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;
29 struct nfs_inode *nfsi;
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 */
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);
44 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
50 delegation = rcu_dereference(nfsi->delegation);
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));
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 trace_nfs4_cb_recall(cps->clp, &args->fh, NULL,
91 &args->stateid, -ntohl(res));
94 /* Set up a helper thread to actually return the delegation */
95 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
100 res = htonl(NFS4ERR_BAD_STATEID);
103 res = htonl(NFS4ERR_RESOURCE);
105 trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
106 &args->stateid, -ntohl(res));
109 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
113 #if defined(CONFIG_NFS_V4_1)
116 * Lookup a layout inode by stateid
118 * Note: returns a refcount on the inode and superblock
120 static struct inode *nfs_layout_find_inode_by_stateid(struct nfs_client *clp,
121 const nfs4_stateid *stateid)
123 struct nfs_server *server;
125 struct pnfs_layout_hdr *lo;
128 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
129 list_for_each_entry(lo, &server->layouts, plh_layouts) {
130 if (!pnfs_layout_is_valid(lo))
132 if (stateid != NULL &&
133 !nfs4_stateid_match_other(stateid, &lo->plh_stateid))
135 inode = igrab(lo->plh_inode);
138 if (!nfs_sb_active(inode->i_sb)) {
140 spin_unlock(&clp->cl_lock);
142 spin_lock(&clp->cl_lock);
154 * Lookup a layout inode by filehandle.
156 * Note: returns a refcount on the inode and superblock
159 static struct inode *nfs_layout_find_inode_by_fh(struct nfs_client *clp,
160 const struct nfs_fh *fh)
162 struct nfs_server *server;
163 struct nfs_inode *nfsi;
165 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))
173 if (nfsi->layout != lo)
175 inode = igrab(lo->plh_inode);
178 if (!nfs_sb_active(inode->i_sb)) {
180 spin_unlock(&clp->cl_lock);
182 spin_lock(&clp->cl_lock);
193 static struct inode *nfs_layout_find_inode(struct nfs_client *clp,
194 const struct nfs_fh *fh,
195 const nfs4_stateid *stateid)
199 spin_lock(&clp->cl_lock);
201 inode = nfs_layout_find_inode_by_stateid(clp, stateid);
203 inode = nfs_layout_find_inode_by_fh(clp, fh);
205 spin_unlock(&clp->cl_lock);
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)
218 /* Is the stateid not initialised? */
219 if (!pnfs_layout_is_valid(lo))
220 return NFS4ERR_NOMATCHING_LAYOUT;
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;
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;
242 if (newseq <= oldseq)
243 return NFS4ERR_OLD_STATEID;
248 static u32 initiate_file_draining(struct nfs_client *clp,
249 struct cb_layoutrecallargs *args)
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);
260 pnfs_layoutcommit_inode(ino, false);
263 spin_lock(&ino->i_lock);
264 lo = NFS_I(ino)->layout;
266 spin_unlock(&ino->i_lock);
269 pnfs_get_layout_hdr(lo);
270 rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid);
273 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
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 if (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
285 be32_to_cpu(args->cbl_stateid.seqid))) {
290 /* Embrace your forgetfulness! */
291 rv = NFS4ERR_NOMATCHING_LAYOUT;
293 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
294 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
298 spin_unlock(&ino->i_lock);
299 pnfs_free_lseg_list(&free_me_list);
300 /* Free all lsegs that are attached to commit buckets */
301 nfs_commit_inode(ino, 0);
302 pnfs_put_layout_hdr(lo);
304 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
305 &args->cbl_stateid, -rv);
306 nfs_iput_and_deactive(ino);
310 static u32 initiate_bulk_draining(struct nfs_client *clp,
311 struct cb_layoutrecallargs *args)
315 if (args->cbl_recall_type == RETURN_FSID)
316 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
318 stat = pnfs_destroy_layouts_byclid(clp, true);
320 return NFS4ERR_DELAY;
321 return NFS4ERR_NOMATCHING_LAYOUT;
324 static u32 do_callback_layoutrecall(struct nfs_client *clp,
325 struct cb_layoutrecallargs *args)
327 if (args->cbl_recall_type == RETURN_FILE)
328 return initiate_file_draining(clp, args);
329 return initiate_bulk_draining(clp, args);
332 __be32 nfs4_callback_layoutrecall(void *argp, void *resp,
333 struct cb_process_state *cps)
335 struct cb_layoutrecallargs *args = argp;
336 u32 res = NFS4ERR_OP_NOT_IN_SESSION;
339 res = do_callback_layoutrecall(cps->clp, args);
340 return cpu_to_be32(res);
343 static void pnfs_recall_all_layouts(struct nfs_client *clp)
345 struct cb_layoutrecallargs args;
347 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
348 memset(&args, 0, sizeof(args));
349 args.cbl_recall_type = RETURN_ALL;
350 /* FIXME we ignore errors, what should we do? */
351 do_callback_layoutrecall(clp, &args);
354 __be32 nfs4_callback_devicenotify(void *argp, void *resp,
355 struct cb_process_state *cps)
357 struct cb_devicenotifyargs *args = argp;
358 const struct pnfs_layoutdriver_type *ld = NULL;
363 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
367 for (i = 0; i < args->ndevs; i++) {
368 struct cb_devicenotifyitem *dev = &args->devs[i];
370 if (!ld || ld->id != dev->cbd_layout_type) {
371 pnfs_put_layoutdriver(ld);
372 ld = pnfs_find_layoutdriver(dev->cbd_layout_type);
376 nfs4_delete_deviceid(ld, cps->clp, &dev->cbd_dev_id);
378 pnfs_put_layoutdriver(ld);
385 * Validate the sequenceID sent by the server.
386 * Return success if the sequenceID is one more than what we last saw on
387 * this slot, accounting for wraparound. Increments the slot's sequence.
389 * We don't yet implement a duplicate request cache, instead we set the
390 * back channel ca_maxresponsesize_cached to zero. This is OK for now
391 * since we only currently implement idempotent callbacks anyway.
393 * We have a single slot backchannel at this time, so we don't bother
394 * checking the used_slots bit array on the table. The lower layer guarantees
395 * a single outstanding callback request at a time.
398 validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
399 const struct cb_sequenceargs * args)
401 if (args->csa_slotid > tbl->server_highest_slotid)
402 return htonl(NFS4ERR_BADSLOT);
405 if (args->csa_sequenceid == slot->seq_nr) {
406 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
407 return htonl(NFS4ERR_DELAY);
408 /* Signal process_op to set this error on next op */
409 if (args->csa_cachethis == 0)
410 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
412 /* Liar! We never allowed you to set csa_cachethis != 0 */
413 return htonl(NFS4ERR_SEQ_FALSE_RETRY);
416 /* Note: wraparound relies on seq_nr being of type u32 */
417 if (likely(args->csa_sequenceid == slot->seq_nr + 1))
418 return htonl(NFS4_OK);
420 /* Misordered request */
421 return htonl(NFS4ERR_SEQ_MISORDERED);
425 * For each referring call triple, check the session's slot table for
426 * a match. If the slot is in use and the sequence numbers match, the
427 * client is still waiting for a response to the original request.
429 static int referring_call_exists(struct nfs_client *clp,
431 struct referring_call_list *rclists,
438 struct nfs4_session *session;
439 struct nfs4_slot_table *tbl;
440 struct referring_call_list *rclist;
441 struct referring_call *ref;
444 * XXX When client trunking is implemented, this becomes
445 * a session lookup from within the loop
447 session = clp->cl_session;
448 tbl = &session->fc_slot_table;
450 for (i = 0; i < nrclists; i++) {
451 rclist = &rclists[i];
452 if (memcmp(session->sess_id.data,
453 rclist->rcl_sessionid.data,
454 NFS4_MAX_SESSIONID_LEN) != 0)
457 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
458 ref = &rclist->rcl_refcalls[j];
460 status = nfs4_slot_wait_on_seqid(tbl, ref->rc_slotid,
461 ref->rc_sequenceid, HZ >> 1) < 0;
472 __be32 nfs4_callback_sequence(void *argp, void *resp,
473 struct cb_process_state *cps)
475 struct cb_sequenceargs *args = argp;
476 struct cb_sequenceres *res = resp;
477 struct nfs4_slot_table *tbl;
478 struct nfs4_slot *slot;
479 struct nfs_client *clp;
481 __be32 status = htonl(NFS4ERR_BADSESSION);
483 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
484 &args->csa_sessionid, cps->minorversion);
488 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
491 tbl = &clp->cl_session->bc_slot_table;
493 /* Set up res before grabbing the spinlock */
494 memcpy(&res->csr_sessionid, &args->csa_sessionid,
495 sizeof(res->csr_sessionid));
496 res->csr_sequenceid = args->csa_sequenceid;
497 res->csr_slotid = args->csa_slotid;
499 spin_lock(&tbl->slot_tbl_lock);
500 /* state manager is resetting the session */
501 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
502 status = htonl(NFS4ERR_DELAY);
503 /* Return NFS4ERR_BADSESSION if we're draining the session
504 * in order to reset it.
506 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
507 status = htonl(NFS4ERR_BADSESSION);
511 status = htonl(NFS4ERR_BADSLOT);
512 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
516 res->csr_highestslotid = tbl->server_highest_slotid;
517 res->csr_target_highestslotid = tbl->target_highest_slotid;
519 status = validate_seqid(tbl, slot, args);
522 if (!nfs4_try_to_lock_slot(tbl, slot)) {
523 status = htonl(NFS4ERR_DELAY);
528 /* The ca_maxresponsesize_cached is 0 with no DRC */
529 if (args->csa_cachethis != 0) {
530 status = htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
535 * Check for pending referring calls. If a match is found, a
536 * related callback was received before the response to the original
539 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists,
540 &tbl->slot_tbl_lock) < 0) {
541 status = htonl(NFS4ERR_DELAY);
547 * If CB_SEQUENCE returns an error, then the state of the slot
548 * (sequence ID, cached reply) MUST NOT change.
550 slot->seq_nr = args->csa_sequenceid;
552 spin_unlock(&tbl->slot_tbl_lock);
555 cps->clp = clp; /* put in nfs4_callback_compound */
556 for (i = 0; i < args->csa_nrclists; i++)
557 kfree(args->csa_rclists[i].rcl_refcalls);
558 kfree(args->csa_rclists);
560 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
561 cps->drc_status = status;
564 res->csr_status = status;
566 trace_nfs4_cb_sequence(args, res, status);
571 validate_bitmap_values(unsigned long mask)
573 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
576 __be32 nfs4_callback_recallany(void *argp, void *resp,
577 struct cb_process_state *cps)
579 struct cb_recallanyargs *args = argp;
583 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
584 if (!cps->clp) /* set in cb_sequence */
587 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
588 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
590 status = cpu_to_be32(NFS4ERR_INVAL);
591 if (!validate_bitmap_values(args->craa_type_mask))
594 status = cpu_to_be32(NFS4_OK);
595 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
596 &args->craa_type_mask))
598 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
599 &args->craa_type_mask))
600 flags |= FMODE_WRITE;
601 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
602 &args->craa_type_mask))
603 pnfs_recall_all_layouts(cps->clp);
605 nfs_expire_unused_delegation_types(cps->clp, flags);
607 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
611 /* Reduce the fore channel's max_slots to the target value */
612 __be32 nfs4_callback_recallslot(void *argp, void *resp,
613 struct cb_process_state *cps)
615 struct cb_recallslotargs *args = argp;
616 struct nfs4_slot_table *fc_tbl;
619 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
620 if (!cps->clp) /* set in cb_sequence */
623 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
624 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
625 args->crsa_target_highest_slotid);
627 fc_tbl = &cps->clp->cl_session->fc_slot_table;
629 status = htonl(NFS4_OK);
631 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
632 nfs41_notify_server(cps->clp);
634 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
638 __be32 nfs4_callback_notify_lock(void *argp, void *resp,
639 struct cb_process_state *cps)
641 struct cb_notify_lock_args *args = argp;
643 if (!cps->clp) /* set in cb_sequence */
644 return htonl(NFS4ERR_OP_NOT_IN_SESSION);
646 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
647 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
649 /* Don't wake anybody if the string looked bogus */
650 if (args->cbnl_valid)
651 __wake_up(&cps->clp->cl_lock_waitq, TASK_NORMAL, 0, args);
653 return htonl(NFS4_OK);
655 #endif /* CONFIG_NFS_V4_1 */