Commit | Line | Data |
---|---|---|
9cf514cc CH |
1 | /* |
2 | * Copyright (c) 2014 Christoph Hellwig. | |
3 | */ | |
c5c707f9 CH |
4 | #include <linux/kmod.h> |
5 | #include <linux/file.h> | |
9cf514cc CH |
6 | #include <linux/jhash.h> |
7 | #include <linux/sched.h> | |
c5c707f9 | 8 | #include <linux/sunrpc/addr.h> |
9cf514cc CH |
9 | |
10 | #include "pnfs.h" | |
11 | #include "netns.h" | |
12 | ||
13 | #define NFSDDBG_FACILITY NFSDDBG_PNFS | |
14 | ||
15 | struct nfs4_layout { | |
16 | struct list_head lo_perstate; | |
17 | struct nfs4_layout_stateid *lo_state; | |
18 | struct nfsd4_layout_seg lo_seg; | |
19 | }; | |
20 | ||
21 | static struct kmem_cache *nfs4_layout_cache; | |
22 | static struct kmem_cache *nfs4_layout_stateid_cache; | |
23 | ||
c5c707f9 CH |
24 | static struct nfsd4_callback_ops nfsd4_cb_layout_ops; |
25 | static const struct lock_manager_operations nfsd4_layouts_lm_ops; | |
26 | ||
9cf514cc CH |
27 | const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] = { |
28 | }; | |
29 | ||
30 | /* pNFS device ID to export fsid mapping */ | |
31 | #define DEVID_HASH_BITS 8 | |
32 | #define DEVID_HASH_SIZE (1 << DEVID_HASH_BITS) | |
33 | #define DEVID_HASH_MASK (DEVID_HASH_SIZE - 1) | |
34 | static u64 nfsd_devid_seq = 1; | |
35 | static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE]; | |
36 | static DEFINE_SPINLOCK(nfsd_devid_lock); | |
37 | ||
38 | static inline u32 devid_hashfn(u64 idx) | |
39 | { | |
40 | return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK; | |
41 | } | |
42 | ||
43 | static void | |
44 | nfsd4_alloc_devid_map(const struct svc_fh *fhp) | |
45 | { | |
46 | const struct knfsd_fh *fh = &fhp->fh_handle; | |
47 | size_t fsid_len = key_len(fh->fh_fsid_type); | |
48 | struct nfsd4_deviceid_map *map, *old; | |
49 | int i; | |
50 | ||
51 | map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL); | |
52 | if (!map) | |
53 | return; | |
54 | ||
55 | map->fsid_type = fh->fh_fsid_type; | |
56 | memcpy(&map->fsid, fh->fh_fsid, fsid_len); | |
57 | ||
58 | spin_lock(&nfsd_devid_lock); | |
59 | if (fhp->fh_export->ex_devid_map) | |
60 | goto out_unlock; | |
61 | ||
62 | for (i = 0; i < DEVID_HASH_SIZE; i++) { | |
63 | list_for_each_entry(old, &nfsd_devid_hash[i], hash) { | |
64 | if (old->fsid_type != fh->fh_fsid_type) | |
65 | continue; | |
66 | if (memcmp(old->fsid, fh->fh_fsid, | |
67 | key_len(old->fsid_type))) | |
68 | continue; | |
69 | ||
70 | fhp->fh_export->ex_devid_map = old; | |
71 | goto out_unlock; | |
72 | } | |
73 | } | |
74 | ||
75 | map->idx = nfsd_devid_seq++; | |
76 | list_add_tail_rcu(&map->hash, &nfsd_devid_hash[devid_hashfn(map->idx)]); | |
77 | fhp->fh_export->ex_devid_map = map; | |
78 | map = NULL; | |
79 | ||
80 | out_unlock: | |
81 | spin_unlock(&nfsd_devid_lock); | |
82 | kfree(map); | |
83 | } | |
84 | ||
85 | struct nfsd4_deviceid_map * | |
86 | nfsd4_find_devid_map(int idx) | |
87 | { | |
88 | struct nfsd4_deviceid_map *map, *ret = NULL; | |
89 | ||
90 | rcu_read_lock(); | |
91 | list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash) | |
92 | if (map->idx == idx) | |
93 | ret = map; | |
94 | rcu_read_unlock(); | |
95 | ||
96 | return ret; | |
97 | } | |
98 | ||
99 | int | |
100 | nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp, | |
101 | u32 device_generation) | |
102 | { | |
103 | if (!fhp->fh_export->ex_devid_map) { | |
104 | nfsd4_alloc_devid_map(fhp); | |
105 | if (!fhp->fh_export->ex_devid_map) | |
106 | return -ENOMEM; | |
107 | } | |
108 | ||
109 | id->fsid_idx = fhp->fh_export->ex_devid_map->idx; | |
110 | id->generation = device_generation; | |
111 | id->pad = 0; | |
112 | return 0; | |
113 | } | |
114 | ||
115 | void nfsd4_setup_layout_type(struct svc_export *exp) | |
116 | { | |
117 | if (exp->ex_flags & NFSEXP_NOPNFS) | |
118 | return; | |
119 | } | |
120 | ||
121 | static void | |
122 | nfsd4_free_layout_stateid(struct nfs4_stid *stid) | |
123 | { | |
124 | struct nfs4_layout_stateid *ls = layoutstateid(stid); | |
125 | struct nfs4_client *clp = ls->ls_stid.sc_client; | |
126 | struct nfs4_file *fp = ls->ls_stid.sc_file; | |
127 | ||
128 | spin_lock(&clp->cl_lock); | |
129 | list_del_init(&ls->ls_perclnt); | |
130 | spin_unlock(&clp->cl_lock); | |
131 | ||
132 | spin_lock(&fp->fi_lock); | |
133 | list_del_init(&ls->ls_perfile); | |
134 | spin_unlock(&fp->fi_lock); | |
135 | ||
c5c707f9 CH |
136 | vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls); |
137 | fput(ls->ls_file); | |
138 | ||
139 | if (ls->ls_recalled) | |
140 | atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls); | |
141 | ||
9cf514cc CH |
142 | kmem_cache_free(nfs4_layout_stateid_cache, ls); |
143 | } | |
144 | ||
c5c707f9 CH |
145 | static int |
146 | nfsd4_layout_setlease(struct nfs4_layout_stateid *ls) | |
147 | { | |
148 | struct file_lock *fl; | |
149 | int status; | |
150 | ||
151 | fl = locks_alloc_lock(); | |
152 | if (!fl) | |
153 | return -ENOMEM; | |
154 | locks_init_lock(fl); | |
155 | fl->fl_lmops = &nfsd4_layouts_lm_ops; | |
156 | fl->fl_flags = FL_LAYOUT; | |
157 | fl->fl_type = F_RDLCK; | |
158 | fl->fl_end = OFFSET_MAX; | |
159 | fl->fl_owner = ls; | |
160 | fl->fl_pid = current->tgid; | |
161 | fl->fl_file = ls->ls_file; | |
162 | ||
163 | status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL); | |
164 | if (status) { | |
165 | locks_free_lock(fl); | |
166 | return status; | |
167 | } | |
168 | BUG_ON(fl != NULL); | |
169 | return 0; | |
170 | } | |
171 | ||
9cf514cc CH |
172 | static struct nfs4_layout_stateid * |
173 | nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate, | |
174 | struct nfs4_stid *parent, u32 layout_type) | |
175 | { | |
176 | struct nfs4_client *clp = cstate->clp; | |
177 | struct nfs4_file *fp = parent->sc_file; | |
178 | struct nfs4_layout_stateid *ls; | |
179 | struct nfs4_stid *stp; | |
180 | ||
181 | stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache); | |
182 | if (!stp) | |
183 | return NULL; | |
184 | stp->sc_free = nfsd4_free_layout_stateid; | |
185 | get_nfs4_file(fp); | |
186 | stp->sc_file = fp; | |
187 | ||
188 | ls = layoutstateid(stp); | |
189 | INIT_LIST_HEAD(&ls->ls_perclnt); | |
190 | INIT_LIST_HEAD(&ls->ls_perfile); | |
191 | spin_lock_init(&ls->ls_lock); | |
192 | INIT_LIST_HEAD(&ls->ls_layouts); | |
193 | ls->ls_layout_type = layout_type; | |
c5c707f9 CH |
194 | nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops, |
195 | NFSPROC4_CLNT_CB_LAYOUT); | |
196 | ||
197 | if (parent->sc_type == NFS4_DELEG_STID) | |
198 | ls->ls_file = get_file(fp->fi_deleg_file); | |
199 | else | |
200 | ls->ls_file = find_any_file(fp); | |
201 | BUG_ON(!ls->ls_file); | |
202 | ||
203 | if (nfsd4_layout_setlease(ls)) { | |
204 | put_nfs4_file(fp); | |
205 | kmem_cache_free(nfs4_layout_stateid_cache, ls); | |
206 | return NULL; | |
207 | } | |
9cf514cc CH |
208 | |
209 | spin_lock(&clp->cl_lock); | |
210 | stp->sc_type = NFS4_LAYOUT_STID; | |
211 | list_add(&ls->ls_perclnt, &clp->cl_lo_states); | |
212 | spin_unlock(&clp->cl_lock); | |
213 | ||
214 | spin_lock(&fp->fi_lock); | |
215 | list_add(&ls->ls_perfile, &fp->fi_lo_states); | |
216 | spin_unlock(&fp->fi_lock); | |
217 | ||
218 | return ls; | |
219 | } | |
220 | ||
221 | __be32 | |
222 | nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp, | |
223 | struct nfsd4_compound_state *cstate, stateid_t *stateid, | |
224 | bool create, u32 layout_type, struct nfs4_layout_stateid **lsp) | |
225 | { | |
226 | struct nfs4_layout_stateid *ls; | |
227 | struct nfs4_stid *stid; | |
228 | unsigned char typemask = NFS4_LAYOUT_STID; | |
229 | __be32 status; | |
230 | ||
231 | if (create) | |
232 | typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID); | |
233 | ||
234 | status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid, | |
235 | net_generic(SVC_NET(rqstp), nfsd_net_id)); | |
236 | if (status) | |
237 | goto out; | |
238 | ||
239 | if (!fh_match(&cstate->current_fh.fh_handle, | |
240 | &stid->sc_file->fi_fhandle)) { | |
241 | status = nfserr_bad_stateid; | |
242 | goto out_put_stid; | |
243 | } | |
244 | ||
245 | if (stid->sc_type != NFS4_LAYOUT_STID) { | |
246 | ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type); | |
247 | nfs4_put_stid(stid); | |
248 | ||
249 | status = nfserr_jukebox; | |
250 | if (!ls) | |
251 | goto out; | |
252 | } else { | |
253 | ls = container_of(stid, struct nfs4_layout_stateid, ls_stid); | |
254 | ||
255 | status = nfserr_bad_stateid; | |
256 | if (stateid->si_generation > stid->sc_stateid.si_generation) | |
257 | goto out_put_stid; | |
258 | if (layout_type != ls->ls_layout_type) | |
259 | goto out_put_stid; | |
260 | } | |
261 | ||
262 | *lsp = ls; | |
263 | return 0; | |
264 | ||
265 | out_put_stid: | |
266 | nfs4_put_stid(stid); | |
267 | out: | |
268 | return status; | |
269 | } | |
270 | ||
c5c707f9 CH |
271 | static void |
272 | nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls) | |
273 | { | |
274 | spin_lock(&ls->ls_lock); | |
275 | if (ls->ls_recalled) | |
276 | goto out_unlock; | |
277 | ||
278 | ls->ls_recalled = true; | |
279 | atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls); | |
280 | if (list_empty(&ls->ls_layouts)) | |
281 | goto out_unlock; | |
282 | ||
283 | atomic_inc(&ls->ls_stid.sc_count); | |
284 | update_stateid(&ls->ls_stid.sc_stateid); | |
285 | memcpy(&ls->ls_recall_sid, &ls->ls_stid.sc_stateid, sizeof(stateid_t)); | |
286 | nfsd4_run_cb(&ls->ls_recall); | |
287 | ||
288 | out_unlock: | |
289 | spin_unlock(&ls->ls_lock); | |
290 | } | |
291 | ||
9cf514cc CH |
292 | static inline u64 |
293 | layout_end(struct nfsd4_layout_seg *seg) | |
294 | { | |
295 | u64 end = seg->offset + seg->length; | |
296 | return end >= seg->offset ? end : NFS4_MAX_UINT64; | |
297 | } | |
298 | ||
299 | static void | |
300 | layout_update_len(struct nfsd4_layout_seg *lo, u64 end) | |
301 | { | |
302 | if (end == NFS4_MAX_UINT64) | |
303 | lo->length = NFS4_MAX_UINT64; | |
304 | else | |
305 | lo->length = end - lo->offset; | |
306 | } | |
307 | ||
308 | static bool | |
309 | layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s) | |
310 | { | |
311 | if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode) | |
312 | return false; | |
313 | if (layout_end(&lo->lo_seg) <= s->offset) | |
314 | return false; | |
315 | if (layout_end(s) <= lo->lo_seg.offset) | |
316 | return false; | |
317 | return true; | |
318 | } | |
319 | ||
320 | static bool | |
321 | layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new) | |
322 | { | |
323 | if (lo->iomode != new->iomode) | |
324 | return false; | |
325 | if (layout_end(new) < lo->offset) | |
326 | return false; | |
327 | if (layout_end(lo) < new->offset) | |
328 | return false; | |
329 | ||
330 | lo->offset = min(lo->offset, new->offset); | |
331 | layout_update_len(lo, max(layout_end(lo), layout_end(new))); | |
332 | return true; | |
333 | } | |
334 | ||
c5c707f9 CH |
335 | static __be32 |
336 | nfsd4_recall_conflict(struct nfs4_layout_stateid *ls) | |
337 | { | |
338 | struct nfs4_file *fp = ls->ls_stid.sc_file; | |
339 | struct nfs4_layout_stateid *l, *n; | |
340 | __be32 nfserr = nfs_ok; | |
341 | ||
342 | assert_spin_locked(&fp->fi_lock); | |
343 | ||
344 | list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) { | |
345 | if (l != ls) { | |
346 | nfsd4_recall_file_layout(l); | |
347 | nfserr = nfserr_recallconflict; | |
348 | } | |
349 | } | |
350 | ||
351 | return nfserr; | |
352 | } | |
353 | ||
9cf514cc CH |
354 | __be32 |
355 | nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls) | |
356 | { | |
357 | struct nfsd4_layout_seg *seg = &lgp->lg_seg; | |
c5c707f9 | 358 | struct nfs4_file *fp = ls->ls_stid.sc_file; |
9cf514cc | 359 | struct nfs4_layout *lp, *new = NULL; |
c5c707f9 | 360 | __be32 nfserr; |
9cf514cc | 361 | |
c5c707f9 CH |
362 | spin_lock(&fp->fi_lock); |
363 | nfserr = nfsd4_recall_conflict(ls); | |
364 | if (nfserr) | |
365 | goto out; | |
9cf514cc CH |
366 | spin_lock(&ls->ls_lock); |
367 | list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) { | |
368 | if (layouts_try_merge(&lp->lo_seg, seg)) | |
369 | goto done; | |
370 | } | |
371 | spin_unlock(&ls->ls_lock); | |
c5c707f9 | 372 | spin_unlock(&fp->fi_lock); |
9cf514cc CH |
373 | |
374 | new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL); | |
375 | if (!new) | |
376 | return nfserr_jukebox; | |
377 | memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg)); | |
378 | new->lo_state = ls; | |
379 | ||
c5c707f9 CH |
380 | spin_lock(&fp->fi_lock); |
381 | nfserr = nfsd4_recall_conflict(ls); | |
382 | if (nfserr) | |
383 | goto out; | |
9cf514cc CH |
384 | spin_lock(&ls->ls_lock); |
385 | list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) { | |
386 | if (layouts_try_merge(&lp->lo_seg, seg)) | |
387 | goto done; | |
388 | } | |
389 | ||
390 | atomic_inc(&ls->ls_stid.sc_count); | |
391 | list_add_tail(&new->lo_perstate, &ls->ls_layouts); | |
392 | new = NULL; | |
393 | done: | |
394 | update_stateid(&ls->ls_stid.sc_stateid); | |
395 | memcpy(&lgp->lg_sid, &ls->ls_stid.sc_stateid, sizeof(stateid_t)); | |
396 | spin_unlock(&ls->ls_lock); | |
c5c707f9 CH |
397 | out: |
398 | spin_unlock(&fp->fi_lock); | |
9cf514cc CH |
399 | if (new) |
400 | kmem_cache_free(nfs4_layout_cache, new); | |
c5c707f9 | 401 | return nfserr; |
9cf514cc CH |
402 | } |
403 | ||
404 | static void | |
405 | nfsd4_free_layouts(struct list_head *reaplist) | |
406 | { | |
407 | while (!list_empty(reaplist)) { | |
408 | struct nfs4_layout *lp = list_first_entry(reaplist, | |
409 | struct nfs4_layout, lo_perstate); | |
410 | ||
411 | list_del(&lp->lo_perstate); | |
412 | nfs4_put_stid(&lp->lo_state->ls_stid); | |
413 | kmem_cache_free(nfs4_layout_cache, lp); | |
414 | } | |
415 | } | |
416 | ||
417 | static void | |
418 | nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg, | |
419 | struct list_head *reaplist) | |
420 | { | |
421 | struct nfsd4_layout_seg *lo = &lp->lo_seg; | |
422 | u64 end = layout_end(lo); | |
423 | ||
424 | if (seg->offset <= lo->offset) { | |
425 | if (layout_end(seg) >= end) { | |
426 | list_move_tail(&lp->lo_perstate, reaplist); | |
427 | return; | |
428 | } | |
429 | end = seg->offset; | |
430 | } else { | |
431 | /* retain the whole layout segment on a split. */ | |
432 | if (layout_end(seg) < end) { | |
433 | dprintk("%s: split not supported\n", __func__); | |
434 | return; | |
435 | } | |
436 | ||
437 | lo->offset = layout_end(seg); | |
438 | } | |
439 | ||
440 | layout_update_len(lo, end); | |
441 | } | |
442 | ||
443 | __be32 | |
444 | nfsd4_return_file_layouts(struct svc_rqst *rqstp, | |
445 | struct nfsd4_compound_state *cstate, | |
446 | struct nfsd4_layoutreturn *lrp) | |
447 | { | |
448 | struct nfs4_layout_stateid *ls; | |
449 | struct nfs4_layout *lp, *n; | |
450 | LIST_HEAD(reaplist); | |
451 | __be32 nfserr; | |
452 | int found = 0; | |
453 | ||
454 | nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid, | |
455 | false, lrp->lr_layout_type, | |
456 | &ls); | |
457 | if (nfserr) | |
458 | return nfserr; | |
459 | ||
460 | spin_lock(&ls->ls_lock); | |
461 | list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) { | |
462 | if (layouts_overlapping(lp, &lrp->lr_seg)) { | |
463 | nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist); | |
464 | found++; | |
465 | } | |
466 | } | |
467 | if (!list_empty(&ls->ls_layouts)) { | |
468 | if (found) { | |
469 | update_stateid(&ls->ls_stid.sc_stateid); | |
470 | memcpy(&lrp->lr_sid, &ls->ls_stid.sc_stateid, | |
471 | sizeof(stateid_t)); | |
472 | } | |
473 | lrp->lrs_present = 1; | |
474 | } else { | |
475 | nfs4_unhash_stid(&ls->ls_stid); | |
476 | lrp->lrs_present = 0; | |
477 | } | |
478 | spin_unlock(&ls->ls_lock); | |
479 | ||
480 | nfs4_put_stid(&ls->ls_stid); | |
481 | nfsd4_free_layouts(&reaplist); | |
482 | return nfs_ok; | |
483 | } | |
484 | ||
485 | __be32 | |
486 | nfsd4_return_client_layouts(struct svc_rqst *rqstp, | |
487 | struct nfsd4_compound_state *cstate, | |
488 | struct nfsd4_layoutreturn *lrp) | |
489 | { | |
490 | struct nfs4_layout_stateid *ls, *n; | |
491 | struct nfs4_client *clp = cstate->clp; | |
492 | struct nfs4_layout *lp, *t; | |
493 | LIST_HEAD(reaplist); | |
494 | ||
495 | lrp->lrs_present = 0; | |
496 | ||
497 | spin_lock(&clp->cl_lock); | |
498 | list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) { | |
499 | if (lrp->lr_return_type == RETURN_FSID && | |
500 | !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle, | |
501 | &cstate->current_fh.fh_handle)) | |
502 | continue; | |
503 | ||
504 | spin_lock(&ls->ls_lock); | |
505 | list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) { | |
506 | if (lrp->lr_seg.iomode == IOMODE_ANY || | |
507 | lrp->lr_seg.iomode == lp->lo_seg.iomode) | |
508 | list_move_tail(&lp->lo_perstate, &reaplist); | |
509 | } | |
510 | spin_unlock(&ls->ls_lock); | |
511 | } | |
512 | spin_unlock(&clp->cl_lock); | |
513 | ||
514 | nfsd4_free_layouts(&reaplist); | |
515 | return 0; | |
516 | } | |
517 | ||
518 | static void | |
519 | nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls, | |
520 | struct list_head *reaplist) | |
521 | { | |
522 | spin_lock(&ls->ls_lock); | |
523 | list_splice_init(&ls->ls_layouts, reaplist); | |
524 | spin_unlock(&ls->ls_lock); | |
525 | } | |
526 | ||
527 | void | |
528 | nfsd4_return_all_client_layouts(struct nfs4_client *clp) | |
529 | { | |
530 | struct nfs4_layout_stateid *ls, *n; | |
531 | LIST_HEAD(reaplist); | |
532 | ||
533 | spin_lock(&clp->cl_lock); | |
534 | list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) | |
535 | nfsd4_return_all_layouts(ls, &reaplist); | |
536 | spin_unlock(&clp->cl_lock); | |
537 | ||
538 | nfsd4_free_layouts(&reaplist); | |
539 | } | |
540 | ||
541 | void | |
542 | nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp) | |
543 | { | |
544 | struct nfs4_layout_stateid *ls, *n; | |
545 | LIST_HEAD(reaplist); | |
546 | ||
547 | spin_lock(&fp->fi_lock); | |
548 | list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) { | |
549 | if (ls->ls_stid.sc_client == clp) | |
550 | nfsd4_return_all_layouts(ls, &reaplist); | |
551 | } | |
552 | spin_unlock(&fp->fi_lock); | |
553 | ||
554 | nfsd4_free_layouts(&reaplist); | |
555 | } | |
556 | ||
c5c707f9 CH |
557 | static void |
558 | nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls) | |
559 | { | |
560 | struct nfs4_client *clp = ls->ls_stid.sc_client; | |
561 | char addr_str[INET6_ADDRSTRLEN]; | |
562 | static char *envp[] = { | |
563 | "HOME=/", | |
564 | "TERM=linux", | |
565 | "PATH=/sbin:/usr/sbin:/bin:/usr/bin", | |
566 | NULL | |
567 | }; | |
568 | char *argv[8]; | |
569 | int error; | |
570 | ||
571 | rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str)); | |
572 | ||
573 | printk(KERN_WARNING | |
574 | "nfsd: client %s failed to respond to layout recall. " | |
575 | " Fencing..\n", addr_str); | |
576 | ||
577 | argv[0] = "/sbin/nfsd-recall-failed"; | |
578 | argv[1] = addr_str; | |
579 | argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id; | |
580 | argv[3] = NULL; | |
581 | ||
582 | error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC); | |
583 | if (error) { | |
584 | printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n", | |
585 | addr_str, error); | |
586 | } | |
587 | } | |
588 | ||
589 | static int | |
590 | nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task) | |
591 | { | |
592 | struct nfs4_layout_stateid *ls = | |
593 | container_of(cb, struct nfs4_layout_stateid, ls_recall); | |
594 | LIST_HEAD(reaplist); | |
595 | ||
596 | switch (task->tk_status) { | |
597 | case 0: | |
598 | return 1; | |
599 | case -NFS4ERR_NOMATCHING_LAYOUT: | |
600 | task->tk_status = 0; | |
601 | return 1; | |
602 | case -NFS4ERR_DELAY: | |
603 | /* Poll the client until it's done with the layout */ | |
604 | /* FIXME: cap number of retries. | |
605 | * The pnfs standard states that we need to only expire | |
606 | * the client after at-least "lease time" .eg lease-time * 2 | |
607 | * when failing to communicate a recall | |
608 | */ | |
609 | rpc_delay(task, HZ/100); /* 10 mili-seconds */ | |
610 | return 0; | |
611 | default: | |
612 | /* | |
613 | * Unknown error or non-responding client, we'll need to fence. | |
614 | */ | |
615 | nfsd4_cb_layout_fail(ls); | |
616 | return -1; | |
617 | } | |
618 | } | |
619 | ||
620 | static void | |
621 | nfsd4_cb_layout_release(struct nfsd4_callback *cb) | |
622 | { | |
623 | struct nfs4_layout_stateid *ls = | |
624 | container_of(cb, struct nfs4_layout_stateid, ls_recall); | |
625 | LIST_HEAD(reaplist); | |
626 | ||
627 | nfsd4_return_all_layouts(ls, &reaplist); | |
628 | nfsd4_free_layouts(&reaplist); | |
629 | nfs4_put_stid(&ls->ls_stid); | |
630 | } | |
631 | ||
632 | static struct nfsd4_callback_ops nfsd4_cb_layout_ops = { | |
633 | .done = nfsd4_cb_layout_done, | |
634 | .release = nfsd4_cb_layout_release, | |
635 | }; | |
636 | ||
637 | static bool | |
638 | nfsd4_layout_lm_break(struct file_lock *fl) | |
639 | { | |
640 | /* | |
641 | * We don't want the locks code to timeout the lease for us; | |
642 | * we'll remove it ourself if a layout isn't returned | |
643 | * in time: | |
644 | */ | |
645 | fl->fl_break_time = 0; | |
646 | nfsd4_recall_file_layout(fl->fl_owner); | |
647 | return false; | |
648 | } | |
649 | ||
650 | static int | |
651 | nfsd4_layout_lm_change(struct file_lock *onlist, int arg, | |
652 | struct list_head *dispose) | |
653 | { | |
654 | BUG_ON(!(arg & F_UNLCK)); | |
655 | return lease_modify(onlist, arg, dispose); | |
656 | } | |
657 | ||
658 | static const struct lock_manager_operations nfsd4_layouts_lm_ops = { | |
659 | .lm_break = nfsd4_layout_lm_break, | |
660 | .lm_change = nfsd4_layout_lm_change, | |
661 | }; | |
662 | ||
9cf514cc CH |
663 | int |
664 | nfsd4_init_pnfs(void) | |
665 | { | |
666 | int i; | |
667 | ||
668 | for (i = 0; i < DEVID_HASH_SIZE; i++) | |
669 | INIT_LIST_HEAD(&nfsd_devid_hash[i]); | |
670 | ||
671 | nfs4_layout_cache = kmem_cache_create("nfs4_layout", | |
672 | sizeof(struct nfs4_layout), 0, 0, NULL); | |
673 | if (!nfs4_layout_cache) | |
674 | return -ENOMEM; | |
675 | ||
676 | nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid", | |
677 | sizeof(struct nfs4_layout_stateid), 0, 0, NULL); | |
678 | if (!nfs4_layout_stateid_cache) { | |
679 | kmem_cache_destroy(nfs4_layout_cache); | |
680 | return -ENOMEM; | |
681 | } | |
682 | return 0; | |
683 | } | |
684 | ||
685 | void | |
686 | nfsd4_exit_pnfs(void) | |
687 | { | |
688 | int i; | |
689 | ||
690 | kmem_cache_destroy(nfs4_layout_cache); | |
691 | kmem_cache_destroy(nfs4_layout_stateid_cache); | |
692 | ||
693 | for (i = 0; i < DEVID_HASH_SIZE; i++) { | |
694 | struct nfsd4_deviceid_map *map, *n; | |
695 | ||
696 | list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash) | |
697 | kfree(map); | |
698 | } | |
699 | } |