locking/refcounts: Include fewer headers in <linux/refcount.h>
[linux/fpc-iii.git] / fs / nfs / callback_proc.c
blob64c214fb9da67d5ba0ce6fb732a6c3df963823cf
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/nfs/callback_proc.c
5 * Copyright (C) 2004 Trond Myklebust
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"
21 #define NFSDBG_FACILITY NFSDBG_CALLBACK
23 __be32 nfs4_callback_getattr(void *argp, void *resp,
24 struct cb_process_state *cps)
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;
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;
36 res->bitmap[0] = res->bitmap[1] = 0;
37 res->status = htonl(NFS4ERR_BADHANDLE);
39 dprintk_rcu("NFS: GETATTR callback request from %s\n",
40 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
42 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
43 if (IS_ERR(inode)) {
44 if (inode == ERR_PTR(-EAGAIN))
45 res->status = htonl(NFS4ERR_DELAY);
46 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
47 -ntohl(res->status));
48 goto out;
50 nfsi = NFS_I(inode);
51 rcu_read_lock();
52 delegation = rcu_dereference(nfsi->delegation);
53 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
54 goto out_iput;
55 res->size = i_size_read(inode);
56 res->change_attr = delegation->change_attr;
57 if (nfs_have_writebacks(inode))
58 res->change_attr++;
59 res->ctime = timespec64_to_timespec(inode->i_ctime);
60 res->mtime = timespec64_to_timespec(inode->i_mtime);
61 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
62 args->bitmap[0];
63 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
64 args->bitmap[1];
65 res->status = 0;
66 out_iput:
67 rcu_read_unlock();
68 trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
69 iput(inode);
70 out:
71 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
72 return res->status;
75 __be32 nfs4_callback_recall(void *argp, void *resp,
76 struct cb_process_state *cps)
78 struct cb_recallargs *args = argp;
79 struct inode *inode;
80 __be32 res;
82 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
83 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
84 goto out;
86 dprintk_rcu("NFS: RECALL callback request from %s\n",
87 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
89 res = htonl(NFS4ERR_BADHANDLE);
90 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
91 if (IS_ERR(inode)) {
92 if (inode == ERR_PTR(-EAGAIN))
93 res = htonl(NFS4ERR_DELAY);
94 trace_nfs4_cb_recall(cps->clp, &args->fh, NULL,
95 &args->stateid, -ntohl(res));
96 goto out;
98 /* Set up a helper thread to actually return the delegation */
99 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
100 case 0:
101 res = 0;
102 break;
103 case -ENOENT:
104 res = htonl(NFS4ERR_BAD_STATEID);
105 break;
106 default:
107 res = htonl(NFS4ERR_RESOURCE);
109 trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
110 &args->stateid, -ntohl(res));
111 iput(inode);
112 out:
113 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
114 return res;
117 #if defined(CONFIG_NFS_V4_1)
120 * Lookup a layout inode by stateid
122 * Note: returns a refcount on the inode and superblock
124 static struct inode *nfs_layout_find_inode_by_stateid(struct nfs_client *clp,
125 const nfs4_stateid *stateid)
127 struct nfs_server *server;
128 struct inode *inode;
129 struct pnfs_layout_hdr *lo;
131 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
132 list_for_each_entry(lo, &server->layouts, plh_layouts) {
133 if (stateid != NULL &&
134 !nfs4_stateid_match_other(stateid, &lo->plh_stateid))
135 continue;
136 inode = igrab(lo->plh_inode);
137 if (!inode)
138 return ERR_PTR(-EAGAIN);
139 if (!nfs_sb_active(inode->i_sb)) {
140 rcu_read_unlock();
141 spin_unlock(&clp->cl_lock);
142 iput(inode);
143 spin_lock(&clp->cl_lock);
144 rcu_read_lock();
145 return ERR_PTR(-EAGAIN);
147 return inode;
151 return ERR_PTR(-ENOENT);
155 * Lookup a layout inode by filehandle.
157 * Note: returns a refcount on the inode and superblock
160 static struct inode *nfs_layout_find_inode_by_fh(struct nfs_client *clp,
161 const struct nfs_fh *fh)
163 struct nfs_server *server;
164 struct nfs_inode *nfsi;
165 struct inode *inode;
166 struct pnfs_layout_hdr *lo;
168 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
169 list_for_each_entry(lo, &server->layouts, plh_layouts) {
170 nfsi = NFS_I(lo->plh_inode);
171 if (nfs_compare_fh(fh, &nfsi->fh))
172 continue;
173 if (nfsi->layout != lo)
174 continue;
175 inode = igrab(lo->plh_inode);
176 if (!inode)
177 return ERR_PTR(-EAGAIN);
178 if (!nfs_sb_active(inode->i_sb)) {
179 rcu_read_unlock();
180 spin_unlock(&clp->cl_lock);
181 iput(inode);
182 spin_lock(&clp->cl_lock);
183 rcu_read_lock();
184 return ERR_PTR(-EAGAIN);
186 return inode;
190 return ERR_PTR(-ENOENT);
193 static struct inode *nfs_layout_find_inode(struct nfs_client *clp,
194 const struct nfs_fh *fh,
195 const nfs4_stateid *stateid)
197 struct inode *inode;
199 spin_lock(&clp->cl_lock);
200 rcu_read_lock();
201 inode = nfs_layout_find_inode_by_stateid(clp, stateid);
202 if (inode == ERR_PTR(-ENOENT))
203 inode = nfs_layout_find_inode_by_fh(clp, fh);
204 rcu_read_unlock();
205 spin_unlock(&clp->cl_lock);
207 return inode;
211 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
213 static u32 pnfs_check_callback_stateid(struct pnfs_layout_hdr *lo,
214 const nfs4_stateid *new)
216 u32 oldseq, newseq;
218 /* Is the stateid still not initialised? */
219 if (!pnfs_layout_is_valid(lo))
220 return NFS4ERR_DELAY;
222 /* Mismatched stateid? */
223 if (!nfs4_stateid_match_other(&lo->plh_stateid, new))
224 return NFS4ERR_BAD_STATEID;
226 newseq = be32_to_cpu(new->seqid);
227 /* Are we already in a layout recall situation? */
228 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
229 lo->plh_return_seq != 0) {
230 if (newseq < lo->plh_return_seq)
231 return NFS4ERR_OLD_STATEID;
232 if (newseq > lo->plh_return_seq)
233 return NFS4ERR_DELAY;
234 goto out;
237 /* Check that the stateid matches what we think it should be. */
238 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
239 if (newseq > oldseq + 1)
240 return NFS4ERR_DELAY;
241 /* Crazy server! */
242 if (newseq <= oldseq)
243 return NFS4ERR_OLD_STATEID;
244 out:
245 return NFS_OK;
248 static u32 initiate_file_draining(struct nfs_client *clp,
249 struct cb_layoutrecallargs *args)
251 struct inode *ino;
252 struct pnfs_layout_hdr *lo;
253 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
254 LIST_HEAD(free_me_list);
256 ino = nfs_layout_find_inode(clp, &args->cbl_fh, &args->cbl_stateid);
257 if (IS_ERR(ino)) {
258 if (ino == ERR_PTR(-EAGAIN))
259 rv = NFS4ERR_DELAY;
260 goto out_noput;
263 pnfs_layoutcommit_inode(ino, false);
266 spin_lock(&ino->i_lock);
267 lo = NFS_I(ino)->layout;
268 if (!lo) {
269 spin_unlock(&ino->i_lock);
270 goto out;
272 pnfs_get_layout_hdr(lo);
273 rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid);
274 if (rv != NFS_OK)
275 goto unlock;
276 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
279 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
281 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
282 rv = NFS4ERR_DELAY;
283 goto unlock;
286 if (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
287 &args->cbl_range,
288 be32_to_cpu(args->cbl_stateid.seqid))) {
289 rv = NFS4_OK;
290 goto unlock;
293 /* Embrace your forgetfulness! */
294 rv = NFS4ERR_NOMATCHING_LAYOUT;
296 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
297 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
298 &args->cbl_range);
300 unlock:
301 spin_unlock(&ino->i_lock);
302 pnfs_free_lseg_list(&free_me_list);
303 /* Free all lsegs that are attached to commit buckets */
304 nfs_commit_inode(ino, 0);
305 pnfs_put_layout_hdr(lo);
306 out:
307 nfs_iput_and_deactive(ino);
308 out_noput:
309 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
310 &args->cbl_stateid, -rv);
311 return rv;
314 static u32 initiate_bulk_draining(struct nfs_client *clp,
315 struct cb_layoutrecallargs *args)
317 int stat;
319 if (args->cbl_recall_type == RETURN_FSID)
320 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
321 else
322 stat = pnfs_destroy_layouts_byclid(clp, true);
323 if (stat != 0)
324 return NFS4ERR_DELAY;
325 return NFS4ERR_NOMATCHING_LAYOUT;
328 static u32 do_callback_layoutrecall(struct nfs_client *clp,
329 struct cb_layoutrecallargs *args)
331 write_seqcount_begin(&clp->cl_callback_count);
332 write_seqcount_end(&clp->cl_callback_count);
333 if (args->cbl_recall_type == RETURN_FILE)
334 return initiate_file_draining(clp, args);
335 return initiate_bulk_draining(clp, args);
338 __be32 nfs4_callback_layoutrecall(void *argp, void *resp,
339 struct cb_process_state *cps)
341 struct cb_layoutrecallargs *args = argp;
342 u32 res = NFS4ERR_OP_NOT_IN_SESSION;
344 if (cps->clp)
345 res = do_callback_layoutrecall(cps->clp, args);
346 return cpu_to_be32(res);
349 static void pnfs_recall_all_layouts(struct nfs_client *clp)
351 struct cb_layoutrecallargs args;
353 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
354 memset(&args, 0, sizeof(args));
355 args.cbl_recall_type = RETURN_ALL;
356 /* FIXME we ignore errors, what should we do? */
357 do_callback_layoutrecall(clp, &args);
360 __be32 nfs4_callback_devicenotify(void *argp, void *resp,
361 struct cb_process_state *cps)
363 struct cb_devicenotifyargs *args = argp;
364 int i;
365 __be32 res = 0;
366 struct nfs_client *clp = cps->clp;
367 struct nfs_server *server = NULL;
369 if (!clp) {
370 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
371 goto out;
374 for (i = 0; i < args->ndevs; i++) {
375 struct cb_devicenotifyitem *dev = &args->devs[i];
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;
386 rcu_read_unlock();
387 continue;
390 found:
391 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
394 out:
395 kfree(args->devs);
396 return res;
400 * Validate the sequenceID sent by the server.
401 * Return success if the sequenceID is one more than what we last saw on
402 * this slot, accounting for wraparound. Increments the slot's sequence.
404 * We don't yet implement a duplicate request cache, instead we set the
405 * back channel ca_maxresponsesize_cached to zero. This is OK for now
406 * since we only currently implement idempotent callbacks anyway.
408 * We have a single slot backchannel at this time, so we don't bother
409 * checking the used_slots bit array on the table. The lower layer guarantees
410 * a single outstanding callback request at a time.
412 static __be32
413 validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
414 const struct cb_sequenceargs * args)
416 if (args->csa_slotid > tbl->server_highest_slotid)
417 return htonl(NFS4ERR_BADSLOT);
419 /* Replay */
420 if (args->csa_sequenceid == slot->seq_nr) {
421 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
422 return htonl(NFS4ERR_DELAY);
423 /* Signal process_op to set this error on next op */
424 if (args->csa_cachethis == 0)
425 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
427 /* Liar! We never allowed you to set csa_cachethis != 0 */
428 return htonl(NFS4ERR_SEQ_FALSE_RETRY);
431 /* Note: wraparound relies on seq_nr being of type u32 */
432 if (likely(args->csa_sequenceid == slot->seq_nr + 1))
433 return htonl(NFS4_OK);
435 /* Misordered request */
436 return htonl(NFS4ERR_SEQ_MISORDERED);
440 * For each referring call triple, check the session's slot table for
441 * a match. If the slot is in use and the sequence numbers match, the
442 * client is still waiting for a response to the original request.
444 static bool referring_call_exists(struct nfs_client *clp,
445 uint32_t nrclists,
446 struct referring_call_list *rclists)
448 bool status = false;
449 int i, j;
450 struct nfs4_session *session;
451 struct nfs4_slot_table *tbl;
452 struct referring_call_list *rclist;
453 struct referring_call *ref;
456 * XXX When client trunking is implemented, this becomes
457 * a session lookup from within the loop
459 session = clp->cl_session;
460 tbl = &session->fc_slot_table;
462 for (i = 0; i < nrclists; i++) {
463 rclist = &rclists[i];
464 if (memcmp(session->sess_id.data,
465 rclist->rcl_sessionid.data,
466 NFS4_MAX_SESSIONID_LEN) != 0)
467 continue;
469 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
470 ref = &rclist->rcl_refcalls[j];
471 status = nfs4_slot_wait_on_seqid(tbl, ref->rc_slotid,
472 ref->rc_sequenceid, HZ >> 1) < 0;
473 if (status)
474 goto out;
478 out:
479 return status;
482 __be32 nfs4_callback_sequence(void *argp, void *resp,
483 struct cb_process_state *cps)
485 struct cb_sequenceargs *args = argp;
486 struct cb_sequenceres *res = resp;
487 struct nfs4_slot_table *tbl;
488 struct nfs4_slot *slot;
489 struct nfs_client *clp;
490 int i;
491 __be32 status = htonl(NFS4ERR_BADSESSION);
493 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
494 &args->csa_sessionid, cps->minorversion);
495 if (clp == NULL)
496 goto out;
498 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
499 goto out;
501 tbl = &clp->cl_session->bc_slot_table;
503 /* Set up res before grabbing the spinlock */
504 memcpy(&res->csr_sessionid, &args->csa_sessionid,
505 sizeof(res->csr_sessionid));
506 res->csr_sequenceid = args->csa_sequenceid;
507 res->csr_slotid = args->csa_slotid;
509 spin_lock(&tbl->slot_tbl_lock);
510 /* state manager is resetting the session */
511 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
512 status = htonl(NFS4ERR_DELAY);
513 /* Return NFS4ERR_BADSESSION if we're draining the session
514 * in order to reset it.
516 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
517 status = htonl(NFS4ERR_BADSESSION);
518 goto out_unlock;
521 status = htonl(NFS4ERR_BADSLOT);
522 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
523 if (IS_ERR(slot))
524 goto out_unlock;
526 res->csr_highestslotid = tbl->server_highest_slotid;
527 res->csr_target_highestslotid = tbl->target_highest_slotid;
529 status = validate_seqid(tbl, slot, args);
530 if (status)
531 goto out_unlock;
532 if (!nfs4_try_to_lock_slot(tbl, slot)) {
533 status = htonl(NFS4ERR_DELAY);
534 goto out_unlock;
536 cps->slot = slot;
538 /* The ca_maxresponsesize_cached is 0 with no DRC */
539 if (args->csa_cachethis != 0) {
540 status = htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
541 goto out_unlock;
545 * Check for pending referring calls. If a match is found, a
546 * related callback was received before the response to the original
547 * call.
549 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
550 status = htonl(NFS4ERR_DELAY);
551 goto out_unlock;
555 * RFC5661 20.9.3
556 * If CB_SEQUENCE returns an error, then the state of the slot
557 * (sequence ID, cached reply) MUST NOT change.
559 slot->seq_nr = args->csa_sequenceid;
560 out_unlock:
561 spin_unlock(&tbl->slot_tbl_lock);
563 out:
564 cps->clp = clp; /* put in nfs4_callback_compound */
565 for (i = 0; i < args->csa_nrclists; i++)
566 kfree(args->csa_rclists[i].rcl_refcalls);
567 kfree(args->csa_rclists);
569 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
570 cps->drc_status = status;
571 status = 0;
572 } else
573 res->csr_status = status;
575 trace_nfs4_cb_sequence(args, res, status);
576 return status;
579 static bool
580 validate_bitmap_values(unsigned int mask)
582 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
585 __be32 nfs4_callback_recallany(void *argp, void *resp,
586 struct cb_process_state *cps)
588 struct cb_recallanyargs *args = argp;
589 __be32 status;
590 fmode_t flags = 0;
592 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
593 if (!cps->clp) /* set in cb_sequence */
594 goto out;
596 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
597 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
599 status = cpu_to_be32(NFS4ERR_INVAL);
600 if (!validate_bitmap_values(args->craa_type_mask))
601 goto out;
603 status = cpu_to_be32(NFS4_OK);
604 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_RDATA_DLG))
605 flags = FMODE_READ;
606 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_WDATA_DLG))
607 flags |= FMODE_WRITE;
608 if (flags)
609 nfs_expire_unused_delegation_types(cps->clp, flags);
611 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_FILE_LAYOUT))
612 pnfs_recall_all_layouts(cps->clp);
613 out:
614 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
615 return status;
618 /* Reduce the fore channel's max_slots to the target value */
619 __be32 nfs4_callback_recallslot(void *argp, void *resp,
620 struct cb_process_state *cps)
622 struct cb_recallslotargs *args = argp;
623 struct nfs4_slot_table *fc_tbl;
624 __be32 status;
626 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
627 if (!cps->clp) /* set in cb_sequence */
628 goto out;
630 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
631 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
632 args->crsa_target_highest_slotid);
634 fc_tbl = &cps->clp->cl_session->fc_slot_table;
636 status = htonl(NFS4_OK);
638 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
639 nfs41_notify_server(cps->clp);
640 out:
641 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
642 return status;
645 __be32 nfs4_callback_notify_lock(void *argp, void *resp,
646 struct cb_process_state *cps)
648 struct cb_notify_lock_args *args = argp;
650 if (!cps->clp) /* set in cb_sequence */
651 return htonl(NFS4ERR_OP_NOT_IN_SESSION);
653 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
654 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
656 /* Don't wake anybody if the string looked bogus */
657 if (args->cbnl_valid)
658 __wake_up(&cps->clp->cl_lock_waitq, TASK_NORMAL, 0, args);
660 return htonl(NFS4_OK);
662 #endif /* CONFIG_NFS_V4_1 */