Merge git://git.kernel.org/pub/scm/linux/kernel/git/joern/logfs

* git://git.kernel.org/pub/scm/linux/kernel/git/joern/logfs:
  [LogFS] Erase new journal segments
  [LogFS] Move reserved segments with journal
  [LogFS] Clear PagePrivate when moving journal
  Simplify and fix pad_wbuf
  Prevent data corruption in logfs_rewrite_block()
  Use deactivate_locked_super
  Fix logfs_get_sb_final error path
  Write out both superblocks on mismatch
  Prevent schedule while atomic in __logfs_readdir
  Plug memory leak in writeseg_end_io
  Limit max_pages for insane devices
  Open segment file before using it

fs/logfs/dev_bdev.c

@@ -80,6 +80,7 @@ static void writeseg_end_io(struct bio *bio, int err)
 			prefetchw(&bvec->bv_page->flags);
 
 		end_page_writeback(page);
+		page_cache_release(page);
 	} while (bvec >= bio->bi_io_vec);
 	bio_put(bio);
 	if (atomic_dec_and_test(&super->s_pending_writes))
@@ -97,8 +98,10 @@ static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
 	unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
 	int i;
 
+	if (max_pages > BIO_MAX_PAGES)
+		max_pages = BIO_MAX_PAGES;
 	bio = bio_alloc(GFP_NOFS, max_pages);
-	BUG_ON(!bio); /* FIXME: handle this */
+	BUG_ON(!bio);
 
 	for (i = 0; i < nr_pages; i++) {
 		if (i >= max_pages) {
@@ -191,8 +194,10 @@ static int do_erase(struct super_block *sb, u64 ofs, pgoff_t index,
 	unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
 	int i;
 
+	if (max_pages > BIO_MAX_PAGES)
+		max_pages = BIO_MAX_PAGES;
 	bio = bio_alloc(GFP_NOFS, max_pages);
-	BUG_ON(!bio); /* FIXME: handle this */
+	BUG_ON(!bio);
 
 	for (i = 0; i < nr_pages; i++) {
 		if (i >= max_pages) {

fs/logfs/dir.c

@@ -303,12 +303,12 @@ static int __logfs_readdir(struct file *file, void *buf, filldir_t filldir)
 				(filler_t *)logfs_readpage, NULL);
 		if (IS_ERR(page))
 			return PTR_ERR(page);
-		dd = kmap_atomic(page, KM_USER0);
+		dd = kmap(page);
 		BUG_ON(dd->namelen == 0);
 
 		full = filldir(buf, (char *)dd->name, be16_to_cpu(dd->namelen),
 				pos, be64_to_cpu(dd->ino), dd->type);
-		kunmap_atomic(dd, KM_USER0);
+		kunmap(page);
 		page_cache_release(page);
 		if (full)
 			break;

fs/logfs/journal.c

@@ -800,6 +800,7 @@ void do_logfs_journal_wl_pass(struct super_block *sb)
 {
 	struct logfs_super *super = logfs_super(sb);
 	struct logfs_area *area = super->s_journal_area;
+	struct btree_head32 *head = &super->s_reserved_segments;
 	u32 segno, ec;
 	int i, err;
 
@@ -807,6 +808,7 @@ void do_logfs_journal_wl_pass(struct super_block *sb)
 	/* Drop old segments */
 	journal_for_each(i)
 		if (super->s_journal_seg[i]) {
+			btree_remove32(head, super->s_journal_seg[i]);
 			logfs_set_segment_unreserved(sb,
 					super->s_journal_seg[i],
 					super->s_journal_ec[i]);
@@ -819,8 +821,13 @@ void do_logfs_journal_wl_pass(struct super_block *sb)
 		super->s_journal_seg[i] = segno;
 		super->s_journal_ec[i] = ec;
 		logfs_set_segment_reserved(sb, segno);
+		err = btree_insert32(head, segno, (void *)1, GFP_KERNEL);
+		BUG_ON(err); /* mempool should prevent this */
+		err = logfs_erase_segment(sb, segno, 1);
+		BUG_ON(err); /* FIXME: remount-ro would be nicer */
 	}
 	/* Manually move journal_area */
+	freeseg(sb, area->a_segno);
 	area->a_segno = super->s_journal_seg[0];
 	area->a_is_open = 0;
 	area->a_used_bytes = 0;

fs/logfs/logfs.h

@@ -587,6 +587,7 @@ void move_page_to_btree(struct page *page);
 int logfs_init_mapping(struct super_block *sb);
 void logfs_sync_area(struct logfs_area *area);
 void logfs_sync_segments(struct super_block *sb);
+void freeseg(struct super_block *sb, u32 segno);
 
 /* area handling */
 int logfs_init_areas(struct super_block *sb);

fs/logfs/readwrite.c

@@ -1594,7 +1594,6 @@ int logfs_delete(struct inode *inode, pgoff_t index,
 	return ret;
 }
 
-/* Rewrite cannot mark the inode dirty but has to write it immediatly. */
 int logfs_rewrite_block(struct inode *inode, u64 bix, u64 ofs,
 		gc_level_t gc_level, long flags)
 {
@@ -1611,6 +1610,18 @@ int logfs_rewrite_block(struct inode *inode, u64 bix, u64 ofs,
 		if (level != 0)
 			alloc_indirect_block(inode, page, 0);
 		err = logfs_write_buf(inode, page, flags);
+		if (!err && shrink_level(gc_level) == 0) {
+			/* Rewrite cannot mark the inode dirty but has to
+			 * write it immediatly.
+			 * Q: Can't we just create an alias for the inode
+			 * instead?  And if not, why not?
+			 */
+			if (inode->i_ino == LOGFS_INO_MASTER)
+				logfs_write_anchor(inode->i_sb);
+			else {
+				err = __logfs_write_inode(inode, flags);
+			}
+		}
 	}
 	logfs_put_write_page(page);
 	return err;

fs/logfs/segment.c

@@ -93,49 +93,57 @@ void __logfs_buf_write(struct logfs_area *area, u64 ofs, void *buf, size_t len,
 	} while (len);
 }
 
-/*
- * bdev_writeseg will write full pages.  Memset the tail to prevent data leaks.
- */
-static void pad_wbuf(struct logfs_area *area, int final)
+static void pad_partial_page(struct logfs_area *area)
 {
 	struct super_block *sb = area->a_sb;
-	struct logfs_super *super = logfs_super(sb);
 	struct page *page;
 	u64 ofs = dev_ofs(sb, area->a_segno, area->a_used_bytes);
 	pgoff_t index = ofs >> PAGE_SHIFT;
 	long offset = ofs & (PAGE_SIZE-1);
 	u32 len = PAGE_SIZE - offset;
 
-	if (len == PAGE_SIZE) {
-		/* The math in this function can surely use some love */
-		len = 0;
-	}
-	if (len) {
-		BUG_ON(area->a_used_bytes >= super->s_segsize);
-
-		page = get_mapping_page(area->a_sb, index, 0);
+	if (len % PAGE_SIZE) {
+		page = get_mapping_page(sb, index, 0);
 		BUG_ON(!page); /* FIXME: reserve a pool */
 		memset(page_address(page) + offset, 0xff, len);
 		SetPagePrivate(page);
 		page_cache_release(page);
 	}
+}
 
-	if (!final)
-		return;
+static void pad_full_pages(struct logfs_area *area)
+{
+	struct super_block *sb = area->a_sb;
+	struct logfs_super *super = logfs_super(sb);
+	u64 ofs = dev_ofs(sb, area->a_segno, area->a_used_bytes);
+	u32 len = super->s_segsize - area->a_used_bytes;
+	pgoff_t index = PAGE_CACHE_ALIGN(ofs) >> PAGE_CACHE_SHIFT;
+	pgoff_t no_indizes = len >> PAGE_CACHE_SHIFT;
+	struct page *page;
 
-	area->a_used_bytes += len;
-	for ( ; area->a_used_bytes < super->s_segsize;
-			area->a_used_bytes += PAGE_SIZE) {
-		/* Memset another page */
-		index++;
-		page = get_mapping_page(area->a_sb, index, 0);
+	while (no_indizes) {
+		page = get_mapping_page(sb, index, 0);
 		BUG_ON(!page); /* FIXME: reserve a pool */
-		memset(page_address(page), 0xff, PAGE_SIZE);
+		SetPageUptodate(page);
+		memset(page_address(page), 0xff, PAGE_CACHE_SIZE);
 		SetPagePrivate(page);
 		page_cache_release(page);
+		index++;
+		no_indizes--;
 	}
 }
 
+/*
+ * bdev_writeseg will write full pages.  Memset the tail to prevent data leaks.
+ * Also make sure we allocate (and memset) all pages for final writeout.
+ */
+static void pad_wbuf(struct logfs_area *area, int final)
+{
+	pad_partial_page(area);
+	if (final)
+		pad_full_pages(area);
+}
+
 /*
  * We have to be careful with the alias tree.  Since lookup is done by bix,
  * it needs to be normalized, so 14, 15, 16, etc. all match when dealing with
@@ -683,7 +691,7 @@ int logfs_segment_delete(struct inode *inode, struct logfs_shadow *shadow)
 	return 0;
 }
 
-static void freeseg(struct super_block *sb, u32 segno)
+void freeseg(struct super_block *sb, u32 segno)
 {
 	struct logfs_super *super = logfs_super(sb);
 	struct address_space *mapping = super->s_mapping_inode->i_mapping;

fs/logfs/super.c

@@ -277,7 +277,7 @@ static int logfs_recover_sb(struct super_block *sb)
 	}
 	if (valid0 && valid1 && ds_cmp(ds0, ds1)) {
 		printk(KERN_INFO"Superblocks don't match - fixing.\n");
-		return write_one_sb(sb, super->s_devops->find_last_sb);
+		return logfs_write_sb(sb);
 	}
 	/* If neither is valid now, something's wrong.  Didn't we properly
 	 * check them before?!? */
@@ -289,6 +289,10 @@ static int logfs_make_writeable(struct super_block *sb)
 {
 	int err;
 
+	err = logfs_open_segfile(sb);
+	if (err)
+		return err;
+
 	/* Repair any broken superblock copies */
 	err = logfs_recover_sb(sb);
 	if (err)
@@ -299,10 +303,6 @@ static int logfs_make_writeable(struct super_block *sb)
 	if (err)
 		return err;
 
-	err = logfs_open_segfile(sb);
-	if (err)
-		return err;
-
 	/* Do one GC pass before any data gets dirtied */
 	logfs_gc_pass(sb);
 
@@ -328,7 +328,7 @@ static int logfs_get_sb_final(struct super_block *sb, struct vfsmount *mnt)
 
 	sb->s_root = d_alloc_root(rootdir);
 	if (!sb->s_root)
-		goto fail;
+		goto fail2;
 
 	super->s_erase_page = alloc_pages(GFP_KERNEL, 0);
 	if (!super->s_erase_page)
@@ -572,8 +572,7 @@ int logfs_get_sb_device(struct file_system_type *type, int flags,
 	return 0;
 
 err1:
-	up_write(&sb->s_umount);
-	deactivate_super(sb);
+	deactivate_locked_super(sb);
 	return err;
 err0:
 	kfree(super);