super.c 26 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051
  1. /*
  2. * Copyright (C) 2005, 2006
  3. * Avishay Traeger (avishay@gmail.com)
  4. * Copyright (C) 2008, 2009
  5. * Boaz Harrosh <bharrosh@panasas.com>
  6. *
  7. * Copyrights for code taken from ext2:
  8. * Copyright (C) 1992, 1993, 1994, 1995
  9. * Remy Card (card@masi.ibp.fr)
  10. * Laboratoire MASI - Institut Blaise Pascal
  11. * Universite Pierre et Marie Curie (Paris VI)
  12. * from
  13. * linux/fs/minix/inode.c
  14. * Copyright (C) 1991, 1992 Linus Torvalds
  15. *
  16. * This file is part of exofs.
  17. *
  18. * exofs is free software; you can redistribute it and/or modify
  19. * it under the terms of the GNU General Public License as published by
  20. * the Free Software Foundation. Since it is based on ext2, and the only
  21. * valid version of GPL for the Linux kernel is version 2, the only valid
  22. * version of GPL for exofs is version 2.
  23. *
  24. * exofs is distributed in the hope that it will be useful,
  25. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  26. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  27. * GNU General Public License for more details.
  28. *
  29. * You should have received a copy of the GNU General Public License
  30. * along with exofs; if not, write to the Free Software
  31. * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  32. */
  33. #include <linux/string.h>
  34. #include <linux/parser.h>
  35. #include <linux/vfs.h>
  36. #include <linux/random.h>
  37. #include <linux/module.h>
  38. #include <linux/exportfs.h>
  39. #include <linux/slab.h>
  40. #include "exofs.h"
  41. #define EXOFS_DBGMSG2(M...) do {} while (0)
  42. /******************************************************************************
  43. * MOUNT OPTIONS
  44. *****************************************************************************/
  45. /*
  46. * struct to hold what we get from mount options
  47. */
  48. struct exofs_mountopt {
  49. bool is_osdname;
  50. const char *dev_name;
  51. uint64_t pid;
  52. int timeout;
  53. };
  54. /*
  55. * exofs-specific mount-time options.
  56. */
  57. enum { Opt_name, Opt_pid, Opt_to, Opt_err };
  58. /*
  59. * Our mount-time options. These should ideally be 64-bit unsigned, but the
  60. * kernel's parsing functions do not currently support that. 32-bit should be
  61. * sufficient for most applications now.
  62. */
  63. static match_table_t tokens = {
  64. {Opt_name, "osdname=%s"},
  65. {Opt_pid, "pid=%u"},
  66. {Opt_to, "to=%u"},
  67. {Opt_err, NULL}
  68. };
  69. /*
  70. * The main option parsing method. Also makes sure that all of the mandatory
  71. * mount options were set.
  72. */
  73. static int parse_options(char *options, struct exofs_mountopt *opts)
  74. {
  75. char *p;
  76. substring_t args[MAX_OPT_ARGS];
  77. int option;
  78. bool s_pid = false;
  79. EXOFS_DBGMSG("parse_options %s\n", options);
  80. /* defaults */
  81. memset(opts, 0, sizeof(*opts));
  82. opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
  83. while ((p = strsep(&options, ",")) != NULL) {
  84. int token;
  85. char str[32];
  86. if (!*p)
  87. continue;
  88. token = match_token(p, tokens, args);
  89. switch (token) {
  90. case Opt_name:
  91. opts->dev_name = match_strdup(&args[0]);
  92. if (unlikely(!opts->dev_name)) {
  93. EXOFS_ERR("Error allocating dev_name");
  94. return -ENOMEM;
  95. }
  96. opts->is_osdname = true;
  97. break;
  98. case Opt_pid:
  99. if (0 == match_strlcpy(str, &args[0], sizeof(str)))
  100. return -EINVAL;
  101. opts->pid = simple_strtoull(str, NULL, 0);
  102. if (opts->pid < EXOFS_MIN_PID) {
  103. EXOFS_ERR("Partition ID must be >= %u",
  104. EXOFS_MIN_PID);
  105. return -EINVAL;
  106. }
  107. s_pid = 1;
  108. break;
  109. case Opt_to:
  110. if (match_int(&args[0], &option))
  111. return -EINVAL;
  112. if (option <= 0) {
  113. EXOFS_ERR("Timout must be > 0");
  114. return -EINVAL;
  115. }
  116. opts->timeout = option * HZ;
  117. break;
  118. }
  119. }
  120. if (!s_pid) {
  121. EXOFS_ERR("Need to specify the following options:\n");
  122. EXOFS_ERR(" -o pid=pid_no_to_use\n");
  123. return -EINVAL;
  124. }
  125. return 0;
  126. }
  127. /******************************************************************************
  128. * INODE CACHE
  129. *****************************************************************************/
  130. /*
  131. * Our inode cache. Isn't it pretty?
  132. */
  133. static struct kmem_cache *exofs_inode_cachep;
  134. /*
  135. * Allocate an inode in the cache
  136. */
  137. static struct inode *exofs_alloc_inode(struct super_block *sb)
  138. {
  139. struct exofs_i_info *oi;
  140. oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
  141. if (!oi)
  142. return NULL;
  143. oi->vfs_inode.i_version = 1;
  144. return &oi->vfs_inode;
  145. }
  146. static void exofs_i_callback(struct rcu_head *head)
  147. {
  148. struct inode *inode = container_of(head, struct inode, i_rcu);
  149. kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
  150. }
  151. /*
  152. * Remove an inode from the cache
  153. */
  154. static void exofs_destroy_inode(struct inode *inode)
  155. {
  156. call_rcu(&inode->i_rcu, exofs_i_callback);
  157. }
  158. /*
  159. * Initialize the inode
  160. */
  161. static void exofs_init_once(void *foo)
  162. {
  163. struct exofs_i_info *oi = foo;
  164. inode_init_once(&oi->vfs_inode);
  165. }
  166. /*
  167. * Create and initialize the inode cache
  168. */
  169. static int init_inodecache(void)
  170. {
  171. exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
  172. sizeof(struct exofs_i_info), 0,
  173. SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
  174. exofs_init_once);
  175. if (exofs_inode_cachep == NULL)
  176. return -ENOMEM;
  177. return 0;
  178. }
  179. /*
  180. * Destroy the inode cache
  181. */
  182. static void destroy_inodecache(void)
  183. {
  184. /*
  185. * Make sure all delayed rcu free inodes are flushed before we
  186. * destroy cache.
  187. */
  188. rcu_barrier();
  189. kmem_cache_destroy(exofs_inode_cachep);
  190. }
  191. /******************************************************************************
  192. * Some osd helpers
  193. *****************************************************************************/
  194. void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
  195. {
  196. osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
  197. }
  198. static int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
  199. u64 offset, void *p, unsigned length)
  200. {
  201. struct osd_request *or = osd_start_request(od, GFP_KERNEL);
  202. /* struct osd_sense_info osi = {.key = 0};*/
  203. int ret;
  204. if (unlikely(!or)) {
  205. EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
  206. return -ENOMEM;
  207. }
  208. ret = osd_req_read_kern(or, obj, offset, p, length);
  209. if (unlikely(ret)) {
  210. EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
  211. goto out;
  212. }
  213. ret = osd_finalize_request(or, 0, cred, NULL);
  214. if (unlikely(ret)) {
  215. EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
  216. goto out;
  217. }
  218. ret = osd_execute_request(or);
  219. if (unlikely(ret))
  220. EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
  221. /* osd_req_decode_sense(or, ret); */
  222. out:
  223. osd_end_request(or);
  224. EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
  225. "length=0x%llx dev=%p ret=>%d\n",
  226. _LLU(obj->id), _LLU(offset), _LLU(length), od, ret);
  227. return ret;
  228. }
  229. static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
  230. EXOFS_APAGE_SB_DATA,
  231. EXOFS_ATTR_SB_STATS,
  232. sizeof(struct exofs_sb_stats));
  233. static int __sbi_read_stats(struct exofs_sb_info *sbi)
  234. {
  235. struct osd_attr attrs[] = {
  236. [0] = g_attr_sb_stats,
  237. };
  238. struct ore_io_state *ios;
  239. int ret;
  240. ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
  241. if (unlikely(ret)) {
  242. EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
  243. return ret;
  244. }
  245. ios->in_attr = attrs;
  246. ios->in_attr_len = ARRAY_SIZE(attrs);
  247. ret = ore_read(ios);
  248. if (unlikely(ret)) {
  249. EXOFS_ERR("Error reading super_block stats => %d\n", ret);
  250. goto out;
  251. }
  252. ret = extract_attr_from_ios(ios, &attrs[0]);
  253. if (ret) {
  254. EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
  255. goto out;
  256. }
  257. if (attrs[0].len) {
  258. struct exofs_sb_stats *ess;
  259. if (unlikely(attrs[0].len != sizeof(*ess))) {
  260. EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
  261. "size(%d) != expected(%zd)\n",
  262. __func__, attrs[0].len, sizeof(*ess));
  263. goto out;
  264. }
  265. ess = attrs[0].val_ptr;
  266. sbi->s_nextid = le64_to_cpu(ess->s_nextid);
  267. sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
  268. }
  269. out:
  270. ore_put_io_state(ios);
  271. return ret;
  272. }
  273. static void stats_done(struct ore_io_state *ios, void *p)
  274. {
  275. ore_put_io_state(ios);
  276. /* Good thanks nothing to do anymore */
  277. }
  278. /* Asynchronously write the stats attribute */
  279. int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
  280. {
  281. struct osd_attr attrs[] = {
  282. [0] = g_attr_sb_stats,
  283. };
  284. struct ore_io_state *ios;
  285. int ret;
  286. ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
  287. if (unlikely(ret)) {
  288. EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
  289. return ret;
  290. }
  291. sbi->s_ess.s_nextid = cpu_to_le64(sbi->s_nextid);
  292. sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
  293. attrs[0].val_ptr = &sbi->s_ess;
  294. ios->done = stats_done;
  295. ios->private = sbi;
  296. ios->out_attr = attrs;
  297. ios->out_attr_len = ARRAY_SIZE(attrs);
  298. ret = ore_write(ios);
  299. if (unlikely(ret)) {
  300. EXOFS_ERR("%s: ore_write failed.\n", __func__);
  301. ore_put_io_state(ios);
  302. }
  303. return ret;
  304. }
  305. /******************************************************************************
  306. * SUPERBLOCK FUNCTIONS
  307. *****************************************************************************/
  308. static const struct super_operations exofs_sops;
  309. static const struct export_operations exofs_export_ops;
  310. /*
  311. * Write the superblock to the OSD
  312. */
  313. static int exofs_sync_fs(struct super_block *sb, int wait)
  314. {
  315. struct exofs_sb_info *sbi;
  316. struct exofs_fscb *fscb;
  317. struct ore_comp one_comp;
  318. struct ore_components oc;
  319. struct ore_io_state *ios;
  320. int ret = -ENOMEM;
  321. fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
  322. if (unlikely(!fscb))
  323. return -ENOMEM;
  324. sbi = sb->s_fs_info;
  325. /* NOTE: We no longer dirty the super_block anywhere in exofs. The
  326. * reason we write the fscb here on unmount is so we can stay backwards
  327. * compatible with fscb->s_version == 1. (What we are not compatible
  328. * with is if a new version FS crashed and then we try to mount an old
  329. * version). Otherwise the exofs_fscb is read-only from mkfs time. All
  330. * the writeable info is set in exofs_sbi_write_stats() above.
  331. */
  332. exofs_init_comps(&oc, &one_comp, sbi, EXOFS_SUPER_ID);
  333. ret = ore_get_io_state(&sbi->layout, &oc, &ios);
  334. if (unlikely(ret))
  335. goto out;
  336. lock_super(sb);
  337. ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
  338. memset(fscb, 0, ios->length);
  339. fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
  340. fscb->s_numfiles = cpu_to_le64(sbi->s_numfiles);
  341. fscb->s_magic = cpu_to_le16(sb->s_magic);
  342. fscb->s_newfs = 0;
  343. fscb->s_version = EXOFS_FSCB_VER;
  344. ios->offset = 0;
  345. ios->kern_buff = fscb;
  346. ret = ore_write(ios);
  347. if (unlikely(ret))
  348. EXOFS_ERR("%s: ore_write failed.\n", __func__);
  349. else
  350. sb->s_dirt = 0;
  351. unlock_super(sb);
  352. out:
  353. EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
  354. ore_put_io_state(ios);
  355. kfree(fscb);
  356. return ret;
  357. }
  358. static void exofs_write_super(struct super_block *sb)
  359. {
  360. if (!(sb->s_flags & MS_RDONLY))
  361. exofs_sync_fs(sb, 1);
  362. else
  363. sb->s_dirt = 0;
  364. }
  365. static void _exofs_print_device(const char *msg, const char *dev_path,
  366. struct osd_dev *od, u64 pid)
  367. {
  368. const struct osd_dev_info *odi = osduld_device_info(od);
  369. printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
  370. msg, dev_path ?: "", odi->osdname, _LLU(pid));
  371. }
  372. static void exofs_free_sbi(struct exofs_sb_info *sbi)
  373. {
  374. unsigned numdevs = sbi->oc.numdevs;
  375. while (numdevs) {
  376. unsigned i = --numdevs;
  377. struct osd_dev *od = ore_comp_dev(&sbi->oc, i);
  378. if (od) {
  379. ore_comp_set_dev(&sbi->oc, i, NULL);
  380. osduld_put_device(od);
  381. }
  382. }
  383. kfree(sbi->oc.ods);
  384. kfree(sbi);
  385. }
  386. /*
  387. * This function is called when the vfs is freeing the superblock. We just
  388. * need to free our own part.
  389. */
  390. static void exofs_put_super(struct super_block *sb)
  391. {
  392. int num_pend;
  393. struct exofs_sb_info *sbi = sb->s_fs_info;
  394. /* make sure there are no pending commands */
  395. for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
  396. num_pend = atomic_read(&sbi->s_curr_pending)) {
  397. wait_queue_head_t wq;
  398. printk(KERN_NOTICE "%s: !!Pending operations in flight. "
  399. "This is a BUG. please report to osd-dev@open-osd.org\n",
  400. __func__);
  401. init_waitqueue_head(&wq);
  402. wait_event_timeout(wq,
  403. (atomic_read(&sbi->s_curr_pending) == 0),
  404. msecs_to_jiffies(100));
  405. }
  406. _exofs_print_device("Unmounting", NULL, ore_comp_dev(&sbi->oc, 0),
  407. sbi->one_comp.obj.partition);
  408. bdi_destroy(&sbi->bdi);
  409. exofs_free_sbi(sbi);
  410. sb->s_fs_info = NULL;
  411. }
  412. static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
  413. struct exofs_device_table *dt)
  414. {
  415. int ret;
  416. sbi->layout.stripe_unit =
  417. le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
  418. sbi->layout.group_width =
  419. le32_to_cpu(dt->dt_data_map.cb_group_width);
  420. sbi->layout.group_depth =
  421. le32_to_cpu(dt->dt_data_map.cb_group_depth);
  422. sbi->layout.mirrors_p1 =
  423. le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
  424. sbi->layout.raid_algorithm =
  425. le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
  426. ret = ore_verify_layout(numdevs, &sbi->layout);
  427. EXOFS_DBGMSG("exofs: layout: "
  428. "num_comps=%u stripe_unit=0x%x group_width=%u "
  429. "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
  430. numdevs,
  431. sbi->layout.stripe_unit,
  432. sbi->layout.group_width,
  433. _LLU(sbi->layout.group_depth),
  434. sbi->layout.mirrors_p1,
  435. sbi->layout.raid_algorithm);
  436. return ret;
  437. }
  438. static unsigned __ra_pages(struct ore_layout *layout)
  439. {
  440. const unsigned _MIN_RA = 32; /* min 128K read-ahead */
  441. unsigned ra_pages = layout->group_width * layout->stripe_unit /
  442. PAGE_SIZE;
  443. unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
  444. ra_pages *= 2; /* two stripes */
  445. if (ra_pages < _MIN_RA)
  446. ra_pages = roundup(_MIN_RA, ra_pages / 2);
  447. if (ra_pages > max_io_pages)
  448. ra_pages = max_io_pages;
  449. return ra_pages;
  450. }
  451. /* @odi is valid only as long as @fscb_dev is valid */
  452. static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
  453. struct osd_dev_info *odi)
  454. {
  455. odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
  456. if (likely(odi->systemid_len))
  457. memcpy(odi->systemid, dt_dev->systemid, OSD_SYSTEMID_LEN);
  458. odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
  459. odi->osdname = dt_dev->osdname;
  460. /* FIXME support long names. Will need a _put function */
  461. if (dt_dev->long_name_offset)
  462. return -EINVAL;
  463. /* Make sure osdname is printable!
  464. * mkexofs should give us space for a null-terminator else the
  465. * device-table is invalid.
  466. */
  467. if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
  468. odi->osdname_len = sizeof(dt_dev->osdname) - 1;
  469. dt_dev->osdname[odi->osdname_len] = 0;
  470. /* If it's all zeros something is bad we read past end-of-obj */
  471. return !(odi->systemid_len || odi->osdname_len);
  472. }
  473. int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
  474. struct exofs_dev **peds)
  475. {
  476. struct __alloc_ore_devs_and_exofs_devs {
  477. /* Twice bigger table: See exofs_init_comps() and comment at
  478. * exofs_read_lookup_dev_table()
  479. */
  480. struct ore_dev *oreds[numdevs * 2 - 1];
  481. struct exofs_dev eds[numdevs];
  482. } *aoded;
  483. struct exofs_dev *eds;
  484. unsigned i;
  485. aoded = kzalloc(sizeof(*aoded), GFP_KERNEL);
  486. if (unlikely(!aoded)) {
  487. EXOFS_ERR("ERROR: failed allocating Device array[%d]\n",
  488. numdevs);
  489. return -ENOMEM;
  490. }
  491. sbi->oc.ods = aoded->oreds;
  492. *peds = eds = aoded->eds;
  493. for (i = 0; i < numdevs; ++i)
  494. aoded->oreds[i] = &eds[i].ored;
  495. return 0;
  496. }
  497. static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
  498. struct osd_dev *fscb_od,
  499. unsigned table_count)
  500. {
  501. struct ore_comp comp;
  502. struct exofs_device_table *dt;
  503. struct exofs_dev *eds;
  504. unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
  505. sizeof(*dt);
  506. unsigned numdevs, i;
  507. int ret;
  508. dt = kmalloc(table_bytes, GFP_KERNEL);
  509. if (unlikely(!dt)) {
  510. EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
  511. table_bytes);
  512. return -ENOMEM;
  513. }
  514. sbi->oc.numdevs = 0;
  515. comp.obj.partition = sbi->one_comp.obj.partition;
  516. comp.obj.id = EXOFS_DEVTABLE_ID;
  517. exofs_make_credential(comp.cred, &comp.obj);
  518. ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
  519. table_bytes);
  520. if (unlikely(ret)) {
  521. EXOFS_ERR("ERROR: reading device table\n");
  522. goto out;
  523. }
  524. numdevs = le64_to_cpu(dt->dt_num_devices);
  525. if (unlikely(!numdevs)) {
  526. ret = -EINVAL;
  527. goto out;
  528. }
  529. WARN_ON(table_count != numdevs);
  530. ret = _read_and_match_data_map(sbi, numdevs, dt);
  531. if (unlikely(ret))
  532. goto out;
  533. ret = __alloc_dev_table(sbi, numdevs, &eds);
  534. if (unlikely(ret))
  535. goto out;
  536. /* exofs round-robins the device table view according to inode
  537. * number. We hold a: twice bigger table hence inodes can point
  538. * to any device and have a sequential view of the table
  539. * starting at this device. See exofs_init_comps()
  540. */
  541. memcpy(&sbi->oc.ods[numdevs], &sbi->oc.ods[0],
  542. (numdevs - 1) * sizeof(sbi->oc.ods[0]));
  543. for (i = 0; i < numdevs; i++) {
  544. struct exofs_fscb fscb;
  545. struct osd_dev_info odi;
  546. struct osd_dev *od;
  547. if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
  548. EXOFS_ERR("ERROR: Read all-zeros device entry\n");
  549. ret = -EINVAL;
  550. goto out;
  551. }
  552. printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
  553. i, odi.osdname);
  554. /* the exofs id is currently the table index */
  555. eds[i].did = i;
  556. /* On all devices the device table is identical. The user can
  557. * specify any one of the participating devices on the command
  558. * line. We always keep them in device-table order.
  559. */
  560. if (fscb_od && osduld_device_same(fscb_od, &odi)) {
  561. eds[i].ored.od = fscb_od;
  562. ++sbi->oc.numdevs;
  563. fscb_od = NULL;
  564. continue;
  565. }
  566. od = osduld_info_lookup(&odi);
  567. if (IS_ERR(od)) {
  568. ret = PTR_ERR(od);
  569. EXOFS_ERR("ERROR: device requested is not found "
  570. "osd_name-%s =>%d\n", odi.osdname, ret);
  571. goto out;
  572. }
  573. eds[i].ored.od = od;
  574. ++sbi->oc.numdevs;
  575. /* Read the fscb of the other devices to make sure the FS
  576. * partition is there.
  577. */
  578. ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
  579. sizeof(fscb));
  580. if (unlikely(ret)) {
  581. EXOFS_ERR("ERROR: Malformed participating device "
  582. "error reading fscb osd_name-%s\n",
  583. odi.osdname);
  584. goto out;
  585. }
  586. /* TODO: verify other information is correct and FS-uuid
  587. * matches. Benny what did you say about device table
  588. * generation and old devices?
  589. */
  590. }
  591. out:
  592. kfree(dt);
  593. if (unlikely(fscb_od && !ret)) {
  594. EXOFS_ERR("ERROR: Bad device-table container device not present\n");
  595. osduld_put_device(fscb_od);
  596. return -EINVAL;
  597. }
  598. return ret;
  599. }
  600. /*
  601. * Read the superblock from the OSD and fill in the fields
  602. */
  603. static int exofs_fill_super(struct super_block *sb, void *data, int silent)
  604. {
  605. struct inode *root;
  606. struct exofs_mountopt *opts = data;
  607. struct exofs_sb_info *sbi; /*extended info */
  608. struct osd_dev *od; /* Master device */
  609. struct exofs_fscb fscb; /*on-disk superblock info */
  610. struct ore_comp comp;
  611. unsigned table_count;
  612. int ret;
  613. sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
  614. if (!sbi)
  615. return -ENOMEM;
  616. /* use mount options to fill superblock */
  617. if (opts->is_osdname) {
  618. struct osd_dev_info odi = {.systemid_len = 0};
  619. odi.osdname_len = strlen(opts->dev_name);
  620. odi.osdname = (u8 *)opts->dev_name;
  621. od = osduld_info_lookup(&odi);
  622. kfree(opts->dev_name);
  623. opts->dev_name = NULL;
  624. } else {
  625. od = osduld_path_lookup(opts->dev_name);
  626. }
  627. if (IS_ERR(od)) {
  628. ret = -EINVAL;
  629. goto free_sbi;
  630. }
  631. /* Default layout in case we do not have a device-table */
  632. sbi->layout.stripe_unit = PAGE_SIZE;
  633. sbi->layout.mirrors_p1 = 1;
  634. sbi->layout.group_width = 1;
  635. sbi->layout.group_depth = -1;
  636. sbi->layout.group_count = 1;
  637. sbi->s_timeout = opts->timeout;
  638. sbi->one_comp.obj.partition = opts->pid;
  639. sbi->one_comp.obj.id = 0;
  640. exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
  641. sbi->oc.single_comp = EC_SINGLE_COMP;
  642. sbi->oc.comps = &sbi->one_comp;
  643. /* fill in some other data by hand */
  644. memset(sb->s_id, 0, sizeof(sb->s_id));
  645. strcpy(sb->s_id, "exofs");
  646. sb->s_blocksize = EXOFS_BLKSIZE;
  647. sb->s_blocksize_bits = EXOFS_BLKSHIFT;
  648. sb->s_maxbytes = MAX_LFS_FILESIZE;
  649. sb->s_max_links = EXOFS_LINK_MAX;
  650. atomic_set(&sbi->s_curr_pending, 0);
  651. sb->s_bdev = NULL;
  652. sb->s_dev = 0;
  653. comp.obj.partition = sbi->one_comp.obj.partition;
  654. comp.obj.id = EXOFS_SUPER_ID;
  655. exofs_make_credential(comp.cred, &comp.obj);
  656. ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
  657. if (unlikely(ret))
  658. goto free_sbi;
  659. sb->s_magic = le16_to_cpu(fscb.s_magic);
  660. /* NOTE: we read below to be backward compatible with old versions */
  661. sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
  662. sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
  663. /* make sure what we read from the object store is correct */
  664. if (sb->s_magic != EXOFS_SUPER_MAGIC) {
  665. if (!silent)
  666. EXOFS_ERR("ERROR: Bad magic value\n");
  667. ret = -EINVAL;
  668. goto free_sbi;
  669. }
  670. if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
  671. EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
  672. EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
  673. ret = -EINVAL;
  674. goto free_sbi;
  675. }
  676. /* start generation numbers from a random point */
  677. get_random_bytes(&sbi->s_next_generation, sizeof(u32));
  678. spin_lock_init(&sbi->s_next_gen_lock);
  679. table_count = le64_to_cpu(fscb.s_dev_table_count);
  680. if (table_count) {
  681. ret = exofs_read_lookup_dev_table(sbi, od, table_count);
  682. if (unlikely(ret))
  683. goto free_sbi;
  684. } else {
  685. struct exofs_dev *eds;
  686. ret = __alloc_dev_table(sbi, 1, &eds);
  687. if (unlikely(ret))
  688. goto free_sbi;
  689. ore_comp_set_dev(&sbi->oc, 0, od);
  690. sbi->oc.numdevs = 1;
  691. }
  692. __sbi_read_stats(sbi);
  693. /* set up operation vectors */
  694. sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
  695. sb->s_bdi = &sbi->bdi;
  696. sb->s_fs_info = sbi;
  697. sb->s_op = &exofs_sops;
  698. sb->s_export_op = &exofs_export_ops;
  699. root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
  700. if (IS_ERR(root)) {
  701. EXOFS_ERR("ERROR: exofs_iget failed\n");
  702. ret = PTR_ERR(root);
  703. goto free_sbi;
  704. }
  705. sb->s_root = d_make_root(root);
  706. if (!sb->s_root) {
  707. EXOFS_ERR("ERROR: get root inode failed\n");
  708. ret = -ENOMEM;
  709. goto free_sbi;
  710. }
  711. if (!S_ISDIR(root->i_mode)) {
  712. dput(sb->s_root);
  713. sb->s_root = NULL;
  714. EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
  715. root->i_mode);
  716. ret = -EINVAL;
  717. goto free_sbi;
  718. }
  719. ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
  720. if (ret) {
  721. EXOFS_DBGMSG("Failed to bdi_setup_and_register\n");
  722. dput(sb->s_root);
  723. sb->s_root = NULL;
  724. goto free_sbi;
  725. }
  726. _exofs_print_device("Mounting", opts->dev_name,
  727. ore_comp_dev(&sbi->oc, 0),
  728. sbi->one_comp.obj.partition);
  729. return 0;
  730. free_sbi:
  731. EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
  732. opts->dev_name, sbi->one_comp.obj.partition, ret);
  733. exofs_free_sbi(sbi);
  734. return ret;
  735. }
  736. /*
  737. * Set up the superblock (calls exofs_fill_super eventually)
  738. */
  739. static struct dentry *exofs_mount(struct file_system_type *type,
  740. int flags, const char *dev_name,
  741. void *data)
  742. {
  743. struct exofs_mountopt opts;
  744. int ret;
  745. ret = parse_options(data, &opts);
  746. if (ret)
  747. return ERR_PTR(ret);
  748. if (!opts.dev_name)
  749. opts.dev_name = dev_name;
  750. return mount_nodev(type, flags, &opts, exofs_fill_super);
  751. }
  752. /*
  753. * Return information about the file system state in the buffer. This is used
  754. * by the 'df' command, for example.
  755. */
  756. static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
  757. {
  758. struct super_block *sb = dentry->d_sb;
  759. struct exofs_sb_info *sbi = sb->s_fs_info;
  760. struct ore_io_state *ios;
  761. struct osd_attr attrs[] = {
  762. ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
  763. OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
  764. ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
  765. OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
  766. };
  767. uint64_t capacity = ULLONG_MAX;
  768. uint64_t used = ULLONG_MAX;
  769. int ret;
  770. ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
  771. if (ret) {
  772. EXOFS_DBGMSG("ore_get_io_state failed.\n");
  773. return ret;
  774. }
  775. ios->in_attr = attrs;
  776. ios->in_attr_len = ARRAY_SIZE(attrs);
  777. ret = ore_read(ios);
  778. if (unlikely(ret))
  779. goto out;
  780. ret = extract_attr_from_ios(ios, &attrs[0]);
  781. if (likely(!ret)) {
  782. capacity = get_unaligned_be64(attrs[0].val_ptr);
  783. if (unlikely(!capacity))
  784. capacity = ULLONG_MAX;
  785. } else
  786. EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
  787. ret = extract_attr_from_ios(ios, &attrs[1]);
  788. if (likely(!ret))
  789. used = get_unaligned_be64(attrs[1].val_ptr);
  790. else
  791. EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
  792. /* fill in the stats buffer */
  793. buf->f_type = EXOFS_SUPER_MAGIC;
  794. buf->f_bsize = EXOFS_BLKSIZE;
  795. buf->f_blocks = capacity >> 9;
  796. buf->f_bfree = (capacity - used) >> 9;
  797. buf->f_bavail = buf->f_bfree;
  798. buf->f_files = sbi->s_numfiles;
  799. buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
  800. buf->f_namelen = EXOFS_NAME_LEN;
  801. out:
  802. ore_put_io_state(ios);
  803. return ret;
  804. }
  805. static const struct super_operations exofs_sops = {
  806. .alloc_inode = exofs_alloc_inode,
  807. .destroy_inode = exofs_destroy_inode,
  808. .write_inode = exofs_write_inode,
  809. .evict_inode = exofs_evict_inode,
  810. .put_super = exofs_put_super,
  811. .write_super = exofs_write_super,
  812. .sync_fs = exofs_sync_fs,
  813. .statfs = exofs_statfs,
  814. };
  815. /******************************************************************************
  816. * EXPORT OPERATIONS
  817. *****************************************************************************/
  818. static struct dentry *exofs_get_parent(struct dentry *child)
  819. {
  820. unsigned long ino = exofs_parent_ino(child);
  821. if (!ino)
  822. return ERR_PTR(-ESTALE);
  823. return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
  824. }
  825. static struct inode *exofs_nfs_get_inode(struct super_block *sb,
  826. u64 ino, u32 generation)
  827. {
  828. struct inode *inode;
  829. inode = exofs_iget(sb, ino);
  830. if (IS_ERR(inode))
  831. return ERR_CAST(inode);
  832. if (generation && inode->i_generation != generation) {
  833. /* we didn't find the right inode.. */
  834. iput(inode);
  835. return ERR_PTR(-ESTALE);
  836. }
  837. return inode;
  838. }
  839. static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
  840. struct fid *fid, int fh_len, int fh_type)
  841. {
  842. return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
  843. exofs_nfs_get_inode);
  844. }
  845. static struct dentry *exofs_fh_to_parent(struct super_block *sb,
  846. struct fid *fid, int fh_len, int fh_type)
  847. {
  848. return generic_fh_to_parent(sb, fid, fh_len, fh_type,
  849. exofs_nfs_get_inode);
  850. }
  851. static const struct export_operations exofs_export_ops = {
  852. .fh_to_dentry = exofs_fh_to_dentry,
  853. .fh_to_parent = exofs_fh_to_parent,
  854. .get_parent = exofs_get_parent,
  855. };
  856. /******************************************************************************
  857. * INSMOD/RMMOD
  858. *****************************************************************************/
  859. /*
  860. * struct that describes this file system
  861. */
  862. static struct file_system_type exofs_type = {
  863. .owner = THIS_MODULE,
  864. .name = "exofs",
  865. .mount = exofs_mount,
  866. .kill_sb = generic_shutdown_super,
  867. };
  868. MODULE_ALIAS_FS("exofs");
  869. static int __init init_exofs(void)
  870. {
  871. int err;
  872. err = init_inodecache();
  873. if (err)
  874. goto out;
  875. err = register_filesystem(&exofs_type);
  876. if (err)
  877. goto out_d;
  878. return 0;
  879. out_d:
  880. destroy_inodecache();
  881. out:
  882. return err;
  883. }
  884. static void __exit exit_exofs(void)
  885. {
  886. unregister_filesystem(&exofs_type);
  887. destroy_inodecache();
  888. }
  889. MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
  890. MODULE_DESCRIPTION("exofs");
  891. MODULE_LICENSE("GPL");
  892. module_init(init_exofs)
  893. module_exit(exit_exofs)