kernel/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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <linux/kernel.h>
  15 #include <linux/mtd/mtd.h>
  16 #include <linux/compiler.h>
  17 #include <linux/sched.h> /* For cond_resched() */
  18 #include "nodelist.h"
  19 #include "debug.h"
  20
  21 /*
  22  * Check whether the user is allowed to write.
  23  */
  24 static int jffs2_rp_can_write(struct jffs2_sb_info *c)
  25 {
  26         uint32_t avail;
  27         struct jffs2_mount_opts *opts = &c->mount_opts;
  28
  29         avail = c->dirty_size + c->free_size + c->unchecked_size +
  30                 c->erasing_size - c->resv_blocks_write * c->sector_size
  31                 - c->nospc_dirty_size;
  32
  33         if (avail < 2 * opts->rp_size)
  34                 jffs2_dbg(1, "rpsize %u, dirty_size %u, free_size %u, "
  35                           "erasing_size %u, unchecked_size %u, "
  36                           "nr_erasing_blocks %u, avail %u, resrv %u\n",
  37                           opts->rp_size, c->dirty_size, c->free_size,
  38                           c->erasing_size, c->unchecked_size,
  39                           c->nr_erasing_blocks, avail, c->nospc_dirty_size);
  40
  41         if (avail > opts->rp_size)
  42                 return 1;
  43
  44         /* Always allow root */
  45         if (capable(CAP_SYS_RESOURCE))
  46                 return 1;
  47
  48         jffs2_dbg(1, "forbid writing\n");
  49         return 0;
  50 }
  51
  52 /**
  53  *      jffs2_reserve_space - request physical space to write nodes to flash
  54  *      @c: superblock info
  55  *      @minsize: Minimum acceptable size of allocation
  56  *      @len: Returned value of allocation length
  57  *      @prio: Allocation type - ALLOC_{NORMAL,DELETION}
  58  *
  59  *      Requests a block of physical space on the flash. Returns zero for success
  60  *      and puts 'len' into the appropriate place, or returns -ENOSPC or other
  61  *      error if appropriate. Doesn't return len since that's
  62  *
  63  *      If it returns zero, jffs2_reserve_space() also downs the per-filesystem
  64  *      allocation semaphore, to prevent more than one allocation from being
  65  *      active at any time. The semaphore is later released by jffs2_commit_allocation()
  66  *
  67  *      jffs2_reserve_space() may trigger garbage collection in order to make room
  68  *      for the requested allocation.
  69  */
  70
  71 static int jffs2_do_reserve_space(struct jffs2_sb_info *c,  uint32_t minsize,
  72                                   uint32_t *len, uint32_t sumsize);
  73
  74 int jffs2_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
  75                         uint32_t *len, int prio, uint32_t sumsize)
  76 {
  77         int ret = -EAGAIN;
  78         int blocksneeded = c->resv_blocks_write;
  79         /* align it */
  80         minsize = PAD(minsize);
  81
  82         jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize);
  83         mutex_lock(&c->alloc_sem);
  84
  85         jffs2_dbg(1, "%s(): alloc sem got\n", __func__);
  86
  87         spin_lock(&c->erase_completion_lock);
  88
  89         /*
  90          * Check if the free space is greater then size of the reserved pool.
  91          * If not, only allow root to proceed with writing.
  92          */
  93         if (prio != ALLOC_DELETION && !jffs2_rp_can_write(c)) {
  94                 ret = -ENOSPC;
  95                 goto out;
  96         }
  97
  98         /* this needs a little more thought (true <tglx> :)) */
  99         while(ret == -EAGAIN) {
 100                 while(c->nr_free_blocks + c->nr_erasing_blocks < blocksneeded) {
 101                         uint32_t dirty, avail;
 102
 103                         /* calculate real dirty size
 104                          * dirty_size contains blocks on erase_pending_list
 105                          * those blocks are counted in c->nr_erasing_blocks.
 106                          * If one block is actually erased, it is not longer counted as dirty_space
 107                          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
 108                          * with c->nr_erasing_blocks * c->sector_size again.
 109                          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
 110                          * This helps us to force gc and pick eventually a clean block to spread the load.
 111                          * We add unchecked_size here, as we hopefully will find some space to use.
 112                          * This will affect the sum only once, as gc first finishes checking
 113                          * of nodes.
 114                          */
 115                         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size + c->unchecked_size;
 116                         if (dirty < c->nospc_dirty_size) {
 117                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
 118                                         jffs2_dbg(1, "%s(): Low on dirty space to GC, but it's a deletion. Allowing...\n",
 119                                                   __func__);
 120                                         break;
 121                                 }
 122                                 jffs2_dbg(1, "dirty size 0x%08x + unchecked_size 0x%08x < nospc_dirty_size 0x%08x, returning -ENOSPC\n",
 123                                           dirty, c->unchecked_size,
 124                                           c->sector_size);
 125
 126                                 spin_unlock(&c->erase_completion_lock);
 127                                 mutex_unlock(&c->alloc_sem);
 128                                 return -ENOSPC;
 129                         }
 130
 131                         /* Calc possibly available space. Possibly available means that we
 132                          * don't know, if unchecked size contains obsoleted nodes, which could give us some
 133                          * more usable space. This will affect the sum only once, as gc first finishes checking
 134                          * of nodes.
 135                          + Return -ENOSPC, if the maximum possibly available space is less or equal than
 136                          * blocksneeded * sector_size.
 137                          * This blocks endless gc looping on a filesystem, which is nearly full, even if
 138                          * the check above passes.
 139                          */
 140                         avail = c->free_size + c->dirty_size + c->erasing_size + c->unchecked_size;
 141                         if ( (avail / c->sector_size) <= blocksneeded) {
 142                                 if (prio == ALLOC_DELETION && c->nr_free_blocks + c->nr_erasing_blocks >= c->resv_blocks_deletion) {
 143                                         jffs2_dbg(1, "%s(): Low on possibly available space, but it's a deletion. Allowing...\n",
 144                                                   __func__);
 145                                         break;
 146                                 }
 147
 148                                 jffs2_dbg(1, "max. available size 0x%08x  < blocksneeded * sector_size 0x%08x, returning -ENOSPC\n",
 149                                           avail, blocksneeded * c->sector_size);
 150                                 spin_unlock(&c->erase_completion_lock);
 151                                 mutex_unlock(&c->alloc_sem);
 152                                 return -ENOSPC;
 153                         }
 154
 155                         mutex_unlock(&c->alloc_sem);
 156
 157                         jffs2_dbg(1, "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",
 158                                   c->nr_free_blocks, c->nr_erasing_blocks,
 159                                   c->free_size, c->dirty_size, c->wasted_size,
 160                                   c->used_size, c->erasing_size, c->bad_size,
 161                                   c->free_size + c->dirty_size +
 162                                   c->wasted_size + c->used_size +
 163                                   c->erasing_size + c->bad_size,
 164                                   c->flash_size);
 165                         spin_unlock(&c->erase_completion_lock);
 166
 167                         ret = jffs2_garbage_collect_pass(c);
 168
 169                         if (ret == -EAGAIN) {
 170                                 spin_lock(&c->erase_completion_lock);
 171                                 if (c->nr_erasing_blocks &&
 172                                     list_empty(&c->erase_pending_list) &&
 173                                     list_empty(&c->erase_complete_list)) {
 174                                         DECLARE_WAITQUEUE(wait, current);
 175                                         set_current_state(TASK_UNINTERRUPTIBLE);
 176                                         add_wait_queue(&c->erase_wait, &wait);
 177                                         jffs2_dbg(1, "%s waiting for erase to complete\n",
 178                                                   __func__);
 179                                         spin_unlock(&c->erase_completion_lock);
 180
 181                                         schedule();
 182                                         remove_wait_queue(&c->erase_wait, &wait);
 183                                 } else
 184                                         spin_unlock(&c->erase_completion_lock);
 185                         } else if (ret)
 186                                 return ret;
 187
 188                         cond_resched();
 189
 190                         if (signal_pending(current))
 191                                 return -EINTR;
 192
 193                         mutex_lock(&c->alloc_sem);
 194                         spin_lock(&c->erase_completion_lock);
 195                 }
 196
 197                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
 198                 if (ret) {
 199                         jffs2_dbg(1, "%s(): ret is %d\n", __func__, ret);
 200                 }
 201         }
 202
 203 out:
 204         spin_unlock(&c->erase_completion_lock);
 205         if (!ret)
 206                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
 207         if (ret)
 208                 mutex_unlock(&c->alloc_sem);
 209         return ret;
 210 }
 211
 212 int jffs2_reserve_space_gc(struct jffs2_sb_info *c, uint32_t minsize,
 213                            uint32_t *len, uint32_t sumsize)
 214 {
 215         int ret;
 216         minsize = PAD(minsize);
 217
 218         jffs2_dbg(1, "%s(): Requested 0x%x bytes\n", __func__, minsize);
 219
 220         while (true) {
 221                 spin_lock(&c->erase_completion_lock);
 222                 ret = jffs2_do_reserve_space(c, minsize, len, sumsize);
 223                 if (ret) {
 224                         jffs2_dbg(1, "%s(): looping, ret is %d\n",
 225                                   __func__, ret);
 226                 }
 227                 spin_unlock(&c->erase_completion_lock);
 228
 229                 if (ret == -EAGAIN)
 230                         cond_resched();
 231                 else
 232                         break;
 233         }
 234         if (!ret)
 235                 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, 1);
 236
 237         return ret;
 238 }
 239
 240
 241 /* Classify nextblock (clean, dirty of verydirty) and force to select an other one */
 242
 243 static void jffs2_close_nextblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
 244 {
 245
 246         if (c->nextblock == NULL) {
 247                 jffs2_dbg(1, "%s(): Erase block at 0x%08x has already been placed in a list\n",
 248                           __func__, jeb->offset);
 249                 return;
 250         }
 251         /* Check, if we have a dirty block now, or if it was dirty already */
 252         if (ISDIRTY (jeb->wasted_size + jeb->dirty_size)) {
 253                 c->dirty_size += jeb->wasted_size;
 254                 c->wasted_size -= jeb->wasted_size;
 255                 jeb->dirty_size += jeb->wasted_size;
 256                 jeb->wasted_size = 0;
 257                 if (VERYDIRTY(c, jeb->dirty_size)) {
 258                         jffs2_dbg(1, "Adding full erase block at 0x%08x to very_dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 259                                   jeb->offset, jeb->free_size, jeb->dirty_size,
 260                                   jeb->used_size);
 261                         list_add_tail(&jeb->list, &c->very_dirty_list);
 262                 } else {
 263                         jffs2_dbg(1, "Adding full erase block at 0x%08x to dirty_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 264                                   jeb->offset, jeb->free_size, jeb->dirty_size,
 265                                   jeb->used_size);
 266                         list_add_tail(&jeb->list, &c->dirty_list);
 267                 }
 268         } else {
 269                 jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 270                           jeb->offset, jeb->free_size, jeb->dirty_size,
 271                           jeb->used_size);
 272                 list_add_tail(&jeb->list, &c->clean_list);
 273         }
 274         c->nextblock = NULL;
 275
 276 }
 277
 278 /* Select a new jeb for nextblock */
 279
 280 static int jffs2_find_nextblock(struct jffs2_sb_info *c)
 281 {
 282         struct list_head *next;
 283
 284         /* Take the next block off the 'free' list */
 285
 286         if (list_empty(&c->free_list)) {
 287
 288                 if (!c->nr_erasing_blocks &&
 289                         !list_empty(&c->erasable_list)) {
 290                         struct jffs2_eraseblock *ejeb;
 291
 292                         ejeb = list_entry(c->erasable_list.next, struct jffs2_eraseblock, list);
 293                         list_move_tail(&ejeb->list, &c->erase_pending_list);
 294                         c->nr_erasing_blocks++;
 295                         jffs2_garbage_collect_trigger(c);
 296                         jffs2_dbg(1, "%s(): Triggering erase of erasable block at 0x%08x\n",
 297                                   __func__, ejeb->offset);
 298                 }
 299
 300                 if (!c->nr_erasing_blocks &&
 301                         !list_empty(&c->erasable_pending_wbuf_list)) {
 302                         jffs2_dbg(1, "%s(): Flushing write buffer\n",
 303                                   __func__);
 304                         /* c->nextblock is NULL, no update to c->nextblock allowed */
 305                         spin_unlock(&c->erase_completion_lock);
 306                         jffs2_flush_wbuf_pad(c);
 307                         spin_lock(&c->erase_completion_lock);
 308                         /* Have another go. It'll be on the erasable_list now */
 309                         return -EAGAIN;
 310                 }
 311
 312                 if (!c->nr_erasing_blocks) {
 313                         /* Ouch. We're in GC, or we wouldn't have got here.
 314                            And there's no space left. At all. */
 315                         pr_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",
 316                                 c->nr_erasing_blocks, c->nr_free_blocks,
 317                                 list_empty(&c->erasable_list) ? "yes" : "no",
 318                                 list_empty(&c->erasing_list) ? "yes" : "no",
 319                                 list_empty(&c->erase_pending_list) ? "yes" : "no");
 320                         return -ENOSPC;
 321                 }
 322
 323                 spin_unlock(&c->erase_completion_lock);
 324                 /* Don't wait for it; just erase one right now */
 325                 jffs2_erase_pending_blocks(c, 1);
 326                 spin_lock(&c->erase_completion_lock);
 327
 328                 /* An erase may have failed, decreasing the
 329                    amount of free space available. So we must
 330                    restart from the beginning */
 331                 return -EAGAIN;
 332         }
 333
 334         next = c->free_list.next;
 335         list_del(next);
 336         c->nextblock = list_entry(next, struct jffs2_eraseblock, list);
 337         c->nr_free_blocks--;
 338
 339         jffs2_sum_reset_collected(c->summary); /* reset collected summary */
 340
 341 #ifdef CONFIG_JFFS2_FS_WRITEBUFFER
 342         /* adjust write buffer offset, else we get a non contiguous write bug */
 343         if (!(c->wbuf_ofs % c->sector_size) && !c->wbuf_len)
 344                 c->wbuf_ofs = 0xffffffff;
 345 #endif
 346
 347         jffs2_dbg(1, "%s(): new nextblock = 0x%08x\n",
 348                   __func__, c->nextblock->offset);
 349
 350         return 0;
 351 }
 352
 353 /* Called with alloc sem _and_ erase_completion_lock */
 354 static int jffs2_do_reserve_space(struct jffs2_sb_info *c, uint32_t minsize,
 355                                   uint32_t *len, uint32_t sumsize)
 356 {
 357         struct jffs2_eraseblock *jeb = c->nextblock;
 358         uint32_t reserved_size;                         /* for summary information at the end of the jeb */
 359         int ret;
 360
 361  restart:
 362         reserved_size = 0;
 363
 364         if (jffs2_sum_active() && (sumsize != JFFS2_SUMMARY_NOSUM_SIZE)) {
 365                                                         /* NOSUM_SIZE means not to generate summary */
 366
 367                 if (jeb) {
 368                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
 369                         dbg_summary("minsize=%d , jeb->free=%d ,"
 370                                                 "summary->size=%d , sumsize=%d\n",
 371                                                 minsize, jeb->free_size,
 372                                                 c->summary->sum_size, sumsize);
 373                 }
 374
 375                 /* Is there enough space for writing out the current node, or we have to
 376                    write out summary information now, close this jeb and select new nextblock? */
 377                 if (jeb && (PAD(minsize) + PAD(c->summary->sum_size + sumsize +
 378                                         JFFS2_SUMMARY_FRAME_SIZE) > jeb->free_size)) {
 379
 380                         /* Has summary been disabled for this jeb? */
 381                         if (jffs2_sum_is_disabled(c->summary)) {
 382                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
 383                                 goto restart;
 384                         }
 385
 386                         /* Writing out the collected summary information */
 387                         dbg_summary("generating summary for 0x%08x.\n", jeb->offset);
 388                         ret = jffs2_sum_write_sumnode(c);
 389
 390                         if (ret)
 391                                 return ret;
 392
 393                         if (jffs2_sum_is_disabled(c->summary)) {
 394                                 /* jffs2_write_sumnode() couldn't write out the summary information
 395                                    diabling summary for this jeb and free the collected information
 396                                  */
 397                                 sumsize = JFFS2_SUMMARY_NOSUM_SIZE;
 398                                 goto restart;
 399                         }
 400
 401                         jffs2_close_nextblock(c, jeb);
 402                         jeb = NULL;
 403                         /* keep always valid value in reserved_size */
 404                         reserved_size = PAD(sumsize + c->summary->sum_size + JFFS2_SUMMARY_FRAME_SIZE);
 405                 }
 406         } else {
 407                 if (jeb && minsize > jeb->free_size) {
 408                         uint32_t waste;
 409
 410                         /* Skip the end of this block and file it as having some dirty space */
 411                         /* If there's a pending write to it, flush now */
 412
 413                         if (jffs2_wbuf_dirty(c)) {
 414                                 spin_unlock(&c->erase_completion_lock);
 415                                 jffs2_dbg(1, "%s(): Flushing write buffer\n",
 416                                           __func__);
 417                                 jffs2_flush_wbuf_pad(c);
 418                                 spin_lock(&c->erase_completion_lock);
 419                                 jeb = c->nextblock;
 420                                 goto restart;
 421                         }
 422
 423                         spin_unlock(&c->erase_completion_lock);
 424
 425                         ret = jffs2_prealloc_raw_node_refs(c, jeb, 1);
 426
 427                         /* Just lock it again and continue. Nothing much can change because
 428                            we hold c->alloc_sem anyway. In fact, it's not entirely clear why
 429                            we hold c->erase_completion_lock in the majority of this function...
 430                            but that's a question for another (more caffeine-rich) day. */
 431                         spin_lock(&c->erase_completion_lock);
 432
 433                         if (ret)
 434                                 return ret;
 435
 436                         waste = jeb->free_size;
 437                         jffs2_link_node_ref(c, jeb,
 438                                             (jeb->offset + c->sector_size - waste) | REF_OBSOLETE,
 439                                             waste, NULL);
 440                         /* FIXME: that made it count as dirty. Convert to wasted */
 441                         jeb->dirty_size -= waste;
 442                         c->dirty_size -= waste;
 443                         jeb->wasted_size += waste;
 444                         c->wasted_size += waste;
 445
 446                         jffs2_close_nextblock(c, jeb);
 447                         jeb = NULL;
 448                 }
 449         }
 450
 451         if (!jeb) {
 452
 453                 ret = jffs2_find_nextblock(c);
 454                 if (ret)
 455                         return ret;
 456
 457                 jeb = c->nextblock;
 458
 459                 if (jeb->free_size != c->sector_size - c->cleanmarker_size) {
 460                         pr_warn("Eep. Block 0x%08x taken from free_list had free_size of 0x%08x!!\n",
 461                                 jeb->offset, jeb->free_size);
 462                         goto restart;
 463                 }
 464         }
 465         /* OK, jeb (==c->nextblock) is now pointing at a block which definitely has
 466            enough space */
 467         *len = jeb->free_size - reserved_size;
 468
 469         if (c->cleanmarker_size && jeb->used_size == c->cleanmarker_size &&
 470             !jeb->first_node->next_in_ino) {
 471                 /* Only node in it beforehand was a CLEANMARKER node (we think).
 472                    So mark it obsolete now that there's going to be another node
 473                    in the block. This will reduce used_size to zero but We've
 474                    already set c->nextblock so that jffs2_mark_node_obsolete()
 475                    won't try to refile it to the dirty_list.
 476                 */
 477                 spin_unlock(&c->erase_completion_lock);
 478                 jffs2_mark_node_obsolete(c, jeb->first_node);
 479                 spin_lock(&c->erase_completion_lock);
 480         }
 481
 482         jffs2_dbg(1, "%s(): Giving 0x%x bytes at 0x%x\n",
 483                   __func__,
 484                   *len, jeb->offset + (c->sector_size - jeb->free_size));
 485         return 0;
 486 }
 487
 488 /**
 489  *      jffs2_add_physical_node_ref - add a physical node reference to the list
 490  *      @c: superblock info
 491  *      @new: new node reference to add
 492  *      @len: length of this physical node
 493  *
 494  *      Should only be used to report nodes for which space has been allocated
 495  *      by jffs2_reserve_space.
 496  *
 497  *      Must be called with the alloc_sem held.
 498  */
 499
 500 struct jffs2_raw_node_ref *jffs2_add_physical_node_ref(struct jffs2_sb_info *c,
 501                                                        uint32_t ofs, uint32_t len,
 502                                                        struct jffs2_inode_cache *ic)
 503 {
 504         struct jffs2_eraseblock *jeb;
 505         struct jffs2_raw_node_ref *new;
 506
 507         jeb = &c->blocks[ofs / c->sector_size];
 508
 509         jffs2_dbg(1, "%s(): Node at 0x%x(%d), size 0x%x\n",
 510                   __func__, ofs & ~3, ofs & 3, len);
 511 #if 1
 512         /* Allow non-obsolete nodes only to be added at the end of c->nextblock,
 513            if c->nextblock is set. Note that wbuf.c will file obsolete nodes
 514            even after refiling c->nextblock */
 515         if ((c->nextblock || ((ofs & 3) != REF_OBSOLETE))
 516             && (jeb != c->nextblock || (ofs & ~3) != jeb->offset + (c->sector_size - jeb->free_size))) {
 517                 pr_warn("argh. node added in wrong place at 0x%08x(%d)\n",
 518                         ofs & ~3, ofs & 3);
 519                 if (c->nextblock)
 520                         pr_warn("nextblock 0x%08x", c->nextblock->offset);
 521                 else
 522                         pr_warn("No nextblock");
 523                 pr_cont(", expected at %08x\n",
 524                         jeb->offset + (c->sector_size - jeb->free_size));
 525                 return ERR_PTR(-EINVAL);
 526         }
 527 #endif
 528         spin_lock(&c->erase_completion_lock);
 529
 530         new = jffs2_link_node_ref(c, jeb, ofs, len, ic);
 531
 532         if (!jeb->free_size && !jeb->dirty_size && !ISDIRTY(jeb->wasted_size)) {
 533                 /* If it lives on the dirty_list, jffs2_reserve_space will put it there */
 534                 jffs2_dbg(1, "Adding full erase block at 0x%08x to clean_list (free 0x%08x, dirty 0x%08x, used 0x%08x\n",
 535                           jeb->offset, jeb->free_size, jeb->dirty_size,
 536                           jeb->used_size);
 537                 if (jffs2_wbuf_dirty(c)) {
 538                         /* Flush the last write in the block if it's outstanding */
 539                         spin_unlock(&c->erase_completion_lock);
 540                         jffs2_flush_wbuf_pad(c);
 541                         spin_lock(&c->erase_completion_lock);
 542                 }
 543
 544                 list_add_tail(&jeb->list, &c->clean_list);
 545                 c->nextblock = NULL;
 546         }
 547         jffs2_dbg_acct_sanity_check_nolock(c,jeb);
 548         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
 549
 550         spin_unlock(&c->erase_completion_lock);
 551
 552         return new;
 553 }
 554
 555
 556 void jffs2_complete_reservation(struct jffs2_sb_info *c)
 557 {
 558         jffs2_dbg(1, "jffs2_complete_reservation()\n");
 559         spin_lock(&c->erase_completion_lock);
 560         jffs2_garbage_collect_trigger(c);
 561         spin_unlock(&c->erase_completion_lock);
 562         mutex_unlock(&c->alloc_sem);
 563 }
 564
 565 static inline int on_list(struct list_head *obj, struct list_head *head)
 566 {
 567         struct list_head *this;
 568
 569         list_for_each(this, head) {
 570                 if (this == obj) {
 571                         jffs2_dbg(1, "%p is on list at %p\n", obj, head);
 572                         return 1;
 573
 574                 }
 575         }
 576         return 0;
 577 }
 578
 579 void jffs2_mark_node_obsolete(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref)
 580 {
 581         struct jffs2_eraseblock *jeb;
 582         int blocknr;
 583         struct jffs2_unknown_node n;
 584         int ret, addedsize;
 585         size_t retlen;
 586         uint32_t freed_len;
 587
 588         if(unlikely(!ref)) {
 589                 pr_notice("EEEEEK. jffs2_mark_node_obsolete called with NULL node\n");
 590                 return;
 591         }
 592         if (ref_obsolete(ref)) {
 593                 jffs2_dbg(1, "%s(): called with already obsolete node at 0x%08x\n",
 594                           __func__, ref_offset(ref));
 595                 return;
 596         }
 597         blocknr = ref->flash_offset / c->sector_size;
 598         if (blocknr >= c->nr_blocks) {
 599                 pr_notice("raw node at 0x%08x is off the end of device!\n",
 600                           ref->flash_offset);
 601                 BUG();
 602         }
 603         jeb = &c->blocks[blocknr];
 604
 605         if (jffs2_can_mark_obsolete(c) && !jffs2_is_readonly(c) &&
 606             !(c->flags & (JFFS2_SB_FLAG_SCANNING | JFFS2_SB_FLAG_BUILDING))) {
 607                 /* Hm. This may confuse static lock analysis. If any of the above
 608                    three conditions is false, we're going to return from this
 609                    function without actually obliterating any nodes or freeing
 610                    any jffs2_raw_node_refs. So we don't need to stop erases from
 611                    happening, or protect against people holding an obsolete
 612                    jffs2_raw_node_ref without the erase_completion_lock. */
 613                 mutex_lock(&c->erase_free_sem);
 614         }
 615
 616         spin_lock(&c->erase_completion_lock);
 617
 618         freed_len = ref_totlen(c, jeb, ref);
 619
 620         if (ref_flags(ref) == REF_UNCHECKED) {
 621                 D1(if (unlikely(jeb->unchecked_size < freed_len)) {
 622                                 pr_notice("raw unchecked node of size 0x%08x freed from erase block %d at 0x%08x, but unchecked_size was already 0x%08x\n",
 623                                           freed_len, blocknr,
 624                                           ref->flash_offset, jeb->used_size);
 625                         BUG();
 626                 })
 627                         jffs2_dbg(1, "Obsoleting previously unchecked node at 0x%08x of len %x\n",
 628                                   ref_offset(ref), freed_len);
 629                 jeb->unchecked_size -= freed_len;
 630                 c->unchecked_size -= freed_len;
 631         } else {
 632                 D1(if (unlikely(jeb->used_size < freed_len)) {
 633                                 pr_notice("raw node of size 0x%08x freed from erase block %d at 0x%08x, but used_size was already 0x%08x\n",
 634                                           freed_len, blocknr,
 635                                           ref->flash_offset, jeb->used_size);
 636                         BUG();
 637                 })
 638                         jffs2_dbg(1, "Obsoleting node at 0x%08x of len %#x: ",
 639                                   ref_offset(ref), freed_len);
 640                 jeb->used_size -= freed_len;
 641                 c->used_size -= freed_len;
 642         }
 643
 644         // Take care, that wasted size is taken into concern
 645         if ((jeb->dirty_size || ISDIRTY(jeb->wasted_size + freed_len)) && jeb != c->nextblock) {
 646                 jffs2_dbg(1, "Dirtying\n");
 647                 addedsize = freed_len;
 648                 jeb->dirty_size += freed_len;
 649                 c->dirty_size += freed_len;
 650
 651                 /* Convert wasted space to dirty, if not a bad block */
 652                 if (jeb->wasted_size) {
 653                         if (on_list(&jeb->list, &c->bad_used_list)) {
 654                                 jffs2_dbg(1, "Leaving block at %08x on the bad_used_list\n",
 655                                           jeb->offset);
 656                                 addedsize = 0; /* To fool the refiling code later */
 657                         } else {
 658                                 jffs2_dbg(1, "Converting %d bytes of wasted space to dirty in block at %08x\n",
 659                                           jeb->wasted_size, jeb->offset);
 660                                 addedsize += jeb->wasted_size;
 661                                 jeb->dirty_size += jeb->wasted_size;
 662                                 c->dirty_size += jeb->wasted_size;
 663                                 c->wasted_size -= jeb->wasted_size;
 664                                 jeb->wasted_size = 0;
 665                         }
 666                 }
 667         } else {
 668                 jffs2_dbg(1, "Wasting\n");
 669                 addedsize = 0;
 670                 jeb->wasted_size += freed_len;
 671                 c->wasted_size += freed_len;
 672         }
 673         ref->flash_offset = ref_offset(ref) | REF_OBSOLETE;
 674
 675         jffs2_dbg_acct_sanity_check_nolock(c, jeb);
 676         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
 677
 678         if (c->flags & JFFS2_SB_FLAG_SCANNING) {
 679                 /* Flash scanning is in progress. Don't muck about with the block
 680                    lists because they're not ready yet, and don't actually
 681                    obliterate nodes that look obsolete. If they weren't
 682                    marked obsolete on the flash at the time they _became_
 683                    obsolete, there was probably a reason for that. */
 684                 spin_unlock(&c->erase_completion_lock);
 685                 /* We didn't lock the erase_free_sem */
 686                 return;
 687         }
 688
 689         if (jeb == c->nextblock) {
 690                 jffs2_dbg(2, "Not moving nextblock 0x%08x to dirty/erase_pending list\n",
 691                           jeb->offset);
 692         } else if (!jeb->used_size && !jeb->unchecked_size) {
 693                 if (jeb == c->gcblock) {
 694                         jffs2_dbg(1, "gcblock at 0x%08x completely dirtied. Clearing gcblock...\n",
 695                                   jeb->offset);
 696                         c->gcblock = NULL;
 697                 } else {
 698                         jffs2_dbg(1, "Eraseblock at 0x%08x completely dirtied. Removing from (dirty?) list...\n",
 699                                   jeb->offset);
 700                         list_del(&jeb->list);
 701                 }
 702                 if (jffs2_wbuf_dirty(c)) {
 703                         jffs2_dbg(1, "...and adding to erasable_pending_wbuf_list\n");
 704                         list_add_tail(&jeb->list, &c->erasable_pending_wbuf_list);
 705                 } else {
 706                         if (jiffies & 127) {
 707                                 /* Most of the time, we just erase it immediately. Otherwise we
 708                                    spend ages scanning it on mount, etc. */
 709                                 jffs2_dbg(1, "...and adding to erase_pending_list\n");
 710                                 list_add_tail(&jeb->list, &c->erase_pending_list);
 711                                 c->nr_erasing_blocks++;
 712                                 jffs2_garbage_collect_trigger(c);
 713                         } else {
 714                                 /* Sometimes, however, we leave it elsewhere so it doesn't get
 715                                    immediately reused, and we spread the load a bit. */
 716                                 jffs2_dbg(1, "...and adding to erasable_list\n");
 717                                 list_add_tail(&jeb->list, &c->erasable_list);
 718                         }
 719                 }
 720                 jffs2_dbg(1, "Done OK\n");
 721         } else if (jeb == c->gcblock) {
 722                 jffs2_dbg(2, "Not moving gcblock 0x%08x to dirty_list\n",
 723                           jeb->offset);
 724         } else if (ISDIRTY(jeb->dirty_size) && !ISDIRTY(jeb->dirty_size - addedsize)) {
 725                 jffs2_dbg(1, "Eraseblock at 0x%08x is freshly dirtied. Removing from clean list...\n",
 726                           jeb->offset);
 727                 list_del(&jeb->list);
 728                 jffs2_dbg(1, "...and adding to dirty_list\n");
 729                 list_add_tail(&jeb->list, &c->dirty_list);
 730         } else if (VERYDIRTY(c, jeb->dirty_size) &&
 731                    !VERYDIRTY(c, jeb->dirty_size - addedsize)) {
 732                 jffs2_dbg(1, "Eraseblock at 0x%08x is now very dirty. Removing from dirty list...\n",
 733                           jeb->offset);
 734                 list_del(&jeb->list);
 735                 jffs2_dbg(1, "...and adding to very_dirty_list\n");
 736                 list_add_tail(&jeb->list, &c->very_dirty_list);
 737         } else {
 738                 jffs2_dbg(1, "Eraseblock at 0x%08x not moved anywhere. (free 0x%08x, dirty 0x%08x, used 0x%08x)\n",
 739                           jeb->offset, jeb->free_size, jeb->dirty_size,
 740                           jeb->used_size);
 741         }
 742
 743         spin_unlock(&c->erase_completion_lock);
 744
 745         if (!jffs2_can_mark_obsolete(c) || jffs2_is_readonly(c) ||
 746                 (c->flags & JFFS2_SB_FLAG_BUILDING)) {
 747                 /* We didn't lock the erase_free_sem */
 748                 return;
 749         }
 750
 751         /* The erase_free_sem is locked, and has been since before we marked the node obsolete
 752            and potentially put its eraseblock onto the erase_pending_list. Thus, we know that
 753            the block hasn't _already_ been erased, and that 'ref' itself hasn't been freed yet
 754            by jffs2_free_jeb_node_refs() in erase.c. Which is nice. */
 755
 756         jffs2_dbg(1, "obliterating obsoleted node at 0x%08x\n",
 757                   ref_offset(ref));
 758         ret = jffs2_flash_read(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
 759         if (ret) {
 760                 pr_warn("Read error reading from obsoleted node at 0x%08x: %d\n",
 761                         ref_offset(ref), ret);
 762                 goto out_erase_sem;
 763         }
 764         if (retlen != sizeof(n)) {
 765                 pr_warn("Short read from obsoleted node at 0x%08x: %zd\n",
 766                         ref_offset(ref), retlen);
 767                 goto out_erase_sem;
 768         }
 769         if (PAD(je32_to_cpu(n.totlen)) != PAD(freed_len)) {
 770                 pr_warn("Node totlen on flash (0x%08x) != totlen from node ref (0x%08x)\n",
 771                         je32_to_cpu(n.totlen), freed_len);
 772                 goto out_erase_sem;
 773         }
 774         if (!(je16_to_cpu(n.nodetype) & JFFS2_NODE_ACCURATE)) {
 775                 jffs2_dbg(1, "Node at 0x%08x was already marked obsolete (nodetype 0x%04x)\n",
 776                           ref_offset(ref), je16_to_cpu(n.nodetype));
 777                 goto out_erase_sem;
 778         }
 779         /* XXX FIXME: This is ugly now */
 780         n.nodetype = cpu_to_je16(je16_to_cpu(n.nodetype) & ~JFFS2_NODE_ACCURATE);
 781         ret = jffs2_flash_write(c, ref_offset(ref), sizeof(n), &retlen, (char *)&n);
 782         if (ret) {
 783                 pr_warn("Write error in obliterating obsoleted node at 0x%08x: %d\n",
 784                         ref_offset(ref), ret);
 785                 goto out_erase_sem;
 786         }
 787         if (retlen != sizeof(n)) {
 788                 pr_warn("Short write in obliterating obsoleted node at 0x%08x: %zd\n",
 789                         ref_offset(ref), retlen);
 790                 goto out_erase_sem;
 791         }
 792
 793         /* Nodes which have been marked obsolete no longer need to be
 794            associated with any inode. Remove them from the per-inode list.
 795
 796            Note we can't do this for NAND at the moment because we need
 797            obsolete dirent nodes to stay on the lists, because of the
 798            horridness in jffs2_garbage_collect_deletion_dirent(). Also
 799            because we delete the inocache, and on NAND we need that to
 800            stay around until all the nodes are actually erased, in order
 801            to stop us from giving the same inode number to another newly
 802            created inode. */
 803         if (ref->next_in_ino) {
 804                 struct jffs2_inode_cache *ic;
 805                 struct jffs2_raw_node_ref **p;
 806
 807                 spin_lock(&c->erase_completion_lock);
 808
 809                 ic = jffs2_raw_ref_to_ic(ref);
 810                 for (p = &ic->nodes; (*p) != ref; p = &((*p)->next_in_ino))
 811                         ;
 812
 813                 *p = ref->next_in_ino;
 814                 ref->next_in_ino = NULL;
 815
 816                 switch (ic->class) {
 817 #ifdef CONFIG_JFFS2_FS_XATTR
 818                         case RAWNODE_CLASS_XATTR_DATUM:
 819                                 jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
 820                                 break;
 821                         case RAWNODE_CLASS_XATTR_REF:
 822                                 jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
 823                                 break;
 824 #endif
 825                         default:
 826                                 if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
 827                                         jffs2_del_ino_cache(c, ic);
 828                                 break;
 829                 }
 830                 spin_unlock(&c->erase_completion_lock);
 831         }
 832
 833  out_erase_sem:
 834         mutex_unlock(&c->erase_free_sem);
 835 }
 836
 837 int jffs2_thread_should_wake(struct jffs2_sb_info *c)
 838 {
 839         int ret = 0;
 840         uint32_t dirty;
 841         int nr_very_dirty = 0;
 842         struct jffs2_eraseblock *jeb;
 843
 844         if (!list_empty(&c->erase_complete_list) ||
 845             !list_empty(&c->erase_pending_list))
 846                 return 1;
 847
 848         if (c->unchecked_size) {
 849                 jffs2_dbg(1, "jffs2_thread_should_wake(): unchecked_size %d, checked_ino #%d\n",
 850                           c->unchecked_size, c->checked_ino);
 851                 return 1;
 852         }
 853
 854         /* dirty_size contains blocks on erase_pending_list
 855          * those blocks are counted in c->nr_erasing_blocks.
 856          * If one block is actually erased, it is not longer counted as dirty_space
 857          * but it is counted in c->nr_erasing_blocks, so we add it and subtract it
 858          * with c->nr_erasing_blocks * c->sector_size again.
 859          * Blocks on erasable_list are counted as dirty_size, but not in c->nr_erasing_blocks
 860          * This helps us to force gc and pick eventually a clean block to spread the load.
 861          */
 862         dirty = c->dirty_size + c->erasing_size - c->nr_erasing_blocks * c->sector_size;
 863
 864         if (c->nr_free_blocks + c->nr_erasing_blocks < c->resv_blocks_gctrigger &&
 865                         (dirty > c->nospc_dirty_size))
 866                 ret = 1;
 867
 868         list_for_each_entry(jeb, &c->very_dirty_list, list) {
 869                 nr_very_dirty++;
 870                 if (nr_very_dirty == c->vdirty_blocks_gctrigger) {
 871                         ret = 1;
 872                         /* In debug mode, actually go through and count them all */
 873                         D1(continue);
 874                         break;
 875                 }
 876         }
 877
 878         jffs2_dbg(1, "%s(): nr_free_blocks %d, nr_erasing_blocks %d, dirty_size 0x%x, vdirty_blocks %d: %s\n",
 879                   __func__, c->nr_free_blocks, c->nr_erasing_blocks,
 880                   c->dirty_size, nr_very_dirty, ret ? "yes" : "no");
 881
 882         return ret;
 883 }