1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/nfs/callback_proc.c
5 * Copyright (C) 2004 Trond Myklebust
7 * NFSv4 callback procedures
9 #include <linux/nfs4.h>
10 #include <linux/nfs_fs.h>
11 #include <linux/slab.h>
12 #include <linux/rcupdate.h>
15 #include "delegation.h"
18 #include "nfs4session.h"
19 #include "nfs4trace.h"
21 #define NFSDBG_FACILITY NFSDBG_CALLBACK
23 __be32
nfs4_callback_getattr(void *argp
, void *resp
,
24 struct cb_process_state
*cps
)
26 struct cb_getattrargs
*args
= argp
;
27 struct cb_getattrres
*res
= resp
;
28 struct nfs_delegation
*delegation
;
31 res
->status
= htonl(NFS4ERR_OP_NOT_IN_SESSION
);
32 if (!cps
->clp
) /* Always set for v4.0. Set in cb_sequence for v4.1 */
35 res
->bitmap
[0] = res
->bitmap
[1] = 0;
36 res
->status
= htonl(NFS4ERR_BADHANDLE
);
38 dprintk_rcu("NFS: GETATTR callback request from %s\n",
39 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
41 inode
= nfs_delegation_find_inode(cps
->clp
, &args
->fh
);
43 if (inode
== ERR_PTR(-EAGAIN
))
44 res
->status
= htonl(NFS4ERR_DELAY
);
45 trace_nfs4_cb_getattr(cps
->clp
, &args
->fh
, NULL
,
50 delegation
= nfs4_get_valid_delegation(inode
);
51 if (delegation
== NULL
|| (delegation
->type
& FMODE_WRITE
) == 0)
53 res
->size
= i_size_read(inode
);
54 res
->change_attr
= delegation
->change_attr
;
55 if (nfs_have_writebacks(inode
))
57 res
->ctime
= inode
->i_ctime
;
58 res
->mtime
= inode
->i_mtime
;
59 res
->bitmap
[0] = (FATTR4_WORD0_CHANGE
|FATTR4_WORD0_SIZE
) &
61 res
->bitmap
[1] = (FATTR4_WORD1_TIME_METADATA
|FATTR4_WORD1_TIME_MODIFY
) &
66 trace_nfs4_cb_getattr(cps
->clp
, &args
->fh
, inode
, -ntohl(res
->status
));
67 nfs_iput_and_deactive(inode
);
69 dprintk("%s: exit with status = %d\n", __func__
, ntohl(res
->status
));
73 __be32
nfs4_callback_recall(void *argp
, void *resp
,
74 struct cb_process_state
*cps
)
76 struct cb_recallargs
*args
= argp
;
80 res
= htonl(NFS4ERR_OP_NOT_IN_SESSION
);
81 if (!cps
->clp
) /* Always set for v4.0. Set in cb_sequence for v4.1 */
84 dprintk_rcu("NFS: RECALL callback request from %s\n",
85 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
87 res
= htonl(NFS4ERR_BADHANDLE
);
88 inode
= nfs_delegation_find_inode(cps
->clp
, &args
->fh
);
90 if (inode
== ERR_PTR(-EAGAIN
))
91 res
= htonl(NFS4ERR_DELAY
);
92 trace_nfs4_cb_recall(cps
->clp
, &args
->fh
, NULL
,
93 &args
->stateid
, -ntohl(res
));
96 /* Set up a helper thread to actually return the delegation */
97 switch (nfs_async_inode_return_delegation(inode
, &args
->stateid
)) {
102 res
= htonl(NFS4ERR_BAD_STATEID
);
105 res
= htonl(NFS4ERR_RESOURCE
);
107 trace_nfs4_cb_recall(cps
->clp
, &args
->fh
, inode
,
108 &args
->stateid
, -ntohl(res
));
109 nfs_iput_and_deactive(inode
);
111 dprintk("%s: exit with status = %d\n", __func__
, ntohl(res
));
115 #if defined(CONFIG_NFS_V4_1)
118 * Lookup a layout inode by stateid
120 * Note: returns a refcount on the inode and superblock
122 static struct inode
*nfs_layout_find_inode_by_stateid(struct nfs_client
*clp
,
123 const nfs4_stateid
*stateid
)
126 struct nfs_server
*server
;
128 struct pnfs_layout_hdr
*lo
;
131 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
132 list_for_each_entry_rcu(lo
, &server
->layouts
, plh_layouts
) {
133 if (!pnfs_layout_is_valid(lo
))
135 if (stateid
!= NULL
&&
136 !nfs4_stateid_match_other(stateid
, &lo
->plh_stateid
))
138 if (!nfs_sb_active(server
->super
))
140 inode
= igrab(lo
->plh_inode
);
144 nfs_sb_deactive(server
->super
);
145 return ERR_PTR(-EAGAIN
);
149 return ERR_PTR(-ENOENT
);
153 * Lookup a layout inode by filehandle.
155 * Note: returns a refcount on the inode and superblock
158 static struct inode
*nfs_layout_find_inode_by_fh(struct nfs_client
*clp
,
159 const struct nfs_fh
*fh
)
161 struct nfs_server
*server
;
162 struct nfs_inode
*nfsi
;
164 struct pnfs_layout_hdr
*lo
;
167 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
168 list_for_each_entry_rcu(lo
, &server
->layouts
, plh_layouts
) {
169 nfsi
= NFS_I(lo
->plh_inode
);
170 if (nfs_compare_fh(fh
, &nfsi
->fh
))
172 if (nfsi
->layout
!= lo
)
174 if (!nfs_sb_active(server
->super
))
176 inode
= igrab(lo
->plh_inode
);
180 nfs_sb_deactive(server
->super
);
181 return ERR_PTR(-EAGAIN
);
185 return ERR_PTR(-ENOENT
);
188 static struct inode
*nfs_layout_find_inode(struct nfs_client
*clp
,
189 const struct nfs_fh
*fh
,
190 const nfs4_stateid
*stateid
)
194 inode
= nfs_layout_find_inode_by_stateid(clp
, stateid
);
195 if (inode
== ERR_PTR(-ENOENT
))
196 inode
= nfs_layout_find_inode_by_fh(clp
, fh
);
201 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
203 static u32
pnfs_check_callback_stateid(struct pnfs_layout_hdr
*lo
,
204 const nfs4_stateid
*new)
208 /* Is the stateid not initialised? */
209 if (!pnfs_layout_is_valid(lo
))
210 return NFS4ERR_NOMATCHING_LAYOUT
;
212 /* Mismatched stateid? */
213 if (!nfs4_stateid_match_other(&lo
->plh_stateid
, new))
214 return NFS4ERR_BAD_STATEID
;
216 newseq
= be32_to_cpu(new->seqid
);
217 /* Are we already in a layout recall situation? */
218 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) &&
219 lo
->plh_return_seq
!= 0) {
220 if (newseq
< lo
->plh_return_seq
)
221 return NFS4ERR_OLD_STATEID
;
222 if (newseq
> lo
->plh_return_seq
)
223 return NFS4ERR_DELAY
;
227 /* Check that the stateid matches what we think it should be. */
228 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
229 if (newseq
> oldseq
+ 1)
230 return NFS4ERR_DELAY
;
232 if (newseq
<= oldseq
)
233 return NFS4ERR_OLD_STATEID
;
238 static u32
initiate_file_draining(struct nfs_client
*clp
,
239 struct cb_layoutrecallargs
*args
)
242 struct pnfs_layout_hdr
*lo
;
243 u32 rv
= NFS4ERR_NOMATCHING_LAYOUT
;
244 LIST_HEAD(free_me_list
);
246 ino
= nfs_layout_find_inode(clp
, &args
->cbl_fh
, &args
->cbl_stateid
);
248 if (ino
== ERR_PTR(-EAGAIN
))
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
);
268 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
270 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
275 pnfs_set_layout_stateid(lo
, &args
->cbl_stateid
, NULL
, true);
276 switch (pnfs_mark_matching_lsegs_return(lo
, &free_me_list
,
278 be32_to_cpu(args
->cbl_stateid
.seqid
))) {
281 /* There are layout segments that need to be returned */
285 /* Embrace your forgetfulness! */
286 rv
= NFS4ERR_NOMATCHING_LAYOUT
;
288 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
) {
289 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
,
294 spin_unlock(&ino
->i_lock
);
295 pnfs_free_lseg_list(&free_me_list
);
296 /* Free all lsegs that are attached to commit buckets */
297 nfs_commit_inode(ino
, 0);
298 pnfs_put_layout_hdr(lo
);
300 nfs_iput_and_deactive(ino
);
302 trace_nfs4_cb_layoutrecall_file(clp
, &args
->cbl_fh
, ino
,
303 &args
->cbl_stateid
, -rv
);
307 static u32
initiate_bulk_draining(struct nfs_client
*clp
,
308 struct cb_layoutrecallargs
*args
)
312 if (args
->cbl_recall_type
== RETURN_FSID
)
313 stat
= pnfs_destroy_layouts_byfsid(clp
, &args
->cbl_fsid
, true);
315 stat
= pnfs_destroy_layouts_byclid(clp
, true);
317 return NFS4ERR_DELAY
;
318 return NFS4ERR_NOMATCHING_LAYOUT
;
321 static u32
do_callback_layoutrecall(struct nfs_client
*clp
,
322 struct cb_layoutrecallargs
*args
)
324 if (args
->cbl_recall_type
== RETURN_FILE
)
325 return initiate_file_draining(clp
, args
);
326 return initiate_bulk_draining(clp
, args
);
329 __be32
nfs4_callback_layoutrecall(void *argp
, void *resp
,
330 struct cb_process_state
*cps
)
332 struct cb_layoutrecallargs
*args
= argp
;
333 u32 res
= NFS4ERR_OP_NOT_IN_SESSION
;
336 res
= do_callback_layoutrecall(cps
->clp
, args
);
337 return cpu_to_be32(res
);
340 static void pnfs_recall_all_layouts(struct nfs_client
*clp
)
342 struct cb_layoutrecallargs args
;
344 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
345 memset(&args
, 0, sizeof(args
));
346 args
.cbl_recall_type
= RETURN_ALL
;
347 /* FIXME we ignore errors, what should we do? */
348 do_callback_layoutrecall(clp
, &args
);
351 __be32
nfs4_callback_devicenotify(void *argp
, void *resp
,
352 struct cb_process_state
*cps
)
354 struct cb_devicenotifyargs
*args
= argp
;
357 struct nfs_client
*clp
= cps
->clp
;
358 struct nfs_server
*server
= NULL
;
361 res
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
365 for (i
= 0; i
< args
->ndevs
; i
++) {
366 struct cb_devicenotifyitem
*dev
= &args
->devs
[i
];
369 server
->pnfs_curr_ld
->id
!= dev
->cbd_layout_type
) {
371 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
)
372 if (server
->pnfs_curr_ld
&&
373 server
->pnfs_curr_ld
->id
== dev
->cbd_layout_type
) {
382 nfs4_delete_deviceid(server
->pnfs_curr_ld
, clp
, &dev
->cbd_dev_id
);
391 * Validate the sequenceID sent by the server.
392 * Return success if the sequenceID is one more than what we last saw on
393 * this slot, accounting for wraparound. Increments the slot's sequence.
395 * We don't yet implement a duplicate request cache, instead we set the
396 * back channel ca_maxresponsesize_cached to zero. This is OK for now
397 * since we only currently implement idempotent callbacks anyway.
399 * We have a single slot backchannel at this time, so we don't bother
400 * checking the used_slots bit array on the table. The lower layer guarantees
401 * a single outstanding callback request at a time.
404 validate_seqid(const struct nfs4_slot_table
*tbl
, const struct nfs4_slot
*slot
,
405 const struct cb_sequenceargs
* args
)
409 ret
= cpu_to_be32(NFS4ERR_BADSLOT
);
410 if (args
->csa_slotid
> tbl
->server_highest_slotid
)
414 if (args
->csa_sequenceid
== slot
->seq_nr
) {
415 ret
= cpu_to_be32(NFS4ERR_DELAY
);
416 if (nfs4_test_locked_slot(tbl
, slot
->slot_nr
))
419 /* Signal process_op to set this error on next op */
420 ret
= cpu_to_be32(NFS4ERR_RETRY_UNCACHED_REP
);
421 if (args
->csa_cachethis
== 0)
424 /* Liar! We never allowed you to set csa_cachethis != 0 */
425 ret
= cpu_to_be32(NFS4ERR_SEQ_FALSE_RETRY
);
429 /* Note: wraparound relies on seq_nr being of type u32 */
430 /* Misordered request */
431 ret
= cpu_to_be32(NFS4ERR_SEQ_MISORDERED
);
432 if (args
->csa_sequenceid
!= slot
->seq_nr
+ 1)
435 return cpu_to_be32(NFS4_OK
);
438 trace_nfs4_cb_seqid_err(args
, ret
);
443 * For each referring call triple, check the session's slot table for
444 * a match. If the slot is in use and the sequence numbers match, the
445 * client is still waiting for a response to the original request.
447 static int referring_call_exists(struct nfs_client
*clp
,
449 struct referring_call_list
*rclists
,
456 struct nfs4_session
*session
;
457 struct nfs4_slot_table
*tbl
;
458 struct referring_call_list
*rclist
;
459 struct referring_call
*ref
;
462 * XXX When client trunking is implemented, this becomes
463 * a session lookup from within the loop
465 session
= clp
->cl_session
;
466 tbl
= &session
->fc_slot_table
;
468 for (i
= 0; i
< nrclists
; i
++) {
469 rclist
= &rclists
[i
];
470 if (memcmp(session
->sess_id
.data
,
471 rclist
->rcl_sessionid
.data
,
472 NFS4_MAX_SESSIONID_LEN
) != 0)
475 for (j
= 0; j
< rclist
->rcl_nrefcalls
; j
++) {
476 ref
= &rclist
->rcl_refcalls
[j
];
478 status
= nfs4_slot_wait_on_seqid(tbl
, ref
->rc_slotid
,
479 ref
->rc_sequenceid
, HZ
>> 1) < 0;
490 __be32
nfs4_callback_sequence(void *argp
, void *resp
,
491 struct cb_process_state
*cps
)
493 struct cb_sequenceargs
*args
= argp
;
494 struct cb_sequenceres
*res
= resp
;
495 struct nfs4_slot_table
*tbl
;
496 struct nfs4_slot
*slot
;
497 struct nfs_client
*clp
;
499 __be32 status
= htonl(NFS4ERR_BADSESSION
);
501 clp
= nfs4_find_client_sessionid(cps
->net
, args
->csa_addr
,
502 &args
->csa_sessionid
, cps
->minorversion
);
506 if (!(clp
->cl_session
->flags
& SESSION4_BACK_CHAN
))
509 tbl
= &clp
->cl_session
->bc_slot_table
;
511 /* Set up res before grabbing the spinlock */
512 memcpy(&res
->csr_sessionid
, &args
->csa_sessionid
,
513 sizeof(res
->csr_sessionid
));
514 res
->csr_sequenceid
= args
->csa_sequenceid
;
515 res
->csr_slotid
= args
->csa_slotid
;
517 spin_lock(&tbl
->slot_tbl_lock
);
518 /* state manager is resetting the session */
519 if (test_bit(NFS4_SLOT_TBL_DRAINING
, &tbl
->slot_tbl_state
)) {
520 status
= htonl(NFS4ERR_DELAY
);
521 /* Return NFS4ERR_BADSESSION if we're draining the session
522 * in order to reset it.
524 if (test_bit(NFS4CLNT_SESSION_RESET
, &clp
->cl_state
))
525 status
= htonl(NFS4ERR_BADSESSION
);
529 status
= htonl(NFS4ERR_BADSLOT
);
530 slot
= nfs4_lookup_slot(tbl
, args
->csa_slotid
);
534 res
->csr_highestslotid
= tbl
->server_highest_slotid
;
535 res
->csr_target_highestslotid
= tbl
->target_highest_slotid
;
537 status
= validate_seqid(tbl
, slot
, args
);
540 if (!nfs4_try_to_lock_slot(tbl
, slot
)) {
541 status
= htonl(NFS4ERR_DELAY
);
546 /* The ca_maxresponsesize_cached is 0 with no DRC */
547 if (args
->csa_cachethis
!= 0) {
548 status
= htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE
);
553 * Check for pending referring calls. If a match is found, a
554 * related callback was received before the response to the original
557 if (referring_call_exists(clp
, args
->csa_nrclists
, args
->csa_rclists
,
558 &tbl
->slot_tbl_lock
) < 0) {
559 status
= htonl(NFS4ERR_DELAY
);
565 * If CB_SEQUENCE returns an error, then the state of the slot
566 * (sequence ID, cached reply) MUST NOT change.
568 slot
->seq_nr
= args
->csa_sequenceid
;
570 spin_unlock(&tbl
->slot_tbl_lock
);
573 cps
->clp
= clp
; /* put in nfs4_callback_compound */
574 for (i
= 0; i
< args
->csa_nrclists
; i
++)
575 kfree(args
->csa_rclists
[i
].rcl_refcalls
);
576 kfree(args
->csa_rclists
);
578 if (status
== htonl(NFS4ERR_RETRY_UNCACHED_REP
)) {
579 cps
->drc_status
= status
;
582 res
->csr_status
= status
;
584 trace_nfs4_cb_sequence(args
, res
, status
);
589 validate_bitmap_values(unsigned int mask
)
591 return (mask
& ~RCA4_TYPE_MASK_ALL
) == 0;
594 __be32
nfs4_callback_recallany(void *argp
, void *resp
,
595 struct cb_process_state
*cps
)
597 struct cb_recallanyargs
*args
= argp
;
600 bool schedule_manager
= false;
602 status
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
603 if (!cps
->clp
) /* set in cb_sequence */
606 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
607 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
609 status
= cpu_to_be32(NFS4ERR_INVAL
);
610 if (!validate_bitmap_values(args
->craa_type_mask
))
613 status
= cpu_to_be32(NFS4_OK
);
614 if (args
->craa_type_mask
& BIT(RCA4_TYPE_MASK_RDATA_DLG
))
616 if (args
->craa_type_mask
& BIT(RCA4_TYPE_MASK_WDATA_DLG
))
617 flags
|= FMODE_WRITE
;
619 nfs_expire_unused_delegation_types(cps
->clp
, flags
);
621 if (args
->craa_type_mask
& BIT(RCA4_TYPE_MASK_FILE_LAYOUT
))
622 pnfs_recall_all_layouts(cps
->clp
);
624 if (args
->craa_type_mask
& BIT(PNFS_FF_RCA4_TYPE_MASK_READ
)) {
625 set_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ
, &cps
->clp
->cl_state
);
626 schedule_manager
= true;
628 if (args
->craa_type_mask
& BIT(PNFS_FF_RCA4_TYPE_MASK_RW
)) {
629 set_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW
, &cps
->clp
->cl_state
);
630 schedule_manager
= true;
632 if (schedule_manager
)
633 nfs4_schedule_state_manager(cps
->clp
);
636 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
640 /* Reduce the fore channel's max_slots to the target value */
641 __be32
nfs4_callback_recallslot(void *argp
, void *resp
,
642 struct cb_process_state
*cps
)
644 struct cb_recallslotargs
*args
= argp
;
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(void *argp
, void *resp
,
668 struct cb_process_state
*cps
)
670 struct cb_notify_lock_args
*args
= argp
;
672 if (!cps
->clp
) /* set in cb_sequence */
673 return htonl(NFS4ERR_OP_NOT_IN_SESSION
);
675 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
676 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
678 /* Don't wake anybody if the string looked bogus */
679 if (args
->cbnl_valid
)
680 __wake_up(&cps
->clp
->cl_lock_waitq
, TASK_NORMAL
, 0, args
);
682 return htonl(NFS4_OK
);
684 #endif /* CONFIG_NFS_V4_1 */
685 #ifdef CONFIG_NFS_V4_2
686 static void nfs4_copy_cb_args(struct nfs4_copy_state
*cp_state
,
687 struct cb_offloadargs
*args
)
689 cp_state
->count
= args
->wr_count
;
690 cp_state
->error
= args
->error
;
692 cp_state
->verf
.committed
= args
->wr_writeverf
.committed
;
693 memcpy(&cp_state
->verf
.verifier
.data
[0],
694 &args
->wr_writeverf
.verifier
.data
[0],
699 __be32
nfs4_callback_offload(void *data
, void *dummy
,
700 struct cb_process_state
*cps
)
702 struct cb_offloadargs
*args
= data
;
703 struct nfs_server
*server
;
704 struct nfs4_copy_state
*copy
, *tmp_copy
;
707 copy
= kzalloc(sizeof(struct nfs4_copy_state
), GFP_NOFS
);
709 return htonl(NFS4ERR_SERVERFAULT
);
711 spin_lock(&cps
->clp
->cl_lock
);
713 list_for_each_entry_rcu(server
, &cps
->clp
->cl_superblocks
,
715 list_for_each_entry(tmp_copy
, &server
->ss_copies
, copies
) {
716 if (memcmp(args
->coa_stateid
.other
,
717 tmp_copy
->stateid
.other
,
718 sizeof(args
->coa_stateid
.other
)))
720 nfs4_copy_cb_args(tmp_copy
, args
);
721 complete(&tmp_copy
->completion
);
729 memcpy(©
->stateid
, &args
->coa_stateid
, NFS4_STATEID_SIZE
);
730 nfs4_copy_cb_args(copy
, args
);
731 list_add_tail(©
->copies
, &cps
->clp
->pending_cb_stateids
);
734 spin_unlock(&cps
->clp
->cl_lock
);
738 #endif /* CONFIG_NFS_V4_2 */