f2fs: support aligned pinned file
authorJaegeuk Kim <jaegeuk@kernel.org>
Fri, 18 Oct 2019 17:06:40 +0000 (10:06 -0700)
committerJaegeuk Kim <jaegeuk@kernel.org>
Thu, 7 Nov 2019 18:40:59 +0000 (10:40 -0800)
This patch supports 2MB-aligned pinned file, which can guarantee no GC at all
by allocating fully valid 2MB segment.

Check free segments by has_not_enough_free_secs() with large budget.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/f2fs.h
fs/f2fs/file.c
fs/f2fs/recovery.c
fs/f2fs/segment.c
fs/f2fs/segment.h
fs/f2fs/super.c
fs/f2fs/sysfs.c

index ca342f4c7db11632795ec7de0e872766ddb0c7d6..c681f51e351b14079d7ff21883bace68c3c9f5e1 100644 (file)
@@ -890,6 +890,7 @@ enum {
        CURSEG_WARM_NODE,       /* direct node blocks of normal files */
        CURSEG_COLD_NODE,       /* indirect node blocks */
        NO_CHECK_TYPE,
+       CURSEG_COLD_DATA_PINNED,/* cold data for pinned file */
 };
 
 struct flush_cmd {
@@ -1301,6 +1302,7 @@ struct f2fs_sb_info {
 
        /* threshold for gc trials on pinned files */
        u64 gc_pin_file_threshold;
+       struct rw_semaphore pin_sem;
 
        /* maximum # of trials to find a victim segment for SSR and GC */
        unsigned int max_victim_search;
@@ -3116,7 +3118,7 @@ void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi);
 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra);
 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
                                        unsigned int start, unsigned int end);
-void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi);
+void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi, int type);
 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range);
 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
                                        struct cp_control *cpc);
index 29bc0a542759a279f994862fe6eb3c8f97eac39d..c31a5bbc8090712ba679666f25d91629b1f054d9 100644 (file)
@@ -1545,12 +1545,44 @@ static int expand_inode_data(struct inode *inode, loff_t offset,
        if (off_end)
                map.m_len++;
 
-       if (f2fs_is_pinned_file(inode))
-               map.m_seg_type = CURSEG_COLD_DATA;
+       if (!map.m_len)
+               return 0;
+
+       if (f2fs_is_pinned_file(inode)) {
+               block_t len = (map.m_len >> sbi->log_blocks_per_seg) <<
+                                       sbi->log_blocks_per_seg;
+               block_t done = 0;
+
+               if (map.m_len % sbi->blocks_per_seg)
+                       len += sbi->blocks_per_seg;
+
+               map.m_len = sbi->blocks_per_seg;
+next_alloc:
+               if (has_not_enough_free_secs(sbi, 0,
+                       GET_SEC_FROM_SEG(sbi, overprovision_segments(sbi)))) {
+                       mutex_lock(&sbi->gc_mutex);
+                       err = f2fs_gc(sbi, true, false, NULL_SEGNO);
+                       if (err && err != -ENODATA && err != -EAGAIN)
+                               goto out_err;
+               }
 
-       err = f2fs_map_blocks(inode, &map, 1, (f2fs_is_pinned_file(inode) ?
-                                               F2FS_GET_BLOCK_PRE_DIO :
-                                               F2FS_GET_BLOCK_PRE_AIO));
+               down_write(&sbi->pin_sem);
+               map.m_seg_type = CURSEG_COLD_DATA_PINNED;
+               f2fs_allocate_new_segments(sbi, CURSEG_COLD_DATA);
+               err = f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_DIO);
+               up_write(&sbi->pin_sem);
+
+               done += map.m_len;
+               len -= map.m_len;
+               map.m_lblk += map.m_len;
+               if (!err && len)
+                       goto next_alloc;
+
+               map.m_len = done;
+       } else {
+               err = f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
+       }
+out_err:
        if (err) {
                pgoff_t last_off;
 
index 783773e4560de98a1db31e48a11c7ed89ccace62..76477f71d4ee9b82123aab7d72ed9f3a0c05e2a1 100644 (file)
@@ -711,7 +711,7 @@ next:
                f2fs_put_page(page, 1);
        }
        if (!err)
-               f2fs_allocate_new_segments(sbi);
+               f2fs_allocate_new_segments(sbi, NO_CHECK_TYPE);
        return err;
 }
 
index 25c750cd0272e80fe7c689bece3e0f4d95bad837..8bb37f8a18456732d48e5ad45fbcdb3b91b9e242 100644 (file)
@@ -2690,7 +2690,7 @@ unlock:
        up_read(&SM_I(sbi)->curseg_lock);
 }
 
-void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
+void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi, int type)
 {
        struct curseg_info *curseg;
        unsigned int old_segno;
@@ -2699,10 +2699,17 @@ void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
        down_write(&SIT_I(sbi)->sentry_lock);
 
        for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
+               if (type != NO_CHECK_TYPE && i != type)
+                       continue;
+
                curseg = CURSEG_I(sbi, i);
-               old_segno = curseg->segno;
-               SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
-               locate_dirty_segment(sbi, old_segno);
+               if (type == NO_CHECK_TYPE || curseg->next_blkoff ||
+                               get_valid_blocks(sbi, curseg->segno, false) ||
+                               get_ckpt_valid_blocks(sbi, curseg->segno)) {
+                       old_segno = curseg->segno;
+                       SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
+                       locate_dirty_segment(sbi, old_segno);
+               }
        }
 
        up_write(&SIT_I(sbi)->sentry_lock);
@@ -3068,6 +3075,19 @@ void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
 {
        struct sit_info *sit_i = SIT_I(sbi);
        struct curseg_info *curseg = CURSEG_I(sbi, type);
+       bool put_pin_sem = false;
+
+       if (type == CURSEG_COLD_DATA) {
+               /* GC during CURSEG_COLD_DATA_PINNED allocation */
+               if (down_read_trylock(&sbi->pin_sem)) {
+                       put_pin_sem = true;
+               } else {
+                       type = CURSEG_WARM_DATA;
+                       curseg = CURSEG_I(sbi, type);
+               }
+       } else if (type == CURSEG_COLD_DATA_PINNED) {
+               type = CURSEG_COLD_DATA;
+       }
 
        down_read(&SM_I(sbi)->curseg_lock);
 
@@ -3133,6 +3153,9 @@ void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
        mutex_unlock(&curseg->curseg_mutex);
 
        up_read(&SM_I(sbi)->curseg_lock);
+
+       if (put_pin_sem)
+               up_read(&sbi->pin_sem);
 }
 
 static void update_device_state(struct f2fs_io_info *fio)
index 325781a1ae4d6d8ce918e5b6b7cf289895024f4a..a95467b202ea2040800f4d87209f317757f62f6a 100644 (file)
@@ -313,6 +313,8 @@ struct sit_entry_set {
  */
 static inline struct curseg_info *CURSEG_I(struct f2fs_sb_info *sbi, int type)
 {
+       if (type == CURSEG_COLD_DATA_PINNED)
+               type = CURSEG_COLD_DATA;
        return (struct curseg_info *)(SM_I(sbi)->curseg_array + type);
 }
 
index f320fd11db48d0898b1768a752ff8fe13c823731..c02a47ce551bc99e1c8470da87d7ea3018e2cb3b 100644 (file)
@@ -2853,6 +2853,7 @@ static void init_sb_info(struct f2fs_sb_info *sbi)
        spin_lock_init(&sbi->dev_lock);
 
        init_rwsem(&sbi->sb_lock);
+       init_rwsem(&sbi->pin_sem);
 }
 
 static int init_percpu_info(struct f2fs_sb_info *sbi)
index b558b64a4c9ca33eee88896bc53ab4f07036a22f..f164959e4224ff7eea9f0ee4adb62b1fb723ae21 100644 (file)
@@ -154,6 +154,8 @@ static ssize_t features_show(struct f2fs_attr *a,
        if (f2fs_sb_has_casefold(sbi))
                len += snprintf(buf + len, PAGE_SIZE - len, "%s%s",
                                len ? ", " : "", "casefold");
+       len += snprintf(buf + len, PAGE_SIZE - len, "%s%s",
+                               len ? ", " : "", "pin_file");
        len += snprintf(buf + len, PAGE_SIZE - len, "\n");
        return len;
 }