2 * Copyright (c) 2014 Christoph Hellwig.
4 #include <linux/kmod.h>
5 #include <linux/file.h>
6 #include <linux/jhash.h>
7 #include <linux/sched.h>
8 #include <linux/sunrpc/addr.h>
14 #define NFSDDBG_FACILITY NFSDDBG_PNFS
17 struct list_head lo_perstate;
18 struct nfs4_layout_stateid *lo_state;
19 struct nfsd4_layout_seg lo_seg;
22 static struct kmem_cache *nfs4_layout_cache;
23 static struct kmem_cache *nfs4_layout_stateid_cache;
25 static struct nfsd4_callback_ops nfsd4_cb_layout_ops;
26 static const struct lock_manager_operations nfsd4_layouts_lm_ops;
28 const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] = {
29 [LAYOUT_BLOCK_VOLUME] = &bl_layout_ops,
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);
40 static inline u32 devid_hashfn(u64 idx)
42 return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK;
46 nfsd4_alloc_devid_map(const struct svc_fh *fhp)
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;
53 map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL);
57 map->fsid_type = fh->fh_fsid_type;
58 memcpy(&map->fsid, fh->fh_fsid, fsid_len);
60 spin_lock(&nfsd_devid_lock);
61 if (fhp->fh_export->ex_devid_map)
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)
68 if (memcmp(old->fsid, fh->fh_fsid,
69 key_len(old->fsid_type)))
72 fhp->fh_export->ex_devid_map = old;
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;
83 spin_unlock(&nfsd_devid_lock);
87 struct nfsd4_deviceid_map *
88 nfsd4_find_devid_map(int idx)
90 struct nfsd4_deviceid_map *map, *ret = NULL;
93 list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash)
102 nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp,
103 u32 device_generation)
105 if (!fhp->fh_export->ex_devid_map) {
106 nfsd4_alloc_devid_map(fhp);
107 if (!fhp->fh_export->ex_devid_map)
111 id->fsid_idx = fhp->fh_export->ex_devid_map->idx;
112 id->generation = device_generation;
117 void nfsd4_setup_layout_type(struct svc_export *exp)
119 struct super_block *sb = exp->ex_path.mnt->mnt_sb;
121 if (exp->ex_flags & NFSEXP_NOPNFS)
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;
131 nfsd4_free_layout_stateid(struct nfs4_stid *stid)
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;
137 trace_layoutstate_free(&ls->ls_stid.sc_stateid);
139 spin_lock(&clp->cl_lock);
140 list_del_init(&ls->ls_perclnt);
141 spin_unlock(&clp->cl_lock);
143 spin_lock(&fp->fi_lock);
144 list_del_init(&ls->ls_perfile);
145 spin_unlock(&fp->fi_lock);
147 vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls);
151 atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
153 kmem_cache_free(nfs4_layout_stateid_cache, ls);
157 nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
159 struct file_lock *fl;
162 fl = locks_alloc_lock();
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;
171 fl->fl_pid = current->tgid;
172 fl->fl_file = ls->ls_file;
174 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL);
183 static struct nfs4_layout_stateid *
184 nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
185 struct nfs4_stid *parent, u32 layout_type)
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;
192 stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache);
195 stp->sc_free = nfsd4_free_layout_stateid;
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;
205 nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
206 NFSPROC4_CLNT_CB_LAYOUT);
208 if (parent->sc_type == NFS4_DELEG_STID)
209 ls->ls_file = get_file(fp->fi_deleg_file);
211 ls->ls_file = find_any_file(fp);
212 BUG_ON(!ls->ls_file);
214 if (nfsd4_layout_setlease(ls)) {
216 kmem_cache_free(nfs4_layout_stateid_cache, ls);
220 spin_lock(&clp->cl_lock);
221 stp->sc_type = NFS4_LAYOUT_STID;
222 list_add(&ls->ls_perclnt, &clp->cl_lo_states);
223 spin_unlock(&clp->cl_lock);
225 spin_lock(&fp->fi_lock);
226 list_add(&ls->ls_perfile, &fp->fi_lo_states);
227 spin_unlock(&fp->fi_lock);
229 trace_layoutstate_alloc(&ls->ls_stid.sc_stateid);
234 nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
235 struct nfsd4_compound_state *cstate, stateid_t *stateid,
236 bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
238 struct nfs4_layout_stateid *ls;
239 struct nfs4_stid *stid;
240 unsigned char typemask = NFS4_LAYOUT_STID;
244 typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID);
246 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid,
247 net_generic(SVC_NET(rqstp), nfsd_net_id));
251 if (!fh_match(&cstate->current_fh.fh_handle,
252 &stid->sc_file->fi_fhandle)) {
253 status = nfserr_bad_stateid;
257 if (stid->sc_type != NFS4_LAYOUT_STID) {
258 ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
261 status = nfserr_jukebox;
265 ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
267 status = nfserr_bad_stateid;
268 if (stateid->si_generation > stid->sc_stateid.si_generation)
270 if (layout_type != ls->ls_layout_type)
284 nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
286 spin_lock(&ls->ls_lock);
290 ls->ls_recalled = true;
291 atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
292 if (list_empty(&ls->ls_layouts))
295 trace_layout_recall(&ls->ls_stid.sc_stateid);
297 atomic_inc(&ls->ls_stid.sc_count);
298 update_stateid(&ls->ls_stid.sc_stateid);
299 memcpy(&ls->ls_recall_sid, &ls->ls_stid.sc_stateid, sizeof(stateid_t));
300 nfsd4_run_cb(&ls->ls_recall);
303 spin_unlock(&ls->ls_lock);
307 layout_end(struct nfsd4_layout_seg *seg)
309 u64 end = seg->offset + seg->length;
310 return end >= seg->offset ? end : NFS4_MAX_UINT64;
314 layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
316 if (end == NFS4_MAX_UINT64)
317 lo->length = NFS4_MAX_UINT64;
319 lo->length = end - lo->offset;
323 layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
325 if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
327 if (layout_end(&lo->lo_seg) <= s->offset)
329 if (layout_end(s) <= lo->lo_seg.offset)
335 layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
337 if (lo->iomode != new->iomode)
339 if (layout_end(new) < lo->offset)
341 if (layout_end(lo) < new->offset)
344 lo->offset = min(lo->offset, new->offset);
345 layout_update_len(lo, max(layout_end(lo), layout_end(new)));
350 nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
352 struct nfs4_file *fp = ls->ls_stid.sc_file;
353 struct nfs4_layout_stateid *l, *n;
354 __be32 nfserr = nfs_ok;
356 assert_spin_locked(&fp->fi_lock);
358 list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
360 nfsd4_recall_file_layout(l);
361 nfserr = nfserr_recallconflict;
369 nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
371 struct nfsd4_layout_seg *seg = &lgp->lg_seg;
372 struct nfs4_file *fp = ls->ls_stid.sc_file;
373 struct nfs4_layout *lp, *new = NULL;
376 spin_lock(&fp->fi_lock);
377 nfserr = nfsd4_recall_conflict(ls);
380 spin_lock(&ls->ls_lock);
381 list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
382 if (layouts_try_merge(&lp->lo_seg, seg))
385 spin_unlock(&ls->ls_lock);
386 spin_unlock(&fp->fi_lock);
388 new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
390 return nfserr_jukebox;
391 memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg));
394 spin_lock(&fp->fi_lock);
395 nfserr = nfsd4_recall_conflict(ls);
398 spin_lock(&ls->ls_lock);
399 list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
400 if (layouts_try_merge(&lp->lo_seg, seg))
404 atomic_inc(&ls->ls_stid.sc_count);
405 list_add_tail(&new->lo_perstate, &ls->ls_layouts);
408 update_stateid(&ls->ls_stid.sc_stateid);
409 memcpy(&lgp->lg_sid, &ls->ls_stid.sc_stateid, sizeof(stateid_t));
410 spin_unlock(&ls->ls_lock);
412 spin_unlock(&fp->fi_lock);
414 kmem_cache_free(nfs4_layout_cache, new);
419 nfsd4_free_layouts(struct list_head *reaplist)
421 while (!list_empty(reaplist)) {
422 struct nfs4_layout *lp = list_first_entry(reaplist,
423 struct nfs4_layout, lo_perstate);
425 list_del(&lp->lo_perstate);
426 nfs4_put_stid(&lp->lo_state->ls_stid);
427 kmem_cache_free(nfs4_layout_cache, lp);
432 nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
433 struct list_head *reaplist)
435 struct nfsd4_layout_seg *lo = &lp->lo_seg;
436 u64 end = layout_end(lo);
438 if (seg->offset <= lo->offset) {
439 if (layout_end(seg) >= end) {
440 list_move_tail(&lp->lo_perstate, reaplist);
445 /* retain the whole layout segment on a split. */
446 if (layout_end(seg) < end) {
447 dprintk("%s: split not supported\n", __func__);
451 lo->offset = layout_end(seg);
454 layout_update_len(lo, end);
458 nfsd4_return_file_layouts(struct svc_rqst *rqstp,
459 struct nfsd4_compound_state *cstate,
460 struct nfsd4_layoutreturn *lrp)
462 struct nfs4_layout_stateid *ls;
463 struct nfs4_layout *lp, *n;
468 nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
469 false, lrp->lr_layout_type,
472 trace_layout_return_lookup_fail(&lrp->lr_sid);
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);
483 if (!list_empty(&ls->ls_layouts)) {
485 update_stateid(&ls->ls_stid.sc_stateid);
486 memcpy(&lrp->lr_sid, &ls->ls_stid.sc_stateid,
489 lrp->lrs_present = 1;
491 trace_layoutstate_unhash(&ls->ls_stid.sc_stateid);
492 nfs4_unhash_stid(&ls->ls_stid);
493 lrp->lrs_present = 0;
495 spin_unlock(&ls->ls_lock);
497 nfs4_put_stid(&ls->ls_stid);
498 nfsd4_free_layouts(&reaplist);
503 nfsd4_return_client_layouts(struct svc_rqst *rqstp,
504 struct nfsd4_compound_state *cstate,
505 struct nfsd4_layoutreturn *lrp)
507 struct nfs4_layout_stateid *ls, *n;
508 struct nfs4_client *clp = cstate->clp;
509 struct nfs4_layout *lp, *t;
512 lrp->lrs_present = 0;
514 spin_lock(&clp->cl_lock);
515 list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
516 if (lrp->lr_return_type == RETURN_FSID &&
517 !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
518 &cstate->current_fh.fh_handle))
521 spin_lock(&ls->ls_lock);
522 list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
523 if (lrp->lr_seg.iomode == IOMODE_ANY ||
524 lrp->lr_seg.iomode == lp->lo_seg.iomode)
525 list_move_tail(&lp->lo_perstate, &reaplist);
527 spin_unlock(&ls->ls_lock);
529 spin_unlock(&clp->cl_lock);
531 nfsd4_free_layouts(&reaplist);
536 nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
537 struct list_head *reaplist)
539 spin_lock(&ls->ls_lock);
540 list_splice_init(&ls->ls_layouts, reaplist);
541 spin_unlock(&ls->ls_lock);
545 nfsd4_return_all_client_layouts(struct nfs4_client *clp)
547 struct nfs4_layout_stateid *ls, *n;
550 spin_lock(&clp->cl_lock);
551 list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
552 nfsd4_return_all_layouts(ls, &reaplist);
553 spin_unlock(&clp->cl_lock);
555 nfsd4_free_layouts(&reaplist);
559 nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
561 struct nfs4_layout_stateid *ls, *n;
564 spin_lock(&fp->fi_lock);
565 list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
566 if (ls->ls_stid.sc_client == clp)
567 nfsd4_return_all_layouts(ls, &reaplist);
569 spin_unlock(&fp->fi_lock);
571 nfsd4_free_layouts(&reaplist);
575 nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls)
577 struct nfs4_client *clp = ls->ls_stid.sc_client;
578 char addr_str[INET6_ADDRSTRLEN];
579 static char *envp[] = {
582 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
588 rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
591 "nfsd: client %s failed to respond to layout recall. "
592 " Fencing..\n", addr_str);
594 argv[0] = "/sbin/nfsd-recall-failed";
596 argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id;
599 error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
601 printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
607 nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
609 struct nfs4_layout_stateid *ls =
610 container_of(cb, struct nfs4_layout_stateid, ls_recall);
613 switch (task->tk_status) {
616 case -NFS4ERR_NOMATCHING_LAYOUT:
617 trace_layout_recall_done(&ls->ls_stid.sc_stateid);
621 /* Poll the client until it's done with the layout */
622 /* FIXME: cap number of retries.
623 * The pnfs standard states that we need to only expire
624 * the client after at-least "lease time" .eg lease-time * 2
625 * when failing to communicate a recall
627 rpc_delay(task, HZ/100); /* 10 mili-seconds */
631 * Unknown error or non-responding client, we'll need to fence.
633 nfsd4_cb_layout_fail(ls);
639 nfsd4_cb_layout_release(struct nfsd4_callback *cb)
641 struct nfs4_layout_stateid *ls =
642 container_of(cb, struct nfs4_layout_stateid, ls_recall);
645 trace_layout_recall_release(&ls->ls_stid.sc_stateid);
647 nfsd4_return_all_layouts(ls, &reaplist);
648 nfsd4_free_layouts(&reaplist);
649 nfs4_put_stid(&ls->ls_stid);
652 static struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
653 .done = nfsd4_cb_layout_done,
654 .release = nfsd4_cb_layout_release,
658 nfsd4_layout_lm_break(struct file_lock *fl)
661 * We don't want the locks code to timeout the lease for us;
662 * we'll remove it ourself if a layout isn't returned
665 fl->fl_break_time = 0;
666 nfsd4_recall_file_layout(fl->fl_owner);
671 nfsd4_layout_lm_change(struct file_lock *onlist, int arg,
672 struct list_head *dispose)
674 BUG_ON(!(arg & F_UNLCK));
675 return lease_modify(onlist, arg, dispose);
678 static const struct lock_manager_operations nfsd4_layouts_lm_ops = {
679 .lm_break = nfsd4_layout_lm_break,
680 .lm_change = nfsd4_layout_lm_change,
684 nfsd4_init_pnfs(void)
688 for (i = 0; i < DEVID_HASH_SIZE; i++)
689 INIT_LIST_HEAD(&nfsd_devid_hash[i]);
691 nfs4_layout_cache = kmem_cache_create("nfs4_layout",
692 sizeof(struct nfs4_layout), 0, 0, NULL);
693 if (!nfs4_layout_cache)
696 nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid",
697 sizeof(struct nfs4_layout_stateid), 0, 0, NULL);
698 if (!nfs4_layout_stateid_cache) {
699 kmem_cache_destroy(nfs4_layout_cache);
706 nfsd4_exit_pnfs(void)
710 kmem_cache_destroy(nfs4_layout_cache);
711 kmem_cache_destroy(nfs4_layout_stateid_cache);
713 for (i = 0; i < DEVID_HASH_SIZE; i++) {
714 struct nfsd4_deviceid_map *map, *n;
716 list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)