lightnvm: physical block device (pblk) target

This patch introduces pblk, a host-side translation layer for
Open-Channel SSDs to expose them like block devices. The translation
layer allows data placement decisions, and I/O scheduling to be
managed by the host, enabling users to optimize the SSD for their
specific workloads.

An open-channel SSD has a set of LUNs (parallel units) and a
collection of blocks. Each block can be read in any order, but
writes must be sequential. Writes may also fail, and if a block
requires it, must also be reset before new writes can be
applied.

To manage the constraints, pblk maintains a logical to
physical address (L2P) table,  write cache, garbage
collection logic, recovery scheme, and logic to rate-limit
user I/Os versus garbage collection I/Os.

The L2P table is fully-associative and manages sectors at a
4KB granularity. Pblk stores the L2P table in two places, in
the out-of-band area of the media and on the last page of a
line. In the cause of a power failure, pblk will perform a
scan to recover the L2P table.

The user data is organized into lines. A line is data
striped across blocks and LUNs. The lines enable the host to
reduce the amount of metadata to maintain besides the user
data and makes it easier to implement RAID or erasure coding
in the future.

pblk implements multi-tenant support and can be instantiated
multiple times on the same drive. Each instance owns a
portion of the SSD - both regarding I/O bandwidth and
capacity - providing I/O isolation for each case.

Finally, pblk also exposes a sysfs interface that allows
user-space to peek into the internals of pblk. The interface
is available at /dev/block/*/pblk/ where * is the block
device name exposed.

This work also contains contributions from:
  Matias Bjørling <matias@cnexlabs.com>
  Simon A. F. Lund <slund@cnexlabs.com>
  Young Tack Jin <youngtack.jin@gmail.com>
  Huaicheng Li <huaicheng@cs.uchicago.edu>

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

Documentation/lightnvm/pblk.txt

@@ -0,0 +1,21 @@
+pblk: Physical Block Device Target
+==================================
+
+pblk implements a fully associative, host-based FTL that exposes a traditional
+block I/O interface. Its primary responsibilities are:
+
+  - Map logical addresses onto physical addresses (4KB granularity) in a
+    logical-to-physical (L2P) table.
+  - Maintain the integrity and consistency of the L2P table as well as its
+    recovery from normal tear down and power outage.
+  - Deal with controller- and media-specific constrains.
+  - Handle I/O errors.
+  - Implement garbage collection.
+  - Maintain consistency across the I/O stack during synchronization points.
+
+For more information please refer to:
+
+  http://lightnvm.io
+
+which maintains updated FAQs, manual pages, technical documentation, tools,
+contacts, etc.

drivers/lightnvm/Kconfig

@@ -33,4 +33,13 @@ config NVM_RRPC
 	host. The target is implemented using a linear mapping table and
 	cost-based garbage collection. It is optimized for 4K IO sizes.
 
+config NVM_PBLK
+	tristate "Physical Block Device Open-Channel SSD target"
+	---help---
+	Allows an open-channel SSD to be exposed as a block device to the
+	host. The target assumes the device exposes raw flash and must be
+	explicitly managed by the host.
+
+	Please note the disk format is considered EXPERIMENTAL for now.
+
 endif # NVM

drivers/lightnvm/Makefile

@@ -4,3 +4,8 @@
 
 obj-$(CONFIG_NVM)		:= core.o
 obj-$(CONFIG_NVM_RRPC)		+= rrpc.o
+obj-$(CONFIG_NVM_PBLK)		+= pblk.o
+pblk-y				:= pblk-init.o pblk-core.o pblk-rb.o \
+				   pblk-write.o pblk-cache.o pblk-read.o \
+				   pblk-gc.o pblk-recovery.o pblk-map.o \
+				   pblk-rl.o pblk-sysfs.o

drivers/lightnvm/pblk-cache.c

@@ -0,0 +1,114 @@
+/*
+ * Copyright (C) 2016 CNEX Labs
+ * Initial release: Javier Gonzalez <javier@cnexlabs.com>
+ *                  Matias Bjorling <matias@cnexlabs.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * pblk-cache.c - pblk's write cache
+ */
+
+#include "pblk.h"
+
+int pblk_write_to_cache(struct pblk *pblk, struct bio *bio, unsigned long flags)
+{
+	struct pblk_w_ctx w_ctx;
+	sector_t lba = pblk_get_lba(bio);
+	unsigned int bpos, pos;
+	int nr_entries = pblk_get_secs(bio);
+	int i, ret;
+
+	/* Update the write buffer head (mem) with the entries that we can
+	 * write. The write in itself cannot fail, so there is no need to
+	 * rollback from here on.
+	 */
+retry:
+	ret = pblk_rb_may_write_user(&pblk->rwb, bio, nr_entries, &bpos);
+	if (ret == NVM_IO_REQUEUE) {
+		io_schedule();
+		goto retry;
+	}
+
+	if (unlikely(!bio_has_data(bio)))
+		goto out;
+
+	w_ctx.flags = flags;
+	pblk_ppa_set_empty(&w_ctx.ppa);
+
+	for (i = 0; i < nr_entries; i++) {
+		void *data = bio_data(bio);
+
+		w_ctx.lba = lba + i;
+
+		pos = pblk_rb_wrap_pos(&pblk->rwb, bpos + i);
+		pblk_rb_write_entry_user(&pblk->rwb, data, w_ctx, pos);
+
+		bio_advance(bio, PBLK_EXPOSED_PAGE_SIZE);
+	}
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_add(nr_entries, &pblk->inflight_writes);
+	atomic_long_add(nr_entries, &pblk->req_writes);
+#endif
+
+out:
+	pblk_write_should_kick(pblk);
+	return ret;
+}
+
+/*
+ * On GC the incoming lbas are not necessarily sequential. Also, some of the
+ * lbas might not be valid entries, which are marked as empty by the GC thread
+ */
+int pblk_write_gc_to_cache(struct pblk *pblk, void *data, u64 *lba_list,
+			   unsigned int nr_entries, unsigned int nr_rec_entries,
+			   struct pblk_line *gc_line, unsigned long flags)
+{
+	struct pblk_w_ctx w_ctx;
+	unsigned int bpos, pos;
+	int i, valid_entries;
+
+	/* Update the write buffer head (mem) with the entries that we can
+	 * write. The write in itself cannot fail, so there is no need to
+	 * rollback from here on.
+	 */
+retry:
+	if (!pblk_rb_may_write_gc(&pblk->rwb, nr_rec_entries, &bpos)) {
+		io_schedule();
+		goto retry;
+	}
+
+	w_ctx.flags = flags;
+	pblk_ppa_set_empty(&w_ctx.ppa);
+
+	for (i = 0, valid_entries = 0; i < nr_entries; i++) {
+		if (lba_list[i] == ADDR_EMPTY)
+			continue;
+
+		w_ctx.lba = lba_list[i];
+
+		pos = pblk_rb_wrap_pos(&pblk->rwb, bpos + valid_entries);
+		pblk_rb_write_entry_gc(&pblk->rwb, data, w_ctx, gc_line, pos);
+
+		data += PBLK_EXPOSED_PAGE_SIZE;
+		valid_entries++;
+	}
+
+	WARN_ONCE(nr_rec_entries != valid_entries,
+					"pblk: inconsistent GC write\n");
+
+#ifdef CONFIG_NVM_DEBUG
+	atomic_long_add(valid_entries, &pblk->inflight_writes);
+	atomic_long_add(valid_entries, &pblk->recov_gc_writes);
+#endif
+
+	pblk_write_should_kick(pblk);
+	return NVM_IO_OK;
+}