inode: Make unused inode LRU per superblock
Dave Chinner [Fri, 8 Jul 2011 04:14:39 +0000 (14:14 +1000)]
The inode unused list is currently a global LRU. This does not match
the other global filesystem cache - the dentry cache - which uses
per-superblock LRU lists. Hence we have related filesystem object
types using different LRU reclaimation schemes.

To enable a per-superblock filesystem cache shrinker, both of these
caches need to have per-sb unused object LRU lists. Hence this patch
converts the global inode LRU to per-sb LRUs.

The patch only does rudimentary per-sb propotioning in the shrinker
infrastructure, as this gets removed when the per-sb shrinker
callouts are introduced later on.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

fs/inode.c
fs/super.c
include/linux/fs.h

index 9a03611..8c34913 100644 (file)
@@ -34,7 +34,7 @@
  * inode->i_lock protects:
  *   inode->i_state, inode->i_hash, __iget()
  * inode_lru_lock protects:
- *   inode_lru, inode->i_lru
+ *   inode->i_sb->s_inode_lru, inode->i_lru
  * inode_sb_list_lock protects:
  *   sb->s_inodes, inode->i_sb_list
  * inode_wb_list_lock protects:
@@ -64,7 +64,6 @@ static unsigned int i_hash_shift __read_mostly;
 static struct hlist_head *inode_hashtable __read_mostly;
 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
 
-static LIST_HEAD(inode_lru);
 static DEFINE_SPINLOCK(inode_lru_lock);
 
 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
@@ -345,7 +344,8 @@ static void inode_lru_list_add(struct inode *inode)
 {
        spin_lock(&inode_lru_lock);
        if (list_empty(&inode->i_lru)) {
-               list_add(&inode->i_lru, &inode_lru);
+               list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
+               inode->i_sb->s_nr_inodes_unused++;
                this_cpu_inc(nr_unused);
        }
        spin_unlock(&inode_lru_lock);
@@ -356,6 +356,7 @@ static void inode_lru_list_del(struct inode *inode)
        spin_lock(&inode_lru_lock);
        if (!list_empty(&inode->i_lru)) {
                list_del_init(&inode->i_lru);
+               inode->i_sb->s_nr_inodes_unused--;
                this_cpu_dec(nr_unused);
        }
        spin_unlock(&inode_lru_lock);
@@ -628,21 +629,20 @@ static int can_unuse(struct inode *inode)
  * LRU does not have strict ordering. Hence we don't want to reclaim inodes
  * with this flag set because they are the inodes that are out of order.
  */
-static void prune_icache(int nr_to_scan)
+static void shrink_icache_sb(struct super_block *sb, int *nr_to_scan)
 {
        LIST_HEAD(freeable);
        int nr_scanned;
        unsigned long reap = 0;
 
-       down_read(&iprune_sem);
        spin_lock(&inode_lru_lock);
-       for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
+       for (nr_scanned = *nr_to_scan; nr_scanned >= 0; nr_scanned--) {
                struct inode *inode;
 
-               if (list_empty(&inode_lru))
+               if (list_empty(&sb->s_inode_lru))
                        break;
 
-               inode = list_entry(inode_lru.prev, struct inode, i_lru);
+               inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
 
                /*
                 * we are inverting the inode_lru_lock/inode->i_lock here,
@@ -650,7 +650,7 @@ static void prune_icache(int nr_to_scan)
                 * inode to the back of the list so we don't spin on it.
                 */
                if (!spin_trylock(&inode->i_lock)) {
-                       list_move(&inode->i_lru, &inode_lru);
+                       list_move(&inode->i_lru, &sb->s_inode_lru);
                        continue;
                }
 
@@ -662,6 +662,7 @@ static void prune_icache(int nr_to_scan)
                    (inode->i_state & ~I_REFERENCED)) {
                        list_del_init(&inode->i_lru);
                        spin_unlock(&inode->i_lock);
+                       sb->s_nr_inodes_unused--;
                        this_cpu_dec(nr_unused);
                        continue;
                }
@@ -669,7 +670,7 @@ static void prune_icache(int nr_to_scan)
                /* recently referenced inodes get one more pass */
                if (inode->i_state & I_REFERENCED) {
                        inode->i_state &= ~I_REFERENCED;
-                       list_move(&inode->i_lru, &inode_lru);
+                       list_move(&inode->i_lru, &sb->s_inode_lru);
                        spin_unlock(&inode->i_lock);
                        continue;
                }
@@ -683,7 +684,7 @@ static void prune_icache(int nr_to_scan)
                        iput(inode);
                        spin_lock(&inode_lru_lock);
 
-                       if (inode != list_entry(inode_lru.next,
+                       if (inode != list_entry(sb->s_inode_lru.next,
                                                struct inode, i_lru))
                                continue;       /* wrong inode or list_empty */
                        /* avoid lock inversions with trylock */
@@ -699,6 +700,7 @@ static void prune_icache(int nr_to_scan)
                spin_unlock(&inode->i_lock);
 
                list_move(&inode->i_lru, &freeable);
+               sb->s_nr_inodes_unused--;
                this_cpu_dec(nr_unused);
        }
        if (current_is_kswapd())
@@ -706,8 +708,75 @@ static void prune_icache(int nr_to_scan)
        else
                __count_vm_events(PGINODESTEAL, reap);
        spin_unlock(&inode_lru_lock);
+       *nr_to_scan = nr_scanned;
 
        dispose_list(&freeable);
+}
+
+static void prune_icache(int count)
+{
+       struct super_block *sb, *p = NULL;
+       int w_count;
+       int unused = inodes_stat.nr_unused;
+       int prune_ratio;
+       int pruned;
+
+       if (unused == 0 || count == 0)
+               return;
+       down_read(&iprune_sem);
+       if (count >= unused)
+               prune_ratio = 1;
+       else
+               prune_ratio = unused / count;
+       spin_lock(&sb_lock);
+       list_for_each_entry(sb, &super_blocks, s_list) {
+               if (list_empty(&sb->s_instances))
+                       continue;
+               if (sb->s_nr_inodes_unused == 0)
+                       continue;
+               sb->s_count++;
+               /* Now, we reclaim unused dentrins with fairness.
+                * We reclaim them same percentage from each superblock.
+                * We calculate number of dentries to scan on this sb
+                * as follows, but the implementation is arranged to avoid
+                * overflows:
+                * number of dentries to scan on this sb =
+                * count * (number of dentries on this sb /
+                * number of dentries in the machine)
+                */
+               spin_unlock(&sb_lock);
+               if (prune_ratio != 1)
+                       w_count = (sb->s_nr_inodes_unused / prune_ratio) + 1;
+               else
+                       w_count = sb->s_nr_inodes_unused;
+               pruned = w_count;
+               /*
+                * We need to be sure this filesystem isn't being unmounted,
+                * otherwise we could race with generic_shutdown_super(), and
+                * end up holding a reference to an inode while the filesystem
+                * is unmounted.  So we try to get s_umount, and make sure
+                * s_root isn't NULL.
+                */
+               if (down_read_trylock(&sb->s_umount)) {
+                       if ((sb->s_root != NULL) &&
+                           (!list_empty(&sb->s_dentry_lru))) {
+                               shrink_icache_sb(sb, &w_count);
+                               pruned -= w_count;
+                       }
+                       up_read(&sb->s_umount);
+               }
+               spin_lock(&sb_lock);
+               if (p)
+                       __put_super(p);
+               count -= pruned;
+               p = sb;
+               /* more work left to do? */
+               if (count <= 0)
+                       break;
+       }
+       if (p)
+               __put_super(p);
+       spin_unlock(&sb_lock);
        up_read(&iprune_sem);
 }
 
index 263edeb..e8e6dbf 100644 (file)
@@ -77,6 +77,7 @@ static struct super_block *alloc_super(struct file_system_type *type)
                INIT_HLIST_BL_HEAD(&s->s_anon);
                INIT_LIST_HEAD(&s->s_inodes);
                INIT_LIST_HEAD(&s->s_dentry_lru);
+               INIT_LIST_HEAD(&s->s_inode_lru);
                init_rwsem(&s->s_umount);
                mutex_init(&s->s_lock);
                lockdep_set_class(&s->s_umount, &type->s_umount_key);
index a0011ae..9724f0a 100644 (file)
@@ -1397,6 +1397,10 @@ struct super_block {
        struct list_head        s_dentry_lru;   /* unused dentry lru */
        int                     s_nr_dentry_unused;     /* # of dentry on lru */
 
+       /* inode_lru_lock protects s_inode_lru and s_nr_inodes_unused */
+       struct list_head        s_inode_lru;            /* unused inode lru */
+       int                     s_nr_inodes_unused;     /* # of inodes on lru */
+
        struct block_device     *s_bdev;
        struct backing_dev_info *s_bdi;
        struct mtd_info         *s_mtd;