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 "nfs4_fs.h"
12#include "callback.h"
13#include "delegation.h"
14#include "internal.h"
15#include "pnfs.h"
16
17#ifdef NFS_DEBUG
18#define NFSDBG_FACILITY NFSDBG_CALLBACK
19#endif
20
21__be32 nfs4_callback_getattr(struct cb_getattrargs *args,
22 struct cb_getattrres *res,
23 struct cb_process_state *cps)
24{
25 struct nfs_delegation *delegation;
26 struct nfs_inode *nfsi;
27 struct inode *inode;
28
29 res->status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
30 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
31 goto out;
32
33 res->bitmap[0] = res->bitmap[1] = 0;
34 res->status = htonl(NFS4ERR_BADHANDLE);
35
36 dprintk("NFS: GETATTR callback request from %s\n",
37 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
38
39 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
40 if (inode == NULL)
41 goto out;
42 nfsi = NFS_I(inode);
43 rcu_read_lock();
44 delegation = rcu_dereference(nfsi->delegation);
45 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
46 goto out_iput;
47 res->size = i_size_read(inode);
48 res->change_attr = delegation->change_attr;
49 if (nfsi->npages != 0)
50 res->change_attr++;
51 res->ctime = inode->i_ctime;
52 res->mtime = inode->i_mtime;
53 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
54 args->bitmap[0];
55 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
56 args->bitmap[1];
57 res->status = 0;
58out_iput:
59 rcu_read_unlock();
60 iput(inode);
61out:
62 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
63 return res->status;
64}
65
66__be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
67 struct cb_process_state *cps)
68{
69 struct inode *inode;
70 __be32 res;
71
72 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
73 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
74 goto out;
75
76 dprintk("NFS: RECALL callback request from %s\n",
77 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
78
79 res = htonl(NFS4ERR_BADHANDLE);
80 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
81 if (inode == NULL)
82 goto out;
83 /* Set up a helper thread to actually return the delegation */
84 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
85 case 0:
86 res = 0;
87 break;
88 case -ENOENT:
89 if (res != 0)
90 res = htonl(NFS4ERR_BAD_STATEID);
91 break;
92 default:
93 res = htonl(NFS4ERR_RESOURCE);
94 }
95 iput(inode);
96out:
97 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
98 return res;
99}
100
101int nfs4_validate_delegation_stateid(struct nfs_delegation *delegation, const nfs4_stateid *stateid)
102{
103 if (delegation == NULL || memcmp(delegation->stateid.data, stateid->data,
104 sizeof(delegation->stateid.data)) != 0)
105 return 0;
106 return 1;
107}
108
109#if defined(CONFIG_NFS_V4_1)
110
111static u32 initiate_file_draining(struct nfs_client *clp,
112 struct cb_layoutrecallargs *args)
113{
114 struct nfs_server *server;
115 struct pnfs_layout_hdr *lo;
116 struct inode *ino;
117 bool found = false;
118 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
119 LIST_HEAD(free_me_list);
120
121 spin_lock(&clp->cl_lock);
122 rcu_read_lock();
123 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
124 list_for_each_entry(lo, &server->layouts, plh_layouts) {
125 if (nfs_compare_fh(&args->cbl_fh,
126 &NFS_I(lo->plh_inode)->fh))
127 continue;
128 ino = igrab(lo->plh_inode);
129 if (!ino)
130 continue;
131 found = true;
132 /* Without this, layout can be freed as soon
133 * as we release cl_lock.
134 */
135 get_layout_hdr(lo);
136 break;
137 }
138 if (found)
139 break;
140 }
141 rcu_read_unlock();
142 spin_unlock(&clp->cl_lock);
143
144 if (!found)
145 return NFS4ERR_NOMATCHING_LAYOUT;
146
147 spin_lock(&ino->i_lock);
148 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
149 mark_matching_lsegs_invalid(lo, &free_me_list,
150 &args->cbl_range))
151 rv = NFS4ERR_DELAY;
152 else
153 rv = NFS4ERR_NOMATCHING_LAYOUT;
154 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
155 spin_unlock(&ino->i_lock);
156 pnfs_free_lseg_list(&free_me_list);
157 put_layout_hdr(lo);
158 iput(ino);
159 return rv;
160}
161
162static u32 initiate_bulk_draining(struct nfs_client *clp,
163 struct cb_layoutrecallargs *args)
164{
165 struct nfs_server *server;
166 struct pnfs_layout_hdr *lo;
167 struct inode *ino;
168 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
169 struct pnfs_layout_hdr *tmp;
170 LIST_HEAD(recall_list);
171 LIST_HEAD(free_me_list);
172 struct pnfs_layout_range range = {
173 .iomode = IOMODE_ANY,
174 .offset = 0,
175 .length = NFS4_MAX_UINT64,
176 };
177
178 spin_lock(&clp->cl_lock);
179 rcu_read_lock();
180 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
181 if ((args->cbl_recall_type == RETURN_FSID) &&
182 memcmp(&server->fsid, &args->cbl_fsid,
183 sizeof(struct nfs_fsid)))
184 continue;
185
186 list_for_each_entry(lo, &server->layouts, plh_layouts) {
187 if (!igrab(lo->plh_inode))
188 continue;
189 get_layout_hdr(lo);
190 BUG_ON(!list_empty(&lo->plh_bulk_recall));
191 list_add(&lo->plh_bulk_recall, &recall_list);
192 }
193 }
194 rcu_read_unlock();
195 spin_unlock(&clp->cl_lock);
196
197 list_for_each_entry_safe(lo, tmp,
198 &recall_list, plh_bulk_recall) {
199 ino = lo->plh_inode;
200 spin_lock(&ino->i_lock);
201 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
202 if (mark_matching_lsegs_invalid(lo, &free_me_list, &range))
203 rv = NFS4ERR_DELAY;
204 list_del_init(&lo->plh_bulk_recall);
205 spin_unlock(&ino->i_lock);
206 pnfs_free_lseg_list(&free_me_list);
207 put_layout_hdr(lo);
208 iput(ino);
209 }
210 return rv;
211}
212
213static u32 do_callback_layoutrecall(struct nfs_client *clp,
214 struct cb_layoutrecallargs *args)
215{
216 u32 res = NFS4ERR_DELAY;
217
218 dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
219 if (test_and_set_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state))
220 goto out;
221 if (args->cbl_recall_type == RETURN_FILE)
222 res = initiate_file_draining(clp, args);
223 else
224 res = initiate_bulk_draining(clp, args);
225 clear_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state);
226out:
227 dprintk("%s returning %i\n", __func__, res);
228 return res;
229
230}
231
232__be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
233 void *dummy, struct cb_process_state *cps)
234{
235 u32 res;
236
237 dprintk("%s: -->\n", __func__);
238
239 if (cps->clp)
240 res = do_callback_layoutrecall(cps->clp, args);
241 else
242 res = NFS4ERR_OP_NOT_IN_SESSION;
243
244 dprintk("%s: exit with status = %d\n", __func__, res);
245 return cpu_to_be32(res);
246}
247
248static void pnfs_recall_all_layouts(struct nfs_client *clp)
249{
250 struct cb_layoutrecallargs args;
251
252 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
253 memset(&args, 0, sizeof(args));
254 args.cbl_recall_type = RETURN_ALL;
255 /* FIXME we ignore errors, what should we do? */
256 do_callback_layoutrecall(clp, &args);
257}
258
259__be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
260 void *dummy, struct cb_process_state *cps)
261{
262 int i;
263 __be32 res = 0;
264 struct nfs_client *clp = cps->clp;
265 struct nfs_server *server = NULL;
266
267 dprintk("%s: -->\n", __func__);
268
269 if (!clp) {
270 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
271 goto out;
272 }
273
274 for (i = 0; i < args->ndevs; i++) {
275 struct cb_devicenotifyitem *dev = &args->devs[i];
276
277 if (!server ||
278 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
279 rcu_read_lock();
280 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
281 if (server->pnfs_curr_ld &&
282 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
283 rcu_read_unlock();
284 goto found;
285 }
286 rcu_read_unlock();
287 dprintk("%s: layout type %u not found\n",
288 __func__, dev->cbd_layout_type);
289 continue;
290 }
291
292 found:
293 if (dev->cbd_notify_type == NOTIFY_DEVICEID4_CHANGE)
294 dprintk("%s: NOTIFY_DEVICEID4_CHANGE not supported, "
295 "deleting instead\n", __func__);
296 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
297 }
298
299out:
300 kfree(args->devs);
301 dprintk("%s: exit with status = %u\n",
302 __func__, be32_to_cpu(res));
303 return res;
304}
305
306int nfs41_validate_delegation_stateid(struct nfs_delegation *delegation, const nfs4_stateid *stateid)
307{
308 if (delegation == NULL)
309 return 0;
310
311 if (stateid->stateid.seqid != 0)
312 return 0;
313 if (memcmp(&delegation->stateid.stateid.other,
314 &stateid->stateid.other,
315 NFS4_STATEID_OTHER_SIZE))
316 return 0;
317
318 return 1;
319}
320
321/*
322 * Validate the sequenceID sent by the server.
323 * Return success if the sequenceID is one more than what we last saw on
324 * this slot, accounting for wraparound. Increments the slot's sequence.
325 *
326 * We don't yet implement a duplicate request cache, instead we set the
327 * back channel ca_maxresponsesize_cached to zero. This is OK for now
328 * since we only currently implement idempotent callbacks anyway.
329 *
330 * We have a single slot backchannel at this time, so we don't bother
331 * checking the used_slots bit array on the table. The lower layer guarantees
332 * a single outstanding callback request at a time.
333 */
334static __be32
335validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
336{
337 struct nfs4_slot *slot;
338
339 dprintk("%s enter. slotid %d seqid %d\n",
340 __func__, args->csa_slotid, args->csa_sequenceid);
341
342 if (args->csa_slotid > NFS41_BC_MAX_CALLBACKS)
343 return htonl(NFS4ERR_BADSLOT);
344
345 slot = tbl->slots + args->csa_slotid;
346 dprintk("%s slot table seqid: %d\n", __func__, slot->seq_nr);
347
348 /* Normal */
349 if (likely(args->csa_sequenceid == slot->seq_nr + 1)) {
350 slot->seq_nr++;
351 goto out_ok;
352 }
353
354 /* Replay */
355 if (args->csa_sequenceid == slot->seq_nr) {
356 dprintk("%s seqid %d is a replay\n",
357 __func__, args->csa_sequenceid);
358 /* Signal process_op to set this error on next op */
359 if (args->csa_cachethis == 0)
360 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
361
362 /* The ca_maxresponsesize_cached is 0 with no DRC */
363 else if (args->csa_cachethis == 1)
364 return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
365 }
366
367 /* Wraparound */
368 if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
369 slot->seq_nr = 1;
370 goto out_ok;
371 }
372
373 /* Misordered request */
374 return htonl(NFS4ERR_SEQ_MISORDERED);
375out_ok:
376 tbl->highest_used_slotid = args->csa_slotid;
377 return htonl(NFS4_OK);
378}
379
380/*
381 * For each referring call triple, check the session's slot table for
382 * a match. If the slot is in use and the sequence numbers match, the
383 * client is still waiting for a response to the original request.
384 */
385static bool referring_call_exists(struct nfs_client *clp,
386 uint32_t nrclists,
387 struct referring_call_list *rclists)
388{
389 bool status = 0;
390 int i, j;
391 struct nfs4_session *session;
392 struct nfs4_slot_table *tbl;
393 struct referring_call_list *rclist;
394 struct referring_call *ref;
395
396 /*
397 * XXX When client trunking is implemented, this becomes
398 * a session lookup from within the loop
399 */
400 session = clp->cl_session;
401 tbl = &session->fc_slot_table;
402
403 for (i = 0; i < nrclists; i++) {
404 rclist = &rclists[i];
405 if (memcmp(session->sess_id.data,
406 rclist->rcl_sessionid.data,
407 NFS4_MAX_SESSIONID_LEN) != 0)
408 continue;
409
410 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
411 ref = &rclist->rcl_refcalls[j];
412
413 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
414 "slotid %u\n", __func__,
415 ((u32 *)&rclist->rcl_sessionid.data)[0],
416 ((u32 *)&rclist->rcl_sessionid.data)[1],
417 ((u32 *)&rclist->rcl_sessionid.data)[2],
418 ((u32 *)&rclist->rcl_sessionid.data)[3],
419 ref->rc_sequenceid, ref->rc_slotid);
420
421 spin_lock(&tbl->slot_tbl_lock);
422 status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
423 tbl->slots[ref->rc_slotid].seq_nr ==
424 ref->rc_sequenceid);
425 spin_unlock(&tbl->slot_tbl_lock);
426 if (status)
427 goto out;
428 }
429 }
430
431out:
432 return status;
433}
434
435__be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
436 struct cb_sequenceres *res,
437 struct cb_process_state *cps)
438{
439 struct nfs4_slot_table *tbl;
440 struct nfs_client *clp;
441 int i;
442 __be32 status = htonl(NFS4ERR_BADSESSION);
443
444 clp = nfs4_find_client_sessionid(args->csa_addr, &args->csa_sessionid);
445 if (clp == NULL)
446 goto out;
447
448 tbl = &clp->cl_session->bc_slot_table;
449
450 spin_lock(&tbl->slot_tbl_lock);
451 /* state manager is resetting the session */
452 if (test_bit(NFS4_SESSION_DRAINING, &clp->cl_session->session_state)) {
453 spin_unlock(&tbl->slot_tbl_lock);
454 status = htonl(NFS4ERR_DELAY);
455 /* Return NFS4ERR_BADSESSION if we're draining the session
456 * in order to reset it.
457 */
458 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
459 status = htonl(NFS4ERR_BADSESSION);
460 goto out;
461 }
462
463 status = validate_seqid(&clp->cl_session->bc_slot_table, args);
464 spin_unlock(&tbl->slot_tbl_lock);
465 if (status)
466 goto out;
467
468 cps->slotid = args->csa_slotid;
469
470 /*
471 * Check for pending referring calls. If a match is found, a
472 * related callback was received before the response to the original
473 * call.
474 */
475 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
476 status = htonl(NFS4ERR_DELAY);
477 goto out;
478 }
479
480 memcpy(&res->csr_sessionid, &args->csa_sessionid,
481 sizeof(res->csr_sessionid));
482 res->csr_sequenceid = args->csa_sequenceid;
483 res->csr_slotid = args->csa_slotid;
484 res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
485 res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
486
487out:
488 cps->clp = clp; /* put in nfs4_callback_compound */
489 for (i = 0; i < args->csa_nrclists; i++)
490 kfree(args->csa_rclists[i].rcl_refcalls);
491 kfree(args->csa_rclists);
492
493 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
494 cps->drc_status = status;
495 status = 0;
496 } else
497 res->csr_status = status;
498
499 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
500 ntohl(status), ntohl(res->csr_status));
501 return status;
502}
503
504static bool
505validate_bitmap_values(unsigned long mask)
506{
507 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
508}
509
510__be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
511 struct cb_process_state *cps)
512{
513 __be32 status;
514 fmode_t flags = 0;
515
516 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
517 if (!cps->clp) /* set in cb_sequence */
518 goto out;
519
520 dprintk("NFS: RECALL_ANY callback request from %s\n",
521 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
522
523 status = cpu_to_be32(NFS4ERR_INVAL);
524 if (!validate_bitmap_values(args->craa_type_mask))
525 goto out;
526
527 status = cpu_to_be32(NFS4_OK);
528 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
529 &args->craa_type_mask))
530 flags = FMODE_READ;
531 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
532 &args->craa_type_mask))
533 flags |= FMODE_WRITE;
534 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
535 &args->craa_type_mask))
536 pnfs_recall_all_layouts(cps->clp);
537 if (flags)
538 nfs_expire_all_delegation_types(cps->clp, flags);
539out:
540 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
541 return status;
542}
543
544/* Reduce the fore channel's max_slots to the target value */
545__be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
546 struct cb_process_state *cps)
547{
548 struct nfs4_slot_table *fc_tbl;
549 __be32 status;
550
551 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
552 if (!cps->clp) /* set in cb_sequence */
553 goto out;
554
555 dprintk("NFS: CB_RECALL_SLOT request from %s target max slots %d\n",
556 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
557 args->crsa_target_max_slots);
558
559 fc_tbl = &cps->clp->cl_session->fc_slot_table;
560
561 status = htonl(NFS4ERR_BAD_HIGH_SLOT);
562 if (args->crsa_target_max_slots > fc_tbl->max_slots ||
563 args->crsa_target_max_slots < 1)
564 goto out;
565
566 status = htonl(NFS4_OK);
567 if (args->crsa_target_max_slots == fc_tbl->max_slots)
568 goto out;
569
570 fc_tbl->target_max_slots = args->crsa_target_max_slots;
571 nfs41_handle_recall_slot(cps->clp);
572out:
573 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
574 return status;
575}
576#endif /* CONFIG_NFS_V4_1 */
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 by filehandle.
114 *
115 * Note: gets a refcount on the layout hdr and on its respective inode.
116 * Caller must put the layout hdr and the inode.
117 *
118 * TODO: keep track of all layouts (and delegations) in a hash table
119 * hashed by filehandle.
120 */
121static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp,
122 struct nfs_fh *fh, nfs4_stateid *stateid)
123{
124 struct nfs_server *server;
125 struct inode *ino;
126 struct pnfs_layout_hdr *lo;
127
128 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
129 list_for_each_entry(lo, &server->layouts, plh_layouts) {
130 if (!nfs4_stateid_match_other(&lo->plh_stateid, stateid))
131 continue;
132 if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
133 continue;
134 ino = igrab(lo->plh_inode);
135 if (!ino)
136 break;
137 spin_lock(&ino->i_lock);
138 /* Is this layout in the process of being freed? */
139 if (NFS_I(ino)->layout != lo) {
140 spin_unlock(&ino->i_lock);
141 iput(ino);
142 break;
143 }
144 pnfs_get_layout_hdr(lo);
145 spin_unlock(&ino->i_lock);
146 return lo;
147 }
148 }
149
150 return NULL;
151}
152
153static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp,
154 struct nfs_fh *fh, nfs4_stateid *stateid)
155{
156 struct pnfs_layout_hdr *lo;
157
158 spin_lock(&clp->cl_lock);
159 rcu_read_lock();
160 lo = get_layout_by_fh_locked(clp, fh, stateid);
161 rcu_read_unlock();
162 spin_unlock(&clp->cl_lock);
163
164 return lo;
165}
166
167/*
168 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
169 */
170static bool pnfs_check_stateid_sequence(struct pnfs_layout_hdr *lo,
171 const nfs4_stateid *new)
172{
173 u32 oldseq, newseq;
174
175 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
176 newseq = be32_to_cpu(new->seqid);
177
178 if (newseq > oldseq + 1)
179 return false;
180 return true;
181}
182
183static u32 initiate_file_draining(struct nfs_client *clp,
184 struct cb_layoutrecallargs *args)
185{
186 struct inode *ino;
187 struct pnfs_layout_hdr *lo;
188 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
189 LIST_HEAD(free_me_list);
190
191 lo = get_layout_by_fh(clp, &args->cbl_fh, &args->cbl_stateid);
192 if (!lo) {
193 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, NULL,
194 &args->cbl_stateid, -rv);
195 goto out;
196 }
197
198 ino = lo->plh_inode;
199
200 spin_lock(&ino->i_lock);
201 if (!pnfs_check_stateid_sequence(lo, &args->cbl_stateid)) {
202 rv = NFS4ERR_DELAY;
203 goto unlock;
204 }
205 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
206 spin_unlock(&ino->i_lock);
207
208 pnfs_layoutcommit_inode(ino, false);
209
210 spin_lock(&ino->i_lock);
211 /*
212 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
213 */
214 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
215 rv = NFS4ERR_DELAY;
216 goto unlock;
217 }
218
219 if (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
220 &args->cbl_range)) {
221 rv = NFS4_OK;
222 goto unlock;
223 }
224
225 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
226 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
227 &args->cbl_range);
228 }
229 pnfs_mark_layout_returned_if_empty(lo);
230unlock:
231 spin_unlock(&ino->i_lock);
232 pnfs_free_lseg_list(&free_me_list);
233 /* Free all lsegs that are attached to commit buckets */
234 nfs_commit_inode(ino, 0);
235 pnfs_put_layout_hdr(lo);
236 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
237 &args->cbl_stateid, -rv);
238 iput(ino);
239out:
240 return rv;
241}
242
243static u32 initiate_bulk_draining(struct nfs_client *clp,
244 struct cb_layoutrecallargs *args)
245{
246 int stat;
247
248 if (args->cbl_recall_type == RETURN_FSID)
249 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
250 else
251 stat = pnfs_destroy_layouts_byclid(clp, true);
252 if (stat != 0)
253 return NFS4ERR_DELAY;
254 return NFS4ERR_NOMATCHING_LAYOUT;
255}
256
257static u32 do_callback_layoutrecall(struct nfs_client *clp,
258 struct cb_layoutrecallargs *args)
259{
260 u32 res;
261
262 dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
263 if (args->cbl_recall_type == RETURN_FILE)
264 res = initiate_file_draining(clp, args);
265 else
266 res = initiate_bulk_draining(clp, args);
267 dprintk("%s returning %i\n", __func__, res);
268 return res;
269
270}
271
272__be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
273 void *dummy, struct cb_process_state *cps)
274{
275 u32 res;
276
277 dprintk("%s: -->\n", __func__);
278
279 if (cps->clp)
280 res = do_callback_layoutrecall(cps->clp, args);
281 else
282 res = NFS4ERR_OP_NOT_IN_SESSION;
283
284 dprintk("%s: exit with status = %d\n", __func__, res);
285 return cpu_to_be32(res);
286}
287
288static void pnfs_recall_all_layouts(struct nfs_client *clp)
289{
290 struct cb_layoutrecallargs args;
291
292 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
293 memset(&args, 0, sizeof(args));
294 args.cbl_recall_type = RETURN_ALL;
295 /* FIXME we ignore errors, what should we do? */
296 do_callback_layoutrecall(clp, &args);
297}
298
299__be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
300 void *dummy, struct cb_process_state *cps)
301{
302 int i;
303 __be32 res = 0;
304 struct nfs_client *clp = cps->clp;
305 struct nfs_server *server = NULL;
306
307 dprintk("%s: -->\n", __func__);
308
309 if (!clp) {
310 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
311 goto out;
312 }
313
314 for (i = 0; i < args->ndevs; i++) {
315 struct cb_devicenotifyitem *dev = &args->devs[i];
316
317 if (!server ||
318 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
319 rcu_read_lock();
320 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
321 if (server->pnfs_curr_ld &&
322 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
323 rcu_read_unlock();
324 goto found;
325 }
326 rcu_read_unlock();
327 dprintk("%s: layout type %u not found\n",
328 __func__, dev->cbd_layout_type);
329 continue;
330 }
331
332 found:
333 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
334 }
335
336out:
337 kfree(args->devs);
338 dprintk("%s: exit with status = %u\n",
339 __func__, be32_to_cpu(res));
340 return res;
341}
342
343/*
344 * Validate the sequenceID sent by the server.
345 * Return success if the sequenceID is one more than what we last saw on
346 * this slot, accounting for wraparound. Increments the slot's sequence.
347 *
348 * We don't yet implement a duplicate request cache, instead we set the
349 * back channel ca_maxresponsesize_cached to zero. This is OK for now
350 * since we only currently implement idempotent callbacks anyway.
351 *
352 * We have a single slot backchannel at this time, so we don't bother
353 * checking the used_slots bit array on the table. The lower layer guarantees
354 * a single outstanding callback request at a time.
355 */
356static __be32
357validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
358 const struct cb_sequenceargs * args)
359{
360 dprintk("%s enter. slotid %u seqid %u, slot table seqid: %u\n",
361 __func__, args->csa_slotid, args->csa_sequenceid, slot->seq_nr);
362
363 if (args->csa_slotid > tbl->server_highest_slotid)
364 return htonl(NFS4ERR_BADSLOT);
365
366 /* Replay */
367 if (args->csa_sequenceid == slot->seq_nr) {
368 dprintk("%s seqid %u is a replay\n",
369 __func__, args->csa_sequenceid);
370 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
371 return htonl(NFS4ERR_DELAY);
372 /* Signal process_op to set this error on next op */
373 if (args->csa_cachethis == 0)
374 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
375
376 /* Liar! We never allowed you to set csa_cachethis != 0 */
377 return htonl(NFS4ERR_SEQ_FALSE_RETRY);
378 }
379
380 /* Wraparound */
381 if (unlikely(slot->seq_nr == 0xFFFFFFFFU)) {
382 if (args->csa_sequenceid == 1)
383 return htonl(NFS4_OK);
384 } else if (likely(args->csa_sequenceid == slot->seq_nr + 1))
385 return htonl(NFS4_OK);
386
387 /* Misordered request */
388 return htonl(NFS4ERR_SEQ_MISORDERED);
389}
390
391/*
392 * For each referring call triple, check the session's slot table for
393 * a match. If the slot is in use and the sequence numbers match, the
394 * client is still waiting for a response to the original request.
395 */
396static bool referring_call_exists(struct nfs_client *clp,
397 uint32_t nrclists,
398 struct referring_call_list *rclists)
399{
400 bool status = 0;
401 int i, j;
402 struct nfs4_session *session;
403 struct nfs4_slot_table *tbl;
404 struct referring_call_list *rclist;
405 struct referring_call *ref;
406
407 /*
408 * XXX When client trunking is implemented, this becomes
409 * a session lookup from within the loop
410 */
411 session = clp->cl_session;
412 tbl = &session->fc_slot_table;
413
414 for (i = 0; i < nrclists; i++) {
415 rclist = &rclists[i];
416 if (memcmp(session->sess_id.data,
417 rclist->rcl_sessionid.data,
418 NFS4_MAX_SESSIONID_LEN) != 0)
419 continue;
420
421 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
422 ref = &rclist->rcl_refcalls[j];
423
424 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
425 "slotid %u\n", __func__,
426 ((u32 *)&rclist->rcl_sessionid.data)[0],
427 ((u32 *)&rclist->rcl_sessionid.data)[1],
428 ((u32 *)&rclist->rcl_sessionid.data)[2],
429 ((u32 *)&rclist->rcl_sessionid.data)[3],
430 ref->rc_sequenceid, ref->rc_slotid);
431
432 spin_lock(&tbl->slot_tbl_lock);
433 status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
434 tbl->slots[ref->rc_slotid].seq_nr ==
435 ref->rc_sequenceid);
436 spin_unlock(&tbl->slot_tbl_lock);
437 if (status)
438 goto out;
439 }
440 }
441
442out:
443 return status;
444}
445
446__be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
447 struct cb_sequenceres *res,
448 struct cb_process_state *cps)
449{
450 struct nfs4_slot_table *tbl;
451 struct nfs4_slot *slot;
452 struct nfs_client *clp;
453 int i;
454 __be32 status = htonl(NFS4ERR_BADSESSION);
455
456 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
457 &args->csa_sessionid, cps->minorversion);
458 if (clp == NULL)
459 goto out;
460
461 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
462 goto out;
463
464 tbl = &clp->cl_session->bc_slot_table;
465 slot = tbl->slots + args->csa_slotid;
466
467 /* Set up res before grabbing the spinlock */
468 memcpy(&res->csr_sessionid, &args->csa_sessionid,
469 sizeof(res->csr_sessionid));
470 res->csr_sequenceid = args->csa_sequenceid;
471 res->csr_slotid = args->csa_slotid;
472
473 spin_lock(&tbl->slot_tbl_lock);
474 /* state manager is resetting the session */
475 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
476 status = htonl(NFS4ERR_DELAY);
477 /* Return NFS4ERR_BADSESSION if we're draining the session
478 * in order to reset it.
479 */
480 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
481 status = htonl(NFS4ERR_BADSESSION);
482 goto out_unlock;
483 }
484
485 status = htonl(NFS4ERR_BADSLOT);
486 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
487 if (IS_ERR(slot))
488 goto out_unlock;
489
490 res->csr_highestslotid = tbl->server_highest_slotid;
491 res->csr_target_highestslotid = tbl->target_highest_slotid;
492
493 status = validate_seqid(tbl, slot, args);
494 if (status)
495 goto out_unlock;
496 if (!nfs4_try_to_lock_slot(tbl, slot)) {
497 status = htonl(NFS4ERR_DELAY);
498 goto out_unlock;
499 }
500 cps->slot = slot;
501
502 /* The ca_maxresponsesize_cached is 0 with no DRC */
503 if (args->csa_cachethis != 0)
504 return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
505
506 /*
507 * Check for pending referring calls. If a match is found, a
508 * related callback was received before the response to the original
509 * call.
510 */
511 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
512 status = htonl(NFS4ERR_DELAY);
513 goto out_unlock;
514 }
515
516 /*
517 * RFC5661 20.9.3
518 * If CB_SEQUENCE returns an error, then the state of the slot
519 * (sequence ID, cached reply) MUST NOT change.
520 */
521 slot->seq_nr = args->csa_sequenceid;
522out_unlock:
523 spin_unlock(&tbl->slot_tbl_lock);
524
525out:
526 cps->clp = clp; /* put in nfs4_callback_compound */
527 for (i = 0; i < args->csa_nrclists; i++)
528 kfree(args->csa_rclists[i].rcl_refcalls);
529 kfree(args->csa_rclists);
530
531 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
532 cps->drc_status = status;
533 status = 0;
534 } else
535 res->csr_status = status;
536
537 trace_nfs4_cb_sequence(args, res, status);
538 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
539 ntohl(status), ntohl(res->csr_status));
540 return status;
541}
542
543static bool
544validate_bitmap_values(unsigned long mask)
545{
546 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
547}
548
549__be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
550 struct cb_process_state *cps)
551{
552 __be32 status;
553 fmode_t flags = 0;
554
555 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
556 if (!cps->clp) /* set in cb_sequence */
557 goto out;
558
559 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
560 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
561
562 status = cpu_to_be32(NFS4ERR_INVAL);
563 if (!validate_bitmap_values(args->craa_type_mask))
564 goto out;
565
566 status = cpu_to_be32(NFS4_OK);
567 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
568 &args->craa_type_mask))
569 flags = FMODE_READ;
570 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
571 &args->craa_type_mask))
572 flags |= FMODE_WRITE;
573 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
574 &args->craa_type_mask))
575 pnfs_recall_all_layouts(cps->clp);
576 if (flags)
577 nfs_expire_unused_delegation_types(cps->clp, flags);
578out:
579 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
580 return status;
581}
582
583/* Reduce the fore channel's max_slots to the target value */
584__be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
585 struct cb_process_state *cps)
586{
587 struct nfs4_slot_table *fc_tbl;
588 __be32 status;
589
590 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
591 if (!cps->clp) /* set in cb_sequence */
592 goto out;
593
594 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
595 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
596 args->crsa_target_highest_slotid);
597
598 fc_tbl = &cps->clp->cl_session->fc_slot_table;
599
600 status = htonl(NFS4_OK);
601
602 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
603 nfs41_notify_server(cps->clp);
604out:
605 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
606 return status;
607}
608#endif /* CONFIG_NFS_V4_1 */