1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/nfs/callback_proc.c
5 * Copyright (C) 2004 Trond Myklebust
7 * NFSv4 callback procedures
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>
20 #include "delegation.h"
23 #include "nfs4session.h"
24 #include "nfs4trace.h"
26 #define NFSDBG_FACILITY NFSDBG_CALLBACK
28 __be32 nfs4_callback_getattr(void *argp, void *resp,
29 struct cb_process_state *cps)
31 struct cb_getattrargs *args = argp;
32 struct cb_getattrres *res = resp;
33 struct nfs_delegation *delegation;
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 */
40 res->bitmap[0] = res->bitmap[1] = 0;
41 res->status = htonl(NFS4ERR_BADHANDLE);
43 dprintk_rcu("NFS: GETATTR callback request from %s\n",
44 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
46 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
48 if (inode == ERR_PTR(-EAGAIN))
49 res->status = htonl(NFS4ERR_DELAY);
50 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
55 delegation = nfs4_get_valid_delegation(inode);
56 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
58 res->size = i_size_read(inode);
59 res->change_attr = delegation->change_attr;
60 if (nfs_have_writebacks(inode))
62 res->ctime = inode->i_ctime;
63 res->mtime = inode->i_mtime;
64 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
66 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
71 trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
72 nfs_iput_and_deactive(inode);
74 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
78 __be32 nfs4_callback_recall(void *argp, void *resp,
79 struct cb_process_state *cps)
81 struct cb_recallargs *args = argp;
85 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
86 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
89 dprintk_rcu("NFS: RECALL callback request from %s\n",
90 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
92 res = htonl(NFS4ERR_BADHANDLE);
93 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
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));
101 /* Set up a helper thread to actually return the delegation */
102 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
107 res = htonl(NFS4ERR_BAD_STATEID);
110 res = htonl(NFS4ERR_RESOURCE);
112 trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
113 &args->stateid, -ntohl(res));
114 nfs_iput_and_deactive(inode);
116 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
120 #if defined(CONFIG_NFS_V4_1)
123 * Lookup a layout inode by stateid
125 * Note: returns a refcount on the inode and superblock
127 static struct inode *nfs_layout_find_inode_by_stateid(struct nfs_client *clp,
128 const nfs4_stateid *stateid)
131 struct nfs_server *server;
133 struct pnfs_layout_hdr *lo;
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))
140 if (!nfs4_stateid_match_other(stateid, &lo->plh_stateid))
142 if (nfs_sb_active(server->super))
143 inode = igrab(lo->plh_inode);
145 inode = ERR_PTR(-EAGAIN);
149 nfs_sb_deactive(server->super);
150 return ERR_PTR(-EAGAIN);
154 return ERR_PTR(-ENOENT);
158 * Lookup a layout inode by filehandle.
160 * Note: returns a refcount on the inode and superblock
163 static struct inode *nfs_layout_find_inode_by_fh(struct nfs_client *clp,
164 const struct nfs_fh *fh)
166 struct nfs_server *server;
167 struct nfs_inode *nfsi;
169 struct pnfs_layout_hdr *lo;
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))
177 if (nfsi->layout != lo)
179 if (nfs_sb_active(server->super))
180 inode = igrab(lo->plh_inode);
182 inode = ERR_PTR(-EAGAIN);
186 nfs_sb_deactive(server->super);
187 return ERR_PTR(-EAGAIN);
191 return ERR_PTR(-ENOENT);
194 static struct inode *nfs_layout_find_inode(struct nfs_client *clp,
195 const struct nfs_fh *fh,
196 const nfs4_stateid *stateid)
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);
207 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
209 static u32 pnfs_check_callback_stateid(struct pnfs_layout_hdr *lo,
210 const nfs4_stateid *new)
214 /* Is the stateid not initialised? */
215 if (!pnfs_layout_is_valid(lo))
216 return NFS4ERR_NOMATCHING_LAYOUT;
218 /* Mismatched stateid? */
219 if (!nfs4_stateid_match_other(&lo->plh_stateid, new))
220 return NFS4ERR_BAD_STATEID;
222 newseq = be32_to_cpu(new->seqid);
223 /* Are we already in a layout recall situation? */
224 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
225 lo->plh_return_seq != 0) {
226 if (newseq < lo->plh_return_seq)
227 return NFS4ERR_OLD_STATEID;
228 if (newseq > lo->plh_return_seq)
229 return NFS4ERR_DELAY;
233 /* Check that the stateid matches what we think it should be. */
234 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
235 if (newseq > oldseq + 1)
236 return NFS4ERR_DELAY;
238 if (newseq <= oldseq)
239 return NFS4ERR_OLD_STATEID;
244 static u32 initiate_file_draining(struct nfs_client *clp,
245 struct cb_layoutrecallargs *args)
248 struct pnfs_layout_hdr *lo;
249 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
250 LIST_HEAD(free_me_list);
252 ino = nfs_layout_find_inode(clp, &args->cbl_fh, &args->cbl_stateid);
254 if (ino == ERR_PTR(-EAGAIN))
259 pnfs_layoutcommit_inode(ino, false);
262 spin_lock(&ino->i_lock);
263 lo = NFS_I(ino)->layout;
265 spin_unlock(&ino->i_lock);
268 pnfs_get_layout_hdr(lo);
269 rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid);
274 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
276 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
281 pnfs_set_layout_stateid(lo, &args->cbl_stateid, NULL, true);
282 switch (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
284 be32_to_cpu(args->cbl_stateid.seqid))) {
287 /* There are layout segments that need to be returned */
291 set_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags);
292 /* Embrace your forgetfulness! */
293 rv = NFS4ERR_NOMATCHING_LAYOUT;
295 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
296 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
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);
307 nfs_iput_and_deactive(ino);
309 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
310 &args->cbl_stateid, -rv);
314 static u32 initiate_bulk_draining(struct nfs_client *clp,
315 struct cb_layoutrecallargs *args)
319 if (args->cbl_recall_type == RETURN_FSID)
320 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
322 stat = pnfs_destroy_layouts_byclid(clp, true);
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 if (args->cbl_recall_type == RETURN_FILE)
332 return initiate_file_draining(clp, args);
333 return initiate_bulk_draining(clp, args);
336 __be32 nfs4_callback_layoutrecall(void *argp, void *resp,
337 struct cb_process_state *cps)
339 struct cb_layoutrecallargs *args = argp;
340 u32 res = NFS4ERR_OP_NOT_IN_SESSION;
343 res = do_callback_layoutrecall(cps->clp, args);
344 return cpu_to_be32(res);
347 static void pnfs_recall_all_layouts(struct nfs_client *clp)
349 struct cb_layoutrecallargs args;
351 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
352 memset(&args, 0, sizeof(args));
353 args.cbl_recall_type = RETURN_ALL;
354 /* FIXME we ignore errors, what should we do? */
355 do_callback_layoutrecall(clp, &args);
358 __be32 nfs4_callback_devicenotify(void *argp, void *resp,
359 struct cb_process_state *cps)
361 struct cb_devicenotifyargs *args = argp;
362 const struct pnfs_layoutdriver_type *ld = NULL;
367 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
371 for (i = 0; i < args->ndevs; i++) {
372 struct cb_devicenotifyitem *dev = &args->devs[i];
374 if (!ld || ld->id != dev->cbd_layout_type) {
375 pnfs_put_layoutdriver(ld);
376 ld = pnfs_find_layoutdriver(dev->cbd_layout_type);
380 nfs4_delete_deviceid(ld, cps->clp, &dev->cbd_dev_id);
382 pnfs_put_layoutdriver(ld);
389 * Validate the sequenceID sent by the server.
390 * Return success if the sequenceID is one more than what we last saw on
391 * this slot, accounting for wraparound. Increments the slot's sequence.
393 * We don't yet implement a duplicate request cache, instead we set the
394 * back channel ca_maxresponsesize_cached to zero. This is OK for now
395 * since we only currently implement idempotent callbacks anyway.
397 * We have a single slot backchannel at this time, so we don't bother
398 * checking the used_slots bit array on the table. The lower layer guarantees
399 * a single outstanding callback request at a time.
402 validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
403 const struct cb_sequenceargs * args)
407 ret = cpu_to_be32(NFS4ERR_BADSLOT);
408 if (args->csa_slotid > tbl->server_highest_slotid)
412 if (args->csa_sequenceid == slot->seq_nr) {
413 ret = cpu_to_be32(NFS4ERR_DELAY);
414 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
417 /* Signal process_op to set this error on next op */
418 ret = cpu_to_be32(NFS4ERR_RETRY_UNCACHED_REP);
419 if (args->csa_cachethis == 0)
422 /* Liar! We never allowed you to set csa_cachethis != 0 */
423 ret = cpu_to_be32(NFS4ERR_SEQ_FALSE_RETRY);
427 /* Note: wraparound relies on seq_nr being of type u32 */
428 /* Misordered request */
429 ret = cpu_to_be32(NFS4ERR_SEQ_MISORDERED);
430 if (args->csa_sequenceid != slot->seq_nr + 1)
433 return cpu_to_be32(NFS4_OK);
436 trace_nfs4_cb_seqid_err(args, ret);
441 * For each referring call triple, check the session's slot table for
442 * a match. If the slot is in use and the sequence numbers match, the
443 * client is still waiting for a response to the original request.
445 static int referring_call_exists(struct nfs_client *clp,
447 struct referring_call_list *rclists,
454 struct nfs4_session *session;
455 struct nfs4_slot_table *tbl;
456 struct referring_call_list *rclist;
457 struct referring_call *ref;
460 * XXX When client trunking is implemented, this becomes
461 * a session lookup from within the loop
463 session = clp->cl_session;
464 tbl = &session->fc_slot_table;
466 for (i = 0; i < nrclists; i++) {
467 rclist = &rclists[i];
468 if (memcmp(session->sess_id.data,
469 rclist->rcl_sessionid.data,
470 NFS4_MAX_SESSIONID_LEN) != 0)
473 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
474 ref = &rclist->rcl_refcalls[j];
476 status = nfs4_slot_wait_on_seqid(tbl, ref->rc_slotid,
477 ref->rc_sequenceid, HZ >> 1) < 0;
488 __be32 nfs4_callback_sequence(void *argp, void *resp,
489 struct cb_process_state *cps)
491 struct cb_sequenceargs *args = argp;
492 struct cb_sequenceres *res = resp;
493 struct nfs4_slot_table *tbl;
494 struct nfs4_slot *slot;
495 struct nfs_client *clp;
497 __be32 status = htonl(NFS4ERR_BADSESSION);
499 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
500 &args->csa_sessionid, cps->minorversion);
504 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
507 tbl = &clp->cl_session->bc_slot_table;
509 /* Set up res before grabbing the spinlock */
510 memcpy(&res->csr_sessionid, &args->csa_sessionid,
511 sizeof(res->csr_sessionid));
512 res->csr_sequenceid = args->csa_sequenceid;
513 res->csr_slotid = args->csa_slotid;
515 spin_lock(&tbl->slot_tbl_lock);
516 /* state manager is resetting the session */
517 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
518 status = htonl(NFS4ERR_DELAY);
519 /* Return NFS4ERR_BADSESSION if we're draining the session
520 * in order to reset it.
522 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
523 status = htonl(NFS4ERR_BADSESSION);
527 status = htonl(NFS4ERR_BADSLOT);
528 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
532 res->csr_highestslotid = tbl->server_highest_slotid;
533 res->csr_target_highestslotid = tbl->target_highest_slotid;
535 status = validate_seqid(tbl, slot, args);
538 if (!nfs4_try_to_lock_slot(tbl, slot)) {
539 status = htonl(NFS4ERR_DELAY);
544 /* The ca_maxresponsesize_cached is 0 with no DRC */
545 if (args->csa_cachethis != 0) {
546 status = htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
551 * Check for pending referring calls. If a match is found, a
552 * related callback was received before the response to the original
555 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists,
556 &tbl->slot_tbl_lock) < 0) {
557 status = htonl(NFS4ERR_DELAY);
563 * If CB_SEQUENCE returns an error, then the state of the slot
564 * (sequence ID, cached reply) MUST NOT change.
566 slot->seq_nr = args->csa_sequenceid;
568 spin_unlock(&tbl->slot_tbl_lock);
571 cps->clp = clp; /* put in nfs4_callback_compound */
572 for (i = 0; i < args->csa_nrclists; i++)
573 kfree(args->csa_rclists[i].rcl_refcalls);
574 kfree(args->csa_rclists);
576 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
577 cps->drc_status = status;
580 res->csr_status = status;
582 trace_nfs4_cb_sequence(args, res, status);
587 validate_bitmap_values(unsigned int mask)
589 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
592 __be32 nfs4_callback_recallany(void *argp, void *resp,
593 struct cb_process_state *cps)
595 struct cb_recallanyargs *args = argp;
598 bool schedule_manager = false;
600 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
601 if (!cps->clp) /* set in cb_sequence */
604 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
605 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
607 status = cpu_to_be32(NFS4ERR_INVAL);
608 if (!validate_bitmap_values(args->craa_type_mask))
611 status = cpu_to_be32(NFS4_OK);
612 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_RDATA_DLG))
614 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_WDATA_DLG))
615 flags |= FMODE_WRITE;
617 nfs_expire_unused_delegation_types(cps->clp, flags);
619 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_FILE_LAYOUT))
620 pnfs_recall_all_layouts(cps->clp);
622 if (args->craa_type_mask & BIT(PNFS_FF_RCA4_TYPE_MASK_READ)) {
623 set_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &cps->clp->cl_state);
624 schedule_manager = true;
626 if (args->craa_type_mask & BIT(PNFS_FF_RCA4_TYPE_MASK_RW)) {
627 set_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &cps->clp->cl_state);
628 schedule_manager = true;
630 if (schedule_manager)
631 nfs4_schedule_state_manager(cps->clp);
634 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
638 /* Reduce the fore channel's max_slots to the target value */
639 __be32 nfs4_callback_recallslot(void *argp, void *resp,
640 struct cb_process_state *cps)
642 struct cb_recallslotargs *args = argp;
643 struct nfs4_slot_table *fc_tbl;
646 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
647 if (!cps->clp) /* set in cb_sequence */
650 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
651 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
652 args->crsa_target_highest_slotid);
654 fc_tbl = &cps->clp->cl_session->fc_slot_table;
656 status = htonl(NFS4_OK);
658 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
659 nfs41_notify_server(cps->clp);
661 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
665 __be32 nfs4_callback_notify_lock(void *argp, void *resp,
666 struct cb_process_state *cps)
668 struct cb_notify_lock_args *args = argp;
670 if (!cps->clp) /* set in cb_sequence */
671 return htonl(NFS4ERR_OP_NOT_IN_SESSION);
673 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
674 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
676 /* Don't wake anybody if the string looked bogus */
677 if (args->cbnl_valid)
678 __wake_up(&cps->clp->cl_lock_waitq, TASK_NORMAL, 0, args);
680 return htonl(NFS4_OK);
682 #endif /* CONFIG_NFS_V4_1 */
683 #ifdef CONFIG_NFS_V4_2
684 static void nfs4_copy_cb_args(struct nfs4_copy_state *cp_state,
685 struct cb_offloadargs *args)
687 cp_state->count = args->wr_count;
688 cp_state->error = args->error;
690 cp_state->verf.committed = args->wr_writeverf.committed;
691 memcpy(&cp_state->verf.verifier.data[0],
692 &args->wr_writeverf.verifier.data[0],
697 __be32 nfs4_callback_offload(void *data, void *dummy,
698 struct cb_process_state *cps)
700 struct cb_offloadargs *args = data;
701 struct nfs_server *server;
702 struct nfs4_copy_state *copy, *tmp_copy;
705 copy = kzalloc(sizeof(struct nfs4_copy_state), GFP_KERNEL);
707 return htonl(NFS4ERR_SERVERFAULT);
709 spin_lock(&cps->clp->cl_lock);
711 list_for_each_entry_rcu(server, &cps->clp->cl_superblocks,
713 list_for_each_entry(tmp_copy, &server->ss_copies, copies) {
714 if (memcmp(args->coa_stateid.other,
715 tmp_copy->stateid.other,
716 sizeof(args->coa_stateid.other)))
718 nfs4_copy_cb_args(tmp_copy, args);
719 complete(&tmp_copy->completion);
727 memcpy(©->stateid, &args->coa_stateid, NFS4_STATEID_SIZE);
728 nfs4_copy_cb_args(copy, args);
729 list_add_tail(©->copies, &cps->clp->pending_cb_stateids);
732 spin_unlock(&cps->clp->cl_lock);
734 trace_nfs4_cb_offload(&args->coa_fh, &args->coa_stateid,
735 args->wr_count, args->error,
736 args->wr_writeverf.committed);
739 #endif /* CONFIG_NFS_V4_2 */