fs/jffs2/build.c

   1 /*
   2  * JFFS2 -- Journalling Flash File System, Version 2.
   3  *
   4  * Copyright © 2001-2007 Red Hat, Inc.
   5  * Copyright © 2004-2010 David Woodhouse <dwmw2@infradead.org>
   6  *
   7  * Created by David Woodhouse <dwmw2@infradead.org>
   8  *
   9  * For licensing information, see the file 'LICENCE' in this directory.
  10  *
  11  */
  12
  13 #include <linux/kernel.h>
  14 #include <linux/sched.h>
  15 #include <linux/slab.h>
  16 #include <linux/vmalloc.h>
  17 #include <linux/mtd/mtd.h>
  18 #include "nodelist.h"
  19
  20 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *,
  21                 struct jffs2_inode_cache *, struct jffs2_full_dirent **);
  22
  23 static inline struct jffs2_inode_cache *
  24 first_inode_chain(int *i, struct jffs2_sb_info *c)
  25 {
  26         for (; *i < c->inocache_hashsize; (*i)++) {
  27                 if (c->inocache_list[*i])
  28                         return c->inocache_list[*i];
  29         }
  30         return NULL;
  31 }
  32
  33 static inline struct jffs2_inode_cache *
  34 next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
  35 {
  36         /* More in this chain? */
  37         if (ic->next)
  38                 return ic->next;
  39         (*i)++;
  40         return first_inode_chain(i, c);
  41 }
  42
  43 #define for_each_inode(i, c, ic)                        \
  44         for (i = 0, ic = first_inode_chain(&i, (c));    \
  45              ic;                                        \
  46              ic = next_inode(&i, ic, (c)))
  47
  48
  49 static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
  50                                     struct jffs2_inode_cache *ic,
  51                                     int *dir_hardlinks)
  52 {
  53         struct jffs2_full_dirent *fd;
  54
  55         dbg_fsbuild("building directory inode #%u\n", ic->ino);
  56
  57         /* For each child, increase nlink */
  58         for(fd = ic->scan_dents; fd; fd = fd->next) {
  59                 struct jffs2_inode_cache *child_ic;
  60                 if (!fd->ino)
  61                         continue;
  62
  63                 /* we can get high latency here with huge directories */
  64
  65                 child_ic = jffs2_get_ino_cache(c, fd->ino);
  66                 if (!child_ic) {
  67                         dbg_fsbuild("child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
  68                                   fd->name, fd->ino, ic->ino);
  69                         jffs2_mark_node_obsolete(c, fd->raw);
  70                         /* Clear the ic/raw union so it doesn't cause problems later. */
  71                         fd->ic = NULL;
  72                         continue;
  73                 }
  74
  75                 /* From this point, fd->raw is no longer used so we can set fd->ic */
  76                 fd->ic = child_ic;
  77                 child_ic->pino_nlink++;
  78                 /* If we appear (at this stage) to have hard-linked directories,
  79                  * set a flag to trigger a scan later */
  80                 if (fd->type == DT_DIR) {
  81                         child_ic->flags |= INO_FLAGS_IS_DIR;
  82                         if (child_ic->pino_nlink > 1)
  83                                 *dir_hardlinks = 1;
  84                 }
  85
  86                 dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
  87                 /* Can't free scan_dents so far. We might need them in pass 2 */
  88         }
  89 }
  90
  91 /* Scan plan:
  92  - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
  93  - Scan directory tree from top down, setting nlink in inocaches
  94  - Scan inocaches for inodes with nlink==0
  95 */
  96 static int jffs2_build_filesystem(struct jffs2_sb_info *c)
  97 {
  98         int ret, i, dir_hardlinks = 0;
  99         struct jffs2_inode_cache *ic;
 100         struct jffs2_full_dirent *fd;
 101         struct jffs2_full_dirent *dead_fds = NULL;
 102
 103         dbg_fsbuild("build FS data structures\n");
 104
 105         /* First, scan the medium and build all the inode caches with
 106            lists of physical nodes */
 107
 108         c->flags |= JFFS2_SB_FLAG_SCANNING;
 109         ret = jffs2_scan_medium(c);
 110         c->flags &= ~JFFS2_SB_FLAG_SCANNING;
 111         if (ret)
 112                 goto exit;
 113
 114         dbg_fsbuild("scanned flash completely\n");
 115         jffs2_dbg_dump_block_lists_nolock(c);
 116
 117         dbg_fsbuild("pass 1 starting\n");
 118         c->flags |= JFFS2_SB_FLAG_BUILDING;
 119         /* Now scan the directory tree, increasing nlink according to every dirent found. */
 120         for_each_inode(i, c, ic) {
 121                 if (ic->scan_dents) {
 122                         jffs2_build_inode_pass1(c, ic, &dir_hardlinks);
 123                         cond_resched();
 124                 }
 125         }
 126
 127         dbg_fsbuild("pass 1 complete\n");
 128
 129         /* Next, scan for inodes with nlink == 0 and remove them. If
 130            they were directories, then decrement the nlink of their
 131            children too, and repeat the scan. As that's going to be
 132            a fairly uncommon occurrence, it's not so evil to do it this
 133            way. Recursion bad. */
 134         dbg_fsbuild("pass 2 starting\n");
 135
 136         for_each_inode(i, c, ic) {
 137                 if (ic->pino_nlink)
 138                         continue;
 139
 140                 jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
 141                 cond_resched();
 142         }
 143
 144         dbg_fsbuild("pass 2a starting\n");
 145
 146         while (dead_fds) {
 147                 fd = dead_fds;
 148                 dead_fds = fd->next;
 149
 150                 ic = jffs2_get_ino_cache(c, fd->ino);
 151
 152                 if (ic)
 153                         jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
 154                 jffs2_free_full_dirent(fd);
 155         }
 156
 157         dbg_fsbuild("pass 2a complete\n");
 158
 159         if (dir_hardlinks) {
 160                 /* If we detected directory hardlinks earlier, *hopefully*
 161                  * they are gone now because some of the links were from
 162                  * dead directories which still had some old dirents lying
 163                  * around and not yet garbage-collected, but which have
 164                  * been discarded above. So clear the pino_nlink field
 165                  * in each directory, so that the final scan below can
 166                  * print appropriate warnings. */
 167                 for_each_inode(i, c, ic) {
 168                         if (ic->flags & INO_FLAGS_IS_DIR)
 169                                 ic->pino_nlink = 0;
 170                 }
 171         }
 172         dbg_fsbuild("freeing temporary data structures\n");
 173
 174         /* Finally, we can scan again and free the dirent structs */
 175         for_each_inode(i, c, ic) {
 176                 while(ic->scan_dents) {
 177                         fd = ic->scan_dents;
 178                         ic->scan_dents = fd->next;
 179                         /* We do use the pino_nlink field to count nlink of
 180                          * directories during fs build, so set it to the
 181                          * parent ino# now. Now that there's hopefully only
 182                          * one. */
 183                         if (fd->type == DT_DIR) {
 184                                 if (!fd->ic) {
 185                                         /* We'll have complained about it and marked the coresponding
 186                                            raw node obsolete already. Just skip it. */
 187                                         continue;
 188                                 }
 189
 190                                 /* We *have* to have set this in jffs2_build_inode_pass1() */
 191                                 BUG_ON(!(fd->ic->flags & INO_FLAGS_IS_DIR));
 192
 193                                 /* We clear ic->pino_nlink ∀ directories' ic *only* if dir_hardlinks
 194                                  * is set. Otherwise, we know this should never trigger anyway, so
 195                                  * we don't do the check. And ic->pino_nlink still contains the nlink
 196                                  * value (which is 1). */
 197                                 if (dir_hardlinks && fd->ic->pino_nlink) {
 198                                         JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u is also hard linked from dir ino #%u\n",
 199                                                     fd->name, fd->ino, ic->ino, fd->ic->pino_nlink);
 200                                         /* Should we unlink it from its previous parent? */
 201                                 }
 202
 203                                 /* For directories, ic->pino_nlink holds that parent inode # */
 204                                 fd->ic->pino_nlink = ic->ino;
 205                         }
 206                         jffs2_free_full_dirent(fd);
 207                 }
 208                 ic->scan_dents = NULL;
 209                 cond_resched();
 210         }
 211         jffs2_build_xattr_subsystem(c);
 212         c->flags &= ~JFFS2_SB_FLAG_BUILDING;
 213
 214         dbg_fsbuild("FS build complete\n");
 215
 216         /* Rotate the lists by some number to ensure wear levelling */
 217         jffs2_rotate_lists(c);
 218
 219         ret = 0;
 220
 221 exit:
 222         if (ret) {
 223                 for_each_inode(i, c, ic) {
 224                         while(ic->scan_dents) {
 225                                 fd = ic->scan_dents;
 226                                 ic->scan_dents = fd->next;
 227                                 jffs2_free_full_dirent(fd);
 228                         }
 229                 }
 230                 jffs2_clear_xattr_subsystem(c);
 231         }
 232
 233         return ret;
 234 }
 235
 236 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *c,
 237                                         struct jffs2_inode_cache *ic,
 238                                         struct jffs2_full_dirent **dead_fds)
 239 {
 240         struct jffs2_raw_node_ref *raw;
 241         struct jffs2_full_dirent *fd;
 242
 243         dbg_fsbuild("removing ino #%u with nlink == zero.\n", ic->ino);
 244
 245         raw = ic->nodes;
 246         while (raw != (void *)ic) {
 247                 struct jffs2_raw_node_ref *next = raw->next_in_ino;
 248                 dbg_fsbuild("obsoleting node at 0x%08x\n", ref_offset(raw));
 249                 jffs2_mark_node_obsolete(c, raw);
 250                 raw = next;
 251         }
 252
 253         if (ic->scan_dents) {
 254                 int whinged = 0;
 255                 dbg_fsbuild("inode #%u was a directory which may have children...\n", ic->ino);
 256
 257                 while(ic->scan_dents) {
 258                         struct jffs2_inode_cache *child_ic;
 259
 260                         fd = ic->scan_dents;
 261                         ic->scan_dents = fd->next;
 262
 263                         if (!fd->ino) {
 264                                 /* It's a deletion dirent. Ignore it */
 265                                 dbg_fsbuild("child \"%s\" is a deletion dirent, skipping...\n", fd->name);
 266                                 jffs2_free_full_dirent(fd);
 267                                 continue;
 268                         }
 269                         if (!whinged)
 270                                 whinged = 1;
 271
 272                         dbg_fsbuild("removing child \"%s\", ino #%u\n", fd->name, fd->ino);
 273
 274                         child_ic = jffs2_get_ino_cache(c, fd->ino);
 275                         if (!child_ic) {
 276                                 dbg_fsbuild("cannot remove child \"%s\", ino #%u, because it doesn't exist\n",
 277                                                 fd->name, fd->ino);
 278                                 jffs2_free_full_dirent(fd);
 279                                 continue;
 280                         }
 281
 282                         /* Reduce nlink of the child. If it's now zero, stick it on the
 283                            dead_fds list to be cleaned up later. Else just free the fd */
 284                         child_ic->pino_nlink--;
 285
 286                         if (!child_ic->pino_nlink) {
 287                                 dbg_fsbuild("inode #%u (\"%s\") now has no links; adding to dead_fds list.\n",
 288                                           fd->ino, fd->name);
 289                                 fd->next = *dead_fds;
 290                                 *dead_fds = fd;
 291                         } else {
 292                                 dbg_fsbuild("inode #%u (\"%s\") has now got nlink %d. Ignoring.\n",
 293                                           fd->ino, fd->name, child_ic->pino_nlink);
 294                                 jffs2_free_full_dirent(fd);
 295                         }
 296                 }
 297         }
 298
 299         /*
 300            We don't delete the inocache from the hash list and free it yet.
 301            The erase code will do that, when all the nodes are completely gone.
 302         */
 303 }
 304
 305 static void jffs2_calc_trigger_levels(struct jffs2_sb_info *c)
 306 {
 307         uint32_t size;
 308
 309         /* Deletion should almost _always_ be allowed. We're fairly
 310            buggered once we stop allowing people to delete stuff
 311            because there's not enough free space... */
 312         c->resv_blocks_deletion = 2;
 313
 314         /* Be conservative about how much space we need before we allow writes.
 315            On top of that which is required for deletia, require an extra 2%
 316            of the medium to be available, for overhead caused by nodes being
 317            split across blocks, etc. */
 318
 319         size = c->flash_size / 50; /* 2% of flash size */
 320         size += c->nr_blocks * 100; /* And 100 bytes per eraseblock */
 321         size += c->sector_size - 1; /* ... and round up */
 322
 323         c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
 324
 325         /* When do we let the GC thread run in the background */
 326
 327         c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
 328
 329         /* When do we allow garbage collection to merge nodes to make
 330            long-term progress at the expense of short-term space exhaustion? */
 331         c->resv_blocks_gcmerge = c->resv_blocks_deletion + 1;
 332
 333         /* When do we allow garbage collection to eat from bad blocks rather
 334            than actually making progress? */
 335         c->resv_blocks_gcbad = 0;//c->resv_blocks_deletion + 2;
 336
 337         /* What number of 'very dirty' eraseblocks do we allow before we
 338            trigger the GC thread even if we don't _need_ the space. When we
 339            can't mark nodes obsolete on the medium, the old dirty nodes cause
 340            performance problems because we have to inspect and discard them. */
 341         c->vdirty_blocks_gctrigger = c->resv_blocks_gctrigger;
 342         if (jffs2_can_mark_obsolete(c))
 343                 c->vdirty_blocks_gctrigger *= 10;
 344
 345         /* If there's less than this amount of dirty space, don't bother
 346            trying to GC to make more space. It'll be a fruitless task */
 347         c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
 348
 349         dbg_fsbuild("JFFS2 trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
 350                   c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks);
 351         dbg_fsbuild("Blocks required to allow deletion:    %d (%d KiB)\n",
 352                   c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024);
 353         dbg_fsbuild("Blocks required to allow writes:      %d (%d KiB)\n",
 354                   c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024);
 355         dbg_fsbuild("Blocks required to quiesce GC thread: %d (%d KiB)\n",
 356                   c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024);
 357         dbg_fsbuild("Blocks required to allow GC merges:   %d (%d KiB)\n",
 358                   c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024);
 359         dbg_fsbuild("Blocks required to GC bad blocks:     %d (%d KiB)\n",
 360                   c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024);
 361         dbg_fsbuild("Amount of dirty space required to GC: %d bytes\n",
 362                   c->nospc_dirty_size);
 363         dbg_fsbuild("Very dirty blocks before GC triggered: %d\n",
 364                   c->vdirty_blocks_gctrigger);
 365 }
 366
 367 int jffs2_do_mount_fs(struct jffs2_sb_info *c)
 368 {
 369         int ret;
 370         int i;
 371         int size;
 372
 373         c->free_size = c->flash_size;
 374         c->nr_blocks = c->flash_size / c->sector_size;
 375         size = sizeof(struct jffs2_eraseblock) * c->nr_blocks;
 376 #ifndef __ECOS
 377         if (jffs2_blocks_use_vmalloc(c))
 378                 c->blocks = vzalloc(size);
 379         else
 380 #endif
 381                 c->blocks = kzalloc(size, GFP_KERNEL);
 382         if (!c->blocks)
 383                 return -ENOMEM;
 384
 385         for (i=0; i<c->nr_blocks; i++) {
 386                 INIT_LIST_HEAD(&c->blocks[i].list);
 387                 c->blocks[i].offset = i * c->sector_size;
 388                 c->blocks[i].free_size = c->sector_size;
 389         }
 390
 391         INIT_LIST_HEAD(&c->clean_list);
 392         INIT_LIST_HEAD(&c->very_dirty_list);
 393         INIT_LIST_HEAD(&c->dirty_list);
 394         INIT_LIST_HEAD(&c->erasable_list);
 395         INIT_LIST_HEAD(&c->erasing_list);
 396         INIT_LIST_HEAD(&c->erase_checking_list);
 397         INIT_LIST_HEAD(&c->erase_pending_list);
 398         INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
 399         INIT_LIST_HEAD(&c->erase_complete_list);
 400         INIT_LIST_HEAD(&c->free_list);
 401         INIT_LIST_HEAD(&c->bad_list);
 402         INIT_LIST_HEAD(&c->bad_used_list);
 403         c->highest_ino = 1;
 404         c->summary = NULL;
 405
 406         ret = jffs2_sum_init(c);
 407         if (ret)
 408                 goto out_free;
 409
 410         if (jffs2_build_filesystem(c)) {
 411                 dbg_fsbuild("build_fs failed\n");
 412                 jffs2_free_ino_caches(c);
 413                 jffs2_free_raw_node_refs(c);
 414                 ret = -EIO;
 415                 goto out_free;
 416         }
 417
 418         jffs2_calc_trigger_levels(c);
 419
 420         return 0;
 421
 422  out_free:
 423 #ifndef __ECOS
 424         if (jffs2_blocks_use_vmalloc(c))
 425                 vfree(c->blocks);
 426         else
 427 #endif
 428                 kfree(c->blocks);
 429
 430         return ret;
 431 }