Loading...
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
10#include <linux/errno.h>
11#include <linux/math.h>
12#include <linux/nfs4.h>
13#include <linux/nfs_fs.h>
14#include <linux/slab.h>
15#include <linux/rcupdate.h>
16#include <linux/types.h>
17
18#include "nfs4_fs.h"
19#include "callback.h"
20#include "delegation.h"
21#include "internal.h"
22#include "pnfs.h"
23#include "nfs4session.h"
24#include "nfs4trace.h"
25
26#define NFSDBG_FACILITY NFSDBG_CALLBACK
27
28__be32 nfs4_callback_getattr(void *argp, void *resp,
29 struct cb_process_state *cps)
30{
31 struct cb_getattrargs *args = argp;
32 struct cb_getattrres *res = resp;
33 struct nfs_delegation *delegation;
34 struct inode *inode;
35
36 res->status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
37 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
38 goto out;
39
40 res->bitmap[0] = res->bitmap[1] = 0;
41 res->status = htonl(NFS4ERR_BADHANDLE);
42
43 dprintk_rcu("NFS: GETATTR callback request from %s\n",
44 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
45
46 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
47 if (IS_ERR(inode)) {
48 if (inode == ERR_PTR(-EAGAIN))
49 res->status = htonl(NFS4ERR_DELAY);
50 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
51 -ntohl(res->status));
52 goto out;
53 }
54 rcu_read_lock();
55 delegation = nfs4_get_valid_delegation(inode);
56 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
57 goto out_iput;
58 res->size = i_size_read(inode);
59 res->change_attr = delegation->change_attr;
60 if (nfs_have_writebacks(inode))
61 res->change_attr++;
62 res->ctime = inode_get_ctime(inode);
63 res->mtime = inode_get_mtime(inode);
64 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
65 args->bitmap[0];
66 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
67 args->bitmap[1];
68 res->status = 0;
69out_iput:
70 rcu_read_unlock();
71 trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
72 nfs_iput_and_deactive(inode);
73out:
74 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
75 return res->status;
76}
77
78__be32 nfs4_callback_recall(void *argp, void *resp,
79 struct cb_process_state *cps)
80{
81 struct cb_recallargs *args = argp;
82 struct inode *inode;
83 __be32 res;
84
85 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
86 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
87 goto out;
88
89 dprintk_rcu("NFS: RECALL callback request from %s\n",
90 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
91
92 res = htonl(NFS4ERR_BADHANDLE);
93 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
94 if (IS_ERR(inode)) {
95 if (inode == ERR_PTR(-EAGAIN))
96 res = htonl(NFS4ERR_DELAY);
97 trace_nfs4_cb_recall(cps->clp, &args->fh, NULL,
98 &args->stateid, -ntohl(res));
99 goto out;
100 }
101 /* Set up a helper thread to actually return the delegation */
102 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
103 case 0:
104 res = 0;
105 break;
106 case -ENOENT:
107 res = htonl(NFS4ERR_BAD_STATEID);
108 break;
109 default:
110 res = htonl(NFS4ERR_RESOURCE);
111 }
112 trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
113 &args->stateid, -ntohl(res));
114 nfs_iput_and_deactive(inode);
115out:
116 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
117 return res;
118}
119
120#if defined(CONFIG_NFS_V4_1)
121
122/*
123 * Lookup a layout inode by stateid
124 *
125 * Note: returns a refcount on the inode and superblock
126 */
127static struct inode *nfs_layout_find_inode_by_stateid(struct nfs_client *clp,
128 const nfs4_stateid *stateid)
129 __must_hold(RCU)
130{
131 struct nfs_server *server;
132 struct inode *inode;
133 struct pnfs_layout_hdr *lo;
134
135 rcu_read_lock();
136 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
137 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
138 if (!pnfs_layout_is_valid(lo))
139 continue;
140 if (!nfs4_stateid_match_other(stateid, &lo->plh_stateid))
141 continue;
142 if (nfs_sb_active(server->super))
143 inode = igrab(lo->plh_inode);
144 else
145 inode = ERR_PTR(-EAGAIN);
146 rcu_read_unlock();
147 if (inode)
148 return inode;
149 nfs_sb_deactive(server->super);
150 return ERR_PTR(-EAGAIN);
151 }
152 }
153 rcu_read_unlock();
154 return ERR_PTR(-ENOENT);
155}
156
157/*
158 * Lookup a layout inode by filehandle.
159 *
160 * Note: returns a refcount on the inode and superblock
161 *
162 */
163static struct inode *nfs_layout_find_inode_by_fh(struct nfs_client *clp,
164 const struct nfs_fh *fh)
165{
166 struct nfs_server *server;
167 struct nfs_inode *nfsi;
168 struct inode *inode;
169 struct pnfs_layout_hdr *lo;
170
171 rcu_read_lock();
172 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
173 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
174 nfsi = NFS_I(lo->plh_inode);
175 if (nfs_compare_fh(fh, &nfsi->fh))
176 continue;
177 if (nfsi->layout != lo)
178 continue;
179 if (nfs_sb_active(server->super))
180 inode = igrab(lo->plh_inode);
181 else
182 inode = ERR_PTR(-EAGAIN);
183 rcu_read_unlock();
184 if (inode)
185 return inode;
186 nfs_sb_deactive(server->super);
187 return ERR_PTR(-EAGAIN);
188 }
189 }
190 rcu_read_unlock();
191 return ERR_PTR(-ENOENT);
192}
193
194static struct inode *nfs_layout_find_inode(struct nfs_client *clp,
195 const struct nfs_fh *fh,
196 const nfs4_stateid *stateid)
197{
198 struct inode *inode;
199
200 inode = nfs_layout_find_inode_by_stateid(clp, stateid);
201 if (inode == ERR_PTR(-ENOENT))
202 inode = nfs_layout_find_inode_by_fh(clp, fh);
203 return inode;
204}
205
206/*
207 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
208 */
209static u32 pnfs_check_callback_stateid(struct pnfs_layout_hdr *lo,
210 const nfs4_stateid *new,
211 struct cb_process_state *cps)
212{
213 u32 oldseq, newseq;
214
215 /* Is the stateid not initialised? */
216 if (!pnfs_layout_is_valid(lo))
217 return NFS4ERR_NOMATCHING_LAYOUT;
218
219 /* Mismatched stateid? */
220 if (!nfs4_stateid_match_other(&lo->plh_stateid, new))
221 return NFS4ERR_BAD_STATEID;
222
223 newseq = be32_to_cpu(new->seqid);
224 /* Are we already in a layout recall situation? */
225 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
226 return NFS4ERR_DELAY;
227
228 /*
229 * Check that the stateid matches what we think it should be.
230 * Note that if the server sent us a list of referring calls,
231 * and we know that those have completed, then we trust the
232 * stateid argument is correct.
233 */
234 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
235 if (newseq > oldseq + 1 && !cps->referring_calls)
236 return NFS4ERR_DELAY;
237
238 /* Crazy server! */
239 if (newseq <= oldseq)
240 return NFS4ERR_OLD_STATEID;
241
242 return NFS_OK;
243}
244
245static u32 initiate_file_draining(struct nfs_client *clp,
246 struct cb_layoutrecallargs *args,
247 struct cb_process_state *cps)
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 (IS_ERR(ino)) {
256 if (ino == ERR_PTR(-EAGAIN))
257 rv = NFS4ERR_DELAY;
258 goto out_noput;
259 }
260
261 pnfs_layoutcommit_inode(ino, false);
262
263
264 spin_lock(&ino->i_lock);
265 lo = NFS_I(ino)->layout;
266 if (!lo) {
267 spin_unlock(&ino->i_lock);
268 goto out;
269 }
270 pnfs_get_layout_hdr(lo);
271 rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid, cps);
272 if (rv != NFS_OK)
273 goto unlock;
274
275 /*
276 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
277 */
278 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
279 rv = NFS4ERR_DELAY;
280 goto unlock;
281 }
282
283 pnfs_set_layout_stateid(lo, &args->cbl_stateid, NULL, true);
284 switch (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
285 &args->cbl_range,
286 be32_to_cpu(args->cbl_stateid.seqid))) {
287 case 0:
288 case -EBUSY:
289 /* There are layout segments that need to be returned */
290 rv = NFS4_OK;
291 break;
292 case -ENOENT:
293 set_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags);
294 /* Embrace your forgetfulness! */
295 rv = NFS4ERR_NOMATCHING_LAYOUT;
296
297 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
298 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
299 &args->cbl_range);
300 }
301 }
302unlock:
303 spin_unlock(&ino->i_lock);
304 pnfs_free_lseg_list(&free_me_list);
305 /* Free all lsegs that are attached to commit buckets */
306 nfs_commit_inode(ino, 0);
307 pnfs_put_layout_hdr(lo);
308out:
309 nfs_iput_and_deactive(ino);
310out_noput:
311 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
312 &args->cbl_stateid, -rv);
313 return rv;
314}
315
316static u32 initiate_bulk_draining(struct nfs_client *clp,
317 struct cb_layoutrecallargs *args)
318{
319 int stat;
320
321 if (args->cbl_recall_type == RETURN_FSID)
322 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
323 else
324 stat = pnfs_destroy_layouts_byclid(clp, true);
325 if (stat != 0)
326 return NFS4ERR_DELAY;
327 return NFS4ERR_NOMATCHING_LAYOUT;
328}
329
330static u32 do_callback_layoutrecall(struct nfs_client *clp,
331 struct cb_layoutrecallargs *args,
332 struct cb_process_state *cps)
333{
334 if (args->cbl_recall_type == RETURN_FILE)
335 return initiate_file_draining(clp, args, cps);
336 return initiate_bulk_draining(clp, args);
337}
338
339__be32 nfs4_callback_layoutrecall(void *argp, void *resp,
340 struct cb_process_state *cps)
341{
342 struct cb_layoutrecallargs *args = argp;
343 u32 res = NFS4ERR_OP_NOT_IN_SESSION;
344
345 if (cps->clp)
346 res = do_callback_layoutrecall(cps->clp, args, cps);
347 return cpu_to_be32(res);
348}
349
350static void pnfs_recall_all_layouts(struct nfs_client *clp,
351 struct cb_process_state *cps)
352{
353 struct cb_layoutrecallargs args;
354
355 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
356 memset(&args, 0, sizeof(args));
357 args.cbl_recall_type = RETURN_ALL;
358 /* FIXME we ignore errors, what should we do? */
359 do_callback_layoutrecall(clp, &args, cps);
360}
361
362__be32 nfs4_callback_devicenotify(void *argp, void *resp,
363 struct cb_process_state *cps)
364{
365 struct cb_devicenotifyargs *args = argp;
366 const struct pnfs_layoutdriver_type *ld = NULL;
367 uint32_t i;
368 __be32 res = 0;
369
370 if (!cps->clp) {
371 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
372 goto out;
373 }
374
375 for (i = 0; i < args->ndevs; i++) {
376 struct cb_devicenotifyitem *dev = &args->devs[i];
377
378 if (!ld || ld->id != dev->cbd_layout_type) {
379 pnfs_put_layoutdriver(ld);
380 ld = pnfs_find_layoutdriver(dev->cbd_layout_type);
381 if (!ld)
382 continue;
383 }
384 nfs4_delete_deviceid(ld, cps->clp, &dev->cbd_dev_id);
385 }
386 pnfs_put_layoutdriver(ld);
387out:
388 kfree(args->devs);
389 return res;
390}
391
392/*
393 * Validate the sequenceID sent by the server.
394 * Return success if the sequenceID is one more than what we last saw on
395 * this slot, accounting for wraparound. Increments the slot's sequence.
396 *
397 * We don't yet implement a duplicate request cache, instead we set the
398 * back channel ca_maxresponsesize_cached to zero. This is OK for now
399 * since we only currently implement idempotent callbacks anyway.
400 *
401 * We have a single slot backchannel at this time, so we don't bother
402 * checking the used_slots bit array on the table. The lower layer guarantees
403 * a single outstanding callback request at a time.
404 */
405static __be32
406validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
407 const struct cb_sequenceargs * args)
408{
409 __be32 ret;
410
411 ret = cpu_to_be32(NFS4ERR_BADSLOT);
412 if (args->csa_slotid > tbl->server_highest_slotid)
413 goto out_err;
414
415 /* Replay */
416 if (args->csa_sequenceid == slot->seq_nr) {
417 ret = cpu_to_be32(NFS4ERR_DELAY);
418 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
419 goto out_err;
420
421 /* Signal process_op to set this error on next op */
422 ret = cpu_to_be32(NFS4ERR_RETRY_UNCACHED_REP);
423 if (args->csa_cachethis == 0)
424 goto out_err;
425
426 /* Liar! We never allowed you to set csa_cachethis != 0 */
427 ret = cpu_to_be32(NFS4ERR_SEQ_FALSE_RETRY);
428 goto out_err;
429 }
430
431 /* Note: wraparound relies on seq_nr being of type u32 */
432 /* Misordered request */
433 ret = cpu_to_be32(NFS4ERR_SEQ_MISORDERED);
434 if (args->csa_sequenceid != slot->seq_nr + 1)
435 goto out_err;
436
437 return cpu_to_be32(NFS4_OK);
438
439out_err:
440 trace_nfs4_cb_seqid_err(args, ret);
441 return ret;
442}
443
444/*
445 * For each referring call triple, check the session's slot table for
446 * a match. If the slot is in use and the sequence numbers match, the
447 * client is still waiting for a response to the original request.
448 */
449static int referring_call_exists(struct nfs_client *clp,
450 uint32_t nrclists,
451 struct referring_call_list *rclists,
452 spinlock_t *lock)
453 __releases(lock)
454 __acquires(lock)
455{
456 int status = 0;
457 int found = 0;
458 int i, j;
459 struct nfs4_session *session;
460 struct nfs4_slot_table *tbl;
461 struct referring_call_list *rclist;
462 struct referring_call *ref;
463
464 /*
465 * XXX When client trunking is implemented, this becomes
466 * a session lookup from within the loop
467 */
468 session = clp->cl_session;
469 tbl = &session->fc_slot_table;
470
471 for (i = 0; i < nrclists; i++) {
472 rclist = &rclists[i];
473 if (memcmp(session->sess_id.data,
474 rclist->rcl_sessionid.data,
475 NFS4_MAX_SESSIONID_LEN) != 0)
476 continue;
477
478 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
479 ref = &rclist->rcl_refcalls[j];
480 spin_unlock(lock);
481 status = nfs4_slot_wait_on_seqid(tbl, ref->rc_slotid,
482 ref->rc_sequenceid, HZ >> 1) < 0;
483 spin_lock(lock);
484 if (status)
485 goto out;
486 found++;
487 }
488 }
489
490out:
491 return status < 0 ? status : found;
492}
493
494__be32 nfs4_callback_sequence(void *argp, void *resp,
495 struct cb_process_state *cps)
496{
497 struct cb_sequenceargs *args = argp;
498 struct cb_sequenceres *res = resp;
499 struct nfs4_slot_table *tbl;
500 struct nfs4_slot *slot;
501 struct nfs_client *clp;
502 int ret;
503 int i;
504 __be32 status = htonl(NFS4ERR_BADSESSION);
505
506 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
507 &args->csa_sessionid, cps->minorversion);
508 if (clp == NULL)
509 goto out;
510
511 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
512 goto out;
513
514 tbl = &clp->cl_session->bc_slot_table;
515
516 /* Set up res before grabbing the spinlock */
517 memcpy(&res->csr_sessionid, &args->csa_sessionid,
518 sizeof(res->csr_sessionid));
519 res->csr_sequenceid = args->csa_sequenceid;
520 res->csr_slotid = args->csa_slotid;
521
522 spin_lock(&tbl->slot_tbl_lock);
523 /* state manager is resetting the session */
524 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
525 status = htonl(NFS4ERR_DELAY);
526 /* Return NFS4ERR_BADSESSION if we're draining the session
527 * in order to reset it.
528 */
529 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
530 status = htonl(NFS4ERR_BADSESSION);
531 goto out_unlock;
532 }
533
534 status = htonl(NFS4ERR_BADSLOT);
535 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
536 if (IS_ERR(slot))
537 goto out_unlock;
538
539 res->csr_highestslotid = tbl->server_highest_slotid;
540 res->csr_target_highestslotid = tbl->target_highest_slotid;
541
542 status = validate_seqid(tbl, slot, args);
543 if (status)
544 goto out_unlock;
545 if (!nfs4_try_to_lock_slot(tbl, slot)) {
546 status = htonl(NFS4ERR_DELAY);
547 goto out_unlock;
548 }
549 cps->slot = slot;
550
551 /* The ca_maxresponsesize_cached is 0 with no DRC */
552 if (args->csa_cachethis != 0) {
553 status = htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
554 goto out_unlock;
555 }
556
557 /*
558 * Check for pending referring calls. If a match is found, a
559 * related callback was received before the response to the original
560 * call.
561 */
562 ret = referring_call_exists(clp, args->csa_nrclists, args->csa_rclists,
563 &tbl->slot_tbl_lock);
564 if (ret < 0) {
565 status = htonl(NFS4ERR_DELAY);
566 goto out_unlock;
567 }
568 cps->referring_calls = ret;
569
570 /*
571 * RFC5661 20.9.3
572 * If CB_SEQUENCE returns an error, then the state of the slot
573 * (sequence ID, cached reply) MUST NOT change.
574 */
575 slot->seq_nr = args->csa_sequenceid;
576out_unlock:
577 spin_unlock(&tbl->slot_tbl_lock);
578
579out:
580 cps->clp = clp; /* put in nfs4_callback_compound */
581 for (i = 0; i < args->csa_nrclists; i++)
582 kfree(args->csa_rclists[i].rcl_refcalls);
583 kfree(args->csa_rclists);
584
585 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
586 cps->drc_status = status;
587 status = 0;
588 } else
589 res->csr_status = status;
590
591 trace_nfs4_cb_sequence(args, res, status);
592 return status;
593}
594
595static bool
596validate_bitmap_values(unsigned int mask)
597{
598 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
599}
600
601__be32 nfs4_callback_recallany(void *argp, void *resp,
602 struct cb_process_state *cps)
603{
604 struct cb_recallanyargs *args = argp;
605 __be32 status;
606 fmode_t flags = 0;
607 bool schedule_manager = false;
608
609 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
610 if (!cps->clp) /* set in cb_sequence */
611 goto out;
612
613 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
614 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
615
616 status = cpu_to_be32(NFS4ERR_INVAL);
617 if (!validate_bitmap_values(args->craa_type_mask))
618 goto out;
619
620 status = cpu_to_be32(NFS4_OK);
621 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_RDATA_DLG))
622 flags = FMODE_READ;
623 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_WDATA_DLG))
624 flags |= FMODE_WRITE;
625 if (flags)
626 nfs_expire_unused_delegation_types(cps->clp, flags);
627
628 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_FILE_LAYOUT))
629 pnfs_recall_all_layouts(cps->clp, cps);
630
631 if (args->craa_type_mask & BIT(PNFS_FF_RCA4_TYPE_MASK_READ)) {
632 set_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &cps->clp->cl_state);
633 schedule_manager = true;
634 }
635 if (args->craa_type_mask & BIT(PNFS_FF_RCA4_TYPE_MASK_RW)) {
636 set_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &cps->clp->cl_state);
637 schedule_manager = true;
638 }
639 if (schedule_manager)
640 nfs4_schedule_state_manager(cps->clp);
641
642out:
643 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
644 return status;
645}
646
647/* Reduce the fore channel's max_slots to the target value */
648__be32 nfs4_callback_recallslot(void *argp, void *resp,
649 struct cb_process_state *cps)
650{
651 struct cb_recallslotargs *args = argp;
652 struct nfs4_slot_table *fc_tbl;
653 __be32 status;
654
655 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
656 if (!cps->clp) /* set in cb_sequence */
657 goto out;
658
659 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
660 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
661 args->crsa_target_highest_slotid);
662
663 fc_tbl = &cps->clp->cl_session->fc_slot_table;
664
665 status = htonl(NFS4_OK);
666
667 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
668 nfs41_notify_server(cps->clp);
669out:
670 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
671 return status;
672}
673
674__be32 nfs4_callback_notify_lock(void *argp, void *resp,
675 struct cb_process_state *cps)
676{
677 struct cb_notify_lock_args *args = argp;
678
679 if (!cps->clp) /* set in cb_sequence */
680 return htonl(NFS4ERR_OP_NOT_IN_SESSION);
681
682 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
683 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
684
685 /* Don't wake anybody if the string looked bogus */
686 if (args->cbnl_valid)
687 __wake_up(&cps->clp->cl_lock_waitq, TASK_NORMAL, 0, args);
688
689 return htonl(NFS4_OK);
690}
691#endif /* CONFIG_NFS_V4_1 */
692#ifdef CONFIG_NFS_V4_2
693static void nfs4_copy_cb_args(struct nfs4_copy_state *cp_state,
694 struct cb_offloadargs *args)
695{
696 cp_state->count = args->wr_count;
697 cp_state->error = args->error;
698 if (!args->error) {
699 cp_state->verf.committed = args->wr_writeverf.committed;
700 memcpy(&cp_state->verf.verifier.data[0],
701 &args->wr_writeverf.verifier.data[0],
702 NFS4_VERIFIER_SIZE);
703 }
704}
705
706__be32 nfs4_callback_offload(void *data, void *dummy,
707 struct cb_process_state *cps)
708{
709 struct cb_offloadargs *args = data;
710 struct nfs_server *server;
711 struct nfs4_copy_state *copy, *tmp_copy;
712 bool found = false;
713
714 copy = kzalloc(sizeof(struct nfs4_copy_state), GFP_KERNEL);
715 if (!copy)
716 return htonl(NFS4ERR_SERVERFAULT);
717
718 spin_lock(&cps->clp->cl_lock);
719 rcu_read_lock();
720 list_for_each_entry_rcu(server, &cps->clp->cl_superblocks,
721 client_link) {
722 list_for_each_entry(tmp_copy, &server->ss_copies, copies) {
723 if (memcmp(args->coa_stateid.other,
724 tmp_copy->stateid.other,
725 sizeof(args->coa_stateid.other)))
726 continue;
727 nfs4_copy_cb_args(tmp_copy, args);
728 complete(&tmp_copy->completion);
729 found = true;
730 goto out;
731 }
732 }
733out:
734 rcu_read_unlock();
735 if (!found) {
736 memcpy(©->stateid, &args->coa_stateid, NFS4_STATEID_SIZE);
737 nfs4_copy_cb_args(copy, args);
738 list_add_tail(©->copies, &cps->clp->pending_cb_stateids);
739 } else
740 kfree(copy);
741 spin_unlock(&cps->clp->cl_lock);
742
743 trace_nfs4_cb_offload(&args->coa_fh, &args->coa_stateid,
744 args->wr_count, args->error,
745 args->wr_writeverf.committed);
746 return 0;
747}
748#endif /* CONFIG_NFS_V4_2 */
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#ifdef NFS_DEBUG
21#define NFSDBG_FACILITY NFSDBG_CALLBACK
22#endif
23
24__be32 nfs4_callback_getattr(struct cb_getattrargs *args,
25 struct cb_getattrres *res,
26 struct cb_process_state *cps)
27{
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 goto out;
45 nfsi = NFS_I(inode);
46 rcu_read_lock();
47 delegation = rcu_dereference(nfsi->delegation);
48 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
49 goto out_iput;
50 res->size = i_size_read(inode);
51 res->change_attr = delegation->change_attr;
52 if (nfsi->npages != 0)
53 res->change_attr++;
54 res->ctime = inode->i_ctime;
55 res->mtime = inode->i_mtime;
56 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
57 args->bitmap[0];
58 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
59 args->bitmap[1];
60 res->status = 0;
61out_iput:
62 rcu_read_unlock();
63 iput(inode);
64out:
65 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
66 return res->status;
67}
68
69__be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
70 struct cb_process_state *cps)
71{
72 struct inode *inode;
73 __be32 res;
74
75 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
76 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
77 goto out;
78
79 dprintk_rcu("NFS: RECALL callback request from %s\n",
80 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
81
82 res = htonl(NFS4ERR_BADHANDLE);
83 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
84 if (inode == NULL)
85 goto out;
86 /* Set up a helper thread to actually return the delegation */
87 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
88 case 0:
89 res = 0;
90 break;
91 case -ENOENT:
92 res = htonl(NFS4ERR_BAD_STATEID);
93 break;
94 default:
95 res = htonl(NFS4ERR_RESOURCE);
96 }
97 trace_nfs4_recall_delegation(inode, -ntohl(res));
98 iput(inode);
99out:
100 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
101 return res;
102}
103
104#if defined(CONFIG_NFS_V4_1)
105
106/*
107 * Lookup a layout by filehandle.
108 *
109 * Note: gets a refcount on the layout hdr and on its respective inode.
110 * Caller must put the layout hdr and the inode.
111 *
112 * TODO: keep track of all layouts (and delegations) in a hash table
113 * hashed by filehandle.
114 */
115static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp,
116 struct nfs_fh *fh, nfs4_stateid *stateid)
117{
118 struct nfs_server *server;
119 struct inode *ino;
120 struct pnfs_layout_hdr *lo;
121
122 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
123 list_for_each_entry(lo, &server->layouts, plh_layouts) {
124 if (!nfs4_stateid_match_other(&lo->plh_stateid, stateid))
125 continue;
126 if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
127 continue;
128 ino = igrab(lo->plh_inode);
129 if (!ino)
130 break;
131 spin_lock(&ino->i_lock);
132 /* Is this layout in the process of being freed? */
133 if (NFS_I(ino)->layout != lo) {
134 spin_unlock(&ino->i_lock);
135 iput(ino);
136 break;
137 }
138 pnfs_get_layout_hdr(lo);
139 spin_unlock(&ino->i_lock);
140 return lo;
141 }
142 }
143
144 return NULL;
145}
146
147static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp,
148 struct nfs_fh *fh, nfs4_stateid *stateid)
149{
150 struct pnfs_layout_hdr *lo;
151
152 spin_lock(&clp->cl_lock);
153 rcu_read_lock();
154 lo = get_layout_by_fh_locked(clp, fh, stateid);
155 rcu_read_unlock();
156 spin_unlock(&clp->cl_lock);
157
158 return lo;
159}
160
161static u32 initiate_file_draining(struct nfs_client *clp,
162 struct cb_layoutrecallargs *args)
163{
164 struct inode *ino;
165 struct pnfs_layout_hdr *lo;
166 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
167 LIST_HEAD(free_me_list);
168
169 lo = get_layout_by_fh(clp, &args->cbl_fh, &args->cbl_stateid);
170 if (!lo)
171 goto out;
172
173 ino = lo->plh_inode;
174 spin_lock(&ino->i_lock);
175 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
176 pnfs_mark_matching_lsegs_invalid(lo, &free_me_list,
177 &args->cbl_range))
178 rv = NFS4ERR_DELAY;
179 else
180 rv = NFS4ERR_NOMATCHING_LAYOUT;
181 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
182 spin_unlock(&ino->i_lock);
183 pnfs_free_lseg_list(&free_me_list);
184 pnfs_put_layout_hdr(lo);
185 iput(ino);
186out:
187 return rv;
188}
189
190static u32 initiate_bulk_draining(struct nfs_client *clp,
191 struct cb_layoutrecallargs *args)
192{
193 int stat;
194
195 if (args->cbl_recall_type == RETURN_FSID)
196 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
197 else
198 stat = pnfs_destroy_layouts_byclid(clp, true);
199 if (stat != 0)
200 return NFS4ERR_DELAY;
201 return NFS4ERR_NOMATCHING_LAYOUT;
202}
203
204static u32 do_callback_layoutrecall(struct nfs_client *clp,
205 struct cb_layoutrecallargs *args)
206{
207 u32 res;
208
209 dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
210 if (args->cbl_recall_type == RETURN_FILE)
211 res = initiate_file_draining(clp, args);
212 else
213 res = initiate_bulk_draining(clp, args);
214 dprintk("%s returning %i\n", __func__, res);
215 return res;
216
217}
218
219__be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
220 void *dummy, struct cb_process_state *cps)
221{
222 u32 res;
223
224 dprintk("%s: -->\n", __func__);
225
226 if (cps->clp)
227 res = do_callback_layoutrecall(cps->clp, args);
228 else
229 res = NFS4ERR_OP_NOT_IN_SESSION;
230
231 dprintk("%s: exit with status = %d\n", __func__, res);
232 return cpu_to_be32(res);
233}
234
235static void pnfs_recall_all_layouts(struct nfs_client *clp)
236{
237 struct cb_layoutrecallargs args;
238
239 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
240 memset(&args, 0, sizeof(args));
241 args.cbl_recall_type = RETURN_ALL;
242 /* FIXME we ignore errors, what should we do? */
243 do_callback_layoutrecall(clp, &args);
244}
245
246__be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
247 void *dummy, struct cb_process_state *cps)
248{
249 int i;
250 __be32 res = 0;
251 struct nfs_client *clp = cps->clp;
252 struct nfs_server *server = NULL;
253
254 dprintk("%s: -->\n", __func__);
255
256 if (!clp) {
257 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
258 goto out;
259 }
260
261 for (i = 0; i < args->ndevs; i++) {
262 struct cb_devicenotifyitem *dev = &args->devs[i];
263
264 if (!server ||
265 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
266 rcu_read_lock();
267 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
268 if (server->pnfs_curr_ld &&
269 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
270 rcu_read_unlock();
271 goto found;
272 }
273 rcu_read_unlock();
274 dprintk("%s: layout type %u not found\n",
275 __func__, dev->cbd_layout_type);
276 continue;
277 }
278
279 found:
280 if (dev->cbd_notify_type == NOTIFY_DEVICEID4_CHANGE)
281 dprintk("%s: NOTIFY_DEVICEID4_CHANGE not supported, "
282 "deleting instead\n", __func__);
283 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
284 }
285
286out:
287 kfree(args->devs);
288 dprintk("%s: exit with status = %u\n",
289 __func__, be32_to_cpu(res));
290 return res;
291}
292
293/*
294 * Validate the sequenceID sent by the server.
295 * Return success if the sequenceID is one more than what we last saw on
296 * this slot, accounting for wraparound. Increments the slot's sequence.
297 *
298 * We don't yet implement a duplicate request cache, instead we set the
299 * back channel ca_maxresponsesize_cached to zero. This is OK for now
300 * since we only currently implement idempotent callbacks anyway.
301 *
302 * We have a single slot backchannel at this time, so we don't bother
303 * checking the used_slots bit array on the table. The lower layer guarantees
304 * a single outstanding callback request at a time.
305 */
306static __be32
307validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
308{
309 struct nfs4_slot *slot;
310
311 dprintk("%s enter. slotid %u seqid %u\n",
312 __func__, args->csa_slotid, args->csa_sequenceid);
313
314 if (args->csa_slotid >= NFS41_BC_MAX_CALLBACKS)
315 return htonl(NFS4ERR_BADSLOT);
316
317 slot = tbl->slots + args->csa_slotid;
318 dprintk("%s slot table seqid: %u\n", __func__, slot->seq_nr);
319
320 /* Normal */
321 if (likely(args->csa_sequenceid == slot->seq_nr + 1)) {
322 slot->seq_nr++;
323 goto out_ok;
324 }
325
326 /* Replay */
327 if (args->csa_sequenceid == slot->seq_nr) {
328 dprintk("%s seqid %u is a replay\n",
329 __func__, args->csa_sequenceid);
330 /* Signal process_op to set this error on next op */
331 if (args->csa_cachethis == 0)
332 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
333
334 /* The ca_maxresponsesize_cached is 0 with no DRC */
335 else if (args->csa_cachethis == 1)
336 return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
337 }
338
339 /* Wraparound */
340 if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
341 slot->seq_nr = 1;
342 goto out_ok;
343 }
344
345 /* Misordered request */
346 return htonl(NFS4ERR_SEQ_MISORDERED);
347out_ok:
348 tbl->highest_used_slotid = args->csa_slotid;
349 return htonl(NFS4_OK);
350}
351
352/*
353 * For each referring call triple, check the session's slot table for
354 * a match. If the slot is in use and the sequence numbers match, the
355 * client is still waiting for a response to the original request.
356 */
357static bool referring_call_exists(struct nfs_client *clp,
358 uint32_t nrclists,
359 struct referring_call_list *rclists)
360{
361 bool status = 0;
362 int i, j;
363 struct nfs4_session *session;
364 struct nfs4_slot_table *tbl;
365 struct referring_call_list *rclist;
366 struct referring_call *ref;
367
368 /*
369 * XXX When client trunking is implemented, this becomes
370 * a session lookup from within the loop
371 */
372 session = clp->cl_session;
373 tbl = &session->fc_slot_table;
374
375 for (i = 0; i < nrclists; i++) {
376 rclist = &rclists[i];
377 if (memcmp(session->sess_id.data,
378 rclist->rcl_sessionid.data,
379 NFS4_MAX_SESSIONID_LEN) != 0)
380 continue;
381
382 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
383 ref = &rclist->rcl_refcalls[j];
384
385 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
386 "slotid %u\n", __func__,
387 ((u32 *)&rclist->rcl_sessionid.data)[0],
388 ((u32 *)&rclist->rcl_sessionid.data)[1],
389 ((u32 *)&rclist->rcl_sessionid.data)[2],
390 ((u32 *)&rclist->rcl_sessionid.data)[3],
391 ref->rc_sequenceid, ref->rc_slotid);
392
393 spin_lock(&tbl->slot_tbl_lock);
394 status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
395 tbl->slots[ref->rc_slotid].seq_nr ==
396 ref->rc_sequenceid);
397 spin_unlock(&tbl->slot_tbl_lock);
398 if (status)
399 goto out;
400 }
401 }
402
403out:
404 return status;
405}
406
407__be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
408 struct cb_sequenceres *res,
409 struct cb_process_state *cps)
410{
411 struct nfs4_slot_table *tbl;
412 struct nfs_client *clp;
413 int i;
414 __be32 status = htonl(NFS4ERR_BADSESSION);
415
416 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
417 &args->csa_sessionid, cps->minorversion);
418 if (clp == NULL)
419 goto out;
420
421 tbl = &clp->cl_session->bc_slot_table;
422
423 spin_lock(&tbl->slot_tbl_lock);
424 /* state manager is resetting the session */
425 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
426 spin_unlock(&tbl->slot_tbl_lock);
427 status = htonl(NFS4ERR_DELAY);
428 /* Return NFS4ERR_BADSESSION if we're draining the session
429 * in order to reset it.
430 */
431 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
432 status = htonl(NFS4ERR_BADSESSION);
433 goto out;
434 }
435
436 status = validate_seqid(&clp->cl_session->bc_slot_table, args);
437 spin_unlock(&tbl->slot_tbl_lock);
438 if (status)
439 goto out;
440
441 cps->slotid = args->csa_slotid;
442
443 /*
444 * Check for pending referring calls. If a match is found, a
445 * related callback was received before the response to the original
446 * call.
447 */
448 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
449 status = htonl(NFS4ERR_DELAY);
450 goto out;
451 }
452
453 memcpy(&res->csr_sessionid, &args->csa_sessionid,
454 sizeof(res->csr_sessionid));
455 res->csr_sequenceid = args->csa_sequenceid;
456 res->csr_slotid = args->csa_slotid;
457 res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
458 res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
459
460out:
461 cps->clp = clp; /* put in nfs4_callback_compound */
462 for (i = 0; i < args->csa_nrclists; i++)
463 kfree(args->csa_rclists[i].rcl_refcalls);
464 kfree(args->csa_rclists);
465
466 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
467 cps->drc_status = status;
468 status = 0;
469 } else
470 res->csr_status = status;
471
472 trace_nfs4_cb_sequence(args, res, status);
473 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
474 ntohl(status), ntohl(res->csr_status));
475 return status;
476}
477
478static bool
479validate_bitmap_values(unsigned long mask)
480{
481 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
482}
483
484__be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
485 struct cb_process_state *cps)
486{
487 __be32 status;
488 fmode_t flags = 0;
489
490 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
491 if (!cps->clp) /* set in cb_sequence */
492 goto out;
493
494 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
495 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
496
497 status = cpu_to_be32(NFS4ERR_INVAL);
498 if (!validate_bitmap_values(args->craa_type_mask))
499 goto out;
500
501 status = cpu_to_be32(NFS4_OK);
502 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
503 &args->craa_type_mask))
504 flags = FMODE_READ;
505 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
506 &args->craa_type_mask))
507 flags |= FMODE_WRITE;
508 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
509 &args->craa_type_mask))
510 pnfs_recall_all_layouts(cps->clp);
511 if (flags)
512 nfs_expire_unused_delegation_types(cps->clp, flags);
513out:
514 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
515 return status;
516}
517
518/* Reduce the fore channel's max_slots to the target value */
519__be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
520 struct cb_process_state *cps)
521{
522 struct nfs4_slot_table *fc_tbl;
523 __be32 status;
524
525 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
526 if (!cps->clp) /* set in cb_sequence */
527 goto out;
528
529 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
530 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
531 args->crsa_target_highest_slotid);
532
533 fc_tbl = &cps->clp->cl_session->fc_slot_table;
534
535 status = htonl(NFS4_OK);
536
537 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
538 nfs41_server_notify_target_slotid_update(cps->clp);
539out:
540 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
541 return status;
542}
543#endif /* CONFIG_NFS_V4_1 */