Commit | Line | Data |
---|---|---|
9cf514cc CH |
1 | /* |
2 | * Copyright (c) 2014 Christoph Hellwig. | |
3 | */ | |
f99d4fbd | 4 | #include <linux/blkdev.h> |
c5c707f9 CH |
5 | #include <linux/kmod.h> |
6 | #include <linux/file.h> | |
9cf514cc CH |
7 | #include <linux/jhash.h> |
8 | #include <linux/sched.h> | |
c5c707f9 | 9 | #include <linux/sunrpc/addr.h> |
9cf514cc CH |
10 | |
11 | #include "pnfs.h" | |
12 | #include "netns.h" | |
31ef83dc | 13 | #include "trace.h" |
9cf514cc CH |
14 | |
15 | #define NFSDDBG_FACILITY NFSDDBG_PNFS | |
16 | ||
17 | struct nfs4_layout { | |
18 | struct list_head lo_perstate; | |
19 | struct nfs4_layout_stateid *lo_state; | |
20 | struct nfsd4_layout_seg lo_seg; | |
21 | }; | |
22 | ||
23 | static struct kmem_cache *nfs4_layout_cache; | |
24 | static struct kmem_cache *nfs4_layout_stateid_cache; | |
25 | ||
c4cb8974 | 26 | static const struct nfsd4_callback_ops nfsd4_cb_layout_ops; |
c5c707f9 CH |
27 | static const struct lock_manager_operations nfsd4_layouts_lm_ops; |
28 | ||
9cf514cc | 29 | const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] = { |
9b9960a0 TH |
30 | #ifdef CONFIG_NFSD_FLEXFILELAYOUT |
31 | [LAYOUT_FLEX_FILES] = &ff_layout_ops, | |
32 | #endif | |
81c39329 | 33 | #ifdef CONFIG_NFSD_BLOCKLAYOUT |
8650b8a0 | 34 | [LAYOUT_BLOCK_VOLUME] = &bl_layout_ops, |
81c39329 | 35 | #endif |
f99d4fbd CH |
36 | #ifdef CONFIG_NFSD_SCSILAYOUT |
37 | [LAYOUT_SCSI] = &scsi_layout_ops, | |
38 | #endif | |
9cf514cc CH |
39 | }; |
40 | ||
41 | /* pNFS device ID to export fsid mapping */ | |
42 | #define DEVID_HASH_BITS 8 | |
43 | #define DEVID_HASH_SIZE (1 << DEVID_HASH_BITS) | |
44 | #define DEVID_HASH_MASK (DEVID_HASH_SIZE - 1) | |
45 | static u64 nfsd_devid_seq = 1; | |
46 | static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE]; | |
47 | static DEFINE_SPINLOCK(nfsd_devid_lock); | |
48 | ||
49 | static inline u32 devid_hashfn(u64 idx) | |
50 | { | |
51 | return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK; | |
52 | } | |
53 | ||
54 | static void | |
55 | nfsd4_alloc_devid_map(const struct svc_fh *fhp) | |
56 | { | |
57 | const struct knfsd_fh *fh = &fhp->fh_handle; | |
58 | size_t fsid_len = key_len(fh->fh_fsid_type); | |
59 | struct nfsd4_deviceid_map *map, *old; | |
60 | int i; | |
61 | ||
62 | map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL); | |
63 | if (!map) | |
64 | return; | |
65 | ||
66 | map->fsid_type = fh->fh_fsid_type; | |
67 | memcpy(&map->fsid, fh->fh_fsid, fsid_len); | |
68 | ||
69 | spin_lock(&nfsd_devid_lock); | |
70 | if (fhp->fh_export->ex_devid_map) | |
71 | goto out_unlock; | |
72 | ||
73 | for (i = 0; i < DEVID_HASH_SIZE; i++) { | |
74 | list_for_each_entry(old, &nfsd_devid_hash[i], hash) { | |
75 | if (old->fsid_type != fh->fh_fsid_type) | |
76 | continue; | |
77 | if (memcmp(old->fsid, fh->fh_fsid, | |
78 | key_len(old->fsid_type))) | |
79 | continue; | |
80 | ||
81 | fhp->fh_export->ex_devid_map = old; | |
82 | goto out_unlock; | |
83 | } | |
84 | } | |
85 | ||
86 | map->idx = nfsd_devid_seq++; | |
87 | list_add_tail_rcu(&map->hash, &nfsd_devid_hash[devid_hashfn(map->idx)]); | |
88 | fhp->fh_export->ex_devid_map = map; | |
89 | map = NULL; | |
90 | ||
91 | out_unlock: | |
92 | spin_unlock(&nfsd_devid_lock); | |
93 | kfree(map); | |
94 | } | |
95 | ||
96 | struct nfsd4_deviceid_map * | |
97 | nfsd4_find_devid_map(int idx) | |
98 | { | |
99 | struct nfsd4_deviceid_map *map, *ret = NULL; | |
100 | ||
101 | rcu_read_lock(); | |
102 | list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash) | |
103 | if (map->idx == idx) | |
104 | ret = map; | |
105 | rcu_read_unlock(); | |
106 | ||
107 | return ret; | |
108 | } | |
109 | ||
110 | int | |
111 | nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp, | |
112 | u32 device_generation) | |
113 | { | |
114 | if (!fhp->fh_export->ex_devid_map) { | |
115 | nfsd4_alloc_devid_map(fhp); | |
116 | if (!fhp->fh_export->ex_devid_map) | |
117 | return -ENOMEM; | |
118 | } | |
119 | ||
120 | id->fsid_idx = fhp->fh_export->ex_devid_map->idx; | |
121 | id->generation = device_generation; | |
122 | id->pad = 0; | |
123 | return 0; | |
124 | } | |
125 | ||
126 | void nfsd4_setup_layout_type(struct svc_export *exp) | |
127 | { | |
9b9960a0 | 128 | #if defined(CONFIG_NFSD_BLOCKLAYOUT) || defined(CONFIG_NFSD_SCSILAYOUT) |
8650b8a0 | 129 | struct super_block *sb = exp->ex_path.mnt->mnt_sb; |
9b9960a0 | 130 | #endif |
8650b8a0 | 131 | |
f3f03330 | 132 | if (!(exp->ex_flags & NFSEXP_PNFS)) |
9cf514cc | 133 | return; |
8650b8a0 | 134 | |
f99d4fbd | 135 | /* |
9b9960a0 TH |
136 | * If flex file is configured, use it by default. Otherwise |
137 | * check if the file system supports exporting a block-like layout. | |
f99d4fbd CH |
138 | * If the block device supports reservations prefer the SCSI layout, |
139 | * otherwise advertise the block layout. | |
140 | */ | |
9b9960a0 | 141 | #ifdef CONFIG_NFSD_FLEXFILELAYOUT |
8a4c3926 | 142 | exp->ex_layout_types |= 1 << LAYOUT_FLEX_FILES; |
9b9960a0 | 143 | #endif |
81c39329 | 144 | #ifdef CONFIG_NFSD_BLOCKLAYOUT |
9b9960a0 | 145 | /* overwrite flex file layout selection if needed */ |
8650b8a0 CH |
146 | if (sb->s_export_op->get_uuid && |
147 | sb->s_export_op->map_blocks && | |
148 | sb->s_export_op->commit_blocks) | |
8a4c3926 | 149 | exp->ex_layout_types |= 1 << LAYOUT_BLOCK_VOLUME; |
81c39329 | 150 | #endif |
f99d4fbd CH |
151 | #ifdef CONFIG_NFSD_SCSILAYOUT |
152 | /* overwrite block layout selection if needed */ | |
153 | if (sb->s_export_op->map_blocks && | |
154 | sb->s_export_op->commit_blocks && | |
155 | sb->s_bdev && sb->s_bdev->bd_disk->fops->pr_ops) | |
8a4c3926 | 156 | exp->ex_layout_types |= 1 << LAYOUT_SCSI; |
f99d4fbd | 157 | #endif |
9cf514cc CH |
158 | } |
159 | ||
160 | static void | |
161 | nfsd4_free_layout_stateid(struct nfs4_stid *stid) | |
162 | { | |
163 | struct nfs4_layout_stateid *ls = layoutstateid(stid); | |
164 | struct nfs4_client *clp = ls->ls_stid.sc_client; | |
165 | struct nfs4_file *fp = ls->ls_stid.sc_file; | |
166 | ||
31ef83dc CH |
167 | trace_layoutstate_free(&ls->ls_stid.sc_stateid); |
168 | ||
9cf514cc CH |
169 | spin_lock(&clp->cl_lock); |
170 | list_del_init(&ls->ls_perclnt); | |
171 | spin_unlock(&clp->cl_lock); | |
172 | ||
173 | spin_lock(&fp->fi_lock); | |
174 | list_del_init(&ls->ls_perfile); | |
175 | spin_unlock(&fp->fi_lock); | |
176 | ||
1983a66f JL |
177 | if (!nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls) |
178 | vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls); | |
c5c707f9 CH |
179 | fput(ls->ls_file); |
180 | ||
181 | if (ls->ls_recalled) | |
182 | atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls); | |
183 | ||
9cf514cc CH |
184 | kmem_cache_free(nfs4_layout_stateid_cache, ls); |
185 | } | |
186 | ||
c5c707f9 CH |
187 | static int |
188 | nfsd4_layout_setlease(struct nfs4_layout_stateid *ls) | |
189 | { | |
190 | struct file_lock *fl; | |
191 | int status; | |
192 | ||
1983a66f JL |
193 | if (nfsd4_layout_ops[ls->ls_layout_type]->disable_recalls) |
194 | return 0; | |
195 | ||
c5c707f9 CH |
196 | fl = locks_alloc_lock(); |
197 | if (!fl) | |
198 | return -ENOMEM; | |
199 | locks_init_lock(fl); | |
200 | fl->fl_lmops = &nfsd4_layouts_lm_ops; | |
201 | fl->fl_flags = FL_LAYOUT; | |
202 | fl->fl_type = F_RDLCK; | |
203 | fl->fl_end = OFFSET_MAX; | |
204 | fl->fl_owner = ls; | |
205 | fl->fl_pid = current->tgid; | |
206 | fl->fl_file = ls->ls_file; | |
207 | ||
208 | status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL); | |
209 | if (status) { | |
210 | locks_free_lock(fl); | |
211 | return status; | |
212 | } | |
213 | BUG_ON(fl != NULL); | |
214 | return 0; | |
215 | } | |
216 | ||
9cf514cc CH |
217 | static struct nfs4_layout_stateid * |
218 | nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate, | |
219 | struct nfs4_stid *parent, u32 layout_type) | |
220 | { | |
221 | struct nfs4_client *clp = cstate->clp; | |
222 | struct nfs4_file *fp = parent->sc_file; | |
223 | struct nfs4_layout_stateid *ls; | |
224 | struct nfs4_stid *stp; | |
225 | ||
d19fb70d KM |
226 | stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache, |
227 | nfsd4_free_layout_stateid); | |
9cf514cc CH |
228 | if (!stp) |
229 | return NULL; | |
d19fb70d | 230 | |
9cf514cc CH |
231 | get_nfs4_file(fp); |
232 | stp->sc_file = fp; | |
233 | ||
234 | ls = layoutstateid(stp); | |
235 | INIT_LIST_HEAD(&ls->ls_perclnt); | |
236 | INIT_LIST_HEAD(&ls->ls_perfile); | |
237 | spin_lock_init(&ls->ls_lock); | |
238 | INIT_LIST_HEAD(&ls->ls_layouts); | |
cc8a5532 | 239 | mutex_init(&ls->ls_mutex); |
9cf514cc | 240 | ls->ls_layout_type = layout_type; |
c5c707f9 CH |
241 | nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops, |
242 | NFSPROC4_CLNT_CB_LAYOUT); | |
243 | ||
244 | if (parent->sc_type == NFS4_DELEG_STID) | |
245 | ls->ls_file = get_file(fp->fi_deleg_file); | |
246 | else | |
247 | ls->ls_file = find_any_file(fp); | |
248 | BUG_ON(!ls->ls_file); | |
249 | ||
250 | if (nfsd4_layout_setlease(ls)) { | |
1ca4b88e | 251 | fput(ls->ls_file); |
c5c707f9 CH |
252 | put_nfs4_file(fp); |
253 | kmem_cache_free(nfs4_layout_stateid_cache, ls); | |
254 | return NULL; | |
255 | } | |
9cf514cc CH |
256 | |
257 | spin_lock(&clp->cl_lock); | |
258 | stp->sc_type = NFS4_LAYOUT_STID; | |
259 | list_add(&ls->ls_perclnt, &clp->cl_lo_states); | |
260 | spin_unlock(&clp->cl_lock); | |
261 | ||
262 | spin_lock(&fp->fi_lock); | |
263 | list_add(&ls->ls_perfile, &fp->fi_lo_states); | |
264 | spin_unlock(&fp->fi_lock); | |
265 | ||
31ef83dc | 266 | trace_layoutstate_alloc(&ls->ls_stid.sc_stateid); |
9cf514cc CH |
267 | return ls; |
268 | } | |
269 | ||
270 | __be32 | |
271 | nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp, | |
272 | struct nfsd4_compound_state *cstate, stateid_t *stateid, | |
273 | bool create, u32 layout_type, struct nfs4_layout_stateid **lsp) | |
274 | { | |
275 | struct nfs4_layout_stateid *ls; | |
276 | struct nfs4_stid *stid; | |
277 | unsigned char typemask = NFS4_LAYOUT_STID; | |
278 | __be32 status; | |
279 | ||
280 | if (create) | |
281 | typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID); | |
282 | ||
283 | status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid, | |
284 | net_generic(SVC_NET(rqstp), nfsd_net_id)); | |
285 | if (status) | |
286 | goto out; | |
287 | ||
288 | if (!fh_match(&cstate->current_fh.fh_handle, | |
289 | &stid->sc_file->fi_fhandle)) { | |
290 | status = nfserr_bad_stateid; | |
291 | goto out_put_stid; | |
292 | } | |
293 | ||
294 | if (stid->sc_type != NFS4_LAYOUT_STID) { | |
295 | ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type); | |
296 | nfs4_put_stid(stid); | |
297 | ||
298 | status = nfserr_jukebox; | |
299 | if (!ls) | |
300 | goto out; | |
cc8a5532 | 301 | mutex_lock(&ls->ls_mutex); |
9cf514cc CH |
302 | } else { |
303 | ls = container_of(stid, struct nfs4_layout_stateid, ls_stid); | |
304 | ||
305 | status = nfserr_bad_stateid; | |
cc8a5532 | 306 | mutex_lock(&ls->ls_mutex); |
14b7f4a1 | 307 | if (nfsd4_stateid_generation_after(stateid, &stid->sc_stateid)) |
cc8a5532 | 308 | goto out_unlock_stid; |
9cf514cc | 309 | if (layout_type != ls->ls_layout_type) |
cc8a5532 | 310 | goto out_unlock_stid; |
9cf514cc CH |
311 | } |
312 | ||
313 | *lsp = ls; | |
314 | return 0; | |
315 | ||
cc8a5532 JL |
316 | out_unlock_stid: |
317 | mutex_unlock(&ls->ls_mutex); | |
9cf514cc CH |
318 | out_put_stid: |
319 | nfs4_put_stid(stid); | |
320 | out: | |
321 | return status; | |
322 | } | |
323 | ||
c5c707f9 CH |
324 | static void |
325 | nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls) | |
326 | { | |
327 | spin_lock(&ls->ls_lock); | |
328 | if (ls->ls_recalled) | |
329 | goto out_unlock; | |
330 | ||
331 | ls->ls_recalled = true; | |
332 | atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls); | |
333 | if (list_empty(&ls->ls_layouts)) | |
334 | goto out_unlock; | |
335 | ||
31ef83dc CH |
336 | trace_layout_recall(&ls->ls_stid.sc_stateid); |
337 | ||
c5c707f9 | 338 | atomic_inc(&ls->ls_stid.sc_count); |
c5c707f9 CH |
339 | nfsd4_run_cb(&ls->ls_recall); |
340 | ||
341 | out_unlock: | |
342 | spin_unlock(&ls->ls_lock); | |
343 | } | |
344 | ||
9cf514cc CH |
345 | static inline u64 |
346 | layout_end(struct nfsd4_layout_seg *seg) | |
347 | { | |
348 | u64 end = seg->offset + seg->length; | |
349 | return end >= seg->offset ? end : NFS4_MAX_UINT64; | |
350 | } | |
351 | ||
352 | static void | |
353 | layout_update_len(struct nfsd4_layout_seg *lo, u64 end) | |
354 | { | |
355 | if (end == NFS4_MAX_UINT64) | |
356 | lo->length = NFS4_MAX_UINT64; | |
357 | else | |
358 | lo->length = end - lo->offset; | |
359 | } | |
360 | ||
361 | static bool | |
362 | layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s) | |
363 | { | |
364 | if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode) | |
365 | return false; | |
366 | if (layout_end(&lo->lo_seg) <= s->offset) | |
367 | return false; | |
368 | if (layout_end(s) <= lo->lo_seg.offset) | |
369 | return false; | |
370 | return true; | |
371 | } | |
372 | ||
373 | static bool | |
374 | layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new) | |
375 | { | |
376 | if (lo->iomode != new->iomode) | |
377 | return false; | |
378 | if (layout_end(new) < lo->offset) | |
379 | return false; | |
380 | if (layout_end(lo) < new->offset) | |
381 | return false; | |
382 | ||
383 | lo->offset = min(lo->offset, new->offset); | |
384 | layout_update_len(lo, max(layout_end(lo), layout_end(new))); | |
385 | return true; | |
386 | } | |
387 | ||
c5c707f9 CH |
388 | static __be32 |
389 | nfsd4_recall_conflict(struct nfs4_layout_stateid *ls) | |
390 | { | |
391 | struct nfs4_file *fp = ls->ls_stid.sc_file; | |
392 | struct nfs4_layout_stateid *l, *n; | |
393 | __be32 nfserr = nfs_ok; | |
394 | ||
395 | assert_spin_locked(&fp->fi_lock); | |
396 | ||
397 | list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) { | |
398 | if (l != ls) { | |
399 | nfsd4_recall_file_layout(l); | |
400 | nfserr = nfserr_recallconflict; | |
401 | } | |
402 | } | |
403 | ||
404 | return nfserr; | |
405 | } | |
406 | ||
9cf514cc CH |
407 | __be32 |
408 | nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls) | |
409 | { | |
410 | struct nfsd4_layout_seg *seg = &lgp->lg_seg; | |
c5c707f9 | 411 | struct nfs4_file *fp = ls->ls_stid.sc_file; |
9cf514cc | 412 | struct nfs4_layout *lp, *new = NULL; |
c5c707f9 | 413 | __be32 nfserr; |
9cf514cc | 414 | |
c5c707f9 CH |
415 | spin_lock(&fp->fi_lock); |
416 | nfserr = nfsd4_recall_conflict(ls); | |
417 | if (nfserr) | |
418 | goto out; | |
9cf514cc CH |
419 | spin_lock(&ls->ls_lock); |
420 | list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) { | |
421 | if (layouts_try_merge(&lp->lo_seg, seg)) | |
422 | goto done; | |
423 | } | |
424 | spin_unlock(&ls->ls_lock); | |
c5c707f9 | 425 | spin_unlock(&fp->fi_lock); |
9cf514cc CH |
426 | |
427 | new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL); | |
428 | if (!new) | |
429 | return nfserr_jukebox; | |
430 | memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg)); | |
431 | new->lo_state = ls; | |
432 | ||
c5c707f9 CH |
433 | spin_lock(&fp->fi_lock); |
434 | nfserr = nfsd4_recall_conflict(ls); | |
435 | if (nfserr) | |
436 | goto out; | |
9cf514cc CH |
437 | spin_lock(&ls->ls_lock); |
438 | list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) { | |
439 | if (layouts_try_merge(&lp->lo_seg, seg)) | |
440 | goto done; | |
441 | } | |
442 | ||
443 | atomic_inc(&ls->ls_stid.sc_count); | |
444 | list_add_tail(&new->lo_perstate, &ls->ls_layouts); | |
445 | new = NULL; | |
446 | done: | |
9767feb2 | 447 | nfs4_inc_and_copy_stateid(&lgp->lg_sid, &ls->ls_stid); |
9cf514cc | 448 | spin_unlock(&ls->ls_lock); |
c5c707f9 CH |
449 | out: |
450 | spin_unlock(&fp->fi_lock); | |
9cf514cc CH |
451 | if (new) |
452 | kmem_cache_free(nfs4_layout_cache, new); | |
c5c707f9 | 453 | return nfserr; |
9cf514cc CH |
454 | } |
455 | ||
456 | static void | |
457 | nfsd4_free_layouts(struct list_head *reaplist) | |
458 | { | |
459 | while (!list_empty(reaplist)) { | |
460 | struct nfs4_layout *lp = list_first_entry(reaplist, | |
461 | struct nfs4_layout, lo_perstate); | |
462 | ||
463 | list_del(&lp->lo_perstate); | |
464 | nfs4_put_stid(&lp->lo_state->ls_stid); | |
465 | kmem_cache_free(nfs4_layout_cache, lp); | |
466 | } | |
467 | } | |
468 | ||
469 | static void | |
470 | nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg, | |
471 | struct list_head *reaplist) | |
472 | { | |
473 | struct nfsd4_layout_seg *lo = &lp->lo_seg; | |
474 | u64 end = layout_end(lo); | |
475 | ||
476 | if (seg->offset <= lo->offset) { | |
477 | if (layout_end(seg) >= end) { | |
478 | list_move_tail(&lp->lo_perstate, reaplist); | |
479 | return; | |
480 | } | |
7890203d | 481 | lo->offset = layout_end(seg); |
9cf514cc CH |
482 | } else { |
483 | /* retain the whole layout segment on a split. */ | |
484 | if (layout_end(seg) < end) { | |
485 | dprintk("%s: split not supported\n", __func__); | |
486 | return; | |
487 | } | |
7890203d | 488 | end = seg->offset; |
9cf514cc CH |
489 | } |
490 | ||
491 | layout_update_len(lo, end); | |
492 | } | |
493 | ||
494 | __be32 | |
495 | nfsd4_return_file_layouts(struct svc_rqst *rqstp, | |
496 | struct nfsd4_compound_state *cstate, | |
497 | struct nfsd4_layoutreturn *lrp) | |
498 | { | |
499 | struct nfs4_layout_stateid *ls; | |
500 | struct nfs4_layout *lp, *n; | |
501 | LIST_HEAD(reaplist); | |
502 | __be32 nfserr; | |
503 | int found = 0; | |
504 | ||
505 | nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid, | |
506 | false, lrp->lr_layout_type, | |
507 | &ls); | |
31ef83dc CH |
508 | if (nfserr) { |
509 | trace_layout_return_lookup_fail(&lrp->lr_sid); | |
9cf514cc | 510 | return nfserr; |
31ef83dc | 511 | } |
9cf514cc CH |
512 | |
513 | spin_lock(&ls->ls_lock); | |
514 | list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) { | |
515 | if (layouts_overlapping(lp, &lrp->lr_seg)) { | |
516 | nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist); | |
517 | found++; | |
518 | } | |
519 | } | |
520 | if (!list_empty(&ls->ls_layouts)) { | |
9767feb2 JL |
521 | if (found) |
522 | nfs4_inc_and_copy_stateid(&lrp->lr_sid, &ls->ls_stid); | |
9cf514cc CH |
523 | lrp->lrs_present = 1; |
524 | } else { | |
31ef83dc | 525 | trace_layoutstate_unhash(&ls->ls_stid.sc_stateid); |
9cf514cc CH |
526 | nfs4_unhash_stid(&ls->ls_stid); |
527 | lrp->lrs_present = 0; | |
528 | } | |
529 | spin_unlock(&ls->ls_lock); | |
530 | ||
cc8a5532 | 531 | mutex_unlock(&ls->ls_mutex); |
9cf514cc CH |
532 | nfs4_put_stid(&ls->ls_stid); |
533 | nfsd4_free_layouts(&reaplist); | |
534 | return nfs_ok; | |
535 | } | |
536 | ||
537 | __be32 | |
538 | nfsd4_return_client_layouts(struct svc_rqst *rqstp, | |
539 | struct nfsd4_compound_state *cstate, | |
540 | struct nfsd4_layoutreturn *lrp) | |
541 | { | |
542 | struct nfs4_layout_stateid *ls, *n; | |
543 | struct nfs4_client *clp = cstate->clp; | |
544 | struct nfs4_layout *lp, *t; | |
545 | LIST_HEAD(reaplist); | |
546 | ||
547 | lrp->lrs_present = 0; | |
548 | ||
549 | spin_lock(&clp->cl_lock); | |
550 | list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) { | |
6f8f28ec KM |
551 | if (ls->ls_layout_type != lrp->lr_layout_type) |
552 | continue; | |
553 | ||
9cf514cc CH |
554 | if (lrp->lr_return_type == RETURN_FSID && |
555 | !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle, | |
556 | &cstate->current_fh.fh_handle)) | |
557 | continue; | |
558 | ||
559 | spin_lock(&ls->ls_lock); | |
560 | list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) { | |
561 | if (lrp->lr_seg.iomode == IOMODE_ANY || | |
562 | lrp->lr_seg.iomode == lp->lo_seg.iomode) | |
563 | list_move_tail(&lp->lo_perstate, &reaplist); | |
564 | } | |
565 | spin_unlock(&ls->ls_lock); | |
566 | } | |
567 | spin_unlock(&clp->cl_lock); | |
568 | ||
569 | nfsd4_free_layouts(&reaplist); | |
570 | return 0; | |
571 | } | |
572 | ||
573 | static void | |
574 | nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls, | |
575 | struct list_head *reaplist) | |
576 | { | |
577 | spin_lock(&ls->ls_lock); | |
578 | list_splice_init(&ls->ls_layouts, reaplist); | |
579 | spin_unlock(&ls->ls_lock); | |
580 | } | |
581 | ||
582 | void | |
583 | nfsd4_return_all_client_layouts(struct nfs4_client *clp) | |
584 | { | |
585 | struct nfs4_layout_stateid *ls, *n; | |
586 | LIST_HEAD(reaplist); | |
587 | ||
588 | spin_lock(&clp->cl_lock); | |
589 | list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) | |
590 | nfsd4_return_all_layouts(ls, &reaplist); | |
591 | spin_unlock(&clp->cl_lock); | |
592 | ||
593 | nfsd4_free_layouts(&reaplist); | |
594 | } | |
595 | ||
596 | void | |
597 | nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp) | |
598 | { | |
599 | struct nfs4_layout_stateid *ls, *n; | |
600 | LIST_HEAD(reaplist); | |
601 | ||
602 | spin_lock(&fp->fi_lock); | |
603 | list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) { | |
604 | if (ls->ls_stid.sc_client == clp) | |
605 | nfsd4_return_all_layouts(ls, &reaplist); | |
606 | } | |
607 | spin_unlock(&fp->fi_lock); | |
608 | ||
609 | nfsd4_free_layouts(&reaplist); | |
610 | } | |
611 | ||
c5c707f9 CH |
612 | static void |
613 | nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls) | |
614 | { | |
615 | struct nfs4_client *clp = ls->ls_stid.sc_client; | |
616 | char addr_str[INET6_ADDRSTRLEN]; | |
377e7a27 | 617 | static char const nfsd_recall_failed[] = "/sbin/nfsd-recall-failed"; |
c5c707f9 CH |
618 | static char *envp[] = { |
619 | "HOME=/", | |
620 | "TERM=linux", | |
621 | "PATH=/sbin:/usr/sbin:/bin:/usr/bin", | |
622 | NULL | |
623 | }; | |
624 | char *argv[8]; | |
625 | int error; | |
626 | ||
627 | rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str)); | |
628 | ||
629 | printk(KERN_WARNING | |
630 | "nfsd: client %s failed to respond to layout recall. " | |
631 | " Fencing..\n", addr_str); | |
632 | ||
377e7a27 | 633 | argv[0] = (char *)nfsd_recall_failed; |
c5c707f9 CH |
634 | argv[1] = addr_str; |
635 | argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id; | |
636 | argv[3] = NULL; | |
637 | ||
377e7a27 GKH |
638 | error = call_usermodehelper(nfsd_recall_failed, argv, envp, |
639 | UMH_WAIT_PROC); | |
c5c707f9 CH |
640 | if (error) { |
641 | printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n", | |
642 | addr_str, error); | |
643 | } | |
644 | } | |
645 | ||
cc8a5532 JL |
646 | static void |
647 | nfsd4_cb_layout_prepare(struct nfsd4_callback *cb) | |
648 | { | |
649 | struct nfs4_layout_stateid *ls = | |
650 | container_of(cb, struct nfs4_layout_stateid, ls_recall); | |
651 | ||
652 | mutex_lock(&ls->ls_mutex); | |
9767feb2 | 653 | nfs4_inc_and_copy_stateid(&ls->ls_recall_sid, &ls->ls_stid); |
be20aa00 | 654 | mutex_unlock(&ls->ls_mutex); |
cc8a5532 JL |
655 | } |
656 | ||
c5c707f9 CH |
657 | static int |
658 | nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task) | |
659 | { | |
660 | struct nfs4_layout_stateid *ls = | |
661 | container_of(cb, struct nfs4_layout_stateid, ls_recall); | |
6b9b2107 JL |
662 | struct nfsd_net *nn; |
663 | ktime_t now, cutoff; | |
f99d4fbd | 664 | const struct nfsd4_layout_ops *ops; |
c5c707f9 CH |
665 | LIST_HEAD(reaplist); |
666 | ||
6b9b2107 | 667 | |
c5c707f9 CH |
668 | switch (task->tk_status) { |
669 | case 0: | |
6b9b2107 JL |
670 | case -NFS4ERR_DELAY: |
671 | /* | |
672 | * Anything left? If not, then call it done. Note that we don't | |
673 | * take the spinlock since this is an optimization and nothing | |
674 | * should get added until the cb counter goes to zero. | |
675 | */ | |
676 | if (list_empty(&ls->ls_layouts)) | |
677 | return 1; | |
678 | ||
679 | /* Poll the client until it's done with the layout */ | |
680 | now = ktime_get(); | |
681 | nn = net_generic(ls->ls_stid.sc_client->net, nfsd_net_id); | |
682 | ||
683 | /* Client gets 2 lease periods to return it */ | |
684 | cutoff = ktime_add_ns(task->tk_start, | |
685 | nn->nfsd4_lease * NSEC_PER_SEC * 2); | |
686 | ||
687 | if (ktime_before(now, cutoff)) { | |
688 | rpc_delay(task, HZ/100); /* 10 mili-seconds */ | |
689 | return 0; | |
690 | } | |
691 | /* Fallthrough */ | |
c5c707f9 CH |
692 | default: |
693 | /* | |
694 | * Unknown error or non-responding client, we'll need to fence. | |
695 | */ | |
f99d4fbd CH |
696 | trace_layout_recall_fail(&ls->ls_stid.sc_stateid); |
697 | ||
698 | ops = nfsd4_layout_ops[ls->ls_layout_type]; | |
699 | if (ops->fence_client) | |
700 | ops->fence_client(ls); | |
701 | else | |
702 | nfsd4_cb_layout_fail(ls); | |
c5c707f9 | 703 | return -1; |
851238a2 JL |
704 | case -NFS4ERR_NOMATCHING_LAYOUT: |
705 | trace_layout_recall_done(&ls->ls_stid.sc_stateid); | |
706 | task->tk_status = 0; | |
707 | return 1; | |
c5c707f9 CH |
708 | } |
709 | } | |
710 | ||
711 | static void | |
712 | nfsd4_cb_layout_release(struct nfsd4_callback *cb) | |
713 | { | |
714 | struct nfs4_layout_stateid *ls = | |
715 | container_of(cb, struct nfs4_layout_stateid, ls_recall); | |
716 | LIST_HEAD(reaplist); | |
717 | ||
31ef83dc CH |
718 | trace_layout_recall_release(&ls->ls_stid.sc_stateid); |
719 | ||
c5c707f9 CH |
720 | nfsd4_return_all_layouts(ls, &reaplist); |
721 | nfsd4_free_layouts(&reaplist); | |
722 | nfs4_put_stid(&ls->ls_stid); | |
723 | } | |
724 | ||
c4cb8974 | 725 | static const struct nfsd4_callback_ops nfsd4_cb_layout_ops = { |
cc8a5532 | 726 | .prepare = nfsd4_cb_layout_prepare, |
c5c707f9 CH |
727 | .done = nfsd4_cb_layout_done, |
728 | .release = nfsd4_cb_layout_release, | |
729 | }; | |
730 | ||
731 | static bool | |
732 | nfsd4_layout_lm_break(struct file_lock *fl) | |
733 | { | |
734 | /* | |
735 | * We don't want the locks code to timeout the lease for us; | |
736 | * we'll remove it ourself if a layout isn't returned | |
737 | * in time: | |
738 | */ | |
739 | fl->fl_break_time = 0; | |
740 | nfsd4_recall_file_layout(fl->fl_owner); | |
741 | return false; | |
742 | } | |
743 | ||
744 | static int | |
745 | nfsd4_layout_lm_change(struct file_lock *onlist, int arg, | |
746 | struct list_head *dispose) | |
747 | { | |
748 | BUG_ON(!(arg & F_UNLCK)); | |
749 | return lease_modify(onlist, arg, dispose); | |
750 | } | |
751 | ||
752 | static const struct lock_manager_operations nfsd4_layouts_lm_ops = { | |
753 | .lm_break = nfsd4_layout_lm_break, | |
754 | .lm_change = nfsd4_layout_lm_change, | |
755 | }; | |
756 | ||
9cf514cc CH |
757 | int |
758 | nfsd4_init_pnfs(void) | |
759 | { | |
760 | int i; | |
761 | ||
762 | for (i = 0; i < DEVID_HASH_SIZE; i++) | |
763 | INIT_LIST_HEAD(&nfsd_devid_hash[i]); | |
764 | ||
765 | nfs4_layout_cache = kmem_cache_create("nfs4_layout", | |
766 | sizeof(struct nfs4_layout), 0, 0, NULL); | |
767 | if (!nfs4_layout_cache) | |
768 | return -ENOMEM; | |
769 | ||
770 | nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid", | |
771 | sizeof(struct nfs4_layout_stateid), 0, 0, NULL); | |
772 | if (!nfs4_layout_stateid_cache) { | |
773 | kmem_cache_destroy(nfs4_layout_cache); | |
774 | return -ENOMEM; | |
775 | } | |
776 | return 0; | |
777 | } | |
778 | ||
779 | void | |
780 | nfsd4_exit_pnfs(void) | |
781 | { | |
782 | int i; | |
783 | ||
784 | kmem_cache_destroy(nfs4_layout_cache); | |
785 | kmem_cache_destroy(nfs4_layout_stateid_cache); | |
786 | ||
787 | for (i = 0; i < DEVID_HASH_SIZE; i++) { | |
788 | struct nfsd4_deviceid_map *map, *n; | |
789 | ||
790 | list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash) | |
791 | kfree(map); | |
792 | } | |
793 | } |