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
;
29 struct nfs_inode
*nfsi
;
32 res
->status
= htonl(NFS4ERR_OP_NOT_IN_SESSION
);
33 if (!cps
->clp
) /* Always set for v4.0. Set in cb_sequence for v4.1 */
36 res
->bitmap
[0] = res
->bitmap
[1] = 0;
37 res
->status
= htonl(NFS4ERR_BADHANDLE
);
39 dprintk_rcu("NFS: GETATTR callback request from %s\n",
40 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
42 inode
= nfs_delegation_find_inode(cps
->clp
, &args
->fh
);
44 trace_nfs4_cb_getattr(cps
->clp
, &args
->fh
, NULL
,
50 delegation
= rcu_dereference(nfsi
->delegation
);
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
));
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 trace_nfs4_cb_recall(cps
->clp
, &args
->fh
, NULL
,
91 &args
->stateid
, -ntohl(res
));
94 /* Set up a helper thread to actually return the delegation */
95 switch (nfs_async_inode_return_delegation(inode
, &args
->stateid
)) {
100 res
= htonl(NFS4ERR_BAD_STATEID
);
103 res
= htonl(NFS4ERR_RESOURCE
);
105 trace_nfs4_cb_recall(cps
->clp
, &args
->fh
, inode
,
106 &args
->stateid
, -ntohl(res
));
109 dprintk("%s: exit with status = %d\n", __func__
, ntohl(res
));
113 #if defined(CONFIG_NFS_V4_1)
116 * Lookup a layout inode by stateid
118 * Note: returns a refcount on the inode and superblock
120 static struct inode
*nfs_layout_find_inode_by_stateid(struct nfs_client
*clp
,
121 const nfs4_stateid
*stateid
)
123 struct nfs_server
*server
;
125 struct pnfs_layout_hdr
*lo
;
128 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
129 list_for_each_entry(lo
, &server
->layouts
, plh_layouts
) {
130 if (stateid
!= NULL
&&
131 !nfs4_stateid_match_other(stateid
, &lo
->plh_stateid
))
133 inode
= igrab(lo
->plh_inode
);
136 if (!nfs_sb_active(inode
->i_sb
)) {
138 spin_unlock(&clp
->cl_lock
);
140 spin_lock(&clp
->cl_lock
);
152 * Lookup a layout inode by filehandle.
154 * Note: returns a refcount on the inode and superblock
157 static struct inode
*nfs_layout_find_inode_by_fh(struct nfs_client
*clp
,
158 const struct nfs_fh
*fh
)
160 struct nfs_server
*server
;
161 struct nfs_inode
*nfsi
;
163 struct pnfs_layout_hdr
*lo
;
166 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
167 list_for_each_entry(lo
, &server
->layouts
, plh_layouts
) {
168 nfsi
= NFS_I(lo
->plh_inode
);
169 if (nfs_compare_fh(fh
, &nfsi
->fh
))
171 if (nfsi
->layout
!= lo
)
173 inode
= igrab(lo
->plh_inode
);
176 if (!nfs_sb_active(inode
->i_sb
)) {
178 spin_unlock(&clp
->cl_lock
);
180 spin_lock(&clp
->cl_lock
);
191 static struct inode
*nfs_layout_find_inode(struct nfs_client
*clp
,
192 const struct nfs_fh
*fh
,
193 const nfs4_stateid
*stateid
)
197 spin_lock(&clp
->cl_lock
);
199 inode
= nfs_layout_find_inode_by_stateid(clp
, stateid
);
201 inode
= nfs_layout_find_inode_by_fh(clp
, fh
);
203 spin_unlock(&clp
->cl_lock
);
209 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
211 static u32
pnfs_check_callback_stateid(struct pnfs_layout_hdr
*lo
,
212 const nfs4_stateid
*new)
216 /* Is the stateid still not initialised? */
217 if (!pnfs_layout_is_valid(lo
))
218 return NFS4ERR_DELAY
;
220 /* Mismatched stateid? */
221 if (!nfs4_stateid_match_other(&lo
->plh_stateid
, new))
222 return NFS4ERR_BAD_STATEID
;
224 newseq
= be32_to_cpu(new->seqid
);
225 /* Are we already in a layout recall situation? */
226 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) &&
227 lo
->plh_return_seq
!= 0) {
228 if (newseq
< lo
->plh_return_seq
)
229 return NFS4ERR_OLD_STATEID
;
230 if (newseq
> lo
->plh_return_seq
)
231 return NFS4ERR_DELAY
;
235 /* Check that the stateid matches what we think it should be. */
236 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
237 if (newseq
> oldseq
+ 1)
238 return NFS4ERR_DELAY
;
240 if (newseq
<= oldseq
)
241 return NFS4ERR_OLD_STATEID
;
246 static u32
initiate_file_draining(struct nfs_client
*clp
,
247 struct cb_layoutrecallargs
*args
)
250 struct pnfs_layout_hdr
*lo
;
251 u32 rv
= NFS4ERR_NOMATCHING_LAYOUT
;
252 LIST_HEAD(free_me_list
);
254 ino
= nfs_layout_find_inode(clp
, &args
->cbl_fh
, &args
->cbl_stateid
);
258 pnfs_layoutcommit_inode(ino
, false);
261 spin_lock(&ino
->i_lock
);
262 lo
= NFS_I(ino
)->layout
;
264 spin_unlock(&ino
->i_lock
);
267 pnfs_get_layout_hdr(lo
);
268 rv
= pnfs_check_callback_stateid(lo
, &args
->cbl_stateid
);
271 pnfs_set_layout_stateid(lo
, &args
->cbl_stateid
, true);
274 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
276 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
281 if (pnfs_mark_matching_lsegs_return(lo
, &free_me_list
,
283 be32_to_cpu(args
->cbl_stateid
.seqid
))) {
288 /* Embrace your forgetfulness! */
289 rv
= NFS4ERR_NOMATCHING_LAYOUT
;
291 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
) {
292 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
,
296 spin_unlock(&ino
->i_lock
);
297 pnfs_free_lseg_list(&free_me_list
);
298 /* Free all lsegs that are attached to commit buckets */
299 nfs_commit_inode(ino
, 0);
300 pnfs_put_layout_hdr(lo
);
302 trace_nfs4_cb_layoutrecall_file(clp
, &args
->cbl_fh
, ino
,
303 &args
->cbl_stateid
, -rv
);
304 nfs_iput_and_deactive(ino
);
308 static u32
initiate_bulk_draining(struct nfs_client
*clp
,
309 struct cb_layoutrecallargs
*args
)
313 if (args
->cbl_recall_type
== RETURN_FSID
)
314 stat
= pnfs_destroy_layouts_byfsid(clp
, &args
->cbl_fsid
, true);
316 stat
= pnfs_destroy_layouts_byclid(clp
, true);
318 return NFS4ERR_DELAY
;
319 return NFS4ERR_NOMATCHING_LAYOUT
;
322 static u32
do_callback_layoutrecall(struct nfs_client
*clp
,
323 struct cb_layoutrecallargs
*args
)
325 if (args
->cbl_recall_type
== RETURN_FILE
)
326 return initiate_file_draining(clp
, args
);
327 return initiate_bulk_draining(clp
, args
);
330 __be32
nfs4_callback_layoutrecall(void *argp
, void *resp
,
331 struct cb_process_state
*cps
)
333 struct cb_layoutrecallargs
*args
= argp
;
334 u32 res
= NFS4ERR_OP_NOT_IN_SESSION
;
337 res
= do_callback_layoutrecall(cps
->clp
, args
);
338 return cpu_to_be32(res
);
341 static void pnfs_recall_all_layouts(struct nfs_client
*clp
)
343 struct cb_layoutrecallargs args
;
345 /* Pretend we got a CB_LAYOUTRECALL(ALL) */
346 memset(&args
, 0, sizeof(args
));
347 args
.cbl_recall_type
= RETURN_ALL
;
348 /* FIXME we ignore errors, what should we do? */
349 do_callback_layoutrecall(clp
, &args
);
352 __be32
nfs4_callback_devicenotify(void *argp
, void *resp
,
353 struct cb_process_state
*cps
)
355 struct cb_devicenotifyargs
*args
= argp
;
358 struct nfs_client
*clp
= cps
->clp
;
359 struct nfs_server
*server
= NULL
;
362 res
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
366 for (i
= 0; i
< args
->ndevs
; i
++) {
367 struct cb_devicenotifyitem
*dev
= &args
->devs
[i
];
370 server
->pnfs_curr_ld
->id
!= dev
->cbd_layout_type
) {
372 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
)
373 if (server
->pnfs_curr_ld
&&
374 server
->pnfs_curr_ld
->id
== dev
->cbd_layout_type
) {
383 nfs4_delete_deviceid(server
->pnfs_curr_ld
, clp
, &dev
->cbd_dev_id
);
392 * Validate the sequenceID sent by the server.
393 * Return success if the sequenceID is one more than what we last saw on
394 * this slot, accounting for wraparound. Increments the slot's sequence.
396 * We don't yet implement a duplicate request cache, instead we set the
397 * back channel ca_maxresponsesize_cached to zero. This is OK for now
398 * since we only currently implement idempotent callbacks anyway.
400 * We have a single slot backchannel at this time, so we don't bother
401 * checking the used_slots bit array on the table. The lower layer guarantees
402 * a single outstanding callback request at a time.
405 validate_seqid(const struct nfs4_slot_table
*tbl
, const struct nfs4_slot
*slot
,
406 const struct cb_sequenceargs
* args
)
408 if (args
->csa_slotid
> tbl
->server_highest_slotid
)
409 return htonl(NFS4ERR_BADSLOT
);
412 if (args
->csa_sequenceid
== slot
->seq_nr
) {
413 if (nfs4_test_locked_slot(tbl
, slot
->slot_nr
))
414 return htonl(NFS4ERR_DELAY
);
415 /* Signal process_op to set this error on next op */
416 if (args
->csa_cachethis
== 0)
417 return htonl(NFS4ERR_RETRY_UNCACHED_REP
);
419 /* Liar! We never allowed you to set csa_cachethis != 0 */
420 return htonl(NFS4ERR_SEQ_FALSE_RETRY
);
424 if (unlikely(slot
->seq_nr
== 0xFFFFFFFFU
)) {
425 if (args
->csa_sequenceid
== 1)
426 return htonl(NFS4_OK
);
427 } else if (likely(args
->csa_sequenceid
== slot
->seq_nr
+ 1))
428 return htonl(NFS4_OK
);
430 /* Misordered request */
431 return htonl(NFS4ERR_SEQ_MISORDERED
);
435 * For each referring call triple, check the session's slot table for
436 * a match. If the slot is in use and the sequence numbers match, the
437 * client is still waiting for a response to the original request.
439 static bool referring_call_exists(struct nfs_client
*clp
,
441 struct referring_call_list
*rclists
)
445 struct nfs4_session
*session
;
446 struct nfs4_slot_table
*tbl
;
447 struct referring_call_list
*rclist
;
448 struct referring_call
*ref
;
451 * XXX When client trunking is implemented, this becomes
452 * a session lookup from within the loop
454 session
= clp
->cl_session
;
455 tbl
= &session
->fc_slot_table
;
457 for (i
= 0; i
< nrclists
; i
++) {
458 rclist
= &rclists
[i
];
459 if (memcmp(session
->sess_id
.data
,
460 rclist
->rcl_sessionid
.data
,
461 NFS4_MAX_SESSIONID_LEN
) != 0)
464 for (j
= 0; j
< rclist
->rcl_nrefcalls
; j
++) {
465 ref
= &rclist
->rcl_refcalls
[j
];
466 status
= nfs4_slot_wait_on_seqid(tbl
, ref
->rc_slotid
,
467 ref
->rc_sequenceid
, HZ
>> 1) < 0;
477 __be32
nfs4_callback_sequence(void *argp
, void *resp
,
478 struct cb_process_state
*cps
)
480 struct cb_sequenceargs
*args
= argp
;
481 struct cb_sequenceres
*res
= resp
;
482 struct nfs4_slot_table
*tbl
;
483 struct nfs4_slot
*slot
;
484 struct nfs_client
*clp
;
486 __be32 status
= htonl(NFS4ERR_BADSESSION
);
488 clp
= nfs4_find_client_sessionid(cps
->net
, args
->csa_addr
,
489 &args
->csa_sessionid
, cps
->minorversion
);
493 if (!(clp
->cl_session
->flags
& SESSION4_BACK_CHAN
))
496 tbl
= &clp
->cl_session
->bc_slot_table
;
498 /* Set up res before grabbing the spinlock */
499 memcpy(&res
->csr_sessionid
, &args
->csa_sessionid
,
500 sizeof(res
->csr_sessionid
));
501 res
->csr_sequenceid
= args
->csa_sequenceid
;
502 res
->csr_slotid
= args
->csa_slotid
;
504 spin_lock(&tbl
->slot_tbl_lock
);
505 /* state manager is resetting the session */
506 if (test_bit(NFS4_SLOT_TBL_DRAINING
, &tbl
->slot_tbl_state
)) {
507 status
= htonl(NFS4ERR_DELAY
);
508 /* Return NFS4ERR_BADSESSION if we're draining the session
509 * in order to reset it.
511 if (test_bit(NFS4CLNT_SESSION_RESET
, &clp
->cl_state
))
512 status
= htonl(NFS4ERR_BADSESSION
);
516 status
= htonl(NFS4ERR_BADSLOT
);
517 slot
= nfs4_lookup_slot(tbl
, args
->csa_slotid
);
521 res
->csr_highestslotid
= tbl
->server_highest_slotid
;
522 res
->csr_target_highestslotid
= tbl
->target_highest_slotid
;
524 status
= validate_seqid(tbl
, slot
, args
);
527 if (!nfs4_try_to_lock_slot(tbl
, slot
)) {
528 status
= htonl(NFS4ERR_DELAY
);
533 /* The ca_maxresponsesize_cached is 0 with no DRC */
534 if (args
->csa_cachethis
!= 0) {
535 status
= htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE
);
540 * Check for pending referring calls. If a match is found, a
541 * related callback was received before the response to the original
544 if (referring_call_exists(clp
, args
->csa_nrclists
, args
->csa_rclists
)) {
545 status
= htonl(NFS4ERR_DELAY
);
551 * If CB_SEQUENCE returns an error, then the state of the slot
552 * (sequence ID, cached reply) MUST NOT change.
554 slot
->seq_nr
= args
->csa_sequenceid
;
556 spin_unlock(&tbl
->slot_tbl_lock
);
559 cps
->clp
= clp
; /* put in nfs4_callback_compound */
560 for (i
= 0; i
< args
->csa_nrclists
; i
++)
561 kfree(args
->csa_rclists
[i
].rcl_refcalls
);
562 kfree(args
->csa_rclists
);
564 if (status
== htonl(NFS4ERR_RETRY_UNCACHED_REP
)) {
565 cps
->drc_status
= status
;
568 res
->csr_status
= status
;
570 trace_nfs4_cb_sequence(args
, res
, status
);
575 validate_bitmap_values(unsigned long mask
)
577 return (mask
& ~RCA4_TYPE_MASK_ALL
) == 0;
580 __be32
nfs4_callback_recallany(void *argp
, void *resp
,
581 struct cb_process_state
*cps
)
583 struct cb_recallanyargs
*args
= argp
;
587 status
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
588 if (!cps
->clp
) /* set in cb_sequence */
591 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
592 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
594 status
= cpu_to_be32(NFS4ERR_INVAL
);
595 if (!validate_bitmap_values(args
->craa_type_mask
))
598 status
= cpu_to_be32(NFS4_OK
);
599 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG
, (const unsigned long *)
600 &args
->craa_type_mask
))
602 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG
, (const unsigned long *)
603 &args
->craa_type_mask
))
604 flags
|= FMODE_WRITE
;
605 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT
, (const unsigned long *)
606 &args
->craa_type_mask
))
607 pnfs_recall_all_layouts(cps
->clp
);
609 nfs_expire_unused_delegation_types(cps
->clp
, flags
);
611 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
615 /* Reduce the fore channel's max_slots to the target value */
616 __be32
nfs4_callback_recallslot(void *argp
, void *resp
,
617 struct cb_process_state
*cps
)
619 struct cb_recallslotargs
*args
= argp
;
620 struct nfs4_slot_table
*fc_tbl
;
623 status
= htonl(NFS4ERR_OP_NOT_IN_SESSION
);
624 if (!cps
->clp
) /* set in cb_sequence */
627 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
628 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
),
629 args
->crsa_target_highest_slotid
);
631 fc_tbl
= &cps
->clp
->cl_session
->fc_slot_table
;
633 status
= htonl(NFS4_OK
);
635 nfs41_set_target_slotid(fc_tbl
, args
->crsa_target_highest_slotid
);
636 nfs41_notify_server(cps
->clp
);
638 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
642 __be32
nfs4_callback_notify_lock(void *argp
, void *resp
,
643 struct cb_process_state
*cps
)
645 struct cb_notify_lock_args
*args
= argp
;
647 if (!cps
->clp
) /* set in cb_sequence */
648 return htonl(NFS4ERR_OP_NOT_IN_SESSION
);
650 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
651 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
653 /* Don't wake anybody if the string looked bogus */
654 if (args
->cbnl_valid
)
655 __wake_up(&cps
->clp
->cl_lock_waitq
, TASK_NORMAL
, 0, args
);
657 return htonl(NFS4_OK
);
659 #endif /* CONFIG_NFS_V4_1 */