xref: /openbmc/linux/drivers/md/raid5-ppl.c (revision f7d84fa7)
1 /*
2  * Partial Parity Log for closing the RAID5 write hole
3  * Copyright (c) 2017, Intel Corporation.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  */
14 
15 #include <linux/kernel.h>
16 #include <linux/blkdev.h>
17 #include <linux/slab.h>
18 #include <linux/crc32c.h>
19 #include <linux/flex_array.h>
20 #include <linux/async_tx.h>
21 #include <linux/raid/md_p.h>
22 #include "md.h"
23 #include "raid5.h"
24 
25 /*
26  * PPL consists of a 4KB header (struct ppl_header) and at least 128KB for
27  * partial parity data. The header contains an array of entries
28  * (struct ppl_header_entry) which describe the logged write requests.
29  * Partial parity for the entries comes after the header, written in the same
30  * sequence as the entries:
31  *
32  * Header
33  *   entry0
34  *   ...
35  *   entryN
36  * PP data
37  *   PP for entry0
38  *   ...
39  *   PP for entryN
40  *
41  * An entry describes one or more consecutive stripe_heads, up to a full
42  * stripe. The modifed raid data chunks form an m-by-n matrix, where m is the
43  * number of stripe_heads in the entry and n is the number of modified data
44  * disks. Every stripe_head in the entry must write to the same data disks.
45  * An example of a valid case described by a single entry (writes to the first
46  * stripe of a 4 disk array, 16k chunk size):
47  *
48  * sh->sector   dd0   dd1   dd2    ppl
49  *            +-----+-----+-----+
50  * 0          | --- | --- | --- | +----+
51  * 8          | -W- | -W- | --- | | pp |   data_sector = 8
52  * 16         | -W- | -W- | --- | | pp |   data_size = 3 * 2 * 4k
53  * 24         | -W- | -W- | --- | | pp |   pp_size = 3 * 4k
54  *            +-----+-----+-----+ +----+
55  *
56  * data_sector is the first raid sector of the modified data, data_size is the
57  * total size of modified data and pp_size is the size of partial parity for
58  * this entry. Entries for full stripe writes contain no partial parity
59  * (pp_size = 0), they only mark the stripes for which parity should be
60  * recalculated after an unclean shutdown. Every entry holds a checksum of its
61  * partial parity, the header also has a checksum of the header itself.
62  *
63  * A write request is always logged to the PPL instance stored on the parity
64  * disk of the corresponding stripe. For each member disk there is one ppl_log
65  * used to handle logging for this disk, independently from others. They are
66  * grouped in child_logs array in struct ppl_conf, which is assigned to
67  * r5conf->log_private.
68  *
69  * ppl_io_unit represents a full PPL write, header_page contains the ppl_header.
70  * PPL entries for logged stripes are added in ppl_log_stripe(). A stripe_head
71  * can be appended to the last entry if it meets the conditions for a valid
72  * entry described above, otherwise a new entry is added. Checksums of entries
73  * are calculated incrementally as stripes containing partial parity are being
74  * added. ppl_submit_iounit() calculates the checksum of the header and submits
75  * a bio containing the header page and partial parity pages (sh->ppl_page) for
76  * all stripes of the io_unit. When the PPL write completes, the stripes
77  * associated with the io_unit are released and raid5d starts writing their data
78  * and parity. When all stripes are written, the io_unit is freed and the next
79  * can be submitted.
80  *
81  * An io_unit is used to gather stripes until it is submitted or becomes full
82  * (if the maximum number of entries or size of PPL is reached). Another io_unit
83  * can't be submitted until the previous has completed (PPL and stripe
84  * data+parity is written). The log->io_list tracks all io_units of a log
85  * (for a single member disk). New io_units are added to the end of the list
86  * and the first io_unit is submitted, if it is not submitted already.
87  * The current io_unit accepting new stripes is always at the end of the list.
88  */
89 
90 struct ppl_conf {
91 	struct mddev *mddev;
92 
93 	/* array of child logs, one for each raid disk */
94 	struct ppl_log *child_logs;
95 	int count;
96 
97 	int block_size;		/* the logical block size used for data_sector
98 				 * in ppl_header_entry */
99 	u32 signature;		/* raid array identifier */
100 	atomic64_t seq;		/* current log write sequence number */
101 
102 	struct kmem_cache *io_kc;
103 	mempool_t *io_pool;
104 	struct bio_set *bs;
105 
106 	/* used only for recovery */
107 	int recovered_entries;
108 	int mismatch_count;
109 
110 	/* stripes to retry if failed to allocate io_unit */
111 	struct list_head no_mem_stripes;
112 	spinlock_t no_mem_stripes_lock;
113 };
114 
115 struct ppl_log {
116 	struct ppl_conf *ppl_conf;	/* shared between all log instances */
117 
118 	struct md_rdev *rdev;		/* array member disk associated with
119 					 * this log instance */
120 	struct mutex io_mutex;
121 	struct ppl_io_unit *current_io;	/* current io_unit accepting new data
122 					 * always at the end of io_list */
123 	spinlock_t io_list_lock;
124 	struct list_head io_list;	/* all io_units of this log */
125 };
126 
127 #define PPL_IO_INLINE_BVECS 32
128 
129 struct ppl_io_unit {
130 	struct ppl_log *log;
131 
132 	struct page *header_page;	/* for ppl_header */
133 
134 	unsigned int entries_count;	/* number of entries in ppl_header */
135 	unsigned int pp_size;		/* total size current of partial parity */
136 
137 	u64 seq;			/* sequence number of this log write */
138 	struct list_head log_sibling;	/* log->io_list */
139 
140 	struct list_head stripe_list;	/* stripes added to the io_unit */
141 	atomic_t pending_stripes;	/* how many stripes not written to raid */
142 
143 	bool submitted;			/* true if write to log started */
144 
145 	/* inline bio and its biovec for submitting the iounit */
146 	struct bio bio;
147 	struct bio_vec biovec[PPL_IO_INLINE_BVECS];
148 };
149 
150 struct dma_async_tx_descriptor *
151 ops_run_partial_parity(struct stripe_head *sh, struct raid5_percpu *percpu,
152 		       struct dma_async_tx_descriptor *tx)
153 {
154 	int disks = sh->disks;
155 	struct page **srcs = flex_array_get(percpu->scribble, 0);
156 	int count = 0, pd_idx = sh->pd_idx, i;
157 	struct async_submit_ctl submit;
158 
159 	pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector);
160 
161 	/*
162 	 * Partial parity is the XOR of stripe data chunks that are not changed
163 	 * during the write request. Depending on available data
164 	 * (read-modify-write vs. reconstruct-write case) we calculate it
165 	 * differently.
166 	 */
167 	if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) {
168 		/*
169 		 * rmw: xor old data and parity from updated disks
170 		 * This is calculated earlier by ops_run_prexor5() so just copy
171 		 * the parity dev page.
172 		 */
173 		srcs[count++] = sh->dev[pd_idx].page;
174 	} else if (sh->reconstruct_state == reconstruct_state_drain_run) {
175 		/* rcw: xor data from all not updated disks */
176 		for (i = disks; i--;) {
177 			struct r5dev *dev = &sh->dev[i];
178 			if (test_bit(R5_UPTODATE, &dev->flags))
179 				srcs[count++] = dev->page;
180 		}
181 	} else {
182 		return tx;
183 	}
184 
185 	init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, tx,
186 			  NULL, sh, flex_array_get(percpu->scribble, 0)
187 			  + sizeof(struct page *) * (sh->disks + 2));
188 
189 	if (count == 1)
190 		tx = async_memcpy(sh->ppl_page, srcs[0], 0, 0, PAGE_SIZE,
191 				  &submit);
192 	else
193 		tx = async_xor(sh->ppl_page, srcs, 0, count, PAGE_SIZE,
194 			       &submit);
195 
196 	return tx;
197 }
198 
199 static void *ppl_io_pool_alloc(gfp_t gfp_mask, void *pool_data)
200 {
201 	struct kmem_cache *kc = pool_data;
202 	struct ppl_io_unit *io;
203 
204 	io = kmem_cache_alloc(kc, gfp_mask);
205 	if (!io)
206 		return NULL;
207 
208 	io->header_page = alloc_page(gfp_mask);
209 	if (!io->header_page) {
210 		kmem_cache_free(kc, io);
211 		return NULL;
212 	}
213 
214 	return io;
215 }
216 
217 static void ppl_io_pool_free(void *element, void *pool_data)
218 {
219 	struct kmem_cache *kc = pool_data;
220 	struct ppl_io_unit *io = element;
221 
222 	__free_page(io->header_page);
223 	kmem_cache_free(kc, io);
224 }
225 
226 static struct ppl_io_unit *ppl_new_iounit(struct ppl_log *log,
227 					  struct stripe_head *sh)
228 {
229 	struct ppl_conf *ppl_conf = log->ppl_conf;
230 	struct ppl_io_unit *io;
231 	struct ppl_header *pplhdr;
232 	struct page *header_page;
233 
234 	io = mempool_alloc(ppl_conf->io_pool, GFP_NOWAIT);
235 	if (!io)
236 		return NULL;
237 
238 	header_page = io->header_page;
239 	memset(io, 0, sizeof(*io));
240 	io->header_page = header_page;
241 
242 	io->log = log;
243 	INIT_LIST_HEAD(&io->log_sibling);
244 	INIT_LIST_HEAD(&io->stripe_list);
245 	atomic_set(&io->pending_stripes, 0);
246 	bio_init(&io->bio, io->biovec, PPL_IO_INLINE_BVECS);
247 
248 	pplhdr = page_address(io->header_page);
249 	clear_page(pplhdr);
250 	memset(pplhdr->reserved, 0xff, PPL_HDR_RESERVED);
251 	pplhdr->signature = cpu_to_le32(ppl_conf->signature);
252 
253 	io->seq = atomic64_add_return(1, &ppl_conf->seq);
254 	pplhdr->generation = cpu_to_le64(io->seq);
255 
256 	return io;
257 }
258 
259 static int ppl_log_stripe(struct ppl_log *log, struct stripe_head *sh)
260 {
261 	struct ppl_io_unit *io = log->current_io;
262 	struct ppl_header_entry *e = NULL;
263 	struct ppl_header *pplhdr;
264 	int i;
265 	sector_t data_sector = 0;
266 	int data_disks = 0;
267 	unsigned int entry_space = (log->rdev->ppl.size << 9) - PPL_HEADER_SIZE;
268 	struct r5conf *conf = sh->raid_conf;
269 
270 	pr_debug("%s: stripe: %llu\n", __func__, (unsigned long long)sh->sector);
271 
272 	/* check if current io_unit is full */
273 	if (io && (io->pp_size == entry_space ||
274 		   io->entries_count == PPL_HDR_MAX_ENTRIES)) {
275 		pr_debug("%s: add io_unit blocked by seq: %llu\n",
276 			 __func__, io->seq);
277 		io = NULL;
278 	}
279 
280 	/* add a new unit if there is none or the current is full */
281 	if (!io) {
282 		io = ppl_new_iounit(log, sh);
283 		if (!io)
284 			return -ENOMEM;
285 		spin_lock_irq(&log->io_list_lock);
286 		list_add_tail(&io->log_sibling, &log->io_list);
287 		spin_unlock_irq(&log->io_list_lock);
288 
289 		log->current_io = io;
290 	}
291 
292 	for (i = 0; i < sh->disks; i++) {
293 		struct r5dev *dev = &sh->dev[i];
294 
295 		if (i != sh->pd_idx && test_bit(R5_Wantwrite, &dev->flags)) {
296 			if (!data_disks || dev->sector < data_sector)
297 				data_sector = dev->sector;
298 			data_disks++;
299 		}
300 	}
301 	BUG_ON(!data_disks);
302 
303 	pr_debug("%s: seq: %llu data_sector: %llu data_disks: %d\n", __func__,
304 		 io->seq, (unsigned long long)data_sector, data_disks);
305 
306 	pplhdr = page_address(io->header_page);
307 
308 	if (io->entries_count > 0) {
309 		struct ppl_header_entry *last =
310 				&pplhdr->entries[io->entries_count - 1];
311 		struct stripe_head *sh_last = list_last_entry(
312 				&io->stripe_list, struct stripe_head, log_list);
313 		u64 data_sector_last = le64_to_cpu(last->data_sector);
314 		u32 data_size_last = le32_to_cpu(last->data_size);
315 
316 		/*
317 		 * Check if we can append the stripe to the last entry. It must
318 		 * be just after the last logged stripe and write to the same
319 		 * disks. Use bit shift and logarithm to avoid 64-bit division.
320 		 */
321 		if ((sh->sector == sh_last->sector + STRIPE_SECTORS) &&
322 		    (data_sector >> ilog2(conf->chunk_sectors) ==
323 		     data_sector_last >> ilog2(conf->chunk_sectors)) &&
324 		    ((data_sector - data_sector_last) * data_disks ==
325 		     data_size_last >> 9))
326 			e = last;
327 	}
328 
329 	if (!e) {
330 		e = &pplhdr->entries[io->entries_count++];
331 		e->data_sector = cpu_to_le64(data_sector);
332 		e->parity_disk = cpu_to_le32(sh->pd_idx);
333 		e->checksum = cpu_to_le32(~0);
334 	}
335 
336 	le32_add_cpu(&e->data_size, data_disks << PAGE_SHIFT);
337 
338 	/* don't write any PP if full stripe write */
339 	if (!test_bit(STRIPE_FULL_WRITE, &sh->state)) {
340 		le32_add_cpu(&e->pp_size, PAGE_SIZE);
341 		io->pp_size += PAGE_SIZE;
342 		e->checksum = cpu_to_le32(crc32c_le(le32_to_cpu(e->checksum),
343 						    page_address(sh->ppl_page),
344 						    PAGE_SIZE));
345 	}
346 
347 	list_add_tail(&sh->log_list, &io->stripe_list);
348 	atomic_inc(&io->pending_stripes);
349 	sh->ppl_io = io;
350 
351 	return 0;
352 }
353 
354 int ppl_write_stripe(struct r5conf *conf, struct stripe_head *sh)
355 {
356 	struct ppl_conf *ppl_conf = conf->log_private;
357 	struct ppl_io_unit *io = sh->ppl_io;
358 	struct ppl_log *log;
359 
360 	if (io || test_bit(STRIPE_SYNCING, &sh->state) || !sh->ppl_page ||
361 	    !test_bit(R5_Wantwrite, &sh->dev[sh->pd_idx].flags) ||
362 	    !test_bit(R5_Insync, &sh->dev[sh->pd_idx].flags)) {
363 		clear_bit(STRIPE_LOG_TRAPPED, &sh->state);
364 		return -EAGAIN;
365 	}
366 
367 	log = &ppl_conf->child_logs[sh->pd_idx];
368 
369 	mutex_lock(&log->io_mutex);
370 
371 	if (!log->rdev || test_bit(Faulty, &log->rdev->flags)) {
372 		mutex_unlock(&log->io_mutex);
373 		return -EAGAIN;
374 	}
375 
376 	set_bit(STRIPE_LOG_TRAPPED, &sh->state);
377 	clear_bit(STRIPE_DELAYED, &sh->state);
378 	atomic_inc(&sh->count);
379 
380 	if (ppl_log_stripe(log, sh)) {
381 		spin_lock_irq(&ppl_conf->no_mem_stripes_lock);
382 		list_add_tail(&sh->log_list, &ppl_conf->no_mem_stripes);
383 		spin_unlock_irq(&ppl_conf->no_mem_stripes_lock);
384 	}
385 
386 	mutex_unlock(&log->io_mutex);
387 
388 	return 0;
389 }
390 
391 static void ppl_log_endio(struct bio *bio)
392 {
393 	struct ppl_io_unit *io = bio->bi_private;
394 	struct ppl_log *log = io->log;
395 	struct ppl_conf *ppl_conf = log->ppl_conf;
396 	struct stripe_head *sh, *next;
397 
398 	pr_debug("%s: seq: %llu\n", __func__, io->seq);
399 
400 	if (bio->bi_error)
401 		md_error(ppl_conf->mddev, log->rdev);
402 
403 	list_for_each_entry_safe(sh, next, &io->stripe_list, log_list) {
404 		list_del_init(&sh->log_list);
405 
406 		set_bit(STRIPE_HANDLE, &sh->state);
407 		raid5_release_stripe(sh);
408 	}
409 }
410 
411 static void ppl_submit_iounit_bio(struct ppl_io_unit *io, struct bio *bio)
412 {
413 	char b[BDEVNAME_SIZE];
414 
415 	pr_debug("%s: seq: %llu size: %u sector: %llu dev: %s\n",
416 		 __func__, io->seq, bio->bi_iter.bi_size,
417 		 (unsigned long long)bio->bi_iter.bi_sector,
418 		 bdevname(bio->bi_bdev, b));
419 
420 	submit_bio(bio);
421 }
422 
423 static void ppl_submit_iounit(struct ppl_io_unit *io)
424 {
425 	struct ppl_log *log = io->log;
426 	struct ppl_conf *ppl_conf = log->ppl_conf;
427 	struct ppl_header *pplhdr = page_address(io->header_page);
428 	struct bio *bio = &io->bio;
429 	struct stripe_head *sh;
430 	int i;
431 
432 	bio->bi_private = io;
433 
434 	if (!log->rdev || test_bit(Faulty, &log->rdev->flags)) {
435 		ppl_log_endio(bio);
436 		return;
437 	}
438 
439 	for (i = 0; i < io->entries_count; i++) {
440 		struct ppl_header_entry *e = &pplhdr->entries[i];
441 
442 		pr_debug("%s: seq: %llu entry: %d data_sector: %llu pp_size: %u data_size: %u\n",
443 			 __func__, io->seq, i, le64_to_cpu(e->data_sector),
444 			 le32_to_cpu(e->pp_size), le32_to_cpu(e->data_size));
445 
446 		e->data_sector = cpu_to_le64(le64_to_cpu(e->data_sector) >>
447 					     ilog2(ppl_conf->block_size >> 9));
448 		e->checksum = cpu_to_le32(~le32_to_cpu(e->checksum));
449 	}
450 
451 	pplhdr->entries_count = cpu_to_le32(io->entries_count);
452 	pplhdr->checksum = cpu_to_le32(~crc32c_le(~0, pplhdr, PPL_HEADER_SIZE));
453 
454 	bio->bi_end_io = ppl_log_endio;
455 	bio->bi_opf = REQ_OP_WRITE | REQ_FUA;
456 	bio->bi_bdev = log->rdev->bdev;
457 	bio->bi_iter.bi_sector = log->rdev->ppl.sector;
458 	bio_add_page(bio, io->header_page, PAGE_SIZE, 0);
459 
460 	list_for_each_entry(sh, &io->stripe_list, log_list) {
461 		/* entries for full stripe writes have no partial parity */
462 		if (test_bit(STRIPE_FULL_WRITE, &sh->state))
463 			continue;
464 
465 		if (!bio_add_page(bio, sh->ppl_page, PAGE_SIZE, 0)) {
466 			struct bio *prev = bio;
467 
468 			bio = bio_alloc_bioset(GFP_NOIO, BIO_MAX_PAGES,
469 					       ppl_conf->bs);
470 			bio->bi_opf = prev->bi_opf;
471 			bio->bi_bdev = prev->bi_bdev;
472 			bio->bi_iter.bi_sector = bio_end_sector(prev);
473 			bio_add_page(bio, sh->ppl_page, PAGE_SIZE, 0);
474 
475 			bio_chain(bio, prev);
476 			ppl_submit_iounit_bio(io, prev);
477 		}
478 	}
479 
480 	ppl_submit_iounit_bio(io, bio);
481 }
482 
483 static void ppl_submit_current_io(struct ppl_log *log)
484 {
485 	struct ppl_io_unit *io;
486 
487 	spin_lock_irq(&log->io_list_lock);
488 
489 	io = list_first_entry_or_null(&log->io_list, struct ppl_io_unit,
490 				      log_sibling);
491 	if (io && io->submitted)
492 		io = NULL;
493 
494 	spin_unlock_irq(&log->io_list_lock);
495 
496 	if (io) {
497 		io->submitted = true;
498 
499 		if (io == log->current_io)
500 			log->current_io = NULL;
501 
502 		ppl_submit_iounit(io);
503 	}
504 }
505 
506 void ppl_write_stripe_run(struct r5conf *conf)
507 {
508 	struct ppl_conf *ppl_conf = conf->log_private;
509 	struct ppl_log *log;
510 	int i;
511 
512 	for (i = 0; i < ppl_conf->count; i++) {
513 		log = &ppl_conf->child_logs[i];
514 
515 		mutex_lock(&log->io_mutex);
516 		ppl_submit_current_io(log);
517 		mutex_unlock(&log->io_mutex);
518 	}
519 }
520 
521 static void ppl_io_unit_finished(struct ppl_io_unit *io)
522 {
523 	struct ppl_log *log = io->log;
524 	struct ppl_conf *ppl_conf = log->ppl_conf;
525 	unsigned long flags;
526 
527 	pr_debug("%s: seq: %llu\n", __func__, io->seq);
528 
529 	local_irq_save(flags);
530 
531 	spin_lock(&log->io_list_lock);
532 	list_del(&io->log_sibling);
533 	spin_unlock(&log->io_list_lock);
534 
535 	mempool_free(io, ppl_conf->io_pool);
536 
537 	spin_lock(&ppl_conf->no_mem_stripes_lock);
538 	if (!list_empty(&ppl_conf->no_mem_stripes)) {
539 		struct stripe_head *sh;
540 
541 		sh = list_first_entry(&ppl_conf->no_mem_stripes,
542 				      struct stripe_head, log_list);
543 		list_del_init(&sh->log_list);
544 		set_bit(STRIPE_HANDLE, &sh->state);
545 		raid5_release_stripe(sh);
546 	}
547 	spin_unlock(&ppl_conf->no_mem_stripes_lock);
548 
549 	local_irq_restore(flags);
550 }
551 
552 void ppl_stripe_write_finished(struct stripe_head *sh)
553 {
554 	struct ppl_io_unit *io;
555 
556 	io = sh->ppl_io;
557 	sh->ppl_io = NULL;
558 
559 	if (io && atomic_dec_and_test(&io->pending_stripes))
560 		ppl_io_unit_finished(io);
561 }
562 
563 static void ppl_xor(int size, struct page *page1, struct page *page2)
564 {
565 	struct async_submit_ctl submit;
566 	struct dma_async_tx_descriptor *tx;
567 	struct page *xor_srcs[] = { page1, page2 };
568 
569 	init_async_submit(&submit, ASYNC_TX_ACK|ASYNC_TX_XOR_DROP_DST,
570 			  NULL, NULL, NULL, NULL);
571 	tx = async_xor(page1, xor_srcs, 0, 2, size, &submit);
572 
573 	async_tx_quiesce(&tx);
574 }
575 
576 /*
577  * PPL recovery strategy: xor partial parity and data from all modified data
578  * disks within a stripe and write the result as the new stripe parity. If all
579  * stripe data disks are modified (full stripe write), no partial parity is
580  * available, so just xor the data disks.
581  *
582  * Recovery of a PPL entry shall occur only if all modified data disks are
583  * available and read from all of them succeeds.
584  *
585  * A PPL entry applies to a stripe, partial parity size for an entry is at most
586  * the size of the chunk. Examples of possible cases for a single entry:
587  *
588  * case 0: single data disk write:
589  *   data0    data1    data2     ppl        parity
590  * +--------+--------+--------+           +--------------------+
591  * | ------ | ------ | ------ | +----+    | (no change)        |
592  * | ------ | -data- | ------ | | pp | -> | data1 ^ pp         |
593  * | ------ | -data- | ------ | | pp | -> | data1 ^ pp         |
594  * | ------ | ------ | ------ | +----+    | (no change)        |
595  * +--------+--------+--------+           +--------------------+
596  * pp_size = data_size
597  *
598  * case 1: more than one data disk write:
599  *   data0    data1    data2     ppl        parity
600  * +--------+--------+--------+           +--------------------+
601  * | ------ | ------ | ------ | +----+    | (no change)        |
602  * | -data- | -data- | ------ | | pp | -> | data0 ^ data1 ^ pp |
603  * | -data- | -data- | ------ | | pp | -> | data0 ^ data1 ^ pp |
604  * | ------ | ------ | ------ | +----+    | (no change)        |
605  * +--------+--------+--------+           +--------------------+
606  * pp_size = data_size / modified_data_disks
607  *
608  * case 2: write to all data disks (also full stripe write):
609  *   data0    data1    data2                parity
610  * +--------+--------+--------+           +--------------------+
611  * | ------ | ------ | ------ |           | (no change)        |
612  * | -data- | -data- | -data- | --------> | xor all data       |
613  * | ------ | ------ | ------ | --------> | (no change)        |
614  * | ------ | ------ | ------ |           | (no change)        |
615  * +--------+--------+--------+           +--------------------+
616  * pp_size = 0
617  *
618  * The following cases are possible only in other implementations. The recovery
619  * code can handle them, but they are not generated at runtime because they can
620  * be reduced to cases 0, 1 and 2:
621  *
622  * case 3:
623  *   data0    data1    data2     ppl        parity
624  * +--------+--------+--------+ +----+    +--------------------+
625  * | ------ | -data- | -data- | | pp |    | data1 ^ data2 ^ pp |
626  * | ------ | -data- | -data- | | pp | -> | data1 ^ data2 ^ pp |
627  * | -data- | -data- | -data- | | -- | -> | xor all data       |
628  * | -data- | -data- | ------ | | pp |    | data0 ^ data1 ^ pp |
629  * +--------+--------+--------+ +----+    +--------------------+
630  * pp_size = chunk_size
631  *
632  * case 4:
633  *   data0    data1    data2     ppl        parity
634  * +--------+--------+--------+ +----+    +--------------------+
635  * | ------ | -data- | ------ | | pp |    | data1 ^ pp         |
636  * | ------ | ------ | ------ | | -- | -> | (no change)        |
637  * | ------ | ------ | ------ | | -- | -> | (no change)        |
638  * | -data- | ------ | ------ | | pp |    | data0 ^ pp         |
639  * +--------+--------+--------+ +----+    +--------------------+
640  * pp_size = chunk_size
641  */
642 static int ppl_recover_entry(struct ppl_log *log, struct ppl_header_entry *e,
643 			     sector_t ppl_sector)
644 {
645 	struct ppl_conf *ppl_conf = log->ppl_conf;
646 	struct mddev *mddev = ppl_conf->mddev;
647 	struct r5conf *conf = mddev->private;
648 	int block_size = ppl_conf->block_size;
649 	struct page *page1;
650 	struct page *page2;
651 	sector_t r_sector_first;
652 	sector_t r_sector_last;
653 	int strip_sectors;
654 	int data_disks;
655 	int i;
656 	int ret = 0;
657 	char b[BDEVNAME_SIZE];
658 	unsigned int pp_size = le32_to_cpu(e->pp_size);
659 	unsigned int data_size = le32_to_cpu(e->data_size);
660 
661 	page1 = alloc_page(GFP_KERNEL);
662 	page2 = alloc_page(GFP_KERNEL);
663 
664 	if (!page1 || !page2) {
665 		ret = -ENOMEM;
666 		goto out;
667 	}
668 
669 	r_sector_first = le64_to_cpu(e->data_sector) * (block_size >> 9);
670 
671 	if ((pp_size >> 9) < conf->chunk_sectors) {
672 		if (pp_size > 0) {
673 			data_disks = data_size / pp_size;
674 			strip_sectors = pp_size >> 9;
675 		} else {
676 			data_disks = conf->raid_disks - conf->max_degraded;
677 			strip_sectors = (data_size >> 9) / data_disks;
678 		}
679 		r_sector_last = r_sector_first +
680 				(data_disks - 1) * conf->chunk_sectors +
681 				strip_sectors;
682 	} else {
683 		data_disks = conf->raid_disks - conf->max_degraded;
684 		strip_sectors = conf->chunk_sectors;
685 		r_sector_last = r_sector_first + (data_size >> 9);
686 	}
687 
688 	pr_debug("%s: array sector first: %llu last: %llu\n", __func__,
689 		 (unsigned long long)r_sector_first,
690 		 (unsigned long long)r_sector_last);
691 
692 	/* if start and end is 4k aligned, use a 4k block */
693 	if (block_size == 512 &&
694 	    (r_sector_first & (STRIPE_SECTORS - 1)) == 0 &&
695 	    (r_sector_last & (STRIPE_SECTORS - 1)) == 0)
696 		block_size = STRIPE_SIZE;
697 
698 	/* iterate through blocks in strip */
699 	for (i = 0; i < strip_sectors; i += (block_size >> 9)) {
700 		bool update_parity = false;
701 		sector_t parity_sector;
702 		struct md_rdev *parity_rdev;
703 		struct stripe_head sh;
704 		int disk;
705 		int indent = 0;
706 
707 		pr_debug("%s:%*s iter %d start\n", __func__, indent, "", i);
708 		indent += 2;
709 
710 		memset(page_address(page1), 0, PAGE_SIZE);
711 
712 		/* iterate through data member disks */
713 		for (disk = 0; disk < data_disks; disk++) {
714 			int dd_idx;
715 			struct md_rdev *rdev;
716 			sector_t sector;
717 			sector_t r_sector = r_sector_first + i +
718 					    (disk * conf->chunk_sectors);
719 
720 			pr_debug("%s:%*s data member disk %d start\n",
721 				 __func__, indent, "", disk);
722 			indent += 2;
723 
724 			if (r_sector >= r_sector_last) {
725 				pr_debug("%s:%*s array sector %llu doesn't need parity update\n",
726 					 __func__, indent, "",
727 					 (unsigned long long)r_sector);
728 				indent -= 2;
729 				continue;
730 			}
731 
732 			update_parity = true;
733 
734 			/* map raid sector to member disk */
735 			sector = raid5_compute_sector(conf, r_sector, 0,
736 						      &dd_idx, NULL);
737 			pr_debug("%s:%*s processing array sector %llu => data member disk %d, sector %llu\n",
738 				 __func__, indent, "",
739 				 (unsigned long long)r_sector, dd_idx,
740 				 (unsigned long long)sector);
741 
742 			rdev = conf->disks[dd_idx].rdev;
743 			if (!rdev) {
744 				pr_debug("%s:%*s data member disk %d missing\n",
745 					 __func__, indent, "", dd_idx);
746 				update_parity = false;
747 				break;
748 			}
749 
750 			pr_debug("%s:%*s reading data member disk %s sector %llu\n",
751 				 __func__, indent, "", bdevname(rdev->bdev, b),
752 				 (unsigned long long)sector);
753 			if (!sync_page_io(rdev, sector, block_size, page2,
754 					REQ_OP_READ, 0, false)) {
755 				md_error(mddev, rdev);
756 				pr_debug("%s:%*s read failed!\n", __func__,
757 					 indent, "");
758 				ret = -EIO;
759 				goto out;
760 			}
761 
762 			ppl_xor(block_size, page1, page2);
763 
764 			indent -= 2;
765 		}
766 
767 		if (!update_parity)
768 			continue;
769 
770 		if (pp_size > 0) {
771 			pr_debug("%s:%*s reading pp disk sector %llu\n",
772 				 __func__, indent, "",
773 				 (unsigned long long)(ppl_sector + i));
774 			if (!sync_page_io(log->rdev,
775 					ppl_sector - log->rdev->data_offset + i,
776 					block_size, page2, REQ_OP_READ, 0,
777 					false)) {
778 				pr_debug("%s:%*s read failed!\n", __func__,
779 					 indent, "");
780 				md_error(mddev, log->rdev);
781 				ret = -EIO;
782 				goto out;
783 			}
784 
785 			ppl_xor(block_size, page1, page2);
786 		}
787 
788 		/* map raid sector to parity disk */
789 		parity_sector = raid5_compute_sector(conf, r_sector_first + i,
790 				0, &disk, &sh);
791 		BUG_ON(sh.pd_idx != le32_to_cpu(e->parity_disk));
792 		parity_rdev = conf->disks[sh.pd_idx].rdev;
793 
794 		BUG_ON(parity_rdev->bdev->bd_dev != log->rdev->bdev->bd_dev);
795 		pr_debug("%s:%*s write parity at sector %llu, disk %s\n",
796 			 __func__, indent, "",
797 			 (unsigned long long)parity_sector,
798 			 bdevname(parity_rdev->bdev, b));
799 		if (!sync_page_io(parity_rdev, parity_sector, block_size,
800 				page1, REQ_OP_WRITE, 0, false)) {
801 			pr_debug("%s:%*s parity write error!\n", __func__,
802 				 indent, "");
803 			md_error(mddev, parity_rdev);
804 			ret = -EIO;
805 			goto out;
806 		}
807 	}
808 out:
809 	if (page1)
810 		__free_page(page1);
811 	if (page2)
812 		__free_page(page2);
813 	return ret;
814 }
815 
816 static int ppl_recover(struct ppl_log *log, struct ppl_header *pplhdr)
817 {
818 	struct ppl_conf *ppl_conf = log->ppl_conf;
819 	struct md_rdev *rdev = log->rdev;
820 	struct mddev *mddev = rdev->mddev;
821 	sector_t ppl_sector = rdev->ppl.sector + (PPL_HEADER_SIZE >> 9);
822 	struct page *page;
823 	int i;
824 	int ret = 0;
825 
826 	page = alloc_page(GFP_KERNEL);
827 	if (!page)
828 		return -ENOMEM;
829 
830 	/* iterate through all PPL entries saved */
831 	for (i = 0; i < le32_to_cpu(pplhdr->entries_count); i++) {
832 		struct ppl_header_entry *e = &pplhdr->entries[i];
833 		u32 pp_size = le32_to_cpu(e->pp_size);
834 		sector_t sector = ppl_sector;
835 		int ppl_entry_sectors = pp_size >> 9;
836 		u32 crc, crc_stored;
837 
838 		pr_debug("%s: disk: %d entry: %d ppl_sector: %llu pp_size: %u\n",
839 			 __func__, rdev->raid_disk, i,
840 			 (unsigned long long)ppl_sector, pp_size);
841 
842 		crc = ~0;
843 		crc_stored = le32_to_cpu(e->checksum);
844 
845 		/* read parial parity for this entry and calculate its checksum */
846 		while (pp_size) {
847 			int s = pp_size > PAGE_SIZE ? PAGE_SIZE : pp_size;
848 
849 			if (!sync_page_io(rdev, sector - rdev->data_offset,
850 					s, page, REQ_OP_READ, 0, false)) {
851 				md_error(mddev, rdev);
852 				ret = -EIO;
853 				goto out;
854 			}
855 
856 			crc = crc32c_le(crc, page_address(page), s);
857 
858 			pp_size -= s;
859 			sector += s >> 9;
860 		}
861 
862 		crc = ~crc;
863 
864 		if (crc != crc_stored) {
865 			/*
866 			 * Don't recover this entry if the checksum does not
867 			 * match, but keep going and try to recover other
868 			 * entries.
869 			 */
870 			pr_debug("%s: ppl entry crc does not match: stored: 0x%x calculated: 0x%x\n",
871 				 __func__, crc_stored, crc);
872 			ppl_conf->mismatch_count++;
873 		} else {
874 			ret = ppl_recover_entry(log, e, ppl_sector);
875 			if (ret)
876 				goto out;
877 			ppl_conf->recovered_entries++;
878 		}
879 
880 		ppl_sector += ppl_entry_sectors;
881 	}
882 
883 	/* flush the disk cache after recovery if necessary */
884 	ret = blkdev_issue_flush(rdev->bdev, GFP_KERNEL, NULL);
885 out:
886 	__free_page(page);
887 	return ret;
888 }
889 
890 static int ppl_write_empty_header(struct ppl_log *log)
891 {
892 	struct page *page;
893 	struct ppl_header *pplhdr;
894 	struct md_rdev *rdev = log->rdev;
895 	int ret = 0;
896 
897 	pr_debug("%s: disk: %d ppl_sector: %llu\n", __func__,
898 		 rdev->raid_disk, (unsigned long long)rdev->ppl.sector);
899 
900 	page = alloc_page(GFP_NOIO | __GFP_ZERO);
901 	if (!page)
902 		return -ENOMEM;
903 
904 	pplhdr = page_address(page);
905 	memset(pplhdr->reserved, 0xff, PPL_HDR_RESERVED);
906 	pplhdr->signature = cpu_to_le32(log->ppl_conf->signature);
907 	pplhdr->checksum = cpu_to_le32(~crc32c_le(~0, pplhdr, PAGE_SIZE));
908 
909 	if (!sync_page_io(rdev, rdev->ppl.sector - rdev->data_offset,
910 			  PPL_HEADER_SIZE, page, REQ_OP_WRITE | REQ_SYNC |
911 			  REQ_FUA, 0, false)) {
912 		md_error(rdev->mddev, rdev);
913 		ret = -EIO;
914 	}
915 
916 	__free_page(page);
917 	return ret;
918 }
919 
920 static int ppl_load_distributed(struct ppl_log *log)
921 {
922 	struct ppl_conf *ppl_conf = log->ppl_conf;
923 	struct md_rdev *rdev = log->rdev;
924 	struct mddev *mddev = rdev->mddev;
925 	struct page *page;
926 	struct ppl_header *pplhdr;
927 	u32 crc, crc_stored;
928 	u32 signature;
929 	int ret = 0;
930 
931 	pr_debug("%s: disk: %d\n", __func__, rdev->raid_disk);
932 
933 	/* read PPL header */
934 	page = alloc_page(GFP_KERNEL);
935 	if (!page)
936 		return -ENOMEM;
937 
938 	if (!sync_page_io(rdev, rdev->ppl.sector - rdev->data_offset,
939 			  PAGE_SIZE, page, REQ_OP_READ, 0, false)) {
940 		md_error(mddev, rdev);
941 		ret = -EIO;
942 		goto out;
943 	}
944 	pplhdr = page_address(page);
945 
946 	/* check header validity */
947 	crc_stored = le32_to_cpu(pplhdr->checksum);
948 	pplhdr->checksum = 0;
949 	crc = ~crc32c_le(~0, pplhdr, PAGE_SIZE);
950 
951 	if (crc_stored != crc) {
952 		pr_debug("%s: ppl header crc does not match: stored: 0x%x calculated: 0x%x\n",
953 			 __func__, crc_stored, crc);
954 		ppl_conf->mismatch_count++;
955 		goto out;
956 	}
957 
958 	signature = le32_to_cpu(pplhdr->signature);
959 
960 	if (mddev->external) {
961 		/*
962 		 * For external metadata the header signature is set and
963 		 * validated in userspace.
964 		 */
965 		ppl_conf->signature = signature;
966 	} else if (ppl_conf->signature != signature) {
967 		pr_debug("%s: ppl header signature does not match: stored: 0x%x configured: 0x%x\n",
968 			 __func__, signature, ppl_conf->signature);
969 		ppl_conf->mismatch_count++;
970 		goto out;
971 	}
972 
973 	/* attempt to recover from log if we are starting a dirty array */
974 	if (!mddev->pers && mddev->recovery_cp != MaxSector)
975 		ret = ppl_recover(log, pplhdr);
976 out:
977 	/* write empty header if we are starting the array */
978 	if (!ret && !mddev->pers)
979 		ret = ppl_write_empty_header(log);
980 
981 	__free_page(page);
982 
983 	pr_debug("%s: return: %d mismatch_count: %d recovered_entries: %d\n",
984 		 __func__, ret, ppl_conf->mismatch_count,
985 		 ppl_conf->recovered_entries);
986 	return ret;
987 }
988 
989 static int ppl_load(struct ppl_conf *ppl_conf)
990 {
991 	int ret = 0;
992 	u32 signature = 0;
993 	bool signature_set = false;
994 	int i;
995 
996 	for (i = 0; i < ppl_conf->count; i++) {
997 		struct ppl_log *log = &ppl_conf->child_logs[i];
998 
999 		/* skip missing drive */
1000 		if (!log->rdev)
1001 			continue;
1002 
1003 		ret = ppl_load_distributed(log);
1004 		if (ret)
1005 			break;
1006 
1007 		/*
1008 		 * For external metadata we can't check if the signature is
1009 		 * correct on a single drive, but we can check if it is the same
1010 		 * on all drives.
1011 		 */
1012 		if (ppl_conf->mddev->external) {
1013 			if (!signature_set) {
1014 				signature = ppl_conf->signature;
1015 				signature_set = true;
1016 			} else if (signature != ppl_conf->signature) {
1017 				pr_warn("md/raid:%s: PPL header signature does not match on all member drives\n",
1018 					mdname(ppl_conf->mddev));
1019 				ret = -EINVAL;
1020 				break;
1021 			}
1022 		}
1023 	}
1024 
1025 	pr_debug("%s: return: %d mismatch_count: %d recovered_entries: %d\n",
1026 		 __func__, ret, ppl_conf->mismatch_count,
1027 		 ppl_conf->recovered_entries);
1028 	return ret;
1029 }
1030 
1031 static void __ppl_exit_log(struct ppl_conf *ppl_conf)
1032 {
1033 	clear_bit(MD_HAS_PPL, &ppl_conf->mddev->flags);
1034 
1035 	kfree(ppl_conf->child_logs);
1036 
1037 	if (ppl_conf->bs)
1038 		bioset_free(ppl_conf->bs);
1039 	mempool_destroy(ppl_conf->io_pool);
1040 	kmem_cache_destroy(ppl_conf->io_kc);
1041 
1042 	kfree(ppl_conf);
1043 }
1044 
1045 void ppl_exit_log(struct r5conf *conf)
1046 {
1047 	struct ppl_conf *ppl_conf = conf->log_private;
1048 
1049 	if (ppl_conf) {
1050 		__ppl_exit_log(ppl_conf);
1051 		conf->log_private = NULL;
1052 	}
1053 }
1054 
1055 static int ppl_validate_rdev(struct md_rdev *rdev)
1056 {
1057 	char b[BDEVNAME_SIZE];
1058 	int ppl_data_sectors;
1059 	int ppl_size_new;
1060 
1061 	/*
1062 	 * The configured PPL size must be enough to store
1063 	 * the header and (at the very least) partial parity
1064 	 * for one stripe. Round it down to ensure the data
1065 	 * space is cleanly divisible by stripe size.
1066 	 */
1067 	ppl_data_sectors = rdev->ppl.size - (PPL_HEADER_SIZE >> 9);
1068 
1069 	if (ppl_data_sectors > 0)
1070 		ppl_data_sectors = rounddown(ppl_data_sectors, STRIPE_SECTORS);
1071 
1072 	if (ppl_data_sectors <= 0) {
1073 		pr_warn("md/raid:%s: PPL space too small on %s\n",
1074 			mdname(rdev->mddev), bdevname(rdev->bdev, b));
1075 		return -ENOSPC;
1076 	}
1077 
1078 	ppl_size_new = ppl_data_sectors + (PPL_HEADER_SIZE >> 9);
1079 
1080 	if ((rdev->ppl.sector < rdev->data_offset &&
1081 	     rdev->ppl.sector + ppl_size_new > rdev->data_offset) ||
1082 	    (rdev->ppl.sector >= rdev->data_offset &&
1083 	     rdev->data_offset + rdev->sectors > rdev->ppl.sector)) {
1084 		pr_warn("md/raid:%s: PPL space overlaps with data on %s\n",
1085 			mdname(rdev->mddev), bdevname(rdev->bdev, b));
1086 		return -EINVAL;
1087 	}
1088 
1089 	if (!rdev->mddev->external &&
1090 	    ((rdev->ppl.offset > 0 && rdev->ppl.offset < (rdev->sb_size >> 9)) ||
1091 	     (rdev->ppl.offset <= 0 && rdev->ppl.offset + ppl_size_new > 0))) {
1092 		pr_warn("md/raid:%s: PPL space overlaps with superblock on %s\n",
1093 			mdname(rdev->mddev), bdevname(rdev->bdev, b));
1094 		return -EINVAL;
1095 	}
1096 
1097 	rdev->ppl.size = ppl_size_new;
1098 
1099 	return 0;
1100 }
1101 
1102 int ppl_init_log(struct r5conf *conf)
1103 {
1104 	struct ppl_conf *ppl_conf;
1105 	struct mddev *mddev = conf->mddev;
1106 	int ret = 0;
1107 	int i;
1108 	bool need_cache_flush = false;
1109 
1110 	pr_debug("md/raid:%s: enabling distributed Partial Parity Log\n",
1111 		 mdname(conf->mddev));
1112 
1113 	if (PAGE_SIZE != 4096)
1114 		return -EINVAL;
1115 
1116 	if (mddev->level != 5) {
1117 		pr_warn("md/raid:%s PPL is not compatible with raid level %d\n",
1118 			mdname(mddev), mddev->level);
1119 		return -EINVAL;
1120 	}
1121 
1122 	if (mddev->bitmap_info.file || mddev->bitmap_info.offset) {
1123 		pr_warn("md/raid:%s PPL is not compatible with bitmap\n",
1124 			mdname(mddev));
1125 		return -EINVAL;
1126 	}
1127 
1128 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
1129 		pr_warn("md/raid:%s PPL is not compatible with journal\n",
1130 			mdname(mddev));
1131 		return -EINVAL;
1132 	}
1133 
1134 	ppl_conf = kzalloc(sizeof(struct ppl_conf), GFP_KERNEL);
1135 	if (!ppl_conf)
1136 		return -ENOMEM;
1137 
1138 	ppl_conf->mddev = mddev;
1139 
1140 	ppl_conf->io_kc = KMEM_CACHE(ppl_io_unit, 0);
1141 	if (!ppl_conf->io_kc) {
1142 		ret = -ENOMEM;
1143 		goto err;
1144 	}
1145 
1146 	ppl_conf->io_pool = mempool_create(conf->raid_disks, ppl_io_pool_alloc,
1147 					   ppl_io_pool_free, ppl_conf->io_kc);
1148 	if (!ppl_conf->io_pool) {
1149 		ret = -ENOMEM;
1150 		goto err;
1151 	}
1152 
1153 	ppl_conf->bs = bioset_create(conf->raid_disks, 0);
1154 	if (!ppl_conf->bs) {
1155 		ret = -ENOMEM;
1156 		goto err;
1157 	}
1158 
1159 	ppl_conf->count = conf->raid_disks;
1160 	ppl_conf->child_logs = kcalloc(ppl_conf->count, sizeof(struct ppl_log),
1161 				       GFP_KERNEL);
1162 	if (!ppl_conf->child_logs) {
1163 		ret = -ENOMEM;
1164 		goto err;
1165 	}
1166 
1167 	atomic64_set(&ppl_conf->seq, 0);
1168 	INIT_LIST_HEAD(&ppl_conf->no_mem_stripes);
1169 	spin_lock_init(&ppl_conf->no_mem_stripes_lock);
1170 
1171 	if (!mddev->external) {
1172 		ppl_conf->signature = ~crc32c_le(~0, mddev->uuid, sizeof(mddev->uuid));
1173 		ppl_conf->block_size = 512;
1174 	} else {
1175 		ppl_conf->block_size = queue_logical_block_size(mddev->queue);
1176 	}
1177 
1178 	for (i = 0; i < ppl_conf->count; i++) {
1179 		struct ppl_log *log = &ppl_conf->child_logs[i];
1180 		struct md_rdev *rdev = conf->disks[i].rdev;
1181 
1182 		mutex_init(&log->io_mutex);
1183 		spin_lock_init(&log->io_list_lock);
1184 		INIT_LIST_HEAD(&log->io_list);
1185 
1186 		log->ppl_conf = ppl_conf;
1187 		log->rdev = rdev;
1188 
1189 		if (rdev) {
1190 			struct request_queue *q;
1191 
1192 			ret = ppl_validate_rdev(rdev);
1193 			if (ret)
1194 				goto err;
1195 
1196 			q = bdev_get_queue(rdev->bdev);
1197 			if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
1198 				need_cache_flush = true;
1199 		}
1200 	}
1201 
1202 	if (need_cache_flush)
1203 		pr_warn("md/raid:%s: Volatile write-back cache should be disabled on all member drives when using PPL!\n",
1204 			mdname(mddev));
1205 
1206 	/* load and possibly recover the logs from the member disks */
1207 	ret = ppl_load(ppl_conf);
1208 
1209 	if (ret) {
1210 		goto err;
1211 	} else if (!mddev->pers &&
1212 		   mddev->recovery_cp == 0 && !mddev->degraded &&
1213 		   ppl_conf->recovered_entries > 0 &&
1214 		   ppl_conf->mismatch_count == 0) {
1215 		/*
1216 		 * If we are starting a dirty array and the recovery succeeds
1217 		 * without any issues, set the array as clean.
1218 		 */
1219 		mddev->recovery_cp = MaxSector;
1220 		set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
1221 	} else if (mddev->pers && ppl_conf->mismatch_count > 0) {
1222 		/* no mismatch allowed when enabling PPL for a running array */
1223 		ret = -EINVAL;
1224 		goto err;
1225 	}
1226 
1227 	conf->log_private = ppl_conf;
1228 	set_bit(MD_HAS_PPL, &ppl_conf->mddev->flags);
1229 
1230 	return 0;
1231 err:
1232 	__ppl_exit_log(ppl_conf);
1233 	return ret;
1234 }
1235 
1236 int ppl_modify_log(struct r5conf *conf, struct md_rdev *rdev, bool add)
1237 {
1238 	struct ppl_conf *ppl_conf = conf->log_private;
1239 	struct ppl_log *log;
1240 	int ret = 0;
1241 	char b[BDEVNAME_SIZE];
1242 
1243 	if (!rdev)
1244 		return -EINVAL;
1245 
1246 	pr_debug("%s: disk: %d operation: %s dev: %s\n",
1247 		 __func__, rdev->raid_disk, add ? "add" : "remove",
1248 		 bdevname(rdev->bdev, b));
1249 
1250 	if (rdev->raid_disk < 0)
1251 		return 0;
1252 
1253 	if (rdev->raid_disk >= ppl_conf->count)
1254 		return -ENODEV;
1255 
1256 	log = &ppl_conf->child_logs[rdev->raid_disk];
1257 
1258 	mutex_lock(&log->io_mutex);
1259 	if (add) {
1260 		ret = ppl_validate_rdev(rdev);
1261 		if (!ret) {
1262 			log->rdev = rdev;
1263 			ret = ppl_write_empty_header(log);
1264 		}
1265 	} else {
1266 		log->rdev = NULL;
1267 	}
1268 	mutex_unlock(&log->io_mutex);
1269 
1270 	return ret;
1271 }
1272