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
 
 
 
 35#include "internal.h"
 
 36#include "nfs4filelayout.h"
 
 37
 38#define NFSDBG_FACILITY         NFSDBG_PNFS_LD
 39
 40MODULE_LICENSE("GPL");
 41MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
 42MODULE_DESCRIPTION("The NFSv4 file layout driver");
 43
 44#define FILELAYOUT_POLL_RETRY_MAX     (15*HZ)
 45
 46static loff_t
 47filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
 48			    loff_t offset)
 49{
 50	u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
 51	u64 tmp;
 
 52
 53	offset -= flseg->pattern_offset;
 54	tmp = offset;
 55	do_div(tmp, stripe_width);
 56
 57	return tmp * flseg->stripe_unit + do_div(offset, flseg->stripe_unit);
 58}
 59
 60/* This function is used by the layout driver to calculate the
 61 * offset of the file on the dserver based on whether the
 62 * layout type is STRIPE_DENSE or STRIPE_SPARSE
 63 */
 64static loff_t
 65filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
 66{
 67	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
 68
 69	switch (flseg->stripe_type) {
 70	case STRIPE_SPARSE:
 71		return offset;
 72
 73	case STRIPE_DENSE:
 74		return filelayout_get_dense_offset(flseg, offset);
 75	}
 76
 77	BUG();
 78}
 79
 80/* For data server errors we don't recover from */
 81static void
 82filelayout_set_lo_fail(struct pnfs_layout_segment *lseg)
 83{
 84	if (lseg->pls_range.iomode == IOMODE_RW) {
 85		dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__);
 86		set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
 87	} else {
 88		dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__);
 89		set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 90	}
 91}
 92
 
 
 
 
 
 
 
 93static int filelayout_async_handle_error(struct rpc_task *task,
 94					 struct nfs4_state *state,
 95					 struct nfs_client *clp,
 96					 int *reset)
 97{
 
 
 
 
 
 
 
 98	if (task->tk_status >= 0)
 99		return 0;
100
101	*reset = 0;
102
103	switch (task->tk_status) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
104	case -NFS4ERR_BADSESSION:
105	case -NFS4ERR_BADSLOT:
106	case -NFS4ERR_BAD_HIGH_SLOT:
107	case -NFS4ERR_DEADSESSION:
108	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
109	case -NFS4ERR_SEQ_FALSE_RETRY:
110	case -NFS4ERR_SEQ_MISORDERED:
111		dprintk("%s ERROR %d, Reset session. Exchangeid "
112			"flags 0x%x\n", __func__, task->tk_status,
113			clp->cl_exchange_flags);
114		nfs4_schedule_session_recovery(clp->cl_session);
115		break;
116	case -NFS4ERR_DELAY:
117	case -NFS4ERR_GRACE:
118	case -EKEYEXPIRED:
119		rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
120		break;
121	case -NFS4ERR_RETRY_UNCACHED_REP:
122		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
123	default:
124		dprintk("%s DS error. Retry through MDS %d\n", __func__,
 
125			task->tk_status);
126		*reset = 1;
127		break;
128	}
 
129	task->tk_status = 0;
130	return -EAGAIN;
 
 
 
 
 
 
 
 
131}
132
133/* NFS_PROTO call done callback routines */
134
135static int filelayout_read_done_cb(struct rpc_task *task,
136				struct nfs_read_data *data)
137{
138	struct nfs_client *clp = data->ds_clp;
139	int reset = 0;
140
141	dprintk("%s DS read\n", __func__);
142
143	if (filelayout_async_handle_error(task, data->args.context->state,
144					  data->ds_clp, &reset) == -EAGAIN) {
145		dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
146			__func__, data->ds_clp, data->ds_clp->cl_session);
147		if (reset) {
148			filelayout_set_lo_fail(data->lseg);
149			nfs4_reset_read(task, data);
150			clp = NFS_SERVER(data->inode)->nfs_client;
151		}
152		nfs_restart_rpc(task, clp);
153		return -EAGAIN;
154	}
155
156	return 0;
157}
158
159/*
160 * We reference the rpc_cred of the first WRITE that triggers the need for
161 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
162 * rfc5661 is not clear about which credential should be used.
163 */
164static void
165filelayout_set_layoutcommit(struct nfs_write_data *wdata)
166{
167	if (FILELAYOUT_LSEG(wdata->lseg)->commit_through_mds ||
 
 
168	    wdata->res.verf->committed == NFS_FILE_SYNC)
169		return;
170
171	pnfs_set_layoutcommit(wdata);
172	dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino,
173		(unsigned long) NFS_I(wdata->inode)->layout->plh_lwb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
174}
175
176/*
177 * Call ops for the async read/write cases
178 * In the case of dense layouts, the offset needs to be reset to its
179 * original value.
180 */
181static void filelayout_read_prepare(struct rpc_task *task, void *data)
182{
183	struct nfs_read_data *rdata = (struct nfs_read_data *)data;
184
 
 
 
 
 
 
 
 
 
 
185	rdata->read_done_cb = filelayout_read_done_cb;
186
187	if (nfs41_setup_sequence(rdata->ds_clp->cl_session,
188				&rdata->args.seq_args, &rdata->res.seq_res,
189				0, task))
 
190		return;
191
192	rpc_call_start(task);
 
193}
194
195static void filelayout_read_call_done(struct rpc_task *task, void *data)
196{
197	struct nfs_read_data *rdata = (struct nfs_read_data *)data;
198
199	dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
200
 
 
 
 
 
 
201	/* Note this may cause RPC to be resent */
202	rdata->mds_ops->rpc_call_done(task, data);
 
 
 
 
 
 
 
203}
204
205static void filelayout_read_release(void *data)
206{
207	struct nfs_read_data *rdata = (struct nfs_read_data *)data;
 
208
209	rdata->mds_ops->rpc_release(data);
 
 
210}
211
212static int filelayout_write_done_cb(struct rpc_task *task,
213				struct nfs_write_data *data)
214{
215	int reset = 0;
 
216
217	if (filelayout_async_handle_error(task, data->args.context->state,
218					  data->ds_clp, &reset) == -EAGAIN) {
219		struct nfs_client *clp;
220
221		dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
222			__func__, data->ds_clp, data->ds_clp->cl_session);
223		if (reset) {
224			filelayout_set_lo_fail(data->lseg);
225			nfs4_reset_write(task, data);
226			clp = NFS_SERVER(data->inode)->nfs_client;
227		} else
228			clp = data->ds_clp;
229		nfs_restart_rpc(task, clp);
230		return -EAGAIN;
231	}
232
233	filelayout_set_layoutcommit(data);
234	return 0;
235}
236
237/* Fake up some data that will cause nfs_commit_release to retry the writes. */
238static void prepare_to_resend_writes(struct nfs_write_data *data)
239{
240	struct nfs_page *first = nfs_list_entry(data->pages.next);
241
242	data->task.tk_status = 0;
243	memcpy(data->verf.verifier, first->wb_verf.verifier,
244	       sizeof(first->wb_verf.verifier));
245	data->verf.verifier[0]++; /* ensure verifier mismatch */
246}
247
248static int filelayout_commit_done_cb(struct rpc_task *task,
249				     struct nfs_write_data *data)
250{
251	int reset = 0;
252
253	if (filelayout_async_handle_error(task, data->args.context->state,
254					  data->ds_clp, &reset) == -EAGAIN) {
255		dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
256			__func__, data->ds_clp, data->ds_clp->cl_session);
257		if (reset) {
258			prepare_to_resend_writes(data);
259			filelayout_set_lo_fail(data->lseg);
260		} else
261			nfs_restart_rpc(task, data->ds_clp);
 
262		return -EAGAIN;
263	}
264
265	return 0;
266}
267
268static void filelayout_write_prepare(struct rpc_task *task, void *data)
269{
270	struct nfs_write_data *wdata = (struct nfs_write_data *)data;
271
 
 
 
 
 
 
 
 
 
 
272	if (nfs41_setup_sequence(wdata->ds_clp->cl_session,
273				&wdata->args.seq_args, &wdata->res.seq_res,
274				0, task))
 
275		return;
276
277	rpc_call_start(task);
 
278}
279
280static void filelayout_write_call_done(struct rpc_task *task, void *data)
281{
282	struct nfs_write_data *wdata = (struct nfs_write_data *)data;
 
 
 
 
 
 
283
284	/* Note this may cause RPC to be resent */
285	wdata->mds_ops->rpc_call_done(task, data);
 
 
 
 
 
 
 
286}
287
288static void filelayout_write_release(void *data)
289{
290	struct nfs_write_data *wdata = (struct nfs_write_data *)data;
 
291
292	wdata->mds_ops->rpc_release(data);
 
 
293}
294
295static void filelayout_commit_release(void *data)
296{
297	struct nfs_write_data *wdata = (struct nfs_write_data *)data;
298
299	nfs_commit_release_pages(wdata);
300	if (atomic_dec_and_test(&NFS_I(wdata->inode)->commits_outstanding))
301		nfs_commit_clear_lock(NFS_I(wdata->inode));
302	nfs_commitdata_release(wdata);
303}
304
305struct rpc_call_ops filelayout_read_call_ops = {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
306	.rpc_call_prepare = filelayout_read_prepare,
307	.rpc_call_done = filelayout_read_call_done,
 
308	.rpc_release = filelayout_read_release,
309};
310
311struct rpc_call_ops filelayout_write_call_ops = {
312	.rpc_call_prepare = filelayout_write_prepare,
313	.rpc_call_done = filelayout_write_call_done,
 
314	.rpc_release = filelayout_write_release,
315};
316
317struct rpc_call_ops filelayout_commit_call_ops = {
318	.rpc_call_prepare = filelayout_write_prepare,
319	.rpc_call_done = filelayout_write_call_done,
 
320	.rpc_release = filelayout_commit_release,
321};
322
323static enum pnfs_try_status
324filelayout_read_pagelist(struct nfs_read_data *data)
325{
326	struct pnfs_layout_segment *lseg = data->lseg;
 
327	struct nfs4_pnfs_ds *ds;
 
328	loff_t offset = data->args.offset;
329	u32 j, idx;
330	struct nfs_fh *fh;
331	int status;
332
333	dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
334		__func__, data->inode->i_ino,
335		data->args.pgbase, (size_t)data->args.count, offset);
336
337	if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags))
338		return PNFS_NOT_ATTEMPTED;
339
340	/* Retrieve the correct rpc_client for the byte range */
341	j = nfs4_fl_calc_j_index(lseg, offset);
342	idx = nfs4_fl_calc_ds_index(lseg, j);
343	ds = nfs4_fl_prepare_ds(lseg, idx);
344	if (!ds) {
345		/* Either layout fh index faulty, or ds connect failed */
346		set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
347		set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
 
348		return PNFS_NOT_ATTEMPTED;
349	}
350	dprintk("%s USE DS: %s\n", __func__, ds->ds_remotestr);
 
351
352	/* No multipath support. Use first DS */
 
353	data->ds_clp = ds->ds_clp;
354	fh = nfs4_fl_select_ds_fh(lseg, j);
355	if (fh)
356		data->args.fh = fh;
357
358	data->args.offset = filelayout_get_dserver_offset(lseg, offset);
359	data->mds_offset = offset;
360
361	/* Perform an asynchronous read to ds */
362	status = nfs_initiate_read(data, ds->ds_clp->cl_rpcclient,
363				   &filelayout_read_call_ops);
364	BUG_ON(status != 0);
365	return PNFS_ATTEMPTED;
366}
367
368/* Perform async writes. */
369static enum pnfs_try_status
370filelayout_write_pagelist(struct nfs_write_data *data, int sync)
371{
372	struct pnfs_layout_segment *lseg = data->lseg;
 
373	struct nfs4_pnfs_ds *ds;
 
374	loff_t offset = data->args.offset;
375	u32 j, idx;
376	struct nfs_fh *fh;
377	int status;
378
379	if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags))
380		return PNFS_NOT_ATTEMPTED;
381
382	/* Retrieve the correct rpc_client for the byte range */
383	j = nfs4_fl_calc_j_index(lseg, offset);
384	idx = nfs4_fl_calc_ds_index(lseg, j);
385	ds = nfs4_fl_prepare_ds(lseg, idx);
386	if (!ds) {
387		printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__);
388		set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
389		set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
390		return PNFS_NOT_ATTEMPTED;
391	}
392	dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s\n", __func__,
393		data->inode->i_ino, sync, (size_t) data->args.count, offset,
394		ds->ds_remotestr);
 
 
 
 
395
396	data->write_done_cb = filelayout_write_done_cb;
 
397	data->ds_clp = ds->ds_clp;
398	fh = nfs4_fl_select_ds_fh(lseg, j);
399	if (fh)
400		data->args.fh = fh;
401	/*
402	 * Get the file offset on the dserver. Set the write offset to
403	 * this offset and save the original offset.
404	 */
405	data->args.offset = filelayout_get_dserver_offset(lseg, offset);
406
407	/* Perform an asynchronous write */
408	status = nfs_initiate_write(data, ds->ds_clp->cl_rpcclient,
409				    &filelayout_write_call_ops, sync);
410	BUG_ON(status != 0);
411	return PNFS_ATTEMPTED;
412}
413
414/*
415 * filelayout_check_layout()
416 *
417 * Make sure layout segment parameters are sane WRT the device.
418 * At this point no generic layer initialization of the lseg has occurred,
419 * and nothing has been added to the layout_hdr cache.
420 *
421 */
422static int
423filelayout_check_layout(struct pnfs_layout_hdr *lo,
424			struct nfs4_filelayout_segment *fl,
425			struct nfs4_layoutget_res *lgr,
426			struct nfs4_deviceid *id,
427			gfp_t gfp_flags)
428{
429	struct nfs4_deviceid_node *d;
430	struct nfs4_file_layout_dsaddr *dsaddr;
431	int status = -EINVAL;
432	struct nfs_server *nfss = NFS_SERVER(lo->plh_inode);
433
434	dprintk("--> %s\n", __func__);
435
436	/* FIXME: remove this check when layout segment support is added */
437	if (lgr->range.offset != 0 ||
438	    lgr->range.length != NFS4_MAX_UINT64) {
439		dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
440			__func__);
441		goto out;
442	}
443
444	if (fl->pattern_offset > lgr->range.offset) {
445		dprintk("%s pattern_offset %lld too large\n",
446				__func__, fl->pattern_offset);
447		goto out;
448	}
449
450	if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) {
451		dprintk("%s Invalid stripe unit (%u)\n",
452			__func__, fl->stripe_unit);
453		goto out;
454	}
455
456	/* find and reference the deviceid */
457	d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld,
458				   NFS_SERVER(lo->plh_inode)->nfs_client, id);
459	if (d == NULL) {
460		dsaddr = get_device_info(lo->plh_inode, id, gfp_flags);
 
461		if (dsaddr == NULL)
462			goto out;
463	} else
464		dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
465	/* Found deviceid is being reaped */
466	if (test_bit(NFS_DEVICEID_INVALID, &dsaddr->id_node.flags))
467			goto out_put;
468
469	fl->dsaddr = dsaddr;
470
471	if (fl->first_stripe_index < 0 ||
472	    fl->first_stripe_index >= dsaddr->stripe_count) {
473		dprintk("%s Bad first_stripe_index %d\n",
474				__func__, fl->first_stripe_index);
475		goto out_put;
476	}
477
478	if ((fl->stripe_type == STRIPE_SPARSE &&
479	    fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
480	    (fl->stripe_type == STRIPE_DENSE &&
481	    fl->num_fh != dsaddr->stripe_count)) {
482		dprintk("%s num_fh %u not valid for given packing\n",
483			__func__, fl->num_fh);
484		goto out_put;
485	}
486
487	if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
488		dprintk("%s Stripe unit (%u) not aligned with rsize %u "
489			"wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
490			nfss->wsize);
491	}
492
493	status = 0;
494out:
495	dprintk("--> %s returns %d\n", __func__, status);
496	return status;
497out_put:
498	nfs4_fl_put_deviceid(dsaddr);
499	goto out;
500}
501
502static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
503{
504	int i;
505
506	for (i = 0; i < fl->num_fh; i++) {
507		if (!fl->fh_array[i])
508			break;
509		kfree(fl->fh_array[i]);
510	}
511	kfree(fl->fh_array);
512	fl->fh_array = NULL;
513}
514
515static void
516_filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
517{
518	filelayout_free_fh_array(fl);
519	kfree(fl);
520}
521
522static int
523filelayout_decode_layout(struct pnfs_layout_hdr *flo,
524			 struct nfs4_filelayout_segment *fl,
525			 struct nfs4_layoutget_res *lgr,
526			 struct nfs4_deviceid *id,
527			 gfp_t gfp_flags)
528{
529	struct xdr_stream stream;
530	struct xdr_buf buf;
531	struct page *scratch;
532	__be32 *p;
533	uint32_t nfl_util;
534	int i;
535
536	dprintk("%s: set_layout_map Begin\n", __func__);
537
538	scratch = alloc_page(gfp_flags);
539	if (!scratch)
540		return -ENOMEM;
541
542	xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
543	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
544
545	/* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
546	 * num_fh (4) */
547	p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
548	if (unlikely(!p))
549		goto out_err;
550
551	memcpy(id, p, sizeof(*id));
552	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
553	nfs4_print_deviceid(id);
554
555	nfl_util = be32_to_cpup(p++);
556	if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
557		fl->commit_through_mds = 1;
558	if (nfl_util & NFL4_UFLG_DENSE)
559		fl->stripe_type = STRIPE_DENSE;
560	else
561		fl->stripe_type = STRIPE_SPARSE;
562	fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
563
564	fl->first_stripe_index = be32_to_cpup(p++);
565	p = xdr_decode_hyper(p, &fl->pattern_offset);
566	fl->num_fh = be32_to_cpup(p++);
567
568	dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
569		__func__, nfl_util, fl->num_fh, fl->first_stripe_index,
570		fl->pattern_offset);
571
572	/* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
573	 * Futher checking is done in filelayout_check_layout */
574	if (fl->num_fh < 0 || fl->num_fh >
575	    max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
576		goto out_err;
577
578	if (fl->num_fh > 0) {
579		fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
580				       gfp_flags);
581		if (!fl->fh_array)
582			goto out_err;
583	}
584
585	for (i = 0; i < fl->num_fh; i++) {
586		/* Do we want to use a mempool here? */
587		fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
588		if (!fl->fh_array[i])
589			goto out_err_free;
590
591		p = xdr_inline_decode(&stream, 4);
592		if (unlikely(!p))
593			goto out_err_free;
594		fl->fh_array[i]->size = be32_to_cpup(p++);
595		if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
596			printk(KERN_ERR "Too big fh %d received %d\n",
597			       i, fl->fh_array[i]->size);
598			goto out_err_free;
599		}
600
601		p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
602		if (unlikely(!p))
603			goto out_err_free;
604		memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
605		dprintk("DEBUG: %s: fh len %d\n", __func__,
606			fl->fh_array[i]->size);
607	}
608
609	__free_page(scratch);
610	return 0;
611
612out_err_free:
613	filelayout_free_fh_array(fl);
614out_err:
615	__free_page(scratch);
616	return -EIO;
617}
618
619static void
620filelayout_free_lseg(struct pnfs_layout_segment *lseg)
621{
622	struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
623
624	dprintk("--> %s\n", __func__);
625	nfs4_fl_put_deviceid(fl->dsaddr);
626	kfree(fl->commit_buckets);
 
 
 
 
 
 
 
 
627	_filelayout_free_lseg(fl);
628}
629
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
630static struct pnfs_layout_segment *
631filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
632		      struct nfs4_layoutget_res *lgr,
633		      gfp_t gfp_flags)
634{
635	struct nfs4_filelayout_segment *fl;
636	int rc;
637	struct nfs4_deviceid id;
638
639	dprintk("--> %s\n", __func__);
640	fl = kzalloc(sizeof(*fl), gfp_flags);
641	if (!fl)
642		return NULL;
643
644	rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
645	if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
646		_filelayout_free_lseg(fl);
647		return NULL;
648	}
649
650	/* This assumes there is only one IOMODE_RW lseg.  What
651	 * we really want to do is have a layout_hdr level
652	 * dictionary of <multipath_list4, fh> keys, each
653	 * associated with a struct list_head, populated by calls
654	 * to filelayout_write_pagelist().
655	 * */
656	if ((!fl->commit_through_mds) && (lgr->range.iomode == IOMODE_RW)) {
657		int i;
658		int size = (fl->stripe_type == STRIPE_SPARSE) ?
659			fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
660
661		fl->commit_buckets = kcalloc(size, sizeof(struct list_head), gfp_flags);
662		if (!fl->commit_buckets) {
663			filelayout_free_lseg(&fl->generic_hdr);
664			return NULL;
665		}
666		fl->number_of_buckets = size;
667		for (i = 0; i < size; i++)
668			INIT_LIST_HEAD(&fl->commit_buckets[i]);
669	}
670	return &fl->generic_hdr;
671}
672
673/*
674 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
675 *
676 * return true  : coalesce page
677 * return false : don't coalesce page
678 */
679static bool
680filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
681		   struct nfs_page *req)
682{
683	u64 p_stripe, r_stripe;
684	u32 stripe_unit;
685
686	if (!pnfs_generic_pg_test(pgio, prev, req) ||
687	    !nfs_generic_pg_test(pgio, prev, req))
688		return false;
689
690	p_stripe = (u64)prev->wb_index << PAGE_CACHE_SHIFT;
691	r_stripe = (u64)req->wb_index << PAGE_CACHE_SHIFT;
692	stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
693
694	do_div(p_stripe, stripe_unit);
695	do_div(r_stripe, stripe_unit);
696
697	return (p_stripe == r_stripe);
698}
699
700void
701filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
702			struct nfs_page *req)
703{
704	BUG_ON(pgio->pg_lseg != NULL);
705
 
 
 
 
 
 
 
 
 
 
706	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
707					   req->wb_context,
708					   0,
709					   NFS4_MAX_UINT64,
710					   IOMODE_READ,
711					   GFP_KERNEL);
712	/* If no lseg, fall back to read through mds */
713	if (pgio->pg_lseg == NULL)
714		nfs_pageio_reset_read_mds(pgio);
715}
716
717void
718filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
719			 struct nfs_page *req)
720{
721	BUG_ON(pgio->pg_lseg != NULL);
 
 
 
722
 
 
723	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
724					   req->wb_context,
725					   0,
726					   NFS4_MAX_UINT64,
727					   IOMODE_RW,
728					   GFP_NOFS);
729	/* If no lseg, fall back to write through mds */
730	if (pgio->pg_lseg == NULL)
731		nfs_pageio_reset_write_mds(pgio);
 
 
 
 
 
 
 
 
 
 
732}
733
734static const struct nfs_pageio_ops filelayout_pg_read_ops = {
735	.pg_init = filelayout_pg_init_read,
736	.pg_test = filelayout_pg_test,
737	.pg_doio = pnfs_generic_pg_readpages,
738};
739
740static const struct nfs_pageio_ops filelayout_pg_write_ops = {
741	.pg_init = filelayout_pg_init_write,
742	.pg_test = filelayout_pg_test,
743	.pg_doio = pnfs_generic_pg_writepages,
744};
745
746static bool filelayout_mark_pnfs_commit(struct pnfs_layout_segment *lseg)
747{
748	return !FILELAYOUT_LSEG(lseg)->commit_through_mds;
749}
750
751static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
752{
753	if (fl->stripe_type == STRIPE_SPARSE)
754		return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
755	else
756		return j;
757}
758
759struct list_head *filelayout_choose_commit_list(struct nfs_page *req)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
760{
761	struct pnfs_layout_segment *lseg = req->wb_commit_lseg;
762	struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
763	u32 i, j;
764	struct list_head *list;
 
 
 
 
765
766	/* Note that we are calling nfs4_fl_calc_j_index on each page
767	 * that ends up being committed to a data server.  An attractive
768	 * alternative is to add a field to nfs_write_data and nfs_page
769	 * to store the value calculated in filelayout_write_pagelist
770	 * and just use that here.
771	 */
772	j = nfs4_fl_calc_j_index(lseg,
773				 (loff_t)req->wb_index << PAGE_CACHE_SHIFT);
774	i = select_bucket_index(fl, j);
775	list = &fl->commit_buckets[i];
 
776	if (list_empty(list)) {
777		/* Non-empty buckets hold a reference on the lseg */
778		get_lseg(lseg);
 
 
 
 
 
779	}
 
 
780	return list;
781}
782
 
 
 
 
 
 
 
 
 
 
 
783static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
784{
785	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
786
787	if (flseg->stripe_type == STRIPE_SPARSE)
788		return i;
789	else
790		return nfs4_fl_calc_ds_index(lseg, i);
791}
792
793static struct nfs_fh *
794select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
795{
796	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
797
798	if (flseg->stripe_type == STRIPE_SPARSE) {
799		if (flseg->num_fh == 1)
800			i = 0;
801		else if (flseg->num_fh == 0)
802			/* Use the MDS OPEN fh set in nfs_read_rpcsetup */
803			return NULL;
804	}
805	return flseg->fh_array[i];
806}
807
808static int filelayout_initiate_commit(struct nfs_write_data *data, int how)
809{
810	struct pnfs_layout_segment *lseg = data->lseg;
811	struct nfs4_pnfs_ds *ds;
 
812	u32 idx;
813	struct nfs_fh *fh;
814
815	idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
816	ds = nfs4_fl_prepare_ds(lseg, idx);
817	if (!ds) {
818		printk(KERN_ERR "%s: prepare_ds failed, use MDS\n", __func__);
819		set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
820		set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
821		prepare_to_resend_writes(data);
822		data->mds_ops->rpc_release(data);
823		return -EAGAIN;
824	}
825	dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how);
826	data->write_done_cb = filelayout_commit_done_cb;
 
827	data->ds_clp = ds->ds_clp;
828	fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
829	if (fh)
830		data->args.fh = fh;
831	return nfs_initiate_commit(data, ds->ds_clp->cl_rpcclient,
832				   &filelayout_commit_call_ops, how);
 
 
 
 
 
833}
834
835/*
836 * This is only useful while we are using whole file layouts.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
837 */
838static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode)
 
839{
840	struct pnfs_layout_segment *lseg, *rv = NULL;
841
842	spin_lock(&inode->i_lock);
843	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
844		if (lseg->pls_range.iomode == IOMODE_RW)
845			rv = get_lseg(lseg);
846	spin_unlock(&inode->i_lock);
 
847	return rv;
848}
849
850static int alloc_ds_commits(struct inode *inode, struct list_head *list)
 
 
851{
852	struct pnfs_layout_segment *lseg;
853	struct nfs4_filelayout_segment *fl;
854	struct nfs_write_data *data;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
855	int i, j;
 
856
857	/* Won't need this when non-whole file layout segments are supported
858	 * instead we will use a pnfs_layout_hdr structure */
859	lseg = find_only_write_lseg(inode);
860	if (!lseg)
861		return 0;
862	fl = FILELAYOUT_LSEG(lseg);
863	for (i = 0; i < fl->number_of_buckets; i++) {
864		if (list_empty(&fl->commit_buckets[i]))
865			continue;
866		data = nfs_commitdata_alloc();
867		if (!data)
868			goto out_bad;
869		data->ds_commit_index = i;
870		data->lseg = lseg;
 
871		list_add(&data->pages, list);
 
872	}
873	put_lseg(lseg);
874	return 0;
875
876out_bad:
877	for (j = i; j < fl->number_of_buckets; j++) {
878		if (list_empty(&fl->commit_buckets[i]))
879			continue;
880		nfs_retry_commit(&fl->commit_buckets[i], lseg);
881		put_lseg(lseg);  /* associated with emptying bucket */
 
882	}
883	put_lseg(lseg);
884	/* Caller will clean up entries put on list */
885	return -ENOMEM;
886}
887
888/* This follows nfs_commit_list pretty closely */
889static int
890filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
891			   int how)
892{
893	struct nfs_write_data	*data, *tmp;
894	LIST_HEAD(list);
 
895
896	if (!list_empty(mds_pages)) {
897		data = nfs_commitdata_alloc();
898		if (!data)
899			goto out_bad;
900		data->lseg = NULL;
901		list_add(&data->pages, &list);
 
 
902	}
903
904	if (alloc_ds_commits(inode, &list))
905		goto out_bad;
 
 
 
 
 
 
906
907	list_for_each_entry_safe(data, tmp, &list, pages) {
908		list_del_init(&data->pages);
909		atomic_inc(&NFS_I(inode)->commits_outstanding);
910		if (!data->lseg) {
911			nfs_init_commit(data, mds_pages, NULL);
912			nfs_initiate_commit(data, NFS_CLIENT(inode),
913					    data->mds_ops, how);
914		} else {
915			nfs_init_commit(data, &FILELAYOUT_LSEG(data->lseg)->commit_buckets[data->ds_commit_index], data->lseg);
 
 
 
916			filelayout_initiate_commit(data, how);
917		}
918	}
919	return 0;
920 out_bad:
921	list_for_each_entry_safe(data, tmp, &list, pages) {
922		nfs_retry_commit(&data->pages, data->lseg);
923		list_del_init(&data->pages);
924		nfs_commit_free(data);
925	}
926	nfs_retry_commit(mds_pages, NULL);
927	nfs_commit_clear_lock(NFS_I(inode));
928	return -ENOMEM;
929}
930
931static void
932filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
933{
934	nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
935}
936
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
937static struct pnfs_layoutdriver_type filelayout_type = {
938	.id			= LAYOUT_NFSV4_1_FILES,
939	.name			= "LAYOUT_NFSV4_1_FILES",
940	.owner			= THIS_MODULE,
 
 
941	.alloc_lseg		= filelayout_alloc_lseg,
942	.free_lseg		= filelayout_free_lseg,
943	.pg_read_ops		= &filelayout_pg_read_ops,
944	.pg_write_ops		= &filelayout_pg_write_ops,
945	.mark_pnfs_commit	= filelayout_mark_pnfs_commit,
946	.choose_commit_list	= filelayout_choose_commit_list,
 
 
 
947	.commit_pagelist	= filelayout_commit_pagelist,
948	.read_pagelist		= filelayout_read_pagelist,
949	.write_pagelist		= filelayout_write_pagelist,
950	.free_deviceid_node	= filelayout_free_deveiceid_node,
951};
952
953static int __init nfs4filelayout_init(void)
954{
955	printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
956	       __func__);
957	return pnfs_register_layoutdriver(&filelayout_type);
958}
959
960static void __exit nfs4filelayout_exit(void)
961{
962	printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
963	       __func__);
964	pnfs_unregister_layoutdriver(&filelayout_type);
965}
966
967MODULE_ALIAS("nfs-layouttype4-1");
968
969module_init(nfs4filelayout_init);
970module_exit(nfs4filelayout_exit);