[PATCH] read-ahead and failfast

Here's the patch to enable failfast flag in the bio submission code, and
use it for multipath and readahead.

drivers/block/ll_rw_blk.c

@@ -689,7 +689,7 @@ void blk_queue_invalidate_tags(request_queue_t *q)
 
 static char *rq_flags[] = {
 	"REQ_RW",
-	"REQ_RW_AHEAD",
+	"REQ_FAILFAST",
 	"REQ_SOFTBARRIER",
 	"REQ_HARDBARRIER",
 	"REQ_CMD",
@@ -706,6 +706,10 @@ static char *rq_flags[] = {
 	"REQ_DRIVE_CMD",
 	"REQ_DRIVE_TASK",
 	"REQ_DRIVE_TASKFILE",
+	"REQ_PREEMPT",
+	"REQ_PM_SUSPEND",
+	"REQ_PM_RESUME",
+	"REQ_PM_SHUTDOWN",
 };
 
 void blk_dump_rq_flags(struct request *rq, char *msg)
@@ -1793,7 +1797,7 @@ void __blk_attempt_remerge(request_queue_t *q, struct request *rq)
 static int __make_request(request_queue_t *q, struct bio *bio)
 {
 	struct request *req, *freereq = NULL;
-	int el_ret, rw, nr_sectors, cur_nr_sectors, barrier;
+	int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, ra;
 	struct list_head *insert_here;
 	sector_t sector;
 
@@ -1814,6 +1818,8 @@ static int __make_request(request_queue_t *q, struct bio *bio)
 
 	barrier = test_bit(BIO_RW_BARRIER, &bio->bi_rw);
 
+	ra = bio_flagged(bio, BIO_RW_AHEAD) || current->flags & PF_READAHEAD;
+
 again:
 	insert_here = NULL;
 	spin_lock_irq(q->queue_lock);
@@ -1901,7 +1907,7 @@ static int __make_request(request_queue_t *q, struct bio *bio)
 			/*
 			 * READA bit set
 			 */
-			if (bio_flagged(bio, BIO_RW_AHEAD))
+			if (ra)
 				goto end_io;
 	
 			freereq = get_request_wait(q, rw);
@@ -1921,6 +1927,12 @@ static int __make_request(request_queue_t *q, struct bio *bio)
 	if (barrier)
 		req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
 
+	/*
+	 * don't stack up retries for read ahead
+	 */
+	if (ra)
+		req->flags |= REQ_FAILFAST;
+
 	req->errors = 0;
 	req->hard_sector = req->sector = sector;
 	req->hard_nr_sectors = req->nr_sectors = nr_sectors;

drivers/ide/ide-cd.c

@@ -767,6 +767,9 @@ static int cdrom_decode_status(ide_drive_t *drive, int good_stat, int *stat_ret)
 	} else if (blk_fs_request(rq)) {
 		/* Handle errors from READ and WRITE requests. */
 
+		if (blk_noretry_request(rq))
+			cdrom_end_request(drive, 0);
+
 		if (sense_key == NOT_READY) {
 			/* Tray open. */
 			cdrom_saw_media_change (drive);

drivers/ide/ide-disk.c

@@ -907,7 +907,7 @@ ide_startstop_t idedisk_error (ide_drive_t *drive, const char *msg, u8 stat)
 		/* force an abort */
 		hwif->OUTB(WIN_IDLEIMMEDIATE,IDE_COMMAND_REG);
 	}
-	if (rq->errors >= ERROR_MAX)
+	if (rq->errors >= ERROR_MAX || blk_noretry_request(rq))
 		DRIVER(drive)->end_request(drive, 0, 0);
 	else {
 		if ((rq->errors & ERROR_RESET) == ERROR_RESET) {

drivers/ide/ide-io.c

@@ -58,8 +58,7 @@
 
 #if (DISK_RECOVERY_TIME > 0)
 
-Error So the User Has To Fix the Compilation And Stop Hacking Port 0x43
-Does anyone ever use this anyway ??
+#error So the User Has To Fix the Compilation And Stop Hacking Port 0x43. Does anyone ever use this anyway ??
 
 /*
  * For really screwy hardware (hey, at least it *can* be used with Linux)
@@ -113,6 +112,13 @@ int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
 	if (!nr_sectors)
 		nr_sectors = rq->hard_cur_sectors;
 
+	/*
+	 * if failfast is set on a request, override number of sectors and
+	 * complete the whole request right now
+	 */
+	if (blk_noretry_request(rq) && !uptodate)
+		nr_sectors = rq->hard_nr_sectors;
+
 	/*
 	 * decide whether to reenable DMA -- 3 is a random magic for now,
 	 * if we DMA timeout more than 3 times, just stay in PIO

drivers/md/multipath.c

@@ -179,6 +179,7 @@ static int multipath_make_request (request_queue_t *q, struct bio * bio)
 
 	mp_bh->bio = *bio;
 	mp_bh->bio.bi_bdev = multipath->rdev->bdev;
+	mp_bh->bio.bi_flags |= (1 << BIO_RW_FAILFAST);
 	mp_bh->bio.bi_end_io = multipath_end_request;
 	mp_bh->bio.bi_private = mp_bh;
 	generic_make_request(&mp_bh->bio);

include/linux/bio.h

@@ -119,10 +119,12 @@ struct bio {
  * bit 0 -- read (not set) or write (set)
  * bit 1 -- rw-ahead when set
  * bit 2 -- barrier
+ * bit 3 -- fail fast, don't want low level driver retries
  */
 #define BIO_RW		0
 #define BIO_RW_AHEAD	1
 #define BIO_RW_BARRIER	2
+#define BIO_RW_FAILFAST	3
 
 /*
  * various member access, note that bio_data should of course not be used

include/linux/blkdev.h

@@ -166,7 +166,7 @@ struct request {
  */
 enum rq_flag_bits {
 	__REQ_RW,		/* not set, read. set, write */
-	__REQ_RW_AHEAD,	/* READA */
+	__REQ_FAILFAST,		/* no low level driver retries */
 	__REQ_SOFTBARRIER,	/* may not be passed by ioscheduler */
 	__REQ_HARDBARRIER,	/* may not be passed by drive either */
 	__REQ_CMD,		/* is a regular fs rw request */
@@ -195,7 +195,7 @@ enum rq_flag_bits {
 };
 
 #define REQ_RW		(1 << __REQ_RW)
-#define REQ_RW_AHEAD	(1 << __REQ_RW_AHEAD)
+#define REQ_FAILFAST	(1 << __REQ_FAILFAST)
 #define REQ_SOFTBARRIER	(1 << __REQ_SOFTBARRIER)
 #define REQ_HARDBARRIER	(1 << __REQ_HARDBARRIER)
 #define REQ_CMD		(1 << __REQ_CMD)
@@ -364,6 +364,7 @@ struct request_queue
 #define blk_queue_tagged(q)	test_bit(QUEUE_FLAG_QUEUED, &(q)->queue_flags)
 #define blk_fs_request(rq)	((rq)->flags & REQ_CMD)
 #define blk_pc_request(rq)	((rq)->flags & REQ_BLOCK_PC)
+#define blk_noretry_request(rq)	((rq)->flags & REQ_FAILFAST)
 
 #define blk_pm_suspend_request(rq)	((rq)->flags & REQ_PM_SUSPEND)
 #define blk_pm_resume_request(rq)	((rq)->flags & REQ_PM_RESUME)

include/linux/sched.h

@@ -487,6 +487,7 @@ do { if (atomic_dec_and_test(&(tsk)->usage)) __put_task_struct(tsk); } while(0)
 #define PF_SWAPOFF	0x00080000	/* I am in swapoff */
 #define PF_LESS_THROTTLE 0x01000000	/* Throttle me less: I clena memory */
 #define PF_SYNCWRITE	0x00200000	/* I am doing a sync write */
+#define PF_READAHEAD	0x00400000	/* I am doing read-ahead */
 
 #ifdef CONFIG_SMP
 extern int set_cpus_allowed(task_t *p, unsigned long new_mask);

mm/readahead.c

@@ -94,9 +94,14 @@ static int read_pages(struct address_space *mapping, struct file *filp,
 {
 	unsigned page_idx;
 	struct pagevec lru_pvec;
+	int ret = 0;
+
+	current->flags |= PF_READAHEAD;
 
-	if (mapping->a_ops->readpages)
-		return mapping->a_ops->readpages(filp, mapping, pages, nr_pages);
+	if (mapping->a_ops->readpages) {
+		ret = mapping->a_ops->readpages(filp, mapping, pages, nr_pages);
+		goto out;
+	}
 
 	pagevec_init(&lru_pvec, 0);
 	for (page_idx = 0; page_idx < nr_pages; page_idx++) {
@@ -112,7 +117,9 @@ static int read_pages(struct address_space *mapping, struct file *filp,
 		}
 	}
 	pagevec_lru_add(&lru_pvec);
-	return 0;
+out:
+	current->flags &= ~PF_READAHEAD;
+	return ret;
 }
 
 /*