2 * Module for the pnfs nfs4 file layout driver.
3 * Defines all I/O and Policy interface operations, plus code
4 * to register itself with the pNFS client.
7 * The Regents of the University of Michigan
10 * Dean Hildebrand <dhildebz@umich.edu>
12 * Permission is granted to use, copy, create derivative works, and
13 * redistribute this software and such derivative works for any purpose,
14 * so long as the name of the University of Michigan is not used in
15 * any advertising or publicity pertaining to the use or distribution
16 * of this software without specific, written prior authorization. If
17 * the above copyright notice or any other identification of the
18 * University of Michigan is included in any copy of any portion of
19 * this software, then the disclaimer below must also be included.
21 * This software is provided as is, without representation or warranty
22 * of any kind either express or implied, including without limitation
23 * the implied warranties of merchantability, fitness for a particular
24 * purpose, or noninfringement. The Regents of the University of
25 * Michigan shall not be liable for any damages, including special,
26 * indirect, incidental, or consequential damages, with respect to any
27 * claim arising out of or in connection with the use of the software,
28 * even if it has been or is hereafter advised of the possibility of
32 #include <linux/nfs_fs.h>
33 #include <linux/nfs_page.h>
34 #include <linux/module.h>
36 #include <linux/sunrpc/metrics.h>
38 #include "nfs4session.h"
40 #include "delegation.h"
41 #include "nfs4filelayout.h"
43 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
45 MODULE_LICENSE("GPL");
46 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
47 MODULE_DESCRIPTION("The NFSv4 file layout driver");
49 #define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
52 filelayout_get_dense_offset(struct nfs4_filelayout_segment
*flseg
,
55 u32 stripe_width
= flseg
->stripe_unit
* flseg
->dsaddr
->stripe_count
;
59 offset
-= flseg
->pattern_offset
;
60 stripe_no
= div_u64(offset
, stripe_width
);
61 div_u64_rem(offset
, flseg
->stripe_unit
, &rem
);
63 return stripe_no
* flseg
->stripe_unit
+ rem
;
66 /* This function is used by the layout driver to calculate the
67 * offset of the file on the dserver based on whether the
68 * layout type is STRIPE_DENSE or STRIPE_SPARSE
71 filelayout_get_dserver_offset(struct pnfs_layout_segment
*lseg
, loff_t offset
)
73 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
75 switch (flseg
->stripe_type
) {
80 return filelayout_get_dense_offset(flseg
, offset
);
86 static void filelayout_reset_write(struct nfs_write_data
*data
)
88 struct nfs_pgio_header
*hdr
= data
->header
;
89 struct rpc_task
*task
= &data
->task
;
91 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
92 dprintk("%s Reset task %5u for i/o through MDS "
93 "(req %s/%lld, %u bytes @ offset %llu)\n", __func__
,
95 hdr
->inode
->i_sb
->s_id
,
96 (long long)NFS_FILEID(hdr
->inode
),
98 (unsigned long long)data
->args
.offset
);
100 task
->tk_status
= pnfs_write_done_resend_to_mds(hdr
->inode
,
107 static void filelayout_reset_read(struct nfs_read_data
*data
)
109 struct nfs_pgio_header
*hdr
= data
->header
;
110 struct rpc_task
*task
= &data
->task
;
112 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
113 dprintk("%s Reset task %5u for i/o through MDS "
114 "(req %s/%lld, %u bytes @ offset %llu)\n", __func__
,
116 hdr
->inode
->i_sb
->s_id
,
117 (long long)NFS_FILEID(hdr
->inode
),
119 (unsigned long long)data
->args
.offset
);
121 task
->tk_status
= pnfs_read_done_resend_to_mds(hdr
->inode
,
128 static void filelayout_fenceme(struct inode
*inode
, struct pnfs_layout_hdr
*lo
)
130 if (!test_and_clear_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
))
132 pnfs_return_layout(inode
);
135 static int filelayout_async_handle_error(struct rpc_task
*task
,
136 struct nfs4_state
*state
,
137 struct nfs_client
*clp
,
138 struct pnfs_layout_segment
*lseg
)
140 struct pnfs_layout_hdr
*lo
= lseg
->pls_layout
;
141 struct inode
*inode
= lo
->plh_inode
;
142 struct nfs_server
*mds_server
= NFS_SERVER(inode
);
143 struct nfs4_deviceid_node
*devid
= FILELAYOUT_DEVID_NODE(lseg
);
144 struct nfs_client
*mds_client
= mds_server
->nfs_client
;
145 struct nfs4_slot_table
*tbl
= &clp
->cl_session
->fc_slot_table
;
147 if (task
->tk_status
>= 0)
150 switch (task
->tk_status
) {
151 /* MDS state errors */
152 case -NFS4ERR_DELEG_REVOKED
:
153 case -NFS4ERR_ADMIN_REVOKED
:
154 case -NFS4ERR_BAD_STATEID
:
157 nfs_remove_bad_delegation(state
->inode
);
158 case -NFS4ERR_OPENMODE
:
161 if (nfs4_schedule_stateid_recovery(mds_server
, state
) < 0)
162 goto out_bad_stateid
;
163 goto wait_on_recovery
;
164 case -NFS4ERR_EXPIRED
:
166 if (nfs4_schedule_stateid_recovery(mds_server
, state
) < 0)
167 goto out_bad_stateid
;
169 nfs4_schedule_lease_recovery(mds_client
);
170 goto wait_on_recovery
;
171 /* DS session errors */
172 case -NFS4ERR_BADSESSION
:
173 case -NFS4ERR_BADSLOT
:
174 case -NFS4ERR_BAD_HIGH_SLOT
:
175 case -NFS4ERR_DEADSESSION
:
176 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION
:
177 case -NFS4ERR_SEQ_FALSE_RETRY
:
178 case -NFS4ERR_SEQ_MISORDERED
:
179 dprintk("%s ERROR %d, Reset session. Exchangeid "
180 "flags 0x%x\n", __func__
, task
->tk_status
,
181 clp
->cl_exchange_flags
);
182 nfs4_schedule_session_recovery(clp
->cl_session
, task
->tk_status
);
186 rpc_delay(task
, FILELAYOUT_POLL_RETRY_MAX
);
188 case -NFS4ERR_RETRY_UNCACHED_REP
:
190 /* Invalidate Layout errors */
191 case -NFS4ERR_PNFS_NO_LAYOUT
:
192 case -ESTALE
: /* mapped NFS4ERR_STALE */
193 case -EBADHANDLE
: /* mapped NFS4ERR_BADHANDLE */
194 case -EISDIR
: /* mapped NFS4ERR_ISDIR */
195 case -NFS4ERR_FHEXPIRED
:
196 case -NFS4ERR_WRONG_TYPE
:
197 dprintk("%s Invalid layout error %d\n", __func__
,
200 * Destroy layout so new i/o will get a new layout.
201 * Layout will not be destroyed until all current lseg
202 * references are put. Mark layout as invalid to resend failed
203 * i/o and all i/o waiting on the slot table to the MDS until
204 * layout is destroyed and a new valid layout is obtained.
206 pnfs_destroy_layout(NFS_I(inode
));
207 rpc_wake_up(&tbl
->slot_tbl_waitq
);
209 /* RPC connection errors */
217 dprintk("%s DS connection error %d\n", __func__
,
219 nfs4_mark_deviceid_unavailable(devid
);
220 set_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
221 rpc_wake_up(&tbl
->slot_tbl_waitq
);
225 dprintk("%s Retry through MDS. Error %d\n", __func__
,
227 return -NFS4ERR_RESET_TO_MDS
;
233 task
->tk_status
= -EIO
;
236 rpc_sleep_on(&mds_client
->cl_rpcwaitq
, task
, NULL
);
237 if (test_bit(NFS4CLNT_MANAGER_RUNNING
, &mds_client
->cl_state
) == 0)
238 rpc_wake_up_queued_task(&mds_client
->cl_rpcwaitq
, task
);
242 /* NFS_PROTO call done callback routines */
244 static int filelayout_read_done_cb(struct rpc_task
*task
,
245 struct nfs_read_data
*data
)
247 struct nfs_pgio_header
*hdr
= data
->header
;
250 err
= filelayout_async_handle_error(task
, data
->args
.context
->state
,
251 data
->ds_clp
, hdr
->lseg
);
254 case -NFS4ERR_RESET_TO_MDS
:
255 filelayout_reset_read(data
);
256 return task
->tk_status
;
258 rpc_restart_call_prepare(task
);
266 * We reference the rpc_cred of the first WRITE that triggers the need for
267 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
268 * rfc5661 is not clear about which credential should be used.
271 filelayout_set_layoutcommit(struct nfs_write_data
*wdata
)
273 struct nfs_pgio_header
*hdr
= wdata
->header
;
275 if (FILELAYOUT_LSEG(hdr
->lseg
)->commit_through_mds
||
276 wdata
->res
.verf
->committed
== NFS_FILE_SYNC
)
279 pnfs_set_layoutcommit(wdata
);
280 dprintk("%s ionde %lu pls_end_pos %lu\n", __func__
, hdr
->inode
->i_ino
,
281 (unsigned long) NFS_I(hdr
->inode
)->layout
->plh_lwb
);
285 filelayout_test_devid_unavailable(struct nfs4_deviceid_node
*node
)
287 return filelayout_test_devid_invalid(node
) ||
288 nfs4_test_deviceid_unavailable(node
);
292 filelayout_reset_to_mds(struct pnfs_layout_segment
*lseg
)
294 struct nfs4_deviceid_node
*node
= FILELAYOUT_DEVID_NODE(lseg
);
296 return filelayout_test_devid_unavailable(node
);
300 * Call ops for the async read/write cases
301 * In the case of dense layouts, the offset needs to be reset to its
304 static void filelayout_read_prepare(struct rpc_task
*task
, void *data
)
306 struct nfs_read_data
*rdata
= data
;
308 if (unlikely(test_bit(NFS_CONTEXT_BAD
, &rdata
->args
.context
->flags
))) {
309 rpc_exit(task
, -EIO
);
312 if (filelayout_reset_to_mds(rdata
->header
->lseg
)) {
313 dprintk("%s task %u reset io to MDS\n", __func__
, task
->tk_pid
);
314 filelayout_reset_read(rdata
);
318 rdata
->read_done_cb
= filelayout_read_done_cb
;
320 if (nfs41_setup_sequence(rdata
->ds_clp
->cl_session
,
321 &rdata
->args
.seq_args
,
325 nfs4_set_rw_stateid(&rdata
->args
.stateid
, rdata
->args
.context
,
326 rdata
->args
.lock_context
, FMODE_READ
);
329 static void filelayout_read_call_done(struct rpc_task
*task
, void *data
)
331 struct nfs_read_data
*rdata
= data
;
333 dprintk("--> %s task->tk_status %d\n", __func__
, task
->tk_status
);
335 if (test_bit(NFS_IOHDR_REDO
, &rdata
->header
->flags
) &&
336 task
->tk_status
== 0)
339 /* Note this may cause RPC to be resent */
340 rdata
->header
->mds_ops
->rpc_call_done(task
, data
);
343 static void filelayout_read_count_stats(struct rpc_task
*task
, void *data
)
345 struct nfs_read_data
*rdata
= data
;
347 rpc_count_iostats(task
, NFS_SERVER(rdata
->header
->inode
)->client
->cl_metrics
);
350 static void filelayout_read_release(void *data
)
352 struct nfs_read_data
*rdata
= data
;
353 struct pnfs_layout_hdr
*lo
= rdata
->header
->lseg
->pls_layout
;
355 filelayout_fenceme(lo
->plh_inode
, lo
);
356 nfs_put_client(rdata
->ds_clp
);
357 rdata
->header
->mds_ops
->rpc_release(data
);
360 static int filelayout_write_done_cb(struct rpc_task
*task
,
361 struct nfs_write_data
*data
)
363 struct nfs_pgio_header
*hdr
= data
->header
;
366 err
= filelayout_async_handle_error(task
, data
->args
.context
->state
,
367 data
->ds_clp
, hdr
->lseg
);
370 case -NFS4ERR_RESET_TO_MDS
:
371 filelayout_reset_write(data
);
372 return task
->tk_status
;
374 rpc_restart_call_prepare(task
);
378 filelayout_set_layoutcommit(data
);
382 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
383 static void prepare_to_resend_writes(struct nfs_commit_data
*data
)
385 struct nfs_page
*first
= nfs_list_entry(data
->pages
.next
);
387 data
->task
.tk_status
= 0;
388 memcpy(&data
->verf
.verifier
, &first
->wb_verf
,
389 sizeof(data
->verf
.verifier
));
390 data
->verf
.verifier
.data
[0]++; /* ensure verifier mismatch */
393 static int filelayout_commit_done_cb(struct rpc_task
*task
,
394 struct nfs_commit_data
*data
)
398 err
= filelayout_async_handle_error(task
, NULL
, data
->ds_clp
,
402 case -NFS4ERR_RESET_TO_MDS
:
403 prepare_to_resend_writes(data
);
406 rpc_restart_call_prepare(task
);
413 static void filelayout_write_prepare(struct rpc_task
*task
, void *data
)
415 struct nfs_write_data
*wdata
= data
;
417 if (unlikely(test_bit(NFS_CONTEXT_BAD
, &wdata
->args
.context
->flags
))) {
418 rpc_exit(task
, -EIO
);
421 if (filelayout_reset_to_mds(wdata
->header
->lseg
)) {
422 dprintk("%s task %u reset io to MDS\n", __func__
, task
->tk_pid
);
423 filelayout_reset_write(wdata
);
427 if (nfs41_setup_sequence(wdata
->ds_clp
->cl_session
,
428 &wdata
->args
.seq_args
,
432 nfs4_set_rw_stateid(&wdata
->args
.stateid
, wdata
->args
.context
,
433 wdata
->args
.lock_context
, FMODE_WRITE
);
436 static void filelayout_write_call_done(struct rpc_task
*task
, void *data
)
438 struct nfs_write_data
*wdata
= data
;
440 if (test_bit(NFS_IOHDR_REDO
, &wdata
->header
->flags
) &&
441 task
->tk_status
== 0)
444 /* Note this may cause RPC to be resent */
445 wdata
->header
->mds_ops
->rpc_call_done(task
, data
);
448 static void filelayout_write_count_stats(struct rpc_task
*task
, void *data
)
450 struct nfs_write_data
*wdata
= data
;
452 rpc_count_iostats(task
, NFS_SERVER(wdata
->header
->inode
)->client
->cl_metrics
);
455 static void filelayout_write_release(void *data
)
457 struct nfs_write_data
*wdata
= data
;
458 struct pnfs_layout_hdr
*lo
= wdata
->header
->lseg
->pls_layout
;
460 filelayout_fenceme(lo
->plh_inode
, lo
);
461 nfs_put_client(wdata
->ds_clp
);
462 wdata
->header
->mds_ops
->rpc_release(data
);
465 static void filelayout_commit_prepare(struct rpc_task
*task
, void *data
)
467 struct nfs_commit_data
*wdata
= data
;
469 nfs41_setup_sequence(wdata
->ds_clp
->cl_session
,
470 &wdata
->args
.seq_args
,
475 static void filelayout_write_commit_done(struct rpc_task
*task
, void *data
)
477 struct nfs_commit_data
*wdata
= data
;
479 /* Note this may cause RPC to be resent */
480 wdata
->mds_ops
->rpc_call_done(task
, data
);
483 static void filelayout_commit_count_stats(struct rpc_task
*task
, void *data
)
485 struct nfs_commit_data
*cdata
= data
;
487 rpc_count_iostats(task
, NFS_SERVER(cdata
->inode
)->client
->cl_metrics
);
490 static void filelayout_commit_release(void *calldata
)
492 struct nfs_commit_data
*data
= calldata
;
494 data
->completion_ops
->completion(data
);
495 pnfs_put_lseg(data
->lseg
);
496 nfs_put_client(data
->ds_clp
);
497 nfs_commitdata_release(data
);
500 static const struct rpc_call_ops filelayout_read_call_ops
= {
501 .rpc_call_prepare
= filelayout_read_prepare
,
502 .rpc_call_done
= filelayout_read_call_done
,
503 .rpc_count_stats
= filelayout_read_count_stats
,
504 .rpc_release
= filelayout_read_release
,
507 static const struct rpc_call_ops filelayout_write_call_ops
= {
508 .rpc_call_prepare
= filelayout_write_prepare
,
509 .rpc_call_done
= filelayout_write_call_done
,
510 .rpc_count_stats
= filelayout_write_count_stats
,
511 .rpc_release
= filelayout_write_release
,
514 static const struct rpc_call_ops filelayout_commit_call_ops
= {
515 .rpc_call_prepare
= filelayout_commit_prepare
,
516 .rpc_call_done
= filelayout_write_commit_done
,
517 .rpc_count_stats
= filelayout_commit_count_stats
,
518 .rpc_release
= filelayout_commit_release
,
521 static enum pnfs_try_status
522 filelayout_read_pagelist(struct nfs_read_data
*data
)
524 struct nfs_pgio_header
*hdr
= data
->header
;
525 struct pnfs_layout_segment
*lseg
= hdr
->lseg
;
526 struct nfs4_pnfs_ds
*ds
;
527 loff_t offset
= data
->args
.offset
;
531 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
532 __func__
, hdr
->inode
->i_ino
,
533 data
->args
.pgbase
, (size_t)data
->args
.count
, offset
);
535 /* Retrieve the correct rpc_client for the byte range */
536 j
= nfs4_fl_calc_j_index(lseg
, offset
);
537 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
538 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
540 return PNFS_NOT_ATTEMPTED
;
541 dprintk("%s USE DS: %s cl_count %d\n", __func__
,
542 ds
->ds_remotestr
, atomic_read(&ds
->ds_clp
->cl_count
));
544 /* No multipath support. Use first DS */
545 atomic_inc(&ds
->ds_clp
->cl_count
);
546 data
->ds_clp
= ds
->ds_clp
;
547 fh
= nfs4_fl_select_ds_fh(lseg
, j
);
551 data
->args
.offset
= filelayout_get_dserver_offset(lseg
, offset
);
552 data
->mds_offset
= offset
;
554 /* Perform an asynchronous read to ds */
555 nfs_initiate_read(ds
->ds_clp
->cl_rpcclient
, data
,
556 &filelayout_read_call_ops
, RPC_TASK_SOFTCONN
);
557 return PNFS_ATTEMPTED
;
560 /* Perform async writes. */
561 static enum pnfs_try_status
562 filelayout_write_pagelist(struct nfs_write_data
*data
, int sync
)
564 struct nfs_pgio_header
*hdr
= data
->header
;
565 struct pnfs_layout_segment
*lseg
= hdr
->lseg
;
566 struct nfs4_pnfs_ds
*ds
;
567 loff_t offset
= data
->args
.offset
;
571 /* Retrieve the correct rpc_client for the byte range */
572 j
= nfs4_fl_calc_j_index(lseg
, offset
);
573 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
574 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
576 return PNFS_NOT_ATTEMPTED
;
577 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n",
578 __func__
, hdr
->inode
->i_ino
, sync
, (size_t) data
->args
.count
,
579 offset
, ds
->ds_remotestr
, atomic_read(&ds
->ds_clp
->cl_count
));
581 data
->write_done_cb
= filelayout_write_done_cb
;
582 atomic_inc(&ds
->ds_clp
->cl_count
);
583 data
->ds_clp
= ds
->ds_clp
;
584 fh
= nfs4_fl_select_ds_fh(lseg
, j
);
588 * Get the file offset on the dserver. Set the write offset to
589 * this offset and save the original offset.
591 data
->args
.offset
= filelayout_get_dserver_offset(lseg
, offset
);
593 /* Perform an asynchronous write */
594 nfs_initiate_write(ds
->ds_clp
->cl_rpcclient
, data
,
595 &filelayout_write_call_ops
, sync
,
597 return PNFS_ATTEMPTED
;
601 * filelayout_check_layout()
603 * Make sure layout segment parameters are sane WRT the device.
604 * At this point no generic layer initialization of the lseg has occurred,
605 * and nothing has been added to the layout_hdr cache.
609 filelayout_check_layout(struct pnfs_layout_hdr
*lo
,
610 struct nfs4_filelayout_segment
*fl
,
611 struct nfs4_layoutget_res
*lgr
,
612 struct nfs4_deviceid
*id
,
615 struct nfs4_deviceid_node
*d
;
616 struct nfs4_file_layout_dsaddr
*dsaddr
;
617 int status
= -EINVAL
;
618 struct nfs_server
*nfss
= NFS_SERVER(lo
->plh_inode
);
620 dprintk("--> %s\n", __func__
);
622 /* FIXME: remove this check when layout segment support is added */
623 if (lgr
->range
.offset
!= 0 ||
624 lgr
->range
.length
!= NFS4_MAX_UINT64
) {
625 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
630 if (fl
->pattern_offset
> lgr
->range
.offset
) {
631 dprintk("%s pattern_offset %lld too large\n",
632 __func__
, fl
->pattern_offset
);
636 if (!fl
->stripe_unit
|| fl
->stripe_unit
% PAGE_SIZE
) {
637 dprintk("%s Invalid stripe unit (%u)\n",
638 __func__
, fl
->stripe_unit
);
642 /* find and reference the deviceid */
643 d
= nfs4_find_get_deviceid(NFS_SERVER(lo
->plh_inode
)->pnfs_curr_ld
,
644 NFS_SERVER(lo
->plh_inode
)->nfs_client
, id
);
646 dsaddr
= filelayout_get_device_info(lo
->plh_inode
, id
,
647 lo
->plh_lc_cred
, gfp_flags
);
651 dsaddr
= container_of(d
, struct nfs4_file_layout_dsaddr
, id_node
);
652 /* Found deviceid is unavailable */
653 if (filelayout_test_devid_unavailable(&dsaddr
->id_node
))
658 if (fl
->first_stripe_index
>= dsaddr
->stripe_count
) {
659 dprintk("%s Bad first_stripe_index %u\n",
660 __func__
, fl
->first_stripe_index
);
664 if ((fl
->stripe_type
== STRIPE_SPARSE
&&
665 fl
->num_fh
> 1 && fl
->num_fh
!= dsaddr
->ds_num
) ||
666 (fl
->stripe_type
== STRIPE_DENSE
&&
667 fl
->num_fh
!= dsaddr
->stripe_count
)) {
668 dprintk("%s num_fh %u not valid for given packing\n",
669 __func__
, fl
->num_fh
);
673 if (fl
->stripe_unit
% nfss
->rsize
|| fl
->stripe_unit
% nfss
->wsize
) {
674 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
675 "wsize %u\n", __func__
, fl
->stripe_unit
, nfss
->rsize
,
681 dprintk("--> %s returns %d\n", __func__
, status
);
684 nfs4_fl_put_deviceid(dsaddr
);
688 static void filelayout_free_fh_array(struct nfs4_filelayout_segment
*fl
)
692 for (i
= 0; i
< fl
->num_fh
; i
++) {
693 if (!fl
->fh_array
[i
])
695 kfree(fl
->fh_array
[i
]);
702 _filelayout_free_lseg(struct nfs4_filelayout_segment
*fl
)
704 filelayout_free_fh_array(fl
);
709 filelayout_decode_layout(struct pnfs_layout_hdr
*flo
,
710 struct nfs4_filelayout_segment
*fl
,
711 struct nfs4_layoutget_res
*lgr
,
712 struct nfs4_deviceid
*id
,
715 struct xdr_stream stream
;
717 struct page
*scratch
;
722 dprintk("%s: set_layout_map Begin\n", __func__
);
724 scratch
= alloc_page(gfp_flags
);
728 xdr_init_decode_pages(&stream
, &buf
, lgr
->layoutp
->pages
, lgr
->layoutp
->len
);
729 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
731 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
733 p
= xdr_inline_decode(&stream
, NFS4_DEVICEID4_SIZE
+ 20);
737 memcpy(id
, p
, sizeof(*id
));
738 p
+= XDR_QUADLEN(NFS4_DEVICEID4_SIZE
);
739 nfs4_print_deviceid(id
);
741 nfl_util
= be32_to_cpup(p
++);
742 if (nfl_util
& NFL4_UFLG_COMMIT_THRU_MDS
)
743 fl
->commit_through_mds
= 1;
744 if (nfl_util
& NFL4_UFLG_DENSE
)
745 fl
->stripe_type
= STRIPE_DENSE
;
747 fl
->stripe_type
= STRIPE_SPARSE
;
748 fl
->stripe_unit
= nfl_util
& ~NFL4_UFLG_MASK
;
750 fl
->first_stripe_index
= be32_to_cpup(p
++);
751 p
= xdr_decode_hyper(p
, &fl
->pattern_offset
);
752 fl
->num_fh
= be32_to_cpup(p
++);
754 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
755 __func__
, nfl_util
, fl
->num_fh
, fl
->first_stripe_index
,
758 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
759 * Futher checking is done in filelayout_check_layout */
761 max(NFS4_PNFS_MAX_STRIPE_CNT
, NFS4_PNFS_MAX_MULTI_CNT
))
764 if (fl
->num_fh
> 0) {
765 fl
->fh_array
= kcalloc(fl
->num_fh
, sizeof(fl
->fh_array
[0]),
771 for (i
= 0; i
< fl
->num_fh
; i
++) {
772 /* Do we want to use a mempool here? */
773 fl
->fh_array
[i
] = kmalloc(sizeof(struct nfs_fh
), gfp_flags
);
774 if (!fl
->fh_array
[i
])
777 p
= xdr_inline_decode(&stream
, 4);
780 fl
->fh_array
[i
]->size
= be32_to_cpup(p
++);
781 if (sizeof(struct nfs_fh
) < fl
->fh_array
[i
]->size
) {
782 printk(KERN_ERR
"NFS: Too big fh %d received %d\n",
783 i
, fl
->fh_array
[i
]->size
);
787 p
= xdr_inline_decode(&stream
, fl
->fh_array
[i
]->size
);
790 memcpy(fl
->fh_array
[i
]->data
, p
, fl
->fh_array
[i
]->size
);
791 dprintk("DEBUG: %s: fh len %d\n", __func__
,
792 fl
->fh_array
[i
]->size
);
795 __free_page(scratch
);
799 filelayout_free_fh_array(fl
);
801 __free_page(scratch
);
806 filelayout_free_lseg(struct pnfs_layout_segment
*lseg
)
808 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
810 dprintk("--> %s\n", __func__
);
811 nfs4_fl_put_deviceid(fl
->dsaddr
);
812 /* This assumes a single RW lseg */
813 if (lseg
->pls_range
.iomode
== IOMODE_RW
) {
814 struct nfs4_filelayout
*flo
;
816 flo
= FILELAYOUT_FROM_HDR(lseg
->pls_layout
);
817 flo
->commit_info
.nbuckets
= 0;
818 kfree(flo
->commit_info
.buckets
);
819 flo
->commit_info
.buckets
= NULL
;
821 _filelayout_free_lseg(fl
);
825 filelayout_alloc_commit_info(struct pnfs_layout_segment
*lseg
,
826 struct nfs_commit_info
*cinfo
,
829 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
830 struct pnfs_commit_bucket
*buckets
;
833 if (fl
->commit_through_mds
)
835 if (cinfo
->ds
->nbuckets
!= 0) {
836 /* This assumes there is only one IOMODE_RW lseg. What
837 * we really want to do is have a layout_hdr level
838 * dictionary of <multipath_list4, fh> keys, each
839 * associated with a struct list_head, populated by calls
840 * to filelayout_write_pagelist().
845 size
= (fl
->stripe_type
== STRIPE_SPARSE
) ?
846 fl
->dsaddr
->ds_num
: fl
->dsaddr
->stripe_count
;
848 buckets
= kcalloc(size
, sizeof(struct pnfs_commit_bucket
),
855 spin_lock(cinfo
->lock
);
856 if (cinfo
->ds
->nbuckets
!= 0)
859 cinfo
->ds
->buckets
= buckets
;
860 cinfo
->ds
->nbuckets
= size
;
861 for (i
= 0; i
< size
; i
++) {
862 INIT_LIST_HEAD(&buckets
[i
].written
);
863 INIT_LIST_HEAD(&buckets
[i
].committing
);
866 spin_unlock(cinfo
->lock
);
871 static struct pnfs_layout_segment
*
872 filelayout_alloc_lseg(struct pnfs_layout_hdr
*layoutid
,
873 struct nfs4_layoutget_res
*lgr
,
876 struct nfs4_filelayout_segment
*fl
;
878 struct nfs4_deviceid id
;
880 dprintk("--> %s\n", __func__
);
881 fl
= kzalloc(sizeof(*fl
), gfp_flags
);
885 rc
= filelayout_decode_layout(layoutid
, fl
, lgr
, &id
, gfp_flags
);
886 if (rc
!= 0 || filelayout_check_layout(layoutid
, fl
, lgr
, &id
, gfp_flags
)) {
887 _filelayout_free_lseg(fl
);
890 return &fl
->generic_hdr
;
894 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
896 * return true : coalesce page
897 * return false : don't coalesce page
900 filelayout_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
901 struct nfs_page
*req
)
903 u64 p_stripe
, r_stripe
;
906 if (!pnfs_generic_pg_test(pgio
, prev
, req
) ||
907 !nfs_generic_pg_test(pgio
, prev
, req
))
910 p_stripe
= (u64
)req_offset(prev
);
911 r_stripe
= (u64
)req_offset(req
);
912 stripe_unit
= FILELAYOUT_LSEG(pgio
->pg_lseg
)->stripe_unit
;
914 do_div(p_stripe
, stripe_unit
);
915 do_div(r_stripe
, stripe_unit
);
917 return (p_stripe
== r_stripe
);
921 filelayout_pg_init_read(struct nfs_pageio_descriptor
*pgio
,
922 struct nfs_page
*req
)
924 WARN_ON_ONCE(pgio
->pg_lseg
!= NULL
);
926 if (req
->wb_offset
!= req
->wb_pgbase
) {
928 * Handling unaligned pages is difficult, because have to
929 * somehow split a req in two in certain cases in the
930 * pg.test code. Avoid this by just not using pnfs
933 nfs_pageio_reset_read_mds(pgio
);
936 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
942 /* If no lseg, fall back to read through mds */
943 if (pgio
->pg_lseg
== NULL
)
944 nfs_pageio_reset_read_mds(pgio
);
948 filelayout_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
949 struct nfs_page
*req
)
951 struct nfs_commit_info cinfo
;
954 WARN_ON_ONCE(pgio
->pg_lseg
!= NULL
);
956 if (req
->wb_offset
!= req
->wb_pgbase
)
958 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
964 /* If no lseg, fall back to write through mds */
965 if (pgio
->pg_lseg
== NULL
)
967 nfs_init_cinfo(&cinfo
, pgio
->pg_inode
, pgio
->pg_dreq
);
968 status
= filelayout_alloc_commit_info(pgio
->pg_lseg
, &cinfo
, GFP_NOFS
);
970 pnfs_put_lseg(pgio
->pg_lseg
);
971 pgio
->pg_lseg
= NULL
;
976 nfs_pageio_reset_write_mds(pgio
);
979 static const struct nfs_pageio_ops filelayout_pg_read_ops
= {
980 .pg_init
= filelayout_pg_init_read
,
981 .pg_test
= filelayout_pg_test
,
982 .pg_doio
= pnfs_generic_pg_readpages
,
985 static const struct nfs_pageio_ops filelayout_pg_write_ops
= {
986 .pg_init
= filelayout_pg_init_write
,
987 .pg_test
= filelayout_pg_test
,
988 .pg_doio
= pnfs_generic_pg_writepages
,
991 static u32
select_bucket_index(struct nfs4_filelayout_segment
*fl
, u32 j
)
993 if (fl
->stripe_type
== STRIPE_SPARSE
)
994 return nfs4_fl_calc_ds_index(&fl
->generic_hdr
, j
);
999 /* The generic layer is about to remove the req from the commit list.
1000 * If this will make the bucket empty, it will need to put the lseg reference.
1003 filelayout_clear_request_commit(struct nfs_page
*req
,
1004 struct nfs_commit_info
*cinfo
)
1006 struct pnfs_layout_segment
*freeme
= NULL
;
1008 spin_lock(cinfo
->lock
);
1009 if (!test_and_clear_bit(PG_COMMIT_TO_DS
, &req
->wb_flags
))
1011 cinfo
->ds
->nwritten
--;
1012 if (list_is_singular(&req
->wb_list
)) {
1013 struct pnfs_commit_bucket
*bucket
;
1015 bucket
= list_first_entry(&req
->wb_list
,
1016 struct pnfs_commit_bucket
,
1018 freeme
= bucket
->wlseg
;
1019 bucket
->wlseg
= NULL
;
1022 nfs_request_remove_commit_list(req
, cinfo
);
1023 spin_unlock(cinfo
->lock
);
1024 pnfs_put_lseg(freeme
);
1027 static struct list_head
*
1028 filelayout_choose_commit_list(struct nfs_page
*req
,
1029 struct pnfs_layout_segment
*lseg
,
1030 struct nfs_commit_info
*cinfo
)
1032 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
1034 struct list_head
*list
;
1035 struct pnfs_commit_bucket
*buckets
;
1037 if (fl
->commit_through_mds
)
1038 return &cinfo
->mds
->list
;
1040 /* Note that we are calling nfs4_fl_calc_j_index on each page
1041 * that ends up being committed to a data server. An attractive
1042 * alternative is to add a field to nfs_write_data and nfs_page
1043 * to store the value calculated in filelayout_write_pagelist
1044 * and just use that here.
1046 j
= nfs4_fl_calc_j_index(lseg
, req_offset(req
));
1047 i
= select_bucket_index(fl
, j
);
1048 buckets
= cinfo
->ds
->buckets
;
1049 list
= &buckets
[i
].written
;
1050 if (list_empty(list
)) {
1051 /* Non-empty buckets hold a reference on the lseg. That ref
1052 * is normally transferred to the COMMIT call and released
1053 * there. It could also be released if the last req is pulled
1054 * off due to a rewrite, in which case it will be done in
1055 * filelayout_clear_request_commit
1057 buckets
[i
].wlseg
= pnfs_get_lseg(lseg
);
1059 set_bit(PG_COMMIT_TO_DS
, &req
->wb_flags
);
1060 cinfo
->ds
->nwritten
++;
1065 filelayout_mark_request_commit(struct nfs_page
*req
,
1066 struct pnfs_layout_segment
*lseg
,
1067 struct nfs_commit_info
*cinfo
)
1069 struct list_head
*list
;
1071 list
= filelayout_choose_commit_list(req
, lseg
, cinfo
);
1072 nfs_request_add_commit_list(req
, list
, cinfo
);
1075 static u32
calc_ds_index_from_commit(struct pnfs_layout_segment
*lseg
, u32 i
)
1077 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
1079 if (flseg
->stripe_type
== STRIPE_SPARSE
)
1082 return nfs4_fl_calc_ds_index(lseg
, i
);
1085 static struct nfs_fh
*
1086 select_ds_fh_from_commit(struct pnfs_layout_segment
*lseg
, u32 i
)
1088 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
1090 if (flseg
->stripe_type
== STRIPE_SPARSE
) {
1091 if (flseg
->num_fh
== 1)
1093 else if (flseg
->num_fh
== 0)
1094 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1097 return flseg
->fh_array
[i
];
1100 static int filelayout_initiate_commit(struct nfs_commit_data
*data
, int how
)
1102 struct pnfs_layout_segment
*lseg
= data
->lseg
;
1103 struct nfs4_pnfs_ds
*ds
;
1107 idx
= calc_ds_index_from_commit(lseg
, data
->ds_commit_index
);
1108 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
1110 prepare_to_resend_writes(data
);
1111 filelayout_commit_release(data
);
1114 dprintk("%s ino %lu, how %d cl_count %d\n", __func__
,
1115 data
->inode
->i_ino
, how
, atomic_read(&ds
->ds_clp
->cl_count
));
1116 data
->commit_done_cb
= filelayout_commit_done_cb
;
1117 atomic_inc(&ds
->ds_clp
->cl_count
);
1118 data
->ds_clp
= ds
->ds_clp
;
1119 fh
= select_ds_fh_from_commit(lseg
, data
->ds_commit_index
);
1122 return nfs_initiate_commit(ds
->ds_clp
->cl_rpcclient
, data
,
1123 &filelayout_commit_call_ops
, how
,
1128 transfer_commit_list(struct list_head
*src
, struct list_head
*dst
,
1129 struct nfs_commit_info
*cinfo
, int max
)
1131 struct nfs_page
*req
, *tmp
;
1134 list_for_each_entry_safe(req
, tmp
, src
, wb_list
) {
1135 if (!nfs_lock_request(req
))
1137 kref_get(&req
->wb_kref
);
1138 if (cond_resched_lock(cinfo
->lock
))
1139 list_safe_reset_next(req
, tmp
, wb_list
);
1140 nfs_request_remove_commit_list(req
, cinfo
);
1141 clear_bit(PG_COMMIT_TO_DS
, &req
->wb_flags
);
1142 nfs_list_add_request(req
, dst
);
1144 if ((ret
== max
) && !cinfo
->dreq
)
1151 filelayout_scan_ds_commit_list(struct pnfs_commit_bucket
*bucket
,
1152 struct nfs_commit_info
*cinfo
,
1155 struct list_head
*src
= &bucket
->written
;
1156 struct list_head
*dst
= &bucket
->committing
;
1159 ret
= transfer_commit_list(src
, dst
, cinfo
, max
);
1161 cinfo
->ds
->nwritten
-= ret
;
1162 cinfo
->ds
->ncommitting
+= ret
;
1163 bucket
->clseg
= bucket
->wlseg
;
1164 if (list_empty(src
))
1165 bucket
->wlseg
= NULL
;
1167 pnfs_get_lseg(bucket
->clseg
);
1172 /* Move reqs from written to committing lists, returning count of number moved.
1173 * Note called with cinfo->lock held.
1175 static int filelayout_scan_commit_lists(struct nfs_commit_info
*cinfo
,
1180 for (i
= 0; i
< cinfo
->ds
->nbuckets
&& max
!= 0; i
++) {
1181 cnt
= filelayout_scan_ds_commit_list(&cinfo
->ds
->buckets
[i
],
1189 /* Pull everything off the committing lists and dump into @dst */
1190 static void filelayout_recover_commit_reqs(struct list_head
*dst
,
1191 struct nfs_commit_info
*cinfo
)
1193 struct pnfs_commit_bucket
*b
;
1196 /* NOTE cinfo->lock is NOT held, relying on fact that this is
1197 * only called on single thread per dreq.
1198 * Can't take the lock because need to do pnfs_put_lseg
1200 for (i
= 0, b
= cinfo
->ds
->buckets
; i
< cinfo
->ds
->nbuckets
; i
++, b
++) {
1201 if (transfer_commit_list(&b
->written
, dst
, cinfo
, 0)) {
1202 pnfs_put_lseg(b
->wlseg
);
1206 cinfo
->ds
->nwritten
= 0;
1210 alloc_ds_commits(struct nfs_commit_info
*cinfo
, struct list_head
*list
)
1212 struct pnfs_ds_commit_info
*fl_cinfo
;
1213 struct pnfs_commit_bucket
*bucket
;
1214 struct nfs_commit_data
*data
;
1216 unsigned int nreq
= 0;
1218 fl_cinfo
= cinfo
->ds
;
1219 bucket
= fl_cinfo
->buckets
;
1220 for (i
= 0; i
< fl_cinfo
->nbuckets
; i
++, bucket
++) {
1221 if (list_empty(&bucket
->committing
))
1223 data
= nfs_commitdata_alloc();
1226 data
->ds_commit_index
= i
;
1227 data
->lseg
= bucket
->clseg
;
1228 bucket
->clseg
= NULL
;
1229 list_add(&data
->pages
, list
);
1233 /* Clean up on error */
1234 for (j
= i
; j
< fl_cinfo
->nbuckets
; j
++, bucket
++) {
1235 if (list_empty(&bucket
->committing
))
1237 nfs_retry_commit(&bucket
->committing
, bucket
->clseg
, cinfo
);
1238 pnfs_put_lseg(bucket
->clseg
);
1239 bucket
->clseg
= NULL
;
1241 /* Caller will clean up entries put on list */
1245 /* This follows nfs_commit_list pretty closely */
1247 filelayout_commit_pagelist(struct inode
*inode
, struct list_head
*mds_pages
,
1248 int how
, struct nfs_commit_info
*cinfo
)
1250 struct nfs_commit_data
*data
, *tmp
;
1252 unsigned int nreq
= 0;
1254 if (!list_empty(mds_pages
)) {
1255 data
= nfs_commitdata_alloc();
1258 list_add(&data
->pages
, &list
);
1261 nfs_retry_commit(mds_pages
, NULL
, cinfo
);
1264 nreq
+= alloc_ds_commits(cinfo
, &list
);
1267 cinfo
->completion_ops
->error_cleanup(NFS_I(inode
));
1271 atomic_add(nreq
, &cinfo
->mds
->rpcs_out
);
1273 list_for_each_entry_safe(data
, tmp
, &list
, pages
) {
1274 list_del_init(&data
->pages
);
1276 nfs_init_commit(data
, mds_pages
, NULL
, cinfo
);
1277 nfs_initiate_commit(NFS_CLIENT(inode
), data
,
1278 data
->mds_ops
, how
, 0);
1280 struct pnfs_commit_bucket
*buckets
;
1282 buckets
= cinfo
->ds
->buckets
;
1283 nfs_init_commit(data
, &buckets
[data
->ds_commit_index
].committing
, data
->lseg
, cinfo
);
1284 filelayout_initiate_commit(data
, how
);
1288 cinfo
->ds
->ncommitting
= 0;
1289 return PNFS_ATTEMPTED
;
1293 filelayout_free_deveiceid_node(struct nfs4_deviceid_node
*d
)
1295 nfs4_fl_free_deviceid(container_of(d
, struct nfs4_file_layout_dsaddr
, id_node
));
1298 static struct pnfs_layout_hdr
*
1299 filelayout_alloc_layout_hdr(struct inode
*inode
, gfp_t gfp_flags
)
1301 struct nfs4_filelayout
*flo
;
1303 flo
= kzalloc(sizeof(*flo
), gfp_flags
);
1304 return &flo
->generic_hdr
;
1308 filelayout_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
1310 kfree(FILELAYOUT_FROM_HDR(lo
));
1313 static struct pnfs_ds_commit_info
*
1314 filelayout_get_ds_info(struct inode
*inode
)
1316 struct pnfs_layout_hdr
*layout
= NFS_I(inode
)->layout
;
1321 return &FILELAYOUT_FROM_HDR(layout
)->commit_info
;
1324 static struct pnfs_layoutdriver_type filelayout_type
= {
1325 .id
= LAYOUT_NFSV4_1_FILES
,
1326 .name
= "LAYOUT_NFSV4_1_FILES",
1327 .owner
= THIS_MODULE
,
1328 .alloc_layout_hdr
= filelayout_alloc_layout_hdr
,
1329 .free_layout_hdr
= filelayout_free_layout_hdr
,
1330 .alloc_lseg
= filelayout_alloc_lseg
,
1331 .free_lseg
= filelayout_free_lseg
,
1332 .pg_read_ops
= &filelayout_pg_read_ops
,
1333 .pg_write_ops
= &filelayout_pg_write_ops
,
1334 .get_ds_info
= &filelayout_get_ds_info
,
1335 .mark_request_commit
= filelayout_mark_request_commit
,
1336 .clear_request_commit
= filelayout_clear_request_commit
,
1337 .scan_commit_lists
= filelayout_scan_commit_lists
,
1338 .recover_commit_reqs
= filelayout_recover_commit_reqs
,
1339 .commit_pagelist
= filelayout_commit_pagelist
,
1340 .read_pagelist
= filelayout_read_pagelist
,
1341 .write_pagelist
= filelayout_write_pagelist
,
1342 .free_deviceid_node
= filelayout_free_deveiceid_node
,
1345 static int __init
nfs4filelayout_init(void)
1347 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Registering...\n",
1349 return pnfs_register_layoutdriver(&filelayout_type
);
1352 static void __exit
nfs4filelayout_exit(void)
1354 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Unregistering...\n",
1356 pnfs_unregister_layoutdriver(&filelayout_type
);
1359 MODULE_ALIAS("nfs-layouttype4-1");
1361 module_init(nfs4filelayout_init
);
1362 module_exit(nfs4filelayout_exit
);