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(void *argp
, void *resp
,
23 struct cb_process_state
*cps
)
25 struct cb_getattrargs
*args
= argp
;
26 struct cb_getattrres
*res
= resp
;
27 struct nfs_delegation
*delegation
;
28 struct nfs_inode
*nfsi
;
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 trace_nfs4_cb_getattr(cps
->clp
, &args
->fh
, NULL
,
49 delegation
= rcu_dereference(nfsi
->delegation
);
50 if (delegation
== NULL
|| (delegation
->type
& FMODE_WRITE
) == 0)
52 res
->size
= i_size_read(inode
);
53 res
->change_attr
= delegation
->change_attr
;
54 if (nfsi
->nrequests
!= 0)
56 res
->ctime
= inode
->i_ctime
;
57 res
->mtime
= inode
->i_mtime
;
58 res
->bitmap
[0] = (FATTR4_WORD0_CHANGE
|FATTR4_WORD0_SIZE
) &
60 res
->bitmap
[1] = (FATTR4_WORD1_TIME_METADATA
|FATTR4_WORD1_TIME_MODIFY
) &
65 trace_nfs4_cb_getattr(cps
->clp
, &args
->fh
, inode
, -ntohl(res
->status
));
68 dprintk("%s: exit with status = %d\n", __func__
, ntohl(res
->status
));
72 __be32
nfs4_callback_recall(void *argp
, void *resp
,
73 struct cb_process_state
*cps
)
75 struct cb_recallargs
*args
= argp
;
79 res
= htonl(NFS4ERR_OP_NOT_IN_SESSION
);
80 if (!cps
->clp
) /* Always set for v4.0. Set in cb_sequence for v4.1 */
83 dprintk_rcu("NFS: RECALL callback request from %s\n",
84 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
86 res
= htonl(NFS4ERR_BADHANDLE
);
87 inode
= nfs_delegation_find_inode(cps
->clp
, &args
->fh
);
89 trace_nfs4_cb_recall(cps
->clp
, &args
->fh
, NULL
,
90 &args
->stateid
, -ntohl(res
));
93 /* Set up a helper thread to actually return the delegation */
94 switch (nfs_async_inode_return_delegation(inode
, &args
->stateid
)) {
99 res
= htonl(NFS4ERR_BAD_STATEID
);
102 res
= htonl(NFS4ERR_RESOURCE
);
104 trace_nfs4_cb_recall(cps
->clp
, &args
->fh
, inode
,
105 &args
->stateid
, -ntohl(res
));
108 dprintk("%s: exit with status = %d\n", __func__
, ntohl(res
));
112 #if defined(CONFIG_NFS_V4_1)
115 * Lookup a layout inode by stateid
117 * Note: returns a refcount on the inode and superblock
119 static struct inode
*nfs_layout_find_inode_by_stateid(struct nfs_client
*clp
,
120 const nfs4_stateid
*stateid
)
122 struct nfs_server
*server
;
124 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 (stateid
!= NULL
&&
130 !nfs4_stateid_match_other(stateid
, &lo
->plh_stateid
))
132 inode
= igrab(lo
->plh_inode
);
135 if (!nfs_sb_active(inode
->i_sb
)) {
137 spin_unlock(&clp
->cl_lock
);
139 spin_lock(&clp
->cl_lock
);
151 * Lookup a layout inode by filehandle.
153 * Note: returns a refcount on the inode and superblock
156 static struct inode
*nfs_layout_find_inode_by_fh(struct nfs_client
*clp
,
157 const struct nfs_fh
*fh
)
159 struct nfs_server
*server
;
160 struct nfs_inode
*nfsi
;
162 struct pnfs_layout_hdr
*lo
;
165 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
166 list_for_each_entry(lo
, &server
->layouts
, plh_layouts
) {
167 nfsi
= NFS_I(lo
->plh_inode
);
168 if (nfs_compare_fh(fh
, &nfsi
->fh
))
170 if (nfsi
->layout
!= lo
)
172 inode
= igrab(lo
->plh_inode
);
175 if (!nfs_sb_active(inode
->i_sb
)) {
177 spin_unlock(&clp
->cl_lock
);
179 spin_lock(&clp
->cl_lock
);
190 static struct inode
*nfs_layout_find_inode(struct nfs_client
*clp
,
191 const struct nfs_fh
*fh
,
192 const nfs4_stateid
*stateid
)
196 spin_lock(&clp
->cl_lock
);
198 inode
= nfs_layout_find_inode_by_stateid(clp
, stateid
);
200 inode
= nfs_layout_find_inode_by_fh(clp
, fh
);
202 spin_unlock(&clp
->cl_lock
);
208 * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
210 static u32
pnfs_check_callback_stateid(struct pnfs_layout_hdr
*lo
,
211 const nfs4_stateid
*new)
215 /* Is the stateid still not initialised? */
216 if (!pnfs_layout_is_valid(lo
))
217 return NFS4ERR_DELAY
;
219 /* Mismatched stateid? */
220 if (!nfs4_stateid_match_other(&lo
->plh_stateid
, new))
221 return NFS4ERR_BAD_STATEID
;
223 newseq
= be32_to_cpu(new->seqid
);
224 /* Are we already in a layout recall situation? */
225 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) &&
226 lo
->plh_return_seq
!= 0) {
227 if (newseq
< lo
->plh_return_seq
)
228 return NFS4ERR_OLD_STATEID
;
229 if (newseq
> lo
->plh_return_seq
)
230 return NFS4ERR_DELAY
;
234 /* Check that the stateid matches what we think it should be. */
235 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
236 if (newseq
> oldseq
+ 1)
237 return NFS4ERR_DELAY
;
239 if (newseq
<= oldseq
)
240 return NFS4ERR_OLD_STATEID
;
245 static u32
initiate_file_draining(struct nfs_client
*clp
,
246 struct cb_layoutrecallargs
*args
)
249 struct pnfs_layout_hdr
*lo
;
250 u32 rv
= NFS4ERR_NOMATCHING_LAYOUT
;
251 LIST_HEAD(free_me_list
);
253 ino
= nfs_layout_find_inode(clp
, &args
->cbl_fh
, &args
->cbl_stateid
);
257 pnfs_layoutcommit_inode(ino
, false);
260 spin_lock(&ino
->i_lock
);
261 lo
= NFS_I(ino
)->layout
;
263 spin_unlock(&ino
->i_lock
);
266 pnfs_get_layout_hdr(lo
);
267 rv
= pnfs_check_callback_stateid(lo
, &args
->cbl_stateid
);
270 pnfs_set_layout_stateid(lo
, &args
->cbl_stateid
, true);
273 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
275 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
280 if (pnfs_mark_matching_lsegs_return(lo
, &free_me_list
,
282 be32_to_cpu(args
->cbl_stateid
.seqid
))) {
287 /* Embrace your forgetfulness! */
288 rv
= NFS4ERR_NOMATCHING_LAYOUT
;
290 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
) {
291 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
,
295 spin_unlock(&ino
->i_lock
);
296 pnfs_free_lseg_list(&free_me_list
);
297 /* Free all lsegs that are attached to commit buckets */
298 nfs_commit_inode(ino
, 0);
299 pnfs_put_layout_hdr(lo
);
301 trace_nfs4_cb_layoutrecall_file(clp
, &args
->cbl_fh
, ino
,
302 &args
->cbl_stateid
, -rv
);
303 nfs_iput_and_deactive(ino
);
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
)
407 if (args
->csa_slotid
> tbl
->server_highest_slotid
)
408 return htonl(NFS4ERR_BADSLOT
);
411 if (args
->csa_sequenceid
== slot
->seq_nr
) {
412 if (nfs4_test_locked_slot(tbl
, slot
->slot_nr
))
413 return htonl(NFS4ERR_DELAY
);
414 /* Signal process_op to set this error on next op */
415 if (args
->csa_cachethis
== 0)
416 return htonl(NFS4ERR_RETRY_UNCACHED_REP
);
418 /* Liar! We never allowed you to set csa_cachethis != 0 */
419 return htonl(NFS4ERR_SEQ_FALSE_RETRY
);
423 if (unlikely(slot
->seq_nr
== 0xFFFFFFFFU
)) {
424 if (args
->csa_sequenceid
== 1)
425 return htonl(NFS4_OK
);
426 } else if (likely(args
->csa_sequenceid
== slot
->seq_nr
+ 1))
427 return htonl(NFS4_OK
);
429 /* Misordered request */
430 return htonl(NFS4ERR_SEQ_MISORDERED
);
434 * For each referring call triple, check the session's slot table for
435 * a match. If the slot is in use and the sequence numbers match, the
436 * client is still waiting for a response to the original request.
438 static bool referring_call_exists(struct nfs_client
*clp
,
440 struct referring_call_list
*rclists
)
444 struct nfs4_session
*session
;
445 struct nfs4_slot_table
*tbl
;
446 struct referring_call_list
*rclist
;
447 struct referring_call
*ref
;
450 * XXX When client trunking is implemented, this becomes
451 * a session lookup from within the loop
453 session
= clp
->cl_session
;
454 tbl
= &session
->fc_slot_table
;
456 for (i
= 0; i
< nrclists
; i
++) {
457 rclist
= &rclists
[i
];
458 if (memcmp(session
->sess_id
.data
,
459 rclist
->rcl_sessionid
.data
,
460 NFS4_MAX_SESSIONID_LEN
) != 0)
463 for (j
= 0; j
< rclist
->rcl_nrefcalls
; j
++) {
464 ref
= &rclist
->rcl_refcalls
[j
];
465 status
= nfs4_slot_wait_on_seqid(tbl
, ref
->rc_slotid
,
466 ref
->rc_sequenceid
, HZ
>> 1) < 0;
476 __be32
nfs4_callback_sequence(void *argp
, void *resp
,
477 struct cb_process_state
*cps
)
479 struct cb_sequenceargs
*args
= argp
;
480 struct cb_sequenceres
*res
= resp
;
481 struct nfs4_slot_table
*tbl
;
482 struct nfs4_slot
*slot
;
483 struct nfs_client
*clp
;
485 __be32 status
= htonl(NFS4ERR_BADSESSION
);
487 clp
= nfs4_find_client_sessionid(cps
->net
, args
->csa_addr
,
488 &args
->csa_sessionid
, cps
->minorversion
);
492 if (!(clp
->cl_session
->flags
& SESSION4_BACK_CHAN
))
495 tbl
= &clp
->cl_session
->bc_slot_table
;
497 /* Set up res before grabbing the spinlock */
498 memcpy(&res
->csr_sessionid
, &args
->csa_sessionid
,
499 sizeof(res
->csr_sessionid
));
500 res
->csr_sequenceid
= args
->csa_sequenceid
;
501 res
->csr_slotid
= args
->csa_slotid
;
503 spin_lock(&tbl
->slot_tbl_lock
);
504 /* state manager is resetting the session */
505 if (test_bit(NFS4_SLOT_TBL_DRAINING
, &tbl
->slot_tbl_state
)) {
506 status
= htonl(NFS4ERR_DELAY
);
507 /* Return NFS4ERR_BADSESSION if we're draining the session
508 * in order to reset it.
510 if (test_bit(NFS4CLNT_SESSION_RESET
, &clp
->cl_state
))
511 status
= htonl(NFS4ERR_BADSESSION
);
515 status
= htonl(NFS4ERR_BADSLOT
);
516 slot
= nfs4_lookup_slot(tbl
, args
->csa_slotid
);
520 res
->csr_highestslotid
= tbl
->server_highest_slotid
;
521 res
->csr_target_highestslotid
= tbl
->target_highest_slotid
;
523 status
= validate_seqid(tbl
, slot
, args
);
526 if (!nfs4_try_to_lock_slot(tbl
, slot
)) {
527 status
= htonl(NFS4ERR_DELAY
);
532 /* The ca_maxresponsesize_cached is 0 with no DRC */
533 if (args
->csa_cachethis
!= 0) {
534 status
= htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE
);
539 * Check for pending referring calls. If a match is found, a
540 * related callback was received before the response to the original
543 if (referring_call_exists(clp
, args
->csa_nrclists
, args
->csa_rclists
)) {
544 status
= htonl(NFS4ERR_DELAY
);
550 * If CB_SEQUENCE returns an error, then the state of the slot
551 * (sequence ID, cached reply) MUST NOT change.
553 slot
->seq_nr
= args
->csa_sequenceid
;
555 spin_unlock(&tbl
->slot_tbl_lock
);
558 cps
->clp
= clp
; /* put in nfs4_callback_compound */
559 for (i
= 0; i
< args
->csa_nrclists
; i
++)
560 kfree(args
->csa_rclists
[i
].rcl_refcalls
);
561 kfree(args
->csa_rclists
);
563 if (status
== htonl(NFS4ERR_RETRY_UNCACHED_REP
)) {
564 cps
->drc_status
= status
;
567 res
->csr_status
= status
;
569 trace_nfs4_cb_sequence(args
, res
, status
);
574 validate_bitmap_values(unsigned long mask
)
576 return (mask
& ~RCA4_TYPE_MASK_ALL
) == 0;
579 __be32
nfs4_callback_recallany(void *argp
, void *resp
,
580 struct cb_process_state
*cps
)
582 struct cb_recallanyargs
*args
= argp
;
586 status
= cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION
);
587 if (!cps
->clp
) /* set in cb_sequence */
590 dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
591 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
593 status
= cpu_to_be32(NFS4ERR_INVAL
);
594 if (!validate_bitmap_values(args
->craa_type_mask
))
597 status
= cpu_to_be32(NFS4_OK
);
598 if (test_bit(RCA4_TYPE_MASK_RDATA_DLG
, (const unsigned long *)
599 &args
->craa_type_mask
))
601 if (test_bit(RCA4_TYPE_MASK_WDATA_DLG
, (const unsigned long *)
602 &args
->craa_type_mask
))
603 flags
|= FMODE_WRITE
;
604 if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT
, (const unsigned long *)
605 &args
->craa_type_mask
))
606 pnfs_recall_all_layouts(cps
->clp
);
608 nfs_expire_unused_delegation_types(cps
->clp
, flags
);
610 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
614 /* Reduce the fore channel's max_slots to the target value */
615 __be32
nfs4_callback_recallslot(void *argp
, void *resp
,
616 struct cb_process_state
*cps
)
618 struct cb_recallslotargs
*args
= argp
;
619 struct nfs4_slot_table
*fc_tbl
;
622 status
= htonl(NFS4ERR_OP_NOT_IN_SESSION
);
623 if (!cps
->clp
) /* set in cb_sequence */
626 dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
627 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
),
628 args
->crsa_target_highest_slotid
);
630 fc_tbl
= &cps
->clp
->cl_session
->fc_slot_table
;
632 status
= htonl(NFS4_OK
);
634 nfs41_set_target_slotid(fc_tbl
, args
->crsa_target_highest_slotid
);
635 nfs41_notify_server(cps
->clp
);
637 dprintk("%s: exit with status = %d\n", __func__
, ntohl(status
));
641 __be32
nfs4_callback_notify_lock(void *argp
, void *resp
,
642 struct cb_process_state
*cps
)
644 struct cb_notify_lock_args
*args
= argp
;
646 if (!cps
->clp
) /* set in cb_sequence */
647 return htonl(NFS4ERR_OP_NOT_IN_SESSION
);
649 dprintk_rcu("NFS: CB_NOTIFY_LOCK request from %s\n",
650 rpc_peeraddr2str(cps
->clp
->cl_rpcclient
, RPC_DISPLAY_ADDR
));
652 /* Don't wake anybody if the string looked bogus */
653 if (args
->cbnl_valid
)
654 __wake_up(&cps
->clp
->cl_lock_waitq
, TASK_NORMAL
, 0, args
);
656 return htonl(NFS4_OK
);
658 #endif /* CONFIG_NFS_V4_1 */