fs/jffs2/nodemgmt.c

   1 /*
   2  * JFFS2 -- Journalling Flash File System, Version 2.
   3  *
   4  * Copyright (C) 2001-2003 Red Hat, Inc.
   5  *
   6  * Created by David Woodhouse <dwmw2@infradead.org>
   7  *
   8  * For licensing information, see the file 'LICENCE' in this directory.
   9  *
  10  * $Id: nodemgmt.c,v 1.122 2005/05/06 09:30:27 dedekind Exp $
  11  *
  12  */
  13
  14 #include <linux/kernel.h>
  15 #include <linux/slab.h>
  16 #include <linux/mtd/mtd.h>
  17 #include <linux/compiler.h>
  18 #include <linux/sched.h> /* For cond_resched() */
  19 #include "nodelist.h"
  20
  21 /**
  22  *      jffs2_reserve_space - request physical space to write nodes to flash
  23  *      @c: superblock info
  24  *      @minsize: Minimum acceptable size of allocation
  25  *      @ofs: Returned value of node offset
  26  *      @len: Returned value of allocation length
  27  *      @prio: Allocation type - ALLOC_{NORMAL,DELETION}
  28  *
  29  *      Requests a block of physical space on the flash. Returns zero for success
  30  *      and puts 'ofs' and 'len' into the appriopriate place, or returns -ENOSPC
  31  *      or other error if appropriate.
  32  *
  33  *      If it returns zero, jffs2_reserve_space() also downs the per-filesystem
  34  *      allocation semaphore, to prevent more than one allocation from being
  35  *      active at any time. The semaphore is later released by jffs2_commit_allocation()
  36  *
  37  *      jffs2_reserve_space() may trigger garbage collection in order to make room
  38  *      for the requested allocation.
  39  */
  40
  41 static int jffs2_do_reserve_space(struct jffs2_sb_info *c,  uint32_t minsize, uint32_t *ofs, uint32_t *len);
  42
  43 int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize, uint32_t *ofs, uint32_t *len, int prio)
  44 {
  45         int ret = -EAGAIN;
  46         int blocksneeded = c->resv_blocks_write;
  47         /* align it */
  48         minsize = PAD(minsize);
  49
  50         D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
  51         down(&c->alloc_sem);
  52
  53         D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
  54
  55         spin_lock(&c->erase_completion_lock);
  56
  57         /* this needs a little more thought (true <tglx> :)) */
  58         while(ret == -EAGAIN) {
  59                 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
  60                         int ret;
  61                         uint32_t dirty, avail;
  62
  63                         /* calculate real dirty size
  64                          * dirty_size contains blocks on erase_pending_list
  65                          * those blocks are counted in c->nr_erasing_blocks.
  66                          * If one block is actually erased, it is not longer counted as dirty_space
  67                          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
  68                          * with c->nr_erasing_blocks * c->sector_size again.
  69                          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
  70                          * This helps us to force gc and pick eventually a clean block to spread the load.
  71                          * We add unchecked_size here, as we hopefully will find some space to use.
  72                          * This will affect the sum only once, as gc first finishes checking
  73                          * of nodes.
  74                          */
  75                         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
  76                         if (dirty < c->nospc_dirty_size) {
  77                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
  78                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
  79                                         break;
  80                                 }
  81                                 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
  82                                           dirty, c->unchecked_size, c->sector_size));
  83
  84                                 spin_unlock(&c->erase_completion_lock);
  85                                 up(&c->alloc_sem);
  86                                 return -ENOSPC;
  87                         }
  88
  89                         /* Calc possibly available space. Possibly available means that we
  90                          * don't know, if unchecked size contains obsoleted nodes, which could give us some
  91                          * more usable space. This will affect the sum only once, as gc first finishes checking
  92                          * of nodes.
  93                          + Return -ENOSPC, if the maximum possibly available space is less or equal than
  94                          * blocksneeded * sector_size.
  95                          * This blocks endless gc looping on a filesystem, which is nearly full, even if
  96                          * the check above passes.
  97                          */
  98                         avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
  99                         if ( (avail / c->sector_size) <= blocksneeded) {
 100                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
 101                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
 102                                         break;
 103                                 }
 104
 105                                 D1(printk(KERN_DEBUG "max. available size 0x%08x  < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
 106                                           avail, blocksneeded * c->sector_size));
 107                                 spin_unlock(&c->erase_completion_lock);
 108                                 up(&c->alloc_sem);
 109                                 return -ENOSPC;
 110                         }
 111
 112                         up(&c->alloc_sem);
 113
 114                         D1(printk(KERN_DEBUG "Triggering GC pass. nr_free_blocks %d, nr_erasing_blocks %d, free_size 0x%08x, dirty_size 0x%08x, wasted_size 0x%08x, used_size 0x%08x, erasing_size 0x%08x, bad_size 0x%08x (total 0x%08x of 0x%08x)\n",
 115                                   c->nr_free_blocks, c->nr_erasing_blocks, c->free_size, c->dirty_size, c->wasted_size, c->used_size, c->erasing_size, c->bad_size,
 116                                   c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
 117                         spin_unlock(&c->erase_completion_lock);
 118
 119                         ret = jffs2_garbage_collect_pass(c);
 120                         if (ret)
 121                                 return ret;
 122
 123                         cond_resched();
 124
 125                         if (signal_pending(current))
 126                                 return -EINTR;
 127
 128                         down(&c->alloc_sem);
 129                         spin_lock(&c->erase_completion_lock);
 130                 }
 131
 132                 ret = jffs2_do_reserve_space(c, minsize, ofs, len);
 133                 if (ret) {
 134                         D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
 135                 }
 136         }
 137         spin_unlock(&c->erase_completion_lock);
 138         if (ret)
 139                 up(&c->alloc_sem);
 140         return ret;
 141 }
 142
 143 int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize, uint32_t *ofs, uint32_t *len)
 144 {
 145         int ret = -EAGAIN;
 146         minsize = PAD(minsize);
 147
 148         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
 149
 150         spin_lock(&c->erase_completion_lock);
 151         while(ret == -EAGAIN) {
 152                 ret = jffs2_do_reserve_space(c, minsize, ofs, len);
 153                 if (ret) {
 154                         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
 155                 }
 156         }
 157         spin_unlock(&c->erase_completion_lock);
 158         return ret;
 159 }
 160
 161 /* Called with alloc sem _and_ erase_completion_lock */
 162 static int jffs2_do_reserve_space(struct jffs2_sb_info *c,  uint32_t minsize, uint32_t *ofs, uint32_t *len)
 163 {
 164         struct jffs2_eraseblock *jeb = c->nextblock;
 165
 166  restart:
 167         if (jeb && minsize > jeb->free_size) {
 168                 /* Skip the end of this block and file it as having some dirty space */
 169                 /* If there's a pending write to it, flush now */
 170                 if (jffs2_wbuf_dirty(c)) {
 171                         spin_unlock(&c->erase_completion_lock);
 172                         D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
 173                         jffs2_flush_wbuf_pad(c);
 174                         spin_lock(&c->erase_completion_lock);
 175                         jeb = c->nextblock;
 176                         goto restart;
 177                 }
 178                 c->wasted_size += jeb->free_size;
 179                 c->free_size -= jeb->free_size;
 180                 jeb->wasted_size += jeb->free_size;
 181                 jeb->free_size = 0;
 182
 183                 /* Check, if we have a dirty block now, or if it was dirty already */
 184                 if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
 185                         c->dirty_size += jeb->wasted_size;
 186                         c->wasted_size -= jeb->wasted_size;
 187                         jeb->dirty_size += jeb->wasted_size;
 188                         jeb->wasted_size = 0;
 189                         if (VERYDIRTY(c, jeb->dirty_size)) {
 190                                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 191                                   jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 192                                 list_add_tail(&jeb->list, &c->very_dirty_list);
 193                         } else {
 194                                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 195                                   jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 196                                 list_add_tail(&jeb->list, &c->dirty_list);
 197                         }
 198                 } else {
 199                         D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 200                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 201                         list_add_tail(&jeb->list, &c->clean_list);
 202                 }
 203                 c->nextblock = jeb = NULL;
 204         }
 205
 206         if (!jeb) {
 207                 struct list_head *next;
 208                 /* Take the next block off the 'free' list */
 209
 210                 if (list_empty(&c->free_list)) {
 211
 212                         if (!c->nr_erasing_blocks &&
 213                             !list_empty(&c->erasable_list)) {
 214                                 struct jffs2_eraseblock *ejeb;
 215
 216                                 ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
 217                                 list_del(&ejeb->list);
 218                                 list_add_tail(&ejeb->list, &c->erase_pending_list);
 219                                 c->nr_erasing_blocks++;
 220                                 jffs2_erase_pending_trigger(c);
 221                                 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Triggering erase of erasable block at 0x%08x\n",
 222                                           ejeb->offset));
 223                         }
 224
 225                         if (!c->nr_erasing_blocks &&
 226                             !list_empty(&c->erasable_pending_wbuf_list)) {
 227                                 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
 228                                 /* c->nextblock is NULL, no update to c->nextblock allowed */
 229                                 spin_unlock(&c->erase_completion_lock);
 230                                 jffs2_flush_wbuf_pad(c);
 231                                 spin_lock(&c->erase_completion_lock);
 232                                 /* Have another go. It'll be on the erasable_list now */
 233                                 return -EAGAIN;
 234                         }
 235
 236                         if (!c->nr_erasing_blocks) {
 237                                 /* Ouch. We're in GC, or we wouldn't have got here.
 238                                    And there's no space left. At all. */
 239                                 printk(KERN_CRIT "Argh. No free space left for GC. nr_erasing_blocks is %d. nr_free_blocks is %d. (erasableempty: %s, erasingempty: %s, erasependingempty: %s)\n",
 240                                        c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
 241                                        list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
 242                                 return -ENOSPC;
 243                         }
 244
 245                         spin_unlock(&c->erase_completion_lock);
 246                         /* Don't wait for it; just erase one right now */
 247                         jffs2_erase_pending_blocks(c, 1);
 248                         spin_lock(&c->erase_completion_lock);
 249
 250                         /* An erase may have failed, decreasing the
 251                            amount of free space available. So we must
 252                            restart from the beginning */
 253                         return -EAGAIN;
 254                 }
 255
 256                 next = c->free_list.next;
 257                 list_del(next);
 258                 c->nextblock = jeb = list_entry(next, struct jffs2_eraseblock, list);
 259                 c->nr_free_blocks--;
 260
 261                 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
 262                         printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
 263                         goto restart;
 264                 }
 265         }
 266         /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
 267            enough space */
 268         *ofs = jeb->offset + (c->sector_size - jeb->free_size);
 269         *len = jeb->free_size;
 270
 271         if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
 272             !jeb->first_node->next_in_ino) {
 273                 /* Only node in it beforehand was a CLEANMARKER node (we think).
 274                    So mark it obsolete now that there's going to be another node
 275                    in the block. This will reduce used_size to zero but We've
 276                    already set c->nextblock so that jffs2_mark_node_obsolete()
 277                    won't try to refile it to the dirty_list.
 278                 */
 279                 spin_unlock(&c->erase_completion_lock);
 280                 jffs2_mark_node_obsolete(c, jeb->first_node);
 281                 spin_lock(&c->erase_completion_lock);
 282         }
 283
 284         D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n", *len, *ofs));
 285         return 0;
 286 }
 287
 288 /**
 289  *      jffs2_add_physical_node_ref - add a physical node reference to the list
 290  *      @c: superblock info
 291  *      @new: new node reference to add
 292  *      @len: length of this physical node
 293  *      @dirty: dirty flag for new node
 294  *
 295  *      Should only be used to report nodes for which space has been allocated
 296  *      by jffs2_reserve_space.
 297  *
 298  *      Must be called with the alloc_sem held.
 299  */
 300
 301 int jffs2_add_physical_node_ref(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *new)
 302 {
 303         struct jffs2_eraseblock *jeb;
 304         uint32_t len;
 305
 306         jeb = &c->blocks[new->flash_offset / c->sector_size];
 307         len = ref_totlen(c, jeb, new);
 308
 309         D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n", ref_offset(new), ref_flags(new), len));
 310 #if 1
 311         /* we could get some obsolete nodes after nextblock was refiled
 312            in wbuf.c */
 313         if ((c->nextblock || !ref_obsolete(new))
 314             &&(jeb != c->nextblock || ref_offset(new) != jeb->offset + (c->sector_size - jeb->free_size))) {
 315                 printk(KERN_WARNING "argh. node added in wrong place\n");
 316                 jffs2_free_raw_node_ref(new);
 317                 return -EINVAL;
 318         }
 319 #endif
 320         spin_lock(&c->erase_completion_lock);
 321
 322         if (!jeb->first_node)
 323                 jeb->first_node = new;
 324         if (jeb->last_node)
 325                 jeb->last_node->next_phys = new;
 326         jeb->last_node = new;
 327
 328         jeb->free_size -= len;
 329         c->free_size -= len;
 330         if (ref_obsolete(new)) {
 331                 jeb->dirty_size += len;
 332                 c->dirty_size += len;
 333         } else {
 334                 jeb->used_size += len;
 335                 c->used_size += len;
 336         }
 337
 338         if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
 339                 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
 340                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 341                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 342                 if (jffs2_wbuf_dirty(c)) {
 343                         /* Flush the last write in the block if it's outstanding */
 344                         spin_unlock(&c->erase_completion_lock);
 345                         jffs2_flush_wbuf_pad(c);
 346                         spin_lock(&c->erase_completion_lock);
 347                 }
 348
 349                 list_add_tail(&jeb->list, &c->clean_list);
 350                 c->nextblock = NULL;
 351         }
 352         ACCT_SANITY_CHECK(c,jeb);
 353         D1(ACCT_PARANOIA_CHECK(jeb));
 354
 355         spin_unlock(&c->erase_completion_lock);
 356
 357         return 0;
 358 }
 359
 360
 361 void jffs2_complete_reservation(struct jffs2_sb_info *c)
 362 {
 363         D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
 364         jffs2_garbage_collect_trigger(c);
 365         up(&c->alloc_sem);
 366 }
 367
 368 static inline int on_list(struct list_head *obj, struct list_head *head)
 369 {
 370         struct list_head *this;
 371
 372         list_for_each(this, head) {
 373                 if (this == obj) {
 374                         D1(printk("%p is on list at %p\n", obj, head));
 375                         return 1;
 376
 377                 }
 378         }
 379         return 0;
 380 }
 381
 382 void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
 383 {
 384         struct jffs2_eraseblock *jeb;
 385         int blocknr;
 386         struct jffs2_unknown_node n;
 387         int ret, addedsize;
 388         size_t retlen;
 389
 390         if(!ref) {
 391                 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
 392                 return;
 393         }
 394         if (ref_obsolete(ref)) {
 395                 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
 396                 return;
 397         }
 398         blocknr = ref->flash_offset / c->sector_size;
 399         if (blocknr >= c->nr_blocks) {
 400                 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
 401                 BUG();
 402         }
 403         jeb = &c->blocks[blocknr];
 404
 405         if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
 406             !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
 407                 /* Hm. This may confuse static lock analysis. If any of the above
 408                    three conditions is false, we're going to return from this
 409                    function without actually obliterating any nodes or freeing
 410                    any jffs2_raw_node_refs. So we don't need to stop erases from
 411                    happening, or protect against people holding an obsolete
 412                    jffs2_raw_node_ref without the erase_completion_lock. */
 413                 down(&c->erase_free_sem);
 414         }
 415
 416         spin_lock(&c->erase_completion_lock);
 417
 418         if (ref_flags(ref) == REF_UNCHECKED) {
 419                 D1(if (unlikely(jeb->unchecked_size < ref_totlen(c, jeb, ref))) {
 420                         printk(KERN_NOTICE "raw unchecked node of size 0x%08x freed from erase block %d at 0x%08x, but unchecked_size was already 0x%08x\n",
 421                                ref_totlen(c, jeb, ref), blocknr, ref->flash_offset, jeb->used_size);
 422                         BUG();
 423                 })
 424                 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), ref_totlen(c, jeb, ref)));
 425                 jeb->unchecked_size -= ref_totlen(c, jeb, ref);
 426                 c->unchecked_size -= ref_totlen(c, jeb, ref);
 427         } else {
 428                 D1(if (unlikely(jeb->used_size < ref_totlen(c, jeb, ref))) {
 429                         printk(KERN_NOTICE "raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n",
 430                                ref_totlen(c, jeb, ref), blocknr, ref->flash_offset, jeb->used_size);
 431                         BUG();
 432                 })
 433                 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %x: ", ref_offset(ref), ref_totlen(c, jeb, ref)));
 434                 jeb->used_size -= ref_totlen(c, jeb, ref);
 435                 c->used_size -= ref_totlen(c, jeb, ref);
 436         }
 437
 438         // Take care, that wasted size is taken into concern
 439         if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + ref_totlen(c, jeb, ref))) && jeb != c->nextblock) {
 440                 D1(printk(KERN_DEBUG "Dirtying\n"));
 441                 addedsize = ref_totlen(c, jeb, ref);
 442                 jeb->dirty_size += ref_totlen(c, jeb, ref);
 443                 c->dirty_size += ref_totlen(c, jeb, ref);
 444
 445                 /* Convert wasted space to dirty, if not a bad block */
 446                 if (jeb->wasted_size) {
 447                         if (on_list(&jeb->list, &c->bad_used_list)) {
 448                                 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
 449                                           jeb->offset));
 450                                 addedsize = 0; /* To fool the refiling code later */
 451                         } else {
 452                                 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
 453                                           jeb->wasted_size, jeb->offset));
 454                                 addedsize += jeb->wasted_size;
 455                                 jeb->dirty_size += jeb->wasted_size;
 456                                 c->dirty_size += jeb->wasted_size;
 457                                 c->wasted_size -= jeb->wasted_size;
 458                                 jeb->wasted_size = 0;
 459                         }
 460                 }
 461         } else {
 462                 D1(printk(KERN_DEBUG "Wasting\n"));
 463                 addedsize = 0;
 464                 jeb->wasted_size += ref_totlen(c, jeb, ref);
 465                 c->wasted_size += ref_totlen(c, jeb, ref);
 466         }
 467         ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
 468
 469         ACCT_SANITY_CHECK(c, jeb);
 470
 471         D1(ACCT_PARANOIA_CHECK(jeb));
 472
 473         if (c->flags & JFFS2_SB_FLAG_SCANNING) {
 474                 /* Flash scanning is in progress. Don't muck about with the block
 475                    lists because they're not ready yet, and don't actually
 476                    obliterate nodes that look obsolete. If they weren't
 477                    marked obsolete on the flash at the time they _became_
 478                    obsolete, there was probably a reason for that. */
 479                 spin_unlock(&c->erase_completion_lock);
 480                 /* We didn't lock the erase_free_sem */
 481                 return;
 482         }
 483
 484         if (jeb == c->nextblock) {
 485                 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
 486         } else if (!jeb->used_size && !jeb->unchecked_size) {
 487                 if (jeb == c->gcblock) {
 488                         D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
 489                         c->gcblock = NULL;
 490                 } else {
 491                         D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
 492                         list_del(&jeb->list);
 493                 }
 494                 if (jffs2_wbuf_dirty(c)) {
 495                         D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
 496                         list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
 497                 } else {
 498                         if (jiffies & 127) {
 499                                 /* Most of the time, we just erase it immediately. Otherwise we
 500                                    spend ages scanning it on mount, etc. */
 501                                 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
 502                                 list_add_tail(&jeb->list, &c->erase_pending_list);
 503                                 c->nr_erasing_blocks++;
 504                                 jffs2_erase_pending_trigger(c);
 505                         } else {
 506                                 /* Sometimes, however, we leave it elsewhere so it doesn't get
 507                                    immediately reused, and we spread the load a bit. */
 508                                 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
 509                                 list_add_tail(&jeb->list, &c->erasable_list);
 510                         }
 511                 }
 512                 D1(printk(KERN_DEBUG "Done OK\n"));
 513         } else if (jeb == c->gcblock) {
 514                 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
 515         } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
 516                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
 517                 list_del(&jeb->list);
 518                 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
 519                 list_add_tail(&jeb->list, &c->dirty_list);
 520         } else if (VERYDIRTY(c, jeb->dirty_size) &&
 521                    !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
 522                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
 523                 list_del(&jeb->list);
 524                 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
 525                 list_add_tail(&jeb->list, &c->very_dirty_list);
 526         } else {
 527                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
 528                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 529         }
 530
 531         spin_unlock(&c->erase_completion_lock);
 532
 533         if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
 534                 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
 535                 /* We didn't lock the erase_free_sem */
 536                 return;
 537         }
 538
 539         /* The erase_free_sem is locked, and has been since before we marked the node obsolete
 540            and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
 541            the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
 542            by jffs2_free_all_node_refs() in erase.c. Which is nice. */
 543
 544         D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
 545         ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
 546         if (ret) {
 547                 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
 548                 goto out_erase_sem;
 549         }
 550         if (retlen != sizeof(n)) {
 551                 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
 552                 goto out_erase_sem;
 553         }
 554         if (PAD(je32_to_cpu(n.totlen)) != PAD(ref_totlen(c, jeb, ref))) {
 555                 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), ref_totlen(c, jeb, ref));
 556                 goto out_erase_sem;
 557         }
 558         if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
 559                 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
 560                 goto out_erase_sem;
 561         }
 562         /* XXX FIXME: This is ugly now */
 563         n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
 564         ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
 565         if (ret) {
 566                 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
 567                 goto out_erase_sem;
 568         }
 569         if (retlen != sizeof(n)) {
 570                 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
 571                 goto out_erase_sem;
 572         }
 573
 574         /* Nodes which have been marked obsolete no longer need to be
 575            associated with any inode. Remove them from the per-inode list.
 576
 577            Note we can't do this for NAND at the moment because we need
 578            obsolete dirent nodes to stay on the lists, because of the
 579            horridness in jffs2_garbage_collect_deletion_dirent(). Also
 580            because we delete the inocache, and on NAND we need that to
 581            stay around until all the nodes are actually erased, in order
 582            to stop us from giving the same inode number to another newly
 583            created inode. */
 584         if (ref->next_in_ino) {
 585                 struct jffs2_inode_cache *ic;
 586                 struct jffs2_raw_node_ref **p;
 587
 588                 spin_lock(&c->erase_completion_lock);
 589
 590                 ic = jffs2_raw_ref_to_ic(ref);
 591                 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
 592                         ;
 593
 594                 *p = ref->next_in_ino;
 595                 ref->next_in_ino = NULL;
 596
 597                 if (ic->nodes == (void *)ic && ic->nlink == 0)
 598                         jffs2_del_ino_cache(c, ic);
 599
 600                 spin_unlock(&c->erase_completion_lock);
 601         }
 602
 603
 604         /* Merge with the next node in the physical list, if there is one
 605            and if it's also obsolete and if it doesn't belong to any inode */
 606         if (ref->next_phys && ref_obsolete(ref->next_phys) &&
 607             !ref->next_phys->next_in_ino) {
 608                 struct jffs2_raw_node_ref *n = ref->next_phys;
 609
 610                 spin_lock(&c->erase_completion_lock);
 611
 612                 ref->__totlen += n->__totlen;
 613                 ref->next_phys = n->next_phys;
 614                 if (jeb->last_node == n) jeb->last_node = ref;
 615                 if (jeb->gc_node == n) {
 616                         /* gc will be happy continuing gc on this node */
 617                         jeb->gc_node=ref;
 618                 }
 619                 spin_unlock(&c->erase_completion_lock);
 620
 621                 jffs2_free_raw_node_ref(n);
 622         }
 623
 624         /* Also merge with the previous node in the list, if there is one
 625            and that one is obsolete */
 626         if (ref != jeb->first_node ) {
 627                 struct jffs2_raw_node_ref *p = jeb->first_node;
 628
 629                 spin_lock(&c->erase_completion_lock);
 630
 631                 while (p->next_phys != ref)
 632                         p = p->next_phys;
 633
 634                 if (ref_obsolete(p) && !ref->next_in_ino) {
 635                         p->__totlen += ref->__totlen;
 636                         if (jeb->last_node == ref) {
 637                                 jeb->last_node = p;
 638                         }
 639                         if (jeb->gc_node == ref) {
 640                                 /* gc will be happy continuing gc on this node */
 641                                 jeb->gc_node=p;
 642                         }
 643                         p->next_phys = ref->next_phys;
 644                         jffs2_free_raw_node_ref(ref);
 645                 }
 646                 spin_unlock(&c->erase_completion_lock);
 647         }
 648  out_erase_sem:
 649         up(&c->erase_free_sem);
 650 }
 651
 652 #if CONFIG_JFFS2_FS_DEBUG >= 2
 653 void jffs2_dump_block_lists(struct jffs2_sb_info *c)
 654 {
 655
 656
 657         printk(KERN_DEBUG "jffs2_dump_block_lists:\n");
 658         printk(KERN_DEBUG "flash_size: %08x\n", c->flash_size);
 659         printk(KERN_DEBUG "used_size: %08x\n", c->used_size);
 660         printk(KERN_DEBUG "dirty_size: %08x\n", c->dirty_size);
 661         printk(KERN_DEBUG "wasted_size: %08x\n", c->wasted_size);
 662         printk(KERN_DEBUG "unchecked_size: %08x\n", c->unchecked_size);
 663         printk(KERN_DEBUG "free_size: %08x\n", c->free_size);
 664         printk(KERN_DEBUG "erasing_size: %08x\n", c->erasing_size);
 665         printk(KERN_DEBUG "bad_size: %08x\n", c->bad_size);
 666         printk(KERN_DEBUG "sector_size: %08x\n", c->sector_size);
 667         printk(KERN_DEBUG "jffs2_reserved_blocks size: %08x\n",c->sector_size * c->resv_blocks_write);
 668
 669         if (c->nextblock) {
 670                 printk(KERN_DEBUG "nextblock: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 671                        c->nextblock->offset, c->nextblock->used_size, c->nextblock->dirty_size, c->nextblock->wasted_size, c->nextblock->unchecked_size, c->nextblock->free_size);
 672         } else {
 673                 printk(KERN_DEBUG "nextblock: NULL\n");
 674         }
 675         if (c->gcblock) {
 676                 printk(KERN_DEBUG "gcblock: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 677                        c->gcblock->offset, c->gcblock->used_size, c->gcblock->dirty_size, c->gcblock->wasted_size, c->gcblock->unchecked_size, c->gcblock->free_size);
 678         } else {
 679                 printk(KERN_DEBUG "gcblock: NULL\n");
 680         }
 681         if (list_empty(&c->clean_list)) {
 682                 printk(KERN_DEBUG "clean_list: empty\n");
 683         } else {
 684                 struct list_head *this;
 685                 int     numblocks = 0;
 686                 uint32_t dirty = 0;
 687
 688                 list_for_each(this, &c->clean_list) {
 689                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 690                         numblocks ++;
 691                         dirty += jeb->wasted_size;
 692                         printk(KERN_DEBUG "clean_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n", jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 693                 }
 694                 printk (KERN_DEBUG "Contains %d blocks with total wasted size %u, average wasted size: %u\n", numblocks, dirty, dirty / numblocks);
 695         }
 696         if (list_empty(&c->very_dirty_list)) {
 697                 printk(KERN_DEBUG "very_dirty_list: empty\n");
 698         } else {
 699                 struct list_head *this;
 700                 int     numblocks = 0;
 701                 uint32_t dirty = 0;
 702
 703                 list_for_each(this, &c->very_dirty_list) {
 704                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 705                         numblocks ++;
 706                         dirty += jeb->dirty_size;
 707                         printk(KERN_DEBUG "very_dirty_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 708                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 709                 }
 710                 printk (KERN_DEBUG "Contains %d blocks with total dirty size %u, average dirty size: %u\n",
 711                         numblocks, dirty, dirty / numblocks);
 712         }
 713         if (list_empty(&c->dirty_list)) {
 714                 printk(KERN_DEBUG "dirty_list: empty\n");
 715         } else {
 716                 struct list_head *this;
 717                 int     numblocks = 0;
 718                 uint32_t dirty = 0;
 719
 720                 list_for_each(this, &c->dirty_list) {
 721                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 722                         numblocks ++;
 723                         dirty += jeb->dirty_size;
 724                         printk(KERN_DEBUG "dirty_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 725                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 726                 }
 727                 printk (KERN_DEBUG "Contains %d blocks with total dirty size %u, average dirty size: %u\n",
 728                         numblocks, dirty, dirty / numblocks);
 729         }
 730         if (list_empty(&c->erasable_list)) {
 731                 printk(KERN_DEBUG "erasable_list: empty\n");
 732         } else {
 733                 struct list_head *this;
 734
 735                 list_for_each(this, &c->erasable_list) {
 736                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 737                         printk(KERN_DEBUG "erasable_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 738                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 739                 }
 740         }
 741         if (list_empty(&c->erasing_list)) {
 742                 printk(KERN_DEBUG "erasing_list: empty\n");
 743         } else {
 744                 struct list_head *this;
 745
 746                 list_for_each(this, &c->erasing_list) {
 747                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 748                         printk(KERN_DEBUG "erasing_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 749                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 750                 }
 751         }
 752         if (list_empty(&c->erase_pending_list)) {
 753                 printk(KERN_DEBUG "erase_pending_list: empty\n");
 754         } else {
 755                 struct list_head *this;
 756
 757                 list_for_each(this, &c->erase_pending_list) {
 758                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 759                         printk(KERN_DEBUG "erase_pending_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 760                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 761                 }
 762         }
 763         if (list_empty(&c->erasable_pending_wbuf_list)) {
 764                 printk(KERN_DEBUG "erasable_pending_wbuf_list: empty\n");
 765         } else {
 766                 struct list_head *this;
 767
 768                 list_for_each(this, &c->erasable_pending_wbuf_list) {
 769                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 770                         printk(KERN_DEBUG "erasable_pending_wbuf_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 771                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 772                 }
 773         }
 774         if (list_empty(&c->free_list)) {
 775                 printk(KERN_DEBUG "free_list: empty\n");
 776         } else {
 777                 struct list_head *this;
 778
 779                 list_for_each(this, &c->free_list) {
 780                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 781                         printk(KERN_DEBUG "free_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 782                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 783                 }
 784         }
 785         if (list_empty(&c->bad_list)) {
 786                 printk(KERN_DEBUG "bad_list: empty\n");
 787         } else {
 788                 struct list_head *this;
 789
 790                 list_for_each(this, &c->bad_list) {
 791                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 792                         printk(KERN_DEBUG "bad_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 793                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 794                 }
 795         }
 796         if (list_empty(&c->bad_used_list)) {
 797                 printk(KERN_DEBUG "bad_used_list: empty\n");
 798         } else {
 799                 struct list_head *this;
 800
 801                 list_for_each(this, &c->bad_used_list) {
 802                         struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
 803                         printk(KERN_DEBUG "bad_used_list: %08x (used %08x, dirty %08x, wasted %08x, unchecked %08x, free %08x)\n",
 804                                jeb->offset, jeb->used_size, jeb->dirty_size, jeb->wasted_size, jeb->unchecked_size, jeb->free_size);
 805                 }
 806         }
 807 }
 808 #endif /* CONFIG_JFFS2_FS_DEBUG */
 809
 810 int jffs2_thread_should_wake(struct jffs2_sb_info *c)
 811 {
 812         int ret = 0;
 813         uint32_t dirty;
 814
 815         if (c->unchecked_size) {
 816                 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
 817                           c->unchecked_size, c->checked_ino));
 818                 return 1;
 819         }
 820
 821         /* dirty_size contains blocks on erase_pending_list
 822          * those blocks are counted in c->nr_erasing_blocks.
 823          * If one block is actually erased, it is not longer counted as dirty_space
 824          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
 825          * with c->nr_erasing_blocks * c->sector_size again.
 826          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
 827          * This helps us to force gc and pick eventually a clean block to spread the load.
 828          */
 829         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
 830
 831         if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
 832                         (dirty > c->nospc_dirty_size))
 833                 ret = 1;
 834
 835         D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x: %s\n",
 836                   c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, ret?"yes":"no"));
 837
 838         return ret;
 839 }