aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/usb/host/ohci-hcd.c
blob: 89901962cbfd821bc454353f61ed1e4e66ecf8db (plain) (blame)
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
/*
 * OHCI HCD (Host Controller Driver) for USB.
 *
 * (C) Copyright 1999 Roman Weissgaerber <weissg@vienna.at>
 * (C) Copyright 2000-2004 David Brownell <dbrownell@users.sourceforge.net>
 *
 * [ Initialisation is based on Linus'  ]
 * [ uhci code and gregs ohci fragments ]
 * [ (C) Copyright 1999 Linus Torvalds  ]
 * [ (C) Copyright 1999 Gregory P. Smith]
 *
 *
 * OHCI is the main "non-Intel/VIA" standard for USB 1.1 host controller
 * interfaces (though some non-x86 Intel chips use it).  It supports
 * smarter hardware than UHCI.  A download link for the spec available
 * through the http://www.usb.org website.
 *
 * This file is licenced under the GPL.
 */

#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/pci.h>
#include <linux/kernel.h>
#include <linux/delay.h>
#include <linux/ioport.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/errno.h>
#include <linux/init.h>
#include <linux/timer.h>
#include <linux/list.h>
#include <linux/usb.h>
#include <linux/usb/otg.h>
#include <linux/dma-mapping.h>
#include <linux/dmapool.h>
#include <linux/reboot.h>
#include <linux/workqueue.h>
#include <linux/debugfs.h>

#include <asm/io.h>
#include <asm/irq.h>
#include <asm/system.h>
#include <asm/unaligned.h>
#include <asm/byteorder.h>

#include "../core/hcd.h"

#define DRIVER_VERSION "2006 August 04"
#define DRIVER_AUTHOR "Roman Weissgaerber, David Brownell"
#define DRIVER_DESC "USB 1.1 'Open' Host Controller (OHCI) Driver"

/*-------------------------------------------------------------------------*/

#undef OHCI_VERBOSE_DEBUG	/* not always helpful */

/* For initializing controller (mask in an HCFS mode too) */
#define	OHCI_CONTROL_INIT	OHCI_CTRL_CBSR
#define	OHCI_INTR_INIT \
		(OHCI_INTR_MIE | OHCI_INTR_RHSC | OHCI_INTR_UE \
		| OHCI_INTR_RD | OHCI_INTR_WDH)

#ifdef __hppa__
/* On PA-RISC, PDC can leave IR set incorrectly; ignore it there. */
#define	IR_DISABLE
#endif

#ifdef CONFIG_ARCH_OMAP
/* OMAP doesn't support IR (no SMM; not needed) */
#define	IR_DISABLE
#endif

/*-------------------------------------------------------------------------*/

static const char	hcd_name [] = "ohci_hcd";

#define	STATECHANGE_DELAY	msecs_to_jiffies(300)

#include "ohci.h"

static void ohci_dump (struct ohci_hcd *ohci, int verbose);
static int ohci_init (struct ohci_hcd *ohci);
static void ohci_stop (struct usb_hcd *hcd);

#if defined(CONFIG_PM) || defined(CONFIG_PCI)
static int ohci_restart (struct ohci_hcd *ohci);
#endif

#ifdef CONFIG_PCI
static void quirk_amd_pll(int state);
static void amd_iso_dev_put(void);
#else
static inline void quirk_amd_pll(int state)
{
	return;
}
static inline void amd_iso_dev_put(void)
{
	return;
}
#endif


#include "ohci-hub.c"
#include "ohci-dbg.c"
#include "ohci-mem.c"
#include "ohci-q.c"


/*
 * On architectures with edge-triggered interrupts we must never return
 * IRQ_NONE.
 */
#if defined(CONFIG_SA1111)  /* ... or other edge-triggered systems */
#define IRQ_NOTMINE	IRQ_HANDLED
#else
#define IRQ_NOTMINE	IRQ_NONE
#endif


/* Some boards misreport power switching/overcurrent */
static int distrust_firmware = 1;
module_param (distrust_firmware, bool, 0);
MODULE_PARM_DESC (distrust_firmware,
	"true to distrust firmware power/overcurrent setup");

/* Some boards leave IR set wrongly, since they fail BIOS/SMM handshakes */
static int no_handshake = 0;
module_param (no_handshake, bool, 0);
MODULE_PARM_DESC (no_handshake, "true (not default) disables BIOS handshake");

/*-------------------------------------------------------------------------*/

/*
 * queue up an urb for anything except the root hub
 */
static int ohci_urb_enqueue (
	struct usb_hcd	*hcd,
	struct urb	*urb,
	gfp_t		mem_flags
) {
	struct ohci_hcd	*ohci = hcd_to_ohci (hcd);
	struct ed	*ed;
	urb_priv_t	*urb_priv;
	unsigned int	pipe = urb->pipe;
	int		i, size = 0;
	unsigned long	flags;
	int		retval = 0;

#ifdef OHCI_VERBOSE_DEBUG
	urb_print(urb, "SUB", usb_pipein(pipe), -EINPROGRESS);
#endif

	/* every endpoint has a ed, locate and maybe (re)initialize it */
	if (! (ed = ed_get (ohci, urb->ep, urb->dev, pipe, urb->interval)))
		return -ENOMEM;

	/* for the private part of the URB we need the number of TDs (size) */
	switch (ed->type) {
		case PIPE_CONTROL:
			/* td_submit_urb() doesn't yet handle these */
			if (urb->transfer_buffer_length > 4096)
				return -EMSGSIZE;

			/* 1 TD for setup, 1 for ACK, plus ... */
			size = 2;
			/* FALLTHROUGH */
		// case PIPE_INTERRUPT:
		// case PIPE_BULK:
		default:
			/* one TD for every 4096 Bytes (can be upto 8K) */
			size += urb->transfer_buffer_length / 4096;
			/* ... and for any remaining bytes ... */
			if ((urb->transfer_buffer_length % 4096) != 0)
				size++;
			/* ... and maybe a zero length packet to wrap it up */
			if (size == 0)
				size++;
			else if ((urb->transfer_flags & URB_ZERO_PACKET) != 0
				&& (urb->transfer_buffer_length
					% usb_maxpacket (urb->dev, pipe,
						usb_pipeout (pipe))) == 0)
				size++;
			break;
		case PIPE_ISOCHRONOUS: /* number of packets from URB */
			size = urb->number_of_packets;
			break;
	}

	/* allocate the private part of the URB */
	urb_priv = kzalloc (sizeof (urb_priv_t) + size * sizeof (struct td *),
			mem_flags);
	if (!urb_priv)
		return -ENOMEM;
	INIT_LIST_HEAD (&urb_priv->pending);
	urb_priv->length = size;
	urb_priv->ed = ed;

	/* allocate the TDs (deferring hash chain updates) */
	for (i = 0; i < size; i++) {
		urb_priv->td [i] = td_alloc (ohci, mem_flags);
		if (!urb_priv->td [i]) {
			urb_priv->length = i;
			urb_free_priv (ohci, urb_priv);
			return -ENOMEM;
		}
	}

	spin_lock_irqsave (&ohci->lock, flags);

	/* don't submit to a dead HC */
	if (!test_bit(HCD_FLAG_HW_ACCESSIBLE, &hcd->flags)) {
		retval = -ENODEV;
		goto fail;
	}
	if (!HC_IS_RUNNING(hcd->state)) {
		retval = -ENODEV;
		goto fail;
	}
	retval = usb_hcd_link_urb_to_ep(hcd, urb);
	if (retval)
		goto fail;

	/* schedule the ed if needed */
	if (ed->state == ED_IDLE) {
		retval = ed_schedule (ohci, ed);
		if (retval < 0) {
			usb_hcd_unlink_urb_from_ep(hcd, urb);
			goto fail;
		}
		if (ed->type == PIPE_ISOCHRONOUS) {
			u16	frame = ohci_frame_no(ohci);

			/* delay a few frames before the first TD */
			frame += max_t (u16, 8, ed->interval);
			frame &= ~(ed->interval - 1);
			frame |= ed->branch;
			urb->start_frame = frame;

			/* yes, only URB_ISO_ASAP is supported, and
			 * urb->start_frame is never used as input.
			 */
		}
	} else if (ed->type == PIPE_ISOCHRONOUS)
		urb->start_frame = ed->last_iso + ed->interval;

	/* fill the TDs and link them to the ed; and
	 * enable that part of the schedule, if needed
	 * and update count of queued periodic urbs
	 */
	urb->hcpriv = urb_priv;
	td_submit_urb (ohci, urb);

fail:
	if (retval)
		urb_free_priv (ohci, urb_priv);
	spin_unlock_irqrestore (&ohci->lock, flags);
	return retval;
}

/*
 * decouple the URB from the HC queues (TDs, urb_priv).
 * reporting is always done
 * asynchronously, and we might be dealing with an urb that's
 * partially transferred, or an ED with other urbs being unlinked.
 */
static int ohci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
{
	struct ohci_hcd		*ohci = hcd_to_ohci (hcd);
	unsigned long		flags;
	int			rc;

#ifdef OHCI_VERBOSE_DEBUG
	urb_print(urb, "UNLINK", 1, status);
#endif

	spin_lock_irqsave (&ohci->lock, flags);
	rc = usb_hcd_check_unlink_urb(hcd, urb, status);
	if (rc) {
		;	/* Do nothing */
	} else if (HC_IS_RUNNING(hcd->state)) {
		urb_priv_t  *urb_priv;

		/* Unless an IRQ completed the unlink while it was being
		 * handed to us, flag it for unlink and giveback, and force
		 * some upcoming INTR_SF to call finish_unlinks()
		 */
		urb_priv = urb->hcpriv;
		if (urb_priv) {
			if (urb_priv->ed->state == ED_OPER)
				start_ed_unlink (ohci, urb_priv->ed);
		}
	} else {
		/*
		 * with HC dead, we won't respect hc queue pointers
		 * any more ... just clean up every urb's memory.
		 */
		if (urb->hcpriv)
			finish_urb(ohci, urb, status);
	}
	spin_unlock_irqrestore (&ohci->lock, flags);
	return rc;
}

/*-------------------------------------------------------------------------*/

/* frees config/altsetting state for endpoints,
 * including ED memory, dummy TD, and bulk/intr data toggle
 */

static void
ohci_endpoint_disable (struct usb_hcd *hcd, struct usb_host_endpoint *ep)
{
	struct ohci_hcd		*ohci = hcd_to_ohci (hcd);
	unsigned long		flags;
	struct ed		*ed = ep->hcpriv;
	unsigned		limit = 1000;

	/* ASSERT:  any requests/urbs are being unlinked */
	/* ASSERT:  nobody can be submitting urbs for this any more */

	if (!ed)
		return;

rescan:
	spin_lock_irqsave (&ohci->lock, flags);

	if (!HC_IS_RUNNING (hcd->state)) {
sanitize:
		ed->state = ED_IDLE;
		if (quirk_zfmicro(ohci) && ed->type == PIPE_INTERRUPT)
			ohci->eds_scheduled--;
		finish_unlinks (ohci, 0);
	}

	switch (ed->state) {
	case ED_UNLINK:		/* wait for hw to finish? */
		/* major IRQ delivery trouble loses INTR_SF too... */
		if (limit-- == 0) {
			ohci_warn(ohci, "ED unlink timeout\n");
			if (quirk_zfmicro(ohci)) {
				ohci_warn(ohci, "Attempting ZF TD recovery\n");
				ohci->ed_to_check = ed;
				ohci->zf_delay = 2;
			}
			goto sanitize;
		}
		spin_unlock_irqrestore (&ohci->lock, flags);
		schedule_timeout_uninterruptible(1);
		goto rescan;
	case ED_IDLE:		/* fully unlinked */
		if (list_empty (&ed->td_list)) {
			td_free (ohci, ed->dummy);
			ed_free (ohci, ed);
			break;
		}
		/* else FALL THROUGH */
	default:
		/* caller was supposed to have unlinked any requests;
		 * that's not our job.  can't recover; must leak ed.
		 */
		ohci_err (ohci, "leak ed %p (#%02x) state %d%s\n",
			ed, ep->desc.bEndpointAddress, ed->state,
			list_empty (&ed->td_list) ? "" : " (has tds)");
		td_free (ohci, ed->dummy);
		break;
	}
	ep->hcpriv = NULL;
	spin_unlock_irqrestore (&ohci->lock, flags);
	return;
}

static int ohci_get_frame (struct usb_hcd *hcd)
{
	struct ohci_hcd		*ohci = hcd_to_ohci (hcd);

	return ohci_frame_no(ohci);
}

static void ohci_usb_reset (struct ohci_hcd *ohci)
{
	ohci->hc_control = ohci_readl (ohci, &ohci->regs->control);
	ohci->hc_control &= OHCI_CTRL_RWC;
	ohci_writel (ohci, ohci->hc_control, &ohci->regs->control);
}

/* ohci_shutdown forcibly disables IRQs and DMA, helping kexec and
 * other cases where the next software may expect clean state from the
 * "firmware".  this is bus-neutral, unlike shutdown() methods.
 */
static void
ohci_shutdown (struct usb_hcd *hcd)
{
	struct ohci_hcd *ohci;

	ohci = hcd_to_ohci (hcd);
	ohci_writel (ohci, OHCI_INTR_MIE, &ohci->regs->intrdisable);
	ohci_usb_reset (ohci);
	/* flush the writes */
	(void) ohci_readl (ohci, &ohci->regs->control);
}

static int check_ed(struct ohci_hcd *ohci, struct ed *ed)
{
	return (hc32_to_cpu(ohci, ed->hwINFO) & ED_IN) != 0
		&& (hc32_to_cpu(ohci, ed->hwHeadP) & TD_MASK)
			== (hc32_to_cpu(ohci, ed->hwTailP) & TD_MASK)
		&& !list_empty(&ed->td_list);
}

/* ZF Micro watchdog timer callback. The ZF Micro chipset sometimes completes
 * an interrupt TD but neglects to add it to the donelist.  On systems with
 * this chipset, we need to periodically check the state of the queues to look
 * for such "lost" TDs.
 */
static void unlink_watchdog_func(unsigned long _ohci)
{
	unsigned long	flags;
	unsigned	max;
	unsigned	seen_count = 0;
	unsigned	i;
	struct ed	**seen = NULL;
	struct ohci_hcd	*ohci = (struct ohci_hcd *) _ohci;

	spin_lock_irqsave(&ohci->lock, flags);
	max = ohci->eds_scheduled;
	if (!max)
		goto done;

	if (ohci->ed_to_check)
		goto out;

	seen = kcalloc(max, sizeof *seen, GFP_ATOMIC);
	if (!seen)
		goto out;

	for (i = 0; i < NUM_INTS; i++) {
		struct ed	*ed = ohci->periodic[i];

		while (ed) {
			unsigned	temp;

			/* scan this branch of the periodic schedule tree */
			for (temp = 0; temp < seen_count; temp++) {
				if (seen[temp] == ed) {
					/* we've checked it and what's after */
					ed = NULL;
					break;
				}
			}
			if (!ed)
				break;
			seen[seen_count++] = ed;
			if (!check_ed(ohci, ed)) {
				ed = ed->ed_next;
				continue;
			}

			/* HC's TD list is empty, but HCD sees at least one
			 * TD that's not been sent through the donelist.
			 */
			ohci->ed_to_check = ed;
			ohci->zf_delay = 2;

			/* The HC may wait until the next frame to report the
			 * TD as done through the donelist and INTR_WDH.  (We
			 * just *assume* it's not a multi-TD interrupt URB;
			 * those could defer the IRQ more than one frame, using
			 * DI...)  Check again after the next INTR_SF.
			 */
			ohci_writel(ohci, OHCI_INTR_SF,
					&ohci->regs->intrstatus);
			ohci_writel(ohci, OHCI_INTR_SF,
					&ohci->regs->intrenable);

			/* flush those writes */
			(void) ohci_readl(ohci, &ohci->regs->control);

			goto out;
		}
	}
out:
	kfree(seen);
	if (ohci->eds_scheduled)
		mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));
done:
	spin_unlock_irqrestore(&ohci->lock, flags);
}

/*-------------------------------------------------------------------------*
 * HC functions
 *-------------------------------------------------------------------------*/

/* init memory, and kick BIOS/SMM off */

static int ohci_init (struct ohci_hcd *ohci)
{
	int ret;
	struct usb_hcd *hcd = ohci_to_hcd(ohci);

	if (distrust_firmware)
		ohci->flags |= OHCI_QUIRK_HUB_POWER;

	disable (ohci);
	ohci->regs = hcd->regs;

	/* REVISIT this BIOS handshake is now moved into PCI "quirks", and
	 * was never needed for most non-PCI systems ... remove the code?
	 */

#ifndef IR_DISABLE
	/* SMM owns the HC?  not for long! */
	if (!no_handshake && ohci_readl (ohci,
					&ohci->regs->control) & OHCI_CTRL_IR) {
		u32 temp;

		ohci_dbg (ohci, "USB HC TakeOver from BIOS/SMM\n");

		/* this timeout is arbitrary.  we make it long, so systems
		 * depending on usb keyboards may be usable even if the
		 * BIOS/SMM code seems pretty broken.
		 */
		temp = 500;	/* arbitrary: five seconds */

		ohci_writel (ohci, OHCI_INTR_OC, &ohci->regs->intrenable);
		ohci_writel (ohci, OHCI_OCR, &ohci->regs->cmdstatus);
		while (ohci_readl (ohci, &ohci->regs->control) & OHCI_CTRL_IR) {
			msleep (10);
			if (--temp == 0) {
				ohci_err (ohci, "USB HC takeover failed!"
					"  (BIOS/SMM bug)\n");
				return -EBUSY;
			}
		}
		ohci_usb_reset (ohci);
	}
#endif

	/* Disable HC interrupts */
	ohci_writel (ohci, OHCI_INTR_MIE, &ohci->regs->intrdisable);

	/* flush the writes, and save key bits like RWC */
	if (ohci_readl (ohci, &ohci->regs->control) & OHCI_CTRL_RWC)
		ohci->hc_control |= OHCI_CTRL_RWC;

	/* Read the number of ports unless overridden */
	if (ohci->num_ports == 0)
		ohci->num_ports = roothub_a(ohci) & RH_A_NDP;

	if (ohci->hcca)
		return 0;

	ohci->hcca = dma_alloc_coherent (hcd->self.controller,
			sizeof *ohci->hcca, &ohci->hcca_dma, 0);
	if (!ohci->hcca)
		return -ENOMEM;

	if ((ret = ohci_mem_init (ohci)) < 0)
		ohci_stop (hcd);
	else {
		create_debug_files (ohci);
	}

	return ret;
}

/*-------------------------------------------------------------------------*/

/* Start an OHCI controller, set the BUS operational
 * resets USB and controller
 * enable interrupts
 */
static int ohci_run (struct ohci_hcd *ohci)
{
	u32			mask, temp;
	int			first = ohci->fminterval == 0;
	struct usb_hcd		*hcd = ohci_to_hcd(ohci);

	disable (ohci);

	/* boot firmware should have set this up (5.1.1.3.1) */
	if (first) {

		temp = ohci_readl (ohci, &ohci->regs->fminterval);
		ohci->fminterval = temp & 0x3fff;
		if (ohci->fminterval != FI)
			ohci_dbg (ohci, "fminterval delta %d\n",
				ohci->fminterval - FI);
		ohci->fminterval |= FSMP (ohci->fminterval) << 16;
		/* also: power/overcurrent flags in roothub.a */
	}

	/* Reset USB nearly "by the book".  RemoteWakeupConnected was
	 * saved if boot firmware (BIOS/SMM/...) told us it's connected,
	 * or if bus glue did the same (e.g. for PCI add-in cards with
	 * PCI PM support).
	 */
	if ((ohci->hc_control & OHCI_CTRL_RWC) != 0
			&& !device_may_wakeup(hcd->self.controller))
		device_init_wakeup(hcd->self.controller, 1);

	switch (ohci->hc_control & OHCI_CTRL_HCFS) {
	case OHCI_USB_OPER:
		temp = 0;
		break;
	case OHCI_USB_SUSPEND:
	case OHCI_USB_RESUME:
		ohci->hc_control &= OHCI_CTRL_RWC;
		ohci->hc_control |= OHCI_USB_RESUME;
		temp = 10 /* msec wait */;
		break;
	// case OHCI_USB_RESET:
	default:
		ohci->hc_control &= OHCI_CTRL_RWC;
		ohci->hc_control |= OHCI_USB_RESET;
		temp = 50 /* msec wait */;
		break;
	}
	ohci_writel (ohci, ohci->hc_control, &ohci->regs->control);
	// flush the writes
	(void) ohci_readl (ohci, &ohci->regs->control);
	msleep(temp);

	memset (ohci->hcca, 0, sizeof (struct ohci_hcca));

	/* 2msec timelimit here means no irqs/preempt */
	spin_lock_irq (&ohci->lock);

retry:
	/* HC Reset requires max 10 us delay */
	ohci_writel (ohci, OHCI_HCR,  &ohci->regs->cmdstatus);
	temp = 30;	/* ... allow extra time */
	while ((ohci_readl (ohci, &ohci->regs->cmdstatus) & OHCI_HCR) != 0) {
		if (--temp == 0) {
			spin_unlock_irq (&ohci->lock);
			ohci_err (ohci, "USB HC reset timed out!\n");
			return -1;
		}
		udelay (1);
	}

	/* now we're in the SUSPEND state ... must go OPERATIONAL
	 * within 2msec else HC enters RESUME
	 *
	 * ... but some hardware won't init fmInterval "by the book"
	 * (SiS, OPTi ...), so reset again instead.  SiS doesn't need
	 * this if we write fmInterval after we're OPERATIONAL.
	 * Unclear about ALi, ServerWorks, and others ... this could
	 * easily be a longstanding bug in chip init on Linux.
	 */
	if (ohci->flags & OHCI_QUIRK_INITRESET) {
		ohci_writel (ohci, ohci->hc_control, &ohci->regs->control);
		// flush those writes
		(void) ohci_readl (ohci, &ohci->regs->control);
	}

	/* Tell the controller where the control and bulk lists are
	 * The lists are empty now. */
	ohci_writel (ohci, 0, &ohci->regs->ed_controlhead);
	ohci_writel (ohci, 0, &ohci->regs->ed_bulkhead);

	/* a reset clears this */
	ohci_writel (ohci, (u32) ohci->hcca_dma, &ohci->regs->hcca);

	periodic_reinit (ohci);

	/* some OHCI implementations are finicky about how they init.
	 * bogus values here mean not even enumeration could work.
	 */
	if ((ohci_readl (ohci, &ohci->regs->fminterval) & 0x3fff0000) == 0
			|| !ohci_readl (ohci, &ohci->regs->periodicstart)) {
		if (!(ohci->flags & OHCI_QUIRK_INITRESET)) {
			ohci->flags |= OHCI_QUIRK_INITRESET;
			ohci_dbg (ohci, "enabling initreset quirk\n");
			goto retry;
		}
		spin_unlock_irq (&ohci->lock);
		ohci_err (ohci, "init err (%08x %04x)\n",
			ohci_readl (ohci, &ohci->regs->fminterval),
			ohci_readl (ohci, &ohci->regs->periodicstart));
		return -EOVERFLOW;
	}

	/* use rhsc irqs after khubd is fully initialized */
	hcd->poll_rh = 1;
	hcd->uses_new_polling = 1;

	/* start controller operations */
	ohci->hc_control &= OHCI_CTRL_RWC;
	ohci->hc_control |= OHCI_CONTROL_INIT | OHCI_USB_OPER;
	ohci_writel (ohci, ohci->hc_control, &ohci->regs->control);
	hcd->state = HC_STATE_RUNNING;

	/* wake on ConnectStatusChange, matching external hubs */
	ohci_writel (ohci, RH_HS_DRWE, &ohci->regs->roothub.status);

	/* Choose the interrupts we care about now, others later on demand */
	mask = OHCI_INTR_INIT;
	ohci_writel (ohci, ~0, &ohci->regs->intrstatus);
	ohci_writel (ohci, mask, &ohci->regs->intrenable);

	/* handle root hub init quirks ... */
	temp = roothub_a (ohci);
	temp &= ~(RH_A_PSM | RH_A_OCPM);
	if (ohci->flags & OHCI_QUIRK_SUPERIO) {
		/* NSC 87560 and maybe others */
		temp |= RH_A_NOCP;
		temp &= ~(RH_A_POTPGT | RH_A_NPS);
		ohci_writel (ohci, temp, &ohci->regs->roothub.a);
	} else if ((ohci->flags & OHCI_QUIRK_AMD756) ||
			(ohci->flags & OHCI_QUIRK_HUB_POWER)) {
		/* hub power always on; required for AMD-756 and some
		 * Mac platforms.  ganged overcurrent reporting, if any.
		 */
		temp |= RH_A_NPS;
		ohci_writel (ohci, temp, &ohci->regs->roothub.a);
	}
	ohci_writel (ohci, RH_HS_LPSC, &ohci->regs->roothub.status);
	ohci_writel (ohci, (temp & RH_A_NPS) ? 0 : RH_B_PPCM,
						&ohci->regs->roothub.b);
	// flush those writes
	(void) ohci_readl (ohci, &ohci->regs->control);

	ohci->next_statechange = jiffies + STATECHANGE_DELAY;
	spin_unlock_irq (&ohci->lock);

	// POTPGT delay is bits 24-31, in 2 ms units.
	mdelay ((temp >> 23) & 0x1fe);
	hcd->state = HC_STATE_RUNNING;

	if (quirk_zfmicro(ohci)) {
		/* Create timer to watch for bad queue state on ZF Micro */
		setup_timer(&ohci->unlink_watchdog, unlink_watchdog_func,
				(unsigned long) ohci);

		ohci->eds_scheduled = 0;
		ohci->ed_to_check = NULL;
	}

	ohci_dump (ohci, 1);

	return 0;
}

/*-------------------------------------------------------------------------*/

/* an interrupt happens */

static irqreturn_t ohci_irq (struct usb_hcd *hcd)
{
	struct ohci_hcd		*ohci = hcd_to_ohci (hcd);
	struct ohci_regs __iomem *regs = ohci->regs;
	int			ints;

	/* Read interrupt status (and flush pending writes).  We ignore the
	 * optimization of checking the LSB of hcca->done_head; it doesn't
	 * work on all systems (edge triggering for OHCI can be a factor).
	 */
	ints = ohci_readl(ohci, &regs->intrstatus);

	/* Check for an all 1's result which is a typical consequence
	 * of dead, unclocked, or unplugged (CardBus...) devices
	 */
	if (ints == ~(u32)0) {
		disable (ohci);
		ohci_dbg (ohci, "device removed!\n");
		return IRQ_HANDLED;
	}

	/* We only care about interrupts that are enabled */
	ints &= ohci_readl(ohci, &regs->intrenable);

	/* interrupt for some other device? */
	if (ints == 0)
		return IRQ_NOTMINE;

	if (ints & OHCI_INTR_UE) {
		// e.g. due to PCI Master/Target Abort
		if (quirk_nec(ohci)) {
			/* Workaround for a silicon bug in some NEC chips used
			 * in Apple's PowerBooks. Adapted from Darwin code.
			 */
			ohci_err (ohci, "OHCI Unrecoverable Error, scheduling NEC chip restart\n");

			ohci_writel (ohci, OHCI_INTR_UE, &regs->intrdisable);

			schedule_work (&ohci->nec_work);
		} else {
			disable (ohci);
			ohci_err (ohci, "OHCI Unrecoverable Error, disabled\n");
		}

		ohci_dump (ohci, 1);
		ohci_usb_reset (ohci);
	}

	if (ints & OHCI_INTR_RHSC) {
		ohci_vdbg(ohci, "rhsc\n");
		ohci->next_statechange = jiffies + STATECHANGE_DELAY;
		ohci_writel(ohci, OHCI_INTR_RD | OHCI_INTR_RHSC,
				&regs->intrstatus);

		/* NOTE: Vendors didn't always make the same implementation
		 * choices for RHSC.  Many followed the spec; RHSC triggers
		 * on an edge, like setting and maybe clearing a port status
		 * change bit.  With others it's level-triggered, active
		 * until khubd clears all the port status change bits.  We'll
		 * always disable it here and rely on polling until khubd
		 * re-enables it.
		 */
		ohci_writel(ohci, OHCI_INTR_RHSC, &regs->intrdisable);
		usb_hcd_poll_rh_status(hcd);
	}

	/* For connect and disconnect events, we expect the controller
	 * to turn on RHSC along with RD.  But for remote wakeup events
	 * this might not happen.
	 */
	else if (ints & OHCI_INTR_RD) {
		ohci_vdbg(ohci, "resume detect\n");
		ohci_writel(ohci, OHCI_INTR_RD, &regs->intrstatus);
		hcd->poll_rh = 1;
		if (ohci->autostop) {
			spin_lock (&ohci->lock);
			ohci_rh_resume (ohci);
			spin_unlock (&ohci->lock);
		} else
			usb_hcd_resume_root_hub(hcd);
	}

	if (ints & OHCI_INTR_WDH) {
		spin_lock (&ohci->lock);
		dl_done_list (ohci);
		spin_unlock (&ohci->lock);
	}

	if (quirk_zfmicro(ohci) && (ints & OHCI_INTR_SF)) {
		spin_lock(&ohci->lock);
		if (ohci->ed_to_check) {
			struct ed *ed = ohci->ed_to_check;

			if (check_ed(ohci, ed)) {
				/* HC thinks the TD list is empty; HCD knows
				 * at least one TD is outstanding
				 */
				if (--ohci->zf_delay == 0) {
					struct td *td = list_entry(
						ed->td_list.next,
						struct td, td_list);
					ohci_warn(ohci,
						  "Reclaiming orphan TD %p\n",
						  td);
					takeback_td(ohci, td);
					ohci->ed_to_check = NULL;
				}
			} else
				ohci->ed_to_check = NULL;
		}
		spin_unlock(&ohci->lock);
	}

	/* could track INTR_SO to reduce available PCI/... bandwidth */

	/* handle any pending URB/ED unlinks, leaving INTR_SF enabled
	 * when there's still unlinking to be done (next frame).
	 */
	spin_lock (&ohci->lock);
	if (ohci->ed_rm_list)
		finish_unlinks (ohci, ohci_frame_no(ohci));
	if ((ints & OHCI_INTR_SF) != 0
			&& !ohci->ed_rm_list
			&& !ohci->ed_to_check
			&& HC_IS_RUNNING(hcd->state))
		ohci_writel (ohci, OHCI_INTR_SF, &regs->intrdisable);
	spin_unlock (&ohci->lock);

	if (HC_IS_RUNNING(hcd->state)) {
		ohci_writel (ohci, ints, &regs->intrstatus);
		ohci_writel (ohci, OHCI_INTR_MIE, &regs->intrenable);
		// flush those writes
		(void) ohci_readl (ohci, &ohci->regs->control);
	}

	return IRQ_HANDLED;
}

/*-------------------------------------------------------------------------*/

static void ohci_stop (struct usb_hcd *hcd)
{
	struct ohci_hcd		*ohci = hcd_to_ohci (hcd);

	ohci_dump (ohci, 1);

	flush_scheduled_work();

	ohci_usb_reset (ohci);
	ohci_writel (ohci, OHCI_INTR_MIE, &ohci->regs->intrdisable);
	free_irq(hcd->irq, hcd);
	hcd->irq = -1;

	if (quirk_zfmicro(ohci))
		del_timer(&ohci->unlink_watchdog);
	if (quirk_amdiso(ohci))
		amd_iso_dev_put();

	remove_debug_files (ohci);
	ohci_mem_cleanup (ohci);
	if (ohci->hcca) {
		dma_free_coherent (hcd->self.controller,
				sizeof *ohci->hcca,
				ohci->hcca, ohci->hcca_dma);
		ohci->hcca = NULL;
		ohci->hcca_dma = 0;
	}
}

/*-------------------------------------------------------------------------*/

#if defined(CONFIG_PM) || defined(CONFIG_PCI)

/* must not be called from interrupt context */
static int ohci_restart (struct ohci_hcd *ohci)
{
	int temp;
	int i;
	struct urb_priv *priv;

	spin_lock_irq(&ohci->lock);
	disable (ohci);

	/* Recycle any "live" eds/tds (and urbs). */
	if (!list_empty (&ohci->pending))
		ohci_dbg(ohci, "abort schedule...\n");
	list_for_each_entry (priv, &ohci->pending, pending) {
		struct urb	*urb = priv->td[0]->urb;
		struct ed	*ed = priv->ed;

		switch (ed->state) {
		case ED_OPER:
			ed->state = ED_UNLINK;
			ed->hwINFO |= cpu_to_hc32(ohci, ED_DEQUEUE);
			ed_deschedule (ohci, ed);

			ed->ed_next = ohci->ed_rm_list;
			ed->ed_prev = NULL;
			ohci->ed_rm_list = ed;
			/* FALLTHROUGH */
		case ED_UNLINK:
			break;
		default:
			ohci_dbg(ohci, "bogus ed %p state %d\n",
					ed, ed->state);
		}

		if (!urb->unlinked)
			urb->unlinked = -ESHUTDOWN;
	}
	finish_unlinks (ohci, 0);
	spin_unlock_irq(&ohci->lock);

	/* paranoia, in case that didn't work: */

	/* empty the interrupt branches */
	for (i = 0; i < NUM_INTS; i++) ohci->load [i] = 0;
	for (i = 0; i < NUM_INTS; i++) ohci->hcca->int_table [i] = 0;

	/* no EDs to remove */
	ohci->ed_rm_list = NULL;

	/* empty control and bulk lists */
	ohci->ed_controltail = NULL;
	ohci->ed_bulktail    = NULL;

	if ((temp = ohci_run (ohci)) < 0) {
		ohci_err (ohci, "can't restart, %d\n", temp);
		return temp;
	}
	ohci_dbg(ohci, "restart complete\n");
	return 0;
}

#endif

/*-------------------------------------------------------------------------*/

#define DRIVER_INFO DRIVER_VERSION " " DRIVER_DESC

MODULE_AUTHOR (DRIVER_AUTHOR);
MODULE_DESCRIPTION (DRIVER_INFO);
MODULE_LICENSE ("GPL");

#ifdef CONFIG_PCI
#include "ohci-pci.c"
#define PCI_DRIVER		ohci_pci_driver
#endif

#if defined(CONFIG_ARCH_SA1100) && defined(CONFIG_SA1111)
#include "ohci-sa1111.c"
#define SA1111_DRIVER		ohci_hcd_sa1111_driver
#endif

#ifdef CONFIG_ARCH_S3C2410
#include "ohci-s3c2410.c"
#define PLATFORM_DRIVER		ohci_hcd_s3c2410_driver
#endif

#ifdef CONFIG_ARCH_OMAP
#include "ohci-omap.c"
#define PLATFORM_DRIVER		ohci_hcd_omap_driver
#endif

#ifdef CONFIG_ARCH_LH7A404
#include "ohci-lh7a404.c"
#define PLATFORM_DRIVER		ohci_hcd_lh7a404_driver
#endif

#if defined(CONFIG_PXA27x) || defined(CONFIG_PXA3xx)
#include "ohci-pxa27x.c"
#define PLATFORM_DRIVER		ohci_hcd_pxa27x_driver
#endif

#ifdef CONFIG_ARCH_EP93XX
#include "ohci-ep93xx.c"
#define PLATFORM_DRIVER		ohci_hcd_ep93xx_driver
#endif

#ifdef CONFIG_SOC_AU1X00
#include "ohci-au1xxx.c"
#define PLATFORM_DRIVER		ohci_hcd_au1xxx_driver
#endif

#ifdef CONFIG_PNX8550
#include "ohci-pnx8550.c"
#define PLATFORM_DRIVER		ohci_hcd_pnx8550_driver
#endif

#ifdef CONFIG_USB_OHCI_HCD_PPC_SOC
#include "ohci-ppc-soc.c"
#define PLATFORM_DRIVER		ohci_hcd_ppc_soc_driver
#endif

#ifdef CONFIG_ARCH_AT91
#include "ohci-at91.c"
#define PLATFORM_DRIVER		ohci_hcd_at91_driver
#endif

#ifdef CONFIG_ARCH_PNX4008
#include "ohci-pnx4008.c"
#define PLATFORM_DRIVER		usb_hcd_pnx4008_driver
#endif

#if defined(CONFIG_CPU_SUBTYPE_SH7720) || \
    defined(CONFIG_CPU_SUBTYPE_SH7721) || \
    defined(CONFIG_CPU_SUBTYPE_SH7763)
#include "ohci-sh.c"
#define PLATFORM_DRIVER		ohci_hcd_sh_driver
#endif


#ifdef CONFIG_USB_OHCI_HCD_PPC_OF
#include "ohci-ppc-of.c"
#define OF_PLATFORM_DRIVER	ohci_hcd_ppc_of_driver
#endif

#ifdef CONFIG_PPC_PS3
#include "ohci-ps3.c"
#define PS3_SYSTEM_BUS_DRIVER	ps3_ohci_driver
#endif

#ifdef CONFIG_USB_OHCI_HCD_SSB
#include "ohci-ssb.c"
#define SSB_OHCI_DRIVER		ssb_ohci_driver
#endif

#ifdef CONFIG_MFD_SM501
#include "ohci-sm501.c"
#define SM501_OHCI_DRIVER	ohci_hcd_sm501_driver
#endif

#if	!defined(PCI_DRIVER) &&		\
	!defined(PLATFORM_DRIVER) &&	\
	!defined(OF_PLATFORM_DRIVER) &&	\
	!defined(SA1111_DRIVER) &&	\
	!defined(PS3_SYSTEM_BUS_DRIVER) && \
	!defined(SM501_OHCI_DRIVER) && \
	!defined(SSB_OHCI_DRIVER)
#error "missing bus glue for ohci-hcd"
#endif

static int __init ohci_hcd_mod_init(void)
{
	int retval = 0;

	if (usb_disabled())
		return -ENODEV;

	printk (KERN_DEBUG "%s: " DRIVER_INFO "\n", hcd_name);
	pr_debug ("%s: block sizes: ed %Zd td %Zd\n", hcd_name,
		sizeof (struct ed), sizeof (struct td));

#ifdef DEBUG
	ohci_debug_root = debugfs_create_dir("ohci", NULL);
	if (!ohci_debug_root) {
		retval = -ENOENT;
		goto error_debug;
	}
#endif

#ifdef PS3_SYSTEM_BUS_DRIVER
	retval = ps3_ohci_driver_register(&PS3_SYSTEM_BUS_DRIVER);
	if (retval < 0)
		goto error_ps3;
#endif

#ifdef PLATFORM_DRIVER
	retval = platform_driver_register(&PLATFORM_DRIVER);
	if (retval < 0)
		goto error_platform;
#endif

#ifdef OF_PLATFORM_DRIVER
	retval = of_register_platform_driver(&OF_PLATFORM_DRIVER);
	if (retval < 0)
		goto error_of_platform;
#endif

#ifdef SA1111_DRIVER
	retval = sa1111_driver_register(&SA1111_DRIVER);
	if (retval < 0)
		goto error_sa1111;
#endif

#ifdef PCI_DRIVER
	retval = pci_register_driver(&PCI_DRIVER);
	if (retval < 0)
		goto error_pci;
#endif

#ifdef SSB_OHCI_DRIVER
	retval = ssb_driver_register(&SSB_OHCI_DRIVER);
	if (retval)
		goto error_ssb;
#endif

#ifdef SM501_OHCI_DRIVER
	retval = platform_driver_register(&SM501_OHCI_DRIVER);
	if (retval < 0)
		goto error_sm501;
#endif

	return retval;

	/* Error path */
#ifdef SM501_OHCI_DRIVER
 error_sm501:
#endif
#ifdef SSB_OHCI_DRIVER
 error_ssb:
#endif
#ifdef PCI_DRIVER
	pci_unregister_driver(&PCI_DRIVER);
 error_pci:
#endif
#ifdef SA1111_DRIVER
	sa1111_driver_unregister(&SA1111_DRIVER);
 error_sa1111:
#endif
#ifdef OF_PLATFORM_DRIVER
	of_unregister_platform_driver(&OF_PLATFORM_DRIVER);
 error_of_platform:
#endif
#ifdef PLATFORM_DRIVER
	platform_driver_unregister(&PLATFORM_DRIVER);
 error_platform:
#endif
#ifdef PS3_SYSTEM_BUS_DRIVER
	ps3_ohci_driver_unregister(&PS3_SYSTEM_BUS_DRIVER);
 error_ps3:
#endif
#ifdef DEBUG
	debugfs_remove(ohci_debug_root);
	ohci_debug_root = NULL;
 error_debug:
#endif

	return retval;
}
module_init(ohci_hcd_mod_init);

static void __exit ohci_hcd_mod_exit(void)
{
#ifdef SM501_OHCI_DRIVER
	platform_driver_unregister(&SM501_OHCI_DRIVER);
#endif
#ifdef SSB_OHCI_DRIVER
	ssb_driver_unregister(&SSB_OHCI_DRIVER);
#endif
#ifdef PCI_DRIVER
	pci_unregister_driver(&PCI_DRIVER);
#endif
#ifdef SA1111_DRIVER
	sa1111_driver_unregister(&SA1111_DRIVER);
#endif
#ifdef OF_PLATFORM_DRIVER
	of_unregister_platform_driver(&OF_PLATFORM_DRIVER);
#endif
#ifdef PLATFORM_DRIVER
	platform_driver_unregister(&PLATFORM_DRIVER);
#endif
#ifdef PS3_SYSTEM_BUS_DRIVER
	ps3_ohci_driver_unregister(&PS3_SYSTEM_BUS_DRIVER);
#endif
#ifdef DEBUG
	debugfs_remove(ohci_debug_root);
#endif
}
module_exit(ohci_hcd_mod_exit);

pan> (lstart >= size) { jfs_err("xtLookup: lstart (0x%lx) >= size (0x%lx)", (ulong) lstart, (ulong) size); return 0; } } /* * search for the xad entry covering the logical extent */ //search: if ((rc = xtSearch(ip, lstart, &next, &cmp, &btstack, 0))) { jfs_err("xtLookup: xtSearch returned %d", rc); return rc; } /* * compute the physical extent covering logical extent * * N.B. search may have failed (e.g., hole in sparse file), * and returned the index of the next entry. */ /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, mp, p, index); /* is xad found covering start of logical extent ? * lstart is a page start address, * i.e., lstart cannot start in a hole; */ if (cmp) { if (next) *plen = min(next - lstart, llen); goto out; } /* * lxd covered by xad */ xad = &p->xad[index]; xoff = offsetXAD(xad); xlen = lengthXAD(xad); xend = xoff + xlen; xaddr = addressXAD(xad); /* initialize new pxd */ *pflag = xad->flag; *paddr = xaddr + (lstart - xoff); /* a page must be fully covered by an xad */ *plen = min(xend - lstart, llen); out: XT_PUTPAGE(mp); return rc; } /* * xtLookupList() * * function: map a single logical extent into a list of physical extent; * * parameter: * struct inode *ip, * struct lxdlist *lxdlist, lxd list (in) * struct xadlist *xadlist, xad list (in/out) * int flag) * * coverage of lxd by xad under assumption of * . lxd's are ordered and disjoint. * . xad's are ordered and disjoint. * * return: * 0: success * * note: a page being written (even a single byte) is backed fully, * except the last page which is only backed with blocks * required to cover the last byte; * the extent backing a page is fully contained within an xad; */ int xtLookupList(struct inode *ip, struct lxdlist * lxdlist, struct xadlist * xadlist, int flag) { int rc = 0; struct btstack btstack; int cmp; s64 bn; struct metapage *mp; xtpage_t *p; int index; lxd_t *lxd; xad_t *xad, *pxd; s64 size, lstart, lend, xstart, xend, pstart; s64 llen, xlen, plen; s64 xaddr, paddr; int nlxd, npxd, maxnpxd; npxd = xadlist->nxad = 0; maxnpxd = xadlist->maxnxad; pxd = xadlist->xad; nlxd = lxdlist->nlxd; lxd = lxdlist->lxd; lstart = offsetLXD(lxd); llen = lengthLXD(lxd); lend = lstart + llen; size = (ip->i_size + (JFS_SBI(ip->i_sb)->bsize - 1)) >> JFS_SBI(ip->i_sb)->l2bsize; /* * search for the xad entry covering the logical extent */ search: if (lstart >= size) return 0; if ((rc = xtSearch(ip, lstart, NULL, &cmp, &btstack, 0))) return rc; /* * compute the physical extent covering logical extent * * N.B. search may have failed (e.g., hole in sparse file), * and returned the index of the next entry. */ //map: /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, mp, p, index); /* is xad on the next sibling page ? */ if (index == le16_to_cpu(p->header.nextindex)) { if (p->header.flag & BT_ROOT) goto mapend; if ((bn = le64_to_cpu(p->header.next)) == 0) goto mapend; XT_PUTPAGE(mp); /* get next sibling page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; index = XTENTRYSTART; } xad = &p->xad[index]; /* * is lxd covered by xad ? */ compare: xstart = offsetXAD(xad); xlen = lengthXAD(xad); xend = xstart + xlen; xaddr = addressXAD(xad); compare1: if (xstart < lstart) goto compare2; /* (lstart <= xstart) */ /* lxd is NOT covered by xad */ if (lend <= xstart) { /* * get next lxd */ if (--nlxd == 0) goto mapend; lxd++; lstart = offsetLXD(lxd); llen = lengthLXD(lxd); lend = lstart + llen; if (lstart >= size) goto mapend; /* compare with the current xad */ goto compare1; } /* lxd is covered by xad */ else { /* (xstart < lend) */ /* initialize new pxd */ pstart = xstart; plen = min(lend - xstart, xlen); paddr = xaddr; goto cover; } /* (xstart < lstart) */ compare2: /* lxd is covered by xad */ if (lstart < xend) { /* initialize new pxd */ pstart = lstart; plen = min(xend - lstart, llen); paddr = xaddr + (lstart - xstart); goto cover; } /* lxd is NOT covered by xad */ else { /* (xend <= lstart) */ /* * get next xad * * linear search next xad covering lxd on * the current xad page, and then tree search */ if (index == le16_to_cpu(p->header.nextindex) - 1) { if (p->header.flag & BT_ROOT) goto mapend; XT_PUTPAGE(mp); goto search; } else { index++; xad++; /* compare with new xad */ goto compare; } } /* * lxd is covered by xad and a new pxd has been initialized * (lstart <= xstart < lend) or (xstart < lstart < xend) */ cover: /* finalize pxd corresponding to current xad */ XT_PUTENTRY(pxd, xad->flag, pstart, plen, paddr); if (++npxd >= maxnpxd) goto mapend; pxd++; /* * lxd is fully covered by xad */ if (lend <= xend) { /* * get next lxd */ if (--nlxd == 0) goto mapend; lxd++; lstart = offsetLXD(lxd); llen = lengthLXD(lxd); lend = lstart + llen; if (lstart >= size) goto mapend; /* * test for old xad covering new lxd * (old xstart < new lstart) */ goto compare2; } /* * lxd is partially covered by xad */ else { /* (xend < lend) */ /* * get next xad * * linear search next xad covering lxd on * the current xad page, and then next xad page search */ if (index == le16_to_cpu(p->header.nextindex) - 1) { if (p->header.flag & BT_ROOT) goto mapend; if ((bn = le64_to_cpu(p->header.next)) == 0) goto mapend; XT_PUTPAGE(mp); /* get next sibling page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; index = XTENTRYSTART; xad = &p->xad[index]; } else { index++; xad++; } /* * test for new xad covering old lxd * (old lstart < new xstart) */ goto compare; } mapend: xadlist->nxad = npxd; //out: XT_PUTPAGE(mp); return rc; } /* * xtSearch() * * function: search for the xad entry covering specified offset. * * parameters: * ip - file object; * xoff - extent offset; * nextp - address of next extent (if any) for search miss * cmpp - comparison result: * btstack - traverse stack; * flag - search process flag (XT_INSERT); * * returns: * btstack contains (bn, index) of search path traversed to the entry. * *cmpp is set to result of comparison with the entry returned. * the page containing the entry is pinned at exit. */ static int xtSearch(struct inode *ip, s64 xoff, s64 *nextp, int *cmpp, struct btstack * btstack, int flag) { struct jfs_inode_info *jfs_ip = JFS_IP(ip); int rc = 0; int cmp = 1; /* init for empty page */ s64 bn; /* block number */ struct metapage *mp; /* page buffer */ xtpage_t *p; /* page */ xad_t *xad; int base, index, lim, btindex; struct btframe *btsp; int nsplit = 0; /* number of pages to split */ s64 t64; s64 next = 0; INCREMENT(xtStat.search); BT_CLR(btstack); btstack->nsplit = 0; /* * search down tree from root: * * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of * internal page, child page Pi contains entry with k, Ki <= K < Kj. * * if entry with search key K is not found * internal page search find the entry with largest key Ki * less than K which point to the child page to search; * leaf page search find the entry with smallest key Kj * greater than K so that the returned index is the position of * the entry to be shifted right for insertion of new entry. * for empty tree, search key is greater than any key of the tree. * * by convention, root bn = 0. */ for (bn = 0;;) { /* get/pin the page to search */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; /* try sequential access heuristics with the previous * access entry in target leaf page: * once search narrowed down into the target leaf, * key must either match an entry in the leaf or * key entry does not exist in the tree; */ //fastSearch: if ((jfs_ip->btorder & BT_SEQUENTIAL) && (p->header.flag & BT_LEAF) && (index = jfs_ip->btindex) < le16_to_cpu(p->header.nextindex)) { xad = &p->xad[index]; t64 = offsetXAD(xad); if (xoff < t64 + lengthXAD(xad)) { if (xoff >= t64) { *cmpp = 0; goto out; } /* stop sequential access heuristics */ goto binarySearch; } else { /* (t64 + lengthXAD(xad)) <= xoff */ /* try next sequential entry */ index++; if (index < le16_to_cpu(p->header.nextindex)) { xad++; t64 = offsetXAD(xad); if (xoff < t64 + lengthXAD(xad)) { if (xoff >= t64) { *cmpp = 0; goto out; } /* miss: key falls between * previous and this entry */ *cmpp = 1; next = t64; goto out; } /* (xoff >= t64 + lengthXAD(xad)); * matching entry may be further out: * stop heuristic search */ /* stop sequential access heuristics */ goto binarySearch; } /* (index == p->header.nextindex); * miss: key entry does not exist in * the target leaf/tree */ *cmpp = 1; goto out; } /* * if hit, return index of the entry found, and * if miss, where new entry with search key is * to be inserted; */ out: /* compute number of pages to split */ if (flag & XT_INSERT) { if (p->header.nextindex == /* little-endian */ p->header.maxentry) nsplit++; else nsplit = 0; btstack->nsplit = nsplit; } /* save search result */ btsp = btstack->top; btsp->bn = bn; btsp->index = index; btsp->mp = mp; /* update sequential access heuristics */ jfs_ip->btindex = index; if (nextp) *nextp = next; INCREMENT(xtStat.fastSearch); return 0; } /* well, ... full search now */ binarySearch: lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART; /* * binary search with search key K on the current page */ for (base = XTENTRYSTART; lim; lim >>= 1) { index = base + (lim >> 1); XT_CMP(cmp, xoff, &p->xad[index], t64); if (cmp == 0) { /* * search hit */ /* search hit - leaf page: * return the entry found */ if (p->header.flag & BT_LEAF) { *cmpp = cmp; /* compute number of pages to split */ if (flag & XT_INSERT) { if (p->header.nextindex == p->header.maxentry) nsplit++; else nsplit = 0; btstack->nsplit = nsplit; } /* save search result */ btsp = btstack->top; btsp->bn = bn; btsp->index = index; btsp->mp = mp; /* init sequential access heuristics */ btindex = jfs_ip->btindex; if (index == btindex || index == btindex + 1) jfs_ip->btorder = BT_SEQUENTIAL; else jfs_ip->btorder = BT_RANDOM; jfs_ip->btindex = index; return 0; } /* search hit - internal page: * descend/search its child page */ if (index < le16_to_cpu(p->header.nextindex)-1) next = offsetXAD(&p->xad[index + 1]); goto next; } if (cmp > 0) { base = index + 1; --lim; } } /* * search miss * * base is the smallest index with key (Kj) greater than * search key (K) and may be zero or maxentry index. */ if (base < le16_to_cpu(p->header.nextindex)) next = offsetXAD(&p->xad[base]); /* * search miss - leaf page: * * return location of entry (base) where new entry with * search key K is to be inserted. */ if (p->header.flag & BT_LEAF) { *cmpp = cmp; /* compute number of pages to split */ if (flag & XT_INSERT) { if (p->header.nextindex == p->header.maxentry) nsplit++; else nsplit = 0; btstack->nsplit = nsplit; } /* save search result */ btsp = btstack->top; btsp->bn = bn; btsp->index = base; btsp->mp = mp; /* init sequential access heuristics */ btindex = jfs_ip->btindex; if (base == btindex || base == btindex + 1) jfs_ip->btorder = BT_SEQUENTIAL; else jfs_ip->btorder = BT_RANDOM; jfs_ip->btindex = base; if (nextp) *nextp = next; return 0; } /* * search miss - non-leaf page: * * if base is non-zero, decrement base by one to get the parent * entry of the child page to search. */ index = base ? base - 1 : base; /* * go down to child page */ next: /* update number of pages to split */ if (p->header.nextindex == p->header.maxentry) nsplit++; else nsplit = 0; /* push (bn, index) of the parent page/entry */ if (BT_STACK_FULL(btstack)) { jfs_error(ip->i_sb, "stack overrun in xtSearch!"); XT_PUTPAGE(mp); return -EIO; } BT_PUSH(btstack, bn, index); /* get the child page block number */ bn = addressXAD(&p->xad[index]); /* unpin the parent page */ XT_PUTPAGE(mp); } } /* * xtInsert() * * function: * * parameter: * tid - transaction id; * ip - file object; * xflag - extent flag (XAD_NOTRECORDED): * xoff - extent offset; * xlen - extent length; * xaddrp - extent address pointer (in/out): * if (*xaddrp) * caller allocated data extent at *xaddrp; * else * allocate data extent and return its xaddr; * flag - * * return: */ int xtInsert(tid_t tid, /* transaction id */ struct inode *ip, int xflag, s64 xoff, s32 xlen, s64 * xaddrp, int flag) { int rc = 0; s64 xaddr, hint; struct metapage *mp; /* meta-page buffer */ xtpage_t *p; /* base B+-tree index page */ s64 bn; int index, nextindex; struct btstack btstack; /* traverse stack */ struct xtsplit split; /* split information */ xad_t *xad; int cmp; s64 next; struct tlock *tlck; struct xtlock *xtlck; jfs_info("xtInsert: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen); /* * search for the entry location at which to insert: * * xtFastSearch() and xtSearch() both returns (leaf page * pinned, index at which to insert). * n.b. xtSearch() may return index of maxentry of * the full page. */ if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT))) return rc; /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, mp, p, index); /* This test must follow XT_GETSEARCH since mp must be valid if * we branch to out: */ if ((cmp == 0) || (next && (xlen > next - xoff))) { rc = -EEXIST; goto out; } /* * allocate data extent requested * * allocation hint: last xad */ if ((xaddr = *xaddrp) == 0) { if (index > XTENTRYSTART) { xad = &p->xad[index - 1]; hint = addressXAD(xad) + lengthXAD(xad) - 1; } else hint = 0; if ((rc = DQUOT_ALLOC_BLOCK(ip, xlen))) goto out; if ((rc = dbAlloc(ip, hint, (s64) xlen, &xaddr))) { DQUOT_FREE_BLOCK(ip, xlen); goto out; } } /* * insert entry for new extent */ xflag |= XAD_NEW; /* * if the leaf page is full, split the page and * propagate up the router entry for the new page from split * * The xtSplitUp() will insert the entry and unpin the leaf page. */ nextindex = le16_to_cpu(p->header.nextindex); if (nextindex == le16_to_cpu(p->header.maxentry)) { split.mp = mp; split.index = index; split.flag = xflag; split.off = xoff; split.len = xlen; split.addr = xaddr; split.pxdlist = NULL; if ((rc = xtSplitUp(tid, ip, &split, &btstack))) { /* undo data extent allocation */ if (*xaddrp == 0) { dbFree(ip, xaddr, (s64) xlen); DQUOT_FREE_BLOCK(ip, xlen); } return rc; } *xaddrp = xaddr; return 0; } /* * insert the new entry into the leaf page */ /* * acquire a transaction lock on the leaf page; * * action: xad insertion/extension; */ BT_MARK_DIRTY(mp, ip); /* if insert into middle, shift right remaining entries. */ if (index < nextindex) memmove(&p->xad[index + 1], &p->xad[index], (nextindex - index) * sizeof(xad_t)); /* insert the new entry: mark the entry NEW */ xad = &p->xad[index]; XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr); /* advance next available entry index */ p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1); /* Don't log it if there are no links to the file */ if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index, (int)xtlck->lwm.offset) : index; xtlck->lwm.length = le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset; } *xaddrp = xaddr; out: /* unpin the leaf page */ XT_PUTPAGE(mp); return rc; } /* * xtSplitUp() * * function: * split full pages as propagating insertion up the tree * * parameter: * tid - transaction id; * ip - file object; * split - entry parameter descriptor; * btstack - traverse stack from xtSearch() * * return: */ static int xtSplitUp(tid_t tid, struct inode *ip, struct xtsplit * split, struct btstack * btstack) { int rc = 0; struct metapage *smp; xtpage_t *sp; /* split page */ struct metapage *rmp; s64 rbn; /* new right page block number */ struct metapage *rcmp; xtpage_t *rcp; /* right child page */ s64 rcbn; /* right child page block number */ int skip; /* index of entry of insertion */ int nextindex; /* next available entry index of p */ struct btframe *parent; /* parent page entry on traverse stack */ xad_t *xad; s64 xaddr; int xlen; int nsplit; /* number of pages split */ struct pxdlist pxdlist; pxd_t *pxd; struct tlock *tlck; struct xtlock *xtlck; smp = split->mp; sp = XT_PAGE(ip, smp); /* is inode xtree root extension/inline EA area free ? */ if ((sp->header.flag & BT_ROOT) && (!S_ISDIR(ip->i_mode)) && (le16_to_cpu(sp->header.maxentry) < XTROOTMAXSLOT) && (JFS_IP(ip)->mode2 & INLINEEA)) { sp->header.maxentry = cpu_to_le16(XTROOTMAXSLOT); JFS_IP(ip)->mode2 &= ~INLINEEA; BT_MARK_DIRTY(smp, ip); /* * acquire a transaction lock on the leaf page; * * action: xad insertion/extension; */ /* if insert into middle, shift right remaining entries. */ skip = split->index; nextindex = le16_to_cpu(sp->header.nextindex); if (skip < nextindex) memmove(&sp->xad[skip + 1], &sp->xad[skip], (nextindex - skip) * sizeof(xad_t)); /* insert the new entry: mark the entry NEW */ xad = &sp->xad[skip]; XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr); /* advance next available entry index */ sp->header.nextindex = cpu_to_le16(le16_to_cpu(sp->header.nextindex) + 1); /* Don't log it if there are no links to the file */ if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; xtlck->lwm.offset = (xtlck->lwm.offset) ? min(skip, (int)xtlck->lwm.offset) : skip; xtlck->lwm.length = le16_to_cpu(sp->header.nextindex) - xtlck->lwm.offset; } return 0; } /* * allocate new index blocks to cover index page split(s) * * allocation hint: ? */ if (split->pxdlist == NULL) { nsplit = btstack->nsplit; split->pxdlist = &pxdlist; pxdlist.maxnpxd = pxdlist.npxd = 0; pxd = &pxdlist.pxd[0]; xlen = JFS_SBI(ip->i_sb)->nbperpage; for (; nsplit > 0; nsplit--, pxd++) { if ((rc = dbAlloc(ip, (s64) 0, (s64) xlen, &xaddr)) == 0) { PXDaddress(pxd, xaddr); PXDlength(pxd, xlen); pxdlist.maxnpxd++; continue; } /* undo allocation */ XT_PUTPAGE(smp); return rc; } } /* * Split leaf page <sp> into <sp> and a new right page <rp>. * * The split routines insert the new entry into the leaf page, * and acquire txLock as appropriate. * return <rp> pinned and its block number <rpbn>. */ rc = (sp->header.flag & BT_ROOT) ? xtSplitRoot(tid, ip, split, &rmp) : xtSplitPage(tid, ip, split, &rmp, &rbn); XT_PUTPAGE(smp); if (rc) return -EIO; /* * propagate up the router entry for the leaf page just split * * insert a router entry for the new page into the parent page, * propagate the insert/split up the tree by walking back the stack * of (bn of parent page, index of child page entry in parent page) * that were traversed during the search for the page that split. * * the propagation of insert/split up the tree stops if the root * splits or the page inserted into doesn't have to split to hold * the new entry. * * the parent entry for the split page remains the same, and * a new entry is inserted at its right with the first key and * block number of the new right page. * * There are a maximum of 3 pages pinned at any time: * right child, left parent and right parent (when the parent splits) * to keep the child page pinned while working on the parent. * make sure that all pins are released at exit. */ while ((parent = BT_POP(btstack)) != NULL) { /* parent page specified by stack frame <parent> */ /* keep current child pages <rcp> pinned */ rcmp = rmp; rcbn = rbn; rcp = XT_PAGE(ip, rcmp); /* * insert router entry in parent for new right child page <rp> */ /* get/pin the parent page <sp> */ XT_GETPAGE(ip, parent->bn, smp, PSIZE, sp, rc); if (rc) { XT_PUTPAGE(rcmp); return rc; } /* * The new key entry goes ONE AFTER the index of parent entry, * because the split was to the right. */ skip = parent->index + 1; /* * split or shift right remaining entries of the parent page */ nextindex = le16_to_cpu(sp->header.nextindex); /* * parent page is full - split the parent page */ if (nextindex == le16_to_cpu(sp->header.maxentry)) { /* init for parent page split */ split->mp = smp; split->index = skip; /* index at insert */ split->flag = XAD_NEW; split->off = offsetXAD(&rcp->xad[XTENTRYSTART]); split->len = JFS_SBI(ip->i_sb)->nbperpage; split->addr = rcbn; /* unpin previous right child page */ XT_PUTPAGE(rcmp); /* The split routines insert the new entry, * and acquire txLock as appropriate. * return <rp> pinned and its block number <rpbn>. */ rc = (sp->header.flag & BT_ROOT) ? xtSplitRoot(tid, ip, split, &rmp) : xtSplitPage(tid, ip, split, &rmp, &rbn); if (rc) { XT_PUTPAGE(smp); return rc; } XT_PUTPAGE(smp); /* keep new child page <rp> pinned */ } /* * parent page is not full - insert in parent page */ else { /* * insert router entry in parent for the right child * page from the first entry of the right child page: */ /* * acquire a transaction lock on the parent page; * * action: router xad insertion; */ BT_MARK_DIRTY(smp, ip); /* * if insert into middle, shift right remaining entries */ if (skip < nextindex) memmove(&sp->xad[skip + 1], &sp->xad[skip], (nextindex - skip) << L2XTSLOTSIZE); /* insert the router entry */ xad = &sp->xad[skip]; XT_PUTENTRY(xad, XAD_NEW, offsetXAD(&rcp->xad[XTENTRYSTART]), JFS_SBI(ip->i_sb)->nbperpage, rcbn); /* advance next available entry index. */ sp->header.nextindex = cpu_to_le16(le16_to_cpu(sp->header.nextindex) + 1); /* Don't log it if there are no links to the file */ if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; xtlck->lwm.offset = (xtlck->lwm.offset) ? min(skip, (int)xtlck->lwm.offset) : skip; xtlck->lwm.length = le16_to_cpu(sp->header.nextindex) - xtlck->lwm.offset; } /* unpin parent page */ XT_PUTPAGE(smp); /* exit propagate up */ break; } } /* unpin current right page */ XT_PUTPAGE(rmp); return 0; } /* * xtSplitPage() * * function: * split a full non-root page into * original/split/left page and new right page * i.e., the original/split page remains as left page. * * parameter: * int tid, * struct inode *ip, * struct xtsplit *split, * struct metapage **rmpp, * u64 *rbnp, * * return: * Pointer to page in which to insert or NULL on error. */ static int xtSplitPage(tid_t tid, struct inode *ip, struct xtsplit * split, struct metapage ** rmpp, s64 * rbnp) { int rc = 0; struct metapage *smp; xtpage_t *sp; struct metapage *rmp; xtpage_t *rp; /* new right page allocated */ s64 rbn; /* new right page block number */ struct metapage *mp; xtpage_t *p; s64 nextbn; int skip, maxentry, middle, righthalf, n; xad_t *xad; struct pxdlist *pxdlist; pxd_t *pxd; struct tlock *tlck; struct xtlock *sxtlck = NULL, *rxtlck = NULL; int quota_allocation = 0; smp = split->mp; sp = XT_PAGE(ip, smp); INCREMENT(xtStat.split); pxdlist = split->pxdlist; pxd = &pxdlist->pxd[pxdlist->npxd]; pxdlist->npxd++; rbn = addressPXD(pxd); /* Allocate blocks to quota. */ if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) { rc = -EDQUOT; goto clean_up; } quota_allocation += lengthPXD(pxd); /* * allocate the new right page for the split */ rmp = get_metapage(ip, rbn, PSIZE, 1); if (rmp == NULL) { rc = -EIO; goto clean_up; } jfs_info("xtSplitPage: ip:0x%p smp:0x%p rmp:0x%p", ip, smp, rmp); BT_MARK_DIRTY(rmp, ip); /* * action: new page; */ rp = (xtpage_t *) rmp->data; rp->header.self = *pxd; rp->header.flag = sp->header.flag & BT_TYPE; rp->header.maxentry = sp->header.maxentry; /* little-endian */ rp->header.nextindex = cpu_to_le16(XTENTRYSTART); BT_MARK_DIRTY(smp, ip); /* Don't log it if there are no links to the file */ if (!test_cflag(COMMIT_Nolink, ip)) { /* * acquire a transaction lock on the new right page; */ tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW); rxtlck = (struct xtlock *) & tlck->lock; rxtlck->lwm.offset = XTENTRYSTART; /* * acquire a transaction lock on the split page */ tlck = txLock(tid, ip, smp, tlckXTREE | tlckGROW); sxtlck = (struct xtlock *) & tlck->lock; } /* * initialize/update sibling pointers of <sp> and <rp> */ nextbn = le64_to_cpu(sp->header.next); rp->header.next = cpu_to_le64(nextbn); rp->header.prev = cpu_to_le64(addressPXD(&sp->header.self)); sp->header.next = cpu_to_le64(rbn); skip = split->index; /* * sequential append at tail (after last entry of last page) * * if splitting the last page on a level because of appending * a entry to it (skip is maxentry), it's likely that the access is * sequential. adding an empty page on the side of the level is less * work and can push the fill factor much higher than normal. * if we're wrong it's no big deal - we will do the split the right * way next time. * (it may look like it's equally easy to do a similar hack for * reverse sorted data, that is, split the tree left, but it's not. * Be my guest.) */ if (nextbn == 0 && skip == le16_to_cpu(sp->header.maxentry)) { /* * acquire a transaction lock on the new/right page; * * action: xad insertion; */ /* insert entry at the first entry of the new right page */ xad = &rp->xad[XTENTRYSTART]; XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr); rp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1); if (!test_cflag(COMMIT_Nolink, ip)) { /* rxtlck->lwm.offset = XTENTRYSTART; */ rxtlck->lwm.length = 1; } *rmpp = rmp; *rbnp = rbn; jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp); return 0; } /* * non-sequential insert (at possibly middle page) */ /* * update previous pointer of old next/right page of <sp> */ if (nextbn != 0) { XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc); if (rc) { XT_PUTPAGE(rmp); goto clean_up; } BT_MARK_DIRTY(mp, ip); /* * acquire a transaction lock on the next page; * * action:sibling pointer update; */ if (!test_cflag(COMMIT_Nolink, ip)) tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK); p->header.prev = cpu_to_le64(rbn); /* sibling page may have been updated previously, or * it may be updated later; */ XT_PUTPAGE(mp); } /* * split the data between the split and new/right pages */ maxentry = le16_to_cpu(sp->header.maxentry); middle = maxentry >> 1; righthalf = maxentry - middle; /* * skip index in old split/left page - insert into left page: */ if (skip <= middle) { /* move right half of split page to the new right page */ memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle], righthalf << L2XTSLOTSIZE); /* shift right tail of left half to make room for new entry */ if (skip < middle) memmove(&sp->xad[skip + 1], &sp->xad[skip], (middle - skip) << L2XTSLOTSIZE); /* insert new entry */ xad = &sp->xad[skip]; XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr); /* update page header */ sp->header.nextindex = cpu_to_le16(middle + 1); if (!test_cflag(COMMIT_Nolink, ip)) { sxtlck->lwm.offset = (sxtlck->lwm.offset) ? min(skip, (int)sxtlck->lwm.offset) : skip; } rp->header.nextindex = cpu_to_le16(XTENTRYSTART + righthalf); } /* * skip index in new right page - insert into right page: */ else { /* move left head of right half to right page */ n = skip - middle; memmove(&rp->xad[XTENTRYSTART], &sp->xad[middle], n << L2XTSLOTSIZE); /* insert new entry */ n += XTENTRYSTART; xad = &rp->xad[n]; XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr); /* move right tail of right half to right page */ if (skip < maxentry) memmove(&rp->xad[n + 1], &sp->xad[skip], (maxentry - skip) << L2XTSLOTSIZE); /* update page header */ sp->header.nextindex = cpu_to_le16(middle); if (!test_cflag(COMMIT_Nolink, ip)) { sxtlck->lwm.offset = (sxtlck->lwm.offset) ? min(middle, (int)sxtlck->lwm.offset) : middle; } rp->header.nextindex = cpu_to_le16(XTENTRYSTART + righthalf + 1); } if (!test_cflag(COMMIT_Nolink, ip)) { sxtlck->lwm.length = le16_to_cpu(sp->header.nextindex) - sxtlck->lwm.offset; /* rxtlck->lwm.offset = XTENTRYSTART; */ rxtlck->lwm.length = le16_to_cpu(rp->header.nextindex) - XTENTRYSTART; } *rmpp = rmp; *rbnp = rbn; jfs_info("xtSplitPage: sp:0x%p rp:0x%p", sp, rp); return rc; clean_up: /* Rollback quota allocation. */ if (quota_allocation) DQUOT_FREE_BLOCK(ip, quota_allocation); return (rc); } /* * xtSplitRoot() * * function: * split the full root page into original/root/split page and new * right page * i.e., root remains fixed in tree anchor (inode) and the root is * copied to a single new right child page since root page << * non-root page, and the split root page contains a single entry * for the new right child page. * * parameter: * int tid, * struct inode *ip, * struct xtsplit *split, * struct metapage **rmpp) * * return: * Pointer to page in which to insert or NULL on error. */ static int xtSplitRoot(tid_t tid, struct inode *ip, struct xtsplit * split, struct metapage ** rmpp) { xtpage_t *sp; struct metapage *rmp; xtpage_t *rp; s64 rbn; int skip, nextindex; xad_t *xad; pxd_t *pxd; struct pxdlist *pxdlist; struct tlock *tlck; struct xtlock *xtlck; sp = &JFS_IP(ip)->i_xtroot; INCREMENT(xtStat.split); /* * allocate a single (right) child page */ pxdlist = split->pxdlist; pxd = &pxdlist->pxd[pxdlist->npxd]; pxdlist->npxd++; rbn = addressPXD(pxd); rmp = get_metapage(ip, rbn, PSIZE, 1); if (rmp == NULL) return -EIO; /* Allocate blocks to quota. */ if (DQUOT_ALLOC_BLOCK(ip, lengthPXD(pxd))) { release_metapage(rmp); return -EDQUOT; } jfs_info("xtSplitRoot: ip:0x%p rmp:0x%p", ip, rmp); /* * acquire a transaction lock on the new right page; * * action: new page; */ BT_MARK_DIRTY(rmp, ip); rp = (xtpage_t *) rmp->data; rp->header.flag = (sp->header.flag & BT_LEAF) ? BT_LEAF : BT_INTERNAL; rp->header.self = *pxd; rp->header.nextindex = cpu_to_le16(XTENTRYSTART); rp->header.maxentry = cpu_to_le16(PSIZE >> L2XTSLOTSIZE); /* initialize sibling pointers */ rp->header.next = 0; rp->header.prev = 0; /* * copy the in-line root page into new right page extent */ nextindex = le16_to_cpu(sp->header.maxentry); memmove(&rp->xad[XTENTRYSTART], &sp->xad[XTENTRYSTART], (nextindex - XTENTRYSTART) << L2XTSLOTSIZE); /* * insert the new entry into the new right/child page * (skip index in the new right page will not change) */ skip = split->index; /* if insert into middle, shift right remaining entries */ if (skip != nextindex) memmove(&rp->xad[skip + 1], &rp->xad[skip], (nextindex - skip) * sizeof(xad_t)); xad = &rp->xad[skip]; XT_PUTENTRY(xad, split->flag, split->off, split->len, split->addr); /* update page header */ rp->header.nextindex = cpu_to_le16(nextindex + 1); if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, rmp, tlckXTREE | tlckNEW); xtlck = (struct xtlock *) & tlck->lock; xtlck->lwm.offset = XTENTRYSTART; xtlck->lwm.length = le16_to_cpu(rp->header.nextindex) - XTENTRYSTART; } /* * reset the root * * init root with the single entry for the new right page * set the 1st entry offset to 0, which force the left-most key * at any level of the tree to be less than any search key. */ /* * acquire a transaction lock on the root page (in-memory inode); * * action: root split; */ BT_MARK_DIRTY(split->mp, ip); xad = &sp->xad[XTENTRYSTART]; XT_PUTENTRY(xad, XAD_NEW, 0, JFS_SBI(ip->i_sb)->nbperpage, rbn); /* update page header of root */ sp->header.flag &= ~BT_LEAF; sp->header.flag |= BT_INTERNAL; sp->header.nextindex = cpu_to_le16(XTENTRYSTART + 1); if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, split->mp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; xtlck->lwm.offset = XTENTRYSTART; xtlck->lwm.length = 1; } *rmpp = rmp; jfs_info("xtSplitRoot: sp:0x%p rp:0x%p", sp, rp); return 0; } /* * xtExtend() * * function: extend in-place; * * note: existing extent may or may not have been committed. * caller is responsible for pager buffer cache update, and * working block allocation map update; * update pmap: alloc whole extended extent; */ int xtExtend(tid_t tid, /* transaction id */ struct inode *ip, s64 xoff, /* delta extent offset */ s32 xlen, /* delta extent length */ int flag) { int rc = 0; int cmp; struct metapage *mp; /* meta-page buffer */ xtpage_t *p; /* base B+-tree index page */ s64 bn; int index, nextindex, len; struct btstack btstack; /* traverse stack */ struct xtsplit split; /* split information */ xad_t *xad; s64 xaddr; struct tlock *tlck; struct xtlock *xtlck = NULL; jfs_info("xtExtend: nxoff:0x%lx nxlen:0x%x", (ulong) xoff, xlen); /* there must exist extent to be extended */ if ((rc = xtSearch(ip, xoff - 1, NULL, &cmp, &btstack, XT_INSERT))) return rc; /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, mp, p, index); if (cmp != 0) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtExtend: xtSearch did not find extent"); return -EIO; } /* extension must be contiguous */ xad = &p->xad[index]; if ((offsetXAD(xad) + lengthXAD(xad)) != xoff) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtExtend: extension is not contiguous"); return -EIO; } /* * acquire a transaction lock on the leaf page; * * action: xad insertion/extension; */ BT_MARK_DIRTY(mp, ip); if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; } /* extend will overflow extent ? */ xlen = lengthXAD(xad) + xlen; if ((len = xlen - MAXXLEN) <= 0) goto extendOld; /* * extent overflow: insert entry for new extent */ //insertNew: xoff = offsetXAD(xad) + MAXXLEN; xaddr = addressXAD(xad) + MAXXLEN; nextindex = le16_to_cpu(p->header.nextindex); /* * if the leaf page is full, insert the new entry and * propagate up the router entry for the new page from split * * The xtSplitUp() will insert the entry and unpin the leaf page. */ if (nextindex == le16_to_cpu(p->header.maxentry)) { /* xtSpliUp() unpins leaf pages */ split.mp = mp; split.index = index + 1; split.flag = XAD_NEW; split.off = xoff; /* split offset */ split.len = len; split.addr = xaddr; split.pxdlist = NULL; if ((rc = xtSplitUp(tid, ip, &split, &btstack))) return rc; /* get back old page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; /* * if leaf root has been split, original root has been * copied to new child page, i.e., original entry now * resides on the new child page; */ if (p->header.flag & BT_INTERNAL) { ASSERT(p->header.nextindex == cpu_to_le16(XTENTRYSTART + 1)); xad = &p->xad[XTENTRYSTART]; bn = addressXAD(xad); XT_PUTPAGE(mp); /* get new child page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; BT_MARK_DIRTY(mp, ip); if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW); xtlck = (struct xtlock *) & tlck->lock; } } } /* * insert the new entry into the leaf page */ else { /* insert the new entry: mark the entry NEW */ xad = &p->xad[index + 1]; XT_PUTENTRY(xad, XAD_NEW, xoff, len, xaddr); /* advance next available entry index */ p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1); } /* get back old entry */ xad = &p->xad[index]; xlen = MAXXLEN; /* * extend old extent */ extendOld: XADlength(xad, xlen); if (!(xad->flag & XAD_NEW)) xad->flag |= XAD_EXTENDED; if (!test_cflag(COMMIT_Nolink, ip)) { xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index, (int)xtlck->lwm.offset) : index; xtlck->lwm.length = le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset; } /* unpin the leaf page */ XT_PUTPAGE(mp); return rc; } #ifdef _NOTYET /* * xtTailgate() * * function: split existing 'tail' extent * (split offset >= start offset of tail extent), and * relocate and extend the split tail half; * * note: existing extent may or may not have been committed. * caller is responsible for pager buffer cache update, and * working block allocation map update; * update pmap: free old split tail extent, alloc new extent; */ int xtTailgate(tid_t tid, /* transaction id */ struct inode *ip, s64 xoff, /* split/new extent offset */ s32 xlen, /* new extent length */ s64 xaddr, /* new extent address */ int flag) { int rc = 0; int cmp; struct metapage *mp; /* meta-page buffer */ xtpage_t *p; /* base B+-tree index page */ s64 bn; int index, nextindex, llen, rlen; struct btstack btstack; /* traverse stack */ struct xtsplit split; /* split information */ xad_t *xad; struct tlock *tlck; struct xtlock *xtlck = 0; struct tlock *mtlck; struct maplock *pxdlock; /* printf("xtTailgate: nxoff:0x%lx nxlen:0x%x nxaddr:0x%lx\n", (ulong)xoff, xlen, (ulong)xaddr); */ /* there must exist extent to be tailgated */ if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, XT_INSERT))) return rc; /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, mp, p, index); if (cmp != 0) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtTailgate: couldn't find extent"); return -EIO; } /* entry found must be last entry */ nextindex = le16_to_cpu(p->header.nextindex); if (index != nextindex - 1) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtTailgate: the entry found is not the last entry"); return -EIO; } BT_MARK_DIRTY(mp, ip); /* * acquire tlock of the leaf page containing original entry */ if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; } /* completely replace extent ? */ xad = &p->xad[index]; /* printf("xtTailgate: xoff:0x%lx xlen:0x%x xaddr:0x%lx\n", (ulong)offsetXAD(xad), lengthXAD(xad), (ulong)addressXAD(xad)); */ if ((llen = xoff - offsetXAD(xad)) == 0) goto updateOld; /* * partially replace extent: insert entry for new extent */ //insertNew: /* * if the leaf page is full, insert the new entry and * propagate up the router entry for the new page from split * * The xtSplitUp() will insert the entry and unpin the leaf page. */ if (nextindex == le16_to_cpu(p->header.maxentry)) { /* xtSpliUp() unpins leaf pages */ split.mp = mp; split.index = index + 1; split.flag = XAD_NEW; split.off = xoff; /* split offset */ split.len = xlen; split.addr = xaddr; split.pxdlist = NULL; if ((rc = xtSplitUp(tid, ip, &split, &btstack))) return rc; /* get back old page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; /* * if leaf root has been split, original root has been * copied to new child page, i.e., original entry now * resides on the new child page; */ if (p->header.flag & BT_INTERNAL) { ASSERT(p->header.nextindex == cpu_to_le16(XTENTRYSTART + 1)); xad = &p->xad[XTENTRYSTART]; bn = addressXAD(xad); XT_PUTPAGE(mp); /* get new child page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; BT_MARK_DIRTY(mp, ip); if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW); xtlck = (struct xtlock *) & tlck->lock; } } } /* * insert the new entry into the leaf page */ else { /* insert the new entry: mark the entry NEW */ xad = &p->xad[index + 1]; XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr); /* advance next available entry index */ p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1); } /* get back old XAD */ xad = &p->xad[index]; /* * truncate/relocate old extent at split offset */ updateOld: /* update dmap for old/committed/truncated extent */ rlen = lengthXAD(xad) - llen; if (!(xad->flag & XAD_NEW)) { /* free from PWMAP at commit */ if (!test_cflag(COMMIT_Nolink, ip)) { mtlck = txMaplock(tid, ip, tlckMAP); pxdlock = (struct maplock *) & mtlck->lock; pxdlock->flag = mlckFREEPXD; PXDaddress(&pxdlock->pxd, addressXAD(xad) + llen); PXDlength(&pxdlock->pxd, rlen); pxdlock->index = 1; } } else /* free from WMAP */ dbFree(ip, addressXAD(xad) + llen, (s64) rlen); if (llen) /* truncate */ XADlength(xad, llen); else /* replace */ XT_PUTENTRY(xad, XAD_NEW, xoff, xlen, xaddr); if (!test_cflag(COMMIT_Nolink, ip)) { xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index, (int)xtlck->lwm.offset) : index; xtlck->lwm.length = le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset; } /* unpin the leaf page */ XT_PUTPAGE(mp); return rc; } #endif /* _NOTYET */ /* * xtUpdate() * * function: update XAD; * * update extent for allocated_but_not_recorded or * compressed extent; * * parameter: * nxad - new XAD; * logical extent of the specified XAD must be completely * contained by an existing XAD; */ int xtUpdate(tid_t tid, struct inode *ip, xad_t * nxad) { /* new XAD */ int rc = 0; int cmp; struct metapage *mp; /* meta-page buffer */ xtpage_t *p; /* base B+-tree index page */ s64 bn; int index0, index, newindex, nextindex; struct btstack btstack; /* traverse stack */ struct xtsplit split; /* split information */ xad_t *xad, *lxad, *rxad; int xflag; s64 nxoff, xoff; int nxlen, xlen, lxlen, rxlen; s64 nxaddr, xaddr; struct tlock *tlck; struct xtlock *xtlck = NULL; int newpage = 0; /* there must exist extent to be tailgated */ nxoff = offsetXAD(nxad); nxlen = lengthXAD(nxad); nxaddr = addressXAD(nxad); if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT))) return rc; /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0); if (cmp != 0) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtUpdate: Could not find extent"); return -EIO; } BT_MARK_DIRTY(mp, ip); /* * acquire tlock of the leaf page containing original entry */ if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; } xad = &p->xad[index0]; xflag = xad->flag; xoff = offsetXAD(xad); xlen = lengthXAD(xad); xaddr = addressXAD(xad); /* nXAD must be completely contained within XAD */ if ((xoff > nxoff) || (nxoff + nxlen > xoff + xlen)) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtUpdate: nXAD in not completely contained within XAD"); return -EIO; } index = index0; newindex = index + 1; nextindex = le16_to_cpu(p->header.nextindex); #ifdef _JFS_WIP_NOCOALESCE if (xoff < nxoff) goto updateRight; /* * replace XAD with nXAD */ replace: /* (nxoff == xoff) */ if (nxlen == xlen) { /* replace XAD with nXAD:recorded */ *xad = *nxad; xad->flag = xflag & ~XAD_NOTRECORDED; goto out; } else /* (nxlen < xlen) */ goto updateLeft; #endif /* _JFS_WIP_NOCOALESCE */ /* #ifdef _JFS_WIP_COALESCE */ if (xoff < nxoff) goto coalesceRight; /* * coalesce with left XAD */ //coalesceLeft: /* (xoff == nxoff) */ /* is XAD first entry of page ? */ if (index == XTENTRYSTART) goto replace; /* is nXAD logically and physically contiguous with lXAD ? */ lxad = &p->xad[index - 1]; lxlen = lengthXAD(lxad); if (!(lxad->flag & XAD_NOTRECORDED) && (nxoff == offsetXAD(lxad) + lxlen) && (nxaddr == addressXAD(lxad) + lxlen) && (lxlen + nxlen < MAXXLEN)) { /* extend right lXAD */ index0 = index - 1; XADlength(lxad, lxlen + nxlen); /* If we just merged two extents together, need to make sure the * right extent gets logged. If the left one is marked XAD_NEW, * then we know it will be logged. Otherwise, mark as * XAD_EXTENDED */ if (!(lxad->flag & XAD_NEW)) lxad->flag |= XAD_EXTENDED; if (xlen > nxlen) { /* truncate XAD */ XADoffset(xad, xoff + nxlen); XADlength(xad, xlen - nxlen); XADaddress(xad, xaddr + nxlen); goto out; } else { /* (xlen == nxlen) */ /* remove XAD */ if (index < nextindex - 1) memmove(&p->xad[index], &p->xad[index + 1], (nextindex - index - 1) << L2XTSLOTSIZE); p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) - 1); index = index0; newindex = index + 1; nextindex = le16_to_cpu(p->header.nextindex); xoff = nxoff = offsetXAD(lxad); xlen = nxlen = lxlen + nxlen; xaddr = nxaddr = addressXAD(lxad); goto coalesceRight; } } /* * replace XAD with nXAD */ replace: /* (nxoff == xoff) */ if (nxlen == xlen) { /* replace XAD with nXAD:recorded */ *xad = *nxad; xad->flag = xflag & ~XAD_NOTRECORDED; goto coalesceRight; } else /* (nxlen < xlen) */ goto updateLeft; /* * coalesce with right XAD */ coalesceRight: /* (xoff <= nxoff) */ /* is XAD last entry of page ? */ if (newindex == nextindex) { if (xoff == nxoff) goto out; goto updateRight; } /* is nXAD logically and physically contiguous with rXAD ? */ rxad = &p->xad[index + 1]; rxlen = lengthXAD(rxad); if (!(rxad->flag & XAD_NOTRECORDED) && (nxoff + nxlen == offsetXAD(rxad)) && (nxaddr + nxlen == addressXAD(rxad)) && (rxlen + nxlen < MAXXLEN)) { /* extend left rXAD */ XADoffset(rxad, nxoff); XADlength(rxad, rxlen + nxlen); XADaddress(rxad, nxaddr); /* If we just merged two extents together, need to make sure * the left extent gets logged. If the right one is marked * XAD_NEW, then we know it will be logged. Otherwise, mark as * XAD_EXTENDED */ if (!(rxad->flag & XAD_NEW)) rxad->flag |= XAD_EXTENDED; if (xlen > nxlen) /* truncate XAD */ XADlength(xad, xlen - nxlen); else { /* (xlen == nxlen) */ /* remove XAD */ memmove(&p->xad[index], &p->xad[index + 1], (nextindex - index - 1) << L2XTSLOTSIZE); p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) - 1); } goto out; } else if (xoff == nxoff) goto out; if (xoff >= nxoff) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtUpdate: xoff >= nxoff"); return -EIO; } /* #endif _JFS_WIP_COALESCE */ /* * split XAD into (lXAD, nXAD): * * |---nXAD---> * --|----------XAD----------|-- * |-lXAD-| */ updateRight: /* (xoff < nxoff) */ /* truncate old XAD as lXAD:not_recorded */ xad = &p->xad[index]; XADlength(xad, nxoff - xoff); /* insert nXAD:recorded */ if (nextindex == le16_to_cpu(p->header.maxentry)) { /* xtSpliUp() unpins leaf pages */ split.mp = mp; split.index = newindex; split.flag = xflag & ~XAD_NOTRECORDED; split.off = nxoff; split.len = nxlen; split.addr = nxaddr; split.pxdlist = NULL; if ((rc = xtSplitUp(tid, ip, &split, &btstack))) return rc; /* get back old page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; /* * if leaf root has been split, original root has been * copied to new child page, i.e., original entry now * resides on the new child page; */ if (p->header.flag & BT_INTERNAL) { ASSERT(p->header.nextindex == cpu_to_le16(XTENTRYSTART + 1)); xad = &p->xad[XTENTRYSTART]; bn = addressXAD(xad); XT_PUTPAGE(mp); /* get new child page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; BT_MARK_DIRTY(mp, ip); if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW); xtlck = (struct xtlock *) & tlck->lock; } } else { /* is nXAD on new page ? */ if (newindex > (le16_to_cpu(p->header.maxentry) >> 1)) { newindex = newindex - le16_to_cpu(p->header.nextindex) + XTENTRYSTART; newpage = 1; } } } else { /* if insert into middle, shift right remaining entries */ if (newindex < nextindex) memmove(&p->xad[newindex + 1], &p->xad[newindex], (nextindex - newindex) << L2XTSLOTSIZE); /* insert the entry */ xad = &p->xad[newindex]; *xad = *nxad; xad->flag = xflag & ~XAD_NOTRECORDED; /* advance next available entry index. */ p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1); } /* * does nXAD force 3-way split ? * * |---nXAD--->| * --|----------XAD-------------|-- * |-lXAD-| |-rXAD -| */ if (nxoff + nxlen == xoff + xlen) goto out; /* reorient nXAD as XAD for further split XAD into (nXAD, rXAD) */ if (newpage) { /* close out old page */ if (!test_cflag(COMMIT_Nolink, ip)) { xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index0, (int)xtlck->lwm.offset) : index0; xtlck->lwm.length = le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset; } bn = le64_to_cpu(p->header.next); XT_PUTPAGE(mp); /* get new right page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; BT_MARK_DIRTY(mp, ip); if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; } index0 = index = newindex; } else index++; newindex = index + 1; nextindex = le16_to_cpu(p->header.nextindex); xlen = xlen - (nxoff - xoff); xoff = nxoff; xaddr = nxaddr; /* recompute split pages */ if (nextindex == le16_to_cpu(p->header.maxentry)) { XT_PUTPAGE(mp); if ((rc = xtSearch(ip, nxoff, NULL, &cmp, &btstack, XT_INSERT))) return rc; /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, mp, p, index0); if (cmp != 0) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtUpdate: xtSearch failed"); return -EIO; } if (index0 != index) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtUpdate: unexpected value of index"); return -EIO; } } /* * split XAD into (nXAD, rXAD) * * ---nXAD---| * --|----------XAD----------|-- * |-rXAD-| */ updateLeft: /* (nxoff == xoff) && (nxlen < xlen) */ /* update old XAD with nXAD:recorded */ xad = &p->xad[index]; *xad = *nxad; xad->flag = xflag & ~XAD_NOTRECORDED; /* insert rXAD:not_recorded */ xoff = xoff + nxlen; xlen = xlen - nxlen; xaddr = xaddr + nxlen; if (nextindex == le16_to_cpu(p->header.maxentry)) { /* printf("xtUpdate.updateLeft.split p:0x%p\n", p); */ /* xtSpliUp() unpins leaf pages */ split.mp = mp; split.index = newindex; split.flag = xflag; split.off = xoff; split.len = xlen; split.addr = xaddr; split.pxdlist = NULL; if ((rc = xtSplitUp(tid, ip, &split, &btstack))) return rc; /* get back old page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; /* * if leaf root has been split, original root has been * copied to new child page, i.e., original entry now * resides on the new child page; */ if (p->header.flag & BT_INTERNAL) { ASSERT(p->header.nextindex == cpu_to_le16(XTENTRYSTART + 1)); xad = &p->xad[XTENTRYSTART]; bn = addressXAD(xad); XT_PUTPAGE(mp); /* get new child page */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; BT_MARK_DIRTY(mp, ip); if (!test_cflag(COMMIT_Nolink, ip)) { tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW); xtlck = (struct xtlock *) & tlck->lock; } } } else { /* if insert into middle, shift right remaining entries */ if (newindex < nextindex) memmove(&p->xad[newindex + 1], &p->xad[newindex], (nextindex - newindex) << L2XTSLOTSIZE); /* insert the entry */ xad = &p->xad[newindex]; XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr); /* advance next available entry index. */ p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1); } out: if (!test_cflag(COMMIT_Nolink, ip)) { xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index0, (int)xtlck->lwm.offset) : index0; xtlck->lwm.length = le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset; } /* unpin the leaf page */ XT_PUTPAGE(mp); return rc; } /* * xtAppend() * * function: grow in append mode from contiguous region specified ; * * parameter: * tid - transaction id; * ip - file object; * xflag - extent flag: * xoff - extent offset; * maxblocks - max extent length; * xlen - extent length (in/out); * xaddrp - extent address pointer (in/out): * flag - * * return: */ int xtAppend(tid_t tid, /* transaction id */ struct inode *ip, int xflag, s64 xoff, s32 maxblocks, s32 * xlenp, /* (in/out) */ s64 * xaddrp, /* (in/out) */ int flag) { int rc = 0; struct metapage *mp; /* meta-page buffer */ xtpage_t *p; /* base B+-tree index page */ s64 bn, xaddr; int index, nextindex; struct btstack btstack; /* traverse stack */ struct xtsplit split; /* split information */ xad_t *xad; int cmp; struct tlock *tlck; struct xtlock *xtlck; int nsplit, nblocks, xlen; struct pxdlist pxdlist; pxd_t *pxd; s64 next; xaddr = *xaddrp; xlen = *xlenp; jfs_info("xtAppend: xoff:0x%lx maxblocks:%d xlen:%d xaddr:0x%lx", (ulong) xoff, maxblocks, xlen, (ulong) xaddr); /* * search for the entry location at which to insert: * * xtFastSearch() and xtSearch() both returns (leaf page * pinned, index at which to insert). * n.b. xtSearch() may return index of maxentry of * the full page. */ if ((rc = xtSearch(ip, xoff, &next, &cmp, &btstack, XT_INSERT))) return rc; /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, mp, p, index); if (cmp == 0) { rc = -EEXIST; goto out; } if (next) xlen = min(xlen, (int)(next - xoff)); //insert: /* * insert entry for new extent */ xflag |= XAD_NEW; /* * if the leaf page is full, split the page and * propagate up the router entry for the new page from split * * The xtSplitUp() will insert the entry and unpin the leaf page. */ nextindex = le16_to_cpu(p->header.nextindex); if (nextindex < le16_to_cpu(p->header.maxentry)) goto insertLeaf; /* * allocate new index blocks to cover index page split(s) */ nsplit = btstack.nsplit; split.pxdlist = &pxdlist; pxdlist.maxnpxd = pxdlist.npxd = 0; pxd = &pxdlist.pxd[0]; nblocks = JFS_SBI(ip->i_sb)->nbperpage; for (; nsplit > 0; nsplit--, pxd++, xaddr += nblocks, maxblocks -= nblocks) { if ((rc = dbAllocBottomUp(ip, xaddr, (s64) nblocks)) == 0) { PXDaddress(pxd, xaddr); PXDlength(pxd, nblocks); pxdlist.maxnpxd++; continue; } /* undo allocation */ goto out; } xlen = min(xlen, maxblocks); /* * allocate data extent requested */ if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen))) goto out; split.mp = mp; split.index = index; split.flag = xflag; split.off = xoff; split.len = xlen; split.addr = xaddr; if ((rc = xtSplitUp(tid, ip, &split, &btstack))) { /* undo data extent allocation */ dbFree(ip, *xaddrp, (s64) * xlenp); return rc; } *xaddrp = xaddr; *xlenp = xlen; return 0; /* * insert the new entry into the leaf page */ insertLeaf: /* * allocate data extent requested */ if ((rc = dbAllocBottomUp(ip, xaddr, (s64) xlen))) goto out; BT_MARK_DIRTY(mp, ip); /* * acquire a transaction lock on the leaf page; * * action: xad insertion/extension; */ tlck = txLock(tid, ip, mp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; /* insert the new entry: mark the entry NEW */ xad = &p->xad[index]; XT_PUTENTRY(xad, xflag, xoff, xlen, xaddr); /* advance next available entry index */ p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) + 1); xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index,(int) xtlck->lwm.offset) : index; xtlck->lwm.length = le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset; *xaddrp = xaddr; *xlenp = xlen; out: /* unpin the leaf page */ XT_PUTPAGE(mp); return rc; } #ifdef _STILL_TO_PORT /* - TBD for defragmentaion/reorganization - * * xtDelete() * * function: * delete the entry with the specified key. * * N.B.: whole extent of the entry is assumed to be deleted. * * parameter: * * return: * ENOENT: if the entry is not found. * * exception: */ int xtDelete(tid_t tid, struct inode *ip, s64 xoff, s32 xlen, int flag) { int rc = 0; struct btstack btstack; int cmp; s64 bn; struct metapage *mp; xtpage_t *p; int index, nextindex; struct tlock *tlck; struct xtlock *xtlck; /* * find the matching entry; xtSearch() pins the page */ if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0))) return rc; XT_GETSEARCH(ip, btstack.top, bn, mp, p, index); if (cmp) { /* unpin the leaf page */ XT_PUTPAGE(mp); return -ENOENT; } /* * delete the entry from the leaf page */ nextindex = le16_to_cpu(p->header.nextindex); p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) - 1); /* * if the leaf page bocome empty, free the page */ if (p->header.nextindex == cpu_to_le16(XTENTRYSTART)) return (xtDeleteUp(tid, ip, mp, p, &btstack)); BT_MARK_DIRTY(mp, ip); /* * acquire a transaction lock on the leaf page; * * action:xad deletion; */ tlck = txLock(tid, ip, mp, tlckXTREE); xtlck = (struct xtlock *) & tlck->lock; xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index, xtlck->lwm.offset) : index; /* if delete from middle, shift left/compact the remaining entries */ if (index < nextindex - 1) memmove(&p->xad[index], &p->xad[index + 1], (nextindex - index - 1) * sizeof(xad_t)); XT_PUTPAGE(mp); return 0; } /* - TBD for defragmentaion/reorganization - * * xtDeleteUp() * * function: * free empty pages as propagating deletion up the tree * * parameter: * * return: */ static int xtDeleteUp(tid_t tid, struct inode *ip, struct metapage * fmp, xtpage_t * fp, struct btstack * btstack) { int rc = 0; struct metapage *mp; xtpage_t *p; int index, nextindex; s64 xaddr; int xlen; struct btframe *parent; struct tlock *tlck; struct xtlock *xtlck; /* * keep root leaf page which has become empty */ if (fp->header.flag & BT_ROOT) { /* keep the root page */ fp->header.flag &= ~BT_INTERNAL; fp->header.flag |= BT_LEAF; fp->header.nextindex = cpu_to_le16(XTENTRYSTART); /* XT_PUTPAGE(fmp); */ return 0; } /* * free non-root leaf page */ if ((rc = xtRelink(tid, ip, fp))) { XT_PUTPAGE(fmp); return rc; } xaddr = addressPXD(&fp->header.self); xlen = lengthPXD(&fp->header.self); /* free the page extent */ dbFree(ip, xaddr, (s64) xlen); /* free the buffer page */ discard_metapage(fmp); /* * propagate page deletion up the index tree * * If the delete from the parent page makes it empty, * continue all the way up the tree. * stop if the root page is reached (which is never deleted) or * if the entry deletion does not empty the page. */ while ((parent = BT_POP(btstack)) != NULL) { /* get/pin the parent page <sp> */ XT_GETPAGE(ip, parent->bn, mp, PSIZE, p, rc); if (rc) return rc; index = parent->index; /* delete the entry for the freed child page from parent. */ nextindex = le16_to_cpu(p->header.nextindex); /* * the parent has the single entry being deleted: * free the parent page which has become empty. */ if (nextindex == 1) { if (p->header.flag & BT_ROOT) { /* keep the root page */ p->header.flag &= ~BT_INTERNAL; p->header.flag |= BT_LEAF; p->header.nextindex = cpu_to_le16(XTENTRYSTART); /* XT_PUTPAGE(mp); */ break; } else { /* free the parent page */ if ((rc = xtRelink(tid, ip, p))) return rc; xaddr = addressPXD(&p->header.self); /* free the page extent */ dbFree(ip, xaddr, (s64) JFS_SBI(ip->i_sb)->nbperpage); /* unpin/free the buffer page */ discard_metapage(mp); /* propagate up */ continue; } } /* * the parent has other entries remaining: * delete the router entry from the parent page. */ else { BT_MARK_DIRTY(mp, ip); /* * acquire a transaction lock on the leaf page; * * action:xad deletion; */ tlck = txLock(tid, ip, mp, tlckXTREE); xtlck = (struct xtlock *) & tlck->lock; xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index, xtlck->lwm. offset) : index; /* if delete from middle, * shift left/compact the remaining entries in the page */ if (index < nextindex - 1) memmove(&p->xad[index], &p->xad[index + 1], (nextindex - index - 1) << L2XTSLOTSIZE); p->header.nextindex = cpu_to_le16(le16_to_cpu(p->header.nextindex) - 1); jfs_info("xtDeleteUp(entry): 0x%lx[%d]", (ulong) parent->bn, index); } /* unpin the parent page */ XT_PUTPAGE(mp); /* exit propagation up */ break; } return 0; } /* * NAME: xtRelocate() * * FUNCTION: relocate xtpage or data extent of regular file; * This function is mainly used by defragfs utility. * * NOTE: This routine does not have the logic to handle * uncommitted allocated extent. The caller should call * txCommit() to commit all the allocation before call * this routine. */ int xtRelocate(tid_t tid, struct inode * ip, xad_t * oxad, /* old XAD */ s64 nxaddr, /* new xaddr */ int xtype) { /* extent type: XTPAGE or DATAEXT */ int rc = 0; struct tblock *tblk; struct tlock *tlck; struct xtlock *xtlck; struct metapage *mp, *pmp, *lmp, *rmp; /* meta-page buffer */ xtpage_t *p, *pp, *rp, *lp; /* base B+-tree index page */ xad_t *xad; pxd_t *pxd; s64 xoff, xsize; int xlen; s64 oxaddr, sxaddr, dxaddr, nextbn, prevbn; cbuf_t *cp; s64 offset, nbytes, nbrd, pno; int nb, npages, nblks; s64 bn; int cmp; int index; struct pxd_lock *pxdlock; struct btstack btstack; /* traverse stack */ xtype = xtype & EXTENT_TYPE; xoff = offsetXAD(oxad); oxaddr = addressXAD(oxad); xlen = lengthXAD(oxad); /* validate extent offset */ offset = xoff << JFS_SBI(ip->i_sb)->l2bsize; if (offset >= ip->i_size) return -ESTALE; /* stale extent */ jfs_info("xtRelocate: xtype:%d xoff:0x%lx xlen:0x%x xaddr:0x%lx:0x%lx", xtype, (ulong) xoff, xlen, (ulong) oxaddr, (ulong) nxaddr); /* * 1. get and validate the parent xtpage/xad entry * covering the source extent to be relocated; */ if (xtype == DATAEXT) { /* search in leaf entry */ rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0); if (rc) return rc; /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index); if (cmp) { XT_PUTPAGE(pmp); return -ESTALE; } /* validate for exact match with a single entry */ xad = &pp->xad[index]; if (addressXAD(xad) != oxaddr || lengthXAD(xad) != xlen) { XT_PUTPAGE(pmp); return -ESTALE; } } else { /* (xtype == XTPAGE) */ /* search in internal entry */ rc = xtSearchNode(ip, oxad, &cmp, &btstack, 0); if (rc) return rc; /* retrieve search result */ XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index); if (cmp) { XT_PUTPAGE(pmp); return -ESTALE; } /* xtSearchNode() validated for exact match with a single entry */ xad = &pp->xad[index]; } jfs_info("xtRelocate: parent xad entry validated."); /* * 2. relocate the extent */ if (xtype == DATAEXT) { /* if the extent is allocated-but-not-recorded * there is no real data to be moved in this extent, */ if (xad->flag & XAD_NOTRECORDED) goto out; else /* release xtpage for cmRead()/xtLookup() */ XT_PUTPAGE(pmp); /* * cmRelocate() * * copy target data pages to be relocated; * * data extent must start at page boundary and * multiple of page size (except the last data extent); * read in each page of the source data extent into cbuf, * update the cbuf extent descriptor of the page to be * homeward bound to new dst data extent * copy the data from the old extent to new extent. * copy is essential for compressed files to avoid problems * that can arise if there was a change in compression * algorithms. * it is a good strategy because it may disrupt cache * policy to keep the pages in memory afterwards. */ offset = xoff << JFS_SBI(ip->i_sb)->l2bsize; assert((offset & CM_OFFSET) == 0); nbytes = xlen << JFS_SBI(ip->i_sb)->l2bsize; pno = offset >> CM_L2BSIZE; npages = (nbytes + (CM_BSIZE - 1)) >> CM_L2BSIZE; /* npages = ((offset + nbytes - 1) >> CM_L2BSIZE) - (offset >> CM_L2BSIZE) + 1; */ sxaddr = oxaddr; dxaddr = nxaddr; /* process the request one cache buffer at a time */ for (nbrd = 0; nbrd < nbytes; nbrd += nb, offset += nb, pno++, npages--) { /* compute page size */ nb = min(nbytes - nbrd, CM_BSIZE); /* get the cache buffer of the page */ if (rc = cmRead(ip, offset, npages, &cp)) break; assert(addressPXD(&cp->cm_pxd) == sxaddr); assert(!cp->cm_modified); /* bind buffer with the new extent address */ nblks = nb >> JFS_IP(ip->i_sb)->l2bsize; cmSetXD(ip, cp, pno, dxaddr, nblks); /* release the cbuf, mark it as modified */ cmPut(cp, true); dxaddr += nblks; sxaddr += nblks; } /* get back parent page */ if ((rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0))) return rc; XT_GETSEARCH(ip, btstack.top, bn, pmp, pp, index); jfs_info("xtRelocate: target data extent relocated."); } else { /* (xtype == XTPAGE) */ /* * read in the target xtpage from the source extent; */ XT_GETPAGE(ip, oxaddr, mp, PSIZE, p, rc); if (rc) { XT_PUTPAGE(pmp); return rc; } /* * read in sibling pages if any to update sibling pointers; */ rmp = NULL; if (p->header.next) { nextbn = le64_to_cpu(p->header.next); XT_GETPAGE(ip, nextbn, rmp, PSIZE, rp, rc); if (rc) { XT_PUTPAGE(pmp); XT_PUTPAGE(mp); return (rc); } } lmp = NULL; if (p->header.prev) { prevbn = le64_to_cpu(p->header.prev); XT_GETPAGE(ip, prevbn, lmp, PSIZE, lp, rc); if (rc) { XT_PUTPAGE(pmp); XT_PUTPAGE(mp); if (rmp) XT_PUTPAGE(rmp); return (rc); } } /* at this point, all xtpages to be updated are in memory */ /* * update sibling pointers of sibling xtpages if any; */ if (lmp) { BT_MARK_DIRTY(lmp, ip); tlck = txLock(tid, ip, lmp, tlckXTREE | tlckRELINK); lp->header.next = cpu_to_le64(nxaddr); XT_PUTPAGE(lmp); } if (rmp) { BT_MARK_DIRTY(rmp, ip); tlck = txLock(tid, ip, rmp, tlckXTREE | tlckRELINK); rp->header.prev = cpu_to_le64(nxaddr); XT_PUTPAGE(rmp); } /* * update the target xtpage to be relocated * * update the self address of the target page * and write to destination extent; * redo image covers the whole xtpage since it is new page * to the destination extent; * update of bmap for the free of source extent * of the target xtpage itself: * update of bmap for the allocation of destination extent * of the target xtpage itself: * update of bmap for the extents covered by xad entries in * the target xtpage is not necessary since they are not * updated; * if not committed before this relocation, * target page may contain XAD_NEW entries which must * be scanned for bmap update (logredo() always * scan xtpage REDOPAGE image for bmap update); * if committed before this relocation (tlckRELOCATE), * scan may be skipped by commit() and logredo(); */ BT_MARK_DIRTY(mp, ip); /* tlckNEW init xtlck->lwm.offset = XTENTRYSTART; */ tlck = txLock(tid, ip, mp, tlckXTREE | tlckNEW); xtlck = (struct xtlock *) & tlck->lock; /* update the self address in the xtpage header */ pxd = &p->header.self; PXDaddress(pxd, nxaddr); /* linelock for the after image of the whole page */ xtlck->lwm.length = le16_to_cpu(p->header.nextindex) - xtlck->lwm.offset; /* update the buffer extent descriptor of target xtpage */ xsize = xlen << JFS_SBI(ip->i_sb)->l2bsize; bmSetXD(mp, nxaddr, xsize); /* unpin the target page to new homeward bound */ XT_PUTPAGE(mp); jfs_info("xtRelocate: target xtpage relocated."); } /* * 3. acquire maplock for the source extent to be freed; * * acquire a maplock saving the src relocated extent address; * to free of the extent at commit time; */ out: /* if DATAEXT relocation, write a LOG_UPDATEMAP record for * free PXD of the source data extent (logredo() will update * bmap for free of source data extent), and update bmap for * free of the source data extent; */ if (xtype == DATAEXT) tlck = txMaplock(tid, ip, tlckMAP); /* if XTPAGE relocation, write a LOG_NOREDOPAGE record * for the source xtpage (logredo() will init NoRedoPage * filter and will also update bmap for free of the source * xtpage), and update bmap for free of the source xtpage; * N.B. We use tlckMAP instead of tlkcXTREE because there * is no buffer associated with this lock since the buffer * has been redirected to the target location. */ else /* (xtype == XTPAGE) */ tlck = txMaplock(tid, ip, tlckMAP | tlckRELOCATE); pxdlock = (struct pxd_lock *) & tlck->lock; pxdlock->flag = mlckFREEPXD; PXDaddress(&pxdlock->pxd, oxaddr); PXDlength(&pxdlock->pxd, xlen); pxdlock->index = 1; /* * 4. update the parent xad entry for relocation; * * acquire tlck for the parent entry with XAD_NEW as entry * update which will write LOG_REDOPAGE and update bmap for * allocation of XAD_NEW destination extent; */ jfs_info("xtRelocate: update parent xad entry."); BT_MARK_DIRTY(pmp, ip); tlck = txLock(tid, ip, pmp, tlckXTREE | tlckGROW); xtlck = (struct xtlock *) & tlck->lock; /* update the XAD with the new destination extent; */ xad = &pp->xad[index]; xad->flag |= XAD_NEW; XADaddress(xad, nxaddr); xtlck->lwm.offset = min(index, xtlck->lwm.offset); xtlck->lwm.length = le16_to_cpu(pp->header.nextindex) - xtlck->lwm.offset; /* unpin the parent xtpage */ XT_PUTPAGE(pmp); return rc; } /* * xtSearchNode() * * function: search for the internal xad entry covering specified extent. * This function is mainly used by defragfs utility. * * parameters: * ip - file object; * xad - extent to find; * cmpp - comparison result: * btstack - traverse stack; * flag - search process flag; * * returns: * btstack contains (bn, index) of search path traversed to the entry. * *cmpp is set to result of comparison with the entry returned. * the page containing the entry is pinned at exit. */ static int xtSearchNode(struct inode *ip, xad_t * xad, /* required XAD entry */ int *cmpp, struct btstack * btstack, int flag) { int rc = 0; s64 xoff, xaddr; int xlen; int cmp = 1; /* init for empty page */ s64 bn; /* block number */ struct metapage *mp; /* meta-page buffer */ xtpage_t *p; /* page */ int base, index, lim; struct btframe *btsp; s64 t64; BT_CLR(btstack); xoff = offsetXAD(xad); xlen = lengthXAD(xad); xaddr = addressXAD(xad); /* * search down tree from root: * * between two consecutive entries of <Ki, Pi> and <Kj, Pj> of * internal page, child page Pi contains entry with k, Ki <= K < Kj. * * if entry with search key K is not found * internal page search find the entry with largest key Ki * less than K which point to the child page to search; * leaf page search find the entry with smallest key Kj * greater than K so that the returned index is the position of * the entry to be shifted right for insertion of new entry. * for empty tree, search key is greater than any key of the tree. * * by convention, root bn = 0. */ for (bn = 0;;) { /* get/pin the page to search */ XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; if (p->header.flag & BT_LEAF) { XT_PUTPAGE(mp); return -ESTALE; } lim = le16_to_cpu(p->header.nextindex) - XTENTRYSTART; /* * binary search with search key K on the current page */ for (base = XTENTRYSTART; lim; lim >>= 1) { index = base + (lim >> 1); XT_CMP(cmp, xoff, &p->xad[index], t64); if (cmp == 0) { /* * search hit * * verify for exact match; */ if (xaddr == addressXAD(&p->xad[index]) && xoff == offsetXAD(&p->xad[index])) { *cmpp = cmp; /* save search result */ btsp = btstack->top; btsp->bn = bn; btsp->index = index; btsp->mp = mp; return 0; } /* descend/search its child page */ goto next; } if (cmp > 0) { base = index + 1; --lim; } } /* * search miss - non-leaf page: * * base is the smallest index with key (Kj) greater than * search key (K) and may be zero or maxentry index. * if base is non-zero, decrement base by one to get the parent * entry of the child page to search. */ index = base ? base - 1 : base; /* * go down to child page */ next: /* get the child page block number */ bn = addressXAD(&p->xad[index]); /* unpin the parent page */ XT_PUTPAGE(mp); } } /* * xtRelink() * * function: * link around a freed page. * * Parameter: * int tid, * struct inode *ip, * xtpage_t *p) * * returns: */ static int xtRelink(tid_t tid, struct inode *ip, xtpage_t * p) { int rc = 0; struct metapage *mp; s64 nextbn, prevbn; struct tlock *tlck; nextbn = le64_to_cpu(p->header.next); prevbn = le64_to_cpu(p->header.prev); /* update prev pointer of the next page */ if (nextbn != 0) { XT_GETPAGE(ip, nextbn, mp, PSIZE, p, rc); if (rc) return rc; /* * acquire a transaction lock on the page; * * action: update prev pointer; */ BT_MARK_DIRTY(mp, ip); tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK); /* the page may already have been tlock'd */ p->header.prev = cpu_to_le64(prevbn); XT_PUTPAGE(mp); } /* update next pointer of the previous page */ if (prevbn != 0) { XT_GETPAGE(ip, prevbn, mp, PSIZE, p, rc); if (rc) return rc; /* * acquire a transaction lock on the page; * * action: update next pointer; */ BT_MARK_DIRTY(mp, ip); tlck = txLock(tid, ip, mp, tlckXTREE | tlckRELINK); /* the page may already have been tlock'd */ p->header.next = le64_to_cpu(nextbn); XT_PUTPAGE(mp); } return 0; } #endif /* _STILL_TO_PORT */ /* * xtInitRoot() * * initialize file root (inline in inode) */ void xtInitRoot(tid_t tid, struct inode *ip) { xtpage_t *p; /* * acquire a transaction lock on the root * * action: */ txLock(tid, ip, (struct metapage *) &JFS_IP(ip)->bxflag, tlckXTREE | tlckNEW); p = &JFS_IP(ip)->i_xtroot; p->header.flag = DXD_INDEX | BT_ROOT | BT_LEAF; p->header.nextindex = cpu_to_le16(XTENTRYSTART); if (S_ISDIR(ip->i_mode)) p->header.maxentry = cpu_to_le16(XTROOTINITSLOT_DIR); else { p->header.maxentry = cpu_to_le16(XTROOTINITSLOT); ip->i_size = 0; } return; } /* * We can run into a deadlock truncating a file with a large number of * xtree pages (large fragmented file). A robust fix would entail a * reservation system where we would reserve a number of metadata pages * and tlocks which we would be guaranteed without a deadlock. Without * this, a partial fix is to limit number of metadata pages we will lock * in a single transaction. Currently we will truncate the file so that * no more than 50 leaf pages will be locked. The caller of xtTruncate * will be responsible for ensuring that the current transaction gets * committed, and that subsequent transactions are created to truncate * the file further if needed. */ #define MAX_TRUNCATE_LEAVES 50 /* * xtTruncate() * * function: * traverse for truncation logging backward bottom up; * terminate at the last extent entry at the current subtree * root page covering new down size. * truncation may occur within the last extent entry. * * parameter: * int tid, * struct inode *ip, * s64 newsize, * int type) {PWMAP, PMAP, WMAP; DELETE, TRUNCATE} * * return: * * note: * PWMAP: * 1. truncate (non-COMMIT_NOLINK file) * by jfs_truncate() or jfs_open(O_TRUNC): * xtree is updated; * 2. truncate index table of directory when last entry removed * map update via tlock at commit time; * PMAP: * Call xtTruncate_pmap instead * WMAP: * 1. remove (free zero link count) on last reference release * (pmap has been freed at commit zero link count); * 2. truncate (COMMIT_NOLINK file, i.e., tmp file): * xtree is updated; * map update directly at truncation time; * * if (DELETE) * no LOG_NOREDOPAGE is required (NOREDOFILE is sufficient); * else if (TRUNCATE) * must write LOG_NOREDOPAGE for deleted index page; * * pages may already have been tlocked by anonymous transactions * during file growth (i.e., write) before truncation; * * except last truncated entry, deleted entries remains as is * in the page (nextindex is updated) for other use * (e.g., log/update allocation map): this avoid copying the page * info but delay free of pages; * */ s64 xtTruncate(tid_t tid, struct inode *ip, s64 newsize, int flag) { int rc = 0; s64 teof; struct metapage *mp; xtpage_t *p; s64 bn; int index, nextindex; xad_t *xad; s64 xoff, xaddr; int xlen, len, freexlen; struct btstack btstack; struct btframe *parent; struct tblock *tblk = NULL; struct tlock *tlck = NULL; struct xtlock *xtlck = NULL; struct xdlistlock xadlock; /* maplock for COMMIT_WMAP */ struct pxd_lock *pxdlock; /* maplock for COMMIT_WMAP */ s64 nfreed; int freed, log; int locked_leaves = 0; /* save object truncation type */ if (tid) { tblk = tid_to_tblock(tid); tblk->xflag |= flag; } nfreed = 0; flag &= COMMIT_MAP; assert(flag != COMMIT_PMAP); if (flag == COMMIT_PWMAP) log = 1; else { log = 0; xadlock.flag = mlckFREEXADLIST; xadlock.index = 1; } /* * if the newsize is not an integral number of pages, * the file between newsize and next page boundary will * be cleared. * if truncating into a file hole, it will cause * a full block to be allocated for the logical block. */ /* * release page blocks of truncated region <teof, eof> * * free the data blocks from the leaf index blocks. * delete the parent index entries corresponding to * the freed child data/index blocks. * free the index blocks themselves which aren't needed * in new sized file. * * index blocks are updated only if the blocks are to be * retained in the new sized file. * if type is PMAP, the data and index pages are NOT * freed, and the data and index blocks are NOT freed * from working map. * (this will allow continued access of data/index of * temporary file (zerolink count file truncated to zero-length)). */ teof = (newsize + (JFS_SBI(ip->i_sb)->bsize - 1)) >> JFS_SBI(ip->i_sb)->l2bsize; /* clear stack */ BT_CLR(&btstack); /* * start with root * * root resides in the inode */ bn = 0; /* * first access of each page: */ getPage: XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; /* process entries backward from last index */ index = le16_to_cpu(p->header.nextindex) - 1; /* Since this is the rightmost page at this level, and we may have * already freed a page that was formerly to the right, let's make * sure that the next pointer is zero. */ if (p->header.next) { if (log) /* * Make sure this change to the header is logged. * If we really truncate this leaf, the flag * will be changed to tlckTRUNCATE */ tlck = txLock(tid, ip, mp, tlckXTREE|tlckGROW); BT_MARK_DIRTY(mp, ip); p->header.next = 0; } if (p->header.flag & BT_INTERNAL) goto getChild; /* * leaf page */ freed = 0; /* does region covered by leaf page precede Teof ? */ xad = &p->xad[index]; xoff = offsetXAD(xad); xlen = lengthXAD(xad); if (teof >= xoff + xlen) { XT_PUTPAGE(mp); goto getParent; } /* (re)acquire tlock of the leaf page */ if (log) { if (++locked_leaves > MAX_TRUNCATE_LEAVES) { /* * We need to limit the size of the transaction * to avoid exhausting pagecache & tlocks */ XT_PUTPAGE(mp); newsize = (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize; goto getParent; } tlck = txLock(tid, ip, mp, tlckXTREE); tlck->type = tlckXTREE | tlckTRUNCATE; xtlck = (struct xtlock *) & tlck->lock; xtlck->hwm.offset = le16_to_cpu(p->header.nextindex) - 1; } BT_MARK_DIRTY(mp, ip); /* * scan backward leaf page entries */ for (; index >= XTENTRYSTART; index--) { xad = &p->xad[index]; xoff = offsetXAD(xad); xlen = lengthXAD(xad); xaddr = addressXAD(xad); /* * The "data" for a directory is indexed by the block * device's address space. This metadata must be invalidated * here */ if (S_ISDIR(ip->i_mode) && (teof == 0)) invalidate_xad_metapages(ip, *xad); /* * entry beyond eof: continue scan of current page * xad * ---|---=======-------> * eof */ if (teof < xoff) { nfreed += xlen; continue; } /* * (xoff <= teof): last entry to be deleted from page; * If other entries remain in page: keep and update the page. */ /* * eof == entry_start: delete the entry * xad * -------|=======-------> * eof * */ if (teof == xoff) { nfreed += xlen; if (index == XTENTRYSTART) break; nextindex = index; } /* * eof within the entry: truncate the entry. * xad * -------===|===-------> * eof */ else if (teof < xoff + xlen) { /* update truncated entry */ len = teof - xoff; freexlen = xlen - len; XADlength(xad, len); /* save pxd of truncated extent in tlck */ xaddr += len; if (log) { /* COMMIT_PWMAP */ xtlck->lwm.offset = (xtlck->lwm.offset) ? min(index, (int)xtlck->lwm.offset) : index; xtlck->lwm.length = index + 1 - xtlck->lwm.offset; xtlck->twm.offset = index; pxdlock = (struct pxd_lock *) & xtlck->pxdlock; pxdlock->flag = mlckFREEPXD; PXDaddress(&pxdlock->pxd, xaddr); PXDlength(&pxdlock->pxd, freexlen); } /* free truncated extent */ else { /* COMMIT_WMAP */ pxdlock = (struct pxd_lock *) & xadlock; pxdlock->flag = mlckFREEPXD; PXDaddress(&pxdlock->pxd, xaddr); PXDlength(&pxdlock->pxd, freexlen); txFreeMap(ip, pxdlock, NULL, COMMIT_WMAP); /* reset map lock */ xadlock.flag = mlckFREEXADLIST; } /* current entry is new last entry; */ nextindex = index + 1; nfreed += freexlen; } /* * eof beyond the entry: * xad * -------=======---|---> * eof */ else { /* (xoff + xlen < teof) */ nextindex = index + 1; } if (nextindex < le16_to_cpu(p->header.nextindex)) { if (!log) { /* COMMIT_WAMP */ xadlock.xdlist = &p->xad[nextindex]; xadlock.count = le16_to_cpu(p->header.nextindex) - nextindex; txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP); } p->header.nextindex = cpu_to_le16(nextindex); } XT_PUTPAGE(mp); /* assert(freed == 0); */ goto getParent; } /* end scan of leaf page entries */ freed = 1; /* * leaf page become empty: free the page if type != PMAP */ if (log) { /* COMMIT_PWMAP */ /* txCommit() with tlckFREE: * free data extents covered by leaf [XTENTRYSTART:hwm); * invalidate leaf if COMMIT_PWMAP; * if (TRUNCATE), will write LOG_NOREDOPAGE; */ tlck->type = tlckXTREE | tlckFREE; } else { /* COMMIT_WAMP */ /* free data extents covered by leaf */ xadlock.xdlist = &p->xad[XTENTRYSTART]; xadlock.count = le16_to_cpu(p->header.nextindex) - XTENTRYSTART; txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP); } if (p->header.flag & BT_ROOT) { p->header.flag &= ~BT_INTERNAL; p->header.flag |= BT_LEAF; p->header.nextindex = cpu_to_le16(XTENTRYSTART); XT_PUTPAGE(mp); /* debug */ goto out; } else { if (log) { /* COMMIT_PWMAP */ /* page will be invalidated at tx completion */ XT_PUTPAGE(mp); } else { /* COMMIT_WMAP */ if (mp->lid) lid_to_tlock(mp->lid)->flag |= tlckFREELOCK; /* invalidate empty leaf page */ discard_metapage(mp); } } /* * the leaf page become empty: delete the parent entry * for the leaf page if the parent page is to be kept * in the new sized file. */ /* * go back up to the parent page */ getParent: /* pop/restore parent entry for the current child page */ if ((parent = BT_POP(&btstack)) == NULL) /* current page must have been root */ goto out; /* get back the parent page */ bn = parent->bn; XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; index = parent->index; /* * child page was not empty: */ if (freed == 0) { /* has any entry deleted from parent ? */ if (index < le16_to_cpu(p->header.nextindex) - 1) { /* (re)acquire tlock on the parent page */ if (log) { /* COMMIT_PWMAP */ /* txCommit() with tlckTRUNCATE: * free child extents covered by parent [); */ tlck = txLock(tid, ip, mp, tlckXTREE); xtlck = (struct xtlock *) & tlck->lock; if (!(tlck->type & tlckTRUNCATE)) { xtlck->hwm.offset = le16_to_cpu(p->header. nextindex) - 1; tlck->type = tlckXTREE | tlckTRUNCATE; } } else { /* COMMIT_WMAP */ /* free child extents covered by parent */ xadlock.xdlist = &p->xad[index + 1]; xadlock.count = le16_to_cpu(p->header.nextindex) - index - 1; txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP); } BT_MARK_DIRTY(mp, ip); p->header.nextindex = cpu_to_le16(index + 1); } XT_PUTPAGE(mp); goto getParent; } /* * child page was empty: */ nfreed += lengthXAD(&p->xad[index]); /* * During working map update, child page's tlock must be handled * before parent's. This is because the parent's tlock will cause * the child's disk space to be marked available in the wmap, so * it's important that the child page be released by that time. * * ToDo: tlocks should be on doubly-linked list, so we can * quickly remove it and add it to the end. */ /* * Move parent page's tlock to the end of the tid's tlock list */ if (log && mp->lid && (tblk->last != mp->lid) && lid_to_tlock(mp->lid)->tid) { lid_t lid = mp->lid; struct tlock *prev; tlck = lid_to_tlock(lid); if (tblk->next == lid) tblk->next = tlck->next; else { for (prev = lid_to_tlock(tblk->next); prev->next != lid; prev = lid_to_tlock(prev->next)) { assert(prev->next); } prev->next = tlck->next; } lid_to_tlock(tblk->last)->next = lid; tlck->next = 0; tblk->last = lid; } /* * parent page become empty: free the page */ if (index == XTENTRYSTART) { if (log) { /* COMMIT_PWMAP */ /* txCommit() with tlckFREE: * free child extents covered by parent; * invalidate parent if COMMIT_PWMAP; */ tlck = txLock(tid, ip, mp, tlckXTREE); xtlck = (struct xtlock *) & tlck->lock; xtlck->hwm.offset = le16_to_cpu(p->header.nextindex) - 1; tlck->type = tlckXTREE | tlckFREE; } else { /* COMMIT_WMAP */ /* free child extents covered by parent */ xadlock.xdlist = &p->xad[XTENTRYSTART]; xadlock.count = le16_to_cpu(p->header.nextindex) - XTENTRYSTART; txFreeMap(ip, (struct maplock *) & xadlock, NULL, COMMIT_WMAP); } BT_MARK_DIRTY(mp, ip); if (p->header.flag & BT_ROOT) { p->header.flag &= ~BT_INTERNAL; p->header.flag |= BT_LEAF; p->header.nextindex = cpu_to_le16(XTENTRYSTART); if (le16_to_cpu(p->header.maxentry) == XTROOTMAXSLOT) { /* * Shrink root down to allow inline * EA (otherwise fsck complains) */ p->header.maxentry = cpu_to_le16(XTROOTINITSLOT); JFS_IP(ip)->mode2 |= INLINEEA; } XT_PUTPAGE(mp); /* debug */ goto out; } else { if (log) { /* COMMIT_PWMAP */ /* page will be invalidated at tx completion */ XT_PUTPAGE(mp); } else { /* COMMIT_WMAP */ if (mp->lid) lid_to_tlock(mp->lid)->flag |= tlckFREELOCK; /* invalidate parent page */ discard_metapage(mp); } /* parent has become empty and freed: * go back up to its parent page */ /* freed = 1; */ goto getParent; } } /* * parent page still has entries for front region; */ else { /* try truncate region covered by preceding entry * (process backward) */ index--; /* go back down to the child page corresponding * to the entry */ goto getChild; } /* * internal page: go down to child page of current entry */ getChild: /* save current parent entry for the child page */ if (BT_STACK_FULL(&btstack)) { jfs_error(ip->i_sb, "stack overrun in xtTruncate!"); XT_PUTPAGE(mp); return -EIO; } BT_PUSH(&btstack, bn, index); /* get child page */ xad = &p->xad[index]; bn = addressXAD(xad); /* * first access of each internal entry: */ /* release parent page */ XT_PUTPAGE(mp); /* process the child page */ goto getPage; out: /* * update file resource stat */ /* set size */ if (S_ISDIR(ip->i_mode) && !newsize) ip->i_size = 1; /* fsck hates zero-length directories */ else ip->i_size = newsize; /* update quota allocation to reflect freed blocks */ DQUOT_FREE_BLOCK(ip, nfreed); /* * free tlock of invalidated pages */ if (flag == COMMIT_WMAP) txFreelock(ip); return newsize; } /* * xtTruncate_pmap() * * function: * Perform truncate to zero lenghth for deleted file, leaving the * the xtree and working map untouched. This allows the file to * be accessed via open file handles, while the delete of the file * is committed to disk. * * parameter: * tid_t tid, * struct inode *ip, * s64 committed_size) * * return: new committed size * * note: * * To avoid deadlock by holding too many transaction locks, the * truncation may be broken up into multiple transactions. * The committed_size keeps track of part of the file has been * freed from the pmaps. */ s64 xtTruncate_pmap(tid_t tid, struct inode *ip, s64 committed_size) { s64 bn; struct btstack btstack; int cmp; int index; int locked_leaves = 0; struct metapage *mp; xtpage_t *p; struct btframe *parent; int rc; struct tblock *tblk; struct tlock *tlck = NULL; xad_t *xad; int xlen; s64 xoff; struct xtlock *xtlck = NULL; /* save object truncation type */ tblk = tid_to_tblock(tid); tblk->xflag |= COMMIT_PMAP; /* clear stack */ BT_CLR(&btstack); if (committed_size) { xoff = (committed_size >> JFS_SBI(ip->i_sb)->l2bsize) - 1; rc = xtSearch(ip, xoff, NULL, &cmp, &btstack, 0); if (rc) return rc; XT_GETSEARCH(ip, btstack.top, bn, mp, p, index); if (cmp != 0) { XT_PUTPAGE(mp); jfs_error(ip->i_sb, "xtTruncate_pmap: did not find extent"); return -EIO; } } else { /* * start with root * * root resides in the inode */ bn = 0; /* * first access of each page: */ getPage: XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; /* process entries backward from last index */ index = le16_to_cpu(p->header.nextindex) - 1; if (p->header.flag & BT_INTERNAL) goto getChild; } /* * leaf page */ if (++locked_leaves > MAX_TRUNCATE_LEAVES) { /* * We need to limit the size of the transaction * to avoid exhausting pagecache & tlocks */ xad = &p->xad[index]; xoff = offsetXAD(xad); xlen = lengthXAD(xad); XT_PUTPAGE(mp); return (xoff + xlen) << JFS_SBI(ip->i_sb)->l2bsize; } tlck = txLock(tid, ip, mp, tlckXTREE); tlck->type = tlckXTREE | tlckFREE; xtlck = (struct xtlock *) & tlck->lock; xtlck->hwm.offset = index; XT_PUTPAGE(mp); /* * go back up to the parent page */ getParent: /* pop/restore parent entry for the current child page */ if ((parent = BT_POP(&btstack)) == NULL) /* current page must have been root */ goto out; /* get back the parent page */ bn = parent->bn; XT_GETPAGE(ip, bn, mp, PSIZE, p, rc); if (rc) return rc; index = parent->index; /* * parent page become empty: free the page */ if (index == XTENTRYSTART) { /* txCommit() with tlckFREE: * free child extents covered by parent; * invalidate parent if COMMIT_PWMAP; */ tlck = txLock(tid, ip, mp, tlckXTREE); xtlck = (struct xtlock *) & tlck->lock; xtlck->hwm.offset = le16_to_cpu(p->header.nextindex) - 1; tlck->type = tlckXTREE | tlckFREE; XT_PUTPAGE(mp); if (p->header.flag & BT_ROOT) { goto out; } else { goto getParent; } } /* * parent page still has entries for front region; */ else index--; /* * internal page: go down to child page of current entry */ getChild: /* save current parent entry for the child page */ if (BT_STACK_FULL(&btstack)) { jfs_error(ip->i_sb, "stack overrun in xtTruncate_pmap!"); XT_PUTPAGE(mp); return -EIO; } BT_PUSH(&btstack, bn, index); /* get child page */ xad = &p->xad[index]; bn = addressXAD(xad); /* * first access of each internal entry: */ /* release parent page */ XT_PUTPAGE(mp); /* process the child page */ goto getPage; out: return 0; } #ifdef CONFIG_JFS_STATISTICS int jfs_xtstat_read(char *buffer, char **start, off_t offset, int length, int *eof, void *data) { int len = 0; off_t begin; len += sprintf(buffer, "JFS Xtree statistics\n" "====================\n" "searches = %d\n" "fast searches = %d\n" "splits = %d\n", xtStat.search, xtStat.fastSearch, xtStat.split); begin = offset; *start = buffer + begin; len -= begin; if (len > length) len = length; else *eof = 1; if (len < 0) len = 0; return len; } #endif