]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - fs/nfs/filelayout/filelayout.c
Merge tag 'for-linus-20170812' of git://git.infradead.org/linux-mtd
[karo-tx-linux.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 nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
130         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
131
132         if (task->tk_status >= 0)
133                 return 0;
134
135         switch (task->tk_status) {
136         /* DS session errors */
137         case -NFS4ERR_BADSESSION:
138         case -NFS4ERR_BADSLOT:
139         case -NFS4ERR_BAD_HIGH_SLOT:
140         case -NFS4ERR_DEADSESSION:
141         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
142         case -NFS4ERR_SEQ_FALSE_RETRY:
143         case -NFS4ERR_SEQ_MISORDERED:
144                 dprintk("%s ERROR %d, Reset session. Exchangeid "
145                         "flags 0x%x\n", __func__, task->tk_status,
146                         clp->cl_exchange_flags);
147                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
148                 break;
149         case -NFS4ERR_DELAY:
150         case -NFS4ERR_GRACE:
151                 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
152                 break;
153         case -NFS4ERR_RETRY_UNCACHED_REP:
154                 break;
155         /* Invalidate Layout errors */
156         case -NFS4ERR_ACCESS:
157         case -NFS4ERR_PNFS_NO_LAYOUT:
158         case -ESTALE:           /* mapped NFS4ERR_STALE */
159         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
160         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
161         case -NFS4ERR_FHEXPIRED:
162         case -NFS4ERR_WRONG_TYPE:
163                 dprintk("%s Invalid layout error %d\n", __func__,
164                         task->tk_status);
165                 /*
166                  * Destroy layout so new i/o will get a new layout.
167                  * Layout will not be destroyed until all current lseg
168                  * references are put. Mark layout as invalid to resend failed
169                  * i/o and all i/o waiting on the slot table to the MDS until
170                  * layout is destroyed and a new valid layout is obtained.
171                  */
172                 pnfs_destroy_layout(NFS_I(inode));
173                 rpc_wake_up(&tbl->slot_tbl_waitq);
174                 goto reset;
175         /* RPC connection errors */
176         case -ECONNREFUSED:
177         case -EHOSTDOWN:
178         case -EHOSTUNREACH:
179         case -ENETUNREACH:
180         case -EIO:
181         case -ETIMEDOUT:
182         case -EPIPE:
183                 dprintk("%s DS connection error %d\n", __func__,
184                         task->tk_status);
185                 nfs4_mark_deviceid_unavailable(devid);
186                 pnfs_error_mark_layout_for_return(inode, lseg);
187                 pnfs_set_lo_fail(lseg);
188                 rpc_wake_up(&tbl->slot_tbl_waitq);
189                 /* fall through */
190         default:
191 reset:
192                 dprintk("%s Retry through MDS. Error %d\n", __func__,
193                         task->tk_status);
194                 return -NFS4ERR_RESET_TO_MDS;
195         }
196         task->tk_status = 0;
197         return -EAGAIN;
198 }
199
200 /* NFS_PROTO call done callback routines */
201
202 static int filelayout_read_done_cb(struct rpc_task *task,
203                                 struct nfs_pgio_header *hdr)
204 {
205         int err;
206
207         trace_nfs4_pnfs_read(hdr, task->tk_status);
208         err = filelayout_async_handle_error(task, hdr->args.context->state,
209                                             hdr->ds_clp, hdr->lseg);
210
211         switch (err) {
212         case -NFS4ERR_RESET_TO_MDS:
213                 filelayout_reset_read(hdr);
214                 return task->tk_status;
215         case -EAGAIN:
216                 rpc_restart_call_prepare(task);
217                 return -EAGAIN;
218         }
219
220         return 0;
221 }
222
223 /*
224  * We reference the rpc_cred of the first WRITE that triggers the need for
225  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
226  * rfc5661 is not clear about which credential should be used.
227  */
228 static void
229 filelayout_set_layoutcommit(struct nfs_pgio_header *hdr)
230 {
231         loff_t end_offs = 0;
232
233         if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds ||
234             hdr->res.verf->committed == NFS_FILE_SYNC)
235                 return;
236         if (hdr->res.verf->committed == NFS_DATA_SYNC)
237                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
238
239         /* Note: if the write is unstable, don't set end_offs until commit */
240         pnfs_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
241         dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
242                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
243 }
244
245 bool
246 filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node)
247 {
248         return filelayout_test_devid_invalid(node) ||
249                 nfs4_test_deviceid_unavailable(node);
250 }
251
252 static bool
253 filelayout_reset_to_mds(struct pnfs_layout_segment *lseg)
254 {
255         struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg);
256
257         return filelayout_test_devid_unavailable(node);
258 }
259
260 /*
261  * Call ops for the async read/write cases
262  * In the case of dense layouts, the offset needs to be reset to its
263  * original value.
264  */
265 static void filelayout_read_prepare(struct rpc_task *task, void *data)
266 {
267         struct nfs_pgio_header *hdr = data;
268
269         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
270                 rpc_exit(task, -EIO);
271                 return;
272         }
273         if (filelayout_reset_to_mds(hdr->lseg)) {
274                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
275                 filelayout_reset_read(hdr);
276                 rpc_exit(task, 0);
277                 return;
278         }
279         hdr->pgio_done_cb = filelayout_read_done_cb;
280
281         if (nfs4_setup_sequence(hdr->ds_clp,
282                         &hdr->args.seq_args,
283                         &hdr->res.seq_res,
284                         task))
285                 return;
286         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
287                         hdr->args.lock_context, FMODE_READ) == -EIO)
288                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
289 }
290
291 static void filelayout_read_call_done(struct rpc_task *task, void *data)
292 {
293         struct nfs_pgio_header *hdr = data;
294
295         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
296
297         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
298             task->tk_status == 0) {
299                 nfs41_sequence_done(task, &hdr->res.seq_res);
300                 return;
301         }
302
303         /* Note this may cause RPC to be resent */
304         hdr->mds_ops->rpc_call_done(task, data);
305 }
306
307 static void filelayout_read_count_stats(struct rpc_task *task, void *data)
308 {
309         struct nfs_pgio_header *hdr = data;
310
311         rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
312 }
313
314 static int filelayout_write_done_cb(struct rpc_task *task,
315                                 struct nfs_pgio_header *hdr)
316 {
317         int err;
318
319         trace_nfs4_pnfs_write(hdr, task->tk_status);
320         err = filelayout_async_handle_error(task, hdr->args.context->state,
321                                             hdr->ds_clp, hdr->lseg);
322
323         switch (err) {
324         case -NFS4ERR_RESET_TO_MDS:
325                 filelayout_reset_write(hdr);
326                 return task->tk_status;
327         case -EAGAIN:
328                 rpc_restart_call_prepare(task);
329                 return -EAGAIN;
330         }
331
332         filelayout_set_layoutcommit(hdr);
333
334         /* zero out the fattr */
335         hdr->fattr.valid = 0;
336         if (task->tk_status >= 0)
337                 nfs_writeback_update_inode(hdr);
338
339         return 0;
340 }
341
342 static int filelayout_commit_done_cb(struct rpc_task *task,
343                                      struct nfs_commit_data *data)
344 {
345         int err;
346
347         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
348         err = filelayout_async_handle_error(task, NULL, data->ds_clp,
349                                             data->lseg);
350
351         switch (err) {
352         case -NFS4ERR_RESET_TO_MDS:
353                 pnfs_generic_prepare_to_resend_writes(data);
354                 return -EAGAIN;
355         case -EAGAIN:
356                 rpc_restart_call_prepare(task);
357                 return -EAGAIN;
358         }
359
360         pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
361
362         return 0;
363 }
364
365 static void filelayout_write_prepare(struct rpc_task *task, void *data)
366 {
367         struct nfs_pgio_header *hdr = data;
368
369         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
370                 rpc_exit(task, -EIO);
371                 return;
372         }
373         if (filelayout_reset_to_mds(hdr->lseg)) {
374                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
375                 filelayout_reset_write(hdr);
376                 rpc_exit(task, 0);
377                 return;
378         }
379         if (nfs4_setup_sequence(hdr->ds_clp,
380                         &hdr->args.seq_args,
381                         &hdr->res.seq_res,
382                         task))
383                 return;
384         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
385                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
386                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
387 }
388
389 static void filelayout_write_call_done(struct rpc_task *task, void *data)
390 {
391         struct nfs_pgio_header *hdr = data;
392
393         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
394             task->tk_status == 0) {
395                 nfs41_sequence_done(task, &hdr->res.seq_res);
396                 return;
397         }
398
399         /* Note this may cause RPC to be resent */
400         hdr->mds_ops->rpc_call_done(task, data);
401 }
402
403 static void filelayout_write_count_stats(struct rpc_task *task, void *data)
404 {
405         struct nfs_pgio_header *hdr = data;
406
407         rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
408 }
409
410 static void filelayout_commit_prepare(struct rpc_task *task, void *data)
411 {
412         struct nfs_commit_data *wdata = data;
413
414         nfs4_setup_sequence(wdata->ds_clp,
415                         &wdata->args.seq_args,
416                         &wdata->res.seq_res,
417                         task);
418 }
419
420 static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
421 {
422         struct nfs_commit_data *cdata = data;
423
424         rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
425 }
426
427 static const struct rpc_call_ops filelayout_read_call_ops = {
428         .rpc_call_prepare = filelayout_read_prepare,
429         .rpc_call_done = filelayout_read_call_done,
430         .rpc_count_stats = filelayout_read_count_stats,
431         .rpc_release = pnfs_generic_rw_release,
432 };
433
434 static const struct rpc_call_ops filelayout_write_call_ops = {
435         .rpc_call_prepare = filelayout_write_prepare,
436         .rpc_call_done = filelayout_write_call_done,
437         .rpc_count_stats = filelayout_write_count_stats,
438         .rpc_release = pnfs_generic_rw_release,
439 };
440
441 static const struct rpc_call_ops filelayout_commit_call_ops = {
442         .rpc_call_prepare = filelayout_commit_prepare,
443         .rpc_call_done = pnfs_generic_write_commit_done,
444         .rpc_count_stats = filelayout_commit_count_stats,
445         .rpc_release = pnfs_generic_commit_release,
446 };
447
448 static enum pnfs_try_status
449 filelayout_read_pagelist(struct nfs_pgio_header *hdr)
450 {
451         struct pnfs_layout_segment *lseg = hdr->lseg;
452         struct nfs4_pnfs_ds *ds;
453         struct rpc_clnt *ds_clnt;
454         loff_t offset = hdr->args.offset;
455         u32 j, idx;
456         struct nfs_fh *fh;
457
458         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
459                 __func__, hdr->inode->i_ino,
460                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
461
462         /* Retrieve the correct rpc_client for the byte range */
463         j = nfs4_fl_calc_j_index(lseg, offset);
464         idx = nfs4_fl_calc_ds_index(lseg, j);
465         ds = nfs4_fl_prepare_ds(lseg, idx);
466         if (!ds)
467                 return PNFS_NOT_ATTEMPTED;
468
469         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
470         if (IS_ERR(ds_clnt))
471                 return PNFS_NOT_ATTEMPTED;
472
473         dprintk("%s USE DS: %s cl_count %d\n", __func__,
474                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
475
476         /* No multipath support. Use first DS */
477         atomic_inc(&ds->ds_clp->cl_count);
478         hdr->ds_clp = ds->ds_clp;
479         hdr->ds_commit_idx = idx;
480         fh = nfs4_fl_select_ds_fh(lseg, j);
481         if (fh)
482                 hdr->args.fh = fh;
483
484         hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
485         hdr->mds_offset = offset;
486
487         /* Perform an asynchronous read to ds */
488         nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
489                           NFS_PROTO(hdr->inode), &filelayout_read_call_ops,
490                           0, RPC_TASK_SOFTCONN);
491         return PNFS_ATTEMPTED;
492 }
493
494 /* Perform async writes. */
495 static enum pnfs_try_status
496 filelayout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
497 {
498         struct pnfs_layout_segment *lseg = hdr->lseg;
499         struct nfs4_pnfs_ds *ds;
500         struct rpc_clnt *ds_clnt;
501         loff_t offset = hdr->args.offset;
502         u32 j, idx;
503         struct nfs_fh *fh;
504
505         /* Retrieve the correct rpc_client for the byte range */
506         j = nfs4_fl_calc_j_index(lseg, offset);
507         idx = nfs4_fl_calc_ds_index(lseg, j);
508         ds = nfs4_fl_prepare_ds(lseg, idx);
509         if (!ds)
510                 return PNFS_NOT_ATTEMPTED;
511
512         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
513         if (IS_ERR(ds_clnt))
514                 return PNFS_NOT_ATTEMPTED;
515
516         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d\n",
517                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
518                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
519
520         hdr->pgio_done_cb = filelayout_write_done_cb;
521         atomic_inc(&ds->ds_clp->cl_count);
522         hdr->ds_clp = ds->ds_clp;
523         hdr->ds_commit_idx = idx;
524         fh = nfs4_fl_select_ds_fh(lseg, j);
525         if (fh)
526                 hdr->args.fh = fh;
527         hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
528
529         /* Perform an asynchronous write */
530         nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
531                           NFS_PROTO(hdr->inode), &filelayout_write_call_ops,
532                           sync, RPC_TASK_SOFTCONN);
533         return PNFS_ATTEMPTED;
534 }
535
536 static int
537 filelayout_check_deviceid(struct pnfs_layout_hdr *lo,
538                           struct nfs4_filelayout_segment *fl,
539                           gfp_t gfp_flags)
540 {
541         struct nfs4_deviceid_node *d;
542         struct nfs4_file_layout_dsaddr *dsaddr;
543         int status = -EINVAL;
544
545         /* Is the deviceid already set? If so, we're good. */
546         if (fl->dsaddr != NULL)
547                 return 0;
548
549         /* find and reference the deviceid */
550         d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &fl->deviceid,
551                         lo->plh_lc_cred, gfp_flags);
552         if (d == NULL)
553                 goto out;
554
555         dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
556         /* Found deviceid is unavailable */
557         if (filelayout_test_devid_unavailable(&dsaddr->id_node))
558                 goto out_put;
559
560         if (fl->first_stripe_index >= dsaddr->stripe_count) {
561                 dprintk("%s Bad first_stripe_index %u\n",
562                                 __func__, fl->first_stripe_index);
563                 goto out_put;
564         }
565
566         if ((fl->stripe_type == STRIPE_SPARSE &&
567             fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
568             (fl->stripe_type == STRIPE_DENSE &&
569             fl->num_fh != dsaddr->stripe_count)) {
570                 dprintk("%s num_fh %u not valid for given packing\n",
571                         __func__, fl->num_fh);
572                 goto out_put;
573         }
574         status = 0;
575
576         /*
577          * Atomic compare and xchange to ensure we don't scribble
578          * over a non-NULL pointer.
579          */
580         if (cmpxchg(&fl->dsaddr, NULL, dsaddr) != NULL)
581                 goto out_put;
582 out:
583         return status;
584 out_put:
585         nfs4_fl_put_deviceid(dsaddr);
586         goto out;
587 }
588
589 /*
590  * filelayout_check_layout()
591  *
592  * Make sure layout segment parameters are sane WRT the device.
593  * At this point no generic layer initialization of the lseg has occurred,
594  * and nothing has been added to the layout_hdr cache.
595  *
596  */
597 static int
598 filelayout_check_layout(struct pnfs_layout_hdr *lo,
599                         struct nfs4_filelayout_segment *fl,
600                         struct nfs4_layoutget_res *lgr,
601                         gfp_t gfp_flags)
602 {
603         int status = -EINVAL;
604
605         dprintk("--> %s\n", __func__);
606
607         /* FIXME: remove this check when layout segment support is added */
608         if (lgr->range.offset != 0 ||
609             lgr->range.length != NFS4_MAX_UINT64) {
610                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
611                         __func__);
612                 goto out;
613         }
614
615         if (fl->pattern_offset > lgr->range.offset) {
616                 dprintk("%s pattern_offset %lld too large\n",
617                                 __func__, fl->pattern_offset);
618                 goto out;
619         }
620
621         if (!fl->stripe_unit) {
622                 dprintk("%s Invalid stripe unit (%u)\n",
623                         __func__, fl->stripe_unit);
624                 goto out;
625         }
626
627         status = 0;
628 out:
629         dprintk("--> %s returns %d\n", __func__, status);
630         return status;
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                          gfp_t gfp_flags)
653 {
654         struct xdr_stream stream;
655         struct xdr_buf buf;
656         struct page *scratch;
657         __be32 *p;
658         uint32_t nfl_util;
659         int i;
660
661         dprintk("%s: set_layout_map Begin\n", __func__);
662
663         scratch = alloc_page(gfp_flags);
664         if (!scratch)
665                 return -ENOMEM;
666
667         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
668         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
669
670         /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
671          * num_fh (4) */
672         p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
673         if (unlikely(!p))
674                 goto out_err;
675
676         memcpy(&fl->deviceid, p, sizeof(fl->deviceid));
677         p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
678         nfs4_print_deviceid(&fl->deviceid);
679
680         nfl_util = be32_to_cpup(p++);
681         if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
682                 fl->commit_through_mds = 1;
683         if (nfl_util & NFL4_UFLG_DENSE)
684                 fl->stripe_type = STRIPE_DENSE;
685         else
686                 fl->stripe_type = STRIPE_SPARSE;
687         fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
688
689         fl->first_stripe_index = be32_to_cpup(p++);
690         p = xdr_decode_hyper(p, &fl->pattern_offset);
691         fl->num_fh = be32_to_cpup(p++);
692
693         dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
694                 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
695                 fl->pattern_offset);
696
697         /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
698          * Futher checking is done in filelayout_check_layout */
699         if (fl->num_fh >
700             max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
701                 goto out_err;
702
703         if (fl->num_fh > 0) {
704                 fl->fh_array = kcalloc(fl->num_fh, sizeof(fl->fh_array[0]),
705                                        gfp_flags);
706                 if (!fl->fh_array)
707                         goto out_err;
708         }
709
710         for (i = 0; i < fl->num_fh; i++) {
711                 /* Do we want to use a mempool here? */
712                 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
713                 if (!fl->fh_array[i])
714                         goto out_err;
715
716                 p = xdr_inline_decode(&stream, 4);
717                 if (unlikely(!p))
718                         goto out_err;
719                 fl->fh_array[i]->size = be32_to_cpup(p++);
720                 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
721                         printk(KERN_ERR "NFS: Too big fh %d received %d\n",
722                                i, fl->fh_array[i]->size);
723                         goto out_err;
724                 }
725
726                 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
727                 if (unlikely(!p))
728                         goto out_err;
729                 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
730                 dprintk("DEBUG: %s: fh len %d\n", __func__,
731                         fl->fh_array[i]->size);
732         }
733
734         __free_page(scratch);
735         return 0;
736
737 out_err:
738         __free_page(scratch);
739         return -EIO;
740 }
741
742 static void
743 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
744 {
745         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
746
747         dprintk("--> %s\n", __func__);
748         nfs4_fl_put_deviceid(fl->dsaddr);
749         /* This assumes a single RW lseg */
750         if (lseg->pls_range.iomode == IOMODE_RW) {
751                 struct nfs4_filelayout *flo;
752
753                 flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
754                 flo->commit_info.nbuckets = 0;
755                 kfree(flo->commit_info.buckets);
756                 flo->commit_info.buckets = NULL;
757         }
758         _filelayout_free_lseg(fl);
759 }
760
761 static int
762 filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg,
763                              struct nfs_commit_info *cinfo,
764                              gfp_t gfp_flags)
765 {
766         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
767         struct pnfs_commit_bucket *buckets;
768         int size, i;
769
770         if (fl->commit_through_mds)
771                 return 0;
772
773         size = (fl->stripe_type == STRIPE_SPARSE) ?
774                 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
775
776         if (cinfo->ds->nbuckets >= size) {
777                 /* This assumes there is only one IOMODE_RW lseg.  What
778                  * we really want to do is have a layout_hdr level
779                  * dictionary of <multipath_list4, fh> keys, each
780                  * associated with a struct list_head, populated by calls
781                  * to filelayout_write_pagelist().
782                  * */
783                 return 0;
784         }
785
786         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
787                           gfp_flags);
788         if (!buckets)
789                 return -ENOMEM;
790         for (i = 0; i < size; i++) {
791                 INIT_LIST_HEAD(&buckets[i].written);
792                 INIT_LIST_HEAD(&buckets[i].committing);
793                 /* mark direct verifier as unset */
794                 buckets[i].direct_verf.committed = NFS_INVALID_STABLE_HOW;
795         }
796
797         spin_lock(&cinfo->inode->i_lock);
798         if (cinfo->ds->nbuckets >= size)
799                 goto out;
800         for (i = 0; i < cinfo->ds->nbuckets; i++) {
801                 list_splice(&cinfo->ds->buckets[i].written,
802                             &buckets[i].written);
803                 list_splice(&cinfo->ds->buckets[i].committing,
804                             &buckets[i].committing);
805                 buckets[i].direct_verf.committed =
806                         cinfo->ds->buckets[i].direct_verf.committed;
807                 buckets[i].wlseg = cinfo->ds->buckets[i].wlseg;
808                 buckets[i].clseg = cinfo->ds->buckets[i].clseg;
809         }
810         swap(cinfo->ds->buckets, buckets);
811         cinfo->ds->nbuckets = size;
812 out:
813         spin_unlock(&cinfo->inode->i_lock);
814         kfree(buckets);
815         return 0;
816 }
817
818 static struct pnfs_layout_segment *
819 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
820                       struct nfs4_layoutget_res *lgr,
821                       gfp_t gfp_flags)
822 {
823         struct nfs4_filelayout_segment *fl;
824         int rc;
825
826         dprintk("--> %s\n", __func__);
827         fl = kzalloc(sizeof(*fl), gfp_flags);
828         if (!fl)
829                 return NULL;
830
831         rc = filelayout_decode_layout(layoutid, fl, lgr, gfp_flags);
832         if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, gfp_flags)) {
833                 _filelayout_free_lseg(fl);
834                 return NULL;
835         }
836         return &fl->generic_hdr;
837 }
838
839 /*
840  * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
841  *
842  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
843  * of bytes (maximum @req->wb_bytes) that can be coalesced.
844  */
845 static size_t
846 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
847                    struct nfs_page *req)
848 {
849         unsigned int size;
850         u64 p_stripe, r_stripe;
851         u32 stripe_offset;
852         u64 segment_offset = pgio->pg_lseg->pls_range.offset;
853         u32 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
854
855         /* calls nfs_generic_pg_test */
856         size = pnfs_generic_pg_test(pgio, prev, req);
857         if (!size)
858                 return 0;
859
860         /* see if req and prev are in the same stripe */
861         if (prev) {
862                 p_stripe = (u64)req_offset(prev) - segment_offset;
863                 r_stripe = (u64)req_offset(req) - segment_offset;
864                 do_div(p_stripe, stripe_unit);
865                 do_div(r_stripe, stripe_unit);
866
867                 if (p_stripe != r_stripe)
868                         return 0;
869         }
870
871         /* calculate remaining bytes in the current stripe */
872         div_u64_rem((u64)req_offset(req) - segment_offset,
873                         stripe_unit,
874                         &stripe_offset);
875         WARN_ON_ONCE(stripe_offset > stripe_unit);
876         if (stripe_offset >= stripe_unit)
877                 return 0;
878         return min(stripe_unit - (unsigned int)stripe_offset, size);
879 }
880
881 static struct pnfs_layout_segment *
882 fl_pnfs_update_layout(struct inode *ino,
883                       struct nfs_open_context *ctx,
884                       loff_t pos,
885                       u64 count,
886                       enum pnfs_iomode iomode,
887                       bool strict_iomode,
888                       gfp_t gfp_flags)
889 {
890         struct pnfs_layout_segment *lseg = NULL;
891         struct pnfs_layout_hdr *lo;
892         struct nfs4_filelayout_segment *fl;
893         int status;
894
895         lseg = pnfs_update_layout(ino, ctx, pos, count, iomode, strict_iomode,
896                                   gfp_flags);
897         if (!lseg)
898                 lseg = ERR_PTR(-ENOMEM);
899         if (IS_ERR(lseg))
900                 goto out;
901
902         lo = NFS_I(ino)->layout;
903         fl = FILELAYOUT_LSEG(lseg);
904
905         status = filelayout_check_deviceid(lo, fl, gfp_flags);
906         if (status) {
907                 pnfs_put_lseg(lseg);
908                 lseg = ERR_PTR(status);
909         }
910 out:
911         return lseg;
912 }
913
914 static void
915 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
916                         struct nfs_page *req)
917 {
918         pnfs_generic_pg_check_layout(pgio);
919         if (!pgio->pg_lseg) {
920                 pgio->pg_lseg = fl_pnfs_update_layout(pgio->pg_inode,
921                                                       req->wb_context,
922                                                       0,
923                                                       NFS4_MAX_UINT64,
924                                                       IOMODE_READ,
925                                                       false,
926                                                       GFP_KERNEL);
927                 if (IS_ERR(pgio->pg_lseg)) {
928                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
929                         pgio->pg_lseg = NULL;
930                         return;
931                 }
932         }
933         /* If no lseg, fall back to read through mds */
934         if (pgio->pg_lseg == NULL)
935                 nfs_pageio_reset_read_mds(pgio);
936 }
937
938 static void
939 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
940                          struct nfs_page *req)
941 {
942         struct nfs_commit_info cinfo;
943         int status;
944
945         pnfs_generic_pg_check_layout(pgio);
946         if (!pgio->pg_lseg) {
947                 pgio->pg_lseg = fl_pnfs_update_layout(pgio->pg_inode,
948                                                       req->wb_context,
949                                                       0,
950                                                       NFS4_MAX_UINT64,
951                                                       IOMODE_RW,
952                                                       false,
953                                                       GFP_NOFS);
954                 if (IS_ERR(pgio->pg_lseg)) {
955                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
956                         pgio->pg_lseg = NULL;
957                         return;
958                 }
959         }
960
961         /* If no lseg, fall back to write through mds */
962         if (pgio->pg_lseg == NULL)
963                 goto out_mds;
964         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
965         status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
966         if (status < 0) {
967                 pnfs_put_lseg(pgio->pg_lseg);
968                 pgio->pg_lseg = NULL;
969                 goto out_mds;
970         }
971         return;
972 out_mds:
973         nfs_pageio_reset_write_mds(pgio);
974 }
975
976 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
977         .pg_init = filelayout_pg_init_read,
978         .pg_test = filelayout_pg_test,
979         .pg_doio = pnfs_generic_pg_readpages,
980         .pg_cleanup = pnfs_generic_pg_cleanup,
981 };
982
983 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
984         .pg_init = filelayout_pg_init_write,
985         .pg_test = filelayout_pg_test,
986         .pg_doio = pnfs_generic_pg_writepages,
987         .pg_cleanup = pnfs_generic_pg_cleanup,
988 };
989
990 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
991 {
992         if (fl->stripe_type == STRIPE_SPARSE)
993                 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
994         else
995                 return j;
996 }
997
998 static void
999 filelayout_mark_request_commit(struct nfs_page *req,
1000                                struct pnfs_layout_segment *lseg,
1001                                struct nfs_commit_info *cinfo,
1002                                u32 ds_commit_idx)
1003
1004 {
1005         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
1006         u32 i, j;
1007
1008         if (fl->commit_through_mds) {
1009                 nfs_request_add_commit_list(req, cinfo);
1010         } else {
1011                 /* Note that we are calling nfs4_fl_calc_j_index on each page
1012                  * that ends up being committed to a data server.  An attractive
1013                  * alternative is to add a field to nfs_write_data and nfs_page
1014                  * to store the value calculated in filelayout_write_pagelist
1015                  * and just use that here.
1016                  */
1017                 j = nfs4_fl_calc_j_index(lseg, req_offset(req));
1018                 i = select_bucket_index(fl, j);
1019                 pnfs_layout_mark_request_commit(req, lseg, cinfo, i);
1020         }
1021 }
1022
1023 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1024 {
1025         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1026
1027         if (flseg->stripe_type == STRIPE_SPARSE)
1028                 return i;
1029         else
1030                 return nfs4_fl_calc_ds_index(lseg, i);
1031 }
1032
1033 static struct nfs_fh *
1034 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1035 {
1036         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1037
1038         if (flseg->stripe_type == STRIPE_SPARSE) {
1039                 if (flseg->num_fh == 1)
1040                         i = 0;
1041                 else if (flseg->num_fh == 0)
1042                         /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1043                         return NULL;
1044         }
1045         return flseg->fh_array[i];
1046 }
1047
1048 static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
1049 {
1050         struct pnfs_layout_segment *lseg = data->lseg;
1051         struct nfs4_pnfs_ds *ds;
1052         struct rpc_clnt *ds_clnt;
1053         u32 idx;
1054         struct nfs_fh *fh;
1055
1056         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1057         ds = nfs4_fl_prepare_ds(lseg, idx);
1058         if (!ds)
1059                 goto out_err;
1060
1061         ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, data->inode);
1062         if (IS_ERR(ds_clnt))
1063                 goto out_err;
1064
1065         dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1066                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count));
1067         data->commit_done_cb = filelayout_commit_done_cb;
1068         atomic_inc(&ds->ds_clp->cl_count);
1069         data->ds_clp = ds->ds_clp;
1070         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1071         if (fh)
1072                 data->args.fh = fh;
1073         return nfs_initiate_commit(ds_clnt, data, NFS_PROTO(data->inode),
1074                                    &filelayout_commit_call_ops, how,
1075                                    RPC_TASK_SOFTCONN);
1076 out_err:
1077         pnfs_generic_prepare_to_resend_writes(data);
1078         pnfs_generic_commit_release(data);
1079         return -EAGAIN;
1080 }
1081
1082 /* filelayout_search_commit_reqs - Search lists in @cinfo for the head reqest
1083  *                                 for @page
1084  * @cinfo - commit info for current inode
1085  * @page - page to search for matching head request
1086  *
1087  * Returns a the head request if one is found, otherwise returns NULL.
1088  */
1089 static struct nfs_page *
1090 filelayout_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page)
1091 {
1092         struct nfs_page *freq, *t;
1093         struct pnfs_commit_bucket *b;
1094         int i;
1095
1096         /* Linearly search the commit lists for each bucket until a matching
1097          * request is found */
1098         for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
1099                 list_for_each_entry_safe(freq, t, &b->written, wb_list) {
1100                         if (freq->wb_page == page)
1101                                 return freq->wb_head;
1102                 }
1103                 list_for_each_entry_safe(freq, t, &b->committing, wb_list) {
1104                         if (freq->wb_page == page)
1105                                 return freq->wb_head;
1106                 }
1107         }
1108
1109         return NULL;
1110 }
1111
1112 static int
1113 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1114                            int how, struct nfs_commit_info *cinfo)
1115 {
1116         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1117                                             filelayout_initiate_commit);
1118 }
1119
1120 static struct nfs4_deviceid_node *
1121 filelayout_alloc_deviceid_node(struct nfs_server *server,
1122                 struct pnfs_device *pdev, gfp_t gfp_flags)
1123 {
1124         struct nfs4_file_layout_dsaddr *dsaddr;
1125
1126         dsaddr = nfs4_fl_alloc_deviceid_node(server, pdev, gfp_flags);
1127         if (!dsaddr)
1128                 return NULL;
1129         return &dsaddr->id_node;
1130 }
1131
1132 static void
1133 filelayout_free_deviceid_node(struct nfs4_deviceid_node *d)
1134 {
1135         nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1136 }
1137
1138 static struct pnfs_layout_hdr *
1139 filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1140 {
1141         struct nfs4_filelayout *flo;
1142
1143         flo = kzalloc(sizeof(*flo), gfp_flags);
1144         return flo != NULL ? &flo->generic_hdr : NULL;
1145 }
1146
1147 static void
1148 filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1149 {
1150         kfree(FILELAYOUT_FROM_HDR(lo));
1151 }
1152
1153 static struct pnfs_ds_commit_info *
1154 filelayout_get_ds_info(struct inode *inode)
1155 {
1156         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1157
1158         if (layout == NULL)
1159                 return NULL;
1160         else
1161                 return &FILELAYOUT_FROM_HDR(layout)->commit_info;
1162 }
1163
1164 static struct pnfs_layoutdriver_type filelayout_type = {
1165         .id                     = LAYOUT_NFSV4_1_FILES,
1166         .name                   = "LAYOUT_NFSV4_1_FILES",
1167         .owner                  = THIS_MODULE,
1168         .alloc_layout_hdr       = filelayout_alloc_layout_hdr,
1169         .free_layout_hdr        = filelayout_free_layout_hdr,
1170         .alloc_lseg             = filelayout_alloc_lseg,
1171         .free_lseg              = filelayout_free_lseg,
1172         .pg_read_ops            = &filelayout_pg_read_ops,
1173         .pg_write_ops           = &filelayout_pg_write_ops,
1174         .get_ds_info            = &filelayout_get_ds_info,
1175         .mark_request_commit    = filelayout_mark_request_commit,
1176         .clear_request_commit   = pnfs_generic_clear_request_commit,
1177         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
1178         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
1179         .search_commit_reqs     = filelayout_search_commit_reqs,
1180         .commit_pagelist        = filelayout_commit_pagelist,
1181         .read_pagelist          = filelayout_read_pagelist,
1182         .write_pagelist         = filelayout_write_pagelist,
1183         .alloc_deviceid_node    = filelayout_alloc_deviceid_node,
1184         .free_deviceid_node     = filelayout_free_deviceid_node,
1185         .sync                   = pnfs_nfs_generic_sync,
1186 };
1187
1188 static int __init nfs4filelayout_init(void)
1189 {
1190         printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1191                __func__);
1192         return pnfs_register_layoutdriver(&filelayout_type);
1193 }
1194
1195 static void __exit nfs4filelayout_exit(void)
1196 {
1197         printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1198                __func__);
1199         pnfs_unregister_layoutdriver(&filelayout_type);
1200 }
1201
1202 MODULE_ALIAS("nfs-layouttype4-1");
1203
1204 module_init(nfs4filelayout_init);
1205 module_exit(nfs4filelayout_exit);