logfs: handle powerfail on NAND flash

The write buffer may not have been written and may no longer be written
due to an interrupted write in the affected page.

Signed-off-by: Joern Engel <joern@logfs.org>

fs/logfs/dev_bdev.c

@@ -303,6 +303,11 @@ static void bdev_put_device(struct super_block *sb)
 	close_bdev_exclusive(logfs_super(sb)->s_bdev, FMODE_READ|FMODE_WRITE);
 }
 
+static int bdev_can_write_buf(struct super_block *sb, u64 ofs)
+{
+	return 0;
+}
+
 static const struct logfs_device_ops bd_devops = {
 	.find_first_sb	= bdev_find_first_sb,
 	.find_last_sb	= bdev_find_last_sb,
@@ -310,6 +315,7 @@ static const struct logfs_device_ops bd_devops = {
 	.readpage	= bdev_readpage,
 	.writeseg	= bdev_writeseg,
 	.erase		= bdev_erase,
+	.can_write_buf	= bdev_can_write_buf,
 	.sync		= bdev_sync,
 	.put_device	= bdev_put_device,
 };

fs/logfs/dev_mtd.c

@@ -9,6 +9,7 @@
 #include <linux/completion.h>
 #include <linux/mount.h>
 #include <linux/sched.h>
+#include <linux/slab.h>
 
 #define PAGE_OFS(ofs) ((ofs) & (PAGE_SIZE-1))
 
@@ -126,7 +127,8 @@ static int mtd_readpage(void *_sb, struct page *page)
 
 	err = mtd_read(sb, page->index << PAGE_SHIFT, PAGE_SIZE,
 			page_address(page));
-	if (err == -EUCLEAN) {
+	if (err == -EUCLEAN || err == -EBADMSG) {
+		/* -EBADMSG happens regularly on power failures */
 		err = 0;
 		/* FIXME: force GC this segment */
 	}
@@ -233,12 +235,32 @@ static void mtd_put_device(struct super_block *sb)
 	put_mtd_device(logfs_super(sb)->s_mtd);
 }
 
+static int mtd_can_write_buf(struct super_block *sb, u64 ofs)
+{
+	struct logfs_super *super = logfs_super(sb);
+	void *buf;
+	int err;
+
+	buf = kmalloc(super->s_writesize, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+	err = mtd_read(sb, ofs, super->s_writesize, buf);
+	if (err)
+		goto out;
+	if (memchr_inv(buf, 0xff, super->s_writesize))
+		err = -EIO;
+	kfree(buf);
+out:
+	return err;
+}
+
 static const struct logfs_device_ops mtd_devops = {
 	.find_first_sb	= mtd_find_first_sb,
 	.find_last_sb	= mtd_find_last_sb,
 	.readpage	= mtd_readpage,
 	.writeseg	= mtd_writeseg,
 	.erase		= mtd_erase,
+	.can_write_buf	= mtd_can_write_buf,
 	.sync		= mtd_sync,
 	.put_device	= mtd_put_device,
 };

fs/logfs/gc.c

@@ -122,7 +122,7 @@ static void logfs_cleanse_block(struct super_block *sb, u64 ofs, u64 ino,
 	logfs_safe_iput(inode, cookie);
 }
 
-static u32 logfs_gc_segment(struct super_block *sb, u32 segno, u8 dist)
+static u32 logfs_gc_segment(struct super_block *sb, u32 segno)
 {
 	struct logfs_super *super = logfs_super(sb);
 	struct logfs_segment_header sh;
@@ -401,7 +401,7 @@ static int __logfs_gc_once(struct super_block *sb, struct gc_candidate *cand)
 			segno, (u64)segno << super->s_segshift,
 			dist, no_free_segments(sb), valid,
 			super->s_free_bytes);
-	cleaned = logfs_gc_segment(sb, segno, dist);
+	cleaned = logfs_gc_segment(sb, segno);
 	log_gc("GC segment #%02x complete - now %x valid\n", segno,
 			valid - cleaned);
 	BUG_ON(cleaned != valid);
@@ -632,38 +632,31 @@ static int check_area(struct super_block *sb, int i)
 {
 	struct logfs_super *super = logfs_super(sb);
 	struct logfs_area *area = super->s_area[i];
-	struct logfs_object_header oh;
+	gc_level_t gc_level;
+	u32 cleaned, valid, ec;
 	u32 segno = area->a_segno;
-	u32 ofs = area->a_used_bytes;
+	u64 ofs = dev_ofs(sb, area->a_segno, area->a_written_bytes);
-	__be32 crc;
-	int err;
 
 	if (!area->a_is_open)
 		return 0;
 
-	for (ofs = area->a_used_bytes;
+	if (super->s_devops->can_write_buf(sb, ofs) == 0)
-	     ofs <= super->s_segsize - sizeof(oh);
+		return 0;
-	     ofs += (u32)be16_to_cpu(oh.len) + sizeof(oh)) {
-		err = wbuf_read(sb, dev_ofs(sb, segno, ofs), sizeof(oh), &oh);
-		if (err)
-			return err;
 
-		if (!memchr_inv(&oh, 0xff, sizeof(oh)))
+	printk(KERN_INFO"LogFS: Possibly incomplete write at %llx\n", ofs);
-			break;
+	/*
-
+	 * The device cannot write back the write buffer.  Most likely the
-		crc = logfs_crc32(&oh, sizeof(oh) - 4, 4);
+	 * wbuf was already written out and the system crashed at some point
-		if (crc != oh.crc) {
+	 * before the journal commit happened.  In that case we wouldn't have
-			printk(KERN_INFO "interrupted header at %llx\n",
+	 * to do anything.  But if the crash happened before the wbuf was
-					dev_ofs(sb, segno, ofs));
+	 * written out correctly, we must GC this segment.  So assume the
-			return 0;
+	 * worst and always do the GC run.
-		}
+	 */
-	}
+	area->a_is_open = 0;
-	if (ofs != area->a_used_bytes) {
+	valid = logfs_valid_bytes(sb, segno, &ec, &gc_level);
-		printk(KERN_INFO "%x bytes unaccounted data found at %llx\n",
+	cleaned = logfs_gc_segment(sb, segno);
-				ofs - area->a_used_bytes,
+	if (cleaned != valid)
-				dev_ofs(sb, segno, area->a_used_bytes));
+		return -EIO;
-		area->a_used_bytes = ofs;
-	}
 	return 0;
 }
 

fs/logfs/logfs.h

@@ -144,6 +144,7 @@ struct logfs_area_ops {
  * @erase:			erase one segment
  * @read:			read from the device
  * @erase:			erase part of the device
+ * @can_write_buf:		decide whether wbuf can be written to ofs
  */
 struct logfs_device_ops {
 	struct page *(*find_first_sb)(struct super_block *sb, u64 *ofs);
@@ -153,6 +154,7 @@ struct logfs_device_ops {
 	void (*writeseg)(struct super_block *sb, u64 ofs, size_t len);
 	int (*erase)(struct super_block *sb, loff_t ofs, size_t len,
 			int ensure_write);
+	int (*can_write_buf)(struct super_block *sb, u64 ofs);
 	void (*sync)(struct super_block *sb);
 	void (*put_device)(struct super_block *sb);
 };