writeback: move bandwidth related fields from backing_dev_info into bdi_writeback
authorTejun Heo <tj@kernel.org>
Fri, 22 May 2015 21:13:28 +0000 (17:13 -0400)
committerJens Axboe <axboe@fb.com>
Tue, 2 Jun 2015 14:33:34 +0000 (08:33 -0600)
Currently, a bdi (backing_dev_info) embeds single wb (bdi_writeback)
and the role of the separation is unclear.  For cgroup support for
writeback IOs, a bdi will be updated to host multiple wb's where each
wb serves writeback IOs of a different cgroup on the bdi.  To achieve
that, a wb should carry all states necessary for servicing writeback
IOs for a cgroup independently.

This patch moves bandwidth related fields from backing_dev_info into
bdi_writeback.

* The moved fields are: bw_time_stamp, dirtied_stamp, written_stamp,
  write_bandwidth, avg_write_bandwidth, dirty_ratelimit,
  balanced_dirty_ratelimit, completions and dirty_exceeded.

* writeback_chunk_size() and over_bground_thresh() now take @wb
  instead of @bdi.

* bdi_writeout_fraction(bdi, ...) -> wb_writeout_fraction(wb, ...)
  bdi_dirty_limit(bdi, ...) -> wb_dirty_limit(wb, ...)
  bdi_position_ration(bdi, ...) -> wb_position_ratio(wb, ...)
  bdi_update_writebandwidth(bdi, ...) -> wb_update_write_bandwidth(wb, ...)
  [__]bdi_update_bandwidth(bdi, ...) -> [__]wb_update_bandwidth(wb, ...)
  bdi_{max|min}_pause(bdi, ...) -> wb_{max|min}_pause(wb, ...)
  bdi_dirty_limits(bdi, ...) -> wb_dirty_limits(wb, ...)

* Init/exits of the relocated fields are moved to bdi_wb_init/exit()
  respectively.  Note that explicit zeroing is dropped in the process
  as wb's are cleared in entirety anyway.

* As there's still only one bdi_writeback per backing_dev_info, all
  uses of bdi->stat[] are mechanically replaced with bdi->wb.stat[]
  introducing no behavior changes.

v2: Typo in description fixed as suggested by Jan.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jaegeuk Kim <jaegeuk@kernel.org>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
fs/f2fs/node.c
fs/f2fs/segment.h
fs/fs-writeback.c
fs/gfs2/super.c
include/linux/backing-dev.h
include/linux/writeback.h
include/trace/events/writeback.h
mm/backing-dev.c
mm/page-writeback.c

index 8ab0cf1..d211602 100644 (file)
@@ -53,7 +53,7 @@ bool available_free_memory(struct f2fs_sb_info *sbi, int type)
                                                        PAGE_CACHE_SHIFT;
                res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
        } else if (type == DIRTY_DENTS) {
-               if (sbi->sb->s_bdi->dirty_exceeded)
+               if (sbi->sb->s_bdi->wb.dirty_exceeded)
                        return false;
                mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
                res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
@@ -70,7 +70,7 @@ bool available_free_memory(struct f2fs_sb_info *sbi, int type)
                                sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
                res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
        } else {
-               if (sbi->sb->s_bdi->dirty_exceeded)
+               if (sbi->sb->s_bdi->wb.dirty_exceeded)
                        return false;
        }
        return res;
index 85d7fa7..6408989 100644 (file)
@@ -713,7 +713,7 @@ static inline unsigned int max_hw_blocks(struct f2fs_sb_info *sbi)
  */
 static inline int nr_pages_to_skip(struct f2fs_sb_info *sbi, int type)
 {
-       if (sbi->sb->s_bdi->dirty_exceeded)
+       if (sbi->sb->s_bdi->wb.dirty_exceeded)
                return 0;
 
        if (type == DATA)
index 8873ecd..1945cb9 100644 (file)
@@ -624,7 +624,7 @@ out:
        return ret;
 }
 
-static long writeback_chunk_size(struct backing_dev_info *bdi,
+static long writeback_chunk_size(struct bdi_writeback *wb,
                                 struct wb_writeback_work *work)
 {
        long pages;
@@ -645,7 +645,7 @@ static long writeback_chunk_size(struct backing_dev_info *bdi,
        if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
                pages = LONG_MAX;
        else {
-               pages = min(bdi->avg_write_bandwidth / 2,
+               pages = min(wb->avg_write_bandwidth / 2,
                            global_dirty_limit / DIRTY_SCOPE);
                pages = min(pages, work->nr_pages);
                pages = round_down(pages + MIN_WRITEBACK_PAGES,
@@ -743,7 +743,7 @@ static long writeback_sb_inodes(struct super_block *sb,
                inode->i_state |= I_SYNC;
                spin_unlock(&inode->i_lock);
 
-               write_chunk = writeback_chunk_size(wb->bdi, work);
+               write_chunk = writeback_chunk_size(wb, work);
                wbc.nr_to_write = write_chunk;
                wbc.pages_skipped = 0;
 
@@ -830,7 +830,7 @@ static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
        return nr_pages - work.nr_pages;
 }
 
-static bool over_bground_thresh(struct backing_dev_info *bdi)
+static bool over_bground_thresh(struct bdi_writeback *wb)
 {
        unsigned long background_thresh, dirty_thresh;
 
@@ -840,8 +840,7 @@ static bool over_bground_thresh(struct backing_dev_info *bdi)
            global_page_state(NR_UNSTABLE_NFS) > background_thresh)
                return true;
 
-       if (wb_stat(&bdi->wb, WB_RECLAIMABLE) >
-                               bdi_dirty_limit(bdi, background_thresh))
+       if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh))
                return true;
 
        return false;
@@ -854,7 +853,7 @@ static bool over_bground_thresh(struct backing_dev_info *bdi)
 static void wb_update_bandwidth(struct bdi_writeback *wb,
                                unsigned long start_time)
 {
-       __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
+       __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time);
 }
 
 /*
@@ -906,7 +905,7 @@ static long wb_writeback(struct bdi_writeback *wb,
                 * For background writeout, stop when we are below the
                 * background dirty threshold
                 */
-               if (work->for_background && !over_bground_thresh(wb->bdi))
+               if (work->for_background && !over_bground_thresh(wb))
                        break;
 
                /*
@@ -998,7 +997,7 @@ static unsigned long get_nr_dirty_pages(void)
 
 static long wb_check_background_flush(struct bdi_writeback *wb)
 {
-       if (over_bground_thresh(wb->bdi)) {
+       if (over_bground_thresh(wb)) {
 
                struct wb_writeback_work work = {
                        .nr_pages       = LONG_MAX,
index 859c6ed..2982445 100644 (file)
@@ -748,7 +748,7 @@ static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
 
        if (wbc->sync_mode == WB_SYNC_ALL)
                gfs2_log_flush(GFS2_SB(inode), ip->i_gl, NORMAL_FLUSH);
-       if (bdi->dirty_exceeded)
+       if (bdi->wb.dirty_exceeded)
                gfs2_ail1_flush(sdp, wbc);
        else
                filemap_fdatawrite(metamapping);
index fe7a907..2ab0604 100644 (file)
@@ -60,16 +60,6 @@ struct bdi_writeback {
        spinlock_t list_lock;           /* protects the b_* lists */
 
        struct percpu_counter stat[NR_WB_STAT_ITEMS];
-};
-
-struct backing_dev_info {
-       struct list_head bdi_list;
-       unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */
-       unsigned int capabilities; /* Device capabilities */
-       congested_fn *congested_fn; /* Function pointer if device is md/dm */
-       void *congested_data;   /* Pointer to aux data for congested func */
-
-       char *name;
 
        unsigned long bw_time_stamp;    /* last time write bw is updated */
        unsigned long dirtied_stamp;
@@ -88,6 +78,16 @@ struct backing_dev_info {
 
        struct fprop_local_percpu completions;
        int dirty_exceeded;
+};
+
+struct backing_dev_info {
+       struct list_head bdi_list;
+       unsigned long ra_pages; /* max readahead in PAGE_CACHE_SIZE units */
+       unsigned int capabilities; /* Device capabilities */
+       congested_fn *congested_fn; /* Function pointer if device is md/dm */
+       void *congested_data;   /* Pointer to aux data for congested func */
+
+       char *name;
 
        unsigned int min_ratio;
        unsigned int max_ratio, max_prop_frac;
index b2dd371..a6b9db7 100644 (file)
@@ -155,16 +155,15 @@ int dirty_writeback_centisecs_handler(struct ctl_table *, int,
                                      void __user *, size_t *, loff_t *);
 
 void global_dirty_limits(unsigned long *pbackground, unsigned long *pdirty);
-unsigned long bdi_dirty_limit(struct backing_dev_info *bdi,
-                              unsigned long dirty);
-
-void __bdi_update_bandwidth(struct backing_dev_info *bdi,
-                           unsigned long thresh,
-                           unsigned long bg_thresh,
-                           unsigned long dirty,
-                           unsigned long bdi_thresh,
-                           unsigned long bdi_dirty,
-                           unsigned long start_time);
+unsigned long wb_dirty_limit(struct bdi_writeback *wb, unsigned long dirty);
+
+void __wb_update_bandwidth(struct bdi_writeback *wb,
+                          unsigned long thresh,
+                          unsigned long bg_thresh,
+                          unsigned long dirty,
+                          unsigned long bdi_thresh,
+                          unsigned long bdi_dirty,
+                          unsigned long start_time);
 
 void page_writeback_init(void);
 void balance_dirty_pages_ratelimited(struct address_space *mapping);
index 880dd74..9b876f6 100644 (file)
@@ -400,13 +400,13 @@ TRACE_EVENT(bdi_dirty_ratelimit,
 
        TP_fast_assign(
                strlcpy(__entry->bdi, dev_name(bdi->dev), 32);
-               __entry->write_bw       = KBps(bdi->write_bandwidth);
-               __entry->avg_write_bw   = KBps(bdi->avg_write_bandwidth);
+               __entry->write_bw       = KBps(bdi->wb.write_bandwidth);
+               __entry->avg_write_bw   = KBps(bdi->wb.avg_write_bandwidth);
                __entry->dirty_rate     = KBps(dirty_rate);
-               __entry->dirty_ratelimit = KBps(bdi->dirty_ratelimit);
+               __entry->dirty_ratelimit = KBps(bdi->wb.dirty_ratelimit);
                __entry->task_ratelimit = KBps(task_ratelimit);
                __entry->balanced_dirty_ratelimit =
-                                         KBps(bdi->balanced_dirty_ratelimit);
+                                       KBps(bdi->wb.balanced_dirty_ratelimit);
        ),
 
        TP_printk("bdi %s: "
index 7b1d191..9a6c472 100644 (file)
@@ -66,7 +66,7 @@ static int bdi_debug_stats_show(struct seq_file *m, void *v)
        spin_unlock(&wb->list_lock);
 
        global_dirty_limits(&background_thresh, &dirty_thresh);
-       bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
+       bdi_thresh = wb_dirty_limit(wb, dirty_thresh);
 
 #define K(x) ((x) << (PAGE_SHIFT - 10))
        seq_printf(m,
@@ -91,7 +91,7 @@ static int bdi_debug_stats_show(struct seq_file *m, void *v)
                   K(background_thresh),
                   (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
                   (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
-                  (unsigned long) K(bdi->write_bandwidth),
+                  (unsigned long) K(wb->write_bandwidth),
                   nr_dirty,
                   nr_io,
                   nr_more_io,
@@ -376,6 +376,11 @@ void bdi_unregister(struct backing_dev_info *bdi)
 }
 EXPORT_SYMBOL(bdi_unregister);
 
+/*
+ * Initial write bandwidth: 100 MB/s
+ */
+#define INIT_BW                (100 << (20 - PAGE_SHIFT))
+
 static int bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
 {
        int i, err;
@@ -391,11 +396,22 @@ static int bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
        spin_lock_init(&wb->list_lock);
        INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
 
+       wb->bw_time_stamp = jiffies;
+       wb->balanced_dirty_ratelimit = INIT_BW;
+       wb->dirty_ratelimit = INIT_BW;
+       wb->write_bandwidth = INIT_BW;
+       wb->avg_write_bandwidth = INIT_BW;
+
+       err = fprop_local_init_percpu(&wb->completions, GFP_KERNEL);
+       if (err)
+               return err;
+
        for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
                err = percpu_counter_init(&wb->stat[i], 0, GFP_KERNEL);
                if (err) {
                        while (--i)
                                percpu_counter_destroy(&wb->stat[i]);
+                       fprop_local_destroy_percpu(&wb->completions);
                        return err;
                }
        }
@@ -411,12 +427,9 @@ static void bdi_wb_exit(struct bdi_writeback *wb)
 
        for (i = 0; i < NR_WB_STAT_ITEMS; i++)
                percpu_counter_destroy(&wb->stat[i]);
-}
 
-/*
- * Initial write bandwidth: 100 MB/s
- */
-#define INIT_BW                (100 << (20 - PAGE_SHIFT))
+       fprop_local_destroy_percpu(&wb->completions);
+}
 
 int bdi_init(struct backing_dev_info *bdi)
 {
@@ -435,22 +448,6 @@ int bdi_init(struct backing_dev_info *bdi)
        if (err)
                return err;
 
-       bdi->dirty_exceeded = 0;
-
-       bdi->bw_time_stamp = jiffies;
-       bdi->written_stamp = 0;
-
-       bdi->balanced_dirty_ratelimit = INIT_BW;
-       bdi->dirty_ratelimit = INIT_BW;
-       bdi->write_bandwidth = INIT_BW;
-       bdi->avg_write_bandwidth = INIT_BW;
-
-       err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL);
-       if (err) {
-               bdi_wb_exit(&bdi->wb);
-               return err;
-       }
-
        return 0;
 }
 EXPORT_SYMBOL(bdi_init);
@@ -468,8 +465,6 @@ void bdi_destroy(struct backing_dev_info *bdi)
        }
 
        bdi_wb_exit(&bdi->wb);
-
-       fprop_local_destroy_percpu(&bdi->completions);
 }
 EXPORT_SYMBOL(bdi_destroy);
 
index dc673a0..cd39ee9 100644 (file)
@@ -399,7 +399,7 @@ static unsigned long wp_next_time(unsigned long cur_time)
 static inline void __wb_writeout_inc(struct bdi_writeback *wb)
 {
        __inc_wb_stat(wb, WB_WRITTEN);
-       __fprop_inc_percpu_max(&writeout_completions, &wb->bdi->completions,
+       __fprop_inc_percpu_max(&writeout_completions, &wb->completions,
                               wb->bdi->max_prop_frac);
        /* First event after period switching was turned off? */
        if (!unlikely(writeout_period_time)) {
@@ -427,10 +427,10 @@ EXPORT_SYMBOL_GPL(wb_writeout_inc);
 /*
  * Obtain an accurate fraction of the BDI's portion.
  */
-static void bdi_writeout_fraction(struct backing_dev_info *bdi,
-               long *numerator, long *denominator)
+static void wb_writeout_fraction(struct bdi_writeback *wb,
+                                long *numerator, long *denominator)
 {
-       fprop_fraction_percpu(&writeout_completions, &bdi->completions,
+       fprop_fraction_percpu(&writeout_completions, &wb->completions,
                                numerator, denominator);
 }
 
@@ -516,11 +516,11 @@ static unsigned long hard_dirty_limit(unsigned long thresh)
 }
 
 /**
- * bdi_dirty_limit - @bdi's share of dirty throttling threshold
- * @bdi: the backing_dev_info to query
+ * wb_dirty_limit - @wb's share of dirty throttling threshold
+ * @wb: bdi_writeback to query
  * @dirty: global dirty limit in pages
  *
- * Returns @bdi's dirty limit in pages. The term "dirty" in the context of
+ * Returns @wb's dirty limit in pages. The term "dirty" in the context of
  * dirty balancing includes all PG_dirty, PG_writeback and NFS unstable pages.
  *
  * Note that balance_dirty_pages() will only seriously take it as a hard limit
@@ -528,34 +528,35 @@ static unsigned long hard_dirty_limit(unsigned long thresh)
  * control. For example, when the device is completely stalled due to some error
  * conditions, or when there are 1000 dd tasks writing to a slow 10MB/s USB key.
  * In the other normal situations, it acts more gently by throttling the tasks
- * more (rather than completely block them) when the bdi dirty pages go high.
+ * more (rather than completely block them) when the wb dirty pages go high.
  *
  * It allocates high/low dirty limits to fast/slow devices, in order to prevent
  * - starving fast devices
  * - piling up dirty pages (that will take long time to sync) on slow devices
  *
- * The bdi's share of dirty limit will be adapting to its throughput and
+ * The wb's share of dirty limit will be adapting to its throughput and
  * bounded by the bdi->min_ratio and/or bdi->max_ratio parameters, if set.
  */
-unsigned long bdi_dirty_limit(struct backing_dev_info *bdi, unsigned long dirty)
+unsigned long wb_dirty_limit(struct bdi_writeback *wb, unsigned long dirty)
 {
-       u64 bdi_dirty;
+       struct backing_dev_info *bdi = wb->bdi;
+       u64 wb_dirty;
        long numerator, denominator;
 
        /*
         * Calculate this BDI's share of the dirty ratio.
         */
-       bdi_writeout_fraction(bdi, &numerator, &denominator);
+       wb_writeout_fraction(wb, &numerator, &denominator);
 
-       bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100;
-       bdi_dirty *= numerator;
-       do_div(bdi_dirty, denominator);
+       wb_dirty = (dirty * (100 - bdi_min_ratio)) / 100;
+       wb_dirty *= numerator;
+       do_div(wb_dirty, denominator);
 
-       bdi_dirty += (dirty * bdi->min_ratio) / 100;
-       if (bdi_dirty > (dirty * bdi->max_ratio) / 100)
-               bdi_dirty = dirty * bdi->max_ratio / 100;
+       wb_dirty += (dirty * bdi->min_ratio) / 100;
+       if (wb_dirty > (dirty * bdi->max_ratio) / 100)
+               wb_dirty = dirty * bdi->max_ratio / 100;
 
-       return bdi_dirty;
+       return wb_dirty;
 }
 
 /*
@@ -664,14 +665,14 @@ static long long pos_ratio_polynom(unsigned long setpoint,
  *   card's bdi_dirty may rush to many times higher than bdi_setpoint.
  * - the bdi dirty thresh drops quickly due to change of JBOD workload
  */
-static unsigned long bdi_position_ratio(struct backing_dev_info *bdi,
-                                       unsigned long thresh,
-                                       unsigned long bg_thresh,
-                                       unsigned long dirty,
-                                       unsigned long bdi_thresh,
-                                       unsigned long bdi_dirty)
+static unsigned long wb_position_ratio(struct bdi_writeback *wb,
+                                      unsigned long thresh,
+                                      unsigned long bg_thresh,
+                                      unsigned long dirty,
+                                      unsigned long bdi_thresh,
+                                      unsigned long bdi_dirty)
 {
-       unsigned long write_bw = bdi->avg_write_bandwidth;
+       unsigned long write_bw = wb->avg_write_bandwidth;
        unsigned long freerun = dirty_freerun_ceiling(thresh, bg_thresh);
        unsigned long limit = hard_dirty_limit(thresh);
        unsigned long x_intercept;
@@ -702,12 +703,12 @@ static unsigned long bdi_position_ratio(struct backing_dev_info *bdi,
         * consume arbitrary amount of RAM because it is accounted in
         * NR_WRITEBACK_TEMP which is not involved in calculating "nr_dirty".
         *
-        * Here, in bdi_position_ratio(), we calculate pos_ratio based on
+        * Here, in wb_position_ratio(), we calculate pos_ratio based on
         * two values: bdi_dirty and bdi_thresh. Let's consider an example:
         * total amount of RAM is 16GB, bdi->max_ratio is equal to 1%, global
         * limits are set by default to 10% and 20% (background and throttle).
         * Then bdi_thresh is 1% of 20% of 16GB. This amounts to ~8K pages.
-        * bdi_dirty_limit(bdi, bg_thresh) is about ~4K pages. bdi_setpoint is
+        * wb_dirty_limit(wb, bg_thresh) is about ~4K pages. bdi_setpoint is
         * about ~6K pages (as the average of background and throttle bdi
         * limits). The 3rd order polynomial will provide positive feedback if
         * bdi_dirty is under bdi_setpoint and vice versa.
@@ -717,7 +718,7 @@ static unsigned long bdi_position_ratio(struct backing_dev_info *bdi,
         * much earlier than global "freerun" is reached (~23MB vs. ~2.3GB
         * in the example above).
         */
-       if (unlikely(bdi->capabilities & BDI_CAP_STRICTLIMIT)) {
+       if (unlikely(wb->bdi->capabilities & BDI_CAP_STRICTLIMIT)) {
                long long bdi_pos_ratio;
                unsigned long bdi_bg_thresh;
 
@@ -842,13 +843,13 @@ static unsigned long bdi_position_ratio(struct backing_dev_info *bdi,
        return pos_ratio;
 }
 
-static void bdi_update_write_bandwidth(struct backing_dev_info *bdi,
-                                      unsigned long elapsed,
-                                      unsigned long written)
+static void wb_update_write_bandwidth(struct bdi_writeback *wb,
+                                     unsigned long elapsed,
+                                     unsigned long written)
 {
        const unsigned long period = roundup_pow_of_two(3 * HZ);
-       unsigned long avg = bdi->avg_write_bandwidth;
-       unsigned long old = bdi->write_bandwidth;
+       unsigned long avg = wb->avg_write_bandwidth;
+       unsigned long old = wb->write_bandwidth;
        u64 bw;
 
        /*
@@ -861,14 +862,14 @@ static void bdi_update_write_bandwidth(struct backing_dev_info *bdi,
         * @written may have decreased due to account_page_redirty().
         * Avoid underflowing @bw calculation.
         */
-       bw = written - min(written, bdi->written_stamp);
+       bw = written - min(written, wb->written_stamp);
        bw *= HZ;
        if (unlikely(elapsed > period)) {
                do_div(bw, elapsed);
                avg = bw;
                goto out;
        }
-       bw += (u64)bdi->write_bandwidth * (period - elapsed);
+       bw += (u64)wb->write_bandwidth * (period - elapsed);
        bw >>= ilog2(period);
 
        /*
@@ -881,8 +882,8 @@ static void bdi_update_write_bandwidth(struct backing_dev_info *bdi,
                avg += (old - avg) >> 3;
 
 out:
-       bdi->write_bandwidth = bw;
-       bdi->avg_write_bandwidth = avg;
+       wb->write_bandwidth = bw;
+       wb->avg_write_bandwidth = avg;
 }
 
 /*
@@ -947,20 +948,20 @@ static void global_update_bandwidth(unsigned long thresh,
  * Normal bdi tasks will be curbed at or below it in long term.
  * Obviously it should be around (write_bw / N) when there are N dd tasks.
  */
-static void bdi_update_dirty_ratelimit(struct backing_dev_info *bdi,
-                                      unsigned long thresh,
-                                      unsigned long bg_thresh,
-                                      unsigned long dirty,
-                                      unsigned long bdi_thresh,
-                                      unsigned long bdi_dirty,
-                                      unsigned long dirtied,
-                                      unsigned long elapsed)
+static void wb_update_dirty_ratelimit(struct bdi_writeback *wb,
+                                     unsigned long thresh,
+                                     unsigned long bg_thresh,
+                                     unsigned long dirty,
+                                     unsigned long bdi_thresh,
+                                     unsigned long bdi_dirty,
+                                     unsigned long dirtied,
+                                     unsigned long elapsed)
 {
        unsigned long freerun = dirty_freerun_ceiling(thresh, bg_thresh);
        unsigned long limit = hard_dirty_limit(thresh);
        unsigned long setpoint = (freerun + limit) / 2;
-       unsigned long write_bw = bdi->avg_write_bandwidth;
-       unsigned long dirty_ratelimit = bdi->dirty_ratelimit;
+       unsigned long write_bw = wb->avg_write_bandwidth;
+       unsigned long dirty_ratelimit = wb->dirty_ratelimit;
        unsigned long dirty_rate;
        unsigned long task_ratelimit;
        unsigned long balanced_dirty_ratelimit;
@@ -972,10 +973,10 @@ static void bdi_update_dirty_ratelimit(struct backing_dev_info *bdi,
         * The dirty rate will match the writeout rate in long term, except
         * when dirty pages are truncated by userspace or re-dirtied by FS.
         */
-       dirty_rate = (dirtied - bdi->dirtied_stamp) * HZ / elapsed;
+       dirty_rate = (dirtied - wb->dirtied_stamp) * HZ / elapsed;
 
-       pos_ratio = bdi_position_ratio(bdi, thresh, bg_thresh, dirty,
-                                      bdi_thresh, bdi_dirty);
+       pos_ratio = wb_position_ratio(wb, thresh, bg_thresh, dirty,
+                                     bdi_thresh, bdi_dirty);
        /*
         * task_ratelimit reflects each dd's dirty rate for the past 200ms.
         */
@@ -1059,31 +1060,31 @@ static void bdi_update_dirty_ratelimit(struct backing_dev_info *bdi,
 
        /*
         * For strictlimit case, calculations above were based on bdi counters
-        * and limits (starting from pos_ratio = bdi_position_ratio() and up to
+        * and limits (starting from pos_ratio = wb_position_ratio() and up to
         * balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate).
         * Hence, to calculate "step" properly, we have to use bdi_dirty as
         * "dirty" and bdi_setpoint as "setpoint".
         *
         * We rampup dirty_ratelimit forcibly if bdi_dirty is low because
         * it's possible that bdi_thresh is close to zero due to inactivity
-        * of backing device (see the implementation of bdi_dirty_limit()).
+        * of backing device (see the implementation of wb_dirty_limit()).
         */
-       if (unlikely(bdi->capabilities & BDI_CAP_STRICTLIMIT)) {
+       if (unlikely(wb->bdi->capabilities & BDI_CAP_STRICTLIMIT)) {
                dirty = bdi_dirty;
                if (bdi_dirty < 8)
                        setpoint = bdi_dirty + 1;
                else
                        setpoint = (bdi_thresh +
-                                   bdi_dirty_limit(bdi, bg_thresh)) / 2;
+                                   wb_dirty_limit(wb, bg_thresh)) / 2;
        }
 
        if (dirty < setpoint) {
-               x = min3(bdi->balanced_dirty_ratelimit,
+               x = min3(wb->balanced_dirty_ratelimit,
                         balanced_dirty_ratelimit, task_ratelimit);
                if (dirty_ratelimit < x)
                        step = x - dirty_ratelimit;
        } else {
-               x = max3(bdi->balanced_dirty_ratelimit,
+               x = max3(wb->balanced_dirty_ratelimit,
                         balanced_dirty_ratelimit, task_ratelimit);
                if (dirty_ratelimit > x)
                        step = dirty_ratelimit - x;
@@ -1105,22 +1106,22 @@ static void bdi_update_dirty_ratelimit(struct backing_dev_info *bdi,
        else
                dirty_ratelimit -= step;
 
-       bdi->dirty_ratelimit = max(dirty_ratelimit, 1UL);
-       bdi->balanced_dirty_ratelimit = balanced_dirty_ratelimit;
+       wb->dirty_ratelimit = max(dirty_ratelimit, 1UL);
+       wb->balanced_dirty_ratelimit = balanced_dirty_ratelimit;
 
-       trace_bdi_dirty_ratelimit(bdi, dirty_rate, task_ratelimit);
+       trace_bdi_dirty_ratelimit(wb->bdi, dirty_rate, task_ratelimit);
 }
 
-void __bdi_update_bandwidth(struct backing_dev_info *bdi,
-                           unsigned long thresh,
-                           unsigned long bg_thresh,
-                           unsigned long dirty,
-                           unsigned long bdi_thresh,
-                           unsigned long bdi_dirty,
-                           unsigned long start_time)
+void __wb_update_bandwidth(struct bdi_writeback *wb,
+                          unsigned long thresh,
+                          unsigned long bg_thresh,
+                          unsigned long dirty,
+                          unsigned long bdi_thresh,
+                          unsigned long bdi_dirty,
+                          unsigned long start_time)
 {
        unsigned long now = jiffies;
-       unsigned long elapsed = now - bdi->bw_time_stamp;
+       unsigned long elapsed = now - wb->bw_time_stamp;
        unsigned long dirtied;
        unsigned long written;
 
@@ -1130,44 +1131,44 @@ void __bdi_update_bandwidth(struct backing_dev_info *bdi,
        if (elapsed < BANDWIDTH_INTERVAL)
                return;
 
-       dirtied = percpu_counter_read(&bdi->wb.stat[WB_DIRTIED]);
-       written = percpu_counter_read(&bdi->wb.stat[WB_WRITTEN]);
+       dirtied = percpu_counter_read(&wb->stat[WB_DIRTIED]);
+       written = percpu_counter_read(&wb->stat[WB_WRITTEN]);
 
        /*
         * Skip quiet periods when disk bandwidth is under-utilized.
         * (at least 1s idle time between two flusher runs)
         */
-       if (elapsed > HZ && time_before(bdi->bw_time_stamp, start_time))
+       if (elapsed > HZ && time_before(wb->bw_time_stamp, start_time))
                goto snapshot;
 
        if (thresh) {
                global_update_bandwidth(thresh, dirty, now);
-               bdi_update_dirty_ratelimit(bdi, thresh, bg_thresh, dirty,
-                                          bdi_thresh, bdi_dirty,
-                                          dirtied, elapsed);
+               wb_update_dirty_ratelimit(wb, thresh, bg_thresh, dirty,
+                                         bdi_thresh, bdi_dirty,
+                                         dirtied, elapsed);
        }
-       bdi_update_write_bandwidth(bdi, elapsed, written);
+       wb_update_write_bandwidth(wb, elapsed, written);
 
 snapshot:
-       bdi->dirtied_stamp = dirtied;
-       bdi->written_stamp = written;
-       bdi->bw_time_stamp = now;
+       wb->dirtied_stamp = dirtied;
+       wb->written_stamp = written;
+       wb->bw_time_stamp = now;
 }
 
-static void bdi_update_bandwidth(struct backing_dev_info *bdi,
-                                unsigned long thresh,
-                                unsigned long bg_thresh,
-                                unsigned long dirty,
-                                unsigned long bdi_thresh,
-                                unsigned long bdi_dirty,
-                                unsigned long start_time)
+static void wb_update_bandwidth(struct bdi_writeback *wb,
+                               unsigned long thresh,
+                               unsigned long bg_thresh,
+                               unsigned long dirty,
+                               unsigned long bdi_thresh,
+                               unsigned long bdi_dirty,
+                               unsigned long start_time)
 {
-       if (time_is_after_eq_jiffies(bdi->bw_time_stamp + BANDWIDTH_INTERVAL))
+       if (time_is_after_eq_jiffies(wb->bw_time_stamp + BANDWIDTH_INTERVAL))
                return;
-       spin_lock(&bdi->wb.list_lock);
-       __bdi_update_bandwidth(bdi, thresh, bg_thresh, dirty,
-                              bdi_thresh, bdi_dirty, start_time);
-       spin_unlock(&bdi->wb.list_lock);
+       spin_lock(&wb->list_lock);
+       __wb_update_bandwidth(wb, thresh, bg_thresh, dirty,
+                             bdi_thresh, bdi_dirty, start_time);
+       spin_unlock(&wb->list_lock);
 }
 
 /*
@@ -1187,10 +1188,10 @@ static unsigned long dirty_poll_interval(unsigned long dirty,
        return 1;
 }
 
-static unsigned long bdi_max_pause(struct backing_dev_info *bdi,
-                                  unsigned long bdi_dirty)
+static unsigned long wb_max_pause(struct bdi_writeback *wb,
+                                     unsigned long bdi_dirty)
 {
-       unsigned long bw = bdi->avg_write_bandwidth;
+       unsigned long bw = wb->avg_write_bandwidth;
        unsigned long t;
 
        /*
@@ -1206,14 +1207,14 @@ static unsigned long bdi_max_pause(struct backing_dev_info *bdi,
        return min_t(unsigned long, t, MAX_PAUSE);
 }
 
-static long bdi_min_pause(struct backing_dev_info *bdi,
-                         long max_pause,
-                         unsigned long task_ratelimit,
-                         unsigned long dirty_ratelimit,
-                         int *nr_dirtied_pause)
+static long wb_min_pause(struct bdi_writeback *wb,
+                        long max_pause,
+                        unsigned long task_ratelimit,
+                        unsigned long dirty_ratelimit,
+                        int *nr_dirtied_pause)
 {
-       long hi = ilog2(bdi->avg_write_bandwidth);
-       long lo = ilog2(bdi->dirty_ratelimit);
+       long hi = ilog2(wb->avg_write_bandwidth);
+       long lo = ilog2(wb->dirty_ratelimit);
        long t;         /* target pause */
        long pause;     /* estimated next pause */
        int pages;      /* target nr_dirtied_pause */
@@ -1281,14 +1282,13 @@ static long bdi_min_pause(struct backing_dev_info *bdi,
        return pages >= DIRTY_POLL_THRESH ? 1 + t / 2 : t;
 }
 
-static inline void bdi_dirty_limits(struct backing_dev_info *bdi,
-                                   unsigned long dirty_thresh,
-                                   unsigned long background_thresh,
-                                   unsigned long *bdi_dirty,
-                                   unsigned long *bdi_thresh,
-                                   unsigned long *bdi_bg_thresh)
+static inline void wb_dirty_limits(struct bdi_writeback *wb,
+                                  unsigned long dirty_thresh,
+                                  unsigned long background_thresh,
+                                  unsigned long *bdi_dirty,
+                                  unsigned long *bdi_thresh,
+                                  unsigned long *bdi_bg_thresh)
 {
-       struct bdi_writeback *wb = &bdi->wb;
        unsigned long wb_reclaimable;
 
        /*
@@ -1301,10 +1301,10 @@ static inline void bdi_dirty_limits(struct backing_dev_info *bdi,
         *   In this case we don't want to hard throttle the USB key
         *   dirtiers for 100 seconds until bdi_dirty drops under
         *   bdi_thresh. Instead the auxiliary bdi control line in
-        *   bdi_position_ratio() will let the dirtier task progress
+        *   wb_position_ratio() will let the dirtier task progress
         *   at some rate <= (write_bw / 2) for bringing down bdi_dirty.
         */
-       *bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
+       *bdi_thresh = wb_dirty_limit(wb, dirty_thresh);
 
        if (bdi_bg_thresh)
                *bdi_bg_thresh = dirty_thresh ? div_u64((u64)*bdi_thresh *
@@ -1354,6 +1354,7 @@ static void balance_dirty_pages(struct address_space *mapping,
        unsigned long dirty_ratelimit;
        unsigned long pos_ratio;
        struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
+       struct bdi_writeback *wb = &bdi->wb;
        bool strictlimit = bdi->capabilities & BDI_CAP_STRICTLIMIT;
        unsigned long start_time = jiffies;
 
@@ -1378,8 +1379,8 @@ static void balance_dirty_pages(struct address_space *mapping,
                global_dirty_limits(&background_thresh, &dirty_thresh);
 
                if (unlikely(strictlimit)) {
-                       bdi_dirty_limits(bdi, dirty_thresh, background_thresh,
-                                        &bdi_dirty, &bdi_thresh, &bg_thresh);
+                       wb_dirty_limits(wb, dirty_thresh, background_thresh,
+                                       &bdi_dirty, &bdi_thresh, &bg_thresh);
 
                        dirty = bdi_dirty;
                        thresh = bdi_thresh;
@@ -1410,28 +1411,28 @@ static void balance_dirty_pages(struct address_space *mapping,
                        bdi_start_background_writeback(bdi);
 
                if (!strictlimit)
-                       bdi_dirty_limits(bdi, dirty_thresh, background_thresh,
-                                        &bdi_dirty, &bdi_thresh, NULL);
+                       wb_dirty_limits(wb, dirty_thresh, background_thresh,
+                                       &bdi_dirty, &bdi_thresh, NULL);
 
                dirty_exceeded = (bdi_dirty > bdi_thresh) &&
                                 ((nr_dirty > dirty_thresh) || strictlimit);
-               if (dirty_exceeded && !bdi->dirty_exceeded)
-                       bdi->dirty_exceeded = 1;
+               if (dirty_exceeded && !wb->dirty_exceeded)
+                       wb->dirty_exceeded = 1;
 
-               bdi_update_bandwidth(bdi, dirty_thresh, background_thresh,
-                                    nr_dirty, bdi_thresh, bdi_dirty,
-                                    start_time);
+               wb_update_bandwidth(wb, dirty_thresh, background_thresh,
+                                   nr_dirty, bdi_thresh, bdi_dirty,
+                                   start_time);
 
-               dirty_ratelimit = bdi->dirty_ratelimit;
-               pos_ratio = bdi_position_ratio(bdi, dirty_thresh,
-                                              background_thresh, nr_dirty,
-                                              bdi_thresh, bdi_dirty);
+               dirty_ratelimit = wb->dirty_ratelimit;
+               pos_ratio = wb_position_ratio(wb, dirty_thresh,
+                                             background_thresh, nr_dirty,
+                                             bdi_thresh, bdi_dirty);
                task_ratelimit = ((u64)dirty_ratelimit * pos_ratio) >>
                                                        RATELIMIT_CALC_SHIFT;
-               max_pause = bdi_max_pause(bdi, bdi_dirty);
-               min_pause = bdi_min_pause(bdi, max_pause,
-                                         task_ratelimit, dirty_ratelimit,
-                                         &nr_dirtied_pause);
+               max_pause = wb_max_pause(wb, bdi_dirty);
+               min_pause = wb_min_pause(wb, max_pause,
+                                        task_ratelimit, dirty_ratelimit,
+                                        &nr_dirtied_pause);
 
                if (unlikely(task_ratelimit == 0)) {
                        period = max_pause;
@@ -1515,15 +1516,15 @@ pause:
                 * more page. However bdi_dirty has accounting errors.  So use
                 * the larger and more IO friendly wb_stat_error.
                 */
-               if (bdi_dirty <= wb_stat_error(&bdi->wb))
+               if (bdi_dirty <= wb_stat_error(wb))
                        break;
 
                if (fatal_signal_pending(current))
                        break;
        }
 
-       if (!dirty_exceeded && bdi->dirty_exceeded)
-               bdi->dirty_exceeded = 0;
+       if (!dirty_exceeded && wb->dirty_exceeded)
+               wb->dirty_exceeded = 0;
 
        if (writeback_in_progress(bdi))
                return;
@@ -1577,6 +1578,7 @@ DEFINE_PER_CPU(int, dirty_throttle_leaks) = 0;
 void balance_dirty_pages_ratelimited(struct address_space *mapping)
 {
        struct backing_dev_info *bdi = inode_to_bdi(mapping->host);
+       struct bdi_writeback *wb = &bdi->wb;
        int ratelimit;
        int *p;
 
@@ -1584,7 +1586,7 @@ void balance_dirty_pages_ratelimited(struct address_space *mapping)
                return;
 
        ratelimit = current->nr_dirtied_pause;
-       if (bdi->dirty_exceeded)
+       if (wb->dirty_exceeded)
                ratelimit = min(ratelimit, 32 >> (PAGE_SHIFT - 10));
 
        preempt_disable();