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>
35 #include "nfs4filelayout.h"
37 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
39 MODULE_LICENSE("GPL");
40 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
41 MODULE_DESCRIPTION("The NFSv4 file layout driver");
43 #define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
46 filelayout_get_dense_offset(struct nfs4_filelayout_segment
*flseg
,
49 u32 stripe_width
= flseg
->stripe_unit
* flseg
->dsaddr
->stripe_count
;
52 offset
-= flseg
->pattern_offset
;
54 do_div(tmp
, stripe_width
);
56 return tmp
* flseg
->stripe_unit
+ do_div(offset
, flseg
->stripe_unit
);
59 /* This function is used by the layout driver to calculate the
60 * offset of the file on the dserver based on whether the
61 * layout type is STRIPE_DENSE or STRIPE_SPARSE
64 filelayout_get_dserver_offset(struct pnfs_layout_segment
*lseg
, loff_t offset
)
66 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
68 switch (flseg
->stripe_type
) {
73 return filelayout_get_dense_offset(flseg
, offset
);
79 /* For data server errors we don't recover from */
81 filelayout_set_lo_fail(struct pnfs_layout_segment
*lseg
)
83 if (lseg
->pls_range
.iomode
== IOMODE_RW
) {
84 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__
);
85 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
87 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__
);
88 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
92 static int filelayout_async_handle_error(struct rpc_task
*task
,
93 struct nfs4_state
*state
,
94 struct nfs_client
*clp
,
97 if (task
->tk_status
>= 0)
102 switch (task
->tk_status
) {
103 case -NFS4ERR_BADSESSION
:
104 case -NFS4ERR_BADSLOT
:
105 case -NFS4ERR_BAD_HIGH_SLOT
:
106 case -NFS4ERR_DEADSESSION
:
107 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION
:
108 case -NFS4ERR_SEQ_FALSE_RETRY
:
109 case -NFS4ERR_SEQ_MISORDERED
:
110 dprintk("%s ERROR %d, Reset session. Exchangeid "
111 "flags 0x%x\n", __func__
, task
->tk_status
,
112 clp
->cl_exchange_flags
);
113 nfs4_schedule_session_recovery(clp
->cl_session
);
118 rpc_delay(task
, FILELAYOUT_POLL_RETRY_MAX
);
120 case -NFS4ERR_RETRY_UNCACHED_REP
:
123 dprintk("%s DS error. Retry through MDS %d\n", __func__
,
132 /* NFS_PROTO call done callback routines */
134 static int filelayout_read_done_cb(struct rpc_task
*task
,
135 struct nfs_read_data
*data
)
137 struct nfs_client
*clp
= data
->ds_clp
;
140 dprintk("%s DS read\n", __func__
);
142 if (filelayout_async_handle_error(task
, data
->args
.context
->state
,
143 data
->ds_clp
, &reset
) == -EAGAIN
) {
144 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
145 __func__
, data
->ds_clp
, data
->ds_clp
->cl_session
);
147 filelayout_set_lo_fail(data
->lseg
);
148 nfs4_reset_read(task
, data
);
149 clp
= NFS_SERVER(data
->inode
)->nfs_client
;
151 nfs_restart_rpc(task
, clp
);
159 * We reference the rpc_cred of the first WRITE that triggers the need for
160 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
161 * rfc5661 is not clear about which credential should be used.
164 filelayout_set_layoutcommit(struct nfs_write_data
*wdata
)
166 if (FILELAYOUT_LSEG(wdata
->lseg
)->commit_through_mds
||
167 wdata
->res
.verf
->committed
== NFS_FILE_SYNC
)
170 pnfs_set_layoutcommit(wdata
);
171 dprintk("%s ionde %lu pls_end_pos %lu\n", __func__
, wdata
->inode
->i_ino
,
172 (unsigned long) wdata
->lseg
->pls_end_pos
);
176 * Call ops for the async read/write cases
177 * In the case of dense layouts, the offset needs to be reset to its
180 static void filelayout_read_prepare(struct rpc_task
*task
, void *data
)
182 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
184 rdata
->read_done_cb
= filelayout_read_done_cb
;
186 if (nfs41_setup_sequence(rdata
->ds_clp
->cl_session
,
187 &rdata
->args
.seq_args
, &rdata
->res
.seq_res
,
191 rpc_call_start(task
);
194 static void filelayout_read_call_done(struct rpc_task
*task
, void *data
)
196 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
198 dprintk("--> %s task->tk_status %d\n", __func__
, task
->tk_status
);
200 /* Note this may cause RPC to be resent */
201 rdata
->mds_ops
->rpc_call_done(task
, data
);
204 static void filelayout_read_release(void *data
)
206 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
208 rdata
->mds_ops
->rpc_release(data
);
211 static int filelayout_write_done_cb(struct rpc_task
*task
,
212 struct nfs_write_data
*data
)
216 if (filelayout_async_handle_error(task
, data
->args
.context
->state
,
217 data
->ds_clp
, &reset
) == -EAGAIN
) {
218 struct nfs_client
*clp
;
220 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
221 __func__
, data
->ds_clp
, data
->ds_clp
->cl_session
);
223 filelayout_set_lo_fail(data
->lseg
);
224 nfs4_reset_write(task
, data
);
225 clp
= NFS_SERVER(data
->inode
)->nfs_client
;
228 nfs_restart_rpc(task
, clp
);
232 filelayout_set_layoutcommit(data
);
236 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
237 static void prepare_to_resend_writes(struct nfs_write_data
*data
)
239 struct nfs_page
*first
= nfs_list_entry(data
->pages
.next
);
241 data
->task
.tk_status
= 0;
242 memcpy(data
->verf
.verifier
, first
->wb_verf
.verifier
,
243 sizeof(first
->wb_verf
.verifier
));
244 data
->verf
.verifier
[0]++; /* ensure verifier mismatch */
247 static int filelayout_commit_done_cb(struct rpc_task
*task
,
248 struct nfs_write_data
*data
)
252 if (filelayout_async_handle_error(task
, data
->args
.context
->state
,
253 data
->ds_clp
, &reset
) == -EAGAIN
) {
254 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
255 __func__
, data
->ds_clp
, data
->ds_clp
->cl_session
);
257 prepare_to_resend_writes(data
);
258 filelayout_set_lo_fail(data
->lseg
);
260 nfs_restart_rpc(task
, data
->ds_clp
);
267 static void filelayout_write_prepare(struct rpc_task
*task
, void *data
)
269 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
271 if (nfs41_setup_sequence(wdata
->ds_clp
->cl_session
,
272 &wdata
->args
.seq_args
, &wdata
->res
.seq_res
,
276 rpc_call_start(task
);
279 static void filelayout_write_call_done(struct rpc_task
*task
, void *data
)
281 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
283 /* Note this may cause RPC to be resent */
284 wdata
->mds_ops
->rpc_call_done(task
, data
);
287 static void filelayout_write_release(void *data
)
289 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
291 wdata
->mds_ops
->rpc_release(data
);
294 static void filelayout_commit_release(void *data
)
296 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
298 nfs_commit_release_pages(wdata
);
299 if (atomic_dec_and_test(&NFS_I(wdata
->inode
)->commits_outstanding
))
300 nfs_commit_clear_lock(NFS_I(wdata
->inode
));
301 nfs_commitdata_release(wdata
);
304 struct rpc_call_ops filelayout_read_call_ops
= {
305 .rpc_call_prepare
= filelayout_read_prepare
,
306 .rpc_call_done
= filelayout_read_call_done
,
307 .rpc_release
= filelayout_read_release
,
310 struct rpc_call_ops filelayout_write_call_ops
= {
311 .rpc_call_prepare
= filelayout_write_prepare
,
312 .rpc_call_done
= filelayout_write_call_done
,
313 .rpc_release
= filelayout_write_release
,
316 struct rpc_call_ops filelayout_commit_call_ops
= {
317 .rpc_call_prepare
= filelayout_write_prepare
,
318 .rpc_call_done
= filelayout_write_call_done
,
319 .rpc_release
= filelayout_commit_release
,
322 static enum pnfs_try_status
323 filelayout_read_pagelist(struct nfs_read_data
*data
)
325 struct pnfs_layout_segment
*lseg
= data
->lseg
;
326 struct nfs4_pnfs_ds
*ds
;
327 loff_t offset
= data
->args
.offset
;
332 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
333 __func__
, data
->inode
->i_ino
,
334 data
->args
.pgbase
, (size_t)data
->args
.count
, offset
);
336 /* Retrieve the correct rpc_client for the byte range */
337 j
= nfs4_fl_calc_j_index(lseg
, offset
);
338 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
339 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
341 /* Either layout fh index faulty, or ds connect failed */
342 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
343 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
344 return PNFS_NOT_ATTEMPTED
;
346 dprintk("%s USE DS:ip %x %hu\n", __func__
,
347 ntohl(ds
->ds_ip_addr
), ntohs(ds
->ds_port
));
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:%x:%hu\n", __func__
,
387 data
->inode
->i_ino
, sync
, (size_t) data
->args
.count
, offset
,
388 ntohl(ds
->ds_ip_addr
), ntohs(ds
->ds_port
));
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
);
400 data
->mds_offset
= offset
;
402 /* Perform an asynchronous write */
403 status
= nfs_initiate_write(data
, ds
->ds_clp
->cl_rpcclient
,
404 &filelayout_write_call_ops
, sync
);
406 return PNFS_ATTEMPTED
;
410 * filelayout_check_layout()
412 * Make sure layout segment parameters are sane WRT the device.
413 * At this point no generic layer initialization of the lseg has occurred,
414 * and nothing has been added to the layout_hdr cache.
418 filelayout_check_layout(struct pnfs_layout_hdr
*lo
,
419 struct nfs4_filelayout_segment
*fl
,
420 struct nfs4_layoutget_res
*lgr
,
421 struct nfs4_deviceid
*id
)
423 struct nfs4_file_layout_dsaddr
*dsaddr
;
424 int status
= -EINVAL
;
425 struct nfs_server
*nfss
= NFS_SERVER(lo
->plh_inode
);
427 dprintk("--> %s\n", __func__
);
429 if (fl
->pattern_offset
> lgr
->range
.offset
) {
430 dprintk("%s pattern_offset %lld to large\n",
431 __func__
, fl
->pattern_offset
);
435 if (!fl
->stripe_unit
|| fl
->stripe_unit
% PAGE_SIZE
) {
436 dprintk("%s Invalid stripe unit (%u)\n",
437 __func__
, fl
->stripe_unit
);
441 /* find and reference the deviceid */
442 dsaddr
= nfs4_fl_find_get_deviceid(id
);
443 if (dsaddr
== NULL
) {
444 dsaddr
= get_device_info(lo
->plh_inode
, id
);
450 if (fl
->first_stripe_index
< 0 ||
451 fl
->first_stripe_index
>= dsaddr
->stripe_count
) {
452 dprintk("%s Bad first_stripe_index %d\n",
453 __func__
, fl
->first_stripe_index
);
457 if ((fl
->stripe_type
== STRIPE_SPARSE
&&
458 fl
->num_fh
> 1 && fl
->num_fh
!= dsaddr
->ds_num
) ||
459 (fl
->stripe_type
== STRIPE_DENSE
&&
460 fl
->num_fh
!= dsaddr
->stripe_count
)) {
461 dprintk("%s num_fh %u not valid for given packing\n",
462 __func__
, fl
->num_fh
);
466 if (fl
->stripe_unit
% nfss
->rsize
|| fl
->stripe_unit
% nfss
->wsize
) {
467 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
468 "wsize %u\n", __func__
, fl
->stripe_unit
, nfss
->rsize
,
474 dprintk("--> %s returns %d\n", __func__
, status
);
477 nfs4_fl_put_deviceid(dsaddr
);
481 static void filelayout_free_fh_array(struct nfs4_filelayout_segment
*fl
)
485 for (i
= 0; i
< fl
->num_fh
; i
++) {
486 if (!fl
->fh_array
[i
])
488 kfree(fl
->fh_array
[i
]);
495 _filelayout_free_lseg(struct nfs4_filelayout_segment
*fl
)
497 filelayout_free_fh_array(fl
);
502 filelayout_decode_layout(struct pnfs_layout_hdr
*flo
,
503 struct nfs4_filelayout_segment
*fl
,
504 struct nfs4_layoutget_res
*lgr
,
505 struct nfs4_deviceid
*id
)
507 struct xdr_stream stream
;
508 struct xdr_buf buf
= {
509 .pages
= lgr
->layoutp
->pages
,
510 .page_len
= lgr
->layoutp
->len
,
511 .buflen
= lgr
->layoutp
->len
,
512 .len
= lgr
->layoutp
->len
,
514 struct page
*scratch
;
519 dprintk("%s: set_layout_map Begin\n", __func__
);
521 scratch
= alloc_page(GFP_KERNEL
);
525 xdr_init_decode(&stream
, &buf
, NULL
);
526 xdr_set_scratch_buffer(&stream
, page_address(scratch
), PAGE_SIZE
);
528 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
530 p
= xdr_inline_decode(&stream
, NFS4_DEVICEID4_SIZE
+ 20);
534 memcpy(id
, p
, sizeof(*id
));
535 p
+= XDR_QUADLEN(NFS4_DEVICEID4_SIZE
);
538 nfl_util
= be32_to_cpup(p
++);
539 if (nfl_util
& NFL4_UFLG_COMMIT_THRU_MDS
)
540 fl
->commit_through_mds
= 1;
541 if (nfl_util
& NFL4_UFLG_DENSE
)
542 fl
->stripe_type
= STRIPE_DENSE
;
544 fl
->stripe_type
= STRIPE_SPARSE
;
545 fl
->stripe_unit
= nfl_util
& ~NFL4_UFLG_MASK
;
547 fl
->first_stripe_index
= be32_to_cpup(p
++);
548 p
= xdr_decode_hyper(p
, &fl
->pattern_offset
);
549 fl
->num_fh
= be32_to_cpup(p
++);
551 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
552 __func__
, nfl_util
, fl
->num_fh
, fl
->first_stripe_index
,
558 fl
->fh_array
= kzalloc(fl
->num_fh
* sizeof(struct nfs_fh
*),
563 for (i
= 0; i
< fl
->num_fh
; i
++) {
564 /* Do we want to use a mempool here? */
565 fl
->fh_array
[i
] = kmalloc(sizeof(struct nfs_fh
), GFP_KERNEL
);
566 if (!fl
->fh_array
[i
])
569 p
= xdr_inline_decode(&stream
, 4);
572 fl
->fh_array
[i
]->size
= be32_to_cpup(p
++);
573 if (sizeof(struct nfs_fh
) < fl
->fh_array
[i
]->size
) {
574 printk(KERN_ERR
"Too big fh %d received %d\n",
575 i
, fl
->fh_array
[i
]->size
);
579 p
= xdr_inline_decode(&stream
, fl
->fh_array
[i
]->size
);
582 memcpy(fl
->fh_array
[i
]->data
, p
, fl
->fh_array
[i
]->size
);
583 dprintk("DEBUG: %s: fh len %d\n", __func__
,
584 fl
->fh_array
[i
]->size
);
587 __free_page(scratch
);
591 filelayout_free_fh_array(fl
);
593 __free_page(scratch
);
598 filelayout_free_lseg(struct pnfs_layout_segment
*lseg
)
600 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
602 dprintk("--> %s\n", __func__
);
603 nfs4_fl_put_deviceid(fl
->dsaddr
);
604 kfree(fl
->commit_buckets
);
605 _filelayout_free_lseg(fl
);
608 static struct pnfs_layout_segment
*
609 filelayout_alloc_lseg(struct pnfs_layout_hdr
*layoutid
,
610 struct nfs4_layoutget_res
*lgr
)
612 struct nfs4_filelayout_segment
*fl
;
614 struct nfs4_deviceid id
;
616 dprintk("--> %s\n", __func__
);
617 fl
= kzalloc(sizeof(*fl
), GFP_KERNEL
);
621 rc
= filelayout_decode_layout(layoutid
, fl
, lgr
, &id
);
622 if (rc
!= 0 || filelayout_check_layout(layoutid
, fl
, lgr
, &id
)) {
623 _filelayout_free_lseg(fl
);
627 /* This assumes there is only one IOMODE_RW lseg. What
628 * we really want to do is have a layout_hdr level
629 * dictionary of <multipath_list4, fh> keys, each
630 * associated with a struct list_head, populated by calls
631 * to filelayout_write_pagelist().
633 if ((!fl
->commit_through_mds
) && (lgr
->range
.iomode
== IOMODE_RW
)) {
635 int size
= (fl
->stripe_type
== STRIPE_SPARSE
) ?
636 fl
->dsaddr
->ds_num
: fl
->dsaddr
->stripe_count
;
638 fl
->commit_buckets
= kcalloc(size
, sizeof(struct list_head
), GFP_KERNEL
);
639 if (!fl
->commit_buckets
) {
640 filelayout_free_lseg(&fl
->generic_hdr
);
643 fl
->number_of_buckets
= size
;
644 for (i
= 0; i
< size
; i
++)
645 INIT_LIST_HEAD(&fl
->commit_buckets
[i
]);
647 return &fl
->generic_hdr
;
651 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
653 * return 1 : coalesce page
654 * return 0 : don't coalesce page
657 filelayout_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
658 struct nfs_page
*req
)
660 u64 p_stripe
, r_stripe
;
665 p_stripe
= (u64
)prev
->wb_index
<< PAGE_CACHE_SHIFT
;
666 r_stripe
= (u64
)req
->wb_index
<< PAGE_CACHE_SHIFT
;
667 stripe_unit
= FILELAYOUT_LSEG(pgio
->pg_lseg
)->stripe_unit
;
669 do_div(p_stripe
, stripe_unit
);
670 do_div(r_stripe
, stripe_unit
);
672 return (p_stripe
== r_stripe
);
675 static bool filelayout_mark_pnfs_commit(struct pnfs_layout_segment
*lseg
)
677 return !FILELAYOUT_LSEG(lseg
)->commit_through_mds
;
680 static u32
select_bucket_index(struct nfs4_filelayout_segment
*fl
, u32 j
)
682 if (fl
->stripe_type
== STRIPE_SPARSE
)
683 return nfs4_fl_calc_ds_index(&fl
->generic_hdr
, j
);
688 struct list_head
*filelayout_choose_commit_list(struct nfs_page
*req
)
690 struct pnfs_layout_segment
*lseg
= req
->wb_commit_lseg
;
691 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
693 struct list_head
*list
;
695 /* Note that we are calling nfs4_fl_calc_j_index on each page
696 * that ends up being committed to a data server. An attractive
697 * alternative is to add a field to nfs_write_data and nfs_page
698 * to store the value calculated in filelayout_write_pagelist
699 * and just use that here.
701 j
= nfs4_fl_calc_j_index(lseg
,
702 (loff_t
)req
->wb_index
<< PAGE_CACHE_SHIFT
);
703 i
= select_bucket_index(fl
, j
);
704 list
= &fl
->commit_buckets
[i
];
705 if (list_empty(list
)) {
706 /* Non-empty buckets hold a reference on the lseg */
712 static u32
calc_ds_index_from_commit(struct pnfs_layout_segment
*lseg
, u32 i
)
714 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
716 if (flseg
->stripe_type
== STRIPE_SPARSE
)
719 return nfs4_fl_calc_ds_index(lseg
, i
);
722 static struct nfs_fh
*
723 select_ds_fh_from_commit(struct pnfs_layout_segment
*lseg
, u32 i
)
725 struct nfs4_filelayout_segment
*flseg
= FILELAYOUT_LSEG(lseg
);
727 if (flseg
->stripe_type
== STRIPE_SPARSE
) {
728 if (flseg
->num_fh
== 1)
730 else if (flseg
->num_fh
== 0)
731 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
734 return flseg
->fh_array
[i
];
737 static int filelayout_initiate_commit(struct nfs_write_data
*data
, int how
)
739 struct pnfs_layout_segment
*lseg
= data
->lseg
;
740 struct nfs4_pnfs_ds
*ds
;
744 idx
= calc_ds_index_from_commit(lseg
, data
->ds_commit_index
);
745 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
747 printk(KERN_ERR
"%s: prepare_ds failed, use MDS\n", __func__
);
748 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
749 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
750 prepare_to_resend_writes(data
);
751 data
->mds_ops
->rpc_release(data
);
754 dprintk("%s ino %lu, how %d\n", __func__
, data
->inode
->i_ino
, how
);
755 data
->write_done_cb
= filelayout_commit_done_cb
;
756 data
->ds_clp
= ds
->ds_clp
;
757 fh
= select_ds_fh_from_commit(lseg
, data
->ds_commit_index
);
760 return nfs_initiate_commit(data
, ds
->ds_clp
->cl_rpcclient
,
761 &filelayout_commit_call_ops
, how
);
765 * This is only useful while we are using whole file layouts.
767 static struct pnfs_layout_segment
*find_only_write_lseg(struct inode
*inode
)
769 struct pnfs_layout_segment
*lseg
, *rv
= NULL
;
771 spin_lock(&inode
->i_lock
);
772 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
)
773 if (lseg
->pls_range
.iomode
== IOMODE_RW
)
775 spin_unlock(&inode
->i_lock
);
779 static int alloc_ds_commits(struct inode
*inode
, struct list_head
*list
)
781 struct pnfs_layout_segment
*lseg
;
782 struct nfs4_filelayout_segment
*fl
;
783 struct nfs_write_data
*data
;
786 /* Won't need this when non-whole file layout segments are supported
787 * instead we will use a pnfs_layout_hdr structure */
788 lseg
= find_only_write_lseg(inode
);
791 fl
= FILELAYOUT_LSEG(lseg
);
792 for (i
= 0; i
< fl
->number_of_buckets
; i
++) {
793 if (list_empty(&fl
->commit_buckets
[i
]))
795 data
= nfs_commitdata_alloc();
798 data
->ds_commit_index
= i
;
800 list_add(&data
->pages
, list
);
806 for (j
= i
; j
< fl
->number_of_buckets
; j
++) {
807 if (list_empty(&fl
->commit_buckets
[i
]))
809 nfs_retry_commit(&fl
->commit_buckets
[i
], lseg
);
810 put_lseg(lseg
); /* associated with emptying bucket */
813 /* Caller will clean up entries put on list */
817 /* This follows nfs_commit_list pretty closely */
819 filelayout_commit_pagelist(struct inode
*inode
, struct list_head
*mds_pages
,
822 struct nfs_write_data
*data
, *tmp
;
825 if (!list_empty(mds_pages
)) {
826 data
= nfs_commitdata_alloc();
830 list_add(&data
->pages
, &list
);
833 if (alloc_ds_commits(inode
, &list
))
836 list_for_each_entry_safe(data
, tmp
, &list
, pages
) {
837 list_del_init(&data
->pages
);
838 atomic_inc(&NFS_I(inode
)->commits_outstanding
);
840 nfs_init_commit(data
, mds_pages
, NULL
);
841 nfs_initiate_commit(data
, NFS_CLIENT(inode
),
844 nfs_init_commit(data
, &FILELAYOUT_LSEG(data
->lseg
)->commit_buckets
[data
->ds_commit_index
], data
->lseg
);
845 filelayout_initiate_commit(data
, how
);
850 list_for_each_entry_safe(data
, tmp
, &list
, pages
) {
851 nfs_retry_commit(&data
->pages
, data
->lseg
);
852 list_del_init(&data
->pages
);
853 nfs_commit_free(data
);
855 nfs_retry_commit(mds_pages
, NULL
);
856 nfs_commit_clear_lock(NFS_I(inode
));
860 static struct pnfs_layoutdriver_type filelayout_type
= {
861 .id
= LAYOUT_NFSV4_1_FILES
,
862 .name
= "LAYOUT_NFSV4_1_FILES",
863 .owner
= THIS_MODULE
,
864 .alloc_lseg
= filelayout_alloc_lseg
,
865 .free_lseg
= filelayout_free_lseg
,
866 .pg_test
= filelayout_pg_test
,
867 .mark_pnfs_commit
= filelayout_mark_pnfs_commit
,
868 .choose_commit_list
= filelayout_choose_commit_list
,
869 .commit_pagelist
= filelayout_commit_pagelist
,
870 .read_pagelist
= filelayout_read_pagelist
,
871 .write_pagelist
= filelayout_write_pagelist
,
874 static int __init
nfs4filelayout_init(void)
876 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Registering...\n",
878 return pnfs_register_layoutdriver(&filelayout_type
);
881 static void __exit
nfs4filelayout_exit(void)
883 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Unregistering...\n",
885 pnfs_unregister_layoutdriver(&filelayout_type
);
888 module_init(nfs4filelayout_init
);
889 module_exit(nfs4filelayout_exit
);