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
);
121 dprintk("%s DS error. Retry through MDS %d\n", __func__
,
130 /* NFS_PROTO call done callback routines */
132 static int filelayout_read_done_cb(struct rpc_task
*task
,
133 struct nfs_read_data
*data
)
135 struct nfs_client
*clp
= data
->ds_clp
;
138 dprintk("%s DS read\n", __func__
);
140 if (filelayout_async_handle_error(task
, data
->args
.context
->state
,
141 data
->ds_clp
, &reset
) == -EAGAIN
) {
142 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
143 __func__
, data
->ds_clp
, data
->ds_clp
->cl_session
);
145 filelayout_set_lo_fail(data
->lseg
);
146 nfs4_reset_read(task
, data
);
147 clp
= NFS_SERVER(data
->inode
)->nfs_client
;
149 nfs_restart_rpc(task
, clp
);
157 * Call ops for the async read/write cases
158 * In the case of dense layouts, the offset needs to be reset to its
161 static void filelayout_read_prepare(struct rpc_task
*task
, void *data
)
163 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
165 rdata
->read_done_cb
= filelayout_read_done_cb
;
167 if (nfs41_setup_sequence(rdata
->ds_clp
->cl_session
,
168 &rdata
->args
.seq_args
, &rdata
->res
.seq_res
,
172 rpc_call_start(task
);
175 static void filelayout_read_call_done(struct rpc_task
*task
, void *data
)
177 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
179 dprintk("--> %s task->tk_status %d\n", __func__
, task
->tk_status
);
181 /* Note this may cause RPC to be resent */
182 rdata
->mds_ops
->rpc_call_done(task
, data
);
185 static void filelayout_read_release(void *data
)
187 struct nfs_read_data
*rdata
= (struct nfs_read_data
*)data
;
189 rdata
->mds_ops
->rpc_release(data
);
192 static int filelayout_write_done_cb(struct rpc_task
*task
,
193 struct nfs_write_data
*data
)
197 if (filelayout_async_handle_error(task
, data
->args
.context
->state
,
198 data
->ds_clp
, &reset
) == -EAGAIN
) {
199 struct nfs_client
*clp
;
201 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
202 __func__
, data
->ds_clp
, data
->ds_clp
->cl_session
);
204 filelayout_set_lo_fail(data
->lseg
);
205 nfs4_reset_write(task
, data
);
206 clp
= NFS_SERVER(data
->inode
)->nfs_client
;
209 nfs_restart_rpc(task
, clp
);
216 static void filelayout_write_prepare(struct rpc_task
*task
, void *data
)
218 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
220 if (nfs41_setup_sequence(wdata
->ds_clp
->cl_session
,
221 &wdata
->args
.seq_args
, &wdata
->res
.seq_res
,
225 rpc_call_start(task
);
228 static void filelayout_write_call_done(struct rpc_task
*task
, void *data
)
230 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
232 /* Note this may cause RPC to be resent */
233 wdata
->mds_ops
->rpc_call_done(task
, data
);
236 static void filelayout_write_release(void *data
)
238 struct nfs_write_data
*wdata
= (struct nfs_write_data
*)data
;
240 wdata
->mds_ops
->rpc_release(data
);
243 struct rpc_call_ops filelayout_read_call_ops
= {
244 .rpc_call_prepare
= filelayout_read_prepare
,
245 .rpc_call_done
= filelayout_read_call_done
,
246 .rpc_release
= filelayout_read_release
,
249 struct rpc_call_ops filelayout_write_call_ops
= {
250 .rpc_call_prepare
= filelayout_write_prepare
,
251 .rpc_call_done
= filelayout_write_call_done
,
252 .rpc_release
= filelayout_write_release
,
255 static enum pnfs_try_status
256 filelayout_read_pagelist(struct nfs_read_data
*data
)
258 struct pnfs_layout_segment
*lseg
= data
->lseg
;
259 struct nfs4_pnfs_ds
*ds
;
260 loff_t offset
= data
->args
.offset
;
265 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
266 __func__
, data
->inode
->i_ino
,
267 data
->args
.pgbase
, (size_t)data
->args
.count
, offset
);
269 /* Retrieve the correct rpc_client for the byte range */
270 j
= nfs4_fl_calc_j_index(lseg
, offset
);
271 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
272 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
274 /* Either layout fh index faulty, or ds connect failed */
275 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
276 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
277 return PNFS_NOT_ATTEMPTED
;
279 dprintk("%s USE DS:ip %x %hu\n", __func__
,
280 ntohl(ds
->ds_ip_addr
), ntohs(ds
->ds_port
));
282 /* No multipath support. Use first DS */
283 data
->ds_clp
= ds
->ds_clp
;
284 fh
= nfs4_fl_select_ds_fh(lseg
, j
);
288 data
->args
.offset
= filelayout_get_dserver_offset(lseg
, offset
);
289 data
->mds_offset
= offset
;
291 /* Perform an asynchronous read to ds */
292 status
= nfs_initiate_read(data
, ds
->ds_clp
->cl_rpcclient
,
293 &filelayout_read_call_ops
);
295 return PNFS_ATTEMPTED
;
298 /* Perform async writes. */
299 static enum pnfs_try_status
300 filelayout_write_pagelist(struct nfs_write_data
*data
, int sync
)
302 struct pnfs_layout_segment
*lseg
= data
->lseg
;
303 struct nfs4_pnfs_ds
*ds
;
304 loff_t offset
= data
->args
.offset
;
309 /* Retrieve the correct rpc_client for the byte range */
310 j
= nfs4_fl_calc_j_index(lseg
, offset
);
311 idx
= nfs4_fl_calc_ds_index(lseg
, j
);
312 ds
= nfs4_fl_prepare_ds(lseg
, idx
);
314 printk(KERN_ERR
"%s: prepare_ds failed, use MDS\n", __func__
);
315 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
316 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
317 return PNFS_NOT_ATTEMPTED
;
319 dprintk("%s ino %lu sync %d req %Zu@%llu DS:%x:%hu\n", __func__
,
320 data
->inode
->i_ino
, sync
, (size_t) data
->args
.count
, offset
,
321 ntohl(ds
->ds_ip_addr
), ntohs(ds
->ds_port
));
323 /* We can't handle commit to ds yet */
324 if (!FILELAYOUT_LSEG(lseg
)->commit_through_mds
)
325 data
->args
.stable
= NFS_FILE_SYNC
;
327 data
->write_done_cb
= filelayout_write_done_cb
;
328 data
->ds_clp
= ds
->ds_clp
;
329 fh
= nfs4_fl_select_ds_fh(lseg
, j
);
333 * Get the file offset on the dserver. Set the write offset to
334 * this offset and save the original offset.
336 data
->args
.offset
= filelayout_get_dserver_offset(lseg
, offset
);
337 data
->mds_offset
= offset
;
339 /* Perform an asynchronous write */
340 status
= nfs_initiate_write(data
, ds
->ds_clp
->cl_rpcclient
,
341 &filelayout_write_call_ops
, sync
);
343 return PNFS_ATTEMPTED
;
347 * filelayout_check_layout()
349 * Make sure layout segment parameters are sane WRT the device.
350 * At this point no generic layer initialization of the lseg has occurred,
351 * and nothing has been added to the layout_hdr cache.
355 filelayout_check_layout(struct pnfs_layout_hdr
*lo
,
356 struct nfs4_filelayout_segment
*fl
,
357 struct nfs4_layoutget_res
*lgr
,
358 struct nfs4_deviceid
*id
)
360 struct nfs4_file_layout_dsaddr
*dsaddr
;
361 int status
= -EINVAL
;
362 struct nfs_server
*nfss
= NFS_SERVER(lo
->plh_inode
);
364 dprintk("--> %s\n", __func__
);
366 if (fl
->pattern_offset
> lgr
->range
.offset
) {
367 dprintk("%s pattern_offset %lld to large\n",
368 __func__
, fl
->pattern_offset
);
372 if (!fl
->stripe_unit
|| fl
->stripe_unit
% PAGE_SIZE
) {
373 dprintk("%s Invalid stripe unit (%u)\n",
374 __func__
, fl
->stripe_unit
);
378 /* find and reference the deviceid */
379 dsaddr
= nfs4_fl_find_get_deviceid(id
);
380 if (dsaddr
== NULL
) {
381 dsaddr
= get_device_info(lo
->plh_inode
, id
);
387 if (fl
->first_stripe_index
< 0 ||
388 fl
->first_stripe_index
>= dsaddr
->stripe_count
) {
389 dprintk("%s Bad first_stripe_index %d\n",
390 __func__
, fl
->first_stripe_index
);
394 if ((fl
->stripe_type
== STRIPE_SPARSE
&&
395 fl
->num_fh
> 1 && fl
->num_fh
!= dsaddr
->ds_num
) ||
396 (fl
->stripe_type
== STRIPE_DENSE
&&
397 fl
->num_fh
!= dsaddr
->stripe_count
)) {
398 dprintk("%s num_fh %u not valid for given packing\n",
399 __func__
, fl
->num_fh
);
403 if (fl
->stripe_unit
% nfss
->rsize
|| fl
->stripe_unit
% nfss
->wsize
) {
404 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
405 "wsize %u\n", __func__
, fl
->stripe_unit
, nfss
->rsize
,
411 dprintk("--> %s returns %d\n", __func__
, status
);
414 nfs4_fl_put_deviceid(dsaddr
);
418 static void filelayout_free_fh_array(struct nfs4_filelayout_segment
*fl
)
422 for (i
= 0; i
< fl
->num_fh
; i
++) {
423 if (!fl
->fh_array
[i
])
425 kfree(fl
->fh_array
[i
]);
432 _filelayout_free_lseg(struct nfs4_filelayout_segment
*fl
)
434 filelayout_free_fh_array(fl
);
439 filelayout_decode_layout(struct pnfs_layout_hdr
*flo
,
440 struct nfs4_filelayout_segment
*fl
,
441 struct nfs4_layoutget_res
*lgr
,
442 struct nfs4_deviceid
*id
)
444 uint32_t *p
= (uint32_t *)lgr
->layout
.buf
;
448 dprintk("%s: set_layout_map Begin\n", __func__
);
450 memcpy(id
, p
, sizeof(*id
));
451 p
+= XDR_QUADLEN(NFS4_DEVICEID4_SIZE
);
454 nfl_util
= be32_to_cpup(p
++);
455 if (nfl_util
& NFL4_UFLG_COMMIT_THRU_MDS
)
456 fl
->commit_through_mds
= 1;
457 if (nfl_util
& NFL4_UFLG_DENSE
)
458 fl
->stripe_type
= STRIPE_DENSE
;
460 fl
->stripe_type
= STRIPE_SPARSE
;
461 fl
->stripe_unit
= nfl_util
& ~NFL4_UFLG_MASK
;
463 fl
->first_stripe_index
= be32_to_cpup(p
++);
464 p
= xdr_decode_hyper(p
, &fl
->pattern_offset
);
465 fl
->num_fh
= be32_to_cpup(p
++);
467 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
468 __func__
, nfl_util
, fl
->num_fh
, fl
->first_stripe_index
,
471 fl
->fh_array
= kzalloc(fl
->num_fh
* sizeof(struct nfs_fh
*),
476 for (i
= 0; i
< fl
->num_fh
; i
++) {
477 /* Do we want to use a mempool here? */
478 fl
->fh_array
[i
] = kmalloc(sizeof(struct nfs_fh
), GFP_KERNEL
);
479 if (!fl
->fh_array
[i
]) {
480 filelayout_free_fh_array(fl
);
483 fl
->fh_array
[i
]->size
= be32_to_cpup(p
++);
484 if (sizeof(struct nfs_fh
) < fl
->fh_array
[i
]->size
) {
485 printk(KERN_ERR
"Too big fh %d received %d\n",
486 i
, fl
->fh_array
[i
]->size
);
487 filelayout_free_fh_array(fl
);
490 memcpy(fl
->fh_array
[i
]->data
, p
, fl
->fh_array
[i
]->size
);
491 p
+= XDR_QUADLEN(fl
->fh_array
[i
]->size
);
492 dprintk("DEBUG: %s: fh len %d\n", __func__
,
493 fl
->fh_array
[i
]->size
);
499 static struct pnfs_layout_segment
*
500 filelayout_alloc_lseg(struct pnfs_layout_hdr
*layoutid
,
501 struct nfs4_layoutget_res
*lgr
)
503 struct nfs4_filelayout_segment
*fl
;
505 struct nfs4_deviceid id
;
507 dprintk("--> %s\n", __func__
);
508 fl
= kzalloc(sizeof(*fl
), GFP_KERNEL
);
512 rc
= filelayout_decode_layout(layoutid
, fl
, lgr
, &id
);
513 if (rc
!= 0 || filelayout_check_layout(layoutid
, fl
, lgr
, &id
)) {
514 _filelayout_free_lseg(fl
);
517 return &fl
->generic_hdr
;
521 filelayout_free_lseg(struct pnfs_layout_segment
*lseg
)
523 struct nfs4_filelayout_segment
*fl
= FILELAYOUT_LSEG(lseg
);
525 dprintk("--> %s\n", __func__
);
526 nfs4_fl_put_deviceid(fl
->dsaddr
);
527 _filelayout_free_lseg(fl
);
531 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
533 * return 1 : coalesce page
534 * return 0 : don't coalesce page
537 filelayout_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
538 struct nfs_page
*req
)
540 u64 p_stripe
, r_stripe
;
545 p_stripe
= (u64
)prev
->wb_index
<< PAGE_CACHE_SHIFT
;
546 r_stripe
= (u64
)req
->wb_index
<< PAGE_CACHE_SHIFT
;
547 stripe_unit
= FILELAYOUT_LSEG(pgio
->pg_lseg
)->stripe_unit
;
549 do_div(p_stripe
, stripe_unit
);
550 do_div(r_stripe
, stripe_unit
);
552 return (p_stripe
== r_stripe
);
555 static struct pnfs_layoutdriver_type filelayout_type
= {
556 .id
= LAYOUT_NFSV4_1_FILES
,
557 .name
= "LAYOUT_NFSV4_1_FILES",
558 .owner
= THIS_MODULE
,
559 .alloc_lseg
= filelayout_alloc_lseg
,
560 .free_lseg
= filelayout_free_lseg
,
561 .pg_test
= filelayout_pg_test
,
562 .read_pagelist
= filelayout_read_pagelist
,
563 .write_pagelist
= filelayout_write_pagelist
,
566 static int __init
nfs4filelayout_init(void)
568 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Registering...\n",
570 return pnfs_register_layoutdriver(&filelayout_type
);
573 static void __exit
nfs4filelayout_exit(void)
575 printk(KERN_INFO
"%s: NFSv4 File Layout Driver Unregistering...\n",
577 pnfs_unregister_layoutdriver(&filelayout_type
);
580 module_init(nfs4filelayout_init
);
581 module_exit(nfs4filelayout_exit
);