Loading...
1/*
2 * linux/fs/nfs/callback_proc.c
3 *
4 * Copyright (C) 2004 Trond Myklebust
5 *
6 * NFSv4 callback procedures
7 */
8#include <linux/nfs4.h>
9#include <linux/nfs_fs.h>
10#include <linux/slab.h>
11#include <linux/rcupdate.h>
12#include "nfs4_fs.h"
13#include "callback.h"
14#include "delegation.h"
15#include "internal.h"
16#include "pnfs.h"
17#include "nfs4session.h"
18#include "nfs4trace.h"
19
20#define NFSDBG_FACILITY NFSDBG_CALLBACK
21
22__be32 nfs4_callback_getattr(struct cb_getattrargs *args,
23 struct cb_getattrres *res,
24 struct cb_process_state *cps)
25{
26 struct nfs_delegation *delegation;
27 struct nfs_inode *nfsi;
28 struct inode *inode;
29
30 res->status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
31 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
32 goto out;
33
34 res->bitmap[0] = res->bitmap[1] = 0;
35 res->status = htonl(NFS4ERR_BADHANDLE);
36
37 dprintk_rcu("NFS: GETATTR callback request from %s\n",
38 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
39
40 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
41 if (inode == NULL) {
42 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
43 -ntohl(res->status));
44 goto out;
45 }
46 nfsi = NFS_I(inode);
47 rcu_read_lock();
48 delegation = rcu_dereference(nfsi->delegation);
49 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
50 goto out_iput;
51 res->size = i_size_read(inode);
52 res->change_attr = delegation->change_attr;
53 if (nfsi->nrequests != 0)
54 res->change_attr++;
55 res->ctime = inode->i_ctime;
56 res->mtime = inode->i_mtime;
57 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
58 args->bitmap[0];
59 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
60 args->bitmap[1];
61 res->status = 0;
62out_iput:
63 rcu_read_unlock();
64 trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
65 iput(inode);
66out:
67 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
68 return res->status;
69}
70
71__be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
72 struct cb_process_state *cps)
73{
74 struct inode *inode;
75 __be32 res;
76
77 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
78 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
79 goto out;
80
81 dprintk_rcu("NFS: RECALL callback request from %s\n",
82 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
83
84 res = htonl(NFS4ERR_BADHANDLE);
85 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
86 if (inode == NULL) {
87 trace_nfs4_cb_recall(cps->clp, &args->fh, NULL,
88 &args->stateid, -ntohl(res));
89 goto out;
90 }
91 /* Set up a helper thread to actually return the delegation */
92 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
93 case 0:
94 res = 0;
95 break;
96 case -ENOENT:
97 res = htonl(NFS4ERR_BAD_STATEID);
98 break;
99 default:
100 res = htonl(NFS4ERR_RESOURCE);
101 }
102 trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
103 &args->stateid, -ntohl(res));
104 iput(inode);
105out:
106 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
107 return res;
108}
109
110#if defined(CONFIG_NFS_V4_1)
111
112/*
113 * Lookup a layout inode by stateid
114 *
115 * Note: returns a refcount on the inode and superblock
116 */
117static struct inode *nfs_layout_find_inode_by_stateid(struct nfs_client *clp,
118 const nfs4_stateid *stateid)
119{
120 struct nfs_server *server;
121 struct inode *inode;
122 struct pnfs_layout_hdr *lo;
123
124restart:
125 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
126 list_for_each_entry(lo, &server->layouts, plh_layouts) {
127 if (stateid != NULL &&
128 !nfs4_stateid_match_other(stateid, &lo->plh_stateid))
129 continue;
130 inode = igrab(lo->plh_inode);
131 if (!inode)
132 continue;
133 if (!nfs_sb_active(inode->i_sb)) {
134 rcu_read_lock();
135 spin_unlock(&clp->cl_lock);
136 iput(inode);
137 spin_lock(&clp->cl_lock);
138 goto restart;
139 }
140 return inode;
141 }
142 }
143
144 return NULL;
145}
146
147/*
148 * Lookup a layout inode by filehandle.
149 *
150 * Note: returns a refcount on the inode and superblock
151 *
152 */
153static struct inode *nfs_layout_find_inode_by_fh(struct nfs_client *clp,
154 const struct nfs_fh *fh)
155{
156 struct nfs_server *server;
157 struct nfs_inode *nfsi;
158 struct inode *inode;
159 struct pnfs_layout_hdr *lo;
160
161restart:
162 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
163 list_for_each_entry(lo, &server->layouts, plh_layouts) {
164 nfsi = NFS_I(lo->plh_inode);
165 if (nfs_compare_fh(fh, &nfsi->fh))
166 continue;
167 if (nfsi->layout != lo)
168 continue;
169 inode = igrab(lo->plh_inode);
170 if (!inode)
171 continue;
172 if (!nfs_sb_active(inode->i_sb)) {
173 rcu_read_lock();
174 spin_unlock(&clp->cl_lock);
175 iput(inode);
176 spin_lock(&clp->cl_lock);
177 goto restart;
178 }
179 return inode;
180 }
181 }
182
183 return NULL;
184}
185
186static struct inode *nfs_layout_find_inode(struct nfs_client *clp,
187 const struct nfs_fh *fh,
188 const nfs4_stateid *stateid)
189{
190 struct inode *inode;
191
192 spin_lock(&clp->cl_lock);
193 rcu_read_lock();
194 inode = nfs_layout_find_inode_by_stateid(clp, stateid);
195 if (!inode)
196 inode = nfs_layout_find_inode_by_fh(clp, fh);
197 rcu_read_unlock();
198 spin_unlock(&clp->cl_lock);
199
200 return inode;
201}
202
203/*
204 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
205 */
206static u32 pnfs_check_callback_stateid(struct pnfs_layout_hdr *lo,
207 const nfs4_stateid *new)
208{
209 u32 oldseq, newseq;
210
211 /* Is the stateid still not initialised? */
212 if (!pnfs_layout_is_valid(lo))
213 return NFS4ERR_DELAY;
214
215 /* Mismatched stateid? */
216 if (!nfs4_stateid_match_other(&lo->plh_stateid, new))
217 return NFS4ERR_BAD_STATEID;
218
219 newseq = be32_to_cpu(new->seqid);
220 /* Are we already in a layout recall situation? */
221 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
222 lo->plh_return_seq != 0) {
223 if (newseq < lo->plh_return_seq)
224 return NFS4ERR_OLD_STATEID;
225 if (newseq > lo->plh_return_seq)
226 return NFS4ERR_DELAY;
227 goto out;
228 }
229
230 /* Check that the stateid matches what we think it should be. */
231 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
232 if (newseq > oldseq + 1)
233 return NFS4ERR_DELAY;
234 /* Crazy server! */
235 if (newseq <= oldseq)
236 return NFS4ERR_OLD_STATEID;
237out:
238 return NFS_OK;
239}
240
241static u32 initiate_file_draining(struct nfs_client *clp,
242 struct cb_layoutrecallargs *args)
243{
244 struct inode *ino;
245 struct pnfs_layout_hdr *lo;
246 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
247 LIST_HEAD(free_me_list);
248
249 ino = nfs_layout_find_inode(clp, &args->cbl_fh, &args->cbl_stateid);
250 if (!ino)
251 goto out;
252
253 pnfs_layoutcommit_inode(ino, false);
254
255
256 spin_lock(&ino->i_lock);
257 lo = NFS_I(ino)->layout;
258 if (!lo) {
259 spin_unlock(&ino->i_lock);
260 goto out;
261 }
262 pnfs_get_layout_hdr(lo);
263 rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid);
264 if (rv != NFS_OK)
265 goto unlock;
266 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
267
268 /*
269 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
270 */
271 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
272 rv = NFS4ERR_DELAY;
273 goto unlock;
274 }
275
276 if (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
277 &args->cbl_range,
278 be32_to_cpu(args->cbl_stateid.seqid))) {
279 rv = NFS4_OK;
280 goto unlock;
281 }
282
283 /* Embrace your forgetfulness! */
284 rv = NFS4ERR_NOMATCHING_LAYOUT;
285
286 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
287 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
288 &args->cbl_range);
289 }
290unlock:
291 spin_unlock(&ino->i_lock);
292 pnfs_free_lseg_list(&free_me_list);
293 /* Free all lsegs that are attached to commit buckets */
294 nfs_commit_inode(ino, 0);
295 pnfs_put_layout_hdr(lo);
296out:
297 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
298 &args->cbl_stateid, -rv);
299 nfs_iput_and_deactive(ino);
300 return rv;
301}
302
303static u32 initiate_bulk_draining(struct nfs_client *clp,
304 struct cb_layoutrecallargs *args)
305{
306 int stat;
307
308 if (args->cbl_recall_type == RETURN_FSID)
309 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
310 else
311 stat = pnfs_destroy_layouts_byclid(clp, true);
312 if (stat != 0)
313 return NFS4ERR_DELAY;
314 return NFS4ERR_NOMATCHING_LAYOUT;
315}
316
317static u32 do_callback_layoutrecall(struct nfs_client *clp,
318 struct cb_layoutrecallargs *args)
319{
320 u32 res;
321
322 dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
323 if (args->cbl_recall_type == RETURN_FILE)
324 res = initiate_file_draining(clp, args);
325 else
326 res = initiate_bulk_draining(clp, args);
327 dprintk("%s returning %i\n", __func__, res);
328 return res;
329
330}
331
332__be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
333 void *dummy, struct cb_process_state *cps)
334{
335 u32 res;
336
337 dprintk("%s: -->\n", __func__);
338
339 if (cps->clp)
340 res = do_callback_layoutrecall(cps->clp, args);
341 else
342 res = NFS4ERR_OP_NOT_IN_SESSION;
343
344 dprintk("%s: exit with status = %d\n", __func__, res);
345 return cpu_to_be32(res);
346}
347
348static void pnfs_recall_all_layouts(struct nfs_client *clp)
349{
350 struct cb_layoutrecallargs args;
351
352 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
353 memset(&args, 0, sizeof(args));
354 args.cbl_recall_type = RETURN_ALL;
355 /* FIXME we ignore errors, what should we do? */
356 do_callback_layoutrecall(clp, &args);
357}
358
359__be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
360 void *dummy, struct cb_process_state *cps)
361{
362 int i;
363 __be32 res = 0;
364 struct nfs_client *clp = cps->clp;
365 struct nfs_server *server = NULL;
366
367 dprintk("%s: -->\n", __func__);
368
369 if (!clp) {
370 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
371 goto out;
372 }
373
374 for (i = 0; i < args->ndevs; i++) {
375 struct cb_devicenotifyitem *dev = &args->devs[i];
376
377 if (!server ||
378 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
379 rcu_read_lock();
380 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
381 if (server->pnfs_curr_ld &&
382 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
383 rcu_read_unlock();
384 goto found;
385 }
386 rcu_read_unlock();
387 dprintk("%s: layout type %u not found\n",
388 __func__, dev->cbd_layout_type);
389 continue;
390 }
391
392 found:
393 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
394 }
395
396out:
397 kfree(args->devs);
398 dprintk("%s: exit with status = %u\n",
399 __func__, be32_to_cpu(res));
400 return res;
401}
402
403/*
404 * Validate the sequenceID sent by the server.
405 * Return success if the sequenceID is one more than what we last saw on
406 * this slot, accounting for wraparound. Increments the slot's sequence.
407 *
408 * We don't yet implement a duplicate request cache, instead we set the
409 * back channel ca_maxresponsesize_cached to zero. This is OK for now
410 * since we only currently implement idempotent callbacks anyway.
411 *
412 * We have a single slot backchannel at this time, so we don't bother
413 * checking the used_slots bit array on the table. The lower layer guarantees
414 * a single outstanding callback request at a time.
415 */
416static __be32
417validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
418 const struct cb_sequenceargs * args)
419{
420 dprintk("%s enter. slotid %u seqid %u, slot table seqid: %u\n",
421 __func__, args->csa_slotid, args->csa_sequenceid, slot->seq_nr);
422
423 if (args->csa_slotid > tbl->server_highest_slotid)
424 return htonl(NFS4ERR_BADSLOT);
425
426 /* Replay */
427 if (args->csa_sequenceid == slot->seq_nr) {
428 dprintk("%s seqid %u is a replay\n",
429 __func__, args->csa_sequenceid);
430 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
431 return htonl(NFS4ERR_DELAY);
432 /* Signal process_op to set this error on next op */
433 if (args->csa_cachethis == 0)
434 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
435
436 /* Liar! We never allowed you to set csa_cachethis != 0 */
437 return htonl(NFS4ERR_SEQ_FALSE_RETRY);
438 }
439
440 /* Wraparound */
441 if (unlikely(slot->seq_nr == 0xFFFFFFFFU)) {
442 if (args->csa_sequenceid == 1)
443 return htonl(NFS4_OK);
444 } else if (likely(args->csa_sequenceid == slot->seq_nr + 1))
445 return htonl(NFS4_OK);
446
447 /* Misordered request */
448 return htonl(NFS4ERR_SEQ_MISORDERED);
449}
450
451/*
452 * For each referring call triple, check the session's slot table for
453 * a match. If the slot is in use and the sequence numbers match, the
454 * client is still waiting for a response to the original request.
455 */
456static bool referring_call_exists(struct nfs_client *clp,
457 uint32_t nrclists,
458 struct referring_call_list *rclists)
459{
460 bool status = 0;
461 int i, j;
462 struct nfs4_session *session;
463 struct nfs4_slot_table *tbl;
464 struct referring_call_list *rclist;
465 struct referring_call *ref;
466
467 /*
468 * XXX When client trunking is implemented, this becomes
469 * a session lookup from within the loop
470 */
471 session = clp->cl_session;
472 tbl = &session->fc_slot_table;
473
474 for (i = 0; i < nrclists; i++) {
475 rclist = &rclists[i];
476 if (memcmp(session->sess_id.data,
477 rclist->rcl_sessionid.data,
478 NFS4_MAX_SESSIONID_LEN) != 0)
479 continue;
480
481 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
482 ref = &rclist->rcl_refcalls[j];
483
484 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
485 "slotid %u\n", __func__,
486 ((u32 *)&rclist->rcl_sessionid.data)[0],
487 ((u32 *)&rclist->rcl_sessionid.data)[1],
488 ((u32 *)&rclist->rcl_sessionid.data)[2],
489 ((u32 *)&rclist->rcl_sessionid.data)[3],
490 ref->rc_sequenceid, ref->rc_slotid);
491
492 status = nfs4_slot_wait_on_seqid(tbl, ref->rc_slotid,
493 ref->rc_sequenceid, HZ >> 1) < 0;
494 if (status)
495 goto out;
496 }
497 }
498
499out:
500 return status;
501}
502
503__be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
504 struct cb_sequenceres *res,
505 struct cb_process_state *cps)
506{
507 struct nfs4_slot_table *tbl;
508 struct nfs4_slot *slot;
509 struct nfs_client *clp;
510 int i;
511 __be32 status = htonl(NFS4ERR_BADSESSION);
512
513 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
514 &args->csa_sessionid, cps->minorversion);
515 if (clp == NULL)
516 goto out;
517
518 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
519 goto out;
520
521 tbl = &clp->cl_session->bc_slot_table;
522
523 /* Set up res before grabbing the spinlock */
524 memcpy(&res->csr_sessionid, &args->csa_sessionid,
525 sizeof(res->csr_sessionid));
526 res->csr_sequenceid = args->csa_sequenceid;
527 res->csr_slotid = args->csa_slotid;
528
529 spin_lock(&tbl->slot_tbl_lock);
530 /* state manager is resetting the session */
531 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
532 status = htonl(NFS4ERR_DELAY);
533 /* Return NFS4ERR_BADSESSION if we're draining the session
534 * in order to reset it.
535 */
536 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
537 status = htonl(NFS4ERR_BADSESSION);
538 goto out_unlock;
539 }
540
541 status = htonl(NFS4ERR_BADSLOT);
542 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
543 if (IS_ERR(slot))
544 goto out_unlock;
545
546 res->csr_highestslotid = tbl->server_highest_slotid;
547 res->csr_target_highestslotid = tbl->target_highest_slotid;
548
549 status = validate_seqid(tbl, slot, args);
550 if (status)
551 goto out_unlock;
552 if (!nfs4_try_to_lock_slot(tbl, slot)) {
553 status = htonl(NFS4ERR_DELAY);
554 goto out_unlock;
555 }
556 cps->slot = slot;
557
558 /* The ca_maxresponsesize_cached is 0 with no DRC */
559 if (args->csa_cachethis != 0) {
560 status = htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
561 goto out_unlock;
562 }
563
564 /*
565 * Check for pending referring calls. If a match is found, a
566 * related callback was received before the response to the original
567 * call.
568 */
569 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
570 status = htonl(NFS4ERR_DELAY);
571 goto out_unlock;
572 }
573
574 /*
575 * RFC5661 20.9.3
576 * If CB_SEQUENCE returns an error, then the state of the slot
577 * (sequence ID, cached reply) MUST NOT change.
578 */
579 slot->seq_nr = args->csa_sequenceid;
580out_unlock:
581 spin_unlock(&tbl->slot_tbl_lock);
582
583out:
584 cps->clp = clp; /* put in nfs4_callback_compound */
585 for (i = 0; i < args->csa_nrclists; i++)
586 kfree(args->csa_rclists[i].rcl_refcalls);
587 kfree(args->csa_rclists);
588
589 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
590 cps->drc_status = status;
591 status = 0;
592 } else
593 res->csr_status = status;
594
595 trace_nfs4_cb_sequence(args, res, status);
596 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
597 ntohl(status), ntohl(res->csr_status));
598 return status;
599}
600
601static bool
602validate_bitmap_values(unsigned long mask)
603{
604 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
605}
606
607__be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
608 struct cb_process_state *cps)
609{
610 __be32 status;
611 fmode_t flags = 0;
612
613 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
614 if (!cps->clp) /* set in cb_sequence */
615 goto out;
616
617 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
618 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
619
620 status = cpu_to_be32(NFS4ERR_INVAL);
621 if (!validate_bitmap_values(args->craa_type_mask))
622 goto out;
623
624 status = cpu_to_be32(NFS4_OK);
625 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
626 &args->craa_type_mask))
627 flags = FMODE_READ;
628 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
629 &args->craa_type_mask))
630 flags |= FMODE_WRITE;
631 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
632 &args->craa_type_mask))
633 pnfs_recall_all_layouts(cps->clp);
634 if (flags)
635 nfs_expire_unused_delegation_types(cps->clp, flags);
636out:
637 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
638 return status;
639}
640
641/* Reduce the fore channel's max_slots to the target value */
642__be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
643 struct cb_process_state *cps)
644{
645 struct nfs4_slot_table *fc_tbl;
646 __be32 status;
647
648 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
649 if (!cps->clp) /* set in cb_sequence */
650 goto out;
651
652 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
653 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
654 args->crsa_target_highest_slotid);
655
656 fc_tbl = &cps->clp->cl_session->fc_slot_table;
657
658 status = htonl(NFS4_OK);
659
660 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
661 nfs41_notify_server(cps->clp);
662out:
663 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
664 return status;
665}
666
667__be32 nfs4_callback_notify_lock(struct cb_notify_lock_args *args, void *dummy,
668 struct cb_process_state *cps)
669{
670 if (!cps->clp) /* set in cb_sequence */
671 return htonl(NFS4ERR_OP_NOT_IN_SESSION);
672
673 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
674 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
675
676 /* Don't wake anybody if the string looked bogus */
677 if (args->cbnl_valid)
678 __wake_up(&cps->clp->cl_lock_waitq, TASK_NORMAL, 0, args);
679
680 return htonl(NFS4_OK);
681}
682#endif /* CONFIG_NFS_V4_1 */
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/nfs/callback_proc.c
4 *
5 * Copyright (C) 2004 Trond Myklebust
6 *
7 * NFSv4 callback procedures
8 */
9#include <linux/nfs4.h>
10#include <linux/nfs_fs.h>
11#include <linux/slab.h>
12#include <linux/rcupdate.h>
13#include "nfs4_fs.h"
14#include "callback.h"
15#include "delegation.h"
16#include "internal.h"
17#include "pnfs.h"
18#include "nfs4session.h"
19#include "nfs4trace.h"
20
21#define NFSDBG_FACILITY NFSDBG_CALLBACK
22
23__be32 nfs4_callback_getattr(void *argp, void *resp,
24 struct cb_process_state *cps)
25{
26 struct cb_getattrargs *args = argp;
27 struct cb_getattrres *res = resp;
28 struct nfs_delegation *delegation;
29 struct nfs_inode *nfsi;
30 struct inode *inode;
31
32 res->status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
33 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
34 goto out;
35
36 res->bitmap[0] = res->bitmap[1] = 0;
37 res->status = htonl(NFS4ERR_BADHANDLE);
38
39 dprintk_rcu("NFS: GETATTR callback request from %s\n",
40 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
41
42 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
43 if (inode == NULL) {
44 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
45 -ntohl(res->status));
46 goto out;
47 }
48 nfsi = NFS_I(inode);
49 rcu_read_lock();
50 delegation = rcu_dereference(nfsi->delegation);
51 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
52 goto out_iput;
53 res->size = i_size_read(inode);
54 res->change_attr = delegation->change_attr;
55 if (nfs_have_writebacks(inode))
56 res->change_attr++;
57 res->ctime = inode->i_ctime;
58 res->mtime = inode->i_mtime;
59 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
60 args->bitmap[0];
61 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
62 args->bitmap[1];
63 res->status = 0;
64out_iput:
65 rcu_read_unlock();
66 trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
67 iput(inode);
68out:
69 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
70 return res->status;
71}
72
73__be32 nfs4_callback_recall(void *argp, void *resp,
74 struct cb_process_state *cps)
75{
76 struct cb_recallargs *args = argp;
77 struct inode *inode;
78 __be32 res;
79
80 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
81 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
82 goto out;
83
84 dprintk_rcu("NFS: RECALL callback request from %s\n",
85 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
86
87 res = htonl(NFS4ERR_BADHANDLE);
88 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
89 if (inode == NULL) {
90 trace_nfs4_cb_recall(cps->clp, &args->fh, NULL,
91 &args->stateid, -ntohl(res));
92 goto out;
93 }
94 /* Set up a helper thread to actually return the delegation */
95 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
96 case 0:
97 res = 0;
98 break;
99 case -ENOENT:
100 res = htonl(NFS4ERR_BAD_STATEID);
101 break;
102 default:
103 res = htonl(NFS4ERR_RESOURCE);
104 }
105 trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
106 &args->stateid, -ntohl(res));
107 iput(inode);
108out:
109 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
110 return res;
111}
112
113#if defined(CONFIG_NFS_V4_1)
114
115/*
116 * Lookup a layout inode by stateid
117 *
118 * Note: returns a refcount on the inode and superblock
119 */
120static struct inode *nfs_layout_find_inode_by_stateid(struct nfs_client *clp,
121 const nfs4_stateid *stateid)
122{
123 struct nfs_server *server;
124 struct inode *inode;
125 struct pnfs_layout_hdr *lo;
126
127restart:
128 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
129 list_for_each_entry(lo, &server->layouts, plh_layouts) {
130 if (stateid != NULL &&
131 !nfs4_stateid_match_other(stateid, &lo->plh_stateid))
132 continue;
133 inode = igrab(lo->plh_inode);
134 if (!inode)
135 continue;
136 if (!nfs_sb_active(inode->i_sb)) {
137 rcu_read_unlock();
138 spin_unlock(&clp->cl_lock);
139 iput(inode);
140 spin_lock(&clp->cl_lock);
141 rcu_read_lock();
142 goto restart;
143 }
144 return inode;
145 }
146 }
147
148 return NULL;
149}
150
151/*
152 * Lookup a layout inode by filehandle.
153 *
154 * Note: returns a refcount on the inode and superblock
155 *
156 */
157static struct inode *nfs_layout_find_inode_by_fh(struct nfs_client *clp,
158 const struct nfs_fh *fh)
159{
160 struct nfs_server *server;
161 struct nfs_inode *nfsi;
162 struct inode *inode;
163 struct pnfs_layout_hdr *lo;
164
165restart:
166 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
167 list_for_each_entry(lo, &server->layouts, plh_layouts) {
168 nfsi = NFS_I(lo->plh_inode);
169 if (nfs_compare_fh(fh, &nfsi->fh))
170 continue;
171 if (nfsi->layout != lo)
172 continue;
173 inode = igrab(lo->plh_inode);
174 if (!inode)
175 continue;
176 if (!nfs_sb_active(inode->i_sb)) {
177 rcu_read_unlock();
178 spin_unlock(&clp->cl_lock);
179 iput(inode);
180 spin_lock(&clp->cl_lock);
181 rcu_read_lock();
182 goto restart;
183 }
184 return inode;
185 }
186 }
187
188 return NULL;
189}
190
191static struct inode *nfs_layout_find_inode(struct nfs_client *clp,
192 const struct nfs_fh *fh,
193 const nfs4_stateid *stateid)
194{
195 struct inode *inode;
196
197 spin_lock(&clp->cl_lock);
198 rcu_read_lock();
199 inode = nfs_layout_find_inode_by_stateid(clp, stateid);
200 if (!inode)
201 inode = nfs_layout_find_inode_by_fh(clp, fh);
202 rcu_read_unlock();
203 spin_unlock(&clp->cl_lock);
204
205 return inode;
206}
207
208/*
209 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
210 */
211static u32 pnfs_check_callback_stateid(struct pnfs_layout_hdr *lo,
212 const nfs4_stateid *new)
213{
214 u32 oldseq, newseq;
215
216 /* Is the stateid still not initialised? */
217 if (!pnfs_layout_is_valid(lo))
218 return NFS4ERR_DELAY;
219
220 /* Mismatched stateid? */
221 if (!nfs4_stateid_match_other(&lo->plh_stateid, new))
222 return NFS4ERR_BAD_STATEID;
223
224 newseq = be32_to_cpu(new->seqid);
225 /* Are we already in a layout recall situation? */
226 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
227 lo->plh_return_seq != 0) {
228 if (newseq < lo->plh_return_seq)
229 return NFS4ERR_OLD_STATEID;
230 if (newseq > lo->plh_return_seq)
231 return NFS4ERR_DELAY;
232 goto out;
233 }
234
235 /* Check that the stateid matches what we think it should be. */
236 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
237 if (newseq > oldseq + 1)
238 return NFS4ERR_DELAY;
239 /* Crazy server! */
240 if (newseq <= oldseq)
241 return NFS4ERR_OLD_STATEID;
242out:
243 return NFS_OK;
244}
245
246static u32 initiate_file_draining(struct nfs_client *clp,
247 struct cb_layoutrecallargs *args)
248{
249 struct inode *ino;
250 struct pnfs_layout_hdr *lo;
251 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
252 LIST_HEAD(free_me_list);
253
254 ino = nfs_layout_find_inode(clp, &args->cbl_fh, &args->cbl_stateid);
255 if (!ino)
256 goto out;
257
258 pnfs_layoutcommit_inode(ino, false);
259
260
261 spin_lock(&ino->i_lock);
262 lo = NFS_I(ino)->layout;
263 if (!lo) {
264 spin_unlock(&ino->i_lock);
265 goto out;
266 }
267 pnfs_get_layout_hdr(lo);
268 rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid);
269 if (rv != NFS_OK)
270 goto unlock;
271 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
272
273 /*
274 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
275 */
276 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
277 rv = NFS4ERR_DELAY;
278 goto unlock;
279 }
280
281 if (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
282 &args->cbl_range,
283 be32_to_cpu(args->cbl_stateid.seqid))) {
284 rv = NFS4_OK;
285 goto unlock;
286 }
287
288 /* Embrace your forgetfulness! */
289 rv = NFS4ERR_NOMATCHING_LAYOUT;
290
291 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
292 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
293 &args->cbl_range);
294 }
295unlock:
296 spin_unlock(&ino->i_lock);
297 pnfs_free_lseg_list(&free_me_list);
298 /* Free all lsegs that are attached to commit buckets */
299 nfs_commit_inode(ino, 0);
300 pnfs_put_layout_hdr(lo);
301out:
302 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
303 &args->cbl_stateid, -rv);
304 nfs_iput_and_deactive(ino);
305 return rv;
306}
307
308static u32 initiate_bulk_draining(struct nfs_client *clp,
309 struct cb_layoutrecallargs *args)
310{
311 int stat;
312
313 if (args->cbl_recall_type == RETURN_FSID)
314 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
315 else
316 stat = pnfs_destroy_layouts_byclid(clp, true);
317 if (stat != 0)
318 return NFS4ERR_DELAY;
319 return NFS4ERR_NOMATCHING_LAYOUT;
320}
321
322static u32 do_callback_layoutrecall(struct nfs_client *clp,
323 struct cb_layoutrecallargs *args)
324{
325 if (args->cbl_recall_type == RETURN_FILE)
326 return initiate_file_draining(clp, args);
327 return initiate_bulk_draining(clp, args);
328}
329
330__be32 nfs4_callback_layoutrecall(void *argp, void *resp,
331 struct cb_process_state *cps)
332{
333 struct cb_layoutrecallargs *args = argp;
334 u32 res = NFS4ERR_OP_NOT_IN_SESSION;
335
336 if (cps->clp)
337 res = do_callback_layoutrecall(cps->clp, args);
338 return cpu_to_be32(res);
339}
340
341static void pnfs_recall_all_layouts(struct nfs_client *clp)
342{
343 struct cb_layoutrecallargs args;
344
345 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
346 memset(&args, 0, sizeof(args));
347 args.cbl_recall_type = RETURN_ALL;
348 /* FIXME we ignore errors, what should we do? */
349 do_callback_layoutrecall(clp, &args);
350}
351
352__be32 nfs4_callback_devicenotify(void *argp, void *resp,
353 struct cb_process_state *cps)
354{
355 struct cb_devicenotifyargs *args = argp;
356 int i;
357 __be32 res = 0;
358 struct nfs_client *clp = cps->clp;
359 struct nfs_server *server = NULL;
360
361 if (!clp) {
362 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
363 goto out;
364 }
365
366 for (i = 0; i < args->ndevs; i++) {
367 struct cb_devicenotifyitem *dev = &args->devs[i];
368
369 if (!server ||
370 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
371 rcu_read_lock();
372 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
373 if (server->pnfs_curr_ld &&
374 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
375 rcu_read_unlock();
376 goto found;
377 }
378 rcu_read_unlock();
379 continue;
380 }
381
382 found:
383 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
384 }
385
386out:
387 kfree(args->devs);
388 return res;
389}
390
391/*
392 * Validate the sequenceID sent by the server.
393 * Return success if the sequenceID is one more than what we last saw on
394 * this slot, accounting for wraparound. Increments the slot's sequence.
395 *
396 * We don't yet implement a duplicate request cache, instead we set the
397 * back channel ca_maxresponsesize_cached to zero. This is OK for now
398 * since we only currently implement idempotent callbacks anyway.
399 *
400 * We have a single slot backchannel at this time, so we don't bother
401 * checking the used_slots bit array on the table. The lower layer guarantees
402 * a single outstanding callback request at a time.
403 */
404static __be32
405validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
406 const struct cb_sequenceargs * args)
407{
408 if (args->csa_slotid > tbl->server_highest_slotid)
409 return htonl(NFS4ERR_BADSLOT);
410
411 /* Replay */
412 if (args->csa_sequenceid == slot->seq_nr) {
413 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
414 return htonl(NFS4ERR_DELAY);
415 /* Signal process_op to set this error on next op */
416 if (args->csa_cachethis == 0)
417 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
418
419 /* Liar! We never allowed you to set csa_cachethis != 0 */
420 return htonl(NFS4ERR_SEQ_FALSE_RETRY);
421 }
422
423 /* Wraparound */
424 if (unlikely(slot->seq_nr == 0xFFFFFFFFU)) {
425 if (args->csa_sequenceid == 1)
426 return htonl(NFS4_OK);
427 } else if (likely(args->csa_sequenceid == slot->seq_nr + 1))
428 return htonl(NFS4_OK);
429
430 /* Misordered request */
431 return htonl(NFS4ERR_SEQ_MISORDERED);
432}
433
434/*
435 * For each referring call triple, check the session's slot table for
436 * a match. If the slot is in use and the sequence numbers match, the
437 * client is still waiting for a response to the original request.
438 */
439static bool referring_call_exists(struct nfs_client *clp,
440 uint32_t nrclists,
441 struct referring_call_list *rclists)
442{
443 bool status = false;
444 int i, j;
445 struct nfs4_session *session;
446 struct nfs4_slot_table *tbl;
447 struct referring_call_list *rclist;
448 struct referring_call *ref;
449
450 /*
451 * XXX When client trunking is implemented, this becomes
452 * a session lookup from within the loop
453 */
454 session = clp->cl_session;
455 tbl = &session->fc_slot_table;
456
457 for (i = 0; i < nrclists; i++) {
458 rclist = &rclists[i];
459 if (memcmp(session->sess_id.data,
460 rclist->rcl_sessionid.data,
461 NFS4_MAX_SESSIONID_LEN) != 0)
462 continue;
463
464 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
465 ref = &rclist->rcl_refcalls[j];
466 status = nfs4_slot_wait_on_seqid(tbl, ref->rc_slotid,
467 ref->rc_sequenceid, HZ >> 1) < 0;
468 if (status)
469 goto out;
470 }
471 }
472
473out:
474 return status;
475}
476
477__be32 nfs4_callback_sequence(void *argp, void *resp,
478 struct cb_process_state *cps)
479{
480 struct cb_sequenceargs *args = argp;
481 struct cb_sequenceres *res = resp;
482 struct nfs4_slot_table *tbl;
483 struct nfs4_slot *slot;
484 struct nfs_client *clp;
485 int i;
486 __be32 status = htonl(NFS4ERR_BADSESSION);
487
488 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
489 &args->csa_sessionid, cps->minorversion);
490 if (clp == NULL)
491 goto out;
492
493 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
494 goto out;
495
496 tbl = &clp->cl_session->bc_slot_table;
497
498 /* Set up res before grabbing the spinlock */
499 memcpy(&res->csr_sessionid, &args->csa_sessionid,
500 sizeof(res->csr_sessionid));
501 res->csr_sequenceid = args->csa_sequenceid;
502 res->csr_slotid = args->csa_slotid;
503
504 spin_lock(&tbl->slot_tbl_lock);
505 /* state manager is resetting the session */
506 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
507 status = htonl(NFS4ERR_DELAY);
508 /* Return NFS4ERR_BADSESSION if we're draining the session
509 * in order to reset it.
510 */
511 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
512 status = htonl(NFS4ERR_BADSESSION);
513 goto out_unlock;
514 }
515
516 status = htonl(NFS4ERR_BADSLOT);
517 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
518 if (IS_ERR(slot))
519 goto out_unlock;
520
521 res->csr_highestslotid = tbl->server_highest_slotid;
522 res->csr_target_highestslotid = tbl->target_highest_slotid;
523
524 status = validate_seqid(tbl, slot, args);
525 if (status)
526 goto out_unlock;
527 if (!nfs4_try_to_lock_slot(tbl, slot)) {
528 status = htonl(NFS4ERR_DELAY);
529 goto out_unlock;
530 }
531 cps->slot = slot;
532
533 /* The ca_maxresponsesize_cached is 0 with no DRC */
534 if (args->csa_cachethis != 0) {
535 status = htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
536 goto out_unlock;
537 }
538
539 /*
540 * Check for pending referring calls. If a match is found, a
541 * related callback was received before the response to the original
542 * call.
543 */
544 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
545 status = htonl(NFS4ERR_DELAY);
546 goto out_unlock;
547 }
548
549 /*
550 * RFC5661 20.9.3
551 * If CB_SEQUENCE returns an error, then the state of the slot
552 * (sequence ID, cached reply) MUST NOT change.
553 */
554 slot->seq_nr = args->csa_sequenceid;
555out_unlock:
556 spin_unlock(&tbl->slot_tbl_lock);
557
558out:
559 cps->clp = clp; /* put in nfs4_callback_compound */
560 for (i = 0; i < args->csa_nrclists; i++)
561 kfree(args->csa_rclists[i].rcl_refcalls);
562 kfree(args->csa_rclists);
563
564 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
565 cps->drc_status = status;
566 status = 0;
567 } else
568 res->csr_status = status;
569
570 trace_nfs4_cb_sequence(args, res, status);
571 return status;
572}
573
574static bool
575validate_bitmap_values(unsigned int mask)
576{
577 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
578}
579
580__be32 nfs4_callback_recallany(void *argp, void *resp,
581 struct cb_process_state *cps)
582{
583 struct cb_recallanyargs *args = argp;
584 __be32 status;
585 fmode_t flags = 0;
586
587 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
588 if (!cps->clp) /* set in cb_sequence */
589 goto out;
590
591 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
592 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
593
594 status = cpu_to_be32(NFS4ERR_INVAL);
595 if (!validate_bitmap_values(args->craa_type_mask))
596 goto out;
597
598 status = cpu_to_be32(NFS4_OK);
599 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_RDATA_DLG))
600 flags = FMODE_READ;
601 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_WDATA_DLG))
602 flags |= FMODE_WRITE;
603 if (flags)
604 nfs_expire_unused_delegation_types(cps->clp, flags);
605
606 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_FILE_LAYOUT))
607 pnfs_recall_all_layouts(cps->clp);
608out:
609 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
610 return status;
611}
612
613/* Reduce the fore channel's max_slots to the target value */
614__be32 nfs4_callback_recallslot(void *argp, void *resp,
615 struct cb_process_state *cps)
616{
617 struct cb_recallslotargs *args = argp;
618 struct nfs4_slot_table *fc_tbl;
619 __be32 status;
620
621 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
622 if (!cps->clp) /* set in cb_sequence */
623 goto out;
624
625 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
626 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
627 args->crsa_target_highest_slotid);
628
629 fc_tbl = &cps->clp->cl_session->fc_slot_table;
630
631 status = htonl(NFS4_OK);
632
633 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
634 nfs41_notify_server(cps->clp);
635out:
636 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
637 return status;
638}
639
640__be32 nfs4_callback_notify_lock(void *argp, void *resp,
641 struct cb_process_state *cps)
642{
643 struct cb_notify_lock_args *args = argp;
644
645 if (!cps->clp) /* set in cb_sequence */
646 return htonl(NFS4ERR_OP_NOT_IN_SESSION);
647
648 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
649 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
650
651 /* Don't wake anybody if the string looked bogus */
652 if (args->cbnl_valid)
653 __wake_up(&cps->clp->cl_lock_waitq, TASK_NORMAL, 0, args);
654
655 return htonl(NFS4_OK);
656}
657#endif /* CONFIG_NFS_V4_1 */