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"
39 #include "../internal.h"
40 #include "../delegation.h"
41 #include "filelayout.h"
42 #include "../nfs4trace.h"
44 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
46 MODULE_LICENSE("GPL");
47 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
48 MODULE_DESCRIPTION("The NFSv4 file layout driver");
50 #define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
53 filelayout_get_dense_offset(struct nfs4_filelayout_segment
*flseg
,
56 u32 stripe_width
= flseg
->stripe_unit
* flseg
->dsaddr
->stripe_count
;
60 offset
-= flseg
->pattern_offset
;
61 stripe_no
= div_u64(offset
, stripe_width
);
62 div_u64_rem(offset
, flseg
->stripe_unit
, &rem
);
64 return stripe_no
* flseg
->stripe_unit
+ rem
;
67 /* This function is used by the layout driver to calculate the
68 * offset of the file on the dserver based on whether the
69 * layout type is STRIPE_DENSE or STRIPE_SPARSE
72 filelayout_get_dserver_offset(struct pnfs_layout_segment
*lseg
, loff_t offset
)
74 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
76 switch (flseg
->stripe_type
) {
81 return filelayout_get_dense_offset(flseg
, offset
);
87 static void filelayout_reset_write(struct nfs_pgio_header
*hdr
)
89 struct rpc_task
*task
= &hdr
->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/%llu, %u bytes @ offset %llu)\n", __func__
,
95 hdr
->inode
->i_sb
->s_id
,
96 (unsigned long long)NFS_FILEID(hdr
->inode
),
98 (unsigned long long)hdr
->args
.offset
);
100 task
->tk_status
= pnfs_write_done_resend_to_mds(hdr
);
104 static void filelayout_reset_read(struct nfs_pgio_header
*hdr
)
106 struct rpc_task
*task
= &hdr
->task
;
108 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
109 dprintk("%s Reset task %5u for i/o through MDS "
110 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__
,
112 hdr
->inode
->i_sb
->s_id
,
113 (unsigned long long)NFS_FILEID(hdr
->inode
),
115 (unsigned long long)hdr
->args
.offset
);
117 task
->tk_status
= pnfs_read_done_resend_to_mds(hdr
);
121 static int filelayout_async_handle_error(struct rpc_task
*task
,
122 struct nfs4_state
*state
,
123 struct nfs_client
*clp
,
124 struct pnfs_layout_segment
*lseg
)
126 struct pnfs_layout_hdr
*lo
= lseg
->pls_layout
;
127 struct inode
*inode
= lo
->plh_inode
;
128 struct nfs_server
*mds_server
= NFS_SERVER(inode
);
129 struct nfs4_deviceid_node
*devid
= FILELAYOUT_DEVID_NODE(lseg
);
130 struct nfs_client
*mds_client
= mds_server
->nfs_client
;
131 struct nfs4_slot_table
*tbl
= &clp
->cl_session
->fc_slot_table
;
133 if (task
->tk_status
>= 0)
136 switch (task
->tk_status
) {
137 /* MDS state errors */
138 case -NFS4ERR_DELEG_REVOKED
:
139 case -NFS4ERR_ADMIN_REVOKED
:
140 case -NFS4ERR_BAD_STATEID
:
141 case -NFS4ERR_OPENMODE
:
144 if (nfs4_schedule_stateid_recovery(mds_server
, state
) < 0)
145 goto out_bad_stateid
;
146 goto wait_on_recovery
;
147 case -NFS4ERR_EXPIRED
:
149 if (nfs4_schedule_stateid_recovery(mds_server
, state
) < 0)
150 goto out_bad_stateid
;
152 nfs4_schedule_lease_recovery(mds_client
);
153 goto wait_on_recovery
;
154 /* DS session errors */
155 case -NFS4ERR_BADSESSION
:
156 case -NFS4ERR_BADSLOT
:
157 case -NFS4ERR_BAD_HIGH_SLOT
:
158 case -NFS4ERR_DEADSESSION
:
159 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION
:
160 case -NFS4ERR_SEQ_FALSE_RETRY
:
161 case -NFS4ERR_SEQ_MISORDERED
:
162 dprintk("%s ERROR %d, Reset session. Exchangeid "
163 "flags 0x%x\n", __func__
, task
->tk_status
,
164 clp
->cl_exchange_flags
);
165 nfs4_schedule_session_recovery(clp
->cl_session
, task
->tk_status
);
169 rpc_delay(task
, FILELAYOUT_POLL_RETRY_MAX
);
171 case -NFS4ERR_RETRY_UNCACHED_REP
:
173 /* Invalidate Layout errors */
174 case -NFS4ERR_PNFS_NO_LAYOUT
:
175 case -ESTALE
: /* mapped NFS4ERR_STALE */
176 case -EBADHANDLE
: /* mapped NFS4ERR_BADHANDLE */
177 case -EISDIR
: /* mapped NFS4ERR_ISDIR */
178 case -NFS4ERR_FHEXPIRED
:
179 case -NFS4ERR_WRONG_TYPE
:
180 dprintk("%s Invalid layout error %d\n", __func__
,
183 * Destroy layout so new i/o will get a new layout.
184 * Layout will not be destroyed until all current lseg
185 * references are put. Mark layout as invalid to resend failed
186 * i/o and all i/o waiting on the slot table to the MDS until
187 * layout is destroyed and a new valid layout is obtained.
189 pnfs_destroy_layout(NFS_I(inode
));
190 rpc_wake_up(&tbl
->slot_tbl_waitq
);
192 /* RPC connection errors */
200 dprintk("%s DS connection error %d\n", __func__
,
202 nfs4_mark_deviceid_unavailable(devid
);
203 pnfs_error_mark_layout_for_return(inode
, lseg
);
204 rpc_wake_up(&tbl
->slot_tbl_waitq
);
208 dprintk("%s Retry through MDS. Error %d\n", __func__
,
210 return -NFS4ERR_RESET_TO_MDS
;
216 task
->tk_status
= -EIO
;
219 rpc_sleep_on(&mds_client
->cl_rpcwaitq
, task
, NULL
);
220 if (test_bit(NFS4CLNT_MANAGER_RUNNING
, &mds_client
->cl_state
) == 0)
221 rpc_wake_up_queued_task(&mds_client
->cl_rpcwaitq
, task
);
225 /* NFS_PROTO call done callback routines */
227 static int filelayout_read_done_cb(struct rpc_task
*task
,
228 struct nfs_pgio_header
*hdr
)
232 trace_nfs4_pnfs_read(hdr
, task
->tk_status
);
233 err
= filelayout_async_handle_error(task
, hdr
->args
.context
->state
,
234 hdr
->ds_clp
, hdr
->lseg
);
237 case -NFS4ERR_RESET_TO_MDS
:
238 filelayout_reset_read(hdr
);
239 return task
->tk_status
;
241 rpc_restart_call_prepare(task
);
249 * We reference the rpc_cred of the first WRITE that triggers the need for
250 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
251 * rfc5661 is not clear about which credential should be used.
254 filelayout_set_layoutcommit(struct nfs_pgio_header
*hdr
)
257 if (FILELAYOUT_LSEG(hdr
->lseg
)->commit_through_mds
||
258 hdr
->res
.verf
->committed
!= NFS_DATA_SYNC
)
261 pnfs_set_layoutcommit(hdr
);
262 dprintk("%s inode %lu pls_end_pos %lu\n", __func__
, hdr
->inode
->i_ino
,
263 (unsigned long) NFS_I(hdr
->inode
)->layout
->plh_lwb
);
267 filelayout_test_devid_unavailable(struct nfs4_deviceid_node
*node
)
269 return filelayout_test_devid_invalid(node
) ||
270 nfs4_test_deviceid_unavailable(node
);
274 filelayout_reset_to_mds(struct pnfs_layout_segment
*lseg
)
276 struct nfs4_deviceid_node
*node
= FILELAYOUT_DEVID_NODE(lseg
);
278 return filelayout_test_devid_unavailable(node
);
282 * Call ops for the async read/write cases
283 * In the case of dense layouts, the offset needs to be reset to its
286 static void filelayout_read_prepare(struct rpc_task
*task
, void *data
)
288 struct nfs_pgio_header
*hdr
= data
;
290 if (unlikely(test_bit(NFS_CONTEXT_BAD
, &hdr
->args
.context
->flags
))) {
291 rpc_exit(task
, -EIO
);
294 if (filelayout_reset_to_mds(hdr
->lseg
)) {
295 dprintk("%s task %u reset io to MDS\n", __func__
, task
->tk_pid
);
296 filelayout_reset_read(hdr
);
300 hdr
->pgio_done_cb
= filelayout_read_done_cb
;
302 if (nfs41_setup_sequence(hdr
->ds_clp
->cl_session
,
307 if (nfs4_set_rw_stateid(&hdr
->args
.stateid
, hdr
->args
.context
,
308 hdr
->args
.lock_context
, FMODE_READ
) == -EIO
)
309 rpc_exit(task
, -EIO
); /* lost lock, terminate I/O */
312 static void filelayout_read_call_done(struct rpc_task
*task
, void *data
)
314 struct nfs_pgio_header
*hdr
= data
;
316 dprintk("--> %s task->tk_status %d\n", __func__
, task
->tk_status
);
318 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
) &&
319 task
->tk_status
== 0) {
320 nfs41_sequence_done(task
, &hdr
->res
.seq_res
);
324 /* Note this may cause RPC to be resent */
325 hdr
->mds_ops
->rpc_call_done(task
, data
);
328 static void filelayout_read_count_stats(struct rpc_task
*task
, void *data
)
330 struct nfs_pgio_header
*hdr
= data
;
332 rpc_count_iostats(task
, NFS_SERVER(hdr
->inode
)->client
->cl_metrics
);
335 static int filelayout_write_done_cb(struct rpc_task
*task
,
336 struct nfs_pgio_header
*hdr
)
340 trace_nfs4_pnfs_write(hdr
, task
->tk_status
);
341 err
= filelayout_async_handle_error(task
, hdr
->args
.context
->state
,
342 hdr
->ds_clp
, hdr
->lseg
);
345 case -NFS4ERR_RESET_TO_MDS
:
346 filelayout_reset_write(hdr
);
347 return task
->tk_status
;
349 rpc_restart_call_prepare(task
);
353 filelayout_set_layoutcommit(hdr
);
357 static int filelayout_commit_done_cb(struct rpc_task
*task
,
358 struct nfs_commit_data
*data
)
362 trace_nfs4_pnfs_commit_ds(data
, task
->tk_status
);
363 err
= filelayout_async_handle_error(task
, NULL
, data
->ds_clp
,
367 case -NFS4ERR_RESET_TO_MDS
:
368 pnfs_generic_prepare_to_resend_writes(data
);
371 rpc_restart_call_prepare(task
);
375 if (data
->verf
.committed
== NFS_UNSTABLE
)
376 pnfs_commit_set_layoutcommit(data
);
381 static void filelayout_write_prepare(struct rpc_task
*task
, void *data
)
383 struct nfs_pgio_header
*hdr
= data
;
385 if (unlikely(test_bit(NFS_CONTEXT_BAD
, &hdr
->args
.context
->flags
))) {
386 rpc_exit(task
, -EIO
);
389 if (filelayout_reset_to_mds(hdr
->lseg
)) {
390 dprintk("%s task %u reset io to MDS\n", __func__
, task
->tk_pid
);
391 filelayout_reset_write(hdr
);
395 if (nfs41_setup_sequence(hdr
->ds_clp
->cl_session
,
400 if (nfs4_set_rw_stateid(&hdr
->args
.stateid
, hdr
->args
.context
,
401 hdr
->args
.lock_context
, FMODE_WRITE
) == -EIO
)
402 rpc_exit(task
, -EIO
); /* lost lock, terminate I/O */
405 static void filelayout_write_call_done(struct rpc_task
*task
, void *data
)
407 struct nfs_pgio_header
*hdr
= data
;
409 if (test_bit(NFS_IOHDR_REDO
, &hdr
->flags
) &&
410 task
->tk_status
== 0) {
411 nfs41_sequence_done(task
, &hdr
->res
.seq_res
);
415 /* Note this may cause RPC to be resent */
416 hdr
->mds_ops
->rpc_call_done(task
, data
);
419 static void filelayout_write_count_stats(struct rpc_task
*task
, void *data
)
421 struct nfs_pgio_header
*hdr
= data
;
423 rpc_count_iostats(task
, NFS_SERVER(hdr
->inode
)->client
->cl_metrics
);
426 static void filelayout_commit_prepare(struct rpc_task
*task
, void *data
)
428 struct nfs_commit_data
*wdata
= data
;
430 nfs41_setup_sequence(wdata
->ds_clp
->cl_session
,
431 &wdata
->args
.seq_args
,
436 static void filelayout_commit_count_stats(struct rpc_task
*task
, void *data
)
438 struct nfs_commit_data
*cdata
= data
;
440 rpc_count_iostats(task
, NFS_SERVER(cdata
->inode
)->client
->cl_metrics
);
443 static const struct rpc_call_ops filelayout_read_call_ops
= {
444 .rpc_call_prepare
= filelayout_read_prepare
,
445 .rpc_call_done
= filelayout_read_call_done
,
446 .rpc_count_stats
= filelayout_read_count_stats
,
447 .rpc_release
= pnfs_generic_rw_release
,
450 static const struct rpc_call_ops filelayout_write_call_ops
= {
451 .rpc_call_prepare
= filelayout_write_prepare
,
452 .rpc_call_done
= filelayout_write_call_done
,
453 .rpc_count_stats
= filelayout_write_count_stats
,
454 .rpc_release
= pnfs_generic_rw_release
,
457 static const struct rpc_call_ops filelayout_commit_call_ops
= {
458 .rpc_call_prepare
= filelayout_commit_prepare
,
459 .rpc_call_done
= pnfs_generic_write_commit_done
,
460 .rpc_count_stats
= filelayout_commit_count_stats
,
461 .rpc_release
= pnfs_generic_commit_release
,
464 static enum pnfs_try_status
465 filelayout_read_pagelist(struct nfs_pgio_header
*hdr
)
467 struct pnfs_layout_segment
*lseg
= hdr
->lseg
;
468 struct nfs4_pnfs_ds
*ds
;
469 struct rpc_clnt
*ds_clnt
;
470 loff_t offset
= hdr
->args
.offset
;
474 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
475 __func__
, hdr
->inode
->i_ino
,
476 hdr
->args
.pgbase
, (size_t)hdr
->args
.count
, offset
);
478 /* Retrieve the correct rpc_client for the byte range */
479 j
= nfs4_fl_calc_j_index(lseg
, offset
);
480 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
481 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
483 return PNFS_NOT_ATTEMPTED
;
485 ds_clnt
= nfs4_find_or_create_ds_client(ds
->ds_clp
, hdr
->inode
);
487 return PNFS_NOT_ATTEMPTED
;
489 dprintk("%s USE DS: %s cl_count %d\n", __func__
,
490 ds
->ds_remotestr
, atomic_read(&ds
->ds_clp
->cl_count
));
492 /* No multipath support. Use first DS */
493 atomic_inc(&ds
->ds_clp
->cl_count
);
494 hdr
->ds_clp
= ds
->ds_clp
;
495 hdr
->ds_commit_idx
= idx
;
496 fh
= nfs4_fl_select_ds_fh(lseg
, j
);
500 hdr
->args
.offset
= filelayout_get_dserver_offset(lseg
, offset
);
501 hdr
->mds_offset
= offset
;
503 /* Perform an asynchronous read to ds */
504 nfs_initiate_pgio(ds_clnt
, hdr
, hdr
->cred
,
505 NFS_PROTO(hdr
->inode
), &filelayout_read_call_ops
,
506 0, RPC_TASK_SOFTCONN
);
507 return PNFS_ATTEMPTED
;
510 /* Perform async writes. */
511 static enum pnfs_try_status
512 filelayout_write_pagelist(struct nfs_pgio_header
*hdr
, int sync
)
514 struct pnfs_layout_segment
*lseg
= hdr
->lseg
;
515 struct nfs4_pnfs_ds
*ds
;
516 struct rpc_clnt
*ds_clnt
;
517 loff_t offset
= hdr
->args
.offset
;
521 /* Retrieve the correct rpc_client for the byte range */
522 j
= nfs4_fl_calc_j_index(lseg
, offset
);
523 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
524 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
526 return PNFS_NOT_ATTEMPTED
;
528 ds_clnt
= nfs4_find_or_create_ds_client(ds
->ds_clp
, hdr
->inode
);
530 return PNFS_NOT_ATTEMPTED
;
532 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n",
533 __func__
, hdr
->inode
->i_ino
, sync
, (size_t) hdr
->args
.count
,
534 offset
, ds
->ds_remotestr
, atomic_read(&ds
->ds_clp
->cl_count
));
536 hdr
->pgio_done_cb
= filelayout_write_done_cb
;
537 atomic_inc(&ds
->ds_clp
->cl_count
);
538 hdr
->ds_clp
= ds
->ds_clp
;
539 hdr
->ds_commit_idx
= idx
;
540 fh
= nfs4_fl_select_ds_fh(lseg
, j
);
543 hdr
->args
.offset
= filelayout_get_dserver_offset(lseg
, offset
);
545 /* Perform an asynchronous write */
546 nfs_initiate_pgio(ds_clnt
, hdr
, hdr
->cred
,
547 NFS_PROTO(hdr
->inode
), &filelayout_write_call_ops
,
548 sync
, RPC_TASK_SOFTCONN
);
549 return PNFS_ATTEMPTED
;
553 * filelayout_check_layout()
555 * Make sure layout segment parameters are sane WRT the device.
556 * At this point no generic layer initialization of the lseg has occurred,
557 * and nothing has been added to the layout_hdr cache.
561 filelayout_check_layout(struct pnfs_layout_hdr
*lo
,
562 struct nfs4_filelayout_segment
*fl
,
563 struct nfs4_layoutget_res
*lgr
,
564 struct nfs4_deviceid
*id
,
567 struct nfs4_deviceid_node
*d
;
568 struct nfs4_file_layout_dsaddr
*dsaddr
;
569 int status
= -EINVAL
;
571 dprintk("--> %s\n", __func__
);
573 /* FIXME: remove this check when layout segment support is added */
574 if (lgr
->range
.offset
!= 0 ||
575 lgr
->range
.length
!= NFS4_MAX_UINT64
) {
576 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
581 if (fl
->pattern_offset
> lgr
->range
.offset
) {
582 dprintk("%s pattern_offset %lld too large\n",
583 __func__
, fl
->pattern_offset
);
587 if (!fl
->stripe_unit
) {
588 dprintk("%s Invalid stripe unit (%u)\n",
589 __func__
, fl
->stripe_unit
);
593 /* find and reference the deviceid */
594 d
= nfs4_find_get_deviceid(NFS_SERVER(lo
->plh_inode
), id
,
595 lo
->plh_lc_cred
, gfp_flags
);
599 dsaddr
= container_of(d
, struct nfs4_file_layout_dsaddr
, id_node
);
600 /* Found deviceid is unavailable */
601 if (filelayout_test_devid_unavailable(&dsaddr
->id_node
))
606 if (fl
->first_stripe_index
>= dsaddr
->stripe_count
) {
607 dprintk("%s Bad first_stripe_index %u\n",
608 __func__
, fl
->first_stripe_index
);
612 if ((fl
->stripe_type
== STRIPE_SPARSE
&&
613 fl
->num_fh
> 1 && fl
->num_fh
!= dsaddr
->ds_num
) ||
614 (fl
->stripe_type
== STRIPE_DENSE
&&
615 fl
->num_fh
!= dsaddr
->stripe_count
)) {
616 dprintk("%s num_fh %u not valid for given packing\n",
617 __func__
, fl
->num_fh
);
623 dprintk("--> %s returns %d\n", __func__
, status
);
626 nfs4_fl_put_deviceid(dsaddr
);
630 static void filelayout_free_fh_array(struct nfs4_filelayout_segment
*fl
)
634 for (i
= 0; i
< fl
->num_fh
; i
++) {
635 if (!fl
->fh_array
[i
])
637 kfree(fl
->fh_array
[i
]);
644 _filelayout_free_lseg(struct nfs4_filelayout_segment
*fl
)
646 filelayout_free_fh_array(fl
);
651 filelayout_decode_layout(struct pnfs_layout_hdr
*flo
,
652 struct nfs4_filelayout_segment
*fl
,
653 struct nfs4_layoutget_res
*lgr
,
654 struct nfs4_deviceid
*id
,
657 struct xdr_stream stream
;
659 struct page
*scratch
;
664 dprintk("%s: set_layout_map Begin\n", __func__
);
666 scratch
= alloc_page(gfp_flags
);
670 xdr_init_decode_pages(&stream
, &buf
, lgr
->layoutp
->pages
, lgr
->layoutp
->len
);
671 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
673 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
675 p
= xdr_inline_decode(&stream
, NFS4_DEVICEID4_SIZE
+ 20);
679 memcpy(id
, p
, sizeof(*id
));
680 p
+= XDR_QUADLEN(NFS4_DEVICEID4_SIZE
);
681 nfs4_print_deviceid(id
);
683 nfl_util
= be32_to_cpup(p
++);
684 if (nfl_util
& NFL4_UFLG_COMMIT_THRU_MDS
)
685 fl
->commit_through_mds
= 1;
686 if (nfl_util
& NFL4_UFLG_DENSE
)
687 fl
->stripe_type
= STRIPE_DENSE
;
689 fl
->stripe_type
= STRIPE_SPARSE
;
690 fl
->stripe_unit
= nfl_util
& ~NFL4_UFLG_MASK
;
692 fl
->first_stripe_index
= be32_to_cpup(p
++);
693 p
= xdr_decode_hyper(p
, &fl
->pattern_offset
);
694 fl
->num_fh
= be32_to_cpup(p
++);
696 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
697 __func__
, nfl_util
, fl
->num_fh
, fl
->first_stripe_index
,
700 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
701 * Futher checking is done in filelayout_check_layout */
703 max(NFS4_PNFS_MAX_STRIPE_CNT
, NFS4_PNFS_MAX_MULTI_CNT
))
706 if (fl
->num_fh
> 0) {
707 fl
->fh_array
= kcalloc(fl
->num_fh
, sizeof(fl
->fh_array
[0]),
713 for (i
= 0; i
< fl
->num_fh
; i
++) {
714 /* Do we want to use a mempool here? */
715 fl
->fh_array
[i
] = kmalloc(sizeof(struct nfs_fh
), gfp_flags
);
716 if (!fl
->fh_array
[i
])
719 p
= xdr_inline_decode(&stream
, 4);
722 fl
->fh_array
[i
]->size
= be32_to_cpup(p
++);
723 if (sizeof(struct nfs_fh
) < fl
->fh_array
[i
]->size
) {
724 printk(KERN_ERR
"NFS: Too big fh %d received %d\n",
725 i
, fl
->fh_array
[i
]->size
);
729 p
= xdr_inline_decode(&stream
, fl
->fh_array
[i
]->size
);
732 memcpy(fl
->fh_array
[i
]->data
, p
, fl
->fh_array
[i
]->size
);
733 dprintk("DEBUG: %s: fh len %d\n", __func__
,
734 fl
->fh_array
[i
]->size
);
737 __free_page(scratch
);
741 filelayout_free_fh_array(fl
);
743 __free_page(scratch
);
748 filelayout_free_lseg(struct pnfs_layout_segment
*lseg
)
750 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
752 dprintk("--> %s\n", __func__
);
753 nfs4_fl_put_deviceid(fl
->dsaddr
);
754 /* This assumes a single RW lseg */
755 if (lseg
->pls_range
.iomode
== IOMODE_RW
) {
756 struct nfs4_filelayout
*flo
;
758 flo
= FILELAYOUT_FROM_HDR(lseg
->pls_layout
);
759 flo
->commit_info
.nbuckets
= 0;
760 kfree(flo
->commit_info
.buckets
);
761 flo
->commit_info
.buckets
= NULL
;
763 _filelayout_free_lseg(fl
);
767 filelayout_alloc_commit_info(struct pnfs_layout_segment
*lseg
,
768 struct nfs_commit_info
*cinfo
,
771 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
772 struct pnfs_commit_bucket
*buckets
;
775 if (fl
->commit_through_mds
)
778 size
= (fl
->stripe_type
== STRIPE_SPARSE
) ?
779 fl
->dsaddr
->ds_num
: fl
->dsaddr
->stripe_count
;
781 if (cinfo
->ds
->nbuckets
>= size
) {
782 /* This assumes there is only one IOMODE_RW lseg. What
783 * we really want to do is have a layout_hdr level
784 * dictionary of <multipath_list4, fh> keys, each
785 * associated with a struct list_head, populated by calls
786 * to filelayout_write_pagelist().
791 buckets
= kcalloc(size
, sizeof(struct pnfs_commit_bucket
),
795 for (i
= 0; i
< size
; i
++) {
796 INIT_LIST_HEAD(&buckets
[i
].written
);
797 INIT_LIST_HEAD(&buckets
[i
].committing
);
798 /* mark direct verifier as unset */
799 buckets
[i
].direct_verf
.committed
= NFS_INVALID_STABLE_HOW
;
802 spin_lock(cinfo
->lock
);
803 if (cinfo
->ds
->nbuckets
>= size
)
805 for (i
= 0; i
< cinfo
->ds
->nbuckets
; i
++) {
806 list_splice(&cinfo
->ds
->buckets
[i
].written
,
807 &buckets
[i
].written
);
808 list_splice(&cinfo
->ds
->buckets
[i
].committing
,
809 &buckets
[i
].committing
);
810 buckets
[i
].direct_verf
.committed
=
811 cinfo
->ds
->buckets
[i
].direct_verf
.committed
;
812 buckets
[i
].wlseg
= cinfo
->ds
->buckets
[i
].wlseg
;
813 buckets
[i
].clseg
= cinfo
->ds
->buckets
[i
].clseg
;
815 swap(cinfo
->ds
->buckets
, buckets
);
816 cinfo
->ds
->nbuckets
= size
;
818 spin_unlock(cinfo
->lock
);
823 static struct pnfs_layout_segment
*
824 filelayout_alloc_lseg(struct pnfs_layout_hdr
*layoutid
,
825 struct nfs4_layoutget_res
*lgr
,
828 struct nfs4_filelayout_segment
*fl
;
830 struct nfs4_deviceid id
;
832 dprintk("--> %s\n", __func__
);
833 fl
= kzalloc(sizeof(*fl
), gfp_flags
);
837 rc
= filelayout_decode_layout(layoutid
, fl
, lgr
, &id
, gfp_flags
);
838 if (rc
!= 0 || filelayout_check_layout(layoutid
, fl
, lgr
, &id
, gfp_flags
)) {
839 _filelayout_free_lseg(fl
);
842 return &fl
->generic_hdr
;
846 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
848 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
849 * of bytes (maximum @req->wb_bytes) that can be coalesced.
852 filelayout_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
853 struct nfs_page
*req
)
856 u64 p_stripe
, r_stripe
;
858 u64 segment_offset
= pgio
->pg_lseg
->pls_range
.offset
;
859 u32 stripe_unit
= FILELAYOUT_LSEG(pgio
->pg_lseg
)->stripe_unit
;
861 /* calls nfs_generic_pg_test */
862 size
= pnfs_generic_pg_test(pgio
, prev
, req
);
866 /* see if req and prev are in the same stripe */
868 p_stripe
= (u64
)req_offset(prev
) - segment_offset
;
869 r_stripe
= (u64
)req_offset(req
) - segment_offset
;
870 do_div(p_stripe
, stripe_unit
);
871 do_div(r_stripe
, stripe_unit
);
873 if (p_stripe
!= r_stripe
)
877 /* calculate remaining bytes in the current stripe */
878 div_u64_rem((u64
)req_offset(req
) - segment_offset
,
881 WARN_ON_ONCE(stripe_offset
> stripe_unit
);
882 if (stripe_offset
>= stripe_unit
)
884 return min(stripe_unit
- (unsigned int)stripe_offset
, size
);
888 filelayout_pg_init_read(struct nfs_pageio_descriptor
*pgio
,
889 struct nfs_page
*req
)
892 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
898 /* If no lseg, fall back to read through mds */
899 if (pgio
->pg_lseg
== NULL
)
900 nfs_pageio_reset_read_mds(pgio
);
904 filelayout_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
905 struct nfs_page
*req
)
907 struct nfs_commit_info cinfo
;
911 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
917 /* If no lseg, fall back to write through mds */
918 if (pgio
->pg_lseg
== NULL
)
920 nfs_init_cinfo(&cinfo
, pgio
->pg_inode
, pgio
->pg_dreq
);
921 status
= filelayout_alloc_commit_info(pgio
->pg_lseg
, &cinfo
, GFP_NOFS
);
923 pnfs_put_lseg(pgio
->pg_lseg
);
924 pgio
->pg_lseg
= NULL
;
929 nfs_pageio_reset_write_mds(pgio
);
932 static const struct nfs_pageio_ops filelayout_pg_read_ops
= {
933 .pg_init
= filelayout_pg_init_read
,
934 .pg_test
= filelayout_pg_test
,
935 .pg_doio
= pnfs_generic_pg_readpages
,
936 .pg_cleanup
= pnfs_generic_pg_cleanup
,
939 static const struct nfs_pageio_ops filelayout_pg_write_ops
= {
940 .pg_init
= filelayout_pg_init_write
,
941 .pg_test
= filelayout_pg_test
,
942 .pg_doio
= pnfs_generic_pg_writepages
,
943 .pg_cleanup
= pnfs_generic_pg_cleanup
,
946 static u32
select_bucket_index(struct nfs4_filelayout_segment
*fl
, u32 j
)
948 if (fl
->stripe_type
== STRIPE_SPARSE
)
949 return nfs4_fl_calc_ds_index(&fl
->generic_hdr
, j
);
955 filelayout_mark_request_commit(struct nfs_page
*req
,
956 struct pnfs_layout_segment
*lseg
,
957 struct nfs_commit_info
*cinfo
,
961 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
963 struct list_head
*list
;
964 struct pnfs_commit_bucket
*buckets
;
966 if (fl
->commit_through_mds
) {
967 list
= &cinfo
->mds
->list
;
968 spin_lock(cinfo
->lock
);
972 /* Note that we are calling nfs4_fl_calc_j_index on each page
973 * that ends up being committed to a data server. An attractive
974 * alternative is to add a field to nfs_write_data and nfs_page
975 * to store the value calculated in filelayout_write_pagelist
976 * and just use that here.
978 j
= nfs4_fl_calc_j_index(lseg
, req_offset(req
));
979 i
= select_bucket_index(fl
, j
);
980 spin_lock(cinfo
->lock
);
981 buckets
= cinfo
->ds
->buckets
;
982 list
= &buckets
[i
].written
;
983 if (list_empty(list
)) {
984 /* Non-empty buckets hold a reference on the lseg. That ref
985 * is normally transferred to the COMMIT call and released
986 * there. It could also be released if the last req is pulled
987 * off due to a rewrite, in which case it will be done in
988 * pnfs_generic_clear_request_commit
990 buckets
[i
].wlseg
= pnfs_get_lseg(lseg
);
992 set_bit(PG_COMMIT_TO_DS
, &req
->wb_flags
);
993 cinfo
->ds
->nwritten
++;
996 /* nfs_request_add_commit_list(). We need to add req to list without
997 * dropping cinfo lock.
999 set_bit(PG_CLEAN
, &(req
)->wb_flags
);
1000 nfs_list_add_request(req
, list
);
1001 cinfo
->mds
->ncommit
++;
1002 spin_unlock(cinfo
->lock
);
1004 inc_zone_page_state(req
->wb_page
, NR_UNSTABLE_NFS
);
1005 inc_bdi_stat(inode_to_bdi(page_file_mapping(req
->wb_page
)->host
),
1007 __mark_inode_dirty(req
->wb_context
->dentry
->d_inode
,
1012 static u32
calc_ds_index_from_commit(struct pnfs_layout_segment
*lseg
, u32 i
)
1014 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
1016 if (flseg
->stripe_type
== STRIPE_SPARSE
)
1019 return nfs4_fl_calc_ds_index(lseg
, i
);
1022 static struct nfs_fh
*
1023 select_ds_fh_from_commit(struct pnfs_layout_segment
*lseg
, u32 i
)
1025 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
1027 if (flseg
->stripe_type
== STRIPE_SPARSE
) {
1028 if (flseg
->num_fh
== 1)
1030 else if (flseg
->num_fh
== 0)
1031 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1034 return flseg
->fh_array
[i
];
1037 static int filelayout_initiate_commit(struct nfs_commit_data
*data
, int how
)
1039 struct pnfs_layout_segment
*lseg
= data
->lseg
;
1040 struct nfs4_pnfs_ds
*ds
;
1041 struct rpc_clnt
*ds_clnt
;
1045 idx
= calc_ds_index_from_commit(lseg
, data
->ds_commit_index
);
1046 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
1050 ds_clnt
= nfs4_find_or_create_ds_client(ds
->ds_clp
, data
->inode
);
1051 if (IS_ERR(ds_clnt
))
1054 dprintk("%s ino %lu, how %d cl_count %d\n", __func__
,
1055 data
->inode
->i_ino
, how
, atomic_read(&ds
->ds_clp
->cl_count
));
1056 data
->commit_done_cb
= filelayout_commit_done_cb
;
1057 atomic_inc(&ds
->ds_clp
->cl_count
);
1058 data
->ds_clp
= ds
->ds_clp
;
1059 fh
= select_ds_fh_from_commit(lseg
, data
->ds_commit_index
);
1062 return nfs_initiate_commit(ds_clnt
, data
, NFS_PROTO(data
->inode
),
1063 &filelayout_commit_call_ops
, how
,
1066 pnfs_generic_prepare_to_resend_writes(data
);
1067 pnfs_generic_commit_release(data
);
1071 /* filelayout_search_commit_reqs - Search lists in @cinfo for the head reqest
1073 * @cinfo - commit info for current inode
1074 * @page - page to search for matching head request
1076 * Returns a the head request if one is found, otherwise returns NULL.
1078 static struct nfs_page
*
1079 filelayout_search_commit_reqs(struct nfs_commit_info
*cinfo
, struct page
*page
)
1081 struct nfs_page
*freq
, *t
;
1082 struct pnfs_commit_bucket
*b
;
1085 /* Linearly search the commit lists for each bucket until a matching
1086 * request is found */
1087 for (i
= 0, b
= cinfo
->ds
->buckets
; i
< cinfo
->ds
->nbuckets
; i
++, b
++) {
1088 list_for_each_entry_safe(freq
, t
, &b
->written
, wb_list
) {
1089 if (freq
->wb_page
== page
)
1090 return freq
->wb_head
;
1092 list_for_each_entry_safe(freq
, t
, &b
->committing
, wb_list
) {
1093 if (freq
->wb_page
== page
)
1094 return freq
->wb_head
;
1102 filelayout_commit_pagelist(struct inode
*inode
, struct list_head
*mds_pages
,
1103 int how
, struct nfs_commit_info
*cinfo
)
1105 return pnfs_generic_commit_pagelist(inode
, mds_pages
, how
, cinfo
,
1106 filelayout_initiate_commit
);
1109 static struct nfs4_deviceid_node
*
1110 filelayout_alloc_deviceid_node(struct nfs_server
*server
,
1111 struct pnfs_device
*pdev
, gfp_t gfp_flags
)
1113 struct nfs4_file_layout_dsaddr
*dsaddr
;
1115 dsaddr
= nfs4_fl_alloc_deviceid_node(server
, pdev
, gfp_flags
);
1118 return &dsaddr
->id_node
;
1122 filelayout_free_deveiceid_node(struct nfs4_deviceid_node
*d
)
1124 nfs4_fl_free_deviceid(container_of(d
, struct nfs4_file_layout_dsaddr
, id_node
));
1127 static struct pnfs_layout_hdr
*
1128 filelayout_alloc_layout_hdr(struct inode
*inode
, gfp_t gfp_flags
)
1130 struct nfs4_filelayout
*flo
;
1132 flo
= kzalloc(sizeof(*flo
), gfp_flags
);
1133 return flo
!= NULL
? &flo
->generic_hdr
: NULL
;
1137 filelayout_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
1139 kfree(FILELAYOUT_FROM_HDR(lo
));
1142 static struct pnfs_ds_commit_info
*
1143 filelayout_get_ds_info(struct inode
*inode
)
1145 struct pnfs_layout_hdr
*layout
= NFS_I(inode
)->layout
;
1150 return &FILELAYOUT_FROM_HDR(layout
)->commit_info
;
1153 static struct pnfs_layoutdriver_type filelayout_type
= {
1154 .id
= LAYOUT_NFSV4_1_FILES
,
1155 .name
= "LAYOUT_NFSV4_1_FILES",
1156 .owner
= THIS_MODULE
,
1157 .alloc_layout_hdr
= filelayout_alloc_layout_hdr
,
1158 .free_layout_hdr
= filelayout_free_layout_hdr
,
1159 .alloc_lseg
= filelayout_alloc_lseg
,
1160 .free_lseg
= filelayout_free_lseg
,
1161 .pg_read_ops
= &filelayout_pg_read_ops
,
1162 .pg_write_ops
= &filelayout_pg_write_ops
,
1163 .get_ds_info
= &filelayout_get_ds_info
,
1164 .mark_request_commit
= filelayout_mark_request_commit
,
1165 .clear_request_commit
= pnfs_generic_clear_request_commit
,
1166 .scan_commit_lists
= pnfs_generic_scan_commit_lists
,
1167 .recover_commit_reqs
= pnfs_generic_recover_commit_reqs
,
1168 .search_commit_reqs
= filelayout_search_commit_reqs
,
1169 .commit_pagelist
= filelayout_commit_pagelist
,
1170 .read_pagelist
= filelayout_read_pagelist
,
1171 .write_pagelist
= filelayout_write_pagelist
,
1172 .alloc_deviceid_node
= filelayout_alloc_deviceid_node
,
1173 .free_deviceid_node
= filelayout_free_deveiceid_node
,
1176 static int __init
nfs4filelayout_init(void)
1178 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Registering...\n",
1180 return pnfs_register_layoutdriver(&filelayout_type
);
1183 static void __exit
nfs4filelayout_exit(void)
1185 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Unregistering...\n",
1187 pnfs_unregister_layoutdriver(&filelayout_type
);
1190 MODULE_ALIAS("nfs-layouttype4-1");
1192 module_init(nfs4filelayout_init
);
1193 module_exit(nfs4filelayout_exit
);