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>
36 #include "nfs4filelayout.h"
38 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
40 MODULE_LICENSE("GPL");
41 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
42 MODULE_DESCRIPTION("The NFSv4 file layout driver");
44 #define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
47 filelayout_get_dense_offset(struct nfs4_filelayout_segment
*flseg
,
50 u32 stripe_width
= flseg
->stripe_unit
* flseg
->dsaddr
->stripe_count
;
53 offset
-= flseg
->pattern_offset
;
55 do_div(tmp
, stripe_width
);
57 return tmp
* flseg
->stripe_unit
+ do_div(offset
, flseg
->stripe_unit
);
60 /* This function is used by the layout driver to calculate the
61 * offset of the file on the dserver based on whether the
62 * layout type is STRIPE_DENSE or STRIPE_SPARSE
65 filelayout_get_dserver_offset(struct pnfs_layout_segment
*lseg
, loff_t offset
)
67 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
69 switch (flseg
->stripe_type
) {
74 return filelayout_get_dense_offset(flseg
, offset
);
80 /* For data server errors we don't recover from */
82 filelayout_set_lo_fail(struct pnfs_layout_segment
*lseg
)
84 if (lseg
->pls_range
.iomode
== IOMODE_RW
) {
85 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__
);
86 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
88 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__
);
89 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
93 static int filelayout_async_handle_error(struct rpc_task
*task
,
94 struct nfs4_state
*state
,
95 struct nfs_client
*clp
,
98 if (task
->tk_status
>= 0)
103 switch (task
->tk_status
) {
104 case -NFS4ERR_BADSESSION
:
105 case -NFS4ERR_BADSLOT
:
106 case -NFS4ERR_BAD_HIGH_SLOT
:
107 case -NFS4ERR_DEADSESSION
:
108 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION
:
109 case -NFS4ERR_SEQ_FALSE_RETRY
:
110 case -NFS4ERR_SEQ_MISORDERED
:
111 dprintk("%s ERROR %d, Reset session. Exchangeid "
112 "flags 0x%x\n", __func__
, task
->tk_status
,
113 clp
->cl_exchange_flags
);
114 nfs4_schedule_session_recovery(clp
->cl_session
);
119 rpc_delay(task
, FILELAYOUT_POLL_RETRY_MAX
);
121 case -NFS4ERR_RETRY_UNCACHED_REP
:
124 dprintk("%s DS error. Retry through MDS %d\n", __func__
,
133 /* NFS_PROTO call done callback routines */
135 static int filelayout_read_done_cb(struct rpc_task
*task
,
136 struct nfs_read_data
*data
)
138 struct nfs_client
*clp
= data
->ds_clp
;
141 dprintk("%s DS read\n", __func__
);
143 if (filelayout_async_handle_error(task
, data
->args
.context
->state
,
144 data
->ds_clp
, &reset
) == -EAGAIN
) {
145 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
146 __func__
, data
->ds_clp
, data
->ds_clp
->cl_session
);
148 filelayout_set_lo_fail(data
->lseg
);
149 nfs4_reset_read(task
, data
);
150 clp
= NFS_SERVER(data
->inode
)->nfs_client
;
152 nfs_restart_rpc(task
, clp
);
160 * We reference the rpc_cred of the first WRITE that triggers the need for
161 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
162 * rfc5661 is not clear about which credential should be used.
165 filelayout_set_layoutcommit(struct nfs_write_data
*wdata
)
167 if (FILELAYOUT_LSEG(wdata
->lseg
)->commit_through_mds
||
168 wdata
->res
.verf
->committed
== NFS_FILE_SYNC
)
171 pnfs_set_layoutcommit(wdata
);
172 dprintk("%s ionde %lu pls_end_pos %lu\n", __func__
, wdata
->inode
->i_ino
,
173 (unsigned long) wdata
->lseg
->pls_end_pos
);
177 * Call ops for the async read/write cases
178 * In the case of dense layouts, the offset needs to be reset to its
181 static void filelayout_read_prepare(struct rpc_task
*task
, void *data
)
183 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
185 rdata
->read_done_cb
= filelayout_read_done_cb
;
187 if (nfs41_setup_sequence(rdata
->ds_clp
->cl_session
,
188 &rdata
->args
.seq_args
, &rdata
->res
.seq_res
,
192 rpc_call_start(task
);
195 static void filelayout_read_call_done(struct rpc_task
*task
, void *data
)
197 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
199 dprintk("--> %s task->tk_status %d\n", __func__
, task
->tk_status
);
201 /* Note this may cause RPC to be resent */
202 rdata
->mds_ops
->rpc_call_done(task
, data
);
205 static void filelayout_read_release(void *data
)
207 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
209 rdata
->mds_ops
->rpc_release(data
);
212 static int filelayout_write_done_cb(struct rpc_task
*task
,
213 struct nfs_write_data
*data
)
217 if (filelayout_async_handle_error(task
, data
->args
.context
->state
,
218 data
->ds_clp
, &reset
) == -EAGAIN
) {
219 struct nfs_client
*clp
;
221 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
222 __func__
, data
->ds_clp
, data
->ds_clp
->cl_session
);
224 filelayout_set_lo_fail(data
->lseg
);
225 nfs4_reset_write(task
, data
);
226 clp
= NFS_SERVER(data
->inode
)->nfs_client
;
229 nfs_restart_rpc(task
, clp
);
233 filelayout_set_layoutcommit(data
);
237 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
238 static void prepare_to_resend_writes(struct nfs_write_data
*data
)
240 struct nfs_page
*first
= nfs_list_entry(data
->pages
.next
);
242 data
->task
.tk_status
= 0;
243 memcpy(data
->verf
.verifier
, first
->wb_verf
.verifier
,
244 sizeof(first
->wb_verf
.verifier
));
245 data
->verf
.verifier
[0]++; /* ensure verifier mismatch */
248 static int filelayout_commit_done_cb(struct rpc_task
*task
,
249 struct nfs_write_data
*data
)
253 if (filelayout_async_handle_error(task
, data
->args
.context
->state
,
254 data
->ds_clp
, &reset
) == -EAGAIN
) {
255 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
256 __func__
, data
->ds_clp
, data
->ds_clp
->cl_session
);
258 prepare_to_resend_writes(data
);
259 filelayout_set_lo_fail(data
->lseg
);
261 nfs_restart_rpc(task
, data
->ds_clp
);
268 static void filelayout_write_prepare(struct rpc_task
*task
, void *data
)
270 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
272 if (nfs41_setup_sequence(wdata
->ds_clp
->cl_session
,
273 &wdata
->args
.seq_args
, &wdata
->res
.seq_res
,
277 rpc_call_start(task
);
280 static void filelayout_write_call_done(struct rpc_task
*task
, void *data
)
282 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
284 /* Note this may cause RPC to be resent */
285 wdata
->mds_ops
->rpc_call_done(task
, data
);
288 static void filelayout_write_release(void *data
)
290 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
292 wdata
->mds_ops
->rpc_release(data
);
295 static void filelayout_commit_release(void *data
)
297 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
299 nfs_commit_release_pages(wdata
);
300 if (atomic_dec_and_test(&NFS_I(wdata
->inode
)->commits_outstanding
))
301 nfs_commit_clear_lock(NFS_I(wdata
->inode
));
302 nfs_commitdata_release(wdata
);
305 struct rpc_call_ops filelayout_read_call_ops
= {
306 .rpc_call_prepare
= filelayout_read_prepare
,
307 .rpc_call_done
= filelayout_read_call_done
,
308 .rpc_release
= filelayout_read_release
,
311 struct rpc_call_ops filelayout_write_call_ops
= {
312 .rpc_call_prepare
= filelayout_write_prepare
,
313 .rpc_call_done
= filelayout_write_call_done
,
314 .rpc_release
= filelayout_write_release
,
317 struct rpc_call_ops filelayout_commit_call_ops
= {
318 .rpc_call_prepare
= filelayout_write_prepare
,
319 .rpc_call_done
= filelayout_write_call_done
,
320 .rpc_release
= filelayout_commit_release
,
323 static enum pnfs_try_status
324 filelayout_read_pagelist(struct nfs_read_data
*data
)
326 struct pnfs_layout_segment
*lseg
= data
->lseg
;
327 struct nfs4_pnfs_ds
*ds
;
328 loff_t offset
= data
->args
.offset
;
333 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
334 __func__
, data
->inode
->i_ino
,
335 data
->args
.pgbase
, (size_t)data
->args
.count
, offset
);
337 /* Retrieve the correct rpc_client for the byte range */
338 j
= nfs4_fl_calc_j_index(lseg
, offset
);
339 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
340 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
342 /* Either layout fh index faulty, or ds connect failed */
343 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
344 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
345 return PNFS_NOT_ATTEMPTED
;
347 dprintk("%s USE DS: %s\n", __func__
, ds
->ds_remotestr
);
349 /* No multipath support. Use first DS */
350 data
->ds_clp
= ds
->ds_clp
;
351 fh
= nfs4_fl_select_ds_fh(lseg
, j
);
355 data
->args
.offset
= filelayout_get_dserver_offset(lseg
, offset
);
356 data
->mds_offset
= offset
;
358 /* Perform an asynchronous read to ds */
359 status
= nfs_initiate_read(data
, ds
->ds_clp
->cl_rpcclient
,
360 &filelayout_read_call_ops
);
362 return PNFS_ATTEMPTED
;
365 /* Perform async writes. */
366 static enum pnfs_try_status
367 filelayout_write_pagelist(struct nfs_write_data
*data
, int sync
)
369 struct pnfs_layout_segment
*lseg
= data
->lseg
;
370 struct nfs4_pnfs_ds
*ds
;
371 loff_t offset
= data
->args
.offset
;
376 /* Retrieve the correct rpc_client for the byte range */
377 j
= nfs4_fl_calc_j_index(lseg
, offset
);
378 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
379 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
381 printk(KERN_ERR
"%s: prepare_ds failed, use MDS\n", __func__
);
382 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
383 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
384 return PNFS_NOT_ATTEMPTED
;
386 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s\n", __func__
,
387 data
->inode
->i_ino
, sync
, (size_t) data
->args
.count
, offset
,
390 data
->write_done_cb
= filelayout_write_done_cb
;
391 data
->ds_clp
= ds
->ds_clp
;
392 fh
= nfs4_fl_select_ds_fh(lseg
, j
);
396 * Get the file offset on the dserver. Set the write offset to
397 * this offset and save the original offset.
399 data
->args
.offset
= filelayout_get_dserver_offset(lseg
, offset
);
401 /* Perform an asynchronous write */
402 status
= nfs_initiate_write(data
, ds
->ds_clp
->cl_rpcclient
,
403 &filelayout_write_call_ops
, sync
);
405 return PNFS_ATTEMPTED
;
409 * filelayout_check_layout()
411 * Make sure layout segment parameters are sane WRT the device.
412 * At this point no generic layer initialization of the lseg has occurred,
413 * and nothing has been added to the layout_hdr cache.
417 filelayout_check_layout(struct pnfs_layout_hdr
*lo
,
418 struct nfs4_filelayout_segment
*fl
,
419 struct nfs4_layoutget_res
*lgr
,
420 struct nfs4_deviceid
*id
,
423 struct nfs4_deviceid_node
*d
;
424 struct nfs4_file_layout_dsaddr
*dsaddr
;
425 int status
= -EINVAL
;
426 struct nfs_server
*nfss
= NFS_SERVER(lo
->plh_inode
);
428 dprintk("--> %s\n", __func__
);
430 /* FIXME: remove this check when layout segment support is added */
431 if (lgr
->range
.offset
!= 0 ||
432 lgr
->range
.length
!= NFS4_MAX_UINT64
) {
433 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
438 if (fl
->pattern_offset
> lgr
->range
.offset
) {
439 dprintk("%s pattern_offset %lld too large\n",
440 __func__
, fl
->pattern_offset
);
444 if (!fl
->stripe_unit
|| fl
->stripe_unit
% PAGE_SIZE
) {
445 dprintk("%s Invalid stripe unit (%u)\n",
446 __func__
, fl
->stripe_unit
);
450 /* find and reference the deviceid */
451 d
= nfs4_find_get_deviceid(NFS_SERVER(lo
->plh_inode
)->pnfs_curr_ld
,
452 NFS_SERVER(lo
->plh_inode
)->nfs_client
, id
);
454 dsaddr
= get_device_info(lo
->plh_inode
, id
, gfp_flags
);
458 dsaddr
= container_of(d
, struct nfs4_file_layout_dsaddr
, id_node
);
461 if (fl
->first_stripe_index
< 0 ||
462 fl
->first_stripe_index
>= dsaddr
->stripe_count
) {
463 dprintk("%s Bad first_stripe_index %d\n",
464 __func__
, fl
->first_stripe_index
);
468 if ((fl
->stripe_type
== STRIPE_SPARSE
&&
469 fl
->num_fh
> 1 && fl
->num_fh
!= dsaddr
->ds_num
) ||
470 (fl
->stripe_type
== STRIPE_DENSE
&&
471 fl
->num_fh
!= dsaddr
->stripe_count
)) {
472 dprintk("%s num_fh %u not valid for given packing\n",
473 __func__
, fl
->num_fh
);
477 if (fl
->stripe_unit
% nfss
->rsize
|| fl
->stripe_unit
% nfss
->wsize
) {
478 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
479 "wsize %u\n", __func__
, fl
->stripe_unit
, nfss
->rsize
,
485 dprintk("--> %s returns %d\n", __func__
, status
);
488 nfs4_fl_put_deviceid(dsaddr
);
492 static void filelayout_free_fh_array(struct nfs4_filelayout_segment
*fl
)
496 for (i
= 0; i
< fl
->num_fh
; i
++) {
497 if (!fl
->fh_array
[i
])
499 kfree(fl
->fh_array
[i
]);
506 _filelayout_free_lseg(struct nfs4_filelayout_segment
*fl
)
508 filelayout_free_fh_array(fl
);
513 filelayout_decode_layout(struct pnfs_layout_hdr
*flo
,
514 struct nfs4_filelayout_segment
*fl
,
515 struct nfs4_layoutget_res
*lgr
,
516 struct nfs4_deviceid
*id
,
519 struct xdr_stream stream
;
521 struct page
*scratch
;
526 dprintk("%s: set_layout_map Begin\n", __func__
);
528 scratch
= alloc_page(gfp_flags
);
532 xdr_init_decode_pages(&stream
, &buf
, lgr
->layoutp
->pages
, lgr
->layoutp
->len
);
533 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
535 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
537 p
= xdr_inline_decode(&stream
, NFS4_DEVICEID4_SIZE
+ 20);
541 memcpy(id
, p
, sizeof(*id
));
542 p
+= XDR_QUADLEN(NFS4_DEVICEID4_SIZE
);
543 nfs4_print_deviceid(id
);
545 nfl_util
= be32_to_cpup(p
++);
546 if (nfl_util
& NFL4_UFLG_COMMIT_THRU_MDS
)
547 fl
->commit_through_mds
= 1;
548 if (nfl_util
& NFL4_UFLG_DENSE
)
549 fl
->stripe_type
= STRIPE_DENSE
;
551 fl
->stripe_type
= STRIPE_SPARSE
;
552 fl
->stripe_unit
= nfl_util
& ~NFL4_UFLG_MASK
;
554 fl
->first_stripe_index
= be32_to_cpup(p
++);
555 p
= xdr_decode_hyper(p
, &fl
->pattern_offset
);
556 fl
->num_fh
= be32_to_cpup(p
++);
558 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
559 __func__
, nfl_util
, fl
->num_fh
, fl
->first_stripe_index
,
562 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
563 * Futher checking is done in filelayout_check_layout */
564 if (fl
->num_fh
< 0 || fl
->num_fh
>
565 max(NFS4_PNFS_MAX_STRIPE_CNT
, NFS4_PNFS_MAX_MULTI_CNT
))
568 if (fl
->num_fh
> 0) {
569 fl
->fh_array
= kzalloc(fl
->num_fh
* sizeof(struct nfs_fh
*),
575 for (i
= 0; i
< fl
->num_fh
; i
++) {
576 /* Do we want to use a mempool here? */
577 fl
->fh_array
[i
] = kmalloc(sizeof(struct nfs_fh
), gfp_flags
);
578 if (!fl
->fh_array
[i
])
581 p
= xdr_inline_decode(&stream
, 4);
584 fl
->fh_array
[i
]->size
= be32_to_cpup(p
++);
585 if (sizeof(struct nfs_fh
) < fl
->fh_array
[i
]->size
) {
586 printk(KERN_ERR
"Too big fh %d received %d\n",
587 i
, fl
->fh_array
[i
]->size
);
591 p
= xdr_inline_decode(&stream
, fl
->fh_array
[i
]->size
);
594 memcpy(fl
->fh_array
[i
]->data
, p
, fl
->fh_array
[i
]->size
);
595 dprintk("DEBUG: %s: fh len %d\n", __func__
,
596 fl
->fh_array
[i
]->size
);
599 __free_page(scratch
);
603 filelayout_free_fh_array(fl
);
605 __free_page(scratch
);
610 filelayout_free_lseg(struct pnfs_layout_segment
*lseg
)
612 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
614 dprintk("--> %s\n", __func__
);
615 nfs4_fl_put_deviceid(fl
->dsaddr
);
616 kfree(fl
->commit_buckets
);
617 _filelayout_free_lseg(fl
);
620 static struct pnfs_layout_segment
*
621 filelayout_alloc_lseg(struct pnfs_layout_hdr
*layoutid
,
622 struct nfs4_layoutget_res
*lgr
,
625 struct nfs4_filelayout_segment
*fl
;
627 struct nfs4_deviceid id
;
629 dprintk("--> %s\n", __func__
);
630 fl
= kzalloc(sizeof(*fl
), gfp_flags
);
634 rc
= filelayout_decode_layout(layoutid
, fl
, lgr
, &id
, gfp_flags
);
635 if (rc
!= 0 || filelayout_check_layout(layoutid
, fl
, lgr
, &id
, gfp_flags
)) {
636 _filelayout_free_lseg(fl
);
640 /* This assumes there is only one IOMODE_RW lseg. What
641 * we really want to do is have a layout_hdr level
642 * dictionary of <multipath_list4, fh> keys, each
643 * associated with a struct list_head, populated by calls
644 * to filelayout_write_pagelist().
646 if ((!fl
->commit_through_mds
) && (lgr
->range
.iomode
== IOMODE_RW
)) {
648 int size
= (fl
->stripe_type
== STRIPE_SPARSE
) ?
649 fl
->dsaddr
->ds_num
: fl
->dsaddr
->stripe_count
;
651 fl
->commit_buckets
= kcalloc(size
, sizeof(struct list_head
), gfp_flags
);
652 if (!fl
->commit_buckets
) {
653 filelayout_free_lseg(&fl
->generic_hdr
);
656 fl
->number_of_buckets
= size
;
657 for (i
= 0; i
< size
; i
++)
658 INIT_LIST_HEAD(&fl
->commit_buckets
[i
]);
660 return &fl
->generic_hdr
;
664 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
666 * return true : coalesce page
667 * return false : don't coalesce page
670 filelayout_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
671 struct nfs_page
*req
)
673 u64 p_stripe
, r_stripe
;
676 if (!pnfs_generic_pg_test(pgio
, prev
, req
) ||
677 !nfs_generic_pg_test(pgio
, prev
, req
))
680 p_stripe
= (u64
)prev
->wb_index
<< PAGE_CACHE_SHIFT
;
681 r_stripe
= (u64
)req
->wb_index
<< PAGE_CACHE_SHIFT
;
682 stripe_unit
= FILELAYOUT_LSEG(pgio
->pg_lseg
)->stripe_unit
;
684 do_div(p_stripe
, stripe_unit
);
685 do_div(r_stripe
, stripe_unit
);
687 return (p_stripe
== r_stripe
);
691 filelayout_pg_init_read(struct nfs_pageio_descriptor
*pgio
,
692 struct nfs_page
*req
)
694 BUG_ON(pgio
->pg_lseg
!= NULL
);
696 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
702 /* If no lseg, fall back to read through mds */
703 if (pgio
->pg_lseg
== NULL
)
704 nfs_pageio_init_read_mds(pgio
, pgio
->pg_inode
);
708 filelayout_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
709 struct nfs_page
*req
)
711 BUG_ON(pgio
->pg_lseg
!= NULL
);
713 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
719 /* If no lseg, fall back to write through mds */
720 if (pgio
->pg_lseg
== NULL
)
721 nfs_pageio_init_write_mds(pgio
, pgio
->pg_inode
,
725 static const struct nfs_pageio_ops filelayout_pg_read_ops
= {
726 .pg_init
= filelayout_pg_init_read
,
727 .pg_test
= filelayout_pg_test
,
728 .pg_doio
= nfs_generic_pg_readpages
,
731 static const struct nfs_pageio_ops filelayout_pg_write_ops
= {
732 .pg_init
= filelayout_pg_init_write
,
733 .pg_test
= filelayout_pg_test
,
734 .pg_doio
= nfs_generic_pg_writepages
,
737 static bool filelayout_mark_pnfs_commit(struct pnfs_layout_segment
*lseg
)
739 return !FILELAYOUT_LSEG(lseg
)->commit_through_mds
;
742 static u32
select_bucket_index(struct nfs4_filelayout_segment
*fl
, u32 j
)
744 if (fl
->stripe_type
== STRIPE_SPARSE
)
745 return nfs4_fl_calc_ds_index(&fl
->generic_hdr
, j
);
750 struct list_head
*filelayout_choose_commit_list(struct nfs_page
*req
)
752 struct pnfs_layout_segment
*lseg
= req
->wb_commit_lseg
;
753 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
755 struct list_head
*list
;
757 /* Note that we are calling nfs4_fl_calc_j_index on each page
758 * that ends up being committed to a data server. An attractive
759 * alternative is to add a field to nfs_write_data and nfs_page
760 * to store the value calculated in filelayout_write_pagelist
761 * and just use that here.
763 j
= nfs4_fl_calc_j_index(lseg
,
764 (loff_t
)req
->wb_index
<< PAGE_CACHE_SHIFT
);
765 i
= select_bucket_index(fl
, j
);
766 list
= &fl
->commit_buckets
[i
];
767 if (list_empty(list
)) {
768 /* Non-empty buckets hold a reference on the lseg */
774 static u32
calc_ds_index_from_commit(struct pnfs_layout_segment
*lseg
, u32 i
)
776 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
778 if (flseg
->stripe_type
== STRIPE_SPARSE
)
781 return nfs4_fl_calc_ds_index(lseg
, i
);
784 static struct nfs_fh
*
785 select_ds_fh_from_commit(struct pnfs_layout_segment
*lseg
, u32 i
)
787 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
789 if (flseg
->stripe_type
== STRIPE_SPARSE
) {
790 if (flseg
->num_fh
== 1)
792 else if (flseg
->num_fh
== 0)
793 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
796 return flseg
->fh_array
[i
];
799 static int filelayout_initiate_commit(struct nfs_write_data
*data
, int how
)
801 struct pnfs_layout_segment
*lseg
= data
->lseg
;
802 struct nfs4_pnfs_ds
*ds
;
806 idx
= calc_ds_index_from_commit(lseg
, data
->ds_commit_index
);
807 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
809 printk(KERN_ERR
"%s: prepare_ds failed, use MDS\n", __func__
);
810 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
811 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
812 prepare_to_resend_writes(data
);
813 data
->mds_ops
->rpc_release(data
);
816 dprintk("%s ino %lu, how %d\n", __func__
, data
->inode
->i_ino
, how
);
817 data
->write_done_cb
= filelayout_commit_done_cb
;
818 data
->ds_clp
= ds
->ds_clp
;
819 fh
= select_ds_fh_from_commit(lseg
, data
->ds_commit_index
);
822 return nfs_initiate_commit(data
, ds
->ds_clp
->cl_rpcclient
,
823 &filelayout_commit_call_ops
, how
);
827 * This is only useful while we are using whole file layouts.
829 static struct pnfs_layout_segment
*find_only_write_lseg(struct inode
*inode
)
831 struct pnfs_layout_segment
*lseg
, *rv
= NULL
;
833 spin_lock(&inode
->i_lock
);
834 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
)
835 if (lseg
->pls_range
.iomode
== IOMODE_RW
)
837 spin_unlock(&inode
->i_lock
);
841 static int alloc_ds_commits(struct inode
*inode
, struct list_head
*list
)
843 struct pnfs_layout_segment
*lseg
;
844 struct nfs4_filelayout_segment
*fl
;
845 struct nfs_write_data
*data
;
848 /* Won't need this when non-whole file layout segments are supported
849 * instead we will use a pnfs_layout_hdr structure */
850 lseg
= find_only_write_lseg(inode
);
853 fl
= FILELAYOUT_LSEG(lseg
);
854 for (i
= 0; i
< fl
->number_of_buckets
; i
++) {
855 if (list_empty(&fl
->commit_buckets
[i
]))
857 data
= nfs_commitdata_alloc();
860 data
->ds_commit_index
= i
;
862 list_add(&data
->pages
, list
);
868 for (j
= i
; j
< fl
->number_of_buckets
; j
++) {
869 if (list_empty(&fl
->commit_buckets
[i
]))
871 nfs_retry_commit(&fl
->commit_buckets
[i
], lseg
);
872 put_lseg(lseg
); /* associated with emptying bucket */
875 /* Caller will clean up entries put on list */
879 /* This follows nfs_commit_list pretty closely */
881 filelayout_commit_pagelist(struct inode
*inode
, struct list_head
*mds_pages
,
884 struct nfs_write_data
*data
, *tmp
;
887 if (!list_empty(mds_pages
)) {
888 data
= nfs_commitdata_alloc();
892 list_add(&data
->pages
, &list
);
895 if (alloc_ds_commits(inode
, &list
))
898 list_for_each_entry_safe(data
, tmp
, &list
, pages
) {
899 list_del_init(&data
->pages
);
900 atomic_inc(&NFS_I(inode
)->commits_outstanding
);
902 nfs_init_commit(data
, mds_pages
, NULL
);
903 nfs_initiate_commit(data
, NFS_CLIENT(inode
),
906 nfs_init_commit(data
, &FILELAYOUT_LSEG(data
->lseg
)->commit_buckets
[data
->ds_commit_index
], data
->lseg
);
907 filelayout_initiate_commit(data
, how
);
912 list_for_each_entry_safe(data
, tmp
, &list
, pages
) {
913 nfs_retry_commit(&data
->pages
, data
->lseg
);
914 list_del_init(&data
->pages
);
915 nfs_commit_free(data
);
917 nfs_retry_commit(mds_pages
, NULL
);
918 nfs_commit_clear_lock(NFS_I(inode
));
923 filelayout_free_deveiceid_node(struct nfs4_deviceid_node
*d
)
925 nfs4_fl_free_deviceid(container_of(d
, struct nfs4_file_layout_dsaddr
, id_node
));
928 static struct pnfs_layoutdriver_type filelayout_type
= {
929 .id
= LAYOUT_NFSV4_1_FILES
,
930 .name
= "LAYOUT_NFSV4_1_FILES",
931 .owner
= THIS_MODULE
,
932 .alloc_lseg
= filelayout_alloc_lseg
,
933 .free_lseg
= filelayout_free_lseg
,
934 .pg_read_ops
= &filelayout_pg_read_ops
,
935 .pg_write_ops
= &filelayout_pg_write_ops
,
936 .mark_pnfs_commit
= filelayout_mark_pnfs_commit
,
937 .choose_commit_list
= filelayout_choose_commit_list
,
938 .commit_pagelist
= filelayout_commit_pagelist
,
939 .read_pagelist
= filelayout_read_pagelist
,
940 .write_pagelist
= filelayout_write_pagelist
,
941 .free_deviceid_node
= filelayout_free_deveiceid_node
,
944 static int __init
nfs4filelayout_init(void)
946 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Registering...\n",
948 return pnfs_register_layoutdriver(&filelayout_type
);
951 static void __exit
nfs4filelayout_exit(void)
953 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Unregistering...\n",
955 pnfs_unregister_layoutdriver(&filelayout_type
);
958 module_init(nfs4filelayout_init
);
959 module_exit(nfs4filelayout_exit
);