pnfs.c 46.6 KB
Newer Older
R
Ricardo Labiaga 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
/*
 *  pNFS functions to call and manage layout drivers.
 *
 *  Copyright (c) 2002 [year of first publication]
 *  The Regents of the University of Michigan
 *  All Rights Reserved
 *
 *  Dean Hildebrand <dhildebz@umich.edu>
 *
 *  Permission is granted to use, copy, create derivative works, and
 *  redistribute this software and such derivative works for any purpose,
 *  so long as the name of the University of Michigan is not used in
 *  any advertising or publicity pertaining to the use or distribution
 *  of this software without specific, written prior authorization. If
 *  the above copyright notice or any other identification of the
 *  University of Michigan is included in any copy of any portion of
 *  this software, then the disclaimer below must also be included.
 *
 *  This software is provided as is, without representation or warranty
 *  of any kind either express or implied, including without limitation
 *  the implied warranties of merchantability, fitness for a particular
 *  purpose, or noninfringement.  The Regents of the University of
 *  Michigan shall not be liable for any damages, including special,
 *  indirect, incidental, or consequential damages, with respect to any
 *  claim arising out of or in connection with the use of the software,
 *  even if it has been or is hereafter advised of the possibility of
 *  such damages.
 */

#include <linux/nfs_fs.h>
31
#include <linux/nfs_page.h>
32
#include <linux/module.h>
33
#include "internal.h"
R
Ricardo Labiaga 已提交
34
#include "pnfs.h"
A
Andy Adamson 已提交
35
#include "iostat.h"
R
Ricardo Labiaga 已提交
36 37

#define NFSDBG_FACILITY		NFSDBG_PNFS
38
#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
R
Ricardo Labiaga 已提交
39

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
/* Locking:
 *
 * pnfs_spinlock:
 *      protects pnfs_modules_tbl.
 */
static DEFINE_SPINLOCK(pnfs_spinlock);

/*
 * pnfs_modules_tbl holds all pnfs modules
 */
static LIST_HEAD(pnfs_modules_tbl);

/* Return the registered pnfs layout driver module matching given id */
static struct pnfs_layoutdriver_type *
find_pnfs_driver_locked(u32 id)
{
	struct pnfs_layoutdriver_type *local;

	list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
		if (local->id == id)
			goto out;
	local = NULL;
out:
	dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
	return local;
}

R
Ricardo Labiaga 已提交
67 68 69
static struct pnfs_layoutdriver_type *
find_pnfs_driver(u32 id)
{
70 71 72 73
	struct pnfs_layoutdriver_type *local;

	spin_lock(&pnfs_spinlock);
	local = find_pnfs_driver_locked(id);
74 75 76 77
	if (local != NULL && !try_module_get(local->owner)) {
		dprintk("%s: Could not grab reference on module\n", __func__);
		local = NULL;
	}
78 79
	spin_unlock(&pnfs_spinlock);
	return local;
R
Ricardo Labiaga 已提交
80 81 82 83 84
}

void
unset_pnfs_layoutdriver(struct nfs_server *nfss)
{
85 86 87
	if (nfss->pnfs_curr_ld) {
		if (nfss->pnfs_curr_ld->clear_layoutdriver)
			nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
88 89 90
		/* Decrement the MDS count. Purge the deviceid cache if zero */
		if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
			nfs4_deviceid_purge_client(nfss->nfs_client);
91
		module_put(nfss->pnfs_curr_ld->owner);
92
	}
R
Ricardo Labiaga 已提交
93 94 95 96 97 98 99 100 101 102
	nfss->pnfs_curr_ld = NULL;
}

/*
 * Try to set the server's pnfs module to the pnfs layout type specified by id.
 * Currently only one pNFS layout driver per filesystem is supported.
 *
 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
 */
void
103 104
set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
		      u32 id)
R
Ricardo Labiaga 已提交
105 106 107 108 109 110 111
{
	struct pnfs_layoutdriver_type *ld_type = NULL;

	if (id == 0)
		goto out_no_driver;
	if (!(server->nfs_client->cl_exchange_flags &
		 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
112 113
		printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
			__func__, id, server->nfs_client->cl_exchange_flags);
R
Ricardo Labiaga 已提交
114 115 116 117 118 119 120 121 122 123 124 125 126
		goto out_no_driver;
	}
	ld_type = find_pnfs_driver(id);
	if (!ld_type) {
		request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
		ld_type = find_pnfs_driver(id);
		if (!ld_type) {
			dprintk("%s: No pNFS module found for %u.\n",
				__func__, id);
			goto out_no_driver;
		}
	}
	server->pnfs_curr_ld = ld_type;
127 128
	if (ld_type->set_layoutdriver
	    && ld_type->set_layoutdriver(server, mntfh)) {
129 130
		printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
			"driver %u.\n", __func__, id);
131 132 133
		module_put(ld_type->owner);
		goto out_no_driver;
	}
134 135
	/* Bump the MDS count */
	atomic_inc(&server->nfs_client->cl_mds_count);
136

R
Ricardo Labiaga 已提交
137 138 139 140 141 142 143
	dprintk("%s: pNFS module for %u set\n", __func__, id);
	return;

out_no_driver:
	dprintk("%s: Using NFSv4 I/O\n", __func__);
	server->pnfs_curr_ld = NULL;
}
144 145 146 147 148 149 150 151

int
pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
{
	int status = -EINVAL;
	struct pnfs_layoutdriver_type *tmp;

	if (ld_type->id == 0) {
152
		printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
153 154
		return status;
	}
155
	if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
156
		printk(KERN_ERR "NFS: %s Layout driver must provide "
157 158 159
		       "alloc_lseg and free_lseg.\n", __func__);
		return status;
	}
160 161 162 163 164 165 166 167 168

	spin_lock(&pnfs_spinlock);
	tmp = find_pnfs_driver_locked(ld_type->id);
	if (!tmp) {
		list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
		status = 0;
		dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
			ld_type->name);
	} else {
169
		printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186
			__func__, ld_type->id);
	}
	spin_unlock(&pnfs_spinlock);

	return status;
}
EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);

void
pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
{
	dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
	spin_lock(&pnfs_spinlock);
	list_del(&ld_type->pnfs_tblid);
	spin_unlock(&pnfs_spinlock);
}
EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
187

188 189 190 191
/*
 * pNFS client layout cache
 */

192
/* Need to hold i_lock if caller does not already hold reference */
F
Fred Isaman 已提交
193
void
194
pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
195
{
196
	atomic_inc(&lo->plh_refcount);
197 198
}

199 200 201 202 203 204 205 206 207 208 209 210
static struct pnfs_layout_hdr *
pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
{
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
	return ld->alloc_layout_hdr ? ld->alloc_layout_hdr(ino, gfp_flags) :
		kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags);
}

static void
pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
{
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(lo->plh_inode)->pnfs_curr_ld;
211
	put_rpccred(lo->plh_lc_cred);
212 213 214
	return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo);
}

215
static void
216
destroy_layout_hdr(struct pnfs_layout_hdr *lo)
217
{
218 219 220
	dprintk("%s: freeing layout cache %p\n", __func__, lo);
	BUG_ON(!list_empty(&lo->plh_layouts));
	NFS_I(lo->plh_inode)->layout = NULL;
221
	pnfs_free_layout_hdr(lo);
222
}
223

224
static void
225
pnfs_put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
226 227 228
{
	if (atomic_dec_and_test(&lo->plh_refcount))
		destroy_layout_hdr(lo);
229 230
}

231
void
232
pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
233
{
234 235 236 237 238 239
	struct inode *inode = lo->plh_inode;

	if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
		destroy_layout_hdr(lo);
		spin_unlock(&inode->i_lock);
	}
240 241
}

242 243 244 245 246 247 248 249 250 251
static int
pnfs_iomode_to_fail_bit(u32 iomode)
{
	return iomode == IOMODE_RW ?
		NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
}

static void
pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
{
252
	lo->plh_retry_timestamp = jiffies;
253 254 255 256 257 258 259 260
	set_bit(pnfs_iomode_to_fail_bit(iomode), &lo->plh_flags);
	dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
			iomode == IOMODE_RW ?  "RW" : "READ");
}

static bool
pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
{
261 262 263 264 265 266 267 268 269 270 271 272
	unsigned long start, end;
	if (test_bit(pnfs_iomode_to_fail_bit(iomode), &lo->plh_flags) == 0)
		return false;
	end = jiffies;
	start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
	if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
		/* It is time to retry the failed layoutgets */
		clear_bit(NFS_LAYOUT_RW_FAILED, &lo->plh_flags);
		clear_bit(NFS_LAYOUT_RO_FAILED, &lo->plh_flags);
		return false;
	}
	return true;
273 274
}

275 276 277
static void
init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
{
278
	INIT_LIST_HEAD(&lseg->pls_list);
279
	INIT_LIST_HEAD(&lseg->pls_lc_list);
280 281 282
	atomic_set(&lseg->pls_refcount, 1);
	smp_mb();
	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
283
	lseg->pls_layout = lo;
284 285
}

286
static void free_lseg(struct pnfs_layout_segment *lseg)
287
{
288
	struct inode *ino = lseg->pls_layout->plh_inode;
289

290
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
291 292
	/* Matched by pnfs_get_layout_hdr in pnfs_insert_layout */
	pnfs_put_layout_hdr(NFS_I(ino)->layout);
293 294
}

F
Fred Isaman 已提交
295
static void
296
pnfs_put_lseg_common(struct pnfs_layout_segment *lseg)
F
Fred Isaman 已提交
297 298 299
{
	struct inode *inode = lseg->pls_layout->plh_inode;

300
	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
F
Fred Isaman 已提交
301 302 303 304
	list_del_init(&lseg->pls_list);
	if (list_empty(&lseg->pls_layout->plh_segs)) {
		set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
		/* Matched by initial refcount set in alloc_init_layout_hdr */
305
		pnfs_put_layout_hdr_locked(lseg->pls_layout);
F
Fred Isaman 已提交
306 307 308 309
	}
	rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
}

310
void
311
pnfs_put_lseg(struct pnfs_layout_segment *lseg)
312
{
F
Fred Isaman 已提交
313 314 315 316 317
	struct inode *inode;

	if (!lseg)
		return;

318 319 320
	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
		atomic_read(&lseg->pls_refcount),
		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
F
Fred Isaman 已提交
321 322 323
	inode = lseg->pls_layout->plh_inode;
	if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
		LIST_HEAD(free_me);
324

325
		pnfs_put_lseg_common(lseg);
F
Fred Isaman 已提交
326 327 328
		list_add(&lseg->pls_list, &free_me);
		spin_unlock(&inode->i_lock);
		pnfs_free_lseg_list(&free_me);
329 330
	}
}
331
EXPORT_SYMBOL_GPL(pnfs_put_lseg);
332

333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391
static inline u64
end_offset(u64 start, u64 len)
{
	u64 end;

	end = start + len;
	return end >= start ? end : NFS4_MAX_UINT64;
}

/* last octet in a range */
static inline u64
last_byte_offset(u64 start, u64 len)
{
	u64 end;

	BUG_ON(!len);
	end = start + len;
	return end > start ? end - 1 : NFS4_MAX_UINT64;
}

/*
 * is l2 fully contained in l1?
 *   start1                             end1
 *   [----------------------------------)
 *           start2           end2
 *           [----------------)
 */
static inline int
lo_seg_contained(struct pnfs_layout_range *l1,
		 struct pnfs_layout_range *l2)
{
	u64 start1 = l1->offset;
	u64 end1 = end_offset(start1, l1->length);
	u64 start2 = l2->offset;
	u64 end2 = end_offset(start2, l2->length);

	return (start1 <= start2) && (end1 >= end2);
}

/*
 * is l1 and l2 intersecting?
 *   start1                             end1
 *   [----------------------------------)
 *                              start2           end2
 *                              [----------------)
 */
static inline int
lo_seg_intersecting(struct pnfs_layout_range *l1,
		    struct pnfs_layout_range *l2)
{
	u64 start1 = l1->offset;
	u64 end1 = end_offset(start1, l1->length);
	u64 start2 = l2->offset;
	u64 end2 = end_offset(start2, l2->length);

	return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
	       (end2 == NFS4_MAX_UINT64 || end2 > start1);
}

392
static bool
393 394
should_free_lseg(struct pnfs_layout_range *lseg_range,
		 struct pnfs_layout_range *recall_range)
395
{
396 397 398
	return (recall_range->iomode == IOMODE_ANY ||
		lseg_range->iomode == recall_range->iomode) &&
	       lo_seg_intersecting(lseg_range, recall_range);
399 400
}

401 402 403 404 405 406 407 408 409 410 411
/* Returns 1 if lseg is removed from list, 0 otherwise */
static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
			     struct list_head *tmp_list)
{
	int rv = 0;

	if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
		/* Remove the reference keeping the lseg in the
		 * list.  It will now be removed when all
		 * outstanding io is finished.
		 */
F
Fred Isaman 已提交
412 413 414
		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
			atomic_read(&lseg->pls_refcount));
		if (atomic_dec_and_test(&lseg->pls_refcount)) {
415
			pnfs_put_lseg_common(lseg);
F
Fred Isaman 已提交
416 417 418
			list_add(&lseg->pls_list, tmp_list);
			rv = 1;
		}
419 420 421 422 423 424 425
	}
	return rv;
}

/* Returns count of number of matching invalid lsegs remaining in list
 * after call.
 */
F
Fred Isaman 已提交
426
int
427
pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
428
			    struct list_head *tmp_list,
429
			    struct pnfs_layout_range *recall_range)
430 431
{
	struct pnfs_layout_segment *lseg, *next;
432
	int invalid = 0, removed = 0;
433 434 435

	dprintk("%s:Begin lo %p\n", __func__, lo);

436
	if (list_empty(&lo->plh_segs)) {
437 438 439
		/* Reset MDS Threshold I/O counters */
		NFS_I(lo->plh_inode)->write_io = 0;
		NFS_I(lo->plh_inode)->read_io = 0;
440
		if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
441
			pnfs_put_layout_hdr_locked(lo);
442 443
		return 0;
	}
444
	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
445 446
		if (!recall_range ||
		    should_free_lseg(&lseg->pls_range, recall_range)) {
447 448 449 450 451 452 453 454 455
			dprintk("%s: freeing lseg %p iomode %d "
				"offset %llu length %llu\n", __func__,
				lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
				lseg->pls_range.length);
			invalid++;
			removed += mark_lseg_invalid(lseg, tmp_list);
		}
	dprintk("%s:Return %i\n", __func__, invalid - removed);
	return invalid - removed;
456 457
}

458
/* note free_me must contain lsegs from a single layout_hdr */
F
Fred Isaman 已提交
459
void
460
pnfs_free_lseg_list(struct list_head *free_me)
461
{
462
	struct pnfs_layout_segment *lseg, *tmp;
463 464 465 466 467 468 469
	struct pnfs_layout_hdr *lo;

	if (list_empty(free_me))
		return;

	lo = list_first_entry(free_me, struct pnfs_layout_segment,
			      pls_list)->pls_layout;
470

471 472 473 474 475 476 477 478
	if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
		struct nfs_client *clp;

		clp = NFS_SERVER(lo->plh_inode)->nfs_client;
		spin_lock(&clp->cl_lock);
		list_del_init(&lo->plh_layouts);
		spin_unlock(&clp->cl_lock);
	}
479
	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
480
		list_del(&lseg->pls_list);
481
		free_lseg(lseg);
482 483 484
	}
}

485 486 487 488
void
pnfs_destroy_layout(struct nfs_inode *nfsi)
{
	struct pnfs_layout_hdr *lo;
489
	LIST_HEAD(tmp_list);
490 491 492 493

	spin_lock(&nfsi->vfs_inode.i_lock);
	lo = nfsi->layout;
	if (lo) {
494
		lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
495
		pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
496 497
	}
	spin_unlock(&nfsi->vfs_inode.i_lock);
498 499
	pnfs_free_lseg_list(&tmp_list);
}
500
EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
501 502 503 504 505 506 507 508

/*
 * Called by the state manger to remove all layouts established under an
 * expired lease.
 */
void
pnfs_destroy_all_layouts(struct nfs_client *clp)
{
509
	struct nfs_server *server;
510 511 512
	struct pnfs_layout_hdr *lo;
	LIST_HEAD(tmp_list);

513 514 515
	nfs4_deviceid_mark_client_invalid(clp);
	nfs4_deviceid_purge_client(clp);

516
	spin_lock(&clp->cl_lock);
517 518 519 520 521 522
	rcu_read_lock();
	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
		if (!list_empty(&server->layouts))
			list_splice_init(&server->layouts, &tmp_list);
	}
	rcu_read_unlock();
523 524 525 526
	spin_unlock(&clp->cl_lock);

	while (!list_empty(&tmp_list)) {
		lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
527
				plh_layouts);
528
		dprintk("%s freeing layout for inode %lu\n", __func__,
529
			lo->plh_inode->i_ino);
530
		list_del_init(&lo->plh_layouts);
531
		pnfs_destroy_layout(NFS_I(lo->plh_inode));
532
	}
533 534
}

535
/* update lo->plh_stateid with new if is more recent */
F
Fred Isaman 已提交
536 537 538
void
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
			bool update_barrier)
539
{
540
	u32 oldseq, newseq;
541

542 543
	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
	newseq = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
544
	if ((int)(newseq - oldseq) > 0) {
545
		nfs4_stateid_copy(&lo->plh_stateid, new);
F
Fred Isaman 已提交
546
		if (update_barrier) {
547
			u32 new_barrier = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
548 549 550 551 552 553 554 555 556 557 558 559 560 561 562

			if ((int)(new_barrier - lo->plh_barrier))
				lo->plh_barrier = new_barrier;
		} else {
			/* Because of wraparound, we want to keep the barrier
			 * "close" to the current seqids.  It needs to be
			 * within 2**31 to count as "behind", so if it
			 * gets too near that limit, give us a litle leeway
			 * and bring it to within 2**30.
			 * NOTE - and yes, this is all unsigned arithmetic.
			 */
			if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
				lo->plh_barrier = newseq - (1 << 30);
		}
	}
563 564
}

565 566
/* lget is set to 1 if called from inside send_layoutget call chain */
static bool
F
Fred Isaman 已提交
567 568 569 570
pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
			int lget)
{
	if ((stateid) &&
571
	    (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0)
F
Fred Isaman 已提交
572
		return true;
F
Fred Isaman 已提交
573
	return lo->plh_block_lgets ||
574
		test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
F
Fred Isaman 已提交
575
		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
F
Fred Isaman 已提交
576
		(list_empty(&lo->plh_segs) &&
577 578 579
		 (atomic_read(&lo->plh_outstanding) > lget));
}

580 581 582
int
pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
			      struct nfs4_state *open_state)
583
{
584
	int status = 0;
585

586
	dprintk("--> %s\n", __func__);
587
	spin_lock(&lo->plh_inode->i_lock);
F
Fred Isaman 已提交
588
	if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
589 590
		status = -EAGAIN;
	} else if (list_empty(&lo->plh_segs)) {
591 592 593 594
		int seq;

		do {
			seq = read_seqbegin(&open_state->seqlock);
595
			nfs4_stateid_copy(dst, &open_state->stateid);
596 597
		} while (read_seqretry(&open_state->seqlock, seq));
	} else
598
		nfs4_stateid_copy(dst, &lo->plh_stateid);
599
	spin_unlock(&lo->plh_inode->i_lock);
600
	dprintk("<-- %s\n", __func__);
601
	return status;
602 603 604 605 606 607 608 609
}

/*
* Get layout from server.
*    for now, assume that whole file layouts are requested.
*    arg->offset: 0
*    arg->length: all ones
*/
610 611 612
static struct pnfs_layout_segment *
send_layoutget(struct pnfs_layout_hdr *lo,
	   struct nfs_open_context *ctx,
613
	   struct pnfs_layout_range *range,
614
	   gfp_t gfp_flags)
615
{
616
	struct inode *ino = lo->plh_inode;
617 618
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs4_layoutget *lgp;
619
	struct pnfs_layout_segment *lseg;
620 621

	dprintk("--> %s\n", __func__);
622

623
	BUG_ON(ctx == NULL);
624
	lgp = kzalloc(sizeof(*lgp), gfp_flags);
625
	if (lgp == NULL)
626
		return NULL;
627

628 629 630
	lgp->args.minlength = PAGE_CACHE_SIZE;
	if (lgp->args.minlength > range->length)
		lgp->args.minlength = range->length;
631
	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
632
	lgp->args.range = *range;
633 634 635
	lgp->args.type = server->pnfs_curr_ld->id;
	lgp->args.inode = ino;
	lgp->args.ctx = get_nfs_open_context(ctx);
636
	lgp->gfp_flags = gfp_flags;
637 638 639 640

	/* Synchronously retrieve layout information from server and
	 * store in lseg.
	 */
641 642 643 644 645 646 647 648
	lseg = nfs4_proc_layoutget(lgp, gfp_flags);
	if (IS_ERR(lseg)) {
		switch (PTR_ERR(lseg)) {
		case -ENOMEM:
		case -ERESTARTSYS:
			break;
		default:
			/* remember that LAYOUTGET failed and suspend trying */
649
			pnfs_layout_io_set_failed(lo, range->iomode);
650 651
		}
		return NULL;
652
	}
653

654 655 656
	return lseg;
}

657 658 659 660 661 662 663 664
/*
 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
 * when the layout segment list is empty.
 *
 * Note that a pnfs_layout_hdr can exist with an empty layout segment
 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
 * deviceid is marked invalid.
 */
B
Benny Halevy 已提交
665 666 667 668 669 670 671 672
int
_pnfs_return_layout(struct inode *ino)
{
	struct pnfs_layout_hdr *lo = NULL;
	struct nfs_inode *nfsi = NFS_I(ino);
	LIST_HEAD(tmp_list);
	struct nfs4_layoutreturn *lrp;
	nfs4_stateid stateid;
673
	int status = 0, empty;
B
Benny Halevy 已提交
674

675
	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
B
Benny Halevy 已提交
676 677 678

	spin_lock(&ino->i_lock);
	lo = nfsi->layout;
679
	if (!lo || pnfs_test_layout_returned(lo)) {
B
Benny Halevy 已提交
680
		spin_unlock(&ino->i_lock);
681 682
		dprintk("NFS: %s no layout to return\n", __func__);
		goto out;
B
Benny Halevy 已提交
683 684 685
	}
	stateid = nfsi->layout->plh_stateid;
	/* Reference matched in nfs4_layoutreturn_release */
686
	pnfs_get_layout_hdr(lo);
687
	empty = list_empty(&lo->plh_segs);
688
	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
689 690 691
	/* Don't send a LAYOUTRETURN if list was initially empty */
	if (empty) {
		spin_unlock(&ino->i_lock);
692
		pnfs_put_layout_hdr(lo);
693 694 695
		dprintk("NFS: %s no layout segments to return\n", __func__);
		goto out;
	}
696
	lo->plh_block_lgets++;
697
	pnfs_mark_layout_returned(lo);
B
Benny Halevy 已提交
698 699 700 701 702 703 704 705
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);

	WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags));

	lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
	if (unlikely(lrp == NULL)) {
		status = -ENOMEM;
706 707
		pnfs_layout_io_set_failed(lo, IOMODE_RW);
		pnfs_layout_io_set_failed(lo, IOMODE_READ);
708
		pnfs_clear_layout_returned(lo);
709
		pnfs_put_layout_hdr(lo);
B
Benny Halevy 已提交
710 711 712 713 714 715
		goto out;
	}

	lrp->args.stateid = stateid;
	lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
	lrp->args.inode = ino;
T
Trond Myklebust 已提交
716
	lrp->args.layout = lo;
B
Benny Halevy 已提交
717 718 719 720 721 722 723
	lrp->clp = NFS_SERVER(ino)->nfs_client;

	status = nfs4_proc_layoutreturn(lrp);
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}
724
EXPORT_SYMBOL_GPL(_pnfs_return_layout);
B
Benny Halevy 已提交
725

F
Fred Isaman 已提交
726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745
bool pnfs_roc(struct inode *ino)
{
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg, *tmp;
	LIST_HEAD(tmp_list);
	bool found = false;

	spin_lock(&ino->i_lock);
	lo = NFS_I(ino)->layout;
	if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
	    test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
		goto out_nolayout;
	list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
		if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
			mark_lseg_invalid(lseg, &tmp_list);
			found = true;
		}
	if (!found)
		goto out_nolayout;
	lo->plh_block_lgets++;
746
	pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
F
Fred Isaman 已提交
747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);
	return true;

out_nolayout:
	spin_unlock(&ino->i_lock);
	return false;
}

void pnfs_roc_release(struct inode *ino)
{
	struct pnfs_layout_hdr *lo;

	spin_lock(&ino->i_lock);
	lo = NFS_I(ino)->layout;
	lo->plh_block_lgets--;
763
	pnfs_put_layout_hdr_locked(lo);
F
Fred Isaman 已提交
764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791
	spin_unlock(&ino->i_lock);
}

void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
{
	struct pnfs_layout_hdr *lo;

	spin_lock(&ino->i_lock);
	lo = NFS_I(ino)->layout;
	if ((int)(barrier - lo->plh_barrier) > 0)
		lo->plh_barrier = barrier;
	spin_unlock(&ino->i_lock);
}

bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
{
	struct nfs_inode *nfsi = NFS_I(ino);
	struct pnfs_layout_segment *lseg;
	bool found = false;

	spin_lock(&ino->i_lock);
	list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
		if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
			found = true;
			break;
		}
	if (!found) {
		struct pnfs_layout_hdr *lo = nfsi->layout;
792
		u32 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
F
Fred Isaman 已提交
793 794 795 796 797 798 799 800 801 802

		/* Since close does not return a layout stateid for use as
		 * a barrier, we choose the worst-case barrier.
		 */
		*barrier = current_seqid + atomic_read(&lo->plh_outstanding);
	}
	spin_unlock(&ino->i_lock);
	return found;
}

803 804 805 806 807 808
/*
 * Compare two layout segments for sorting into layout cache.
 * We want to preferentially return RW over RO layouts, so ensure those
 * are seen first.
 */
static s64
809 810
cmp_layout(struct pnfs_layout_range *l1,
	   struct pnfs_layout_range *l2)
811
{
812 813 814 815 816 817 818 819 820 821 822 823
	s64 d;

	/* high offset > low offset */
	d = l1->offset - l2->offset;
	if (d)
		return d;

	/* short length > long length */
	d = l2->length - l1->length;
	if (d)
		return d;

824
	/* read > read/write */
825
	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
826 827
}

828 829 830 831
static void
pnfs_insert_layout(struct pnfs_layout_hdr *lo,
		   struct pnfs_layout_segment *lseg)
{
832 833
	struct pnfs_layout_segment *lp;

834 835
	dprintk("%s:Begin\n", __func__);

836 837
	assert_spin_locked(&lo->plh_inode->i_lock);
	list_for_each_entry(lp, &lo->plh_segs, pls_list) {
838
		if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
839
			continue;
840
		list_add_tail(&lseg->pls_list, &lp->pls_list);
841 842 843
		dprintk("%s: inserted lseg %p "
			"iomode %d offset %llu length %llu before "
			"lp %p iomode %d offset %llu length %llu\n",
844 845 846 847
			__func__, lseg, lseg->pls_range.iomode,
			lseg->pls_range.offset, lseg->pls_range.length,
			lp, lp->pls_range.iomode, lp->pls_range.offset,
			lp->pls_range.length);
848
		goto out;
849
	}
850 851 852 853 854 855
	list_add_tail(&lseg->pls_list, &lo->plh_segs);
	dprintk("%s: inserted lseg %p "
		"iomode %d offset %llu length %llu at tail\n",
		__func__, lseg, lseg->pls_range.iomode,
		lseg->pls_range.offset, lseg->pls_range.length);
out:
856
	pnfs_get_layout_hdr(lo);
857 858

	dprintk("%s:Return\n", __func__);
859 860 861
}

static struct pnfs_layout_hdr *
862 863 864
alloc_init_layout_hdr(struct inode *ino,
		      struct nfs_open_context *ctx,
		      gfp_t gfp_flags)
865 866 867
{
	struct pnfs_layout_hdr *lo;

868
	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
869 870
	if (!lo)
		return NULL;
871
	atomic_set(&lo->plh_refcount, 1);
872 873
	INIT_LIST_HEAD(&lo->plh_layouts);
	INIT_LIST_HEAD(&lo->plh_segs);
F
Fred Isaman 已提交
874
	INIT_LIST_HEAD(&lo->plh_bulk_recall);
875
	lo->plh_inode = ino;
876
	lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
877 878 879 880
	return lo;
}

static struct pnfs_layout_hdr *
881 882 883
pnfs_find_alloc_layout(struct inode *ino,
		       struct nfs_open_context *ctx,
		       gfp_t gfp_flags)
884 885 886 887 888 889 890
{
	struct nfs_inode *nfsi = NFS_I(ino);
	struct pnfs_layout_hdr *new = NULL;

	dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);

	assert_spin_locked(&ino->i_lock);
891 892 893 894 895 896
	if (nfsi->layout) {
		if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
			return NULL;
		else
			return nfsi->layout;
	}
897
	spin_unlock(&ino->i_lock);
898
	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
899 900 901 902 903
	spin_lock(&ino->i_lock);

	if (likely(nfsi->layout == NULL))	/* Won the race? */
		nfsi->layout = new;
	else
904
		pnfs_free_layout_hdr(new);
905 906 907
	return nfsi->layout;
}

908 909 910 911 912 913 914 915 916 917 918 919
/*
 * iomode matching rules:
 * iomode	lseg	match
 * -----	-----	-----
 * ANY		READ	true
 * ANY		RW	true
 * RW		READ	false
 * RW		RW	true
 * READ		READ	true
 * READ		RW	true
 */
static int
920 921
is_matching_lseg(struct pnfs_layout_range *ls_range,
		 struct pnfs_layout_range *range)
922
{
923 924 925 926 927 928 929 930 931 932 933
	struct pnfs_layout_range range1;

	if ((range->iomode == IOMODE_RW &&
	     ls_range->iomode != IOMODE_RW) ||
	    !lo_seg_intersecting(ls_range, range))
		return 0;

	/* range1 covers only the first byte in the range */
	range1 = *range;
	range1.length = 1;
	return lo_seg_contained(ls_range, &range1);
934 935 936 937 938
}

/*
 * lookup range in layout
 */
939
static struct pnfs_layout_segment *
940 941
pnfs_find_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_range *range)
942
{
943 944 945 946
	struct pnfs_layout_segment *lseg, *ret = NULL;

	dprintk("%s:Begin\n", __func__);

947 948
	assert_spin_locked(&lo->plh_inode->i_lock);
	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
949
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
950
		    is_matching_lseg(&lseg->pls_range, range)) {
951
			ret = pnfs_get_lseg(lseg);
952 953
			break;
		}
954
		if (lseg->pls_range.offset > range->offset)
955 956 957 958
			break;
	}

	dprintk("%s:Return lseg %p ref %d\n",
959
		__func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
960
	return ret;
961 962
}

963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
/*
 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
 * to the MDS or over pNFS
 *
 * The nfs_inode read_io and write_io fields are cumulative counters reset
 * when there are no layout segments. Note that in pnfs_update_layout iomode
 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
 * WRITE request.
 *
 * A return of true means use MDS I/O.
 *
 * From rfc 5661:
 * If a file's size is smaller than the file size threshold, data accesses
 * SHOULD be sent to the metadata server.  If an I/O request has a length that
 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
 * server.  If both file size and I/O size are provided, the client SHOULD
 * reach or exceed  both thresholds before sending its read or write
 * requests to the data server.
 */
static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
				     struct inode *ino, int iomode)
{
	struct nfs4_threshold *t = ctx->mdsthreshold;
	struct nfs_inode *nfsi = NFS_I(ino);
	loff_t fsize = i_size_read(ino);
	bool size = false, size_set = false, io = false, io_set = false, ret = false;

	if (t == NULL)
		return ret;

	dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
		__func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);

	switch (iomode) {
	case IOMODE_READ:
		if (t->bm & THRESHOLD_RD) {
			dprintk("%s fsize %llu\n", __func__, fsize);
			size_set = true;
			if (fsize < t->rd_sz)
				size = true;
		}
		if (t->bm & THRESHOLD_RD_IO) {
			dprintk("%s nfsi->read_io %llu\n", __func__,
				nfsi->read_io);
			io_set = true;
			if (nfsi->read_io < t->rd_io_sz)
				io = true;
		}
		break;
	case IOMODE_RW:
		if (t->bm & THRESHOLD_WR) {
			dprintk("%s fsize %llu\n", __func__, fsize);
			size_set = true;
			if (fsize < t->wr_sz)
				size = true;
		}
		if (t->bm & THRESHOLD_WR_IO) {
			dprintk("%s nfsi->write_io %llu\n", __func__,
				nfsi->write_io);
			io_set = true;
			if (nfsi->write_io < t->wr_io_sz)
				io = true;
		}
		break;
	}
	if (size_set && io_set) {
		if (size && io)
			ret = true;
	} else if (size || io)
		ret = true;

	dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
	return ret;
}

1038 1039 1040 1041
/*
 * Layout segment is retreived from the server if not cached.
 * The appropriate layout segment is referenced and returned to the caller.
 */
1042
struct pnfs_layout_segment *
1043 1044
pnfs_update_layout(struct inode *ino,
		   struct nfs_open_context *ctx,
1045 1046
		   loff_t pos,
		   u64 count,
1047 1048
		   enum pnfs_iomode iomode,
		   gfp_t gfp_flags)
1049
{
1050 1051 1052 1053 1054
	struct pnfs_layout_range arg = {
		.iomode = iomode,
		.offset = pos,
		.length = count,
	};
1055
	unsigned pg_offset;
1056 1057
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs_client *clp = server->nfs_client;
1058 1059
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg = NULL;
1060
	bool first = false;
1061 1062

	if (!pnfs_enabled_sb(NFS_SERVER(ino)))
1063
		goto out;
1064 1065

	if (pnfs_within_mdsthreshold(ctx, ino, iomode))
1066
		goto out;
1067

1068
	spin_lock(&ino->i_lock);
1069
	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1070 1071 1072 1073
	if (lo == NULL) {
		spin_unlock(&ino->i_lock);
		goto out;
	}
1074

F
Fred Isaman 已提交
1075
	/* Do we even need to bother with this? */
1076
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1077
		dprintk("%s matches recall, use MDS\n", __func__);
1078 1079 1080 1081
		goto out_unlock;
	}

	/* if LAYOUTGET already failed once we don't try again */
1082
	if (pnfs_layout_io_test_failed(lo, iomode))
1083 1084
		goto out_unlock;

1085
	/* Check to see if the layout for the given range already exists */
1086
	lseg = pnfs_find_lseg(lo, &arg);
1087 1088 1089
	if (lseg)
		goto out_unlock;

F
Fred Isaman 已提交
1090
	if (pnfs_layoutgets_blocked(lo, NULL, 0))
1091 1092 1093
		goto out_unlock;
	atomic_inc(&lo->plh_outstanding);

1094
	pnfs_get_layout_hdr(lo);
1095 1096
	if (list_empty(&lo->plh_segs))
		first = true;
1097 1098 1099 1100

	/* Enable LAYOUTRETURNs */
	pnfs_clear_layout_returned(lo);

1101 1102
	spin_unlock(&ino->i_lock);
	if (first) {
1103 1104 1105 1106 1107
		/* The lo must be on the clp list if there is any
		 * chance of a CB_LAYOUTRECALL(FILE) coming in.
		 */
		spin_lock(&clp->cl_lock);
		BUG_ON(!list_empty(&lo->plh_layouts));
1108
		list_add_tail(&lo->plh_layouts, &server->layouts);
1109 1110
		spin_unlock(&clp->cl_lock);
	}
1111

1112 1113 1114 1115 1116
	pg_offset = arg.offset & ~PAGE_CACHE_MASK;
	if (pg_offset) {
		arg.offset -= pg_offset;
		arg.length += pg_offset;
	}
1117 1118
	if (arg.length != NFS4_MAX_UINT64)
		arg.length = PAGE_CACHE_ALIGN(arg.length);
1119

1120
	lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1121 1122 1123 1124
	if (!lseg && first) {
		spin_lock(&clp->cl_lock);
		list_del_init(&lo->plh_layouts);
		spin_unlock(&clp->cl_lock);
1125
	}
1126
	atomic_dec(&lo->plh_outstanding);
1127
out_put_layout_hdr:
1128
	pnfs_put_layout_hdr(lo);
1129
out:
1130 1131 1132 1133 1134 1135 1136 1137
	dprintk("%s: inode %s/%llu pNFS layout segment %s for "
			"(%s, offset: %llu, length: %llu)\n",
			__func__, ino->i_sb->s_id,
			(unsigned long long)NFS_FILEID(ino),
			lseg == NULL ? "not found" : "found",
			iomode==IOMODE_RW ?  "read/write" : "read-only",
			(unsigned long long)pos,
			(unsigned long long)count);
1138 1139 1140
	return lseg;
out_unlock:
	spin_unlock(&ino->i_lock);
1141
	goto out_put_layout_hdr;
1142
}
1143
EXPORT_SYMBOL_GPL(pnfs_update_layout);
1144

1145
struct pnfs_layout_segment *
1146 1147 1148 1149 1150
pnfs_layout_process(struct nfs4_layoutget *lgp)
{
	struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
	struct nfs4_layoutget_res *res = &lgp->res;
	struct pnfs_layout_segment *lseg;
1151
	struct inode *ino = lo->plh_inode;
1152 1153 1154
	int status = 0;

	/* Inject layout blob into I/O device driver */
1155
	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
	if (!lseg || IS_ERR(lseg)) {
		if (!lseg)
			status = -ENOMEM;
		else
			status = PTR_ERR(lseg);
		dprintk("%s: Could not allocate layout: error %d\n",
		       __func__, status);
		goto out;
	}

	spin_lock(&ino->i_lock);
1167
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1168 1169 1170 1171 1172 1173 1174 1175
		dprintk("%s forget reply due to recall\n", __func__);
		goto out_forget_reply;
	}

	if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
		dprintk("%s forget reply due to state\n", __func__);
		goto out_forget_reply;
	}
1176
	init_lseg(lo, lseg);
1177
	lseg->pls_range = res->range;
1178
	pnfs_get_lseg(lseg);
1179 1180
	pnfs_insert_layout(lo, lseg);

F
Fred Isaman 已提交
1181 1182 1183 1184 1185
	if (res->return_on_close) {
		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
		set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
	}

1186
	/* Done processing layoutget. Set the layout stateid */
F
Fred Isaman 已提交
1187
	pnfs_set_layout_stateid(lo, &res->stateid, false);
1188
	spin_unlock(&ino->i_lock);
1189
	return lseg;
1190
out:
1191
	return ERR_PTR(status);
F
Fred Isaman 已提交
1192 1193 1194 1195 1196 1197

out_forget_reply:
	spin_unlock(&ino->i_lock);
	lseg->pls_layout = lo;
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
	goto out;
1198 1199
}

1200 1201 1202 1203 1204
void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
	BUG_ON(pgio->pg_lseg != NULL);

1205 1206 1207 1208
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_read_mds(pgio);
		return;
	}
1209 1210 1211 1212 1213 1214
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
					   req->wb_bytes,
					   IOMODE_READ,
					   GFP_KERNEL);
1215 1216
	/* If no lseg, fall back to read through mds */
	if (pgio->pg_lseg == NULL)
1217
		nfs_pageio_reset_read_mds(pgio);
1218

1219 1220 1221 1222 1223 1224 1225 1226
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);

void
pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
	BUG_ON(pgio->pg_lseg != NULL);

1227 1228 1229 1230
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_write_mds(pgio);
		return;
	}
1231 1232 1233 1234 1235 1236
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
					   req->wb_bytes,
					   IOMODE_RW,
					   GFP_NOFS);
1237 1238
	/* If no lseg, fall back to write through mds */
	if (pgio->pg_lseg == NULL)
1239
		nfs_pageio_reset_write_mds(pgio);
1240 1241 1242
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);

1243
void
1244 1245
pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode,
		      const struct nfs_pgio_completion_ops *compl_ops)
1246 1247 1248 1249 1250
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (ld == NULL)
1251 1252 1253
		nfs_pageio_init_read(pgio, inode, compl_ops);
	else
		nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0);
1254 1255
}

1256
void
1257 1258 1259
pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode,
		       int ioflags,
		       const struct nfs_pgio_completion_ops *compl_ops)
1260 1261 1262 1263 1264
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (ld == NULL)
1265 1266 1267
		nfs_pageio_init_write(pgio, inode, ioflags, compl_ops);
	else
		nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags);
1268 1269
}

1270
bool
1271 1272
pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
		     struct nfs_page *req)
1273
{
1274 1275
	if (pgio->pg_lseg == NULL)
		return nfs_generic_pg_test(pgio, prev, req);
1276

1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292
	/*
	 * Test if a nfs_page is fully contained in the pnfs_layout_range.
	 * Note that this test makes several assumptions:
	 * - that the previous nfs_page in the struct nfs_pageio_descriptor
	 *   is known to lie within the range.
	 *   - that the nfs_page being tested is known to be contiguous with the
	 *   previous nfs_page.
	 *   - Layout ranges are page aligned, so we only have to test the
	 *   start offset of the request.
	 *
	 * Please also note that 'end_offset' is actually the offset of the
	 * first byte that lies outside the pnfs_layout_range. FIXME?
	 *
	 */
	return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
					 pgio->pg_lseg->pls_range.length);
1293
}
1294
EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1295

1296
int pnfs_write_done_resend_to_mds(struct inode *inode,
1297 1298
				struct list_head *head,
				const struct nfs_pgio_completion_ops *compl_ops)
1299 1300 1301 1302 1303
{
	struct nfs_pageio_descriptor pgio;
	LIST_HEAD(failed);

	/* Resend all requests through the MDS */
1304
	nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops);
1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);

		nfs_list_remove_request(req);
		if (!nfs_pageio_add_request(&pgio, req))
			nfs_list_add_request(req, &failed);
	}
	nfs_pageio_complete(&pgio);

	if (!list_empty(&failed)) {
		/* For some reason our attempt to resend pages. Mark the
		 * overall send request as having failed, and let
		 * nfs_writeback_release_full deal with the error.
		 */
		list_move(&failed, head);
		return -EIO;
	}
	return 0;
}
1324
EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1325

1326 1327
static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
{
1328 1329 1330 1331
	struct nfs_pgio_header *hdr = data->header;

	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1332
	    PNFS_LAYOUTRET_ON_ERROR) {
1333 1334
		clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
		pnfs_return_layout(hdr->inode);
1335
	}
1336 1337
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
1338 1339
							&hdr->pages,
							hdr->completion_ops);
1340 1341
}

1342 1343 1344
/*
 * Called by non rpc-based layout drivers
 */
1345
void pnfs_ld_write_done(struct nfs_write_data *data)
1346
{
1347 1348 1349
	struct nfs_pgio_header *hdr = data->header;

	if (!hdr->pnfs_error) {
1350
		pnfs_set_layoutcommit(data);
1351
		hdr->mds_ops->rpc_call_done(&data->task, data);
1352 1353
	} else
		pnfs_ld_handle_write_error(data);
1354
	hdr->mds_ops->rpc_release(data);
1355
}
1356
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1357

1358 1359 1360 1361
static void
pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
		struct nfs_write_data *data)
{
1362 1363
	struct nfs_pgio_header *hdr = data->header;

1364 1365 1366 1367 1368
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
		list_splice_tail_init(&hdr->pages, &desc->pg_list);
		nfs_pageio_reset_write_mds(desc);
		desc->pg_recoalesce = 1;
	}
1369 1370 1371 1372
	nfs_writedata_release(data);
}

static enum pnfs_try_status
1373
pnfs_try_to_write_data(struct nfs_write_data *wdata,
1374 1375 1376
			const struct rpc_call_ops *call_ops,
			struct pnfs_layout_segment *lseg,
			int how)
1377
{
1378 1379
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
1380 1381 1382
	enum pnfs_try_status trypnfs;
	struct nfs_server *nfss = NFS_SERVER(inode);

1383
	hdr->mds_ops = call_ops;
1384 1385 1386 1387

	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
		inode->i_ino, wdata->args.count, wdata->args.offset, how);
	trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
1388
	if (trypnfs != PNFS_NOT_ATTEMPTED)
1389 1390 1391 1392 1393
		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}

1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404
static void
pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
{
	struct nfs_write_data *data;
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;

	desc->pg_lseg = NULL;
	while (!list_empty(head)) {
		enum pnfs_try_status trypnfs;

1405
		data = list_first_entry(head, struct nfs_write_data, list);
1406 1407 1408 1409 1410 1411
		list_del_init(&data->list);

		trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
		if (trypnfs == PNFS_NOT_ATTEMPTED)
			pnfs_write_through_mds(desc, data);
	}
1412
	pnfs_put_lseg(lseg);
1413 1414
}

1415 1416
static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
{
1417
	pnfs_put_lseg(hdr->lseg);
1418 1419
	nfs_writehdr_free(hdr);
}
B
Bryan Schumaker 已提交
1420
EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
1421

1422 1423 1424
int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
{
1425 1426
	struct nfs_write_header *whdr;
	struct nfs_pgio_header *hdr;
1427 1428
	int ret;

1429 1430
	whdr = nfs_writehdr_alloc();
	if (!whdr) {
1431
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1432
		pnfs_put_lseg(desc->pg_lseg);
1433
		desc->pg_lseg = NULL;
1434
		return -ENOMEM;
1435
	}
1436 1437
	hdr = &whdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1438
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1439 1440 1441
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_flush(desc, hdr);
	if (ret != 0) {
1442
		pnfs_put_lseg(desc->pg_lseg);
1443 1444 1445 1446
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
	if (atomic_dec_and_test(&hdr->refcnt))
1447
		hdr->completion_ops->completion(hdr);
1448
	return ret;
1449 1450 1451
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);

1452
int pnfs_read_done_resend_to_mds(struct inode *inode,
1453 1454
				struct list_head *head,
				const struct nfs_pgio_completion_ops *compl_ops)
1455 1456
{
	struct nfs_pageio_descriptor pgio;
1457
	LIST_HEAD(failed);
1458

1459
	/* Resend all requests through the MDS */
1460
	nfs_pageio_init_read(&pgio, inode, compl_ops);
1461 1462
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);
1463 1464

		nfs_list_remove_request(req);
1465 1466
		if (!nfs_pageio_add_request(&pgio, req))
			nfs_list_add_request(req, &failed);
1467 1468
	}
	nfs_pageio_complete(&pgio);
1469 1470 1471 1472 1473 1474 1475

	if (!list_empty(&failed)) {
		list_move(&failed, head);
		return -EIO;
	}
	return 0;
}
1476
EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1477 1478 1479

static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
{
1480 1481 1482 1483
	struct nfs_pgio_header *hdr = data->header;

	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1484
	    PNFS_LAYOUTRET_ON_ERROR) {
1485 1486
		clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
		pnfs_return_layout(hdr->inode);
1487
	}
1488 1489
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
1490 1491
							&hdr->pages,
							hdr->completion_ops);
1492 1493
}

1494 1495 1496
/*
 * Called by non rpc-based layout drivers
 */
1497
void pnfs_ld_read_done(struct nfs_read_data *data)
1498
{
1499 1500 1501
	struct nfs_pgio_header *hdr = data->header;

	if (likely(!hdr->pnfs_error)) {
1502
		__nfs4_read_done_cb(data);
1503
		hdr->mds_ops->rpc_call_done(&data->task, data);
1504 1505
	} else
		pnfs_ld_handle_read_error(data);
1506
	hdr->mds_ops->rpc_release(data);
1507 1508 1509
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

1510 1511 1512 1513
static void
pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
		struct nfs_read_data *data)
{
1514 1515
	struct nfs_pgio_header *hdr = data->header;

1516 1517 1518 1519 1520
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
		list_splice_tail_init(&hdr->pages, &desc->pg_list);
		nfs_pageio_reset_read_mds(desc);
		desc->pg_recoalesce = 1;
	}
1521 1522 1523
	nfs_readdata_release(data);
}

A
Andy Adamson 已提交
1524 1525 1526
/*
 * Call the appropriate parallel I/O subsystem read function.
 */
1527
static enum pnfs_try_status
A
Andy Adamson 已提交
1528
pnfs_try_to_read_data(struct nfs_read_data *rdata,
1529 1530
		       const struct rpc_call_ops *call_ops,
		       struct pnfs_layout_segment *lseg)
A
Andy Adamson 已提交
1531
{
1532 1533
	struct nfs_pgio_header *hdr = rdata->header;
	struct inode *inode = hdr->inode;
A
Andy Adamson 已提交
1534 1535 1536
	struct nfs_server *nfss = NFS_SERVER(inode);
	enum pnfs_try_status trypnfs;

1537
	hdr->mds_ops = call_ops;
A
Andy Adamson 已提交
1538 1539 1540 1541 1542

	dprintk("%s: Reading ino:%lu %u@%llu\n",
		__func__, inode->i_ino, rdata->args.count, rdata->args.offset);

	trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
1543
	if (trypnfs != PNFS_NOT_ATTEMPTED)
A
Andy Adamson 已提交
1544 1545 1546 1547
		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}
A
Andy Adamson 已提交
1548

1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559
static void
pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
{
	struct nfs_read_data *data;
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;

	desc->pg_lseg = NULL;
	while (!list_empty(head)) {
		enum pnfs_try_status trypnfs;

1560
		data = list_first_entry(head, struct nfs_read_data, list);
1561 1562 1563 1564 1565 1566
		list_del_init(&data->list);

		trypnfs = pnfs_try_to_read_data(data, call_ops, lseg);
		if (trypnfs == PNFS_NOT_ATTEMPTED)
			pnfs_read_through_mds(desc, data);
	}
1567
	pnfs_put_lseg(lseg);
1568 1569
}

1570 1571
static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
{
1572
	pnfs_put_lseg(hdr->lseg);
1573 1574
	nfs_readhdr_free(hdr);
}
B
Bryan Schumaker 已提交
1575
EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
1576

1577 1578 1579
int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
{
1580 1581
	struct nfs_read_header *rhdr;
	struct nfs_pgio_header *hdr;
1582 1583
	int ret;

1584 1585
	rhdr = nfs_readhdr_alloc();
	if (!rhdr) {
1586
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1587
		ret = -ENOMEM;
1588
		pnfs_put_lseg(desc->pg_lseg);
1589 1590 1591
		desc->pg_lseg = NULL;
		return ret;
	}
1592 1593
	hdr = &rhdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1594
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1595 1596 1597
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_pagein(desc, hdr);
	if (ret != 0) {
1598
		pnfs_put_lseg(desc->pg_lseg);
1599 1600 1601 1602
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_reads(desc, &hdr->rpc_list);
	if (atomic_dec_and_test(&hdr->refcnt))
1603
		hdr->completion_ops->completion(hdr);
1604
	return ret;
1605 1606 1607
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);

A
Andy Adamson 已提交
1608
/*
1609
 * There can be multiple RW segments.
A
Andy Adamson 已提交
1610
 */
1611
static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
A
Andy Adamson 已提交
1612
{
1613
	struct pnfs_layout_segment *lseg;
A
Andy Adamson 已提交
1614

1615 1616 1617 1618 1619
	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
		if (lseg->pls_range.iomode == IOMODE_RW &&
		    test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
			list_add(&lseg->pls_lc_list, listp);
	}
A
Andy Adamson 已提交
1620 1621
}

P
Peng Tao 已提交
1622 1623
void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
{
1624
	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
P
Peng Tao 已提交
1625 1626 1627
}
EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);

A
Andy Adamson 已提交
1628 1629 1630
void
pnfs_set_layoutcommit(struct nfs_write_data *wdata)
{
1631 1632 1633
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
	struct nfs_inode *nfsi = NFS_I(inode);
1634
	loff_t end_pos = wdata->mds_offset + wdata->res.count;
1635
	bool mark_as_dirty = false;
A
Andy Adamson 已提交
1636

1637
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1638
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1639
		mark_as_dirty = true;
A
Andy Adamson 已提交
1640
		dprintk("%s: Set layoutcommit for inode %lu ",
1641
			__func__, inode->i_ino);
A
Andy Adamson 已提交
1642
	}
1643
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
1644
		/* references matched in nfs4_layoutcommit_release */
1645
		pnfs_get_lseg(hdr->lseg);
1646
	}
1647 1648
	if (end_pos > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = end_pos;
1649
	spin_unlock(&inode->i_lock);
1650
	dprintk("%s: lseg %p end_pos %llu\n",
1651
		__func__, hdr->lseg, nfsi->layout->plh_lwb);
1652 1653 1654 1655

	/* if pnfs_layoutcommit_inode() runs between inode locks, the next one
	 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
	if (mark_as_dirty)
1656
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1657 1658 1659
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

A
Andy Adamson 已提交
1660 1661 1662 1663 1664 1665 1666 1667
void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
{
	struct nfs_server *nfss = NFS_SERVER(data->args.inode);

	if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
		nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
}

1668 1669 1670 1671 1672 1673 1674 1675
/*
 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
 * data to disk to allow the server to recover the data if it crashes.
 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
 * is off, and a COMMIT is sent to a data server, or
 * if WRITEs to a data server return NFS_DATA_SYNC.
 */
A
Andy Adamson 已提交
1676
int
1677
pnfs_layoutcommit_inode(struct inode *inode, bool sync)
A
Andy Adamson 已提交
1678 1679 1680 1681 1682 1683 1684 1685
{
	struct nfs4_layoutcommit_data *data;
	struct nfs_inode *nfsi = NFS_I(inode);
	loff_t end_pos;
	int status = 0;

	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);

1686 1687 1688
	if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		return 0;

A
Andy Adamson 已提交
1689 1690
	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
	data = kzalloc(sizeof(*data), GFP_NOFS);
1691 1692 1693 1694
	if (!data) {
		status = -ENOMEM;
		goto out;
	}
A
Andy Adamson 已提交
1695

P
Peng Tao 已提交
1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709
	if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		goto out_free;

	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
		if (!sync) {
			status = -EAGAIN;
			goto out_free;
		}
		status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING,
					nfs_wait_bit_killable, TASK_KILLABLE);
		if (status)
			goto out_free;
	}

1710
	INIT_LIST_HEAD(&data->lseg_list);
1711
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1712
	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
P
Peng Tao 已提交
1713
		clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
A
Andy Adamson 已提交
1714
		spin_unlock(&inode->i_lock);
P
Peng Tao 已提交
1715 1716
		wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
		goto out_free;
A
Andy Adamson 已提交
1717
	}
1718 1719

	pnfs_list_write_lseg(inode, &data->lseg_list);
A
Andy Adamson 已提交
1720

1721 1722
	end_pos = nfsi->layout->plh_lwb;
	nfsi->layout->plh_lwb = 0;
A
Andy Adamson 已提交
1723

1724
	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
A
Andy Adamson 已提交
1725 1726 1727
	spin_unlock(&inode->i_lock);

	data->args.inode = inode;
1728
	data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
A
Andy Adamson 已提交
1729 1730 1731 1732 1733 1734 1735 1736
	nfs_fattr_init(&data->fattr);
	data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
	data->res.fattr = &data->fattr;
	data->args.lastbytewritten = end_pos - 1;
	data->res.server = NFS_SERVER(inode);

	status = nfs4_proc_layoutcommit(data, sync);
out:
P
Peng Tao 已提交
1737 1738
	if (status)
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1739 1740
	dprintk("<-- %s status %d\n", __func__, status);
	return status;
P
Peng Tao 已提交
1741 1742 1743
out_free:
	kfree(data);
	goto out;
A
Andy Adamson 已提交
1744
}
1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756

struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
{
	struct nfs4_threshold *thp;

	thp = kzalloc(sizeof(*thp), GFP_NOFS);
	if (!thp) {
		dprintk("%s mdsthreshold allocation failed\n", __func__);
		return NULL;
	}
	return thp;
}