gc.c 42.9 KB
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378
/*
 * JFFS2 -- Journalling Flash File System, Version 2.
 *
 * Copyright © 2001-2007 Red Hat, Inc.
 * Copyright © 2004-2010 David Woodhouse <dwmw2@infradead.org>
 *
 * Created by David Woodhouse <dwmw2@infradead.org>
 *
 * For licensing information, see the file 'LICENCE' in this directory.
 *
 */

#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

#include <linux/kernel.h>
#include <linux/mtd/mtd.h>
#include <linux/slab.h>
#include <linux/pagemap.h>
#include <linux/crc32.h>
#include <linux/compiler.h>
#include <linux/stat.h>
#include "nodelist.h"
#include "compr.h"

static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
					  struct jffs2_inode_cache *ic,
					  struct jffs2_raw_node_ref *raw);
static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
					struct jffs2_inode_info *f, struct jffs2_full_dnode *fd);
static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
					struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
					struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
				      struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
				      uint32_t start, uint32_t end);
static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
				       struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
				       uint32_t start, uint32_t end);
static int jffs2_garbage_collect_live(struct jffs2_sb_info *c,  struct jffs2_eraseblock *jeb,
			       struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f);

/* Called with erase_completion_lock held */
static struct jffs2_eraseblock *jffs2_find_gc_block(struct jffs2_sb_info *c)
{
	struct jffs2_eraseblock *ret;
	struct list_head *nextlist = NULL;
	int n = jiffies % 128;

	/* Pick an eraseblock to garbage collect next. This is where we'll
	   put the clever wear-levelling algorithms. Eventually.  */
	/* We possibly want to favour the dirtier blocks more when the
	   number of free blocks is low. */
again:
	if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) {
		jffs2_dbg(1, "Picking block from bad_used_list to GC next\n");
		nextlist = &c->bad_used_list;
	} else if (n < 50 && !list_empty(&c->erasable_list)) {
		/* Note that most of them will have gone directly to be erased.
		   So don't favour the erasable_list _too_ much. */
		jffs2_dbg(1, "Picking block from erasable_list to GC next\n");
		nextlist = &c->erasable_list;
	} else if (n < 110 && !list_empty(&c->very_dirty_list)) {
		/* Most of the time, pick one off the very_dirty list */
		jffs2_dbg(1, "Picking block from very_dirty_list to GC next\n");
		nextlist = &c->very_dirty_list;
	} else if (n < 126 && !list_empty(&c->dirty_list)) {
		jffs2_dbg(1, "Picking block from dirty_list to GC next\n");
		nextlist = &c->dirty_list;
	} else if (!list_empty(&c->clean_list)) {
		jffs2_dbg(1, "Picking block from clean_list to GC next\n");
		nextlist = &c->clean_list;
	} else if (!list_empty(&c->dirty_list)) {
		jffs2_dbg(1, "Picking block from dirty_list to GC next (clean_list was empty)\n");

		nextlist = &c->dirty_list;
	} else if (!list_empty(&c->very_dirty_list)) {
		jffs2_dbg(1, "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n");
		nextlist = &c->very_dirty_list;
	} else if (!list_empty(&c->erasable_list)) {
		jffs2_dbg(1, "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n");

		nextlist = &c->erasable_list;
	} else if (!list_empty(&c->erasable_pending_wbuf_list)) {
		/* There are blocks are wating for the wbuf sync */
		jffs2_dbg(1, "Synching wbuf in order to reuse erasable_pending_wbuf_list blocks\n");
		spin_unlock(&c->erase_completion_lock);
		jffs2_flush_wbuf_pad(c);
		spin_lock(&c->erase_completion_lock);
		goto again;
	} else {
		/* Eep. All were empty */
		jffs2_dbg(1, "No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n");
		return NULL;
	}

	ret = list_entry(nextlist->next, struct jffs2_eraseblock, list);
	list_del(&ret->list);
	c->gcblock = ret;
	ret->gc_node = ret->first_node;
	if (!ret->gc_node) {
		pr_warn("Eep. ret->gc_node for block at 0x%08x is NULL\n",
			ret->offset);
		BUG();
	}

	/* Have we accidentally picked a clean block with wasted space ? */
	if (ret->wasted_size) {
		jffs2_dbg(1, "Converting wasted_size %08x to dirty_size\n",
			  ret->wasted_size);
		ret->dirty_size += ret->wasted_size;
		c->wasted_size -= ret->wasted_size;
		c->dirty_size += ret->wasted_size;
		ret->wasted_size = 0;
	}

	return ret;
}

/* jffs2_garbage_collect_pass
 * Make a single attempt to progress GC. Move one node, and possibly
 * start erasing one eraseblock.
 */
int jffs2_garbage_collect_pass(struct jffs2_sb_info *c)
{
	struct jffs2_inode_info *f;
	struct jffs2_inode_cache *ic;
	struct jffs2_eraseblock *jeb;
	struct jffs2_raw_node_ref *raw;
	uint32_t gcblock_dirty;
	int ret = 0, inum, nlink;
	int xattr = 0;

	if (mutex_lock_interruptible(&c->alloc_sem))
		return -EINTR;

	for (;;) {
		spin_lock(&c->erase_completion_lock);
		if (!c->unchecked_size)
			break;

		/* We can't start doing GC yet. We haven't finished checking
		   the node CRCs etc. Do it now. */

		/* checked_ino is protected by the alloc_sem */
		if (c->checked_ino > c->highest_ino && xattr) {
			pr_crit("Checked all inodes but still 0x%x bytes of unchecked space?\n",
				c->unchecked_size);
			jffs2_dbg_dump_block_lists_nolock(c);
			spin_unlock(&c->erase_completion_lock);
			mutex_unlock(&c->alloc_sem);
			return -ENOSPC;
		}

		spin_unlock(&c->erase_completion_lock);

		if (!xattr)
			xattr = jffs2_verify_xattr(c);

		spin_lock(&c->inocache_lock);

		ic = jffs2_get_ino_cache(c, c->checked_ino++);

		if (!ic) {
			spin_unlock(&c->inocache_lock);
			continue;
		}

		if (!ic->pino_nlink) {
			jffs2_dbg(1, "Skipping check of ino #%d with nlink/pino zero\n",
				  ic->ino);
			spin_unlock(&c->inocache_lock);
			jffs2_xattr_delete_inode(c, ic);
			continue;
		}
		switch(ic->state) {
		case INO_STATE_CHECKEDABSENT:
		case INO_STATE_PRESENT:
			jffs2_dbg(1, "Skipping ino #%u already checked\n",
				  ic->ino);
			spin_unlock(&c->inocache_lock);
			continue;

		case INO_STATE_GC:
		case INO_STATE_CHECKING:
			pr_warn("Inode #%u is in state %d during CRC check phase!\n",
				ic->ino, ic->state);
			spin_unlock(&c->inocache_lock);
			BUG();

		case INO_STATE_READING:
			/* We need to wait for it to finish, lest we move on
			   and trigger the BUG() above while we haven't yet
			   finished checking all its nodes */
			jffs2_dbg(1, "Waiting for ino #%u to finish reading\n",
				  ic->ino);
			/* We need to come back again for the _same_ inode. We've
			 made no progress in this case, but that should be OK */
			c->checked_ino--;

			mutex_unlock(&c->alloc_sem);
			sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
			return 0;

		default:
			BUG();

		case INO_STATE_UNCHECKED:
			;
		}
		ic->state = INO_STATE_CHECKING;
		spin_unlock(&c->inocache_lock);

		jffs2_dbg(1, "%s(): triggering inode scan of ino#%u\n",
			  __func__, ic->ino);

		ret = jffs2_do_crccheck_inode(c, ic);
		if (ret)
			pr_warn("Returned error for crccheck of ino #%u. Expect badness...\n",
				ic->ino);

		jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT);
		mutex_unlock(&c->alloc_sem);
		return ret;
	}

	/* If there are any blocks which need erasing, erase them now */
	if (!list_empty(&c->erase_complete_list) ||
	    !list_empty(&c->erase_pending_list)) {
		spin_unlock(&c->erase_completion_lock);
		mutex_unlock(&c->alloc_sem);
		jffs2_dbg(1, "%s(): erasing pending blocks\n", __func__);
		if (jffs2_erase_pending_blocks(c, 1))
			return 0;

		jffs2_dbg(1, "No progress from erasing block; doing GC anyway\n");
		mutex_lock(&c->alloc_sem);
		spin_lock(&c->erase_completion_lock);
	}

	/* First, work out which block we're garbage-collecting */
	jeb = c->gcblock;

	if (!jeb)
		jeb = jffs2_find_gc_block(c);

	if (!jeb) {
		/* Couldn't find a free block. But maybe we can just erase one and make 'progress'? */
		if (c->nr_erasing_blocks) {
			spin_unlock(&c->erase_completion_lock);
			mutex_unlock(&c->alloc_sem);
			return -EAGAIN;
		}
		jffs2_dbg(1, "Couldn't find erase block to garbage collect!\n");
		spin_unlock(&c->erase_completion_lock);
		mutex_unlock(&c->alloc_sem);
		return -EIO;
	}

	jffs2_dbg(1, "GC from block %08x, used_size %08x, dirty_size %08x, free_size %08x\n",
		  jeb->offset, jeb->used_size, jeb->dirty_size, jeb->free_size);
	D1(if (c->nextblock)
	   printk(KERN_DEBUG "Nextblock at  %08x, used_size %08x, dirty_size %08x, wasted_size %08x, free_size %08x\n", c->nextblock->offset, c->nextblock->used_size, c->nextblock->dirty_size, c->nextblock->wasted_size, c->nextblock->free_size));

	if (!jeb->used_size) {
		mutex_unlock(&c->alloc_sem);
		goto eraseit;
	}

	raw = jeb->gc_node;
	gcblock_dirty = jeb->dirty_size;

	while(ref_obsolete(raw)) {
		jffs2_dbg(1, "Node at 0x%08x is obsolete... skipping\n",
			  ref_offset(raw));
		raw = ref_next(raw);
		if (unlikely(!raw)) {
			pr_warn("eep. End of raw list while still supposedly nodes to GC\n");
			pr_warn("erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n",
				jeb->offset, jeb->free_size,
				jeb->dirty_size, jeb->used_size);
			jeb->gc_node = raw;
			spin_unlock(&c->erase_completion_lock);
			mutex_unlock(&c->alloc_sem);
			BUG();
		}
	}
	jeb->gc_node = raw;

	jffs2_dbg(1, "Going to garbage collect node at 0x%08x\n",
		  ref_offset(raw));

	if (!raw->next_in_ino) {
		/* Inode-less node. Clean marker, snapshot or something like that */
		spin_unlock(&c->erase_completion_lock);
		if (ref_flags(raw) == REF_PRISTINE) {
			/* It's an unknown node with JFFS2_FEATURE_RWCOMPAT_COPY */
			jffs2_garbage_collect_pristine(c, NULL, raw);
		} else {
			/* Just mark it obsolete */
			jffs2_mark_node_obsolete(c, raw);
		}
		mutex_unlock(&c->alloc_sem);
		goto eraseit_lock;
	}

	ic = jffs2_raw_ref_to_ic(raw);

#ifdef CONFIG_JFFS2_FS_XATTR
	/* When 'ic' refers xattr_datum/xattr_ref, this node is GCed as xattr.
	 * We can decide whether this node is inode or xattr by ic->class.     */
	if (ic->class == RAWNODE_CLASS_XATTR_DATUM
	    || ic->class == RAWNODE_CLASS_XATTR_REF) {
		spin_unlock(&c->erase_completion_lock);

		if (ic->class == RAWNODE_CLASS_XATTR_DATUM) {
			ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic, raw);
		} else {
			ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic, raw);
		}
		goto test_gcnode;
	}
#endif

	/* We need to hold the inocache. Either the erase_completion_lock or
	   the inocache_lock are sufficient; we trade down since the inocache_lock
	   causes less contention. */
	spin_lock(&c->inocache_lock);

	spin_unlock(&c->erase_completion_lock);

	jffs2_dbg(1, "%s(): collecting from block @0x%08x. Node @0x%08x(%d), ino #%u\n",
		  __func__, jeb->offset, ref_offset(raw), ref_flags(raw),
		  ic->ino);

	/* Three possibilities:
	   1. Inode is already in-core. We must iget it and do proper
	      updating to its fragtree, etc.
	   2. Inode is not in-core, node is REF_PRISTINE. We lock the
	      inocache to prevent a read_inode(), copy the node intact.
	   3. Inode is not in-core, node is not pristine. We must iget()
	      and take the slow path.
	*/

	switch(ic->state) {
	case INO_STATE_CHECKEDABSENT:
		/* It's been checked, but it's not currently in-core.
		   We can just copy any pristine nodes, but have
		   to prevent anyone else from doing read_inode() while
		   we're at it, so we set the state accordingly */
		if (ref_flags(raw) == REF_PRISTINE)
			ic->state = INO_STATE_GC;
		else {
			jffs2_dbg(1, "Ino #%u is absent but node not REF_PRISTINE. Reading.\n",
				  ic->ino);
		}
		break;

	case INO_STATE_PRESENT:
		/* It's in-core. GC must iget() it. */
		break;

	case INO_STATE_UNCHECKED:
	case INO_STATE_CHECKING:
	case INO_STATE_GC:
		/* Should never happen. We should have finished checking
		   by the time we actually start doing any GC, and since
		   we're holding the alloc_sem, no other garbage collection
		   can happen.
		*/
		pr_crit("Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n",
			ic->ino, ic->state);
		mutex_unlock(&c->alloc_sem);
		spin_unlock(&c->inocache_lock);
		BUG();

	case INO_STATE_READING:
		/* Someone's currently trying to read it. We must wait for
		   them to finish and then go through the full iget() route
		   to do the GC. However, sometimes read_inode() needs to get
		   the alloc_sem() (for marking nodes invalid) so we must
		   drop the alloc_sem before sleeping. */

		mutex_unlock(&c->alloc_sem);
		jffs2_dbg(1, "%s(): waiting for ino #%u in state %d\n",
			  __func__, ic->ino, ic->state);
		sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
		/* And because we dropped the alloc_sem we must start again from the
		   beginning. Ponder chance of livelock here -- we're returning success
		   without actually making any progress.

		   Q: What are the chances that the inode is back in INO_STATE_READING
		   again by the time we next enter this function? And that this happens
		   enough times to cause a real delay?

		   A: Small enough that I don't care :)
		*/
		return 0;
	}

	/* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the
	   node intact, and we don't have to muck about with the fragtree etc.
	   because we know it's not in-core. If it _was_ in-core, we go through
	   all the iget() crap anyway */

	if (ic->state == INO_STATE_GC) {
		spin_unlock(&c->inocache_lock);

		ret = jffs2_garbage_collect_pristine(c, ic, raw);

		spin_lock(&c->inocache_lock);
		ic->state = INO_STATE_CHECKEDABSENT;
		wake_up(&c->inocache_wq);

		if (ret != -EBADFD) {
			spin_unlock(&c->inocache_lock);
			goto test_gcnode;
		}

		/* Fall through if it wanted us to, with inocache_lock held */
	}

	/* Prevent the fairly unlikely race where the gcblock is
	   entirely obsoleted by the final close of a file which had
	   the only valid nodes in the block, followed by erasure,
	   followed by freeing of the ic because the erased block(s)
	   held _all_ the nodes of that inode.... never been seen but
	   it's vaguely possible. */

	inum = ic->ino;
	nlink = ic->pino_nlink;
	spin_unlock(&c->inocache_lock);

	f = jffs2_gc_fetch_inode(c, inum, !nlink);
	if (IS_ERR(f)) {
		ret = PTR_ERR(f);
		goto release_sem;
	}
	if (!f) {
		ret = 0;
		goto release_sem;
	}

	ret = jffs2_garbage_collect_live(c, jeb, raw, f);

	jffs2_gc_release_inode(c, f);

 test_gcnode:
	if (jeb->dirty_size == gcblock_dirty && !ref_obsolete(jeb->gc_node)) {
		/* Eep. This really should never happen. GC is broken */
		pr_err("Error garbage collecting node at %08x!\n",
		       ref_offset(jeb->gc_node));
		ret = -ENOSPC;
	}
 release_sem:
	mutex_unlock(&c->alloc_sem);

 eraseit_lock:
	/* If we've finished this block, start it erasing */
	spin_lock(&c->erase_completion_lock);

 eraseit:
	if (c->gcblock && !c->gcblock->used_size) {
		jffs2_dbg(1, "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n",
			  c->gcblock->offset);
		/* We're GC'ing an empty block? */
		list_add_tail(&c->gcblock->list, &c->erase_pending_list);
		c->gcblock = NULL;
		c->nr_erasing_blocks++;
		jffs2_garbage_collect_trigger(c);
	}
	spin_unlock(&c->erase_completion_lock);

	return ret;
}

static int jffs2_garbage_collect_live(struct jffs2_sb_info *c,  struct jffs2_eraseblock *jeb,
				      struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f)
{
	struct jffs2_node_frag *frag;
	struct jffs2_full_dnode *fn = NULL;
	struct jffs2_full_dirent *fd;
	uint32_t start = 0, end = 0, nrfrags = 0;
	int ret = 0;

	mutex_lock(&f->sem);

	/* Now we have the lock for this inode. Check that it's still the one at the head
	   of the list. */

	spin_lock(&c->erase_completion_lock);

	if (c->gcblock != jeb) {
		spin_unlock(&c->erase_completion_lock);
		jffs2_dbg(1, "GC block is no longer gcblock. Restart\n");
		goto upnout;
	}
	if (ref_obsolete(raw)) {
		spin_unlock(&c->erase_completion_lock);
		jffs2_dbg(1, "node to be GC'd was obsoleted in the meantime.\n");
		/* They'll call again */
		goto upnout;
	}
	spin_unlock(&c->erase_completion_lock);

	/* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */
	if (f->metadata && f->metadata->raw == raw) {
		fn = f->metadata;
		ret = jffs2_garbage_collect_metadata(c, jeb, f, fn);
		goto upnout;
	}

	/* FIXME. Read node and do lookup? */
	for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) {
		if (frag->node && frag->node->raw == raw) {
			fn = frag->node;
			end = frag->ofs + frag->size;
			if (!nrfrags++)
				start = frag->ofs;
			if (nrfrags == frag->node->frags)
				break; /* We've found them all */
		}
	}
	if (fn) {
		if (ref_flags(raw) == REF_PRISTINE) {
			ret = jffs2_garbage_collect_pristine(c, f->inocache, raw);
			if (!ret) {
				/* Urgh. Return it sensibly. */
				frag->node->raw = f->inocache->nodes;
			}
			if (ret != -EBADFD)
				goto upnout;
		}
		/* We found a datanode. Do the GC */
		if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) {
			/* It crosses a page boundary. Therefore, it must be a hole. */
			ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end);
		} else {
			/* It could still be a hole. But we GC the page this way anyway */
			ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end);
		}
		goto upnout;
	}

	/* Wasn't a dnode. Try dirent */
	for (fd = f->dents; fd; fd=fd->next) {
		if (fd->raw == raw)
			break;
	}

	if (fd && fd->ino) {
		ret = jffs2_garbage_collect_dirent(c, jeb, f, fd);
	} else if (fd) {
		ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd);
	} else {
		pr_warn("Raw node at 0x%08x wasn't in node lists for ino #%u\n",
			ref_offset(raw), f->inocache->ino);
		if (ref_obsolete(raw)) {
			pr_warn("But it's obsolete so we don't mind too much\n");
		} else {
			jffs2_dbg_dump_node(c, ref_offset(raw));
			BUG();
		}
	}
 upnout:
	mutex_unlock(&f->sem);

	return ret;
}

static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
					  struct jffs2_inode_cache *ic,
					  struct jffs2_raw_node_ref *raw)
{
	union jffs2_node_union *node;
	size_t retlen;
	int ret;
	uint32_t phys_ofs, alloclen;
	uint32_t crc, rawlen;
	int retried = 0;

	jffs2_dbg(1, "Going to GC REF_PRISTINE node at 0x%08x\n",
		  ref_offset(raw));

	alloclen = rawlen = ref_totlen(c, c->gcblock, raw);

	/* Ask for a small amount of space (or the totlen if smaller) because we
	   don't want to force wastage of the end of a block if splitting would
	   work. */
	if (ic && alloclen > sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN)
		alloclen = sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN;

	ret = jffs2_reserve_space_gc(c, alloclen, &alloclen, rawlen);
	/* 'rawlen' is not the exact summary size; it is only an upper estimation */

	if (ret)
		return ret;

	if (alloclen < rawlen) {
		/* Doesn't fit untouched. We'll go the old route and split it */
		return -EBADFD;
	}

	node = kmalloc(rawlen, GFP_KERNEL);
	if (!node)
		return -ENOMEM;

	ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node);
	if (!ret && retlen != rawlen)
		ret = -EIO;
	if (ret)
		goto out_node;

	crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4);
	if (je32_to_cpu(node->u.hdr_crc) != crc) {
		pr_warn("Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
			ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc);
		goto bail;
	}

	switch(je16_to_cpu(node->u.nodetype)) {
	case JFFS2_NODETYPE_INODE:
		crc = crc32(0, node, sizeof(node->i)-8);
		if (je32_to_cpu(node->i.node_crc) != crc) {
			pr_warn("Node CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
				ref_offset(raw), je32_to_cpu(node->i.node_crc),
				crc);
			goto bail;
		}

		if (je32_to_cpu(node->i.dsize)) {
			crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize));
			if (je32_to_cpu(node->i.data_crc) != crc) {
				pr_warn("Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
					ref_offset(raw),
					je32_to_cpu(node->i.data_crc), crc);
				goto bail;
			}
		}
		break;

	case JFFS2_NODETYPE_DIRENT:
		crc = crc32(0, node, sizeof(node->d)-8);
		if (je32_to_cpu(node->d.node_crc) != crc) {
			pr_warn("Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
				ref_offset(raw),
				je32_to_cpu(node->d.node_crc), crc);
			goto bail;
		}

		if (strnlen(node->d.name, node->d.nsize) != node->d.nsize) {
			pr_warn("Name in dirent node at 0x%08x contains zeroes\n",
				ref_offset(raw));
			goto bail;
		}

		if (node->d.nsize) {
			crc = crc32(0, node->d.name, node->d.nsize);
			if (je32_to_cpu(node->d.name_crc) != crc) {
				pr_warn("Name CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
					ref_offset(raw),
					je32_to_cpu(node->d.name_crc), crc);
				goto bail;
			}
		}
		break;
	default:
		/* If it's inode-less, we don't _know_ what it is. Just copy it intact */
		if (ic) {
			pr_warn("Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n",
				ref_offset(raw), je16_to_cpu(node->u.nodetype));
			goto bail;
		}
	}

	/* OK, all the CRCs are good; this node can just be copied as-is. */
 retry:
	phys_ofs = write_ofs(c);

	ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node);

	if (ret || (retlen != rawlen)) {
		pr_notice("Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n",
			  rawlen, phys_ofs, ret, retlen);
		if (retlen) {
			jffs2_add_physical_node_ref(c, phys_ofs | REF_OBSOLETE, rawlen, NULL);
		} else {
			pr_notice("Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n",
				  phys_ofs);
		}
		if (!retried) {
			/* Try to reallocate space and retry */
			uint32_t dummy;
			struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size];

			retried = 1;

			jffs2_dbg(1, "Retrying failed write of REF_PRISTINE node.\n");

			jffs2_dbg_acct_sanity_check(c,jeb);
			jffs2_dbg_acct_paranoia_check(c, jeb);

			ret = jffs2_reserve_space_gc(c, rawlen, &dummy, rawlen);
						/* this is not the exact summary size of it,
							it is only an upper estimation */

			if (!ret) {
				jffs2_dbg(1, "Allocated space at 0x%08x to retry failed write.\n",
					  phys_ofs);

				jffs2_dbg_acct_sanity_check(c,jeb);
				jffs2_dbg_acct_paranoia_check(c, jeb);

				goto retry;
			}
			jffs2_dbg(1, "Failed to allocate space to retry failed write: %d!\n",
				  ret);
		}

		if (!ret)
			ret = -EIO;
		goto out_node;
	}
	jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, rawlen, ic);

	jffs2_mark_node_obsolete(c, raw);
	jffs2_dbg(1, "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n",
		  ref_offset(raw));

 out_node:
	kfree(node);
	return ret;
 bail:
	ret = -EBADFD;
	goto out_node;
}

static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
					struct jffs2_inode_info *f, struct jffs2_full_dnode *fn)
{
	struct jffs2_full_dnode *new_fn;
	struct jffs2_raw_inode ri;
	struct jffs2_node_frag *last_frag;
	union jffs2_device_node dev;
	char *mdata = NULL;
	int mdatalen = 0;
	uint32_t alloclen, ilen;
	int ret;

	if (S_ISBLK(JFFS2_F_I_MODE(f)) ||
	    S_ISCHR(JFFS2_F_I_MODE(f)) ) {
		/* For these, we don't actually need to read the old node */
		mdatalen = jffs2_encode_dev(&dev, JFFS2_F_I_RDEV(f));
		mdata = (char *)&dev;
		jffs2_dbg(1, "%s(): Writing %d bytes of kdev_t\n",
			  __func__, mdatalen);
	} else if (S_ISLNK(JFFS2_F_I_MODE(f))) {
		mdatalen = fn->size;
		mdata = kmalloc(fn->size, GFP_KERNEL);
		if (!mdata) {
			pr_warn("kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n");
			return -ENOMEM;
		}
		ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen);
		if (ret) {
			pr_warn("read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n",
				ret);
			kfree(mdata);
			return ret;
		}
		jffs2_dbg(1, "%s(): Writing %d bites of symlink target\n",
			  __func__, mdatalen);

	}

	ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &alloclen,
				JFFS2_SUMMARY_INODE_SIZE);
	if (ret) {
		pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n",
			sizeof(ri) + mdatalen, ret);
		goto out;
	}

	last_frag = frag_last(&f->fragtree);
	if (last_frag)
		/* Fetch the inode length from the fragtree rather then
		 * from i_size since i_size may have not been updated yet */
		ilen = last_frag->ofs + last_frag->size;
	else
		ilen = JFFS2_F_I_SIZE(f);

	memset(&ri, 0, sizeof(ri));
	ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
	ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
	ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen);
	ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));

	ri.ino = cpu_to_je32(f->inocache->ino);
	ri.version = cpu_to_je32(++f->highest_version);
	ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
	ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
	ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
	ri.isize = cpu_to_je32(ilen);
	ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
	ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
	ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
	ri.offset = cpu_to_je32(0);
	ri.csize = cpu_to_je32(mdatalen);
	ri.dsize = cpu_to_je32(mdatalen);
	ri.compr = JFFS2_COMPR_NONE;
	ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
	ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen));

	new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, ALLOC_GC);

	if (IS_ERR(new_fn)) {
		pr_warn("Error writing new dnode: %ld\n", PTR_ERR(new_fn));
		ret = PTR_ERR(new_fn);
		goto out;
	}
	jffs2_mark_node_obsolete(c, fn->raw);
	jffs2_free_full_dnode(fn);
	f->metadata = new_fn;
 out:
	if (S_ISLNK(JFFS2_F_I_MODE(f)))
		kfree(mdata);
	return ret;
}

static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
					struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
{
	struct jffs2_full_dirent *new_fd;
	struct jffs2_raw_dirent rd;
	uint32_t alloclen;
	int ret;

	rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
	rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT);
	rd.nsize = strlen(fd->name);
	rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize);
	rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4));

	rd.pino = cpu_to_je32(f->inocache->ino);
	rd.version = cpu_to_je32(++f->highest_version);
	rd.ino = cpu_to_je32(fd->ino);
	/* If the times on this inode were set by explicit utime() they can be different,
	   so refrain from splatting them. */
	if (JFFS2_F_I_MTIME(f) == JFFS2_F_I_CTIME(f))
		rd.mctime = cpu_to_je32(JFFS2_F_I_MTIME(f));
	else
		rd.mctime = cpu_to_je32(0);
	rd.type = fd->type;
	rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8));
	rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize));

	ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &alloclen,
				JFFS2_SUMMARY_DIRENT_SIZE(rd.nsize));
	if (ret) {
		pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n",
			sizeof(rd)+rd.nsize, ret);
		return ret;
	}
	new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, ALLOC_GC);

	if (IS_ERR(new_fd)) {
		pr_warn("jffs2_write_dirent in garbage_collect_dirent failed: %ld\n",
			PTR_ERR(new_fd));
		return PTR_ERR(new_fd);
	}
	jffs2_add_fd_to_list(c, new_fd, &f->dents);
	return 0;
}

static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
					struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
{
	struct jffs2_full_dirent **fdp = &f->dents;
	int found = 0;

	/* On a medium where we can't actually mark nodes obsolete
	   pernamently, such as NAND flash, we need to work out
	   whether this deletion dirent is still needed to actively
	   delete a 'real' dirent with the same name that's still
	   somewhere else on the flash. */
	if (!jffs2_can_mark_obsolete(c)) {
		struct jffs2_raw_dirent *rd;
		struct jffs2_raw_node_ref *raw;
		int ret;
		size_t retlen;
		int name_len = strlen(fd->name);
		uint32_t name_crc = crc32(0, fd->name, name_len);
		uint32_t rawlen = ref_totlen(c, jeb, fd->raw);

		rd = kmalloc(rawlen, GFP_KERNEL);
		if (!rd)
			return -ENOMEM;

		/* Prevent the erase code from nicking the obsolete node refs while
		   we're looking at them. I really don't like this extra lock but
		   can't see any alternative. Suggestions on a postcard to... */
		mutex_lock(&c->erase_free_sem);

		for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) {

			cond_resched();

			/* We only care about obsolete ones */
			if (!(ref_obsolete(raw)))
				continue;

			/* Any dirent with the same name is going to have the same length... */
			if (ref_totlen(c, NULL, raw) != rawlen)
				continue;

			/* Doesn't matter if there's one in the same erase block. We're going to
			   delete it too at the same time. */
			if (SECTOR_ADDR(raw->flash_offset) == SECTOR_ADDR(fd->raw->flash_offset))
				continue;

			jffs2_dbg(1, "Check potential deletion dirent at %08x\n",
				  ref_offset(raw));

			/* This is an obsolete node belonging to the same directory, and it's of the right
			   length. We need to take a closer look...*/
			ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd);
			if (ret) {
				pr_warn("%s(): Read error (%d) reading obsolete node at %08x\n",
					__func__, ret, ref_offset(raw));
				/* If we can't read it, we don't need to continue to obsolete it. Continue */
				continue;
			}
			if (retlen != rawlen) {
				pr_warn("%s(): Short read (%zd not %u) reading header from obsolete node at %08x\n",
					__func__, retlen, rawlen,
					ref_offset(raw));
				continue;
			}

			if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT)
				continue;

			/* If the name CRC doesn't match, skip */
			if (je32_to_cpu(rd->name_crc) != name_crc)
				continue;

			/* If the name length doesn't match, or it's another deletion dirent, skip */
			if (rd->nsize != name_len || !je32_to_cpu(rd->ino))
				continue;

			/* OK, check the actual name now */
			if (memcmp(rd->name, fd->name, name_len))
				continue;

			/* OK. The name really does match. There really is still an older node on
			   the flash which our deletion dirent obsoletes. So we have to write out
			   a new deletion dirent to replace it */
			mutex_unlock(&c->erase_free_sem);

			jffs2_dbg(1, "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n",
				  ref_offset(fd->raw), fd->name,
				  ref_offset(raw), je32_to_cpu(rd->ino));
			kfree(rd);

			return jffs2_garbage_collect_dirent(c, jeb, f, fd);
		}

		mutex_unlock(&c->erase_free_sem);
		kfree(rd);
	}

	/* FIXME: If we're deleting a dirent which contains the current mtime and ctime,
	   we should update the metadata node with those times accordingly */

	/* No need for it any more. Just mark it obsolete and remove it from the list */
	while (*fdp) {
		if ((*fdp) == fd) {
			found = 1;
			*fdp = fd->next;
			break;
		}
		fdp = &(*fdp)->next;
	}
	if (!found) {
		pr_warn("Deletion dirent \"%s\" not found in list for ino #%u\n",
			fd->name, f->inocache->ino);
	}
	jffs2_mark_node_obsolete(c, fd->raw);
	jffs2_free_full_dirent(fd);
	return 0;
}

static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
				      struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
				      uint32_t start, uint32_t end)
{
	struct jffs2_raw_inode ri;
	struct jffs2_node_frag *frag;
	struct jffs2_full_dnode *new_fn;
	uint32_t alloclen, ilen;
	int ret;

	jffs2_dbg(1, "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n",
		  f->inocache->ino, start, end);

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

	if(fn->frags > 1) {
		size_t readlen;
		uint32_t crc;
		/* It's partially obsoleted by a later write. So we have to
		   write it out again with the _same_ version as before */
		ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri);
		if (readlen != sizeof(ri) || ret) {
			pr_warn("Node read failed in jffs2_garbage_collect_hole. Ret %d, retlen %zd. Data will be lost by writing new hole node\n",
				ret, readlen);
			goto fill;
		}
		if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) {
			pr_warn("%s(): Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n",
				__func__, ref_offset(fn->raw),
				je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE);
			return -EIO;
		}
		if (je32_to_cpu(ri.totlen) != sizeof(ri)) {
			pr_warn("%s(): Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n",
				__func__, ref_offset(fn->raw),
				je32_to_cpu(ri.totlen), sizeof(ri));
			return -EIO;
		}
		crc = crc32(0, &ri, sizeof(ri)-8);
		if (crc != je32_to_cpu(ri.node_crc)) {
			pr_warn("%s: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n",
				__func__, ref_offset(fn->raw),
				je32_to_cpu(ri.node_crc), crc);
			/* FIXME: We could possibly deal with this by writing new holes for each frag */
			pr_warn("Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
				start, end, f->inocache->ino);
			goto fill;
		}
		if (ri.compr != JFFS2_COMPR_ZERO) {
			pr_warn("%s(): Node 0x%08x wasn't a hole node!\n",
				__func__, ref_offset(fn->raw));
			pr_warn("Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
				start, end, f->inocache->ino);
			goto fill;
		}
	} else {
	fill:
		ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
		ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
		ri.totlen = cpu_to_je32(sizeof(ri));
		ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));

		ri.ino = cpu_to_je32(f->inocache->ino);
		ri.version = cpu_to_je32(++f->highest_version);
		ri.offset = cpu_to_je32(start);
		ri.dsize = cpu_to_je32(end - start);
		ri.csize = cpu_to_je32(0);
		ri.compr = JFFS2_COMPR_ZERO;
	}

	frag = frag_last(&f->fragtree);
	if (frag)
		/* Fetch the inode length from the fragtree rather then
		 * from i_size since i_size may have not been updated yet */
		ilen = frag->ofs + frag->size;
	else
		ilen = JFFS2_F_I_SIZE(f);

	ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
	ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
	ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
	ri.isize = cpu_to_je32(ilen);
	ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
	ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
	ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
	ri.data_crc = cpu_to_je32(0);
	ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));

	ret = jffs2_reserve_space_gc(c, sizeof(ri), &alloclen,
				     JFFS2_SUMMARY_INODE_SIZE);
	if (ret) {
		pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n",
			sizeof(ri), ret);
		return ret;
	}
	new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, ALLOC_GC);

	if (IS_ERR(new_fn)) {
		pr_warn("Error writing new hole node: %ld\n", PTR_ERR(new_fn));
		return PTR_ERR(new_fn);
	}
	if (je32_to_cpu(ri.version) == f->highest_version) {
		jffs2_add_full_dnode_to_inode(c, f, new_fn);
		if (f->metadata) {
			jffs2_mark_node_obsolete(c, f->metadata->raw);
			jffs2_free_full_dnode(f->metadata);
			f->metadata = NULL;
		}
		return 0;
	}

	/*
	 * We should only get here in the case where the node we are
	 * replacing had more than one frag, so we kept the same version
	 * number as before. (Except in case of error -- see 'goto fill;'
	 * above.)
	 */
	D1(if(unlikely(fn->frags <= 1)) {
			pr_warn("%s(): Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n",
				__func__, fn->frags, je32_to_cpu(ri.version),
				f->highest_version, je32_to_cpu(ri.ino));
	});

	/* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */
	mark_ref_normal(new_fn->raw);

	for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs);
	     frag; frag = frag_next(frag)) {
		if (frag->ofs > fn->size + fn->ofs)
			break;
		if (frag->node == fn) {
			frag->node = new_fn;
			new_fn->frags++;
			fn->frags--;
		}
	}
	if (fn->frags) {
		pr_warn("%s(): Old node still has frags!\n", __func__);
		BUG();
	}
	if (!new_fn->frags) {
		pr_warn("%s(): New node has no frags!\n", __func__);
		BUG();
	}

	jffs2_mark_node_obsolete(c, fn->raw);
	jffs2_free_full_dnode(fn);

	return 0;
}

static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *orig_jeb,
				       struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
				       uint32_t start, uint32_t end)
{
	struct jffs2_full_dnode *new_fn;
	struct jffs2_raw_inode ri;
	uint32_t alloclen, offset, orig_end, orig_start;
	int ret = 0;
	unsigned char *comprbuf = NULL, *writebuf;
	unsigned long pg;
	unsigned char *pg_ptr;

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

	jffs2_dbg(1, "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n",
		  f->inocache->ino, start, end);

	orig_end = end;
	orig_start = start;

	if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) {
		/* Attempt to do some merging. But only expand to cover logically
		   adjacent frags if the block containing them is already considered
		   to be dirty. Otherwise we end up with GC just going round in
		   circles dirtying the nodes it already wrote out, especially
		   on NAND where we have small eraseblocks and hence a much higher
		   chance of nodes having to be split to cross boundaries. */

		struct jffs2_node_frag *frag;
		uint32_t min, max;

		min = start & ~(PAGE_CACHE_SIZE-1);
		max = min + PAGE_CACHE_SIZE;

		frag = jffs2_lookup_node_frag(&f->fragtree, start);

		/* BUG_ON(!frag) but that'll happen anyway... */

		BUG_ON(frag->ofs != start);

		/* First grow down... */
		while((frag = frag_prev(frag)) && frag->ofs >= min) {

			/* If the previous frag doesn't even reach the beginning, there's
			   excessive fragmentation. Just merge. */
			if (frag->ofs > min) {
				jffs2_dbg(1, "Expanding down to cover partial frag (0x%x-0x%x)\n",
					  frag->ofs, frag->ofs+frag->size);
				start = frag->ofs;
				continue;
			}
			/* OK. This frag holds the first byte of the page. */
			if (!frag->node || !frag->node->raw) {
				jffs2_dbg(1, "First frag in page is hole (0x%x-0x%x). Not expanding down.\n",
					  frag->ofs, frag->ofs+frag->size);
				break;
			} else {

				/* OK, it's a frag which extends to the beginning of the page. Does it live
				   in a block which is still considered clean? If so, don't obsolete it.
				   If not, cover it anyway. */

				struct jffs2_raw_node_ref *raw = frag->node->raw;
				struct jffs2_eraseblock *jeb;

				jeb = &c->blocks[raw->flash_offset / c->sector_size];

				if (jeb == c->gcblock) {
					jffs2_dbg(1, "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n",
						  frag->ofs,
						  frag->ofs + frag->size,
						  ref_offset(raw));
					start = frag->ofs;
					break;
				}
				if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
					jffs2_dbg(1, "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n",
						  frag->ofs,
						  frag->ofs + frag->size,
						  jeb->offset);
					break;
				}

				jffs2_dbg(1, "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n",
					  frag->ofs,
					  frag->ofs + frag->size,
					  jeb->offset);
				start = frag->ofs;
				break;
			}
		}

		/* ... then up */

		/* Find last frag which is actually part of the node we're to GC. */
		frag = jffs2_lookup_node_frag(&f->fragtree, end-1);

		while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) {

			/* If the previous frag doesn't even reach the beginning, there's lots
			   of fragmentation. Just merge. */
			if (frag->ofs+frag->size < max) {
				jffs2_dbg(1, "Expanding up to cover partial frag (0x%x-0x%x)\n",
					  frag->ofs, frag->ofs+frag->size);
				end = frag->ofs + frag->size;
				continue;
			}

			if (!frag->node || !frag->node->raw) {
				jffs2_dbg(1, "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n",
					  frag->ofs, frag->ofs+frag->size);
				break;
			} else {

				/* OK, it's a frag which extends to the beginning of the page. Does it live
				   in a block which is still considered clean? If so, don't obsolete it.
				   If not, cover it anyway. */

				struct jffs2_raw_node_ref *raw = frag->node->raw;
				struct jffs2_eraseblock *jeb;

				jeb = &c->blocks[raw->flash_offset / c->sector_size];

				if (jeb == c->gcblock) {
					jffs2_dbg(1, "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n",
						  frag->ofs,
						  frag->ofs + frag->size,
						  ref_offset(raw));
					end = frag->ofs + frag->size;
					break;
				}
				if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
					jffs2_dbg(1, "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n",
						  frag->ofs,
						  frag->ofs + frag->size,
						  jeb->offset);
					break;
				}

				jffs2_dbg(1, "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n",
					  frag->ofs,
					  frag->ofs + frag->size,
					  jeb->offset);
				end = frag->ofs + frag->size;
				break;
			}
		}
		jffs2_dbg(1, "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n",
			  orig_start, orig_end, start, end);

		D1(BUG_ON(end > frag_last(&f->fragtree)->ofs + frag_last(&f->fragtree)->size));
		BUG_ON(end < orig_end);
		BUG_ON(start > orig_start);
	}

	/* First, use readpage() to read the appropriate page into the page cache */
	/* Q: What happens if we actually try to GC the _same_ page for which commit_write()
	 *    triggered garbage collection in the first place?
	 * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the
	 *    page OK. We'll actually write it out again in commit_write, which is a little
	 *    suboptimal, but at least we're correct.
	 */
	pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg);

	if (IS_ERR(pg_ptr)) {
		pr_warn("read_cache_page() returned error: %ld\n",
			PTR_ERR(pg_ptr));
		return PTR_ERR(pg_ptr);
	}

	offset = start;
	while(offset < orig_end) {
		uint32_t datalen;
		uint32_t cdatalen;
		uint16_t comprtype = JFFS2_COMPR_NONE;

		ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN,
					&alloclen, JFFS2_SUMMARY_INODE_SIZE);

		if (ret) {
			pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n",
				sizeof(ri) + JFFS2_MIN_DATA_LEN, ret);
			break;
		}
		cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset);
		datalen = end - offset;

		writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1));

		comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen);

		ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
		ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
		ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen);
		ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));

		ri.ino = cpu_to_je32(f->inocache->ino);
		ri.version = cpu_to_je32(++f->highest_version);
		ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
		ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
		ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
		ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
		ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
		ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
		ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
		ri.offset = cpu_to_je32(offset);
		ri.csize = cpu_to_je32(cdatalen);
		ri.dsize = cpu_to_je32(datalen);
		ri.compr = comprtype & 0xff;
		ri.usercompr = (comprtype >> 8) & 0xff;
		ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
		ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen));

		new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, ALLOC_GC);

		jffs2_free_comprbuf(comprbuf, writebuf);

		if (IS_ERR(new_fn)) {
			pr_warn("Error writing new dnode: %ld\n",
				PTR_ERR(new_fn));
			ret = PTR_ERR(new_fn);
			break;
		}
		ret = jffs2_add_full_dnode_to_inode(c, f, new_fn);
		offset += datalen;
		if (f->metadata) {
			jffs2_mark_node_obsolete(c, f->metadata->raw);
			jffs2_free_full_dnode(f->metadata);
			f->metadata = NULL;
		}
	}

	jffs2_gc_release_page(c, pg_ptr, &pg);
	return ret;
}