[PATCH 0/4] better discard support

[PATCH 0/4] better discard support

[Christoph Hellwig]

Trying to make progress on discard again.  This is mostly aimed at the
batched discard and SCSI arrays, but the ATA work should also fit in.

Since last time the first two patches got merged in mainline so the
patch series is a bit smaller now.  The first one is new and based on
a patch from Willy - we now remove the prepare_discard function, and instead
pass the discard request down as normal FS request.  To make that
actually work nicely with the block layer we have to allocate the payload
in the caller.  From the layering POV that makes a lot more sense as that
is where the payload is allocated for all other request, and in practice
it works because the single page and single sector is enough for all
implementations that we care about now.  But it could become quite
nasty if we get an implementation that doesn't fir into this scheme.

Second patch is the separate limit for the length of discard requests
as they can and should be much larger than normal I/O requests.

Patch 3 and 4 are only included for reference to show a working setup
wit hthe block bits.  The SD patch will need quite a bit of work to
look at the unmap limits that are now in SPC.

[PATCH 1/4] block: use normal I/O path for discard requests

[Christoph Hellwig]

[-- Attachment #1: discard-remove-prepare-discard --]
[-- Type: text/plain, Size: 9875 bytes --]

prepare_discard_fn() was being called in a place where memory allocation
was effectively impossible.  This makes it inappropriate for all but
the most trivial translations of Linux's DISCARD operation to the block
command set.  Additionally adding a payload there makes the ownership
of the bio backing unclear as it's now allocated by the device driver
and not the submitter as usual.

It is replaced with QUEUE_FLAG_DISCARD which is used to indicate whether
the queue supports discard operations or not.  blkdev_issue_discard now
allocates a one-page, sector-length payload which is the right thing
for the common ATA and SCSI implementations.

The mtd implementation of prepare_discard_fn() is replaced with simply
checking for the request being a discard.

Largely based on a previous patch from Matthew Wilcox <matthew@wil.cx>
which did the prepare_discard_fn but not the different payload allocation
yet.

Signed-off-by: Christoph Hellwig <hch@lst.de>

-- 

 block/blk-barrier.c         |   37 +++++++++++++++++++++++++++++++------
 block/blk-core.c            |    3 +--
 block/blk-settings.c        |   17 -----------------
 drivers/mtd/mtd_blkdevs.c   |   19 +++++--------------
 drivers/staging/dst/dcore.c |    2 +-
 include/linux/blkdev.h      |    6 ++----
6 files changed, 40 insertions(+), 44 deletions(-)

Index: linux-2.6/block/blk-barrier.c
===================================================================
--- linux-2.6.orig/block/blk-barrier.c	2009-09-17 11:47:20.557003984 -0300
+++ linux-2.6/block/blk-barrier.c	2009-09-17 12:19:18.659011091 -0300
@@ -350,6 +350,7 @@ static void blkdev_discard_end_io(struct
 
 	if (bio->bi_private)
 		complete(bio->bi_private);
+	__free_page(bio_page(bio));
 
 	bio_put(bio);
 }
@@ -372,26 +373,44 @@ int blkdev_issue_discard(struct block_de
 	struct request_queue *q = bdev_get_queue(bdev);
 	int type = flags & DISCARD_FL_BARRIER ?
 		DISCARD_BARRIER : DISCARD_NOBARRIER;
+	struct bio *bio;
+	struct page *page;
 	int ret = 0;
 
 	if (!q)
 		return -ENXIO;
 
-	if (!q->prepare_discard_fn)
+	if (!blk_queue_discard(q))
 		return -EOPNOTSUPP;
 
 	while (nr_sects && !ret) {
-		struct bio *bio = bio_alloc(gfp_mask, 0);
-		if (!bio)
-			return -ENOMEM;
+		unsigned int sector_size = q->limits.logical_block_size;
 
+		bio = bio_alloc(gfp_mask, 1);
+		if (!bio)
+			goto out;
+		bio->bi_sector = sector;
 		bio->bi_end_io = blkdev_discard_end_io;
 		bio->bi_bdev = bdev;
 		if (flags & DISCARD_FL_WAIT)
 			bio->bi_private = &wait;
 
-		bio->bi_sector = sector;
-
+		/*
+		 * Add a zeroed one-sector payload as that's what
+		 * our current implementations need.  If we'll ever need
+		 * more the interface will need revisiting.
+		 */
+		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+		if (!page)
+			goto out_free_bio;
+		if (bio_add_pc_page(q, bio, page, sector_size, 0) < sector_size)
+			goto out_free_page;
+
+		/*
+		 * And override the bio size - the way discard works we
+		 * touch many more blocks on disk than the actual payload
+		 * length.
+		 */
 		if (nr_sects > queue_max_hw_sectors(q)) {
 			bio->bi_size = queue_max_hw_sectors(q) << 9;
 			nr_sects -= queue_max_hw_sectors(q);
@@ -414,5 +433,11 @@ int blkdev_issue_discard(struct block_de
 		bio_put(bio);
 	}
 	return ret;
+out_free_page:
+	__free_page(page);
+out_free_bio:
+	bio_put(bio);
+out:
+	return -ENOMEM;
 }
 EXPORT_SYMBOL(blkdev_issue_discard);
Index: linux-2.6/block/blk-core.c
===================================================================
--- linux-2.6.orig/block/blk-core.c	2009-09-17 11:47:20.594004647 -0300
+++ linux-2.6/block/blk-core.c	2009-09-17 12:15:20.485003934 -0300
@@ -1124,7 +1124,6 @@ void init_request_from_bio(struct reques
 		req->cmd_flags |= REQ_DISCARD;
 		if (bio_rw_flagged(bio, BIO_RW_BARRIER))
 			req->cmd_flags |= REQ_SOFTBARRIER;
-		req->q->prepare_discard_fn(req->q, req);
 	} else if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)))
 		req->cmd_flags |= REQ_HARDBARRIER;
 
@@ -1470,7 +1469,7 @@ static inline void __generic_make_reques
 			goto end_io;
 
 		if (bio_rw_flagged(bio, BIO_RW_DISCARD) &&
-		    !q->prepare_discard_fn) {
+		    !blk_queue_discard(q)) {
 			err = -EOPNOTSUPP;
 			goto end_io;
 		}
Index: linux-2.6/block/blk-settings.c
===================================================================
--- linux-2.6.orig/block/blk-settings.c	2009-09-17 11:47:20.601004002 -0300
+++ linux-2.6/block/blk-settings.c	2009-09-17 12:15:20.494003963 -0300
@@ -34,23 +34,6 @@ void blk_queue_prep_rq(struct request_qu
 EXPORT_SYMBOL(blk_queue_prep_rq);
 
 /**
- * blk_queue_set_discard - set a discard_sectors function for queue
- * @q:		queue
- * @dfn:	prepare_discard function
- *
- * It's possible for a queue to register a discard callback which is used
- * to transform a discard request into the appropriate type for the
- * hardware. If none is registered, then discard requests are failed
- * with %EOPNOTSUPP.
- *
- */
-void blk_queue_set_discard(struct request_queue *q, prepare_discard_fn *dfn)
-{
-	q->prepare_discard_fn = dfn;
-}
-EXPORT_SYMBOL(blk_queue_set_discard);
-
-/**
  * blk_queue_merge_bvec - set a merge_bvec function for queue
  * @q:		queue
  * @mbfn:	merge_bvec_fn
Index: linux-2.6/drivers/mtd/mtd_blkdevs.c
===================================================================
--- linux-2.6.orig/drivers/mtd/mtd_blkdevs.c	2009-09-17 11:47:20.607004137 -0300
+++ linux-2.6/drivers/mtd/mtd_blkdevs.c	2009-09-17 11:47:34.453004247 -0300
@@ -32,14 +32,6 @@ struct mtd_blkcore_priv {
 	spinlock_t queue_lock;
 };
 
-static int blktrans_discard_request(struct request_queue *q,
-				    struct request *req)
-{
-	req->cmd_type = REQ_TYPE_LINUX_BLOCK;
-	req->cmd[0] = REQ_LB_OP_DISCARD;
-	return 0;
-}
-
 static int do_blktrans_request(struct mtd_blktrans_ops *tr,
 			       struct mtd_blktrans_dev *dev,
 			       struct request *req)
@@ -52,10 +44,6 @@ static int do_blktrans_request(struct mt
 
 	buf = req->buffer;
 
-	if (req->cmd_type == REQ_TYPE_LINUX_BLOCK &&
-	    req->cmd[0] == REQ_LB_OP_DISCARD)
-		return tr->discard(dev, block, nsect);
-
 	if (!blk_fs_request(req))
 		return -EIO;
 
@@ -63,6 +51,9 @@ static int do_blktrans_request(struct mt
 	    get_capacity(req->rq_disk))
 		return -EIO;
 
+	if (blk_discard_rq(req))
+		return tr->discard(dev, block, nsect);
+
 	switch(rq_data_dir(req)) {
 	case READ:
 		for (; nsect > 0; nsect--, block++, buf += tr->blksize)
@@ -380,8 +371,8 @@ int register_mtd_blktrans(struct mtd_blk
 	tr->blkcore_priv->rq->queuedata = tr;
 	blk_queue_logical_block_size(tr->blkcore_priv->rq, tr->blksize);
 	if (tr->discard)
-		blk_queue_set_discard(tr->blkcore_priv->rq,
-				      blktrans_discard_request);
+		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
+					tr->blkcore_priv->rq);
 
 	tr->blkshift = ffs(tr->blksize) - 1;
 
Index: linux-2.6/drivers/staging/dst/dcore.c
===================================================================
--- linux-2.6.orig/drivers/staging/dst/dcore.c	2009-09-17 11:47:20.615002991 -0300
+++ linux-2.6/drivers/staging/dst/dcore.c	2009-09-17 11:47:34.456003163 -0300
@@ -102,7 +102,7 @@ static int dst_request(struct request_qu
 	struct dst_node *n = q->queuedata;
 	int err = -EIO;
 
-	if (bio_empty_barrier(bio) && !q->prepare_discard_fn) {
+	if (bio_empty_barrier(bio) && !blk_queue_discard(q)) {
 		/*
 		 * This is a dirty^Wnice hack, but if we complete this
 		 * operation with -EOPNOTSUPP like intended, XFS
Index: linux-2.6/include/linux/blkdev.h
===================================================================
--- linux-2.6.orig/include/linux/blkdev.h	2009-09-17 11:47:20.622004162 -0300
+++ linux-2.6/include/linux/blkdev.h	2009-09-17 12:15:20.542254662 -0300
@@ -82,7 +82,6 @@ enum rq_cmd_type_bits {
 enum {
 	REQ_LB_OP_EJECT	= 0x40,		/* eject request */
 	REQ_LB_OP_FLUSH = 0x41,		/* flush request */
-	REQ_LB_OP_DISCARD = 0x42,	/* discard sectors */
 };
 
 /*
@@ -261,7 +260,6 @@ typedef void (request_fn_proc) (struct r
 typedef int (make_request_fn) (struct request_queue *q, struct bio *bio);
 typedef int (prep_rq_fn) (struct request_queue *, struct request *);
 typedef void (unplug_fn) (struct request_queue *);
-typedef int (prepare_discard_fn) (struct request_queue *, struct request *);
 
 struct bio_vec;
 struct bvec_merge_data {
@@ -340,7 +338,6 @@ struct request_queue
 	make_request_fn		*make_request_fn;
 	prep_rq_fn		*prep_rq_fn;
 	unplug_fn		*unplug_fn;
-	prepare_discard_fn	*prepare_discard_fn;
 	merge_bvec_fn		*merge_bvec_fn;
 	prepare_flush_fn	*prepare_flush_fn;
 	softirq_done_fn		*softirq_done_fn;
@@ -460,6 +457,7 @@ struct request_queue
 #define QUEUE_FLAG_VIRT        QUEUE_FLAG_NONROT /* paravirt device */
 #define QUEUE_FLAG_IO_STAT     15	/* do IO stats */
 #define QUEUE_FLAG_CQ	       16	/* hardware does queuing */
+#define QUEUE_FLAG_DISCARD     17	/* supports DISCARD */
 
 #define QUEUE_FLAG_DEFAULT	((1 << QUEUE_FLAG_IO_STAT) |		\
 				 (1 << QUEUE_FLAG_CLUSTER) |		\
@@ -591,6 +589,7 @@ enum {
 #define blk_queue_flushing(q)	((q)->ordseq)
 #define blk_queue_stackable(q)	\
 	test_bit(QUEUE_FLAG_STACKABLE, &(q)->queue_flags)
+#define blk_queue_discard(q)	test_bit(QUEUE_FLAG_DISCARD, &(q)->queue_flags)
 
 #define blk_fs_request(rq)	((rq)->cmd_type == REQ_TYPE_FS)
 #define blk_pc_request(rq)	((rq)->cmd_type == REQ_TYPE_BLOCK_PC)
@@ -955,7 +954,6 @@ extern void blk_queue_merge_bvec(struct 
 extern void blk_queue_dma_alignment(struct request_queue *, int);
 extern void blk_queue_update_dma_alignment(struct request_queue *, int);
 extern void blk_queue_softirq_done(struct request_queue *, softirq_done_fn *);
-extern void blk_queue_set_discard(struct request_queue *, prepare_discard_fn *);
 extern void blk_queue_rq_timed_out(struct request_queue *, rq_timed_out_fn *);
 extern void blk_queue_rq_timeout(struct request_queue *, unsigned int);
 extern struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev);

[PATCH 2/4] block: allow large discard requests

[Christoph Hellwig]

[-- Attachment #1: discard-allow-large-requests --]
[-- Type: text/plain, Size: 4775 bytes --]

Currently we set the bio size to the byte equivalent of the blocks to
be trimmed when submitting the initial DISCARD ioctl.  That means it
is subject to the max_hw_sectors limitation of the HBA which is
much lower than the size of a DISCARD request we can support.
Add a separate max_discard_sectors tunable to limit the size for discard
requests. 

We limit the max discard request size in bytes to 32bit as that is the
limit for bio->bi_size.  This could be much larger if we had a way to pass
that information through the block layer.

Signed-off-by: Christoph Hellwig <hch@lst.de>

Index: linux-2.6/block/blk-core.c
===================================================================
--- linux-2.6.orig/block/blk-core.c	2009-09-17 12:15:20.485003934 -0300
+++ linux-2.6/block/blk-core.c	2009-09-17 12:23:33.018028602 -0300
@@ -1436,7 +1436,8 @@ static inline void __generic_make_reques
 			goto end_io;
 		}
 
-		if (unlikely(nr_sectors > queue_max_hw_sectors(q))) {
+		if (unlikely(!bio_rw_flagged(bio, BIO_RW_DISCARD) &&
+			     nr_sectors > queue_max_hw_sectors(q))) {
 			printk(KERN_ERR "bio too big device %s (%u > %u)\n",
 			       bdevname(bio->bi_bdev, b),
 			       bio_sectors(bio),
Index: linux-2.6/block/blk-settings.c
===================================================================
--- linux-2.6.orig/block/blk-settings.c	2009-09-17 12:15:20.494003963 -0300
+++ linux-2.6/block/blk-settings.c	2009-09-17 13:12:00.770254159 -0300
@@ -95,6 +95,7 @@ void blk_set_default_limits(struct queue
 	lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
 	lim->max_segment_size = MAX_SEGMENT_SIZE;
 	lim->max_sectors = lim->max_hw_sectors = SAFE_MAX_SECTORS;
+	lim->max_discard_sectors = SAFE_MAX_SECTORS;
 	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
 	lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
 	lim->alignment_offset = 0;
@@ -237,6 +238,18 @@ void blk_queue_max_hw_sectors(struct req
 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
 
 /**
+ * blk_queue_max_discard_sectors - set max sectors for a single discard
+ * @q:  the request queue for the device
+ * @max_discard: maximum number of sectors to discard
+ **/
+void blk_queue_max_discard_sectors(struct request_queue *q,
+		unsigned int max_discard_sectors)
+{
+	q->limits.max_discard_sectors = max_discard_sectors;
+}
+EXPORT_SYMBOL(blk_queue_max_discard_sectors);
+
+/**
  * blk_queue_max_phys_segments - set max phys segments for a request for this queue
  * @q:  the request queue for the device
  * @max_segments:  max number of segments
Index: linux-2.6/include/linux/blkdev.h
===================================================================
--- linux-2.6.orig/include/linux/blkdev.h	2009-09-17 12:15:20.542254662 -0300
+++ linux-2.6/include/linux/blkdev.h	2009-09-17 13:12:15.542006586 -0300
@@ -311,6 +311,7 @@ struct queue_limits {
 	unsigned int		alignment_offset;
 	unsigned int		io_min;
 	unsigned int		io_opt;
+	unsigned int		max_discard_sectors;
 
 	unsigned short		logical_block_size;
 	unsigned short		max_hw_segments;
@@ -928,6 +929,8 @@ extern void blk_queue_max_hw_sectors(str
 extern void blk_queue_max_phys_segments(struct request_queue *, unsigned short);
 extern void blk_queue_max_hw_segments(struct request_queue *, unsigned short);
 extern void blk_queue_max_segment_size(struct request_queue *, unsigned int);
+extern void blk_queue_max_discard_sectors(struct request_queue *q,
+		unsigned int max_discard_sectors);
 extern void blk_queue_logical_block_size(struct request_queue *, unsigned short);
 extern void blk_queue_physical_block_size(struct request_queue *, unsigned short);
 extern void blk_queue_alignment_offset(struct request_queue *q,
Index: linux-2.6/block/blk-barrier.c
===================================================================
--- linux-2.6.orig/block/blk-barrier.c	2009-09-17 12:19:18.659011091 -0300
+++ linux-2.6/block/blk-barrier.c	2009-09-17 12:31:05.660256438 -0300
@@ -385,6 +385,8 @@ int blkdev_issue_discard(struct block_de
 
 	while (nr_sects && !ret) {
 		unsigned int sector_size = q->limits.logical_block_size;
+		unsigned int max_discard_sectors =
+			min(q->limits.max_discard_sectors, UINT_MAX >> 9);
 
 		bio = bio_alloc(gfp_mask, 1);
 		if (!bio)
@@ -411,10 +413,10 @@ int blkdev_issue_discard(struct block_de
 		 * touch many more blocks on disk than the actual payload
 		 * length.
 		 */
-		if (nr_sects > queue_max_hw_sectors(q)) {
-			bio->bi_size = queue_max_hw_sectors(q) << 9;
-			nr_sects -= queue_max_hw_sectors(q);
-			sector += queue_max_hw_sectors(q);
+		if (nr_sects > max_discard_sectors) {
+			bio->bi_size = max_discard_sectors << 9;
+			nr_sects -= max_discard_sectors;
+			sector += max_discard_sectors;
 		} else {
 			bio->bi_size = nr_sects << 9;
 			nr_sects = 0;

[PATCH 3/4] sd: add support for WRITE SAME (16) with unmap bit

[Christoph Hellwig]

[-- Attachment #1: discard-add-scsi-write-same-support --]
[-- Type: text/plain, Size: 4178 bytes --]

Send  WRITE SAME request with the unmap bit set to the device if it
advertises thin provisioning support.  Still pretty hacky and not
actually looking at the INQUIRY data for the unmap parameters for
example.


Signed-off-by: Christoph Hellwig <hch@lst.de>

Index: linux-2.6/drivers/scsi/sd.c
===================================================================
--- linux-2.6.orig/drivers/scsi/sd.c	2009-09-17 13:12:00.743253792 -0300
+++ linux-2.6/drivers/scsi/sd.c	2009-09-17 13:18:31.185005932 -0300
@@ -370,6 +370,35 @@ static void scsi_disk_put(struct scsi_di
 	mutex_unlock(&sd_ref_mutex);
 }
 
+static void sd_prepare_discard(struct request_queue *q, struct request *rq)
+{
+	struct bio *bio = rq->bio;
+
+	rq->cmd_type = REQ_TYPE_BLOCK_PC;
+	rq->timeout = SD_TIMEOUT;
+	rq->cmd[0] = WRITE_SAME_16;
+	rq->cmd[1] = 0x8; /* UNMAP bit */
+	rq->cmd[2] = sizeof(bio->bi_sector) > 4 ?
+			(unsigned char) (bio->bi_sector >> 56) & 0xff : 0;
+	rq->cmd[3] = sizeof(bio->bi_sector) > 4 ?
+			(unsigned char) (bio->bi_sector >> 48) & 0xff : 0;
+	rq->cmd[4] = sizeof(bio->bi_sector) > 4 ?
+			(unsigned char) (bio->bi_sector >> 40) & 0xff : 0;
+	rq->cmd[5] = sizeof(bio->bi_sector) > 4 ?
+			(unsigned char) (bio->bi_sector >> 32) & 0xff : 0;
+	rq->cmd[6] = (unsigned char) (bio->bi_sector >> 24) & 0xff;
+	rq->cmd[7] = (unsigned char) (bio->bi_sector >> 16) & 0xff;
+	rq->cmd[8] = (unsigned char) (bio->bi_sector >> 8) & 0xff;
+	rq->cmd[9] = (unsigned char) bio->bi_sector & 0xff;
+	rq->cmd[10] = (unsigned char) (bio_sectors(bio) >> 24) & 0xff;
+	rq->cmd[11] = (unsigned char) (bio_sectors(bio) >> 16) & 0xff;
+	rq->cmd[12] = (unsigned char) (bio_sectors(bio) >> 8) & 0xff;
+	rq->cmd[13] = (unsigned char) bio_sectors(bio) & 0xff;
+	rq->cmd[14] = 0;
+	rq->cmd[15] = 0;
+	rq->cmd_len = 16;
+}
+
 /**
  *	sd_init_command - build a scsi (read or write) command from
  *	information in the request structure.
@@ -389,6 +418,13 @@ static int sd_prep_fn(struct request_que
 	unsigned int this_count = blk_rq_sectors(rq);
 	int ret, host_dif;
 
+	/*
+	 * Discard request come in as REQ_TYPE_FS but we turn them into
+	 * block PC requests to make life easier.
+	 */
+	if (blk_discard_rq(rq))
+		sd_prepare_discard(q, rq);
+
 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 		ret = scsi_setup_blk_pc_cmnd(sdp, rq);
 		goto out;
@@ -396,6 +432,7 @@ static int sd_prep_fn(struct request_que
 		ret = BLKPREP_KILL;
 		goto out;
 	}
+
 	ret = scsi_setup_fs_cmnd(sdp, rq);
 	if (ret != BLKPREP_OK)
 		goto out;
@@ -1369,6 +1406,9 @@ static int read_capacity_16(struct scsi_
 		sd_printk(KERN_NOTICE, sdkp,
 			  "physical block alignment offset: %u\n", alignment);
 
+	if (buffer[14] & 0x80)
+		sdkp->thin_provisioning = 1;
+
 	sdkp->capacity = lba + 1;
 	return sector_size;
 }
@@ -1916,6 +1956,11 @@ static int sd_revalidate_disk(struct gen
 
 	blk_queue_ordered(sdkp->disk->queue, ordered, sd_prepare_flush);
 
+	if (sdkp->thin_provisioning) {
+		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, sdkp->disk->queue);
+		blk_queue_max_discard_sectors(sdp->request_queue, UINT_MAX);
+	}
+
 	set_capacity(disk, sdkp->capacity);
 	kfree(buffer);
 
Index: linux-2.6/include/scsi/scsi.h
===================================================================
--- linux-2.6.orig/include/scsi/scsi.h	2009-09-17 13:12:00.756253702 -0300
+++ linux-2.6/include/scsi/scsi.h	2009-09-17 13:12:21.031004671 -0300
@@ -122,6 +122,8 @@ struct scsi_cmnd;
 #define READ_16               0x88
 #define WRITE_16              0x8a
 #define VERIFY_16	      0x8f
+#define WRITE_SAME_16	      0x93
+
 #define SERVICE_ACTION_IN     0x9e
 /* values for service action in */
 #define	SAI_READ_CAPACITY_16  0x10
Index: linux-2.6/drivers/scsi/sd.h
===================================================================
--- linux-2.6.orig/drivers/scsi/sd.h	2009-09-17 13:12:00.750254405 -0300
+++ linux-2.6/drivers/scsi/sd.h	2009-09-17 13:12:21.034025726 -0300
@@ -55,6 +55,7 @@ struct scsi_disk {
 	unsigned	RCD : 1;	/* state of disk RCD bit, unused */
 	unsigned	DPOFUA : 1;	/* state of disk DPOFUA bit */
 	unsigned	first_scan : 1;
+	unsigned	thin_provisioning : 1;
 };
 #define to_scsi_disk(obj) container_of(obj,struct scsi_disk,dev)
 

[PATCH 4/4] xfs: add batches discard support

[Christoph Hellwig]

[-- Attachment #1: xfs-add-trim-support-2 --]
[-- Type: text/plain, Size: 6249 bytes --]

Add support for discarding all currently unused space by an ioctl.  Only
intended as demonstration and not for merging.

Use the following small tool to exercise it:


#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdint.h>
#include <sys/ioctl.h>

#define XFS_IOC_TRIM                 _IOR ('X', 126, uint32_t)


int main(int argc, char **argv)
{
	int minsize = 4096;
	int fd;

	if (argc != 2) {
		fprintf(stderr, "usage: %s mountpoint\n", argv[0]);
		return 1;
	}

	fd = open(argv[1], O_RDONLY);
	if (fd < 0) {
		perror("open");
		return 1;
	}

	if (ioctl(fd, XFS_IOC_TRIM, &minsize)) {
		if (errno == EOPNOTSUPP)
			fprintf(stderr, "TRIM not supported\n");
		else
			perror("XFS_IOC_TRIM");
		return 1;
	}

	return 0;
}


Signed-off-by: Christoph Hellwig <hch@lst.de>

Index: linux-2.6/fs/xfs/linux-2.6/xfs_ioctl.c
===================================================================
--- linux-2.6.orig/fs/xfs/linux-2.6/xfs_ioctl.c	2009-08-29 15:53:27.319844716 -0300
+++ linux-2.6/fs/xfs/linux-2.6/xfs_ioctl.c	2009-08-29 16:51:56.271867967 -0300
@@ -1274,6 +1274,31 @@ xfs_ioc_getbmapx(
 	return 0;
 }
 
+STATIC int
+xfs_ioc_trim(
+	struct xfs_mount	*mp,
+	__uint32_t		*argp)
+{
+	xfs_agnumber_t		agno;
+	int			error = 0;
+	__uint32_t		minlen;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+	if (get_user(minlen, argp))
+		return -EFAULT;
+
+	down_read(&mp->m_peraglock);
+	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
+		error = -xfs_trim_extents(mp, agno, minlen);
+		if (error)
+			break;
+	}
+	up_read(&mp->m_peraglock);
+
+	return error;
+}
+
 /*
  * Note: some of the ioctl's return positive numbers as a
  * byte count indicating success, such as readlink_by_handle.
@@ -1523,6 +1548,9 @@ xfs_file_ioctl(
 		error = xfs_errortag_clearall(mp, 1);
 		return -error;
 
+	case XFS_IOC_TRIM:
+		return xfs_ioc_trim(mp, arg);
+
 	default:
 		return -ENOTTY;
 	}
Index: linux-2.6/fs/xfs/xfs_alloc.c
===================================================================
--- linux-2.6.orig/fs/xfs/xfs_alloc.c	2009-08-29 15:53:27.355845733 -0300
+++ linux-2.6/fs/xfs/xfs_alloc.c	2009-08-29 16:59:20.451343922 -0300
@@ -2609,6 +2609,96 @@ error0:
 	return error;
 }
 
+STATIC int
+xfs_trim_extent(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno,
+	xfs_agblock_t		fbno,
+	xfs_extlen_t		flen)
+{
+	xfs_daddr_t		blkno = XFS_AGB_TO_DADDR(mp, agno, fbno);
+	sector_t		nblks = XFS_FSB_TO_BB(mp, flen);
+	int			error;
+
+	xfs_fs_cmn_err(CE_NOTE, mp, "discarding sectors [0x%llx-0x%llx]",
+			blkno, nblks);
+
+	error = -blkdev_issue_discard(mp->m_ddev_targp->bt_bdev, blkno, nblks,
+				      GFP_NOFS, DISCARD_FL_WAIT);
+	if (error && error != EOPNOTSUPP)
+		xfs_fs_cmn_err(CE_NOTE, mp, "discard failed, error %d", error);
+	return error;
+}
+
+/*
+ * Notify the underlying block device about our free extent map.
+ *
+ * This walks all free extents above a minimum threshold and notifies the
+ * underlying device that these blocks are unused.  That information is
+ * useful for SSDs or thinly provisioned storage in high end arrays or
+ * virtualization scenarios.
+ */
+int
+xfs_trim_extents(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno,
+	xfs_extlen_t		minlen)	/* minimum extent size to bother */
+{
+	struct xfs_btree_cur	*cur;	/* cursor for the by-block btree */
+	struct xfs_buf		*agbp;	/* AGF buffer pointer */
+	xfs_agblock_t		bno;	/* block the for next search */
+	xfs_agblock_t		fbno;	/* start block of found extent */
+	xfs_extlen_t		flen;	/* length of found extent */
+	int			error;
+	int			i;
+
+	error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
+	if (error)
+		return error;
+
+	bno = 0;
+	for (;;) {
+		cur = xfs_allocbt_init_cursor(mp, NULL, agbp, agno,
+					      XFS_BTNUM_BNO);
+
+		error = xfs_alloc_lookup_ge(cur, bno, minlen, &i);
+		if (error)
+			goto error0;
+		if (!i) {
+			/*
+			 * No more free extents found: done.
+			 */
+			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
+			break;
+		}
+
+		error = xfs_alloc_get_rec(cur, &fbno, &flen, &i);
+		if (error)
+			goto error0;
+		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
+
+		/*
+		 * Pass if the freespace extent isn't long enough to bother.
+		 */
+		if (flen >= minlen) {
+			error = xfs_trim_extent(mp, agno, fbno, flen);
+			if (error) {
+				xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
+				break;
+			}
+		}
+
+		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
+		bno = fbno + flen;
+	}
+
+out:
+	xfs_buf_relse(agbp);
+	return error;
+error0:
+	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
+	goto out;
+}
 
 /*
  * AG Busy list management
Index: linux-2.6/fs/xfs/xfs_alloc.h
===================================================================
--- linux-2.6.orig/fs/xfs/xfs_alloc.h	2009-08-29 15:53:27.371844485 -0300
+++ linux-2.6/fs/xfs/xfs_alloc.h	2009-08-29 16:51:56.271867967 -0300
@@ -215,4 +215,7 @@ xfs_free_extent(
 	xfs_fsblock_t	bno,	/* starting block number of extent */
 	xfs_extlen_t	len);	/* length of extent */
 
+int xfs_trim_extents(struct xfs_mount *mp, xfs_agnumber_t agno,
+	xfs_extlen_t minlen);
+
 #endif	/* __XFS_ALLOC_H__ */
Index: linux-2.6/fs/xfs/xfs_fs.h
===================================================================
--- linux-2.6.orig/fs/xfs/xfs_fs.h	2009-08-29 15:53:27.391844445 -0300
+++ linux-2.6/fs/xfs/xfs_fs.h	2009-08-29 16:51:56.279865211 -0300
@@ -475,6 +475,7 @@ typedef struct xfs_handle {
 #define XFS_IOC_ATTRMULTI_BY_HANDLE  _IOW ('X', 123, struct xfs_fsop_attrmulti_handlereq)
 #define XFS_IOC_FSGEOMETRY	     _IOR ('X', 124, struct xfs_fsop_geom)
 #define XFS_IOC_GOINGDOWN	     _IOR ('X', 125, __uint32_t)
+#define XFS_IOC_TRIM		     _IOR ('X', 126, __uint32_t)
 /*	XFS_IOC_GETFSUUID ---------- deprecated 140	 */
 
 
Index: linux-2.6/fs/xfs/linux-2.6/xfs_ioctl32.c
===================================================================
--- linux-2.6.orig/fs/xfs/linux-2.6/xfs_ioctl32.c	2009-08-29 15:53:27.339845024 -0300
+++ linux-2.6/fs/xfs/linux-2.6/xfs_ioctl32.c	2009-08-29 16:51:56.283864672 -0300
@@ -563,6 +563,7 @@ xfs_file_compat_ioctl(
 	case XFS_IOC_GOINGDOWN:
 	case XFS_IOC_ERROR_INJECTION:
 	case XFS_IOC_ERROR_CLEARALL:
+	case XFS_IOC_TRIM:
 		return xfs_file_ioctl(filp, cmd, p);
 #ifndef BROKEN_X86_ALIGNMENT
 	/* These are handled fine if no alignment issues */