xref: /openbmc/linux/fs/nfs/callback_proc.c (revision 171f1bc7)
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;
58 out_iput:
59 	rcu_read_unlock();
60 	iput(inode);
61 out:
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);
96 out:
97 	dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
98 	return res;
99 }
100 
101 int 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 
111 static 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 
162 static 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 
213 static 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);
226 out:
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 
248 static 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 
299 out:
300 	kfree(args->devs);
301 	dprintk("%s: exit with status = %u\n",
302 		__func__, be32_to_cpu(res));
303 	return res;
304 }
305 
306 int 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  */
334 static __be32
335 validate_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);
375 out_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  */
385 static 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 
431 out:
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 
487 out:
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 
504 static bool
505 validate_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);
539 out:
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);
572 out:
573 	dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
574 	return status;
575 }
576 #endif /* CONFIG_NFS_V4_1 */
577