X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=sidebyside;f=ldiskfs%2Fkernel_patches%2Fpatches%2Fext4-prealloc-rhel5.patch;h=dda25ed27fdfa0f527776a41a5133f9e53d29c79;hb=898d16e4a9265be32af41bea56f96ddf6e83877d;hp=34d0472fa01e01d0fb8c51b333d726b1ac2b2e8b;hpb=b175e2441b0cd9fae60341ba92b0f7f192e71446;p=fs%2Flustre-release.git

diff --git a/ldiskfs/kernel_patches/patches/ext4-prealloc-rhel5.patch b/ldiskfs/kernel_patches/patches/ext4-prealloc-rhel5.patch
index 34d0472..dda25ed 100644
--- a/ldiskfs/kernel_patches/patches/ext4-prealloc-rhel5.patch
+++ b/ldiskfs/kernel_patches/patches/ext4-prealloc-rhel5.patch
@@ -1,38 +1,53 @@
-Index: linux-2.6.18-128.1.6/fs/ext4/ext4_sb.h
+Index: linux-2.6.18-128.1.6/fs/ext4/super.c
 ===================================================================
---- linux-2.6.18-128.1.6.orig/fs/ext4/ext4_sb.h	2009-05-28 17:16:51.000000000 +0530
-+++ linux-2.6.18-128.1.6/fs/ext4/ext4_sb.h	2009-05-28 17:16:52.000000000 +0530
+--- linux-2.6.18-128.1.6.orig/fs/ext4/super.c
++++ linux-2.6.18-128.1.6/fs/ext4/super.c
+@@ -108,7 +108,8 @@
+ EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
+ EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
+ EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
+-EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
++EXT4_RW_ATTR_SBI_UI(mb_small_req, s_mb_small_req);
++EXT4_RW_ATTR_SBI_UI(mb_large_req, s_mb_large_req);
+ EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
+ EXT4_RW_ATTR_SBI_UI(max_dir_size, s_max_dir_size);
+ 
+@@ -108,7 +108,8 @@
+	ATTR_LIST(mb_max_to_scan),
+	ATTR_LIST(mb_min_to_scan),
+	ATTR_LIST(mb_order2_req),
+-	ATTR_LIST(mb_stream_req),
++	ATTR_LIST(mb_small_req),
++	ATTR_LIST(mb_large_req),
+	ATTR_LIST(mb_group_prealloc),
+	ATTR_LIST(max_dir_size),
+	NULL,
+Index: linux-2.6.18-128.1.6/fs/ext4/ext4.h
+===================================================================
+--- linux-2.6.18-128.1.6.orig/fs/ext4/ext4.h	2009-05-28 17:16:51.000000000 +0530
++++ linux-2.6.18-128.1.6/fs/ext4/ext4.h	2009-05-28 17:16:52.000000000 +0530
 @@ -108,11 +108,14 @@
  
  	/* tunables */
  	unsigned long s_stripe;
--	unsigned long s_mb_stream_request;
+-	unsigned int s_mb_stream_request;
 +	unsigned long s_mb_small_req;
 +	unsigned long s_mb_large_req;
- 	unsigned long s_mb_max_to_scan;
- 	unsigned long s_mb_min_to_scan;
- 	unsigned long s_mb_stats;
- 	unsigned long s_mb_order2_reqs;
+ 	unsigned int s_mb_max_to_scan;
+ 	unsigned int s_mb_min_to_scan;
+ 	unsigned int s_mb_stats;
+ 	unsigned int s_mb_order2_reqs;
 +	unsigned long *s_mb_prealloc_table;
 +	unsigned long s_mb_prealloc_table_size;
- 	unsigned long s_mb_group_prealloc;
+ 	unsigned int s_mb_group_prealloc;
  	/* where last allocation was done - for stream allocation */
  	unsigned long s_mb_last_group;
 Index: linux-2.6.18-128.1.6/fs/ext4/mballoc.c
 ===================================================================
 --- linux-2.6.18-128.1.6.orig/fs/ext4/mballoc.c	2009-05-28 17:16:51.000000000 +0530
 +++ linux-2.6.18-128.1.6/fs/ext4/mballoc.c	2009-05-28 17:19:57.000000000 +0530
-@@ -1744,7 +1744,7 @@
- 	if (size < isize)
- 		size = isize;
- 
--	if (size < sbi->s_mb_stream_request &&
-+	if ((ac->ac_g_ex.fe_len < sbi->s_mb_large_req) &&
- 			(ac->ac_flags & EXT4_MB_HINT_DATA)) {
- 		/* TBD: may be hot point */
- 		spin_lock(&sbi->s_md_lock);
-@@ -2484,6 +2484,26 @@
- 	return -ENOMEM;
+@@ -2284,6 +2284,26 @@
+	}
  }
  
 +static void ext4_mb_prealloc_table_add(struct ext4_sb_info *sbi, int value)
@@ -55,81 +70,16 @@ Index: linux-2.6.18-128.1.6/fs/ext4/mballoc.c
 +}
 +
 +
- int ext4_mb_init(struct super_block *sb, int needs_recovery)
+ static int ext4_mb_good_group(struct ext4_allocation_context *ac,
+				ext4_group_t group, int cr)
  {
- 	struct ext4_sb_info *sbi = EXT4_SB(sb);
-@@ -2542,15 +2562,59 @@
- 	sbi->s_mb_max_to_scan = MB_DEFAULT_MAX_TO_SCAN;
- 	sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
- 	sbi->s_mb_stats = MB_DEFAULT_STATS;
--	sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
- 	sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
- 	sbi->s_mb_history_filter = EXT4_MB_HISTORY_DEFAULT;
--	sbi->s_mb_group_prealloc = MB_DEFAULT_GROUP_PREALLOC;
-+
-+	if (sbi->s_stripe == 0) {
-+		sbi->s_mb_prealloc_table_size = 8;
-+		i = sbi->s_mb_prealloc_table_size * sizeof(unsigned long);
-+		sbi->s_mb_prealloc_table = kmalloc(i, GFP_NOFS);
-+		if (sbi->s_mb_prealloc_table == NULL) {
-+				clear_opt(sbi->s_mount_opt, MBALLOC);
-+				kfree(sbi->s_mb_offsets);
-+				kfree(sbi->s_mb_maxs);
-+				return -ENOMEM;
-+		}
-+		memset(sbi->s_mb_prealloc_table, 0, i);
+@@ -2325,6 +2389,80 @@
+	.llseek		= seq_lseek,
+	.release	= seq_release,
+ };
 +
-+		ext4_mb_prealloc_table_add(sbi, 4);
-+		ext4_mb_prealloc_table_add(sbi, 8);
-+		ext4_mb_prealloc_table_add(sbi, 16);
-+		ext4_mb_prealloc_table_add(sbi, 32);
-+		ext4_mb_prealloc_table_add(sbi, 64);
-+		ext4_mb_prealloc_table_add(sbi, 128);
-+		ext4_mb_prealloc_table_add(sbi, 256);
-+		ext4_mb_prealloc_table_add(sbi, 512);
-+
-+		sbi->s_mb_small_req = 256;
-+		sbi->s_mb_large_req = 1024;
-+		sbi->s_mb_group_prealloc = 512;
-+	} else {
-+		sbi->s_mb_prealloc_table_size = 3;
-+		i = sbi->s_mb_prealloc_table_size * sizeof(unsigned long);
-+		sbi->s_mb_prealloc_table = kmalloc(i, GFP_NOFS);
-+		if (sbi->s_mb_prealloc_table == NULL) {
-+			clear_opt(sbi->s_mount_opt, MBALLOC);
-+			kfree(sbi->s_mb_offsets);
-+			kfree(sbi->s_mb_maxs);
-+			return -ENOMEM;
-+		}
-+		memset(sbi->s_mb_prealloc_table, 0, i);
-+
-+		ext4_mb_prealloc_table_add(sbi, sbi->s_stripe);
-+		ext4_mb_prealloc_table_add(sbi, sbi->s_stripe * 2);
-+		ext4_mb_prealloc_table_add(sbi, sbi->s_stripe * 4);
-+
-+		sbi->s_mb_small_req = sbi->s_stripe;
-+		sbi->s_mb_large_req = sbi->s_stripe * 8;
-+		sbi->s_mb_group_prealloc = sbi->s_stripe * 4;
-+	}
- 
- 	i = sizeof(struct ext4_locality_group) * num_possible_cpus();
- 	sbi->s_locality_groups = kmalloc(i, GFP_KERNEL);
- 	if (sbi->s_locality_groups == NULL) {
- 		clear_opt(sbi->s_mount_opt, MBALLOC);
-+		kfree(sbi->s_mb_prealloc_table);
- 		kfree(sbi->s_mb_offsets);
- 		kfree(sbi->s_mb_maxs);
- 		return -ENOMEM;
-@@ -2725,10 +2789,82 @@
- #define EXT4_MB_MAX_TO_SCAN_NAME	"max_to_scan"
- #define EXT4_MB_MIN_TO_SCAN_NAME	"min_to_scan"
- #define EXT4_MB_ORDER2_REQ		"order2_req"
--#define EXT4_MB_STREAM_REQ		"stream_req"
-+#define EXT4_MB_SMALL_REQ          	"small_req"
-+#define EXT4_MB_LARGE_REQ            	"large_req"
 +#define EXT4_MB_PREALLOC_TABLE          "prealloc_table"
- #define EXT4_MB_GROUP_PREALLOC		"group_prealloc"
- 
++
 +static int ext4_mb_prealloc_table_proc_read(char *page, char **start, off_t off,
 +					    int count, int *eof, void *data)
 +{
@@ -182,7 +132,7 @@ Index: linux-2.6.18-128.1.6/fs/ext4/mballoc.c
 +		prev = value;
 +		num++;
 +	}
- 
++
 +	new_table = kmalloc(num * sizeof(*new_table), GFP_KERNEL);
 +	if (new_table == NULL)
 +		return -ENOMEM;
@@ -202,69 +152,90 @@ Index: linux-2.6.18-128.1.6/fs/ext4/mballoc.c
 +	return cnt;
 +}
  
- #define MB_PROC_FOPS(name)					\
- static int ext4_mb_##name##_proc_show(struct seq_file *m, void *v)	\
-@@ -2774,7 +2910,8 @@
- MB_PROC_FOPS(max_to_scan);
- MB_PROC_FOPS(min_to_scan);
- MB_PROC_FOPS(order2_reqs);
--MB_PROC_FOPS(stream_request);
-+MB_PROC_FOPS(small_req);
-+MB_PROC_FOPS(large_req);
- MB_PROC_FOPS(group_prealloc);
- 
- #define	MB_PROC_HANDLER(name, var)					\
-@@ -2795,6 +2932,7 @@
- 	mode_t mode = S_IFREG | S_IRUGO | S_IWUSR;
- 	struct ext4_sb_info *sbi = EXT4_SB(sb);
- 	struct proc_dir_entry *proc;
-+	struct proc_dir_entry *proc_entry;
- 	char devname[64];
- 
- 	if (proc_root_ext4 == NULL) {
-@@ -2808,15 +2946,29 @@
- 	MB_PROC_HANDLER(EXT4_MB_MAX_TO_SCAN_NAME, max_to_scan);
- 	MB_PROC_HANDLER(EXT4_MB_MIN_TO_SCAN_NAME, min_to_scan);
- 	MB_PROC_HANDLER(EXT4_MB_ORDER2_REQ, order2_reqs);
--	MB_PROC_HANDLER(EXT4_MB_STREAM_REQ, stream_request);
-+	MB_PROC_HANDLER(EXT4_MB_SMALL_REQ, small_req);
-+	MB_PROC_HANDLER(EXT4_MB_LARGE_REQ, large_req);
- 	MB_PROC_HANDLER(EXT4_MB_GROUP_PREALLOC, group_prealloc);
- 
-+	proc_entry = create_proc_entry(EXT4_MB_PREALLOC_TABLE, S_IFREG |
-+				       S_IRUGO | S_IWUSR, sbi->s_mb_proc);
-+	if (proc_entry == NULL) {
-+		printk(KERN_ERR "EXT4-fs: unable to create %s\n",
-+		       EXT4_MB_PREALLOC_TABLE);
-+		goto err_out;
-+	}
-+	proc_entry->data = sbi;
-+	proc_entry->read_proc = ext4_mb_prealloc_table_proc_read;
-+	proc_entry->write_proc = ext4_mb_prealloc_table_proc_write;
+ static void ext4_mb_history_release(struct super_block *sb)
+ {
+@@ -2400,6 +2400,7 @@
+		remove_proc_entry("mb_groups", sbi->s_proc);
+		if (sbi->s_mb_history_max)
+			remove_proc_entry("mb_history", sbi->s_proc);
++		remove_proc_entry(EXT4_MB_PREALLOC_TABLE, sbi->s_proc);
+	}
+	kfree(sbi->s_mb_history);
+ }
+@@ -2408,6 +2446,13 @@
+			p->proc_fops = &ext4_mb_seq_groups_fops;
+			p->data = sb;
+		}
++		p = create_proc_entry(EXT4_MB_PREALLOC_TABLE, S_IFREG |
++				      S_IRUGO | S_IWUSR, sbi->s_proc);
++		if (p) {
++			p->data = sbi;
++			p->read_proc = ext4_mb_prealloc_table_proc_read;
++			p->write_proc = ext4_mb_prealloc_table_proc_write;
++		}
+	}
+
+	sbi->s_mb_history_cur = 0;
+@@ -2542,13 +2562,57 @@
+ 	sbi->s_mb_max_to_scan = MB_DEFAULT_MAX_TO_SCAN;
+ 	sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
+ 	sbi->s_mb_stats = MB_DEFAULT_STATS;
+-	sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
+ 	sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
+ 	sbi->s_mb_history_filter = EXT4_MB_HISTORY_DEFAULT;
+-	sbi->s_mb_group_prealloc = MB_DEFAULT_GROUP_PREALLOC;
 +
- 	return 0;
- 
- err_out:
- 	printk(KERN_ERR "EXT4-fs: Unable to create %s\n", devname);
- 	remove_proc_entry(EXT4_MB_GROUP_PREALLOC, sbi->s_mb_proc);
--	remove_proc_entry(EXT4_MB_STREAM_REQ, sbi->s_mb_proc);
-+	remove_proc_entry(EXT4_MB_PREALLOC_TABLE, sbi->s_mb_proc);
-+	remove_proc_entry(EXT4_MB_LARGE_REQ, sbi->s_mb_proc);
-+	remove_proc_entry(EXT4_MB_SMALL_REQ, sbi->s_mb_proc);
- 	remove_proc_entry(EXT4_MB_ORDER2_REQ, sbi->s_mb_proc);
- 	remove_proc_entry(EXT4_MB_MIN_TO_SCAN_NAME, sbi->s_mb_proc);
- 	remove_proc_entry(EXT4_MB_MAX_TO_SCAN_NAME, sbi->s_mb_proc);
-@@ -2838,7 +2990,9 @@
++	if (sbi->s_stripe == 0) {
++		sbi->s_mb_prealloc_table_size = 10;
++		i = sbi->s_mb_prealloc_table_size * sizeof(unsigned long);
++		sbi->s_mb_prealloc_table = kmalloc(i, GFP_NOFS);
++		if (sbi->s_mb_prealloc_table == NULL) {
++				kfree(sbi->s_mb_offsets);
++				kfree(sbi->s_mb_maxs);
++				return -ENOMEM;
++		}
++		memset(sbi->s_mb_prealloc_table, 0, i);
++
++		ext4_mb_prealloc_table_add(sbi, 4);
++		ext4_mb_prealloc_table_add(sbi, 8);
++		ext4_mb_prealloc_table_add(sbi, 16);
++		ext4_mb_prealloc_table_add(sbi, 32);
++		ext4_mb_prealloc_table_add(sbi, 64);
++		ext4_mb_prealloc_table_add(sbi, 128);
++		ext4_mb_prealloc_table_add(sbi, 256);
++		ext4_mb_prealloc_table_add(sbi, 512);
++		ext4_mb_prealloc_table_add(sbi, 1024);
++		ext4_mb_prealloc_table_add(sbi, 2048);
++
++		sbi->s_mb_small_req = 256;
++		sbi->s_mb_large_req = 1024;
++		sbi->s_mb_group_prealloc = 512;
++	} else {
++		sbi->s_mb_prealloc_table_size = 3;
++		i = sbi->s_mb_prealloc_table_size * sizeof(unsigned long);
++		sbi->s_mb_prealloc_table = kmalloc(i, GFP_NOFS);
++		if (sbi->s_mb_prealloc_table == NULL) {
++			kfree(sbi->s_mb_offsets);
++			kfree(sbi->s_mb_maxs);
++			return -ENOMEM;
++		}
++		memset(sbi->s_mb_prealloc_table, 0, i);
++
++		ext4_mb_prealloc_table_add(sbi, sbi->s_stripe);
++		ext4_mb_prealloc_table_add(sbi, sbi->s_stripe * 2);
++		ext4_mb_prealloc_table_add(sbi, sbi->s_stripe * 4);
++
++		sbi->s_mb_small_req = sbi->s_stripe;
++		sbi->s_mb_large_req = sbi->s_stripe * 8;
++		sbi->s_mb_group_prealloc = sbi->s_stripe * 4;
++	}
  
- 	bdevname(sb->s_bdev, devname);
- 	remove_proc_entry(EXT4_MB_GROUP_PREALLOC, sbi->s_mb_proc);
--	remove_proc_entry(EXT4_MB_STREAM_REQ, sbi->s_mb_proc);
-+	remove_proc_entry(EXT4_MB_PREALLOC_TABLE, sbi->s_mb_proc);
-+	remove_proc_entry(EXT4_MB_LARGE_REQ, sbi->s_mb_proc);
-+	remove_proc_entry(EXT4_MB_SMALL_REQ, sbi->s_mb_proc);
- 	remove_proc_entry(EXT4_MB_ORDER2_REQ, sbi->s_mb_proc);
- 	remove_proc_entry(EXT4_MB_MIN_TO_SCAN_NAME, sbi->s_mb_proc);
- 	remove_proc_entry(EXT4_MB_MAX_TO_SCAN_NAME, sbi->s_mb_proc);
+	sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
+ 	if (sbi->s_locality_groups == NULL) {
++		kfree(sbi->s_mb_prealloc_table);
+ 		kfree(sbi->s_mb_offsets);
+ 		kfree(sbi->s_mb_maxs);
+ 		return -ENOMEM;
 @@ -3032,11 +3186,12 @@
  ext4_mb_normalize_request(struct ext4_allocation_context *ac,
  				struct ext4_allocation_request *ar)
@@ -358,48 +329,50 @@ Index: linux-2.6.18-128.1.6/fs/ext4/mballoc.c
 @@ -3185,7 +3326,6 @@
  	}
  	BUG_ON(start + size <= ac->ac_o_ex.fe_logical &&
- 			start > ac->ac_o_ex.fe_logical);
--	BUG_ON(size <= 0 || size >= EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
+			start > ac->ac_o_ex.fe_logical);
+-	BUG_ON(size <= 0 || size > EXT4_BLOCKS_PER_GROUP(ac->ac_sb));
  
  	/* now prepare goal request */
  
-@@ -4077,22 +4217,32 @@
- {
- 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
- 	int bsbits = ac->ac_sb->s_blocksize_bits;
--	loff_t size, isize;
-+	loff_t size;
- 
- 	if (!(ac->ac_flags & EXT4_MB_HINT_DATA))
- 		return;
- 
--	size = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len;
--	isize = i_size_read(ac->ac_inode) >> bsbits;
--	size = max(size, isize);
--
--	/* don't use group allocation for large files */
--	if (size >= sbi->s_mb_stream_request)
-+	if (ac->ac_o_ex.fe_len >= sbi->s_mb_small_req)
- 		return;
+@@ -4077,11 +4217,17 @@
  
- 	if (unlikely(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY))
- 		return;
+	/* don't use group allocation for large files */
+	size = max(size, isize);
+-	if (size >= sbi->s_mb_stream_request) {
++	if ((ac->ac_o_ex.fe_len >= sbi->s_mb_small_req) ||
++	    (size >= sbi->s_mb_large_req)) {
+		ac->ac_flags |= EXT4_MB_STREAM_ALLOC;
+		return;
+	}
  
 +	/* request is so large that we don't care about
 +	 * streaming - it overweights any possible seek */
 +	if (ac->ac_o_ex.fe_len >= sbi->s_mb_large_req)
 +		return;
 +
-+	size = ac->ac_o_ex.fe_logical + ac->ac_o_ex.fe_len;
-+	size = size << bsbits;
-+	if (size < i_size_read(ac->ac_inode))
-+		size = i_size_read(ac->ac_inode);
-+	size = (size + ac->ac_sb->s_blocksize - 1) >> bsbits;
-+
-+	/* don't use group allocation for large files */
-+	if (size >= sbi->s_mb_large_req)
-+		return;
-+
  	BUG_ON(ac->ac_lg != NULL);
  	/*
  	 * locality group prealloc space are per cpu. The reason for having
+Index: linux-2.6.27.21-0.1/fs/ext4/inode.c
+===================================================================
+--- linux-2.6.27.21-0.1.orig/fs/ext4/inode.c	2009-05-28 11:12:42.000000000 +0530
++++ linux-2.6.27.21-0.1/fs/ext4/inode.c	2009-05-28 11:16:48.000000000 +0530
+@@ -2442,14 +2442,14 @@
+ 		return -EROFS;
+ 
+ 	/*
+-	 * Make sure nr_to_write is >= sbi->s_mb_stream_request
++	 * Make sure nr_to_write is >= sbi->s_mb_small_req
+ 	 * This make sure small files blocks are allocated in
+ 	 * single attempt. This ensure that small files
+ 	 * get less fragmented.
+ 	 */
+-	if (wbc->nr_to_write < sbi->s_mb_stream_request) {
+-		nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
+-		wbc->nr_to_write = sbi->s_mb_stream_request;
++	if (wbc->nr_to_write < sbi->s_mb_small_req) {
++		nr_to_writebump = sbi->s_mb_small_req - wbc->nr_to_write;
++		wbc->nr_to_write = sbi->s_mb_small_req;
+ 	}
+ 	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
+ 		range_whole = 1;