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/mtd/mtd.h>
  14 #include <linux/compiler.h>
  15 #include <linux/sched.h> /* For cond_resched() */
  16 #include "nodelist.h"
  17 #include "debug.h"
  18
  19 /**
  20  *      jffs2_reserve_space - request physical space to write nodes to flash
  21  *      @c: superblock info
  22  *      @minsize: Minimum acceptable size of allocation
  23  *      @len: Returned value of allocation length
  24  *      @prio: Allocation type - ALLOC_{NORMAL,DELETION}
  25  *
  26  *      Requests a block of physical space on the flash. Returns zero for success
  27  *      and puts 'len' into the appropriate place, or returns -ENOSPC or other
  28  *      error if appropriate. Doesn't return len since that's
  29  *
  30  *      If it returns zero, jffs2_reserve_space() also downs the per-filesystem
  31  *      allocation semaphore, to prevent more than one allocation from being
  32  *      active at any time. The semaphore is later released by jffs2_commit_allocation()
  33  *
  34  *      jffs2_reserve_space() may trigger garbage collection in order to make room
  35  *      for the requested allocation.
  36  */
  37
  38 static int jffs2_do_reserve_space(struct jffs2_sb_info *c,  uint32_t minsize,
  39                                   uint32_t *len, uint32_t sumsize);
  40
  41 int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
  42                         uint32_t *len, int prio, uint32_t sumsize)
  43 {
  44         int ret = -EAGAIN;
  45         int blocksneeded = c->resv_blocks_write;
  46         /* align it */
  47         minsize = PAD(minsize);
  48
  49         D1(printk(KERN_DEBUG "jffs2_reserve_space(): Requested 0x%x bytes\n", minsize));
  50         mutex_lock(&c->alloc_sem);
  51
  52         D1(printk(KERN_DEBUG "jffs2_reserve_space(): alloc sem got\n"));
  53
  54         spin_lock(&c->erase_completion_lock);
  55
  56         /* this needs a little more thought (true <tglx> :)) */
  57         while(ret == -EAGAIN) {
  58                 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
  59                         uint32_t dirty, avail;
  60
  61                         /* calculate real dirty size
  62                          * dirty_size contains blocks on erase_pending_list
  63                          * those blocks are counted in c->nr_erasing_blocks.
  64                          * If one block is actually erased, it is not longer counted as dirty_space
  65                          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
  66                          * with c->nr_erasing_blocks * c->sector_size again.
  67                          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
  68                          * This helps us to force gc and pick eventually a clean block to spread the load.
  69                          * We add unchecked_size here, as we hopefully will find some space to use.
  70                          * This will affect the sum only once, as gc first finishes checking
  71                          * of nodes.
  72                          */
  73                         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
  74                         if (dirty < c->nospc_dirty_size) {
  75                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
  76                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on dirty space to GC, but it's a deletion. Allowing...\n"));
  77                                         break;
  78                                 }
  79                                 D1(printk(KERN_DEBUG "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
  80                                           dirty, c->unchecked_size, c->sector_size));
  81
  82                                 spin_unlock(&c->erase_completion_lock);
  83                                 mutex_unlock(&c->alloc_sem);
  84                                 return -ENOSPC;
  85                         }
  86
  87                         /* Calc possibly available space. Possibly available means that we
  88                          * don't know, if unchecked size contains obsoleted nodes, which could give us some
  89                          * more usable space. This will affect the sum only once, as gc first finishes checking
  90                          * of nodes.
  91                          + Return -ENOSPC, if the maximum possibly available space is less or equal than
  92                          * blocksneeded * sector_size.
  93                          * This blocks endless gc looping on a filesystem, which is nearly full, even if
  94                          * the check above passes.
  95                          */
  96                         avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
  97                         if ( (avail / c->sector_size) <= blocksneeded) {
  98                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
  99                                         D1(printk(KERN_NOTICE "jffs2_reserve_space(): Low on possibly available space, but it's a deletion. Allowing...\n"));
 100                                         break;
 101                                 }
 102
 103                                 D1(printk(KERN_DEBUG "max. available size 0x%08x  < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
 104                                           avail, blocksneeded * c->sector_size));
 105                                 spin_unlock(&c->erase_completion_lock);
 106                                 mutex_unlock(&c->alloc_sem);
 107                                 return -ENOSPC;
 108                         }
 109
 110                         mutex_unlock(&c->alloc_sem);
 111
 112                         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",
 113                                   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,
 114                                   c->free_size + c->dirty_size + c->wasted_size + c->used_size + c->erasing_size + c->bad_size, c->flash_size));
 115                         spin_unlock(&c->erase_completion_lock);
 116
 117                         ret = jffs2_garbage_collect_pass(c);
 118
 119                         if (ret == -EAGAIN) {
 120                                 spin_lock(&c->erase_completion_lock);
 121                                 if (c->nr_erasing_blocks &&
 122                                     list_empty(&c->erase_pending_list) &&
 123                                     list_empty(&c->erase_complete_list)) {
 124                                         DECLARE_WAITQUEUE(wait, current);
 125                                         set_current_state(TASK_UNINTERRUPTIBLE);
 126                                         add_wait_queue(&c->erase_wait, &wait);
 127                                         D1(printk(KERN_DEBUG "%s waiting for erase to complete\n", __func__));
 128                                         spin_unlock(&c->erase_completion_lock);
 129
 130                                         schedule();
 131                                         remove_wait_queue(&c->erase_wait, &wait);
 132                                 } else
 133                                         spin_unlock(&c->erase_completion_lock);
 134                         } else if (ret)
 135                                 return ret;
 136
 137                         cond_resched();
 138
 139                         if (signal_pending(current))
 140                                 return -EINTR;
 141
 142                         mutex_lock(&c->alloc_sem);
 143                         spin_lock(&c->erase_completion_lock);
 144                 }
 145
 146                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
 147                 if (ret) {
 148                         D1(printk(KERN_DEBUG "jffs2_reserve_space: ret is %d\n", ret));
 149                 }
 150         }
 151         spin_unlock(&c->erase_completion_lock);
 152         if (!ret)
 153                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
 154         if (ret)
 155                 mutex_unlock(&c->alloc_sem);
 156         return ret;
 157 }
 158
 159 int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
 160                            uint32_t *len, uint32_t sumsize)
 161 {
 162         int ret;
 163         minsize = PAD(minsize);
 164
 165         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc(): Requested 0x%x bytes\n", minsize));
 166
 167         while (true) {
 168                 spin_lock(&c->erase_completion_lock);
 169                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
 170                 if (ret) {
 171                         D1(printk(KERN_DEBUG "jffs2_reserve_space_gc: looping, ret is %d\n", ret));
 172                 }
 173                 spin_unlock(&c->erase_completion_lock);
 174
 175                 if (ret == -EAGAIN)
 176                         cond_resched();
 177                 else
 178                         break;
 179         }
 180         if (!ret)
 181                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
 182
 183         return ret;
 184 }
 185
 186
 187 /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
 188
 189 static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
 190 {
 191
 192         if (c->nextblock == NULL) {
 193                 D1(printk(KERN_DEBUG "jffs2_close_nextblock: Erase block at 0x%08x has already been placed in a list\n",
 194                   jeb->offset));
 195                 return;
 196         }
 197         /* Check, if we have a dirty block now, or if it was dirty already */
 198         if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
 199                 c->dirty_size += jeb->wasted_size;
 200                 c->wasted_size -= jeb->wasted_size;
 201                 jeb->dirty_size += jeb->wasted_size;
 202                 jeb->wasted_size = 0;
 203                 if (VERYDIRTY(c, jeb->dirty_size)) {
 204                         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",
 205                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 206                         list_add_tail(&jeb->list, &c->very_dirty_list);
 207                 } else {
 208                         D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 209                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 210                         list_add_tail(&jeb->list, &c->dirty_list);
 211                 }
 212         } else {
 213                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 214                   jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 215                 list_add_tail(&jeb->list, &c->clean_list);
 216         }
 217         c->nextblock = NULL;
 218
 219 }
 220
 221 /* Select a new jeb for nextblock */
 222
 223 static int jffs2_find_nextblock(struct jffs2_sb_info *c)
 224 {
 225         struct list_head *next;
 226
 227         /* Take the next block off the 'free' list */
 228
 229         if (list_empty(&c->free_list)) {
 230
 231                 if (!c->nr_erasing_blocks &&
 232                         !list_empty(&c->erasable_list)) {
 233                         struct jffs2_eraseblock *ejeb;
 234
 235                         ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
 236                         list_move_tail(&ejeb->list, &c->erase_pending_list);
 237                         c->nr_erasing_blocks++;
 238                         jffs2_garbage_collect_trigger(c);
 239                         D1(printk(KERN_DEBUG "jffs2_find_nextblock: Triggering erase of erasable block at 0x%08x\n",
 240                                   ejeb->offset));
 241                 }
 242
 243                 if (!c->nr_erasing_blocks &&
 244                         !list_empty(&c->erasable_pending_wbuf_list)) {
 245                         D1(printk(KERN_DEBUG "jffs2_find_nextblock: Flushing write buffer\n"));
 246                         /* c->nextblock is NULL, no update to c->nextblock allowed */
 247                         spin_unlock(&c->erase_completion_lock);
 248                         jffs2_flush_wbuf_pad(c);
 249                         spin_lock(&c->erase_completion_lock);
 250                         /* Have another go. It'll be on the erasable_list now */
 251                         return -EAGAIN;
 252                 }
 253
 254                 if (!c->nr_erasing_blocks) {
 255                         /* Ouch. We're in GC, or we wouldn't have got here.
 256                            And there's no space left. At all. */
 257                         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",
 258                                    c->nr_erasing_blocks, c->nr_free_blocks, list_empty(&c->erasable_list)?"yes":"no",
 259                                    list_empty(&c->erasing_list)?"yes":"no", list_empty(&c->erase_pending_list)?"yes":"no");
 260                         return -ENOSPC;
 261                 }
 262
 263                 spin_unlock(&c->erase_completion_lock);
 264                 /* Don't wait for it; just erase one right now */
 265                 jffs2_erase_pending_blocks(c, 1);
 266                 spin_lock(&c->erase_completion_lock);
 267
 268                 /* An erase may have failed, decreasing the
 269                    amount of free space available. So we must
 270                    restart from the beginning */
 271                 return -EAGAIN;
 272         }
 273
 274         next = c->free_list.next;
 275         list_del(next);
 276         c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
 277         c->nr_free_blocks--;
 278
 279         jffs2_sum_reset_collected(c->summary); /* reset collected summary */
 280
 281 #ifdef CONFIG_JFFS2_FS_WRITEBUFFER
 282         /* adjust write buffer offset, else we get a non contiguous write bug */
 283         if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
 284                 c->wbuf_ofs = 0xffffffff;
 285 #endif
 286
 287         D1(printk(KERN_DEBUG "jffs2_find_nextblock(): new nextblock = 0x%08x\n", c->nextblock->offset));
 288
 289         return 0;
 290 }
 291
 292 /* Called with alloc sem _and_ erase_completion_lock */
 293 static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
 294                                   uint32_t *len, uint32_t sumsize)
 295 {
 296         struct jffs2_eraseblock *jeb = c->nextblock;
 297         uint32_t reserved_size;                         /* for summary information at the end of the jeb */
 298         int ret;
 299
 300  restart:
 301         reserved_size = 0;
 302
 303         if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
 304                                                         /* NOSUM_SIZE means not to generate summary */
 305
 306                 if (jeb) {
 307                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
 308                         dbg_summary("minsize=%d , jeb->free=%d ,"
 309                                                 "summary->size=%d , sumsize=%d\n",
 310                                                 minsize, jeb->free_size,
 311                                                 c->summary->sum_size, sumsize);
 312                 }
 313
 314                 /* Is there enough space for writing out the current node, or we have to
 315                    write out summary information now, close this jeb and select new nextblock? */
 316                 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
 317                                         JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
 318
 319                         /* Has summary been disabled for this jeb? */
 320                         if (jffs2_sum_is_disabled(c->summary)) {
 321                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
 322                                 goto restart;
 323                         }
 324
 325                         /* Writing out the collected summary information */
 326                         dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
 327                         ret = jffs2_sum_write_sumnode(c);
 328
 329                         if (ret)
 330                                 return ret;
 331
 332                         if (jffs2_sum_is_disabled(c->summary)) {
 333                                 /* jffs2_write_sumnode() couldn't write out the summary information
 334                                    diabling summary for this jeb and free the collected information
 335                                  */
 336                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
 337                                 goto restart;
 338                         }
 339
 340                         jffs2_close_nextblock(c, jeb);
 341                         jeb = NULL;
 342                         /* keep always valid value in reserved_size */
 343                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
 344                 }
 345         } else {
 346                 if (jeb && minsize > jeb->free_size) {
 347                         uint32_t waste;
 348
 349                         /* Skip the end of this block and file it as having some dirty space */
 350                         /* If there's a pending write to it, flush now */
 351
 352                         if (jffs2_wbuf_dirty(c)) {
 353                                 spin_unlock(&c->erase_completion_lock);
 354                                 D1(printk(KERN_DEBUG "jffs2_do_reserve_space: Flushing write buffer\n"));
 355                                 jffs2_flush_wbuf_pad(c);
 356                                 spin_lock(&c->erase_completion_lock);
 357                                 jeb = c->nextblock;
 358                                 goto restart;
 359                         }
 360
 361                         spin_unlock(&c->erase_completion_lock);
 362
 363                         ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
 364                         if (ret)
 365                                 return ret;
 366                         /* Just lock it again and continue. Nothing much can change because
 367                            we hold c->alloc_sem anyway. In fact, it's not entirely clear why
 368                            we hold c->erase_completion_lock in the majority of this function...
 369                            but that's a question for another (more caffeine-rich) day. */
 370                         spin_lock(&c->erase_completion_lock);
 371
 372                         waste = jeb->free_size;
 373                         jffs2_link_node_ref(c, jeb,
 374                                             (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
 375                                             waste, NULL);
 376                         /* FIXME: that made it count as dirty. Convert to wasted */
 377                         jeb->dirty_size -= waste;
 378                         c->dirty_size -= waste;
 379                         jeb->wasted_size += waste;
 380                         c->wasted_size += waste;
 381
 382                         jffs2_close_nextblock(c, jeb);
 383                         jeb = NULL;
 384                 }
 385         }
 386
 387         if (!jeb) {
 388
 389                 ret = jffs2_find_nextblock(c);
 390                 if (ret)
 391                         return ret;
 392
 393                 jeb = c->nextblock;
 394
 395                 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
 396                         printk(KERN_WARNING "Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n", jeb->offset, jeb->free_size);
 397                         goto restart;
 398                 }
 399         }
 400         /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
 401            enough space */
 402         *len = jeb->free_size - reserved_size;
 403
 404         if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
 405             !jeb->first_node->next_in_ino) {
 406                 /* Only node in it beforehand was a CLEANMARKER node (we think).
 407                    So mark it obsolete now that there's going to be another node
 408                    in the block. This will reduce used_size to zero but We've
 409                    already set c->nextblock so that jffs2_mark_node_obsolete()
 410                    won't try to refile it to the dirty_list.
 411                 */
 412                 spin_unlock(&c->erase_completion_lock);
 413                 jffs2_mark_node_obsolete(c, jeb->first_node);
 414                 spin_lock(&c->erase_completion_lock);
 415         }
 416
 417         D1(printk(KERN_DEBUG "jffs2_do_reserve_space(): Giving 0x%x bytes at 0x%x\n",
 418                   *len, jeb->offset + (c->sector_size - jeb->free_size)));
 419         return 0;
 420 }
 421
 422 /**
 423  *      jffs2_add_physical_node_ref - add a physical node reference to the list
 424  *      @c: superblock info
 425  *      @new: new node reference to add
 426  *      @len: length of this physical node
 427  *
 428  *      Should only be used to report nodes for which space has been allocated
 429  *      by jffs2_reserve_space.
 430  *
 431  *      Must be called with the alloc_sem held.
 432  */
 433
 434 struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
 435                                                        uint32_t ofs, uint32_t len,
 436                                                        struct jffs2_inode_cache *ic)
 437 {
 438         struct jffs2_eraseblock *jeb;
 439         struct jffs2_raw_node_ref *new;
 440
 441         jeb = &c->blocks[ofs / c->sector_size];
 442
 443         D1(printk(KERN_DEBUG "jffs2_add_physical_node_ref(): Node at 0x%x(%d), size 0x%x\n",
 444                   ofs & ~3, ofs & 3, len));
 445 #if 1
 446         /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
 447            if c->nextblock is set. Note that wbuf.c will file obsolete nodes
 448            even after refiling c->nextblock */
 449         if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
 450             && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
 451                 printk(KERN_WARNING "argh. node added in wrong place at 0x%08x(%d)\n", ofs & ~3, ofs & 3);
 452                 if (c->nextblock)
 453                         printk(KERN_WARNING "nextblock 0x%08x", c->nextblock->offset);
 454                 else
 455                         printk(KERN_WARNING "No nextblock");
 456                 printk(", expected at %08x\n", jeb->offset + (c->sector_size - jeb->free_size));
 457                 return ERR_PTR(-EINVAL);
 458         }
 459 #endif
 460         spin_lock(&c->erase_completion_lock);
 461
 462         new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
 463
 464         if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
 465                 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
 466                 D1(printk(KERN_DEBUG "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 467                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 468                 if (jffs2_wbuf_dirty(c)) {
 469                         /* Flush the last write in the block if it's outstanding */
 470                         spin_unlock(&c->erase_completion_lock);
 471                         jffs2_flush_wbuf_pad(c);
 472                         spin_lock(&c->erase_completion_lock);
 473                 }
 474
 475                 list_add_tail(&jeb->list, &c->clean_list);
 476                 c->nextblock = NULL;
 477         }
 478         jffs2_dbg_acct_sanity_check_nolock(c,jeb);
 479         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
 480
 481         spin_unlock(&c->erase_completion_lock);
 482
 483         return new;
 484 }
 485
 486
 487 void jffs2_complete_reservation(struct jffs2_sb_info *c)
 488 {
 489         D1(printk(KERN_DEBUG "jffs2_complete_reservation()\n"));
 490         spin_lock(&c->erase_completion_lock);
 491         jffs2_garbage_collect_trigger(c);
 492         spin_unlock(&c->erase_completion_lock);
 493         mutex_unlock(&c->alloc_sem);
 494 }
 495
 496 static inline int on_list(struct list_head *obj, struct list_head *head)
 497 {
 498         struct list_head *this;
 499
 500         list_for_each(this, head) {
 501                 if (this == obj) {
 502                         D1(printk("%p is on list at %p\n", obj, head));
 503                         return 1;
 504
 505                 }
 506         }
 507         return 0;
 508 }
 509
 510 void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
 511 {
 512         struct jffs2_eraseblock *jeb;
 513         int blocknr;
 514         struct jffs2_unknown_node n;
 515         int ret, addedsize;
 516         size_t retlen;
 517         uint32_t freed_len;
 518
 519         if(unlikely(!ref)) {
 520                 printk(KERN_NOTICE "EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
 521                 return;
 522         }
 523         if (ref_obsolete(ref)) {
 524                 D1(printk(KERN_DEBUG "jffs2_mark_node_obsolete called with already obsolete node at 0x%08x\n", ref_offset(ref)));
 525                 return;
 526         }
 527         blocknr = ref->flash_offset / c->sector_size;
 528         if (blocknr >= c->nr_blocks) {
 529                 printk(KERN_NOTICE "raw node at 0x%08x is off the end of device!\n", ref->flash_offset);
 530                 BUG();
 531         }
 532         jeb = &c->blocks[blocknr];
 533
 534         if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
 535             !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
 536                 /* Hm. This may confuse static lock analysis. If any of the above
 537                    three conditions is false, we're going to return from this
 538                    function without actually obliterating any nodes or freeing
 539                    any jffs2_raw_node_refs. So we don't need to stop erases from
 540                    happening, or protect against people holding an obsolete
 541                    jffs2_raw_node_ref without the erase_completion_lock. */
 542                 mutex_lock(&c->erase_free_sem);
 543         }
 544
 545         spin_lock(&c->erase_completion_lock);
 546
 547         freed_len = ref_totlen(c, jeb, ref);
 548
 549         if (ref_flags(ref) == REF_UNCHECKED) {
 550                 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
 551                         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",
 552                                freed_len, blocknr, ref->flash_offset, jeb->used_size);
 553                         BUG();
 554                 })
 555                 D1(printk(KERN_DEBUG "Obsoleting previously unchecked node at 0x%08x of len %x: ", ref_offset(ref), freed_len));
 556                 jeb->unchecked_size -= freed_len;
 557                 c->unchecked_size -= freed_len;
 558         } else {
 559                 D1(if (unlikely(jeb->used_size < freed_len)) {
 560                         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",
 561                                freed_len, blocknr, ref->flash_offset, jeb->used_size);
 562                         BUG();
 563                 })
 564                 D1(printk(KERN_DEBUG "Obsoleting node at 0x%08x of len %#x: ", ref_offset(ref), freed_len));
 565                 jeb->used_size -= freed_len;
 566                 c->used_size -= freed_len;
 567         }
 568
 569         // Take care, that wasted size is taken into concern
 570         if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
 571                 D1(printk("Dirtying\n"));
 572                 addedsize = freed_len;
 573                 jeb->dirty_size += freed_len;
 574                 c->dirty_size += freed_len;
 575
 576                 /* Convert wasted space to dirty, if not a bad block */
 577                 if (jeb->wasted_size) {
 578                         if (on_list(&jeb->list, &c->bad_used_list)) {
 579                                 D1(printk(KERN_DEBUG "Leaving block at %08x on the bad_used_list\n",
 580                                           jeb->offset));
 581                                 addedsize = 0; /* To fool the refiling code later */
 582                         } else {
 583                                 D1(printk(KERN_DEBUG "Converting %d bytes of wasted space to dirty in block at %08x\n",
 584                                           jeb->wasted_size, jeb->offset));
 585                                 addedsize += jeb->wasted_size;
 586                                 jeb->dirty_size += jeb->wasted_size;
 587                                 c->dirty_size += jeb->wasted_size;
 588                                 c->wasted_size -= jeb->wasted_size;
 589                                 jeb->wasted_size = 0;
 590                         }
 591                 }
 592         } else {
 593                 D1(printk("Wasting\n"));
 594                 addedsize = 0;
 595                 jeb->wasted_size += freed_len;
 596                 c->wasted_size += freed_len;
 597         }
 598         ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
 599
 600         jffs2_dbg_acct_sanity_check_nolock(c, jeb);
 601         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
 602
 603         if (c->flags & JFFS2_SB_FLAG_SCANNING) {
 604                 /* Flash scanning is in progress. Don't muck about with the block
 605                    lists because they're not ready yet, and don't actually
 606                    obliterate nodes that look obsolete. If they weren't
 607                    marked obsolete on the flash at the time they _became_
 608                    obsolete, there was probably a reason for that. */
 609                 spin_unlock(&c->erase_completion_lock);
 610                 /* We didn't lock the erase_free_sem */
 611                 return;
 612         }
 613
 614         if (jeb == c->nextblock) {
 615                 D2(printk(KERN_DEBUG "Not moving nextblock 0x%08x to dirty/erase_pending list\n", jeb->offset));
 616         } else if (!jeb->used_size && !jeb->unchecked_size) {
 617                 if (jeb == c->gcblock) {
 618                         D1(printk(KERN_DEBUG "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n", jeb->offset));
 619                         c->gcblock = NULL;
 620                 } else {
 621                         D1(printk(KERN_DEBUG "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n", jeb->offset));
 622                         list_del(&jeb->list);
 623                 }
 624                 if (jffs2_wbuf_dirty(c)) {
 625                         D1(printk(KERN_DEBUG "...and adding to erasable_pending_wbuf_list\n"));
 626                         list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
 627                 } else {
 628                         if (jiffies & 127) {
 629                                 /* Most of the time, we just erase it immediately. Otherwise we
 630                                    spend ages scanning it on mount, etc. */
 631                                 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
 632                                 list_add_tail(&jeb->list, &c->erase_pending_list);
 633                                 c->nr_erasing_blocks++;
 634                                 jffs2_garbage_collect_trigger(c);
 635                         } else {
 636                                 /* Sometimes, however, we leave it elsewhere so it doesn't get
 637                                    immediately reused, and we spread the load a bit. */
 638                                 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
 639                                 list_add_tail(&jeb->list, &c->erasable_list);
 640                         }
 641                 }
 642                 D1(printk(KERN_DEBUG "Done OK\n"));
 643         } else if (jeb == c->gcblock) {
 644                 D2(printk(KERN_DEBUG "Not moving gcblock 0x%08x to dirty_list\n", jeb->offset));
 645         } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
 646                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n", jeb->offset));
 647                 list_del(&jeb->list);
 648                 D1(printk(KERN_DEBUG "...and adding to dirty_list\n"));
 649                 list_add_tail(&jeb->list, &c->dirty_list);
 650         } else if (VERYDIRTY(c, jeb->dirty_size) &&
 651                    !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
 652                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n", jeb->offset));
 653                 list_del(&jeb->list);
 654                 D1(printk(KERN_DEBUG "...and adding to very_dirty_list\n"));
 655                 list_add_tail(&jeb->list, &c->very_dirty_list);
 656         } else {
 657                 D1(printk(KERN_DEBUG "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
 658                           jeb->offset, jeb->free_size, jeb->dirty_size, jeb->used_size));
 659         }
 660
 661         spin_unlock(&c->erase_completion_lock);
 662
 663         if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
 664                 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
 665                 /* We didn't lock the erase_free_sem */
 666                 return;
 667         }
 668
 669         /* The erase_free_sem is locked, and has been since before we marked the node obsolete
 670            and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
 671            the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
 672            by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
 673
 674         D1(printk(KERN_DEBUG "obliterating obsoleted node at 0x%08x\n", ref_offset(ref)));
 675         ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
 676         if (ret) {
 677                 printk(KERN_WARNING "Read error reading from obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
 678                 goto out_erase_sem;
 679         }
 680         if (retlen != sizeof(n)) {
 681                 printk(KERN_WARNING "Short read from obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
 682                 goto out_erase_sem;
 683         }
 684         if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
 685                 printk(KERN_WARNING "Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n", je32_to_cpu(n.totlen), freed_len);
 686                 goto out_erase_sem;
 687         }
 688         if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
 689                 D1(printk(KERN_DEBUG "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n", ref_offset(ref), je16_to_cpu(n.nodetype)));
 690                 goto out_erase_sem;
 691         }
 692         /* XXX FIXME: This is ugly now */
 693         n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
 694         ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
 695         if (ret) {
 696                 printk(KERN_WARNING "Write error in obliterating obsoleted node at 0x%08x: %d\n", ref_offset(ref), ret);
 697                 goto out_erase_sem;
 698         }
 699         if (retlen != sizeof(n)) {
 700                 printk(KERN_WARNING "Short write in obliterating obsoleted node at 0x%08x: %zd\n", ref_offset(ref), retlen);
 701                 goto out_erase_sem;
 702         }
 703
 704         /* Nodes which have been marked obsolete no longer need to be
 705            associated with any inode. Remove them from the per-inode list.
 706
 707            Note we can't do this for NAND at the moment because we need
 708            obsolete dirent nodes to stay on the lists, because of the
 709            horridness in jffs2_garbage_collect_deletion_dirent(). Also
 710            because we delete the inocache, and on NAND we need that to
 711            stay around until all the nodes are actually erased, in order
 712            to stop us from giving the same inode number to another newly
 713            created inode. */
 714         if (ref->next_in_ino) {
 715                 struct jffs2_inode_cache *ic;
 716                 struct jffs2_raw_node_ref **p;
 717
 718                 spin_lock(&c->erase_completion_lock);
 719
 720                 ic = jffs2_raw_ref_to_ic(ref);
 721                 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
 722                         ;
 723
 724                 *p = ref->next_in_ino;
 725                 ref->next_in_ino = NULL;
 726
 727                 switch (ic->class) {
 728 #ifdef CONFIG_JFFS2_FS_XATTR
 729                         case RAWNODE_CLASS_XATTR_DATUM:
 730                                 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
 731                                 break;
 732                         case RAWNODE_CLASS_XATTR_REF:
 733                                 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
 734                                 break;
 735 #endif
 736                         default:
 737                                 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
 738                                         jffs2_del_ino_cache(c, ic);
 739                                 break;
 740                 }
 741                 spin_unlock(&c->erase_completion_lock);
 742         }
 743
 744  out_erase_sem:
 745         mutex_unlock(&c->erase_free_sem);
 746 }
 747
 748 int jffs2_thread_should_wake(struct jffs2_sb_info *c)
 749 {
 750         int ret = 0;
 751         uint32_t dirty;
 752         int nr_very_dirty = 0;
 753         struct jffs2_eraseblock *jeb;
 754
 755         if (!list_empty(&c->erase_complete_list) ||
 756             !list_empty(&c->erase_pending_list))
 757                 return 1;
 758
 759         if (c->unchecked_size) {
 760                 D1(printk(KERN_DEBUG "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
 761                           c->unchecked_size, c->checked_ino));
 762                 return 1;
 763         }
 764
 765         /* dirty_size contains blocks on erase_pending_list
 766          * those blocks are counted in c->nr_erasing_blocks.
 767          * If one block is actually erased, it is not longer counted as dirty_space
 768          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
 769          * with c->nr_erasing_blocks * c->sector_size again.
 770          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
 771          * This helps us to force gc and pick eventually a clean block to spread the load.
 772          */
 773         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
 774
 775         if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
 776                         (dirty > c->nospc_dirty_size))
 777                 ret = 1;
 778
 779         list_for_each_entry(jeb, &c->very_dirty_list, list) {
 780                 nr_very_dirty++;
 781                 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
 782                         ret = 1;
 783                         /* In debug mode, actually go through and count them all */
 784                         D1(continue);
 785                         break;
 786                 }
 787         }
 788
 789         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",
 790                   c->nr_free_blocks, c->nr_erasing_blocks, c->dirty_size, nr_very_dirty, ret?"yes":"no"));
 791
 792         return ret;
 793 }