ACPI: thinkpad-acpi: add development version tag
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / jffs2 / gc.c
blob3b6f2fa12cff16df800dee5ebb9b89fe93698eb5
1 /*
2 * JFFS2 -- Journalling Flash File System, Version 2.
4 * Copyright © 2001-2007 Red Hat, Inc.
6 * Created by David Woodhouse <dwmw2@infradead.org>
8 * For licensing information, see the file 'LICENCE' in this directory.
12 #include <linux/kernel.h>
13 #include <linux/mtd/mtd.h>
14 #include <linux/slab.h>
15 #include <linux/pagemap.h>
16 #include <linux/crc32.h>
17 #include <linux/compiler.h>
18 #include <linux/stat.h>
19 #include "nodelist.h"
20 #include "compr.h"
22 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
23 struct jffs2_inode_cache *ic,
24 struct jffs2_raw_node_ref *raw);
25 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
26 struct jffs2_inode_info *f, struct jffs2_full_dnode *fd);
27 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
28 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
29 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
30 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd);
31 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
32 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
33 uint32_t start, uint32_t end);
34 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
35 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
36 uint32_t start, uint32_t end);
37 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
38 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f);
40 /* Called with erase_completion_lock held */
41 static struct jffs2_eraseblock *jffs2_find_gc_block(struct jffs2_sb_info *c)
43 struct jffs2_eraseblock *ret;
44 struct list_head *nextlist = NULL;
45 int n = jiffies % 128;
47 /* Pick an eraseblock to garbage collect next. This is where we'll
48 put the clever wear-levelling algorithms. Eventually. */
49 /* We possibly want to favour the dirtier blocks more when the
50 number of free blocks is low. */
51 again:
52 if (!list_empty(&c->bad_used_list) && c->nr_free_blocks > c->resv_blocks_gcbad) {
53 D1(printk(KERN_DEBUG "Picking block from bad_used_list to GC next\n"));
54 nextlist = &c->bad_used_list;
55 } else if (n < 50 && !list_empty(&c->erasable_list)) {
56 /* Note that most of them will have gone directly to be erased.
57 So don't favour the erasable_list _too_ much. */
58 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next\n"));
59 nextlist = &c->erasable_list;
60 } else if (n < 110 && !list_empty(&c->very_dirty_list)) {
61 /* Most of the time, pick one off the very_dirty list */
62 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next\n"));
63 nextlist = &c->very_dirty_list;
64 } else if (n < 126 && !list_empty(&c->dirty_list)) {
65 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next\n"));
66 nextlist = &c->dirty_list;
67 } else if (!list_empty(&c->clean_list)) {
68 D1(printk(KERN_DEBUG "Picking block from clean_list to GC next\n"));
69 nextlist = &c->clean_list;
70 } else if (!list_empty(&c->dirty_list)) {
71 D1(printk(KERN_DEBUG "Picking block from dirty_list to GC next (clean_list was empty)\n"));
73 nextlist = &c->dirty_list;
74 } else if (!list_empty(&c->very_dirty_list)) {
75 D1(printk(KERN_DEBUG "Picking block from very_dirty_list to GC next (clean_list and dirty_list were empty)\n"));
76 nextlist = &c->very_dirty_list;
77 } else if (!list_empty(&c->erasable_list)) {
78 D1(printk(KERN_DEBUG "Picking block from erasable_list to GC next (clean_list and {very_,}dirty_list were empty)\n"));
80 nextlist = &c->erasable_list;
81 } else if (!list_empty(&c->erasable_pending_wbuf_list)) {
82 /* There are blocks are wating for the wbuf sync */
83 D1(printk(KERN_DEBUG "Synching wbuf in order to reuse erasable_pending_wbuf_list blocks\n"));
84 spin_unlock(&c->erase_completion_lock);
85 jffs2_flush_wbuf_pad(c);
86 spin_lock(&c->erase_completion_lock);
87 goto again;
88 } else {
89 /* Eep. All were empty */
90 D1(printk(KERN_NOTICE "jffs2: No clean, dirty _or_ erasable blocks to GC from! Where are they all?\n"));
91 return NULL;
94 ret = list_entry(nextlist->next, struct jffs2_eraseblock, list);
95 list_del(&ret->list);
96 c->gcblock = ret;
97 ret->gc_node = ret->first_node;
98 if (!ret->gc_node) {
99 printk(KERN_WARNING "Eep. ret->gc_node for block at 0x%08x is NULL\n", ret->offset);
100 BUG();
103 /* Have we accidentally picked a clean block with wasted space ? */
104 if (ret->wasted_size) {
105 D1(printk(KERN_DEBUG "Converting wasted_size %08x to dirty_size\n", ret->wasted_size));
106 ret->dirty_size += ret->wasted_size;
107 c->wasted_size -= ret->wasted_size;
108 c->dirty_size += ret->wasted_size;
109 ret->wasted_size = 0;
112 return ret;
115 /* jffs2_garbage_collect_pass
116 * Make a single attempt to progress GC. Move one node, and possibly
117 * start erasing one eraseblock.
119 int jffs2_garbage_collect_pass(struct jffs2_sb_info *c)
121 struct jffs2_inode_info *f;
122 struct jffs2_inode_cache *ic;
123 struct jffs2_eraseblock *jeb;
124 struct jffs2_raw_node_ref *raw;
125 uint32_t gcblock_dirty;
126 int ret = 0, inum, nlink;
127 int xattr = 0;
129 if (mutex_lock_interruptible(&c->alloc_sem))
130 return -EINTR;
132 for (;;) {
133 spin_lock(&c->erase_completion_lock);
134 if (!c->unchecked_size)
135 break;
137 /* We can't start doing GC yet. We haven't finished checking
138 the node CRCs etc. Do it now. */
140 /* checked_ino is protected by the alloc_sem */
141 if (c->checked_ino > c->highest_ino && xattr) {
142 printk(KERN_CRIT "Checked all inodes but still 0x%x bytes of unchecked space?\n",
143 c->unchecked_size);
144 jffs2_dbg_dump_block_lists_nolock(c);
145 spin_unlock(&c->erase_completion_lock);
146 mutex_unlock(&c->alloc_sem);
147 return -ENOSPC;
150 spin_unlock(&c->erase_completion_lock);
152 if (!xattr)
153 xattr = jffs2_verify_xattr(c);
155 spin_lock(&c->inocache_lock);
157 ic = jffs2_get_ino_cache(c, c->checked_ino++);
159 if (!ic) {
160 spin_unlock(&c->inocache_lock);
161 continue;
164 if (!ic->pino_nlink) {
165 D1(printk(KERN_DEBUG "Skipping check of ino #%d with nlink/pino zero\n",
166 ic->ino));
167 spin_unlock(&c->inocache_lock);
168 jffs2_xattr_delete_inode(c, ic);
169 continue;
171 switch(ic->state) {
172 case INO_STATE_CHECKEDABSENT:
173 case INO_STATE_PRESENT:
174 D1(printk(KERN_DEBUG "Skipping ino #%u already checked\n", ic->ino));
175 spin_unlock(&c->inocache_lock);
176 continue;
178 case INO_STATE_GC:
179 case INO_STATE_CHECKING:
180 printk(KERN_WARNING "Inode #%u is in state %d during CRC check phase!\n", ic->ino, ic->state);
181 spin_unlock(&c->inocache_lock);
182 BUG();
184 case INO_STATE_READING:
185 /* We need to wait for it to finish, lest we move on
186 and trigger the BUG() above while we haven't yet
187 finished checking all its nodes */
188 D1(printk(KERN_DEBUG "Waiting for ino #%u to finish reading\n", ic->ino));
189 /* We need to come back again for the _same_ inode. We've
190 made no progress in this case, but that should be OK */
191 c->checked_ino--;
193 mutex_unlock(&c->alloc_sem);
194 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
195 return 0;
197 default:
198 BUG();
200 case INO_STATE_UNCHECKED:
203 ic->state = INO_STATE_CHECKING;
204 spin_unlock(&c->inocache_lock);
206 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() triggering inode scan of ino#%u\n", ic->ino));
208 ret = jffs2_do_crccheck_inode(c, ic);
209 if (ret)
210 printk(KERN_WARNING "Returned error for crccheck of ino #%u. Expect badness...\n", ic->ino);
212 jffs2_set_inocache_state(c, ic, INO_STATE_CHECKEDABSENT);
213 mutex_unlock(&c->alloc_sem);
214 return ret;
217 /* First, work out which block we're garbage-collecting */
218 jeb = c->gcblock;
220 if (!jeb)
221 jeb = jffs2_find_gc_block(c);
223 if (!jeb) {
224 /* Couldn't find a free block. But maybe we can just erase one and make 'progress'? */
225 if (!list_empty(&c->erase_pending_list)) {
226 spin_unlock(&c->erase_completion_lock);
227 mutex_unlock(&c->alloc_sem);
228 return -EAGAIN;
230 D1(printk(KERN_NOTICE "jffs2: Couldn't find erase block to garbage collect!\n"));
231 spin_unlock(&c->erase_completion_lock);
232 mutex_unlock(&c->alloc_sem);
233 return -EIO;
236 D1(printk(KERN_DEBUG "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));
237 D1(if (c->nextblock)
238 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));
240 if (!jeb->used_size) {
241 mutex_unlock(&c->alloc_sem);
242 goto eraseit;
245 raw = jeb->gc_node;
246 gcblock_dirty = jeb->dirty_size;
248 while(ref_obsolete(raw)) {
249 D1(printk(KERN_DEBUG "Node at 0x%08x is obsolete... skipping\n", ref_offset(raw)));
250 raw = ref_next(raw);
251 if (unlikely(!raw)) {
252 printk(KERN_WARNING "eep. End of raw list while still supposedly nodes to GC\n");
253 printk(KERN_WARNING "erase block at 0x%08x. free_size 0x%08x, dirty_size 0x%08x, used_size 0x%08x\n",
254 jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size);
255 jeb->gc_node = raw;
256 spin_unlock(&c->erase_completion_lock);
257 mutex_unlock(&c->alloc_sem);
258 BUG();
261 jeb->gc_node = raw;
263 D1(printk(KERN_DEBUG "Going to garbage collect node at 0x%08x\n", ref_offset(raw)));
265 if (!raw->next_in_ino) {
266 /* Inode-less node. Clean marker, snapshot or something like that */
267 spin_unlock(&c->erase_completion_lock);
268 if (ref_flags(raw) == REF_PRISTINE) {
269 /* It's an unknown node with JFFS2_FEATURE_RWCOMPAT_COPY */
270 jffs2_garbage_collect_pristine(c, NULL, raw);
271 } else {
272 /* Just mark it obsolete */
273 jffs2_mark_node_obsolete(c, raw);
275 mutex_unlock(&c->alloc_sem);
276 goto eraseit_lock;
279 ic = jffs2_raw_ref_to_ic(raw);
281 #ifdef CONFIG_JFFS2_FS_XATTR
282 /* When 'ic' refers xattr_datum/xattr_ref, this node is GCed as xattr.
283 * We can decide whether this node is inode or xattr by ic->class. */
284 if (ic->class == RAWNODE_CLASS_XATTR_DATUM
285 || ic->class == RAWNODE_CLASS_XATTR_REF) {
286 spin_unlock(&c->erase_completion_lock);
288 if (ic->class == RAWNODE_CLASS_XATTR_DATUM) {
289 ret = jffs2_garbage_collect_xattr_datum(c, (struct jffs2_xattr_datum *)ic, raw);
290 } else {
291 ret = jffs2_garbage_collect_xattr_ref(c, (struct jffs2_xattr_ref *)ic, raw);
293 goto test_gcnode;
295 #endif
297 /* We need to hold the inocache. Either the erase_completion_lock or
298 the inocache_lock are sufficient; we trade down since the inocache_lock
299 causes less contention. */
300 spin_lock(&c->inocache_lock);
302 spin_unlock(&c->erase_completion_lock);
304 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass collecting from block @0x%08x. Node @0x%08x(%d), ino #%u\n", jeb->offset, ref_offset(raw), ref_flags(raw), ic->ino));
306 /* Three possibilities:
307 1. Inode is already in-core. We must iget it and do proper
308 updating to its fragtree, etc.
309 2. Inode is not in-core, node is REF_PRISTINE. We lock the
310 inocache to prevent a read_inode(), copy the node intact.
311 3. Inode is not in-core, node is not pristine. We must iget()
312 and take the slow path.
315 switch(ic->state) {
316 case INO_STATE_CHECKEDABSENT:
317 /* It's been checked, but it's not currently in-core.
318 We can just copy any pristine nodes, but have
319 to prevent anyone else from doing read_inode() while
320 we're at it, so we set the state accordingly */
321 if (ref_flags(raw) == REF_PRISTINE)
322 ic->state = INO_STATE_GC;
323 else {
324 D1(printk(KERN_DEBUG "Ino #%u is absent but node not REF_PRISTINE. Reading.\n",
325 ic->ino));
327 break;
329 case INO_STATE_PRESENT:
330 /* It's in-core. GC must iget() it. */
331 break;
333 case INO_STATE_UNCHECKED:
334 case INO_STATE_CHECKING:
335 case INO_STATE_GC:
336 /* Should never happen. We should have finished checking
337 by the time we actually start doing any GC, and since
338 we're holding the alloc_sem, no other garbage collection
339 can happen.
341 printk(KERN_CRIT "Inode #%u already in state %d in jffs2_garbage_collect_pass()!\n",
342 ic->ino, ic->state);
343 mutex_unlock(&c->alloc_sem);
344 spin_unlock(&c->inocache_lock);
345 BUG();
347 case INO_STATE_READING:
348 /* Someone's currently trying to read it. We must wait for
349 them to finish and then go through the full iget() route
350 to do the GC. However, sometimes read_inode() needs to get
351 the alloc_sem() (for marking nodes invalid) so we must
352 drop the alloc_sem before sleeping. */
354 mutex_unlock(&c->alloc_sem);
355 D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() waiting for ino #%u in state %d\n",
356 ic->ino, ic->state));
357 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
358 /* And because we dropped the alloc_sem we must start again from the
359 beginning. Ponder chance of livelock here -- we're returning success
360 without actually making any progress.
362 Q: What are the chances that the inode is back in INO_STATE_READING
363 again by the time we next enter this function? And that this happens
364 enough times to cause a real delay?
366 A: Small enough that I don't care :)
368 return 0;
371 /* OK. Now if the inode is in state INO_STATE_GC, we are going to copy the
372 node intact, and we don't have to muck about with the fragtree etc.
373 because we know it's not in-core. If it _was_ in-core, we go through
374 all the iget() crap anyway */
376 if (ic->state == INO_STATE_GC) {
377 spin_unlock(&c->inocache_lock);
379 ret = jffs2_garbage_collect_pristine(c, ic, raw);
381 spin_lock(&c->inocache_lock);
382 ic->state = INO_STATE_CHECKEDABSENT;
383 wake_up(&c->inocache_wq);
385 if (ret != -EBADFD) {
386 spin_unlock(&c->inocache_lock);
387 goto test_gcnode;
390 /* Fall through if it wanted us to, with inocache_lock held */
393 /* Prevent the fairly unlikely race where the gcblock is
394 entirely obsoleted by the final close of a file which had
395 the only valid nodes in the block, followed by erasure,
396 followed by freeing of the ic because the erased block(s)
397 held _all_ the nodes of that inode.... never been seen but
398 it's vaguely possible. */
400 inum = ic->ino;
401 nlink = ic->pino_nlink;
402 spin_unlock(&c->inocache_lock);
404 f = jffs2_gc_fetch_inode(c, inum, !nlink);
405 if (IS_ERR(f)) {
406 ret = PTR_ERR(f);
407 goto release_sem;
409 if (!f) {
410 ret = 0;
411 goto release_sem;
414 ret = jffs2_garbage_collect_live(c, jeb, raw, f);
416 jffs2_gc_release_inode(c, f);
418 test_gcnode:
419 if (jeb->dirty_size == gcblock_dirty && !ref_obsolete(jeb->gc_node)) {
420 /* Eep. This really should never happen. GC is broken */
421 printk(KERN_ERR "Error garbage collecting node at %08x!\n", ref_offset(jeb->gc_node));
422 ret = -ENOSPC;
424 release_sem:
425 mutex_unlock(&c->alloc_sem);
427 eraseit_lock:
428 /* If we've finished this block, start it erasing */
429 spin_lock(&c->erase_completion_lock);
431 eraseit:
432 if (c->gcblock && !c->gcblock->used_size) {
433 D1(printk(KERN_DEBUG "Block at 0x%08x completely obsoleted by GC. Moving to erase_pending_list\n", c->gcblock->offset));
434 /* We're GC'ing an empty block? */
435 list_add_tail(&c->gcblock->list, &c->erase_pending_list);
436 c->gcblock = NULL;
437 c->nr_erasing_blocks++;
438 jffs2_erase_pending_trigger(c);
440 spin_unlock(&c->erase_completion_lock);
442 return ret;
445 static int jffs2_garbage_collect_live(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
446 struct jffs2_raw_node_ref *raw, struct jffs2_inode_info *f)
448 struct jffs2_node_frag *frag;
449 struct jffs2_full_dnode *fn = NULL;
450 struct jffs2_full_dirent *fd;
451 uint32_t start = 0, end = 0, nrfrags = 0;
452 int ret = 0;
454 mutex_lock(&f->sem);
456 /* Now we have the lock for this inode. Check that it's still the one at the head
457 of the list. */
459 spin_lock(&c->erase_completion_lock);
461 if (c->gcblock != jeb) {
462 spin_unlock(&c->erase_completion_lock);
463 D1(printk(KERN_DEBUG "GC block is no longer gcblock. Restart\n"));
464 goto upnout;
466 if (ref_obsolete(raw)) {
467 spin_unlock(&c->erase_completion_lock);
468 D1(printk(KERN_DEBUG "node to be GC'd was obsoleted in the meantime.\n"));
469 /* They'll call again */
470 goto upnout;
472 spin_unlock(&c->erase_completion_lock);
474 /* OK. Looks safe. And nobody can get us now because we have the semaphore. Move the block */
475 if (f->metadata && f->metadata->raw == raw) {
476 fn = f->metadata;
477 ret = jffs2_garbage_collect_metadata(c, jeb, f, fn);
478 goto upnout;
481 /* FIXME. Read node and do lookup? */
482 for (frag = frag_first(&f->fragtree); frag; frag = frag_next(frag)) {
483 if (frag->node && frag->node->raw == raw) {
484 fn = frag->node;
485 end = frag->ofs + frag->size;
486 if (!nrfrags++)
487 start = frag->ofs;
488 if (nrfrags == frag->node->frags)
489 break; /* We've found them all */
492 if (fn) {
493 if (ref_flags(raw) == REF_PRISTINE) {
494 ret = jffs2_garbage_collect_pristine(c, f->inocache, raw);
495 if (!ret) {
496 /* Urgh. Return it sensibly. */
497 frag->node->raw = f->inocache->nodes;
499 if (ret != -EBADFD)
500 goto upnout;
502 /* We found a datanode. Do the GC */
503 if((start >> PAGE_CACHE_SHIFT) < ((end-1) >> PAGE_CACHE_SHIFT)) {
504 /* It crosses a page boundary. Therefore, it must be a hole. */
505 ret = jffs2_garbage_collect_hole(c, jeb, f, fn, start, end);
506 } else {
507 /* It could still be a hole. But we GC the page this way anyway */
508 ret = jffs2_garbage_collect_dnode(c, jeb, f, fn, start, end);
510 goto upnout;
513 /* Wasn't a dnode. Try dirent */
514 for (fd = f->dents; fd; fd=fd->next) {
515 if (fd->raw == raw)
516 break;
519 if (fd && fd->ino) {
520 ret = jffs2_garbage_collect_dirent(c, jeb, f, fd);
521 } else if (fd) {
522 ret = jffs2_garbage_collect_deletion_dirent(c, jeb, f, fd);
523 } else {
524 printk(KERN_WARNING "Raw node at 0x%08x wasn't in node lists for ino #%u\n",
525 ref_offset(raw), f->inocache->ino);
526 if (ref_obsolete(raw)) {
527 printk(KERN_WARNING "But it's obsolete so we don't mind too much\n");
528 } else {
529 jffs2_dbg_dump_node(c, ref_offset(raw));
530 BUG();
533 upnout:
534 mutex_unlock(&f->sem);
536 return ret;
539 static int jffs2_garbage_collect_pristine(struct jffs2_sb_info *c,
540 struct jffs2_inode_cache *ic,
541 struct jffs2_raw_node_ref *raw)
543 union jffs2_node_union *node;
544 size_t retlen;
545 int ret;
546 uint32_t phys_ofs, alloclen;
547 uint32_t crc, rawlen;
548 int retried = 0;
550 D1(printk(KERN_DEBUG "Going to GC REF_PRISTINE node at 0x%08x\n", ref_offset(raw)));
552 alloclen = rawlen = ref_totlen(c, c->gcblock, raw);
554 /* Ask for a small amount of space (or the totlen if smaller) because we
555 don't want to force wastage of the end of a block if splitting would
556 work. */
557 if (ic && alloclen > sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN)
558 alloclen = sizeof(struct jffs2_raw_inode) + JFFS2_MIN_DATA_LEN;
560 ret = jffs2_reserve_space_gc(c, alloclen, &alloclen, rawlen);
561 /* 'rawlen' is not the exact summary size; it is only an upper estimation */
563 if (ret)
564 return ret;
566 if (alloclen < rawlen) {
567 /* Doesn't fit untouched. We'll go the old route and split it */
568 return -EBADFD;
571 node = kmalloc(rawlen, GFP_KERNEL);
572 if (!node)
573 return -ENOMEM;
575 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)node);
576 if (!ret && retlen != rawlen)
577 ret = -EIO;
578 if (ret)
579 goto out_node;
581 crc = crc32(0, node, sizeof(struct jffs2_unknown_node)-4);
582 if (je32_to_cpu(node->u.hdr_crc) != crc) {
583 printk(KERN_WARNING "Header CRC failed on REF_PRISTINE node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
584 ref_offset(raw), je32_to_cpu(node->u.hdr_crc), crc);
585 goto bail;
588 switch(je16_to_cpu(node->u.nodetype)) {
589 case JFFS2_NODETYPE_INODE:
590 crc = crc32(0, node, sizeof(node->i)-8);
591 if (je32_to_cpu(node->i.node_crc) != crc) {
592 printk(KERN_WARNING "Node CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
593 ref_offset(raw), je32_to_cpu(node->i.node_crc), crc);
594 goto bail;
597 if (je32_to_cpu(node->i.dsize)) {
598 crc = crc32(0, node->i.data, je32_to_cpu(node->i.csize));
599 if (je32_to_cpu(node->i.data_crc) != crc) {
600 printk(KERN_WARNING "Data CRC failed on REF_PRISTINE data node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
601 ref_offset(raw), je32_to_cpu(node->i.data_crc), crc);
602 goto bail;
605 break;
607 case JFFS2_NODETYPE_DIRENT:
608 crc = crc32(0, node, sizeof(node->d)-8);
609 if (je32_to_cpu(node->d.node_crc) != crc) {
610 printk(KERN_WARNING "Node CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
611 ref_offset(raw), je32_to_cpu(node->d.node_crc), crc);
612 goto bail;
615 if (strnlen(node->d.name, node->d.nsize) != node->d.nsize) {
616 printk(KERN_WARNING "Name in dirent node at 0x%08x contains zeroes\n", ref_offset(raw));
617 goto bail;
620 if (node->d.nsize) {
621 crc = crc32(0, node->d.name, node->d.nsize);
622 if (je32_to_cpu(node->d.name_crc) != crc) {
623 printk(KERN_WARNING "Name CRC failed on REF_PRISTINE dirent node at 0x%08x: Read 0x%08x, calculated 0x%08x\n",
624 ref_offset(raw), je32_to_cpu(node->d.name_crc), crc);
625 goto bail;
628 break;
629 default:
630 /* If it's inode-less, we don't _know_ what it is. Just copy it intact */
631 if (ic) {
632 printk(KERN_WARNING "Unknown node type for REF_PRISTINE node at 0x%08x: 0x%04x\n",
633 ref_offset(raw), je16_to_cpu(node->u.nodetype));
634 goto bail;
638 /* OK, all the CRCs are good; this node can just be copied as-is. */
639 retry:
640 phys_ofs = write_ofs(c);
642 ret = jffs2_flash_write(c, phys_ofs, rawlen, &retlen, (char *)node);
644 if (ret || (retlen != rawlen)) {
645 printk(KERN_NOTICE "Write of %d bytes at 0x%08x failed. returned %d, retlen %zd\n",
646 rawlen, phys_ofs, ret, retlen);
647 if (retlen) {
648 jffs2_add_physical_node_ref(c, phys_ofs | REF_OBSOLETE, rawlen, NULL);
649 } else {
650 printk(KERN_NOTICE "Not marking the space at 0x%08x as dirty because the flash driver returned retlen zero\n", phys_ofs);
652 if (!retried) {
653 /* Try to reallocate space and retry */
654 uint32_t dummy;
655 struct jffs2_eraseblock *jeb = &c->blocks[phys_ofs / c->sector_size];
657 retried = 1;
659 D1(printk(KERN_DEBUG "Retrying failed write of REF_PRISTINE node.\n"));
661 jffs2_dbg_acct_sanity_check(c,jeb);
662 jffs2_dbg_acct_paranoia_check(c, jeb);
664 ret = jffs2_reserve_space_gc(c, rawlen, &dummy, rawlen);
665 /* this is not the exact summary size of it,
666 it is only an upper estimation */
668 if (!ret) {
669 D1(printk(KERN_DEBUG "Allocated space at 0x%08x to retry failed write.\n", phys_ofs));
671 jffs2_dbg_acct_sanity_check(c,jeb);
672 jffs2_dbg_acct_paranoia_check(c, jeb);
674 goto retry;
676 D1(printk(KERN_DEBUG "Failed to allocate space to retry failed write: %d!\n", ret));
679 if (!ret)
680 ret = -EIO;
681 goto out_node;
683 jffs2_add_physical_node_ref(c, phys_ofs | REF_PRISTINE, rawlen, ic);
685 jffs2_mark_node_obsolete(c, raw);
686 D1(printk(KERN_DEBUG "WHEEE! GC REF_PRISTINE node at 0x%08x succeeded\n", ref_offset(raw)));
688 out_node:
689 kfree(node);
690 return ret;
691 bail:
692 ret = -EBADFD;
693 goto out_node;
696 static int jffs2_garbage_collect_metadata(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
697 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn)
699 struct jffs2_full_dnode *new_fn;
700 struct jffs2_raw_inode ri;
701 struct jffs2_node_frag *last_frag;
702 union jffs2_device_node dev;
703 char *mdata = NULL;
704 int mdatalen = 0;
705 uint32_t alloclen, ilen;
706 int ret;
708 if (S_ISBLK(JFFS2_F_I_MODE(f)) ||
709 S_ISCHR(JFFS2_F_I_MODE(f)) ) {
710 /* For these, we don't actually need to read the old node */
711 mdatalen = jffs2_encode_dev(&dev, JFFS2_F_I_RDEV(f));
712 mdata = (char *)&dev;
713 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bytes of kdev_t\n", mdatalen));
714 } else if (S_ISLNK(JFFS2_F_I_MODE(f))) {
715 mdatalen = fn->size;
716 mdata = kmalloc(fn->size, GFP_KERNEL);
717 if (!mdata) {
718 printk(KERN_WARNING "kmalloc of mdata failed in jffs2_garbage_collect_metadata()\n");
719 return -ENOMEM;
721 ret = jffs2_read_dnode(c, f, fn, mdata, 0, mdatalen);
722 if (ret) {
723 printk(KERN_WARNING "read of old metadata failed in jffs2_garbage_collect_metadata(): %d\n", ret);
724 kfree(mdata);
725 return ret;
727 D1(printk(KERN_DEBUG "jffs2_garbage_collect_metadata(): Writing %d bites of symlink target\n", mdatalen));
731 ret = jffs2_reserve_space_gc(c, sizeof(ri) + mdatalen, &alloclen,
732 JFFS2_SUMMARY_INODE_SIZE);
733 if (ret) {
734 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_metadata failed: %d\n",
735 sizeof(ri)+ mdatalen, ret);
736 goto out;
739 last_frag = frag_last(&f->fragtree);
740 if (last_frag)
741 /* Fetch the inode length from the fragtree rather then
742 * from i_size since i_size may have not been updated yet */
743 ilen = last_frag->ofs + last_frag->size;
744 else
745 ilen = JFFS2_F_I_SIZE(f);
747 memset(&ri, 0, sizeof(ri));
748 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
749 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
750 ri.totlen = cpu_to_je32(sizeof(ri) + mdatalen);
751 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
753 ri.ino = cpu_to_je32(f->inocache->ino);
754 ri.version = cpu_to_je32(++f->highest_version);
755 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
756 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
757 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
758 ri.isize = cpu_to_je32(ilen);
759 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
760 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
761 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
762 ri.offset = cpu_to_je32(0);
763 ri.csize = cpu_to_je32(mdatalen);
764 ri.dsize = cpu_to_je32(mdatalen);
765 ri.compr = JFFS2_COMPR_NONE;
766 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
767 ri.data_crc = cpu_to_je32(crc32(0, mdata, mdatalen));
769 new_fn = jffs2_write_dnode(c, f, &ri, mdata, mdatalen, ALLOC_GC);
771 if (IS_ERR(new_fn)) {
772 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn));
773 ret = PTR_ERR(new_fn);
774 goto out;
776 jffs2_mark_node_obsolete(c, fn->raw);
777 jffs2_free_full_dnode(fn);
778 f->metadata = new_fn;
779 out:
780 if (S_ISLNK(JFFS2_F_I_MODE(f)))
781 kfree(mdata);
782 return ret;
785 static int jffs2_garbage_collect_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
786 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
788 struct jffs2_full_dirent *new_fd;
789 struct jffs2_raw_dirent rd;
790 uint32_t alloclen;
791 int ret;
793 rd.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
794 rd.nodetype = cpu_to_je16(JFFS2_NODETYPE_DIRENT);
795 rd.nsize = strlen(fd->name);
796 rd.totlen = cpu_to_je32(sizeof(rd) + rd.nsize);
797 rd.hdr_crc = cpu_to_je32(crc32(0, &rd, sizeof(struct jffs2_unknown_node)-4));
799 rd.pino = cpu_to_je32(f->inocache->ino);
800 rd.version = cpu_to_je32(++f->highest_version);
801 rd.ino = cpu_to_je32(fd->ino);
802 /* If the times on this inode were set by explicit utime() they can be different,
803 so refrain from splatting them. */
804 if (JFFS2_F_I_MTIME(f) == JFFS2_F_I_CTIME(f))
805 rd.mctime = cpu_to_je32(JFFS2_F_I_MTIME(f));
806 else
807 rd.mctime = cpu_to_je32(0);
808 rd.type = fd->type;
809 rd.node_crc = cpu_to_je32(crc32(0, &rd, sizeof(rd)-8));
810 rd.name_crc = cpu_to_je32(crc32(0, fd->name, rd.nsize));
812 ret = jffs2_reserve_space_gc(c, sizeof(rd)+rd.nsize, &alloclen,
813 JFFS2_SUMMARY_DIRENT_SIZE(rd.nsize));
814 if (ret) {
815 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dirent failed: %d\n",
816 sizeof(rd)+rd.nsize, ret);
817 return ret;
819 new_fd = jffs2_write_dirent(c, f, &rd, fd->name, rd.nsize, ALLOC_GC);
821 if (IS_ERR(new_fd)) {
822 printk(KERN_WARNING "jffs2_write_dirent in garbage_collect_dirent failed: %ld\n", PTR_ERR(new_fd));
823 return PTR_ERR(new_fd);
825 jffs2_add_fd_to_list(c, new_fd, &f->dents);
826 return 0;
829 static int jffs2_garbage_collect_deletion_dirent(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
830 struct jffs2_inode_info *f, struct jffs2_full_dirent *fd)
832 struct jffs2_full_dirent **fdp = &f->dents;
833 int found = 0;
835 /* On a medium where we can't actually mark nodes obsolete
836 pernamently, such as NAND flash, we need to work out
837 whether this deletion dirent is still needed to actively
838 delete a 'real' dirent with the same name that's still
839 somewhere else on the flash. */
840 if (!jffs2_can_mark_obsolete(c)) {
841 struct jffs2_raw_dirent *rd;
842 struct jffs2_raw_node_ref *raw;
843 int ret;
844 size_t retlen;
845 int name_len = strlen(fd->name);
846 uint32_t name_crc = crc32(0, fd->name, name_len);
847 uint32_t rawlen = ref_totlen(c, jeb, fd->raw);
849 rd = kmalloc(rawlen, GFP_KERNEL);
850 if (!rd)
851 return -ENOMEM;
853 /* Prevent the erase code from nicking the obsolete node refs while
854 we're looking at them. I really don't like this extra lock but
855 can't see any alternative. Suggestions on a postcard to... */
856 mutex_lock(&c->erase_free_sem);
858 for (raw = f->inocache->nodes; raw != (void *)f->inocache; raw = raw->next_in_ino) {
860 cond_resched();
862 /* We only care about obsolete ones */
863 if (!(ref_obsolete(raw)))
864 continue;
866 /* Any dirent with the same name is going to have the same length... */
867 if (ref_totlen(c, NULL, raw) != rawlen)
868 continue;
870 /* Doesn't matter if there's one in the same erase block. We're going to
871 delete it too at the same time. */
872 if (SECTOR_ADDR(raw->flash_offset) == SECTOR_ADDR(fd->raw->flash_offset))
873 continue;
875 D1(printk(KERN_DEBUG "Check potential deletion dirent at %08x\n", ref_offset(raw)));
877 /* This is an obsolete node belonging to the same directory, and it's of the right
878 length. We need to take a closer look...*/
879 ret = jffs2_flash_read(c, ref_offset(raw), rawlen, &retlen, (char *)rd);
880 if (ret) {
881 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Read error (%d) reading obsolete node at %08x\n", ret, ref_offset(raw));
882 /* If we can't read it, we don't need to continue to obsolete it. Continue */
883 continue;
885 if (retlen != rawlen) {
886 printk(KERN_WARNING "jffs2_g_c_deletion_dirent(): Short read (%zd not %u) reading header from obsolete node at %08x\n",
887 retlen, rawlen, ref_offset(raw));
888 continue;
891 if (je16_to_cpu(rd->nodetype) != JFFS2_NODETYPE_DIRENT)
892 continue;
894 /* If the name CRC doesn't match, skip */
895 if (je32_to_cpu(rd->name_crc) != name_crc)
896 continue;
898 /* If the name length doesn't match, or it's another deletion dirent, skip */
899 if (rd->nsize != name_len || !je32_to_cpu(rd->ino))
900 continue;
902 /* OK, check the actual name now */
903 if (memcmp(rd->name, fd->name, name_len))
904 continue;
906 /* OK. The name really does match. There really is still an older node on
907 the flash which our deletion dirent obsoletes. So we have to write out
908 a new deletion dirent to replace it */
909 mutex_unlock(&c->erase_free_sem);
911 D1(printk(KERN_DEBUG "Deletion dirent at %08x still obsoletes real dirent \"%s\" at %08x for ino #%u\n",
912 ref_offset(fd->raw), fd->name, ref_offset(raw), je32_to_cpu(rd->ino)));
913 kfree(rd);
915 return jffs2_garbage_collect_dirent(c, jeb, f, fd);
918 mutex_unlock(&c->erase_free_sem);
919 kfree(rd);
922 /* FIXME: If we're deleting a dirent which contains the current mtime and ctime,
923 we should update the metadata node with those times accordingly */
925 /* No need for it any more. Just mark it obsolete and remove it from the list */
926 while (*fdp) {
927 if ((*fdp) == fd) {
928 found = 1;
929 *fdp = fd->next;
930 break;
932 fdp = &(*fdp)->next;
934 if (!found) {
935 printk(KERN_WARNING "Deletion dirent \"%s\" not found in list for ino #%u\n", fd->name, f->inocache->ino);
937 jffs2_mark_node_obsolete(c, fd->raw);
938 jffs2_free_full_dirent(fd);
939 return 0;
942 static int jffs2_garbage_collect_hole(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb,
943 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
944 uint32_t start, uint32_t end)
946 struct jffs2_raw_inode ri;
947 struct jffs2_node_frag *frag;
948 struct jffs2_full_dnode *new_fn;
949 uint32_t alloclen, ilen;
950 int ret;
952 D1(printk(KERN_DEBUG "Writing replacement hole node for ino #%u from offset 0x%x to 0x%x\n",
953 f->inocache->ino, start, end));
955 memset(&ri, 0, sizeof(ri));
957 if(fn->frags > 1) {
958 size_t readlen;
959 uint32_t crc;
960 /* It's partially obsoleted by a later write. So we have to
961 write it out again with the _same_ version as before */
962 ret = jffs2_flash_read(c, ref_offset(fn->raw), sizeof(ri), &readlen, (char *)&ri);
963 if (readlen != sizeof(ri) || ret) {
964 printk(KERN_WARNING "Node read failed in jffs2_garbage_collect_hole. Ret %d, retlen %zd. Data will be lost by writing new hole node\n", ret, readlen);
965 goto fill;
967 if (je16_to_cpu(ri.nodetype) != JFFS2_NODETYPE_INODE) {
968 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had node type 0x%04x instead of JFFS2_NODETYPE_INODE(0x%04x)\n",
969 ref_offset(fn->raw),
970 je16_to_cpu(ri.nodetype), JFFS2_NODETYPE_INODE);
971 return -EIO;
973 if (je32_to_cpu(ri.totlen) != sizeof(ri)) {
974 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had totlen 0x%x instead of expected 0x%zx\n",
975 ref_offset(fn->raw),
976 je32_to_cpu(ri.totlen), sizeof(ri));
977 return -EIO;
979 crc = crc32(0, &ri, sizeof(ri)-8);
980 if (crc != je32_to_cpu(ri.node_crc)) {
981 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node at 0x%08x had CRC 0x%08x which doesn't match calculated CRC 0x%08x\n",
982 ref_offset(fn->raw),
983 je32_to_cpu(ri.node_crc), crc);
984 /* FIXME: We could possibly deal with this by writing new holes for each frag */
985 printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
986 start, end, f->inocache->ino);
987 goto fill;
989 if (ri.compr != JFFS2_COMPR_ZERO) {
990 printk(KERN_WARNING "jffs2_garbage_collect_hole: Node 0x%08x wasn't a hole node!\n", ref_offset(fn->raw));
991 printk(KERN_WARNING "Data in the range 0x%08x to 0x%08x of inode #%u will be lost\n",
992 start, end, f->inocache->ino);
993 goto fill;
995 } else {
996 fill:
997 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
998 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
999 ri.totlen = cpu_to_je32(sizeof(ri));
1000 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
1002 ri.ino = cpu_to_je32(f->inocache->ino);
1003 ri.version = cpu_to_je32(++f->highest_version);
1004 ri.offset = cpu_to_je32(start);
1005 ri.dsize = cpu_to_je32(end - start);
1006 ri.csize = cpu_to_je32(0);
1007 ri.compr = JFFS2_COMPR_ZERO;
1010 frag = frag_last(&f->fragtree);
1011 if (frag)
1012 /* Fetch the inode length from the fragtree rather then
1013 * from i_size since i_size may have not been updated yet */
1014 ilen = frag->ofs + frag->size;
1015 else
1016 ilen = JFFS2_F_I_SIZE(f);
1018 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
1019 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
1020 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
1021 ri.isize = cpu_to_je32(ilen);
1022 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
1023 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
1024 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
1025 ri.data_crc = cpu_to_je32(0);
1026 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
1028 ret = jffs2_reserve_space_gc(c, sizeof(ri), &alloclen,
1029 JFFS2_SUMMARY_INODE_SIZE);
1030 if (ret) {
1031 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_hole failed: %d\n",
1032 sizeof(ri), ret);
1033 return ret;
1035 new_fn = jffs2_write_dnode(c, f, &ri, NULL, 0, ALLOC_GC);
1037 if (IS_ERR(new_fn)) {
1038 printk(KERN_WARNING "Error writing new hole node: %ld\n", PTR_ERR(new_fn));
1039 return PTR_ERR(new_fn);
1041 if (je32_to_cpu(ri.version) == f->highest_version) {
1042 jffs2_add_full_dnode_to_inode(c, f, new_fn);
1043 if (f->metadata) {
1044 jffs2_mark_node_obsolete(c, f->metadata->raw);
1045 jffs2_free_full_dnode(f->metadata);
1046 f->metadata = NULL;
1048 return 0;
1052 * We should only get here in the case where the node we are
1053 * replacing had more than one frag, so we kept the same version
1054 * number as before. (Except in case of error -- see 'goto fill;'
1055 * above.)
1057 D1(if(unlikely(fn->frags <= 1)) {
1058 printk(KERN_WARNING "jffs2_garbage_collect_hole: Replacing fn with %d frag(s) but new ver %d != highest_version %d of ino #%d\n",
1059 fn->frags, je32_to_cpu(ri.version), f->highest_version,
1060 je32_to_cpu(ri.ino));
1063 /* This is a partially-overlapped hole node. Mark it REF_NORMAL not REF_PRISTINE */
1064 mark_ref_normal(new_fn->raw);
1066 for (frag = jffs2_lookup_node_frag(&f->fragtree, fn->ofs);
1067 frag; frag = frag_next(frag)) {
1068 if (frag->ofs > fn->size + fn->ofs)
1069 break;
1070 if (frag->node == fn) {
1071 frag->node = new_fn;
1072 new_fn->frags++;
1073 fn->frags--;
1076 if (fn->frags) {
1077 printk(KERN_WARNING "jffs2_garbage_collect_hole: Old node still has frags!\n");
1078 BUG();
1080 if (!new_fn->frags) {
1081 printk(KERN_WARNING "jffs2_garbage_collect_hole: New node has no frags!\n");
1082 BUG();
1085 jffs2_mark_node_obsolete(c, fn->raw);
1086 jffs2_free_full_dnode(fn);
1088 return 0;
1091 static int jffs2_garbage_collect_dnode(struct jffs2_sb_info *c, struct jffs2_eraseblock *orig_jeb,
1092 struct jffs2_inode_info *f, struct jffs2_full_dnode *fn,
1093 uint32_t start, uint32_t end)
1095 struct jffs2_full_dnode *new_fn;
1096 struct jffs2_raw_inode ri;
1097 uint32_t alloclen, offset, orig_end, orig_start;
1098 int ret = 0;
1099 unsigned char *comprbuf = NULL, *writebuf;
1100 unsigned long pg;
1101 unsigned char *pg_ptr;
1103 memset(&ri, 0, sizeof(ri));
1105 D1(printk(KERN_DEBUG "Writing replacement dnode for ino #%u from offset 0x%x to 0x%x\n",
1106 f->inocache->ino, start, end));
1108 orig_end = end;
1109 orig_start = start;
1111 if (c->nr_free_blocks + c->nr_erasing_blocks > c->resv_blocks_gcmerge) {
1112 /* Attempt to do some merging. But only expand to cover logically
1113 adjacent frags if the block containing them is already considered
1114 to be dirty. Otherwise we end up with GC just going round in
1115 circles dirtying the nodes it already wrote out, especially
1116 on NAND where we have small eraseblocks and hence a much higher
1117 chance of nodes having to be split to cross boundaries. */
1119 struct jffs2_node_frag *frag;
1120 uint32_t min, max;
1122 min = start & ~(PAGE_CACHE_SIZE-1);
1123 max = min + PAGE_CACHE_SIZE;
1125 frag = jffs2_lookup_node_frag(&f->fragtree, start);
1127 /* BUG_ON(!frag) but that'll happen anyway... */
1129 BUG_ON(frag->ofs != start);
1131 /* First grow down... */
1132 while((frag = frag_prev(frag)) && frag->ofs >= min) {
1134 /* If the previous frag doesn't even reach the beginning, there's
1135 excessive fragmentation. Just merge. */
1136 if (frag->ofs > min) {
1137 D1(printk(KERN_DEBUG "Expanding down to cover partial frag (0x%x-0x%x)\n",
1138 frag->ofs, frag->ofs+frag->size));
1139 start = frag->ofs;
1140 continue;
1142 /* OK. This frag holds the first byte of the page. */
1143 if (!frag->node || !frag->node->raw) {
1144 D1(printk(KERN_DEBUG "First frag in page is hole (0x%x-0x%x). Not expanding down.\n",
1145 frag->ofs, frag->ofs+frag->size));
1146 break;
1147 } else {
1149 /* OK, it's a frag which extends to the beginning of the page. Does it live
1150 in a block which is still considered clean? If so, don't obsolete it.
1151 If not, cover it anyway. */
1153 struct jffs2_raw_node_ref *raw = frag->node->raw;
1154 struct jffs2_eraseblock *jeb;
1156 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1158 if (jeb == c->gcblock) {
1159 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1160 frag->ofs, frag->ofs+frag->size, ref_offset(raw)));
1161 start = frag->ofs;
1162 break;
1164 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1165 D1(printk(KERN_DEBUG "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n",
1166 frag->ofs, frag->ofs+frag->size, jeb->offset));
1167 break;
1170 D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n",
1171 frag->ofs, frag->ofs+frag->size, jeb->offset));
1172 start = frag->ofs;
1173 break;
1177 /* ... then up */
1179 /* Find last frag which is actually part of the node we're to GC. */
1180 frag = jffs2_lookup_node_frag(&f->fragtree, end-1);
1182 while((frag = frag_next(frag)) && frag->ofs+frag->size <= max) {
1184 /* If the previous frag doesn't even reach the beginning, there's lots
1185 of fragmentation. Just merge. */
1186 if (frag->ofs+frag->size < max) {
1187 D1(printk(KERN_DEBUG "Expanding up to cover partial frag (0x%x-0x%x)\n",
1188 frag->ofs, frag->ofs+frag->size));
1189 end = frag->ofs + frag->size;
1190 continue;
1193 if (!frag->node || !frag->node->raw) {
1194 D1(printk(KERN_DEBUG "Last frag in page is hole (0x%x-0x%x). Not expanding up.\n",
1195 frag->ofs, frag->ofs+frag->size));
1196 break;
1197 } else {
1199 /* OK, it's a frag which extends to the beginning of the page. Does it live
1200 in a block which is still considered clean? If so, don't obsolete it.
1201 If not, cover it anyway. */
1203 struct jffs2_raw_node_ref *raw = frag->node->raw;
1204 struct jffs2_eraseblock *jeb;
1206 jeb = &c->blocks[raw->flash_offset / c->sector_size];
1208 if (jeb == c->gcblock) {
1209 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in gcblock at %08x\n",
1210 frag->ofs, frag->ofs+frag->size, ref_offset(raw)));
1211 end = frag->ofs + frag->size;
1212 break;
1214 if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {
1215 D1(printk(KERN_DEBUG "Not expanding up to cover frag (0x%x-0x%x) in clean block %08x\n",
1216 frag->ofs, frag->ofs+frag->size, jeb->offset));
1217 break;
1220 D1(printk(KERN_DEBUG "Expanding up to cover frag (0x%x-0x%x) in dirty block %08x\n",
1221 frag->ofs, frag->ofs+frag->size, jeb->offset));
1222 end = frag->ofs + frag->size;
1223 break;
1226 D1(printk(KERN_DEBUG "Expanded dnode to write from (0x%x-0x%x) to (0x%x-0x%x)\n",
1227 orig_start, orig_end, start, end));
1229 D1(BUG_ON(end > frag_last(&f->fragtree)->ofs + frag_last(&f->fragtree)->size));
1230 BUG_ON(end < orig_end);
1231 BUG_ON(start > orig_start);
1234 /* First, use readpage() to read the appropriate page into the page cache */
1235 /* Q: What happens if we actually try to GC the _same_ page for which commit_write()
1236 * triggered garbage collection in the first place?
1237 * A: I _think_ it's OK. read_cache_page shouldn't deadlock, we'll write out the
1238 * page OK. We'll actually write it out again in commit_write, which is a little
1239 * suboptimal, but at least we're correct.
1241 pg_ptr = jffs2_gc_fetch_page(c, f, start, &pg);
1243 if (IS_ERR(pg_ptr)) {
1244 printk(KERN_WARNING "read_cache_page() returned error: %ld\n", PTR_ERR(pg_ptr));
1245 return PTR_ERR(pg_ptr);
1248 offset = start;
1249 while(offset < orig_end) {
1250 uint32_t datalen;
1251 uint32_t cdatalen;
1252 uint16_t comprtype = JFFS2_COMPR_NONE;
1254 ret = jffs2_reserve_space_gc(c, sizeof(ri) + JFFS2_MIN_DATA_LEN,
1255 &alloclen, JFFS2_SUMMARY_INODE_SIZE);
1257 if (ret) {
1258 printk(KERN_WARNING "jffs2_reserve_space_gc of %zd bytes for garbage_collect_dnode failed: %d\n",
1259 sizeof(ri)+ JFFS2_MIN_DATA_LEN, ret);
1260 break;
1262 cdatalen = min_t(uint32_t, alloclen - sizeof(ri), end - offset);
1263 datalen = end - offset;
1265 writebuf = pg_ptr + (offset & (PAGE_CACHE_SIZE -1));
1267 comprtype = jffs2_compress(c, f, writebuf, &comprbuf, &datalen, &cdatalen);
1269 ri.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1270 ri.nodetype = cpu_to_je16(JFFS2_NODETYPE_INODE);
1271 ri.totlen = cpu_to_je32(sizeof(ri) + cdatalen);
1272 ri.hdr_crc = cpu_to_je32(crc32(0, &ri, sizeof(struct jffs2_unknown_node)-4));
1274 ri.ino = cpu_to_je32(f->inocache->ino);
1275 ri.version = cpu_to_je32(++f->highest_version);
1276 ri.mode = cpu_to_jemode(JFFS2_F_I_MODE(f));
1277 ri.uid = cpu_to_je16(JFFS2_F_I_UID(f));
1278 ri.gid = cpu_to_je16(JFFS2_F_I_GID(f));
1279 ri.isize = cpu_to_je32(JFFS2_F_I_SIZE(f));
1280 ri.atime = cpu_to_je32(JFFS2_F_I_ATIME(f));
1281 ri.ctime = cpu_to_je32(JFFS2_F_I_CTIME(f));
1282 ri.mtime = cpu_to_je32(JFFS2_F_I_MTIME(f));
1283 ri.offset = cpu_to_je32(offset);
1284 ri.csize = cpu_to_je32(cdatalen);
1285 ri.dsize = cpu_to_je32(datalen);
1286 ri.compr = comprtype & 0xff;
1287 ri.usercompr = (comprtype >> 8) & 0xff;
1288 ri.node_crc = cpu_to_je32(crc32(0, &ri, sizeof(ri)-8));
1289 ri.data_crc = cpu_to_je32(crc32(0, comprbuf, cdatalen));
1291 new_fn = jffs2_write_dnode(c, f, &ri, comprbuf, cdatalen, ALLOC_GC);
1293 jffs2_free_comprbuf(comprbuf, writebuf);
1295 if (IS_ERR(new_fn)) {
1296 printk(KERN_WARNING "Error writing new dnode: %ld\n", PTR_ERR(new_fn));
1297 ret = PTR_ERR(new_fn);
1298 break;
1300 ret = jffs2_add_full_dnode_to_inode(c, f, new_fn);
1301 offset += datalen;
1302 if (f->metadata) {
1303 jffs2_mark_node_obsolete(c, f->metadata->raw);
1304 jffs2_free_full_dnode(f->metadata);
1305 f->metadata = NULL;
1309 jffs2_gc_release_page(c, pg_ptr, &pg);
1310 return ret;