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 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
,
124 struct nfs_server
*server
;
125 struct nfs_inode
*nfsi
;
127 struct pnfs_layout_hdr
*lo
;
130 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
131 list_for_each_entry(lo
, &server
->layouts
, plh_layouts
) {
132 nfsi
= NFS_I(lo
->plh_inode
);
133 if (nfs_compare_fh(fh
, &nfsi
->fh
))
135 if (nfsi
->layout
!= lo
)
137 ino
= igrab(lo
->plh_inode
);
140 spin_lock(&ino
->i_lock
);
141 /* Is this layout in the process of being freed? */
142 if (nfsi
->layout
!= lo
) {
143 spin_unlock(&ino
->i_lock
);
147 pnfs_get_layout_hdr(lo
);
148 spin_unlock(&ino
->i_lock
);
156 static struct pnfs_layout_hdr
* get_layout_by_fh(struct nfs_client
*clp
,
159 struct pnfs_layout_hdr
*lo
;
161 spin_lock(&clp
->cl_lock
);
163 lo
= get_layout_by_fh_locked(clp
, fh
);
165 spin_unlock(&clp
->cl_lock
);
171 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
173 static u32
pnfs_check_callback_stateid(struct pnfs_layout_hdr
*lo
,
174 const nfs4_stateid
*new)
178 /* Is the stateid still not initialised? */
179 if (!pnfs_layout_is_valid(lo
))
180 return NFS4ERR_DELAY
;
182 /* Mismatched stateid? */
183 if (!nfs4_stateid_match_other(&lo
->plh_stateid
, new))
184 return NFS4ERR_BAD_STATEID
;
186 newseq
= be32_to_cpu(new->seqid
);
187 /* Are we already in a layout recall situation? */
188 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) &&
189 lo
->plh_return_seq
!= 0) {
190 if (newseq
< lo
->plh_return_seq
)
191 return NFS4ERR_OLD_STATEID
;
192 if (newseq
> lo
->plh_return_seq
)
193 return NFS4ERR_DELAY
;
197 /* Check that the stateid matches what we think it should be. */
198 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
199 if (newseq
> oldseq
+ 1)
200 return NFS4ERR_DELAY
;
202 if (newseq
<= oldseq
)
203 return NFS4ERR_OLD_STATEID
;
208 static u32
initiate_file_draining(struct nfs_client
*clp
,
209 struct cb_layoutrecallargs
*args
)
212 struct pnfs_layout_hdr
*lo
;
213 u32 rv
= NFS4ERR_NOMATCHING_LAYOUT
;
214 LIST_HEAD(free_me_list
);
216 lo
= get_layout_by_fh(clp
, &args
->cbl_fh
);
218 trace_nfs4_cb_layoutrecall_file(clp
, &args
->cbl_fh
, NULL
,
219 &args
->cbl_stateid
, -rv
);
224 pnfs_layoutcommit_inode(ino
, false);
227 spin_lock(&ino
->i_lock
);
228 rv
= pnfs_check_callback_stateid(lo
, &args
->cbl_stateid
);
231 pnfs_set_layout_stateid(lo
, &args
->cbl_stateid
, true);
234 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
236 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
241 if (pnfs_mark_matching_lsegs_return(lo
, &free_me_list
,
243 be32_to_cpu(args
->cbl_stateid
.seqid
))) {
248 /* Embrace your forgetfulness! */
249 rv
= NFS4ERR_NOMATCHING_LAYOUT
;
251 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
) {
252 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
,
256 spin_unlock(&ino
->i_lock
);
257 pnfs_free_lseg_list(&free_me_list
);
258 /* Free all lsegs that are attached to commit buckets */
259 nfs_commit_inode(ino
, 0);
260 pnfs_put_layout_hdr(lo
);
261 trace_nfs4_cb_layoutrecall_file(clp
, &args
->cbl_fh
, ino
,
262 &args
->cbl_stateid
, -rv
);
268 static u32
initiate_bulk_draining(struct nfs_client
*clp
,
269 struct cb_layoutrecallargs
*args
)
273 if (args
->cbl_recall_type
== RETURN_FSID
)
274 stat
= pnfs_destroy_layouts_byfsid(clp
, &args
->cbl_fsid
, true);
276 stat
= pnfs_destroy_layouts_byclid(clp
, true);
278 return NFS4ERR_DELAY
;
279 return NFS4ERR_NOMATCHING_LAYOUT
;
282 static u32
do_callback_layoutrecall(struct nfs_client
*clp
,
283 struct cb_layoutrecallargs
*args
)
287 dprintk("%s enter, type=%i\n", __func__
, args
->cbl_recall_type
);
288 if (args
->cbl_recall_type
== RETURN_FILE
)
289 res
= initiate_file_draining(clp
, args
);
291 res
= initiate_bulk_draining(clp
, args
);
292 dprintk("%s returning %i\n", __func__
, res
);
297 __be32
nfs4_callback_layoutrecall(struct cb_layoutrecallargs
*args
,
298 void *dummy
, struct cb_process_state
*cps
)
302 dprintk("%s: -->\n", __func__
);
305 res
= do_callback_layoutrecall(cps
->clp
, args
);
307 res
= NFS4ERR_OP_NOT_IN_SESSION
;
309 dprintk("%s: exit with status = %d\n", __func__
, res
);
310 return cpu_to_be32(res
);
313 static void pnfs_recall_all_layouts(struct nfs_client
*clp
)
315 struct cb_layoutrecallargs args
;
317 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
318 memset(&args
, 0, sizeof(args
));
319 args
.cbl_recall_type
= RETURN_ALL
;
320 /* FIXME we ignore errors, what should we do? */
321 do_callback_layoutrecall(clp
, &args
);
324 __be32
nfs4_callback_devicenotify(struct cb_devicenotifyargs
*args
,
325 void *dummy
, struct cb_process_state
*cps
)
329 struct nfs_client
*clp
= cps
->clp
;
330 struct nfs_server
*server
= NULL
;
332 dprintk("%s: -->\n", __func__
);
335 res
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
339 for (i
= 0; i
< args
->ndevs
; i
++) {
340 struct cb_devicenotifyitem
*dev
= &args
->devs
[i
];
343 server
->pnfs_curr_ld
->id
!= dev
->cbd_layout_type
) {
345 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
)
346 if (server
->pnfs_curr_ld
&&
347 server
->pnfs_curr_ld
->id
== dev
->cbd_layout_type
) {
352 dprintk("%s: layout type %u not found\n",
353 __func__
, dev
->cbd_layout_type
);
358 nfs4_delete_deviceid(server
->pnfs_curr_ld
, clp
, &dev
->cbd_dev_id
);
363 dprintk("%s: exit with status = %u\n",
364 __func__
, be32_to_cpu(res
));
369 * Validate the sequenceID sent by the server.
370 * Return success if the sequenceID is one more than what we last saw on
371 * this slot, accounting for wraparound. Increments the slot's sequence.
373 * We don't yet implement a duplicate request cache, instead we set the
374 * back channel ca_maxresponsesize_cached to zero. This is OK for now
375 * since we only currently implement idempotent callbacks anyway.
377 * We have a single slot backchannel at this time, so we don't bother
378 * checking the used_slots bit array on the table. The lower layer guarantees
379 * a single outstanding callback request at a time.
382 validate_seqid(const struct nfs4_slot_table
*tbl
, const struct nfs4_slot
*slot
,
383 const struct cb_sequenceargs
* args
)
385 dprintk("%s enter. slotid %u seqid %u, slot table seqid: %u\n",
386 __func__
, args
->csa_slotid
, args
->csa_sequenceid
, slot
->seq_nr
);
388 if (args
->csa_slotid
> tbl
->server_highest_slotid
)
389 return htonl(NFS4ERR_BADSLOT
);
392 if (args
->csa_sequenceid
== slot
->seq_nr
) {
393 dprintk("%s seqid %u is a replay\n",
394 __func__
, args
->csa_sequenceid
);
395 if (nfs4_test_locked_slot(tbl
, slot
->slot_nr
))
396 return htonl(NFS4ERR_DELAY
);
397 /* Signal process_op to set this error on next op */
398 if (args
->csa_cachethis
== 0)
399 return htonl(NFS4ERR_RETRY_UNCACHED_REP
);
401 /* Liar! We never allowed you to set csa_cachethis != 0 */
402 return htonl(NFS4ERR_SEQ_FALSE_RETRY
);
406 if (unlikely(slot
->seq_nr
== 0xFFFFFFFFU
)) {
407 if (args
->csa_sequenceid
== 1)
408 return htonl(NFS4_OK
);
409 } else if (likely(args
->csa_sequenceid
== slot
->seq_nr
+ 1))
410 return htonl(NFS4_OK
);
412 /* Misordered request */
413 return htonl(NFS4ERR_SEQ_MISORDERED
);
417 * For each referring call triple, check the session's slot table for
418 * a match. If the slot is in use and the sequence numbers match, the
419 * client is still waiting for a response to the original request.
421 static bool referring_call_exists(struct nfs_client
*clp
,
423 struct referring_call_list
*rclists
)
427 struct nfs4_session
*session
;
428 struct nfs4_slot_table
*tbl
;
429 struct referring_call_list
*rclist
;
430 struct referring_call
*ref
;
433 * XXX When client trunking is implemented, this becomes
434 * a session lookup from within the loop
436 session
= clp
->cl_session
;
437 tbl
= &session
->fc_slot_table
;
439 for (i
= 0; i
< nrclists
; i
++) {
440 rclist
= &rclists
[i
];
441 if (memcmp(session
->sess_id
.data
,
442 rclist
->rcl_sessionid
.data
,
443 NFS4_MAX_SESSIONID_LEN
) != 0)
446 for (j
= 0; j
< rclist
->rcl_nrefcalls
; j
++) {
447 ref
= &rclist
->rcl_refcalls
[j
];
449 dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
450 "slotid %u\n", __func__
,
451 ((u32
*)&rclist
->rcl_sessionid
.data
)[0],
452 ((u32
*)&rclist
->rcl_sessionid
.data
)[1],
453 ((u32
*)&rclist
->rcl_sessionid
.data
)[2],
454 ((u32
*)&rclist
->rcl_sessionid
.data
)[3],
455 ref
->rc_sequenceid
, ref
->rc_slotid
);
457 status
= nfs4_slot_wait_on_seqid(tbl
, ref
->rc_slotid
,
458 ref
->rc_sequenceid
, HZ
>> 1) < 0;
468 __be32
nfs4_callback_sequence(struct cb_sequenceargs
*args
,
469 struct cb_sequenceres
*res
,
470 struct cb_process_state
*cps
)
472 struct nfs4_slot_table
*tbl
;
473 struct nfs4_slot
*slot
;
474 struct nfs_client
*clp
;
476 __be32 status
= htonl(NFS4ERR_BADSESSION
);
478 clp
= nfs4_find_client_sessionid(cps
->net
, args
->csa_addr
,
479 &args
->csa_sessionid
, cps
->minorversion
);
483 if (!(clp
->cl_session
->flags
& SESSION4_BACK_CHAN
))
486 tbl
= &clp
->cl_session
->bc_slot_table
;
488 /* Set up res before grabbing the spinlock */
489 memcpy(&res
->csr_sessionid
, &args
->csa_sessionid
,
490 sizeof(res
->csr_sessionid
));
491 res
->csr_sequenceid
= args
->csa_sequenceid
;
492 res
->csr_slotid
= args
->csa_slotid
;
494 spin_lock(&tbl
->slot_tbl_lock
);
495 /* state manager is resetting the session */
496 if (test_bit(NFS4_SLOT_TBL_DRAINING
, &tbl
->slot_tbl_state
)) {
497 status
= htonl(NFS4ERR_DELAY
);
498 /* Return NFS4ERR_BADSESSION if we're draining the session
499 * in order to reset it.
501 if (test_bit(NFS4CLNT_SESSION_RESET
, &clp
->cl_state
))
502 status
= htonl(NFS4ERR_BADSESSION
);
506 status
= htonl(NFS4ERR_BADSLOT
);
507 slot
= nfs4_lookup_slot(tbl
, args
->csa_slotid
);
511 res
->csr_highestslotid
= tbl
->server_highest_slotid
;
512 res
->csr_target_highestslotid
= tbl
->target_highest_slotid
;
514 status
= validate_seqid(tbl
, slot
, args
);
517 if (!nfs4_try_to_lock_slot(tbl
, slot
)) {
518 status
= htonl(NFS4ERR_DELAY
);
523 /* The ca_maxresponsesize_cached is 0 with no DRC */
524 if (args
->csa_cachethis
!= 0) {
525 status
= htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE
);
530 * Check for pending referring calls. If a match is found, a
531 * related callback was received before the response to the original
534 if (referring_call_exists(clp
, args
->csa_nrclists
, args
->csa_rclists
)) {
535 status
= htonl(NFS4ERR_DELAY
);
541 * If CB_SEQUENCE returns an error, then the state of the slot
542 * (sequence ID, cached reply) MUST NOT change.
544 slot
->seq_nr
= args
->csa_sequenceid
;
546 spin_unlock(&tbl
->slot_tbl_lock
);
549 cps
->clp
= clp
; /* put in nfs4_callback_compound */
550 for (i
= 0; i
< args
->csa_nrclists
; i
++)
551 kfree(args
->csa_rclists
[i
].rcl_refcalls
);
552 kfree(args
->csa_rclists
);
554 if (status
== htonl(NFS4ERR_RETRY_UNCACHED_REP
)) {
555 cps
->drc_status
= status
;
558 res
->csr_status
= status
;
560 trace_nfs4_cb_sequence(args
, res
, status
);
561 dprintk("%s: exit with status = %d res->csr_status %d\n", __func__
,
562 ntohl(status
), ntohl(res
->csr_status
));
567 validate_bitmap_values(unsigned long mask
)
569 return (mask
& ~RCA4_TYPE_MASK_ALL
) == 0;
572 __be32
nfs4_callback_recallany(struct cb_recallanyargs
*args
, void *dummy
,
573 struct cb_process_state
*cps
)
578 status
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
579 if (!cps
->clp
) /* set in cb_sequence */
582 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
583 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
585 status
= cpu_to_be32(NFS4ERR_INVAL
);
586 if (!validate_bitmap_values(args
->craa_type_mask
))
589 status
= cpu_to_be32(NFS4_OK
);
590 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG
, (const unsigned long *)
591 &args
->craa_type_mask
))
593 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG
, (const unsigned long *)
594 &args
->craa_type_mask
))
595 flags
|= FMODE_WRITE
;
596 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT
, (const unsigned long *)
597 &args
->craa_type_mask
))
598 pnfs_recall_all_layouts(cps
->clp
);
600 nfs_expire_unused_delegation_types(cps
->clp
, flags
);
602 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
606 /* Reduce the fore channel's max_slots to the target value */
607 __be32
nfs4_callback_recallslot(struct cb_recallslotargs
*args
, void *dummy
,
608 struct cb_process_state
*cps
)
610 struct nfs4_slot_table
*fc_tbl
;
613 status
= htonl(NFS4ERR_OP_NOT_IN_SESSION
);
614 if (!cps
->clp
) /* set in cb_sequence */
617 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
618 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
),
619 args
->crsa_target_highest_slotid
);
621 fc_tbl
= &cps
->clp
->cl_session
->fc_slot_table
;
623 status
= htonl(NFS4_OK
);
625 nfs41_set_target_slotid(fc_tbl
, args
->crsa_target_highest_slotid
);
626 nfs41_notify_server(cps
->clp
);
628 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
632 __be32
nfs4_callback_notify_lock(struct cb_notify_lock_args
*args
, void *dummy
,
633 struct cb_process_state
*cps
)
635 if (!cps
->clp
) /* set in cb_sequence */
636 return htonl(NFS4ERR_OP_NOT_IN_SESSION
);
638 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
639 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
641 /* Don't wake anybody if the string looked bogus */
642 if (args
->cbnl_valid
)
643 __wake_up(&cps
->clp
->cl_lock_waitq
, TASK_NORMAL
, 0, args
);
645 return htonl(NFS4_OK
);
647 #endif /* CONFIG_NFS_V4_1 */