pnfs: Prepare for flexfiles by pulling out common code
[firefly-linux-kernel-4.4.55.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
36 #include <linux/sunrpc/metrics.h>
37
38 #include "../nfs4session.h"
39 #include "../internal.h"
40 #include "../delegation.h"
41 #include "filelayout.h"
42 #include "../nfs4trace.h"
43
44 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
45
46 MODULE_LICENSE("GPL");
47 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
48 MODULE_DESCRIPTION("The NFSv4 file layout driver");
49
50 #define FILELAYOUT_POLL_RETRY_MAX     (15*HZ)
51
52 static loff_t
53 filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
54                             loff_t offset)
55 {
56         u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
57         u64 stripe_no;
58         u32 rem;
59
60         offset -= flseg->pattern_offset;
61         stripe_no = div_u64(offset, stripe_width);
62         div_u64_rem(offset, flseg->stripe_unit, &rem);
63
64         return stripe_no * flseg->stripe_unit + rem;
65 }
66
67 /* This function is used by the layout driver to calculate the
68  * offset of the file on the dserver based on whether the
69  * layout type is STRIPE_DENSE or STRIPE_SPARSE
70  */
71 static loff_t
72 filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
73 {
74         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
75
76         switch (flseg->stripe_type) {
77         case STRIPE_SPARSE:
78                 return offset;
79
80         case STRIPE_DENSE:
81                 return filelayout_get_dense_offset(flseg, offset);
82         }
83
84         BUG();
85 }
86
87 static void filelayout_reset_write(struct nfs_pgio_header *hdr)
88 {
89         struct rpc_task *task = &hdr->task;
90
91         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
92                 dprintk("%s Reset task %5u for i/o through MDS "
93                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
94                         hdr->task.tk_pid,
95                         hdr->inode->i_sb->s_id,
96                         (unsigned long long)NFS_FILEID(hdr->inode),
97                         hdr->args.count,
98                         (unsigned long long)hdr->args.offset);
99
100                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
101         }
102 }
103
104 static void filelayout_reset_read(struct nfs_pgio_header *hdr)
105 {
106         struct rpc_task *task = &hdr->task;
107
108         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
109                 dprintk("%s Reset task %5u for i/o through MDS "
110                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
111                         hdr->task.tk_pid,
112                         hdr->inode->i_sb->s_id,
113                         (unsigned long long)NFS_FILEID(hdr->inode),
114                         hdr->args.count,
115                         (unsigned long long)hdr->args.offset);
116
117                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
118         }
119 }
120
121 static int filelayout_async_handle_error(struct rpc_task *task,
122                                          struct nfs4_state *state,
123                                          struct nfs_client *clp,
124                                          struct pnfs_layout_segment *lseg)
125 {
126         struct pnfs_layout_hdr *lo = lseg->pls_layout;
127         struct inode *inode = lo->plh_inode;
128         struct nfs_server *mds_server = NFS_SERVER(inode);
129         struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
130         struct nfs_client *mds_client = mds_server->nfs_client;
131         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
132
133         if (task->tk_status >= 0)
134                 return 0;
135
136         switch (task->tk_status) {
137         /* MDS state errors */
138         case -NFS4ERR_DELEG_REVOKED:
139         case -NFS4ERR_ADMIN_REVOKED:
140         case -NFS4ERR_BAD_STATEID:
141         case -NFS4ERR_OPENMODE:
142                 if (state == NULL)
143                         break;
144                 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
145                         goto out_bad_stateid;
146                 goto wait_on_recovery;
147         case -NFS4ERR_EXPIRED:
148                 if (state != NULL) {
149                         if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
150                                 goto out_bad_stateid;
151                 }
152                 nfs4_schedule_lease_recovery(mds_client);
153                 goto wait_on_recovery;
154         /* DS session errors */
155         case -NFS4ERR_BADSESSION:
156         case -NFS4ERR_BADSLOT:
157         case -NFS4ERR_BAD_HIGH_SLOT:
158         case -NFS4ERR_DEADSESSION:
159         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
160         case -NFS4ERR_SEQ_FALSE_RETRY:
161         case -NFS4ERR_SEQ_MISORDERED:
162                 dprintk("%s ERROR %d, Reset session. Exchangeid "
163                         "flags 0x%x\n", __func__, task->tk_status,
164                         clp->cl_exchange_flags);
165                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
166                 break;
167         case -NFS4ERR_DELAY:
168         case -NFS4ERR_GRACE:
169                 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
170                 break;
171         case -NFS4ERR_RETRY_UNCACHED_REP:
172                 break;
173         /* Invalidate Layout errors */
174         case -NFS4ERR_PNFS_NO_LAYOUT:
175         case -ESTALE:           /* mapped NFS4ERR_STALE */
176         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
177         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
178         case -NFS4ERR_FHEXPIRED:
179         case -NFS4ERR_WRONG_TYPE:
180                 dprintk("%s Invalid layout error %d\n", __func__,
181                         task->tk_status);
182                 /*
183                  * Destroy layout so new i/o will get a new layout.
184                  * Layout will not be destroyed until all current lseg
185                  * references are put. Mark layout as invalid to resend failed
186                  * i/o and all i/o waiting on the slot table to the MDS until
187                  * layout is destroyed and a new valid layout is obtained.
188                  */
189                 pnfs_destroy_layout(NFS_I(inode));
190                 rpc_wake_up(&tbl->slot_tbl_waitq);
191                 goto reset;
192         /* RPC connection errors */
193         case -ECONNREFUSED:
194         case -EHOSTDOWN:
195         case -EHOSTUNREACH:
196         case -ENETUNREACH:
197         case -EIO:
198         case -ETIMEDOUT:
199         case -EPIPE:
200                 dprintk("%s DS connection error %d\n", __func__,
201                         task->tk_status);
202                 nfs4_mark_deviceid_unavailable(devid);
203                 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
204                 rpc_wake_up(&tbl->slot_tbl_waitq);
205                 /* fall through */
206         default:
207 reset:
208                 dprintk("%s Retry through MDS. Error %d\n", __func__,
209                         task->tk_status);
210                 return -NFS4ERR_RESET_TO_MDS;
211         }
212 out:
213         task->tk_status = 0;
214         return -EAGAIN;
215 out_bad_stateid:
216         task->tk_status = -EIO;
217         return 0;
218 wait_on_recovery:
219         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
220         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
221                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
222         goto out;
223 }
224
225 /* NFS_PROTO call done callback routines */
226
227 static int filelayout_read_done_cb(struct rpc_task *task,
228                                 struct nfs_pgio_header *hdr)
229 {
230         int err;
231
232         trace_nfs4_pnfs_read(hdr, task->tk_status);
233         err = filelayout_async_handle_error(task, hdr->args.context->state,
234                                             hdr->ds_clp, hdr->lseg);
235
236         switch (err) {
237         case -NFS4ERR_RESET_TO_MDS:
238                 filelayout_reset_read(hdr);
239                 return task->tk_status;
240         case -EAGAIN:
241                 rpc_restart_call_prepare(task);
242                 return -EAGAIN;
243         }
244
245         return 0;
246 }
247
248 /*
249  * We reference the rpc_cred of the first WRITE that triggers the need for
250  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
251  * rfc5661 is not clear about which credential should be used.
252  */
253 static void
254 filelayout_set_layoutcommit(struct nfs_pgio_header *hdr)
255 {
256
257         if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds ||
258             hdr->res.verf->committed != NFS_DATA_SYNC)
259                 return;
260
261         pnfs_set_layoutcommit(hdr);
262         dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
263                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
264 }
265
266 bool
267 filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node)
268 {
269         return filelayout_test_devid_invalid(node) ||
270                 nfs4_test_deviceid_unavailable(node);
271 }
272
273 static bool
274 filelayout_reset_to_mds(struct pnfs_layout_segment *lseg)
275 {
276         struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg);
277
278         return filelayout_test_devid_unavailable(node);
279 }
280
281 /*
282  * Call ops for the async read/write cases
283  * In the case of dense layouts, the offset needs to be reset to its
284  * original value.
285  */
286 static void filelayout_read_prepare(struct rpc_task *task, void *data)
287 {
288         struct nfs_pgio_header *hdr = data;
289
290         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
291                 rpc_exit(task, -EIO);
292                 return;
293         }
294         if (filelayout_reset_to_mds(hdr->lseg)) {
295                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
296                 filelayout_reset_read(hdr);
297                 rpc_exit(task, 0);
298                 return;
299         }
300         hdr->pgio_done_cb = filelayout_read_done_cb;
301
302         if (nfs41_setup_sequence(hdr->ds_clp->cl_session,
303                         &hdr->args.seq_args,
304                         &hdr->res.seq_res,
305                         task))
306                 return;
307         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
308                         hdr->args.lock_context, FMODE_READ) == -EIO)
309                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
310 }
311
312 static void filelayout_read_call_done(struct rpc_task *task, void *data)
313 {
314         struct nfs_pgio_header *hdr = data;
315
316         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
317
318         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
319             task->tk_status == 0) {
320                 nfs41_sequence_done(task, &hdr->res.seq_res);
321                 return;
322         }
323
324         /* Note this may cause RPC to be resent */
325         hdr->mds_ops->rpc_call_done(task, data);
326 }
327
328 static void filelayout_read_count_stats(struct rpc_task *task, void *data)
329 {
330         struct nfs_pgio_header *hdr = data;
331
332         rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
333 }
334
335 static int filelayout_write_done_cb(struct rpc_task *task,
336                                 struct nfs_pgio_header *hdr)
337 {
338         int err;
339
340         trace_nfs4_pnfs_write(hdr, task->tk_status);
341         err = filelayout_async_handle_error(task, hdr->args.context->state,
342                                             hdr->ds_clp, hdr->lseg);
343
344         switch (err) {
345         case -NFS4ERR_RESET_TO_MDS:
346                 filelayout_reset_write(hdr);
347                 return task->tk_status;
348         case -EAGAIN:
349                 rpc_restart_call_prepare(task);
350                 return -EAGAIN;
351         }
352
353         filelayout_set_layoutcommit(hdr);
354         return 0;
355 }
356
357 static int filelayout_commit_done_cb(struct rpc_task *task,
358                                      struct nfs_commit_data *data)
359 {
360         int err;
361
362         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
363         err = filelayout_async_handle_error(task, NULL, data->ds_clp,
364                                             data->lseg);
365
366         switch (err) {
367         case -NFS4ERR_RESET_TO_MDS:
368                 pnfs_generic_prepare_to_resend_writes(data);
369                 return -EAGAIN;
370         case -EAGAIN:
371                 rpc_restart_call_prepare(task);
372                 return -EAGAIN;
373         }
374
375         if (data->verf.committed == NFS_UNSTABLE)
376                 pnfs_commit_set_layoutcommit(data);
377
378         return 0;
379 }
380
381 static void filelayout_write_prepare(struct rpc_task *task, void *data)
382 {
383         struct nfs_pgio_header *hdr = data;
384
385         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
386                 rpc_exit(task, -EIO);
387                 return;
388         }
389         if (filelayout_reset_to_mds(hdr->lseg)) {
390                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
391                 filelayout_reset_write(hdr);
392                 rpc_exit(task, 0);
393                 return;
394         }
395         if (nfs41_setup_sequence(hdr->ds_clp->cl_session,
396                         &hdr->args.seq_args,
397                         &hdr->res.seq_res,
398                         task))
399                 return;
400         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
401                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
402                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
403 }
404
405 static void filelayout_write_call_done(struct rpc_task *task, void *data)
406 {
407         struct nfs_pgio_header *hdr = data;
408
409         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
410             task->tk_status == 0) {
411                 nfs41_sequence_done(task, &hdr->res.seq_res);
412                 return;
413         }
414
415         /* Note this may cause RPC to be resent */
416         hdr->mds_ops->rpc_call_done(task, data);
417 }
418
419 static void filelayout_write_count_stats(struct rpc_task *task, void *data)
420 {
421         struct nfs_pgio_header *hdr = data;
422
423         rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
424 }
425
426 static void filelayout_commit_prepare(struct rpc_task *task, void *data)
427 {
428         struct nfs_commit_data *wdata = data;
429
430         nfs41_setup_sequence(wdata->ds_clp->cl_session,
431                         &wdata->args.seq_args,
432                         &wdata->res.seq_res,
433                         task);
434 }
435
436 static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
437 {
438         struct nfs_commit_data *cdata = data;
439
440         rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
441 }
442
443 static const struct rpc_call_ops filelayout_read_call_ops = {
444         .rpc_call_prepare = filelayout_read_prepare,
445         .rpc_call_done = filelayout_read_call_done,
446         .rpc_count_stats = filelayout_read_count_stats,
447         .rpc_release = pnfs_generic_rw_release,
448 };
449
450 static const struct rpc_call_ops filelayout_write_call_ops = {
451         .rpc_call_prepare = filelayout_write_prepare,
452         .rpc_call_done = filelayout_write_call_done,
453         .rpc_count_stats = filelayout_write_count_stats,
454         .rpc_release = pnfs_generic_rw_release,
455 };
456
457 static const struct rpc_call_ops filelayout_commit_call_ops = {
458         .rpc_call_prepare = filelayout_commit_prepare,
459         .rpc_call_done = pnfs_generic_write_commit_done,
460         .rpc_count_stats = filelayout_commit_count_stats,
461         .rpc_release = pnfs_generic_commit_release,
462 };
463
464 static enum pnfs_try_status
465 filelayout_read_pagelist(struct nfs_pgio_header *hdr)
466 {
467         struct pnfs_layout_segment *lseg = hdr->lseg;
468         struct nfs4_pnfs_ds *ds;
469         struct rpc_clnt *ds_clnt;
470         loff_t offset = hdr->args.offset;
471         u32 j, idx;
472         struct nfs_fh *fh;
473
474         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
475                 __func__, hdr->inode->i_ino,
476                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
477
478         /* Retrieve the correct rpc_client for the byte range */
479         j = nfs4_fl_calc_j_index(lseg, offset);
480         idx = nfs4_fl_calc_ds_index(lseg, j);
481         ds = nfs4_fl_prepare_ds(lseg, idx);
482         if (!ds)
483                 return PNFS_NOT_ATTEMPTED;
484
485         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
486         if (IS_ERR(ds_clnt))
487                 return PNFS_NOT_ATTEMPTED;
488
489         dprintk("%s USE DS: %s cl_count %d\n", __func__,
490                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
491
492         /* No multipath support. Use first DS */
493         atomic_inc(&ds->ds_clp->cl_count);
494         hdr->ds_clp = ds->ds_clp;
495         hdr->ds_idx = idx;
496         fh = nfs4_fl_select_ds_fh(lseg, j);
497         if (fh)
498                 hdr->args.fh = fh;
499
500         hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
501         hdr->mds_offset = offset;
502
503         /* Perform an asynchronous read to ds */
504         nfs_initiate_pgio(ds_clnt, hdr,
505                             &filelayout_read_call_ops, 0, RPC_TASK_SOFTCONN);
506         return PNFS_ATTEMPTED;
507 }
508
509 /* Perform async writes. */
510 static enum pnfs_try_status
511 filelayout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
512 {
513         struct pnfs_layout_segment *lseg = hdr->lseg;
514         struct nfs4_pnfs_ds *ds;
515         struct rpc_clnt *ds_clnt;
516         loff_t offset = hdr->args.offset;
517         u32 j, idx;
518         struct nfs_fh *fh;
519
520         /* Retrieve the correct rpc_client for the byte range */
521         j = nfs4_fl_calc_j_index(lseg, offset);
522         idx = nfs4_fl_calc_ds_index(lseg, j);
523         ds = nfs4_fl_prepare_ds(lseg, idx);
524         if (!ds)
525                 return PNFS_NOT_ATTEMPTED;
526
527         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
528         if (IS_ERR(ds_clnt))
529                 return PNFS_NOT_ATTEMPTED;
530
531         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n",
532                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
533                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
534
535         hdr->pgio_done_cb = filelayout_write_done_cb;
536         atomic_inc(&ds->ds_clp->cl_count);
537         hdr->ds_clp = ds->ds_clp;
538         hdr->ds_idx = idx;
539         fh = nfs4_fl_select_ds_fh(lseg, j);
540         if (fh)
541                 hdr->args.fh = fh;
542         hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
543
544         /* Perform an asynchronous write */
545         nfs_initiate_pgio(ds_clnt, hdr,
546                                     &filelayout_write_call_ops, sync,
547                                     RPC_TASK_SOFTCONN);
548         return PNFS_ATTEMPTED;
549 }
550
551 /*
552  * filelayout_check_layout()
553  *
554  * Make sure layout segment parameters are sane WRT the device.
555  * At this point no generic layer initialization of the lseg has occurred,
556  * and nothing has been added to the layout_hdr cache.
557  *
558  */
559 static int
560 filelayout_check_layout(struct pnfs_layout_hdr *lo,
561                         struct nfs4_filelayout_segment *fl,
562                         struct nfs4_layoutget_res *lgr,
563                         struct nfs4_deviceid *id,
564                         gfp_t gfp_flags)
565 {
566         struct nfs4_deviceid_node *d;
567         struct nfs4_file_layout_dsaddr *dsaddr;
568         int status = -EINVAL;
569
570         dprintk("--> %s\n", __func__);
571
572         /* FIXME: remove this check when layout segment support is added */
573         if (lgr->range.offset != 0 ||
574             lgr->range.length != NFS4_MAX_UINT64) {
575                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
576                         __func__);
577                 goto out;
578         }
579
580         if (fl->pattern_offset > lgr->range.offset) {
581                 dprintk("%s pattern_offset %lld too large\n",
582                                 __func__, fl->pattern_offset);
583                 goto out;
584         }
585
586         if (!fl->stripe_unit) {
587                 dprintk("%s Invalid stripe unit (%u)\n",
588                         __func__, fl->stripe_unit);
589                 goto out;
590         }
591
592         /* find and reference the deviceid */
593         d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), id,
594                         lo->plh_lc_cred, gfp_flags);
595         if (d == NULL)
596                 goto out;
597
598         dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
599         /* Found deviceid is unavailable */
600         if (filelayout_test_devid_unavailable(&dsaddr->id_node))
601                 goto out_put;
602
603         fl->dsaddr = dsaddr;
604
605         if (fl->first_stripe_index >= dsaddr->stripe_count) {
606                 dprintk("%s Bad first_stripe_index %u\n",
607                                 __func__, fl->first_stripe_index);
608                 goto out_put;
609         }
610
611         if ((fl->stripe_type == STRIPE_SPARSE &&
612             fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
613             (fl->stripe_type == STRIPE_DENSE &&
614             fl->num_fh != dsaddr->stripe_count)) {
615                 dprintk("%s num_fh %u not valid for given packing\n",
616                         __func__, fl->num_fh);
617                 goto out_put;
618         }
619
620         status = 0;
621 out:
622         dprintk("--> %s returns %d\n", __func__, status);
623         return status;
624 out_put:
625         nfs4_fl_put_deviceid(dsaddr);
626         goto out;
627 }
628
629 static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
630 {
631         int i;
632
633         for (i = 0; i < fl->num_fh; i++) {
634                 if (!fl->fh_array[i])
635                         break;
636                 kfree(fl->fh_array[i]);
637         }
638         kfree(fl->fh_array);
639         fl->fh_array = NULL;
640 }
641
642 static void
643 _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
644 {
645         filelayout_free_fh_array(fl);
646         kfree(fl);
647 }
648
649 static int
650 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
651                          struct nfs4_filelayout_segment *fl,
652                          struct nfs4_layoutget_res *lgr,
653                          struct nfs4_deviceid *id,
654                          gfp_t gfp_flags)
655 {
656         struct xdr_stream stream;
657         struct xdr_buf buf;
658         struct page *scratch;
659         __be32 *p;
660         uint32_t nfl_util;
661         int i;
662
663         dprintk("%s: set_layout_map Begin\n", __func__);
664
665         scratch = alloc_page(gfp_flags);
666         if (!scratch)
667                 return -ENOMEM;
668
669         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
670         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
671
672         /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
673          * num_fh (4) */
674         p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
675         if (unlikely(!p))
676                 goto out_err;
677
678         memcpy(id, p, sizeof(*id));
679         p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
680         nfs4_print_deviceid(id);
681
682         nfl_util = be32_to_cpup(p++);
683         if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
684                 fl->commit_through_mds = 1;
685         if (nfl_util & NFL4_UFLG_DENSE)
686                 fl->stripe_type = STRIPE_DENSE;
687         else
688                 fl->stripe_type = STRIPE_SPARSE;
689         fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
690
691         fl->first_stripe_index = be32_to_cpup(p++);
692         p = xdr_decode_hyper(p, &fl->pattern_offset);
693         fl->num_fh = be32_to_cpup(p++);
694
695         dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
696                 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
697                 fl->pattern_offset);
698
699         /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
700          * Futher checking is done in filelayout_check_layout */
701         if (fl->num_fh >
702             max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
703                 goto out_err;
704
705         if (fl->num_fh > 0) {
706                 fl->fh_array = kcalloc(fl->num_fh, sizeof(fl->fh_array[0]),
707                                        gfp_flags);
708                 if (!fl->fh_array)
709                         goto out_err;
710         }
711
712         for (i = 0; i < fl->num_fh; i++) {
713                 /* Do we want to use a mempool here? */
714                 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
715                 if (!fl->fh_array[i])
716                         goto out_err_free;
717
718                 p = xdr_inline_decode(&stream, 4);
719                 if (unlikely(!p))
720                         goto out_err_free;
721                 fl->fh_array[i]->size = be32_to_cpup(p++);
722                 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
723                         printk(KERN_ERR "NFS: Too big fh %d received %d\n",
724                                i, fl->fh_array[i]->size);
725                         goto out_err_free;
726                 }
727
728                 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
729                 if (unlikely(!p))
730                         goto out_err_free;
731                 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
732                 dprintk("DEBUG: %s: fh len %d\n", __func__,
733                         fl->fh_array[i]->size);
734         }
735
736         __free_page(scratch);
737         return 0;
738
739 out_err_free:
740         filelayout_free_fh_array(fl);
741 out_err:
742         __free_page(scratch);
743         return -EIO;
744 }
745
746 static void
747 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
748 {
749         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
750
751         dprintk("--> %s\n", __func__);
752         nfs4_fl_put_deviceid(fl->dsaddr);
753         /* This assumes a single RW lseg */
754         if (lseg->pls_range.iomode == IOMODE_RW) {
755                 struct nfs4_filelayout *flo;
756
757                 flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
758                 flo->commit_info.nbuckets = 0;
759                 kfree(flo->commit_info.buckets);
760                 flo->commit_info.buckets = NULL;
761         }
762         _filelayout_free_lseg(fl);
763 }
764
765 static int
766 filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg,
767                              struct nfs_commit_info *cinfo,
768                              gfp_t gfp_flags)
769 {
770         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
771         struct pnfs_commit_bucket *buckets;
772         int size, i;
773
774         if (fl->commit_through_mds)
775                 return 0;
776
777         size = (fl->stripe_type == STRIPE_SPARSE) ?
778                 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
779
780         if (cinfo->ds->nbuckets >= size) {
781                 /* This assumes there is only one IOMODE_RW lseg.  What
782                  * we really want to do is have a layout_hdr level
783                  * dictionary of <multipath_list4, fh> keys, each
784                  * associated with a struct list_head, populated by calls
785                  * to filelayout_write_pagelist().
786                  * */
787                 return 0;
788         }
789
790         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
791                           gfp_flags);
792         if (!buckets)
793                 return -ENOMEM;
794         for (i = 0; i < size; i++) {
795                 INIT_LIST_HEAD(&buckets[i].written);
796                 INIT_LIST_HEAD(&buckets[i].committing);
797                 /* mark direct verifier as unset */
798                 buckets[i].direct_verf.committed = NFS_INVALID_STABLE_HOW;
799         }
800
801         spin_lock(cinfo->lock);
802         if (cinfo->ds->nbuckets >= size)
803                 goto out;
804         for (i = 0; i < cinfo->ds->nbuckets; i++) {
805                 list_splice(&cinfo->ds->buckets[i].written,
806                             &buckets[i].written);
807                 list_splice(&cinfo->ds->buckets[i].committing,
808                             &buckets[i].committing);
809                 buckets[i].direct_verf.committed =
810                         cinfo->ds->buckets[i].direct_verf.committed;
811                 buckets[i].wlseg = cinfo->ds->buckets[i].wlseg;
812                 buckets[i].clseg = cinfo->ds->buckets[i].clseg;
813         }
814         swap(cinfo->ds->buckets, buckets);
815         cinfo->ds->nbuckets = size;
816 out:
817         spin_unlock(cinfo->lock);
818         kfree(buckets);
819         return 0;
820 }
821
822 static struct pnfs_layout_segment *
823 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
824                       struct nfs4_layoutget_res *lgr,
825                       gfp_t gfp_flags)
826 {
827         struct nfs4_filelayout_segment *fl;
828         int rc;
829         struct nfs4_deviceid id;
830
831         dprintk("--> %s\n", __func__);
832         fl = kzalloc(sizeof(*fl), gfp_flags);
833         if (!fl)
834                 return NULL;
835
836         rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
837         if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
838                 _filelayout_free_lseg(fl);
839                 return NULL;
840         }
841         return &fl->generic_hdr;
842 }
843
844 /*
845  * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
846  *
847  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
848  * of bytes (maximum @req->wb_bytes) that can be coalesced.
849  */
850 static size_t
851 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
852                    struct nfs_page *req)
853 {
854         unsigned int size;
855         u64 p_stripe, r_stripe;
856         u32 stripe_offset;
857         u64 segment_offset = pgio->pg_lseg->pls_range.offset;
858         u32 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
859
860         /* calls nfs_generic_pg_test */
861         size = pnfs_generic_pg_test(pgio, prev, req);
862         if (!size)
863                 return 0;
864
865         /* see if req and prev are in the same stripe */
866         if (prev) {
867                 p_stripe = (u64)req_offset(prev) - segment_offset;
868                 r_stripe = (u64)req_offset(req) - segment_offset;
869                 do_div(p_stripe, stripe_unit);
870                 do_div(r_stripe, stripe_unit);
871
872                 if (p_stripe != r_stripe)
873                         return 0;
874         }
875
876         /* calculate remaining bytes in the current stripe */
877         div_u64_rem((u64)req_offset(req) - segment_offset,
878                         stripe_unit,
879                         &stripe_offset);
880         WARN_ON_ONCE(stripe_offset > stripe_unit);
881         if (stripe_offset >= stripe_unit)
882                 return 0;
883         return min(stripe_unit - (unsigned int)stripe_offset, size);
884 }
885
886 static void
887 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
888                         struct nfs_page *req)
889 {
890         if (!pgio->pg_lseg)
891                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
892                                            req->wb_context,
893                                            0,
894                                            NFS4_MAX_UINT64,
895                                            IOMODE_READ,
896                                            GFP_KERNEL);
897         /* If no lseg, fall back to read through mds */
898         if (pgio->pg_lseg == NULL)
899                 nfs_pageio_reset_read_mds(pgio);
900 }
901
902 static void
903 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
904                          struct nfs_page *req)
905 {
906         struct nfs_commit_info cinfo;
907         int status;
908
909         if (!pgio->pg_lseg)
910                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
911                                            req->wb_context,
912                                            0,
913                                            NFS4_MAX_UINT64,
914                                            IOMODE_RW,
915                                            GFP_NOFS);
916         /* If no lseg, fall back to write through mds */
917         if (pgio->pg_lseg == NULL)
918                 goto out_mds;
919         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
920         status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
921         if (status < 0) {
922                 pnfs_put_lseg(pgio->pg_lseg);
923                 pgio->pg_lseg = NULL;
924                 goto out_mds;
925         }
926         return;
927 out_mds:
928         nfs_pageio_reset_write_mds(pgio);
929 }
930
931 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
932         .pg_init = filelayout_pg_init_read,
933         .pg_test = filelayout_pg_test,
934         .pg_doio = pnfs_generic_pg_readpages,
935 };
936
937 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
938         .pg_init = filelayout_pg_init_write,
939         .pg_test = filelayout_pg_test,
940         .pg_doio = pnfs_generic_pg_writepages,
941 };
942
943 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
944 {
945         if (fl->stripe_type == STRIPE_SPARSE)
946                 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
947         else
948                 return j;
949 }
950
951 static void
952 filelayout_mark_request_commit(struct nfs_page *req,
953                                struct pnfs_layout_segment *lseg,
954                                struct nfs_commit_info *cinfo)
955
956 {
957         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
958         u32 i, j;
959         struct list_head *list;
960         struct pnfs_commit_bucket *buckets;
961
962         if (fl->commit_through_mds) {
963                 list = &cinfo->mds->list;
964                 spin_lock(cinfo->lock);
965                 goto mds_commit;
966         }
967
968         /* Note that we are calling nfs4_fl_calc_j_index on each page
969          * that ends up being committed to a data server.  An attractive
970          * alternative is to add a field to nfs_write_data and nfs_page
971          * to store the value calculated in filelayout_write_pagelist
972          * and just use that here.
973          */
974         j = nfs4_fl_calc_j_index(lseg, req_offset(req));
975         i = select_bucket_index(fl, j);
976         spin_lock(cinfo->lock);
977         buckets = cinfo->ds->buckets;
978         list = &buckets[i].written;
979         if (list_empty(list)) {
980                 /* Non-empty buckets hold a reference on the lseg.  That ref
981                  * is normally transferred to the COMMIT call and released
982                  * there.  It could also be released if the last req is pulled
983                  * off due to a rewrite, in which case it will be done in
984                  * pnfs_generic_clear_request_commit
985                  */
986                 buckets[i].wlseg = pnfs_get_lseg(lseg);
987         }
988         set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
989         cinfo->ds->nwritten++;
990
991 mds_commit:
992         /* nfs_request_add_commit_list(). We need to add req to list without
993          * dropping cinfo lock.
994          */
995         set_bit(PG_CLEAN, &(req)->wb_flags);
996         nfs_list_add_request(req, list);
997         cinfo->mds->ncommit++;
998         spin_unlock(cinfo->lock);
999         if (!cinfo->dreq) {
1000                 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1001                 inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
1002                              BDI_RECLAIMABLE);
1003                 __mark_inode_dirty(req->wb_context->dentry->d_inode,
1004                                    I_DIRTY_DATASYNC);
1005         }
1006 }
1007
1008 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1009 {
1010         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1011
1012         if (flseg->stripe_type == STRIPE_SPARSE)
1013                 return i;
1014         else
1015                 return nfs4_fl_calc_ds_index(lseg, i);
1016 }
1017
1018 static struct nfs_fh *
1019 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1020 {
1021         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1022
1023         if (flseg->stripe_type == STRIPE_SPARSE) {
1024                 if (flseg->num_fh == 1)
1025                         i = 0;
1026                 else if (flseg->num_fh == 0)
1027                         /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1028                         return NULL;
1029         }
1030         return flseg->fh_array[i];
1031 }
1032
1033 static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
1034 {
1035         struct pnfs_layout_segment *lseg = data->lseg;
1036         struct nfs4_pnfs_ds *ds;
1037         struct rpc_clnt *ds_clnt;
1038         u32 idx;
1039         struct nfs_fh *fh;
1040
1041         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1042         ds = nfs4_fl_prepare_ds(lseg, idx);
1043         if (!ds)
1044                 goto out_err;
1045
1046         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, data->inode);
1047         if (IS_ERR(ds_clnt))
1048                 goto out_err;
1049
1050         dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1051                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count));
1052         data->commit_done_cb = filelayout_commit_done_cb;
1053         atomic_inc(&ds->ds_clp->cl_count);
1054         data->ds_clp = ds->ds_clp;
1055         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1056         if (fh)
1057                 data->args.fh = fh;
1058         return nfs_initiate_commit(ds_clnt, data,
1059                                    &filelayout_commit_call_ops, how,
1060                                    RPC_TASK_SOFTCONN);
1061 out_err:
1062         pnfs_generic_prepare_to_resend_writes(data);
1063         pnfs_generic_commit_release(data);
1064         return -EAGAIN;
1065 }
1066
1067 /* filelayout_search_commit_reqs - Search lists in @cinfo for the head reqest
1068  *                                 for @page
1069  * @cinfo - commit info for current inode
1070  * @page - page to search for matching head request
1071  *
1072  * Returns a the head request if one is found, otherwise returns NULL.
1073  */
1074 static struct nfs_page *
1075 filelayout_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page)
1076 {
1077         struct nfs_page *freq, *t;
1078         struct pnfs_commit_bucket *b;
1079         int i;
1080
1081         /* Linearly search the commit lists for each bucket until a matching
1082          * request is found */
1083         for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
1084                 list_for_each_entry_safe(freq, t, &b->written, wb_list) {
1085                         if (freq->wb_page == page)
1086                                 return freq->wb_head;
1087                 }
1088                 list_for_each_entry_safe(freq, t, &b->committing, wb_list) {
1089                         if (freq->wb_page == page)
1090                                 return freq->wb_head;
1091                 }
1092         }
1093
1094         return NULL;
1095 }
1096
1097 static int
1098 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1099                            int how, struct nfs_commit_info *cinfo)
1100 {
1101         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1102                                             filelayout_initiate_commit);
1103 }
1104
1105 static struct nfs4_deviceid_node *
1106 filelayout_alloc_deviceid_node(struct nfs_server *server,
1107                 struct pnfs_device *pdev, gfp_t gfp_flags)
1108 {
1109         struct nfs4_file_layout_dsaddr *dsaddr;
1110
1111         dsaddr = nfs4_fl_alloc_deviceid_node(server, pdev, gfp_flags);
1112         if (!dsaddr)
1113                 return NULL;
1114         return &dsaddr->id_node;
1115 }
1116
1117 static void
1118 filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
1119 {
1120         nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1121 }
1122
1123 static struct pnfs_layout_hdr *
1124 filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1125 {
1126         struct nfs4_filelayout *flo;
1127
1128         flo = kzalloc(sizeof(*flo), gfp_flags);
1129         return flo != NULL ? &flo->generic_hdr : NULL;
1130 }
1131
1132 static void
1133 filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1134 {
1135         kfree(FILELAYOUT_FROM_HDR(lo));
1136 }
1137
1138 static struct pnfs_ds_commit_info *
1139 filelayout_get_ds_info(struct inode *inode)
1140 {
1141         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1142
1143         if (layout == NULL)
1144                 return NULL;
1145         else
1146                 return &FILELAYOUT_FROM_HDR(layout)->commit_info;
1147 }
1148
1149 static struct pnfs_layoutdriver_type filelayout_type = {
1150         .id                     = LAYOUT_NFSV4_1_FILES,
1151         .name                   = "LAYOUT_NFSV4_1_FILES",
1152         .owner                  = THIS_MODULE,
1153         .alloc_layout_hdr       = filelayout_alloc_layout_hdr,
1154         .free_layout_hdr        = filelayout_free_layout_hdr,
1155         .alloc_lseg             = filelayout_alloc_lseg,
1156         .free_lseg              = filelayout_free_lseg,
1157         .pg_read_ops            = &filelayout_pg_read_ops,
1158         .pg_write_ops           = &filelayout_pg_write_ops,
1159         .get_ds_info            = &filelayout_get_ds_info,
1160         .mark_request_commit    = filelayout_mark_request_commit,
1161         .clear_request_commit   = pnfs_generic_clear_request_commit,
1162         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
1163         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
1164         .search_commit_reqs     = filelayout_search_commit_reqs,
1165         .commit_pagelist        = filelayout_commit_pagelist,
1166         .read_pagelist          = filelayout_read_pagelist,
1167         .write_pagelist         = filelayout_write_pagelist,
1168         .alloc_deviceid_node    = filelayout_alloc_deviceid_node,
1169         .free_deviceid_node     = filelayout_free_deveiceid_node,
1170 };
1171
1172 static int __init nfs4filelayout_init(void)
1173 {
1174         printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1175                __func__);
1176         return pnfs_register_layoutdriver(&filelayout_type);
1177 }
1178
1179 static void __exit nfs4filelayout_exit(void)
1180 {
1181         printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1182                __func__);
1183         pnfs_unregister_layoutdriver(&filelayout_type);
1184 }
1185
1186 MODULE_ALIAS("nfs-layouttype4-1");
1187
1188 module_init(nfs4filelayout_init);
1189 module_exit(nfs4filelayout_exit);