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
47MODULE_LICENSE("GPL");
48MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
49MODULE_DESCRIPTION("The NFSv4 file layout driver");
50
51#define FILELAYOUT_POLL_RETRY_MAX     (15*HZ)
52static const struct pnfs_commit_ops filelayout_commit_ops;
53
54static loff_t
55filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
56			    loff_t offset)
57{
58	u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
59	u64 stripe_no;
60	u32 rem;
61
62	offset -= flseg->pattern_offset;
63	stripe_no = div_u64(offset, stripe_width);
64	div_u64_rem(offset, flseg->stripe_unit, &rem);
65
66	return stripe_no * flseg->stripe_unit + rem;
67}
68
69/* This function is used by the layout driver to calculate the
70 * offset of the file on the dserver based on whether the
71 * layout type is STRIPE_DENSE or STRIPE_SPARSE
72 */
73static loff_t
74filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
75{
76	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
77
78	switch (flseg->stripe_type) {
79	case STRIPE_SPARSE:
80		return offset;
81
82	case STRIPE_DENSE:
83		return filelayout_get_dense_offset(flseg, offset);
84	}
85
86	BUG();
87}
88
89static void filelayout_reset_write(struct nfs_pgio_header *hdr)
90{
91	struct rpc_task *task = &hdr->task;
92
93	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
94		dprintk("%s Reset task %5u for i/o through MDS "
95			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
96			hdr->task.tk_pid,
97			hdr->inode->i_sb->s_id,
98			(unsigned long long)NFS_FILEID(hdr->inode),
99			hdr->args.count,
100			(unsigned long long)hdr->args.offset);
101
102		task->tk_status = pnfs_write_done_resend_to_mds(hdr);
103	}
104}
105
106static void filelayout_reset_read(struct nfs_pgio_header *hdr)
107{
108	struct rpc_task *task = &hdr->task;
109
110	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
111		dprintk("%s Reset task %5u for i/o through MDS "
112			"(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
113			hdr->task.tk_pid,
114			hdr->inode->i_sb->s_id,
115			(unsigned long long)NFS_FILEID(hdr->inode),
116			hdr->args.count,
117			(unsigned long long)hdr->args.offset);
118
119		task->tk_status = pnfs_read_done_resend_to_mds(hdr);
120	}
121}
122
123static int filelayout_async_handle_error(struct rpc_task *task,
124					 struct nfs4_state *state,
125					 struct nfs_client *clp,
126					 struct pnfs_layout_segment *lseg)
127{
128	struct pnfs_layout_hdr *lo = lseg->pls_layout;
129	struct inode *inode = lo->plh_inode;
130	struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
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	/* DS session errors */
138	case -NFS4ERR_BADSESSION:
139	case -NFS4ERR_BADSLOT:
140	case -NFS4ERR_BAD_HIGH_SLOT:
141	case -NFS4ERR_DEADSESSION:
142	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
143	case -NFS4ERR_SEQ_FALSE_RETRY:
144	case -NFS4ERR_SEQ_MISORDERED:
145		dprintk("%s ERROR %d, Reset session. Exchangeid "
146			"flags 0x%x\n", __func__, task->tk_status,
147			clp->cl_exchange_flags);
148		nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
149		break;
150	case -NFS4ERR_DELAY:
151	case -NFS4ERR_GRACE:
152		rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
153		break;
154	case -NFS4ERR_RETRY_UNCACHED_REP:
155		break;
156	/* Invalidate Layout errors */
157	case -NFS4ERR_ACCESS:
158	case -NFS4ERR_PNFS_NO_LAYOUT:
159	case -ESTALE:           /* mapped NFS4ERR_STALE */
160	case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
161	case -EISDIR:           /* mapped NFS4ERR_ISDIR */
162	case -NFS4ERR_FHEXPIRED:
163	case -NFS4ERR_WRONG_TYPE:
164		dprintk("%s Invalid layout error %d\n", __func__,
165			task->tk_status);
166		/*
167		 * Destroy layout so new i/o will get a new layout.
168		 * Layout will not be destroyed until all current lseg
169		 * references are put. Mark layout as invalid to resend failed
170		 * i/o and all i/o waiting on the slot table to the MDS until
171		 * layout is destroyed and a new valid layout is obtained.
172		 */
173		pnfs_destroy_layout(NFS_I(inode));
174		rpc_wake_up(&tbl->slot_tbl_waitq);
175		goto reset;
176	/* RPC connection errors */
177	case -ECONNREFUSED:
178	case -EHOSTDOWN:
179	case -EHOSTUNREACH:
180	case -ENETUNREACH:
181	case -EIO:
182	case -ETIMEDOUT:
183	case -EPIPE:
184		dprintk("%s DS connection error %d\n", __func__,
185			task->tk_status);
186		nfs4_mark_deviceid_unavailable(devid);
187		pnfs_error_mark_layout_for_return(inode, lseg);
188		pnfs_set_lo_fail(lseg);
189		rpc_wake_up(&tbl->slot_tbl_waitq);
190		fallthrough;
191	default:
192reset:
193		dprintk("%s Retry through MDS. Error %d\n", __func__,
194			task->tk_status);
195		return -NFS4ERR_RESET_TO_MDS;
196	}
197	task->tk_status = 0;
198	return -EAGAIN;
199}
200
201/* NFS_PROTO call done callback routines */
202
203static int filelayout_read_done_cb(struct rpc_task *task,
204				struct nfs_pgio_header *hdr)
205{
206	int err;
207
208	trace_nfs4_pnfs_read(hdr, task->tk_status);
209	err = filelayout_async_handle_error(task, hdr->args.context->state,
210					    hdr->ds_clp, hdr->lseg);
211
212	switch (err) {
213	case -NFS4ERR_RESET_TO_MDS:
214		filelayout_reset_read(hdr);
215		return task->tk_status;
216	case -EAGAIN:
217		rpc_restart_call_prepare(task);
218		return -EAGAIN;
219	}
220
221	return 0;
222}
223
224/*
225 * We reference the rpc_cred of the first WRITE that triggers the need for
226 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
227 * rfc5661 is not clear about which credential should be used.
228 */
229static void
230filelayout_set_layoutcommit(struct nfs_pgio_header *hdr)
231{
232	loff_t end_offs = 0;
233
234	if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds ||
235	    hdr->res.verf->committed == NFS_FILE_SYNC)
236		return;
237	if (hdr->res.verf->committed == NFS_DATA_SYNC)
238		end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
239
240	/* Note: if the write is unstable, don't set end_offs until commit */
241	pnfs_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
242	dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
243		(unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
244}
245
246bool
247filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node)
248{
249	return filelayout_test_devid_invalid(node) ||
250		nfs4_test_deviceid_unavailable(node);
251}
252
253static bool
254filelayout_reset_to_mds(struct pnfs_layout_segment *lseg)
255{
256	struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg);
257
258	return filelayout_test_devid_unavailable(node);
259}
260
261/*
262 * Call ops for the async read/write cases
263 * In the case of dense layouts, the offset needs to be reset to its
264 * original value.
265 */
266static void filelayout_read_prepare(struct rpc_task *task, void *data)
267{
268	struct nfs_pgio_header *hdr = data;
269
270	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
271		rpc_exit(task, -EIO);
272		return;
273	}
274	if (filelayout_reset_to_mds(hdr->lseg)) {
275		dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
276		filelayout_reset_read(hdr);
277		rpc_exit(task, 0);
278		return;
279	}
280	hdr->pgio_done_cb = filelayout_read_done_cb;
281
282	if (nfs4_setup_sequence(hdr->ds_clp,
283			&hdr->args.seq_args,
284			&hdr->res.seq_res,
285			task))
286		return;
287	if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
288			hdr->args.lock_context, FMODE_READ) == -EIO)
289		rpc_exit(task, -EIO); /* lost lock, terminate I/O */
290}
291
292static void filelayout_read_call_done(struct rpc_task *task, void *data)
293{
294	struct nfs_pgio_header *hdr = data;
295
296	dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
297
298	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
299	    task->tk_status == 0) {
300		nfs41_sequence_done(task, &hdr->res.seq_res);
301		return;
302	}
303
304	/* Note this may cause RPC to be resent */
305	hdr->mds_ops->rpc_call_done(task, data);
306}
307
308static void filelayout_read_count_stats(struct rpc_task *task, void *data)
309{
310	struct nfs_pgio_header *hdr = data;
311
312	rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
313}
314
315static int filelayout_write_done_cb(struct rpc_task *task,
316				struct nfs_pgio_header *hdr)
317{
318	int err;
319
320	trace_nfs4_pnfs_write(hdr, task->tk_status);
321	err = filelayout_async_handle_error(task, hdr->args.context->state,
322					    hdr->ds_clp, hdr->lseg);
323
324	switch (err) {
325	case -NFS4ERR_RESET_TO_MDS:
326		filelayout_reset_write(hdr);
327		return task->tk_status;
328	case -EAGAIN:
329		rpc_restart_call_prepare(task);
330		return -EAGAIN;
331	}
332
333	filelayout_set_layoutcommit(hdr);
334
335	/* zero out the fattr */
336	hdr->fattr.valid = 0;
337	if (task->tk_status >= 0)
338		nfs_writeback_update_inode(hdr);
339
340	return 0;
341}
342
343static int filelayout_commit_done_cb(struct rpc_task *task,
344				     struct nfs_commit_data *data)
345{
346	int err;
347
348	trace_nfs4_pnfs_commit_ds(data, task->tk_status);
349	err = filelayout_async_handle_error(task, NULL, data->ds_clp,
350					    data->lseg);
351
352	switch (err) {
353	case -NFS4ERR_RESET_TO_MDS:
354		pnfs_generic_prepare_to_resend_writes(data);
355		return -EAGAIN;
356	case -EAGAIN:
357		rpc_restart_call_prepare(task);
358		return -EAGAIN;
359	}
360
361	pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
362
363	return 0;
364}
365
366static void filelayout_write_prepare(struct rpc_task *task, void *data)
367{
368	struct nfs_pgio_header *hdr = data;
369
370	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
371		rpc_exit(task, -EIO);
372		return;
373	}
374	if (filelayout_reset_to_mds(hdr->lseg)) {
375		dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
376		filelayout_reset_write(hdr);
377		rpc_exit(task, 0);
378		return;
379	}
380	if (nfs4_setup_sequence(hdr->ds_clp,
381			&hdr->args.seq_args,
382			&hdr->res.seq_res,
383			task))
384		return;
385	if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
386			hdr->args.lock_context, FMODE_WRITE) == -EIO)
387		rpc_exit(task, -EIO); /* lost lock, terminate I/O */
388}
389
390static void filelayout_write_call_done(struct rpc_task *task, void *data)
391{
392	struct nfs_pgio_header *hdr = data;
393
394	if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
395	    task->tk_status == 0) {
396		nfs41_sequence_done(task, &hdr->res.seq_res);
397		return;
398	}
399
400	/* Note this may cause RPC to be resent */
401	hdr->mds_ops->rpc_call_done(task, data);
402}
403
404static void filelayout_write_count_stats(struct rpc_task *task, void *data)
405{
406	struct nfs_pgio_header *hdr = data;
407
408	rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
409}
410
411static void filelayout_commit_prepare(struct rpc_task *task, void *data)
412{
413	struct nfs_commit_data *wdata = data;
414
415	nfs4_setup_sequence(wdata->ds_clp,
416			&wdata->args.seq_args,
417			&wdata->res.seq_res,
418			task);
419}
420
421static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
422{
423	struct nfs_commit_data *cdata = data;
424
425	rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
426}
427
428static const struct rpc_call_ops filelayout_read_call_ops = {
429	.rpc_call_prepare = filelayout_read_prepare,
430	.rpc_call_done = filelayout_read_call_done,
431	.rpc_count_stats = filelayout_read_count_stats,
432	.rpc_release = pnfs_generic_rw_release,
433};
434
435static const struct rpc_call_ops filelayout_write_call_ops = {
436	.rpc_call_prepare = filelayout_write_prepare,
437	.rpc_call_done = filelayout_write_call_done,
438	.rpc_count_stats = filelayout_write_count_stats,
439	.rpc_release = pnfs_generic_rw_release,
440};
441
442static const struct rpc_call_ops filelayout_commit_call_ops = {
443	.rpc_call_prepare = filelayout_commit_prepare,
444	.rpc_call_done = pnfs_generic_write_commit_done,
445	.rpc_count_stats = filelayout_commit_count_stats,
446	.rpc_release = pnfs_generic_commit_release,
447};
448
449static enum pnfs_try_status
450filelayout_read_pagelist(struct nfs_pgio_header *hdr)
451{
452	struct pnfs_layout_segment *lseg = hdr->lseg;
453	struct nfs4_pnfs_ds *ds;
454	struct rpc_clnt *ds_clnt;
455	loff_t offset = hdr->args.offset;
456	u32 j, idx;
457	struct nfs_fh *fh;
458
459	dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
460		__func__, hdr->inode->i_ino,
461		hdr->args.pgbase, (size_t)hdr->args.count, offset);
462
463	/* Retrieve the correct rpc_client for the byte range */
464	j = nfs4_fl_calc_j_index(lseg, offset);
465	idx = nfs4_fl_calc_ds_index(lseg, j);
466	ds = nfs4_fl_prepare_ds(lseg, idx);
467	if (!ds)
468		return PNFS_NOT_ATTEMPTED;
469
470	ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
471	if (IS_ERR(ds_clnt))
472		return PNFS_NOT_ATTEMPTED;
473
474	dprintk("%s USE DS: %s cl_count %d\n", __func__,
475		ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count));
476
477	/* No multipath support. Use first DS */
478	refcount_inc(&ds->ds_clp->cl_count);
479	hdr->ds_clp = ds->ds_clp;
480	hdr->ds_commit_idx = idx;
481	fh = nfs4_fl_select_ds_fh(lseg, j);
482	if (fh)
483		hdr->args.fh = fh;
484
485	hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
486	hdr->mds_offset = offset;
487
488	/* Perform an asynchronous read to ds */
489	nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
490			  NFS_PROTO(hdr->inode), &filelayout_read_call_ops,
491			  0, RPC_TASK_SOFTCONN);
492	return PNFS_ATTEMPTED;
493}
494
495/* Perform async writes. */
496static enum pnfs_try_status
497filelayout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
498{
499	struct pnfs_layout_segment *lseg = hdr->lseg;
500	struct nfs4_pnfs_ds *ds;
501	struct rpc_clnt *ds_clnt;
502	loff_t offset = hdr->args.offset;
503	u32 j, idx;
504	struct nfs_fh *fh;
505
506	/* Retrieve the correct rpc_client for the byte range */
507	j = nfs4_fl_calc_j_index(lseg, offset);
508	idx = nfs4_fl_calc_ds_index(lseg, j);
509	ds = nfs4_fl_prepare_ds(lseg, idx);
510	if (!ds)
511		return PNFS_NOT_ATTEMPTED;
512
513	ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, hdr->inode);
514	if (IS_ERR(ds_clnt))
515		return PNFS_NOT_ATTEMPTED;
516
517	dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d\n",
518		__func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
519		offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count));
520
521	hdr->pgio_done_cb = filelayout_write_done_cb;
522	refcount_inc(&ds->ds_clp->cl_count);
523	hdr->ds_clp = ds->ds_clp;
524	hdr->ds_commit_idx = idx;
525	fh = nfs4_fl_select_ds_fh(lseg, j);
526	if (fh)
527		hdr->args.fh = fh;
528	hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
529
530	/* Perform an asynchronous write */
531	nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
532			  NFS_PROTO(hdr->inode), &filelayout_write_call_ops,
533			  sync, RPC_TASK_SOFTCONN);
534	return PNFS_ATTEMPTED;
535}
536
537static int
538filelayout_check_deviceid(struct pnfs_layout_hdr *lo,
539			  struct nfs4_filelayout_segment *fl,
540			  gfp_t gfp_flags)
541{
542	struct nfs4_deviceid_node *d;
543	struct nfs4_file_layout_dsaddr *dsaddr;
544	int status = -EINVAL;
545
546	/* Is the deviceid already set? If so, we're good. */
547	if (fl->dsaddr != NULL)
548		return 0;
549
550	/* find and reference the deviceid */
551	d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &fl->deviceid,
552			lo->plh_lc_cred, gfp_flags);
553	if (d == NULL)
554		goto out;
555
556	dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
557	/* Found deviceid is unavailable */
558	if (filelayout_test_devid_unavailable(&dsaddr->id_node))
559		goto out_put;
560
561	if (fl->first_stripe_index >= dsaddr->stripe_count) {
562		dprintk("%s Bad first_stripe_index %u\n",
563				__func__, fl->first_stripe_index);
564		goto out_put;
565	}
566
567	if ((fl->stripe_type == STRIPE_SPARSE &&
568	    fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
569	    (fl->stripe_type == STRIPE_DENSE &&
570	    fl->num_fh != dsaddr->stripe_count)) {
571		dprintk("%s num_fh %u not valid for given packing\n",
572			__func__, fl->num_fh);
573		goto out_put;
574	}
575	status = 0;
576
577	/*
578	 * Atomic compare and xchange to ensure we don't scribble
579	 * over a non-NULL pointer.
580	 */
581	if (cmpxchg(&fl->dsaddr, NULL, dsaddr) != NULL)
582		goto out_put;
583out:
584	return status;
585out_put:
586	nfs4_fl_put_deviceid(dsaddr);
587	goto out;
588}
589
590/*
591 * filelayout_check_layout()
592 *
593 * Make sure layout segment parameters are sane WRT the device.
594 * At this point no generic layer initialization of the lseg has occurred,
595 * and nothing has been added to the layout_hdr cache.
596 *
597 */
598static int
599filelayout_check_layout(struct pnfs_layout_hdr *lo,
600			struct nfs4_filelayout_segment *fl,
601			struct nfs4_layoutget_res *lgr,
602			gfp_t gfp_flags)
603{
604	int status = -EINVAL;
605
606	dprintk("--> %s\n", __func__);
607
608	/* FIXME: remove this check when layout segment support is added */
609	if (lgr->range.offset != 0 ||
610	    lgr->range.length != NFS4_MAX_UINT64) {
611		dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
612			__func__);
613		goto out;
614	}
615
616	if (fl->pattern_offset > lgr->range.offset) {
617		dprintk("%s pattern_offset %lld too large\n",
618				__func__, fl->pattern_offset);
619		goto out;
620	}
621
622	if (!fl->stripe_unit) {
623		dprintk("%s Invalid stripe unit (%u)\n",
624			__func__, fl->stripe_unit);
625		goto out;
626	}
627
628	status = 0;
629out:
630	dprintk("--> %s returns %d\n", __func__, status);
631	return status;
632}
633
634static void _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
635{
636	int i;
637
638	if (fl->fh_array) {
639		for (i = 0; i < fl->num_fh; i++) {
640			if (!fl->fh_array[i])
641				break;
642			kfree(fl->fh_array[i]);
643		}
644		kfree(fl->fh_array);
645	}
646	kfree(fl);
647}
648
649static int
650filelayout_decode_layout(struct pnfs_layout_hdr *flo,
651			 struct nfs4_filelayout_segment *fl,
652			 struct nfs4_layoutget_res *lgr,
653			 gfp_t gfp_flags)
654{
655	struct xdr_stream stream;
656	struct xdr_buf buf;
657	struct page *scratch;
658	__be32 *p;
659	uint32_t nfl_util;
660	int i;
661
662	dprintk("%s: set_layout_map Begin\n", __func__);
663
664	scratch = alloc_page(gfp_flags);
665	if (!scratch)
666		return -ENOMEM;
667
668	xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
669	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
670
671	/* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
672	 * num_fh (4) */
673	p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
674	if (unlikely(!p))
675		goto out_err;
676
677	memcpy(&fl->deviceid, p, sizeof(fl->deviceid));
678	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
679	nfs4_print_deviceid(&fl->deviceid);
680
681	nfl_util = be32_to_cpup(p++);
682	if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
683		fl->commit_through_mds = 1;
684	if (nfl_util & NFL4_UFLG_DENSE)
685		fl->stripe_type = STRIPE_DENSE;
686	else
687		fl->stripe_type = STRIPE_SPARSE;
688	fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
689
690	fl->first_stripe_index = be32_to_cpup(p++);
691	p = xdr_decode_hyper(p, &fl->pattern_offset);
692	fl->num_fh = be32_to_cpup(p++);
693
694	dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
695		__func__, nfl_util, fl->num_fh, fl->first_stripe_index,
696		fl->pattern_offset);
697
698	/* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
699	 * Futher checking is done in filelayout_check_layout */
700	if (fl->num_fh >
701	    max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
702		goto out_err;
703
704	if (fl->num_fh > 0) {
705		fl->fh_array = kcalloc(fl->num_fh, sizeof(fl->fh_array[0]),
706				       gfp_flags);
707		if (!fl->fh_array)
708			goto out_err;
709	}
710
711	for (i = 0; i < fl->num_fh; i++) {
712		/* Do we want to use a mempool here? */
713		fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
714		if (!fl->fh_array[i])
715			goto out_err;
716
717		p = xdr_inline_decode(&stream, 4);
718		if (unlikely(!p))
719			goto out_err;
720		fl->fh_array[i]->size = be32_to_cpup(p++);
721		if (fl->fh_array[i]->size > NFS_MAXFHSIZE) {
722			printk(KERN_ERR "NFS: Too big fh %d received %d\n",
723			       i, fl->fh_array[i]->size);
724			goto out_err;
725		}
726
727		p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
728		if (unlikely(!p))
729			goto out_err;
730		memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
731		dprintk("DEBUG: %s: fh len %d\n", __func__,
732			fl->fh_array[i]->size);
733	}
734
735	__free_page(scratch);
736	return 0;
737
738out_err:
739	__free_page(scratch);
740	return -EIO;
741}
742
743static void
744filelayout_free_lseg(struct pnfs_layout_segment *lseg)
745{
746	struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
747
748	dprintk("--> %s\n", __func__);
749	if (fl->dsaddr != NULL)
750		nfs4_fl_put_deviceid(fl->dsaddr);
751	/* This assumes a single RW lseg */
752	if (lseg->pls_range.iomode == IOMODE_RW) {
753		struct nfs4_filelayout *flo;
754		struct inode *inode;
755
756		flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
757		inode = flo->generic_hdr.plh_inode;
758		spin_lock(&inode->i_lock);
759		pnfs_generic_ds_cinfo_release_lseg(&flo->commit_info, lseg);
760		spin_unlock(&inode->i_lock);
761	}
762	_filelayout_free_lseg(fl);
763}
764
765static struct pnfs_layout_segment *
766filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
767		      struct nfs4_layoutget_res *lgr,
768		      gfp_t gfp_flags)
769{
770	struct nfs4_filelayout_segment *fl;
771	int rc;
772
773	dprintk("--> %s\n", __func__);
774	fl = kzalloc(sizeof(*fl), gfp_flags);
775	if (!fl)
776		return NULL;
777
778	rc = filelayout_decode_layout(layoutid, fl, lgr, gfp_flags);
779	if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, gfp_flags)) {
780		_filelayout_free_lseg(fl);
781		return NULL;
782	}
783	return &fl->generic_hdr;
784}
785
786static bool
787filelayout_lseg_is_striped(const struct nfs4_filelayout_segment *flseg)
788{
789	return flseg->num_fh > 1;
790}
791
792/*
793 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
794 *
795 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
796 * of bytes (maximum @req->wb_bytes) that can be coalesced.
797 */
798static size_t
799filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
800		   struct nfs_page *req)
801{
802	unsigned int size;
803	u64 p_stripe, r_stripe;
804	u32 stripe_offset;
805	u64 segment_offset = pgio->pg_lseg->pls_range.offset;
806	u32 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
807
808	/* calls nfs_generic_pg_test */
809	size = pnfs_generic_pg_test(pgio, prev, req);
810	if (!size)
811		return 0;
812	else if (!filelayout_lseg_is_striped(FILELAYOUT_LSEG(pgio->pg_lseg)))
813		return size;
814
815	/* see if req and prev are in the same stripe */
816	if (prev) {
817		p_stripe = (u64)req_offset(prev) - segment_offset;
818		r_stripe = (u64)req_offset(req) - segment_offset;
819		do_div(p_stripe, stripe_unit);
820		do_div(r_stripe, stripe_unit);
821
822		if (p_stripe != r_stripe)
823			return 0;
824	}
825
826	/* calculate remaining bytes in the current stripe */
827	div_u64_rem((u64)req_offset(req) - segment_offset,
828			stripe_unit,
829			&stripe_offset);
830	WARN_ON_ONCE(stripe_offset > stripe_unit);
831	if (stripe_offset >= stripe_unit)
832		return 0;
833	return min(stripe_unit - (unsigned int)stripe_offset, size);
834}
835
836static struct pnfs_layout_segment *
837fl_pnfs_update_layout(struct inode *ino,
838		      struct nfs_open_context *ctx,
839		      loff_t pos,
840		      u64 count,
841		      enum pnfs_iomode iomode,
842		      bool strict_iomode,
843		      gfp_t gfp_flags)
844{
845	struct pnfs_layout_segment *lseg = NULL;
846	struct pnfs_layout_hdr *lo;
847	struct nfs4_filelayout_segment *fl;
848	int status;
849
850	lseg = pnfs_update_layout(ino, ctx, pos, count, iomode, strict_iomode,
851				  gfp_flags);
852	if (IS_ERR_OR_NULL(lseg))
853		goto out;
854
855	lo = NFS_I(ino)->layout;
856	fl = FILELAYOUT_LSEG(lseg);
857
858	status = filelayout_check_deviceid(lo, fl, gfp_flags);
859	if (status) {
860		pnfs_put_lseg(lseg);
861		lseg = NULL;
862	}
863out:
864	return lseg;
865}
866
867static void
868filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
869			struct nfs_page *req)
870{
871	pnfs_generic_pg_check_layout(pgio);
872	if (!pgio->pg_lseg) {
873		pgio->pg_lseg = fl_pnfs_update_layout(pgio->pg_inode,
874						      nfs_req_openctx(req),
875						      0,
876						      NFS4_MAX_UINT64,
877						      IOMODE_READ,
878						      false,
879						      GFP_KERNEL);
880		if (IS_ERR(pgio->pg_lseg)) {
881			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
882			pgio->pg_lseg = NULL;
883			return;
884		}
885	}
886	/* If no lseg, fall back to read through mds */
887	if (pgio->pg_lseg == NULL)
888		nfs_pageio_reset_read_mds(pgio);
889}
890
891static void
892filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
893			 struct nfs_page *req)
894{
895	pnfs_generic_pg_check_layout(pgio);
896	if (!pgio->pg_lseg) {
897		pgio->pg_lseg = fl_pnfs_update_layout(pgio->pg_inode,
898						      nfs_req_openctx(req),
899						      0,
900						      NFS4_MAX_UINT64,
901						      IOMODE_RW,
902						      false,
903						      GFP_NOFS);
904		if (IS_ERR(pgio->pg_lseg)) {
905			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
906			pgio->pg_lseg = NULL;
907			return;
908		}
909	}
910
911	/* If no lseg, fall back to write through mds */
912	if (pgio->pg_lseg == NULL)
913		nfs_pageio_reset_write_mds(pgio);
914}
915
916static const struct nfs_pageio_ops filelayout_pg_read_ops = {
917	.pg_init = filelayout_pg_init_read,
918	.pg_test = filelayout_pg_test,
919	.pg_doio = pnfs_generic_pg_readpages,
920	.pg_cleanup = pnfs_generic_pg_cleanup,
921};
922
923static const struct nfs_pageio_ops filelayout_pg_write_ops = {
924	.pg_init = filelayout_pg_init_write,
925	.pg_test = filelayout_pg_test,
926	.pg_doio = pnfs_generic_pg_writepages,
927	.pg_cleanup = pnfs_generic_pg_cleanup,
928};
929
930static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
931{
932	if (fl->stripe_type == STRIPE_SPARSE)
933		return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
934	else
935		return j;
936}
937
938static void
939filelayout_mark_request_commit(struct nfs_page *req,
940			       struct pnfs_layout_segment *lseg,
941			       struct nfs_commit_info *cinfo,
942			       u32 ds_commit_idx)
943
944{
945	struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
946	u32 i, j;
947
948	if (fl->commit_through_mds) {
949		nfs_request_add_commit_list(req, cinfo);
950	} else {
951		/* Note that we are calling nfs4_fl_calc_j_index on each page
952		 * that ends up being committed to a data server.  An attractive
953		 * alternative is to add a field to nfs_write_data and nfs_page
954		 * to store the value calculated in filelayout_write_pagelist
955		 * and just use that here.
956		 */
957		j = nfs4_fl_calc_j_index(lseg, req_offset(req));
958		i = select_bucket_index(fl, j);
959		pnfs_layout_mark_request_commit(req, lseg, cinfo, i);
960	}
961}
962
963static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
964{
965	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
966
967	if (flseg->stripe_type == STRIPE_SPARSE)
968		return i;
969	else
970		return nfs4_fl_calc_ds_index(lseg, i);
971}
972
973static struct nfs_fh *
974select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
975{
976	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
977
978	if (flseg->stripe_type == STRIPE_SPARSE) {
979		if (flseg->num_fh == 1)
980			i = 0;
981		else if (flseg->num_fh == 0)
982			/* Use the MDS OPEN fh set in nfs_read_rpcsetup */
983			return NULL;
984	}
985	return flseg->fh_array[i];
986}
987
988static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
989{
990	struct pnfs_layout_segment *lseg = data->lseg;
991	struct nfs4_pnfs_ds *ds;
992	struct rpc_clnt *ds_clnt;
993	u32 idx;
994	struct nfs_fh *fh;
995
996	idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
997	ds = nfs4_fl_prepare_ds(lseg, idx);
998	if (!ds)
999		goto out_err;
1000
1001	ds_clnt = nfs4_find_or_create_ds_client(ds->ds_clp, data->inode);
1002	if (IS_ERR(ds_clnt))
1003		goto out_err;
1004
1005	dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1006		data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count));
1007	data->commit_done_cb = filelayout_commit_done_cb;
1008	refcount_inc(&ds->ds_clp->cl_count);
1009	data->ds_clp = ds->ds_clp;
1010	fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1011	if (fh)
1012		data->args.fh = fh;
1013	return nfs_initiate_commit(ds_clnt, data, NFS_PROTO(data->inode),
1014				   &filelayout_commit_call_ops, how,
1015				   RPC_TASK_SOFTCONN);
1016out_err:
1017	pnfs_generic_prepare_to_resend_writes(data);
1018	pnfs_generic_commit_release(data);
1019	return -EAGAIN;
1020}
1021
1022static int
1023filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1024			   int how, struct nfs_commit_info *cinfo)
1025{
1026	return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1027					    filelayout_initiate_commit);
1028}
1029
1030static struct nfs4_deviceid_node *
1031filelayout_alloc_deviceid_node(struct nfs_server *server,
1032		struct pnfs_device *pdev, gfp_t gfp_flags)
1033{
1034	struct nfs4_file_layout_dsaddr *dsaddr;
1035
1036	dsaddr = nfs4_fl_alloc_deviceid_node(server, pdev, gfp_flags);
1037	if (!dsaddr)
1038		return NULL;
1039	return &dsaddr->id_node;
1040}
1041
1042static void
1043filelayout_free_deviceid_node(struct nfs4_deviceid_node *d)
1044{
1045	nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1046}
1047
1048static struct pnfs_layout_hdr *
1049filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1050{
1051	struct nfs4_filelayout *flo;
1052
1053	flo = kzalloc(sizeof(*flo), gfp_flags);
1054	if (flo == NULL)
1055		return NULL;
1056	pnfs_init_ds_commit_info(&flo->commit_info);
1057	flo->commit_info.ops = &filelayout_commit_ops;
1058	return &flo->generic_hdr;
1059}
1060
1061static void
1062filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1063{
1064	kfree_rcu(FILELAYOUT_FROM_HDR(lo), generic_hdr.plh_rcu);
1065}
1066
1067static struct pnfs_ds_commit_info *
1068filelayout_get_ds_info(struct inode *inode)
1069{
1070	struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1071
1072	if (layout == NULL)
1073		return NULL;
1074	else
1075		return &FILELAYOUT_FROM_HDR(layout)->commit_info;
1076}
1077
1078static void
1079filelayout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1080		struct pnfs_layout_segment *lseg)
1081{
1082	struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
1083	struct inode *inode = lseg->pls_layout->plh_inode;
1084	struct pnfs_commit_array *array, *new;
1085	unsigned int size = (fl->stripe_type == STRIPE_SPARSE) ?
1086		fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
1087
1088	new = pnfs_alloc_commit_array(size, GFP_NOIO);
1089	if (new) {
1090		spin_lock(&inode->i_lock);
1091		array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1092		spin_unlock(&inode->i_lock);
1093		if (array != new)
1094			pnfs_free_commit_array(new);
1095	}
1096}
1097
1098static void
1099filelayout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1100		struct inode *inode)
1101{
1102	spin_lock(&inode->i_lock);
1103	pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1104	spin_unlock(&inode->i_lock);
1105}
1106
1107static const struct pnfs_commit_ops filelayout_commit_ops = {
1108	.setup_ds_info		= filelayout_setup_ds_info,
1109	.release_ds_info	= filelayout_release_ds_info,
1110	.mark_request_commit	= filelayout_mark_request_commit,
1111	.clear_request_commit	= pnfs_generic_clear_request_commit,
1112	.scan_commit_lists	= pnfs_generic_scan_commit_lists,
1113	.recover_commit_reqs	= pnfs_generic_recover_commit_reqs,
1114	.search_commit_reqs	= pnfs_generic_search_commit_reqs,
1115	.commit_pagelist	= filelayout_commit_pagelist,
1116};
1117
1118static struct pnfs_layoutdriver_type filelayout_type = {
1119	.id			= LAYOUT_NFSV4_1_FILES,
1120	.name			= "LAYOUT_NFSV4_1_FILES",
1121	.owner			= THIS_MODULE,
1122	.flags			= PNFS_LAYOUTGET_ON_OPEN,
1123	.max_layoutget_response	= 4096, /* 1 page or so... */
1124	.alloc_layout_hdr	= filelayout_alloc_layout_hdr,
1125	.free_layout_hdr	= filelayout_free_layout_hdr,
1126	.alloc_lseg		= filelayout_alloc_lseg,
1127	.free_lseg		= filelayout_free_lseg,
1128	.pg_read_ops		= &filelayout_pg_read_ops,
1129	.pg_write_ops		= &filelayout_pg_write_ops,
1130	.get_ds_info		= &filelayout_get_ds_info,
1131	.read_pagelist		= filelayout_read_pagelist,
1132	.write_pagelist		= filelayout_write_pagelist,
1133	.alloc_deviceid_node	= filelayout_alloc_deviceid_node,
1134	.free_deviceid_node	= filelayout_free_deviceid_node,
1135	.sync			= pnfs_nfs_generic_sync,
1136};
1137
1138static int __init nfs4filelayout_init(void)
1139{
1140	printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1141	       __func__);
1142	return pnfs_register_layoutdriver(&filelayout_type);
1143}
1144
1145static void __exit nfs4filelayout_exit(void)
1146{
1147	printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1148	       __func__);
1149	pnfs_unregister_layoutdriver(&filelayout_type);
1150}
1151
1152MODULE_ALIAS("nfs-layouttype4-1");
1153
1154module_init(nfs4filelayout_init);
1155module_exit(nfs4filelayout_exit);
1156