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);
88 /* Set up a helper thread to actually return the delegation */
89 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
94 res = htonl(NFS4ERR_BAD_STATEID);
97 res = htonl(NFS4ERR_RESOURCE);
99 trace_nfs4_recall_delegation(inode, -ntohl(res));
102 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
106 #if defined(CONFIG_NFS_V4_1)
109 * Lookup a layout by filehandle.
111 * Note: gets a refcount on the layout hdr and on its respective inode.
112 * Caller must put the layout hdr and the inode.
114 * TODO: keep track of all layouts (and delegations) in a hash table
115 * hashed by filehandle.
117 static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp,
118 struct nfs_fh *fh, nfs4_stateid *stateid)
120 struct nfs_server *server;
122 struct pnfs_layout_hdr *lo;
124 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
125 list_for_each_entry(lo, &server->layouts, plh_layouts) {
126 if (!nfs4_stateid_match_other(&lo->plh_stateid, stateid))
128 if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
130 ino = igrab(lo->plh_inode);
133 spin_lock(&ino->i_lock);
134 /* Is this layout in the process of being freed? */
135 if (NFS_I(ino)->layout != lo) {
136 spin_unlock(&ino->i_lock);
140 pnfs_get_layout_hdr(lo);
141 spin_unlock(&ino->i_lock);
149 static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp,
150 struct nfs_fh *fh, nfs4_stateid *stateid)
152 struct pnfs_layout_hdr *lo;
154 spin_lock(&clp->cl_lock);
156 lo = get_layout_by_fh_locked(clp, fh, stateid);
158 spin_unlock(&clp->cl_lock);
164 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
166 static bool pnfs_check_stateid_sequence(struct pnfs_layout_hdr *lo,
167 const nfs4_stateid *new)
171 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
172 newseq = be32_to_cpu(new->seqid);
174 if (newseq > oldseq + 1)
179 static u32 initiate_file_draining(struct nfs_client *clp,
180 struct cb_layoutrecallargs *args)
183 struct pnfs_layout_hdr *lo;
184 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
185 LIST_HEAD(free_me_list);
187 lo = get_layout_by_fh(clp, &args->cbl_fh, &args->cbl_stateid);
193 spin_lock(&ino->i_lock);
194 if (!pnfs_check_stateid_sequence(lo, &args->cbl_stateid)) {
198 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
199 spin_unlock(&ino->i_lock);
201 pnfs_layoutcommit_inode(ino, false);
203 spin_lock(&ino->i_lock);
205 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
207 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
212 if (pnfs_mark_matching_lsegs_invalid(lo, &free_me_list,
214 pnfs_mark_matching_lsegs_return(lo,
221 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
222 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
225 pnfs_mark_layout_returned_if_empty(lo);
227 spin_unlock(&ino->i_lock);
228 pnfs_free_lseg_list(&free_me_list);
229 pnfs_put_layout_hdr(lo);
230 trace_nfs4_cb_layoutrecall_inode(clp, &args->cbl_fh, ino, -rv);
236 static u32 initiate_bulk_draining(struct nfs_client *clp,
237 struct cb_layoutrecallargs *args)
241 if (args->cbl_recall_type == RETURN_FSID)
242 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
244 stat = pnfs_destroy_layouts_byclid(clp, true);
246 return NFS4ERR_DELAY;
247 return NFS4ERR_NOMATCHING_LAYOUT;
250 static u32 do_callback_layoutrecall(struct nfs_client *clp,
251 struct cb_layoutrecallargs *args)
255 dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
256 if (args->cbl_recall_type == RETURN_FILE)
257 res = initiate_file_draining(clp, args);
259 res = initiate_bulk_draining(clp, args);
260 dprintk("%s returning %i\n", __func__, res);
265 __be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
266 void *dummy, struct cb_process_state *cps)
270 dprintk("%s: -->\n", __func__);
273 res = do_callback_layoutrecall(cps->clp, args);
275 res = NFS4ERR_OP_NOT_IN_SESSION;
277 dprintk("%s: exit with status = %d\n", __func__, res);
278 return cpu_to_be32(res);
281 static void pnfs_recall_all_layouts(struct nfs_client *clp)
283 struct cb_layoutrecallargs args;
285 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
286 memset(&args, 0, sizeof(args));
287 args.cbl_recall_type = RETURN_ALL;
288 /* FIXME we ignore errors, what should we do? */
289 do_callback_layoutrecall(clp, &args);
292 __be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
293 void *dummy, struct cb_process_state *cps)
297 struct nfs_client *clp = cps->clp;
298 struct nfs_server *server = NULL;
300 dprintk("%s: -->\n", __func__);
303 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
307 for (i = 0; i < args->ndevs; i++) {
308 struct cb_devicenotifyitem *dev = &args->devs[i];
311 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
313 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
314 if (server->pnfs_curr_ld &&
315 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
320 dprintk("%s: layout type %u not found\n",
321 __func__, dev->cbd_layout_type);
326 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
331 dprintk("%s: exit with status = %u\n",
332 __func__, be32_to_cpu(res));
337 * Validate the sequenceID sent by the server.
338 * Return success if the sequenceID is one more than what we last saw on
339 * this slot, accounting for wraparound. Increments the slot's sequence.
341 * We don't yet implement a duplicate request cache, instead we set the
342 * back channel ca_maxresponsesize_cached to zero. This is OK for now
343 * since we only currently implement idempotent callbacks anyway.
345 * We have a single slot backchannel at this time, so we don't bother
346 * checking the used_slots bit array on the table. The lower layer guarantees
347 * a single outstanding callback request at a time.
350 validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
352 struct nfs4_slot *slot;
354 dprintk("%s enter. slotid %u seqid %u\n",
355 __func__, args->csa_slotid, args->csa_sequenceid);
357 if (args->csa_slotid >= NFS41_BC_MAX_CALLBACKS)
358 return htonl(NFS4ERR_BADSLOT);
360 slot = tbl->slots + args->csa_slotid;
361 dprintk("%s slot table seqid: %u\n", __func__, slot->seq_nr);
364 if (likely(args->csa_sequenceid == slot->seq_nr + 1))
368 if (args->csa_sequenceid == slot->seq_nr) {
369 dprintk("%s seqid %u is a replay\n",
370 __func__, args->csa_sequenceid);
371 /* Signal process_op to set this error on next op */
372 if (args->csa_cachethis == 0)
373 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
375 /* The ca_maxresponsesize_cached is 0 with no DRC */
376 else if (args->csa_cachethis == 1)
377 return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
381 if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
386 /* Misordered request */
387 return htonl(NFS4ERR_SEQ_MISORDERED);
389 tbl->highest_used_slotid = args->csa_slotid;
390 return htonl(NFS4_OK);
394 * For each referring call triple, check the session's slot table for
395 * a match. If the slot is in use and the sequence numbers match, the
396 * client is still waiting for a response to the original request.
398 static bool referring_call_exists(struct nfs_client *clp,
400 struct referring_call_list *rclists)
404 struct nfs4_session *session;
405 struct nfs4_slot_table *tbl;
406 struct referring_call_list *rclist;
407 struct referring_call *ref;
410 * XXX When client trunking is implemented, this becomes
411 * a session lookup from within the loop
413 session = clp->cl_session;
414 tbl = &session->fc_slot_table;
416 for (i = 0; i < nrclists; i++) {
417 rclist = &rclists[i];
418 if (memcmp(session->sess_id.data,
419 rclist->rcl_sessionid.data,
420 NFS4_MAX_SESSIONID_LEN) != 0)
423 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
424 ref = &rclist->rcl_refcalls[j];
426 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
427 "slotid %u\n", __func__,
428 ((u32 *)&rclist->rcl_sessionid.data)[0],
429 ((u32 *)&rclist->rcl_sessionid.data)[1],
430 ((u32 *)&rclist->rcl_sessionid.data)[2],
431 ((u32 *)&rclist->rcl_sessionid.data)[3],
432 ref->rc_sequenceid, ref->rc_slotid);
434 spin_lock(&tbl->slot_tbl_lock);
435 status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
436 tbl->slots[ref->rc_slotid].seq_nr ==
438 spin_unlock(&tbl->slot_tbl_lock);
448 __be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
449 struct cb_sequenceres *res,
450 struct cb_process_state *cps)
452 struct nfs4_slot_table *tbl;
453 struct nfs4_slot *slot;
454 struct nfs_client *clp;
456 __be32 status = htonl(NFS4ERR_BADSESSION);
458 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
459 &args->csa_sessionid, cps->minorversion);
463 if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
466 tbl = &clp->cl_session->bc_slot_table;
467 slot = tbl->slots + args->csa_slotid;
469 spin_lock(&tbl->slot_tbl_lock);
470 /* state manager is resetting the session */
471 if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
472 status = htonl(NFS4ERR_DELAY);
473 /* Return NFS4ERR_BADSESSION if we're draining the session
474 * in order to reset it.
476 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
477 status = htonl(NFS4ERR_BADSESSION);
481 memcpy(&res->csr_sessionid, &args->csa_sessionid,
482 sizeof(res->csr_sessionid));
483 res->csr_sequenceid = args->csa_sequenceid;
484 res->csr_slotid = args->csa_slotid;
485 res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
486 res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
488 status = validate_seqid(tbl, args);
492 cps->slotid = args->csa_slotid;
495 * Check for pending referring calls. If a match is found, a
496 * related callback was received before the response to the original
499 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
500 status = htonl(NFS4ERR_DELAY);
506 * If CB_SEQUENCE returns an error, then the state of the slot
507 * (sequence ID, cached reply) MUST NOT change.
511 spin_unlock(&tbl->slot_tbl_lock);
514 cps->clp = clp; /* put in nfs4_callback_compound */
515 for (i = 0; i < args->csa_nrclists; i++)
516 kfree(args->csa_rclists[i].rcl_refcalls);
517 kfree(args->csa_rclists);
519 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
520 cps->drc_status = status;
523 res->csr_status = status;
525 trace_nfs4_cb_sequence(args, res, status);
526 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
527 ntohl(status), ntohl(res->csr_status));
532 validate_bitmap_values(unsigned long mask)
534 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
537 __be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
538 struct cb_process_state *cps)
543 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
544 if (!cps->clp) /* set in cb_sequence */
547 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
548 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
550 status = cpu_to_be32(NFS4ERR_INVAL);
551 if (!validate_bitmap_values(args->craa_type_mask))
554 status = cpu_to_be32(NFS4_OK);
555 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
556 &args->craa_type_mask))
558 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
559 &args->craa_type_mask))
560 flags |= FMODE_WRITE;
561 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
562 &args->craa_type_mask))
563 pnfs_recall_all_layouts(cps->clp);
565 nfs_expire_unused_delegation_types(cps->clp, flags);
567 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
571 /* Reduce the fore channel's max_slots to the target value */
572 __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
573 struct cb_process_state *cps)
575 struct nfs4_slot_table *fc_tbl;
578 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
579 if (!cps->clp) /* set in cb_sequence */
582 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
583 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
584 args->crsa_target_highest_slotid);
586 fc_tbl = &cps->clp->cl_session->fc_slot_table;
588 status = htonl(NFS4_OK);
590 nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
591 nfs41_notify_server(cps->clp);
593 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
596 #endif /* CONFIG_NFS_V4_1 */