pnfs.c 51.0 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
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;
203
	return ld->alloc_layout_hdr(ino, gfp_flags);
204 205 206 207 208
}

static void
pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
{
209 210 211 212 213 214 215 216 217 218
	struct nfs_server *server = NFS_SERVER(lo->plh_inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (!list_empty(&lo->plh_layouts)) {
		struct nfs_client *clp = server->nfs_client;

		spin_lock(&clp->cl_lock);
		list_del_init(&lo->plh_layouts);
		spin_unlock(&clp->cl_lock);
	}
219
	put_rpccred(lo->plh_lc_cred);
220
	return ld->free_layout_hdr(lo);
221 222
}

223
static void
224
pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
225
{
226
	struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
227
	dprintk("%s: freeing layout cache %p\n", __func__, lo);
228 229 230 231
	nfsi->layout = NULL;
	/* Reset MDS Threshold I/O counters */
	nfsi->write_io = 0;
	nfsi->read_io = 0;
232 233
}

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

	if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
240
		pnfs_detach_layout_hdr(lo);
241
		spin_unlock(&inode->i_lock);
242
		pnfs_free_layout_hdr(lo);
243
	}
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
	if (!test_and_set_bit(fail_bit, &lo->plh_flags))
258 259 260 261 262 263 264 265 266 267 268 269 270 271
		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 pnfs_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);
322 323
}

F
Fred Isaman 已提交
324
static void
325 326
pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_segment *lseg)
F
Fred Isaman 已提交
327
{
328
	struct inode *inode = lo->plh_inode;
F
Fred Isaman 已提交
329

330
	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
F
Fred Isaman 已提交
331
	list_del_init(&lseg->pls_list);
332 333
	/* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
	atomic_dec(&lo->plh_refcount);
334 335
	if (list_empty(&lo->plh_segs))
		clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
F
Fred Isaman 已提交
336 337 338
	rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
}

339
void
340
pnfs_put_lseg(struct pnfs_layout_segment *lseg)
341
{
342
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
343 344 345 346 347
	struct inode *inode;

	if (!lseg)
		return;

348 349 350
	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
		atomic_read(&lseg->pls_refcount),
		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
351 352
	lo = lseg->pls_layout;
	inode = lo->plh_inode;
F
Fred Isaman 已提交
353
	if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
354
		pnfs_get_layout_hdr(lo);
355
		pnfs_layout_remove_lseg(lo, lseg);
F
Fred Isaman 已提交
356
		spin_unlock(&inode->i_lock);
357
		pnfs_free_lseg(lseg);
358
		pnfs_put_layout_hdr(lo);
359 360
	}
}
361
EXPORT_SYMBOL_GPL(pnfs_put_lseg);
362

363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410
static inline u64
end_offset(u64 start, u64 len)
{
	u64 end;

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

411
static bool
412 413
should_free_lseg(struct pnfs_layout_range *lseg_range,
		 struct pnfs_layout_range *recall_range)
414
{
415 416 417
	return (recall_range->iomode == IOMODE_ANY ||
		lseg_range->iomode == recall_range->iomode) &&
	       lo_seg_intersecting(lseg_range, recall_range);
418 419
}

420 421 422 423 424 425 426 427 428 429
static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
		struct list_head *tmp_list)
{
	if (!atomic_dec_and_test(&lseg->pls_refcount))
		return false;
	pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
	list_add(&lseg->pls_list, tmp_list);
	return true;
}

430 431 432 433 434 435 436 437 438 439 440
/* 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
441 442
		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
			atomic_read(&lseg->pls_refcount));
443
		if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
F
Fred Isaman 已提交
444
			rv = 1;
445 446 447 448 449 450 451
	}
	return rv;
}

/* Returns count of number of matching invalid lsegs remaining in list
 * after call.
 */
F
Fred Isaman 已提交
452
int
453
pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
454
			    struct list_head *tmp_list,
455
			    struct pnfs_layout_range *recall_range)
456 457
{
	struct pnfs_layout_segment *lseg, *next;
458
	int invalid = 0, removed = 0;
459 460 461

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

462
	if (list_empty(&lo->plh_segs))
463
		return 0;
464
	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
465 466
		if (!recall_range ||
		    should_free_lseg(&lseg->pls_range, recall_range)) {
467 468 469 470 471 472 473 474 475
			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;
476 477
}

478
/* note free_me must contain lsegs from a single layout_hdr */
F
Fred Isaman 已提交
479
void
480
pnfs_free_lseg_list(struct list_head *free_me)
481
{
482
	struct pnfs_layout_segment *lseg, *tmp;
483 484 485 486

	if (list_empty(free_me))
		return;

487
	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
488
		list_del(&lseg->pls_list);
489
		pnfs_free_lseg(lseg);
490 491 492
	}
}

493 494 495 496
void
pnfs_destroy_layout(struct nfs_inode *nfsi)
{
	struct pnfs_layout_hdr *lo;
497
	LIST_HEAD(tmp_list);
498 499 500 501

	spin_lock(&nfsi->vfs_inode.i_lock);
	lo = nfsi->layout;
	if (lo) {
502
		lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
503
		pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
504 505 506 507 508 509 510 511
		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);
512
}
513
EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
514

515 516 517
static bool
pnfs_layout_add_bulk_destroy_list(struct inode *inode,
		struct list_head *layout_list)
518 519
{
	struct pnfs_layout_hdr *lo;
520
	bool ret = false;
521

522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600
	spin_lock(&inode->i_lock);
	lo = NFS_I(inode)->layout;
	if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
		pnfs_get_layout_hdr(lo);
		list_add(&lo->plh_bulk_destroy, layout_list);
		ret = true;
	}
	spin_unlock(&inode->i_lock);
	return ret;
}

/* Caller must hold rcu_read_lock and clp->cl_lock */
static int
pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
		struct nfs_server *server,
		struct list_head *layout_list)
{
	struct pnfs_layout_hdr *lo, *next;
	struct inode *inode;

	list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
		inode = igrab(lo->plh_inode);
		if (inode == NULL)
			continue;
		list_del_init(&lo->plh_layouts);
		if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
			continue;
		rcu_read_unlock();
		spin_unlock(&clp->cl_lock);
		iput(inode);
		spin_lock(&clp->cl_lock);
		rcu_read_lock();
		return -EAGAIN;
	}
	return 0;
}

static int
pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
		bool is_bulk_recall)
{
	struct pnfs_layout_hdr *lo;
	struct inode *inode;
	struct pnfs_layout_range range = {
		.iomode = IOMODE_ANY,
		.offset = 0,
		.length = NFS4_MAX_UINT64,
	};
	LIST_HEAD(lseg_list);
	int ret = 0;

	while (!list_empty(layout_list)) {
		lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
				plh_bulk_destroy);
		dprintk("%s freeing layout for inode %lu\n", __func__,
			lo->plh_inode->i_ino);
		inode = lo->plh_inode;
		spin_lock(&inode->i_lock);
		list_del_init(&lo->plh_bulk_destroy);
		lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
		if (is_bulk_recall)
			set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
		if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range))
			ret = -EAGAIN;
		spin_unlock(&inode->i_lock);
		pnfs_free_lseg_list(&lseg_list);
		pnfs_put_layout_hdr(lo);
		iput(inode);
	}
	return ret;
}

int
pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
		struct nfs_fsid *fsid,
		bool is_recall)
{
	struct nfs_server *server;
	LIST_HEAD(layout_list);
601

602
	spin_lock(&clp->cl_lock);
603
	rcu_read_lock();
604
restart:
605
	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
606 607 608 609 610 611
		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
			continue;
		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
				server,
				&layout_list) != 0)
			goto restart;
612 613
	}
	rcu_read_unlock();
614 615
	spin_unlock(&clp->cl_lock);

616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635
	if (list_empty(&layout_list))
		return 0;
	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
}

int
pnfs_destroy_layouts_byclid(struct nfs_client *clp,
		bool is_recall)
{
	struct nfs_server *server;
	LIST_HEAD(layout_list);

	spin_lock(&clp->cl_lock);
	rcu_read_lock();
restart:
	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
					server,
					&layout_list) != 0)
			goto restart;
636
	}
637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655
	rcu_read_unlock();
	spin_unlock(&clp->cl_lock);

	if (list_empty(&layout_list))
		return 0;
	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
}

/*
 * Called by the state manger to remove all layouts established under an
 * expired lease.
 */
void
pnfs_destroy_all_layouts(struct nfs_client *clp)
{
	nfs4_deviceid_mark_client_invalid(clp);
	nfs4_deviceid_purge_client(clp);

	pnfs_destroy_layouts_byclid(clp, false);
656 657
}

658 659 660 661 662 663 664 665 666
/*
 * Compare 2 layout stateid sequence ids, to see which is newer,
 * taking into account wraparound issues.
 */
static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
{
	return (s32)s1 - (s32)s2 > 0;
}

667
/* update lo->plh_stateid with new if is more recent */
F
Fred Isaman 已提交
668 669 670
void
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
			bool update_barrier)
671
{
672 673
	u32 oldseq, newseq, new_barrier;
	int empty = list_empty(&lo->plh_segs);
674

675 676
	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
	newseq = be32_to_cpu(new->seqid);
677
	if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
678
		nfs4_stateid_copy(&lo->plh_stateid, new);
F
Fred Isaman 已提交
679
		if (update_barrier) {
680
			new_barrier = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
681 682
		} else {
			/* Because of wraparound, we want to keep the barrier
683
			 * "close" to the current seqids.
F
Fred Isaman 已提交
684
			 */
685
			new_barrier = newseq - atomic_read(&lo->plh_outstanding);
F
Fred Isaman 已提交
686
		}
687 688
		if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
			lo->plh_barrier = new_barrier;
F
Fred Isaman 已提交
689
	}
690 691
}

692
static bool
693 694
pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
		const nfs4_stateid *stateid)
F
Fred Isaman 已提交
695
{
696
	u32 seqid = be32_to_cpu(stateid->seqid);
697

698 699 700 701 702 703 704
	return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
}

/* lget is set to 1 if called from inside send_layoutget call chain */
static bool
pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget)
{
F
Fred Isaman 已提交
705 706
	return lo->plh_block_lgets ||
		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
F
Fred Isaman 已提交
707
		(list_empty(&lo->plh_segs) &&
708 709 710
		 (atomic_read(&lo->plh_outstanding) > lget));
}

711 712 713
int
pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
			      struct nfs4_state *open_state)
714
{
715
	int status = 0;
716

717
	dprintk("--> %s\n", __func__);
718
	spin_lock(&lo->plh_inode->i_lock);
719
	if (pnfs_layoutgets_blocked(lo, 1)) {
720
		status = -EAGAIN;
721 722
	} else if (!nfs4_valid_open_stateid(open_state)) {
		status = -EBADF;
723
	} else if (list_empty(&lo->plh_segs)) {
724 725 726 727
		int seq;

		do {
			seq = read_seqbegin(&open_state->seqlock);
728
			nfs4_stateid_copy(dst, &open_state->stateid);
729 730
		} while (read_seqretry(&open_state->seqlock, seq));
	} else
731
		nfs4_stateid_copy(dst, &lo->plh_stateid);
732
	spin_unlock(&lo->plh_inode->i_lock);
733
	dprintk("<-- %s\n", __func__);
734
	return status;
735 736 737 738 739 740 741 742
}

/*
* Get layout from server.
*    for now, assume that whole file layouts are requested.
*    arg->offset: 0
*    arg->length: all ones
*/
743 744 745
static struct pnfs_layout_segment *
send_layoutget(struct pnfs_layout_hdr *lo,
	   struct nfs_open_context *ctx,
746
	   struct pnfs_layout_range *range,
747
	   gfp_t gfp_flags)
748
{
749
	struct inode *ino = lo->plh_inode;
750 751
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs4_layoutget *lgp;
752
	struct pnfs_layout_segment *lseg;
753 754

	dprintk("--> %s\n", __func__);
755

756
	lgp = kzalloc(sizeof(*lgp), gfp_flags);
757
	if (lgp == NULL)
758
		return NULL;
759

760 761 762
	lgp->args.minlength = PAGE_CACHE_SIZE;
	if (lgp->args.minlength > range->length)
		lgp->args.minlength = range->length;
763
	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
764
	lgp->args.range = *range;
765 766 767
	lgp->args.type = server->pnfs_curr_ld->id;
	lgp->args.inode = ino;
	lgp->args.ctx = get_nfs_open_context(ctx);
768
	lgp->gfp_flags = gfp_flags;
769 770 771 772

	/* Synchronously retrieve layout information from server and
	 * store in lseg.
	 */
773 774 775 776 777 778 779 780
	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 */
781
			pnfs_layout_io_set_failed(lo, range->iomode);
782 783
		}
		return NULL;
784
	}
785

786 787 788
	return lseg;
}

789 790 791 792 793 794 795 796 797 798 799 800 801 802 803
static void pnfs_clear_layoutcommit(struct inode *inode,
		struct list_head *head)
{
	struct nfs_inode *nfsi = NFS_I(inode);
	struct pnfs_layout_segment *lseg, *tmp;

	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		return;
	list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
		if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
			continue;
		pnfs_lseg_dec_and_remove_zero(lseg, head);
	}
}

804 805 806 807 808 809 810 811
/*
 * 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
812 813 814 815 816 817 818 819
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;
820
	int status = 0, empty;
B
Benny Halevy 已提交
821

822
	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
B
Benny Halevy 已提交
823 824 825

	spin_lock(&ino->i_lock);
	lo = nfsi->layout;
826
	if (!lo) {
B
Benny Halevy 已提交
827
		spin_unlock(&ino->i_lock);
828 829
		dprintk("NFS: %s no layout to return\n", __func__);
		goto out;
B
Benny Halevy 已提交
830 831 832
	}
	stateid = nfsi->layout->plh_stateid;
	/* Reference matched in nfs4_layoutreturn_release */
833
	pnfs_get_layout_hdr(lo);
834
	empty = list_empty(&lo->plh_segs);
835
	pnfs_clear_layoutcommit(ino, &tmp_list);
836
	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
837 838 839
	/* Don't send a LAYOUTRETURN if list was initially empty */
	if (empty) {
		spin_unlock(&ino->i_lock);
840
		pnfs_put_layout_hdr(lo);
841 842 843
		dprintk("NFS: %s no layout segments to return\n", __func__);
		goto out;
	}
844
	lo->plh_block_lgets++;
B
Benny Halevy 已提交
845 846 847 848 849 850
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);

	lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
	if (unlikely(lrp == NULL)) {
		status = -ENOMEM;
851 852 853
		spin_lock(&ino->i_lock);
		lo->plh_block_lgets--;
		spin_unlock(&ino->i_lock);
854
		pnfs_put_layout_hdr(lo);
B
Benny Halevy 已提交
855 856 857 858 859 860
		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
861
	lrp->args.layout = lo;
B
Benny Halevy 已提交
862 863 864 865 866 867 868
	lrp->clp = NFS_SERVER(ino)->nfs_client;

	status = nfs4_proc_layoutreturn(lrp);
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}
869
EXPORT_SYMBOL_GPL(_pnfs_return_layout);
B
Benny Halevy 已提交
870

871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897
int
pnfs_commit_and_return_layout(struct inode *inode)
{
	struct pnfs_layout_hdr *lo;
	int ret;

	spin_lock(&inode->i_lock);
	lo = NFS_I(inode)->layout;
	if (lo == NULL) {
		spin_unlock(&inode->i_lock);
		return 0;
	}
	pnfs_get_layout_hdr(lo);
	/* Block new layoutgets and read/write to ds */
	lo->plh_block_lgets++;
	spin_unlock(&inode->i_lock);
	filemap_fdatawait(inode->i_mapping);
	ret = pnfs_layoutcommit_inode(inode, true);
	if (ret == 0)
		ret = _pnfs_return_layout(inode);
	spin_lock(&inode->i_lock);
	lo->plh_block_lgets--;
	spin_unlock(&inode->i_lock);
	pnfs_put_layout_hdr(lo);
	return ret;
}

F
Fred Isaman 已提交
898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917
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++;
918
	pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
F
Fred Isaman 已提交
919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934
	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--;
935 936 937 938 939 940
	if (atomic_dec_and_test(&lo->plh_refcount)) {
		pnfs_detach_layout_hdr(lo);
		spin_unlock(&ino->i_lock);
		pnfs_free_layout_hdr(lo);
	} else
		spin_unlock(&ino->i_lock);
F
Fred Isaman 已提交
941 942 943 944 945 946 947 948
}

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;
949
	if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
F
Fred Isaman 已提交
950 951 952 953
		lo->plh_barrier = barrier;
	spin_unlock(&ino->i_lock);
}

954
bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
F
Fred Isaman 已提交
955 956
{
	struct nfs_inode *nfsi = NFS_I(ino);
957
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
958
	struct pnfs_layout_segment *lseg;
959
	u32 current_seqid;
F
Fred Isaman 已提交
960 961 962 963 964
	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)) {
965
			rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
F
Fred Isaman 已提交
966
			found = true;
967
			goto out;
F
Fred Isaman 已提交
968
		}
969 970
	lo = nfsi->layout;
	current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
F
Fred Isaman 已提交
971

972 973 974 975 976
	/* 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
977 978 979 980
	spin_unlock(&ino->i_lock);
	return found;
}

981 982 983 984 985 986
/*
 * 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
987 988
cmp_layout(struct pnfs_layout_range *l1,
	   struct pnfs_layout_range *l2)
989
{
990 991 992 993 994 995 996 997 998 999 1000 1001
	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;

1002
	/* read > read/write */
1003
	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1004 1005
}

1006
static void
1007
pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1008 1009
		   struct pnfs_layout_segment *lseg)
{
1010 1011
	struct pnfs_layout_segment *lp;

1012 1013
	dprintk("%s:Begin\n", __func__);

1014
	list_for_each_entry(lp, &lo->plh_segs, pls_list) {
1015
		if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
1016
			continue;
1017
		list_add_tail(&lseg->pls_list, &lp->pls_list);
1018 1019 1020
		dprintk("%s: inserted lseg %p "
			"iomode %d offset %llu length %llu before "
			"lp %p iomode %d offset %llu length %llu\n",
1021 1022 1023 1024
			__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);
1025
		goto out;
1026
	}
1027 1028 1029 1030 1031 1032
	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:
1033
	pnfs_get_layout_hdr(lo);
1034 1035

	dprintk("%s:Return\n", __func__);
1036 1037 1038
}

static struct pnfs_layout_hdr *
1039 1040 1041
alloc_init_layout_hdr(struct inode *ino,
		      struct nfs_open_context *ctx,
		      gfp_t gfp_flags)
1042 1043 1044
{
	struct pnfs_layout_hdr *lo;

1045
	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1046 1047
	if (!lo)
		return NULL;
1048
	atomic_set(&lo->plh_refcount, 1);
1049 1050
	INIT_LIST_HEAD(&lo->plh_layouts);
	INIT_LIST_HEAD(&lo->plh_segs);
1051
	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1052
	lo->plh_inode = ino;
1053
	lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
1054 1055 1056 1057
	return lo;
}

static struct pnfs_layout_hdr *
1058 1059 1060
pnfs_find_alloc_layout(struct inode *ino,
		       struct nfs_open_context *ctx,
		       gfp_t gfp_flags)
1061 1062 1063 1064 1065 1066
{
	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);

1067 1068
	if (nfsi->layout != NULL)
		goto out_existing;
1069
	spin_unlock(&ino->i_lock);
1070
	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1071 1072
	spin_lock(&ino->i_lock);

1073
	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1074
		nfsi->layout = new;
1075
		return new;
1076 1077
	} else if (new != NULL)
		pnfs_free_layout_hdr(new);
1078 1079
out_existing:
	pnfs_get_layout_hdr(nfsi->layout);
1080 1081 1082
	return nfsi->layout;
}

1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094
/*
 * 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
1095 1096
is_matching_lseg(struct pnfs_layout_range *ls_range,
		 struct pnfs_layout_range *range)
1097
{
1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
	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);
1109 1110 1111 1112 1113
}

/*
 * lookup range in layout
 */
1114
static struct pnfs_layout_segment *
1115 1116
pnfs_find_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_range *range)
1117
{
1118 1119 1120 1121
	struct pnfs_layout_segment *lseg, *ret = NULL;

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

1122
	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1123
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1124
		    is_matching_lseg(&lseg->pls_range, range)) {
1125
			ret = pnfs_get_lseg(lseg);
1126 1127
			break;
		}
1128
		if (lseg->pls_range.offset > range->offset)
1129 1130 1131 1132
			break;
	}

	dprintk("%s:Return lseg %p ref %d\n",
1133
		__func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1134
	return ret;
1135 1136
}

1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211
/*
 * 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;
}

1212 1213 1214 1215
/*
 * Layout segment is retreived from the server if not cached.
 * The appropriate layout segment is referenced and returned to the caller.
 */
1216
struct pnfs_layout_segment *
1217 1218
pnfs_update_layout(struct inode *ino,
		   struct nfs_open_context *ctx,
1219 1220
		   loff_t pos,
		   u64 count,
1221 1222
		   enum pnfs_iomode iomode,
		   gfp_t gfp_flags)
1223
{
1224 1225 1226 1227 1228
	struct pnfs_layout_range arg = {
		.iomode = iomode,
		.offset = pos,
		.length = count,
	};
1229
	unsigned pg_offset;
1230 1231
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs_client *clp = server->nfs_client;
1232 1233
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg = NULL;
1234
	bool first;
1235 1236

	if (!pnfs_enabled_sb(NFS_SERVER(ino)))
1237
		goto out;
1238 1239

	if (pnfs_within_mdsthreshold(ctx, ino, iomode))
1240
		goto out;
1241

1242
	spin_lock(&ino->i_lock);
1243
	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1244 1245 1246 1247
	if (lo == NULL) {
		spin_unlock(&ino->i_lock);
		goto out;
	}
1248

F
Fred Isaman 已提交
1249
	/* Do we even need to bother with this? */
1250
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1251
		dprintk("%s matches recall, use MDS\n", __func__);
1252 1253 1254 1255
		goto out_unlock;
	}

	/* if LAYOUTGET already failed once we don't try again */
1256
	if (pnfs_layout_io_test_failed(lo, iomode))
1257 1258
		goto out_unlock;

1259
	/* Check to see if the layout for the given range already exists */
1260
	lseg = pnfs_find_lseg(lo, &arg);
1261 1262 1263
	if (lseg)
		goto out_unlock;

1264
	if (pnfs_layoutgets_blocked(lo, 0))
1265 1266 1267
		goto out_unlock;
	atomic_inc(&lo->plh_outstanding);

1268
	first = list_empty(&lo->plh_layouts) ? true : false;
1269
	spin_unlock(&ino->i_lock);
1270

1271
	if (first) {
1272 1273 1274 1275
		/* 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);
1276
		list_add_tail(&lo->plh_layouts, &server->layouts);
1277 1278
		spin_unlock(&clp->cl_lock);
	}
1279

1280 1281 1282 1283 1284
	pg_offset = arg.offset & ~PAGE_CACHE_MASK;
	if (pg_offset) {
		arg.offset -= pg_offset;
		arg.length += pg_offset;
	}
1285 1286
	if (arg.length != NFS4_MAX_UINT64)
		arg.length = PAGE_CACHE_ALIGN(arg.length);
1287

1288
	lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1289
	atomic_dec(&lo->plh_outstanding);
1290
out_put_layout_hdr:
1291
	pnfs_put_layout_hdr(lo);
1292
out:
1293 1294 1295 1296 1297 1298 1299 1300
	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);
1301 1302 1303
	return lseg;
out_unlock:
	spin_unlock(&ino->i_lock);
1304
	goto out_put_layout_hdr;
1305
}
1306
EXPORT_SYMBOL_GPL(pnfs_update_layout);
1307

1308
struct pnfs_layout_segment *
1309 1310 1311 1312 1313
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;
1314
	struct inode *ino = lo->plh_inode;
1315 1316 1317
	int status = 0;

	/* Inject layout blob into I/O device driver */
1318
	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
	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);
1330
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1331 1332 1333 1334
		dprintk("%s forget reply due to recall\n", __func__);
		goto out_forget_reply;
	}

1335 1336
	if (pnfs_layoutgets_blocked(lo, 1) ||
	    pnfs_layout_stateid_blocked(lo, &res->stateid)) {
F
Fred Isaman 已提交
1337 1338 1339
		dprintk("%s forget reply due to state\n", __func__);
		goto out_forget_reply;
	}
1340 1341 1342 1343

	/* Done processing layoutget. Set the layout stateid */
	pnfs_set_layout_stateid(lo, &res->stateid, false);

1344
	init_lseg(lo, lseg);
1345
	lseg->pls_range = res->range;
1346
	pnfs_get_lseg(lseg);
1347
	pnfs_layout_insert_lseg(lo, lseg);
1348

F
Fred Isaman 已提交
1349 1350 1351 1352 1353
	if (res->return_on_close) {
		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
		set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
	}

1354
	spin_unlock(&ino->i_lock);
1355
	return lseg;
1356
out:
1357
	return ERR_PTR(status);
F
Fred Isaman 已提交
1358 1359 1360 1361 1362 1363

out_forget_reply:
	spin_unlock(&ino->i_lock);
	lseg->pls_layout = lo;
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
	goto out;
1364 1365
}

1366 1367 1368
void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
P
Peng Tao 已提交
1369 1370
	u64 rd_size = req->wb_bytes;

1371
	WARN_ON_ONCE(pgio->pg_lseg != NULL);
1372

1373 1374 1375 1376
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_read_mds(pgio);
		return;
	}
P
Peng Tao 已提交
1377 1378 1379 1380 1381 1382

	if (pgio->pg_dreq == NULL)
		rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
	else
		rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);

1383 1384 1385
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
P
Peng Tao 已提交
1386
					   rd_size,
1387 1388
					   IOMODE_READ,
					   GFP_KERNEL);
1389 1390
	/* If no lseg, fall back to read through mds */
	if (pgio->pg_lseg == NULL)
1391
		nfs_pageio_reset_read_mds(pgio);
1392

1393 1394 1395 1396
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);

void
1397 1398
pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
			   struct nfs_page *req, u64 wb_size)
1399
{
1400
	WARN_ON_ONCE(pgio->pg_lseg != NULL);
1401

1402 1403 1404 1405
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_write_mds(pgio);
		return;
	}
1406

1407 1408 1409
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
1410
					   wb_size,
1411 1412
					   IOMODE_RW,
					   GFP_NOFS);
1413 1414
	/* If no lseg, fall back to write through mds */
	if (pgio->pg_lseg == NULL)
1415
		nfs_pageio_reset_write_mds(pgio);
1416 1417 1418
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);

1419
void
1420 1421
pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode,
		      const struct nfs_pgio_completion_ops *compl_ops)
1422 1423 1424 1425 1426
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (ld == NULL)
1427 1428 1429
		nfs_pageio_init_read(pgio, inode, compl_ops);
	else
		nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0);
1430 1431
}

1432
void
1433 1434 1435
pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode,
		       int ioflags,
		       const struct nfs_pgio_completion_ops *compl_ops)
1436 1437 1438 1439 1440
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (ld == NULL)
1441 1442 1443
		nfs_pageio_init_write(pgio, inode, ioflags, compl_ops);
	else
		nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags);
1444 1445
}

1446
bool
1447 1448
pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
		     struct nfs_page *req)
1449
{
1450 1451
	if (pgio->pg_lseg == NULL)
		return nfs_generic_pg_test(pgio, prev, req);
1452

1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468
	/*
	 * 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);
1469
}
1470
EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1471

1472
int pnfs_write_done_resend_to_mds(struct inode *inode,
1473
				struct list_head *head,
1474 1475
				const struct nfs_pgio_completion_ops *compl_ops,
				struct nfs_direct_req *dreq)
1476 1477 1478 1479 1480
{
	struct nfs_pageio_descriptor pgio;
	LIST_HEAD(failed);

	/* Resend all requests through the MDS */
1481
	nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops);
1482
	pgio.pg_dreq = dreq;
1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501
	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;
}
1502
EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1503

1504 1505
static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
{
1506 1507 1508 1509
	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 &
1510
	    PNFS_LAYOUTRET_ON_ERROR) {
1511
		pnfs_return_layout(hdr->inode);
1512
	}
1513 1514
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
1515
							&hdr->pages,
1516 1517
							hdr->completion_ops,
							hdr->dreq);
1518 1519
}

1520 1521 1522
/*
 * Called by non rpc-based layout drivers
 */
1523
void pnfs_ld_write_done(struct nfs_write_data *data)
1524
{
1525 1526 1527
	struct nfs_pgio_header *hdr = data->header;

	if (!hdr->pnfs_error) {
1528
		pnfs_set_layoutcommit(data);
1529
		hdr->mds_ops->rpc_call_done(&data->task, data);
1530 1531
	} else
		pnfs_ld_handle_write_error(data);
1532
	hdr->mds_ops->rpc_release(data);
1533
}
1534
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1535

1536 1537 1538 1539
static void
pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
		struct nfs_write_data *data)
{
1540 1541
	struct nfs_pgio_header *hdr = data->header;

1542 1543 1544 1545 1546
	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;
	}
1547 1548 1549 1550
	nfs_writedata_release(data);
}

static enum pnfs_try_status
1551
pnfs_try_to_write_data(struct nfs_write_data *wdata,
1552 1553 1554
			const struct rpc_call_ops *call_ops,
			struct pnfs_layout_segment *lseg,
			int how)
1555
{
1556 1557
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
1558 1559 1560
	enum pnfs_try_status trypnfs;
	struct nfs_server *nfss = NFS_SERVER(inode);

1561
	hdr->mds_ops = call_ops;
1562 1563 1564 1565

	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);
1566
	if (trypnfs != PNFS_NOT_ATTEMPTED)
1567 1568 1569 1570 1571
		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}

1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582
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;

1583
		data = list_first_entry(head, struct nfs_write_data, list);
1584 1585 1586 1587 1588 1589
		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);
	}
1590
	pnfs_put_lseg(lseg);
1591 1592
}

1593 1594
static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
{
1595
	pnfs_put_lseg(hdr->lseg);
1596 1597
	nfs_writehdr_free(hdr);
}
B
Bryan Schumaker 已提交
1598
EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
1599

1600 1601 1602
int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
{
1603 1604
	struct nfs_write_header *whdr;
	struct nfs_pgio_header *hdr;
1605 1606
	int ret;

1607 1608
	whdr = nfs_writehdr_alloc();
	if (!whdr) {
1609
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1610
		pnfs_put_lseg(desc->pg_lseg);
1611
		desc->pg_lseg = NULL;
1612
		return -ENOMEM;
1613
	}
1614 1615
	hdr = &whdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1616
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1617 1618 1619
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_flush(desc, hdr);
	if (ret != 0) {
1620
		pnfs_put_lseg(desc->pg_lseg);
1621 1622 1623 1624
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
	if (atomic_dec_and_test(&hdr->refcnt))
1625
		hdr->completion_ops->completion(hdr);
1626
	return ret;
1627 1628 1629
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);

1630
int pnfs_read_done_resend_to_mds(struct inode *inode,
1631
				struct list_head *head,
1632 1633
				const struct nfs_pgio_completion_ops *compl_ops,
				struct nfs_direct_req *dreq)
1634 1635
{
	struct nfs_pageio_descriptor pgio;
1636
	LIST_HEAD(failed);
1637

1638
	/* Resend all requests through the MDS */
1639
	nfs_pageio_init_read(&pgio, inode, compl_ops);
1640
	pgio.pg_dreq = dreq;
1641 1642
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);
1643 1644

		nfs_list_remove_request(req);
1645 1646
		if (!nfs_pageio_add_request(&pgio, req))
			nfs_list_add_request(req, &failed);
1647 1648
	}
	nfs_pageio_complete(&pgio);
1649 1650 1651 1652 1653 1654 1655

	if (!list_empty(&failed)) {
		list_move(&failed, head);
		return -EIO;
	}
	return 0;
}
1656
EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1657 1658 1659

static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
{
1660 1661 1662 1663
	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 &
1664
	    PNFS_LAYOUTRET_ON_ERROR) {
1665
		pnfs_return_layout(hdr->inode);
1666
	}
1667 1668
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
1669
							&hdr->pages,
1670 1671
							hdr->completion_ops,
							hdr->dreq);
1672 1673
}

1674 1675 1676
/*
 * Called by non rpc-based layout drivers
 */
1677
void pnfs_ld_read_done(struct nfs_read_data *data)
1678
{
1679 1680 1681
	struct nfs_pgio_header *hdr = data->header;

	if (likely(!hdr->pnfs_error)) {
1682
		__nfs4_read_done_cb(data);
1683
		hdr->mds_ops->rpc_call_done(&data->task, data);
1684 1685
	} else
		pnfs_ld_handle_read_error(data);
1686
	hdr->mds_ops->rpc_release(data);
1687 1688 1689
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

1690 1691 1692 1693
static void
pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
		struct nfs_read_data *data)
{
1694 1695
	struct nfs_pgio_header *hdr = data->header;

1696 1697 1698 1699 1700
	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;
	}
1701 1702 1703
	nfs_readdata_release(data);
}

A
Andy Adamson 已提交
1704 1705 1706
/*
 * Call the appropriate parallel I/O subsystem read function.
 */
1707
static enum pnfs_try_status
A
Andy Adamson 已提交
1708
pnfs_try_to_read_data(struct nfs_read_data *rdata,
1709 1710
		       const struct rpc_call_ops *call_ops,
		       struct pnfs_layout_segment *lseg)
A
Andy Adamson 已提交
1711
{
1712 1713
	struct nfs_pgio_header *hdr = rdata->header;
	struct inode *inode = hdr->inode;
A
Andy Adamson 已提交
1714 1715 1716
	struct nfs_server *nfss = NFS_SERVER(inode);
	enum pnfs_try_status trypnfs;

1717
	hdr->mds_ops = call_ops;
A
Andy Adamson 已提交
1718 1719 1720 1721 1722

	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);
1723
	if (trypnfs != PNFS_NOT_ATTEMPTED)
A
Andy Adamson 已提交
1724 1725 1726 1727
		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}
A
Andy Adamson 已提交
1728

1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739
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;

1740
		data = list_first_entry(head, struct nfs_read_data, list);
1741 1742 1743 1744 1745 1746
		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);
	}
1747
	pnfs_put_lseg(lseg);
1748 1749
}

1750 1751
static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
{
1752
	pnfs_put_lseg(hdr->lseg);
1753 1754
	nfs_readhdr_free(hdr);
}
B
Bryan Schumaker 已提交
1755
EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
1756

1757 1758 1759
int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
{
1760 1761
	struct nfs_read_header *rhdr;
	struct nfs_pgio_header *hdr;
1762 1763
	int ret;

1764 1765
	rhdr = nfs_readhdr_alloc();
	if (!rhdr) {
1766
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1767
		ret = -ENOMEM;
1768
		pnfs_put_lseg(desc->pg_lseg);
1769 1770 1771
		desc->pg_lseg = NULL;
		return ret;
	}
1772 1773
	hdr = &rhdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1774
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1775 1776 1777
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_pagein(desc, hdr);
	if (ret != 0) {
1778
		pnfs_put_lseg(desc->pg_lseg);
1779 1780 1781 1782
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_reads(desc, &hdr->rpc_list);
	if (atomic_dec_and_test(&hdr->refcnt))
1783
		hdr->completion_ops->completion(hdr);
1784
	return ret;
1785 1786 1787
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);

A
Andy Adamson 已提交
1788
/*
1789
 * There can be multiple RW segments.
A
Andy Adamson 已提交
1790
 */
1791
static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
A
Andy Adamson 已提交
1792
{
1793
	struct pnfs_layout_segment *lseg;
A
Andy Adamson 已提交
1794

1795 1796
	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
		if (lseg->pls_range.iomode == IOMODE_RW &&
1797
		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1798 1799
			list_add(&lseg->pls_lc_list, listp);
	}
A
Andy Adamson 已提交
1800 1801
}

1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817
static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
{
	struct pnfs_layout_segment *lseg, *tmp;
	unsigned long *bitlock = &NFS_I(inode)->flags;

	/* Matched by references in pnfs_set_layoutcommit */
	list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
		list_del_init(&lseg->pls_lc_list);
		pnfs_put_lseg(lseg);
	}

	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
	smp_mb__after_clear_bit();
	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
}

P
Peng Tao 已提交
1818 1819
void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
{
1820
	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
P
Peng Tao 已提交
1821 1822 1823
}
EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);

A
Andy Adamson 已提交
1824 1825 1826
void
pnfs_set_layoutcommit(struct nfs_write_data *wdata)
{
1827 1828 1829
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
	struct nfs_inode *nfsi = NFS_I(inode);
1830
	loff_t end_pos = wdata->mds_offset + wdata->res.count;
1831
	bool mark_as_dirty = false;
A
Andy Adamson 已提交
1832

1833
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1834
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1835
		mark_as_dirty = true;
A
Andy Adamson 已提交
1836
		dprintk("%s: Set layoutcommit for inode %lu ",
1837
			__func__, inode->i_ino);
A
Andy Adamson 已提交
1838
	}
1839
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
1840
		/* references matched in nfs4_layoutcommit_release */
1841
		pnfs_get_lseg(hdr->lseg);
1842
	}
1843 1844
	if (end_pos > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = end_pos;
1845
	spin_unlock(&inode->i_lock);
1846
	dprintk("%s: lseg %p end_pos %llu\n",
1847
		__func__, hdr->lseg, nfsi->layout->plh_lwb);
1848 1849 1850 1851

	/* 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)
1852
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1853 1854 1855
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

A
Andy Adamson 已提交
1856 1857 1858 1859 1860 1861
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);
1862
	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
A
Andy Adamson 已提交
1863 1864
}

1865 1866 1867 1868 1869 1870 1871 1872
/*
 * 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
1873
int
1874
pnfs_layoutcommit_inode(struct inode *inode, bool sync)
A
Andy Adamson 已提交
1875 1876 1877 1878 1879 1880 1881 1882
{
	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);

1883 1884 1885
	if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		return 0;

A
Andy Adamson 已提交
1886 1887
	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
	data = kzalloc(sizeof(*data), GFP_NOFS);
1888 1889 1890 1891
	if (!data) {
		status = -ENOMEM;
		goto out;
	}
A
Andy Adamson 已提交
1892

P
Peng Tao 已提交
1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906
	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;
	}

1907
	INIT_LIST_HEAD(&data->lseg_list);
1908
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1909
	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
P
Peng Tao 已提交
1910
		clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
A
Andy Adamson 已提交
1911
		spin_unlock(&inode->i_lock);
P
Peng Tao 已提交
1912 1913
		wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
		goto out_free;
A
Andy Adamson 已提交
1914
	}
1915 1916

	pnfs_list_write_lseg(inode, &data->lseg_list);
A
Andy Adamson 已提交
1917

1918 1919
	end_pos = nfsi->layout->plh_lwb;
	nfsi->layout->plh_lwb = 0;
A
Andy Adamson 已提交
1920

1921
	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
A
Andy Adamson 已提交
1922 1923 1924
	spin_unlock(&inode->i_lock);

	data->args.inode = inode;
1925
	data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
A
Andy Adamson 已提交
1926 1927 1928 1929 1930 1931 1932 1933
	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
1934 1935
	if (status)
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1936 1937
	dprintk("<-- %s status %d\n", __func__, status);
	return status;
P
Peng Tao 已提交
1938 1939 1940
out_free:
	kfree(data);
	goto out;
A
Andy Adamson 已提交
1941
}
1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953

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;
}