext2_test_bit