linux - drivers/md/dm-writecache.c

2773 lines · 69.2 KiB

1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* Copyright (C) 2018 Red Hat. All rights reserved.
4	*
5	* This file is released under the GPL.
6	*/
7
8	#include <linux/device-mapper.h>
9	#include <linux/module.h>
10	#include <linux/init.h>
11	#include <linux/vmalloc.h>
12	#include <linux/kthread.h>
13	#include <linux/dm-io.h>
14	#include <linux/dm-kcopyd.h>
15	#include <linux/dax.h>
16	#include <linux/libnvdimm.h>
17	#include <linux/delay.h>
18	#include "dm-io-tracker.h"
19
20	#define DM_MSG_PREFIX "writecache"
21
22	#define HIGH_WATERMARK 50
23	#define LOW_WATERMARK 45
24	#define MAX_WRITEBACK_JOBS min(0x10000000 / PAGE_SIZE, totalram_pages() / 16)
25	#define ENDIO_LATENCY 16
26	#define WRITEBACK_LATENCY 64
27	#define AUTOCOMMIT_BLOCKS_SSD 65536
28	#define AUTOCOMMIT_BLOCKS_PMEM 64
29	#define AUTOCOMMIT_MSEC 1000
30	#define MAX_AGE_DIV 16
31	#define MAX_AGE_UNSPECIFIED -1UL
32	#define PAUSE_WRITEBACK (HZ * 3)
33
34	#define BITMAP_GRANULARITY 65536
35	#if BITMAP_GRANULARITY < PAGE_SIZE
36	#undef BITMAP_GRANULARITY
37	#define BITMAP_GRANULARITY PAGE_SIZE
38	#endif
39
40	#if IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API) && IS_ENABLED(CONFIG_FS_DAX)
41	#define DM_WRITECACHE_HAS_PMEM
42	#endif
43
44	#ifdef DM_WRITECACHE_HAS_PMEM
45	#define pmem_assign(dest, src) \
46	do { \
47	typeof(dest) uniq = (src); \
48	memcpy_flushcache(&(dest), &uniq, sizeof(dest)); \
49	} while (0)
50	#else
51	#define pmem_assign(dest, src) ((dest) = (src))
52	#endif
53
54	#if IS_ENABLED(CONFIG_ARCH_HAS_COPY_MC) && defined(DM_WRITECACHE_HAS_PMEM)
55	#define DM_WRITECACHE_HANDLE_HARDWARE_ERRORS
56	#endif
57
58	#define MEMORY_SUPERBLOCK_MAGIC 0x23489321
59	#define MEMORY_SUPERBLOCK_VERSION 1
60
61	struct wc_memory_entry {
62	__le64 original_sector;
63	__le64 seq_count;
64	};
65
66	struct wc_memory_superblock {
67	union {
68	struct {
69	__le32 magic;
70	__le32 version;
71	__le32 block_size;
72	__le32 pad;
73	__le64 n_blocks;
74	__le64 seq_count;
75	};
76	__le64 padding[8];
77	};
78	struct wc_memory_entry entries[];
79	};
80
81	struct wc_entry {
82	struct rb_node rb_node;
83	struct list_head lru;
84	unsigned short wc_list_contiguous;
85	#if BITS_PER_LONG == 64
86	bool write_in_progress : 1;
87	unsigned long index : 47;
88	#else
89	bool write_in_progress;
90	unsigned long index;
91	#endif
92	unsigned long age;
93	#ifdef DM_WRITECACHE_HANDLE_HARDWARE_ERRORS
94	uint64_t original_sector;
95	uint64_t seq_count;
96	#endif
97	};
98
99	#ifdef DM_WRITECACHE_HAS_PMEM
100	#define WC_MODE_PMEM(wc) ((wc)->pmem_mode)
101	#define WC_MODE_FUA(wc) ((wc)->writeback_fua)
102	#else
103	#define WC_MODE_PMEM(wc) false
104	#define WC_MODE_FUA(wc) false
105	#endif
106	#define WC_MODE_SORT_FREELIST(wc) (!WC_MODE_PMEM(wc))
107
108	struct dm_writecache {
109	struct mutex lock;
110	struct list_head lru;
111	union {
112	struct list_head freelist;
113	struct {
114	struct rb_root freetree;
115	struct wc_entry *current_free;
116	};
117	};
118	struct rb_root tree;
119
120	size_t freelist_size;
121	size_t writeback_size;
122	size_t freelist_high_watermark;
123	size_t freelist_low_watermark;
124	unsigned long max_age;
125	unsigned long pause;
126
127	unsigned int uncommitted_blocks;
128	unsigned int autocommit_blocks;
129	unsigned int max_writeback_jobs;
130
131	int error;
132
133	unsigned long autocommit_jiffies;
134	struct timer_list autocommit_timer;
135	struct wait_queue_head freelist_wait;
136
137	struct timer_list max_age_timer;
138
139	atomic_t bio_in_progress[2];
140	struct wait_queue_head bio_in_progress_wait[2];
141
142	struct dm_target *ti;
143	struct dm_dev *dev;
144	struct dm_dev *ssd_dev;
145	sector_t start_sector;
146	void *memory_map;
147	uint64_t memory_map_size;
148	size_t metadata_sectors;
149	size_t n_blocks;
150	uint64_t seq_count;
151	sector_t data_device_sectors;
152	void *block_start;
153	struct wc_entry *entries;
154	unsigned int block_size;
155	unsigned char block_size_bits;
156
157	bool pmem_mode:1;
158	bool writeback_fua:1;
159
160	bool overwrote_committed:1;
161	bool memory_vmapped:1;
162
163	bool start_sector_set:1;
164	bool high_wm_percent_set:1;
165	bool low_wm_percent_set:1;
166	bool max_writeback_jobs_set:1;
167	bool autocommit_blocks_set:1;
168	bool autocommit_time_set:1;
169	bool max_age_set:1;
170	bool writeback_fua_set:1;
171	bool flush_on_suspend:1;
172	bool cleaner:1;
173	bool cleaner_set:1;
174	bool metadata_only:1;
175	bool pause_set:1;
176
177	unsigned int high_wm_percent_value;
178	unsigned int low_wm_percent_value;
179	unsigned int autocommit_time_value;
180	unsigned int max_age_value;
181	unsigned int pause_value;
182
183	unsigned int writeback_all;
184	struct workqueue_struct *writeback_wq;
185	struct work_struct writeback_work;
186	struct work_struct flush_work;
187
188	struct dm_io_tracker iot;
189
190	struct dm_io_client *dm_io;
191
192	raw_spinlock_t endio_list_lock;
193	struct list_head endio_list;
194	struct task_struct *endio_thread;
195
196	struct task_struct *flush_thread;
197	struct bio_list flush_list;
198
199	struct dm_kcopyd_client *dm_kcopyd;
200	unsigned long *dirty_bitmap;
201	unsigned int dirty_bitmap_size;
202
203	struct bio_set bio_set;
204	mempool_t copy_pool;
205
206	struct {
207	unsigned long long reads;
208	unsigned long long read_hits;
209	unsigned long long writes;
210	unsigned long long write_hits_uncommitted;
211	unsigned long long write_hits_committed;
212	unsigned long long writes_around;
213	unsigned long long writes_allocate;
214	unsigned long long writes_blocked_on_freelist;
215	unsigned long long flushes;
216	unsigned long long discards;
217	} stats;
218	};
219
220	#define WB_LIST_INLINE 16
221
222	struct writeback_struct {
223	struct list_head endio_entry;
224	struct dm_writecache *wc;
225	struct wc_entry **wc_list;
226	unsigned int wc_list_n;
227	struct wc_entry *wc_list_inline[WB_LIST_INLINE];
228	struct bio bio;
229	};
230
231	struct copy_struct {
232	struct list_head endio_entry;
233	struct dm_writecache *wc;
234	struct wc_entry *e;
235	unsigned int n_entries;
236	int error;
237	};
238
239	DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(dm_writecache_throttle,
240	"A percentage of time allocated for data copying");
241
242	static void wc_lock(struct dm_writecache *wc)
243	{
244	mutex_lock(&wc->lock);
245	}
246
247	static void wc_unlock(struct dm_writecache *wc)
248	{
249	mutex_unlock(&wc->lock);
250	}
251
252	#ifdef DM_WRITECACHE_HAS_PMEM
253	static int persistent_memory_claim(struct dm_writecache *wc)
254	{
255	int r;
256	loff_t s;
257	long p, da;
258	unsigned long pfn;
259	int id;
260	struct page **pages;
261	sector_t offset;
262
263	wc->memory_vmapped = false;
264
265	s = wc->memory_map_size;
266	p = s >> PAGE_SHIFT;
267	if (!p) {
268	r = -EINVAL;
269	goto err1;
270	}
271	if (p != s >> PAGE_SHIFT) {
272	r = -EOVERFLOW;
273	goto err1;
274	}
275
276	offset = get_start_sect(wc->ssd_dev->bdev);
277	if (offset & (PAGE_SIZE / 512 - 1)) {
278	r = -EINVAL;
279	goto err1;
280	}
281	offset >>= PAGE_SHIFT - 9;
282
283	id = dax_read_lock();
284
285	da = dax_direct_access(wc->ssd_dev->dax_dev, offset, p, DAX_ACCESS,
286	&wc->memory_map, &pfn);
287	if (da < 0) {
288	wc->memory_map = NULL;
289	r = da;
290	goto err2;
291	}
292	if (!pfn_valid(pfn)) {
293	wc->memory_map = NULL;
294	r = -EOPNOTSUPP;
295	goto err2;
296	}
297	if (da != p) {
298	long i;
299
300	wc->memory_map = NULL;
301	pages = vmalloc_array(p, sizeof(struct page *));
302	if (!pages) {
303	r = -ENOMEM;
304	goto err2;
305	}
306	i = 0;
307	do {
308	long daa;
309
310	daa = dax_direct_access(wc->ssd_dev->dax_dev, offset + i,
311	p - i, DAX_ACCESS, NULL, &pfn);
312	if (daa <= 0) {
313	r = daa ? daa : -EINVAL;
314	goto err3;
315	}
316	if (!pfn_valid(pfn)) {
317	r = -EOPNOTSUPP;
318	goto err3;
319	}
320	while (daa-- && i < p) {
321	pages[i++] = pfn_to_page(pfn);
322	pfn++;
323	if (!(i & 15))
324	cond_resched();
325	}
326	} while (i < p);
327	wc->memory_map = vmap(pages, p, VM_MAP, PAGE_KERNEL);
328	if (!wc->memory_map) {
329	r = -ENOMEM;
330	goto err3;
331	}
332	vfree(pages);
333	wc->memory_vmapped = true;
334	}
335
336	dax_read_unlock(id);
337
338	wc->memory_map += (size_t)wc->start_sector << SECTOR_SHIFT;
339	wc->memory_map_size -= (size_t)wc->start_sector << SECTOR_SHIFT;
340
341	return 0;
342	err3:
343	vfree(pages);
344	err2:
345	dax_read_unlock(id);
346	err1:
347	return r;
348	}
349	#else
350	static int persistent_memory_claim(struct dm_writecache *wc)
351	{
352	return -EOPNOTSUPP;
353	}
354	#endif
355
356	static void persistent_memory_release(struct dm_writecache *wc)
357	{
358	if (wc->memory_vmapped)
359	vunmap(wc->memory_map - ((size_t)wc->start_sector << SECTOR_SHIFT));
360	}
361
362	static struct page persistent_memory_page(void addr)
363	{
364	if (is_vmalloc_addr(addr))
365	return vmalloc_to_page(addr);
366	else
367	return virt_to_page(addr);
368	}
369
370	static unsigned int persistent_memory_page_offset(void *addr)
371	{
372	return (unsigned long)addr & (PAGE_SIZE - 1);
373	}
374
375	static void persistent_memory_flush_cache(void *ptr, size_t size)
376	{
377	if (is_vmalloc_addr(ptr))
378	flush_kernel_vmap_range(ptr, size);
379	}
380
381	static void persistent_memory_invalidate_cache(void *ptr, size_t size)
382	{
383	if (is_vmalloc_addr(ptr))
384	invalidate_kernel_vmap_range(ptr, size);
385	}
386
387	static struct wc_memory_superblock sb(struct dm_writecache wc)
388	{
389	return wc->memory_map;
390	}
391
392	static struct wc_memory_entry memory_entry(struct dm_writecache wc, struct wc_entry *e)
393	{
394	return &sb(wc)->entries[e->index];
395	}
396
397	static void memory_data(struct dm_writecache wc, struct wc_entry *e)
398	{
399	return (char *)wc->block_start + (e->index << wc->block_size_bits);
400	}
401
402	static sector_t cache_sector(struct dm_writecache wc, struct wc_entry e)
403	{
404	return wc->start_sector + wc->metadata_sectors +
405	((sector_t)e->index << (wc->block_size_bits - SECTOR_SHIFT));
406	}
407
408	static uint64_t read_original_sector(struct dm_writecache wc, struct wc_entry e)
409	{
410	#ifdef DM_WRITECACHE_HANDLE_HARDWARE_ERRORS
411	return e->original_sector;
412	#else
413	return le64_to_cpu(memory_entry(wc, e)->original_sector);
414	#endif
415	}
416
417	static uint64_t read_seq_count(struct dm_writecache wc, struct wc_entry e)
418	{
419	#ifdef DM_WRITECACHE_HANDLE_HARDWARE_ERRORS
420	return e->seq_count;
421	#else
422	return le64_to_cpu(memory_entry(wc, e)->seq_count);
423	#endif
424	}
425
426	static void clear_seq_count(struct dm_writecache wc, struct wc_entry e)
427	{
428	#ifdef DM_WRITECACHE_HANDLE_HARDWARE_ERRORS
429	e->seq_count = -1;
430	#endif
431	pmem_assign(memory_entry(wc, e)->seq_count, cpu_to_le64(-1));
432	}
433
434	static void write_original_sector_seq_count(struct dm_writecache wc, struct wc_entry e,
435	uint64_t original_sector, uint64_t seq_count)
436	{
437	struct wc_memory_entry me;
438	#ifdef DM_WRITECACHE_HANDLE_HARDWARE_ERRORS
439	e->original_sector = original_sector;
440	e->seq_count = seq_count;
441	#endif
442	me.original_sector = cpu_to_le64(original_sector);
443	me.seq_count = cpu_to_le64(seq_count);
444	pmem_assign(*memory_entry(wc, e), me);
445	}
446
447	#define writecache_error(wc, err, msg, arg...) \
448	do { \
449	if (!cmpxchg(&(wc)->error, 0, err)) \
450	DMERR(msg, ##arg); \
451	wake_up(&(wc)->freelist_wait); \
452	} while (0)
453
454	#define writecache_has_error(wc) (unlikely(READ_ONCE((wc)->error)))
455
456	static void writecache_flush_all_metadata(struct dm_writecache *wc)
457	{
458	if (!WC_MODE_PMEM(wc))
459	memset(wc->dirty_bitmap, -1, wc->dirty_bitmap_size);
460	}
461
462	static void writecache_flush_region(struct dm_writecache wc, void ptr, size_t size)
463	{
464	if (!WC_MODE_PMEM(wc))
465	__set_bit(((char )ptr - (char )wc->memory_map) / BITMAP_GRANULARITY,
466	wc->dirty_bitmap);
467	}
468
469	static void writecache_disk_flush(struct dm_writecache wc, struct dm_dev dev);
470
471	struct io_notify {
472	struct dm_writecache *wc;
473	struct completion c;
474	atomic_t count;
475	};
476
477	static void writecache_notify_io(unsigned long error, unsigned long unsup, void *context)
478	{
479	struct io_notify *endio = context;
480
481	if (unlikely(error != 0))
482	writecache_error(endio->wc, -EIO, "error writing metadata");
483	else if (unlikely(unsup != 0))
484	writecache_error(endio->wc, -EOPNOTSUPP, "error writing metadata");
485	BUG_ON(atomic_read(&endio->count) <= 0);
486	if (atomic_dec_and_test(&endio->count))
487	complete(&endio->c);
488	}
489
490	static void writecache_wait_for_ios(struct dm_writecache *wc, int direction)
491	{
492	wait_event(wc->bio_in_progress_wait[direction],
493	!atomic_read(&wc->bio_in_progress[direction]));
494	}
495
496	static void ssd_commit_flushed(struct dm_writecache *wc, bool wait_for_ios)
497	{
498	struct dm_io_region region;
499	struct dm_io_request req;
500	struct io_notify endio = {
501	wc,
502	COMPLETION_INITIALIZER_ONSTACK(endio.c),
503	ATOMIC_INIT(1),
504	};
505	unsigned int bitmap_bits = wc->dirty_bitmap_size * 8;
506	unsigned int i = 0;
507
508	while (1) {
509	unsigned int j;
510
511	i = find_next_bit(wc->dirty_bitmap, bitmap_bits, i);
512	if (unlikely(i == bitmap_bits))
513	break;
514	j = find_next_zero_bit(wc->dirty_bitmap, bitmap_bits, i);
515
516	region.bdev = wc->ssd_dev->bdev;
517	region.sector = (sector_t)i * (BITMAP_GRANULARITY >> SECTOR_SHIFT);
518	region.count = (sector_t)(j - i) * (BITMAP_GRANULARITY >> SECTOR_SHIFT);
519
520	if (unlikely(region.sector >= wc->metadata_sectors))
521	break;
522	if (unlikely(region.sector + region.count > wc->metadata_sectors))
523	region.count = wc->metadata_sectors - region.sector;
524
525	region.sector += wc->start_sector;
526	atomic_inc(&endio.count);
527	req.bi_opf = REQ_OP_WRITE \| REQ_SYNC;
528	req.mem.type = DM_IO_VMA;
529	req.mem.ptr.vma = (char )wc->memory_map + (size_t)i BITMAP_GRANULARITY;
530	req.client = wc->dm_io;
531	req.notify.fn = writecache_notify_io;
532	req.notify.context = &endio;
533
534	/* writing via async dm-io (implied by notify.fn above) won't return an error */
535	(void) dm_io(&req, 1, &region, NULL, NULL, IOPRIO_DEFAULT);
536	i = j;
537	}
538
539	writecache_notify_io(0, 0, &endio);
540	wait_for_completion_io(&endio.c);
541
542	if (wait_for_ios)
543	writecache_wait_for_ios(wc, WRITE);
544
545	writecache_disk_flush(wc, wc->ssd_dev);
546
547	memset(wc->dirty_bitmap, 0, wc->dirty_bitmap_size);
548	}
549
550	static void ssd_commit_superblock(struct dm_writecache *wc)
551	{
552	int r;
553	struct dm_io_region region;
554	struct dm_io_request req;
555
556	region.bdev = wc->ssd_dev->bdev;
557	region.sector = 0;
558	region.count = max(4096U, wc->block_size) >> SECTOR_SHIFT;
559
560	if (unlikely(region.sector + region.count > wc->metadata_sectors))
561	region.count = wc->metadata_sectors - region.sector;
562
563	region.sector += wc->start_sector;
564
565	req.bi_opf = REQ_OP_WRITE \| REQ_SYNC \| REQ_FUA;
566	req.mem.type = DM_IO_VMA;
567	req.mem.ptr.vma = (char *)wc->memory_map;
568	req.client = wc->dm_io;
569	req.notify.fn = NULL;
570	req.notify.context = NULL;
571
572	r = dm_io(&req, 1, &region, NULL, NULL, IOPRIO_DEFAULT);
573	if (unlikely(r))
574	writecache_error(wc, r, "error writing superblock");
575	}
576
577	static void writecache_commit_flushed(struct dm_writecache *wc, bool wait_for_ios)
578	{
579	if (WC_MODE_PMEM(wc))
580	pmem_wmb();
581	else
582	ssd_commit_flushed(wc, wait_for_ios);
583	}
584
585	static void writecache_disk_flush(struct dm_writecache wc, struct dm_dev dev)
586	{
587	int r;
588	struct dm_io_region region;
589	struct dm_io_request req;
590
591	region.bdev = dev->bdev;
592	region.sector = 0;
593	region.count = 0;
594	req.bi_opf = REQ_OP_WRITE \| REQ_PREFLUSH;
595	req.mem.type = DM_IO_KMEM;
596	req.mem.ptr.addr = NULL;
597	req.client = wc->dm_io;
598	req.notify.fn = NULL;
599
600	r = dm_io(&req, 1, &region, NULL, NULL, IOPRIO_DEFAULT);
601	if (unlikely(r))
602	writecache_error(wc, r, "error flushing metadata: %d", r);
603	}
604
605	#define WFE_RETURN_FOLLOWING 1
606	#define WFE_LOWEST_SEQ 2
607
608	static struct wc_entry writecache_find_entry(struct dm_writecache wc,
609	uint64_t block, int flags)
610	{
611	struct wc_entry *e;
612	struct rb_node *node = wc->tree.rb_node;
613
614	if (unlikely(!node))
615	return NULL;
616
617	while (1) {
618	e = container_of(node, struct wc_entry, rb_node);
619	if (read_original_sector(wc, e) == block)
620	break;
621
622	node = (read_original_sector(wc, e) >= block ?
623	e->rb_node.rb_left : e->rb_node.rb_right);
624	if (unlikely(!node)) {
625	if (!(flags & WFE_RETURN_FOLLOWING))
626	return NULL;
627	if (read_original_sector(wc, e) >= block)
628	return e;
629
630	node = rb_next(&e->rb_node);
631	if (unlikely(!node))
632	return NULL;
633
634	e = container_of(node, struct wc_entry, rb_node);
635	return e;
636	}
637	}
638
639	while (1) {
640	struct wc_entry *e2;
641
642	if (flags & WFE_LOWEST_SEQ)
643	node = rb_prev(&e->rb_node);
644	else
645	node = rb_next(&e->rb_node);
646	if (unlikely(!node))
647	return e;
648	e2 = container_of(node, struct wc_entry, rb_node);
649	if (read_original_sector(wc, e2) != block)
650	return e;
651	e = e2;
652	}
653	}
654
655	static void writecache_insert_entry(struct dm_writecache wc, struct wc_entry ins)
656	{
657	struct wc_entry *e;
658	struct rb_node *node = &wc->tree.rb_node, parent = NULL;
659
660	while (*node) {
661	e = container_of(*node, struct wc_entry, rb_node);
662	parent = &e->rb_node;
663	if (read_original_sector(wc, e) > read_original_sector(wc, ins))
664	node = &parent->rb_left;
665	else
666	node = &parent->rb_right;
667	}
668	rb_link_node(&ins->rb_node, parent, node);
669	rb_insert_color(&ins->rb_node, &wc->tree);
670	list_add(&ins->lru, &wc->lru);
671	ins->age = jiffies;
672	}
673
674	static void writecache_unlink(struct dm_writecache wc, struct wc_entry e)
675	{
676	list_del(&e->lru);
677	rb_erase(&e->rb_node, &wc->tree);
678	}
679
680	static void writecache_add_to_freelist(struct dm_writecache wc, struct wc_entry e)
681	{
682	if (WC_MODE_SORT_FREELIST(wc)) {
683	struct rb_node *node = &wc->freetree.rb_node, parent = NULL;
684
685	if (unlikely(!*node))
686	wc->current_free = e;
687	while (*node) {
688	parent = *node;
689	if (&e->rb_node < *node)
690	node = &parent->rb_left;
691	else
692	node = &parent->rb_right;
693	}
694	rb_link_node(&e->rb_node, parent, node);
695	rb_insert_color(&e->rb_node, &wc->freetree);
696	} else {
697	list_add_tail(&e->lru, &wc->freelist);
698	}
699	wc->freelist_size++;
700	}
701
702	static inline void writecache_verify_watermark(struct dm_writecache *wc)
703	{
704	if (unlikely(wc->freelist_size + wc->writeback_size <= wc->freelist_high_watermark))
705	queue_work(wc->writeback_wq, &wc->writeback_work);
706	}
707
708	static void writecache_max_age_timer(struct timer_list *t)
709	{
710	struct dm_writecache *wc = timer_container_of(wc, t, max_age_timer);
711
712	if (!dm_suspended(wc->ti) && !writecache_has_error(wc)) {
713	queue_work(wc->writeback_wq, &wc->writeback_work);
714	mod_timer(&wc->max_age_timer, jiffies + wc->max_age / MAX_AGE_DIV);
715	}
716	}
717
718	static struct wc_entry writecache_pop_from_freelist(struct dm_writecache wc, sector_t expected_sector)
719	{
720	struct wc_entry *e;
721
722	if (WC_MODE_SORT_FREELIST(wc)) {
723	struct rb_node *next;
724
725	if (unlikely(!wc->current_free))
726	return NULL;
727	e = wc->current_free;
728	if (expected_sector != (sector_t)-1 && unlikely(cache_sector(wc, e) != expected_sector))
729	return NULL;
730	next = rb_next(&e->rb_node);
731	rb_erase(&e->rb_node, &wc->freetree);
732	if (unlikely(!next))
733	next = rb_first(&wc->freetree);
734	wc->current_free = next ? container_of(next, struct wc_entry, rb_node) : NULL;
735	} else {
736	if (unlikely(list_empty(&wc->freelist)))
737	return NULL;
738	e = container_of(wc->freelist.next, struct wc_entry, lru);
739	if (expected_sector != (sector_t)-1 && unlikely(cache_sector(wc, e) != expected_sector))
740	return NULL;
741	list_del(&e->lru);
742	}
743	wc->freelist_size--;
744
745	writecache_verify_watermark(wc);
746
747	return e;
748	}
749
750	static void writecache_free_entry(struct dm_writecache wc, struct wc_entry e)
751	{
752	writecache_unlink(wc, e);
753	writecache_add_to_freelist(wc, e);
754	clear_seq_count(wc, e);
755	writecache_flush_region(wc, memory_entry(wc, e), sizeof(struct wc_memory_entry));
756	if (unlikely(waitqueue_active(&wc->freelist_wait)))
757	wake_up(&wc->freelist_wait);
758	}
759
760	static void writecache_wait_on_freelist(struct dm_writecache *wc)
761	{
762	DEFINE_WAIT(wait);
763
764	prepare_to_wait(&wc->freelist_wait, &wait, TASK_UNINTERRUPTIBLE);
765	wc_unlock(wc);
766	io_schedule();
767	finish_wait(&wc->freelist_wait, &wait);
768	wc_lock(wc);
769	}
770
771	static void writecache_poison_lists(struct dm_writecache *wc)
772	{
773	/*
774	* Catch incorrect access to these values while the device is suspended.
775	*/
776	memset(&wc->tree, -1, sizeof(wc->tree));
777	wc->lru.next = LIST_POISON1;
778	wc->lru.prev = LIST_POISON2;
779	wc->freelist.next = LIST_POISON1;
780	wc->freelist.prev = LIST_POISON2;
781	}
782
783	static void writecache_flush_entry(struct dm_writecache wc, struct wc_entry e)
784	{
785	writecache_flush_region(wc, memory_entry(wc, e), sizeof(struct wc_memory_entry));
786	if (WC_MODE_PMEM(wc))
787	writecache_flush_region(wc, memory_data(wc, e), wc->block_size);
788	}
789
790	static bool writecache_entry_is_committed(struct dm_writecache wc, struct wc_entry e)
791	{
792	return read_seq_count(wc, e) < wc->seq_count;
793	}
794
795	static void writecache_flush(struct dm_writecache *wc)
796	{
797	struct wc_entry e, e2;
798	bool need_flush_after_free;
799
800	wc->uncommitted_blocks = 0;
801	timer_delete(&wc->autocommit_timer);
802
803	if (list_empty(&wc->lru))
804	return;
805
806	e = container_of(wc->lru.next, struct wc_entry, lru);
807	if (writecache_entry_is_committed(wc, e)) {
808	if (wc->overwrote_committed) {
809	writecache_wait_for_ios(wc, WRITE);
810	writecache_disk_flush(wc, wc->ssd_dev);
811	wc->overwrote_committed = false;
812	}
813	return;
814	}
815	while (1) {
816	writecache_flush_entry(wc, e);
817	if (unlikely(e->lru.next == &wc->lru))
818	break;
819	e2 = container_of(e->lru.next, struct wc_entry, lru);
820	if (writecache_entry_is_committed(wc, e2))
821	break;
822	e = e2;
823	cond_resched();
824	}
825	writecache_commit_flushed(wc, true);
826
827	wc->seq_count++;
828	pmem_assign(sb(wc)->seq_count, cpu_to_le64(wc->seq_count));
829	if (WC_MODE_PMEM(wc))
830	writecache_commit_flushed(wc, false);
831	else
832	ssd_commit_superblock(wc);
833
834	wc->overwrote_committed = false;
835
836	need_flush_after_free = false;
837	while (1) {
838	/* Free another committed entry with lower seq-count */
839	struct rb_node *rb_node = rb_prev(&e->rb_node);
840
841	if (rb_node) {
842	e2 = container_of(rb_node, struct wc_entry, rb_node);
843	if (read_original_sector(wc, e2) == read_original_sector(wc, e) &&
844	likely(!e2->write_in_progress)) {
845	writecache_free_entry(wc, e2);
846	need_flush_after_free = true;
847	}
848	}
849	if (unlikely(e->lru.prev == &wc->lru))
850	break;
851	e = container_of(e->lru.prev, struct wc_entry, lru);
852	cond_resched();
853	}
854
855	if (need_flush_after_free)
856	writecache_commit_flushed(wc, false);
857	}
858
859	static void writecache_flush_work(struct work_struct *work)
860	{
861	struct dm_writecache *wc = container_of(work, struct dm_writecache, flush_work);
862
863	wc_lock(wc);
864	writecache_flush(wc);
865	wc_unlock(wc);
866	}
867
868	static void writecache_autocommit_timer(struct timer_list *t)
869	{
870	struct dm_writecache *wc = timer_container_of(wc, t, autocommit_timer);
871
872	if (!writecache_has_error(wc))
873	queue_work(wc->writeback_wq, &wc->flush_work);
874	}
875
876	static void writecache_schedule_autocommit(struct dm_writecache *wc)
877	{
878	if (!timer_pending(&wc->autocommit_timer))
879	mod_timer(&wc->autocommit_timer, jiffies + wc->autocommit_jiffies);
880	}
881
882	static void writecache_discard(struct dm_writecache *wc, sector_t start, sector_t end)
883	{
884	struct wc_entry *e;
885	bool discarded_something = false;
886
887	e = writecache_find_entry(wc, start, WFE_RETURN_FOLLOWING \| WFE_LOWEST_SEQ);
888	if (unlikely(!e))
889	return;
890
891	while (read_original_sector(wc, e) < end) {
892	struct rb_node *node = rb_next(&e->rb_node);
893
894	if (likely(!e->write_in_progress)) {
895	if (!discarded_something) {
896	if (!WC_MODE_PMEM(wc)) {
897	writecache_wait_for_ios(wc, READ);
898	writecache_wait_for_ios(wc, WRITE);
899	}
900	discarded_something = true;
901	}
902	if (!writecache_entry_is_committed(wc, e))
903	wc->uncommitted_blocks--;
904	writecache_free_entry(wc, e);
905	}
906
907	if (unlikely(!node))
908	break;
909
910	e = container_of(node, struct wc_entry, rb_node);
911	}
912
913	if (discarded_something)
914	writecache_commit_flushed(wc, false);
915	}
916
917	static bool writecache_wait_for_writeback(struct dm_writecache *wc)
918	{
919	if (wc->writeback_size) {
920	writecache_wait_on_freelist(wc);
921	return true;
922	}
923	return false;
924	}
925
926	static void writecache_suspend(struct dm_target *ti)
927	{
928	struct dm_writecache *wc = ti->private;
929	bool flush_on_suspend;
930
931	timer_delete_sync(&wc->autocommit_timer);
932	timer_delete_sync(&wc->max_age_timer);
933
934	wc_lock(wc);
935	writecache_flush(wc);
936	flush_on_suspend = wc->flush_on_suspend;
937	if (flush_on_suspend) {
938	wc->flush_on_suspend = false;
939	wc->writeback_all++;
940	queue_work(wc->writeback_wq, &wc->writeback_work);
941	}
942	wc_unlock(wc);
943
944	drain_workqueue(wc->writeback_wq);
945
946	wc_lock(wc);
947	if (flush_on_suspend)
948	wc->writeback_all--;
949	while (writecache_wait_for_writeback(wc))
950	;
951
952	if (WC_MODE_PMEM(wc))
953	persistent_memory_flush_cache(wc->memory_map, wc->memory_map_size);
954
955	writecache_poison_lists(wc);
956
957	wc_unlock(wc);
958	}
959
960	static int writecache_alloc_entries(struct dm_writecache *wc)
961	{
962	size_t b;
963
964	if (wc->entries)
965	return 0;
966	wc->entries = vmalloc_array(wc->n_blocks, sizeof(struct wc_entry));
967	if (!wc->entries)
968	return -ENOMEM;
969	for (b = 0; b < wc->n_blocks; b++) {
970	struct wc_entry *e = &wc->entries[b];
971
972	e->index = b;
973	e->write_in_progress = false;
974	cond_resched();
975	}
976
977	return 0;
978	}
979
980	static int writecache_read_metadata(struct dm_writecache *wc, sector_t n_sectors)
981	{
982	struct dm_io_region region;
983	struct dm_io_request req;
984
985	region.bdev = wc->ssd_dev->bdev;
986	region.sector = wc->start_sector;
987	region.count = n_sectors;
988	req.bi_opf = REQ_OP_READ \| REQ_SYNC;
989	req.mem.type = DM_IO_VMA;
990	req.mem.ptr.vma = (char *)wc->memory_map;
991	req.client = wc->dm_io;
992	req.notify.fn = NULL;
993
994	return dm_io(&req, 1, &region, NULL, NULL, IOPRIO_DEFAULT);
995	}
996
997	static void writecache_resume(struct dm_target *ti)
998	{
999	struct dm_writecache *wc = ti->private;
1000	size_t b;
1001	bool need_flush = false;
1002	__le64 sb_seq_count;
1003	int r;
1004
1005	wc_lock(wc);
1006
1007	wc->data_device_sectors = bdev_nr_sectors(wc->dev->bdev);
1008
1009	if (WC_MODE_PMEM(wc)) {
1010	persistent_memory_invalidate_cache(wc->memory_map, wc->memory_map_size);
1011	} else {
1012	r = writecache_read_metadata(wc, wc->metadata_sectors);
1013	if (r) {
1014	size_t sb_entries_offset;
1015
1016	writecache_error(wc, r, "unable to read metadata: %d", r);
1017	sb_entries_offset = offsetof(struct wc_memory_superblock, entries);
1018	memset((char *)wc->memory_map + sb_entries_offset, -1,
1019	(wc->metadata_sectors << SECTOR_SHIFT) - sb_entries_offset);
1020	}
1021	}
1022
1023	wc->tree = RB_ROOT;
1024	INIT_LIST_HEAD(&wc->lru);
1025	if (WC_MODE_SORT_FREELIST(wc)) {
1026	wc->freetree = RB_ROOT;
1027	wc->current_free = NULL;
1028	} else {
1029	INIT_LIST_HEAD(&wc->freelist);
1030	}
1031	wc->freelist_size = 0;
1032
1033	r = copy_mc_to_kernel(&sb_seq_count, &sb(wc)->seq_count,
1034	sizeof(uint64_t));
1035	if (r) {
1036	writecache_error(wc, r, "hardware memory error when reading superblock: %d", r);
1037	sb_seq_count = cpu_to_le64(0);
1038	}
1039	wc->seq_count = le64_to_cpu(sb_seq_count);
1040
1041	#ifdef DM_WRITECACHE_HANDLE_HARDWARE_ERRORS
1042	for (b = 0; b < wc->n_blocks; b++) {
1043	struct wc_entry *e = &wc->entries[b];
1044	struct wc_memory_entry wme;
1045
1046	if (writecache_has_error(wc)) {
1047	e->original_sector = -1;
1048	e->seq_count = -1;
1049	continue;
1050	}
1051	r = copy_mc_to_kernel(&wme, memory_entry(wc, e),
1052	sizeof(struct wc_memory_entry));
1053	if (r) {
1054	writecache_error(wc, r, "hardware memory error when reading metadata entry %lu: %d",
1055	(unsigned long)b, r);
1056	e->original_sector = -1;
1057	e->seq_count = -1;
1058	} else {
1059	e->original_sector = le64_to_cpu(wme.original_sector);
1060	e->seq_count = le64_to_cpu(wme.seq_count);
1061	}
1062	cond_resched();
1063	}
1064	#endif
1065	for (b = 0; b < wc->n_blocks; b++) {
1066	struct wc_entry *e = &wc->entries[b];
1067
1068	if (!writecache_entry_is_committed(wc, e)) {
1069	if (read_seq_count(wc, e) != -1) {
1070	erase_this:
1071	clear_seq_count(wc, e);
1072	need_flush = true;
1073	}
1074	writecache_add_to_freelist(wc, e);
1075	} else {
1076	struct wc_entry *old;
1077
1078	old = writecache_find_entry(wc, read_original_sector(wc, e), 0);
1079	if (!old) {
1080	writecache_insert_entry(wc, e);
1081	} else {
1082	if (read_seq_count(wc, old) == read_seq_count(wc, e)) {
1083	writecache_error(wc, -EINVAL,
1084	"two identical entries, position %llu, sector %llu, sequence %llu",
1085	(unsigned long long)b, (unsigned long long)read_original_sector(wc, e),
1086	(unsigned long long)read_seq_count(wc, e));
1087	}
1088	if (read_seq_count(wc, old) > read_seq_count(wc, e)) {
1089	goto erase_this;
1090	} else {
1091	writecache_free_entry(wc, old);
1092	writecache_insert_entry(wc, e);
1093	need_flush = true;
1094	}
1095	}
1096	}
1097	cond_resched();
1098	}
1099
1100	if (need_flush) {
1101	writecache_flush_all_metadata(wc);
1102	writecache_commit_flushed(wc, false);
1103	}
1104
1105	writecache_verify_watermark(wc);
1106
1107	if (wc->max_age != MAX_AGE_UNSPECIFIED)
1108	mod_timer(&wc->max_age_timer, jiffies + wc->max_age / MAX_AGE_DIV);
1109
1110	wc_unlock(wc);
1111	}
1112
1113	static int process_flush_mesg(unsigned int argc, char *argv, struct dm_writecache wc)
1114	{
1115	if (argc != 1)
1116	return -EINVAL;
1117
1118	wc_lock(wc);
1119	if (dm_suspended(wc->ti)) {
1120	wc_unlock(wc);
1121	return -EBUSY;
1122	}
1123	if (writecache_has_error(wc)) {
1124	wc_unlock(wc);
1125	return -EIO;
1126	}
1127
1128	writecache_flush(wc);
1129	wc->writeback_all++;
1130	queue_work(wc->writeback_wq, &wc->writeback_work);
1131	wc_unlock(wc);
1132
1133	flush_workqueue(wc->writeback_wq);
1134
1135	wc_lock(wc);
1136	wc->writeback_all--;
1137	if (writecache_has_error(wc)) {
1138	wc_unlock(wc);
1139	return -EIO;
1140	}
1141	wc_unlock(wc);
1142
1143	return 0;
1144	}
1145
1146	static int process_flush_on_suspend_mesg(unsigned int argc, char *argv, struct dm_writecache wc)
1147	{
1148	if (argc != 1)
1149	return -EINVAL;
1150
1151	wc_lock(wc);
1152	wc->flush_on_suspend = true;
1153	wc_unlock(wc);
1154
1155	return 0;
1156	}
1157
1158	static void activate_cleaner(struct dm_writecache *wc)
1159	{
1160	wc->flush_on_suspend = true;
1161	wc->cleaner = true;
1162	wc->freelist_high_watermark = wc->n_blocks;
1163	wc->freelist_low_watermark = wc->n_blocks;
1164	}
1165
1166	static int process_cleaner_mesg(unsigned int argc, char *argv, struct dm_writecache wc)
1167	{
1168	if (argc != 1)
1169	return -EINVAL;
1170
1171	wc_lock(wc);
1172	activate_cleaner(wc);
1173	if (!dm_suspended(wc->ti))
1174	writecache_verify_watermark(wc);
1175	wc_unlock(wc);
1176
1177	return 0;
1178	}
1179
1180	static int process_clear_stats_mesg(unsigned int argc, char *argv, struct dm_writecache wc)
1181	{
1182	if (argc != 1)
1183	return -EINVAL;
1184
1185	wc_lock(wc);
1186	memset(&wc->stats, 0, sizeof(wc->stats));
1187	wc_unlock(wc);
1188
1189	return 0;
1190	}
1191
1192	static int writecache_message(struct dm_target ti, unsigned int argc, char *argv,
1193	char *result, unsigned int maxlen)
1194	{
1195	int r = -EINVAL;
1196	struct dm_writecache *wc = ti->private;
1197
1198	if (!strcasecmp(argv[0], "flush"))
1199	r = process_flush_mesg(argc, argv, wc);
1200	else if (!strcasecmp(argv[0], "flush_on_suspend"))
1201	r = process_flush_on_suspend_mesg(argc, argv, wc);
1202	else if (!strcasecmp(argv[0], "cleaner"))
1203	r = process_cleaner_mesg(argc, argv, wc);
1204	else if (!strcasecmp(argv[0], "clear_stats"))
1205	r = process_clear_stats_mesg(argc, argv, wc);
1206	else
1207	DMERR("unrecognised message received: %s", argv[0]);
1208
1209	return r;
1210	}
1211
1212	static void memcpy_flushcache_optimized(void dest, void source, size_t size)
1213	{
1214	/*
1215	* clflushopt performs better with block size 1024, 2048, 4096
1216	* non-temporal stores perform better with block size 512
1217	*
1218	* block size 512 1024 2048 4096
1219	* movnti 496 MB/s 642 MB/s 725 MB/s 744 MB/s
1220	* clflushopt 373 MB/s 688 MB/s 1.1 GB/s 1.2 GB/s
1221	*
1222	* We see that movnti performs better for 512-byte blocks, and
1223	* clflushopt performs better for 1024-byte and larger blocks. So, we
1224	* prefer clflushopt for sizes >= 768.
1225	*
1226	* NOTE: this happens to be the case now (with dm-writecache's single
1227	* threaded model) but re-evaluate this once memcpy_flushcache() is
1228	* enabled to use movdir64b which might invalidate this performance
1229	* advantage seen with cache-allocating-writes plus flushing.
1230	*/
1231	#ifdef CONFIG_X86
1232	if (cpu_feature_enabled(X86_FEATURE_CLFLUSHOPT) &&
1233	likely(boot_cpu_data.x86_clflush_size == 64) &&
1234	likely(size >= 768)) {
1235	do {
1236	memcpy((void )dest, (void )source, 64);
1237	clflushopt((void *)dest);
1238	dest += 64;
1239	source += 64;
1240	size -= 64;
1241	} while (size >= 64);
1242	return;
1243	}
1244	#endif
1245	memcpy_flushcache(dest, source, size);
1246	}
1247
1248	static void bio_copy_block(struct dm_writecache wc, struct bio bio, void *data)
1249	{
1250	void *buf;
1251	unsigned int size;
1252	int rw = bio_data_dir(bio);
1253	unsigned int remaining_size = wc->block_size;
1254
1255	do {
1256	struct bio_vec bv = bio_iter_iovec(bio, bio->bi_iter);
1257
1258	buf = bvec_kmap_local(&bv);
1259	size = bv.bv_len;
1260	if (unlikely(size > remaining_size))
1261	size = remaining_size;
1262
1263	if (rw == READ) {
1264	int r;
1265
1266	r = copy_mc_to_kernel(buf, data, size);
1267	flush_dcache_page(bio_page(bio));
1268	if (unlikely(r)) {
1269	writecache_error(wc, r, "hardware memory error when reading data: %d", r);
1270	bio->bi_status = BLK_STS_IOERR;
1271	}
1272	} else {
1273	flush_dcache_page(bio_page(bio));
1274	memcpy_flushcache_optimized(data, buf, size);
1275	}
1276
1277	kunmap_local(buf);
1278
1279	data = (char *)data + size;
1280	remaining_size -= size;
1281	bio_advance(bio, size);
1282	} while (unlikely(remaining_size));
1283	}
1284
1285	static int writecache_flush_thread(void *data)
1286	{
1287	struct dm_writecache *wc = data;
1288
1289	while (1) {
1290	struct bio *bio;
1291
1292	wc_lock(wc);
1293	bio = bio_list_pop(&wc->flush_list);
1294	if (!bio) {
1295	set_current_state(TASK_INTERRUPTIBLE);
1296	wc_unlock(wc);
1297
1298	if (unlikely(kthread_should_stop())) {
1299	set_current_state(TASK_RUNNING);
1300	break;
1301	}
1302
1303	schedule();
1304	continue;
1305	}
1306
1307	if (bio_op(bio) == REQ_OP_DISCARD) {
1308	writecache_discard(wc, bio->bi_iter.bi_sector,
1309	bio_end_sector(bio));
1310	wc_unlock(wc);
1311	bio_set_dev(bio, wc->dev->bdev);
1312	submit_bio_noacct(bio);
1313	} else {
1314	writecache_flush(wc);
1315	wc_unlock(wc);
1316	if (writecache_has_error(wc))
1317	bio->bi_status = BLK_STS_IOERR;
1318	bio_endio(bio);
1319	}
1320	}
1321
1322	return 0;
1323	}
1324
1325	static void writecache_offload_bio(struct dm_writecache wc, struct bio bio)
1326	{
1327	if (bio_list_empty(&wc->flush_list))
1328	wake_up_process(wc->flush_thread);
1329	bio_list_add(&wc->flush_list, bio);
1330	}
1331
1332	enum wc_map_op {
1333	WC_MAP_SUBMIT,
1334	WC_MAP_REMAP,
1335	WC_MAP_REMAP_ORIGIN,
1336	WC_MAP_RETURN,
1337	WC_MAP_ERROR,
1338	};
1339
1340	static void writecache_map_remap_origin(struct dm_writecache wc, struct bio bio,
1341	struct wc_entry *e)
1342	{
1343	if (e) {
1344	sector_t next_boundary =
1345	read_original_sector(wc, e) - bio->bi_iter.bi_sector;
1346	if (next_boundary < bio->bi_iter.bi_size >> SECTOR_SHIFT)
1347	dm_accept_partial_bio(bio, next_boundary);
1348	}
1349	}
1350
1351	static enum wc_map_op writecache_map_read(struct dm_writecache wc, struct bio bio)
1352	{
1353	enum wc_map_op map_op;
1354	struct wc_entry *e;
1355
1356	read_next_block:
1357	wc->stats.reads++;
1358	e = writecache_find_entry(wc, bio->bi_iter.bi_sector, WFE_RETURN_FOLLOWING);
1359	if (e && read_original_sector(wc, e) == bio->bi_iter.bi_sector) {
1360	wc->stats.read_hits++;
1361	if (WC_MODE_PMEM(wc)) {
1362	bio_copy_block(wc, bio, memory_data(wc, e));
1363	if (bio->bi_iter.bi_size)
1364	goto read_next_block;
1365	map_op = WC_MAP_SUBMIT;
1366	} else {
1367	dm_accept_partial_bio(bio, wc->block_size >> SECTOR_SHIFT);
1368	bio_set_dev(bio, wc->ssd_dev->bdev);
1369	bio->bi_iter.bi_sector = cache_sector(wc, e);
1370	if (!writecache_entry_is_committed(wc, e))
1371	writecache_wait_for_ios(wc, WRITE);
1372	map_op = WC_MAP_REMAP;
1373	}
1374	} else {
1375	writecache_map_remap_origin(wc, bio, e);
1376	wc->stats.reads += (bio->bi_iter.bi_size - wc->block_size) >> wc->block_size_bits;
1377	map_op = WC_MAP_REMAP_ORIGIN;
1378	}
1379
1380	return map_op;
1381	}
1382
1383	static void writecache_bio_copy_ssd(struct dm_writecache wc, struct bio bio,
1384	struct wc_entry *e, bool search_used)
1385	{
1386	unsigned int bio_size = wc->block_size;
1387	sector_t start_cache_sec = cache_sector(wc, e);
1388	sector_t current_cache_sec = start_cache_sec + (bio_size >> SECTOR_SHIFT);
1389
1390	while (bio_size < bio->bi_iter.bi_size) {
1391	if (!search_used) {
1392	struct wc_entry *f = writecache_pop_from_freelist(wc, current_cache_sec);
1393
1394	if (!f)
1395	break;
1396	write_original_sector_seq_count(wc, f, bio->bi_iter.bi_sector +
1397	(bio_size >> SECTOR_SHIFT), wc->seq_count);
1398	writecache_insert_entry(wc, f);
1399	wc->uncommitted_blocks++;
1400	} else {
1401	struct wc_entry *f;
1402	struct rb_node *next = rb_next(&e->rb_node);
1403
1404	if (!next)
1405	break;
1406	f = container_of(next, struct wc_entry, rb_node);
1407	if (f != e + 1)
1408	break;
1409	if (read_original_sector(wc, f) !=
1410	read_original_sector(wc, e) + (wc->block_size >> SECTOR_SHIFT))
1411	break;
1412	if (unlikely(f->write_in_progress))
1413	break;
1414	if (writecache_entry_is_committed(wc, f))
1415	wc->overwrote_committed = true;
1416	e = f;
1417	}
1418	bio_size += wc->block_size;
1419	current_cache_sec += wc->block_size >> SECTOR_SHIFT;
1420	}
1421
1422	bio_set_dev(bio, wc->ssd_dev->bdev);
1423	bio->bi_iter.bi_sector = start_cache_sec;
1424	dm_accept_partial_bio(bio, bio_size >> SECTOR_SHIFT);
1425
1426	wc->stats.writes += bio->bi_iter.bi_size >> wc->block_size_bits;
1427	wc->stats.writes_allocate += (bio->bi_iter.bi_size - wc->block_size) >> wc->block_size_bits;
1428
1429	if (unlikely(wc->uncommitted_blocks >= wc->autocommit_blocks)) {
1430	wc->uncommitted_blocks = 0;
1431	queue_work(wc->writeback_wq, &wc->flush_work);
1432	} else {
1433	writecache_schedule_autocommit(wc);
1434	}
1435	}
1436
1437	static enum wc_map_op writecache_map_write(struct dm_writecache wc, struct bio bio)
1438	{
1439	struct wc_entry *e;
1440
1441	do {
1442	bool found_entry = false;
1443	bool search_used = false;
1444
1445	if (writecache_has_error(wc)) {
1446	wc->stats.writes += bio->bi_iter.bi_size >> wc->block_size_bits;
1447	return WC_MAP_ERROR;
1448	}
1449	e = writecache_find_entry(wc, bio->bi_iter.bi_sector, 0);
1450	if (e) {
1451	if (!writecache_entry_is_committed(wc, e)) {
1452	wc->stats.write_hits_uncommitted++;
1453	search_used = true;
1454	goto bio_copy;
1455	}
1456	wc->stats.write_hits_committed++;
1457	if (!WC_MODE_PMEM(wc) && !e->write_in_progress) {
1458	wc->overwrote_committed = true;
1459	search_used = true;
1460	goto bio_copy;
1461	}
1462	found_entry = true;
1463	} else {
1464	if (unlikely(wc->cleaner) \|\|
1465	(wc->metadata_only && !(bio->bi_opf & REQ_META)))
1466	goto direct_write;
1467	}
1468	e = writecache_pop_from_freelist(wc, (sector_t)-1);
1469	if (unlikely(!e)) {
1470	if (!WC_MODE_PMEM(wc) && !found_entry) {
1471	direct_write:
1472	e = writecache_find_entry(wc, bio->bi_iter.bi_sector, WFE_RETURN_FOLLOWING);
1473	writecache_map_remap_origin(wc, bio, e);
1474	wc->stats.writes_around += bio->bi_iter.bi_size >> wc->block_size_bits;
1475	wc->stats.writes += bio->bi_iter.bi_size >> wc->block_size_bits;
1476	return WC_MAP_REMAP_ORIGIN;
1477	}
1478	wc->stats.writes_blocked_on_freelist++;
1479	writecache_wait_on_freelist(wc);
1480	continue;
1481	}
1482	write_original_sector_seq_count(wc, e, bio->bi_iter.bi_sector, wc->seq_count);
1483	writecache_insert_entry(wc, e);
1484	wc->uncommitted_blocks++;
1485	wc->stats.writes_allocate++;
1486	bio_copy:
1487	if (WC_MODE_PMEM(wc)) {
1488	bio_copy_block(wc, bio, memory_data(wc, e));
1489	wc->stats.writes++;
1490	} else {
1491	writecache_bio_copy_ssd(wc, bio, e, search_used);
1492	return WC_MAP_REMAP;
1493	}
1494	} while (bio->bi_iter.bi_size);
1495
1496	if (unlikely(bio->bi_opf & REQ_FUA \|\| wc->uncommitted_blocks >= wc->autocommit_blocks))
1497	writecache_flush(wc);
1498	else
1499	writecache_schedule_autocommit(wc);
1500
1501	return WC_MAP_SUBMIT;
1502	}
1503
1504	static enum wc_map_op writecache_map_flush(struct dm_writecache wc, struct bio bio)
1505	{
1506	if (writecache_has_error(wc))
1507	return WC_MAP_ERROR;
1508
1509	if (WC_MODE_PMEM(wc)) {
1510	wc->stats.flushes++;
1511	writecache_flush(wc);
1512	if (writecache_has_error(wc))
1513	return WC_MAP_ERROR;
1514	else if (unlikely(wc->cleaner) \|\| unlikely(wc->metadata_only))
1515	return WC_MAP_REMAP_ORIGIN;
1516	return WC_MAP_SUBMIT;
1517	}
1518	/* SSD: */
1519	if (dm_bio_get_target_bio_nr(bio))
1520	return WC_MAP_REMAP_ORIGIN;
1521	wc->stats.flushes++;
1522	writecache_offload_bio(wc, bio);
1523	return WC_MAP_RETURN;
1524	}
1525
1526	static enum wc_map_op writecache_map_discard(struct dm_writecache wc, struct bio bio)
1527	{
1528	wc->stats.discards += bio->bi_iter.bi_size >> wc->block_size_bits;
1529
1530	if (writecache_has_error(wc))
1531	return WC_MAP_ERROR;
1532
1533	if (WC_MODE_PMEM(wc)) {
1534	writecache_discard(wc, bio->bi_iter.bi_sector, bio_end_sector(bio));
1535	return WC_MAP_REMAP_ORIGIN;
1536	}
1537	/* SSD: */
1538	writecache_offload_bio(wc, bio);
1539	return WC_MAP_RETURN;
1540	}
1541
1542	static int writecache_map(struct dm_target ti, struct bio bio)
1543	{
1544	struct dm_writecache *wc = ti->private;
1545	enum wc_map_op map_op;
1546
1547	bio->bi_private = NULL;
1548
1549	wc_lock(wc);
1550
1551	if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1552	map_op = writecache_map_flush(wc, bio);
1553	goto done;
1554	}
1555
1556	bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1557
1558	if (unlikely((((unsigned int)bio->bi_iter.bi_sector \| bio_sectors(bio)) &
1559	(wc->block_size / 512 - 1)) != 0)) {
1560	DMERR("I/O is not aligned, sector %llu, size %u, block size %u",
1561	(unsigned long long)bio->bi_iter.bi_sector,
1562	bio->bi_iter.bi_size, wc->block_size);
1563	map_op = WC_MAP_ERROR;
1564	goto done;
1565	}
1566
1567	if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) {
1568	map_op = writecache_map_discard(wc, bio);
1569	goto done;
1570	}
1571
1572	if (bio_data_dir(bio) == READ)
1573	map_op = writecache_map_read(wc, bio);
1574	else
1575	map_op = writecache_map_write(wc, bio);
1576	done:
1577	switch (map_op) {
1578	case WC_MAP_REMAP_ORIGIN:
1579	if (likely(wc->pause != 0)) {
1580	if (bio_op(bio) == REQ_OP_WRITE) {
1581	dm_iot_io_begin(&wc->iot, 1);
1582	bio->bi_private = (void *)2;
1583	}
1584	}
1585	bio_set_dev(bio, wc->dev->bdev);
1586	wc_unlock(wc);
1587	return DM_MAPIO_REMAPPED;
1588
1589	case WC_MAP_REMAP:
1590	/* make sure that writecache_end_io decrements bio_in_progress: */
1591	bio->bi_private = (void *)1;
1592	atomic_inc(&wc->bio_in_progress[bio_data_dir(bio)]);
1593	wc_unlock(wc);
1594	return DM_MAPIO_REMAPPED;
1595
1596	case WC_MAP_SUBMIT:
1597	wc_unlock(wc);
1598	bio_endio(bio);
1599	return DM_MAPIO_SUBMITTED;
1600
1601	case WC_MAP_RETURN:
1602	wc_unlock(wc);
1603	return DM_MAPIO_SUBMITTED;
1604
1605	case WC_MAP_ERROR:
1606	wc_unlock(wc);
1607	bio_io_error(bio);
1608	return DM_MAPIO_SUBMITTED;
1609
1610	default:
1611	BUG();
1612	wc_unlock(wc);
1613	return DM_MAPIO_KILL;
1614	}
1615	}
1616
1617	static int writecache_end_io(struct dm_target ti, struct bio bio, blk_status_t *status)
1618	{
1619	struct dm_writecache *wc = ti->private;
1620
1621	if (bio->bi_private == (void *)1) {
1622	int dir = bio_data_dir(bio);
1623
1624	if (atomic_dec_and_test(&wc->bio_in_progress[dir]))
1625	if (unlikely(waitqueue_active(&wc->bio_in_progress_wait[dir])))
1626	wake_up(&wc->bio_in_progress_wait[dir]);
1627	} else if (bio->bi_private == (void *)2) {
1628	dm_iot_io_end(&wc->iot, 1);
1629	}
1630	return 0;
1631	}
1632
1633	static int writecache_iterate_devices(struct dm_target *ti,
1634	iterate_devices_callout_fn fn, void *data)
1635	{
1636	struct dm_writecache *wc = ti->private;
1637
1638	return fn(ti, wc->dev, 0, ti->len, data);
1639	}
1640
1641	static void writecache_io_hints(struct dm_target ti, struct queue_limits limits)
1642	{
1643	struct dm_writecache *wc = ti->private;
1644
1645	dm_stack_bs_limits(limits, wc->block_size);
1646	}
1647
1648	static void writecache_writeback_endio(struct bio *bio)
1649	{
1650	struct writeback_struct *wb = container_of(bio, struct writeback_struct, bio);
1651	struct dm_writecache *wc = wb->wc;
1652	unsigned long flags;
1653
1654	raw_spin_lock_irqsave(&wc->endio_list_lock, flags);
1655	if (unlikely(list_empty(&wc->endio_list)))
1656	wake_up_process(wc->endio_thread);
1657	list_add_tail(&wb->endio_entry, &wc->endio_list);
1658	raw_spin_unlock_irqrestore(&wc->endio_list_lock, flags);
1659	}
1660
1661	static void writecache_copy_endio(int read_err, unsigned long write_err, void *ptr)
1662	{
1663	struct copy_struct *c = ptr;
1664	struct dm_writecache *wc = c->wc;
1665
1666	c->error = likely(!(read_err \| write_err)) ? 0 : -EIO;
1667
1668	raw_spin_lock_irq(&wc->endio_list_lock);
1669	if (unlikely(list_empty(&wc->endio_list)))
1670	wake_up_process(wc->endio_thread);
1671	list_add_tail(&c->endio_entry, &wc->endio_list);
1672	raw_spin_unlock_irq(&wc->endio_list_lock);
1673	}
1674
1675	static void __writecache_endio_pmem(struct dm_writecache wc, struct list_head list)
1676	{
1677	unsigned int i;
1678	struct writeback_struct *wb;
1679	struct wc_entry *e;
1680	unsigned long n_walked = 0;
1681
1682	do {
1683	wb = list_entry(list->next, struct writeback_struct, endio_entry);
1684	list_del(&wb->endio_entry);
1685
1686	if (unlikely(wb->bio.bi_status != BLK_STS_OK))
1687	writecache_error(wc, blk_status_to_errno(wb->bio.bi_status),
1688	"write error %d", wb->bio.bi_status);
1689	i = 0;
1690	do {
1691	e = wb->wc_list[i];
1692	BUG_ON(!e->write_in_progress);
1693	e->write_in_progress = false;
1694	INIT_LIST_HEAD(&e->lru);
1695	if (!writecache_has_error(wc))
1696	writecache_free_entry(wc, e);
1697	BUG_ON(!wc->writeback_size);
1698	wc->writeback_size--;
1699	n_walked++;
1700	if (unlikely(n_walked >= ENDIO_LATENCY)) {
1701	writecache_commit_flushed(wc, false);
1702	wc_unlock(wc);
1703	wc_lock(wc);
1704	n_walked = 0;
1705	}
1706	} while (++i < wb->wc_list_n);
1707
1708	if (wb->wc_list != wb->wc_list_inline)
1709	kfree(wb->wc_list);
1710	bio_put(&wb->bio);
1711	} while (!list_empty(list));
1712	}
1713
1714	static void __writecache_endio_ssd(struct dm_writecache wc, struct list_head list)
1715	{
1716	struct copy_struct *c;
1717	struct wc_entry *e;
1718
1719	do {
1720	c = list_entry(list->next, struct copy_struct, endio_entry);
1721	list_del(&c->endio_entry);
1722
1723	if (unlikely(c->error))
1724	writecache_error(wc, c->error, "copy error");
1725
1726	e = c->e;
1727	do {
1728	BUG_ON(!e->write_in_progress);
1729	e->write_in_progress = false;
1730	INIT_LIST_HEAD(&e->lru);
1731	if (!writecache_has_error(wc))
1732	writecache_free_entry(wc, e);
1733
1734	BUG_ON(!wc->writeback_size);
1735	wc->writeback_size--;
1736	e++;
1737	} while (--c->n_entries);
1738	mempool_free(c, &wc->copy_pool);
1739	} while (!list_empty(list));
1740	}
1741
1742	static int writecache_endio_thread(void *data)
1743	{
1744	struct dm_writecache *wc = data;
1745
1746	while (1) {
1747	struct list_head list;
1748
1749	raw_spin_lock_irq(&wc->endio_list_lock);
1750	if (!list_empty(&wc->endio_list))
1751	goto pop_from_list;
1752	set_current_state(TASK_INTERRUPTIBLE);
1753	raw_spin_unlock_irq(&wc->endio_list_lock);
1754
1755	if (unlikely(kthread_should_stop())) {
1756	set_current_state(TASK_RUNNING);
1757	break;
1758	}
1759
1760	schedule();
1761
1762	continue;
1763
1764	pop_from_list:
1765	list = wc->endio_list;
1766	list.next->prev = list.prev->next = &list;
1767	INIT_LIST_HEAD(&wc->endio_list);
1768	raw_spin_unlock_irq(&wc->endio_list_lock);
1769
1770	if (!WC_MODE_FUA(wc))
1771	writecache_disk_flush(wc, wc->dev);
1772
1773	wc_lock(wc);
1774
1775	if (WC_MODE_PMEM(wc)) {
1776	__writecache_endio_pmem(wc, &list);
1777	} else {
1778	__writecache_endio_ssd(wc, &list);
1779	writecache_wait_for_ios(wc, READ);
1780	}
1781
1782	writecache_commit_flushed(wc, false);
1783
1784	wc_unlock(wc);
1785	}
1786
1787	return 0;
1788	}
1789
1790	static bool wc_add_block(struct writeback_struct wb, struct wc_entry e)
1791	{
1792	struct dm_writecache *wc = wb->wc;
1793	unsigned int block_size = wc->block_size;
1794	void *address = memory_data(wc, e);
1795
1796	persistent_memory_flush_cache(address, block_size);
1797
1798	if (unlikely(bio_end_sector(&wb->bio) >= wc->data_device_sectors))
1799	return true;
1800
1801	return bio_add_page(&wb->bio, persistent_memory_page(address),
1802	block_size, persistent_memory_page_offset(address)) != 0;
1803	}
1804
1805	struct writeback_list {
1806	struct list_head list;
1807	size_t size;
1808	};
1809
1810	static void __writeback_throttle(struct dm_writecache wc, struct writeback_list wbl)
1811	{
1812	if (unlikely(wc->max_writeback_jobs)) {
1813	if (READ_ONCE(wc->writeback_size) - wbl->size >= wc->max_writeback_jobs) {
1814	wc_lock(wc);
1815	while (wc->writeback_size - wbl->size >= wc->max_writeback_jobs)
1816	writecache_wait_on_freelist(wc);
1817	wc_unlock(wc);
1818	}
1819	}
1820	cond_resched();
1821	}
1822
1823	static void __writecache_writeback_pmem(struct dm_writecache wc, struct writeback_list wbl)
1824	{
1825	struct wc_entry e, f;
1826	struct bio *bio;
1827	struct writeback_struct *wb;
1828	unsigned int max_pages;
1829
1830	while (wbl->size) {
1831	wbl->size--;
1832	e = container_of(wbl->list.prev, struct wc_entry, lru);
1833	list_del(&e->lru);
1834
1835	max_pages = e->wc_list_contiguous;
1836
1837	bio = bio_alloc_bioset(wc->dev->bdev, max_pages, REQ_OP_WRITE,
1838	GFP_NOIO, &wc->bio_set);
1839	wb = container_of(bio, struct writeback_struct, bio);
1840	wb->wc = wc;
1841	bio->bi_end_io = writecache_writeback_endio;
1842	bio->bi_iter.bi_sector = read_original_sector(wc, e);
1843
1844	if (unlikely(max_pages > WB_LIST_INLINE))
1845	wb->wc_list = kmalloc_objs(struct wc_entry *, max_pages,
1846	GFP_NOIO \| __GFP_NORETRY \| __GFP_NOMEMALLOC \| __GFP_NOWARN);
1847
1848	if (likely(max_pages <= WB_LIST_INLINE) \|\| unlikely(!wb->wc_list)) {
1849	wb->wc_list = wb->wc_list_inline;
1850	max_pages = WB_LIST_INLINE;
1851	}
1852
1853	BUG_ON(!wc_add_block(wb, e));
1854
1855	wb->wc_list[0] = e;
1856	wb->wc_list_n = 1;
1857
1858	while (wbl->size && wb->wc_list_n < max_pages) {
1859	f = container_of(wbl->list.prev, struct wc_entry, lru);
1860	if (read_original_sector(wc, f) !=
1861	read_original_sector(wc, e) + (wc->block_size >> SECTOR_SHIFT))
1862	break;
1863	if (!wc_add_block(wb, f))
1864	break;
1865	wbl->size--;
1866	list_del(&f->lru);
1867	wb->wc_list[wb->wc_list_n++] = f;
1868	e = f;
1869	}
1870	if (WC_MODE_FUA(wc))
1871	bio->bi_opf \|= REQ_FUA;
1872	if (writecache_has_error(wc)) {
1873	bio->bi_status = BLK_STS_IOERR;
1874	bio_endio(bio);
1875	} else if (unlikely(!bio_sectors(bio))) {
1876	bio->bi_status = BLK_STS_OK;
1877	bio_endio(bio);
1878	} else {
1879	submit_bio(bio);
1880	}
1881
1882	__writeback_throttle(wc, wbl);
1883	}
1884	}
1885
1886	static void __writecache_writeback_ssd(struct dm_writecache wc, struct writeback_list wbl)
1887	{
1888	struct wc_entry e, f;
1889	struct dm_io_region from, to;
1890	struct copy_struct *c;
1891
1892	while (wbl->size) {
1893	unsigned int n_sectors;
1894
1895	wbl->size--;
1896	e = container_of(wbl->list.prev, struct wc_entry, lru);
1897	list_del(&e->lru);
1898
1899	n_sectors = e->wc_list_contiguous << (wc->block_size_bits - SECTOR_SHIFT);
1900
1901	from.bdev = wc->ssd_dev->bdev;
1902	from.sector = cache_sector(wc, e);
1903	from.count = n_sectors;
1904	to.bdev = wc->dev->bdev;
1905	to.sector = read_original_sector(wc, e);
1906	to.count = n_sectors;
1907
1908	c = mempool_alloc(&wc->copy_pool, GFP_NOIO);
1909	c->wc = wc;
1910	c->e = e;
1911	c->n_entries = e->wc_list_contiguous;
1912
1913	while ((n_sectors -= wc->block_size >> SECTOR_SHIFT)) {
1914	wbl->size--;
1915	f = container_of(wbl->list.prev, struct wc_entry, lru);
1916	BUG_ON(f != e + 1);
1917	list_del(&f->lru);
1918	e = f;
1919	}
1920
1921	if (unlikely(to.sector + to.count > wc->data_device_sectors)) {
1922	if (to.sector >= wc->data_device_sectors) {
1923	writecache_copy_endio(0, 0, c);
1924	continue;
1925	}
1926	from.count = to.count = wc->data_device_sectors - to.sector;
1927	}
1928
1929	dm_kcopyd_copy(wc->dm_kcopyd, &from, 1, &to, 0, writecache_copy_endio, c);
1930
1931	__writeback_throttle(wc, wbl);
1932	}
1933	}
1934
1935	static void writecache_writeback(struct work_struct *work)
1936	{
1937	struct dm_writecache *wc = container_of(work, struct dm_writecache, writeback_work);
1938	struct blk_plug plug;
1939	struct wc_entry f, g, *e = NULL;
1940	struct rb_node node, next_node;
1941	struct list_head skipped;
1942	struct writeback_list wbl;
1943	unsigned long n_walked;
1944
1945	if (!WC_MODE_PMEM(wc)) {
1946	/* Wait for any active kcopyd work on behalf of ssd writeback */
1947	dm_kcopyd_client_flush(wc->dm_kcopyd);
1948	}
1949
1950	if (likely(wc->pause != 0)) {
1951	while (1) {
1952	unsigned long idle;
1953
1954	if (unlikely(wc->cleaner) \|\| unlikely(wc->writeback_all) \|\|
1955	unlikely(dm_suspended(wc->ti)))
1956	break;
1957	idle = dm_iot_idle_time(&wc->iot);
1958	if (idle >= wc->pause)
1959	break;
1960	idle = wc->pause - idle;
1961	if (idle > HZ)
1962	idle = HZ;
1963	schedule_timeout_idle(idle);
1964	}
1965	}
1966
1967	wc_lock(wc);
1968	restart:
1969	if (writecache_has_error(wc)) {
1970	wc_unlock(wc);
1971	return;
1972	}
1973
1974	if (unlikely(wc->writeback_all)) {
1975	if (writecache_wait_for_writeback(wc))
1976	goto restart;
1977	}
1978
1979	if (wc->overwrote_committed)
1980	writecache_wait_for_ios(wc, WRITE);
1981
1982	n_walked = 0;
1983	INIT_LIST_HEAD(&skipped);
1984	INIT_LIST_HEAD(&wbl.list);
1985	wbl.size = 0;
1986	while (!list_empty(&wc->lru) &&
1987	(wc->writeback_all \|\|
1988	wc->freelist_size + wc->writeback_size <= wc->freelist_low_watermark \|\|
1989	(jiffies - container_of(wc->lru.prev, struct wc_entry, lru)->age >=
1990	wc->max_age - wc->max_age / MAX_AGE_DIV))) {
1991
1992	n_walked++;
1993	if (unlikely(n_walked > WRITEBACK_LATENCY) &&
1994	likely(!wc->writeback_all)) {
1995	if (likely(!dm_suspended(wc->ti)))
1996	queue_work(wc->writeback_wq, &wc->writeback_work);
1997	break;
1998	}
1999
2000	if (unlikely(wc->writeback_all)) {
2001	if (unlikely(!e)) {
2002	writecache_flush(wc);
2003	e = container_of(rb_first(&wc->tree), struct wc_entry, rb_node);
2004	} else
2005	e = g;
2006	} else
2007	e = container_of(wc->lru.prev, struct wc_entry, lru);
2008	BUG_ON(e->write_in_progress);
2009	if (unlikely(!writecache_entry_is_committed(wc, e)))
2010	writecache_flush(wc);
2011
2012	node = rb_prev(&e->rb_node);
2013	if (node) {
2014	f = container_of(node, struct wc_entry, rb_node);
2015	if (unlikely(read_original_sector(wc, f) ==
2016	read_original_sector(wc, e))) {
2017	BUG_ON(!f->write_in_progress);
2018	list_move(&e->lru, &skipped);
2019	cond_resched();
2020	continue;
2021	}
2022	}
2023	wc->writeback_size++;
2024	list_move(&e->lru, &wbl.list);
2025	wbl.size++;
2026	e->write_in_progress = true;
2027	e->wc_list_contiguous = 1;
2028
2029	f = e;
2030
2031	while (1) {
2032	next_node = rb_next(&f->rb_node);
2033	if (unlikely(!next_node))
2034	break;
2035	g = container_of(next_node, struct wc_entry, rb_node);
2036	if (unlikely(read_original_sector(wc, g) ==
2037	read_original_sector(wc, f))) {
2038	f = g;
2039	continue;
2040	}
2041	if (read_original_sector(wc, g) !=
2042	read_original_sector(wc, f) + (wc->block_size >> SECTOR_SHIFT))
2043	break;
2044	if (unlikely(g->write_in_progress))
2045	break;
2046	if (unlikely(!writecache_entry_is_committed(wc, g)))
2047	break;
2048
2049	if (!WC_MODE_PMEM(wc)) {
2050	if (g != f + 1)
2051	break;
2052	}
2053
2054	n_walked++;
2055	//if (unlikely(n_walked > WRITEBACK_LATENCY) && likely(!wc->writeback_all))
2056	// break;
2057
2058	wc->writeback_size++;
2059	list_move(&g->lru, &wbl.list);
2060	wbl.size++;
2061	g->write_in_progress = true;
2062	g->wc_list_contiguous = BIO_MAX_VECS;
2063	f = g;
2064	e->wc_list_contiguous++;
2065	if (unlikely(e->wc_list_contiguous == BIO_MAX_VECS)) {
2066	if (unlikely(wc->writeback_all)) {
2067	next_node = rb_next(&f->rb_node);
2068	if (likely(next_node))
2069	g = container_of(next_node, struct wc_entry, rb_node);
2070	}
2071	break;
2072	}
2073	}
2074	cond_resched();
2075	}
2076
2077	if (!list_empty(&skipped)) {
2078	list_splice_tail(&skipped, &wc->lru);
2079	/*
2080	* If we didn't do any progress, we must wait until some
2081	* writeback finishes to avoid burning CPU in a loop
2082	*/
2083	if (unlikely(!wbl.size))
2084	writecache_wait_for_writeback(wc);
2085	}
2086
2087	wc_unlock(wc);
2088
2089	blk_start_plug(&plug);
2090
2091	if (WC_MODE_PMEM(wc))
2092	__writecache_writeback_pmem(wc, &wbl);
2093	else
2094	__writecache_writeback_ssd(wc, &wbl);
2095
2096	blk_finish_plug(&plug);
2097
2098	if (unlikely(wc->writeback_all)) {
2099	wc_lock(wc);
2100	while (writecache_wait_for_writeback(wc))
2101	;
2102	wc_unlock(wc);
2103	}
2104	}
2105
2106	static int calculate_memory_size(uint64_t device_size, unsigned int block_size,
2107	size_t n_blocks_p, size_t n_metadata_blocks_p)
2108	{
2109	uint64_t n_blocks, offset;
2110	struct wc_entry e;
2111
2112	n_blocks = device_size;
2113	do_div(n_blocks, block_size + sizeof(struct wc_memory_entry));
2114
2115	while (1) {
2116	if (!n_blocks)
2117	return -ENOSPC;
2118	/* Verify the following entries[n_blocks] won't overflow */
2119	if (n_blocks >= ((size_t)-sizeof(struct wc_memory_superblock) /
2120	sizeof(struct wc_memory_entry)))
2121	return -EFBIG;
2122	offset = offsetof(struct wc_memory_superblock, entries[n_blocks]);
2123	offset = (offset + block_size - 1) & ~(uint64_t)(block_size - 1);
2124	if (offset + n_blocks * block_size <= device_size)
2125	break;
2126	n_blocks--;
2127	}
2128
2129	/* check if the bit field overflows */
2130	e.index = n_blocks;
2131	if (e.index != n_blocks)
2132	return -EFBIG;
2133
2134	if (n_blocks_p)
2135	*n_blocks_p = n_blocks;
2136	if (n_metadata_blocks_p)
2137	*n_metadata_blocks_p = offset >> __ffs(block_size);
2138	return 0;
2139	}
2140
2141	static int init_memory(struct dm_writecache *wc)
2142	{
2143	size_t b;
2144	int r;
2145
2146	r = calculate_memory_size(wc->memory_map_size, wc->block_size, &wc->n_blocks, NULL);
2147	if (r)
2148	return r;
2149
2150	r = writecache_alloc_entries(wc);
2151	if (r)
2152	return r;
2153
2154	for (b = 0; b < ARRAY_SIZE(sb(wc)->padding); b++)
2155	pmem_assign(sb(wc)->padding[b], cpu_to_le64(0));
2156	pmem_assign(sb(wc)->version, cpu_to_le32(MEMORY_SUPERBLOCK_VERSION));
2157	pmem_assign(sb(wc)->block_size, cpu_to_le32(wc->block_size));
2158	pmem_assign(sb(wc)->n_blocks, cpu_to_le64(wc->n_blocks));
2159	pmem_assign(sb(wc)->seq_count, cpu_to_le64(0));
2160
2161	for (b = 0; b < wc->n_blocks; b++) {
2162	write_original_sector_seq_count(wc, &wc->entries[b], -1, -1);
2163	cond_resched();
2164	}
2165
2166	writecache_flush_all_metadata(wc);
2167	writecache_commit_flushed(wc, false);
2168	pmem_assign(sb(wc)->magic, cpu_to_le32(MEMORY_SUPERBLOCK_MAGIC));
2169	writecache_flush_region(wc, &sb(wc)->magic, sizeof(sb(wc)->magic));
2170	writecache_commit_flushed(wc, false);
2171
2172	return 0;
2173	}
2174
2175	static void writecache_dtr(struct dm_target *ti)
2176	{
2177	struct dm_writecache *wc = ti->private;
2178
2179	if (!wc)
2180	return;
2181
2182	if (wc->endio_thread)
2183	kthread_stop(wc->endio_thread);
2184
2185	if (wc->flush_thread)
2186	kthread_stop(wc->flush_thread);
2187
2188	bioset_exit(&wc->bio_set);
2189
2190	mempool_exit(&wc->copy_pool);
2191
2192	if (wc->writeback_wq)
2193	destroy_workqueue(wc->writeback_wq);
2194
2195	if (wc->dev)
2196	dm_put_device(ti, wc->dev);
2197
2198	if (wc->ssd_dev)
2199	dm_put_device(ti, wc->ssd_dev);
2200
2201	vfree(wc->entries);
2202
2203	if (wc->memory_map) {
2204	if (WC_MODE_PMEM(wc))
2205	persistent_memory_release(wc);
2206	else
2207	vfree(wc->memory_map);
2208	}
2209
2210	if (wc->dm_kcopyd)
2211	dm_kcopyd_client_destroy(wc->dm_kcopyd);
2212
2213	if (wc->dm_io)
2214	dm_io_client_destroy(wc->dm_io);
2215
2216	vfree(wc->dirty_bitmap);
2217
2218	kfree(wc);
2219	}
2220
2221	static int writecache_ctr(struct dm_target ti, unsigned int argc, char *argv)
2222	{
2223	struct dm_writecache *wc;
2224	struct dm_arg_set as;
2225	const char *string;
2226	unsigned int opt_params;
2227	size_t offset, data_size;
2228	int i, r;
2229	char dummy;
2230	int high_wm_percent = HIGH_WATERMARK;
2231	int low_wm_percent = LOW_WATERMARK;
2232	uint64_t x;
2233	struct wc_memory_superblock s;
2234
2235	static struct dm_arg _args[] = {
2236	{0, 18, "Invalid number of feature args"},
2237	};
2238
2239	as.argc = argc;
2240	as.argv = argv;
2241
2242	wc = kzalloc_obj(struct dm_writecache);
2243	if (!wc) {
2244	ti->error = "Cannot allocate writecache structure";
2245	r = -ENOMEM;
2246	goto bad;
2247	}
2248	ti->private = wc;
2249	wc->ti = ti;
2250
2251	mutex_init(&wc->lock);
2252	wc->max_age = MAX_AGE_UNSPECIFIED;
2253	writecache_poison_lists(wc);
2254	init_waitqueue_head(&wc->freelist_wait);
2255	timer_setup(&wc->autocommit_timer, writecache_autocommit_timer, 0);
2256	timer_setup(&wc->max_age_timer, writecache_max_age_timer, 0);
2257
2258	for (i = 0; i < 2; i++) {
2259	atomic_set(&wc->bio_in_progress[i], 0);
2260	init_waitqueue_head(&wc->bio_in_progress_wait[i]);
2261	}
2262
2263	wc->dm_io = dm_io_client_create();
2264	if (IS_ERR(wc->dm_io)) {
2265	r = PTR_ERR(wc->dm_io);
2266	ti->error = "Unable to allocate dm-io client";
2267	wc->dm_io = NULL;
2268	goto bad;
2269	}
2270
2271	wc->writeback_wq = alloc_workqueue("writecache-writeback",
2272	WQ_MEM_RECLAIM \| WQ_PERCPU, 1);
2273	if (!wc->writeback_wq) {
2274	r = -ENOMEM;
2275	ti->error = "Could not allocate writeback workqueue";
2276	goto bad;
2277	}
2278	INIT_WORK(&wc->writeback_work, writecache_writeback);
2279	INIT_WORK(&wc->flush_work, writecache_flush_work);
2280
2281	dm_iot_init(&wc->iot);
2282
2283	raw_spin_lock_init(&wc->endio_list_lock);
2284	INIT_LIST_HEAD(&wc->endio_list);
2285	wc->endio_thread = kthread_run(writecache_endio_thread, wc, "writecache_endio");
2286	if (IS_ERR(wc->endio_thread)) {
2287	r = PTR_ERR(wc->endio_thread);
2288	wc->endio_thread = NULL;
2289	ti->error = "Couldn't spawn endio thread";
2290	goto bad;
2291	}
2292
2293	/*
2294	* Parse the mode (pmem or ssd)
2295	*/
2296	string = dm_shift_arg(&as);
2297	if (!string)
2298	goto bad_arguments;
2299
2300	if (!strcasecmp(string, "s")) {
2301	wc->pmem_mode = false;
2302	} else if (!strcasecmp(string, "p")) {
2303	#ifdef DM_WRITECACHE_HAS_PMEM
2304	wc->pmem_mode = true;
2305	wc->writeback_fua = true;
2306	#else
2307	/*
2308	* If the architecture doesn't support persistent memory or
2309	* the kernel doesn't support any DAX drivers, this driver can
2310	* only be used in SSD-only mode.
2311	*/
2312	r = -EOPNOTSUPP;
2313	ti->error = "Persistent memory or DAX not supported on this system";
2314	goto bad;
2315	#endif
2316	} else {
2317	goto bad_arguments;
2318	}
2319
2320	if (WC_MODE_PMEM(wc)) {
2321	r = bioset_init(&wc->bio_set, BIO_POOL_SIZE,
2322	offsetof(struct writeback_struct, bio),
2323	BIOSET_NEED_BVECS);
2324	if (r) {
2325	ti->error = "Could not allocate bio set";
2326	goto bad;
2327	}
2328	} else {
2329	wc->pause = PAUSE_WRITEBACK;
2330	r = mempool_init_kmalloc_pool(&wc->copy_pool, 1, sizeof(struct copy_struct));
2331	if (r) {
2332	ti->error = "Could not allocate mempool";
2333	goto bad;
2334	}
2335	}
2336
2337	/*
2338	* Parse the origin data device
2339	*/
2340	string = dm_shift_arg(&as);
2341	if (!string)
2342	goto bad_arguments;
2343	r = dm_get_device(ti, string, dm_table_get_mode(ti->table), &wc->dev);
2344	if (r) {
2345	ti->error = "Origin data device lookup failed";
2346	goto bad;
2347	}
2348
2349	/*
2350	* Parse cache data device (be it pmem or ssd)
2351	*/
2352	string = dm_shift_arg(&as);
2353	if (!string)
2354	goto bad_arguments;
2355
2356	r = dm_get_device(ti, string, dm_table_get_mode(ti->table), &wc->ssd_dev);
2357	if (r) {
2358	ti->error = "Cache data device lookup failed";
2359	goto bad;
2360	}
2361	wc->memory_map_size = bdev_nr_bytes(wc->ssd_dev->bdev);
2362
2363	/*
2364	* Parse the cache block size
2365	*/
2366	string = dm_shift_arg(&as);
2367	if (!string)
2368	goto bad_arguments;
2369	if (sscanf(string, "%u%c", &wc->block_size, &dummy) != 1 \|\|
2370	wc->block_size < 512 \|\| wc->block_size > PAGE_SIZE \|\|
2371	(wc->block_size & (wc->block_size - 1))) {
2372	r = -EINVAL;
2373	ti->error = "Invalid block size";
2374	goto bad;
2375	}
2376	if (wc->block_size < bdev_logical_block_size(wc->dev->bdev) \|\|
2377	wc->block_size < bdev_logical_block_size(wc->ssd_dev->bdev)) {
2378	r = -EINVAL;
2379	ti->error = "Block size is smaller than device logical block size";
2380	goto bad;
2381	}
2382	wc->block_size_bits = __ffs(wc->block_size);
2383
2384	wc->max_writeback_jobs = MAX_WRITEBACK_JOBS;
2385	wc->autocommit_blocks = !WC_MODE_PMEM(wc) ? AUTOCOMMIT_BLOCKS_SSD : AUTOCOMMIT_BLOCKS_PMEM;
2386	wc->autocommit_jiffies = msecs_to_jiffies(AUTOCOMMIT_MSEC);
2387
2388	/*
2389	* Parse optional arguments
2390	*/
2391	r = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
2392	if (r)
2393	goto bad;
2394
2395	while (opt_params) {
2396	string = dm_shift_arg(&as), opt_params--;
2397	if (!strcasecmp(string, "start_sector") && opt_params >= 1) {
2398	unsigned long long start_sector;
2399
2400	string = dm_shift_arg(&as), opt_params--;
2401	if (sscanf(string, "%llu%c", &start_sector, &dummy) != 1)
2402	goto invalid_optional;
2403	wc->start_sector = start_sector;
2404	wc->start_sector_set = true;
2405	if (wc->start_sector != start_sector \|\|
2406	wc->start_sector >= wc->memory_map_size >> SECTOR_SHIFT)
2407	goto invalid_optional;
2408	} else if (!strcasecmp(string, "high_watermark") && opt_params >= 1) {
2409	string = dm_shift_arg(&as), opt_params--;
2410	if (sscanf(string, "%d%c", &high_wm_percent, &dummy) != 1)
2411	goto invalid_optional;
2412	if (high_wm_percent < 0 \|\| high_wm_percent > 100)
2413	goto invalid_optional;
2414	wc->high_wm_percent_value = high_wm_percent;
2415	wc->high_wm_percent_set = true;
2416	} else if (!strcasecmp(string, "low_watermark") && opt_params >= 1) {
2417	string = dm_shift_arg(&as), opt_params--;
2418	if (sscanf(string, "%d%c", &low_wm_percent, &dummy) != 1)
2419	goto invalid_optional;
2420	if (low_wm_percent < 0 \|\| low_wm_percent > 100)
2421	goto invalid_optional;
2422	wc->low_wm_percent_value = low_wm_percent;
2423	wc->low_wm_percent_set = true;
2424	} else if (!strcasecmp(string, "writeback_jobs") && opt_params >= 1) {
2425	string = dm_shift_arg(&as), opt_params--;
2426	if (sscanf(string, "%u%c", &wc->max_writeback_jobs, &dummy) != 1)
2427	goto invalid_optional;
2428	wc->max_writeback_jobs_set = true;
2429	} else if (!strcasecmp(string, "autocommit_blocks") && opt_params >= 1) {
2430	string = dm_shift_arg(&as), opt_params--;
2431	if (sscanf(string, "%u%c", &wc->autocommit_blocks, &dummy) != 1)
2432	goto invalid_optional;
2433	wc->autocommit_blocks_set = true;
2434	} else if (!strcasecmp(string, "autocommit_time") && opt_params >= 1) {
2435	unsigned int autocommit_msecs;
2436
2437	string = dm_shift_arg(&as), opt_params--;
2438	if (sscanf(string, "%u%c", &autocommit_msecs, &dummy) != 1)
2439	goto invalid_optional;
2440	if (autocommit_msecs > 3600000)
2441	goto invalid_optional;
2442	wc->autocommit_jiffies = msecs_to_jiffies(autocommit_msecs);
2443	wc->autocommit_time_value = autocommit_msecs;
2444	wc->autocommit_time_set = true;
2445	} else if (!strcasecmp(string, "max_age") && opt_params >= 1) {
2446	unsigned int max_age_msecs;
2447
2448	string = dm_shift_arg(&as), opt_params--;
2449	if (sscanf(string, "%u%c", &max_age_msecs, &dummy) != 1)
2450	goto invalid_optional;
2451	if (max_age_msecs > 86400000)
2452	goto invalid_optional;
2453	wc->max_age = msecs_to_jiffies(max_age_msecs);
2454	wc->max_age_set = true;
2455	wc->max_age_value = max_age_msecs;
2456	} else if (!strcasecmp(string, "cleaner")) {
2457	wc->cleaner_set = true;
2458	wc->cleaner = true;
2459	} else if (!strcasecmp(string, "fua")) {
2460	if (WC_MODE_PMEM(wc)) {
2461	wc->writeback_fua = true;
2462	wc->writeback_fua_set = true;
2463	} else
2464	goto invalid_optional;
2465	} else if (!strcasecmp(string, "nofua")) {
2466	if (WC_MODE_PMEM(wc)) {
2467	wc->writeback_fua = false;
2468	wc->writeback_fua_set = true;
2469	} else
2470	goto invalid_optional;
2471	} else if (!strcasecmp(string, "metadata_only")) {
2472	wc->metadata_only = true;
2473	} else if (!strcasecmp(string, "pause_writeback") && opt_params >= 1) {
2474	unsigned int pause_msecs;
2475
2476	if (WC_MODE_PMEM(wc))
2477	goto invalid_optional;
2478	string = dm_shift_arg(&as), opt_params--;
2479	if (sscanf(string, "%u%c", &pause_msecs, &dummy) != 1)
2480	goto invalid_optional;
2481	if (pause_msecs > 60000)
2482	goto invalid_optional;
2483	wc->pause = msecs_to_jiffies(pause_msecs);
2484	wc->pause_set = true;
2485	wc->pause_value = pause_msecs;
2486	} else {
2487	invalid_optional:
2488	r = -EINVAL;
2489	ti->error = "Invalid optional argument";
2490	goto bad;
2491	}
2492	}
2493
2494	if (high_wm_percent < low_wm_percent) {
2495	r = -EINVAL;
2496	ti->error = "High watermark must be greater than or equal to low watermark";
2497	goto bad;
2498	}
2499
2500	if (WC_MODE_PMEM(wc)) {
2501	if (!dax_synchronous(wc->ssd_dev->dax_dev)) {
2502	r = -EOPNOTSUPP;
2503	ti->error = "Asynchronous persistent memory not supported as pmem cache";
2504	goto bad;
2505	}
2506
2507	r = persistent_memory_claim(wc);
2508	if (r) {
2509	ti->error = "Unable to map persistent memory for cache";
2510	goto bad;
2511	}
2512	} else {
2513	size_t n_blocks, n_metadata_blocks;
2514	uint64_t n_bitmap_bits;
2515
2516	wc->memory_map_size -= (uint64_t)wc->start_sector << SECTOR_SHIFT;
2517
2518	bio_list_init(&wc->flush_list);
2519	wc->flush_thread = kthread_run(writecache_flush_thread, wc, "dm_writecache_flush");
2520	if (IS_ERR(wc->flush_thread)) {
2521	r = PTR_ERR(wc->flush_thread);
2522	wc->flush_thread = NULL;
2523	ti->error = "Couldn't spawn flush thread";
2524	goto bad;
2525	}
2526
2527	r = calculate_memory_size(wc->memory_map_size, wc->block_size,
2528	&n_blocks, &n_metadata_blocks);
2529	if (r) {
2530	ti->error = "Invalid device size";
2531	goto bad;
2532	}
2533
2534	n_bitmap_bits = (((uint64_t)n_metadata_blocks << wc->block_size_bits) +
2535	BITMAP_GRANULARITY - 1) / BITMAP_GRANULARITY;
2536	/* this is limitation of test_bit functions */
2537	if (n_bitmap_bits > 1U << 31) {
2538	r = -EFBIG;
2539	ti->error = "Invalid device size";
2540	goto bad;
2541	}
2542
2543	wc->memory_map = vmalloc(n_metadata_blocks << wc->block_size_bits);
2544	if (!wc->memory_map) {
2545	r = -ENOMEM;
2546	ti->error = "Unable to allocate memory for metadata";
2547	goto bad;
2548	}
2549
2550	wc->dm_kcopyd = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2551	if (IS_ERR(wc->dm_kcopyd)) {
2552	r = PTR_ERR(wc->dm_kcopyd);
2553	ti->error = "Unable to allocate dm-kcopyd client";
2554	wc->dm_kcopyd = NULL;
2555	goto bad;
2556	}
2557
2558	wc->metadata_sectors = n_metadata_blocks << (wc->block_size_bits - SECTOR_SHIFT);
2559	wc->dirty_bitmap_size = (n_bitmap_bits + BITS_PER_LONG - 1) /
2560	BITS_PER_LONG * sizeof(unsigned long);
2561	wc->dirty_bitmap = vzalloc(wc->dirty_bitmap_size);
2562	if (!wc->dirty_bitmap) {
2563	r = -ENOMEM;
2564	ti->error = "Unable to allocate dirty bitmap";
2565	goto bad;
2566	}
2567
2568	r = writecache_read_metadata(wc, wc->block_size >> SECTOR_SHIFT);
2569	if (r) {
2570	ti->error = "Unable to read first block of metadata";
2571	goto bad;
2572	}
2573	}
2574
2575	r = copy_mc_to_kernel(&s, sb(wc), sizeof(struct wc_memory_superblock));
2576	if (r) {
2577	ti->error = "Hardware memory error when reading superblock";
2578	goto bad;
2579	}
2580	if (!le32_to_cpu(s.magic) && !le32_to_cpu(s.version)) {
2581	r = init_memory(wc);
2582	if (r) {
2583	ti->error = "Unable to initialize device";
2584	goto bad;
2585	}
2586	r = copy_mc_to_kernel(&s, sb(wc),
2587	sizeof(struct wc_memory_superblock));
2588	if (r) {
2589	ti->error = "Hardware memory error when reading superblock";
2590	goto bad;
2591	}
2592	}
2593
2594	if (le32_to_cpu(s.magic) != MEMORY_SUPERBLOCK_MAGIC) {
2595	ti->error = "Invalid magic in the superblock";
2596	r = -EINVAL;
2597	goto bad;
2598	}
2599
2600	if (le32_to_cpu(s.version) != MEMORY_SUPERBLOCK_VERSION) {
2601	ti->error = "Invalid version in the superblock";
2602	r = -EINVAL;
2603	goto bad;
2604	}
2605
2606	if (le32_to_cpu(s.block_size) != wc->block_size) {
2607	ti->error = "Block size does not match superblock";
2608	r = -EINVAL;
2609	goto bad;
2610	}
2611
2612	wc->n_blocks = le64_to_cpu(s.n_blocks);
2613
2614	offset = wc->n_blocks * sizeof(struct wc_memory_entry);
2615	if (offset / sizeof(struct wc_memory_entry) != le64_to_cpu(sb(wc)->n_blocks)) {
2616	overflow:
2617	ti->error = "Overflow in size calculation";
2618	r = -EINVAL;
2619	goto bad;
2620	}
2621	offset += sizeof(struct wc_memory_superblock);
2622	if (offset < sizeof(struct wc_memory_superblock))
2623	goto overflow;
2624	offset = (offset + wc->block_size - 1) & ~(size_t)(wc->block_size - 1);
2625	data_size = wc->n_blocks * (size_t)wc->block_size;
2626	if (!offset \|\| (data_size / wc->block_size != wc->n_blocks) \|\|
2627	(offset + data_size < offset))
2628	goto overflow;
2629	if (offset + data_size > wc->memory_map_size) {
2630	ti->error = "Memory area is too small";
2631	r = -EINVAL;
2632	goto bad;
2633	}
2634
2635	wc->metadata_sectors = offset >> SECTOR_SHIFT;
2636	wc->block_start = (char *)sb(wc) + offset;
2637
2638	x = (uint64_t)wc->n_blocks * (100 - high_wm_percent);
2639	x += 50;
2640	do_div(x, 100);
2641	wc->freelist_high_watermark = x;
2642	x = (uint64_t)wc->n_blocks * (100 - low_wm_percent);
2643	x += 50;
2644	do_div(x, 100);
2645	wc->freelist_low_watermark = x;
2646
2647	if (wc->cleaner)
2648	activate_cleaner(wc);
2649
2650	r = writecache_alloc_entries(wc);
2651	if (r) {
2652	ti->error = "Cannot allocate memory";
2653	goto bad;
2654	}
2655
2656	ti->num_flush_bios = WC_MODE_PMEM(wc) ? 1 : 2;
2657	ti->flush_supported = true;
2658	ti->num_discard_bios = 1;
2659
2660	if (WC_MODE_PMEM(wc))
2661	persistent_memory_flush_cache(wc->memory_map, wc->memory_map_size);
2662
2663	return 0;
2664
2665	bad_arguments:
2666	r = -EINVAL;
2667	ti->error = "Bad arguments";
2668	bad:
2669	writecache_dtr(ti);
2670	return r;
2671	}
2672
2673	static void writecache_status(struct dm_target *ti, status_type_t type,
2674	unsigned int status_flags, char *result, unsigned int maxlen)
2675	{
2676	struct dm_writecache *wc = ti->private;
2677	unsigned int extra_args;
2678	unsigned int sz = 0;
2679
2680	switch (type) {
2681	case STATUSTYPE_INFO:
2682	DMEMIT("%ld %llu %llu %llu %llu %llu %llu %llu %llu %llu %llu %llu %llu %llu",
2683	writecache_has_error(wc),
2684	(unsigned long long)wc->n_blocks, (unsigned long long)wc->freelist_size,
2685	(unsigned long long)wc->writeback_size,
2686	wc->stats.reads,
2687	wc->stats.read_hits,
2688	wc->stats.writes,
2689	wc->stats.write_hits_uncommitted,
2690	wc->stats.write_hits_committed,
2691	wc->stats.writes_around,
2692	wc->stats.writes_allocate,
2693	wc->stats.writes_blocked_on_freelist,
2694	wc->stats.flushes,
2695	wc->stats.discards);
2696	break;
2697	case STATUSTYPE_TABLE:
2698	DMEMIT("%c %s %s %u ", WC_MODE_PMEM(wc) ? 'p' : 's',
2699	wc->dev->name, wc->ssd_dev->name, wc->block_size);
2700	extra_args = 0;
2701	if (wc->start_sector_set)
2702	extra_args += 2;
2703	if (wc->high_wm_percent_set)
2704	extra_args += 2;
2705	if (wc->low_wm_percent_set)
2706	extra_args += 2;
2707	if (wc->max_writeback_jobs_set)
2708	extra_args += 2;
2709	if (wc->autocommit_blocks_set)
2710	extra_args += 2;
2711	if (wc->autocommit_time_set)
2712	extra_args += 2;
2713	if (wc->max_age_set)
2714	extra_args += 2;
2715	if (wc->cleaner_set)
2716	extra_args++;
2717	if (wc->writeback_fua_set)
2718	extra_args++;
2719	if (wc->metadata_only)
2720	extra_args++;
2721	if (wc->pause_set)
2722	extra_args += 2;
2723
2724	DMEMIT("%u", extra_args);
2725	if (wc->start_sector_set)
2726	DMEMIT(" start_sector %llu", (unsigned long long)wc->start_sector);
2727	if (wc->high_wm_percent_set)
2728	DMEMIT(" high_watermark %u", wc->high_wm_percent_value);
2729	if (wc->low_wm_percent_set)
2730	DMEMIT(" low_watermark %u", wc->low_wm_percent_value);
2731	if (wc->max_writeback_jobs_set)
2732	DMEMIT(" writeback_jobs %u", wc->max_writeback_jobs);
2733	if (wc->autocommit_blocks_set)
2734	DMEMIT(" autocommit_blocks %u", wc->autocommit_blocks);
2735	if (wc->autocommit_time_set)
2736	DMEMIT(" autocommit_time %u", wc->autocommit_time_value);
2737	if (wc->max_age_set)
2738	DMEMIT(" max_age %u", wc->max_age_value);
2739	if (wc->cleaner_set)
2740	DMEMIT(" cleaner");
2741	if (wc->writeback_fua_set)
2742	DMEMIT(" %sfua", wc->writeback_fua ? "" : "no");
2743	if (wc->metadata_only)
2744	DMEMIT(" metadata_only");
2745	if (wc->pause_set)
2746	DMEMIT(" pause_writeback %u", wc->pause_value);
2747	break;
2748	case STATUSTYPE_IMA:
2749	*result = '\0';
2750	break;
2751	}
2752	}
2753
2754	static struct target_type writecache_target = {
2755	.name = "writecache",
2756	.version = {1, 6, 0},
2757	.module = THIS_MODULE,
2758	.ctr = writecache_ctr,
2759	.dtr = writecache_dtr,
2760	.status = writecache_status,
2761	.postsuspend = writecache_suspend,
2762	.resume = writecache_resume,
2763	.message = writecache_message,
2764	.map = writecache_map,
2765	.end_io = writecache_end_io,
2766	.iterate_devices = writecache_iterate_devices,
2767	.io_hints = writecache_io_hints,
2768	};
2769	module_dm(writecache);
2770
2771	MODULE_DESCRIPTION(DM_NAME " writecache target");
2772	MODULE_AUTHOR("Mikulas Patocka <[email protected]>");
2773	MODULE_LICENSE("GPL");