1 files changed, 153 insertions, 112 deletions

diff --git a/fs/logfs/readwrite.c b/fs/logfs/readwrite.c
index bff40253dfb..48140315f62 100644
--- a/fs/logfs/readwrite.c
+++ b/fs/logfs/readwrite.c
@@ -119,14 +119,14 @@ static void logfs_disk_to_inode(struct logfs_disk_inode *di, struct inode*inode)
 	inode->i_mode	= be16_to_cpu(di->di_mode);
 	li->li_height	= di->di_height;
 	li->li_flags	= be32_to_cpu(di->di_flags);
-	inode->i_uid	= be32_to_cpu(di->di_uid);
-	inode->i_gid	= be32_to_cpu(di->di_gid);
+	i_uid_write(inode, be32_to_cpu(di->di_uid));
+	i_gid_write(inode, be32_to_cpu(di->di_gid));
 	inode->i_size	= be64_to_cpu(di->di_size);
 	logfs_set_blocks(inode, be64_to_cpu(di->di_used_bytes));
 	inode->i_atime	= be64_to_timespec(di->di_atime);
 	inode->i_ctime	= be64_to_timespec(di->di_ctime);
 	inode->i_mtime	= be64_to_timespec(di->di_mtime);
-	inode->i_nlink	= be32_to_cpu(di->di_refcount);
+	set_nlink(inode, be32_to_cpu(di->di_refcount));
 	inode->i_generation = be32_to_cpu(di->di_generation);
 
 	switch (inode->i_mode & S_IFMT) {
@@ -156,8 +156,8 @@ static void logfs_inode_to_disk(struct inode *inode, struct logfs_disk_inode*di)
 	di->di_height	= li->li_height;
 	di->di_pad	= 0;
 	di->di_flags	= cpu_to_be32(li->li_flags);
-	di->di_uid	= cpu_to_be32(inode->i_uid);
-	di->di_gid	= cpu_to_be32(inode->i_gid);
+	di->di_uid	= cpu_to_be32(i_uid_read(inode));
+	di->di_gid	= cpu_to_be32(i_gid_read(inode));
 	di->di_size	= cpu_to_be64(i_size_read(inode));
 	di->di_used_bytes = cpu_to_be64(li->li_used_bytes);
 	di->di_atime	= timespec_to_be64(inode->i_atime);
@@ -244,8 +244,7 @@ static void preunlock_page(struct super_block *sb, struct page *page, int lock)
  * is waiting for s_write_mutex.  We annotate this fact by setting PG_pre_locked
  * in addition to PG_locked.
  */
-static void logfs_get_wblocks(struct super_block *sb, struct page *page,
-		int lock)
+void logfs_get_wblocks(struct super_block *sb, struct page *page, int lock)
 {
 	struct logfs_super *super = logfs_super(sb);
 
@@ -260,8 +259,7 @@ static void logfs_get_wblocks(struct super_block *sb, struct page *page,
 	}
 }
 
-static void logfs_put_wblocks(struct super_block *sb, struct page *page,
-		int lock)
+void logfs_put_wblocks(struct super_block *sb, struct page *page, int lock)
 {
 	struct logfs_super *super = logfs_super(sb);
 
@@ -424,31 +422,12 @@ static void inode_write_block(struct logfs_block *block)
 	if (inode->i_ino == LOGFS_INO_MASTER)
 		logfs_write_anchor(inode->i_sb);
 	else {
-		ret = __logfs_write_inode(inode, 0);
+		ret = __logfs_write_inode(inode, NULL, 0);
 		/* see indirect_write_block comment */
 		BUG_ON(ret);
 	}
 }
 
-static gc_level_t inode_block_level(struct logfs_block *block)
-{
-	BUG_ON(block->inode->i_ino == LOGFS_INO_MASTER);
-	return GC_LEVEL(LOGFS_MAX_LEVELS);
-}
-
-static gc_level_t indirect_block_level(struct logfs_block *block)
-{
-	struct page *page;
-	struct inode *inode;
-	u64 bix;
-	level_t level;
-
-	page = block->page;
-	inode = page->mapping->host;
-	logfs_unpack_index(page->index, &bix, &level);
-	return expand_level(inode->i_ino, level);
-}
-
 /*
  * This silences a false, yet annoying gcc warning.  I hate it when my editor
  * jumps into bitops.h each time I recompile this file.
@@ -500,7 +479,7 @@ static int inode_write_alias(struct super_block *sb,
 			val = inode_val0(inode);
 			break;
 		case INODE_USED_OFS:
-			val = cpu_to_be64(li->li_used_bytes);;
+			val = cpu_to_be64(li->li_used_bytes);
 			break;
 		case INODE_SIZE_OFS:
 			val = cpu_to_be64(i_size_read(inode));
@@ -538,9 +517,9 @@ static int indirect_write_alias(struct super_block *sb,
 
 		ino = page->mapping->host->i_ino;
 		logfs_unpack_index(page->index, &bix, &level);
-		child = kmap_atomic(page, KM_USER0);
+		child = kmap_atomic(page);
 		val = child[pos];
-		kunmap_atomic(child, KM_USER0);
+		kunmap_atomic(child);
 		err = write_one_alias(sb, ino, bix, level, pos, val);
 		if (err)
 			return err;
@@ -579,22 +558,25 @@ static void inode_free_block(struct super_block *sb, struct logfs_block *block)
 static void indirect_free_block(struct super_block *sb,
 		struct logfs_block *block)
 {
-	ClearPagePrivate(block->page);
-	block->page->private = 0;
+	struct page *page = block->page;
+
+	if (PagePrivate(page)) {
+		ClearPagePrivate(page);
+		page_cache_release(page);
+		set_page_private(page, 0);
+	}
 	__free_block(sb, block);
 }
 
 
 static struct logfs_block_ops inode_block_ops = {
 	.write_block = inode_write_block,
-	.block_level = inode_block_level,
 	.free_block = inode_free_block,
 	.write_alias = inode_write_alias,
 };
 
 struct logfs_block_ops indirect_block_ops = {
 	.write_block = indirect_write_block,
-	.block_level = indirect_block_level,
 	.free_block = indirect_free_block,
 	.write_alias = indirect_write_alias,
 };
@@ -671,8 +653,11 @@ static void alloc_data_block(struct inode *inode, struct page *page)
 	logfs_unpack_index(page->index, &bix, &level);
 	block = __alloc_block(inode->i_sb, inode->i_ino, bix, level);
 	block->page = page;
+
 	SetPagePrivate(page);
-	page->private = (unsigned long)block;
+	page_cache_get(page);
+	set_page_private(page, (unsigned long) block);
+
 	block->ops = &indirect_block_ops;
 }
 
@@ -688,9 +673,9 @@ static void alloc_indirect_block(struct inode *inode, struct page *page,
 	alloc_data_block(inode, page);
 
 	block = logfs_block(page);
-	array = kmap_atomic(page, KM_USER0);
+	array = kmap_atomic(page);
 	initialize_block_counters(page, block, array, page_is_empty);
-	kunmap_atomic(array, KM_USER0);
+	kunmap_atomic(array);
 }
 
 static void block_set_pointer(struct page *page, int index, u64 ptr)
@@ -700,10 +685,10 @@ static void block_set_pointer(struct page *page, int index, u64 ptr)
 	u64 oldptr;
 
 	BUG_ON(!block);
-	array = kmap_atomic(page, KM_USER0);
+	array = kmap_atomic(page);
 	oldptr = be64_to_cpu(array[index]);
 	array[index] = cpu_to_be64(ptr);
-	kunmap_atomic(array, KM_USER0);
+	kunmap_atomic(array);
 	SetPageUptodate(page);
 
 	block->full += !!(ptr & LOGFS_FULLY_POPULATED)
@@ -716,9 +701,9 @@ static u64 block_get_pointer(struct page *page, int index)
 	__be64 *block;
 	u64 ptr;
 
-	block = kmap_atomic(page, KM_USER0);
+	block = kmap_atomic(page);
 	ptr = be64_to_cpu(block[index]);
-	kunmap_atomic(block, KM_USER0);
+	kunmap_atomic(block);
 	return ptr;
 }
 
@@ -865,7 +850,7 @@ static u64 seek_holedata_loop(struct inode *inode, u64 bix, int data)
 		}
 
 		slot = get_bits(bix, SUBLEVEL(level));
-		rblock = kmap_atomic(page, KM_USER0);
+		rblock = kmap_atomic(page);
 		while (slot < LOGFS_BLOCK_FACTOR) {
 			if (data && (rblock[slot] != 0))
 				break;
@@ -876,12 +861,12 @@ static u64 seek_holedata_loop(struct inode *inode, u64 bix, int data)
 			bix &= ~(increment - 1);
 		}
 		if (slot >= LOGFS_BLOCK_FACTOR) {
-			kunmap_atomic(rblock, KM_USER0);
+			kunmap_atomic(rblock);
 			logfs_put_read_page(page);
 			return bix;
 		}
 		bofs = be64_to_cpu(rblock[slot]);
-		kunmap_atomic(rblock, KM_USER0);
+		kunmap_atomic(rblock);
 		logfs_put_read_page(page);
 		if (!bofs) {
 			BUG_ON(data);
@@ -913,6 +898,8 @@ u64 logfs_seek_hole(struct inode *inode, u64 bix)
 		return bix;
 	else if (li->li_data[INDIRECT_INDEX] & LOGFS_FULLY_POPULATED)
 		bix = maxbix(li->li_height);
+	else if (bix >= maxbix(li->li_height))
+		return bix;
 	else {
 		bix = seek_holedata_loop(inode, bix, 0);
 		if (bix < maxbix(li->li_height))
@@ -1114,17 +1101,25 @@ static int logfs_reserve_bytes(struct inode *inode, int bytes)
 int get_page_reserve(struct inode *inode, struct page *page)
 {
 	struct logfs_super *super = logfs_super(inode->i_sb);
+	struct logfs_block *block = logfs_block(page);
 	int ret;
 
-	if (logfs_block(page) && logfs_block(page)->reserved_bytes)
+	if (block && block->reserved_bytes)
 		return 0;
 
 	logfs_get_wblocks(inode->i_sb, page, WF_LOCK);
-	ret = logfs_reserve_bytes(inode, 6 * LOGFS_MAX_OBJECTSIZE);
+	while ((ret = logfs_reserve_bytes(inode, 6 * LOGFS_MAX_OBJECTSIZE)) &&
+			!list_empty(&super->s_writeback_list)) {
+		block = list_entry(super->s_writeback_list.next,
+				struct logfs_block, alias_list);
+		block->ops->write_block(block);
+	}
 	if (!ret) {
 		alloc_data_block(inode, page);
-		logfs_block(page)->reserved_bytes += 6 * LOGFS_MAX_OBJECTSIZE;
+		block = logfs_block(page);
+		block->reserved_bytes += 6 * LOGFS_MAX_OBJECTSIZE;
 		super->s_dirty_pages += 6 * LOGFS_MAX_OBJECTSIZE;
+		list_move_tail(&block->alias_list, &super->s_writeback_list);
 	}
 	logfs_put_wblocks(inode->i_sb, page, WF_LOCK);
 	return ret;
@@ -1241,6 +1236,18 @@ static void free_shadow(struct inode *inode, struct logfs_shadow *shadow)
 	mempool_free(shadow, super->s_shadow_pool);
 }
 
+static void mark_segment(struct shadow_tree *tree, u32 segno)
+{
+	int err;
+
+	if (!btree_lookup32(&tree->segment_map, segno)) {
+		err = btree_insert32(&tree->segment_map, segno, (void *)1,
+				GFP_NOFS);
+		BUG_ON(err);
+		tree->no_shadowed_segments++;
+	}
+}
+
 /**
  * fill_shadow_tree - Propagate shadow tree changes due to a write
  * @inode:	Inode owning the page
@@ -1288,6 +1295,8 @@ static void fill_shadow_tree(struct inode *inode, struct page *page,
 
 		super->s_dirty_used_bytes += shadow->new_len;
 		super->s_dirty_free_bytes += shadow->old_len;
+		mark_segment(tree, shadow->old_ofs >> super->s_segshift);
+		mark_segment(tree, shadow->new_ofs >> super->s_segshift);
 	}
 }
 
@@ -1567,11 +1576,15 @@ int logfs_write_buf(struct inode *inode, struct page *page, long flags)
 static int __logfs_delete(struct inode *inode, struct page *page)
 {
 	long flags = WF_DELETE;
+	int err;
 
 	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 
 	if (page->index < I0_BLOCKS)
 		return logfs_write_direct(inode, page, flags);
+	err = grow_inode(inode, page->index, 0);
+	if (err)
+		return err;
 	return logfs_write_rec(inode, page, page->index, 0, flags);
 }
 
@@ -1613,14 +1626,14 @@ int logfs_rewrite_block(struct inode *inode, u64 bix, u64 ofs,
 		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.
+			 * write it immediately.
 			 * 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);
+				err = __logfs_write_inode(inode, page, flags);
 			}
 		}
 	}
@@ -1845,20 +1858,44 @@ static int __logfs_truncate(struct inode *inode, u64 size)
 	return logfs_truncate_direct(inode, size);
 }
 
-int logfs_truncate(struct inode *inode, u64 size)
+/*
+ * Truncate, by changing the segment file, can consume a fair amount
+ * of resources.  So back off from time to time and do some GC.
+ * 8 or 2048 blocks should be well within safety limits even if
+ * every single block resided in a different segment.
+ */
+#define TRUNCATE_STEP	(8 * 1024 * 1024)
+int logfs_truncate(struct inode *inode, u64 target)
 {
 	struct super_block *sb = inode->i_sb;
-	int err;
+	u64 size = i_size_read(inode);
+	int err = 0;
 
-	logfs_get_wblocks(sb, NULL, 1);
-	err = __logfs_truncate(inode, size);
-	if (!err)
-		err = __logfs_write_inode(inode, 0);
-	logfs_put_wblocks(sb, NULL, 1);
+	size = ALIGN(size, TRUNCATE_STEP);
+	while (size > target) {
+		if (size > TRUNCATE_STEP)
+			size -= TRUNCATE_STEP;
+		else
+			size = 0;
+		if (size < target)
+			size = target;
+
+		logfs_get_wblocks(sb, NULL, 1);
+		err = __logfs_truncate(inode, size);
+		if (!err)
+			err = __logfs_write_inode(inode, NULL, 0);
+		logfs_put_wblocks(sb, NULL, 1);
+	}
 
-	if (!err)
-		err = vmtruncate(inode, size);
+	if (!err) {
+		err = inode_newsize_ok(inode, target);
+		if (err)
+			goto out;
+
+		truncate_setsize(inode, target);
+	}
 
+ out:
 	/* I don't trust error recovery yet. */
 	WARN_ON(err);
 	return err;
@@ -1880,8 +1917,11 @@ static void move_page_to_inode(struct inode *inode, struct page *page)
 	li->li_block = block;
 
 	block->page = NULL;
-	page->private = 0;
-	ClearPagePrivate(page);
+	if (PagePrivate(page)) {
+		ClearPagePrivate(page);
+		page_cache_release(page);
+		set_page_private(page, 0);
+	}
 }
 
 static void move_inode_to_page(struct page *page, struct inode *inode)
@@ -1897,8 +1937,12 @@ static void move_inode_to_page(struct page *page, struct inode *inode)
 	BUG_ON(PagePrivate(page));
 	block->ops = &indirect_block_ops;
 	block->page = page;
-	page->private = (unsigned long)block;
-	SetPagePrivate(page);
+
+	if (!PagePrivate(page)) {
+		SetPagePrivate(page);
+		page_cache_get(page);
+		set_page_private(page, (unsigned long) block);
+	}
 
 	block->inode = NULL;
 	li->li_block = NULL;
@@ -1923,9 +1967,9 @@ int logfs_read_inode(struct inode *inode)
 	if (IS_ERR(page))
 		return PTR_ERR(page);
 
-	di = kmap_atomic(page, KM_USER0);
+	di = kmap_atomic(page);
 	logfs_disk_to_inode(di, inode);
-	kunmap_atomic(di, KM_USER0);
+	kunmap_atomic(di);
 	move_page_to_inode(inode, page);
 	page_cache_release(page);
 	return 0;
@@ -1944,38 +1988,13 @@ static struct page *inode_to_page(struct inode *inode)
 	if (!page)
 		return NULL;
 
-	di = kmap_atomic(page, KM_USER0);
+	di = kmap_atomic(page);
 	logfs_inode_to_disk(inode, di);
-	kunmap_atomic(di, KM_USER0);
+	kunmap_atomic(di);
 	move_inode_to_page(page, inode);
 	return page;
 }
 
-/* Cheaper version of write_inode.  All changes are concealed in
- * aliases, which are moved back.  No write to the medium happens.
- */
-void logfs_clear_inode(struct inode *inode)
-{
-	struct super_block *sb = inode->i_sb;
-	struct logfs_inode *li = logfs_inode(inode);
-	struct logfs_block *block = li->li_block;
-	struct page *page;
-
-	/* Only deleted files may be dirty at this point */
-	BUG_ON(inode->i_state & I_DIRTY && inode->i_nlink);
-	if (!block)
-		return;
-	if ((logfs_super(sb)->s_flags & LOGFS_SB_FLAG_SHUTDOWN)) {
-		block->ops->free_block(inode->i_sb, block);
-		return;
-	}
-
-	BUG_ON(inode->i_ino < LOGFS_RESERVED_INOS);
-	page = inode_to_page(inode);
-	BUG_ON(!page); /* FIXME: Use emergency page */
-	logfs_put_write_page(page);
-}
-
 static int do_write_inode(struct inode *inode)
 {
 	struct super_block *sb = inode->i_sb;
@@ -1998,6 +2017,9 @@ static int do_write_inode(struct inode *inode)
 
 	/* FIXME: transaction is part of logfs_block now.  Is that enough? */
 	err = logfs_write_buf(master_inode, page, 0);
+	if (err)
+		move_page_to_inode(inode, page);
+
 	logfs_put_write_page(page);
 	return err;
 }
@@ -2025,13 +2047,13 @@ static void logfs_mod_segment_entry(struct super_block *sb, u32 segno,
 
 	if (write)
 		alloc_indirect_block(inode, page, 0);
-	se = kmap_atomic(page, KM_USER0);
+	se = kmap_atomic(page);
 	change_se(se + child_no, arg);
 	if (write) {
 		logfs_set_alias(sb, logfs_block(page), child_no);
 		BUG_ON((int)be32_to_cpu(se[child_no].valid) > super->s_segsize);
 	}
-	kunmap_atomic(se, KM_USER0);
+	kunmap_atomic(se);
 
 	logfs_put_write_page(page);
 }
@@ -2107,14 +2129,14 @@ void logfs_set_segment_unreserved(struct super_block *sb, u32 segno, u32 ec)
 			ec_level);
 }
 
-int __logfs_write_inode(struct inode *inode, long flags)
+int __logfs_write_inode(struct inode *inode, struct page *page, long flags)
 {
 	struct super_block *sb = inode->i_sb;
 	int ret;
 
-	logfs_get_wblocks(sb, NULL, flags & WF_LOCK);
+	logfs_get_wblocks(sb, page, flags & WF_LOCK);
 	ret = do_write_inode(inode);
-	logfs_put_wblocks(sb, NULL, flags & WF_LOCK);
+	logfs_put_wblocks(sb, page, flags & WF_LOCK);
 	return ret;
 }
 
@@ -2143,18 +2165,39 @@ static int do_delete_inode(struct inode *inode)
  * ZOMBIE inodes have already been deleted before and should remain dead,
  * if it weren't for valid checking.  No need to kill them again here.
  */
-void logfs_delete_inode(struct inode *inode)
+void logfs_evict_inode(struct inode *inode)
 {
+	struct super_block *sb = inode->i_sb;
 	struct logfs_inode *li = logfs_inode(inode);
+	struct logfs_block *block = li->li_block;
+	struct page *page;
 
-	if (!(li->li_flags & LOGFS_IF_ZOMBIE)) {
-		li->li_flags |= LOGFS_IF_ZOMBIE;
-		if (i_size_read(inode) > 0)
-			logfs_truncate(inode, 0);
-		do_delete_inode(inode);
+	if (!inode->i_nlink) {
+		if (!(li->li_flags & LOGFS_IF_ZOMBIE)) {
+			li->li_flags |= LOGFS_IF_ZOMBIE;
+			if (i_size_read(inode) > 0)
+				logfs_truncate(inode, 0);
+			do_delete_inode(inode);
+		}
 	}
-	truncate_inode_pages(&inode->i_data, 0);
+	truncate_inode_pages_final(&inode->i_data);
 	clear_inode(inode);
+
+	/* Cheaper version of write_inode.  All changes are concealed in
+	 * aliases, which are moved back.  No write to the medium happens.
+	 */
+	/* Only deleted files may be dirty at this point */
+	BUG_ON(inode->i_state & I_DIRTY && inode->i_nlink);
+	if (!block)
+		return;
+	if ((logfs_super(sb)->s_flags & LOGFS_SB_FLAG_SHUTDOWN)) {
+		block->ops->free_block(inode->i_sb, block);
+		return;
+	}
+
+	page = inode_to_page(inode);
+	BUG_ON(!page); /* FIXME: Use emergency page */
+	logfs_put_write_page(page);
 }
 
 void btree_write_block(struct logfs_block *block)
@@ -2207,10 +2250,10 @@ int logfs_inode_write(struct inode *inode, const void *buf, size_t count,
 	if (!page)
 		return -ENOMEM;
 
-	pagebuf = kmap_atomic(page, KM_USER0);
+	pagebuf = kmap_atomic(page);
 	memcpy(pagebuf, buf, count);
 	flush_dcache_page(page);
-	kunmap_atomic(pagebuf, KM_USER0);
+	kunmap_atomic(pagebuf);
 
 	if (i_size_read(inode) < pos + LOGFS_BLOCKSIZE)
 		i_size_write(inode, pos + LOGFS_BLOCKSIZE);
@@ -2238,6 +2281,7 @@ int logfs_init_rw(struct super_block *sb)
 	int min_fill = 3 * super->s_no_blocks;
 
 	INIT_LIST_HEAD(&super->s_object_alias);
+	INIT_LIST_HEAD(&super->s_writeback_list);
 	mutex_init(&super->s_write_mutex);
 	super->s_block_pool = mempool_create_kmalloc_pool(min_fill,
 			sizeof(struct logfs_block));
@@ -2250,9 +2294,6 @@ void logfs_cleanup_rw(struct super_block *sb)
 {
 	struct logfs_super *super = logfs_super(sb);
 
-	destroy_meta_inode(super->s_segfile_inode);
-	if (super->s_block_pool)
-		mempool_destroy(super->s_block_pool);
-	if (super->s_shadow_pool)
-		mempool_destroy(super->s_shadow_pool);
+	logfs_mempool_destroy(super->s_block_pool);
+	logfs_mempool_destroy(super->s_shadow_pool);
 }