err.no Git - linux-2.6/blob - fs/jffs2/nodemgmt.c

   1 /*
   2  * JFFS2 -- Journalling Flash File System, Version 2.
   3  *
   4  * Copyright © 2001-2007 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  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/slab.h>
  14 #include <linux/mtd/mtd.h>
  15 #include <linux/compiler.h>
  16 #include <linux/sched.h> /* For cond_resched() */
  17 #include "nodelist.h"
  18 #include "debug.h"
  19
  20 /**
  21  *      jffs2_reserve_space - request physical space to write nodes to flash
  22  *      @c: superblock info
  23  *      @minsize: Minimum acceptable size of allocation
  24  *      @len: Returned value of allocation length
  25  *      @prio: Allocation type - ALLOC_{NORMAL,DELETION}
  26  *
  27  *      Requests a block of physical space on the flash. Returns zero for success
  28  *      and puts 'len' into the appropriate place, or returns -ENOSPC or other
  29  *      error if appropriate. Doesn't return len since that's
  30  *
  31  *      If it returns zero, jffs2_reserve_space() also downs the per-filesystem
  32  *      allocation semaphore, to prevent more than one allocation from being
  33  *      active at any time. The semaphore is later released by jffs2_commit_allocation()
  34  *
  35  *      jffs2_reserve_space() may trigger garbage collection in order to make room
  36  *      for the requested allocation.
  37  */
  38
  39 static int jffs2_do_reserve_space(struct jffs2_sb_info *c,  uint32_t minsize,
  40                                   uint32_t *len, uint32_t sumsize);
  41
  42 int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
  43                         uint32_t *len, int prio, uint32_t sumsize)
  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         mutex_lock(&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                         uint32_t dirty, avail;
  61
  62                         /* calculate real dirty size
  63                          * dirty_size contains blocks on erase_pending_list
  64                          * those blocks are counted in c->nr_erasing_blocks.
  65                          * If one block is actually erased, it is not longer counted as dirty_space
  66                          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
  67                          * with c->nr_erasing_blocks * c->sector_size again.
  68                          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
  69                          * This helps us to force gc and pick eventually a clean block to spread the load.
  70                          * We add unchecked_size here, as we hopefully will find some space to use.
  71                          * This will affect the sum only once, as gc first finishes checking
  72                          * of nodes.
  73                          */
  74                         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
  75                         if (dirty < c->nospc_dirty_size) {
  76                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
  77                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
  78                                         break;
  79                                 }
  80                                 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
  81                                           dirty, c->unchecked_size, c->sector_size));
  82
  83                                 spin_unlock(&c->erase_completion_lock);
  84                                 mutex_unlock(&c->alloc_sem);
  85                                 return -ENOSPC;
  86                         }
  87
  88                         /* Calc possibly available space. Possibly available means that we
  89                          * don't know, if unchecked size contains obsoleted nodes, which could give us some
  90                          * more usable space. This will affect the sum only once, as gc first finishes checking
  91                          * of nodes.
  92                          + Return -ENOSPC, if the maximum possibly available space is less or equal than
  93                          * blocksneeded * sector_size.
  94                          * This blocks endless gc looping on a filesystem, which is nearly full, even if
  95                          * the check above passes.
  96                          */
  97                         avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
  98                         if ( (avail / c->sector_size) <= blocksneeded) {
  99                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
 100                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
 101                                         break;
 102                                 }
 103
 104                                 D1(printk(KERN_DEBUG "max. available size 0x%08x  < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
 105                                           avail, blocksneeded * c->sector_size));
 106                                 spin_unlock(&c->erase_completion_lock);
 107                                 mutex_unlock(&c->alloc_sem);
 108                                 return -ENOSPC;
 109                         }
 110
 111                         mutex_unlock(&c->alloc_sem);
 112
 113                         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",
 114                                   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,
 115                                   c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
 116                         spin_unlock(&c->erase_completion_lock);
 117
 118                         ret = jffs2_garbage_collect_pass(c);
 119
 120                         if (ret == -EAGAIN)
 121                                 jffs2_erase_pending_blocks(c, 1);
 122                         else if (ret)
 123                                 return ret;
 124
 125                         cond_resched();
 126
 127                         if (signal_pending(current))
 128                                 return -EINTR;
 129
 130                         mutex_lock(&c->alloc_sem);
 131                         spin_lock(&c->erase_completion_lock);
 132                 }
 133
 134                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
 135                 if (ret) {
 136                         D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
 137                 }
 138         }
 139         spin_unlock(&c->erase_completion_lock);
 140         if (!ret)
 141                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
 142         if (ret)
 143                 mutex_unlock(&c->alloc_sem);
 144         return ret;
 145 }
 146
 147 int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
 148                            uint32_t *len, uint32_t sumsize)
 149 {
 150         int ret = -EAGAIN;
 151         minsize = PAD(minsize);
 152
 153         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
 154
 155         spin_lock(&c->erase_completion_lock);
 156         while(ret == -EAGAIN) {
 157                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
 158                 if (ret) {
 159                         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
 160                 }
 161         }
 162         spin_unlock(&c->erase_completion_lock);
 163         if (!ret)
 164                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
 165
 166         return ret;
 167 }
 168
 169
 170 /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
 171
 172 static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
 173 {
 174
 175         if (c->nextblock == NULL) {
 176                 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
 177                   jeb->offset));
 178                 return;
 179         }
 180         /* Check, if we have a dirty block now, or if it was dirty already */
 181         if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
 182                 c->dirty_size += jeb->wasted_size;
 183                 c->wasted_size -= jeb->wasted_size;
 184                 jeb->dirty_size += jeb->wasted_size;
 185                 jeb->wasted_size = 0;
 186                 if (VERYDIRTY(c, jeb->dirty_size)) {
 187                         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",
 188                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 189                         list_add_tail(&jeb->list, &c->very_dirty_list);
 190                 } else {
 191                         D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 192                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 193                         list_add_tail(&jeb->list, &c->dirty_list);
 194                 }
 195         } else {
 196                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 197                   jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 198                 list_add_tail(&jeb->list, &c->clean_list);
 199         }
 200         c->nextblock = NULL;
 201
 202 }
 203
 204 /* Select a new jeb for nextblock */
 205
 206 static int jffs2_find_nextblock(struct jffs2_sb_info *c)
 207 {
 208         struct list_head *next;
 209
 210         /* Take the next block off the 'free' list */
 211
 212         if (list_empty(&c->free_list)) {
 213
 214                 if (!c->nr_erasing_blocks &&
 215                         !list_empty(&c->erasable_list)) {
 216                         struct jffs2_eraseblock *ejeb;
 217
 218                         ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
 219                         list_move_tail(&ejeb->list, &c->erase_pending_list);
 220                         c->nr_erasing_blocks++;
 221                         jffs2_erase_pending_trigger(c);
 222                         D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
 223                                   ejeb->offset));
 224                 }
 225
 226                 if (!c->nr_erasing_blocks &&
 227                         !list_empty(&c->erasable_pending_wbuf_list)) {
 228                         D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
 229                         /* c->nextblock is NULL, no update to c->nextblock allowed */
 230                         spin_unlock(&c->erase_completion_lock);
 231                         jffs2_flush_wbuf_pad(c);
 232                         spin_lock(&c->erase_completion_lock);
 233                         /* Have another go. It'll be on the erasable_list now */
 234                         return -EAGAIN;
 235                 }
 236
 237                 if (!c->nr_erasing_blocks) {
 238                         /* Ouch. We're in GC, or we wouldn't have got here.
 239                            And there's no space left. At all. */
 240                         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",
 241                                    c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
 242                                    list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
 243                         return -ENOSPC;
 244                 }
 245
 246                 spin_unlock(&c->erase_completion_lock);
 247                 /* Don't wait for it; just erase one right now */
 248                 jffs2_erase_pending_blocks(c, 1);
 249                 spin_lock(&c->erase_completion_lock);
 250
 251                 /* An erase may have failed, decreasing the
 252                    amount of free space available. So we must
 253                    restart from the beginning */
 254                 return -EAGAIN;
 255         }
 256
 257         next = c->free_list.next;
 258         list_del(next);
 259         c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
 260         c->nr_free_blocks--;
 261
 262         jffs2_sum_reset_collected(c->summary); /* reset collected summary */
 263
 264         D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
 265
 266         return 0;
 267 }
 268
 269 /* Called with alloc sem _and_ erase_completion_lock */
 270 static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
 271                                   uint32_t *len, uint32_t sumsize)
 272 {
 273         struct jffs2_eraseblock *jeb = c->nextblock;
 274         uint32_t reserved_size;                         /* for summary information at the end of the jeb */
 275         int ret;
 276
 277  restart:
 278         reserved_size = 0;
 279
 280         if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
 281                                                         /* NOSUM_SIZE means not to generate summary */
 282
 283                 if (jeb) {
 284                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
 285                         dbg_summary("minsize=%d , jeb->free=%d ,"
 286                                                 "summary->size=%d , sumsize=%d\n",
 287                                                 minsize, jeb->free_size,
 288                                                 c->summary->sum_size, sumsize);
 289                 }
 290
 291                 /* Is there enough space for writing out the current node, or we have to
 292                    write out summary information now, close this jeb and select new nextblock? */
 293                 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
 294                                         JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
 295
 296                         /* Has summary been disabled for this jeb? */
 297                         if (jffs2_sum_is_disabled(c->summary)) {
 298                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
 299                                 goto restart;
 300                         }
 301
 302                         /* Writing out the collected summary information */
 303                         dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
 304                         ret = jffs2_sum_write_sumnode(c);
 305
 306                         if (ret)
 307                                 return ret;
 308
 309                         if (jffs2_sum_is_disabled(c->summary)) {
 310                                 /* jffs2_write_sumnode() couldn't write out the summary information
 311                                    diabling summary for this jeb and free the collected information
 312                                  */
 313                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
 314                                 goto restart;
 315                         }
 316
 317                         jffs2_close_nextblock(c, jeb);
 318                         jeb = NULL;
 319                         /* keep always valid value in reserved_size */
 320                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
 321                 }
 322         } else {
 323                 if (jeb && minsize > jeb->free_size) {
 324                         uint32_t waste;
 325
 326                         /* Skip the end of this block and file it as having some dirty space */
 327                         /* If there's a pending write to it, flush now */
 328
 329                         if (jffs2_wbuf_dirty(c)) {
 330                                 spin_unlock(&c->erase_completion_lock);
 331                                 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
 332                                 jffs2_flush_wbuf_pad(c);
 333                                 spin_lock(&c->erase_completion_lock);
 334                                 jeb = c->nextblock;
 335                                 goto restart;
 336                         }
 337
 338                         spin_unlock(&c->erase_completion_lock);
 339
 340                         ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
 341                         if (ret)
 342                                 return ret;
 343                         /* Just lock it again and continue. Nothing much can change because
 344                            we hold c->alloc_sem anyway. In fact, it's not entirely clear why
 345                            we hold c->erase_completion_lock in the majority of this function...
 346                            but that's a question for another (more caffeine-rich) day. */
 347                         spin_lock(&c->erase_completion_lock);
 348
 349                         waste = jeb->free_size;
 350                         jffs2_link_node_ref(c, jeb,
 351                                             (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
 352                                             waste, NULL);
 353                         /* FIXME: that made it count as dirty. Convert to wasted */
 354                         jeb->dirty_size -= waste;
 355                         c->dirty_size -= waste;
 356                         jeb->wasted_size += waste;
 357                         c->wasted_size += waste;
 358
 359                         jffs2_close_nextblock(c, jeb);
 360                         jeb = NULL;
 361                 }
 362         }
 363
 364         if (!jeb) {
 365
 366                 ret = jffs2_find_nextblock(c);
 367                 if (ret)
 368                         return ret;
 369
 370                 jeb = c->nextblock;
 371
 372                 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
 373                         printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
 374                         goto restart;
 375                 }
 376         }
 377         /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
 378            enough space */
 379         *len = jeb->free_size - reserved_size;
 380
 381         if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
 382             !jeb->first_node->next_in_ino) {
 383                 /* Only node in it beforehand was a CLEANMARKER node (we think).
 384                    So mark it obsolete now that there's going to be another node
 385                    in the block. This will reduce used_size to zero but We've
 386                    already set c->nextblock so that jffs2_mark_node_obsolete()
 387                    won't try to refile it to the dirty_list.
 388                 */
 389                 spin_unlock(&c->erase_completion_lock);
 390                 jffs2_mark_node_obsolete(c, jeb->first_node);
 391                 spin_lock(&c->erase_completion_lock);
 392         }
 393
 394         D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
 395                   *len, jeb->offset + (c->sector_size - jeb->free_size)));
 396         return 0;
 397 }
 398
 399 /**
 400  *      jffs2_add_physical_node_ref - add a physical node reference to the list
 401  *      @c: superblock info
 402  *      @new: new node reference to add
 403  *      @len: length of this physical node
 404  *
 405  *      Should only be used to report nodes for which space has been allocated
 406  *      by jffs2_reserve_space.
 407  *
 408  *      Must be called with the alloc_sem held.
 409  */
 410
 411 struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
 412                                                        uint32_t ofs, uint32_t len,
 413                                                        struct jffs2_inode_cache *ic)
 414 {
 415         struct jffs2_eraseblock *jeb;
 416         struct jffs2_raw_node_ref *new;
 417
 418         jeb = &c->blocks[ofs / c->sector_size];
 419
 420         D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
 421                   ofs & ~3, ofs & 3, len));
 422 #if 1
 423         /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
 424            if c->nextblock is set. Note that wbuf.c will file obsolete nodes
 425            even after refiling c->nextblock */
 426         if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
 427             && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
 428                 printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3);
 429                 if (c->nextblock)
 430                         printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset);
 431                 else
 432                         printk(KERN_WARNING "No nextblock");
 433                 printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size));
 434                 return ERR_PTR(-EINVAL);
 435         }
 436 #endif
 437         spin_lock(&c->erase_completion_lock);
 438
 439         new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
 440
 441         if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
 442                 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
 443                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 444                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 445                 if (jffs2_wbuf_dirty(c)) {
 446                         /* Flush the last write in the block if it's outstanding */
 447                         spin_unlock(&c->erase_completion_lock);
 448                         jffs2_flush_wbuf_pad(c);
 449                         spin_lock(&c->erase_completion_lock);
 450                 }
 451
 452                 list_add_tail(&jeb->list, &c->clean_list);
 453                 c->nextblock = NULL;
 454         }
 455         jffs2_dbg_acct_sanity_check_nolock(c,jeb);
 456         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
 457
 458         spin_unlock(&c->erase_completion_lock);
 459
 460         return new;
 461 }
 462
 463
 464 void jffs2_complete_reservation(struct jffs2_sb_info *c)
 465 {
 466         D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
 467         jffs2_garbage_collect_trigger(c);
 468         mutex_unlock(&c->alloc_sem);
 469 }
 470
 471 static inline int on_list(struct list_head *obj, struct list_head *head)
 472 {
 473         struct list_head *this;
 474
 475         list_for_each(this, head) {
 476                 if (this == obj) {
 477                         D1(printk("%p is on list at %p\n", obj, head));
 478                         return 1;
 479
 480                 }
 481         }
 482         return 0;
 483 }
 484
 485 void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
 486 {
 487         struct jffs2_eraseblock *jeb;
 488         int blocknr;
 489         struct jffs2_unknown_node n;
 490         int ret, addedsize;
 491         size_t retlen;
 492         uint32_t freed_len;
 493
 494         if(unlikely(!ref)) {
 495                 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
 496                 return;
 497         }
 498         if (ref_obsolete(ref)) {
 499                 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
 500                 return;
 501         }
 502         blocknr = ref->flash_offset / c->sector_size;
 503         if (blocknr >= c->nr_blocks) {
 504                 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
 505                 BUG();
 506         }
 507         jeb = &c->blocks[blocknr];
 508
 509         if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
 510             !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
 511                 /* Hm. This may confuse static lock analysis. If any of the above
 512                    three conditions is false, we're going to return from this
 513                    function without actually obliterating any nodes or freeing
 514                    any jffs2_raw_node_refs. So we don't need to stop erases from
 515                    happening, or protect against people holding an obsolete
 516                    jffs2_raw_node_ref without the erase_completion_lock. */
 517                 mutex_lock(&c->erase_free_sem);
 518         }
 519
 520         spin_lock(&c->erase_completion_lock);
 521
 522         freed_len = ref_totlen(c, jeb, ref);
 523
 524         if (ref_flags(ref) == REF_UNCHECKED) {
 525                 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
 526                         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",
 527                                freed_len, blocknr, ref->flash_offset, jeb->used_size);
 528                         BUG();
 529                 })
 530                 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
 531                 jeb->unchecked_size -= freed_len;
 532                 c->unchecked_size -= freed_len;
 533         } else {
 534                 D1(if (unlikely(jeb->used_size < freed_len)) {
 535                         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",
 536                                freed_len, blocknr, ref->flash_offset, jeb->used_size);
 537                         BUG();
 538                 })
 539                 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
 540                 jeb->used_size -= freed_len;
 541                 c->used_size -= freed_len;
 542         }
 543
 544         // Take care, that wasted size is taken into concern
 545         if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
 546                 D1(printk("Dirtying\n"));
 547                 addedsize = freed_len;
 548                 jeb->dirty_size += freed_len;
 549                 c->dirty_size += freed_len;
 550
 551                 /* Convert wasted space to dirty, if not a bad block */
 552                 if (jeb->wasted_size) {
 553                         if (on_list(&jeb->list, &c->bad_used_list)) {
 554                                 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
 555                                           jeb->offset));
 556                                 addedsize = 0; /* To fool the refiling code later */
 557                         } else {
 558                                 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
 559                                           jeb->wasted_size, jeb->offset));
 560                                 addedsize += jeb->wasted_size;
 561                                 jeb->dirty_size += jeb->wasted_size;
 562                                 c->dirty_size += jeb->wasted_size;
 563                                 c->wasted_size -= jeb->wasted_size;
 564                                 jeb->wasted_size = 0;
 565                         }
 566                 }
 567         } else {
 568                 D1(printk("Wasting\n"));
 569                 addedsize = 0;
 570                 jeb->wasted_size += freed_len;
 571                 c->wasted_size += freed_len;
 572         }
 573         ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
 574
 575         jffs2_dbg_acct_sanity_check_nolock(c, jeb);
 576         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
 577
 578         if (c->flags & JFFS2_SB_FLAG_SCANNING) {
 579                 /* Flash scanning is in progress. Don't muck about with the block
 580                    lists because they're not ready yet, and don't actually
 581                    obliterate nodes that look obsolete. If they weren't
 582                    marked obsolete on the flash at the time they _became_
 583                    obsolete, there was probably a reason for that. */
 584                 spin_unlock(&c->erase_completion_lock);
 585                 /* We didn't lock the erase_free_sem */
 586                 return;
 587         }
 588
 589         if (jeb == c->nextblock) {
 590                 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
 591         } else if (!jeb->used_size && !jeb->unchecked_size) {
 592                 if (jeb == c->gcblock) {
 593                         D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
 594                         c->gcblock = NULL;
 595                 } else {
 596                         D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
 597                         list_del(&jeb->list);
 598                 }
 599                 if (jffs2_wbuf_dirty(c)) {
 600                         D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
 601                         list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
 602                 } else {
 603                         if (jiffies & 127) {
 604                                 /* Most of the time, we just erase it immediately. Otherwise we
 605                                    spend ages scanning it on mount, etc. */
 606                                 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
 607                                 list_add_tail(&jeb->list, &c->erase_pending_list);
 608                                 c->nr_erasing_blocks++;
 609                                 jffs2_erase_pending_trigger(c);
 610                         } else {
 611                                 /* Sometimes, however, we leave it elsewhere so it doesn't get
 612                                    immediately reused, and we spread the load a bit. */
 613                                 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
 614                                 list_add_tail(&jeb->list, &c->erasable_list);
 615                         }
 616                 }
 617                 D1(printk(KERN_DEBUG "Done OK\n"));
 618         } else if (jeb == c->gcblock) {
 619                 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
 620         } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
 621                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
 622                 list_del(&jeb->list);
 623                 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
 624                 list_add_tail(&jeb->list, &c->dirty_list);
 625         } else if (VERYDIRTY(c, jeb->dirty_size) &&
 626                    !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
 627                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
 628                 list_del(&jeb->list);
 629                 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
 630                 list_add_tail(&jeb->list, &c->very_dirty_list);
 631         } else {
 632                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
 633                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 634         }
 635
 636         spin_unlock(&c->erase_completion_lock);
 637
 638         if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
 639                 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
 640                 /* We didn't lock the erase_free_sem */
 641                 return;
 642         }
 643
 644         /* The erase_free_sem is locked, and has been since before we marked the node obsolete
 645            and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
 646            the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
 647            by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
 648
 649         D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
 650         ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
 651         if (ret) {
 652                 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
 653                 goto out_erase_sem;
 654         }
 655         if (retlen != sizeof(n)) {
 656                 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
 657                 goto out_erase_sem;
 658         }
 659         if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
 660                 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len);
 661                 goto out_erase_sem;
 662         }
 663         if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
 664                 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
 665                 goto out_erase_sem;
 666         }
 667         /* XXX FIXME: This is ugly now */
 668         n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
 669         ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
 670         if (ret) {
 671                 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
 672                 goto out_erase_sem;
 673         }
 674         if (retlen != sizeof(n)) {
 675                 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
 676                 goto out_erase_sem;
 677         }
 678
 679         /* Nodes which have been marked obsolete no longer need to be
 680            associated with any inode. Remove them from the per-inode list.
 681
 682            Note we can't do this for NAND at the moment because we need
 683            obsolete dirent nodes to stay on the lists, because of the
 684            horridness in jffs2_garbage_collect_deletion_dirent(). Also
 685            because we delete the inocache, and on NAND we need that to
 686            stay around until all the nodes are actually erased, in order
 687            to stop us from giving the same inode number to another newly
 688            created inode. */
 689         if (ref->next_in_ino) {
 690                 struct jffs2_inode_cache *ic;
 691                 struct jffs2_raw_node_ref **p;
 692
 693                 spin_lock(&c->erase_completion_lock);
 694
 695                 ic = jffs2_raw_ref_to_ic(ref);
 696                 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
 697                         ;
 698
 699                 *p = ref->next_in_ino;
 700                 ref->next_in_ino = NULL;
 701
 702                 switch (ic->class) {
 703 #ifdef CONFIG_JFFS2_FS_XATTR
 704                         case RAWNODE_CLASS_XATTR_DATUM:
 705                                 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
 706                                 break;
 707                         case RAWNODE_CLASS_XATTR_REF:
 708                                 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
 709                                 break;
 710 #endif
 711                         default:
 712                                 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
 713                                         jffs2_del_ino_cache(c, ic);
 714                                 break;
 715                 }
 716                 spin_unlock(&c->erase_completion_lock);
 717         }
 718
 719  out_erase_sem:
 720         mutex_unlock(&c->erase_free_sem);
 721 }
 722
 723 int jffs2_thread_should_wake(struct jffs2_sb_info *c)
 724 {
 725         int ret = 0;
 726         uint32_t dirty;
 727         int nr_very_dirty = 0;
 728         struct jffs2_eraseblock *jeb;
 729
 730         if (c->unchecked_size) {
 731                 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
 732                           c->unchecked_size, c->checked_ino));
 733                 return 1;
 734         }
 735
 736         /* dirty_size contains blocks on erase_pending_list
 737          * those blocks are counted in c->nr_erasing_blocks.
 738          * If one block is actually erased, it is not longer counted as dirty_space
 739          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
 740          * with c->nr_erasing_blocks * c->sector_size again.
 741          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
 742          * This helps us to force gc and pick eventually a clean block to spread the load.
 743          */
 744         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
 745
 746         if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
 747                         (dirty > c->nospc_dirty_size))
 748                 ret = 1;
 749
 750         list_for_each_entry(jeb, &c->very_dirty_list, list) {
 751                 nr_very_dirty++;
 752                 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
 753                         ret = 1;
 754                         /* In debug mode, actually go through and count them all */
 755                         D1(continue);
 756                         break;
 757                 }
 758         }
 759
 760         D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
 761                   c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no"));
 762
 763         return ret;
 764 }