gc.c 43.1 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3
/*
 * JFFS2 -- Journalling Flash File System, Version 2.
 *
4
 * Copyright © 2001-2007 Red Hat, Inc.
5
 * Copyright © 2004-2010 David Woodhouse <dwmw2@infradead.org>
Linus Torvalds's avatar
Linus Torvalds committed
6 7 8 9 10 11 12
 *
 * Created by David Woodhouse <dwmw2@infradead.org>
 *
 * For licensing information, see the file 'LICENCE' in this directory.
 *
 */

13 14
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

Linus Torvalds's avatar
Linus Torvalds committed
15 16 17 18 19 20 21 22 23 24
#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"

25
static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
Linus Torvalds's avatar
Linus Torvalds committed
26 27
					  struct jffs2_inode_cache *ic,
					  struct jffs2_raw_node_ref *raw);
28
static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds's avatar
Linus Torvalds committed
29
					struct jffs2_inode_info *f, struct jffs2_full_dnode *fd);
30
static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds's avatar
Linus Torvalds committed
31
					struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
32
static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds's avatar
Linus Torvalds committed
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53
					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. */
54
again:
Linus Torvalds's avatar
Linus Torvalds committed
55
	if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) {
56
		jffs2_dbg(1, "Picking block from bad_used_list to GC next\n");
Linus Torvalds's avatar
Linus Torvalds committed
57 58
		nextlist = &c->bad_used_list;
	} else if (n < 50 && !list_empty(&c->erasable_list)) {
59
		/* Note that most of them will have gone directly to be erased.
Linus Torvalds's avatar
Linus Torvalds committed
60
		   So don't favour the erasable_list _too_ much. */
61
		jffs2_dbg(1, "Picking block from erasable_list to GC next\n");
Linus Torvalds's avatar
Linus Torvalds committed
62 63 64
		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 */
65
		jffs2_dbg(1, "Picking block from very_dirty_list to GC next\n");
Linus Torvalds's avatar
Linus Torvalds committed
66 67
		nextlist = &c->very_dirty_list;
	} else if (n < 126 && !list_empty(&c->dirty_list)) {
68
		jffs2_dbg(1, "Picking block from dirty_list to GC next\n");
Linus Torvalds's avatar
Linus Torvalds committed
69 70
		nextlist = &c->dirty_list;
	} else if (!list_empty(&c->clean_list)) {
71
		jffs2_dbg(1, "Picking block from clean_list to GC next\n");
Linus Torvalds's avatar
Linus Torvalds committed
72 73
		nextlist = &c->clean_list;
	} else if (!list_empty(&c->dirty_list)) {
74
		jffs2_dbg(1, "Picking block from dirty_list to GC next (clean_list was empty)\n");
Linus Torvalds's avatar
Linus Torvalds committed
75 76 77

		nextlist = &c->dirty_list;
	} else if (!list_empty(&c->very_dirty_list)) {
78
		jffs2_dbg(1, "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n");
Linus Torvalds's avatar
Linus Torvalds committed
79 80
		nextlist = &c->very_dirty_list;
	} else if (!list_empty(&c->erasable_list)) {
81
		jffs2_dbg(1, "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n");
Linus Torvalds's avatar
Linus Torvalds committed
82 83

		nextlist = &c->erasable_list;
84 85
	} else if (!list_empty(&c->erasable_pending_wbuf_list)) {
		/* There are blocks are wating for the wbuf sync */
86
		jffs2_dbg(1, "Synching wbuf in order to reuse erasable_pending_wbuf_list blocks\n");
87
		spin_unlock(&c->erase_completion_lock);
88
		jffs2_flush_wbuf_pad(c);
89
		spin_lock(&c->erase_completion_lock);
90
		goto again;
Linus Torvalds's avatar
Linus Torvalds committed
91 92
	} else {
		/* Eep. All were empty */
93
		jffs2_dbg(1, "No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n");
Linus Torvalds's avatar
Linus Torvalds committed
94 95 96 97 98 99 100 101
		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) {
102 103
		pr_warn("Eep. ret->gc_node for block at 0x%08x is NULL\n",
			ret->offset);
Linus Torvalds's avatar
Linus Torvalds committed
104 105
		BUG();
	}
106

Linus Torvalds's avatar
Linus Torvalds committed
107 108
	/* Have we accidentally picked a clean block with wasted space ? */
	if (ret->wasted_size) {
109 110
		jffs2_dbg(1, "Converting wasted_size %08x to dirty_size\n",
			  ret->wasted_size);
Linus Torvalds's avatar
Linus Torvalds committed
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
		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;
130
	uint32_t gcblock_dirty;
Linus Torvalds's avatar
Linus Torvalds committed
131
	int ret = 0, inum, nlink;
132
	int xattr = 0;
Linus Torvalds's avatar
Linus Torvalds committed
133

134
	if (mutex_lock_interruptible(&c->alloc_sem))
Linus Torvalds's avatar
Linus Torvalds committed
135 136 137 138 139 140 141 142 143
		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. */
144

Linus Torvalds's avatar
Linus Torvalds committed
145
		/* checked_ino is protected by the alloc_sem */
146
		if (c->checked_ino > c->highest_ino && xattr) {
147 148
			pr_crit("Checked all inodes but still 0x%x bytes of unchecked space?\n",
				c->unchecked_size);
149
			jffs2_dbg_dump_block_lists_nolock(c);
Linus Torvalds's avatar
Linus Torvalds committed
150
			spin_unlock(&c->erase_completion_lock);
151
			mutex_unlock(&c->alloc_sem);
152
			return -ENOSPC;
Linus Torvalds's avatar
Linus Torvalds committed
153 154 155 156
		}

		spin_unlock(&c->erase_completion_lock);

157 158 159
		if (!xattr)
			xattr = jffs2_verify_xattr(c);

Linus Torvalds's avatar
Linus Torvalds committed
160 161 162 163 164 165 166 167 168
		spin_lock(&c->inocache_lock);

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

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

169
		if (!ic->pino_nlink) {
170 171
			jffs2_dbg(1, "Skipping check of ino #%d with nlink/pino zero\n",
				  ic->ino);
Linus Torvalds's avatar
Linus Torvalds committed
172
			spin_unlock(&c->inocache_lock);
173
			jffs2_xattr_delete_inode(c, ic);
Linus Torvalds's avatar
Linus Torvalds committed
174 175 176 177 178
			continue;
		}
		switch(ic->state) {
		case INO_STATE_CHECKEDABSENT:
		case INO_STATE_PRESENT:
179 180
			jffs2_dbg(1, "Skipping ino #%u already checked\n",
				  ic->ino);
Linus Torvalds's avatar
Linus Torvalds committed
181 182 183 184 185
			spin_unlock(&c->inocache_lock);
			continue;

		case INO_STATE_GC:
		case INO_STATE_CHECKING:
186 187
			pr_warn("Inode #%u is in state %d during CRC check phase!\n",
				ic->ino, ic->state);
Linus Torvalds's avatar
Linus Torvalds committed
188 189 190 191 192
			spin_unlock(&c->inocache_lock);
			BUG();

		case INO_STATE_READING:
			/* We need to wait for it to finish, lest we move on
193
			   and trigger the BUG() above while we haven't yet
Linus Torvalds's avatar
Linus Torvalds committed
194
			   finished checking all its nodes */
195 196
			jffs2_dbg(1, "Waiting for ino #%u to finish reading\n",
				  ic->ino);
197 198 199 200
			/* 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--;

201
			mutex_unlock(&c->alloc_sem);
Linus Torvalds's avatar
Linus Torvalds committed
202 203 204 205 206 207 208 209 210 211 212 213
			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);

214 215
		jffs2_dbg(1, "%s(): triggering inode scan of ino#%u\n",
			  __func__, ic->ino);
Linus Torvalds's avatar
Linus Torvalds committed
216 217 218

		ret = jffs2_do_crccheck_inode(c, ic);
		if (ret)
219 220
			pr_warn("Returned error for crccheck of ino #%u. Expect badness...\n",
				ic->ino);
Linus Torvalds's avatar
Linus Torvalds committed
221 222

		jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT);
223
		mutex_unlock(&c->alloc_sem);
Linus Torvalds's avatar
Linus Torvalds committed
224 225 226
		return ret;
	}

227 228 229 230
	/* 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);
231
		mutex_unlock(&c->alloc_sem);
232
		jffs2_dbg(1, "%s(): erasing pending blocks\n", __func__);
233
		if (jffs2_erase_pending_blocks(c, 1))
234
			return 0;
235

236
		jffs2_dbg(1, "No progress from erasing block; doing GC anyway\n");
237
		mutex_lock(&c->alloc_sem);
238
		spin_lock(&c->erase_completion_lock);
239 240
	}

Linus Torvalds's avatar
Linus Torvalds committed
241 242 243 244 245 246 247
	/* First, work out which block we're garbage-collecting */
	jeb = c->gcblock;

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

	if (!jeb) {
248
		/* Couldn't find a free block. But maybe we can just erase one and make 'progress'? */
249
		if (c->nr_erasing_blocks) {
250 251 252 253
			spin_unlock(&c->erase_completion_lock);
			mutex_unlock(&c->alloc_sem);
			return -EAGAIN;
		}
254
		jffs2_dbg(1, "Couldn't find erase block to garbage collect!\n");
Linus Torvalds's avatar
Linus Torvalds committed
255
		spin_unlock(&c->erase_completion_lock);
256
		mutex_unlock(&c->alloc_sem);
Linus Torvalds's avatar
Linus Torvalds committed
257 258 259
		return -EIO;
	}

260 261
	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);
Linus Torvalds's avatar
Linus Torvalds committed
262 263 264 265
	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) {
266
		mutex_unlock(&c->alloc_sem);
Linus Torvalds's avatar
Linus Torvalds committed
267 268 269 270
		goto eraseit;
	}

	raw = jeb->gc_node;
271
	gcblock_dirty = jeb->dirty_size;
272

Linus Torvalds's avatar
Linus Torvalds committed
273
	while(ref_obsolete(raw)) {
274 275
		jffs2_dbg(1, "Node at 0x%08x is obsolete... skipping\n",
			  ref_offset(raw));
276
		raw = ref_next(raw);
Linus Torvalds's avatar
Linus Torvalds committed
277
		if (unlikely(!raw)) {
278 279 280 281
			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);
Linus Torvalds's avatar
Linus Torvalds committed
282 283
			jeb->gc_node = raw;
			spin_unlock(&c->erase_completion_lock);
284
			mutex_unlock(&c->alloc_sem);
Linus Torvalds's avatar
Linus Torvalds committed
285 286 287 288 289
			BUG();
		}
	}
	jeb->gc_node = raw;

290 291
	jffs2_dbg(1, "Going to garbage collect node at 0x%08x\n",
		  ref_offset(raw));
Linus Torvalds's avatar
Linus Torvalds committed
292 293 294 295

	if (!raw->next_in_ino) {
		/* Inode-less node. Clean marker, snapshot or something like that */
		spin_unlock(&c->erase_completion_lock);
296 297 298 299 300 301 302
		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);
		}
303
		mutex_unlock(&c->alloc_sem);
Linus Torvalds's avatar
Linus Torvalds committed
304 305 306 307 308
		goto eraseit_lock;
	}

	ic = jffs2_raw_ref_to_ic(raw);

309
#ifdef CONFIG_JFFS2_FS_XATTR
310
	/* When 'ic' refers xattr_datum/xattr_ref, this node is GCed as xattr.
311 312 313 314 315 316
	 * 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) {
317
			ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic, raw);
318
		} else {
319
			ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic, raw);
320
		}
321
		goto test_gcnode;
322 323
	}
#endif
324

Linus Torvalds's avatar
Linus Torvalds committed
325
	/* We need to hold the inocache. Either the erase_completion_lock or
326
	   the inocache_lock are sufficient; we trade down since the inocache_lock
Linus Torvalds's avatar
Linus Torvalds committed
327 328 329 330 331
	   causes less contention. */
	spin_lock(&c->inocache_lock);

	spin_unlock(&c->erase_completion_lock);

332 333 334
	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);
Linus Torvalds's avatar
Linus Torvalds committed
335 336 337 338 339 340 341 342 343 344 345 346

	/* 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:
347
		/* It's been checked, but it's not currently in-core.
Linus Torvalds's avatar
Linus Torvalds committed
348 349 350 351 352 353
		   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 {
354 355
			jffs2_dbg(1, "Ino #%u is absent but node not REF_PRISTINE. Reading.\n",
				  ic->ino);
Linus Torvalds's avatar
Linus Torvalds committed
356 357 358 359 360 361 362 363 364 365 366
		}
		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
367 368
		   by the time we actually start doing any GC, and since
		   we're holding the alloc_sem, no other garbage collection
Linus Torvalds's avatar
Linus Torvalds committed
369 370
		   can happen.
		*/
371 372
		pr_crit("Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n",
			ic->ino, ic->state);
373
		mutex_unlock(&c->alloc_sem);
Linus Torvalds's avatar
Linus Torvalds committed
374 375 376 377 378 379 380 381 382 383
		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. */

384
		mutex_unlock(&c->alloc_sem);
385 386
		jffs2_dbg(1, "%s(): waiting for ino #%u in state %d\n",
			  __func__, ic->ino, ic->state);
Linus Torvalds's avatar
Linus Torvalds committed
387
		sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
388
		/* And because we dropped the alloc_sem we must start again from the
Linus Torvalds's avatar
Linus Torvalds committed
389 390 391
		   beginning. Ponder chance of livelock here -- we're returning success
		   without actually making any progress.

392
		   Q: What are the chances that the inode is back in INO_STATE_READING
Linus Torvalds's avatar
Linus Torvalds committed
393 394 395
		   again by the time we next enter this function? And that this happens
		   enough times to cause a real delay?

396
		   A: Small enough that I don't care :)
Linus Torvalds's avatar
Linus Torvalds committed
397 398 399 400 401
		*/
		return 0;
	}

	/* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the
402
	   node intact, and we don't have to muck about with the fragtree etc.
Linus Torvalds's avatar
Linus Torvalds committed
403 404 405 406 407 408 409 410 411 412 413 414 415 416
	   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);
417
			goto test_gcnode;
Linus Torvalds's avatar
Linus Torvalds committed
418 419 420 421 422 423 424 425 426 427 428 429 430
		}

		/* 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;
431
	nlink = ic->pino_nlink;
Linus Torvalds's avatar
Linus Torvalds committed
432 433
	spin_unlock(&c->inocache_lock);

434
	f = jffs2_gc_fetch_inode(c, inum, !nlink);
Linus Torvalds's avatar
Linus Torvalds committed
435 436 437 438 439 440 441 442 443 444 445 446 447
	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);

448 449 450
 test_gcnode:
	if (jeb->dirty_size == gcblock_dirty && !ref_obsolete(jeb->gc_node)) {
		/* Eep. This really should never happen. GC is broken */
451 452
		pr_err("Error garbage collecting node at %08x!\n",
		       ref_offset(jeb->gc_node));
453
		ret = -ENOSPC;
454
	}
Linus Torvalds's avatar
Linus Torvalds committed
455
 release_sem:
456
	mutex_unlock(&c->alloc_sem);
Linus Torvalds's avatar
Linus Torvalds committed
457 458 459 460 461 462 463

 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) {
464 465
		jffs2_dbg(1, "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n",
			  c->gcblock->offset);
Linus Torvalds's avatar
Linus Torvalds committed
466 467 468 469
		/* We're GC'ing an empty block? */
		list_add_tail(&c->gcblock->list, &c->erase_pending_list);
		c->gcblock = NULL;
		c->nr_erasing_blocks++;
470
		jffs2_garbage_collect_trigger(c);
Linus Torvalds's avatar
Linus Torvalds committed
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485
	}
	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;

486
	mutex_lock(&f->sem);
Linus Torvalds's avatar
Linus Torvalds committed
487 488 489 490 491 492 493 494

	/* 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);
495
		jffs2_dbg(1, "GC block is no longer gcblock. Restart\n");
Linus Torvalds's avatar
Linus Torvalds committed
496 497 498 499
		goto upnout;
	}
	if (ref_obsolete(raw)) {
		spin_unlock(&c->erase_completion_lock);
500
		jffs2_dbg(1, "node to be GC'd was obsoleted in the meantime.\n");
Linus Torvalds's avatar
Linus Torvalds committed
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
		/* 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;
530
			}
Linus Torvalds's avatar
Linus Torvalds committed
531 532 533 534 535 536 537 538 539 540 541 542 543
			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;
	}
544

Linus Torvalds's avatar
Linus Torvalds committed
545 546 547 548 549 550 551 552 553 554 555
	/* 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 {
556 557
		pr_warn("Raw node at 0x%08x wasn't in node lists for ino #%u\n",
			ref_offset(raw), f->inocache->ino);
Linus Torvalds's avatar
Linus Torvalds committed
558
		if (ref_obsolete(raw)) {
559
			pr_warn("But it's obsolete so we don't mind too much\n");
Linus Torvalds's avatar
Linus Torvalds committed
560
		} else {
561 562
			jffs2_dbg_dump_node(c, ref_offset(raw));
			BUG();
Linus Torvalds's avatar
Linus Torvalds committed
563 564 565
		}
	}
 upnout:
566
	mutex_unlock(&f->sem);
Linus Torvalds's avatar
Linus Torvalds committed
567 568 569 570

	return ret;
}

571
static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
Linus Torvalds's avatar
Linus Torvalds committed
572 573 574 575 576 577 578 579 580 581
					  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;

582 583
	jffs2_dbg(1, "Going to GC REF_PRISTINE node at 0x%08x\n",
		  ref_offset(raw));
Linus Torvalds's avatar
Linus Torvalds committed
584

585
	alloclen = rawlen = ref_totlen(c, c->gcblock, raw);
Linus Torvalds's avatar
Linus Torvalds committed
586 587 588 589

	/* 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. */
590 591 592
	if (ic && alloclen > sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN)
		alloclen = sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN;

593
	ret = jffs2_reserve_space_gc(c, alloclen, &alloclen, rawlen);
594
	/* 'rawlen' is not the exact summary size; it is only an upper estimation */
595

Linus Torvalds's avatar
Linus Torvalds committed
596 597 598 599 600 601 602 603 604 605
	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)
606
		return -ENOMEM;
Linus Torvalds's avatar
Linus Torvalds committed
607 608 609 610 611 612 613 614 615

	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) {
616 617
		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);
Linus Torvalds's avatar
Linus Torvalds committed
618 619 620 621 622 623 624
		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) {
625 626 627
			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);
Linus Torvalds's avatar
Linus Torvalds committed
628 629 630 631 632 633
			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) {
634 635 636
				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);
Linus Torvalds's avatar
Linus Torvalds committed
637 638 639 640 641 642 643 644
				goto bail;
			}
		}
		break;

	case JFFS2_NODETYPE_DIRENT:
		crc = crc32(0, node, sizeof(node->d)-8);
		if (je32_to_cpu(node->d.node_crc) != crc) {
645 646 647
			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);
Linus Torvalds's avatar
Linus Torvalds committed
648 649 650
			goto bail;
		}

651
		if (strnlen(node->d.name, node->d.nsize) != node->d.nsize) {
652 653
			pr_warn("Name in dirent node at 0x%08x contains zeroes\n",
				ref_offset(raw));
654 655 656
			goto bail;
		}

Linus Torvalds's avatar
Linus Torvalds committed
657 658 659
		if (node->d.nsize) {
			crc = crc32(0, node->d.name, node->d.nsize);
			if (je32_to_cpu(node->d.name_crc) != crc) {
660 661 662
				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);
Linus Torvalds's avatar
Linus Torvalds committed
663 664 665 666 667
				goto bail;
			}
		}
		break;
	default:
668 669
		/* If it's inode-less, we don't _know_ what it is. Just copy it intact */
		if (ic) {
670 671
			pr_warn("Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n",
				ref_offset(raw), je16_to_cpu(node->u.nodetype));
672 673
			goto bail;
		}
Linus Torvalds's avatar
Linus Torvalds committed
674 675 676 677
	}

	/* OK, all the CRCs are good; this node can just be copied as-is. */
 retry:
678
	phys_ofs = write_ofs(c);
Linus Torvalds's avatar
Linus Torvalds committed
679 680 681 682

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

	if (ret || (retlen != rawlen)) {
683 684
		pr_notice("Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n",
			  rawlen, phys_ofs, ret, retlen);
Linus Torvalds's avatar
Linus Torvalds committed
685
		if (retlen) {
686
			jffs2_add_physical_node_ref(c, phys_ofs | REF_OBSOLETE, rawlen, NULL);
Linus Torvalds's avatar
Linus Torvalds committed
687
		} else {
688 689
			pr_notice("Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n",
				  phys_ofs);
Linus Torvalds's avatar
Linus Torvalds committed
690
		}
691
		if (!retried) {
Linus Torvalds's avatar
Linus Torvalds committed
692 693 694 695 696 697
			/* Try to reallocate space and retry */
			uint32_t dummy;
			struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size];

			retried = 1;

698
			jffs2_dbg(1, "Retrying failed write of REF_PRISTINE node.\n");
699

700 701
			jffs2_dbg_acct_sanity_check(c,jeb);
			jffs2_dbg_acct_paranoia_check(c, jeb);
Linus Torvalds's avatar
Linus Torvalds committed
702

703
			ret = jffs2_reserve_space_gc(c, rawlen, &dummy, rawlen);
704 705
						/* this is not the exact summary size of it,
							it is only an upper estimation */
Linus Torvalds's avatar
Linus Torvalds committed
706 707

			if (!ret) {
708 709
				jffs2_dbg(1, "Allocated space at 0x%08x to retry failed write.\n",
					  phys_ofs);
Linus Torvalds's avatar
Linus Torvalds committed
710

711 712
				jffs2_dbg_acct_sanity_check(c,jeb);
				jffs2_dbg_acct_paranoia_check(c, jeb);
Linus Torvalds's avatar
Linus Torvalds committed
713 714 715

				goto retry;
			}
716 717
			jffs2_dbg(1, "Failed to allocate space to retry failed write: %d!\n",
				  ret);
Linus Torvalds's avatar
Linus Torvalds committed
718 719 720 721 722 723
		}

		if (!ret)
			ret = -EIO;
		goto out_node;
	}
724
	jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, rawlen, ic);
Linus Torvalds's avatar
Linus Torvalds committed
725 726

	jffs2_mark_node_obsolete(c, raw);
727 728
	jffs2_dbg(1, "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n",
		  ref_offset(raw));
Linus Torvalds's avatar
Linus Torvalds committed
729 730 731 732 733 734 735 736 737

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

738
static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds's avatar
Linus Torvalds committed
739 740 741 742
					struct jffs2_inode_info *f, struct jffs2_full_dnode *fn)
{
	struct jffs2_full_dnode *new_fn;
	struct jffs2_raw_inode ri;
743
	struct jffs2_node_frag *last_frag;
744
	union jffs2_device_node dev;
745 746
	char *mdata = NULL;
	int mdatalen = 0;
747
	uint32_t alloclen, ilen;
Linus Torvalds's avatar
Linus Torvalds committed
748 749 750 751 752
	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 */
753
		mdatalen = jffs2_encode_dev(&dev, JFFS2_F_I_RDEV(f));
Linus Torvalds's avatar
Linus Torvalds committed
754
		mdata = (char *)&dev;
755 756
		jffs2_dbg(1, "%s(): Writing %d bytes of kdev_t\n",
			  __func__, mdatalen);
Linus Torvalds's avatar
Linus Torvalds committed
757 758 759 760
	} else if (S_ISLNK(JFFS2_F_I_MODE(f))) {
		mdatalen = fn->size;
		mdata = kmalloc(fn->size, GFP_KERNEL);
		if (!mdata) {
761
			pr_warn("kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n");
Linus Torvalds's avatar
Linus Torvalds committed
762 763 764 765
			return -ENOMEM;
		}
		ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen);
		if (ret) {
766 767
			pr_warn("read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n",
				ret);
Linus Torvalds's avatar
Linus Torvalds committed
768 769 770
			kfree(mdata);
			return ret;
		}
771 772
		jffs2_dbg(1, "%s(): Writing %d bites of symlink target\n",
			  __func__, mdatalen);
Linus Torvalds's avatar
Linus Torvalds committed
773 774

	}
775

776
	ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &alloclen,
777
				JFFS2_SUMMARY_INODE_SIZE);
Linus Torvalds's avatar
Linus Torvalds committed
778
	if (ret) {
779 780
		pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n",
			sizeof(ri) + mdatalen, ret);
Linus Torvalds's avatar
Linus Torvalds committed
781 782
		goto out;
	}
783

784 785 786 787 788 789 790
	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);
791

Linus Torvalds's avatar
Linus Torvalds committed
792 793 794 795 796 797 798 799 800 801 802
	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));
803
	ri.isize = cpu_to_je32(ilen);
Linus Torvalds's avatar
Linus Torvalds committed
804 805 806 807 808 809 810 811 812 813
	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));

814
	new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, ALLOC_GC);
Linus Torvalds's avatar
Linus Torvalds committed
815 816

	if (IS_ERR(new_fn)) {
817
		pr_warn("Error writing new dnode: %ld\n", PTR_ERR(new_fn));
Linus Torvalds's avatar
Linus Torvalds committed
818 819 820 821 822 823 824 825 826 827 828 829
		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;
}

830
static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds's avatar
Linus Torvalds committed
831 832 833 834
					struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
{
	struct jffs2_full_dirent *new_fd;
	struct jffs2_raw_dirent rd;
835
	uint32_t alloclen;
Linus Torvalds's avatar
Linus Torvalds committed
836 837 838 839 840 841 842 843 844 845 846
	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);
847 848 849 850
	/* 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));
851
	else
852
		rd.mctime = cpu_to_je32(0);
Linus Torvalds's avatar
Linus Torvalds committed
853 854 855
	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));
856

857
	ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &alloclen,
858
				JFFS2_SUMMARY_DIRENT_SIZE(rd.nsize));
Linus Torvalds's avatar
Linus Torvalds committed
859
	if (ret) {
860 861
		pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n",
			sizeof(rd)+rd.nsize, ret);
Linus Torvalds's avatar
Linus Torvalds committed
862 863
		return ret;
	}
864
	new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, ALLOC_GC);
Linus Torvalds's avatar
Linus Torvalds committed
865 866

	if (IS_ERR(new_fd)) {
867 868
		pr_warn("jffs2_write_dirent in garbage_collect_dirent failed: %ld\n",
			PTR_ERR(new_fd));
Linus Torvalds's avatar
Linus Torvalds committed
869 870 871 872 873 874
		return PTR_ERR(new_fd);
	}
	jffs2_add_fd_to_list(c, new_fd, &f->dents);
	return 0;
}

875
static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
Linus Torvalds's avatar
Linus Torvalds committed
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
					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... */
902
		mutex_lock(&c->erase_free_sem);
Linus Torvalds's avatar
Linus Torvalds committed
903 904 905

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

906 907
			cond_resched();

Linus Torvalds's avatar
Linus Torvalds committed
908 909 910 911 912 913 914 915
			/* 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;

916
			/* Doesn't matter if there's one in the same erase block. We're going to
Linus Torvalds's avatar
Linus Torvalds committed
917
			   delete it too at the same time. */
918
			if (SECTOR_ADDR(raw->flash_offset) == SECTOR_ADDR(fd->raw->flash_offset))
Linus Torvalds's avatar
Linus Torvalds committed
919 920
				continue;

921 922
			jffs2_dbg(1, "Check potential deletion dirent at %08x\n",
				  ref_offset(raw));
Linus Torvalds's avatar
Linus Torvalds committed
923 924 925 926 927

			/* 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) {
928 929
				pr_warn("%s(): Read error (%d) reading obsolete node at %08x\n",
					__func__, ret, ref_offset(raw));
Linus Torvalds's avatar
Linus Torvalds committed
930 931 932 933
				/* If we can't read it, we don't need to continue to obsolete it. Continue */
				continue;
			}
			if (retlen != rawlen) {
934 935 936
				pr_warn("%s(): Short read (%zd not %u) reading header from obsolete node at %08x\n",
					__func__, retlen, rawlen,
					ref_offset(raw));
Linus Torvalds's avatar
Linus Torvalds committed
937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
				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 */
958
			mutex_unlock(&c->erase_free_sem);
Linus Torvalds's avatar
Linus Torvalds committed
959

960 961 962
			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));
Linus Torvalds's avatar
Linus Torvalds committed
963 964 965 966 967
			kfree(rd);

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

968
		mutex_unlock(&c->erase_free_sem);
Linus Torvalds's avatar
Linus Torvalds committed
969 970 971
		kfree(rd);
	}

972
	/* FIXME: If we're deleting a dirent which contains the current mtime and ctime,
973 974
	   we should update the metadata node with those times accordingly */

Linus Torvalds's avatar
Linus Torvalds committed
975 976 977 978 979 980 981 982 983 984
	/* 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) {
985 986
		pr_warn("Deletion dirent \"%s\" not found in list for ino #%u\n",
			fd->name, f->inocache->ino);
Linus Torvalds's avatar
Linus Torvalds committed
987 988 989 990 991 992 993 994 995 996 997 998 999
	}
	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;
1000
	uint32_t alloclen, ilen;
Linus Torvalds's avatar
Linus Torvalds committed
1001 1002
	int ret;

1003 1004
	jffs2_dbg(1, "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n",
		  f->inocache->ino, start, end);
1005

Linus Torvalds's avatar
Linus Torvalds committed
1006 1007 1008 1009 1010
	memset(&ri, 0, sizeof(ri));

	if(fn->frags > 1) {
		size_t readlen;
		uint32_t crc;
1011
		/* It's partially obsoleted by a later write. So we have to
Linus Torvalds's avatar
Linus Torvalds committed
1012 1013 1014
		   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) {
1015 1016
			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);
Linus Torvalds's avatar
Linus Torvalds committed
1017 1018 1019
			goto fill;
		}
		if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) {
1020 1021 1022
			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);
Linus Torvalds's avatar
Linus Torvalds committed
1023 1024 1025
			return -EIO;
		}
		if (je32_to_cpu(ri.totlen) != sizeof(ri)) {
1026 1027 1028
			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));
Linus Torvalds's avatar
Linus Torvalds committed
1029 1030 1031 1032
			return -EIO;
		}
		crc = crc32(0, &ri, sizeof(ri)-8);
		if (crc != je32_to_cpu(ri.node_crc)) {
1033 1034 1035
			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);
Linus Torvalds's avatar
Linus Torvalds committed
1036
			/* FIXME: We could possibly deal with this by writing new holes for each frag */
1037 1038
			pr_warn("Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
				start, end, f->inocache->ino);
Linus Torvalds's avatar
Linus Torvalds committed
1039 1040 1041
			goto fill;
		}
		if (ri.compr != JFFS2_COMPR_ZERO) {
1042 1043 1044 1045
			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);
Linus Torvalds's avatar
Linus Torvalds committed
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061
			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;
	}
1062

1063 1064 1065 1066 1067 1068 1069 1070
	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);

Linus Torvalds's avatar
Linus Torvalds committed
1071 1072 1073
	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));
1074
	ri.isize = cpu_to_je32(ilen);
Linus Torvalds's avatar
Linus Torvalds committed
1075 1076 1077 1078 1079 1080
	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));

1081 1082
	ret = jffs2_reserve_space_gc(c, sizeof(ri), &alloclen,
				     JFFS2_SUMMARY_INODE_SIZE);
Linus Torvalds's avatar
Linus Torvalds committed
1083
	if (ret) {
1084 1085
		pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n",
			sizeof(ri), ret);
Linus Torvalds's avatar
Linus Torvalds committed
1086 1087
		return ret;
	}
1088
	new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, ALLOC_GC);
Linus Torvalds's avatar
Linus Torvalds committed
1089 1090

	if (IS_ERR(new_fn)) {
1091
		pr_warn("Error writing new hole node: %ld\n", PTR_ERR(new_fn));
Linus Torvalds's avatar
Linus Torvalds committed
1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
		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;
	}

1104
	/*
Linus Torvalds's avatar
Linus Torvalds committed
1105 1106
	 * 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
1107
	 * number as before. (Except in case of error -- see 'goto fill;'
Linus Torvalds's avatar
Linus Torvalds committed
1108 1109 1110
	 * above.)
	 */
	D1(if(unlikely(fn->frags <= 1)) {
1111 1112 1113
			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));
Linus Torvalds's avatar
Linus Torvalds committed
1114 1115 1116 1117 1118
	});

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

1119
	for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs);
Linus Torvalds's avatar
Linus Torvalds committed
1120 1121 1122 1123 1124 1125 1126 1127 1128 1129
	     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) {
1130
		pr_warn("%s(): Old node still has frags!\n", __func__);
Linus Torvalds's avatar
Linus Torvalds committed
1131 1132 1133
		BUG();
	}
	if (!new_fn->frags) {
1134
		pr_warn("%s(): New node has no frags!\n", __func__);
Linus Torvalds's avatar
Linus Torvalds committed
1135 1136
		BUG();
	}
1137

Linus Torvalds's avatar
Linus Torvalds committed
1138 1139
	jffs2_mark_node_obsolete(c, fn->raw);
	jffs2_free_full_dnode(fn);
1140

Linus Torvalds's avatar
Linus Torvalds committed
1141 1142 1143
	return 0;
}

1144
static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *orig_jeb,
Linus Torvalds's avatar
Linus Torvalds committed
1145 1146 1147 1148 1149
				       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;
1150
	uint32_t alloclen, offset, orig_end, orig_start;
Linus Torvalds's avatar
Linus Torvalds committed
1151 1152 1153 1154
	int ret = 0;
	unsigned char *comprbuf = NULL, *writebuf;
	unsigned long pg;
	unsigned char *pg_ptr;
1155

Linus Torvalds's avatar
Linus Torvalds committed
1156 1157
	memset(&ri, 0, sizeof(ri));

1158 1159
	jffs2_dbg(1, "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n",
		  f->inocache->ino, start, end);
Linus Torvalds's avatar
Linus Torvalds committed
1160 1161 1162 1163 1164 1165 1166

	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
1167 1168
		   to be dirty. Otherwise we end up with GC just going round in
		   circles dirtying the nodes it already wrote out, especially
Linus Torvalds's avatar
Linus Torvalds committed
1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189
		   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) {
1190 1191
				jffs2_dbg(1, "Expanding down to cover partial frag (0x%x-0x%x)\n",
					  frag->ofs, frag->ofs+frag->size);
Linus Torvalds's avatar
Linus Torvalds committed
1192 1193 1194 1195 1196
				start = frag->ofs;
				continue;
			}
			/* OK. This frag holds the first byte of the page. */
			if (!frag->node || !frag->node->raw) {
1197 1198
				jffs2_dbg(1, "First frag in page is hole (0x%x-0x%x). Not expanding down.\n",
					  frag->ofs, frag->ofs+frag->size);
Linus Torvalds's avatar
Linus Torvalds committed
1199 1200 1201
				break;
			} else {

1202
				/* OK, it's a frag which extends to the beginning of the page. Does it live
Linus Torvalds's avatar
Linus Torvalds committed
1203 1204 1205 1206 1207 1208 1209 1210 1211
				   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) {
1212 1213 1214 1215
					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));
Linus Torvalds's avatar
Linus Torvalds committed
1216 1217 1218 1219
					start = frag->ofs;
					break;
				}
				if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1220 1221 1222 1223
					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);
Linus Torvalds's avatar
Linus Torvalds committed
1224 1225 1226
					break;
				}

1227 1228 1229 1230
				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);
Linus Torvalds's avatar
Linus Torvalds committed
1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245
				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) {
1246 1247
				jffs2_dbg(1, "Expanding up to cover partial frag (0x%x-0x%x)\n",
					  frag->ofs, frag->ofs+frag->size);
Linus Torvalds's avatar
Linus Torvalds committed
1248 1249 1250 1251 1252
				end = frag->ofs + frag->size;
				continue;
			}

			if (!frag->node || !frag->node->raw) {
1253 1254
				jffs2_dbg(1, "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n",
					  frag->ofs, frag->ofs+frag->size);
Linus Torvalds's avatar
Linus Torvalds committed
1255 1256 1257
				break;
			} else {

1258
				/* OK, it's a frag which extends to the beginning of the page. Does it live
Linus Torvalds's avatar
Linus Torvalds committed
1259 1260 1261 1262 1263 1264 1265 1266 1267
				   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) {
1268 1269 1270 1271
					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));
Linus Torvalds's avatar
Linus Torvalds committed
1272 1273 1274 1275
					end = frag->ofs + frag->size;
					break;
				}
				if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1276 1277 1278 1279
					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);
Linus Torvalds's avatar
Linus Torvalds committed
1280 1281 1282
					break;
				}

1283 1284 1285 1286
				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);
Linus Torvalds's avatar
Linus Torvalds committed
1287 1288 1289 1290
				end = frag->ofs + frag->size;
				break;
			}
		}
1291 1292
		jffs2_dbg(1, "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n",
			  orig_start, orig_end, start, end);
Linus Torvalds's avatar
Linus Torvalds committed
1293

1294
		D1(BUG_ON(end > frag_last(&f->fragtree)->ofs + frag_last(&f->fragtree)->size));
Linus Torvalds's avatar
Linus Torvalds committed
1295 1296 1297
		BUG_ON(end < orig_end);
		BUG_ON(start > orig_start);
	}
1298

1299 1300 1301 1302 1303 1304 1305 1306 1307
	/* The rules state that we must obtain the page lock *before* f->sem, so
	 * drop f->sem temporarily. Since we also hold c->alloc_sem, nothing's
	 * actually going to *change* so we're safe; we only allow reading.
	 *
	 * It is important to note that jffs2_write_begin() will ensure that its
	 * page is marked Uptodate before allocating space. That means that if we
	 * end up here trying to GC the *same* page that jffs2_write_begin() is
	 * trying to write out, read_cache_page() will not deadlock. */
	mutex_unlock(&f->sem);
Linus Torvalds's avatar
Linus Torvalds committed
1308
	pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg);
1309
	mutex_lock(&f->sem);
Linus Torvalds's avatar
Linus Torvalds committed
1310 1311

	if (IS_ERR(pg_ptr)) {
1312 1313
		pr_warn("read_cache_page() returned error: %ld\n",
			PTR_ERR(pg_ptr));
Linus Torvalds's avatar
Linus Torvalds committed
1314 1315 1316 1317 1318 1319 1320 1321 1322
		return PTR_ERR(pg_ptr);
	}

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

1323
		ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN,
1324
					&alloclen, JFFS2_SUMMARY_INODE_SIZE);
Linus Torvalds's avatar
Linus Torvalds committed
1325 1326

		if (ret) {
1327 1328
			pr_warn("jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n",
				sizeof(ri) + JFFS2_MIN_DATA_LEN, ret);
Linus Torvalds's avatar
Linus Torvalds committed
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
			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));
1359

1360
		new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, ALLOC_GC);
Linus Torvalds's avatar
Linus Torvalds committed
1361 1362 1363 1364

		jffs2_free_comprbuf(comprbuf, writebuf);

		if (IS_ERR(new_fn)) {
1365 1366
			pr_warn("Error writing new dnode: %ld\n",
				PTR_ERR(new_fn));
Linus Torvalds's avatar
Linus Torvalds committed
1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381
			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;
}