forked from ~ljy/RK356X_SDK_RELEASE

blame | history | patch | commit | commitdiff
hc
2024-01-31 f9004dbfff8a3fbbd7e2a88c8a4327c7f2f8e5b2 
 kernel/drivers/md/md-cluster.c
....@@ -1,11 +1,6 @@
1
+// SPDX-License-Identifier: GPL-2.0-or-later
12 /*
23  * Copyright (C) 2015, SUSE
3
- *
4
- * This program is free software; you can redistribute it and/or modify
5
- * it under the terms of the GNU General Public License as published by
6
- * the Free Software Foundation; either version 2, or (at your option)
7
- * any later version.
8
- *
94  */
105 
116 
....@@ -31,13 +26,6 @@
3126 	void (*bast)(void *arg, int mode); /* blocking AST function pointer*/
3227 	struct mddev *mddev; /* pointing back to mddev. */
3328 	int mode;
34
-};
35
-
36
-struct suspend_info {
37
-	int slot;
38
-	sector_t lo;
39
-	sector_t hi;
40
-	struct list_head list;
4129 };
4230 
4331 struct resync_info {
....@@ -80,7 +68,13 @@
8068 	struct dlm_lock_resource **other_bitmap_lockres;
8169 	struct dlm_lock_resource *resync_lockres;
8270 	struct list_head suspend_list;
71
+
8372 	spinlock_t suspend_lock;
73
+	/* record the region which write should be suspended */
74
+	sector_t suspend_lo;
75
+	sector_t suspend_hi;
76
+	int suspend_from; /* the slot which broadcast suspend_lo/hi */
77
+
8478 	struct md_thread *recovery_thread;
8579 	unsigned long recovery_map;
8680 	/* communication loc resources */
....@@ -105,6 +99,7 @@
10599 	RE_ADD,
106100 	BITMAP_NEEDS_SYNC,
107101 	CHANGE_CAPACITY,
102
+	BITMAP_RESIZE,
108103 };
109104 
110105 struct cluster_msg {
....@@ -270,25 +265,22 @@
270265 	ri->hi = cpu_to_le64(hi);
271266 }
272267 
273
-static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres)
268
+static int read_resync_info(struct mddev *mddev,
269
+			    struct dlm_lock_resource *lockres)
274270 {
275271 	struct resync_info ri;
276
-	struct suspend_info *s = NULL;
277
-	sector_t hi = 0;
272
+	struct md_cluster_info *cinfo = mddev->cluster_info;
273
+	int ret = 0;
278274 
279275 	dlm_lock_sync(lockres, DLM_LOCK_CR);
280276 	memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
281
-	hi = le64_to_cpu(ri.hi);
282
-	if (hi > 0) {
283
-		s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
284
-		if (!s)
285
-			goto out;
286
-		s->hi = hi;
287
-		s->lo = le64_to_cpu(ri.lo);
277
+	if (le64_to_cpu(ri.hi) > 0) {
278
+		cinfo->suspend_hi = le64_to_cpu(ri.hi);
279
+		cinfo->suspend_lo = le64_to_cpu(ri.lo);
280
+		ret = 1;
288281 	}
289282 	dlm_unlock_sync(lockres);
290
-out:
291
-	return s;
283
+	return ret;
292284 }
293285 
294286 static void recover_bitmaps(struct md_thread *thread)
....@@ -298,7 +290,6 @@
298290 	struct dlm_lock_resource *bm_lockres;
299291 	char str[64];
300292 	int slot, ret;
301
-	struct suspend_info *s, *tmp;
302293 	sector_t lo, hi;
303294 
304295 	while (cinfo->recovery_map) {
....@@ -325,12 +316,16 @@
325316 
326317 		/* Clear suspend_area associated with the bitmap */
327318 		spin_lock_irq(&cinfo->suspend_lock);
328
-		list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
329
-			if (slot == s->slot) {
330
-				list_del(&s->list);
331
-				kfree(s);
332
-			}
319
+		cinfo->suspend_hi = 0;
320
+		cinfo->suspend_lo = 0;
321
+		cinfo->suspend_from = -1;
333322 		spin_unlock_irq(&cinfo->suspend_lock);
323
+
324
+		/* Kick off a reshape if needed */
325
+		if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
326
+		    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
327
+		    mddev->reshape_position != MaxSector)
328
+			md_wakeup_thread(mddev->sync_thread);
334329 
335330 		if (hi > 0) {
336331 			if (lo < mddev->recovery_cp)
....@@ -434,34 +429,23 @@
434429 	}
435430 }
436431 
437
-static void __remove_suspend_info(struct md_cluster_info *cinfo, int slot)
438
-{
439
-	struct suspend_info *s, *tmp;
440
-
441
-	list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
442
-		if (slot == s->slot) {
443
-			list_del(&s->list);
444
-			kfree(s);
445
-			break;
446
-		}
447
-}
448
-
449432 static void remove_suspend_info(struct mddev *mddev, int slot)
450433 {
451434 	struct md_cluster_info *cinfo = mddev->cluster_info;
452435 	mddev->pers->quiesce(mddev, 1);
453436 	spin_lock_irq(&cinfo->suspend_lock);
454
-	__remove_suspend_info(cinfo, slot);
437
+	cinfo->suspend_hi = 0;
438
+	cinfo->suspend_lo = 0;
455439 	spin_unlock_irq(&cinfo->suspend_lock);
456440 	mddev->pers->quiesce(mddev, 0);
457441 }
458
-
459442 
460443 static void process_suspend_info(struct mddev *mddev,
461444 		int slot, sector_t lo, sector_t hi)
462445 {
463446 	struct md_cluster_info *cinfo = mddev->cluster_info;
464
-	struct suspend_info *s;
447
+	struct mdp_superblock_1 *sb = NULL;
448
+	struct md_rdev *rdev;
465449 
466450 	if (!hi) {
467451 		/*
....@@ -475,6 +459,12 @@
475459 		return;
476460 	}
477461 
462
+	rdev_for_each(rdev, mddev)
463
+		if (rdev->raid_disk > -1 && !test_bit(Faulty, &rdev->flags)) {
464
+			sb = page_address(rdev->sb_page);
465
+			break;
466
+		}
467
+
478468 	/*
479469 	 * The bitmaps are not same for different nodes
480470 	 * if RESYNCING is happening in one node, then
....@@ -487,26 +477,26 @@
487477 	 * sync_low/hi is used to record the region which
488478 	 * arrived in the previous RESYNCING message,
489479 	 *
490
-	 * Call bitmap_sync_with_cluster to clear
491
-	 * NEEDED_MASK and set RESYNC_MASK since
492
-	 * resync thread is running in another node,
493
-	 * so we don't need to do the resync again
494
-	 * with the same section */
495
-	md_bitmap_sync_with_cluster(mddev, cinfo->sync_low, cinfo->sync_hi, lo, hi);
480
+	 * Call md_bitmap_sync_with_cluster to clear NEEDED_MASK
481
+	 * and set RESYNC_MASK since  resync thread is running
482
+	 * in another node, so we don't need to do the resync
483
+	 * again with the same section.
484
+	 *
485
+	 * Skip md_bitmap_sync_with_cluster in case reshape
486
+	 * happening, because reshaping region is small and
487
+	 * we don't want to trigger lots of WARN.
488
+	 */
489
+	if (sb && !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE))
490
+		md_bitmap_sync_with_cluster(mddev, cinfo->sync_low,
491
+					    cinfo->sync_hi, lo, hi);
496492 	cinfo->sync_low = lo;
497493 	cinfo->sync_hi = hi;
498494 
499
-	s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
500
-	if (!s)
501
-		return;
502
-	s->slot = slot;
503
-	s->lo = lo;
504
-	s->hi = hi;
505495 	mddev->pers->quiesce(mddev, 1);
506496 	spin_lock_irq(&cinfo->suspend_lock);
507
-	/* Remove existing entry (if exists) before adding */
508
-	__remove_suspend_info(cinfo, slot);
509
-	list_add(&s->list, &cinfo->suspend_list);
497
+	cinfo->suspend_from = slot;
498
+	cinfo->suspend_lo = lo;
499
+	cinfo->suspend_hi = hi;
510500 	spin_unlock_irq(&cinfo->suspend_lock);
511501 	mddev->pers->quiesce(mddev, 0);
512502 }
....@@ -592,7 +582,7 @@
592582 		break;
593583 	case CHANGE_CAPACITY:
594584 		set_capacity(mddev->gendisk, mddev->array_sectors);
595
-		revalidate_disk(mddev->gendisk);
585
+		revalidate_disk_size(mddev->gendisk, true);
596586 		break;
597587 	case RESYNCING:
598588 		set_bit(MD_RESYNCING_REMOTE, &mddev->recovery);
....@@ -611,6 +601,11 @@
611601 		break;
612602 	case BITMAP_NEEDS_SYNC:
613603 		__recover_slot(mddev, le32_to_cpu(msg->slot));
604
+		break;
605
+	case BITMAP_RESIZE:
606
+		if (le64_to_cpu(msg->high) != mddev->pers->size(mddev, 0, 0))
607
+			ret = md_bitmap_resize(mddev->bitmap,
608
+					    le64_to_cpu(msg->high), 0, 0);
614609 		break;
615610 	default:
616611 		ret = -1;
....@@ -805,7 +800,6 @@
805800 	struct md_cluster_info *cinfo = mddev->cluster_info;
806801 	int i, ret = 0;
807802 	struct dlm_lock_resource *bm_lockres;
808
-	struct suspend_info *s;
809803 	char str[64];
810804 	sector_t lo, hi;
811805 
....@@ -824,16 +818,13 @@
824818 		bm_lockres->flags |= DLM_LKF_NOQUEUE;
825819 		ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
826820 		if (ret == -EAGAIN) {
827
-			s = read_resync_info(mddev, bm_lockres);
828
-			if (s) {
821
+			if (read_resync_info(mddev, bm_lockres)) {
829822 				pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n",
830823 						__func__, __LINE__,
831
-						(unsigned long long) s->lo,
832
-						(unsigned long long) s->hi, i);
833
-				spin_lock_irq(&cinfo->suspend_lock);
834
-				s->slot = i;
835
-				list_add(&s->list, &cinfo->suspend_list);
836
-				spin_unlock_irq(&cinfo->suspend_lock);
824
+					(unsigned long long) cinfo->suspend_lo,
825
+					(unsigned long long) cinfo->suspend_hi,
826
+					i);
827
+				cinfo->suspend_from = i;
837828 			}
838829 			ret = 0;
839830 			lockres_free(bm_lockres);
....@@ -1006,10 +997,17 @@
1006997 	if (!cinfo)
1007998 		return 0;
1008999 
1009
-	/* BITMAP_NEEDS_SYNC message should be sent when node
1000
+	/*
1001
+	 * BITMAP_NEEDS_SYNC message should be sent when node
10101002 	 * is leaving the cluster with dirty bitmap, also we
1011
-	 * can only deliver it when dlm connection is available */
1012
-	if (cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector)
1003
+	 * can only deliver it when dlm connection is available.
1004
+	 *
1005
+	 * Also, we should send BITMAP_NEEDS_SYNC message in
1006
+	 * case reshaping is interrupted.
1007
+	 */
1008
+	if ((cinfo->slot_number > 0 && mddev->recovery_cp != MaxSector) ||
1009
+	    (mddev->reshape_position != MaxSector &&
1010
+	     test_bit(MD_CLOSING, &mddev->flags)))
10131011 		resync_bitmap(mddev);
10141012 
10151013 	set_bit(MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD, &cinfo->state);
....@@ -1105,6 +1103,82 @@
11051103 	struct md_cluster_info *cinfo = mddev->cluster_info;
11061104 	clear_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
11071105 	unlock_comm(cinfo);
1106
+}
1107
+
1108
+static int update_bitmap_size(struct mddev *mddev, sector_t size)
1109
+{
1110
+	struct md_cluster_info *cinfo = mddev->cluster_info;
1111
+	struct cluster_msg cmsg = {0};
1112
+	int ret;
1113
+
1114
+	cmsg.type = cpu_to_le32(BITMAP_RESIZE);
1115
+	cmsg.high = cpu_to_le64(size);
1116
+	ret = sendmsg(cinfo, &cmsg, 0);
1117
+	if (ret)
1118
+		pr_err("%s:%d: failed to send BITMAP_RESIZE message (%d)\n",
1119
+			__func__, __LINE__, ret);
1120
+	return ret;
1121
+}
1122
+
1123
+static int resize_bitmaps(struct mddev *mddev, sector_t newsize, sector_t oldsize)
1124
+{
1125
+	struct bitmap_counts *counts;
1126
+	char str[64];
1127
+	struct dlm_lock_resource *bm_lockres;
1128
+	struct bitmap *bitmap = mddev->bitmap;
1129
+	unsigned long my_pages = bitmap->counts.pages;
1130
+	int i, rv;
1131
+
1132
+	/*
1133
+	 * We need to ensure all the nodes can grow to a larger
1134
+	 * bitmap size before make the reshaping.
1135
+	 */
1136
+	rv = update_bitmap_size(mddev, newsize);
1137
+	if (rv)
1138
+		return rv;
1139
+
1140
+	for (i = 0; i < mddev->bitmap_info.nodes; i++) {
1141
+		if (i == md_cluster_ops->slot_number(mddev))
1142
+			continue;
1143
+
1144
+		bitmap = get_bitmap_from_slot(mddev, i);
1145
+		if (IS_ERR(bitmap)) {
1146
+			pr_err("can't get bitmap from slot %d\n", i);
1147
+			bitmap = NULL;
1148
+			goto out;
1149
+		}
1150
+		counts = &bitmap->counts;
1151
+
1152
+		/*
1153
+		 * If we can hold the bitmap lock of one node then
1154
+		 * the slot is not occupied, update the pages.
1155
+		 */
1156
+		snprintf(str, 64, "bitmap%04d", i);
1157
+		bm_lockres = lockres_init(mddev, str, NULL, 1);
1158
+		if (!bm_lockres) {
1159
+			pr_err("Cannot initialize %s lock\n", str);
1160
+			goto out;
1161
+		}
1162
+		bm_lockres->flags |= DLM_LKF_NOQUEUE;
1163
+		rv = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
1164
+		if (!rv)
1165
+			counts->pages = my_pages;
1166
+		lockres_free(bm_lockres);
1167
+
1168
+		if (my_pages != counts->pages)
1169
+			/*
1170
+			 * Let's revert the bitmap size if one node
1171
+			 * can't resize bitmap
1172
+			 */
1173
+			goto out;
1174
+		md_bitmap_free(bitmap);
1175
+	}
1176
+
1177
+	return 0;
1178
+out:
1179
+	md_bitmap_free(bitmap);
1180
+	update_bitmap_size(mddev, oldsize);
1181
+	return -1;
11081182 }
11091183 
11101184 /*
....@@ -1231,12 +1305,12 @@
12311305 			pr_err("%s:%d: failed to send CHANGE_CAPACITY msg\n",
12321306 			       __func__, __LINE__);
12331307 		set_capacity(mddev->gendisk, mddev->array_sectors);
1234
-		revalidate_disk(mddev->gendisk);
1308
+		revalidate_disk_size(mddev->gendisk, true);
12351309 	} else {
12361310 		/* revert to previous sectors */
12371311 		ret = mddev->pers->resize(mddev, old_dev_sectors);
12381312 		if (!ret)
1239
-			revalidate_disk(mddev->gendisk);
1313
+			revalidate_disk_size(mddev->gendisk, true);
12401314 		ret = __sendmsg(cinfo, &cmsg);
12411315 		if (ret)
12421316 			pr_err("%s:%d: failed to send METADATA_UPDATED msg\n",
....@@ -1249,6 +1323,16 @@
12491323 {
12501324 	struct md_cluster_info *cinfo = mddev->cluster_info;
12511325 	return dlm_lock_sync_interruptible(cinfo->resync_lockres, DLM_LOCK_EX, mddev);
1326
+}
1327
+
1328
+static void resync_info_get(struct mddev *mddev, sector_t *lo, sector_t *hi)
1329
+{
1330
+	struct md_cluster_info *cinfo = mddev->cluster_info;
1331
+
1332
+	spin_lock_irq(&cinfo->suspend_lock);
1333
+	*lo = cinfo->suspend_lo;
1334
+	*hi = cinfo->suspend_hi;
1335
+	spin_unlock_irq(&cinfo->suspend_lock);
12521336 }
12531337 
12541338 static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi)
....@@ -1303,21 +1387,14 @@
13031387 {
13041388 	struct md_cluster_info *cinfo = mddev->cluster_info;
13051389 	int ret = 0;
1306
-	struct suspend_info *s;
13071390 
13081391 	if ((direction == READ) &&
13091392 		test_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state))
13101393 		return 1;
13111394 
13121395 	spin_lock_irq(&cinfo->suspend_lock);
1313
-	if (list_empty(&cinfo->suspend_list))
1314
-		goto out;
1315
-	list_for_each_entry(s, &cinfo->suspend_list, list)
1316
-		if (hi > s->lo && lo < s->hi) {
1317
-			ret = 1;
1318
-			break;
1319
-		}
1320
-out:
1396
+	if (hi > cinfo->suspend_lo && lo < cinfo->suspend_hi)
1397
+		ret = 1;
13211398 	spin_unlock_irq(&cinfo->suspend_lock);
13221399 	return ret;
13231400 }
....@@ -1492,6 +1569,7 @@
14921569 	.resync_start = resync_start,
14931570 	.resync_finish = resync_finish,
14941571 	.resync_info_update = resync_info_update,
1572
+	.resync_info_get = resync_info_get,
14951573 	.metadata_update_start = metadata_update_start,
14961574 	.metadata_update_finish = metadata_update_finish,
14971575 	.metadata_update_cancel = metadata_update_cancel,
....@@ -1502,6 +1580,7 @@
15021580 	.remove_disk = remove_disk,
15031581 	.load_bitmaps = load_bitmaps,
15041582 	.gather_bitmaps = gather_bitmaps,
1583
+	.resize_bitmaps = resize_bitmaps,
15051584 	.lock_all_bitmaps = lock_all_bitmaps,
15061585 	.unlock_all_bitmaps = unlock_all_bitmaps,
15071586 	.update_size = update_size,