dir.c 51 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984
  1. // SPDX-License-Identifier: GPL-2.0
  2. #include <linux/ceph/ceph_debug.h>
  3. #include <linux/spinlock.h>
  4. #include <linux/namei.h>
  5. #include <linux/slab.h>
  6. #include <linux/sched.h>
  7. #include <linux/xattr.h>
  8. #include "super.h"
  9. #include "mds_client.h"
  10. /*
  11. * Directory operations: readdir, lookup, create, link, unlink,
  12. * rename, etc.
  13. */
  14. /*
  15. * Ceph MDS operations are specified in terms of a base ino and
  16. * relative path. Thus, the client can specify an operation on a
  17. * specific inode (e.g., a getattr due to fstat(2)), or as a path
  18. * relative to, say, the root directory.
  19. *
  20. * Normally, we limit ourselves to strict inode ops (no path component)
  21. * or dentry operations (a single path component relative to an ino). The
  22. * exception to this is open_root_dentry(), which will open the mount
  23. * point by name.
  24. */
  25. const struct dentry_operations ceph_dentry_ops;
  26. static bool __dentry_lease_is_valid(struct ceph_dentry_info *di);
  27. static int __dir_lease_try_check(const struct dentry *dentry);
  28. /*
  29. * Initialize ceph dentry state.
  30. */
  31. static int ceph_d_init(struct dentry *dentry)
  32. {
  33. struct ceph_dentry_info *di;
  34. struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb);
  35. di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
  36. if (!di)
  37. return -ENOMEM; /* oh well */
  38. di->dentry = dentry;
  39. di->lease_session = NULL;
  40. di->time = jiffies;
  41. dentry->d_fsdata = di;
  42. INIT_LIST_HEAD(&di->lease_list);
  43. atomic64_inc(&mdsc->metric.total_dentries);
  44. return 0;
  45. }
  46. /*
  47. * for f_pos for readdir:
  48. * - hash order:
  49. * (0xff << 52) | ((24 bits hash) << 28) |
  50. * (the nth entry has hash collision);
  51. * - frag+name order;
  52. * ((frag value) << 28) | (the nth entry in frag);
  53. */
  54. #define OFFSET_BITS 28
  55. #define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
  56. #define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
  57. loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
  58. {
  59. loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
  60. if (hash_order)
  61. fpos |= HASH_ORDER;
  62. return fpos;
  63. }
  64. static bool is_hash_order(loff_t p)
  65. {
  66. return (p & HASH_ORDER) == HASH_ORDER;
  67. }
  68. static unsigned fpos_frag(loff_t p)
  69. {
  70. return p >> OFFSET_BITS;
  71. }
  72. static unsigned fpos_hash(loff_t p)
  73. {
  74. return ceph_frag_value(fpos_frag(p));
  75. }
  76. static unsigned fpos_off(loff_t p)
  77. {
  78. return p & OFFSET_MASK;
  79. }
  80. static int fpos_cmp(loff_t l, loff_t r)
  81. {
  82. int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
  83. if (v)
  84. return v;
  85. return (int)(fpos_off(l) - fpos_off(r));
  86. }
  87. /*
  88. * make note of the last dentry we read, so we can
  89. * continue at the same lexicographical point,
  90. * regardless of what dir changes take place on the
  91. * server.
  92. */
  93. static int note_last_dentry(struct ceph_dir_file_info *dfi, const char *name,
  94. int len, unsigned next_offset)
  95. {
  96. char *buf = kmalloc(len+1, GFP_KERNEL);
  97. if (!buf)
  98. return -ENOMEM;
  99. kfree(dfi->last_name);
  100. dfi->last_name = buf;
  101. memcpy(dfi->last_name, name, len);
  102. dfi->last_name[len] = 0;
  103. dfi->next_offset = next_offset;
  104. dout("note_last_dentry '%s'\n", dfi->last_name);
  105. return 0;
  106. }
  107. static struct dentry *
  108. __dcache_find_get_entry(struct dentry *parent, u64 idx,
  109. struct ceph_readdir_cache_control *cache_ctl)
  110. {
  111. struct inode *dir = d_inode(parent);
  112. struct dentry *dentry;
  113. unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
  114. loff_t ptr_pos = idx * sizeof(struct dentry *);
  115. pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
  116. if (ptr_pos >= i_size_read(dir))
  117. return NULL;
  118. if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
  119. ceph_readdir_cache_release(cache_ctl);
  120. cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
  121. if (!cache_ctl->page) {
  122. dout(" page %lu not found\n", ptr_pgoff);
  123. return ERR_PTR(-EAGAIN);
  124. }
  125. /* reading/filling the cache are serialized by
  126. i_mutex, no need to use page lock */
  127. unlock_page(cache_ctl->page);
  128. cache_ctl->dentries = kmap(cache_ctl->page);
  129. }
  130. cache_ctl->index = idx & idx_mask;
  131. rcu_read_lock();
  132. spin_lock(&parent->d_lock);
  133. /* check i_size again here, because empty directory can be
  134. * marked as complete while not holding the i_mutex. */
  135. if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
  136. dentry = cache_ctl->dentries[cache_ctl->index];
  137. else
  138. dentry = NULL;
  139. spin_unlock(&parent->d_lock);
  140. if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
  141. dentry = NULL;
  142. rcu_read_unlock();
  143. return dentry ? : ERR_PTR(-EAGAIN);
  144. }
  145. /*
  146. * When possible, we try to satisfy a readdir by peeking at the
  147. * dcache. We make this work by carefully ordering dentries on
  148. * d_child when we initially get results back from the MDS, and
  149. * falling back to a "normal" sync readdir if any dentries in the dir
  150. * are dropped.
  151. *
  152. * Complete dir indicates that we have all dentries in the dir. It is
  153. * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
  154. * the MDS if/when the directory is modified).
  155. */
  156. static int __dcache_readdir(struct file *file, struct dir_context *ctx,
  157. int shared_gen)
  158. {
  159. struct ceph_dir_file_info *dfi = file->private_data;
  160. struct dentry *parent = file->f_path.dentry;
  161. struct inode *dir = d_inode(parent);
  162. struct dentry *dentry, *last = NULL;
  163. struct ceph_dentry_info *di;
  164. struct ceph_readdir_cache_control cache_ctl = {};
  165. u64 idx = 0;
  166. int err = 0;
  167. dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos);
  168. /* search start position */
  169. if (ctx->pos > 2) {
  170. u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
  171. while (count > 0) {
  172. u64 step = count >> 1;
  173. dentry = __dcache_find_get_entry(parent, idx + step,
  174. &cache_ctl);
  175. if (!dentry) {
  176. /* use linar search */
  177. idx = 0;
  178. break;
  179. }
  180. if (IS_ERR(dentry)) {
  181. err = PTR_ERR(dentry);
  182. goto out;
  183. }
  184. di = ceph_dentry(dentry);
  185. spin_lock(&dentry->d_lock);
  186. if (fpos_cmp(di->offset, ctx->pos) < 0) {
  187. idx += step + 1;
  188. count -= step + 1;
  189. } else {
  190. count = step;
  191. }
  192. spin_unlock(&dentry->d_lock);
  193. dput(dentry);
  194. }
  195. dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
  196. }
  197. for (;;) {
  198. bool emit_dentry = false;
  199. dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
  200. if (!dentry) {
  201. dfi->file_info.flags |= CEPH_F_ATEND;
  202. err = 0;
  203. break;
  204. }
  205. if (IS_ERR(dentry)) {
  206. err = PTR_ERR(dentry);
  207. goto out;
  208. }
  209. spin_lock(&dentry->d_lock);
  210. di = ceph_dentry(dentry);
  211. if (d_unhashed(dentry) ||
  212. d_really_is_negative(dentry) ||
  213. di->lease_shared_gen != shared_gen) {
  214. spin_unlock(&dentry->d_lock);
  215. dput(dentry);
  216. err = -EAGAIN;
  217. goto out;
  218. }
  219. if (fpos_cmp(ctx->pos, di->offset) <= 0) {
  220. __ceph_dentry_dir_lease_touch(di);
  221. emit_dentry = true;
  222. }
  223. spin_unlock(&dentry->d_lock);
  224. if (emit_dentry) {
  225. dout(" %llx dentry %p %pd %p\n", di->offset,
  226. dentry, dentry, d_inode(dentry));
  227. ctx->pos = di->offset;
  228. if (!dir_emit(ctx, dentry->d_name.name,
  229. dentry->d_name.len, ceph_present_inode(d_inode(dentry)),
  230. d_inode(dentry)->i_mode >> 12)) {
  231. dput(dentry);
  232. err = 0;
  233. break;
  234. }
  235. ctx->pos++;
  236. if (last)
  237. dput(last);
  238. last = dentry;
  239. } else {
  240. dput(dentry);
  241. }
  242. }
  243. out:
  244. ceph_readdir_cache_release(&cache_ctl);
  245. if (last) {
  246. int ret;
  247. di = ceph_dentry(last);
  248. ret = note_last_dentry(dfi, last->d_name.name, last->d_name.len,
  249. fpos_off(di->offset) + 1);
  250. if (ret < 0)
  251. err = ret;
  252. dput(last);
  253. /* last_name no longer match cache index */
  254. if (dfi->readdir_cache_idx >= 0) {
  255. dfi->readdir_cache_idx = -1;
  256. dfi->dir_release_count = 0;
  257. }
  258. }
  259. return err;
  260. }
  261. static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos)
  262. {
  263. if (!dfi->last_readdir)
  264. return true;
  265. if (is_hash_order(pos))
  266. return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos));
  267. else
  268. return dfi->frag != fpos_frag(pos);
  269. }
  270. static int ceph_readdir(struct file *file, struct dir_context *ctx)
  271. {
  272. struct ceph_dir_file_info *dfi = file->private_data;
  273. struct inode *inode = file_inode(file);
  274. struct ceph_inode_info *ci = ceph_inode(inode);
  275. struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
  276. struct ceph_mds_client *mdsc = fsc->mdsc;
  277. int i;
  278. int err;
  279. unsigned frag = -1;
  280. struct ceph_mds_reply_info_parsed *rinfo;
  281. dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
  282. if (dfi->file_info.flags & CEPH_F_ATEND)
  283. return 0;
  284. /* always start with . and .. */
  285. if (ctx->pos == 0) {
  286. dout("readdir off 0 -> '.'\n");
  287. if (!dir_emit(ctx, ".", 1, ceph_present_inode(inode),
  288. inode->i_mode >> 12))
  289. return 0;
  290. ctx->pos = 1;
  291. }
  292. if (ctx->pos == 1) {
  293. u64 ino;
  294. struct dentry *dentry = file->f_path.dentry;
  295. spin_lock(&dentry->d_lock);
  296. ino = ceph_present_inode(dentry->d_parent->d_inode);
  297. spin_unlock(&dentry->d_lock);
  298. dout("readdir off 1 -> '..'\n");
  299. if (!dir_emit(ctx, "..", 2, ino, inode->i_mode >> 12))
  300. return 0;
  301. ctx->pos = 2;
  302. }
  303. spin_lock(&ci->i_ceph_lock);
  304. /* request Fx cap. if have Fx, we don't need to release Fs cap
  305. * for later create/unlink. */
  306. __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_WR);
  307. /* can we use the dcache? */
  308. if (ceph_test_mount_opt(fsc, DCACHE) &&
  309. !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
  310. ceph_snap(inode) != CEPH_SNAPDIR &&
  311. __ceph_dir_is_complete_ordered(ci) &&
  312. __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
  313. int shared_gen = atomic_read(&ci->i_shared_gen);
  314. spin_unlock(&ci->i_ceph_lock);
  315. err = __dcache_readdir(file, ctx, shared_gen);
  316. if (err != -EAGAIN)
  317. return err;
  318. } else {
  319. spin_unlock(&ci->i_ceph_lock);
  320. }
  321. /* proceed with a normal readdir */
  322. more:
  323. /* do we have the correct frag content buffered? */
  324. if (need_send_readdir(dfi, ctx->pos)) {
  325. struct ceph_mds_request *req;
  326. int op = ceph_snap(inode) == CEPH_SNAPDIR ?
  327. CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
  328. /* discard old result, if any */
  329. if (dfi->last_readdir) {
  330. ceph_mdsc_put_request(dfi->last_readdir);
  331. dfi->last_readdir = NULL;
  332. }
  333. if (is_hash_order(ctx->pos)) {
  334. /* fragtree isn't always accurate. choose frag
  335. * based on previous reply when possible. */
  336. if (frag == (unsigned)-1)
  337. frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
  338. NULL, NULL);
  339. } else {
  340. frag = fpos_frag(ctx->pos);
  341. }
  342. dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
  343. ceph_vinop(inode), frag, dfi->last_name);
  344. req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
  345. if (IS_ERR(req))
  346. return PTR_ERR(req);
  347. err = ceph_alloc_readdir_reply_buffer(req, inode);
  348. if (err) {
  349. ceph_mdsc_put_request(req);
  350. return err;
  351. }
  352. /* hints to request -> mds selection code */
  353. req->r_direct_mode = USE_AUTH_MDS;
  354. if (op == CEPH_MDS_OP_READDIR) {
  355. req->r_direct_hash = ceph_frag_value(frag);
  356. __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
  357. req->r_inode_drop = CEPH_CAP_FILE_EXCL;
  358. }
  359. if (dfi->last_name) {
  360. req->r_path2 = kstrdup(dfi->last_name, GFP_KERNEL);
  361. if (!req->r_path2) {
  362. ceph_mdsc_put_request(req);
  363. return -ENOMEM;
  364. }
  365. } else if (is_hash_order(ctx->pos)) {
  366. req->r_args.readdir.offset_hash =
  367. cpu_to_le32(fpos_hash(ctx->pos));
  368. }
  369. req->r_dir_release_cnt = dfi->dir_release_count;
  370. req->r_dir_ordered_cnt = dfi->dir_ordered_count;
  371. req->r_readdir_cache_idx = dfi->readdir_cache_idx;
  372. req->r_readdir_offset = dfi->next_offset;
  373. req->r_args.readdir.frag = cpu_to_le32(frag);
  374. req->r_args.readdir.flags =
  375. cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
  376. req->r_inode = inode;
  377. ihold(inode);
  378. req->r_dentry = dget(file->f_path.dentry);
  379. err = ceph_mdsc_do_request(mdsc, NULL, req);
  380. if (err < 0) {
  381. ceph_mdsc_put_request(req);
  382. return err;
  383. }
  384. dout("readdir got and parsed readdir result=%d on "
  385. "frag %x, end=%d, complete=%d, hash_order=%d\n",
  386. err, frag,
  387. (int)req->r_reply_info.dir_end,
  388. (int)req->r_reply_info.dir_complete,
  389. (int)req->r_reply_info.hash_order);
  390. rinfo = &req->r_reply_info;
  391. if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
  392. frag = le32_to_cpu(rinfo->dir_dir->frag);
  393. if (!rinfo->hash_order) {
  394. dfi->next_offset = req->r_readdir_offset;
  395. /* adjust ctx->pos to beginning of frag */
  396. ctx->pos = ceph_make_fpos(frag,
  397. dfi->next_offset,
  398. false);
  399. }
  400. }
  401. dfi->frag = frag;
  402. dfi->last_readdir = req;
  403. if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
  404. dfi->readdir_cache_idx = req->r_readdir_cache_idx;
  405. if (dfi->readdir_cache_idx < 0) {
  406. /* preclude from marking dir ordered */
  407. dfi->dir_ordered_count = 0;
  408. } else if (ceph_frag_is_leftmost(frag) &&
  409. dfi->next_offset == 2) {
  410. /* note dir version at start of readdir so
  411. * we can tell if any dentries get dropped */
  412. dfi->dir_release_count = req->r_dir_release_cnt;
  413. dfi->dir_ordered_count = req->r_dir_ordered_cnt;
  414. }
  415. } else {
  416. dout("readdir !did_prepopulate\n");
  417. /* disable readdir cache */
  418. dfi->readdir_cache_idx = -1;
  419. /* preclude from marking dir complete */
  420. dfi->dir_release_count = 0;
  421. }
  422. /* note next offset and last dentry name */
  423. if (rinfo->dir_nr > 0) {
  424. struct ceph_mds_reply_dir_entry *rde =
  425. rinfo->dir_entries + (rinfo->dir_nr-1);
  426. unsigned next_offset = req->r_reply_info.dir_end ?
  427. 2 : (fpos_off(rde->offset) + 1);
  428. err = note_last_dentry(dfi, rde->name, rde->name_len,
  429. next_offset);
  430. if (err) {
  431. ceph_mdsc_put_request(dfi->last_readdir);
  432. dfi->last_readdir = NULL;
  433. return err;
  434. }
  435. } else if (req->r_reply_info.dir_end) {
  436. dfi->next_offset = 2;
  437. /* keep last name */
  438. }
  439. }
  440. rinfo = &dfi->last_readdir->r_reply_info;
  441. dout("readdir frag %x num %d pos %llx chunk first %llx\n",
  442. dfi->frag, rinfo->dir_nr, ctx->pos,
  443. rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
  444. i = 0;
  445. /* search start position */
  446. if (rinfo->dir_nr > 0) {
  447. int step, nr = rinfo->dir_nr;
  448. while (nr > 0) {
  449. step = nr >> 1;
  450. if (rinfo->dir_entries[i + step].offset < ctx->pos) {
  451. i += step + 1;
  452. nr -= step + 1;
  453. } else {
  454. nr = step;
  455. }
  456. }
  457. }
  458. for (; i < rinfo->dir_nr; i++) {
  459. struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
  460. BUG_ON(rde->offset < ctx->pos);
  461. ctx->pos = rde->offset;
  462. dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
  463. i, rinfo->dir_nr, ctx->pos,
  464. rde->name_len, rde->name, &rde->inode.in);
  465. BUG_ON(!rde->inode.in);
  466. if (!dir_emit(ctx, rde->name, rde->name_len,
  467. ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)),
  468. le32_to_cpu(rde->inode.in->mode) >> 12)) {
  469. /*
  470. * NOTE: Here no need to put the 'dfi->last_readdir',
  471. * because when dir_emit stops us it's most likely
  472. * doesn't have enough memory, etc. So for next readdir
  473. * it will continue.
  474. */
  475. dout("filldir stopping us...\n");
  476. return 0;
  477. }
  478. ctx->pos++;
  479. }
  480. ceph_mdsc_put_request(dfi->last_readdir);
  481. dfi->last_readdir = NULL;
  482. if (dfi->next_offset > 2) {
  483. frag = dfi->frag;
  484. goto more;
  485. }
  486. /* more frags? */
  487. if (!ceph_frag_is_rightmost(dfi->frag)) {
  488. frag = ceph_frag_next(dfi->frag);
  489. if (is_hash_order(ctx->pos)) {
  490. loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
  491. dfi->next_offset, true);
  492. if (new_pos > ctx->pos)
  493. ctx->pos = new_pos;
  494. /* keep last_name */
  495. } else {
  496. ctx->pos = ceph_make_fpos(frag, dfi->next_offset,
  497. false);
  498. kfree(dfi->last_name);
  499. dfi->last_name = NULL;
  500. }
  501. dout("readdir next frag is %x\n", frag);
  502. goto more;
  503. }
  504. dfi->file_info.flags |= CEPH_F_ATEND;
  505. /*
  506. * if dir_release_count still matches the dir, no dentries
  507. * were released during the whole readdir, and we should have
  508. * the complete dir contents in our cache.
  509. */
  510. if (atomic64_read(&ci->i_release_count) ==
  511. dfi->dir_release_count) {
  512. spin_lock(&ci->i_ceph_lock);
  513. if (dfi->dir_ordered_count ==
  514. atomic64_read(&ci->i_ordered_count)) {
  515. dout(" marking %p complete and ordered\n", inode);
  516. /* use i_size to track number of entries in
  517. * readdir cache */
  518. BUG_ON(dfi->readdir_cache_idx < 0);
  519. i_size_write(inode, dfi->readdir_cache_idx *
  520. sizeof(struct dentry*));
  521. } else {
  522. dout(" marking %p complete\n", inode);
  523. }
  524. __ceph_dir_set_complete(ci, dfi->dir_release_count,
  525. dfi->dir_ordered_count);
  526. spin_unlock(&ci->i_ceph_lock);
  527. }
  528. dout("readdir %p file %p done.\n", inode, file);
  529. return 0;
  530. }
  531. static void reset_readdir(struct ceph_dir_file_info *dfi)
  532. {
  533. if (dfi->last_readdir) {
  534. ceph_mdsc_put_request(dfi->last_readdir);
  535. dfi->last_readdir = NULL;
  536. }
  537. kfree(dfi->last_name);
  538. dfi->last_name = NULL;
  539. dfi->dir_release_count = 0;
  540. dfi->readdir_cache_idx = -1;
  541. dfi->next_offset = 2; /* compensate for . and .. */
  542. dfi->file_info.flags &= ~CEPH_F_ATEND;
  543. }
  544. /*
  545. * discard buffered readdir content on seekdir(0), or seek to new frag,
  546. * or seek prior to current chunk
  547. */
  548. static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos)
  549. {
  550. struct ceph_mds_reply_info_parsed *rinfo;
  551. loff_t chunk_offset;
  552. if (new_pos == 0)
  553. return true;
  554. if (is_hash_order(new_pos)) {
  555. /* no need to reset last_name for a forward seek when
  556. * dentries are sotred in hash order */
  557. } else if (dfi->frag != fpos_frag(new_pos)) {
  558. return true;
  559. }
  560. rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL;
  561. if (!rinfo || !rinfo->dir_nr)
  562. return true;
  563. chunk_offset = rinfo->dir_entries[0].offset;
  564. return new_pos < chunk_offset ||
  565. is_hash_order(new_pos) != is_hash_order(chunk_offset);
  566. }
  567. static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
  568. {
  569. struct ceph_dir_file_info *dfi = file->private_data;
  570. struct inode *inode = file->f_mapping->host;
  571. loff_t retval;
  572. inode_lock(inode);
  573. retval = -EINVAL;
  574. switch (whence) {
  575. case SEEK_CUR:
  576. offset += file->f_pos;
  577. case SEEK_SET:
  578. break;
  579. case SEEK_END:
  580. retval = -EOPNOTSUPP;
  581. default:
  582. goto out;
  583. }
  584. if (offset >= 0) {
  585. if (need_reset_readdir(dfi, offset)) {
  586. dout("dir_llseek dropping %p content\n", file);
  587. reset_readdir(dfi);
  588. } else if (is_hash_order(offset) && offset > file->f_pos) {
  589. /* for hash offset, we don't know if a forward seek
  590. * is within same frag */
  591. dfi->dir_release_count = 0;
  592. dfi->readdir_cache_idx = -1;
  593. }
  594. if (offset != file->f_pos) {
  595. file->f_pos = offset;
  596. file->f_version = 0;
  597. dfi->file_info.flags &= ~CEPH_F_ATEND;
  598. }
  599. retval = offset;
  600. }
  601. out:
  602. inode_unlock(inode);
  603. return retval;
  604. }
  605. /*
  606. * Handle lookups for the hidden .snap directory.
  607. */
  608. int ceph_handle_snapdir(struct ceph_mds_request *req,
  609. struct dentry *dentry, int err)
  610. {
  611. struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
  612. struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
  613. /* .snap dir? */
  614. if (err == -ENOENT &&
  615. ceph_snap(parent) == CEPH_NOSNAP &&
  616. strcmp(dentry->d_name.name,
  617. fsc->mount_options->snapdir_name) == 0) {
  618. struct inode *inode = ceph_get_snapdir(parent);
  619. dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
  620. dentry, dentry, inode);
  621. BUG_ON(!d_unhashed(dentry));
  622. d_add(dentry, inode);
  623. err = 0;
  624. }
  625. return err;
  626. }
  627. /*
  628. * Figure out final result of a lookup/open request.
  629. *
  630. * Mainly, make sure we return the final req->r_dentry (if it already
  631. * existed) in place of the original VFS-provided dentry when they
  632. * differ.
  633. *
  634. * Gracefully handle the case where the MDS replies with -ENOENT and
  635. * no trace (which it may do, at its discretion, e.g., if it doesn't
  636. * care to issue a lease on the negative dentry).
  637. */
  638. struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
  639. struct dentry *dentry, int err)
  640. {
  641. if (err == -ENOENT) {
  642. /* no trace? */
  643. err = 0;
  644. if (!req->r_reply_info.head->is_dentry) {
  645. dout("ENOENT and no trace, dentry %p inode %p\n",
  646. dentry, d_inode(dentry));
  647. if (d_really_is_positive(dentry)) {
  648. d_drop(dentry);
  649. err = -ENOENT;
  650. } else {
  651. d_add(dentry, NULL);
  652. }
  653. }
  654. }
  655. if (err)
  656. dentry = ERR_PTR(err);
  657. else if (dentry != req->r_dentry)
  658. dentry = dget(req->r_dentry); /* we got spliced */
  659. else
  660. dentry = NULL;
  661. return dentry;
  662. }
  663. static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
  664. {
  665. return ceph_ino(inode) == CEPH_INO_ROOT &&
  666. strncmp(dentry->d_name.name, ".ceph", 5) == 0;
  667. }
  668. /*
  669. * Look up a single dir entry. If there is a lookup intent, inform
  670. * the MDS so that it gets our 'caps wanted' value in a single op.
  671. */
  672. static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
  673. unsigned int flags)
  674. {
  675. struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
  676. struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
  677. struct ceph_mds_request *req;
  678. int op;
  679. int mask;
  680. int err;
  681. dout("lookup %p dentry %p '%pd'\n",
  682. dir, dentry, dentry);
  683. if (dentry->d_name.len > NAME_MAX)
  684. return ERR_PTR(-ENAMETOOLONG);
  685. /* can we conclude ENOENT locally? */
  686. if (d_really_is_negative(dentry)) {
  687. struct ceph_inode_info *ci = ceph_inode(dir);
  688. struct ceph_dentry_info *di = ceph_dentry(dentry);
  689. spin_lock(&ci->i_ceph_lock);
  690. dout(" dir %p flags are 0x%lx\n", dir, ci->i_ceph_flags);
  691. if (strncmp(dentry->d_name.name,
  692. fsc->mount_options->snapdir_name,
  693. dentry->d_name.len) &&
  694. !is_root_ceph_dentry(dir, dentry) &&
  695. ceph_test_mount_opt(fsc, DCACHE) &&
  696. __ceph_dir_is_complete(ci) &&
  697. __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
  698. __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
  699. spin_unlock(&ci->i_ceph_lock);
  700. dout(" dir %p complete, -ENOENT\n", dir);
  701. d_add(dentry, NULL);
  702. di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
  703. return NULL;
  704. }
  705. spin_unlock(&ci->i_ceph_lock);
  706. }
  707. op = ceph_snap(dir) == CEPH_SNAPDIR ?
  708. CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
  709. req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
  710. if (IS_ERR(req))
  711. return ERR_CAST(req);
  712. req->r_dentry = dget(dentry);
  713. req->r_num_caps = 2;
  714. mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
  715. if (ceph_security_xattr_wanted(dir))
  716. mask |= CEPH_CAP_XATTR_SHARED;
  717. req->r_args.getattr.mask = cpu_to_le32(mask);
  718. req->r_parent = dir;
  719. set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
  720. err = ceph_mdsc_do_request(mdsc, NULL, req);
  721. err = ceph_handle_snapdir(req, dentry, err);
  722. dentry = ceph_finish_lookup(req, dentry, err);
  723. ceph_mdsc_put_request(req); /* will dput(dentry) */
  724. dout("lookup result=%p\n", dentry);
  725. return dentry;
  726. }
  727. /*
  728. * If we do a create but get no trace back from the MDS, follow up with
  729. * a lookup (the VFS expects us to link up the provided dentry).
  730. */
  731. int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
  732. {
  733. struct dentry *result = ceph_lookup(dir, dentry, 0);
  734. if (result && !IS_ERR(result)) {
  735. /*
  736. * We created the item, then did a lookup, and found
  737. * it was already linked to another inode we already
  738. * had in our cache (and thus got spliced). To not
  739. * confuse VFS (especially when inode is a directory),
  740. * we don't link our dentry to that inode, return an
  741. * error instead.
  742. *
  743. * This event should be rare and it happens only when
  744. * we talk to old MDS. Recent MDS does not send traceless
  745. * reply for request that creates new inode.
  746. */
  747. d_drop(result);
  748. return -ESTALE;
  749. }
  750. return PTR_ERR(result);
  751. }
  752. static int ceph_mknod(struct inode *dir, struct dentry *dentry,
  753. umode_t mode, dev_t rdev)
  754. {
  755. struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
  756. struct ceph_mds_request *req;
  757. struct ceph_acl_sec_ctx as_ctx = {};
  758. int err;
  759. if (ceph_snap(dir) != CEPH_NOSNAP)
  760. return -EROFS;
  761. if (ceph_quota_is_max_files_exceeded(dir)) {
  762. err = -EDQUOT;
  763. goto out;
  764. }
  765. err = ceph_pre_init_acls(dir, &mode, &as_ctx);
  766. if (err < 0)
  767. goto out;
  768. err = ceph_security_init_secctx(dentry, mode, &as_ctx);
  769. if (err < 0)
  770. goto out;
  771. dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
  772. dir, dentry, mode, rdev);
  773. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
  774. if (IS_ERR(req)) {
  775. err = PTR_ERR(req);
  776. goto out;
  777. }
  778. req->r_dentry = dget(dentry);
  779. req->r_num_caps = 2;
  780. req->r_parent = dir;
  781. set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
  782. req->r_args.mknod.mode = cpu_to_le32(mode);
  783. req->r_args.mknod.rdev = cpu_to_le32(rdev);
  784. req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
  785. req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
  786. if (as_ctx.pagelist) {
  787. req->r_pagelist = as_ctx.pagelist;
  788. as_ctx.pagelist = NULL;
  789. }
  790. err = ceph_mdsc_do_request(mdsc, dir, req);
  791. if (!err && !req->r_reply_info.head->is_dentry)
  792. err = ceph_handle_notrace_create(dir, dentry);
  793. ceph_mdsc_put_request(req);
  794. out:
  795. if (!err)
  796. ceph_init_inode_acls(d_inode(dentry), &as_ctx);
  797. else
  798. d_drop(dentry);
  799. ceph_release_acl_sec_ctx(&as_ctx);
  800. return err;
  801. }
  802. static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
  803. bool excl)
  804. {
  805. return ceph_mknod(dir, dentry, mode, 0);
  806. }
  807. static int ceph_symlink(struct inode *dir, struct dentry *dentry,
  808. const char *dest)
  809. {
  810. struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
  811. struct ceph_mds_request *req;
  812. struct ceph_acl_sec_ctx as_ctx = {};
  813. int err;
  814. if (ceph_snap(dir) != CEPH_NOSNAP)
  815. return -EROFS;
  816. if (ceph_quota_is_max_files_exceeded(dir)) {
  817. err = -EDQUOT;
  818. goto out;
  819. }
  820. err = ceph_security_init_secctx(dentry, S_IFLNK | 0777, &as_ctx);
  821. if (err < 0)
  822. goto out;
  823. dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
  824. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
  825. if (IS_ERR(req)) {
  826. err = PTR_ERR(req);
  827. goto out;
  828. }
  829. req->r_path2 = kstrdup(dest, GFP_KERNEL);
  830. if (!req->r_path2) {
  831. err = -ENOMEM;
  832. ceph_mdsc_put_request(req);
  833. goto out;
  834. }
  835. req->r_parent = dir;
  836. set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
  837. req->r_dentry = dget(dentry);
  838. req->r_num_caps = 2;
  839. req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
  840. req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
  841. if (as_ctx.pagelist) {
  842. req->r_pagelist = as_ctx.pagelist;
  843. as_ctx.pagelist = NULL;
  844. }
  845. err = ceph_mdsc_do_request(mdsc, dir, req);
  846. if (!err && !req->r_reply_info.head->is_dentry)
  847. err = ceph_handle_notrace_create(dir, dentry);
  848. ceph_mdsc_put_request(req);
  849. out:
  850. if (err)
  851. d_drop(dentry);
  852. ceph_release_acl_sec_ctx(&as_ctx);
  853. return err;
  854. }
  855. static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
  856. {
  857. struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
  858. struct ceph_mds_request *req;
  859. struct ceph_acl_sec_ctx as_ctx = {};
  860. int err = -EROFS;
  861. int op;
  862. if (ceph_snap(dir) == CEPH_SNAPDIR) {
  863. /* mkdir .snap/foo is a MKSNAP */
  864. op = CEPH_MDS_OP_MKSNAP;
  865. dout("mksnap dir %p snap '%pd' dn %p\n", dir,
  866. dentry, dentry);
  867. } else if (ceph_snap(dir) == CEPH_NOSNAP) {
  868. dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
  869. op = CEPH_MDS_OP_MKDIR;
  870. } else {
  871. goto out;
  872. }
  873. if (op == CEPH_MDS_OP_MKDIR &&
  874. ceph_quota_is_max_files_exceeded(dir)) {
  875. err = -EDQUOT;
  876. goto out;
  877. }
  878. mode |= S_IFDIR;
  879. err = ceph_pre_init_acls(dir, &mode, &as_ctx);
  880. if (err < 0)
  881. goto out;
  882. err = ceph_security_init_secctx(dentry, mode, &as_ctx);
  883. if (err < 0)
  884. goto out;
  885. req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
  886. if (IS_ERR(req)) {
  887. err = PTR_ERR(req);
  888. goto out;
  889. }
  890. req->r_dentry = dget(dentry);
  891. req->r_num_caps = 2;
  892. req->r_parent = dir;
  893. set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
  894. req->r_args.mkdir.mode = cpu_to_le32(mode);
  895. req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
  896. req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
  897. if (as_ctx.pagelist) {
  898. req->r_pagelist = as_ctx.pagelist;
  899. as_ctx.pagelist = NULL;
  900. }
  901. err = ceph_mdsc_do_request(mdsc, dir, req);
  902. if (!err &&
  903. !req->r_reply_info.head->is_target &&
  904. !req->r_reply_info.head->is_dentry)
  905. err = ceph_handle_notrace_create(dir, dentry);
  906. ceph_mdsc_put_request(req);
  907. out:
  908. if (!err)
  909. ceph_init_inode_acls(d_inode(dentry), &as_ctx);
  910. else
  911. d_drop(dentry);
  912. ceph_release_acl_sec_ctx(&as_ctx);
  913. return err;
  914. }
  915. static int ceph_link(struct dentry *old_dentry, struct inode *dir,
  916. struct dentry *dentry)
  917. {
  918. struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
  919. struct ceph_mds_request *req;
  920. int err;
  921. if (ceph_snap(dir) != CEPH_NOSNAP)
  922. return -EROFS;
  923. dout("link in dir %p old_dentry %p dentry %p\n", dir,
  924. old_dentry, dentry);
  925. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
  926. if (IS_ERR(req)) {
  927. d_drop(dentry);
  928. return PTR_ERR(req);
  929. }
  930. req->r_dentry = dget(dentry);
  931. req->r_num_caps = 2;
  932. req->r_old_dentry = dget(old_dentry);
  933. req->r_parent = dir;
  934. set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
  935. req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
  936. req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
  937. /* release LINK_SHARED on source inode (mds will lock it) */
  938. req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
  939. err = ceph_mdsc_do_request(mdsc, dir, req);
  940. if (err) {
  941. d_drop(dentry);
  942. } else if (!req->r_reply_info.head->is_dentry) {
  943. ihold(d_inode(old_dentry));
  944. d_instantiate(dentry, d_inode(old_dentry));
  945. }
  946. ceph_mdsc_put_request(req);
  947. return err;
  948. }
  949. static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc,
  950. struct ceph_mds_request *req)
  951. {
  952. int result = req->r_err ? req->r_err :
  953. le32_to_cpu(req->r_reply_info.head->result);
  954. if (result == -EJUKEBOX)
  955. goto out;
  956. /* If op failed, mark everyone involved for errors */
  957. if (result) {
  958. int pathlen = 0;
  959. u64 base = 0;
  960. char *path = ceph_mdsc_build_path(req->r_dentry, &pathlen,
  961. &base, 0);
  962. /* mark error on parent + clear complete */
  963. mapping_set_error(req->r_parent->i_mapping, result);
  964. ceph_dir_clear_complete(req->r_parent);
  965. /* drop the dentry -- we don't know its status */
  966. if (!d_unhashed(req->r_dentry))
  967. d_drop(req->r_dentry);
  968. /* mark inode itself for an error (since metadata is bogus) */
  969. mapping_set_error(req->r_old_inode->i_mapping, result);
  970. pr_warn("ceph: async unlink failure path=(%llx)%s result=%d!\n",
  971. base, IS_ERR(path) ? "<<bad>>" : path, result);
  972. ceph_mdsc_free_path(path, pathlen);
  973. }
  974. out:
  975. iput(req->r_old_inode);
  976. ceph_mdsc_release_dir_caps(req);
  977. }
  978. static int get_caps_for_async_unlink(struct inode *dir, struct dentry *dentry)
  979. {
  980. struct ceph_inode_info *ci = ceph_inode(dir);
  981. struct ceph_dentry_info *di;
  982. int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_UNLINK;
  983. spin_lock(&ci->i_ceph_lock);
  984. if ((__ceph_caps_issued(ci, NULL) & want) == want) {
  985. ceph_take_cap_refs(ci, want, false);
  986. got = want;
  987. }
  988. spin_unlock(&ci->i_ceph_lock);
  989. /* If we didn't get anything, return 0 */
  990. if (!got)
  991. return 0;
  992. spin_lock(&dentry->d_lock);
  993. di = ceph_dentry(dentry);
  994. /*
  995. * - We are holding Fx, which implies Fs caps.
  996. * - Only support async unlink for primary linkage
  997. */
  998. if (atomic_read(&ci->i_shared_gen) != di->lease_shared_gen ||
  999. !(di->flags & CEPH_DENTRY_PRIMARY_LINK))
  1000. want = 0;
  1001. spin_unlock(&dentry->d_lock);
  1002. /* Do we still want what we've got? */
  1003. if (want == got)
  1004. return got;
  1005. ceph_put_cap_refs(ci, got);
  1006. return 0;
  1007. }
  1008. /*
  1009. * rmdir and unlink are differ only by the metadata op code
  1010. */
  1011. static int ceph_unlink(struct inode *dir, struct dentry *dentry)
  1012. {
  1013. struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
  1014. struct ceph_mds_client *mdsc = fsc->mdsc;
  1015. struct inode *inode = d_inode(dentry);
  1016. struct ceph_mds_request *req;
  1017. bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
  1018. int err = -EROFS;
  1019. int op;
  1020. if (ceph_snap(dir) == CEPH_SNAPDIR) {
  1021. /* rmdir .snap/foo is RMSNAP */
  1022. dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
  1023. op = CEPH_MDS_OP_RMSNAP;
  1024. } else if (ceph_snap(dir) == CEPH_NOSNAP) {
  1025. dout("unlink/rmdir dir %p dn %p inode %p\n",
  1026. dir, dentry, inode);
  1027. op = d_is_dir(dentry) ?
  1028. CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
  1029. } else
  1030. goto out;
  1031. retry:
  1032. req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
  1033. if (IS_ERR(req)) {
  1034. err = PTR_ERR(req);
  1035. goto out;
  1036. }
  1037. req->r_dentry = dget(dentry);
  1038. req->r_num_caps = 2;
  1039. req->r_parent = dir;
  1040. req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
  1041. req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
  1042. req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
  1043. if (try_async && op == CEPH_MDS_OP_UNLINK &&
  1044. (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) {
  1045. dout("async unlink on %llu/%.*s caps=%s", ceph_ino(dir),
  1046. dentry->d_name.len, dentry->d_name.name,
  1047. ceph_cap_string(req->r_dir_caps));
  1048. set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
  1049. req->r_callback = ceph_async_unlink_cb;
  1050. req->r_old_inode = d_inode(dentry);
  1051. ihold(req->r_old_inode);
  1052. err = ceph_mdsc_submit_request(mdsc, dir, req);
  1053. if (!err) {
  1054. /*
  1055. * We have enough caps, so we assume that the unlink
  1056. * will succeed. Fix up the target inode and dcache.
  1057. */
  1058. drop_nlink(inode);
  1059. d_delete(dentry);
  1060. } else if (err == -EJUKEBOX) {
  1061. try_async = false;
  1062. ceph_mdsc_put_request(req);
  1063. goto retry;
  1064. }
  1065. } else {
  1066. set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
  1067. err = ceph_mdsc_do_request(mdsc, dir, req);
  1068. if (!err && !req->r_reply_info.head->is_dentry)
  1069. d_delete(dentry);
  1070. }
  1071. ceph_mdsc_put_request(req);
  1072. out:
  1073. return err;
  1074. }
  1075. static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
  1076. struct inode *new_dir, struct dentry *new_dentry,
  1077. unsigned int flags)
  1078. {
  1079. struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old_dir->i_sb);
  1080. struct ceph_mds_request *req;
  1081. int op = CEPH_MDS_OP_RENAME;
  1082. int err;
  1083. if (flags)
  1084. return -EINVAL;
  1085. if (ceph_snap(old_dir) != ceph_snap(new_dir))
  1086. return -EXDEV;
  1087. if (ceph_snap(old_dir) != CEPH_NOSNAP) {
  1088. if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
  1089. op = CEPH_MDS_OP_RENAMESNAP;
  1090. else
  1091. return -EROFS;
  1092. } else if (old_dir != new_dir) {
  1093. err = ceph_quota_check_rename(mdsc, d_inode(old_dentry),
  1094. new_dir);
  1095. if (err)
  1096. return err;
  1097. }
  1098. dout("rename dir %p dentry %p to dir %p dentry %p\n",
  1099. old_dir, old_dentry, new_dir, new_dentry);
  1100. req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
  1101. if (IS_ERR(req))
  1102. return PTR_ERR(req);
  1103. ihold(old_dir);
  1104. req->r_dentry = dget(new_dentry);
  1105. req->r_num_caps = 2;
  1106. req->r_old_dentry = dget(old_dentry);
  1107. req->r_old_dentry_dir = old_dir;
  1108. req->r_parent = new_dir;
  1109. set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
  1110. req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
  1111. req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
  1112. req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
  1113. req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
  1114. /* release LINK_RDCACHE on source inode (mds will lock it) */
  1115. req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
  1116. if (d_really_is_positive(new_dentry)) {
  1117. req->r_inode_drop =
  1118. ceph_drop_caps_for_unlink(d_inode(new_dentry));
  1119. }
  1120. err = ceph_mdsc_do_request(mdsc, old_dir, req);
  1121. if (!err && !req->r_reply_info.head->is_dentry) {
  1122. /*
  1123. * Normally d_move() is done by fill_trace (called by
  1124. * do_request, above). If there is no trace, we need
  1125. * to do it here.
  1126. */
  1127. d_move(old_dentry, new_dentry);
  1128. }
  1129. ceph_mdsc_put_request(req);
  1130. return err;
  1131. }
  1132. /*
  1133. * Move dentry to tail of mdsc->dentry_leases list when lease is updated.
  1134. * Leases at front of the list will expire first. (Assume all leases have
  1135. * similar duration)
  1136. *
  1137. * Called under dentry->d_lock.
  1138. */
  1139. void __ceph_dentry_lease_touch(struct ceph_dentry_info *di)
  1140. {
  1141. struct dentry *dn = di->dentry;
  1142. struct ceph_mds_client *mdsc;
  1143. dout("dentry_lease_touch %p %p '%pd'\n", di, dn, dn);
  1144. di->flags |= CEPH_DENTRY_LEASE_LIST;
  1145. if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
  1146. di->flags |= CEPH_DENTRY_REFERENCED;
  1147. return;
  1148. }
  1149. mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
  1150. spin_lock(&mdsc->dentry_list_lock);
  1151. list_move_tail(&di->lease_list, &mdsc->dentry_leases);
  1152. spin_unlock(&mdsc->dentry_list_lock);
  1153. }
  1154. static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc,
  1155. struct ceph_dentry_info *di)
  1156. {
  1157. di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED);
  1158. di->lease_gen = 0;
  1159. di->time = jiffies;
  1160. list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases);
  1161. }
  1162. /*
  1163. * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases
  1164. * list if it's not in the list, otherwise set 'referenced' flag.
  1165. *
  1166. * Called under dentry->d_lock.
  1167. */
  1168. void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di)
  1169. {
  1170. struct dentry *dn = di->dentry;
  1171. struct ceph_mds_client *mdsc;
  1172. dout("dentry_dir_lease_touch %p %p '%pd' (offset 0x%llx)\n",
  1173. di, dn, dn, di->offset);
  1174. if (!list_empty(&di->lease_list)) {
  1175. if (di->flags & CEPH_DENTRY_LEASE_LIST) {
  1176. /* don't remove dentry from dentry lease list
  1177. * if its lease is valid */
  1178. if (__dentry_lease_is_valid(di))
  1179. return;
  1180. } else {
  1181. di->flags |= CEPH_DENTRY_REFERENCED;
  1182. return;
  1183. }
  1184. }
  1185. if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
  1186. di->flags |= CEPH_DENTRY_REFERENCED;
  1187. di->flags &= ~CEPH_DENTRY_LEASE_LIST;
  1188. return;
  1189. }
  1190. mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
  1191. spin_lock(&mdsc->dentry_list_lock);
  1192. __dentry_dir_lease_touch(mdsc, di),
  1193. spin_unlock(&mdsc->dentry_list_lock);
  1194. }
  1195. static void __dentry_lease_unlist(struct ceph_dentry_info *di)
  1196. {
  1197. struct ceph_mds_client *mdsc;
  1198. if (di->flags & CEPH_DENTRY_SHRINK_LIST)
  1199. return;
  1200. if (list_empty(&di->lease_list))
  1201. return;
  1202. mdsc = ceph_sb_to_client(di->dentry->d_sb)->mdsc;
  1203. spin_lock(&mdsc->dentry_list_lock);
  1204. list_del_init(&di->lease_list);
  1205. spin_unlock(&mdsc->dentry_list_lock);
  1206. }
  1207. enum {
  1208. KEEP = 0,
  1209. DELETE = 1,
  1210. TOUCH = 2,
  1211. STOP = 4,
  1212. };
  1213. struct ceph_lease_walk_control {
  1214. bool dir_lease;
  1215. bool expire_dir_lease;
  1216. unsigned long nr_to_scan;
  1217. unsigned long dir_lease_ttl;
  1218. };
  1219. static unsigned long
  1220. __dentry_leases_walk(struct ceph_mds_client *mdsc,
  1221. struct ceph_lease_walk_control *lwc,
  1222. int (*check)(struct dentry*, void*))
  1223. {
  1224. struct ceph_dentry_info *di, *tmp;
  1225. struct dentry *dentry, *last = NULL;
  1226. struct list_head* list;
  1227. LIST_HEAD(dispose);
  1228. unsigned long freed = 0;
  1229. int ret = 0;
  1230. list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases;
  1231. spin_lock(&mdsc->dentry_list_lock);
  1232. list_for_each_entry_safe(di, tmp, list, lease_list) {
  1233. if (!lwc->nr_to_scan)
  1234. break;
  1235. --lwc->nr_to_scan;
  1236. dentry = di->dentry;
  1237. if (last == dentry)
  1238. break;
  1239. if (!spin_trylock(&dentry->d_lock))
  1240. continue;
  1241. if (__lockref_is_dead(&dentry->d_lockref)) {
  1242. list_del_init(&di->lease_list);
  1243. goto next;
  1244. }
  1245. ret = check(dentry, lwc);
  1246. if (ret & TOUCH) {
  1247. /* move it into tail of dir lease list */
  1248. __dentry_dir_lease_touch(mdsc, di);
  1249. if (!last)
  1250. last = dentry;
  1251. }
  1252. if (ret & DELETE) {
  1253. /* stale lease */
  1254. di->flags &= ~CEPH_DENTRY_REFERENCED;
  1255. if (dentry->d_lockref.count > 0) {
  1256. /* update_dentry_lease() will re-add
  1257. * it to lease list, or
  1258. * ceph_d_delete() will return 1 when
  1259. * last reference is dropped */
  1260. list_del_init(&di->lease_list);
  1261. } else {
  1262. di->flags |= CEPH_DENTRY_SHRINK_LIST;
  1263. list_move_tail(&di->lease_list, &dispose);
  1264. dget_dlock(dentry);
  1265. }
  1266. }
  1267. next:
  1268. spin_unlock(&dentry->d_lock);
  1269. if (ret & STOP)
  1270. break;
  1271. }
  1272. spin_unlock(&mdsc->dentry_list_lock);
  1273. while (!list_empty(&dispose)) {
  1274. di = list_first_entry(&dispose, struct ceph_dentry_info,
  1275. lease_list);
  1276. dentry = di->dentry;
  1277. spin_lock(&dentry->d_lock);
  1278. list_del_init(&di->lease_list);
  1279. di->flags &= ~CEPH_DENTRY_SHRINK_LIST;
  1280. if (di->flags & CEPH_DENTRY_REFERENCED) {
  1281. spin_lock(&mdsc->dentry_list_lock);
  1282. if (di->flags & CEPH_DENTRY_LEASE_LIST) {
  1283. list_add_tail(&di->lease_list,
  1284. &mdsc->dentry_leases);
  1285. } else {
  1286. __dentry_dir_lease_touch(mdsc, di);
  1287. }
  1288. spin_unlock(&mdsc->dentry_list_lock);
  1289. } else {
  1290. freed++;
  1291. }
  1292. spin_unlock(&dentry->d_lock);
  1293. /* ceph_d_delete() does the trick */
  1294. dput(dentry);
  1295. }
  1296. return freed;
  1297. }
  1298. static int __dentry_lease_check(struct dentry *dentry, void *arg)
  1299. {
  1300. struct ceph_dentry_info *di = ceph_dentry(dentry);
  1301. int ret;
  1302. if (__dentry_lease_is_valid(di))
  1303. return STOP;
  1304. ret = __dir_lease_try_check(dentry);
  1305. if (ret == -EBUSY)
  1306. return KEEP;
  1307. if (ret > 0)
  1308. return TOUCH;
  1309. return DELETE;
  1310. }
  1311. static int __dir_lease_check(struct dentry *dentry, void *arg)
  1312. {
  1313. struct ceph_lease_walk_control *lwc = arg;
  1314. struct ceph_dentry_info *di = ceph_dentry(dentry);
  1315. int ret = __dir_lease_try_check(dentry);
  1316. if (ret == -EBUSY)
  1317. return KEEP;
  1318. if (ret > 0) {
  1319. if (time_before(jiffies, di->time + lwc->dir_lease_ttl))
  1320. return STOP;
  1321. /* Move dentry to tail of dir lease list if we don't want
  1322. * to delete it. So dentries in the list are checked in a
  1323. * round robin manner */
  1324. if (!lwc->expire_dir_lease)
  1325. return TOUCH;
  1326. if (dentry->d_lockref.count > 0 ||
  1327. (di->flags & CEPH_DENTRY_REFERENCED))
  1328. return TOUCH;
  1329. /* invalidate dir lease */
  1330. di->lease_shared_gen = 0;
  1331. }
  1332. return DELETE;
  1333. }
  1334. int ceph_trim_dentries(struct ceph_mds_client *mdsc)
  1335. {
  1336. struct ceph_lease_walk_control lwc;
  1337. unsigned long count;
  1338. unsigned long freed;
  1339. spin_lock(&mdsc->caps_list_lock);
  1340. if (mdsc->caps_use_max > 0 &&
  1341. mdsc->caps_use_count > mdsc->caps_use_max)
  1342. count = mdsc->caps_use_count - mdsc->caps_use_max;
  1343. else
  1344. count = 0;
  1345. spin_unlock(&mdsc->caps_list_lock);
  1346. lwc.dir_lease = false;
  1347. lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE * 2;
  1348. freed = __dentry_leases_walk(mdsc, &lwc, __dentry_lease_check);
  1349. if (!lwc.nr_to_scan) /* more invalid leases */
  1350. return -EAGAIN;
  1351. if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE)
  1352. lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE;
  1353. lwc.dir_lease = true;
  1354. lwc.expire_dir_lease = freed < count;
  1355. lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ;
  1356. freed +=__dentry_leases_walk(mdsc, &lwc, __dir_lease_check);
  1357. if (!lwc.nr_to_scan) /* more to check */
  1358. return -EAGAIN;
  1359. return freed > 0 ? 1 : 0;
  1360. }
  1361. /*
  1362. * Ensure a dentry lease will no longer revalidate.
  1363. */
  1364. void ceph_invalidate_dentry_lease(struct dentry *dentry)
  1365. {
  1366. struct ceph_dentry_info *di = ceph_dentry(dentry);
  1367. spin_lock(&dentry->d_lock);
  1368. di->time = jiffies;
  1369. di->lease_shared_gen = 0;
  1370. di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
  1371. __dentry_lease_unlist(di);
  1372. spin_unlock(&dentry->d_lock);
  1373. }
  1374. /*
  1375. * Check if dentry lease is valid. If not, delete the lease. Try to
  1376. * renew if the least is more than half up.
  1377. */
  1378. static bool __dentry_lease_is_valid(struct ceph_dentry_info *di)
  1379. {
  1380. struct ceph_mds_session *session;
  1381. if (!di->lease_gen)
  1382. return false;
  1383. session = di->lease_session;
  1384. if (session) {
  1385. u32 gen;
  1386. unsigned long ttl;
  1387. spin_lock(&session->s_gen_ttl_lock);
  1388. gen = session->s_cap_gen;
  1389. ttl = session->s_cap_ttl;
  1390. spin_unlock(&session->s_gen_ttl_lock);
  1391. if (di->lease_gen == gen &&
  1392. time_before(jiffies, ttl) &&
  1393. time_before(jiffies, di->time))
  1394. return true;
  1395. }
  1396. di->lease_gen = 0;
  1397. return false;
  1398. }
  1399. static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags)
  1400. {
  1401. struct ceph_dentry_info *di;
  1402. struct ceph_mds_session *session = NULL;
  1403. u32 seq = 0;
  1404. int valid = 0;
  1405. spin_lock(&dentry->d_lock);
  1406. di = ceph_dentry(dentry);
  1407. if (di && __dentry_lease_is_valid(di)) {
  1408. valid = 1;
  1409. if (di->lease_renew_after &&
  1410. time_after(jiffies, di->lease_renew_after)) {
  1411. /*
  1412. * We should renew. If we're in RCU walk mode
  1413. * though, we can't do that so just return
  1414. * -ECHILD.
  1415. */
  1416. if (flags & LOOKUP_RCU) {
  1417. valid = -ECHILD;
  1418. } else {
  1419. session = ceph_get_mds_session(di->lease_session);
  1420. seq = di->lease_seq;
  1421. di->lease_renew_after = 0;
  1422. di->lease_renew_from = jiffies;
  1423. }
  1424. }
  1425. }
  1426. spin_unlock(&dentry->d_lock);
  1427. if (session) {
  1428. ceph_mdsc_lease_send_msg(session, dentry,
  1429. CEPH_MDS_LEASE_RENEW, seq);
  1430. ceph_put_mds_session(session);
  1431. }
  1432. dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
  1433. return valid;
  1434. }
  1435. /*
  1436. * Called under dentry->d_lock.
  1437. */
  1438. static int __dir_lease_try_check(const struct dentry *dentry)
  1439. {
  1440. struct ceph_dentry_info *di = ceph_dentry(dentry);
  1441. struct inode *dir;
  1442. struct ceph_inode_info *ci;
  1443. int valid = 0;
  1444. if (!di->lease_shared_gen)
  1445. return 0;
  1446. if (IS_ROOT(dentry))
  1447. return 0;
  1448. dir = d_inode(dentry->d_parent);
  1449. ci = ceph_inode(dir);
  1450. if (spin_trylock(&ci->i_ceph_lock)) {
  1451. if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen &&
  1452. __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0))
  1453. valid = 1;
  1454. spin_unlock(&ci->i_ceph_lock);
  1455. } else {
  1456. valid = -EBUSY;
  1457. }
  1458. if (!valid)
  1459. di->lease_shared_gen = 0;
  1460. return valid;
  1461. }
  1462. /*
  1463. * Check if directory-wide content lease/cap is valid.
  1464. */
  1465. static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry,
  1466. struct ceph_mds_client *mdsc)
  1467. {
  1468. struct ceph_inode_info *ci = ceph_inode(dir);
  1469. int valid;
  1470. int shared_gen;
  1471. spin_lock(&ci->i_ceph_lock);
  1472. valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
  1473. if (valid) {
  1474. __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
  1475. shared_gen = atomic_read(&ci->i_shared_gen);
  1476. }
  1477. spin_unlock(&ci->i_ceph_lock);
  1478. if (valid) {
  1479. struct ceph_dentry_info *di;
  1480. spin_lock(&dentry->d_lock);
  1481. di = ceph_dentry(dentry);
  1482. if (dir == d_inode(dentry->d_parent) &&
  1483. di && di->lease_shared_gen == shared_gen)
  1484. __ceph_dentry_dir_lease_touch(di);
  1485. else
  1486. valid = 0;
  1487. spin_unlock(&dentry->d_lock);
  1488. }
  1489. dout("dir_lease_is_valid dir %p v%u dentry %p = %d\n",
  1490. dir, (unsigned)atomic_read(&ci->i_shared_gen), dentry, valid);
  1491. return valid;
  1492. }
  1493. /*
  1494. * Check if cached dentry can be trusted.
  1495. */
  1496. static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
  1497. {
  1498. int valid = 0;
  1499. struct dentry *parent;
  1500. struct inode *dir, *inode;
  1501. struct ceph_mds_client *mdsc;
  1502. if (flags & LOOKUP_RCU) {
  1503. parent = READ_ONCE(dentry->d_parent);
  1504. dir = d_inode_rcu(parent);
  1505. if (!dir)
  1506. return -ECHILD;
  1507. inode = d_inode_rcu(dentry);
  1508. } else {
  1509. parent = dget_parent(dentry);
  1510. dir = d_inode(parent);
  1511. inode = d_inode(dentry);
  1512. }
  1513. dout("d_revalidate %p '%pd' inode %p offset 0x%llx\n", dentry,
  1514. dentry, inode, ceph_dentry(dentry)->offset);
  1515. mdsc = ceph_sb_to_client(dir->i_sb)->mdsc;
  1516. /* always trust cached snapped dentries, snapdir dentry */
  1517. if (ceph_snap(dir) != CEPH_NOSNAP) {
  1518. dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
  1519. dentry, inode);
  1520. valid = 1;
  1521. } else if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
  1522. valid = 1;
  1523. } else {
  1524. valid = dentry_lease_is_valid(dentry, flags);
  1525. if (valid == -ECHILD)
  1526. return valid;
  1527. if (valid || dir_lease_is_valid(dir, dentry, mdsc)) {
  1528. if (inode)
  1529. valid = ceph_is_any_caps(inode);
  1530. else
  1531. valid = 1;
  1532. }
  1533. }
  1534. if (!valid) {
  1535. struct ceph_mds_request *req;
  1536. int op, err;
  1537. u32 mask;
  1538. if (flags & LOOKUP_RCU)
  1539. return -ECHILD;
  1540. percpu_counter_inc(&mdsc->metric.d_lease_mis);
  1541. op = ceph_snap(dir) == CEPH_SNAPDIR ?
  1542. CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
  1543. req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
  1544. if (!IS_ERR(req)) {
  1545. req->r_dentry = dget(dentry);
  1546. req->r_num_caps = 2;
  1547. req->r_parent = dir;
  1548. mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
  1549. if (ceph_security_xattr_wanted(dir))
  1550. mask |= CEPH_CAP_XATTR_SHARED;
  1551. req->r_args.getattr.mask = cpu_to_le32(mask);
  1552. err = ceph_mdsc_do_request(mdsc, NULL, req);
  1553. switch (err) {
  1554. case 0:
  1555. if (d_really_is_positive(dentry) &&
  1556. d_inode(dentry) == req->r_target_inode)
  1557. valid = 1;
  1558. break;
  1559. case -ENOENT:
  1560. if (d_really_is_negative(dentry))
  1561. valid = 1;
  1562. fallthrough;
  1563. default:
  1564. break;
  1565. }
  1566. ceph_mdsc_put_request(req);
  1567. dout("d_revalidate %p lookup result=%d\n",
  1568. dentry, err);
  1569. }
  1570. } else {
  1571. percpu_counter_inc(&mdsc->metric.d_lease_hit);
  1572. }
  1573. dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
  1574. if (!valid)
  1575. ceph_dir_clear_complete(dir);
  1576. if (!(flags & LOOKUP_RCU))
  1577. dput(parent);
  1578. return valid;
  1579. }
  1580. /*
  1581. * Delete unused dentry that doesn't have valid lease
  1582. *
  1583. * Called under dentry->d_lock.
  1584. */
  1585. static int ceph_d_delete(const struct dentry *dentry)
  1586. {
  1587. struct ceph_dentry_info *di;
  1588. /* won't release caps */
  1589. if (d_really_is_negative(dentry))
  1590. return 0;
  1591. if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
  1592. return 0;
  1593. /* vaild lease? */
  1594. di = ceph_dentry(dentry);
  1595. if (di) {
  1596. if (__dentry_lease_is_valid(di))
  1597. return 0;
  1598. if (__dir_lease_try_check(dentry))
  1599. return 0;
  1600. }
  1601. return 1;
  1602. }
  1603. /*
  1604. * Release our ceph_dentry_info.
  1605. */
  1606. static void ceph_d_release(struct dentry *dentry)
  1607. {
  1608. struct ceph_dentry_info *di = ceph_dentry(dentry);
  1609. struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
  1610. dout("d_release %p\n", dentry);
  1611. atomic64_dec(&fsc->mdsc->metric.total_dentries);
  1612. spin_lock(&dentry->d_lock);
  1613. __dentry_lease_unlist(di);
  1614. dentry->d_fsdata = NULL;
  1615. spin_unlock(&dentry->d_lock);
  1616. ceph_put_mds_session(di->lease_session);
  1617. kmem_cache_free(ceph_dentry_cachep, di);
  1618. }
  1619. /*
  1620. * When the VFS prunes a dentry from the cache, we need to clear the
  1621. * complete flag on the parent directory.
  1622. *
  1623. * Called under dentry->d_lock.
  1624. */
  1625. static void ceph_d_prune(struct dentry *dentry)
  1626. {
  1627. struct ceph_inode_info *dir_ci;
  1628. struct ceph_dentry_info *di;
  1629. dout("ceph_d_prune %pd %p\n", dentry, dentry);
  1630. /* do we have a valid parent? */
  1631. if (IS_ROOT(dentry))
  1632. return;
  1633. /* we hold d_lock, so d_parent is stable */
  1634. dir_ci = ceph_inode(d_inode(dentry->d_parent));
  1635. if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
  1636. return;
  1637. /* who calls d_delete() should also disable dcache readdir */
  1638. if (d_really_is_negative(dentry))
  1639. return;
  1640. /* d_fsdata does not get cleared until d_release */
  1641. if (!d_unhashed(dentry)) {
  1642. __ceph_dir_clear_complete(dir_ci);
  1643. return;
  1644. }
  1645. /* Disable dcache readdir just in case that someone called d_drop()
  1646. * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
  1647. * properly (dcache readdir is still enabled) */
  1648. di = ceph_dentry(dentry);
  1649. if (di->offset > 0 &&
  1650. di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
  1651. __ceph_dir_clear_ordered(dir_ci);
  1652. }
  1653. /*
  1654. * read() on a dir. This weird interface hack only works if mounted
  1655. * with '-o dirstat'.
  1656. */
  1657. static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
  1658. loff_t *ppos)
  1659. {
  1660. struct ceph_dir_file_info *dfi = file->private_data;
  1661. struct inode *inode = file_inode(file);
  1662. struct ceph_inode_info *ci = ceph_inode(inode);
  1663. int left;
  1664. const int bufsize = 1024;
  1665. if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
  1666. return -EISDIR;
  1667. if (!dfi->dir_info) {
  1668. dfi->dir_info = kmalloc(bufsize, GFP_KERNEL);
  1669. if (!dfi->dir_info)
  1670. return -ENOMEM;
  1671. dfi->dir_info_len =
  1672. snprintf(dfi->dir_info, bufsize,
  1673. "entries: %20lld\n"
  1674. " files: %20lld\n"
  1675. " subdirs: %20lld\n"
  1676. "rentries: %20lld\n"
  1677. " rfiles: %20lld\n"
  1678. " rsubdirs: %20lld\n"
  1679. "rbytes: %20lld\n"
  1680. "rctime: %10lld.%09ld\n",
  1681. ci->i_files + ci->i_subdirs,
  1682. ci->i_files,
  1683. ci->i_subdirs,
  1684. ci->i_rfiles + ci->i_rsubdirs,
  1685. ci->i_rfiles,
  1686. ci->i_rsubdirs,
  1687. ci->i_rbytes,
  1688. ci->i_rctime.tv_sec,
  1689. ci->i_rctime.tv_nsec);
  1690. }
  1691. if (*ppos >= dfi->dir_info_len)
  1692. return 0;
  1693. size = min_t(unsigned, size, dfi->dir_info_len-*ppos);
  1694. left = copy_to_user(buf, dfi->dir_info + *ppos, size);
  1695. if (left == size)
  1696. return -EFAULT;
  1697. *ppos += (size - left);
  1698. return size - left;
  1699. }
  1700. /*
  1701. * Return name hash for a given dentry. This is dependent on
  1702. * the parent directory's hash function.
  1703. */
  1704. unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
  1705. {
  1706. struct ceph_inode_info *dci = ceph_inode(dir);
  1707. unsigned hash;
  1708. switch (dci->i_dir_layout.dl_dir_hash) {
  1709. case 0: /* for backward compat */
  1710. case CEPH_STR_HASH_LINUX:
  1711. return dn->d_name.hash;
  1712. default:
  1713. spin_lock(&dn->d_lock);
  1714. hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
  1715. dn->d_name.name, dn->d_name.len);
  1716. spin_unlock(&dn->d_lock);
  1717. return hash;
  1718. }
  1719. }
  1720. const struct file_operations ceph_dir_fops = {
  1721. .read = ceph_read_dir,
  1722. .iterate = ceph_readdir,
  1723. .llseek = ceph_dir_llseek,
  1724. .open = ceph_open,
  1725. .release = ceph_release,
  1726. .unlocked_ioctl = ceph_ioctl,
  1727. .compat_ioctl = compat_ptr_ioctl,
  1728. .fsync = ceph_fsync,
  1729. .lock = ceph_lock,
  1730. .flock = ceph_flock,
  1731. };
  1732. const struct file_operations ceph_snapdir_fops = {
  1733. .iterate = ceph_readdir,
  1734. .llseek = ceph_dir_llseek,
  1735. .open = ceph_open,
  1736. .release = ceph_release,
  1737. };
  1738. const struct inode_operations ceph_dir_iops = {
  1739. .lookup = ceph_lookup,
  1740. .permission = ceph_permission,
  1741. .getattr = ceph_getattr,
  1742. .setattr = ceph_setattr,
  1743. .listxattr = ceph_listxattr,
  1744. .get_acl = ceph_get_acl,
  1745. .set_acl = ceph_set_acl,
  1746. .mknod = ceph_mknod,
  1747. .symlink = ceph_symlink,
  1748. .mkdir = ceph_mkdir,
  1749. .link = ceph_link,
  1750. .unlink = ceph_unlink,
  1751. .rmdir = ceph_unlink,
  1752. .rename = ceph_rename,
  1753. .create = ceph_create,
  1754. .atomic_open = ceph_atomic_open,
  1755. };
  1756. const struct inode_operations ceph_snapdir_iops = {
  1757. .lookup = ceph_lookup,
  1758. .permission = ceph_permission,
  1759. .getattr = ceph_getattr,
  1760. .mkdir = ceph_mkdir,
  1761. .rmdir = ceph_unlink,
  1762. .rename = ceph_rename,
  1763. };
  1764. const struct dentry_operations ceph_dentry_ops = {
  1765. .d_revalidate = ceph_d_revalidate,
  1766. .d_delete = ceph_d_delete,
  1767. .d_release = ceph_d_release,
  1768. .d_prune = ceph_d_prune,
  1769. .d_init = ceph_d_init,
  1770. };