summaryrefslogtreecommitdiff
path: root/libbcache/extents.c
blob: 76b55f64813d8acafe73e4d9286454ed96dfe2a0 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
/*
 * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
 *
 * Code for managing the extent btree and dynamically updating the writeback
 * dirty sector count.
 */

#include "bcache.h"
#include "bkey_methods.h"
#include "btree_gc.h"
#include "btree_update.h"
#include "checksum.h"
#include "debug.h"
#include "dirent.h"
#include "error.h"
#include "extents.h"
#include "inode.h"
#include "journal.h"
#include "super-io.h"
#include "writeback.h"
#include "xattr.h"

#include <trace/events/bcache.h>

static enum merge_result bch_extent_merge(struct bch_fs *, struct btree *,
					  struct bkey_i *, struct bkey_i *);

static void sort_key_next(struct btree_node_iter *iter,
			  struct btree *b,
			  struct btree_node_iter_set *i)
{
	i->k += __btree_node_offset_to_key(b, i->k)->u64s;

	if (i->k == i->end)
		*i = iter->data[--iter->used];
}

/*
 * Returns true if l > r - unless l == r, in which case returns true if l is
 * older than r.
 *
 * Necessary for btree_sort_fixup() - if there are multiple keys that compare
 * equal in different sets, we have to process them newest to oldest.
 */
#define key_sort_cmp(l, r)						\
({									\
	int _c = bkey_cmp_packed(b,					\
				 __btree_node_offset_to_key(b, (l).k),	\
				 __btree_node_offset_to_key(b, (r).k));	\
									\
	_c ? _c > 0 : (l).k > (r).k;					\
})

static inline bool should_drop_next_key(struct btree_node_iter *iter,
					struct btree *b)
{
	struct btree_node_iter_set *l = iter->data, *r = iter->data + 1;
	struct bkey_packed *k = __btree_node_offset_to_key(b, l->k);

	if (bkey_whiteout(k))
		return true;

	if (iter->used < 2)
		return false;

	if (iter->used > 2 &&
	    key_sort_cmp(r[0], r[1]))
		r++;

	/*
	 * key_sort_cmp() ensures that when keys compare equal the older key
	 * comes first; so if l->k compares equal to r->k then l->k is older and
	 * should be dropped.
	 */
	return !bkey_cmp_packed(b,
				__btree_node_offset_to_key(b, l->k),
				__btree_node_offset_to_key(b, r->k));
}

struct btree_nr_keys bch_key_sort_fix_overlapping(struct bset *dst,
						  struct btree *b,
						  struct btree_node_iter *iter)
{
	struct bkey_packed *out = dst->start;
	struct btree_nr_keys nr;

	memset(&nr, 0, sizeof(nr));

	heap_resort(iter, key_sort_cmp);

	while (!bch_btree_node_iter_end(iter)) {
		if (!should_drop_next_key(iter, b)) {
			struct bkey_packed *k =
				__btree_node_offset_to_key(b, iter->data->k);

			bkey_copy(out, k);
			btree_keys_account_key_add(&nr, 0, out);
			out = bkey_next(out);
		}

		sort_key_next(iter, b, iter->data);
		heap_sift(iter, 0, key_sort_cmp);
	}

	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
	return nr;
}

/* Common among btree and extent ptrs */

const struct bch_extent_ptr *
bch_extent_has_device(struct bkey_s_c_extent e, unsigned dev)
{
	const struct bch_extent_ptr *ptr;

	extent_for_each_ptr(e, ptr)
		if (ptr->dev == dev)
			return ptr;

	return NULL;
}

unsigned bch_extent_nr_ptrs(struct bkey_s_c_extent e)
{
	const struct bch_extent_ptr *ptr;
	unsigned nr_ptrs = 0;

	extent_for_each_ptr(e, ptr)
		nr_ptrs++;

	return nr_ptrs;
}

unsigned bch_extent_nr_dirty_ptrs(struct bkey_s_c k)
{
	struct bkey_s_c_extent e;
	const struct bch_extent_ptr *ptr;
	unsigned nr_ptrs = 0;

	switch (k.k->type) {
	case BCH_EXTENT:
	case BCH_EXTENT_CACHED:
		e = bkey_s_c_to_extent(k);

		extent_for_each_ptr(e, ptr)
			nr_ptrs += !ptr->cached;
		break;

	case BCH_RESERVATION:
		nr_ptrs = bkey_s_c_to_reservation(k).v->nr_replicas;
		break;
	}

	return nr_ptrs;
}

/* returns true if equal */
static bool crc_cmp(union bch_extent_crc *l, union bch_extent_crc *r)
{
	return extent_crc_type(l) == extent_crc_type(r) &&
		!memcmp(l, r, extent_entry_bytes(to_entry(l)));
}

/* Increment pointers after @crc by crc's offset until the next crc entry: */
void bch_extent_crc_narrow_pointers(struct bkey_s_extent e, union bch_extent_crc *crc)
{
	union bch_extent_entry *entry;

	extent_for_each_entry_from(e, entry, extent_entry_next(to_entry(crc))) {
		if (!extent_entry_is_ptr(entry))
			return;

		entry->ptr.offset += crc_offset(crc);
	}
}

/*
 * We're writing another replica for this extent, so while we've got the data in
 * memory we'll be computing a new checksum for the currently live data.
 *
 * If there are other replicas we aren't moving, and they are checksummed but
 * not compressed, we can modify them to point to only the data that is
 * currently live (so that readers won't have to bounce) while we've got the
 * checksum we need:
 *
 * XXX: to guard against data being corrupted while in memory, instead of
 * recomputing the checksum here, it would be better in the read path to instead
 * of computing the checksum of the entire extent:
 *
 * | extent                              |
 *
 * compute the checksums of the live and dead data separately
 * | dead data || live data || dead data |
 *
 * and then verify that crc_dead1 + crc_live + crc_dead2 == orig_crc, and then
 * use crc_live here (that we verified was correct earlier)
 *
 * note: doesn't work with encryption
 */
void bch_extent_narrow_crcs(struct bkey_s_extent e)
{
	union bch_extent_crc *crc;
	bool have_wide = false, have_narrow = false;
	struct bch_csum csum = { 0 };
	unsigned csum_type = 0;

	extent_for_each_crc(e, crc) {
		if (crc_compression_type(crc) ||
		    bch_csum_type_is_encryption(crc_csum_type(crc)))
			continue;

		if (crc_uncompressed_size(e.k, crc) != e.k->size) {
			have_wide = true;
		} else {
			have_narrow = true;
			csum = crc_csum(crc);
			csum_type = crc_csum_type(crc);
		}
	}

	if (!have_wide || !have_narrow)
		return;

	extent_for_each_crc(e, crc) {
		if (crc_compression_type(crc))
			continue;

		if (crc_uncompressed_size(e.k, crc) != e.k->size) {
			switch (extent_crc_type(crc)) {
			case BCH_EXTENT_CRC_NONE:
				BUG();
			case BCH_EXTENT_CRC32:
				if (bch_crc_bytes[csum_type] > 4)
					continue;

				bch_extent_crc_narrow_pointers(e, crc);
				crc->crc32._compressed_size	= e.k->size - 1;
				crc->crc32._uncompressed_size	= e.k->size - 1;
				crc->crc32.offset		= 0;
				crc->crc32.csum_type		= csum_type;
				crc->crc32.csum			= csum.lo;
				break;
			case BCH_EXTENT_CRC64:
				if (bch_crc_bytes[csum_type] > 10)
					continue;

				bch_extent_crc_narrow_pointers(e, crc);
				crc->crc64._compressed_size	= e.k->size - 1;
				crc->crc64._uncompressed_size	= e.k->size - 1;
				crc->crc64.offset		= 0;
				crc->crc64.csum_type		= csum_type;
				crc->crc64.csum_lo		= csum.lo;
				crc->crc64.csum_hi		= csum.hi;
				break;
			case BCH_EXTENT_CRC128:
				if (bch_crc_bytes[csum_type] > 16)
					continue;

				bch_extent_crc_narrow_pointers(e, crc);
				crc->crc128._compressed_size	= e.k->size - 1;
				crc->crc128._uncompressed_size	= e.k->size - 1;
				crc->crc128.offset		= 0;
				crc->crc128.csum_type		= csum_type;
				crc->crc128.csum		= csum;
				break;
			}
		}
	}
}

void bch_extent_drop_redundant_crcs(struct bkey_s_extent e)
{
	union bch_extent_entry *entry = e.v->start;
	union bch_extent_crc *crc, *prev = NULL;

	while (entry != extent_entry_last(e)) {
		union bch_extent_entry *next = extent_entry_next(entry);
		size_t crc_u64s = extent_entry_u64s(entry);

		if (!extent_entry_is_crc(entry))
			goto next;

		crc = entry_to_crc(entry);

		if (next == extent_entry_last(e)) {
			/* crc entry with no pointers after it: */
			goto drop;
		}

		if (extent_entry_is_crc(next)) {
			/* no pointers before next crc entry: */
			goto drop;
		}

		if (prev && crc_cmp(crc, prev)) {
			/* identical to previous crc entry: */
			goto drop;
		}

		if (!prev &&
		    !crc_csum_type(crc) &&
		    !crc_compression_type(crc)) {
			/* null crc entry: */
			bch_extent_crc_narrow_pointers(e, crc);
			goto drop;
		}

		prev = crc;
next:
		entry = next;
		continue;
drop:
		memmove_u64s_down(crc, next,
				  (u64 *) extent_entry_last(e) - (u64 *) next);
		e.k->u64s -= crc_u64s;
	}

	EBUG_ON(bkey_val_u64s(e.k) && !bch_extent_nr_ptrs(e.c));
}

static bool should_drop_ptr(const struct bch_fs *c,
			    struct bkey_s_c_extent e,
			    const struct bch_extent_ptr *ptr)
{
	struct bch_dev *ca = c->devs[ptr->dev];

	return ptr_stale(ca, ptr);
}

static void bch_extent_drop_stale(struct bch_fs *c, struct bkey_s_extent e)
{
	struct bch_extent_ptr *ptr = &e.v->start->ptr;
	bool dropped = false;

	while ((ptr = extent_ptr_next(e, ptr)))
		if (should_drop_ptr(c, e.c, ptr)) {
			__bch_extent_drop_ptr(e, ptr);
			dropped = true;
		} else
			ptr++;

	if (dropped)
		bch_extent_drop_redundant_crcs(e);
}

static bool bch_ptr_normalize(struct bch_fs *c, struct btree *bk,
			      struct bkey_s k)
{
	return bch_extent_normalize(c, k);
}

static void bch_ptr_swab(const struct bkey_format *f, struct bkey_packed *k)
{
	switch (k->type) {
	case BCH_EXTENT:
	case BCH_EXTENT_CACHED: {
		union bch_extent_entry *entry;
		u64 *d = (u64 *) bkeyp_val(f, k);
		unsigned i;

		for (i = 0; i < bkeyp_val_u64s(f, k); i++)
			d[i] = swab64(d[i]);

		for (entry = (union bch_extent_entry *) d;
		     entry < (union bch_extent_entry *) (d + bkeyp_val_u64s(f, k));
		     entry = extent_entry_next(entry)) {
			switch (extent_entry_type(entry)) {
			case BCH_EXTENT_ENTRY_crc32:
				entry->crc32.csum = swab32(entry->crc32.csum);
				break;
			case BCH_EXTENT_ENTRY_crc64:
				entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
				entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
				break;
			case BCH_EXTENT_ENTRY_crc128:
				entry->crc128.csum.hi = swab64(entry->crc64.csum_hi);
				entry->crc128.csum.lo = swab64(entry->crc64.csum_lo);
				break;
			case BCH_EXTENT_ENTRY_ptr:
				break;
			}
		}
		break;
	}
	}
}

static const char *extent_ptr_invalid(const struct bch_fs *c,
				      struct bkey_s_c_extent e,
				      const struct bch_extent_ptr *ptr,
				      unsigned size_ondisk,
				      bool metadata)
{
	const struct bch_extent_ptr *ptr2;
	struct bch_dev *ca;

	if (ptr->dev >= c->sb.nr_devices)
		return "pointer to invalid device";

	ca = c->devs[ptr->dev];
	if (!ca)
		return "pointer to invalid device";

	extent_for_each_ptr(e, ptr2)
		if (ptr != ptr2 && ptr->dev == ptr2->dev)
			return "multiple pointers to same device";

	if (ptr->offset + size_ondisk > ca->mi.bucket_size * ca->mi.nbuckets)
		return "offset past end of device";

	if (ptr->offset < ca->mi.bucket_size * ca->mi.first_bucket)
		return "offset before first bucket";

	if ((ptr->offset & (ca->mi.bucket_size - 1)) +
	    size_ondisk > ca->mi.bucket_size)
		return "spans multiple buckets";

	if (!(metadata ? ca->mi.has_metadata : ca->mi.has_data))
		return "device not marked as containing data";

	return NULL;
}

static size_t extent_print_ptrs(struct bch_fs *c, char *buf,
				size_t size, struct bkey_s_c_extent e)
{
	char *out = buf, *end = buf + size;
	const union bch_extent_entry *entry;
	const union bch_extent_crc *crc;
	const struct bch_extent_ptr *ptr;
	struct bch_dev *ca;
	bool first = true;

#define p(...)	(out += scnprintf(out, end - out, __VA_ARGS__))

	extent_for_each_entry(e, entry) {
		if (!first)
			p(" ");

		switch (__extent_entry_type(entry)) {
		case BCH_EXTENT_ENTRY_crc32:
		case BCH_EXTENT_ENTRY_crc64:
		case BCH_EXTENT_ENTRY_crc128:
			crc = entry_to_crc(entry);

			p("crc: c_size %u size %u offset %u csum %u compress %u",
			  crc_compressed_size(e.k, crc),
			  crc_uncompressed_size(e.k, crc),
			  crc_offset(crc), crc_csum_type(crc),
			  crc_compression_type(crc));
			break;
		case BCH_EXTENT_ENTRY_ptr:
			ptr = entry_to_ptr(entry);
			ca = c->devs[ptr->dev];

			p("ptr: %u:%llu gen %u%s", ptr->dev,
			  (u64) ptr->offset, ptr->gen,
			  ca && ptr_stale(ca, ptr)
			  ? " stale" : "");
			break;
		default:
			p("(invalid extent entry %.16llx)", *((u64 *) entry));
			goto out;
		}

		first = false;
	}
out:
	if (bkey_extent_is_cached(e.k))
		p(" cached");
#undef p
	return out - buf;
}

/* Btree ptrs */

static const char *bch_btree_ptr_invalid(const struct bch_fs *c,
					 struct bkey_s_c k)
{
	if (bkey_extent_is_cached(k.k))
		return "cached";

	if (k.k->size)
		return "nonzero key size";

	if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
		return "value too big";

	switch (k.k->type) {
	case BCH_EXTENT: {
		struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
		const union bch_extent_entry *entry;
		const struct bch_extent_ptr *ptr;
		const union bch_extent_crc *crc;
		const char *reason;

		extent_for_each_entry(e, entry)
			if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX)
				return "invalid extent entry type";

		extent_for_each_ptr_crc(e, ptr, crc) {
			reason = extent_ptr_invalid(c, e, ptr,
						    c->sb.btree_node_size,
						    true);
			if (reason)
				return reason;
		}

		if (crc)
			return "has crc field";

		return NULL;
	}

	default:
		return "invalid value type";
	}
}

static void btree_ptr_debugcheck(struct bch_fs *c, struct btree *b,
				 struct bkey_s_c k)
{
	struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
	const struct bch_extent_ptr *ptr;
	unsigned seq;
	const char *err;
	char buf[160];
	struct bucket *g;
	struct bch_dev *ca;
	unsigned replicas = 0;
	bool bad;

	extent_for_each_ptr(e, ptr) {
		ca = c->devs[ptr->dev];
		g = PTR_BUCKET(ca, ptr);
		replicas++;

		err = "stale";
		if (ptr_stale(ca, ptr))
			goto err;

		do {
			seq = read_seqcount_begin(&c->gc_pos_lock);
			bad = gc_pos_cmp(c->gc_pos, gc_pos_btree_node(b)) > 0 &&
				g->mark.data_type != BUCKET_BTREE;
		} while (read_seqcount_retry(&c->gc_pos_lock, seq));

		err = "inconsistent";
		if (bad)
			goto err;
	}

	if (replicas < c->sb.meta_replicas_have) {
		bch_bkey_val_to_text(c, btree_node_type(b),
				     buf, sizeof(buf), k);
		bch_fs_bug(c,
			"btree key bad (too few replicas, %u < %u): %s",
			replicas, c->sb.meta_replicas_have, buf);
		return;
	}

	return;
err:
	bch_bkey_val_to_text(c, btree_node_type(b), buf, sizeof(buf), k);
	bch_fs_bug(c, "%s btree pointer %s: bucket %zi prio %i "
		      "gen %i last_gc %i mark %08x",
		      err, buf, PTR_BUCKET_NR(ca, ptr),
		      g->read_prio, PTR_BUCKET(ca, ptr)->mark.gen,
		      ca->oldest_gens[PTR_BUCKET_NR(ca, ptr)],
		      (unsigned) g->mark.counter);
}

static void bch_btree_ptr_to_text(struct bch_fs *c, char *buf,
				  size_t size, struct bkey_s_c k)
{
	char *out = buf, *end = buf + size;
	const char *invalid;

#define p(...)	(out += scnprintf(out, end - out, __VA_ARGS__))

	if (bkey_extent_is_data(k.k))
		out += extent_print_ptrs(c, buf, size, bkey_s_c_to_extent(k));

	invalid = bch_btree_ptr_invalid(c, k);
	if (invalid)
		p(" invalid: %s", invalid);
#undef p
}

struct extent_pick_ptr
bch_btree_pick_ptr(struct bch_fs *c, const struct btree *b)
{
	struct bkey_s_c_extent e = bkey_i_to_s_c_extent(&b->key);
	const union bch_extent_crc *crc;
	const struct bch_extent_ptr *ptr;
	struct extent_pick_ptr pick = { .ca = NULL };

	extent_for_each_ptr_crc(e, ptr, crc) {
		struct bch_dev *ca = c->devs[ptr->dev];
		struct btree *root = btree_node_root(c, b);

		if (bch_fs_inconsistent_on(crc, c,
				"btree node pointer with crc at btree %u level %u/%u bucket %zu",
				b->btree_id, b->level, root ? root->level : -1,
				PTR_BUCKET_NR(ca, ptr)))
			break;

		if (bch_dev_inconsistent_on(ptr_stale(ca, ptr), ca,
				"stale btree node pointer at btree %u level %u/%u bucket %zu",
				b->btree_id, b->level, root ? root->level : -1,
				PTR_BUCKET_NR(ca, ptr)))
			continue;

		if (ca->mi.state == BCH_MEMBER_STATE_FAILED)
			continue;

		if (pick.ca && pick.ca->mi.tier < ca->mi.tier)
			continue;

		if (!percpu_ref_tryget(&ca->io_ref))
			continue;

		if (pick.ca)
			percpu_ref_put(&pick.ca->io_ref);

		pick.ca		= ca;
		pick.ptr	= *ptr;
	}

	return pick;
}

const struct bkey_ops bch_bkey_btree_ops = {
	.key_invalid	= bch_btree_ptr_invalid,
	.key_debugcheck	= btree_ptr_debugcheck,
	.val_to_text	= bch_btree_ptr_to_text,
	.swab		= bch_ptr_swab,
};

/* Extents */

static bool __bch_cut_front(struct bpos where, struct bkey_s k)
{
	u64 len = 0;

	if (bkey_cmp(where, bkey_start_pos(k.k)) <= 0)
		return false;

	EBUG_ON(bkey_cmp(where, k.k->p) > 0);

	len = k.k->p.offset - where.offset;

	BUG_ON(len > k.k->size);

	/*
	 * Don't readjust offset if the key size is now 0, because that could
	 * cause offset to point to the next bucket:
	 */
	if (!len)
		__set_bkey_deleted(k.k);
	else if (bkey_extent_is_data(k.k)) {
		struct bkey_s_extent e = bkey_s_to_extent(k);
		struct bch_extent_ptr *ptr;
		union bch_extent_crc *crc, *prev_crc = NULL;

		extent_for_each_ptr_crc(e, ptr, crc) {
			switch (extent_crc_type(crc)) {
			case BCH_EXTENT_CRC_NONE:
				ptr->offset += e.k->size - len;
				break;
			case BCH_EXTENT_CRC32:
				if (prev_crc != crc)
					crc->crc32.offset += e.k->size - len;
				break;
			case BCH_EXTENT_CRC64:
				if (prev_crc != crc)
					crc->crc64.offset += e.k->size - len;
				break;
			case BCH_EXTENT_CRC128:
				if (prev_crc != crc)
					crc->crc128.offset += e.k->size - len;
				break;
			}
			prev_crc = crc;
		}
	}

	k.k->size = len;

	return true;
}

bool bch_cut_front(struct bpos where, struct bkey_i *k)
{
	return __bch_cut_front(where, bkey_i_to_s(k));
}

bool bch_cut_back(struct bpos where, struct bkey *k)
{
	u64 len = 0;

	if (bkey_cmp(where, k->p) >= 0)
		return false;

	EBUG_ON(bkey_cmp(where, bkey_start_pos(k)) < 0);

	len = where.offset - bkey_start_offset(k);

	BUG_ON(len > k->size);

	k->p = where;
	k->size = len;

	if (!len)
		__set_bkey_deleted(k);

	return true;
}

/**
 * bch_key_resize - adjust size of @k
 *
 * bkey_start_offset(k) will be preserved, modifies where the extent ends
 */
void bch_key_resize(struct bkey *k,
		    unsigned new_size)
{
	k->p.offset -= k->size;
	k->p.offset += new_size;
	k->size = new_size;
}

/*
 * In extent_sort_fix_overlapping(), insert_fixup_extent(),
 * extent_merge_inline() - we're modifying keys in place that are packed. To do
 * that we have to unpack the key, modify the unpacked key - then this
 * copies/repacks the unpacked to the original as necessary.
 */
static bool __extent_save(struct btree *b, struct btree_node_iter *iter,
			  struct bkey_packed *dst, struct bkey *src)
{
	struct bkey_format *f = &b->format;
	struct bkey_i *dst_unpacked;
	bool ret;

	if ((dst_unpacked = packed_to_bkey(dst))) {
		dst_unpacked->k = *src;
		ret = true;
	} else {
		ret = bkey_pack_key(dst, src, f);
	}

	if (ret && iter)
		bch_verify_key_order(b, iter, dst);

	return ret;
}

static void extent_save(struct btree *b, struct btree_node_iter *iter,
			struct bkey_packed *dst, struct bkey *src)
{
	BUG_ON(!__extent_save(b, iter, dst, src));
}

/*
 * Returns true if l > r - unless l == r, in which case returns true if l is
 * older than r.
 *
 * Necessary for sort_fix_overlapping() - if there are multiple keys that
 * compare equal in different sets, we have to process them newest to oldest.
 */
#define extent_sort_cmp(l, r)						\
({									\
	struct bkey _ul = bkey_unpack_key(b,				\
				__btree_node_offset_to_key(b, (l).k));	\
	struct bkey _ur = bkey_unpack_key(b,				\
				__btree_node_offset_to_key(b, (r).k));	\
									\
	int _c = bkey_cmp(bkey_start_pos(&_ul), bkey_start_pos(&_ur));	\
	_c ? _c > 0 : (l).k < (r).k;					\
})

static inline void extent_sort_sift(struct btree_node_iter *iter,
				    struct btree *b, size_t i)
{
	heap_sift(iter, i, extent_sort_cmp);
}

static inline void extent_sort_next(struct btree_node_iter *iter,
				    struct btree *b,
				    struct btree_node_iter_set *i)
{
	sort_key_next(iter, b, i);
	heap_sift(iter, i - iter->data, extent_sort_cmp);
}

static void extent_sort_append(struct bch_fs *c,
			       struct btree *b,
			       struct btree_nr_keys *nr,
			       struct bkey_packed *start,
			       struct bkey_packed **prev,
			       struct bkey_packed *k)
{
	struct bkey_format *f = &b->format;
	BKEY_PADDED(k) tmp;

	if (bkey_whiteout(k))
		return;

	bkey_unpack(b, &tmp.k, k);

	if (*prev &&
	    bch_extent_merge(c, b, (void *) *prev, &tmp.k))
		return;

	if (*prev) {
		bkey_pack(*prev, (void *) *prev, f);

		btree_keys_account_key_add(nr, 0, *prev);
		*prev = bkey_next(*prev);
	} else {
		*prev = start;
	}

	bkey_copy(*prev, &tmp.k);
}

struct btree_nr_keys bch_extent_sort_fix_overlapping(struct bch_fs *c,
					struct bset *dst,
					struct btree *b,
					struct btree_node_iter *iter)
{
	struct bkey_format *f = &b->format;
	struct btree_node_iter_set *_l = iter->data, *_r;
	struct bkey_packed *prev = NULL, *out, *lk, *rk;
	struct bkey l_unpacked, r_unpacked;
	struct bkey_s l, r;
	struct btree_nr_keys nr;

	memset(&nr, 0, sizeof(nr));

	heap_resort(iter, extent_sort_cmp);

	while (!bch_btree_node_iter_end(iter)) {
		lk = __btree_node_offset_to_key(b, _l->k);

		if (iter->used == 1) {
			extent_sort_append(c, b, &nr, dst->start, &prev, lk);
			extent_sort_next(iter, b, _l);
			continue;
		}

		_r = iter->data + 1;
		if (iter->used > 2 &&
		    extent_sort_cmp(_r[0], _r[1]))
			_r++;

		rk = __btree_node_offset_to_key(b, _r->k);

		l = __bkey_disassemble(b, lk, &l_unpacked);
		r = __bkey_disassemble(b, rk, &r_unpacked);

		/* If current key and next key don't overlap, just append */
		if (bkey_cmp(l.k->p, bkey_start_pos(r.k)) <= 0) {
			extent_sort_append(c, b, &nr, dst->start, &prev, lk);
			extent_sort_next(iter, b, _l);
			continue;
		}

		/* Skip 0 size keys */
		if (!r.k->size) {
			extent_sort_next(iter, b, _r);
			continue;
		}

		/*
		 * overlap: keep the newer key and trim the older key so they
		 * don't overlap. comparing pointers tells us which one is
		 * newer, since the bsets are appended one after the other.
		 */

		/* can't happen because of comparison func */
		BUG_ON(_l->k < _r->k &&
		       !bkey_cmp(bkey_start_pos(l.k), bkey_start_pos(r.k)));

		if (_l->k > _r->k) {
			/* l wins, trim r */
			if (bkey_cmp(l.k->p, r.k->p) >= 0) {
				sort_key_next(iter, b, _r);
			} else {
				__bch_cut_front(l.k->p, r);
				extent_save(b, NULL, rk, r.k);
			}

			extent_sort_sift(iter, b, _r - iter->data);
		} else if (bkey_cmp(l.k->p, r.k->p) > 0) {
			BKEY_PADDED(k) tmp;

			/*
			 * r wins, but it overlaps in the middle of l - split l:
			 */
			bkey_reassemble(&tmp.k, l.s_c);
			bch_cut_back(bkey_start_pos(r.k), &tmp.k.k);

			__bch_cut_front(r.k->p, l);
			extent_save(b, NULL, lk, l.k);

			extent_sort_sift(iter, b, 0);

			extent_sort_append(c, b, &nr, dst->start, &prev,
					   bkey_to_packed(&tmp.k));
		} else {
			bch_cut_back(bkey_start_pos(r.k), l.k);
			extent_save(b, NULL, lk, l.k);
		}
	}

	if (prev) {
		bkey_pack(prev, (void *) prev, f);
		btree_keys_account_key_add(&nr, 0, prev);
		out = bkey_next(prev);
	} else {
		out = dst->start;
	}

	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
	return nr;
}

struct extent_insert_state {
	struct btree_insert		*trans;
	struct btree_insert_entry	*insert;
	struct bpos			committed;
	struct bch_fs_usage		stats;

	/* for deleting: */
	struct bkey_i			whiteout;
	bool				do_journal;
	bool				deleting;
};

static void bch_add_sectors(struct extent_insert_state *s,
			    struct bkey_s_c k, u64 offset, s64 sectors)
{
	struct bch_fs *c = s->trans->c;
	struct btree *b = s->insert->iter->nodes[0];

	EBUG_ON(bkey_cmp(bkey_start_pos(k.k), b->data->min_key) < 0);

	if (!sectors)
		return;

	bch_mark_key(c, k, sectors, false, gc_pos_btree_node(b),
		     &s->stats, s->trans->journal_res.seq);

	if (bkey_extent_is_data(k.k) &&
	    !bkey_extent_is_cached(k.k))
		bcache_dev_sectors_dirty_add(c, k.k->p.inode, offset, sectors);
}

static void bch_subtract_sectors(struct extent_insert_state *s,
				 struct bkey_s_c k, u64 offset, s64 sectors)
{
	bch_add_sectors(s, k, offset, -sectors);
}

/* These wrappers subtract exactly the sectors that we're removing from @k */
static void bch_cut_subtract_back(struct extent_insert_state *s,
				  struct bpos where, struct bkey_s k)
{
	bch_subtract_sectors(s, k.s_c, where.offset,
			     k.k->p.offset - where.offset);
	bch_cut_back(where, k.k);
}

static void bch_cut_subtract_front(struct extent_insert_state *s,
				   struct bpos where, struct bkey_s k)
{
	bch_subtract_sectors(s, k.s_c, bkey_start_offset(k.k),
			     where.offset - bkey_start_offset(k.k));
	__bch_cut_front(where, k);
}

static void bch_drop_subtract(struct extent_insert_state *s, struct bkey_s k)
{
	if (k.k->size)
		bch_subtract_sectors(s, k.s_c,
				     bkey_start_offset(k.k), k.k->size);
	k.k->size = 0;
	__set_bkey_deleted(k.k);
}

/*
 * Note: If this returns true because only some pointers matched,
 * we can lose some caching that had happened in the interim.
 * Because cache promotion only promotes the part of the extent
 * actually read, and not the whole extent, and due to the key
 * splitting done in bch_extent_insert_fixup, preserving such
 * caching is difficult.
 */
static bool bch_extent_cmpxchg_cmp(struct bkey_s_c l, struct bkey_s_c r)
{
	struct bkey_s_c_extent le, re;
	const struct bch_extent_ptr *lp, *rp;
	s64 offset;

	BUG_ON(!l.k->size || !r.k->size);

	if (l.k->type != r.k->type ||
	    bversion_cmp(l.k->version, r.k->version))
		return false;

	switch (l.k->type) {
	case KEY_TYPE_COOKIE:
		return !memcmp(bkey_s_c_to_cookie(l).v,
			       bkey_s_c_to_cookie(r).v,
			       sizeof(struct bch_cookie));

	case BCH_EXTENT:
	case BCH_EXTENT_CACHED:
		le = bkey_s_c_to_extent(l);
		re = bkey_s_c_to_extent(r);

		/*
		 * bkey_cmpxchg() handles partial matches - when either l or r
		 * has been trimmed - so we need just to handle l or r not
		 * starting at the same place when checking for a match here.
		 *
		 * If the starts of the keys are different, we just apply that
		 * offset to the device pointer offsets when checking those -
		 * matching how bch_cut_front() adjusts device pointer offsets
		 * when adjusting the start of a key:
		 */
		offset = bkey_start_offset(l.k) - bkey_start_offset(r.k);

		/*
		 * XXX: perhaps we only raced with copygc or tiering replacing
		 * one of the pointers: it should suffice to find _any_ matching
		 * pointer
		 */

		if (bkey_val_u64s(le.k) != bkey_val_u64s(re.k))
			return false;

		extent_for_each_ptr(le, lp) {
			const union bch_extent_entry *entry =
				vstruct_idx(re.v, (u64 *) lp - le.v->_data);

			if (!extent_entry_is_ptr(entry))
				return false;

			rp = &entry->ptr;

			if (lp->offset	!= rp->offset + offset ||
			    lp->dev	!= rp->dev ||
			    lp->gen	!= rp->gen)
				return false;
		}

		return true;
	default:
		return false;
	}

}

/*
 * Returns true on success, false on failure (and false means @new no longer
 * overlaps with @k)
 *
 * If returned true, we may have inserted up to one key in @b.
 * If returned false, we may have inserted up to two keys in @b.
 *
 * On return, there is room in @res for at least one more key of the same size
 * as @new.
 */
enum extent_insert_hook_ret bch_extent_cmpxchg(struct extent_insert_hook *hook,
					       struct bpos committed_pos,
					       struct bpos next_pos,
					       struct bkey_s_c k,
					       const struct bkey_i *new)
{
	struct bch_replace_info *replace = container_of(hook,
					struct bch_replace_info, hook);
	struct bkey_i *old = &replace->key;

	EBUG_ON(bkey_cmp(committed_pos, bkey_start_pos(&new->k)) < 0);

	/* must have something to compare against */
	EBUG_ON(!bkey_val_u64s(&old->k));

	/* new must be a subset of old */
	EBUG_ON(bkey_cmp(new->k.p, old->k.p) > 0 ||
		bkey_cmp(bkey_start_pos(&new->k), bkey_start_pos(&old->k)) < 0);

	if (k.k && bch_extent_cmpxchg_cmp(k, bkey_i_to_s_c(old))) {
		replace->successes++;
		return BTREE_HOOK_DO_INSERT;
	} else {
		replace->failures++;
		return BTREE_HOOK_NO_INSERT;
	}
}

static bool bch_extent_merge_inline(struct bch_fs *,
				    struct btree_iter *,
				    struct bkey_packed *,
				    struct bkey_packed *,
				    bool);

#define MAX_LOCK_HOLD_TIME	(5 * NSEC_PER_MSEC)

static enum btree_insert_ret
extent_insert_should_stop(struct extent_insert_state *s)
{
	struct btree *b = s->insert->iter->nodes[0];

	/*
	 * Check if we have sufficient space in both the btree node and the
	 * journal reservation:
	 *
	 * Each insert checks for room in the journal entry, but we check for
	 * room in the btree node up-front. In the worst case, bkey_cmpxchg()
	 * will insert two keys, and one iteration of this room will insert one
	 * key, so we need room for three keys.
	 */
	if (!bch_btree_node_insert_fits(s->trans->c, b, s->insert->k->k.u64s))
		return BTREE_INSERT_BTREE_NODE_FULL;
	else if (!journal_res_insert_fits(s->trans, s->insert))
		return BTREE_INSERT_JOURNAL_RES_FULL; /* XXX worth tracing */
	else
		return BTREE_INSERT_OK;
}

static void extent_bset_insert(struct bch_fs *c, struct btree_iter *iter,
			       struct bkey_i *insert)
{
	struct btree *b = iter->nodes[0];
	struct btree_node_iter *node_iter = &iter->node_iters[0];
	struct bset_tree *t = bset_tree_last(b);
	struct bkey_packed *where =
		bch_btree_node_iter_bset_pos(node_iter, b, t);
	struct bkey_packed *prev = bkey_prev(b, t, where);
	struct bkey_packed *next_live_key = where;
	unsigned clobber_u64s;

	if (prev)
		where = bkey_next(prev);

	while (next_live_key != btree_bkey_last(b, t) &&
	       bkey_deleted(next_live_key))
		next_live_key = bkey_next(next_live_key);

	/*
	 * Everything between where and next_live_key is now deleted keys, and
	 * is overwritten:
	 */
	clobber_u64s = (u64 *) next_live_key - (u64 *) where;

	if (prev &&
	    bch_extent_merge_inline(c, iter, prev, bkey_to_packed(insert), true))
		goto drop_deleted_keys;

	if (next_live_key != btree_bkey_last(b, t) &&
	    bch_extent_merge_inline(c, iter, bkey_to_packed(insert),
				    next_live_key, false))
		goto drop_deleted_keys;

	bch_bset_insert(b, node_iter, where, insert, clobber_u64s);
	bch_btree_node_iter_fix(iter, b, node_iter, t, where,
				clobber_u64s, where->u64s);
	return;
drop_deleted_keys:
	bch_bset_delete(b, where, clobber_u64s);
	bch_btree_node_iter_fix(iter, b, node_iter, t, where, clobber_u64s, 0);
}

static void extent_insert_committed(struct extent_insert_state *s)
{
	struct bch_fs *c = s->trans->c;
	struct btree_iter *iter = s->insert->iter;
	struct bkey_i *insert = !s->deleting
		? s->insert->k
		: &s->whiteout;
	BKEY_PADDED(k) split;

	EBUG_ON(bkey_cmp(insert->k.p, s->committed) < 0);
	EBUG_ON(bkey_cmp(s->committed, bkey_start_pos(&insert->k)) < 0);

	if (!bkey_cmp(s->committed, bkey_start_pos(&insert->k)))
		return;

	if (s->deleting && !s->do_journal) {
		bch_cut_front(s->committed, insert);
		goto done;
	}

	EBUG_ON(bkey_deleted(&insert->k) || !insert->k.size);

	bkey_copy(&split.k, insert);

	if (!(s->trans->flags & BTREE_INSERT_JOURNAL_REPLAY) &&
	    bkey_cmp(s->committed, insert->k.p) &&
	    bkey_extent_is_compressed(bkey_i_to_s_c(insert))) {
		/* XXX: possibly need to increase our reservation? */
		bch_cut_subtract_back(s, s->committed,
				      bkey_i_to_s(&split.k));
		bch_cut_front(s->committed, insert);
		bch_add_sectors(s, bkey_i_to_s_c(insert),
				bkey_start_offset(&insert->k),
				insert->k.size);
	} else {
		bch_cut_back(s->committed, &split.k.k);
		bch_cut_front(s->committed, insert);
	}

	if (debug_check_bkeys(c))
		bkey_debugcheck(c, iter->nodes[iter->level],
				bkey_i_to_s_c(&split.k));

	bch_btree_journal_key(s->trans, iter, &split.k);

	if (!s->deleting)
		extent_bset_insert(c, iter, &split.k);
done:
	bch_btree_iter_set_pos_same_leaf(iter, s->committed);

	insert->k.needs_whiteout	= false;
	s->do_journal			= false;
	s->trans->did_work		= true;
}

static enum extent_insert_hook_ret
__extent_insert_advance_pos(struct extent_insert_state *s,
			    struct bpos next_pos,
			    struct bkey_s_c k)
{
	struct extent_insert_hook *hook = s->trans->hook;
	enum extent_insert_hook_ret ret;
#if 0
	/*
	 * Currently disabled for encryption - broken with fcollapse. Will have
	 * to reenable when versions are exposed for send/receive - versions
	 * will have to be monotonic then:
	 */
	if (k.k && k.k->size &&
	    !bversion_zero(s->insert->k->k.version) &&
	    bversion_cmp(k.k->version, s->insert->k->k.version) > 0) {
		ret = BTREE_HOOK_NO_INSERT;
	} else
#endif
	if (hook)
		ret = hook->fn(hook, s->committed, next_pos, k, s->insert->k);
	else
		ret = BTREE_HOOK_DO_INSERT;

	EBUG_ON(bkey_deleted(&s->insert->k->k) || !s->insert->k->k.size);

	switch (ret) {
	case BTREE_HOOK_DO_INSERT:
		break;
	case BTREE_HOOK_NO_INSERT:
		extent_insert_committed(s);
		bch_cut_subtract_front(s, next_pos, bkey_i_to_s(s->insert->k));

		bch_btree_iter_set_pos_same_leaf(s->insert->iter, next_pos);
		break;
	case BTREE_HOOK_RESTART_TRANS:
		return ret;
	}

	s->committed = next_pos;
	return ret;
}

/*
 * Update iter->pos, marking how much of @insert we've processed, and call hook
 * fn:
 */
static enum extent_insert_hook_ret
extent_insert_advance_pos(struct extent_insert_state *s, struct bkey_s_c k)
{
	struct btree *b = s->insert->iter->nodes[0];
	struct bpos next_pos = bpos_min(s->insert->k->k.p,
					k.k ? k.k->p : b->key.k.p);

	/* hole? */
	if (k.k && bkey_cmp(s->committed, bkey_start_pos(k.k)) < 0) {
		bool have_uncommitted = bkey_cmp(s->committed,
				bkey_start_pos(&s->insert->k->k)) > 0;

		switch (__extent_insert_advance_pos(s, bkey_start_pos(k.k),
						    bkey_s_c_null)) {
		case BTREE_HOOK_DO_INSERT:
			break;
		case BTREE_HOOK_NO_INSERT:
			/*
			 * we had to split @insert and insert the committed
			 * part - need to bail out and recheck journal
			 * reservation/btree node before we advance pos past @k:
			 */
			if (have_uncommitted)
				return BTREE_HOOK_NO_INSERT;
			break;
		case BTREE_HOOK_RESTART_TRANS:
			return BTREE_HOOK_RESTART_TRANS;
		}
	}

	/* avoid redundant calls to hook fn: */
	if (!bkey_cmp(s->committed, next_pos))
		return BTREE_HOOK_DO_INSERT;

	return __extent_insert_advance_pos(s, next_pos, k);
}

static enum btree_insert_ret
extent_insert_check_split_compressed(struct extent_insert_state *s,
				     struct bkey_s_c k,
				     enum bch_extent_overlap overlap)
{
	struct bch_fs *c = s->trans->c;
	unsigned sectors;

	if (overlap == BCH_EXTENT_OVERLAP_MIDDLE &&
	    (sectors = bkey_extent_is_compressed(k))) {
		int flags = BCH_DISK_RESERVATION_BTREE_LOCKS_HELD;

		if (s->trans->flags & BTREE_INSERT_NOFAIL)
			flags |= BCH_DISK_RESERVATION_NOFAIL;

		switch (bch_disk_reservation_add(c,
				s->trans->disk_res,
				sectors, flags)) {
		case 0:
			break;
		case -ENOSPC:
			return BTREE_INSERT_ENOSPC;
		case -EINTR:
			return BTREE_INSERT_NEED_GC_LOCK;
		default:
			BUG();
		}
	}

	return BTREE_INSERT_OK;
}

static enum btree_insert_ret
extent_squash(struct extent_insert_state *s, struct bkey_i *insert,
	      struct bset_tree *t, struct bkey_packed *_k, struct bkey_s k,
	      enum bch_extent_overlap overlap)
{
	struct bch_fs *c = s->trans->c;
	struct btree_iter *iter = s->insert->iter;
	struct btree *b = iter->nodes[0];
	struct btree_node_iter *node_iter = &iter->node_iters[0];

	switch (overlap) {
	case BCH_EXTENT_OVERLAP_FRONT:
		/* insert overlaps with start of k: */
		bch_cut_subtract_front(s, insert->k.p, k);
		BUG_ON(bkey_deleted(k.k));
		extent_save(b, node_iter, _k, k.k);
		break;

	case BCH_EXTENT_OVERLAP_BACK:
		/* insert overlaps with end of k: */
		bch_cut_subtract_back(s, bkey_start_pos(&insert->k), k);
		BUG_ON(bkey_deleted(k.k));
		extent_save(b, node_iter, _k, k.k);

		/*
		 * As the auxiliary tree is indexed by the end of the
		 * key and we've just changed the end, update the
		 * auxiliary tree.
		 */
		bch_bset_fix_invalidated_key(b, t, _k);
		bch_btree_node_iter_fix(iter, b, node_iter, t,
					_k, _k->u64s, _k->u64s);
		break;

	case BCH_EXTENT_OVERLAP_ALL: {
		struct bpos orig_pos = k.k->p;

		/* The insert key completely covers k, invalidate k */
		if (!bkey_whiteout(k.k))
			btree_keys_account_key_drop(&b->nr,
						t - b->set, _k);

		bch_drop_subtract(s, k);
		k.k->p = bkey_start_pos(&insert->k);
		if (!__extent_save(b, node_iter, _k, k.k)) {
			/*
			 * Couldn't repack: we aren't necessarily able
			 * to repack if the new key is outside the range
			 * of the old extent, so we have to split
			 * @insert:
			 */
			k.k->p = orig_pos;
			extent_save(b, node_iter, _k, k.k);

			if (extent_insert_advance_pos(s, k.s_c) ==
			    BTREE_HOOK_RESTART_TRANS)
				return BTREE_INSERT_NEED_TRAVERSE;

			extent_insert_committed(s);
			/*
			 * We split and inserted upto at k.k->p - that
			 * has to coincide with iter->pos, so that we
			 * don't have anything more we have to insert
			 * until we recheck our journal reservation:
			 */
			EBUG_ON(bkey_cmp(s->committed, k.k->p));
		} else {
			bch_bset_fix_invalidated_key(b, t, _k);
			bch_btree_node_iter_fix(iter, b, node_iter, t,
						_k, _k->u64s, _k->u64s);
		}

		break;
	}
	case BCH_EXTENT_OVERLAP_MIDDLE: {
		BKEY_PADDED(k) split;
		/*
		 * The insert key falls 'in the middle' of k
		 * The insert key splits k in 3:
		 * - start only in k, preserve
		 * - middle common section, invalidate in k
		 * - end only in k, preserve
		 *
		 * We update the old key to preserve the start,
		 * insert will be the new common section,
		 * we manually insert the end that we are preserving.
		 *
		 * modify k _before_ doing the insert (which will move
		 * what k points to)
		 */
		bkey_reassemble(&split.k, k.s_c);
		split.k.k.needs_whiteout |= bset_written(b, bset(b, t));

		bch_cut_back(bkey_start_pos(&insert->k), &split.k.k);
		BUG_ON(bkey_deleted(&split.k.k));

		bch_cut_subtract_front(s, insert->k.p, k);
		BUG_ON(bkey_deleted(k.k));
		extent_save(b, node_iter, _k, k.k);

		bch_add_sectors(s, bkey_i_to_s_c(&split.k),
				bkey_start_offset(&split.k.k),
				split.k.k.size);
		extent_bset_insert(c, iter, &split.k);
		break;
	}
	}

	return BTREE_INSERT_OK;
}

static enum btree_insert_ret
bch_delete_fixup_extent(struct extent_insert_state *s)
{
	struct bch_fs *c = s->trans->c;
	struct btree_iter *iter = s->insert->iter;
	struct btree *b = iter->nodes[0];
	struct btree_node_iter *node_iter = &iter->node_iters[0];
	struct bkey_packed *_k;
	struct bkey unpacked;
	struct bkey_i *insert = s->insert->k;
	enum btree_insert_ret ret = BTREE_INSERT_OK;

	EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&insert->k)));

	s->whiteout	= *insert;
	s->do_journal	= false;

	while (bkey_cmp(s->committed, insert->k.p) < 0 &&
	       (ret = extent_insert_should_stop(s)) == BTREE_INSERT_OK &&
	       (_k = bch_btree_node_iter_peek_all(node_iter, b))) {
		struct bset_tree *t = bch_bkey_to_bset(b, _k);
		struct bkey_s k = __bkey_disassemble(b, _k, &unpacked);
		enum bch_extent_overlap overlap;

		EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&insert->k)));
		EBUG_ON(bkey_cmp(iter->pos, k.k->p) >= 0);

		if (bkey_cmp(bkey_start_pos(k.k), insert->k.p) >= 0)
			break;

		if (bkey_whiteout(k.k)) {
			s->committed = bpos_min(insert->k.p, k.k->p);
			goto next;
		}

		overlap = bch_extent_overlap(&insert->k, k.k);

		ret = extent_insert_check_split_compressed(s, k.s_c, overlap);
		if (ret != BTREE_INSERT_OK)
			goto stop;

		switch (extent_insert_advance_pos(s, k.s_c)) {
		case BTREE_HOOK_DO_INSERT:
			break;
		case BTREE_HOOK_NO_INSERT:
			continue;
		case BTREE_HOOK_RESTART_TRANS:
			ret = BTREE_INSERT_NEED_TRAVERSE;
			goto stop;
		}

		s->do_journal = true;

		if (overlap == BCH_EXTENT_OVERLAP_ALL) {
			btree_keys_account_key_drop(&b->nr,
						t - b->set, _k);
			bch_subtract_sectors(s, k.s_c,
					     bkey_start_offset(k.k), k.k->size);
			_k->type = KEY_TYPE_DISCARD;
			reserve_whiteout(b, t, _k);
		} else if (k.k->needs_whiteout ||
			   bset_written(b, bset(b, t))) {
			struct bkey_i discard = *insert;

			switch (overlap) {
			case BCH_EXTENT_OVERLAP_FRONT:
				bch_cut_front(bkey_start_pos(k.k), &discard);
				break;
			case BCH_EXTENT_OVERLAP_BACK:
				bch_cut_back(k.k->p, &discard.k);
				break;
			default:
				break;
			}

			discard.k.needs_whiteout = true;

			ret = extent_squash(s, insert, t, _k, k, overlap);
			BUG_ON(ret != BTREE_INSERT_OK);

			extent_bset_insert(c, iter, &discard);
		} else {
			ret = extent_squash(s, insert, t, _k, k, overlap);
			BUG_ON(ret != BTREE_INSERT_OK);
		}
next:
		bch_cut_front(s->committed, insert);
		bch_btree_iter_set_pos_same_leaf(iter, s->committed);
	}

	if (bkey_cmp(s->committed, insert->k.p) < 0 &&
	    ret == BTREE_INSERT_OK &&
	    extent_insert_advance_pos(s, bkey_s_c_null) == BTREE_HOOK_RESTART_TRANS)
		ret = BTREE_INSERT_NEED_TRAVERSE;
stop:
	extent_insert_committed(s);

	bch_fs_usage_apply(c, &s->stats, s->trans->disk_res,
			   gc_pos_btree_node(b));

	EBUG_ON(bkey_cmp(iter->pos, s->committed));
	EBUG_ON((bkey_cmp(iter->pos, b->key.k.p) == 0) != iter->at_end_of_leaf);

	bch_cut_front(iter->pos, insert);

	if (insert->k.size && iter->at_end_of_leaf)
		ret = BTREE_INSERT_NEED_TRAVERSE;

	EBUG_ON(insert->k.size && ret == BTREE_INSERT_OK);

	return ret;
}

/**
 * bch_extent_insert_fixup - insert a new extent and deal with overlaps
 *
 * this may result in not actually doing the insert, or inserting some subset
 * of the insert key. For cmpxchg operations this is where that logic lives.
 *
 * All subsets of @insert that need to be inserted are inserted using
 * bch_btree_insert_and_journal(). If @b or @res fills up, this function
 * returns false, setting @iter->pos for the prefix of @insert that actually got
 * inserted.
 *
 * BSET INVARIANTS: this function is responsible for maintaining all the
 * invariants for bsets of extents in memory. things get really hairy with 0
 * size extents
 *
 * within one bset:
 *
 * bkey_start_pos(bkey_next(k)) >= k
 * or bkey_start_offset(bkey_next(k)) >= k->offset
 *
 * i.e. strict ordering, no overlapping extents.
 *
 * multiple bsets (i.e. full btree node):
 *
 * ∀ k, j
 *   k.size != 0 ∧ j.size != 0 →
 *     ¬ (k > bkey_start_pos(j) ∧ k < j)
 *
 * i.e. no two overlapping keys _of nonzero size_
 *
 * We can't realistically maintain this invariant for zero size keys because of
 * the key merging done in bch_btree_insert_key() - for two mergeable keys k, j
 * there may be another 0 size key between them in another bset, and it will
 * thus overlap with the merged key.
 *
 * In addition, the end of iter->pos indicates how much has been processed.
 * If the end of iter->pos is not the same as the end of insert, then
 * key insertion needs to continue/be retried.
 */
enum btree_insert_ret
bch_insert_fixup_extent(struct btree_insert *trans,
			struct btree_insert_entry *insert)
{
	struct bch_fs *c = trans->c;
	struct btree_iter *iter = insert->iter;
	struct btree *b = iter->nodes[0];
	struct btree_node_iter *node_iter = &iter->node_iters[0];
	struct bkey_packed *_k;
	struct bkey unpacked;
	enum btree_insert_ret ret = BTREE_INSERT_OK;

	struct extent_insert_state s = {
		.trans		= trans,
		.insert		= insert,
		.committed	= insert->iter->pos,
		.deleting	= bkey_whiteout(&insert->k->k),
	};

	EBUG_ON(iter->level);
	EBUG_ON(bkey_deleted(&insert->k->k) || !insert->k->k.size);

	if (s.deleting)
		return bch_delete_fixup_extent(&s);

	/*
	 * As we process overlapping extents, we advance @iter->pos both to
	 * signal to our caller (btree_insert_key()) how much of @insert->k has
	 * been inserted, and also to keep @iter->pos consistent with
	 * @insert->k and the node iterator that we're advancing:
	 */
	EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&insert->k->k)));

	if (!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))
		bch_add_sectors(&s, bkey_i_to_s_c(insert->k),
				bkey_start_offset(&insert->k->k),
				insert->k->k.size);

	while (bkey_cmp(s.committed, insert->k->k.p) < 0 &&
	       (ret = extent_insert_should_stop(&s)) == BTREE_INSERT_OK &&
	       (_k = bch_btree_node_iter_peek_all(node_iter, b))) {
		struct bset_tree *t = bch_bkey_to_bset(b, _k);
		struct bkey_s k = __bkey_disassemble(b, _k, &unpacked);
		enum bch_extent_overlap overlap;

		EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&insert->k->k)));
		EBUG_ON(bkey_cmp(iter->pos, k.k->p) >= 0);

		if (bkey_cmp(bkey_start_pos(k.k), insert->k->k.p) >= 0)
			break;

		overlap = bch_extent_overlap(&insert->k->k, k.k);

		ret = extent_insert_check_split_compressed(&s, k.s_c, overlap);
		if (ret != BTREE_INSERT_OK)
			goto stop;

		if (!k.k->size)
			goto squash;

		/*
		 * Only call advance pos & call hook for nonzero size extents:
		 * If hook returned BTREE_HOOK_NO_INSERT, @insert->k no longer
		 * overlaps with @k:
		 */
		switch (extent_insert_advance_pos(&s, k.s_c)) {
		case BTREE_HOOK_DO_INSERT:
			break;
		case BTREE_HOOK_NO_INSERT:
			continue;
		case BTREE_HOOK_RESTART_TRANS:
			ret = BTREE_INSERT_NEED_TRAVERSE;
			goto stop;
		}

		if (k.k->size &&
		    (k.k->needs_whiteout || bset_written(b, bset(b, t))))
			insert->k->k.needs_whiteout = true;

		if (overlap == BCH_EXTENT_OVERLAP_ALL &&
		    bkey_whiteout(k.k) &&
		    k.k->needs_whiteout) {
			unreserve_whiteout(b, t, _k);
			_k->needs_whiteout = false;
		}
squash:
		ret = extent_squash(&s, insert->k, t, _k, k, overlap);
		if (ret != BTREE_INSERT_OK)
			goto stop;
	}

	if (bkey_cmp(s.committed, insert->k->k.p) < 0 &&
	    ret == BTREE_INSERT_OK &&
	    extent_insert_advance_pos(&s, bkey_s_c_null) == BTREE_HOOK_RESTART_TRANS)
		ret = BTREE_INSERT_NEED_TRAVERSE;
stop:
	extent_insert_committed(&s);
	/*
	 * Subtract any remaining sectors from @insert, if we bailed out early
	 * and didn't fully insert @insert:
	 */
	if (insert->k->k.size &&
	    !(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))
		bch_subtract_sectors(&s, bkey_i_to_s_c(insert->k),
				     bkey_start_offset(&insert->k->k),
				     insert->k->k.size);

	bch_fs_usage_apply(c, &s.stats, trans->disk_res,
			   gc_pos_btree_node(b));

	EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&insert->k->k)));
	EBUG_ON(bkey_cmp(iter->pos, s.committed));
	EBUG_ON((bkey_cmp(iter->pos, b->key.k.p) == 0) != iter->at_end_of_leaf);

	if (insert->k->k.size && iter->at_end_of_leaf)
		ret = BTREE_INSERT_NEED_TRAVERSE;

	EBUG_ON(insert->k->k.size && ret == BTREE_INSERT_OK);

	return ret;
}

static const char *bch_extent_invalid(const struct bch_fs *c,
				      struct bkey_s_c k)
{
	if (bkey_val_u64s(k.k) > BKEY_EXTENT_VAL_U64s_MAX)
		return "value too big";

	if (!k.k->size)
		return "zero key size";

	switch (k.k->type) {
	case BCH_EXTENT:
	case BCH_EXTENT_CACHED: {
		struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
		const union bch_extent_entry *entry;
		const union bch_extent_crc *crc;
		const struct bch_extent_ptr *ptr;
		unsigned size_ondisk = e.k->size;
		const char *reason;

		extent_for_each_entry(e, entry) {
			if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX)
				return "invalid extent entry type";

			if (extent_entry_is_crc(entry)) {
				crc = entry_to_crc(entry);

				if (crc_offset(crc) + e.k->size >
				    crc_uncompressed_size(e.k, crc))
					return "checksum offset + key size > uncompressed size";

				size_ondisk = crc_compressed_size(e.k, crc);

				if (!bch_checksum_type_valid(c, crc_csum_type(crc)))
					return "invalid checksum type";

				if (crc_compression_type(crc) >= BCH_COMPRESSION_NR)
					return "invalid compression type";
			} else {
				ptr = entry_to_ptr(entry);

				reason = extent_ptr_invalid(c, e, &entry->ptr,
							    size_ondisk, false);
				if (reason)
					return reason;
			}
		}

		return NULL;
	}

	case BCH_RESERVATION: {
		struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);

		if (bkey_val_bytes(k.k) != sizeof(struct bch_reservation))
			return "incorrect value size";

		if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX)
			return "invalid nr_replicas";

		return NULL;
	}

	default:
		return "invalid value type";
	}
}

static void bch_extent_debugcheck_extent(struct bch_fs *c, struct btree *b,
					 struct bkey_s_c_extent e)
{
	const struct bch_extent_ptr *ptr;
	struct bch_dev *ca;
	struct bucket *g;
	unsigned seq, stale;
	char buf[160];
	bool bad;
	unsigned ptrs_per_tier[BCH_TIER_MAX];
	unsigned replicas = 0;

	/*
	 * XXX: we should be doing most/all of these checks at startup time,
	 * where we check bkey_invalid() in btree_node_read_done()
	 *
	 * But note that we can't check for stale pointers or incorrect gc marks
	 * until after journal replay is done (it might be an extent that's
	 * going to get overwritten during replay)
	 */

	memset(ptrs_per_tier, 0, sizeof(ptrs_per_tier));

	extent_for_each_ptr(e, ptr) {
		ca = c->devs[ptr->dev];
		g = PTR_BUCKET(ca, ptr);
		replicas++;
		ptrs_per_tier[ca->mi.tier]++;

		/*
		 * If journal replay hasn't finished, we might be seeing keys
		 * that will be overwritten by the time journal replay is done:
		 */
		if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
			continue;

		stale = 0;

		do {
			struct bucket_mark mark;

			seq = read_seqcount_begin(&c->gc_pos_lock);
			mark = READ_ONCE(g->mark);

			/* between mark and bucket gen */
			smp_rmb();

			stale = ptr_stale(ca, ptr);

			bch_fs_bug_on(stale && !ptr->cached, c,
					 "stale dirty pointer");

			bch_fs_bug_on(stale > 96, c,
					 "key too stale: %i",
					 stale);

			if (stale)
				break;

			bad = (mark.data_type != BUCKET_DATA ||
			       (gc_pos_cmp(c->gc_pos, gc_pos_btree_node(b)) > 0 &&
				!mark.owned_by_allocator &&
				!(ptr->cached
				  ? mark.cached_sectors
				  : mark.dirty_sectors)));
		} while (read_seqcount_retry(&c->gc_pos_lock, seq));

		if (bad)
			goto bad_ptr;
	}

	if (replicas > BCH_REPLICAS_MAX) {
		bch_bkey_val_to_text(c, btree_node_type(b), buf,
				     sizeof(buf), e.s_c);
		bch_fs_bug(c,
			"extent key bad (too many replicas: %u): %s",
			replicas, buf);
		return;
	}

	if (!bkey_extent_is_cached(e.k) &&
	    replicas < c->sb.data_replicas_have) {
		bch_bkey_val_to_text(c, btree_node_type(b), buf,
				     sizeof(buf), e.s_c);
		bch_fs_bug(c,
			"extent key bad (too few replicas, %u < %u): %s",
			replicas, c->sb.data_replicas_have, buf);
		return;
	}

	return;

bad_ptr:
	bch_bkey_val_to_text(c, btree_node_type(b), buf,
			     sizeof(buf), e.s_c);
	bch_fs_bug(c, "extent pointer bad gc mark: %s:\nbucket %zu prio %i "
		   "gen %i last_gc %i mark 0x%08x",
		   buf, PTR_BUCKET_NR(ca, ptr),
		   g->read_prio, PTR_BUCKET(ca, ptr)->mark.gen,
		   ca->oldest_gens[PTR_BUCKET_NR(ca, ptr)],
		   (unsigned) g->mark.counter);
	return;
}

static void bch_extent_debugcheck(struct bch_fs *c, struct btree *b,
				  struct bkey_s_c k)
{
	switch (k.k->type) {
	case BCH_EXTENT:
	case BCH_EXTENT_CACHED:
		bch_extent_debugcheck_extent(c, b, bkey_s_c_to_extent(k));
		break;
	case BCH_RESERVATION:
		break;
	default:
		BUG();
	}
}

static void bch_extent_to_text(struct bch_fs *c, char *buf,
			       size_t size, struct bkey_s_c k)
{
	char *out = buf, *end = buf + size;
	const char *invalid;

#define p(...)	(out += scnprintf(out, end - out, __VA_ARGS__))

	if (bkey_extent_is_data(k.k))
		out += extent_print_ptrs(c, buf, size, bkey_s_c_to_extent(k));

	invalid = bch_extent_invalid(c, k);
	if (invalid)
		p(" invalid: %s", invalid);
#undef p
}

static unsigned PTR_TIER(struct bch_fs *c,
			 const struct bch_extent_ptr *ptr)
{
	return c->devs[ptr->dev]->mi.tier;
}

static void bch_extent_crc_init(union bch_extent_crc *crc,
				unsigned compressed_size,
				unsigned uncompressed_size,
				unsigned compression_type,
				unsigned nonce,
				struct bch_csum csum, unsigned csum_type)
{
	if (bch_crc_bytes[csum_type]	<= 4 &&
	    uncompressed_size		<= CRC32_SIZE_MAX &&
	    nonce			<= CRC32_NONCE_MAX) {
		crc->crc32 = (struct bch_extent_crc32) {
			.type = 1 << BCH_EXTENT_ENTRY_crc32,
			._compressed_size	= compressed_size - 1,
			._uncompressed_size	= uncompressed_size - 1,
			.offset			= 0,
			.compression_type	= compression_type,
			.csum_type		= csum_type,
			.csum			= *((__le32 *) &csum.lo),
		};
		return;
	}

	if (bch_crc_bytes[csum_type]	<= 10 &&
	    uncompressed_size		<= CRC64_SIZE_MAX &&
	    nonce			<= CRC64_NONCE_MAX) {
		crc->crc64 = (struct bch_extent_crc64) {
			.type = 1 << BCH_EXTENT_ENTRY_crc64,
			._compressed_size	= compressed_size - 1,
			._uncompressed_size	= uncompressed_size - 1,
			.offset			= 0,
			.nonce			= nonce,
			.compression_type	= compression_type,
			.csum_type		= csum_type,
			.csum_lo		= csum.lo,
			.csum_hi		= *((__le16 *) &csum.hi),
		};
		return;
	}

	if (bch_crc_bytes[csum_type]	<= 16 &&
	    uncompressed_size		<= CRC128_SIZE_MAX &&
	    nonce			<= CRC128_NONCE_MAX) {
		crc->crc128 = (struct bch_extent_crc128) {
			.type = 1 << BCH_EXTENT_ENTRY_crc128,
			._compressed_size	= compressed_size - 1,
			._uncompressed_size	= uncompressed_size - 1,
			.offset			= 0,
			.nonce			= nonce,
			.compression_type	= compression_type,
			.csum_type		= csum_type,
			.csum			= csum,
		};
		return;
	}

	BUG();
}

void bch_extent_crc_append(struct bkey_i_extent *e,
			   unsigned compressed_size,
			   unsigned uncompressed_size,
			   unsigned compression_type,
			   unsigned nonce,
			   struct bch_csum csum, unsigned csum_type)
{
	union bch_extent_crc *crc;

	BUG_ON(compressed_size > uncompressed_size);
	BUG_ON(uncompressed_size != e->k.size);
	BUG_ON(!compressed_size || !uncompressed_size);

	/*
	 * Look up the last crc entry, so we can check if we need to add
	 * another:
	 */
	extent_for_each_crc(extent_i_to_s(e), crc)
		;

	if (!crc && !csum_type && !compression_type)
		return;

	if (crc &&
	    crc_compressed_size(&e->k, crc)	== compressed_size &&
	    crc_uncompressed_size(&e->k, crc)	== uncompressed_size &&
	    crc_offset(crc)			== 0 &&
	    crc_nonce(crc)			== nonce &&
	    crc_csum_type(crc)			== csum_type &&
	    crc_compression_type(crc)		== compression_type &&
	    crc_csum(crc).lo			== csum.lo &&
	    crc_csum(crc).hi			== csum.hi)
		return;

	bch_extent_crc_init((void *) extent_entry_last(extent_i_to_s(e)),
			    compressed_size,
			    uncompressed_size,
			    compression_type,
			    nonce, csum, csum_type);
	__extent_entry_push(e);
}

/*
 * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
 *
 * Returns true if @k should be dropped entirely
 *
 * For existing keys, only called when btree nodes are being rewritten, not when
 * they're merely being compacted/resorted in memory.
 */
bool bch_extent_normalize(struct bch_fs *c, struct bkey_s k)
{
	struct bkey_s_extent e;

	switch (k.k->type) {
	case KEY_TYPE_ERROR:
		return false;

	case KEY_TYPE_DELETED:
	case KEY_TYPE_COOKIE:
		return true;

	case KEY_TYPE_DISCARD:
		return bversion_zero(k.k->version);

	case BCH_EXTENT:
	case BCH_EXTENT_CACHED:
		e = bkey_s_to_extent(k);

		bch_extent_drop_stale(c, e);

		if (!bkey_val_u64s(e.k)) {
			if (bkey_extent_is_cached(e.k)) {
				k.k->type = KEY_TYPE_DISCARD;
				if (bversion_zero(k.k->version))
					return true;
			} else {
				k.k->type = KEY_TYPE_ERROR;
			}
		}

		return false;
	case BCH_RESERVATION:
		return false;
	default:
		BUG();
	}
}

void bch_extent_mark_replicas_cached(struct bch_fs *c,
				     struct bkey_s_extent e,
				     unsigned nr_cached)
{
	struct bch_extent_ptr *ptr;
	bool have_higher_tier;
	unsigned tier = 0;

	if (!nr_cached)
		return;

	do {
		have_higher_tier = false;

		extent_for_each_ptr(e, ptr) {
			if (!ptr->cached &&
			    PTR_TIER(c, ptr) == tier) {
				ptr->cached = true;
				nr_cached--;
				if (!nr_cached)
					return;
			}

			if (PTR_TIER(c, ptr) > tier)
				have_higher_tier = true;
		}

		tier++;
	} while (have_higher_tier);
}

/*
 * This picks a non-stale pointer, preferabbly from a device other than
 * avoid.  Avoid can be NULL, meaning pick any.  If there are no non-stale
 * pointers to other devices, it will still pick a pointer from avoid.
 * Note that it prefers lowered-numbered pointers to higher-numbered pointers
 * as the pointers are sorted by tier, hence preferring pointers to tier 0
 * rather than pointers to tier 1.
 */
void bch_extent_pick_ptr_avoiding(struct bch_fs *c, struct bkey_s_c k,
				  struct bch_dev *avoid,
				  struct extent_pick_ptr *ret)
{
	struct bkey_s_c_extent e;
	const union bch_extent_crc *crc;
	const struct bch_extent_ptr *ptr;

	switch (k.k->type) {
	case KEY_TYPE_DELETED:
	case KEY_TYPE_DISCARD:
	case KEY_TYPE_COOKIE:
		ret->ca = NULL;
		return;

	case KEY_TYPE_ERROR:
		ret->ca = ERR_PTR(-EIO);
		return;

	case BCH_EXTENT:
	case BCH_EXTENT_CACHED:
		e = bkey_s_c_to_extent(k);
		ret->ca = NULL;

		extent_for_each_ptr_crc(e, ptr, crc) {
			struct bch_dev *ca = c->devs[ptr->dev];

			if (ptr_stale(ca, ptr))
				continue;

			if (ca->mi.state == BCH_MEMBER_STATE_FAILED)
				continue;

			if (ret->ca &&
			    (ca == avoid ||
			     ret->ca->mi.tier < ca->mi.tier))
				continue;

			if (!percpu_ref_tryget(&ca->io_ref))
				continue;

			if (ret->ca)
				percpu_ref_put(&ret->ca->io_ref);

			*ret = (struct extent_pick_ptr) {
				.crc = crc_to_128(e.k, crc),
				.ptr = *ptr,
				.ca = ca,
			};
		}

		if (!ret->ca && !bkey_extent_is_cached(e.k))
			ret->ca = ERR_PTR(-EIO);
		return;

	case BCH_RESERVATION:
		ret->ca = NULL;
		return;

	default:
		BUG();
	}
}

static enum merge_result bch_extent_merge(struct bch_fs *c,
					  struct btree *bk,
					  struct bkey_i *l, struct bkey_i *r)
{
	struct bkey_s_extent el, er;
	union bch_extent_entry *en_l, *en_r;

	if (key_merging_disabled(c))
		return BCH_MERGE_NOMERGE;

	/*
	 * Generic header checks
	 * Assumes left and right are in order
	 * Left and right must be exactly aligned
	 */

	if (l->k.u64s		!= r->k.u64s ||
	    l->k.type		!= r->k.type ||
	    bversion_cmp(l->k.version, r->k.version) ||
	    bkey_cmp(l->k.p, bkey_start_pos(&r->k)))
		return BCH_MERGE_NOMERGE;

	switch (l->k.type) {
	case KEY_TYPE_DELETED:
	case KEY_TYPE_DISCARD:
	case KEY_TYPE_ERROR:
		/* These types are mergeable, and no val to check */
		break;

	case BCH_EXTENT:
	case BCH_EXTENT_CACHED:
		el = bkey_i_to_s_extent(l);
		er = bkey_i_to_s_extent(r);

		extent_for_each_entry(el, en_l) {
			struct bch_extent_ptr *lp, *rp;
			unsigned bucket_size;

			en_r = vstruct_idx(er.v, (u64 *) en_l - el.v->_data);

			if ((extent_entry_type(en_l) !=
			     extent_entry_type(en_r)) ||
			    extent_entry_is_crc(en_l))
				return BCH_MERGE_NOMERGE;

			lp = &en_l->ptr;
			rp = &en_r->ptr;

			if (lp->offset + el.k->size	!= rp->offset ||
			    lp->dev			!= rp->dev ||
			    lp->gen			!= rp->gen)
				return BCH_MERGE_NOMERGE;

			/* We don't allow extents to straddle buckets: */
			bucket_size = c->devs[lp->dev]->mi.bucket_size;

			if ((lp->offset & ~((u64) bucket_size - 1)) !=
			    (rp->offset & ~((u64) bucket_size - 1)))
				return BCH_MERGE_NOMERGE;
		}

		break;
	case BCH_RESERVATION: {
		struct bkey_i_reservation *li = bkey_i_to_reservation(l);
		struct bkey_i_reservation *ri = bkey_i_to_reservation(r);

		if (li->v.generation != ri->v.generation ||
		    li->v.nr_replicas != ri->v.nr_replicas)
			return BCH_MERGE_NOMERGE;
		break;
	}
	default:
		return BCH_MERGE_NOMERGE;
	}

	l->k.needs_whiteout |= r->k.needs_whiteout;

	/* Keys with no pointers aren't restricted to one bucket and could
	 * overflow KEY_SIZE
	 */
	if ((u64) l->k.size + r->k.size > KEY_SIZE_MAX) {
		bch_key_resize(&l->k, KEY_SIZE_MAX);
		bch_cut_front(l->k.p, r);
		return BCH_MERGE_PARTIAL;
	}

	bch_key_resize(&l->k, l->k.size + r->k.size);

	return BCH_MERGE_MERGE;
}

static void extent_i_save(struct btree *b, struct bkey_packed *dst,
			  struct bkey_i *src)
{
	struct bkey_format *f = &b->format;
	struct bkey_i *dst_unpacked;

	BUG_ON(bkeyp_val_u64s(f, dst) != bkey_val_u64s(&src->k));

	/*
	 * We don't want the bch_verify_key_order() call in extent_save(),
	 * because we may be out of order with deleted keys that are about to be
	 * removed by extent_bset_insert()
	 */

	if ((dst_unpacked = packed_to_bkey(dst)))
		bkey_copy(dst_unpacked, src);
	else
		BUG_ON(!bkey_pack(dst, src, f));
}

static bool extent_merge_one_overlapping(struct btree_iter *iter,
					 struct bpos new_pos,
					 struct bset_tree *t,
					 struct bkey_packed *k, struct bkey uk,
					 bool check, bool could_pack)
{
	struct btree *b = iter->nodes[0];
	struct btree_node_iter *node_iter = &iter->node_iters[0];

	BUG_ON(!bkey_deleted(k));

	if (check) {
		return !bkey_packed(k) || could_pack;
	} else {
		uk.p = new_pos;
		extent_save(b, node_iter, k, &uk);
		bch_bset_fix_invalidated_key(b, t, k);
		bch_btree_node_iter_fix(iter, b, node_iter, t,
					k, k->u64s, k->u64s);
		return true;
	}
}

static bool extent_merge_do_overlapping(struct btree_iter *iter,
					struct bkey *m, bool back_merge)
{
	struct btree *b = iter->nodes[0];
	struct btree_node_iter *node_iter = &iter->node_iters[0];
	struct bset_tree *t;
	struct bkey_packed *k;
	struct bkey uk;
	struct bpos new_pos = back_merge ? m->p : bkey_start_pos(m);
	bool could_pack = bkey_pack_pos((void *) &uk, new_pos, b);
	bool check = true;

	/*
	 * @m is the new merged extent:
	 *
	 * The merge took place in the last bset; we know there can't be any 0
	 * size extents overlapping with m there because if so they would have
	 * been between the two extents we merged.
	 *
	 * But in the other bsets, we have to check for and fix such extents:
	 */
do_fixup:
	for_each_bset(b, t) {
		if (t == bset_tree_last(b))
			break;

		/*
		 * if we don't find this bset in the iterator we already got to
		 * the end of that bset, so start searching from the end.
		 */
		k = bch_btree_node_iter_bset_pos(node_iter, b, t);

		if (k == btree_bkey_last(b, t))
			k = bkey_prev_all(b, t, k);
		if (!k)
			continue;

		if (back_merge) {
			/*
			 * Back merge: 0 size extents will be before the key
			 * that was just inserted (and thus the iterator
			 * position) - walk backwards to find them
			 */
			for (;
			     k &&
			     (uk = bkey_unpack_key(b, k),
			      bkey_cmp(uk.p, bkey_start_pos(m)) > 0);
			     k = bkey_prev_all(b, t, k)) {
				if (bkey_cmp(uk.p, m->p) >= 0)
					continue;

				if (!extent_merge_one_overlapping(iter, new_pos,
						t, k, uk, check, could_pack))
					return false;
			}
		} else {
			/* Front merge - walk forwards */
			for (;
			     k != btree_bkey_last(b, t) &&
			     (uk = bkey_unpack_key(b, k),
			      bkey_cmp(uk.p, m->p) < 0);
			     k = bkey_next(k)) {
				if (bkey_cmp(uk.p,
					     bkey_start_pos(m)) <= 0)
					continue;

				if (!extent_merge_one_overlapping(iter, new_pos,
						t, k, uk, check, could_pack))
					return false;
			}
		}
	}

	if (check) {
		check = false;
		goto do_fixup;
	}

	return true;
}

/*
 * When merging an extent that we're inserting into a btree node, the new merged
 * extent could overlap with an existing 0 size extent - if we don't fix that,
 * it'll break the btree node iterator so this code finds those 0 size extents
 * and shifts them out of the way.
 *
 * Also unpacks and repacks.
 */
static bool bch_extent_merge_inline(struct bch_fs *c,
				    struct btree_iter *iter,
				    struct bkey_packed *l,
				    struct bkey_packed *r,
				    bool back_merge)
{
	struct btree *b = iter->nodes[0];
	struct btree_node_iter *node_iter = &iter->node_iters[0];
	const struct bkey_format *f = &b->format;
	struct bset_tree *t = bset_tree_last(b);
	struct bkey_packed *m;
	BKEY_PADDED(k) li;
	BKEY_PADDED(k) ri;
	struct bkey_i *mi;
	struct bkey tmp;

	/*
	 * We need to save copies of both l and r, because we might get a
	 * partial merge (which modifies both) and then fails to repack
	 */
	bkey_unpack(b, &li.k, l);
	bkey_unpack(b, &ri.k, r);

	m = back_merge ? l : r;
	mi = back_merge ? &li.k : &ri.k;

	/* l & r should be in last bset: */
	EBUG_ON(bch_bkey_to_bset(b, m) != t);

	switch (bch_extent_merge(c, b, &li.k, &ri.k)) {
	case BCH_MERGE_NOMERGE:
		return false;
	case BCH_MERGE_PARTIAL:
		if (bkey_packed(m) && !bkey_pack_key((void *) &tmp, &mi->k, f))
			return false;

		if (!extent_merge_do_overlapping(iter, &li.k.k, back_merge))
			return false;

		extent_i_save(b, m, mi);
		bch_bset_fix_invalidated_key(b, t, m);

		/*
		 * Update iterator to reflect what we just inserted - otherwise,
		 * the iter_fix() call is going to put us _before_ the key we
		 * just partially merged with:
		 */
		if (back_merge)
			bch_btree_iter_set_pos_same_leaf(iter, li.k.k.p);

		bch_btree_node_iter_fix(iter, iter->nodes[0], node_iter,
					t, m, m->u64s, m->u64s);

		if (!back_merge)
			bkey_copy(packed_to_bkey(l), &li.k);
		else
			bkey_copy(packed_to_bkey(r), &ri.k);
		return false;
	case BCH_MERGE_MERGE:
		if (bkey_packed(m) && !bkey_pack_key((void *) &tmp, &li.k.k, f))
			return false;

		if (!extent_merge_do_overlapping(iter, &li.k.k, back_merge))
			return false;

		extent_i_save(b, m, &li.k);
		bch_bset_fix_invalidated_key(b, t, m);

		bch_btree_node_iter_fix(iter, iter->nodes[0], node_iter,
					t, m, m->u64s, m->u64s);
		return true;
	default:
		BUG();
	}
}

const struct bkey_ops bch_bkey_extent_ops = {
	.key_invalid	= bch_extent_invalid,
	.key_debugcheck	= bch_extent_debugcheck,
	.val_to_text	= bch_extent_to_text,
	.swab		= bch_ptr_swab,
	.key_normalize	= bch_ptr_normalize,
	.key_merge	= bch_extent_merge,
	.is_extents	= true,
};