aboutsummaryrefslogtreecommitdiff
path: root/fs/jfs/jfs_dmap.c
diff options
context:
space:
mode:
authorTino Reichardt <milky-kernel@mcmilk.de>2012-09-17 11:58:19 -0500
committerDave Kleikamp <dave.kleikamp@oracle.com>2012-09-17 11:58:19 -0500
commitb40c2e665cd552eae5fbdbb878bc29a34357668e (patch)
tree23714e5773a77abf1daf71ca7958dcd8dfa688ea /fs/jfs/jfs_dmap.c
parentfbcbe2b3c92ee1c930dcfcf8bb764074c100fd63 (diff)
fs/jfs: TRIM support for JFS Filesystem
This patch adds support for the two linux interfaces of the discard/TRIM command for SSD devices and sparse/thinly-provisioned LUNs. JFS will support batched discard via FITRIM ioctl and online discard with the discard mount option. Signed-off-by: Tino Reichardt <list-jfs@mcmilk.de> Signed-off-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Diffstat (limited to 'fs/jfs/jfs_dmap.c')
-rw-r--r--fs/jfs/jfs_dmap.c125
1 files changed, 121 insertions, 4 deletions
diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
index 9cbd11a3f80..174feb6a73c 100644
--- a/fs/jfs/jfs_dmap.c
+++ b/fs/jfs/jfs_dmap.c
@@ -1,5 +1,6 @@
/*
* Copyright (C) International Business Machines Corp., 2000-2004
+ * Portions Copyright (C) Tino Reichardt, 2012
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
@@ -25,6 +26,7 @@
#include "jfs_lock.h"
#include "jfs_metapage.h"
#include "jfs_debug.h"
+#include "jfs_discard.h"
/*
* SERIALIZATION of the Block Allocation Map.
@@ -104,7 +106,6 @@ static int dbFreeBits(struct bmap * bmp, struct dmap * dp, s64 blkno,
static int dbFreeDmap(struct bmap * bmp, struct dmap * dp, s64 blkno,
int nblocks);
static int dbMaxBud(u8 * cp);
-s64 dbMapFileSizeToMapSize(struct inode *ipbmap);
static int blkstol2(s64 nb);
static int cntlz(u32 value);
@@ -145,7 +146,6 @@ static const s8 budtab[256] = {
2, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, -1
};
-
/*
* NAME: dbMount()
*
@@ -310,7 +310,6 @@ int dbSync(struct inode *ipbmap)
return (0);
}
-
/*
* NAME: dbFree()
*
@@ -337,6 +336,7 @@ int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
s64 lblkno, rem;
struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
+ struct super_block *sb = ipbmap->i_sb;
IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
@@ -351,6 +351,13 @@ int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
return -EIO;
}
+ /**
+ * TRIM the blocks, when mounted with discard option
+ */
+ if (JFS_SBI(sb)->flag & JFS_DISCARD)
+ if (JFS_SBI(sb)->minblks_trim <= nblocks)
+ jfs_issue_discard(ipbmap, blkno, nblocks);
+
/*
* free the blocks a dmap at a time.
*/
@@ -1095,7 +1102,6 @@ static int dbExtend(struct inode *ip, s64 blkno, s64 nblocks, s64 addnblocks)
/* we were not successful */
release_metapage(mp);
-
return (rc);
}
@@ -1590,6 +1596,117 @@ static int dbAllocAny(struct bmap * bmp, s64 nblocks, int l2nb, s64 * results)
/*
+ * NAME: dbDiscardAG()
+ *
+ * FUNCTION: attempt to discard (TRIM) all free blocks of specific AG
+ *
+ * algorithm:
+ * 1) allocate blocks, as large as possible and save them
+ * while holding IWRITE_LOCK on ipbmap
+ * 2) trim all these saved block/length values
+ * 3) mark the blocks free again
+ *
+ * benefit:
+ * - we work only on one ag at some time, minimizing how long we
+ * need to lock ipbmap
+ * - reading / writing the fs is possible most time, even on
+ * trimming
+ *
+ * downside:
+ * - we write two times to the dmapctl and dmap pages
+ * - but for me, this seems the best way, better ideas?
+ * /TR 2012
+ *
+ * PARAMETERS:
+ * ip - pointer to in-core inode
+ * agno - ag to trim
+ * minlen - minimum value of contiguous blocks
+ *
+ * RETURN VALUES:
+ * s64 - actual number of blocks trimmed
+ */
+s64 dbDiscardAG(struct inode *ip, int agno, s64 minlen)
+{
+ struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
+ struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
+ s64 nblocks, blkno;
+ u64 trimmed = 0;
+ int rc, l2nb;
+ struct super_block *sb = ipbmap->i_sb;
+
+ struct range2trim {
+ u64 blkno;
+ u64 nblocks;
+ } *totrim, *tt;
+
+ /* max blkno / nblocks pairs to trim */
+ int count = 0, range_cnt;
+
+ /* prevent others from writing new stuff here, while trimming */
+ IWRITE_LOCK(ipbmap, RDWRLOCK_DMAP);
+
+ nblocks = bmp->db_agfree[agno];
+ range_cnt = nblocks;
+ do_div(range_cnt, (int)minlen);
+ range_cnt = min(range_cnt + 1, 32 * 1024);
+ totrim = kmalloc(sizeof(struct range2trim) * range_cnt, GFP_NOFS);
+ if (totrim == NULL) {
+ jfs_error(bmp->db_ipbmap->i_sb,
+ "dbDiscardAG: no memory for trim array");
+ IWRITE_UNLOCK(ipbmap);
+ return 0;
+ }
+
+ tt = totrim;
+ while (nblocks >= minlen) {
+ l2nb = BLKSTOL2(nblocks);
+
+ /* 0 = okay, -EIO = fatal, -ENOSPC -> try smaller block */
+ rc = dbAllocAG(bmp, agno, nblocks, l2nb, &blkno);
+ if (rc == 0) {
+ tt->blkno = blkno;
+ tt->nblocks = nblocks;
+ tt++; count++;
+
+ /* the whole ag is free, trim now */
+ if (bmp->db_agfree[agno] == 0)
+ break;
+
+ /* give a hint for the next while */
+ nblocks = bmp->db_agfree[agno];
+ continue;
+ } else if (rc == -ENOSPC) {
+ /* search for next smaller log2 block */
+ l2nb = BLKSTOL2(nblocks) - 1;
+ nblocks = 1 << l2nb;
+ } else {
+ /* Trim any already allocated blocks */
+ jfs_error(bmp->db_ipbmap->i_sb,
+ "dbDiscardAG: -EIO");
+ break;
+ }
+
+ /* check, if our trim array is full */
+ if (unlikely(count >= range_cnt - 1))
+ break;
+ }
+ IWRITE_UNLOCK(ipbmap);
+
+ tt->nblocks = 0; /* mark the current end */
+ for (tt = totrim; tt->nblocks != 0; tt++) {
+ /* when mounted with online discard, dbFree() will
+ * call jfs_issue_discard() itself */
+ if (!(JFS_SBI(sb)->flag & JFS_DISCARD))
+ jfs_issue_discard(ip, tt->blkno, tt->nblocks);
+ dbFree(ip, tt->blkno, tt->nblocks);
+ trimmed += tt->nblocks;
+ }
+ kfree(totrim);
+
+ return trimmed;
+}
+
+/*
* NAME: dbFindCtl()
*
* FUNCTION: starting at a specified dmap control page level and block