pnfs.c 47.5 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
	struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
219 220
	dprintk("%s: freeing layout cache %p\n", __func__, lo);
	BUG_ON(!list_empty(&lo->plh_layouts));
221 222 223 224
	nfsi->layout = NULL;
	/* Reset MDS Threshold I/O counters */
	nfsi->write_io = 0;
	nfsi->read_io = 0;
225
	pnfs_free_layout_hdr(lo);
226
}
227

228
static void
229
pnfs_put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
230 231 232
{
	if (atomic_dec_and_test(&lo->plh_refcount))
		destroy_layout_hdr(lo);
233 234
}

235
void
236
pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
237
{
238 239 240 241 242 243
	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);
	}
244 245
}

246 247 248 249 250 251 252 253
static int
pnfs_iomode_to_fail_bit(u32 iomode)
{
	return iomode == IOMODE_RW ?
		NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
}

static void
254
pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
255
{
256
	lo->plh_retry_timestamp = jiffies;
257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
	if (test_and_set_bit(fail_bit, &lo->plh_flags))
		atomic_inc(&lo->plh_refcount);
}

static void
pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
{
	if (test_and_clear_bit(fail_bit, &lo->plh_flags))
		atomic_dec(&lo->plh_refcount);
}

static void
pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
{
	struct inode *inode = lo->plh_inode;
272 273 274 275 276 277
	struct pnfs_layout_range range = {
		.iomode = iomode,
		.offset = 0,
		.length = NFS4_MAX_UINT64,
	};
	LIST_HEAD(head);
278 279 280

	spin_lock(&inode->i_lock);
	pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
281
	pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
282
	spin_unlock(&inode->i_lock);
283
	pnfs_free_lseg_list(&head);
284 285 286 287 288 289 290
	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)
{
291
	unsigned long start, end;
292 293 294
	int fail_bit = pnfs_iomode_to_fail_bit(iomode);

	if (test_bit(fail_bit, &lo->plh_flags) == 0)
295 296 297 298 299
		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 */
300
		pnfs_layout_clear_fail_bit(lo, fail_bit);
301 302 303
		return false;
	}
	return true;
304 305
}

306 307 308
static void
init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
{
309
	INIT_LIST_HEAD(&lseg->pls_list);
310
	INIT_LIST_HEAD(&lseg->pls_lc_list);
311 312 313
	atomic_set(&lseg->pls_refcount, 1);
	smp_mb();
	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
314
	lseg->pls_layout = lo;
315 316
}

317
static void free_lseg(struct pnfs_layout_segment *lseg)
318
{
319
	struct inode *ino = lseg->pls_layout->plh_inode;
320

321
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
322
	/* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
323
	pnfs_put_layout_hdr(NFS_I(ino)->layout);
324 325
}

F
Fred Isaman 已提交
326
static void
327 328
pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_segment *lseg)
F
Fred Isaman 已提交
329
{
330
	struct inode *inode = lo->plh_inode;
F
Fred Isaman 已提交
331

332
	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
F
Fred Isaman 已提交
333
	list_del_init(&lseg->pls_list);
334 335
	if (list_empty(&lo->plh_segs)) {
		set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags);
F
Fred Isaman 已提交
336
		/* Matched by initial refcount set in alloc_init_layout_hdr */
337
		pnfs_put_layout_hdr_locked(lo);
F
Fred Isaman 已提交
338 339 340 341
	}
	rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
}

342
void
343
pnfs_put_lseg(struct pnfs_layout_segment *lseg)
344
{
345
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
346 347 348 349 350
	struct inode *inode;

	if (!lseg)
		return;

351 352 353
	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
		atomic_read(&lseg->pls_refcount),
		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
354 355
	lo = lseg->pls_layout;
	inode = lo->plh_inode;
F
Fred Isaman 已提交
356 357
	if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
		LIST_HEAD(free_me);
358

359
		pnfs_layout_remove_lseg(lo, lseg);
F
Fred Isaman 已提交
360
		spin_unlock(&inode->i_lock);
361
		list_add(&lseg->pls_list, &free_me);
F
Fred Isaman 已提交
362
		pnfs_free_lseg_list(&free_me);
363 364
	}
}
365
EXPORT_SYMBOL_GPL(pnfs_put_lseg);
366

367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
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);
}

426
static bool
427 428
should_free_lseg(struct pnfs_layout_range *lseg_range,
		 struct pnfs_layout_range *recall_range)
429
{
430 431 432
	return (recall_range->iomode == IOMODE_ANY ||
		lseg_range->iomode == recall_range->iomode) &&
	       lo_seg_intersecting(lseg_range, recall_range);
433 434
}

435 436 437 438 439 440 441 442 443 444 445
/* 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
446 447 448
		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
			atomic_read(&lseg->pls_refcount));
		if (atomic_dec_and_test(&lseg->pls_refcount)) {
449
			pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
F
Fred Isaman 已提交
450 451 452
			list_add(&lseg->pls_list, tmp_list);
			rv = 1;
		}
453 454 455 456 457 458 459
	}
	return rv;
}

/* Returns count of number of matching invalid lsegs remaining in list
 * after call.
 */
F
Fred Isaman 已提交
460
int
461
pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
462
			    struct list_head *tmp_list,
463
			    struct pnfs_layout_range *recall_range)
464 465
{
	struct pnfs_layout_segment *lseg, *next;
466
	int invalid = 0, removed = 0;
467 468 469

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

470 471
	if (list_empty(&lo->plh_segs)) {
		if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
472
			pnfs_put_layout_hdr_locked(lo);
473 474
		return 0;
	}
475
	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
476 477
		if (!recall_range ||
		    should_free_lseg(&lseg->pls_range, recall_range)) {
478 479 480 481 482 483 484 485 486
			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;
487 488
}

489
/* note free_me must contain lsegs from a single layout_hdr */
F
Fred Isaman 已提交
490
void
491
pnfs_free_lseg_list(struct list_head *free_me)
492
{
493
	struct pnfs_layout_segment *lseg, *tmp;
494 495 496 497 498 499 500
	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;
501

502 503 504 505 506 507 508 509
	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);
	}
510
	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
511
		list_del(&lseg->pls_list);
512
		free_lseg(lseg);
513 514 515
	}
}

516 517 518 519
void
pnfs_destroy_layout(struct nfs_inode *nfsi)
{
	struct pnfs_layout_hdr *lo;
520
	LIST_HEAD(tmp_list);
521 522 523 524

	spin_lock(&nfsi->vfs_inode.i_lock);
	lo = nfsi->layout;
	if (lo) {
525
		lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
526
		pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
527 528 529 530 531 532 533 534
		pnfs_get_layout_hdr(lo);
		pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
		pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
		spin_unlock(&nfsi->vfs_inode.i_lock);
		pnfs_free_lseg_list(&tmp_list);
		pnfs_put_layout_hdr(lo);
	} else
		spin_unlock(&nfsi->vfs_inode.i_lock);
535
}
536
EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
537 538 539 540 541 542 543 544

/*
 * Called by the state manger to remove all layouts established under an
 * expired lease.
 */
void
pnfs_destroy_all_layouts(struct nfs_client *clp)
{
545
	struct nfs_server *server;
546 547 548
	struct pnfs_layout_hdr *lo;
	LIST_HEAD(tmp_list);

549 550 551
	nfs4_deviceid_mark_client_invalid(clp);
	nfs4_deviceid_purge_client(clp);

552
	spin_lock(&clp->cl_lock);
553 554 555 556 557 558
	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();
559 560 561 562
	spin_unlock(&clp->cl_lock);

	while (!list_empty(&tmp_list)) {
		lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
563
				plh_layouts);
564
		dprintk("%s freeing layout for inode %lu\n", __func__,
565
			lo->plh_inode->i_ino);
566
		list_del_init(&lo->plh_layouts);
567
		pnfs_destroy_layout(NFS_I(lo->plh_inode));
568
	}
569 570
}

571
/* update lo->plh_stateid with new if is more recent */
F
Fred Isaman 已提交
572 573 574
void
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
			bool update_barrier)
575
{
576
	u32 oldseq, newseq;
577

578 579
	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
	newseq = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
580
	if ((int)(newseq - oldseq) > 0) {
581
		nfs4_stateid_copy(&lo->plh_stateid, new);
F
Fred Isaman 已提交
582
		if (update_barrier) {
583
			u32 new_barrier = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
584 585 586 587 588 589 590 591 592 593 594 595 596 597 598

			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);
		}
	}
599 600
}

601 602
/* lget is set to 1 if called from inside send_layoutget call chain */
static bool
F
Fred Isaman 已提交
603 604 605 606
pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
			int lget)
{
	if ((stateid) &&
607
	    (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0)
F
Fred Isaman 已提交
608
		return true;
F
Fred Isaman 已提交
609
	return lo->plh_block_lgets ||
610
		test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
F
Fred Isaman 已提交
611
		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
F
Fred Isaman 已提交
612
		(list_empty(&lo->plh_segs) &&
613 614 615
		 (atomic_read(&lo->plh_outstanding) > lget));
}

616 617 618
int
pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
			      struct nfs4_state *open_state)
619
{
620
	int status = 0;
621

622
	dprintk("--> %s\n", __func__);
623
	spin_lock(&lo->plh_inode->i_lock);
F
Fred Isaman 已提交
624
	if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
625 626
		status = -EAGAIN;
	} else if (list_empty(&lo->plh_segs)) {
627 628 629 630
		int seq;

		do {
			seq = read_seqbegin(&open_state->seqlock);
631
			nfs4_stateid_copy(dst, &open_state->stateid);
632 633
		} while (read_seqretry(&open_state->seqlock, seq));
	} else
634
		nfs4_stateid_copy(dst, &lo->plh_stateid);
635
	spin_unlock(&lo->plh_inode->i_lock);
636
	dprintk("<-- %s\n", __func__);
637
	return status;
638 639 640 641 642 643 644 645
}

/*
* Get layout from server.
*    for now, assume that whole file layouts are requested.
*    arg->offset: 0
*    arg->length: all ones
*/
646 647 648
static struct pnfs_layout_segment *
send_layoutget(struct pnfs_layout_hdr *lo,
	   struct nfs_open_context *ctx,
649
	   struct pnfs_layout_range *range,
650
	   gfp_t gfp_flags)
651
{
652
	struct inode *ino = lo->plh_inode;
653 654
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs4_layoutget *lgp;
655
	struct pnfs_layout_segment *lseg;
656 657

	dprintk("--> %s\n", __func__);
658

659
	BUG_ON(ctx == NULL);
660
	lgp = kzalloc(sizeof(*lgp), gfp_flags);
661
	if (lgp == NULL)
662
		return NULL;
663

664 665 666
	lgp->args.minlength = PAGE_CACHE_SIZE;
	if (lgp->args.minlength > range->length)
		lgp->args.minlength = range->length;
667
	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
668
	lgp->args.range = *range;
669 670 671
	lgp->args.type = server->pnfs_curr_ld->id;
	lgp->args.inode = ino;
	lgp->args.ctx = get_nfs_open_context(ctx);
672
	lgp->gfp_flags = gfp_flags;
673 674 675 676

	/* Synchronously retrieve layout information from server and
	 * store in lseg.
	 */
677 678 679 680 681 682 683 684
	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 */
685
			pnfs_layout_io_set_failed(lo, range->iomode);
686 687
		}
		return NULL;
688
	}
689

690 691 692
	return lseg;
}

693 694 695 696 697 698 699 700
/*
 * 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
701 702 703 704 705 706 707 708
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;
709
	int status = 0, empty;
B
Benny Halevy 已提交
710

711
	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
B
Benny Halevy 已提交
712 713 714

	spin_lock(&ino->i_lock);
	lo = nfsi->layout;
715
	if (!lo || pnfs_test_layout_returned(lo)) {
B
Benny Halevy 已提交
716
		spin_unlock(&ino->i_lock);
717 718
		dprintk("NFS: %s no layout to return\n", __func__);
		goto out;
B
Benny Halevy 已提交
719 720 721
	}
	stateid = nfsi->layout->plh_stateid;
	/* Reference matched in nfs4_layoutreturn_release */
722
	pnfs_get_layout_hdr(lo);
723
	empty = list_empty(&lo->plh_segs);
724
	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
725 726 727
	/* Don't send a LAYOUTRETURN if list was initially empty */
	if (empty) {
		spin_unlock(&ino->i_lock);
728
		pnfs_put_layout_hdr(lo);
729 730 731
		dprintk("NFS: %s no layout segments to return\n", __func__);
		goto out;
	}
732
	lo->plh_block_lgets++;
733
	pnfs_mark_layout_returned(lo);
B
Benny Halevy 已提交
734 735 736 737 738 739 740 741
	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;
742 743
		pnfs_layout_io_set_failed(lo, IOMODE_RW);
		pnfs_layout_io_set_failed(lo, IOMODE_READ);
744
		pnfs_clear_layout_returned(lo);
745
		pnfs_put_layout_hdr(lo);
B
Benny Halevy 已提交
746 747 748 749 750 751
		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
752
	lrp->args.layout = lo;
B
Benny Halevy 已提交
753 754 755 756 757 758 759
	lrp->clp = NFS_SERVER(ino)->nfs_client;

	status = nfs4_proc_layoutreturn(lrp);
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}
760
EXPORT_SYMBOL_GPL(_pnfs_return_layout);
B
Benny Halevy 已提交
761

F
Fred Isaman 已提交
762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781
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++;
782
	pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
F
Fred Isaman 已提交
783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798
	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--;
799
	pnfs_put_layout_hdr_locked(lo);
F
Fred Isaman 已提交
800 801 802 803 804 805 806 807 808 809 810 811 812 813
	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);
}

814
bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
F
Fred Isaman 已提交
815 816
{
	struct nfs_inode *nfsi = NFS_I(ino);
817
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
818
	struct pnfs_layout_segment *lseg;
819
	u32 current_seqid;
F
Fred Isaman 已提交
820 821 822 823 824
	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)) {
825
			rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
F
Fred Isaman 已提交
826
			found = true;
827
			goto out;
F
Fred Isaman 已提交
828
		}
829 830
	lo = nfsi->layout;
	current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
F
Fred Isaman 已提交
831

832 833 834 835 836
	/* 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);
out:
F
Fred Isaman 已提交
837 838 839 840
	spin_unlock(&ino->i_lock);
	return found;
}

841 842 843 844 845 846
/*
 * 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
847 848
cmp_layout(struct pnfs_layout_range *l1,
	   struct pnfs_layout_range *l2)
849
{
850 851 852 853 854 855 856 857 858 859 860 861
	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;

862
	/* read > read/write */
863
	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
864 865
}

866
static void
867
pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
868 869
		   struct pnfs_layout_segment *lseg)
{
870 871
	struct pnfs_layout_segment *lp;

872 873
	dprintk("%s:Begin\n", __func__);

874 875
	assert_spin_locked(&lo->plh_inode->i_lock);
	list_for_each_entry(lp, &lo->plh_segs, pls_list) {
876
		if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
877
			continue;
878
		list_add_tail(&lseg->pls_list, &lp->pls_list);
879 880 881
		dprintk("%s: inserted lseg %p "
			"iomode %d offset %llu length %llu before "
			"lp %p iomode %d offset %llu length %llu\n",
882 883 884 885
			__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);
886
		goto out;
887
	}
888 889 890 891 892 893
	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:
894
	pnfs_get_layout_hdr(lo);
895 896

	dprintk("%s:Return\n", __func__);
897 898 899
}

static struct pnfs_layout_hdr *
900 901 902
alloc_init_layout_hdr(struct inode *ino,
		      struct nfs_open_context *ctx,
		      gfp_t gfp_flags)
903 904 905
{
	struct pnfs_layout_hdr *lo;

906
	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
907 908
	if (!lo)
		return NULL;
909
	atomic_set(&lo->plh_refcount, 1);
910 911
	INIT_LIST_HEAD(&lo->plh_layouts);
	INIT_LIST_HEAD(&lo->plh_segs);
F
Fred Isaman 已提交
912
	INIT_LIST_HEAD(&lo->plh_bulk_recall);
913
	lo->plh_inode = ino;
914
	lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
915 916 917 918
	return lo;
}

static struct pnfs_layout_hdr *
919 920 921
pnfs_find_alloc_layout(struct inode *ino,
		       struct nfs_open_context *ctx,
		       gfp_t gfp_flags)
922 923 924 925 926 927 928
{
	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);
929 930 931 932 933 934
	if (nfsi->layout) {
		if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
			return NULL;
		else
			return nfsi->layout;
	}
935
	spin_unlock(&ino->i_lock);
936
	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
937 938 939 940 941
	spin_lock(&ino->i_lock);

	if (likely(nfsi->layout == NULL))	/* Won the race? */
		nfsi->layout = new;
	else
942
		pnfs_free_layout_hdr(new);
943 944 945
	return nfsi->layout;
}

946 947 948 949 950 951 952 953 954 955 956 957
/*
 * 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
958 959
is_matching_lseg(struct pnfs_layout_range *ls_range,
		 struct pnfs_layout_range *range)
960
{
961 962 963 964 965 966 967 968 969 970 971
	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);
972 973 974 975 976
}

/*
 * lookup range in layout
 */
977
static struct pnfs_layout_segment *
978 979
pnfs_find_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_range *range)
980
{
981 982 983 984
	struct pnfs_layout_segment *lseg, *ret = NULL;

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

985 986
	assert_spin_locked(&lo->plh_inode->i_lock);
	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
987
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
988
		    is_matching_lseg(&lseg->pls_range, range)) {
989
			ret = pnfs_get_lseg(lseg);
990 991
			break;
		}
992
		if (lseg->pls_range.offset > range->offset)
993 994 995 996
			break;
	}

	dprintk("%s:Return lseg %p ref %d\n",
997
		__func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
998
	return ret;
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 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
/*
 * 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;
}

1076 1077 1078 1079
/*
 * Layout segment is retreived from the server if not cached.
 * The appropriate layout segment is referenced and returned to the caller.
 */
1080
struct pnfs_layout_segment *
1081 1082
pnfs_update_layout(struct inode *ino,
		   struct nfs_open_context *ctx,
1083 1084
		   loff_t pos,
		   u64 count,
1085 1086
		   enum pnfs_iomode iomode,
		   gfp_t gfp_flags)
1087
{
1088 1089 1090 1091 1092
	struct pnfs_layout_range arg = {
		.iomode = iomode,
		.offset = pos,
		.length = count,
	};
1093
	unsigned pg_offset;
1094 1095
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs_client *clp = server->nfs_client;
1096 1097
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg = NULL;
1098
	bool first = false;
1099 1100

	if (!pnfs_enabled_sb(NFS_SERVER(ino)))
1101
		goto out;
1102 1103

	if (pnfs_within_mdsthreshold(ctx, ino, iomode))
1104
		goto out;
1105

1106
	spin_lock(&ino->i_lock);
1107
	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1108 1109 1110 1111
	if (lo == NULL) {
		spin_unlock(&ino->i_lock);
		goto out;
	}
1112

F
Fred Isaman 已提交
1113
	/* Do we even need to bother with this? */
1114
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1115
		dprintk("%s matches recall, use MDS\n", __func__);
1116 1117 1118 1119
		goto out_unlock;
	}

	/* if LAYOUTGET already failed once we don't try again */
1120
	if (pnfs_layout_io_test_failed(lo, iomode))
1121 1122
		goto out_unlock;

1123
	/* Check to see if the layout for the given range already exists */
1124
	lseg = pnfs_find_lseg(lo, &arg);
1125 1126 1127
	if (lseg)
		goto out_unlock;

F
Fred Isaman 已提交
1128
	if (pnfs_layoutgets_blocked(lo, NULL, 0))
1129 1130 1131
		goto out_unlock;
	atomic_inc(&lo->plh_outstanding);

1132
	pnfs_get_layout_hdr(lo);
1133 1134
	if (list_empty(&lo->plh_segs))
		first = true;
1135 1136 1137 1138

	/* Enable LAYOUTRETURNs */
	pnfs_clear_layout_returned(lo);

1139 1140
	spin_unlock(&ino->i_lock);
	if (first) {
1141 1142 1143 1144 1145
		/* 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));
1146
		list_add_tail(&lo->plh_layouts, &server->layouts);
1147 1148
		spin_unlock(&clp->cl_lock);
	}
1149

1150 1151 1152 1153 1154
	pg_offset = arg.offset & ~PAGE_CACHE_MASK;
	if (pg_offset) {
		arg.offset -= pg_offset;
		arg.length += pg_offset;
	}
1155 1156
	if (arg.length != NFS4_MAX_UINT64)
		arg.length = PAGE_CACHE_ALIGN(arg.length);
1157

1158
	lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1159 1160 1161 1162
	if (!lseg && first) {
		spin_lock(&clp->cl_lock);
		list_del_init(&lo->plh_layouts);
		spin_unlock(&clp->cl_lock);
1163
	}
1164
	atomic_dec(&lo->plh_outstanding);
1165
out_put_layout_hdr:
1166
	pnfs_put_layout_hdr(lo);
1167
out:
1168 1169 1170 1171 1172 1173 1174 1175
	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);
1176 1177 1178
	return lseg;
out_unlock:
	spin_unlock(&ino->i_lock);
1179
	goto out_put_layout_hdr;
1180
}
1181
EXPORT_SYMBOL_GPL(pnfs_update_layout);
1182

1183
struct pnfs_layout_segment *
1184 1185 1186 1187 1188
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;
1189
	struct inode *ino = lo->plh_inode;
1190 1191 1192
	int status = 0;

	/* Inject layout blob into I/O device driver */
1193
	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204
	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);
1205
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1206 1207 1208 1209 1210 1211 1212 1213
		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;
	}
1214
	init_lseg(lo, lseg);
1215
	lseg->pls_range = res->range;
1216
	pnfs_get_lseg(lseg);
1217
	pnfs_layout_insert_lseg(lo, lseg);
1218

F
Fred Isaman 已提交
1219 1220 1221 1222 1223
	if (res->return_on_close) {
		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
		set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
	}

1224
	/* Done processing layoutget. Set the layout stateid */
F
Fred Isaman 已提交
1225
	pnfs_set_layout_stateid(lo, &res->stateid, false);
1226
	spin_unlock(&ino->i_lock);
1227
	return lseg;
1228
out:
1229
	return ERR_PTR(status);
F
Fred Isaman 已提交
1230 1231 1232 1233 1234 1235

out_forget_reply:
	spin_unlock(&ino->i_lock);
	lseg->pls_layout = lo;
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
	goto out;
1236 1237
}

1238 1239 1240 1241 1242
void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
	BUG_ON(pgio->pg_lseg != NULL);

1243 1244 1245 1246
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_read_mds(pgio);
		return;
	}
1247 1248 1249 1250 1251 1252
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
					   req->wb_bytes,
					   IOMODE_READ,
					   GFP_KERNEL);
1253 1254
	/* If no lseg, fall back to read through mds */
	if (pgio->pg_lseg == NULL)
1255
		nfs_pageio_reset_read_mds(pgio);
1256

1257 1258 1259 1260 1261 1262 1263 1264
}
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);

1265 1266 1267 1268
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_write_mds(pgio);
		return;
	}
1269 1270 1271 1272 1273 1274
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
					   req->wb_bytes,
					   IOMODE_RW,
					   GFP_NOFS);
1275 1276
	/* If no lseg, fall back to write through mds */
	if (pgio->pg_lseg == NULL)
1277
		nfs_pageio_reset_write_mds(pgio);
1278 1279 1280
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);

1281
void
1282 1283
pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode,
		      const struct nfs_pgio_completion_ops *compl_ops)
1284 1285 1286 1287 1288
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (ld == NULL)
1289 1290 1291
		nfs_pageio_init_read(pgio, inode, compl_ops);
	else
		nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0);
1292 1293
}

1294
void
1295 1296 1297
pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode,
		       int ioflags,
		       const struct nfs_pgio_completion_ops *compl_ops)
1298 1299 1300 1301 1302
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (ld == NULL)
1303 1304 1305
		nfs_pageio_init_write(pgio, inode, ioflags, compl_ops);
	else
		nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags);
1306 1307
}

1308
bool
1309 1310
pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
		     struct nfs_page *req)
1311
{
1312 1313
	if (pgio->pg_lseg == NULL)
		return nfs_generic_pg_test(pgio, prev, req);
1314

1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330
	/*
	 * 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);
1331
}
1332
EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1333

1334
int pnfs_write_done_resend_to_mds(struct inode *inode,
1335 1336
				struct list_head *head,
				const struct nfs_pgio_completion_ops *compl_ops)
1337 1338 1339 1340 1341
{
	struct nfs_pageio_descriptor pgio;
	LIST_HEAD(failed);

	/* Resend all requests through the MDS */
1342
	nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops);
1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361
	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;
}
1362
EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1363

1364 1365
static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
{
1366 1367 1368 1369
	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 &
1370
	    PNFS_LAYOUTRET_ON_ERROR) {
1371 1372
		clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
		pnfs_return_layout(hdr->inode);
1373
	}
1374 1375
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
1376 1377
							&hdr->pages,
							hdr->completion_ops);
1378 1379
}

1380 1381 1382
/*
 * Called by non rpc-based layout drivers
 */
1383
void pnfs_ld_write_done(struct nfs_write_data *data)
1384
{
1385 1386 1387
	struct nfs_pgio_header *hdr = data->header;

	if (!hdr->pnfs_error) {
1388
		pnfs_set_layoutcommit(data);
1389
		hdr->mds_ops->rpc_call_done(&data->task, data);
1390 1391
	} else
		pnfs_ld_handle_write_error(data);
1392
	hdr->mds_ops->rpc_release(data);
1393
}
1394
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1395

1396 1397 1398 1399
static void
pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
		struct nfs_write_data *data)
{
1400 1401
	struct nfs_pgio_header *hdr = data->header;

1402 1403 1404 1405 1406
	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;
	}
1407 1408 1409 1410
	nfs_writedata_release(data);
}

static enum pnfs_try_status
1411
pnfs_try_to_write_data(struct nfs_write_data *wdata,
1412 1413 1414
			const struct rpc_call_ops *call_ops,
			struct pnfs_layout_segment *lseg,
			int how)
1415
{
1416 1417
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
1418 1419 1420
	enum pnfs_try_status trypnfs;
	struct nfs_server *nfss = NFS_SERVER(inode);

1421
	hdr->mds_ops = call_ops;
1422 1423 1424 1425

	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);
1426
	if (trypnfs != PNFS_NOT_ATTEMPTED)
1427 1428 1429 1430 1431
		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}

1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442
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;

1443
		data = list_first_entry(head, struct nfs_write_data, list);
1444 1445 1446 1447 1448 1449
		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);
	}
1450
	pnfs_put_lseg(lseg);
1451 1452
}

1453 1454
static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
{
1455
	pnfs_put_lseg(hdr->lseg);
1456 1457
	nfs_writehdr_free(hdr);
}
B
Bryan Schumaker 已提交
1458
EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
1459

1460 1461 1462
int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
{
1463 1464
	struct nfs_write_header *whdr;
	struct nfs_pgio_header *hdr;
1465 1466
	int ret;

1467 1468
	whdr = nfs_writehdr_alloc();
	if (!whdr) {
1469
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1470
		pnfs_put_lseg(desc->pg_lseg);
1471
		desc->pg_lseg = NULL;
1472
		return -ENOMEM;
1473
	}
1474 1475
	hdr = &whdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1476
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1477 1478 1479
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_flush(desc, hdr);
	if (ret != 0) {
1480
		pnfs_put_lseg(desc->pg_lseg);
1481 1482 1483 1484
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
	if (atomic_dec_and_test(&hdr->refcnt))
1485
		hdr->completion_ops->completion(hdr);
1486
	return ret;
1487 1488 1489
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);

1490
int pnfs_read_done_resend_to_mds(struct inode *inode,
1491 1492
				struct list_head *head,
				const struct nfs_pgio_completion_ops *compl_ops)
1493 1494
{
	struct nfs_pageio_descriptor pgio;
1495
	LIST_HEAD(failed);
1496

1497
	/* Resend all requests through the MDS */
1498
	nfs_pageio_init_read(&pgio, inode, compl_ops);
1499 1500
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);
1501 1502

		nfs_list_remove_request(req);
1503 1504
		if (!nfs_pageio_add_request(&pgio, req))
			nfs_list_add_request(req, &failed);
1505 1506
	}
	nfs_pageio_complete(&pgio);
1507 1508 1509 1510 1511 1512 1513

	if (!list_empty(&failed)) {
		list_move(&failed, head);
		return -EIO;
	}
	return 0;
}
1514
EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1515 1516 1517

static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
{
1518 1519 1520 1521
	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 &
1522
	    PNFS_LAYOUTRET_ON_ERROR) {
1523 1524
		clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
		pnfs_return_layout(hdr->inode);
1525
	}
1526 1527
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
1528 1529
							&hdr->pages,
							hdr->completion_ops);
1530 1531
}

1532 1533 1534
/*
 * Called by non rpc-based layout drivers
 */
1535
void pnfs_ld_read_done(struct nfs_read_data *data)
1536
{
1537 1538 1539
	struct nfs_pgio_header *hdr = data->header;

	if (likely(!hdr->pnfs_error)) {
1540
		__nfs4_read_done_cb(data);
1541
		hdr->mds_ops->rpc_call_done(&data->task, data);
1542 1543
	} else
		pnfs_ld_handle_read_error(data);
1544
	hdr->mds_ops->rpc_release(data);
1545 1546 1547
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

1548 1549 1550 1551
static void
pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
		struct nfs_read_data *data)
{
1552 1553
	struct nfs_pgio_header *hdr = data->header;

1554 1555 1556 1557 1558
	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;
	}
1559 1560 1561
	nfs_readdata_release(data);
}

A
Andy Adamson 已提交
1562 1563 1564
/*
 * Call the appropriate parallel I/O subsystem read function.
 */
1565
static enum pnfs_try_status
A
Andy Adamson 已提交
1566
pnfs_try_to_read_data(struct nfs_read_data *rdata,
1567 1568
		       const struct rpc_call_ops *call_ops,
		       struct pnfs_layout_segment *lseg)
A
Andy Adamson 已提交
1569
{
1570 1571
	struct nfs_pgio_header *hdr = rdata->header;
	struct inode *inode = hdr->inode;
A
Andy Adamson 已提交
1572 1573 1574
	struct nfs_server *nfss = NFS_SERVER(inode);
	enum pnfs_try_status trypnfs;

1575
	hdr->mds_ops = call_ops;
A
Andy Adamson 已提交
1576 1577 1578 1579 1580

	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);
1581
	if (trypnfs != PNFS_NOT_ATTEMPTED)
A
Andy Adamson 已提交
1582 1583 1584 1585
		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}
A
Andy Adamson 已提交
1586

1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597
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;

1598
		data = list_first_entry(head, struct nfs_read_data, list);
1599 1600 1601 1602 1603 1604
		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);
	}
1605
	pnfs_put_lseg(lseg);
1606 1607
}

1608 1609
static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
{
1610
	pnfs_put_lseg(hdr->lseg);
1611 1612
	nfs_readhdr_free(hdr);
}
B
Bryan Schumaker 已提交
1613
EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
1614

1615 1616 1617
int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
{
1618 1619
	struct nfs_read_header *rhdr;
	struct nfs_pgio_header *hdr;
1620 1621
	int ret;

1622 1623
	rhdr = nfs_readhdr_alloc();
	if (!rhdr) {
1624
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1625
		ret = -ENOMEM;
1626
		pnfs_put_lseg(desc->pg_lseg);
1627 1628 1629
		desc->pg_lseg = NULL;
		return ret;
	}
1630 1631
	hdr = &rhdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1632
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1633 1634 1635
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_pagein(desc, hdr);
	if (ret != 0) {
1636
		pnfs_put_lseg(desc->pg_lseg);
1637 1638 1639 1640
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_reads(desc, &hdr->rpc_list);
	if (atomic_dec_and_test(&hdr->refcnt))
1641
		hdr->completion_ops->completion(hdr);
1642
	return ret;
1643 1644 1645
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);

A
Andy Adamson 已提交
1646
/*
1647
 * There can be multiple RW segments.
A
Andy Adamson 已提交
1648
 */
1649
static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
A
Andy Adamson 已提交
1650
{
1651
	struct pnfs_layout_segment *lseg;
A
Andy Adamson 已提交
1652

1653 1654 1655 1656 1657
	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
1658 1659
}

P
Peng Tao 已提交
1660 1661
void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
{
1662
	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
P
Peng Tao 已提交
1663 1664 1665
}
EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);

A
Andy Adamson 已提交
1666 1667 1668
void
pnfs_set_layoutcommit(struct nfs_write_data *wdata)
{
1669 1670 1671
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
	struct nfs_inode *nfsi = NFS_I(inode);
1672
	loff_t end_pos = wdata->mds_offset + wdata->res.count;
1673
	bool mark_as_dirty = false;
A
Andy Adamson 已提交
1674

1675
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1676
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1677
		mark_as_dirty = true;
A
Andy Adamson 已提交
1678
		dprintk("%s: Set layoutcommit for inode %lu ",
1679
			__func__, inode->i_ino);
A
Andy Adamson 已提交
1680
	}
1681
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
1682
		/* references matched in nfs4_layoutcommit_release */
1683
		pnfs_get_lseg(hdr->lseg);
1684
	}
1685 1686
	if (end_pos > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = end_pos;
1687
	spin_unlock(&inode->i_lock);
1688
	dprintk("%s: lseg %p end_pos %llu\n",
1689
		__func__, hdr->lseg, nfsi->layout->plh_lwb);
1690 1691 1692 1693

	/* 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)
1694
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1695 1696 1697
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

A
Andy Adamson 已提交
1698 1699 1700 1701 1702 1703 1704 1705
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);
}

1706 1707 1708 1709 1710 1711 1712 1713
/*
 * 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
1714
int
1715
pnfs_layoutcommit_inode(struct inode *inode, bool sync)
A
Andy Adamson 已提交
1716 1717 1718 1719 1720 1721 1722 1723
{
	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);

1724 1725 1726
	if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		return 0;

A
Andy Adamson 已提交
1727 1728
	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
	data = kzalloc(sizeof(*data), GFP_NOFS);
1729 1730 1731 1732
	if (!data) {
		status = -ENOMEM;
		goto out;
	}
A
Andy Adamson 已提交
1733

P
Peng Tao 已提交
1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747
	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;
	}

1748
	INIT_LIST_HEAD(&data->lseg_list);
1749
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1750
	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
P
Peng Tao 已提交
1751
		clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
A
Andy Adamson 已提交
1752
		spin_unlock(&inode->i_lock);
P
Peng Tao 已提交
1753 1754
		wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
		goto out_free;
A
Andy Adamson 已提交
1755
	}
1756 1757

	pnfs_list_write_lseg(inode, &data->lseg_list);
A
Andy Adamson 已提交
1758

1759 1760
	end_pos = nfsi->layout->plh_lwb;
	nfsi->layout->plh_lwb = 0;
A
Andy Adamson 已提交
1761

1762
	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
A
Andy Adamson 已提交
1763 1764 1765
	spin_unlock(&inode->i_lock);

	data->args.inode = inode;
1766
	data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
A
Andy Adamson 已提交
1767 1768 1769 1770 1771 1772 1773 1774
	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
1775 1776
	if (status)
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1777 1778
	dprintk("<-- %s status %d\n", __func__, status);
	return status;
P
Peng Tao 已提交
1779 1780 1781
out_free:
	kfree(data);
	goto out;
A
Andy Adamson 已提交
1782
}
1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794

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;
}