UBI: do not switch to R/O mode on read errors

This patch improves UBI errors handling. ATM UBI switches to
R/O mode when the WL worker fails to read the source PEB.
This means that the upper layers (e.g., UBIFS) has no
chances to unmap the erroneous PEB and fix the error.
This patch changes this behaviour and makes UBI put PEBs
like this into a separate RB-tree, thus preventing the
WL worker from hitting the same read errors again and
again.

But there is a 10% limit on a maximum amount of PEBs like this.
If there are too much of them, UBI switches to R/O mode.

Additionally, this patch teaches UBI not to panic and
switch to R/O mode if after a PEB has been copied, the
target LEB cannot be read back. Instead, now UBI cancels
the operation and schedules the target PEB for torturing.

The error paths has been tested by ingecting errors
into 'ubi_eba_copy_leb()'.

Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>

drivers/mtd/ubi/build.c

@@ -632,6 +632,15 @@ static int io_init(struct ubi_device *ubi)
 		return -EINVAL;
 	}
 
+	/*
+	 * Set maximum amount of physical erroneous eraseblocks to be 10%.
+	 * Erroneous PEB are those which have read errors.
+	 */
+	ubi->max_erroneous = ubi->peb_count / 10;
+	if (ubi->max_erroneous < 16)
+		ubi->max_erroneous = 16;
+	dbg_msg("max_erroneous    %d", ubi->max_erroneous);
+
 	/*
 	 * It may happen that EC and VID headers are situated in one minimal
 	 * I/O unit. In this case we can only accept this UBI image in

drivers/mtd/ubi/eba.c

@@ -419,8 +419,9 @@ int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
 				 * not implemented.
 				 */
 				if (err == UBI_IO_BAD_VID_HDR) {
-					ubi_warn("bad VID header at PEB %d, LEB"
-						 "%d:%d", pnum, vol_id, lnum);
+					ubi_warn("corrupted VID header at PEB "
+						 "%d, LEB %d:%d", pnum, vol_id,
+						 lnum);
 					err = -EBADMSG;
 				} else
 					ubi_ro_mode(ubi);
@@ -1032,6 +1033,8 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
 	if (err && err != UBI_IO_BITFLIPS) {
 		ubi_warn("error %d while reading data from PEB %d",
 			 err, from);
+		if (err == -EIO)
+			err = MOVE_SOURCE_RD_ERR;
 		goto out_unlock_buf;
 	}
 
@@ -1078,9 +1081,11 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
 	/* Read the VID header back and check if it was written correctly */
 	err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
 	if (err) {
-		if (err != UBI_IO_BITFLIPS)
+		if (err != UBI_IO_BITFLIPS) {
 			ubi_warn("cannot read VID header back from PEB %d", to);
-		else
+			if (err == -EIO)
+				err = MOVE_TARGET_RD_ERR;
+		} else
 			err = MOVE_CANCEL_BITFLIPS;
 		goto out_unlock_buf;
 	}
@@ -1102,10 +1107,12 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
 
 		err = ubi_io_read_data(ubi, ubi->peb_buf2, to, 0, aldata_size);
 		if (err) {
-			if (err != UBI_IO_BITFLIPS)
+			if (err != UBI_IO_BITFLIPS) {
 				ubi_warn("cannot read data back from PEB %d",
 					 to);
-			else
+				if (err == -EIO)
+					err = MOVE_TARGET_RD_ERR;
+			} else
 				err = MOVE_CANCEL_BITFLIPS;
 			goto out_unlock_buf;
 		}

drivers/mtd/ubi/ubi.h

@@ -105,6 +105,10 @@ enum {
  *
  * MOVE_CANCEL_RACE: canceled because the volume is being deleted, the source
  *                   PEB was put meanwhile, or there is I/O on the source PEB
+ * MOVE_SOURCE_RD_ERR: canceled because there was a read error from the source
+ *                     PEB
+ * MOVE_TARGET_RD_ERR: canceled because there was a read error from the target
+ *                     PEB
  * MOVE_TARGET_WR_ERR: canceled because there was a write error to the target
  *                     PEB
  * MOVE_CANCEL_BITFLIPS: canceled because a bit-flip was detected in the
@@ -112,6 +116,8 @@ enum {
  */
 enum {
 	MOVE_CANCEL_RACE = 1,
+	MOVE_SOURCE_RD_ERR,
+	MOVE_TARGET_RD_ERR,
 	MOVE_TARGET_WR_ERR,
 	MOVE_CANCEL_BITFLIPS,
 };
@@ -334,14 +340,15 @@ struct ubi_wl_entry;
  * @alc_mutex: serializes "atomic LEB change" operations
  *
  * @used: RB-tree of used physical eraseblocks
+ * @erroneous: RB-tree of erroneous used physical eraseblocks
  * @free: RB-tree of free physical eraseblocks
  * @scrub: RB-tree of physical eraseblocks which need scrubbing
  * @pq: protection queue (contain physical eraseblocks which are temporarily
  *      protected from the wear-leveling worker)
  * @pq_head: protection queue head
  * @wl_lock: protects the @used, @free, @pq, @pq_head, @lookuptbl, @move_from,
- * 	     @move_to, @move_to_put @erase_pending, @wl_scheduled and @works
- * 	     fields
+ * 	     @move_to, @move_to_put @erase_pending, @wl_scheduled, @works and
+ * 	     @erroneous_peb_count fields
  * @move_mutex: serializes eraseblock moves
  * @work_sem: synchronizes the WL worker with use tasks
  * @wl_scheduled: non-zero if the wear-leveling was scheduled
@@ -361,6 +368,8 @@ struct ubi_wl_entry;
  * @peb_size: physical eraseblock size
  * @bad_peb_count: count of bad physical eraseblocks
  * @good_peb_count: count of good physical eraseblocks
+ * @erroneous_peb_count: count of erroneous physical eraseblocks in @erroneous
+ * @max_erroneous: maximum allowed amount of erroneous physical eraseblocks
  * @min_io_size: minimal input/output unit size of the underlying MTD device
  * @hdrs_min_io_size: minimal I/O unit size used for VID and EC headers
  * @ro_mode: if the UBI device is in read-only mode
@@ -418,6 +427,7 @@ struct ubi_device {
 
 	/* Wear-leveling sub-system's stuff */
 	struct rb_root used;
+	struct rb_root erroneous;
 	struct rb_root free;
 	struct rb_root scrub;
 	struct list_head pq[UBI_PROT_QUEUE_LEN];
@@ -442,6 +452,8 @@ struct ubi_device {
 	int peb_size;
 	int bad_peb_count;
 	int good_peb_count;
+	int erroneous_peb_count;
+	int max_erroneous;
 	int min_io_size;
 	int hdrs_min_io_size;
 	int ro_mode;

drivers/mtd/ubi/wl.c

@@ -55,8 +55,8 @@
  *
  * As it was said, for the UBI sub-system all physical eraseblocks are either
  * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
- * used eraseblocks are kept in @wl->used or @wl->scrub RB-trees, or
- * (temporarily) in the @wl->pq queue.
+ * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
+ * RB-trees, as well as (temporarily) in the @wl->pq queue.
  *
  * When the WL sub-system returns a physical eraseblock, the physical
  * eraseblock is protected from being moved for some "time". For this reason,
@@ -83,6 +83,8 @@
  * used. The former state corresponds to the @wl->free tree. The latter state
  * is split up on several sub-states:
  * o the WL movement is allowed (@wl->used tree);
+ * o the WL movement is disallowed (@wl->erroneous) becouse the PEB is
+ *   erroneous - e.g., there was a read error;
  * o the WL movement is temporarily prohibited (@wl->pq queue);
  * o scrubbing is needed (@wl->scrub tree).
  *
@@ -653,7 +655,7 @@ static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
 				int cancel)
 {
-	int err, scrubbing = 0, torture = 0, protect = 0;
+	int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
 	struct ubi_wl_entry *e1, *e2;
 	struct ubi_vid_hdr *vid_hdr;
 
@@ -769,13 +771,31 @@ static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
 			goto out_not_moved;
 		}
 
-		if (err == MOVE_CANCEL_BITFLIPS ||
-		    err == MOVE_TARGET_WR_ERR) {
+		if (err == MOVE_CANCEL_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
+		    err == MOVE_TARGET_RD_ERR) {
 			/* Target PEB bit-flips or write error, torture it */
 			torture = 1;
 			goto out_not_moved;
 		}
 
+		if (err == MOVE_SOURCE_RD_ERR) {
+			/*
+			 * An error happened while reading the source PEB. Do
+			 * not switch to R/O mode in this case, and give the
+			 * upper layers a possibility to recover from this,
+			 * e.g. by unmapping corresponding LEB. Instead, just
+			 * put thie PEB to the @ubi->erroneus list to prevent
+			 * UBI from trying to move the over and over again.
+			 */
+			if (ubi->erroneous_peb_count > ubi->max_erroneous) {
+				ubi_err("too many erroneous eraseblocks (%d)",
+					ubi->erroneous_peb_count);
+				goto out_error;
+			}
+			erroneous = 1;
+			goto out_not_moved;
+		}
+
 		if (err < 0)
 			goto out_error;
 
@@ -832,7 +852,10 @@ static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
 	spin_lock(&ubi->wl_lock);
 	if (protect)
 		prot_queue_add(ubi, e1);
-	else if (scrubbing)
+	else if (erroneous) {
+		wl_tree_add(e1, &ubi->erroneous);
+		ubi->erroneous_peb_count += 1;
+	} else if (scrubbing)
 		wl_tree_add(e1, &ubi->scrub);
 	else
 		wl_tree_add(e1, &ubi->used);
@@ -1116,6 +1139,13 @@ int ubi_wl_put_peb(struct ubi_device *ubi, int pnum, int torture)
 		} else if (in_wl_tree(e, &ubi->scrub)) {
 			paranoid_check_in_wl_tree(e, &ubi->scrub);
 			rb_erase(&e->u.rb, &ubi->scrub);
+		} else if (in_wl_tree(e, &ubi->erroneous)) {
+			paranoid_check_in_wl_tree(e, &ubi->erroneous);
+			rb_erase(&e->u.rb, &ubi->erroneous);
+			ubi->erroneous_peb_count -= 1;
+			ubi_assert(ubi->erroneous_peb_count >= 0);
+			/* Erronious PEBs should be tortured */
+			torture = 1;
 		} else {
 			err = prot_queue_del(ubi, e->pnum);
 			if (err) {
@@ -1364,7 +1394,7 @@ int ubi_wl_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si)
 	struct ubi_scan_leb *seb, *tmp;
 	struct ubi_wl_entry *e;
 
-	ubi->used = ubi->free = ubi->scrub = RB_ROOT;
+	ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
 	spin_lock_init(&ubi->wl_lock);
 	mutex_init(&ubi->move_mutex);
 	init_rwsem(&ubi->work_sem);
@@ -1502,6 +1532,7 @@ void ubi_wl_close(struct ubi_device *ubi)
 	cancel_pending(ubi);
 	protection_queue_destroy(ubi);
 	tree_destroy(&ubi->used);
+	tree_destroy(&ubi->erroneous);
 	tree_destroy(&ubi->free);
 	tree_destroy(&ubi->scrub);
 	kfree(ubi->lookuptbl);