1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/nfs/callback_proc.c
5 * Copyright (C) 2004 Trond Myklebust
7 * NFSv4 callback procedures
9 #include <linux/nfs4.h>
10 #include <linux/nfs_fs.h>
11 #include <linux/slab.h>
12 #include <linux/rcupdate.h>
15 #include "delegation.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;
31 res->status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
32 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
35 res->bitmap[0] = res->bitmap[1] = 0;
36 res->status = htonl(NFS4ERR_BADHANDLE);
38 dprintk_rcu("NFS: GETATTR callback request from %s\n",
39 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
41 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
43 if (inode == ERR_PTR(-EAGAIN))
44 res->status = htonl(NFS4ERR_DELAY);
45 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
50 delegation = nfs4_get_valid_delegation(inode);
51 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
53 res->size = i_size_read(inode);
54 res->change_attr = delegation->change_attr;
55 if (nfs_have_writebacks(inode))
57 res->ctime = inode->i_ctime;
58 res->mtime = inode->i_mtime;
59 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
61 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
66 trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
67 nfs_iput_and_deactive(inode);
69 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
73 __be32 nfs4_callback_recall(void *argp, void *resp,
74 struct cb_process_state *cps)
76 struct cb_recallargs *args = argp;
80 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
81 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
84 dprintk_rcu("NFS: RECALL callback request from %s\n",
85 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
87 res = htonl(NFS4ERR_BADHANDLE);
88 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
90 if (inode == ERR_PTR(-EAGAIN))
91 res = htonl(NFS4ERR_DELAY);
92 trace_nfs4_cb_recall(cps->clp, &args->fh, NULL,
93 &args->stateid, -ntohl(res));
96 /* Set up a helper thread to actually return the delegation */
97 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
102 res = htonl(NFS4ERR_BAD_STATEID);
105 res = htonl(NFS4ERR_RESOURCE);
107 trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
108 &args->stateid, -ntohl(res));
109 nfs_iput_and_deactive(inode);
111 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
115 #if defined(CONFIG_NFS_V4_1)
118 * Lookup a layout inode by stateid
120 * Note: returns a refcount on the inode and superblock
122 static struct inode *nfs_layout_find_inode_by_stateid(struct nfs_client *clp,
123 const nfs4_stateid *stateid)
125 struct nfs_server *server;
127 struct pnfs_layout_hdr *lo;
129 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
130 list_for_each_entry(lo, &server->layouts, plh_layouts) {
131 if (stateid != NULL &&
132 !nfs4_stateid_match_other(stateid, &lo->plh_stateid))
134 inode = igrab(lo->plh_inode);
136 return ERR_PTR(-EAGAIN);
137 if (!nfs_sb_active(inode->i_sb)) {
139 spin_unlock(&clp->cl_lock);
141 spin_lock(&clp->cl_lock);
143 return ERR_PTR(-EAGAIN);
149 return ERR_PTR(-ENOENT);
153 * Lookup a layout inode by filehandle.
155 * Note: returns a refcount on the inode and superblock
158 static struct inode *nfs_layout_find_inode_by_fh(struct nfs_client *clp,
159 const struct nfs_fh *fh)
161 struct nfs_server *server;
162 struct nfs_inode *nfsi;
164 struct pnfs_layout_hdr *lo;
166 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
167 list_for_each_entry(lo, &server->layouts, plh_layouts) {
168 nfsi = NFS_I(lo->plh_inode);
169 if (nfs_compare_fh(fh, &nfsi->fh))
171 if (nfsi->layout != lo)
173 inode = igrab(lo->plh_inode);
175 return ERR_PTR(-EAGAIN);
176 if (!nfs_sb_active(inode->i_sb)) {
178 spin_unlock(&clp->cl_lock);
180 spin_lock(&clp->cl_lock);
182 return ERR_PTR(-EAGAIN);
188 return ERR_PTR(-ENOENT);
191 static struct inode *nfs_layout_find_inode(struct nfs_client *clp,
192 const struct nfs_fh *fh,
193 const nfs4_stateid *stateid)
197 spin_lock(&clp->cl_lock);
199 inode = nfs_layout_find_inode_by_stateid(clp, stateid);
200 if (inode == ERR_PTR(-ENOENT))
201 inode = nfs_layout_find_inode_by_fh(clp, fh);
203 spin_unlock(&clp->cl_lock);
209 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
211 static u32 pnfs_check_callback_stateid(struct pnfs_layout_hdr *lo,
212 const nfs4_stateid *new)
216 /* Is the stateid not initialised? */
217 if (!pnfs_layout_is_valid(lo))
218 return NFS4ERR_NOMATCHING_LAYOUT;
220 /* Mismatched stateid? */
221 if (!nfs4_stateid_match_other(&lo->plh_stateid, new))
222 return NFS4ERR_BAD_STATEID;
224 newseq = be32_to_cpu(new->seqid);
225 /* Are we already in a layout recall situation? */
226 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
227 lo->plh_return_seq != 0) {
228 if (newseq < lo->plh_return_seq)
229 return NFS4ERR_OLD_STATEID;
230 if (newseq > lo->plh_return_seq)
231 return NFS4ERR_DELAY;
235 /* Check that the stateid matches what we think it should be. */
236 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
237 if (newseq > oldseq + 1)
238 return NFS4ERR_DELAY;
240 if (newseq <= oldseq)
241 return NFS4ERR_OLD_STATEID;
246 static u32 initiate_file_draining(struct nfs_client *clp,
247 struct cb_layoutrecallargs *args)
250 struct pnfs_layout_hdr *lo;
251 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
252 LIST_HEAD(free_me_list);
254 ino = nfs_layout_find_inode(clp, &args->cbl_fh, &args->cbl_stateid);
256 if (ino == ERR_PTR(-EAGAIN))
261 pnfs_layoutcommit_inode(ino, false);
264 spin_lock(&ino->i_lock);
265 lo = NFS_I(ino)->layout;
267 spin_unlock(&ino->i_lock);
270 pnfs_get_layout_hdr(lo);
271 rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid);
276 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
278 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
283 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
284 switch (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
286 be32_to_cpu(args->cbl_stateid.seqid))) {
289 /* There are layout segments that need to be returned */
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,
302 spin_unlock(&ino->i_lock);
303 pnfs_free_lseg_list(&free_me_list);
304 /* Free all lsegs that are attached to commit buckets */
305 nfs_commit_inode(ino, 0);
306 pnfs_put_layout_hdr(lo);
308 nfs_iput_and_deactive(ino);
310 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
311 &args->cbl_stateid, -rv);
315 static u32 initiate_bulk_draining(struct nfs_client *clp,
316 struct cb_layoutrecallargs *args)
320 if (args->cbl_recall_type == RETURN_FSID)
321 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
323 stat = pnfs_destroy_layouts_byclid(clp, true);
325 return NFS4ERR_DELAY;
326 return NFS4ERR_NOMATCHING_LAYOUT;
329 static u32 do_callback_layoutrecall(struct nfs_client *clp,
330 struct cb_layoutrecallargs *args)
332 if (args->cbl_recall_type == RETURN_FILE)
333 return initiate_file_draining(clp, args);
334 return initiate_bulk_draining(clp, args);
337 __be32 nfs4_callback_layoutrecall(void *argp, void *resp,
338 struct cb_process_state *cps)
340 struct cb_layoutrecallargs *args = argp;
341 u32 res = NFS4ERR_OP_NOT_IN_SESSION;
344 res = do_callback_layoutrecall(cps->clp, args);
345 return cpu_to_be32(res);
348 static void pnfs_recall_all_layouts(struct nfs_client *clp)
350 struct cb_layoutrecallargs args;
352 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
353 memset(&args, 0, sizeof(args));
354 args.cbl_recall_type = RETURN_ALL;
355 /* FIXME we ignore errors, what should we do? */
356 do_callback_layoutrecall(clp, &args);
359 __be32 nfs4_callback_devicenotify(void *argp, void *resp,
360 struct cb_process_state *cps)
362 struct cb_devicenotifyargs *args = argp;
365 struct nfs_client *clp = cps->clp;
366 struct nfs_server *server = NULL;
369 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
373 for (i = 0; i < args->ndevs; i++) {
374 struct cb_devicenotifyitem *dev = &args->devs[i];
377 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
379 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
380 if (server->pnfs_curr_ld &&
381 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
390 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
399 * Validate the sequenceID sent by the server.
400 * Return success if the sequenceID is one more than what we last saw on
401 * this slot, accounting for wraparound. Increments the slot's sequence.
403 * We don't yet implement a duplicate request cache, instead we set the
404 * back channel ca_maxresponsesize_cached to zero. This is OK for now
405 * since we only currently implement idempotent callbacks anyway.
407 * We have a single slot backchannel at this time, so we don't bother
408 * checking the used_slots bit array on the table. The lower layer guarantees
409 * a single outstanding callback request at a time.
412 validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
413 const struct cb_sequenceargs * args)
417 ret = cpu_to_be32(NFS4ERR_BADSLOT);
418 if (args->csa_slotid > tbl->server_highest_slotid)
422 if (args->csa_sequenceid == slot->seq_nr) {
423 ret = cpu_to_be32(NFS4ERR_DELAY);
424 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
427 /* Signal process_op to set this error on next op */
428 ret = cpu_to_be32(NFS4ERR_RETRY_UNCACHED_REP);
429 if (args->csa_cachethis == 0)
432 /* Liar! We never allowed you to set csa_cachethis != 0 */
433 ret = cpu_to_be32(NFS4ERR_SEQ_FALSE_RETRY);
437 /* Note: wraparound relies on seq_nr being of type u32 */
438 /* Misordered request */
439 ret = cpu_to_be32(NFS4ERR_SEQ_MISORDERED);
440 if (args->csa_sequenceid != slot->seq_nr + 1)
443 return cpu_to_be32(NFS4_OK);
446 trace_nfs4_cb_seqid_err(args, ret);
451 * For each referring call triple, check the session's slot table for
452 * a match. If the slot is in use and the sequence numbers match, the
453 * client is still waiting for a response to the original request.
455 static int referring_call_exists(struct nfs_client *clp,
457 struct referring_call_list *rclists,
464 struct nfs4_session *session;
465 struct nfs4_slot_table *tbl;
466 struct referring_call_list *rclist;
467 struct referring_call *ref;
470 * XXX When client trunking is implemented, this becomes
471 * a session lookup from within the loop
473 session = clp->cl_session;
474 tbl = &session->fc_slot_table;
476 for (i = 0; i < nrclists; i++) {
477 rclist = &rclists[i];
478 if (memcmp(session->sess_id.data,
479 rclist->rcl_sessionid.data,
480 NFS4_MAX_SESSIONID_LEN) != 0)
483 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
484 ref = &rclist->rcl_refcalls[j];
486 status = nfs4_slot_wait_on_seqid(tbl, ref->rc_slotid,
487 ref->rc_sequenceid, HZ >> 1) < 0;
498 __be32 nfs4_callback_sequence(void *argp, void *resp,
499 struct cb_process_state *cps)
501 struct cb_sequenceargs *args = argp;
502 struct cb_sequenceres *res = resp;
503 struct nfs4_slot_table *tbl;
504 struct nfs4_slot *slot;
505 struct nfs_client *clp;
507 __be32 status = htonl(NFS4ERR_BADSESSION);
509 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
510 &args->csa_sessionid, cps->minorversion);
514 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
517 tbl = &clp->cl_session->bc_slot_table;
519 /* Set up res before grabbing the spinlock */
520 memcpy(&res->csr_sessionid, &args->csa_sessionid,
521 sizeof(res->csr_sessionid));
522 res->csr_sequenceid = args->csa_sequenceid;
523 res->csr_slotid = args->csa_slotid;
525 spin_lock(&tbl->slot_tbl_lock);
526 /* state manager is resetting the session */
527 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
528 status = htonl(NFS4ERR_DELAY);
529 /* Return NFS4ERR_BADSESSION if we're draining the session
530 * in order to reset it.
532 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
533 status = htonl(NFS4ERR_BADSESSION);
537 status = htonl(NFS4ERR_BADSLOT);
538 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
542 res->csr_highestslotid = tbl->server_highest_slotid;
543 res->csr_target_highestslotid = tbl->target_highest_slotid;
545 status = validate_seqid(tbl, slot, args);
548 if (!nfs4_try_to_lock_slot(tbl, slot)) {
549 status = htonl(NFS4ERR_DELAY);
554 /* The ca_maxresponsesize_cached is 0 with no DRC */
555 if (args->csa_cachethis != 0) {
556 status = htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
561 * Check for pending referring calls. If a match is found, a
562 * related callback was received before the response to the original
565 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists,
566 &tbl->slot_tbl_lock) < 0) {
567 status = htonl(NFS4ERR_DELAY);
573 * If CB_SEQUENCE returns an error, then the state of the slot
574 * (sequence ID, cached reply) MUST NOT change.
576 slot->seq_nr = args->csa_sequenceid;
578 spin_unlock(&tbl->slot_tbl_lock);
581 cps->clp = clp; /* put in nfs4_callback_compound */
582 for (i = 0; i < args->csa_nrclists; i++)
583 kfree(args->csa_rclists[i].rcl_refcalls);
584 kfree(args->csa_rclists);
586 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
587 cps->drc_status = status;
590 res->csr_status = status;
592 trace_nfs4_cb_sequence(args, res, status);
597 validate_bitmap_values(unsigned int mask)
599 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
602 __be32 nfs4_callback_recallany(void *argp, void *resp,
603 struct cb_process_state *cps)
605 struct cb_recallanyargs *args = argp;
609 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
610 if (!cps->clp) /* set in cb_sequence */
613 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
614 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
616 status = cpu_to_be32(NFS4ERR_INVAL);
617 if (!validate_bitmap_values(args->craa_type_mask))
620 status = cpu_to_be32(NFS4_OK);
621 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_RDATA_DLG))
623 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_WDATA_DLG))
624 flags |= FMODE_WRITE;
626 nfs_expire_unused_delegation_types(cps->clp, flags);
628 if (args->craa_type_mask & BIT(RCA4_TYPE_MASK_FILE_LAYOUT))
629 pnfs_recall_all_layouts(cps->clp);
631 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
635 /* Reduce the fore channel's max_slots to the target value */
636 __be32 nfs4_callback_recallslot(void *argp, void *resp,
637 struct cb_process_state *cps)
639 struct cb_recallslotargs *args = argp;
640 struct nfs4_slot_table *fc_tbl;
643 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
644 if (!cps->clp) /* set in cb_sequence */
647 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
648 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
649 args->crsa_target_highest_slotid);
651 fc_tbl = &cps->clp->cl_session->fc_slot_table;
653 status = htonl(NFS4_OK);
655 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
656 nfs41_notify_server(cps->clp);
658 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
662 __be32 nfs4_callback_notify_lock(void *argp, void *resp,
663 struct cb_process_state *cps)
665 struct cb_notify_lock_args *args = argp;
667 if (!cps->clp) /* set in cb_sequence */
668 return htonl(NFS4ERR_OP_NOT_IN_SESSION);
670 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
671 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
673 /* Don't wake anybody if the string looked bogus */
674 if (args->cbnl_valid)
675 __wake_up(&cps->clp->cl_lock_waitq, TASK_NORMAL, 0, args);
677 return htonl(NFS4_OK);
679 #endif /* CONFIG_NFS_V4_1 */
680 #ifdef CONFIG_NFS_V4_2
681 static void nfs4_copy_cb_args(struct nfs4_copy_state *cp_state,
682 struct cb_offloadargs *args)
684 cp_state->count = args->wr_count;
685 cp_state->error = args->error;
687 cp_state->verf.committed = args->wr_writeverf.committed;
688 memcpy(&cp_state->verf.verifier.data[0],
689 &args->wr_writeverf.verifier.data[0],
694 __be32 nfs4_callback_offload(void *data, void *dummy,
695 struct cb_process_state *cps)
697 struct cb_offloadargs *args = data;
698 struct nfs_server *server;
699 struct nfs4_copy_state *copy, *tmp_copy;
702 copy = kzalloc(sizeof(struct nfs4_copy_state), GFP_NOFS);
704 return htonl(NFS4ERR_SERVERFAULT);
706 spin_lock(&cps->clp->cl_lock);
708 list_for_each_entry_rcu(server, &cps->clp->cl_superblocks,
710 list_for_each_entry(tmp_copy, &server->ss_copies, copies) {
711 if (memcmp(args->coa_stateid.other,
712 tmp_copy->stateid.other,
713 sizeof(args->coa_stateid.other)))
715 nfs4_copy_cb_args(tmp_copy, args);
716 complete(&tmp_copy->completion);
724 memcpy(©->stateid, &args->coa_stateid, NFS4_STATEID_SIZE);
725 nfs4_copy_cb_args(copy, args);
726 list_add_tail(©->copies, &cps->clp->pending_cb_stateids);
729 spin_unlock(&cps->clp->cl_lock);
733 #endif /* CONFIG_NFS_V4_2 */