build.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395
  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. #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  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. static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *,
  20. struct jffs2_inode_cache *, struct jffs2_full_dirent **);
  21. static inline struct jffs2_inode_cache *
  22. first_inode_chain(int *i, struct jffs2_sb_info *c)
  23. {
  24. for (; *i < c->inocache_hashsize; (*i)++) {
  25. if (c->inocache_list[*i])
  26. return c->inocache_list[*i];
  27. }
  28. return NULL;
  29. }
  30. static inline struct jffs2_inode_cache *
  31. next_inode(int *i, struct jffs2_inode_cache *ic, struct jffs2_sb_info *c)
  32. {
  33. /* More in this chain? */
  34. if (ic->next)
  35. return ic->next;
  36. (*i)++;
  37. return first_inode_chain(i, c);
  38. }
  39. #define for_each_inode(i, c, ic) \
  40. for (i = 0, ic = first_inode_chain(&i, (c)); \
  41. ic; \
  42. ic = next_inode(&i, ic, (c)))
  43. static void jffs2_build_inode_pass1(struct jffs2_sb_info *c,
  44. struct jffs2_inode_cache *ic)
  45. {
  46. struct jffs2_full_dirent *fd;
  47. dbg_fsbuild("building directory inode #%u\n", ic->ino);
  48. /* For each child, increase nlink */
  49. for(fd = ic->scan_dents; fd; fd = fd->next) {
  50. struct jffs2_inode_cache *child_ic;
  51. if (!fd->ino)
  52. continue;
  53. /* we can get high latency here with huge directories */
  54. child_ic = jffs2_get_ino_cache(c, fd->ino);
  55. if (!child_ic) {
  56. dbg_fsbuild("child \"%s\" (ino #%u) of dir ino #%u doesn't exist!\n",
  57. fd->name, fd->ino, ic->ino);
  58. jffs2_mark_node_obsolete(c, fd->raw);
  59. continue;
  60. }
  61. if (fd->type == DT_DIR) {
  62. if (child_ic->pino_nlink) {
  63. JFFS2_ERROR("child dir \"%s\" (ino #%u) of dir ino #%u appears to be a hard link\n",
  64. fd->name, fd->ino, ic->ino);
  65. /* TODO: What do we do about it? */
  66. } else {
  67. child_ic->pino_nlink = ic->ino;
  68. }
  69. } else
  70. child_ic->pino_nlink++;
  71. dbg_fsbuild("increased nlink for child \"%s\" (ino #%u)\n", fd->name, fd->ino);
  72. /* Can't free scan_dents so far. We might need them in pass 2 */
  73. }
  74. }
  75. /* Scan plan:
  76. - Scan physical nodes. Build map of inodes/dirents. Allocate inocaches as we go
  77. - Scan directory tree from top down, setting nlink in inocaches
  78. - Scan inocaches for inodes with nlink==0
  79. */
  80. static int jffs2_build_filesystem(struct jffs2_sb_info *c)
  81. {
  82. int ret;
  83. int i;
  84. struct jffs2_inode_cache *ic;
  85. struct jffs2_full_dirent *fd;
  86. struct jffs2_full_dirent *dead_fds = NULL;
  87. dbg_fsbuild("build FS data structures\n");
  88. /* First, scan the medium and build all the inode caches with
  89. lists of physical nodes */
  90. c->flags |= JFFS2_SB_FLAG_SCANNING;
  91. ret = jffs2_scan_medium(c);
  92. c->flags &= ~JFFS2_SB_FLAG_SCANNING;
  93. if (ret)
  94. goto exit;
  95. dbg_fsbuild("scanned flash completely\n");
  96. jffs2_dbg_dump_block_lists_nolock(c);
  97. dbg_fsbuild("pass 1 starting\n");
  98. c->flags |= JFFS2_SB_FLAG_BUILDING;
  99. /* Now scan the directory tree, increasing nlink according to every dirent found. */
  100. for_each_inode(i, c, ic) {
  101. if (ic->scan_dents) {
  102. jffs2_build_inode_pass1(c, ic);
  103. cond_resched();
  104. }
  105. }
  106. dbg_fsbuild("pass 1 complete\n");
  107. /* Next, scan for inodes with nlink == 0 and remove them. If
  108. they were directories, then decrement the nlink of their
  109. children too, and repeat the scan. As that's going to be
  110. a fairly uncommon occurrence, it's not so evil to do it this
  111. way. Recursion bad. */
  112. dbg_fsbuild("pass 2 starting\n");
  113. for_each_inode(i, c, ic) {
  114. if (ic->pino_nlink)
  115. continue;
  116. jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
  117. cond_resched();
  118. }
  119. dbg_fsbuild("pass 2a starting\n");
  120. while (dead_fds) {
  121. fd = dead_fds;
  122. dead_fds = fd->next;
  123. ic = jffs2_get_ino_cache(c, fd->ino);
  124. if (ic)
  125. jffs2_build_remove_unlinked_inode(c, ic, &dead_fds);
  126. jffs2_free_full_dirent(fd);
  127. }
  128. dbg_fsbuild("pass 2a complete\n");
  129. dbg_fsbuild("freeing temporary data structures\n");
  130. /* Finally, we can scan again and free the dirent structs */
  131. for_each_inode(i, c, ic) {
  132. while(ic->scan_dents) {
  133. fd = ic->scan_dents;
  134. ic->scan_dents = fd->next;
  135. jffs2_free_full_dirent(fd);
  136. }
  137. ic->scan_dents = NULL;
  138. cond_resched();
  139. }
  140. jffs2_build_xattr_subsystem(c);
  141. c->flags &= ~JFFS2_SB_FLAG_BUILDING;
  142. dbg_fsbuild("FS build complete\n");
  143. /* Rotate the lists by some number to ensure wear levelling */
  144. jffs2_rotate_lists(c);
  145. ret = 0;
  146. exit:
  147. if (ret) {
  148. for_each_inode(i, c, ic) {
  149. while(ic->scan_dents) {
  150. fd = ic->scan_dents;
  151. ic->scan_dents = fd->next;
  152. jffs2_free_full_dirent(fd);
  153. }
  154. }
  155. jffs2_clear_xattr_subsystem(c);
  156. }
  157. return ret;
  158. }
  159. static void jffs2_build_remove_unlinked_inode(struct jffs2_sb_info *c,
  160. struct jffs2_inode_cache *ic,
  161. struct jffs2_full_dirent **dead_fds)
  162. {
  163. struct jffs2_raw_node_ref *raw;
  164. struct jffs2_full_dirent *fd;
  165. dbg_fsbuild("removing ino #%u with nlink == zero.\n", ic->ino);
  166. raw = ic->nodes;
  167. while (raw != (void *)ic) {
  168. struct jffs2_raw_node_ref *next = raw->next_in_ino;
  169. dbg_fsbuild("obsoleting node at 0x%08x\n", ref_offset(raw));
  170. jffs2_mark_node_obsolete(c, raw);
  171. raw = next;
  172. }
  173. if (ic->scan_dents) {
  174. int whinged = 0;
  175. dbg_fsbuild("inode #%u was a directory which may have children...\n", ic->ino);
  176. while(ic->scan_dents) {
  177. struct jffs2_inode_cache *child_ic;
  178. fd = ic->scan_dents;
  179. ic->scan_dents = fd->next;
  180. if (!fd->ino) {
  181. /* It's a deletion dirent. Ignore it */
  182. dbg_fsbuild("child \"%s\" is a deletion dirent, skipping...\n", fd->name);
  183. jffs2_free_full_dirent(fd);
  184. continue;
  185. }
  186. if (!whinged)
  187. whinged = 1;
  188. dbg_fsbuild("removing child \"%s\", ino #%u\n", fd->name, fd->ino);
  189. child_ic = jffs2_get_ino_cache(c, fd->ino);
  190. if (!child_ic) {
  191. dbg_fsbuild("cannot remove child \"%s\", ino #%u, because it doesn't exist\n",
  192. fd->name, fd->ino);
  193. jffs2_free_full_dirent(fd);
  194. continue;
  195. }
  196. /* Reduce nlink of the child. If it's now zero, stick it on the
  197. dead_fds list to be cleaned up later. Else just free the fd */
  198. if (fd->type == DT_DIR)
  199. child_ic->pino_nlink = 0;
  200. else
  201. child_ic->pino_nlink--;
  202. if (!child_ic->pino_nlink) {
  203. dbg_fsbuild("inode #%u (\"%s\") now has no links; adding to dead_fds list.\n",
  204. fd->ino, fd->name);
  205. fd->next = *dead_fds;
  206. *dead_fds = fd;
  207. } else {
  208. dbg_fsbuild("inode #%u (\"%s\") has now got nlink %d. Ignoring.\n",
  209. fd->ino, fd->name, child_ic->pino_nlink);
  210. jffs2_free_full_dirent(fd);
  211. }
  212. }
  213. }
  214. /*
  215. We don't delete the inocache from the hash list and free it yet.
  216. The erase code will do that, when all the nodes are completely gone.
  217. */
  218. }
  219. static void jffs2_calc_trigger_levels(struct jffs2_sb_info *c)
  220. {
  221. uint32_t size;
  222. /* Deletion should almost _always_ be allowed. We're fairly
  223. buggered once we stop allowing people to delete stuff
  224. because there's not enough free space... */
  225. c->resv_blocks_deletion = 2;
  226. /* Be conservative about how much space we need before we allow writes.
  227. On top of that which is required for deletia, require an extra 2%
  228. of the medium to be available, for overhead caused by nodes being
  229. split across blocks, etc. */
  230. size = c->flash_size / 50; /* 2% of flash size */
  231. size += c->nr_blocks * 100; /* And 100 bytes per eraseblock */
  232. size += c->sector_size - 1; /* ... and round up */
  233. c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
  234. /* When do we let the GC thread run in the background */
  235. c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
  236. /* When do we allow garbage collection to merge nodes to make
  237. long-term progress at the expense of short-term space exhaustion? */
  238. c->resv_blocks_gcmerge = c->resv_blocks_deletion + 1;
  239. /* When do we allow garbage collection to eat from bad blocks rather
  240. than actually making progress? */
  241. c->resv_blocks_gcbad = 0;//c->resv_blocks_deletion + 2;
  242. /* What number of 'very dirty' eraseblocks do we allow before we
  243. trigger the GC thread even if we don't _need_ the space. When we
  244. can't mark nodes obsolete on the medium, the old dirty nodes cause
  245. performance problems because we have to inspect and discard them. */
  246. c->vdirty_blocks_gctrigger = c->resv_blocks_gctrigger;
  247. if (jffs2_can_mark_obsolete(c))
  248. c->vdirty_blocks_gctrigger *= 10;
  249. /* If there's less than this amount of dirty space, don't bother
  250. trying to GC to make more space. It'll be a fruitless task */
  251. c->nospc_dirty_size = c->sector_size + (c->flash_size / 100);
  252. dbg_fsbuild("trigger levels (size %d KiB, block size %d KiB, %d blocks)\n",
  253. c->flash_size / 1024, c->sector_size / 1024, c->nr_blocks);
  254. dbg_fsbuild("Blocks required to allow deletion: %d (%d KiB)\n",
  255. c->resv_blocks_deletion, c->resv_blocks_deletion*c->sector_size/1024);
  256. dbg_fsbuild("Blocks required to allow writes: %d (%d KiB)\n",
  257. c->resv_blocks_write, c->resv_blocks_write*c->sector_size/1024);
  258. dbg_fsbuild("Blocks required to quiesce GC thread: %d (%d KiB)\n",
  259. c->resv_blocks_gctrigger, c->resv_blocks_gctrigger*c->sector_size/1024);
  260. dbg_fsbuild("Blocks required to allow GC merges: %d (%d KiB)\n",
  261. c->resv_blocks_gcmerge, c->resv_blocks_gcmerge*c->sector_size/1024);
  262. dbg_fsbuild("Blocks required to GC bad blocks: %d (%d KiB)\n",
  263. c->resv_blocks_gcbad, c->resv_blocks_gcbad*c->sector_size/1024);
  264. dbg_fsbuild("Amount of dirty space required to GC: %d bytes\n",
  265. c->nospc_dirty_size);
  266. dbg_fsbuild("Very dirty blocks before GC triggered: %d\n",
  267. c->vdirty_blocks_gctrigger);
  268. }
  269. int jffs2_do_mount_fs(struct jffs2_sb_info *c)
  270. {
  271. int ret;
  272. int i;
  273. int size;
  274. c->free_size = c->flash_size;
  275. c->nr_blocks = c->flash_size / c->sector_size;
  276. size = sizeof(struct jffs2_eraseblock) * c->nr_blocks;
  277. #ifndef __ECOS
  278. if (jffs2_blocks_use_vmalloc(c))
  279. c->blocks = vzalloc(size);
  280. else
  281. #endif
  282. c->blocks = kzalloc(size, GFP_KERNEL);
  283. if (!c->blocks)
  284. return -ENOMEM;
  285. for (i=0; i<c->nr_blocks; i++) {
  286. INIT_LIST_HEAD(&c->blocks[i].list);
  287. c->blocks[i].offset = i * c->sector_size;
  288. c->blocks[i].free_size = c->sector_size;
  289. }
  290. INIT_LIST_HEAD(&c->clean_list);
  291. INIT_LIST_HEAD(&c->very_dirty_list);
  292. INIT_LIST_HEAD(&c->dirty_list);
  293. INIT_LIST_HEAD(&c->erasable_list);
  294. INIT_LIST_HEAD(&c->erasing_list);
  295. INIT_LIST_HEAD(&c->erase_checking_list);
  296. INIT_LIST_HEAD(&c->erase_pending_list);
  297. INIT_LIST_HEAD(&c->erasable_pending_wbuf_list);
  298. INIT_LIST_HEAD(&c->erase_complete_list);
  299. INIT_LIST_HEAD(&c->free_list);
  300. INIT_LIST_HEAD(&c->bad_list);
  301. INIT_LIST_HEAD(&c->bad_used_list);
  302. c->highest_ino = 1;
  303. c->summary = NULL;
  304. ret = jffs2_sum_init(c);
  305. if (ret)
  306. goto out_free;
  307. if (jffs2_build_filesystem(c)) {
  308. dbg_fsbuild("build_fs failed\n");
  309. jffs2_free_ino_caches(c);
  310. jffs2_free_raw_node_refs(c);
  311. ret = -EIO;
  312. goto out_free;
  313. }
  314. jffs2_calc_trigger_levels(c);
  315. return 0;
  316. out_free:
  317. #ifndef __ECOS
  318. if (jffs2_blocks_use_vmalloc(c))
  319. vfree(c->blocks);
  320. else
  321. #endif
  322. kfree(c->blocks);
  323. return ret;
  324. }