]>
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" | |
31ef83dc | 12 | #include "trace.h" |
9cf514cc CH |
13 | |
14 | #define NFSDDBG_FACILITY NFSDDBG_PNFS | |
15 | ||
16 | struct nfs4_layout { | |
17 | struct list_head lo_perstate; | |
18 | struct nfs4_layout_stateid *lo_state; | |
19 | struct nfsd4_layout_seg lo_seg; | |
20 | }; | |
21 | ||
22 | static struct kmem_cache *nfs4_layout_cache; | |
23 | static struct kmem_cache *nfs4_layout_stateid_cache; | |
24 | ||
c5c707f9 CH |
25 | static struct nfsd4_callback_ops nfsd4_cb_layout_ops; |
26 | static const struct lock_manager_operations nfsd4_layouts_lm_ops; | |
27 | ||
9cf514cc | 28 | const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] = { |
8650b8a0 | 29 | [LAYOUT_BLOCK_VOLUME] = &bl_layout_ops, |
9cf514cc CH |
30 | }; |
31 | ||
32 | /* pNFS device ID to export fsid mapping */ | |
33 | #define DEVID_HASH_BITS 8 | |
34 | #define DEVID_HASH_SIZE (1 << DEVID_HASH_BITS) | |
35 | #define DEVID_HASH_MASK (DEVID_HASH_SIZE - 1) | |
36 | static u64 nfsd_devid_seq = 1; | |
37 | static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE]; | |
38 | static DEFINE_SPINLOCK(nfsd_devid_lock); | |
39 | ||
40 | static inline u32 devid_hashfn(u64 idx) | |
41 | { | |
42 | return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK; | |
43 | } | |
44 | ||
45 | static void | |
46 | nfsd4_alloc_devid_map(const struct svc_fh *fhp) | |
47 | { | |
48 | const struct knfsd_fh *fh = &fhp->fh_handle; | |
49 | size_t fsid_len = key_len(fh->fh_fsid_type); | |
50 | struct nfsd4_deviceid_map *map, *old; | |
51 | int i; | |
52 | ||
53 | map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL); | |
54 | if (!map) | |
55 | return; | |
56 | ||
57 | map->fsid_type = fh->fh_fsid_type; | |
58 | memcpy(&map->fsid, fh->fh_fsid, fsid_len); | |
59 | ||
60 | spin_lock(&nfsd_devid_lock); | |
61 | if (fhp->fh_export->ex_devid_map) | |
62 | goto out_unlock; | |
63 | ||
64 | for (i = 0; i < DEVID_HASH_SIZE; i++) { | |
65 | list_for_each_entry(old, &nfsd_devid_hash[i], hash) { | |
66 | if (old->fsid_type != fh->fh_fsid_type) | |
67 | continue; | |
68 | if (memcmp(old->fsid, fh->fh_fsid, | |
69 | key_len(old->fsid_type))) | |
70 | continue; | |
71 | ||
72 | fhp->fh_export->ex_devid_map = old; | |
73 | goto out_unlock; | |
74 | } | |
75 | } | |
76 | ||
77 | map->idx = nfsd_devid_seq++; | |
78 | list_add_tail_rcu(&map->hash, &nfsd_devid_hash[devid_hashfn(map->idx)]); | |
79 | fhp->fh_export->ex_devid_map = map; | |
80 | map = NULL; | |
81 | ||
82 | out_unlock: | |
83 | spin_unlock(&nfsd_devid_lock); | |
84 | kfree(map); | |
85 | } | |
86 | ||
87 | struct nfsd4_deviceid_map * | |
88 | nfsd4_find_devid_map(int idx) | |
89 | { | |
90 | struct nfsd4_deviceid_map *map, *ret = NULL; | |
91 | ||
92 | rcu_read_lock(); | |
93 | list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash) | |
94 | if (map->idx == idx) | |
95 | ret = map; | |
96 | rcu_read_unlock(); | |
97 | ||
98 | return ret; | |
99 | } | |
100 | ||
101 | int | |
102 | nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp, | |
103 | u32 device_generation) | |
104 | { | |
105 | if (!fhp->fh_export->ex_devid_map) { | |
106 | nfsd4_alloc_devid_map(fhp); | |
107 | if (!fhp->fh_export->ex_devid_map) | |
108 | return -ENOMEM; | |
109 | } | |
110 | ||
111 | id->fsid_idx = fhp->fh_export->ex_devid_map->idx; | |
112 | id->generation = device_generation; | |
113 | id->pad = 0; | |
114 | return 0; | |
115 | } | |
116 | ||
117 | void nfsd4_setup_layout_type(struct svc_export *exp) | |
118 | { | |
8650b8a0 CH |
119 | struct super_block *sb = exp->ex_path.mnt->mnt_sb; |
120 | ||
f3f03330 | 121 | if (!(exp->ex_flags & NFSEXP_PNFS)) |
9cf514cc | 122 | return; |
8650b8a0 CH |
123 | |
124 | if (sb->s_export_op->get_uuid && | |
125 | sb->s_export_op->map_blocks && | |
126 | sb->s_export_op->commit_blocks) | |
127 | exp->ex_layout_type = LAYOUT_BLOCK_VOLUME; | |
9cf514cc CH |
128 | } |
129 | ||
130 | static void | |
131 | nfsd4_free_layout_stateid(struct nfs4_stid *stid) | |
132 | { | |
133 | struct nfs4_layout_stateid *ls = layoutstateid(stid); | |
134 | struct nfs4_client *clp = ls->ls_stid.sc_client; | |
135 | struct nfs4_file *fp = ls->ls_stid.sc_file; | |
136 | ||
31ef83dc CH |
137 | trace_layoutstate_free(&ls->ls_stid.sc_stateid); |
138 | ||
9cf514cc CH |
139 | spin_lock(&clp->cl_lock); |
140 | list_del_init(&ls->ls_perclnt); | |
141 | spin_unlock(&clp->cl_lock); | |
142 | ||
143 | spin_lock(&fp->fi_lock); | |
144 | list_del_init(&ls->ls_perfile); | |
145 | spin_unlock(&fp->fi_lock); | |
146 | ||
c5c707f9 CH |
147 | vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls); |
148 | fput(ls->ls_file); | |
149 | ||
150 | if (ls->ls_recalled) | |
151 | atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls); | |
152 | ||
9cf514cc CH |
153 | kmem_cache_free(nfs4_layout_stateid_cache, ls); |
154 | } | |
155 | ||
c5c707f9 CH |
156 | static int |
157 | nfsd4_layout_setlease(struct nfs4_layout_stateid *ls) | |
158 | { | |
159 | struct file_lock *fl; | |
160 | int status; | |
161 | ||
162 | fl = locks_alloc_lock(); | |
163 | if (!fl) | |
164 | return -ENOMEM; | |
165 | locks_init_lock(fl); | |
166 | fl->fl_lmops = &nfsd4_layouts_lm_ops; | |
167 | fl->fl_flags = FL_LAYOUT; | |
168 | fl->fl_type = F_RDLCK; | |
169 | fl->fl_end = OFFSET_MAX; | |
170 | fl->fl_owner = ls; | |
171 | fl->fl_pid = current->tgid; | |
172 | fl->fl_file = ls->ls_file; | |
173 | ||
174 | status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL); | |
175 | if (status) { | |
176 | locks_free_lock(fl); | |
177 | return status; | |
178 | } | |
179 | BUG_ON(fl != NULL); | |
180 | return 0; | |
181 | } | |
182 | ||
9cf514cc CH |
183 | static struct nfs4_layout_stateid * |
184 | nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate, | |
185 | struct nfs4_stid *parent, u32 layout_type) | |
186 | { | |
187 | struct nfs4_client *clp = cstate->clp; | |
188 | struct nfs4_file *fp = parent->sc_file; | |
189 | struct nfs4_layout_stateid *ls; | |
190 | struct nfs4_stid *stp; | |
191 | ||
192 | stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache); | |
193 | if (!stp) | |
194 | return NULL; | |
195 | stp->sc_free = nfsd4_free_layout_stateid; | |
196 | get_nfs4_file(fp); | |
197 | stp->sc_file = fp; | |
198 | ||
199 | ls = layoutstateid(stp); | |
200 | INIT_LIST_HEAD(&ls->ls_perclnt); | |
201 | INIT_LIST_HEAD(&ls->ls_perfile); | |
202 | spin_lock_init(&ls->ls_lock); | |
203 | INIT_LIST_HEAD(&ls->ls_layouts); | |
204 | ls->ls_layout_type = layout_type; | |
c5c707f9 CH |
205 | nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops, |
206 | NFSPROC4_CLNT_CB_LAYOUT); | |
207 | ||
208 | if (parent->sc_type == NFS4_DELEG_STID) | |
209 | ls->ls_file = get_file(fp->fi_deleg_file); | |
210 | else | |
211 | ls->ls_file = find_any_file(fp); | |
212 | BUG_ON(!ls->ls_file); | |
213 | ||
214 | if (nfsd4_layout_setlease(ls)) { | |
1ca4b88e | 215 | fput(ls->ls_file); |
c5c707f9 CH |
216 | put_nfs4_file(fp); |
217 | kmem_cache_free(nfs4_layout_stateid_cache, ls); | |
218 | return NULL; | |
219 | } | |
9cf514cc CH |
220 | |
221 | spin_lock(&clp->cl_lock); | |
222 | stp->sc_type = NFS4_LAYOUT_STID; | |
223 | list_add(&ls->ls_perclnt, &clp->cl_lo_states); | |
224 | spin_unlock(&clp->cl_lock); | |
225 | ||
226 | spin_lock(&fp->fi_lock); | |
227 | list_add(&ls->ls_perfile, &fp->fi_lo_states); | |
228 | spin_unlock(&fp->fi_lock); | |
229 | ||
31ef83dc | 230 | trace_layoutstate_alloc(&ls->ls_stid.sc_stateid); |
9cf514cc CH |
231 | return ls; |
232 | } | |
233 | ||
234 | __be32 | |
235 | nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp, | |
236 | struct nfsd4_compound_state *cstate, stateid_t *stateid, | |
237 | bool create, u32 layout_type, struct nfs4_layout_stateid **lsp) | |
238 | { | |
239 | struct nfs4_layout_stateid *ls; | |
240 | struct nfs4_stid *stid; | |
241 | unsigned char typemask = NFS4_LAYOUT_STID; | |
242 | __be32 status; | |
243 | ||
244 | if (create) | |
245 | typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID); | |
246 | ||
247 | status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid, | |
248 | net_generic(SVC_NET(rqstp), nfsd_net_id)); | |
249 | if (status) | |
250 | goto out; | |
251 | ||
252 | if (!fh_match(&cstate->current_fh.fh_handle, | |
253 | &stid->sc_file->fi_fhandle)) { | |
254 | status = nfserr_bad_stateid; | |
255 | goto out_put_stid; | |
256 | } | |
257 | ||
258 | if (stid->sc_type != NFS4_LAYOUT_STID) { | |
259 | ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type); | |
260 | nfs4_put_stid(stid); | |
261 | ||
262 | status = nfserr_jukebox; | |
263 | if (!ls) | |
264 | goto out; | |
265 | } else { | |
266 | ls = container_of(stid, struct nfs4_layout_stateid, ls_stid); | |
267 | ||
268 | status = nfserr_bad_stateid; | |
269 | if (stateid->si_generation > stid->sc_stateid.si_generation) | |
270 | goto out_put_stid; | |
271 | if (layout_type != ls->ls_layout_type) | |
272 | goto out_put_stid; | |
273 | } | |
274 | ||
275 | *lsp = ls; | |
276 | return 0; | |
277 | ||
278 | out_put_stid: | |
279 | nfs4_put_stid(stid); | |
280 | out: | |
281 | return status; | |
282 | } | |
283 | ||
c5c707f9 CH |
284 | static void |
285 | nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls) | |
286 | { | |
287 | spin_lock(&ls->ls_lock); | |
288 | if (ls->ls_recalled) | |
289 | goto out_unlock; | |
290 | ||
291 | ls->ls_recalled = true; | |
292 | atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls); | |
293 | if (list_empty(&ls->ls_layouts)) | |
294 | goto out_unlock; | |
295 | ||
31ef83dc CH |
296 | trace_layout_recall(&ls->ls_stid.sc_stateid); |
297 | ||
c5c707f9 CH |
298 | atomic_inc(&ls->ls_stid.sc_count); |
299 | update_stateid(&ls->ls_stid.sc_stateid); | |
300 | memcpy(&ls->ls_recall_sid, &ls->ls_stid.sc_stateid, sizeof(stateid_t)); | |
301 | nfsd4_run_cb(&ls->ls_recall); | |
302 | ||
303 | out_unlock: | |
304 | spin_unlock(&ls->ls_lock); | |
305 | } | |
306 | ||
9cf514cc CH |
307 | static inline u64 |
308 | layout_end(struct nfsd4_layout_seg *seg) | |
309 | { | |
310 | u64 end = seg->offset + seg->length; | |
311 | return end >= seg->offset ? end : NFS4_MAX_UINT64; | |
312 | } | |
313 | ||
314 | static void | |
315 | layout_update_len(struct nfsd4_layout_seg *lo, u64 end) | |
316 | { | |
317 | if (end == NFS4_MAX_UINT64) | |
318 | lo->length = NFS4_MAX_UINT64; | |
319 | else | |
320 | lo->length = end - lo->offset; | |
321 | } | |
322 | ||
323 | static bool | |
324 | layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s) | |
325 | { | |
326 | if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode) | |
327 | return false; | |
328 | if (layout_end(&lo->lo_seg) <= s->offset) | |
329 | return false; | |
330 | if (layout_end(s) <= lo->lo_seg.offset) | |
331 | return false; | |
332 | return true; | |
333 | } | |
334 | ||
335 | static bool | |
336 | layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new) | |
337 | { | |
338 | if (lo->iomode != new->iomode) | |
339 | return false; | |
340 | if (layout_end(new) < lo->offset) | |
341 | return false; | |
342 | if (layout_end(lo) < new->offset) | |
343 | return false; | |
344 | ||
345 | lo->offset = min(lo->offset, new->offset); | |
346 | layout_update_len(lo, max(layout_end(lo), layout_end(new))); | |
347 | return true; | |
348 | } | |
349 | ||
c5c707f9 CH |
350 | static __be32 |
351 | nfsd4_recall_conflict(struct nfs4_layout_stateid *ls) | |
352 | { | |
353 | struct nfs4_file *fp = ls->ls_stid.sc_file; | |
354 | struct nfs4_layout_stateid *l, *n; | |
355 | __be32 nfserr = nfs_ok; | |
356 | ||
357 | assert_spin_locked(&fp->fi_lock); | |
358 | ||
359 | list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) { | |
360 | if (l != ls) { | |
361 | nfsd4_recall_file_layout(l); | |
362 | nfserr = nfserr_recallconflict; | |
363 | } | |
364 | } | |
365 | ||
366 | return nfserr; | |
367 | } | |
368 | ||
9cf514cc CH |
369 | __be32 |
370 | nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls) | |
371 | { | |
372 | struct nfsd4_layout_seg *seg = &lgp->lg_seg; | |
c5c707f9 | 373 | struct nfs4_file *fp = ls->ls_stid.sc_file; |
9cf514cc | 374 | struct nfs4_layout *lp, *new = NULL; |
c5c707f9 | 375 | __be32 nfserr; |
9cf514cc | 376 | |
c5c707f9 CH |
377 | spin_lock(&fp->fi_lock); |
378 | nfserr = nfsd4_recall_conflict(ls); | |
379 | if (nfserr) | |
380 | goto out; | |
9cf514cc CH |
381 | spin_lock(&ls->ls_lock); |
382 | list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) { | |
383 | if (layouts_try_merge(&lp->lo_seg, seg)) | |
384 | goto done; | |
385 | } | |
386 | spin_unlock(&ls->ls_lock); | |
c5c707f9 | 387 | spin_unlock(&fp->fi_lock); |
9cf514cc CH |
388 | |
389 | new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL); | |
390 | if (!new) | |
391 | return nfserr_jukebox; | |
392 | memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg)); | |
393 | new->lo_state = ls; | |
394 | ||
c5c707f9 CH |
395 | spin_lock(&fp->fi_lock); |
396 | nfserr = nfsd4_recall_conflict(ls); | |
397 | if (nfserr) | |
398 | goto out; | |
9cf514cc CH |
399 | spin_lock(&ls->ls_lock); |
400 | list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) { | |
401 | if (layouts_try_merge(&lp->lo_seg, seg)) | |
402 | goto done; | |
403 | } | |
404 | ||
405 | atomic_inc(&ls->ls_stid.sc_count); | |
406 | list_add_tail(&new->lo_perstate, &ls->ls_layouts); | |
407 | new = NULL; | |
408 | done: | |
409 | update_stateid(&ls->ls_stid.sc_stateid); | |
410 | memcpy(&lgp->lg_sid, &ls->ls_stid.sc_stateid, sizeof(stateid_t)); | |
411 | spin_unlock(&ls->ls_lock); | |
c5c707f9 CH |
412 | out: |
413 | spin_unlock(&fp->fi_lock); | |
9cf514cc CH |
414 | if (new) |
415 | kmem_cache_free(nfs4_layout_cache, new); | |
c5c707f9 | 416 | return nfserr; |
9cf514cc CH |
417 | } |
418 | ||
419 | static void | |
420 | nfsd4_free_layouts(struct list_head *reaplist) | |
421 | { | |
422 | while (!list_empty(reaplist)) { | |
423 | struct nfs4_layout *lp = list_first_entry(reaplist, | |
424 | struct nfs4_layout, lo_perstate); | |
425 | ||
426 | list_del(&lp->lo_perstate); | |
427 | nfs4_put_stid(&lp->lo_state->ls_stid); | |
428 | kmem_cache_free(nfs4_layout_cache, lp); | |
429 | } | |
430 | } | |
431 | ||
432 | static void | |
433 | nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg, | |
434 | struct list_head *reaplist) | |
435 | { | |
436 | struct nfsd4_layout_seg *lo = &lp->lo_seg; | |
437 | u64 end = layout_end(lo); | |
438 | ||
439 | if (seg->offset <= lo->offset) { | |
440 | if (layout_end(seg) >= end) { | |
441 | list_move_tail(&lp->lo_perstate, reaplist); | |
442 | return; | |
443 | } | |
7890203d | 444 | lo->offset = layout_end(seg); |
9cf514cc CH |
445 | } else { |
446 | /* retain the whole layout segment on a split. */ | |
447 | if (layout_end(seg) < end) { | |
448 | dprintk("%s: split not supported\n", __func__); | |
449 | return; | |
450 | } | |
7890203d | 451 | end = seg->offset; |
9cf514cc CH |
452 | } |
453 | ||
454 | layout_update_len(lo, end); | |
455 | } | |
456 | ||
457 | __be32 | |
458 | nfsd4_return_file_layouts(struct svc_rqst *rqstp, | |
459 | struct nfsd4_compound_state *cstate, | |
460 | struct nfsd4_layoutreturn *lrp) | |
461 | { | |
462 | struct nfs4_layout_stateid *ls; | |
463 | struct nfs4_layout *lp, *n; | |
464 | LIST_HEAD(reaplist); | |
465 | __be32 nfserr; | |
466 | int found = 0; | |
467 | ||
468 | nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid, | |
469 | false, lrp->lr_layout_type, | |
470 | &ls); | |
31ef83dc CH |
471 | if (nfserr) { |
472 | trace_layout_return_lookup_fail(&lrp->lr_sid); | |
9cf514cc | 473 | return nfserr; |
31ef83dc | 474 | } |
9cf514cc CH |
475 | |
476 | spin_lock(&ls->ls_lock); | |
477 | list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) { | |
478 | if (layouts_overlapping(lp, &lrp->lr_seg)) { | |
479 | nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist); | |
480 | found++; | |
481 | } | |
482 | } | |
483 | if (!list_empty(&ls->ls_layouts)) { | |
484 | if (found) { | |
485 | update_stateid(&ls->ls_stid.sc_stateid); | |
486 | memcpy(&lrp->lr_sid, &ls->ls_stid.sc_stateid, | |
487 | sizeof(stateid_t)); | |
488 | } | |
489 | lrp->lrs_present = 1; | |
490 | } else { | |
31ef83dc | 491 | trace_layoutstate_unhash(&ls->ls_stid.sc_stateid); |
9cf514cc CH |
492 | nfs4_unhash_stid(&ls->ls_stid); |
493 | lrp->lrs_present = 0; | |
494 | } | |
495 | spin_unlock(&ls->ls_lock); | |
496 | ||
497 | nfs4_put_stid(&ls->ls_stid); | |
498 | nfsd4_free_layouts(&reaplist); | |
499 | return nfs_ok; | |
500 | } | |
501 | ||
502 | __be32 | |
503 | nfsd4_return_client_layouts(struct svc_rqst *rqstp, | |
504 | struct nfsd4_compound_state *cstate, | |
505 | struct nfsd4_layoutreturn *lrp) | |
506 | { | |
507 | struct nfs4_layout_stateid *ls, *n; | |
508 | struct nfs4_client *clp = cstate->clp; | |
509 | struct nfs4_layout *lp, *t; | |
510 | LIST_HEAD(reaplist); | |
511 | ||
512 | lrp->lrs_present = 0; | |
513 | ||
514 | spin_lock(&clp->cl_lock); | |
515 | list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) { | |
6f8f28ec KM |
516 | if (ls->ls_layout_type != lrp->lr_layout_type) |
517 | continue; | |
518 | ||
9cf514cc CH |
519 | if (lrp->lr_return_type == RETURN_FSID && |
520 | !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle, | |
521 | &cstate->current_fh.fh_handle)) | |
522 | continue; | |
523 | ||
524 | spin_lock(&ls->ls_lock); | |
525 | list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) { | |
526 | if (lrp->lr_seg.iomode == IOMODE_ANY || | |
527 | lrp->lr_seg.iomode == lp->lo_seg.iomode) | |
528 | list_move_tail(&lp->lo_perstate, &reaplist); | |
529 | } | |
530 | spin_unlock(&ls->ls_lock); | |
531 | } | |
532 | spin_unlock(&clp->cl_lock); | |
533 | ||
534 | nfsd4_free_layouts(&reaplist); | |
535 | return 0; | |
536 | } | |
537 | ||
538 | static void | |
539 | nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls, | |
540 | struct list_head *reaplist) | |
541 | { | |
542 | spin_lock(&ls->ls_lock); | |
543 | list_splice_init(&ls->ls_layouts, reaplist); | |
544 | spin_unlock(&ls->ls_lock); | |
545 | } | |
546 | ||
547 | void | |
548 | nfsd4_return_all_client_layouts(struct nfs4_client *clp) | |
549 | { | |
550 | struct nfs4_layout_stateid *ls, *n; | |
551 | LIST_HEAD(reaplist); | |
552 | ||
553 | spin_lock(&clp->cl_lock); | |
554 | list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) | |
555 | nfsd4_return_all_layouts(ls, &reaplist); | |
556 | spin_unlock(&clp->cl_lock); | |
557 | ||
558 | nfsd4_free_layouts(&reaplist); | |
559 | } | |
560 | ||
561 | void | |
562 | nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp) | |
563 | { | |
564 | struct nfs4_layout_stateid *ls, *n; | |
565 | LIST_HEAD(reaplist); | |
566 | ||
567 | spin_lock(&fp->fi_lock); | |
568 | list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) { | |
569 | if (ls->ls_stid.sc_client == clp) | |
570 | nfsd4_return_all_layouts(ls, &reaplist); | |
571 | } | |
572 | spin_unlock(&fp->fi_lock); | |
573 | ||
574 | nfsd4_free_layouts(&reaplist); | |
575 | } | |
576 | ||
c5c707f9 CH |
577 | static void |
578 | nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls) | |
579 | { | |
580 | struct nfs4_client *clp = ls->ls_stid.sc_client; | |
581 | char addr_str[INET6_ADDRSTRLEN]; | |
582 | static char *envp[] = { | |
583 | "HOME=/", | |
584 | "TERM=linux", | |
585 | "PATH=/sbin:/usr/sbin:/bin:/usr/bin", | |
586 | NULL | |
587 | }; | |
588 | char *argv[8]; | |
589 | int error; | |
590 | ||
591 | rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str)); | |
592 | ||
715a03d2 KM |
593 | trace_layout_recall_fail(&ls->ls_stid.sc_stateid); |
594 | ||
c5c707f9 CH |
595 | printk(KERN_WARNING |
596 | "nfsd: client %s failed to respond to layout recall. " | |
597 | " Fencing..\n", addr_str); | |
598 | ||
599 | argv[0] = "/sbin/nfsd-recall-failed"; | |
600 | argv[1] = addr_str; | |
601 | argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id; | |
602 | argv[3] = NULL; | |
603 | ||
604 | error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC); | |
605 | if (error) { | |
606 | printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n", | |
607 | addr_str, error); | |
608 | } | |
609 | } | |
610 | ||
611 | static int | |
612 | nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task) | |
613 | { | |
614 | struct nfs4_layout_stateid *ls = | |
615 | container_of(cb, struct nfs4_layout_stateid, ls_recall); | |
616 | LIST_HEAD(reaplist); | |
617 | ||
618 | switch (task->tk_status) { | |
619 | case 0: | |
620 | return 1; | |
621 | case -NFS4ERR_NOMATCHING_LAYOUT: | |
31ef83dc | 622 | trace_layout_recall_done(&ls->ls_stid.sc_stateid); |
c5c707f9 CH |
623 | task->tk_status = 0; |
624 | return 1; | |
625 | case -NFS4ERR_DELAY: | |
626 | /* Poll the client until it's done with the layout */ | |
627 | /* FIXME: cap number of retries. | |
628 | * The pnfs standard states that we need to only expire | |
629 | * the client after at-least "lease time" .eg lease-time * 2 | |
630 | * when failing to communicate a recall | |
631 | */ | |
632 | rpc_delay(task, HZ/100); /* 10 mili-seconds */ | |
633 | return 0; | |
634 | default: | |
635 | /* | |
636 | * Unknown error or non-responding client, we'll need to fence. | |
637 | */ | |
638 | nfsd4_cb_layout_fail(ls); | |
639 | return -1; | |
640 | } | |
641 | } | |
642 | ||
643 | static void | |
644 | nfsd4_cb_layout_release(struct nfsd4_callback *cb) | |
645 | { | |
646 | struct nfs4_layout_stateid *ls = | |
647 | container_of(cb, struct nfs4_layout_stateid, ls_recall); | |
648 | LIST_HEAD(reaplist); | |
649 | ||
31ef83dc CH |
650 | trace_layout_recall_release(&ls->ls_stid.sc_stateid); |
651 | ||
c5c707f9 CH |
652 | nfsd4_return_all_layouts(ls, &reaplist); |
653 | nfsd4_free_layouts(&reaplist); | |
654 | nfs4_put_stid(&ls->ls_stid); | |
655 | } | |
656 | ||
657 | static struct nfsd4_callback_ops nfsd4_cb_layout_ops = { | |
658 | .done = nfsd4_cb_layout_done, | |
659 | .release = nfsd4_cb_layout_release, | |
660 | }; | |
661 | ||
662 | static bool | |
663 | nfsd4_layout_lm_break(struct file_lock *fl) | |
664 | { | |
665 | /* | |
666 | * We don't want the locks code to timeout the lease for us; | |
667 | * we'll remove it ourself if a layout isn't returned | |
668 | * in time: | |
669 | */ | |
670 | fl->fl_break_time = 0; | |
671 | nfsd4_recall_file_layout(fl->fl_owner); | |
672 | return false; | |
673 | } | |
674 | ||
675 | static int | |
676 | nfsd4_layout_lm_change(struct file_lock *onlist, int arg, | |
677 | struct list_head *dispose) | |
678 | { | |
679 | BUG_ON(!(arg & F_UNLCK)); | |
680 | return lease_modify(onlist, arg, dispose); | |
681 | } | |
682 | ||
683 | static const struct lock_manager_operations nfsd4_layouts_lm_ops = { | |
684 | .lm_break = nfsd4_layout_lm_break, | |
685 | .lm_change = nfsd4_layout_lm_change, | |
686 | }; | |
687 | ||
9cf514cc CH |
688 | int |
689 | nfsd4_init_pnfs(void) | |
690 | { | |
691 | int i; | |
692 | ||
693 | for (i = 0; i < DEVID_HASH_SIZE; i++) | |
694 | INIT_LIST_HEAD(&nfsd_devid_hash[i]); | |
695 | ||
696 | nfs4_layout_cache = kmem_cache_create("nfs4_layout", | |
697 | sizeof(struct nfs4_layout), 0, 0, NULL); | |
698 | if (!nfs4_layout_cache) | |
699 | return -ENOMEM; | |
700 | ||
701 | nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid", | |
702 | sizeof(struct nfs4_layout_stateid), 0, 0, NULL); | |
703 | if (!nfs4_layout_stateid_cache) { | |
704 | kmem_cache_destroy(nfs4_layout_cache); | |
705 | return -ENOMEM; | |
706 | } | |
707 | return 0; | |
708 | } | |
709 | ||
710 | void | |
711 | nfsd4_exit_pnfs(void) | |
712 | { | |
713 | int i; | |
714 | ||
715 | kmem_cache_destroy(nfs4_layout_cache); | |
716 | kmem_cache_destroy(nfs4_layout_stateid_cache); | |
717 | ||
718 | for (i = 0; i < DEVID_HASH_SIZE; i++) { | |
719 | struct nfsd4_deviceid_map *map, *n; | |
720 | ||
721 | list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash) | |
722 | kfree(map); | |
723 | } | |
724 | } |