kernel/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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  15 #include <linux/kernel.h>
  16 #include <linux/sched.h>
  17 #include <linux/slab.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/mtd/mtd.h>
  20 #include "nodelist.h"
  21
  22 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *,
  23                 struct jffs2_inode_cache *, struct jffs2_full_dirent **);
  24
  25 static inline struct jffs2_inode_cache *
  26 first_inode_chain(int *i, struct jffs2_sb_info *c)
  27 {
  28         for (; *i < c->inocache_hashsize; (*i)++) {
  29                 if (c->inocache_list[*i])
  30                         return c->inocache_list[*i];
  31         }
  32         return NULL;
  33 }
  34
  35 static inline struct jffs2_inode_cache *
  36 next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
  37 {
  38         /* More in this chain? */
  39         if (ic->next)
  40                 return ic->next;
  41         (*i)++;
  42         return first_inode_chain(i, c);
  43 }
  44
  45 #define for_each_inode(i, c, ic)                        \
  46         for (i = 0, ic = first_inode_chain(&i, (c));    \
  47              ic;                                        \
  48              ic = next_inode(&i, ic, (c)))
  49
  50
  51 static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
  52                                     struct jffs2_inode_cache *ic,
  53                                     int *dir_hardlinks)
  54 {
  55         struct jffs2_full_dirent *fd;
  56
  57         dbg_fsbuild("building directory inode #%u\n", ic->ino);
  58
  59         /* For each child, increase nlink */
  60         for(fd = ic->scan_dents; fd; fd = fd->next) {
  61                 struct jffs2_inode_cache *child_ic;
  62                 if (!fd->ino)
  63                         continue;
  64
  65                 /* we can get high latency here with huge directories */
  66
  67                 child_ic = jffs2_get_ino_cache(c, fd->ino);
  68                 if (!child_ic) {
  69                         dbg_fsbuild("child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
  70                                   fd->name, fd->ino, ic->ino);
  71                         jffs2_mark_node_obsolete(c, fd->raw);
  72                         /* Clear the ic/raw union so it doesn't cause problems later. */
  73                         fd->ic = NULL;
  74                         continue;
  75                 }
  76
  77                 /* From this point, fd->raw is no longer used so we can set fd->ic */
  78                 fd->ic = child_ic;
  79                 child_ic->pino_nlink++;
  80                 /* If we appear (at this stage) to have hard-linked directories,
  81                  * set a flag to trigger a scan later */
  82                 if (fd->type == DT_DIR) {
  83                         child_ic->flags |= INO_FLAGS_IS_DIR;
  84                         if (child_ic->pino_nlink > 1)
  85                                 *dir_hardlinks = 1;
  86                 }
  87
  88                 dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
  89                 /* Can't free scan_dents so far. We might need them in pass 2 */
  90         }
  91 }
  92
  93 /* Scan plan:
  94  - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
  95  - Scan directory tree from top down, setting nlink in inocaches
  96  - Scan inocaches for inodes with nlink==0
  97 */
  98 static int jffs2_build_filesystem(struct jffs2_sb_info *c)
  99 {
 100         int ret, i, dir_hardlinks = 0;
 101         struct jffs2_inode_cache *ic;
 102         struct jffs2_full_dirent *fd;
 103         struct jffs2_full_dirent *dead_fds = NULL;
 104
 105         dbg_fsbuild("build FS data structures\n");
 106
 107         /* First, scan the medium and build all the inode caches with
 108            lists of physical nodes */
 109
 110         c->flags |= JFFS2_SB_FLAG_SCANNING;
 111         ret = jffs2_scan_medium(c);
 112         c->flags &= ~JFFS2_SB_FLAG_SCANNING;
 113         if (ret)
 114                 goto exit;
 115
 116         dbg_fsbuild("scanned flash completely\n");
 117         jffs2_dbg_dump_block_lists_nolock(c);
 118
 119         dbg_fsbuild("pass 1 starting\n");
 120         c->flags |= JFFS2_SB_FLAG_BUILDING;
 121         /* Now scan the directory tree, increasing nlink according to every dirent found. */
 122         for_each_inode(i, c, ic) {
 123                 if (ic->scan_dents) {
 124                         jffs2_build_inode_pass1(c, ic, &dir_hardlinks);
 125                         cond_resched();
 126                 }
 127         }
 128
 129         dbg_fsbuild("pass 1 complete\n");
 130
 131         /* Next, scan for inodes with nlink == 0 and remove them. If
 132            they were directories, then decrement the nlink of their
 133            children too, and repeat the scan. As that's going to be
 134            a fairly uncommon occurrence, it's not so evil to do it this
 135            way. Recursion bad. */
 136         dbg_fsbuild("pass 2 starting\n");
 137
 138         for_each_inode(i, c, ic) {
 139                 if (ic->pino_nlink)
 140                         continue;
 141
 142                 jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
 143                 cond_resched();
 144         }
 145
 146         dbg_fsbuild("pass 2a starting\n");
 147
 148         while (dead_fds) {
 149                 fd = dead_fds;
 150                 dead_fds = fd->next;
 151
 152                 ic = jffs2_get_ino_cache(c, fd->ino);
 153
 154                 if (ic)
 155                         jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
 156                 jffs2_free_full_dirent(fd);
 157         }
 158
 159         dbg_fsbuild("pass 2a complete\n");
 160
 161         if (dir_hardlinks) {
 162                 /* If we detected directory hardlinks earlier, *hopefully*
 163                  * they are gone now because some of the links were from
 164                  * dead directories which still had some old dirents lying
 165                  * around and not yet garbage-collected, but which have
 166                  * been discarded above. So clear the pino_nlink field
 167                  * in each directory, so that the final scan below can
 168                  * print appropriate warnings. */
 169                 for_each_inode(i, c, ic) {
 170                         if (ic->flags & INO_FLAGS_IS_DIR)
 171                                 ic->pino_nlink = 0;
 172                 }
 173         }
 174         dbg_fsbuild("freeing temporary data structures\n");
 175
 176         /* Finally, we can scan again and free the dirent structs */
 177         for_each_inode(i, c, ic) {
 178                 while(ic->scan_dents) {
 179                         fd = ic->scan_dents;
 180                         ic->scan_dents = fd->next;
 181                         /* We do use the pino_nlink field to count nlink of
 182                          * directories during fs build, so set it to the
 183                          * parent ino# now. Now that there's hopefully only
 184                          * one. */
 185                         if (fd->type == DT_DIR) {
 186                                 if (!fd->ic) {
 187                                         /* We'll have complained about it and marked the coresponding
 188                                            raw node obsolete already. Just skip it. */
 189                                         continue;
 190                                 }
 191
 192                                 /* We *have* to have set this in jffs2_build_inode_pass1() */
 193                                 BUG_ON(!(fd->ic->flags & INO_FLAGS_IS_DIR));
 194
 195                                 /* We clear ic->pino_nlink ∀ directories' ic *only* if dir_hardlinks
 196                                  * is set. Otherwise, we know this should never trigger anyway, so
 197                                  * we don't do the check. And ic->pino_nlink still contains the nlink
 198                                  * value (which is 1). */
 199                                 if (dir_hardlinks && fd->ic->pino_nlink) {
 200                                         JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u is also hard linked from dir ino #%u\n",
 201                                                     fd->name, fd->ino, ic->ino, fd->ic->pino_nlink);
 202                                         /* Should we unlink it from its previous parent? */
 203                                 }
 204
 205                                 /* For directories, ic->pino_nlink holds that parent inode # */
 206                                 fd->ic->pino_nlink = ic->ino;
 207                         }
 208                         jffs2_free_full_dirent(fd);
 209                 }
 210                 ic->scan_dents = NULL;
 211                 cond_resched();
 212         }
 213         jffs2_build_xattr_subsystem(c);
 214         c->flags &= ~JFFS2_SB_FLAG_BUILDING;
 215
 216         dbg_fsbuild("FS build complete\n");
 217
 218         /* Rotate the lists by some number to ensure wear levelling */
 219         jffs2_rotate_lists(c);
 220
 221         ret = 0;
 222
 223 exit:
 224         if (ret) {
 225                 for_each_inode(i, c, ic) {
 226                         while(ic->scan_dents) {
 227                                 fd = ic->scan_dents;
 228                                 ic->scan_dents = fd->next;
 229                                 jffs2_free_full_dirent(fd);
 230                         }
 231                 }
 232                 jffs2_clear_xattr_subsystem(c);
 233         }
 234
 235         return ret;
 236 }
 237
 238 static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *c,
 239                                         struct jffs2_inode_cache *ic,
 240                                         struct jffs2_full_dirent **dead_fds)
 241 {
 242         struct jffs2_raw_node_ref *raw;
 243         struct jffs2_full_dirent *fd;
 244
 245         dbg_fsbuild("removing ino #%u with nlink == zero.\n", ic->ino);
 246
 247         raw = ic->nodes;
 248         while (raw != (void *)ic) {
 249                 struct jffs2_raw_node_ref *next = raw->next_in_ino;
 250                 dbg_fsbuild("obsoleting node at 0x%08x\n", ref_offset(raw));
 251                 jffs2_mark_node_obsolete(c, raw);
 252                 raw = next;
 253         }
 254
 255         if (ic->scan_dents) {
 256                 int whinged = 0;
 257                 dbg_fsbuild("inode #%u was a directory which may have children...\n", ic->ino);
 258
 259                 while(ic->scan_dents) {
 260                         struct jffs2_inode_cache *child_ic;
 261
 262                         fd = ic->scan_dents;
 263                         ic->scan_dents = fd->next;
 264
 265                         if (!fd->ino) {
 266                                 /* It's a deletion dirent. Ignore it */
 267                                 dbg_fsbuild("child \"%s\" is a deletion dirent, skipping...\n", fd->name);
 268                                 jffs2_free_full_dirent(fd);
 269                                 continue;
 270                         }
 271                         if (!whinged)
 272                                 whinged = 1;
 273
 274                         dbg_fsbuild("removing child \"%s\", ino #%u\n", fd->name, fd->ino);
 275
 276                         child_ic = jffs2_get_ino_cache(c, fd->ino);
 277                         if (!child_ic) {
 278                                 dbg_fsbuild("cannot remove child \"%s\", ino #%u, because it doesn't exist\n",
 279                                                 fd->name, fd->ino);
 280                                 jffs2_free_full_dirent(fd);
 281                                 continue;
 282                         }
 283
 284                         /* Reduce nlink of the child. If it's now zero, stick it on the
 285                            dead_fds list to be cleaned up later. Else just free the fd */
 286                         child_ic->pino_nlink--;
 287
 288                         if (!child_ic->pino_nlink) {
 289                                 dbg_fsbuild("inode #%u (\"%s\") now has no links; adding to dead_fds list.\n",
 290                                           fd->ino, fd->name);
 291                                 fd->next = *dead_fds;
 292                                 *dead_fds = fd;
 293                         } else {
 294                                 dbg_fsbuild("inode #%u (\"%s\") has now got nlink %d. Ignoring.\n",
 295                                           fd->ino, fd->name, child_ic->pino_nlink);
 296                                 jffs2_free_full_dirent(fd);
 297                         }
 298                 }
 299         }
 300
 301         /*
 302            We don't delete the inocache from the hash list and free it yet.
 303            The erase code will do that, when all the nodes are completely gone.
 304         */
 305 }
 306
 307 static void jffs2_calc_trigger_levels(struct jffs2_sb_info *c)
 308 {
 309         uint32_t size;
 310
 311         /* Deletion should almost _always_ be allowed. We're fairly
 312            buggered once we stop allowing people to delete stuff
 313            because there's not enough free space... */
 314         c->resv_blocks_deletion = 2;
 315
 316         /* Be conservative about how much space we need before we allow writes.
 317            On top of that which is required for deletia, require an extra 2%
 318            of the medium to be available, for overhead caused by nodes being
 319            split across blocks, etc. */
 320
 321         size = c->flash_size / 50; /* 2% of flash size */
 322         size += c->nr_blocks * 100; /* And 100 bytes per eraseblock */
 323         size += c->sector_size - 1; /* ... and round up */
 324
 325         c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
 326
 327         /* When do we let the GC thread run in the background */
 328
 329         c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
 330
 331         /* When do we allow garbage collection to merge nodes to make
 332            long-term progress at the expense of short-term space exhaustion? */
 333         c->resv_blocks_gcmerge = c->resv_blocks_deletion + 1;
 334
 335         /* When do we allow garbage collection to eat from bad blocks rather
 336            than actually making progress? */
 337         c->resv_blocks_gcbad = 0;//c->resv_blocks_deletion + 2;
 338
 339         /* What number of 'very dirty' eraseblocks do we allow before we
 340            trigger the GC thread even if we don't _need_ the space. When we
 341            can't mark nodes obsolete on the medium, the old dirty nodes cause
 342            performance problems because we have to inspect and discard them. */
 343         c->vdirty_blocks_gctrigger = c->resv_blocks_gctrigger;
 344         if (jffs2_can_mark_obsolete(c))
 345                 c->vdirty_blocks_gctrigger *= 10;
 346
 347         /* If there's less than this amount of dirty space, don't bother
 348            trying to GC to make more space. It'll be a fruitless task */
 349         c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
 350
 351         dbg_fsbuild("trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
 352                     c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks);
 353         dbg_fsbuild("Blocks required to allow deletion:    %d (%d KiB)\n",
 354                   c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024);
 355         dbg_fsbuild("Blocks required to allow writes:      %d (%d KiB)\n",
 356                   c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024);
 357         dbg_fsbuild("Blocks required to quiesce GC thread: %d (%d KiB)\n",
 358                   c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024);
 359         dbg_fsbuild("Blocks required to allow GC merges:   %d (%d KiB)\n",
 360                   c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024);
 361         dbg_fsbuild("Blocks required to GC bad blocks:     %d (%d KiB)\n",
 362                   c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024);
 363         dbg_fsbuild("Amount of dirty space required to GC: %d bytes\n",
 364                   c->nospc_dirty_size);
 365         dbg_fsbuild("Very dirty blocks before GC triggered: %d\n",
 366                   c->vdirty_blocks_gctrigger);
 367 }
 368
 369 int jffs2_do_mount_fs(struct jffs2_sb_info *c)
 370 {
 371         int ret;
 372         int i;
 373         int size;
 374
 375         c->free_size = c->flash_size;
 376         c->nr_blocks = c->flash_size / c->sector_size;
 377         size = sizeof(struct jffs2_eraseblock) * c->nr_blocks;
 378 #ifndef __ECOS
 379         if (jffs2_blocks_use_vmalloc(c))
 380                 c->blocks = vzalloc(size);
 381         else
 382 #endif
 383                 c->blocks = kzalloc(size, GFP_KERNEL);
 384         if (!c->blocks)
 385                 return -ENOMEM;
 386
 387         for (i=0; i<c->nr_blocks; i++) {
 388                 INIT_LIST_HEAD(&c->blocks[i].list);
 389                 c->blocks[i].offset = i * c->sector_size;
 390                 c->blocks[i].free_size = c->sector_size;
 391         }
 392
 393         INIT_LIST_HEAD(&c->clean_list);
 394         INIT_LIST_HEAD(&c->very_dirty_list);
 395         INIT_LIST_HEAD(&c->dirty_list);
 396         INIT_LIST_HEAD(&c->erasable_list);
 397         INIT_LIST_HEAD(&c->erasing_list);
 398         INIT_LIST_HEAD(&c->erase_checking_list);
 399         INIT_LIST_HEAD(&c->erase_pending_list);
 400         INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
 401         INIT_LIST_HEAD(&c->erase_complete_list);
 402         INIT_LIST_HEAD(&c->free_list);
 403         INIT_LIST_HEAD(&c->bad_list);
 404         INIT_LIST_HEAD(&c->bad_used_list);
 405         c->highest_ino = 1;
 406         c->summary = NULL;
 407
 408         ret = jffs2_sum_init(c);
 409         if (ret)
 410                 goto out_free;
 411
 412         if (jffs2_build_filesystem(c)) {
 413                 dbg_fsbuild("build_fs failed\n");
 414                 jffs2_free_ino_caches(c);
 415                 jffs2_free_raw_node_refs(c);
 416                 ret = -EIO;
 417                 goto out_free;
 418         }
 419
 420         jffs2_calc_trigger_levels(c);
 421
 422         return 0;
 423
 424  out_free:
 425 #ifndef __ECOS
 426         if (jffs2_blocks_use_vmalloc(c))
 427                 vfree(c->blocks);
 428         else
 429 #endif
 430                 kfree(c->blocks);
 431
 432         return ret;
 433 }