2 * linux/fs/nfs/callback_proc.c
4 * Copyright (C) 2004 Trond Myklebust
6 * NFSv4 callback procedures
8 #include <linux/nfs4.h>
9 #include <linux/nfs_fs.h>
10 #include <linux/slab.h>
11 #include <linux/rcupdate.h>
14 #include "delegation.h"
17 #include "nfs4session.h"
18 #include "nfs4trace.h"
20 #define NFSDBG_FACILITY NFSDBG_CALLBACK
22 __be32 nfs4_callback_getattr(struct cb_getattrargs *args,
23 struct cb_getattrres *res,
24 struct cb_process_state *cps)
26 struct nfs_delegation *delegation;
27 struct nfs_inode *nfsi;
30 res->status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
31 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
34 res->bitmap[0] = res->bitmap[1] = 0;
35 res->status = htonl(NFS4ERR_BADHANDLE);
37 dprintk_rcu("NFS: GETATTR callback request from %s\n",
38 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
40 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
42 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
48 delegation = rcu_dereference(nfsi->delegation);
49 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
51 res->size = i_size_read(inode);
52 res->change_attr = delegation->change_attr;
53 if (nfsi->nrequests != 0)
55 res->ctime = inode->i_ctime;
56 res->mtime = inode->i_mtime;
57 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
59 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
64 trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
67 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
71 __be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
72 struct cb_process_state *cps)
77 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
78 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
81 dprintk_rcu("NFS: RECALL callback request from %s\n",
82 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
84 res = htonl(NFS4ERR_BADHANDLE);
85 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
87 trace_nfs4_cb_recall(cps->clp, &args->fh, NULL,
88 &args->stateid, -ntohl(res));
91 /* Set up a helper thread to actually return the delegation */
92 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
97 res = htonl(NFS4ERR_BAD_STATEID);
100 res = htonl(NFS4ERR_RESOURCE);
102 trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
103 &args->stateid, -ntohl(res));
106 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
110 #if defined(CONFIG_NFS_V4_1)
113 * Lookup a layout inode by stateid
115 * Note: returns a refcount on the inode and superblock
117 static struct inode *nfs_layout_find_inode_by_stateid(struct nfs_client *clp,
118 const nfs4_stateid *stateid)
120 struct nfs_server *server;
122 struct pnfs_layout_hdr *lo;
125 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
126 list_for_each_entry(lo, &server->layouts, plh_layouts) {
127 if (stateid != NULL &&
128 !nfs4_stateid_match_other(stateid, &lo->plh_stateid))
130 inode = igrab(lo->plh_inode);
133 if (!nfs_sb_active(inode->i_sb)) {
135 spin_unlock(&clp->cl_lock);
137 spin_lock(&clp->cl_lock);
148 * Lookup a layout inode by filehandle.
150 * Note: returns a refcount on the inode and superblock
153 static struct inode *nfs_layout_find_inode_by_fh(struct nfs_client *clp,
154 const struct nfs_fh *fh)
156 struct nfs_server *server;
157 struct nfs_inode *nfsi;
159 struct pnfs_layout_hdr *lo;
162 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
163 list_for_each_entry(lo, &server->layouts, plh_layouts) {
164 nfsi = NFS_I(lo->plh_inode);
165 if (nfs_compare_fh(fh, &nfsi->fh))
167 if (nfsi->layout != lo)
169 inode = igrab(lo->plh_inode);
172 if (!nfs_sb_active(inode->i_sb)) {
174 spin_unlock(&clp->cl_lock);
176 spin_lock(&clp->cl_lock);
186 static struct inode *nfs_layout_find_inode(struct nfs_client *clp,
187 const struct nfs_fh *fh,
188 const nfs4_stateid *stateid)
192 spin_lock(&clp->cl_lock);
194 inode = nfs_layout_find_inode_by_stateid(clp, stateid);
196 inode = nfs_layout_find_inode_by_fh(clp, fh);
198 spin_unlock(&clp->cl_lock);
204 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
206 static u32 pnfs_check_callback_stateid(struct pnfs_layout_hdr *lo,
207 const nfs4_stateid *new)
211 /* Is the stateid still not initialised? */
212 if (!pnfs_layout_is_valid(lo))
213 return NFS4ERR_DELAY;
215 /* Mismatched stateid? */
216 if (!nfs4_stateid_match_other(&lo->plh_stateid, new))
217 return NFS4ERR_BAD_STATEID;
219 newseq = be32_to_cpu(new->seqid);
220 /* Are we already in a layout recall situation? */
221 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
222 lo->plh_return_seq != 0) {
223 if (newseq < lo->plh_return_seq)
224 return NFS4ERR_OLD_STATEID;
225 if (newseq > lo->plh_return_seq)
226 return NFS4ERR_DELAY;
230 /* Check that the stateid matches what we think it should be. */
231 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
232 if (newseq > oldseq + 1)
233 return NFS4ERR_DELAY;
235 if (newseq <= oldseq)
236 return NFS4ERR_OLD_STATEID;
241 static u32 initiate_file_draining(struct nfs_client *clp,
242 struct cb_layoutrecallargs *args)
245 struct pnfs_layout_hdr *lo;
246 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
247 LIST_HEAD(free_me_list);
249 ino = nfs_layout_find_inode(clp, &args->cbl_fh, &args->cbl_stateid);
253 pnfs_layoutcommit_inode(ino, false);
256 spin_lock(&ino->i_lock);
257 lo = NFS_I(ino)->layout;
259 spin_unlock(&ino->i_lock);
262 pnfs_get_layout_hdr(lo);
263 rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid);
266 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
269 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
271 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
276 if (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
278 be32_to_cpu(args->cbl_stateid.seqid))) {
283 /* Embrace your forgetfulness! */
284 rv = NFS4ERR_NOMATCHING_LAYOUT;
286 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
287 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
291 spin_unlock(&ino->i_lock);
292 pnfs_free_lseg_list(&free_me_list);
293 /* Free all lsegs that are attached to commit buckets */
294 nfs_commit_inode(ino, 0);
295 pnfs_put_layout_hdr(lo);
297 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
298 &args->cbl_stateid, -rv);
299 nfs_iput_and_deactive(ino);
303 static u32 initiate_bulk_draining(struct nfs_client *clp,
304 struct cb_layoutrecallargs *args)
308 if (args->cbl_recall_type == RETURN_FSID)
309 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
311 stat = pnfs_destroy_layouts_byclid(clp, true);
313 return NFS4ERR_DELAY;
314 return NFS4ERR_NOMATCHING_LAYOUT;
317 static u32 do_callback_layoutrecall(struct nfs_client *clp,
318 struct cb_layoutrecallargs *args)
320 if (args->cbl_recall_type == RETURN_FILE)
321 return initiate_file_draining(clp, args);
322 return initiate_bulk_draining(clp, args);
325 __be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
326 void *dummy, struct cb_process_state *cps)
330 dprintk("%s: -->\n", __func__);
333 res = do_callback_layoutrecall(cps->clp, args);
335 res = NFS4ERR_OP_NOT_IN_SESSION;
337 dprintk("%s: exit with status = %d\n", __func__, res);
338 return cpu_to_be32(res);
341 static void pnfs_recall_all_layouts(struct nfs_client *clp)
343 struct cb_layoutrecallargs args;
345 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
346 memset(&args, 0, sizeof(args));
347 args.cbl_recall_type = RETURN_ALL;
348 /* FIXME we ignore errors, what should we do? */
349 do_callback_layoutrecall(clp, &args);
352 __be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
353 void *dummy, struct cb_process_state *cps)
357 struct nfs_client *clp = cps->clp;
358 struct nfs_server *server = NULL;
360 dprintk("%s: -->\n", __func__);
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];
371 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
373 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
374 if (server->pnfs_curr_ld &&
375 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
380 dprintk("%s: layout type %u not found\n",
381 __func__, dev->cbd_layout_type);
386 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
391 dprintk("%s: exit with status = %u\n",
392 __func__, be32_to_cpu(res));
397 * Validate the sequenceID sent by the server.
398 * Return success if the sequenceID is one more than what we last saw on
399 * this slot, accounting for wraparound. Increments the slot's sequence.
401 * We don't yet implement a duplicate request cache, instead we set the
402 * back channel ca_maxresponsesize_cached to zero. This is OK for now
403 * since we only currently implement idempotent callbacks anyway.
405 * We have a single slot backchannel at this time, so we don't bother
406 * checking the used_slots bit array on the table. The lower layer guarantees
407 * a single outstanding callback request at a time.
410 validate_seqid(const struct nfs4_slot_table *tbl, const struct nfs4_slot *slot,
411 const struct cb_sequenceargs * args)
413 dprintk("%s enter. slotid %u seqid %u, slot table seqid: %u\n",
414 __func__, args->csa_slotid, args->csa_sequenceid, slot->seq_nr);
416 if (args->csa_slotid > tbl->server_highest_slotid)
417 return htonl(NFS4ERR_BADSLOT);
420 if (args->csa_sequenceid == slot->seq_nr) {
421 dprintk("%s seqid %u is a replay\n",
422 __func__, args->csa_sequenceid);
423 if (nfs4_test_locked_slot(tbl, slot->slot_nr))
424 return htonl(NFS4ERR_DELAY);
425 /* Signal process_op to set this error on next op */
426 if (args->csa_cachethis == 0)
427 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
429 /* Liar! We never allowed you to set csa_cachethis != 0 */
430 return htonl(NFS4ERR_SEQ_FALSE_RETRY);
434 if (unlikely(slot->seq_nr == 0xFFFFFFFFU)) {
435 if (args->csa_sequenceid == 1)
436 return htonl(NFS4_OK);
437 } else if (likely(args->csa_sequenceid == slot->seq_nr + 1))
438 return htonl(NFS4_OK);
440 /* Misordered request */
441 return htonl(NFS4ERR_SEQ_MISORDERED);
445 * For each referring call triple, check the session's slot table for
446 * a match. If the slot is in use and the sequence numbers match, the
447 * client is still waiting for a response to the original request.
449 static bool referring_call_exists(struct nfs_client *clp,
451 struct referring_call_list *rclists)
455 struct nfs4_session *session;
456 struct nfs4_slot_table *tbl;
457 struct referring_call_list *rclist;
458 struct referring_call *ref;
461 * XXX When client trunking is implemented, this becomes
462 * a session lookup from within the loop
464 session = clp->cl_session;
465 tbl = &session->fc_slot_table;
467 for (i = 0; i < nrclists; i++) {
468 rclist = &rclists[i];
469 if (memcmp(session->sess_id.data,
470 rclist->rcl_sessionid.data,
471 NFS4_MAX_SESSIONID_LEN) != 0)
474 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
475 ref = &rclist->rcl_refcalls[j];
477 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
478 "slotid %u\n", __func__,
479 ((u32 *)&rclist->rcl_sessionid.data)[0],
480 ((u32 *)&rclist->rcl_sessionid.data)[1],
481 ((u32 *)&rclist->rcl_sessionid.data)[2],
482 ((u32 *)&rclist->rcl_sessionid.data)[3],
483 ref->rc_sequenceid, ref->rc_slotid);
485 status = nfs4_slot_wait_on_seqid(tbl, ref->rc_slotid,
486 ref->rc_sequenceid, HZ >> 1) < 0;
496 __be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
497 struct cb_sequenceres *res,
498 struct cb_process_state *cps)
500 struct nfs4_slot_table *tbl;
501 struct nfs4_slot *slot;
502 struct nfs_client *clp;
504 __be32 status = htonl(NFS4ERR_BADSESSION);
506 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
507 &args->csa_sessionid, cps->minorversion);
511 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
514 tbl = &clp->cl_session->bc_slot_table;
516 /* Set up res before grabbing the spinlock */
517 memcpy(&res->csr_sessionid, &args->csa_sessionid,
518 sizeof(res->csr_sessionid));
519 res->csr_sequenceid = args->csa_sequenceid;
520 res->csr_slotid = args->csa_slotid;
522 spin_lock(&tbl->slot_tbl_lock);
523 /* state manager is resetting the session */
524 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
525 status = htonl(NFS4ERR_DELAY);
526 /* Return NFS4ERR_BADSESSION if we're draining the session
527 * in order to reset it.
529 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
530 status = htonl(NFS4ERR_BADSESSION);
534 status = htonl(NFS4ERR_BADSLOT);
535 slot = nfs4_lookup_slot(tbl, args->csa_slotid);
539 res->csr_highestslotid = tbl->server_highest_slotid;
540 res->csr_target_highestslotid = tbl->target_highest_slotid;
542 status = validate_seqid(tbl, slot, args);
545 if (!nfs4_try_to_lock_slot(tbl, slot)) {
546 status = htonl(NFS4ERR_DELAY);
551 /* The ca_maxresponsesize_cached is 0 with no DRC */
552 if (args->csa_cachethis != 0) {
553 status = htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
558 * Check for pending referring calls. If a match is found, a
559 * related callback was received before the response to the original
562 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
563 status = htonl(NFS4ERR_DELAY);
569 * If CB_SEQUENCE returns an error, then the state of the slot
570 * (sequence ID, cached reply) MUST NOT change.
572 slot->seq_nr = args->csa_sequenceid;
574 spin_unlock(&tbl->slot_tbl_lock);
577 cps->clp = clp; /* put in nfs4_callback_compound */
578 for (i = 0; i < args->csa_nrclists; i++)
579 kfree(args->csa_rclists[i].rcl_refcalls);
580 kfree(args->csa_rclists);
582 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
583 cps->drc_status = status;
586 res->csr_status = status;
588 trace_nfs4_cb_sequence(args, res, status);
589 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
590 ntohl(status), ntohl(res->csr_status));
595 validate_bitmap_values(unsigned long mask)
597 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
600 __be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
601 struct cb_process_state *cps)
606 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
607 if (!cps->clp) /* set in cb_sequence */
610 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
611 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
613 status = cpu_to_be32(NFS4ERR_INVAL);
614 if (!validate_bitmap_values(args->craa_type_mask))
617 status = cpu_to_be32(NFS4_OK);
618 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
619 &args->craa_type_mask))
621 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
622 &args->craa_type_mask))
623 flags |= FMODE_WRITE;
624 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
625 &args->craa_type_mask))
626 pnfs_recall_all_layouts(cps->clp);
628 nfs_expire_unused_delegation_types(cps->clp, flags);
630 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
634 /* Reduce the fore channel's max_slots to the target value */
635 __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
636 struct cb_process_state *cps)
638 struct nfs4_slot_table *fc_tbl;
641 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
642 if (!cps->clp) /* set in cb_sequence */
645 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
646 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
647 args->crsa_target_highest_slotid);
649 fc_tbl = &cps->clp->cl_session->fc_slot_table;
651 status = htonl(NFS4_OK);
653 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
654 nfs41_notify_server(cps->clp);
656 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
660 __be32 nfs4_callback_notify_lock(struct cb_notify_lock_args *args, void *dummy,
661 struct cb_process_state *cps)
663 if (!cps->clp) /* set in cb_sequence */
664 return htonl(NFS4ERR_OP_NOT_IN_SESSION);
666 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
667 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
669 /* Don't wake anybody if the string looked bogus */
670 if (args->cbnl_valid)
671 __wake_up(&cps->clp->cl_lock_waitq, TASK_NORMAL, 0, args);
673 return htonl(NFS4_OK);
675 #endif /* CONFIG_NFS_V4_1 */