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"
19 #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);
45 delegation = rcu_dereference(nfsi->delegation);
46 if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
48 res->size = i_size_read(inode);
49 res->change_attr = delegation->change_attr;
50 if (nfsi->npages != 0)
52 res->ctime = inode->i_ctime;
53 res->mtime = inode->i_mtime;
54 res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
56 res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
63 dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
67 __be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
68 struct cb_process_state *cps)
73 res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
74 if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
77 dprintk_rcu("NFS: RECALL callback request from %s\n",
78 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
80 res = htonl(NFS4ERR_BADHANDLE);
81 inode = nfs_delegation_find_inode(cps->clp, &args->fh);
84 /* Set up a helper thread to actually return the delegation */
85 switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
91 res = htonl(NFS4ERR_BAD_STATEID);
94 res = htonl(NFS4ERR_RESOURCE);
98 dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
102 int nfs4_validate_delegation_stateid(struct nfs_delegation *delegation, const nfs4_stateid *stateid)
104 if (delegation == NULL || memcmp(delegation->stateid.data, stateid->data,
105 sizeof(delegation->stateid.data)) != 0)
110 #if defined(CONFIG_NFS_V4_1)
113 * Lookup a layout by filehandle.
115 * Note: gets a refcount on the layout hdr and on its respective inode.
116 * Caller must put the layout hdr and the inode.
118 * TODO: keep track of all layouts (and delegations) in a hash table
119 * hashed by filehandle.
121 static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp, struct nfs_fh *fh)
123 struct nfs_server *server;
125 struct pnfs_layout_hdr *lo;
127 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
128 list_for_each_entry(lo, &server->layouts, plh_layouts) {
129 if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
131 ino = igrab(lo->plh_inode);
142 static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp, struct nfs_fh *fh)
144 struct pnfs_layout_hdr *lo;
146 spin_lock(&clp->cl_lock);
148 lo = get_layout_by_fh_locked(clp, fh);
150 spin_unlock(&clp->cl_lock);
155 static u32 initiate_file_draining(struct nfs_client *clp,
156 struct cb_layoutrecallargs *args)
159 struct pnfs_layout_hdr *lo;
160 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
161 LIST_HEAD(free_me_list);
163 lo = get_layout_by_fh(clp, &args->cbl_fh);
165 return NFS4ERR_NOMATCHING_LAYOUT;
168 spin_lock(&ino->i_lock);
169 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
170 mark_matching_lsegs_invalid(lo, &free_me_list,
174 rv = NFS4ERR_NOMATCHING_LAYOUT;
175 pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
176 spin_unlock(&ino->i_lock);
177 pnfs_free_lseg_list(&free_me_list);
183 static u32 initiate_bulk_draining(struct nfs_client *clp,
184 struct cb_layoutrecallargs *args)
186 struct nfs_server *server;
187 struct pnfs_layout_hdr *lo;
189 u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
190 struct pnfs_layout_hdr *tmp;
191 LIST_HEAD(recall_list);
192 LIST_HEAD(free_me_list);
193 struct pnfs_layout_range range = {
194 .iomode = IOMODE_ANY,
196 .length = NFS4_MAX_UINT64,
199 spin_lock(&clp->cl_lock);
201 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
202 if ((args->cbl_recall_type == RETURN_FSID) &&
203 memcmp(&server->fsid, &args->cbl_fsid,
204 sizeof(struct nfs_fsid)))
207 list_for_each_entry(lo, &server->layouts, plh_layouts) {
208 if (!igrab(lo->plh_inode))
211 BUG_ON(!list_empty(&lo->plh_bulk_recall));
212 list_add(&lo->plh_bulk_recall, &recall_list);
216 spin_unlock(&clp->cl_lock);
218 list_for_each_entry_safe(lo, tmp,
219 &recall_list, plh_bulk_recall) {
221 spin_lock(&ino->i_lock);
222 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
223 if (mark_matching_lsegs_invalid(lo, &free_me_list, &range))
225 list_del_init(&lo->plh_bulk_recall);
226 spin_unlock(&ino->i_lock);
227 pnfs_free_lseg_list(&free_me_list);
234 static u32 do_callback_layoutrecall(struct nfs_client *clp,
235 struct cb_layoutrecallargs *args)
239 dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
240 if (args->cbl_recall_type == RETURN_FILE)
241 res = initiate_file_draining(clp, args);
243 res = initiate_bulk_draining(clp, args);
244 dprintk("%s returning %i\n", __func__, res);
249 __be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
250 void *dummy, struct cb_process_state *cps)
254 dprintk("%s: -->\n", __func__);
257 res = do_callback_layoutrecall(cps->clp, args);
259 res = NFS4ERR_OP_NOT_IN_SESSION;
261 dprintk("%s: exit with status = %d\n", __func__, res);
262 return cpu_to_be32(res);
265 static void pnfs_recall_all_layouts(struct nfs_client *clp)
267 struct cb_layoutrecallargs args;
269 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
270 memset(&args, 0, sizeof(args));
271 args.cbl_recall_type = RETURN_ALL;
272 /* FIXME we ignore errors, what should we do? */
273 do_callback_layoutrecall(clp, &args);
276 __be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
277 void *dummy, struct cb_process_state *cps)
281 struct nfs_client *clp = cps->clp;
282 struct nfs_server *server = NULL;
284 dprintk("%s: -->\n", __func__);
287 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
291 for (i = 0; i < args->ndevs; i++) {
292 struct cb_devicenotifyitem *dev = &args->devs[i];
295 server->pnfs_curr_ld->id != dev->cbd_layout_type) {
297 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
298 if (server->pnfs_curr_ld &&
299 server->pnfs_curr_ld->id == dev->cbd_layout_type) {
304 dprintk("%s: layout type %u not found\n",
305 __func__, dev->cbd_layout_type);
310 if (dev->cbd_notify_type == NOTIFY_DEVICEID4_CHANGE)
311 dprintk("%s: NOTIFY_DEVICEID4_CHANGE not supported, "
312 "deleting instead\n", __func__);
313 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
318 dprintk("%s: exit with status = %u\n",
319 __func__, be32_to_cpu(res));
323 int nfs41_validate_delegation_stateid(struct nfs_delegation *delegation, const nfs4_stateid *stateid)
325 if (delegation == NULL)
328 if (stateid->stateid.seqid != 0)
330 if (memcmp(&delegation->stateid.stateid.other,
331 &stateid->stateid.other,
332 NFS4_STATEID_OTHER_SIZE))
339 * Validate the sequenceID sent by the server.
340 * Return success if the sequenceID is one more than what we last saw on
341 * this slot, accounting for wraparound. Increments the slot's sequence.
343 * We don't yet implement a duplicate request cache, instead we set the
344 * back channel ca_maxresponsesize_cached to zero. This is OK for now
345 * since we only currently implement idempotent callbacks anyway.
347 * We have a single slot backchannel at this time, so we don't bother
348 * checking the used_slots bit array on the table. The lower layer guarantees
349 * a single outstanding callback request at a time.
352 validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
354 struct nfs4_slot *slot;
356 dprintk("%s enter. slotid %d seqid %d\n",
357 __func__, args->csa_slotid, args->csa_sequenceid);
359 if (args->csa_slotid >= NFS41_BC_MAX_CALLBACKS)
360 return htonl(NFS4ERR_BADSLOT);
362 slot = tbl->slots + args->csa_slotid;
363 dprintk("%s slot table seqid: %d\n", __func__, slot->seq_nr);
366 if (likely(args->csa_sequenceid == slot->seq_nr + 1)) {
372 if (args->csa_sequenceid == slot->seq_nr) {
373 dprintk("%s seqid %d is a replay\n",
374 __func__, args->csa_sequenceid);
375 /* Signal process_op to set this error on next op */
376 if (args->csa_cachethis == 0)
377 return htonl(NFS4ERR_RETRY_UNCACHED_REP);
379 /* The ca_maxresponsesize_cached is 0 with no DRC */
380 else if (args->csa_cachethis == 1)
381 return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
385 if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
390 /* Misordered request */
391 return htonl(NFS4ERR_SEQ_MISORDERED);
393 tbl->highest_used_slotid = args->csa_slotid;
394 return htonl(NFS4_OK);
398 * For each referring call triple, check the session's slot table for
399 * a match. If the slot is in use and the sequence numbers match, the
400 * client is still waiting for a response to the original request.
402 static bool referring_call_exists(struct nfs_client *clp,
404 struct referring_call_list *rclists)
408 struct nfs4_session *session;
409 struct nfs4_slot_table *tbl;
410 struct referring_call_list *rclist;
411 struct referring_call *ref;
414 * XXX When client trunking is implemented, this becomes
415 * a session lookup from within the loop
417 session = clp->cl_session;
418 tbl = &session->fc_slot_table;
420 for (i = 0; i < nrclists; i++) {
421 rclist = &rclists[i];
422 if (memcmp(session->sess_id.data,
423 rclist->rcl_sessionid.data,
424 NFS4_MAX_SESSIONID_LEN) != 0)
427 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
428 ref = &rclist->rcl_refcalls[j];
430 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
431 "slotid %u\n", __func__,
432 ((u32 *)&rclist->rcl_sessionid.data)[0],
433 ((u32 *)&rclist->rcl_sessionid.data)[1],
434 ((u32 *)&rclist->rcl_sessionid.data)[2],
435 ((u32 *)&rclist->rcl_sessionid.data)[3],
436 ref->rc_sequenceid, ref->rc_slotid);
438 spin_lock(&tbl->slot_tbl_lock);
439 status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
440 tbl->slots[ref->rc_slotid].seq_nr ==
442 spin_unlock(&tbl->slot_tbl_lock);
452 __be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
453 struct cb_sequenceres *res,
454 struct cb_process_state *cps)
456 struct nfs4_slot_table *tbl;
457 struct nfs_client *clp;
459 __be32 status = htonl(NFS4ERR_BADSESSION);
461 clp = nfs4_find_client_sessionid(cps->net, args->csa_addr, &args->csa_sessionid);
465 tbl = &clp->cl_session->bc_slot_table;
467 spin_lock(&tbl->slot_tbl_lock);
468 /* state manager is resetting the session */
469 if (test_bit(NFS4_SESSION_DRAINING, &clp->cl_session->session_state)) {
470 spin_unlock(&tbl->slot_tbl_lock);
471 status = htonl(NFS4ERR_DELAY);
472 /* Return NFS4ERR_BADSESSION if we're draining the session
473 * in order to reset it.
475 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
476 status = htonl(NFS4ERR_BADSESSION);
480 status = validate_seqid(&clp->cl_session->bc_slot_table, args);
481 spin_unlock(&tbl->slot_tbl_lock);
485 cps->slotid = args->csa_slotid;
488 * Check for pending referring calls. If a match is found, a
489 * related callback was received before the response to the original
492 if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
493 status = htonl(NFS4ERR_DELAY);
497 memcpy(&res->csr_sessionid, &args->csa_sessionid,
498 sizeof(res->csr_sessionid));
499 res->csr_sequenceid = args->csa_sequenceid;
500 res->csr_slotid = args->csa_slotid;
501 res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
502 res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
505 cps->clp = clp; /* put in nfs4_callback_compound */
506 for (i = 0; i < args->csa_nrclists; i++)
507 kfree(args->csa_rclists[i].rcl_refcalls);
508 kfree(args->csa_rclists);
510 if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
511 cps->drc_status = status;
514 res->csr_status = status;
516 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
517 ntohl(status), ntohl(res->csr_status));
522 validate_bitmap_values(unsigned long mask)
524 return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
527 __be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
528 struct cb_process_state *cps)
533 status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
534 if (!cps->clp) /* set in cb_sequence */
537 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
538 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
540 status = cpu_to_be32(NFS4ERR_INVAL);
541 if (!validate_bitmap_values(args->craa_type_mask))
544 status = cpu_to_be32(NFS4_OK);
545 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
546 &args->craa_type_mask))
548 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
549 &args->craa_type_mask))
550 flags |= FMODE_WRITE;
551 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
552 &args->craa_type_mask))
553 pnfs_recall_all_layouts(cps->clp);
555 nfs_expire_all_delegation_types(cps->clp, flags);
557 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
561 /* Reduce the fore channel's max_slots to the target value */
562 __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
563 struct cb_process_state *cps)
565 struct nfs4_slot_table *fc_tbl;
568 status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
569 if (!cps->clp) /* set in cb_sequence */
572 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target max slots %d\n",
573 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
574 args->crsa_target_max_slots);
576 fc_tbl = &cps->clp->cl_session->fc_slot_table;
578 status = htonl(NFS4ERR_BAD_HIGH_SLOT);
579 if (args->crsa_target_max_slots > fc_tbl->max_slots ||
580 args->crsa_target_max_slots < 1)
583 status = htonl(NFS4_OK);
584 if (args->crsa_target_max_slots == fc_tbl->max_slots)
587 fc_tbl->target_max_slots = args->crsa_target_max_slots;
588 nfs41_handle_recall_slot(cps->clp);
590 dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
593 #endif /* CONFIG_NFS_V4_1 */