Merge tag 'for-4.5' of git://git.osdn.jp/gitroot/uclinux-h8/linux
[linux-drm-fsl-dcu.git] / fs / nfs / filelayout / filelayout.c
1 /*
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.
5  *
6  *  Copyright (c) 2002
7  *  The Regents of the University of Michigan
8  *  All Rights Reserved
9  *
10  *  Dean Hildebrand <dhildebz@umich.edu>
11  *
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.
20  *
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
29  *  such damages.
30  */
31
32 #include <linux/nfs_fs.h>
33 #include <linux/nfs_page.h>
34 #include <linux/module.h>
35 #include <linux/backing-dev.h>
36
37 #include <linux/sunrpc/metrics.h>
38
39 #include "../nfs4session.h"
40 #include "../internal.h"
41 #include "../delegation.h"
42 #include "filelayout.h"
43 #include "../nfs4trace.h"
44
45 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
46
47 MODULE_LICENSE("GPL");
48 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
49 MODULE_DESCRIPTION("The NFSv4 file layout driver");
50
51 #define FILELAYOUT_POLL_RETRY_MAX     (15*HZ)
52
53 static loff_t
54 filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
55                             loff_t offset)
56 {
57         u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
58         u64 stripe_no;
59         u32 rem;
60
61         offset -= flseg->pattern_offset;
62         stripe_no = div_u64(offset, stripe_width);
63         div_u64_rem(offset, flseg->stripe_unit, &rem);
64
65         return stripe_no * flseg->stripe_unit + rem;
66 }
67
68 /* This function is used by the layout driver to calculate the
69  * offset of the file on the dserver based on whether the
70  * layout type is STRIPE_DENSE or STRIPE_SPARSE
71  */
72 static loff_t
73 filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
74 {
75         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
76
77         switch (flseg->stripe_type) {
78         case STRIPE_SPARSE:
79                 return offset;
80
81         case STRIPE_DENSE:
82                 return filelayout_get_dense_offset(flseg, offset);
83         }
84
85         BUG();
86 }
87
88 static void filelayout_reset_write(struct nfs_pgio_header *hdr)
89 {
90         struct rpc_task *task = &hdr->task;
91
92         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
93                 dprintk("%s Reset task %5u for i/o through MDS "
94                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
95                         hdr->task.tk_pid,
96                         hdr->inode->i_sb->s_id,
97                         (unsigned long long)NFS_FILEID(hdr->inode),
98                         hdr->args.count,
99                         (unsigned long long)hdr->args.offset);
100
101                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
102         }
103 }
104
105 static void filelayout_reset_read(struct nfs_pgio_header *hdr)
106 {
107         struct rpc_task *task = &hdr->task;
108
109         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
110                 dprintk("%s Reset task %5u for i/o through MDS "
111                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
112                         hdr->task.tk_pid,
113                         hdr->inode->i_sb->s_id,
114                         (unsigned long long)NFS_FILEID(hdr->inode),
115                         hdr->args.count,
116                         (unsigned long long)hdr->args.offset);
117
118                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
119         }
120 }
121
122 static int filelayout_async_handle_error(struct rpc_task *task,
123                                          struct nfs4_state *state,
124                                          struct nfs_client *clp,
125                                          struct pnfs_layout_segment *lseg)
126 {
127         struct pnfs_layout_hdr *lo = lseg->pls_layout;
128         struct inode *inode = lo->plh_inode;
129         struct nfs_server *mds_server = NFS_SERVER(inode);
130         struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
131         struct nfs_client *mds_client = mds_server->nfs_client;
132         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
133
134         if (task->tk_status >= 0)
135                 return 0;
136
137         switch (task->tk_status) {
138         /* MDS state errors */
139         case -NFS4ERR_DELEG_REVOKED:
140         case -NFS4ERR_ADMIN_REVOKED:
141         case -NFS4ERR_BAD_STATEID:
142         case -NFS4ERR_OPENMODE:
143                 if (state == NULL)
144                         break;
145                 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
146                         goto out_bad_stateid;
147                 goto wait_on_recovery;
148         case -NFS4ERR_EXPIRED:
149                 if (state != NULL) {
150                         if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
151                                 goto out_bad_stateid;
152                 }
153                 nfs4_schedule_lease_recovery(mds_client);
154                 goto wait_on_recovery;
155         /* DS session errors */
156         case -NFS4ERR_BADSESSION:
157         case -NFS4ERR_BADSLOT:
158         case -NFS4ERR_BAD_HIGH_SLOT:
159         case -NFS4ERR_DEADSESSION:
160         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
161         case -NFS4ERR_SEQ_FALSE_RETRY:
162         case -NFS4ERR_SEQ_MISORDERED:
163                 dprintk("%s ERROR %d, Reset session. Exchangeid "
164                         "flags 0x%x\n", __func__, task->tk_status,
165                         clp->cl_exchange_flags);
166                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
167                 break;
168         case -NFS4ERR_DELAY:
169         case -NFS4ERR_GRACE:
170                 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
171                 break;
172         case -NFS4ERR_RETRY_UNCACHED_REP:
173                 break;
174         /* Invalidate Layout errors */
175         case -NFS4ERR_PNFS_NO_LAYOUT:
176         case -ESTALE:           /* mapped NFS4ERR_STALE */
177         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
178         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
179         case -NFS4ERR_FHEXPIRED:
180         case -NFS4ERR_WRONG_TYPE:
181                 dprintk("%s Invalid layout error %d\n", __func__,
182                         task->tk_status);
183                 /*
184                  * Destroy layout so new i/o will get a new layout.
185                  * Layout will not be destroyed until all current lseg
186                  * references are put. Mark layout as invalid to resend failed
187                  * i/o and all i/o waiting on the slot table to the MDS until
188                  * layout is destroyed and a new valid layout is obtained.
189                  */
190                 pnfs_destroy_layout(NFS_I(inode));
191                 rpc_wake_up(&tbl->slot_tbl_waitq);
192                 goto reset;
193         /* RPC connection errors */
194         case -ECONNREFUSED:
195         case -EHOSTDOWN:
196         case -EHOSTUNREACH:
197         case -ENETUNREACH:
198         case -EIO:
199         case -ETIMEDOUT:
200         case -EPIPE:
201                 dprintk("%s DS connection error %d\n", __func__,
202                         task->tk_status);
203                 nfs4_mark_deviceid_unavailable(devid);
204                 pnfs_error_mark_layout_for_return(inode, lseg);
205                 pnfs_set_lo_fail(lseg);
206                 rpc_wake_up(&tbl->slot_tbl_waitq);
207                 /* fall through */
208         default:
209 reset:
210                 dprintk("%s Retry through MDS. Error %d\n", __func__,
211                         task->tk_status);
212                 return -NFS4ERR_RESET_TO_MDS;
213         }
214 out:
215         task->tk_status = 0;
216         return -EAGAIN;
217 out_bad_stateid:
218         task->tk_status = -EIO;
219         return 0;
220 wait_on_recovery:
221         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
222         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
223                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
224         goto out;
225 }
226
227 /* NFS_PROTO call done callback routines */
228
229 static int filelayout_read_done_cb(struct rpc_task *task,
230                                 struct nfs_pgio_header *hdr)
231 {
232         int err;
233
234         trace_nfs4_pnfs_read(hdr, task->tk_status);
235         err = filelayout_async_handle_error(task, hdr->args.context->state,
236                                             hdr->ds_clp, hdr->lseg);
237
238         switch (err) {
239         case -NFS4ERR_RESET_TO_MDS:
240                 filelayout_reset_read(hdr);
241                 return task->tk_status;
242         case -EAGAIN:
243                 rpc_restart_call_prepare(task);
244                 return -EAGAIN;
245         }
246
247         return 0;
248 }
249
250 /*
251  * We reference the rpc_cred of the first WRITE that triggers the need for
252  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
253  * rfc5661 is not clear about which credential should be used.
254  */
255 static void
256 filelayout_set_layoutcommit(struct nfs_pgio_header *hdr)
257 {
258
259         if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds ||
260             hdr->res.verf->committed != NFS_DATA_SYNC)
261                 return;
262
263         pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
264                         hdr->mds_offset + hdr->res.count);
265         dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
266                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
267 }
268
269 bool
270 filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node)
271 {
272         return filelayout_test_devid_invalid(node) ||
273                 nfs4_test_deviceid_unavailable(node);
274 }
275
276 static bool
277 filelayout_reset_to_mds(struct pnfs_layout_segment *lseg)
278 {
279         struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg);
280
281         return filelayout_test_devid_unavailable(node);
282 }
283
284 /*
285  * Call ops for the async read/write cases
286  * In the case of dense layouts, the offset needs to be reset to its
287  * original value.
288  */
289 static void filelayout_read_prepare(struct rpc_task *task, void *data)
290 {
291         struct nfs_pgio_header *hdr = data;
292
293         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
294                 rpc_exit(task, -EIO);
295                 return;
296         }
297         if (filelayout_reset_to_mds(hdr->lseg)) {
298                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
299                 filelayout_reset_read(hdr);
300                 rpc_exit(task, 0);
301                 return;
302         }
303         hdr->pgio_done_cb = filelayout_read_done_cb;
304
305         if (nfs41_setup_sequence(hdr->ds_clp->cl_session,
306                         &hdr->args.seq_args,
307                         &hdr->res.seq_res,
308                         task))
309                 return;
310         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
311                         hdr->args.lock_context, FMODE_READ) == -EIO)
312                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
313 }
314
315 static void filelayout_read_call_done(struct rpc_task *task, void *data)
316 {
317         struct nfs_pgio_header *hdr = data;
318
319         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
320
321         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
322             task->tk_status == 0) {
323                 nfs41_sequence_done(task, &hdr->res.seq_res);
324                 return;
325         }
326
327         /* Note this may cause RPC to be resent */
328         hdr->mds_ops->rpc_call_done(task, data);
329 }
330
331 static void filelayout_read_count_stats(struct rpc_task *task, void *data)
332 {
333         struct nfs_pgio_header *hdr = data;
334
335         rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
336 }
337
338 static int filelayout_write_done_cb(struct rpc_task *task,
339                                 struct nfs_pgio_header *hdr)
340 {
341         int err;
342
343         trace_nfs4_pnfs_write(hdr, task->tk_status);
344         err = filelayout_async_handle_error(task, hdr->args.context->state,
345                                             hdr->ds_clp, hdr->lseg);
346
347         switch (err) {
348         case -NFS4ERR_RESET_TO_MDS:
349                 filelayout_reset_write(hdr);
350                 return task->tk_status;
351         case -EAGAIN:
352                 rpc_restart_call_prepare(task);
353                 return -EAGAIN;
354         }
355
356         filelayout_set_layoutcommit(hdr);
357         return 0;
358 }
359
360 static int filelayout_commit_done_cb(struct rpc_task *task,
361                                      struct nfs_commit_data *data)
362 {
363         int err;
364
365         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
366         err = filelayout_async_handle_error(task, NULL, data->ds_clp,
367                                             data->lseg);
368
369         switch (err) {
370         case -NFS4ERR_RESET_TO_MDS:
371                 pnfs_generic_prepare_to_resend_writes(data);
372                 return -EAGAIN;
373         case -EAGAIN:
374                 rpc_restart_call_prepare(task);
375                 return -EAGAIN;
376         }
377
378         if (data->verf.committed == NFS_UNSTABLE)
379                 pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
380
381         return 0;
382 }
383
384 static void filelayout_write_prepare(struct rpc_task *task, void *data)
385 {
386         struct nfs_pgio_header *hdr = data;
387
388         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
389                 rpc_exit(task, -EIO);
390                 return;
391         }
392         if (filelayout_reset_to_mds(hdr->lseg)) {
393                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
394                 filelayout_reset_write(hdr);
395                 rpc_exit(task, 0);
396                 return;
397         }
398         if (nfs41_setup_sequence(hdr->ds_clp->cl_session,
399                         &hdr->args.seq_args,
400                         &hdr->res.seq_res,
401                         task))
402                 return;
403         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
404                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
405                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
406 }
407
408 static void filelayout_write_call_done(struct rpc_task *task, void *data)
409 {
410         struct nfs_pgio_header *hdr = data;
411
412         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
413             task->tk_status == 0) {
414                 nfs41_sequence_done(task, &hdr->res.seq_res);
415                 return;
416         }
417
418         /* Note this may cause RPC to be resent */
419         hdr->mds_ops->rpc_call_done(task, data);
420 }
421
422 static void filelayout_write_count_stats(struct rpc_task *task, void *data)
423 {
424         struct nfs_pgio_header *hdr = data;
425
426         rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
427 }
428
429 static void filelayout_commit_prepare(struct rpc_task *task, void *data)
430 {
431         struct nfs_commit_data *wdata = data;
432
433         nfs41_setup_sequence(wdata->ds_clp->cl_session,
434                         &wdata->args.seq_args,
435                         &wdata->res.seq_res,
436                         task);
437 }
438
439 static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
440 {
441         struct nfs_commit_data *cdata = data;
442
443         rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
444 }
445
446 static const struct rpc_call_ops filelayout_read_call_ops = {
447         .rpc_call_prepare = filelayout_read_prepare,
448         .rpc_call_done = filelayout_read_call_done,
449         .rpc_count_stats = filelayout_read_count_stats,
450         .rpc_release = pnfs_generic_rw_release,
451 };
452
453 static const struct rpc_call_ops filelayout_write_call_ops = {
454         .rpc_call_prepare = filelayout_write_prepare,
455         .rpc_call_done = filelayout_write_call_done,
456         .rpc_count_stats = filelayout_write_count_stats,
457         .rpc_release = pnfs_generic_rw_release,
458 };
459
460 static const struct rpc_call_ops filelayout_commit_call_ops = {
461         .rpc_call_prepare = filelayout_commit_prepare,
462         .rpc_call_done = pnfs_generic_write_commit_done,
463         .rpc_count_stats = filelayout_commit_count_stats,
464         .rpc_release = pnfs_generic_commit_release,
465 };
466
467 static enum pnfs_try_status
468 filelayout_read_pagelist(struct nfs_pgio_header *hdr)
469 {
470         struct pnfs_layout_segment *lseg = hdr->lseg;
471         struct nfs4_pnfs_ds *ds;
472         struct rpc_clnt *ds_clnt;
473         loff_t offset = hdr->args.offset;
474         u32 j, idx;
475         struct nfs_fh *fh;
476
477         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
478                 __func__, hdr->inode->i_ino,
479                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
480
481         /* Retrieve the correct rpc_client for the byte range */
482         j = nfs4_fl_calc_j_index(lseg, offset);
483         idx = nfs4_fl_calc_ds_index(lseg, j);
484         ds = nfs4_fl_prepare_ds(lseg, idx);
485         if (!ds)
486                 return PNFS_NOT_ATTEMPTED;
487
488         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
489         if (IS_ERR(ds_clnt))
490                 return PNFS_NOT_ATTEMPTED;
491
492         dprintk("%s USE DS: %s cl_count %d\n", __func__,
493                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
494
495         /* No multipath support. Use first DS */
496         atomic_inc(&ds->ds_clp->cl_count);
497         hdr->ds_clp = ds->ds_clp;
498         hdr->ds_commit_idx = idx;
499         fh = nfs4_fl_select_ds_fh(lseg, j);
500         if (fh)
501                 hdr->args.fh = fh;
502
503         hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
504         hdr->mds_offset = offset;
505
506         /* Perform an asynchronous read to ds */
507         nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
508                           NFS_PROTO(hdr->inode), &filelayout_read_call_ops,
509                           0, RPC_TASK_SOFTCONN);
510         return PNFS_ATTEMPTED;
511 }
512
513 /* Perform async writes. */
514 static enum pnfs_try_status
515 filelayout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
516 {
517         struct pnfs_layout_segment *lseg = hdr->lseg;
518         struct nfs4_pnfs_ds *ds;
519         struct rpc_clnt *ds_clnt;
520         loff_t offset = hdr->args.offset;
521         u32 j, idx;
522         struct nfs_fh *fh;
523
524         /* Retrieve the correct rpc_client for the byte range */
525         j = nfs4_fl_calc_j_index(lseg, offset);
526         idx = nfs4_fl_calc_ds_index(lseg, j);
527         ds = nfs4_fl_prepare_ds(lseg, idx);
528         if (!ds)
529                 return PNFS_NOT_ATTEMPTED;
530
531         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
532         if (IS_ERR(ds_clnt))
533                 return PNFS_NOT_ATTEMPTED;
534
535         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n",
536                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
537                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
538
539         hdr->pgio_done_cb = filelayout_write_done_cb;
540         atomic_inc(&ds->ds_clp->cl_count);
541         hdr->ds_clp = ds->ds_clp;
542         hdr->ds_commit_idx = idx;
543         fh = nfs4_fl_select_ds_fh(lseg, j);
544         if (fh)
545                 hdr->args.fh = fh;
546         hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
547
548         /* Perform an asynchronous write */
549         nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
550                           NFS_PROTO(hdr->inode), &filelayout_write_call_ops,
551                           sync, RPC_TASK_SOFTCONN);
552         return PNFS_ATTEMPTED;
553 }
554
555 /*
556  * filelayout_check_layout()
557  *
558  * Make sure layout segment parameters are sane WRT the device.
559  * At this point no generic layer initialization of the lseg has occurred,
560  * and nothing has been added to the layout_hdr cache.
561  *
562  */
563 static int
564 filelayout_check_layout(struct pnfs_layout_hdr *lo,
565                         struct nfs4_filelayout_segment *fl,
566                         struct nfs4_layoutget_res *lgr,
567                         struct nfs4_deviceid *id,
568                         gfp_t gfp_flags)
569 {
570         struct nfs4_deviceid_node *d;
571         struct nfs4_file_layout_dsaddr *dsaddr;
572         int status = -EINVAL;
573
574         dprintk("--> %s\n", __func__);
575
576         /* FIXME: remove this check when layout segment support is added */
577         if (lgr->range.offset != 0 ||
578             lgr->range.length != NFS4_MAX_UINT64) {
579                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
580                         __func__);
581                 goto out;
582         }
583
584         if (fl->pattern_offset > lgr->range.offset) {
585                 dprintk("%s pattern_offset %lld too large\n",
586                                 __func__, fl->pattern_offset);
587                 goto out;
588         }
589
590         if (!fl->stripe_unit) {
591                 dprintk("%s Invalid stripe unit (%u)\n",
592                         __func__, fl->stripe_unit);
593                 goto out;
594         }
595
596         /* find and reference the deviceid */
597         d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), id,
598                         lo->plh_lc_cred, gfp_flags);
599         if (d == NULL)
600                 goto out;
601
602         dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
603         /* Found deviceid is unavailable */
604         if (filelayout_test_devid_unavailable(&dsaddr->id_node))
605                 goto out_put;
606
607         fl->dsaddr = dsaddr;
608
609         if (fl->first_stripe_index >= dsaddr->stripe_count) {
610                 dprintk("%s Bad first_stripe_index %u\n",
611                                 __func__, fl->first_stripe_index);
612                 goto out_put;
613         }
614
615         if ((fl->stripe_type == STRIPE_SPARSE &&
616             fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
617             (fl->stripe_type == STRIPE_DENSE &&
618             fl->num_fh != dsaddr->stripe_count)) {
619                 dprintk("%s num_fh %u not valid for given packing\n",
620                         __func__, fl->num_fh);
621                 goto out_put;
622         }
623
624         status = 0;
625 out:
626         dprintk("--> %s returns %d\n", __func__, status);
627         return status;
628 out_put:
629         nfs4_fl_put_deviceid(dsaddr);
630         goto out;
631 }
632
633 static void _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
634 {
635         int i;
636
637         if (fl->fh_array) {
638                 for (i = 0; i < fl->num_fh; i++) {
639                         if (!fl->fh_array[i])
640                                 break;
641                         kfree(fl->fh_array[i]);
642                 }
643                 kfree(fl->fh_array);
644         }
645         kfree(fl);
646 }
647
648 static int
649 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
650                          struct nfs4_filelayout_segment *fl,
651                          struct nfs4_layoutget_res *lgr,
652                          struct nfs4_deviceid *id,
653                          gfp_t gfp_flags)
654 {
655         struct xdr_stream stream;
656         struct xdr_buf buf;
657         struct page *scratch;
658         __be32 *p;
659         uint32_t nfl_util;
660         int i;
661
662         dprintk("%s: set_layout_map Begin\n", __func__);
663
664         scratch = alloc_page(gfp_flags);
665         if (!scratch)
666                 return -ENOMEM;
667
668         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
669         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
670
671         /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
672          * num_fh (4) */
673         p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
674         if (unlikely(!p))
675                 goto out_err;
676
677         memcpy(id, p, sizeof(*id));
678         p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
679         nfs4_print_deviceid(id);
680
681         nfl_util = be32_to_cpup(p++);
682         if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
683                 fl->commit_through_mds = 1;
684         if (nfl_util & NFL4_UFLG_DENSE)
685                 fl->stripe_type = STRIPE_DENSE;
686         else
687                 fl->stripe_type = STRIPE_SPARSE;
688         fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
689
690         fl->first_stripe_index = be32_to_cpup(p++);
691         p = xdr_decode_hyper(p, &fl->pattern_offset);
692         fl->num_fh = be32_to_cpup(p++);
693
694         dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
695                 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
696                 fl->pattern_offset);
697
698         /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
699          * Futher checking is done in filelayout_check_layout */
700         if (fl->num_fh >
701             max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
702                 goto out_err;
703
704         if (fl->num_fh > 0) {
705                 fl->fh_array = kcalloc(fl->num_fh, sizeof(fl->fh_array[0]),
706                                        gfp_flags);
707                 if (!fl->fh_array)
708                         goto out_err;
709         }
710
711         for (i = 0; i < fl->num_fh; i++) {
712                 /* Do we want to use a mempool here? */
713                 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
714                 if (!fl->fh_array[i])
715                         goto out_err;
716
717                 p = xdr_inline_decode(&stream, 4);
718                 if (unlikely(!p))
719                         goto out_err;
720                 fl->fh_array[i]->size = be32_to_cpup(p++);
721                 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
722                         printk(KERN_ERR "NFS: Too big fh %d received %d\n",
723                                i, fl->fh_array[i]->size);
724                         goto out_err;
725                 }
726
727                 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
728                 if (unlikely(!p))
729                         goto out_err;
730                 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
731                 dprintk("DEBUG: %s: fh len %d\n", __func__,
732                         fl->fh_array[i]->size);
733         }
734
735         __free_page(scratch);
736         return 0;
737
738 out_err:
739         __free_page(scratch);
740         return -EIO;
741 }
742
743 static void
744 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
745 {
746         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
747
748         dprintk("--> %s\n", __func__);
749         nfs4_fl_put_deviceid(fl->dsaddr);
750         /* This assumes a single RW lseg */
751         if (lseg->pls_range.iomode == IOMODE_RW) {
752                 struct nfs4_filelayout *flo;
753
754                 flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
755                 flo->commit_info.nbuckets = 0;
756                 kfree(flo->commit_info.buckets);
757                 flo->commit_info.buckets = NULL;
758         }
759         _filelayout_free_lseg(fl);
760 }
761
762 static int
763 filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg,
764                              struct nfs_commit_info *cinfo,
765                              gfp_t gfp_flags)
766 {
767         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
768         struct pnfs_commit_bucket *buckets;
769         int size, i;
770
771         if (fl->commit_through_mds)
772                 return 0;
773
774         size = (fl->stripe_type == STRIPE_SPARSE) ?
775                 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
776
777         if (cinfo->ds->nbuckets >= size) {
778                 /* This assumes there is only one IOMODE_RW lseg.  What
779                  * we really want to do is have a layout_hdr level
780                  * dictionary of <multipath_list4, fh> keys, each
781                  * associated with a struct list_head, populated by calls
782                  * to filelayout_write_pagelist().
783                  * */
784                 return 0;
785         }
786
787         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
788                           gfp_flags);
789         if (!buckets)
790                 return -ENOMEM;
791         for (i = 0; i < size; i++) {
792                 INIT_LIST_HEAD(&buckets[i].written);
793                 INIT_LIST_HEAD(&buckets[i].committing);
794                 /* mark direct verifier as unset */
795                 buckets[i].direct_verf.committed = NFS_INVALID_STABLE_HOW;
796         }
797
798         spin_lock(cinfo->lock);
799         if (cinfo->ds->nbuckets >= size)
800                 goto out;
801         for (i = 0; i < cinfo->ds->nbuckets; i++) {
802                 list_splice(&cinfo->ds->buckets[i].written,
803                             &buckets[i].written);
804                 list_splice(&cinfo->ds->buckets[i].committing,
805                             &buckets[i].committing);
806                 buckets[i].direct_verf.committed =
807                         cinfo->ds->buckets[i].direct_verf.committed;
808                 buckets[i].wlseg = cinfo->ds->buckets[i].wlseg;
809                 buckets[i].clseg = cinfo->ds->buckets[i].clseg;
810         }
811         swap(cinfo->ds->buckets, buckets);
812         cinfo->ds->nbuckets = size;
813 out:
814         spin_unlock(cinfo->lock);
815         kfree(buckets);
816         return 0;
817 }
818
819 static struct pnfs_layout_segment *
820 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
821                       struct nfs4_layoutget_res *lgr,
822                       gfp_t gfp_flags)
823 {
824         struct nfs4_filelayout_segment *fl;
825         int rc;
826         struct nfs4_deviceid id;
827
828         dprintk("--> %s\n", __func__);
829         fl = kzalloc(sizeof(*fl), gfp_flags);
830         if (!fl)
831                 return NULL;
832
833         rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
834         if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
835                 _filelayout_free_lseg(fl);
836                 return NULL;
837         }
838         return &fl->generic_hdr;
839 }
840
841 /*
842  * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
843  *
844  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
845  * of bytes (maximum @req->wb_bytes) that can be coalesced.
846  */
847 static size_t
848 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
849                    struct nfs_page *req)
850 {
851         unsigned int size;
852         u64 p_stripe, r_stripe;
853         u32 stripe_offset;
854         u64 segment_offset = pgio->pg_lseg->pls_range.offset;
855         u32 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
856
857         /* calls nfs_generic_pg_test */
858         size = pnfs_generic_pg_test(pgio, prev, req);
859         if (!size)
860                 return 0;
861
862         /* see if req and prev are in the same stripe */
863         if (prev) {
864                 p_stripe = (u64)req_offset(prev) - segment_offset;
865                 r_stripe = (u64)req_offset(req) - segment_offset;
866                 do_div(p_stripe, stripe_unit);
867                 do_div(r_stripe, stripe_unit);
868
869                 if (p_stripe != r_stripe)
870                         return 0;
871         }
872
873         /* calculate remaining bytes in the current stripe */
874         div_u64_rem((u64)req_offset(req) - segment_offset,
875                         stripe_unit,
876                         &stripe_offset);
877         WARN_ON_ONCE(stripe_offset > stripe_unit);
878         if (stripe_offset >= stripe_unit)
879                 return 0;
880         return min(stripe_unit - (unsigned int)stripe_offset, size);
881 }
882
883 static void
884 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
885                         struct nfs_page *req)
886 {
887         if (!pgio->pg_lseg) {
888                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
889                                            req->wb_context,
890                                            0,
891                                            NFS4_MAX_UINT64,
892                                            IOMODE_READ,
893                                            GFP_KERNEL);
894                 if (IS_ERR(pgio->pg_lseg)) {
895                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
896                         pgio->pg_lseg = NULL;
897                         return;
898                 }
899         }
900         /* If no lseg, fall back to read through mds */
901         if (pgio->pg_lseg == NULL)
902                 nfs_pageio_reset_read_mds(pgio);
903 }
904
905 static void
906 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
907                          struct nfs_page *req)
908 {
909         struct nfs_commit_info cinfo;
910         int status;
911
912         if (!pgio->pg_lseg) {
913                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
914                                            req->wb_context,
915                                            0,
916                                            NFS4_MAX_UINT64,
917                                            IOMODE_RW,
918                                            GFP_NOFS);
919                 if (IS_ERR(pgio->pg_lseg)) {
920                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
921                         pgio->pg_lseg = NULL;
922                         return;
923                 }
924         }
925
926         /* If no lseg, fall back to write through mds */
927         if (pgio->pg_lseg == NULL)
928                 goto out_mds;
929         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
930         status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
931         if (status < 0) {
932                 pnfs_put_lseg(pgio->pg_lseg);
933                 pgio->pg_lseg = NULL;
934                 goto out_mds;
935         }
936         return;
937 out_mds:
938         nfs_pageio_reset_write_mds(pgio);
939 }
940
941 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
942         .pg_init = filelayout_pg_init_read,
943         .pg_test = filelayout_pg_test,
944         .pg_doio = pnfs_generic_pg_readpages,
945         .pg_cleanup = pnfs_generic_pg_cleanup,
946 };
947
948 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
949         .pg_init = filelayout_pg_init_write,
950         .pg_test = filelayout_pg_test,
951         .pg_doio = pnfs_generic_pg_writepages,
952         .pg_cleanup = pnfs_generic_pg_cleanup,
953 };
954
955 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
956 {
957         if (fl->stripe_type == STRIPE_SPARSE)
958                 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
959         else
960                 return j;
961 }
962
963 static void
964 filelayout_mark_request_commit(struct nfs_page *req,
965                                struct pnfs_layout_segment *lseg,
966                                struct nfs_commit_info *cinfo,
967                                u32 ds_commit_idx)
968
969 {
970         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
971         u32 i, j;
972
973         if (fl->commit_through_mds) {
974                 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
975         } else {
976                 /* Note that we are calling nfs4_fl_calc_j_index on each page
977                  * that ends up being committed to a data server.  An attractive
978                  * alternative is to add a field to nfs_write_data and nfs_page
979                  * to store the value calculated in filelayout_write_pagelist
980                  * and just use that here.
981                  */
982                 j = nfs4_fl_calc_j_index(lseg, req_offset(req));
983                 i = select_bucket_index(fl, j);
984                 pnfs_layout_mark_request_commit(req, lseg, cinfo, i);
985         }
986 }
987
988 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
989 {
990         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
991
992         if (flseg->stripe_type == STRIPE_SPARSE)
993                 return i;
994         else
995                 return nfs4_fl_calc_ds_index(lseg, i);
996 }
997
998 static struct nfs_fh *
999 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1000 {
1001         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1002
1003         if (flseg->stripe_type == STRIPE_SPARSE) {
1004                 if (flseg->num_fh == 1)
1005                         i = 0;
1006                 else if (flseg->num_fh == 0)
1007                         /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1008                         return NULL;
1009         }
1010         return flseg->fh_array[i];
1011 }
1012
1013 static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
1014 {
1015         struct pnfs_layout_segment *lseg = data->lseg;
1016         struct nfs4_pnfs_ds *ds;
1017         struct rpc_clnt *ds_clnt;
1018         u32 idx;
1019         struct nfs_fh *fh;
1020
1021         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1022         ds = nfs4_fl_prepare_ds(lseg, idx);
1023         if (!ds)
1024                 goto out_err;
1025
1026         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, data->inode);
1027         if (IS_ERR(ds_clnt))
1028                 goto out_err;
1029
1030         dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1031                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count));
1032         data->commit_done_cb = filelayout_commit_done_cb;
1033         atomic_inc(&ds->ds_clp->cl_count);
1034         data->ds_clp = ds->ds_clp;
1035         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1036         if (fh)
1037                 data->args.fh = fh;
1038         return nfs_initiate_commit(ds_clnt, data, NFS_PROTO(data->inode),
1039                                    &filelayout_commit_call_ops, how,
1040                                    RPC_TASK_SOFTCONN);
1041 out_err:
1042         pnfs_generic_prepare_to_resend_writes(data);
1043         pnfs_generic_commit_release(data);
1044         return -EAGAIN;
1045 }
1046
1047 /* filelayout_search_commit_reqs - Search lists in @cinfo for the head reqest
1048  *                                 for @page
1049  * @cinfo - commit info for current inode
1050  * @page - page to search for matching head request
1051  *
1052  * Returns a the head request if one is found, otherwise returns NULL.
1053  */
1054 static struct nfs_page *
1055 filelayout_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page)
1056 {
1057         struct nfs_page *freq, *t;
1058         struct pnfs_commit_bucket *b;
1059         int i;
1060
1061         /* Linearly search the commit lists for each bucket until a matching
1062          * request is found */
1063         for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
1064                 list_for_each_entry_safe(freq, t, &b->written, wb_list) {
1065                         if (freq->wb_page == page)
1066                                 return freq->wb_head;
1067                 }
1068                 list_for_each_entry_safe(freq, t, &b->committing, wb_list) {
1069                         if (freq->wb_page == page)
1070                                 return freq->wb_head;
1071                 }
1072         }
1073
1074         return NULL;
1075 }
1076
1077 static int
1078 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1079                            int how, struct nfs_commit_info *cinfo)
1080 {
1081         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1082                                             filelayout_initiate_commit);
1083 }
1084
1085 static struct nfs4_deviceid_node *
1086 filelayout_alloc_deviceid_node(struct nfs_server *server,
1087                 struct pnfs_device *pdev, gfp_t gfp_flags)
1088 {
1089         struct nfs4_file_layout_dsaddr *dsaddr;
1090
1091         dsaddr = nfs4_fl_alloc_deviceid_node(server, pdev, gfp_flags);
1092         if (!dsaddr)
1093                 return NULL;
1094         return &dsaddr->id_node;
1095 }
1096
1097 static void
1098 filelayout_free_deviceid_node(struct nfs4_deviceid_node *d)
1099 {
1100         nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1101 }
1102
1103 static struct pnfs_layout_hdr *
1104 filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1105 {
1106         struct nfs4_filelayout *flo;
1107
1108         flo = kzalloc(sizeof(*flo), gfp_flags);
1109         return flo != NULL ? &flo->generic_hdr : NULL;
1110 }
1111
1112 static void
1113 filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1114 {
1115         kfree(FILELAYOUT_FROM_HDR(lo));
1116 }
1117
1118 static struct pnfs_ds_commit_info *
1119 filelayout_get_ds_info(struct inode *inode)
1120 {
1121         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1122
1123         if (layout == NULL)
1124                 return NULL;
1125         else
1126                 return &FILELAYOUT_FROM_HDR(layout)->commit_info;
1127 }
1128
1129 static struct pnfs_layoutdriver_type filelayout_type = {
1130         .id                     = LAYOUT_NFSV4_1_FILES,
1131         .name                   = "LAYOUT_NFSV4_1_FILES",
1132         .owner                  = THIS_MODULE,
1133         .alloc_layout_hdr       = filelayout_alloc_layout_hdr,
1134         .free_layout_hdr        = filelayout_free_layout_hdr,
1135         .alloc_lseg             = filelayout_alloc_lseg,
1136         .free_lseg              = filelayout_free_lseg,
1137         .pg_read_ops            = &filelayout_pg_read_ops,
1138         .pg_write_ops           = &filelayout_pg_write_ops,
1139         .get_ds_info            = &filelayout_get_ds_info,
1140         .mark_request_commit    = filelayout_mark_request_commit,
1141         .clear_request_commit   = pnfs_generic_clear_request_commit,
1142         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
1143         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
1144         .search_commit_reqs     = filelayout_search_commit_reqs,
1145         .commit_pagelist        = filelayout_commit_pagelist,
1146         .read_pagelist          = filelayout_read_pagelist,
1147         .write_pagelist         = filelayout_write_pagelist,
1148         .alloc_deviceid_node    = filelayout_alloc_deviceid_node,
1149         .free_deviceid_node     = filelayout_free_deviceid_node,
1150         .sync                   = pnfs_nfs_generic_sync,
1151 };
1152
1153 static int __init nfs4filelayout_init(void)
1154 {
1155         printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1156                __func__);
1157         return pnfs_register_layoutdriver(&filelayout_type);
1158 }
1159
1160 static void __exit nfs4filelayout_exit(void)
1161 {
1162         printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1163                __func__);
1164         pnfs_unregister_layoutdriver(&filelayout_type);
1165 }
1166
1167 MODULE_ALIAS("nfs-layouttype4-1");
1168
1169 module_init(nfs4filelayout_init);
1170 module_exit(nfs4filelayout_exit);