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
)
322 dprintk("%s enter, type=%i\n", __func__
, args
->cbl_recall_type
);
323 if (args
->cbl_recall_type
== RETURN_FILE
)
324 res
= initiate_file_draining(clp
, args
);
326 res
= initiate_bulk_draining(clp
, args
);
327 dprintk("%s returning %i\n", __func__
, res
);
332 __be32
nfs4_callback_layoutrecall(struct cb_layoutrecallargs
*args
,
333 void *dummy
, struct cb_process_state
*cps
)
337 dprintk("%s: -->\n", __func__
);
340 res
= do_callback_layoutrecall(cps
->clp
, args
);
342 res
= NFS4ERR_OP_NOT_IN_SESSION
;
344 dprintk("%s: exit with status = %d\n", __func__
, res
);
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(struct cb_devicenotifyargs
*args
,
360 void *dummy
, struct cb_process_state
*cps
)
364 struct nfs_client
*clp
= cps
->clp
;
365 struct nfs_server
*server
= NULL
;
367 dprintk("%s: -->\n", __func__
);
370 res
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
374 for (i
= 0; i
< args
->ndevs
; i
++) {
375 struct cb_devicenotifyitem
*dev
= &args
->devs
[i
];
378 server
->pnfs_curr_ld
->id
!= dev
->cbd_layout_type
) {
380 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
)
381 if (server
->pnfs_curr_ld
&&
382 server
->pnfs_curr_ld
->id
== dev
->cbd_layout_type
) {
387 dprintk("%s: layout type %u not found\n",
388 __func__
, dev
->cbd_layout_type
);
393 nfs4_delete_deviceid(server
->pnfs_curr_ld
, clp
, &dev
->cbd_dev_id
);
398 dprintk("%s: exit with status = %u\n",
399 __func__
, be32_to_cpu(res
));
404 * Validate the sequenceID sent by the server.
405 * Return success if the sequenceID is one more than what we last saw on
406 * this slot, accounting for wraparound. Increments the slot's sequence.
408 * We don't yet implement a duplicate request cache, instead we set the
409 * back channel ca_maxresponsesize_cached to zero. This is OK for now
410 * since we only currently implement idempotent callbacks anyway.
412 * We have a single slot backchannel at this time, so we don't bother
413 * checking the used_slots bit array on the table. The lower layer guarantees
414 * a single outstanding callback request at a time.
417 validate_seqid(const struct nfs4_slot_table
*tbl
, const struct nfs4_slot
*slot
,
418 const struct cb_sequenceargs
* args
)
420 dprintk("%s enter. slotid %u seqid %u, slot table seqid: %u\n",
421 __func__
, args
->csa_slotid
, args
->csa_sequenceid
, slot
->seq_nr
);
423 if (args
->csa_slotid
> tbl
->server_highest_slotid
)
424 return htonl(NFS4ERR_BADSLOT
);
427 if (args
->csa_sequenceid
== slot
->seq_nr
) {
428 dprintk("%s seqid %u is a replay\n",
429 __func__
, args
->csa_sequenceid
);
430 if (nfs4_test_locked_slot(tbl
, slot
->slot_nr
))
431 return htonl(NFS4ERR_DELAY
);
432 /* Signal process_op to set this error on next op */
433 if (args
->csa_cachethis
== 0)
434 return htonl(NFS4ERR_RETRY_UNCACHED_REP
);
436 /* Liar! We never allowed you to set csa_cachethis != 0 */
437 return htonl(NFS4ERR_SEQ_FALSE_RETRY
);
441 if (unlikely(slot
->seq_nr
== 0xFFFFFFFFU
)) {
442 if (args
->csa_sequenceid
== 1)
443 return htonl(NFS4_OK
);
444 } else if (likely(args
->csa_sequenceid
== slot
->seq_nr
+ 1))
445 return htonl(NFS4_OK
);
447 /* Misordered request */
448 return htonl(NFS4ERR_SEQ_MISORDERED
);
452 * For each referring call triple, check the session's slot table for
453 * a match. If the slot is in use and the sequence numbers match, the
454 * client is still waiting for a response to the original request.
456 static bool referring_call_exists(struct nfs_client
*clp
,
458 struct referring_call_list
*rclists
)
462 struct nfs4_session
*session
;
463 struct nfs4_slot_table
*tbl
;
464 struct referring_call_list
*rclist
;
465 struct referring_call
*ref
;
468 * XXX When client trunking is implemented, this becomes
469 * a session lookup from within the loop
471 session
= clp
->cl_session
;
472 tbl
= &session
->fc_slot_table
;
474 for (i
= 0; i
< nrclists
; i
++) {
475 rclist
= &rclists
[i
];
476 if (memcmp(session
->sess_id
.data
,
477 rclist
->rcl_sessionid
.data
,
478 NFS4_MAX_SESSIONID_LEN
) != 0)
481 for (j
= 0; j
< rclist
->rcl_nrefcalls
; j
++) {
482 ref
= &rclist
->rcl_refcalls
[j
];
484 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
485 "slotid %u\n", __func__
,
486 ((u32
*)&rclist
->rcl_sessionid
.data
)[0],
487 ((u32
*)&rclist
->rcl_sessionid
.data
)[1],
488 ((u32
*)&rclist
->rcl_sessionid
.data
)[2],
489 ((u32
*)&rclist
->rcl_sessionid
.data
)[3],
490 ref
->rc_sequenceid
, ref
->rc_slotid
);
492 status
= nfs4_slot_wait_on_seqid(tbl
, ref
->rc_slotid
,
493 ref
->rc_sequenceid
, HZ
>> 1) < 0;
503 __be32
nfs4_callback_sequence(struct cb_sequenceargs
*args
,
504 struct cb_sequenceres
*res
,
505 struct cb_process_state
*cps
)
507 struct nfs4_slot_table
*tbl
;
508 struct nfs4_slot
*slot
;
509 struct nfs_client
*clp
;
511 __be32 status
= htonl(NFS4ERR_BADSESSION
);
513 clp
= nfs4_find_client_sessionid(cps
->net
, args
->csa_addr
,
514 &args
->csa_sessionid
, cps
->minorversion
);
518 if (!(clp
->cl_session
->flags
& SESSION4_BACK_CHAN
))
521 tbl
= &clp
->cl_session
->bc_slot_table
;
523 /* Set up res before grabbing the spinlock */
524 memcpy(&res
->csr_sessionid
, &args
->csa_sessionid
,
525 sizeof(res
->csr_sessionid
));
526 res
->csr_sequenceid
= args
->csa_sequenceid
;
527 res
->csr_slotid
= args
->csa_slotid
;
529 spin_lock(&tbl
->slot_tbl_lock
);
530 /* state manager is resetting the session */
531 if (test_bit(NFS4_SLOT_TBL_DRAINING
, &tbl
->slot_tbl_state
)) {
532 status
= htonl(NFS4ERR_DELAY
);
533 /* Return NFS4ERR_BADSESSION if we're draining the session
534 * in order to reset it.
536 if (test_bit(NFS4CLNT_SESSION_RESET
, &clp
->cl_state
))
537 status
= htonl(NFS4ERR_BADSESSION
);
541 status
= htonl(NFS4ERR_BADSLOT
);
542 slot
= nfs4_lookup_slot(tbl
, args
->csa_slotid
);
546 res
->csr_highestslotid
= tbl
->server_highest_slotid
;
547 res
->csr_target_highestslotid
= tbl
->target_highest_slotid
;
549 status
= validate_seqid(tbl
, slot
, args
);
552 if (!nfs4_try_to_lock_slot(tbl
, slot
)) {
553 status
= htonl(NFS4ERR_DELAY
);
558 /* The ca_maxresponsesize_cached is 0 with no DRC */
559 if (args
->csa_cachethis
!= 0) {
560 status
= htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE
);
565 * Check for pending referring calls. If a match is found, a
566 * related callback was received before the response to the original
569 if (referring_call_exists(clp
, args
->csa_nrclists
, args
->csa_rclists
)) {
570 status
= htonl(NFS4ERR_DELAY
);
576 * If CB_SEQUENCE returns an error, then the state of the slot
577 * (sequence ID, cached reply) MUST NOT change.
579 slot
->seq_nr
= args
->csa_sequenceid
;
581 spin_unlock(&tbl
->slot_tbl_lock
);
584 cps
->clp
= clp
; /* put in nfs4_callback_compound */
585 for (i
= 0; i
< args
->csa_nrclists
; i
++)
586 kfree(args
->csa_rclists
[i
].rcl_refcalls
);
587 kfree(args
->csa_rclists
);
589 if (status
== htonl(NFS4ERR_RETRY_UNCACHED_REP
)) {
590 cps
->drc_status
= status
;
593 res
->csr_status
= status
;
595 trace_nfs4_cb_sequence(args
, res
, status
);
596 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__
,
597 ntohl(status
), ntohl(res
->csr_status
));
602 validate_bitmap_values(unsigned long mask
)
604 return (mask
& ~RCA4_TYPE_MASK_ALL
) == 0;
607 __be32
nfs4_callback_recallany(struct cb_recallanyargs
*args
, void *dummy
,
608 struct cb_process_state
*cps
)
613 status
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
614 if (!cps
->clp
) /* set in cb_sequence */
617 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
618 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
620 status
= cpu_to_be32(NFS4ERR_INVAL
);
621 if (!validate_bitmap_values(args
->craa_type_mask
))
624 status
= cpu_to_be32(NFS4_OK
);
625 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG
, (const unsigned long *)
626 &args
->craa_type_mask
))
628 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG
, (const unsigned long *)
629 &args
->craa_type_mask
))
630 flags
|= FMODE_WRITE
;
631 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT
, (const unsigned long *)
632 &args
->craa_type_mask
))
633 pnfs_recall_all_layouts(cps
->clp
);
635 nfs_expire_unused_delegation_types(cps
->clp
, flags
);
637 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
641 /* Reduce the fore channel's max_slots to the target value */
642 __be32
nfs4_callback_recallslot(struct cb_recallslotargs
*args
, void *dummy
,
643 struct cb_process_state
*cps
)
645 struct nfs4_slot_table
*fc_tbl
;
648 status
= htonl(NFS4ERR_OP_NOT_IN_SESSION
);
649 if (!cps
->clp
) /* set in cb_sequence */
652 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
653 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
),
654 args
->crsa_target_highest_slotid
);
656 fc_tbl
= &cps
->clp
->cl_session
->fc_slot_table
;
658 status
= htonl(NFS4_OK
);
660 nfs41_set_target_slotid(fc_tbl
, args
->crsa_target_highest_slotid
);
661 nfs41_notify_server(cps
->clp
);
663 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
667 __be32
nfs4_callback_notify_lock(struct cb_notify_lock_args
*args
, void *dummy
,
668 struct cb_process_state
*cps
)
670 if (!cps
->clp
) /* set in cb_sequence */
671 return htonl(NFS4ERR_OP_NOT_IN_SESSION
);
673 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
674 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
676 /* Don't wake anybody if the string looked bogus */
677 if (args
->cbnl_valid
)
678 __wake_up(&cps
->clp
->cl_lock_waitq
, TASK_NORMAL
, 0, args
);
680 return htonl(NFS4_OK
);
682 #endif /* CONFIG_NFS_V4_1 */