f2fs: change to use rwsem for gc_mutex

Mutex lock won't serialize callers, in order to avoid starving of unlucky
caller, let's use rwsem lock instead.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

fs/f2fs/f2fs.h

@@ -1391,7 +1391,10 @@ struct f2fs_sb_info {
 	struct f2fs_mount_info mount_opt;	/* mount options */
 
 	/* for cleaning operations */
-	struct mutex gc_mutex;			/* mutex for GC */
+	struct rw_semaphore gc_lock;		/*
+						 * semaphore for GC, avoid
+						 * race between GC and GC or CP
+						 */
 	struct f2fs_gc_kthread	*gc_thread;	/* GC thread */
 	unsigned int cur_victim_sec;		/* current victim section num */
 	unsigned int gc_mode;			/* current GC state */

fs/f2fs/file.c

@@ -1642,7 +1642,7 @@ static int expand_inode_data(struct inode *inode, loff_t offset,
 next_alloc:
 		if (has_not_enough_free_secs(sbi, 0,
 			GET_SEC_FROM_SEG(sbi, overprovision_segments(sbi)))) {
-			mutex_lock(&sbi->gc_mutex);
+			down_write(&sbi->gc_lock);
 			err = f2fs_gc(sbi, true, false, NULL_SEGNO);
 			if (err && err != -ENODATA && err != -EAGAIN)
 				goto out_err;
@@ -2450,12 +2450,12 @@ static int f2fs_ioc_gc(struct file *filp, unsigned long arg)
 		return ret;
 
 	if (!sync) {
-		if (!mutex_trylock(&sbi->gc_mutex)) {
+		if (!down_write_trylock(&sbi->gc_lock)) {
 			ret = -EBUSY;
 			goto out;
 		}
 	} else {
-		mutex_lock(&sbi->gc_mutex);
+		down_write(&sbi->gc_lock);
 	}
 
 	ret = f2fs_gc(sbi, sync, true, NULL_SEGNO);
@@ -2493,12 +2493,12 @@ static int f2fs_ioc_gc_range(struct file *filp, unsigned long arg)
 
 do_more:
 	if (!range.sync) {
-		if (!mutex_trylock(&sbi->gc_mutex)) {
+		if (!down_write_trylock(&sbi->gc_lock)) {
 			ret = -EBUSY;
 			goto out;
 		}
 	} else {
-		mutex_lock(&sbi->gc_mutex);
+		down_write(&sbi->gc_lock);
 	}
 
 	ret = f2fs_gc(sbi, range.sync, true, GET_SEGNO(sbi, range.start));
@@ -2929,7 +2929,7 @@ static int f2fs_ioc_flush_device(struct file *filp, unsigned long arg)
 	end_segno = min(start_segno + range.segments, dev_end_segno);
 
 	while (start_segno < end_segno) {
-		if (!mutex_trylock(&sbi->gc_mutex)) {
+		if (!down_write_trylock(&sbi->gc_lock)) {
 			ret = -EBUSY;
 			goto out;
 		}

fs/f2fs/gc.c

@@ -78,18 +78,18 @@ static int gc_thread_func(void *data)
 		 */
 		if (sbi->gc_mode == GC_URGENT) {
 			wait_ms = gc_th->urgent_sleep_time;
-			mutex_lock(&sbi->gc_mutex);
+			down_write(&sbi->gc_lock);
 			goto do_gc;
 		}
 
-		if (!mutex_trylock(&sbi->gc_mutex)) {
+		if (!down_write_trylock(&sbi->gc_lock)) {
 			stat_other_skip_bggc_count(sbi);
 			goto next;
 		}
 
 		if (!is_idle(sbi, GC_TIME)) {
 			increase_sleep_time(gc_th, &wait_ms);
-			mutex_unlock(&sbi->gc_mutex);
+			up_write(&sbi->gc_lock);
 			stat_io_skip_bggc_count(sbi);
 			goto next;
 		}
@@ -1370,7 +1370,7 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
 				reserved_segments(sbi),
 				prefree_segments(sbi));
 
-	mutex_unlock(&sbi->gc_mutex);
+	up_write(&sbi->gc_lock);
 
 	put_gc_inode(&gc_list);
 
@@ -1409,9 +1409,9 @@ static int free_segment_range(struct f2fs_sb_info *sbi, unsigned int start,
 			.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
 		};
 
-		mutex_lock(&sbi->gc_mutex);
+		down_write(&sbi->gc_lock);
 		do_garbage_collect(sbi, segno, &gc_list, FG_GC);
-		mutex_unlock(&sbi->gc_mutex);
+		up_write(&sbi->gc_lock);
 		put_gc_inode(&gc_list);
 
 		if (get_valid_blocks(sbi, segno, true))

fs/f2fs/segment.c

@@ -504,7 +504,7 @@ void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 	 * dir/node pages without enough free segments.
 	 */
 	if (has_not_enough_free_secs(sbi, 0, 0)) {
-		mutex_lock(&sbi->gc_mutex);
+		down_write(&sbi->gc_lock);
 		f2fs_gc(sbi, false, false, NULL_SEGNO);
 	}
 }
@@ -2860,9 +2860,9 @@ int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
 	if (sbi->discard_blks == 0)
 		goto out;
 
-	mutex_lock(&sbi->gc_mutex);
+	down_write(&sbi->gc_lock);
 	err = f2fs_write_checkpoint(sbi, &cpc);
-	mutex_unlock(&sbi->gc_mutex);
+	up_write(&sbi->gc_lock);
 	if (err)
 		goto out;
 

fs/f2fs/super.c

@@ -1238,9 +1238,9 @@ int f2fs_sync_fs(struct super_block *sb, int sync)
 
 		cpc.reason = __get_cp_reason(sbi);
 
-		mutex_lock(&sbi->gc_mutex);
+		down_write(&sbi->gc_lock);
 		err = f2fs_write_checkpoint(sbi, &cpc);
-		mutex_unlock(&sbi->gc_mutex);
+		up_write(&sbi->gc_lock);
 	}
 	f2fs_trace_ios(NULL, 1);
 
@@ -1621,7 +1621,7 @@ static int f2fs_disable_checkpoint(struct f2fs_sb_info *sbi)
 	f2fs_update_time(sbi, DISABLE_TIME);
 
 	while (!f2fs_time_over(sbi, DISABLE_TIME)) {
-		mutex_lock(&sbi->gc_mutex);
+		down_write(&sbi->gc_lock);
 		err = f2fs_gc(sbi, true, false, NULL_SEGNO);
 		if (err == -ENODATA) {
 			err = 0;
@@ -1643,7 +1643,7 @@ static int f2fs_disable_checkpoint(struct f2fs_sb_info *sbi)
 		goto restore_flag;
 	}
 
-	mutex_lock(&sbi->gc_mutex);
+	down_write(&sbi->gc_lock);
 	cpc.reason = CP_PAUSE;
 	set_sbi_flag(sbi, SBI_CP_DISABLED);
 	err = f2fs_write_checkpoint(sbi, &cpc);
@@ -1655,7 +1655,7 @@ static int f2fs_disable_checkpoint(struct f2fs_sb_info *sbi)
 	spin_unlock(&sbi->stat_lock);
 
 out_unlock:
-	mutex_unlock(&sbi->gc_mutex);
+	up_write(&sbi->gc_lock);
 restore_flag:
 	sbi->sb->s_flags = s_flags;	/* Restore MS_RDONLY status */
 	return err;
@@ -1663,12 +1663,12 @@ static int f2fs_disable_checkpoint(struct f2fs_sb_info *sbi)
 
 static void f2fs_enable_checkpoint(struct f2fs_sb_info *sbi)
 {
-	mutex_lock(&sbi->gc_mutex);
+	down_write(&sbi->gc_lock);
 	f2fs_dirty_to_prefree(sbi);
 
 	clear_sbi_flag(sbi, SBI_CP_DISABLED);
 	set_sbi_flag(sbi, SBI_IS_DIRTY);
-	mutex_unlock(&sbi->gc_mutex);
+	up_write(&sbi->gc_lock);
 
 	f2fs_sync_fs(sbi->sb, 1);
 }
@@ -3398,7 +3398,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
 
 	/* init f2fs-specific super block info */
 	sbi->valid_super_block = valid_super_block;
-	mutex_init(&sbi->gc_mutex);
+	init_rwsem(&sbi->gc_lock);
 	mutex_init(&sbi->writepages);
 	mutex_init(&sbi->cp_mutex);
 	mutex_init(&sbi->resize_mutex);