raid5-ppl: Partial Parity Log write logging implementation

Implement the calculation of partial parity for a stripe and PPL write
logging functionality. The description of PPL is added to the
documentation. More details can be found in the comments in raid5-ppl.c.

Attach a page for holding the partial parity data to stripe_head.
Allocate it only if mddev has the MD_HAS_PPL flag set.

Partial parity is the xor of not modified data chunks of a stripe and is
calculated as follows:

- reconstruct-write case:
  xor data from all not updated disks in a stripe

- read-modify-write case:
  xor old data and parity from all updated disks in a stripe

Implement it using the async_tx API and integrate into raid_run_ops().
It must be called when we still have access to old data, so do it when
STRIPE_OP_BIODRAIN is set, but before ops_run_prexor5(). The result is
stored into sh->ppl_page.

Partial parity is not meaningful for full stripe write and is not stored
in the log or used for recovery, so don't attempt to calculate it when
stripe has STRIPE_FULL_WRITE.

Put the PPL metadata structures to md_p.h because userspace tools
(mdadm) will also need to read/write PPL.

Warn about using PPL with enabled disk volatile write-back cache for
now. It can be removed once disk cache flushing before writing PPL is
implemented.
Signed-off-by: Artur Paszkiewicz <artur.paszkiewicz@intel.com>
Signed-off-by: Shaohua Li <shli@fb.com>

Documentation/md/raid5-ppl.txt

+Partial Parity Log
+
+Partial Parity Log (PPL) is a feature available for RAID5 arrays. The issue
+addressed by PPL is that after a dirty shutdown, parity of a particular stripe
+may become inconsistent with data on other member disks. If the array is also
+in degraded state, there is no way to recalculate parity, because one of the
+disks is missing. This can lead to silent data corruption when rebuilding the
+array or using it is as degraded - data calculated from parity for array blocks
+that have not been touched by a write request during the unclean shutdown can
+be incorrect. Such condition is known as the RAID5 Write Hole. Because of
+this, md by default does not allow starting a dirty degraded array.
+
+Partial parity for a write operation is the XOR of stripe data chunks not
+modified by this write. It is just enough data needed for recovering from the
+write hole. XORing partial parity with the modified chunks produces parity for
+the stripe, consistent with its state before the write operation, regardless of
+which chunk writes have completed. If one of the not modified data disks of
+this stripe is missing, this updated parity can be used to recover its
+contents. PPL recovery is also performed when starting an array after an
+unclean shutdown and all disks are available, eliminating the need to resync
+the array. Because of this, using write-intent bitmap and PPL together is not
+supported.
+
+When handling a write request PPL writes partial parity before new data and
+parity are dispatched to disks. PPL is a distributed log - it is stored on
+array member drives in the metadata area, on the parity drive of a particular
+stripe.  It does not require a dedicated journaling drive. Write performance is
+reduced by up to 30%-40% but it scales with the number of drives in the array
+and the journaling drive does not become a bottleneck or a single point of
+failure.
+
+Unlike raid5-cache, the other solution in md for closing the write hole, PPL is
+not a true journal. It does not protect from losing in-flight data, only from
+silent data corruption. If a dirty disk of a stripe is lost, no PPL recovery is
+performed for this stripe (parity is not updated). So it is possible to have
+arbitrary data in the written part of a stripe if that disk is lost. In such
+case the behavior is the same as in plain raid5.
+
+PPL is available for md version-1 metadata and external (specifically IMSM)
+metadata arrays. It can be enabled using mdadm option --consistency-policy=ppl.
+
+Currently, volatile write-back cache should be disabled on all member drives
+when using PPL. Otherwise it cannot guarantee consistency in case of power
+failure.

drivers/md/Makefile

@@ -18,7 +18,7 @@ dm-cache-cleaner-y += dm-cache-policy-cleaner.o
 dm-era-y	+= dm-era-target.o
 dm-verity-y	+= dm-verity-target.o
 md-mod-y	+= md.o bitmap.o
-raid456-y	+= raid5.o raid5-cache.o
+raid456-y	+= raid5.o raid5-cache.o raid5-ppl.o
 
 # Note: link order is important.  All raid personalities
 # and must come before md.o, as they each initialise 

drivers/md/raid5-log.h

@@ -31,6 +31,20 @@ extern struct md_sysfs_entry r5c_journal_mode;
 extern void r5c_update_on_rdev_error(struct mddev *mddev);
 extern bool r5c_big_stripe_cached(struct r5conf *conf, sector_t sect);
 
+extern struct dma_async_tx_descriptor *
+ops_run_partial_parity(struct stripe_head *sh, struct raid5_percpu *percpu,
+		       struct dma_async_tx_descriptor *tx);
+extern int ppl_init_log(struct r5conf *conf);
+extern void ppl_exit_log(struct r5conf *conf);
+extern int ppl_write_stripe(struct r5conf *conf, struct stripe_head *sh);
+extern void ppl_write_stripe_run(struct r5conf *conf);
+extern void ppl_stripe_write_finished(struct stripe_head *sh);
+
+static inline bool raid5_has_ppl(struct r5conf *conf)
+{
+	return test_bit(MD_HAS_PPL, &conf->mddev->flags);
+}
+
 static inline int log_stripe(struct stripe_head *sh, struct stripe_head_state *s)
 {
 	struct r5conf *conf = sh->raid_conf;
@@ -45,6 +59,8 @@ static inline int log_stripe(struct stripe_head *sh, struct stripe_head_state *s
 			/* caching phase */
 			return r5c_cache_data(conf->log, sh);
 		}
+	} else if (raid5_has_ppl(conf)) {
+		return ppl_write_stripe(conf, sh);
 	}
 
 	return -EAGAIN;
@@ -56,24 +72,32 @@ static inline void log_stripe_write_finished(struct stripe_head *sh)
 
 	if (conf->log)
 		r5l_stripe_write_finished(sh);
+	else if (raid5_has_ppl(conf))
+		ppl_stripe_write_finished(sh);
 }
 
 static inline void log_write_stripe_run(struct r5conf *conf)
 {
 	if (conf->log)
 		r5l_write_stripe_run(conf->log);
+	else if (raid5_has_ppl(conf))
+		ppl_write_stripe_run(conf);
 }
 
 static inline void log_exit(struct r5conf *conf)
 {
 	if (conf->log)
 		r5l_exit_log(conf);
+	else if (raid5_has_ppl(conf))
+		ppl_exit_log(conf);
 }
 
 static inline int log_init(struct r5conf *conf, struct md_rdev *journal_dev)
 {
 	if (journal_dev)
 		return r5l_init_log(conf, journal_dev);
+	else if (raid5_has_ppl(conf))
+		return ppl_init_log(conf);
 
 	return 0;
 }

drivers/md/raid5.c

@@ -482,6 +482,11 @@ static void shrink_buffers(struct stripe_head *sh)
 		sh->dev[i].page = NULL;
 		put_page(p);
 	}
+
+	if (sh->ppl_page) {
+		put_page(sh->ppl_page);
+		sh->ppl_page = NULL;
+	}
 }
 
 static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
@@ -498,6 +503,13 @@ static int grow_buffers(struct stripe_head *sh, gfp_t gfp)
 		sh->dev[i].page = page;
 		sh->dev[i].orig_page = page;
 	}
+
+	if (raid5_has_ppl(sh->raid_conf)) {
+		sh->ppl_page = alloc_page(gfp);
+		if (!sh->ppl_page)
+			return 1;
+	}
+
 	return 0;
 }
 
@@ -746,7 +758,7 @@ static bool stripe_can_batch(struct stripe_head *sh)
 {
 	struct r5conf *conf = sh->raid_conf;
 
-	if (conf->log)
+	if (conf->log || raid5_has_ppl(conf))
 		return false;
 	return test_bit(STRIPE_BATCH_READY, &sh->state) &&
 		!test_bit(STRIPE_BITMAP_PENDING, &sh->state) &&
@@ -2093,6 +2105,9 @@ static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
 			async_tx_ack(tx);
 	}
 
+	if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request))
+		tx = ops_run_partial_parity(sh, percpu, tx);
+
 	if (test_bit(STRIPE_OP_PREXOR, &ops_request)) {
 		if (level < 6)
 			tx = ops_run_prexor5(sh, percpu, tx);
@@ -3168,6 +3183,12 @@ schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s,
 		s->locked++;
 	}
 
+	if (raid5_has_ppl(sh->raid_conf) &&
+	    test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) &&
+	    !test_bit(STRIPE_FULL_WRITE, &sh->state) &&
+	    test_bit(R5_Insync, &sh->dev[pd_idx].flags))
+		set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request);
+
 	pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n",
 		__func__, (unsigned long long)sh->sector,
 		s->locked, s->ops_request);
@@ -3215,6 +3236,36 @@ static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx,
 	if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi))
 		goto overlap;
 
+	if (forwrite && raid5_has_ppl(conf)) {
+		/*
+		 * With PPL only writes to consecutive data chunks within a
+		 * stripe are allowed because for a single stripe_head we can
+		 * only have one PPL entry at a time, which describes one data
+		 * range. Not really an overlap, but wait_for_overlap can be
+		 * used to handle this.
+		 */
+		sector_t sector;
+		sector_t first = 0;
+		sector_t last = 0;
+		int count = 0;
+		int i;
+
+		for (i = 0; i < sh->disks; i++) {
+			if (i != sh->pd_idx &&
+			    (i == dd_idx || sh->dev[i].towrite)) {
+				sector = sh->dev[i].sector;
+				if (count == 0 || sector < first)
+					first = sector;
+				if (sector > last)
+					last = sector;
+				count++;
+			}
+		}
+
+		if (first + conf->chunk_sectors * (count - 1) != last)
+			goto overlap;
+	}
+
 	if (!forwrite || previous)
 		clear_bit(STRIPE_BATCH_READY, &sh->state);
 
@@ -7208,6 +7259,13 @@ static int raid5_run(struct mddev *mddev)
 		BUG_ON(mddev->delta_disks != 0);
 	}
 
+	if (test_bit(MD_HAS_JOURNAL, &mddev->flags) &&
+	    test_bit(MD_HAS_PPL, &mddev->flags)) {
+		pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n",
+			mdname(mddev));
+		clear_bit(MD_HAS_PPL, &mddev->flags);
+	}
+
 	if (mddev->private == NULL)
 		conf = setup_conf(mddev);
 	else
@@ -7689,7 +7747,7 @@ static int raid5_resize(struct mddev *mddev, sector_t sectors)
 	sector_t newsize;
 	struct r5conf *conf = mddev->private;
 
-	if (conf->log)
+	if (conf->log || raid5_has_ppl(conf))
 		return -EINVAL;
 	sectors &= ~((sector_t)conf->chunk_sectors - 1);
 	newsize = raid5_size(mddev, sectors, mddev->raid_disks);
@@ -7740,7 +7798,7 @@ static int check_reshape(struct mddev *mddev)
 {
 	struct r5conf *conf = mddev->private;
 
-	if (conf->log)
+	if (conf->log || raid5_has_ppl(conf))
 		return -EINVAL;
 	if (mddev->delta_disks == 0 &&
 	    mddev->new_layout == mddev->layout &&

drivers/md/raid5.h

@@ -224,10 +224,16 @@ struct stripe_head {
 	spinlock_t		batch_lock; /* only header's lock is useful */
 	struct list_head	batch_list; /* protected by head's batch lock*/
 
+	union {
 		struct r5l_io_unit	*log_io;
+		struct ppl_io_unit	*ppl_io;
+	};
+
 	struct list_head	log_list;
 	sector_t		log_start; /* first meta block on the journal */
 	struct list_head	r5c; /* for r5c_cache->stripe_in_journal */
+
+	struct page		*ppl_page; /* partial parity of this stripe */
 	/**
 	 * struct stripe_operations
 	 * @target - STRIPE_OP_COMPUTE_BLK target
@@ -400,6 +406,7 @@ enum {
 	STRIPE_OP_BIODRAIN,
 	STRIPE_OP_RECONSTRUCT,
 	STRIPE_OP_CHECK,
+	STRIPE_OP_PARTIAL_PARITY,
 };
 
 /*
@@ -696,6 +703,7 @@ struct r5conf {
 	int			group_cnt;
 	int			worker_cnt_per_group;
 	struct r5l_log		*log;
+	void			*log_private;
 
 	spinlock_t		pending_bios_lock;
 	bool			batch_bio_dispatch;

include/uapi/linux/raid/md_p.h

@@ -398,4 +398,31 @@ struct r5l_meta_block {
 
 #define R5LOG_VERSION 0x1
 #define R5LOG_MAGIC 0x6433c509
+
+struct ppl_header_entry {
+	__le64 data_sector;	/* raid sector of the new data */
+	__le32 pp_size;		/* length of partial parity */
+	__le32 data_size;	/* length of data */
+	__le32 parity_disk;	/* member disk containing parity */
+	__le32 checksum;	/* checksum of partial parity data for this
+				 * entry (~crc32c) */
+} __attribute__ ((__packed__));
+
+#define PPL_HEADER_SIZE 4096
+#define PPL_HDR_RESERVED 512
+#define PPL_HDR_ENTRY_SPACE \
+	(PPL_HEADER_SIZE - PPL_HDR_RESERVED - 4 * sizeof(u32) - sizeof(u64))
+#define PPL_HDR_MAX_ENTRIES \
+	(PPL_HDR_ENTRY_SPACE / sizeof(struct ppl_header_entry))
+
+struct ppl_header {
+	__u8 reserved[PPL_HDR_RESERVED];/* reserved space, fill with 0xff */
+	__le32 signature;		/* signature (family number of volume) */
+	__le32 padding;			/* zero pad */
+	__le64 generation;		/* generation number of the header */
+	__le32 entries_count;		/* number of entries in entry array */
+	__le32 checksum;		/* checksum of the header (~crc32c) */
+	struct ppl_header_entry entries[PPL_HDR_MAX_ENTRIES];
+} __attribute__ ((__packed__));
+
 #endif