pnfs.c 51.1 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"
36
#include "nfs4trace.h"
R
Ricardo Labiaga 已提交
37 38

#define NFSDBG_FACILITY		NFSDBG_PNFS
39
#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
R
Ricardo Labiaga 已提交
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 67
/* 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
68 69 70
static struct pnfs_layoutdriver_type *
find_pnfs_driver(u32 id)
{
71 72 73 74
	struct pnfs_layoutdriver_type *local;

	spin_lock(&pnfs_spinlock);
	local = find_pnfs_driver_locked(id);
75 76 77 78
	if (local != NULL && !try_module_get(local->owner)) {
		dprintk("%s: Could not grab reference on module\n", __func__);
		local = NULL;
	}
79 80
	spin_unlock(&pnfs_spinlock);
	return local;
R
Ricardo Labiaga 已提交
81 82 83 84 85
}

void
unset_pnfs_layoutdriver(struct nfs_server *nfss)
{
86 87 88
	if (nfss->pnfs_curr_ld) {
		if (nfss->pnfs_curr_ld->clear_layoutdriver)
			nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
89 90 91
		/* 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);
92
		module_put(nfss->pnfs_curr_ld->owner);
93
	}
R
Ricardo Labiaga 已提交
94 95 96 97 98 99 100 101 102 103
	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
104 105
set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
		      u32 id)
R
Ricardo Labiaga 已提交
106 107 108 109 110 111 112
{
	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))) {
113 114
		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
115 116 117 118 119 120 121 122 123 124 125 126 127
		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;
128 129
	if (ld_type->set_layoutdriver
	    && ld_type->set_layoutdriver(server, mntfh)) {
130 131
		printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
			"driver %u.\n", __func__, id);
132 133 134
		module_put(ld_type->owner);
		goto out_no_driver;
	}
135 136
	/* Bump the MDS count */
	atomic_inc(&server->nfs_client->cl_mds_count);
137

R
Ricardo Labiaga 已提交
138 139 140 141 142 143 144
	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;
}
145 146 147 148 149 150 151 152

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

	if (ld_type->id == 0) {
153
		printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
154 155
		return status;
	}
156
	if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
157
		printk(KERN_ERR "NFS: %s Layout driver must provide "
158 159 160
		       "alloc_lseg and free_lseg.\n", __func__);
		return status;
	}
161 162 163 164 165 166 167 168 169

	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 {
170
		printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187
			__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);
188

189 190 191 192
/*
 * pNFS client layout cache
 */

193
/* Need to hold i_lock if caller does not already hold reference */
F
Fred Isaman 已提交
194
void
195
pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
196
{
197
	atomic_inc(&lo->plh_refcount);
198 199
}

200 201 202 203
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;
204
	return ld->alloc_layout_hdr(ino, gfp_flags);
205 206 207 208 209
}

static void
pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
{
210 211 212 213 214 215 216 217 218 219
	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);
	}
220
	put_rpccred(lo->plh_lc_cred);
221
	return ld->free_layout_hdr(lo);
222 223
}

224
static void
225
pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
226
{
227
	struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
228
	dprintk("%s: freeing layout cache %p\n", __func__, lo);
229 230 231 232
	nfsi->layout = NULL;
	/* Reset MDS Threshold I/O counters */
	nfsi->write_io = 0;
	nfsi->read_io = 0;
233 234
}

235
void
236
pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
237
{
238 239 240
	struct inode *inode = lo->plh_inode;

	if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
241
		pnfs_detach_layout_hdr(lo);
242
		spin_unlock(&inode->i_lock);
243
		pnfs_free_layout_hdr(lo);
244
	}
245 246
}

247 248 249 250 251 252 253 254
static int
pnfs_iomode_to_fail_bit(u32 iomode)
{
	return iomode == IOMODE_RW ?
		NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
}

static void
255
pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
256
{
257
	lo->plh_retry_timestamp = jiffies;
258
	if (!test_and_set_bit(fail_bit, &lo->plh_flags))
259 260 261 262 263 264 265 266 267 268 269 270 271 272
		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;
273 274 275 276 277 278
	struct pnfs_layout_range range = {
		.iomode = iomode,
		.offset = 0,
		.length = NFS4_MAX_UINT64,
	};
	LIST_HEAD(head);
279 280 281

	spin_lock(&inode->i_lock);
	pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
282
	pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
283
	spin_unlock(&inode->i_lock);
284
	pnfs_free_lseg_list(&head);
285 286 287 288 289 290 291
	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)
{
292
	unsigned long start, end;
293 294 295
	int fail_bit = pnfs_iomode_to_fail_bit(iomode);

	if (test_bit(fail_bit, &lo->plh_flags) == 0)
296 297 298 299 300
		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 */
301
		pnfs_layout_clear_fail_bit(lo, fail_bit);
302 303 304
		return false;
	}
	return true;
305 306
}

307 308 309
static void
init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
{
310
	INIT_LIST_HEAD(&lseg->pls_list);
311
	INIT_LIST_HEAD(&lseg->pls_lc_list);
312 313 314
	atomic_set(&lseg->pls_refcount, 1);
	smp_mb();
	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
315
	lseg->pls_layout = lo;
316 317
}

318
static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
319
{
320
	struct inode *ino = lseg->pls_layout->plh_inode;
321

322
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
323 324
}

F
Fred Isaman 已提交
325
static void
326 327
pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_segment *lseg)
F
Fred Isaman 已提交
328
{
329
	struct inode *inode = lo->plh_inode;
F
Fred Isaman 已提交
330

331
	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
F
Fred Isaman 已提交
332
	list_del_init(&lseg->pls_list);
333 334
	/* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
	atomic_dec(&lo->plh_refcount);
335 336
	if (list_empty(&lo->plh_segs))
		clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
F
Fred Isaman 已提交
337 338 339
	rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
}

340
void
341
pnfs_put_lseg(struct pnfs_layout_segment *lseg)
342
{
343
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
344 345 346 347 348
	struct inode *inode;

	if (!lseg)
		return;

349 350 351
	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
		atomic_read(&lseg->pls_refcount),
		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
352 353
	lo = lseg->pls_layout;
	inode = lo->plh_inode;
F
Fred Isaman 已提交
354
	if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
355
		pnfs_get_layout_hdr(lo);
356
		pnfs_layout_remove_lseg(lo, lseg);
F
Fred Isaman 已提交
357
		spin_unlock(&inode->i_lock);
358
		pnfs_free_lseg(lseg);
359
		pnfs_put_layout_hdr(lo);
360 361
	}
}
362
EXPORT_SYMBOL_GPL(pnfs_put_lseg);
363

364
static u64
365 366 367 368 369 370 371 372 373 374 375 376 377 378 379
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
 *           [----------------)
 */
380
static bool
381
pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
382
		 const struct pnfs_layout_range *l2)
383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398
{
	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
 *                              [----------------)
 */
399
static bool
400
pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
401
		    const struct pnfs_layout_range *l2)
402 403 404 405 406 407 408 409 410 411
{
	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);
}

412
static bool
413 414
should_free_lseg(const struct pnfs_layout_range *lseg_range,
		 const struct pnfs_layout_range *recall_range)
415
{
416 417
	return (recall_range->iomode == IOMODE_ANY ||
		lseg_range->iomode == recall_range->iomode) &&
418
	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
419 420
}

421 422 423 424 425 426 427 428 429 430
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;
}

431 432 433 434 435 436 437 438 439 440 441
/* 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
442 443
		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
			atomic_read(&lseg->pls_refcount));
444
		if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
F
Fred Isaman 已提交
445
			rv = 1;
446 447 448 449 450 451 452
	}
	return rv;
}

/* Returns count of number of matching invalid lsegs remaining in list
 * after call.
 */
F
Fred Isaman 已提交
453
int
454
pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
455
			    struct list_head *tmp_list,
456
			    struct pnfs_layout_range *recall_range)
457 458
{
	struct pnfs_layout_segment *lseg, *next;
459
	int invalid = 0, removed = 0;
460 461 462

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

463
	if (list_empty(&lo->plh_segs))
464
		return 0;
465
	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
466 467
		if (!recall_range ||
		    should_free_lseg(&lseg->pls_range, recall_range)) {
468 469 470 471 472 473 474 475 476
			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;
477 478
}

479
/* note free_me must contain lsegs from a single layout_hdr */
F
Fred Isaman 已提交
480
void
481
pnfs_free_lseg_list(struct list_head *free_me)
482
{
483
	struct pnfs_layout_segment *lseg, *tmp;
484 485 486 487

	if (list_empty(free_me))
		return;

488
	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
489
		list_del(&lseg->pls_list);
490
		pnfs_free_lseg(lseg);
491 492 493
	}
}

494 495 496 497
void
pnfs_destroy_layout(struct nfs_inode *nfsi)
{
	struct pnfs_layout_hdr *lo;
498
	LIST_HEAD(tmp_list);
499 500 501 502

	spin_lock(&nfsi->vfs_inode.i_lock);
	lo = nfsi->layout;
	if (lo) {
503
		lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
504
		pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
505 506 507 508 509 510 511 512
		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);
513
}
514
EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
515

516 517 518
static bool
pnfs_layout_add_bulk_destroy_list(struct inode *inode,
		struct list_head *layout_list)
519 520
{
	struct pnfs_layout_hdr *lo;
521
	bool ret = false;
522

523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601
	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);
602

603
	spin_lock(&clp->cl_lock);
604
	rcu_read_lock();
605
restart:
606
	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
607 608 609 610 611 612
		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
			continue;
		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
				server,
				&layout_list) != 0)
			goto restart;
613 614
	}
	rcu_read_unlock();
615 616
	spin_unlock(&clp->cl_lock);

617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636
	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;
637
	}
638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656
	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);
657 658
}

659 660 661 662 663 664
/*
 * 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)
{
665
	return (s32)(s1 - s2) > 0;
666 667
}

668 669 670 671 672 673 674 675 676 677 678
static void
pnfs_verify_layout_stateid(struct pnfs_layout_hdr *lo,
		const nfs4_stateid *new,
		struct list_head *free_me_list)
{
	if (nfs4_stateid_match_other(&lo->plh_stateid, new))
		return;
	/* Layout is new! Kill existing layout segments */
	pnfs_mark_matching_lsegs_invalid(lo, free_me_list, NULL);
}

679
/* update lo->plh_stateid with new if is more recent */
F
Fred Isaman 已提交
680 681 682
void
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
			bool update_barrier)
683
{
684 685
	u32 oldseq, newseq, new_barrier;
	int empty = list_empty(&lo->plh_segs);
686

687 688
	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
	newseq = be32_to_cpu(new->seqid);
689
	if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
690
		nfs4_stateid_copy(&lo->plh_stateid, new);
F
Fred Isaman 已提交
691
		if (update_barrier) {
692
			new_barrier = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
693 694
		} else {
			/* Because of wraparound, we want to keep the barrier
695
			 * "close" to the current seqids.
F
Fred Isaman 已提交
696
			 */
697
			new_barrier = newseq - atomic_read(&lo->plh_outstanding);
F
Fred Isaman 已提交
698
		}
699 700
		if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
			lo->plh_barrier = new_barrier;
F
Fred Isaman 已提交
701
	}
702 703
}

704
static bool
705 706
pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
		const nfs4_stateid *stateid)
F
Fred Isaman 已提交
707
{
708
	u32 seqid = be32_to_cpu(stateid->seqid);
709

710 711 712 713 714 715 716
	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
717 718
	return lo->plh_block_lgets ||
		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
F
Fred Isaman 已提交
719
		(list_empty(&lo->plh_segs) &&
720 721 722
		 (atomic_read(&lo->plh_outstanding) > lget));
}

723 724 725
int
pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
			      struct nfs4_state *open_state)
726
{
727
	int status = 0;
728

729
	dprintk("--> %s\n", __func__);
730
	spin_lock(&lo->plh_inode->i_lock);
731
	if (pnfs_layoutgets_blocked(lo, 1)) {
732
		status = -EAGAIN;
733 734
	} else if (!nfs4_valid_open_stateid(open_state)) {
		status = -EBADF;
735
	} else if (list_empty(&lo->plh_segs)) {
736 737 738 739
		int seq;

		do {
			seq = read_seqbegin(&open_state->seqlock);
740
			nfs4_stateid_copy(dst, &open_state->stateid);
741 742
		} while (read_seqretry(&open_state->seqlock, seq));
	} else
743
		nfs4_stateid_copy(dst, &lo->plh_stateid);
744
	spin_unlock(&lo->plh_inode->i_lock);
745
	dprintk("<-- %s\n", __func__);
746
	return status;
747 748 749 750 751 752 753 754
}

/*
* Get layout from server.
*    for now, assume that whole file layouts are requested.
*    arg->offset: 0
*    arg->length: all ones
*/
755 756 757
static struct pnfs_layout_segment *
send_layoutget(struct pnfs_layout_hdr *lo,
	   struct nfs_open_context *ctx,
758
	   struct pnfs_layout_range *range,
759
	   gfp_t gfp_flags)
760
{
761
	struct inode *ino = lo->plh_inode;
762 763
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs4_layoutget *lgp;
764
	struct pnfs_layout_segment *lseg;
765 766

	dprintk("--> %s\n", __func__);
767

768
	lgp = kzalloc(sizeof(*lgp), gfp_flags);
769
	if (lgp == NULL)
770
		return NULL;
771

772 773 774
	lgp->args.minlength = PAGE_CACHE_SIZE;
	if (lgp->args.minlength > range->length)
		lgp->args.minlength = range->length;
775
	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
776
	lgp->args.range = *range;
777 778 779
	lgp->args.type = server->pnfs_curr_ld->id;
	lgp->args.inode = ino;
	lgp->args.ctx = get_nfs_open_context(ctx);
780
	lgp->gfp_flags = gfp_flags;
781
	lgp->cred = lo->plh_lc_cred;
782 783 784 785

	/* Synchronously retrieve layout information from server and
	 * store in lseg.
	 */
786 787 788 789 790 791 792 793
	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 */
794
			pnfs_layout_io_set_failed(lo, range->iomode);
795 796
		}
		return NULL;
797
	}
798

799 800 801
	return lseg;
}

802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
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);
	}
}

817 818 819 820 821 822 823 824
/*
 * 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
825 826 827 828 829 830 831 832
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;
833
	int status = 0, empty;
B
Benny Halevy 已提交
834

835
	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
B
Benny Halevy 已提交
836 837 838

	spin_lock(&ino->i_lock);
	lo = nfsi->layout;
839
	if (!lo) {
B
Benny Halevy 已提交
840
		spin_unlock(&ino->i_lock);
841 842
		dprintk("NFS: %s no layout to return\n", __func__);
		goto out;
B
Benny Halevy 已提交
843 844 845
	}
	stateid = nfsi->layout->plh_stateid;
	/* Reference matched in nfs4_layoutreturn_release */
846
	pnfs_get_layout_hdr(lo);
847
	empty = list_empty(&lo->plh_segs);
848
	pnfs_clear_layoutcommit(ino, &tmp_list);
849
	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
850 851 852
	/* Don't send a LAYOUTRETURN if list was initially empty */
	if (empty) {
		spin_unlock(&ino->i_lock);
853
		pnfs_put_layout_hdr(lo);
854 855 856
		dprintk("NFS: %s no layout segments to return\n", __func__);
		goto out;
	}
857
	lo->plh_block_lgets++;
B
Benny Halevy 已提交
858 859 860 861 862 863
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);

	lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
	if (unlikely(lrp == NULL)) {
		status = -ENOMEM;
864 865 866
		spin_lock(&ino->i_lock);
		lo->plh_block_lgets--;
		spin_unlock(&ino->i_lock);
867
		pnfs_put_layout_hdr(lo);
B
Benny Halevy 已提交
868 869 870 871 872 873
		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
874
	lrp->args.layout = lo;
B
Benny Halevy 已提交
875
	lrp->clp = NFS_SERVER(ino)->nfs_client;
876
	lrp->cred = lo->plh_lc_cred;
B
Benny Halevy 已提交
877 878 879 880 881 882

	status = nfs4_proc_layoutreturn(lrp);
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}
883
EXPORT_SYMBOL_GPL(_pnfs_return_layout);
B
Benny Halevy 已提交
884

885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
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
912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931
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++;
932
	pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
F
Fred Isaman 已提交
933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948
	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--;
949 950 951 952 953 954
	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
955 956 957 958 959 960 961 962
}

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;
963
	if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
F
Fred Isaman 已提交
964 965 966 967
		lo->plh_barrier = barrier;
	spin_unlock(&ino->i_lock);
}

968
bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
F
Fred Isaman 已提交
969 970
{
	struct nfs_inode *nfsi = NFS_I(ino);
971
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
972
	struct pnfs_layout_segment *lseg;
973
	u32 current_seqid;
F
Fred Isaman 已提交
974 975 976 977 978
	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)) {
979
			rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
F
Fred Isaman 已提交
980
			found = true;
981
			goto out;
F
Fred Isaman 已提交
982
		}
983 984
	lo = nfsi->layout;
	current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
F
Fred Isaman 已提交
985

986 987 988 989 990
	/* 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
991 992 993 994
	spin_unlock(&ino->i_lock);
	return found;
}

995 996 997 998 999 1000
/*
 * 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
1001
pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1002
	   const struct pnfs_layout_range *l2)
1003
{
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015
	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;

1016
	/* read > read/write */
1017
	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1018 1019
}

1020
static void
1021
pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1022 1023
		   struct pnfs_layout_segment *lseg)
{
1024 1025
	struct pnfs_layout_segment *lp;

1026 1027
	dprintk("%s:Begin\n", __func__);

1028
	list_for_each_entry(lp, &lo->plh_segs, pls_list) {
1029
		if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0)
1030
			continue;
1031
		list_add_tail(&lseg->pls_list, &lp->pls_list);
1032 1033 1034
		dprintk("%s: inserted lseg %p "
			"iomode %d offset %llu length %llu before "
			"lp %p iomode %d offset %llu length %llu\n",
1035 1036 1037 1038
			__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);
1039
		goto out;
1040
	}
1041 1042 1043 1044 1045 1046
	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:
1047
	pnfs_get_layout_hdr(lo);
1048 1049

	dprintk("%s:Return\n", __func__);
1050 1051 1052
}

static struct pnfs_layout_hdr *
1053 1054 1055
alloc_init_layout_hdr(struct inode *ino,
		      struct nfs_open_context *ctx,
		      gfp_t gfp_flags)
1056 1057 1058
{
	struct pnfs_layout_hdr *lo;

1059
	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1060 1061
	if (!lo)
		return NULL;
1062
	atomic_set(&lo->plh_refcount, 1);
1063 1064
	INIT_LIST_HEAD(&lo->plh_layouts);
	INIT_LIST_HEAD(&lo->plh_segs);
1065
	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1066
	lo->plh_inode = ino;
1067
	lo->plh_lc_cred = get_rpccred(ctx->cred);
1068 1069 1070 1071
	return lo;
}

static struct pnfs_layout_hdr *
1072 1073 1074
pnfs_find_alloc_layout(struct inode *ino,
		       struct nfs_open_context *ctx,
		       gfp_t gfp_flags)
1075 1076 1077 1078 1079 1080
{
	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);

1081 1082
	if (nfsi->layout != NULL)
		goto out_existing;
1083
	spin_unlock(&ino->i_lock);
1084
	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1085 1086
	spin_lock(&ino->i_lock);

1087
	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1088
		nfsi->layout = new;
1089
		return new;
1090 1091
	} else if (new != NULL)
		pnfs_free_layout_hdr(new);
1092 1093
out_existing:
	pnfs_get_layout_hdr(nfsi->layout);
1094 1095 1096
	return nfsi->layout;
}

1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107
/*
 * iomode matching rules:
 * iomode	lseg	match
 * -----	-----	-----
 * ANY		READ	true
 * ANY		RW	true
 * RW		READ	false
 * RW		RW	true
 * READ		READ	true
 * READ		RW	true
 */
1108
static bool
1109
pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1110
		 const struct pnfs_layout_range *range)
1111
{
1112 1113 1114 1115
	struct pnfs_layout_range range1;

	if ((range->iomode == IOMODE_RW &&
	     ls_range->iomode != IOMODE_RW) ||
1116
	    !pnfs_lseg_range_intersecting(ls_range, range))
1117 1118 1119 1120 1121
		return 0;

	/* range1 covers only the first byte in the range */
	range1 = *range;
	range1.length = 1;
1122
	return pnfs_lseg_range_contained(ls_range, &range1);
1123 1124 1125 1126 1127
}

/*
 * lookup range in layout
 */
1128
static struct pnfs_layout_segment *
1129 1130
pnfs_find_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_range *range)
1131
{
1132 1133 1134 1135
	struct pnfs_layout_segment *lseg, *ret = NULL;

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

1136
	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1137
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1138
		    pnfs_lseg_range_match(&lseg->pls_range, range)) {
1139
			ret = pnfs_get_lseg(lseg);
1140 1141
			break;
		}
1142
		if (lseg->pls_range.offset > range->offset)
1143 1144 1145 1146
			break;
	}

	dprintk("%s:Return lseg %p ref %d\n",
1147
		__func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1148
	return ret;
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 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225
/*
 * 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;
}

1226 1227 1228 1229
/*
 * Layout segment is retreived from the server if not cached.
 * The appropriate layout segment is referenced and returned to the caller.
 */
1230
struct pnfs_layout_segment *
1231 1232
pnfs_update_layout(struct inode *ino,
		   struct nfs_open_context *ctx,
1233 1234
		   loff_t pos,
		   u64 count,
1235 1236
		   enum pnfs_iomode iomode,
		   gfp_t gfp_flags)
1237
{
1238 1239 1240 1241 1242
	struct pnfs_layout_range arg = {
		.iomode = iomode,
		.offset = pos,
		.length = count,
	};
1243
	unsigned pg_offset;
1244 1245
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs_client *clp = server->nfs_client;
1246 1247
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg = NULL;
1248
	bool first;
1249 1250

	if (!pnfs_enabled_sb(NFS_SERVER(ino)))
1251
		goto out;
1252 1253

	if (pnfs_within_mdsthreshold(ctx, ino, iomode))
1254
		goto out;
1255

1256
	spin_lock(&ino->i_lock);
1257
	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1258 1259 1260 1261
	if (lo == NULL) {
		spin_unlock(&ino->i_lock);
		goto out;
	}
1262

F
Fred Isaman 已提交
1263
	/* Do we even need to bother with this? */
1264
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1265
		dprintk("%s matches recall, use MDS\n", __func__);
1266 1267 1268 1269
		goto out_unlock;
	}

	/* if LAYOUTGET already failed once we don't try again */
1270
	if (pnfs_layout_io_test_failed(lo, iomode))
1271 1272
		goto out_unlock;

1273
	/* Check to see if the layout for the given range already exists */
1274
	lseg = pnfs_find_lseg(lo, &arg);
1275 1276 1277
	if (lseg)
		goto out_unlock;

1278
	if (pnfs_layoutgets_blocked(lo, 0))
1279 1280 1281
		goto out_unlock;
	atomic_inc(&lo->plh_outstanding);

1282
	first = list_empty(&lo->plh_layouts) ? true : false;
1283
	spin_unlock(&ino->i_lock);
1284

1285
	if (first) {
1286 1287 1288 1289
		/* 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);
1290
		list_add_tail(&lo->plh_layouts, &server->layouts);
1291 1292
		spin_unlock(&clp->cl_lock);
	}
1293

1294 1295 1296 1297 1298
	pg_offset = arg.offset & ~PAGE_CACHE_MASK;
	if (pg_offset) {
		arg.offset -= pg_offset;
		arg.length += pg_offset;
	}
1299 1300
	if (arg.length != NFS4_MAX_UINT64)
		arg.length = PAGE_CACHE_ALIGN(arg.length);
1301

1302
	lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1303
	atomic_dec(&lo->plh_outstanding);
1304
out_put_layout_hdr:
1305
	pnfs_put_layout_hdr(lo);
1306
out:
1307 1308 1309 1310 1311 1312 1313 1314
	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);
1315 1316 1317
	return lseg;
out_unlock:
	spin_unlock(&ino->i_lock);
1318
	goto out_put_layout_hdr;
1319
}
1320
EXPORT_SYMBOL_GPL(pnfs_update_layout);
1321

1322
struct pnfs_layout_segment *
1323 1324 1325 1326 1327
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;
1328
	struct inode *ino = lo->plh_inode;
1329
	LIST_HEAD(free_me);
1330 1331 1332
	int status = 0;

	/* Inject layout blob into I/O device driver */
1333
	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344
	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);
1345
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1346 1347 1348 1349
		dprintk("%s forget reply due to recall\n", __func__);
		goto out_forget_reply;
	}

1350 1351
	if (pnfs_layoutgets_blocked(lo, 1) ||
	    pnfs_layout_stateid_blocked(lo, &res->stateid)) {
F
Fred Isaman 已提交
1352 1353 1354
		dprintk("%s forget reply due to state\n", __func__);
		goto out_forget_reply;
	}
1355

1356 1357
	/* Check that the new stateid matches the old stateid */
	pnfs_verify_layout_stateid(lo, &res->stateid, &free_me);
1358 1359 1360
	/* Done processing layoutget. Set the layout stateid */
	pnfs_set_layout_stateid(lo, &res->stateid, false);

1361
	init_lseg(lo, lseg);
1362
	lseg->pls_range = res->range;
1363
	pnfs_get_lseg(lseg);
1364
	pnfs_layout_insert_lseg(lo, lseg);
1365

F
Fred Isaman 已提交
1366 1367 1368 1369 1370
	if (res->return_on_close) {
		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
		set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
	}

1371
	spin_unlock(&ino->i_lock);
1372
	pnfs_free_lseg_list(&free_me);
1373
	return lseg;
1374
out:
1375
	return ERR_PTR(status);
F
Fred Isaman 已提交
1376 1377 1378 1379 1380 1381

out_forget_reply:
	spin_unlock(&ino->i_lock);
	lseg->pls_layout = lo;
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
	goto out;
1382 1383
}

1384 1385 1386
void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
P
Peng Tao 已提交
1387 1388
	u64 rd_size = req->wb_bytes;

1389
	WARN_ON_ONCE(pgio->pg_lseg != NULL);
1390

1391 1392 1393 1394
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_read_mds(pgio);
		return;
	}
P
Peng Tao 已提交
1395 1396 1397 1398 1399 1400

	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);

1401 1402 1403
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
P
Peng Tao 已提交
1404
					   rd_size,
1405 1406
					   IOMODE_READ,
					   GFP_KERNEL);
1407 1408
	/* If no lseg, fall back to read through mds */
	if (pgio->pg_lseg == NULL)
1409
		nfs_pageio_reset_read_mds(pgio);
1410

1411 1412 1413 1414
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);

void
1415 1416
pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
			   struct nfs_page *req, u64 wb_size)
1417
{
1418
	WARN_ON_ONCE(pgio->pg_lseg != NULL);
1419

1420 1421 1422 1423
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_write_mds(pgio);
		return;
	}
1424

1425 1426 1427
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
1428
					   wb_size,
1429 1430
					   IOMODE_RW,
					   GFP_NOFS);
1431 1432
	/* If no lseg, fall back to write through mds */
	if (pgio->pg_lseg == NULL)
1433
		nfs_pageio_reset_write_mds(pgio);
1434 1435 1436
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);

1437 1438 1439 1440 1441
/*
 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
 * of bytes (maximum @req->wb_bytes) that can be coalesced.
 */
size_t
1442 1443
pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
		     struct nfs_page *req)
1444
{
1445 1446
	if (pgio->pg_lseg == NULL)
		return nfs_generic_pg_test(pgio, prev, req);
1447

1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461
	/*
	 * 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?
	 *
	 */
1462 1463 1464 1465
	if (req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
					 pgio->pg_lseg->pls_range.length))
		return req->wb_bytes;
	return 0;
1466
}
1467
EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1468

1469
int pnfs_write_done_resend_to_mds(struct inode *inode,
1470
				struct list_head *head,
1471 1472
				const struct nfs_pgio_completion_ops *compl_ops,
				struct nfs_direct_req *dreq)
1473 1474 1475 1476 1477
{
	struct nfs_pageio_descriptor pgio;
	LIST_HEAD(failed);

	/* Resend all requests through the MDS */
1478
	nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, true, compl_ops);
1479
	pgio.pg_dreq = dreq;
1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498
	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;
}
1499
EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1500

1501
static void pnfs_ld_handle_write_error(struct nfs_pgio_data *data)
1502
{
1503 1504 1505 1506
	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 &
1507
	    PNFS_LAYOUTRET_ON_ERROR) {
1508
		pnfs_return_layout(hdr->inode);
1509
	}
1510 1511
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
1512
							&hdr->pages,
1513 1514
							hdr->completion_ops,
							hdr->dreq);
1515 1516
}

1517 1518 1519
/*
 * Called by non rpc-based layout drivers
 */
1520
void pnfs_ld_write_done(struct nfs_pgio_data *data)
1521
{
1522 1523
	struct nfs_pgio_header *hdr = data->header;

1524
	trace_nfs4_pnfs_write(data, hdr->pnfs_error);
1525
	if (!hdr->pnfs_error) {
1526
		pnfs_set_layoutcommit(data);
1527
		hdr->mds_ops->rpc_call_done(&data->task, data);
1528 1529
	} else
		pnfs_ld_handle_write_error(data);
1530
	hdr->mds_ops->rpc_release(data);
1531
}
1532
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1533

1534 1535
static void
pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1536
		struct nfs_pgio_data *data)
1537
{
1538 1539
	struct nfs_pgio_header *hdr = data->header;

1540 1541 1542 1543 1544
	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;
	}
1545
	nfs_pgio_data_release(data);
1546 1547 1548
}

static enum pnfs_try_status
1549
pnfs_try_to_write_data(struct nfs_pgio_data *wdata,
1550 1551 1552
			const struct rpc_call_ops *call_ops,
			struct pnfs_layout_segment *lseg,
			int how)
1553
{
1554 1555
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
1556 1557 1558
	enum pnfs_try_status trypnfs;
	struct nfs_server *nfss = NFS_SERVER(inode);

1559
	hdr->mds_ops = call_ops;
1560 1561 1562 1563

	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);
1564
	if (trypnfs != PNFS_NOT_ATTEMPTED)
1565 1566 1567 1568 1569
		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}

1570 1571 1572
static void
pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
{
1573
	struct nfs_pgio_data *data;
1574 1575 1576 1577 1578 1579 1580
	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;

1581
		data = list_first_entry(head, struct nfs_pgio_data, list);
1582 1583 1584 1585 1586 1587
		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);
	}
1588
	pnfs_put_lseg(lseg);
1589 1590
}

1591 1592
static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
{
1593
	pnfs_put_lseg(hdr->lseg);
1594
	nfs_rw_header_free(hdr);
1595
}
B
Bryan Schumaker 已提交
1596
EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
1597

1598 1599 1600
int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
{
1601
	struct nfs_rw_header *whdr;
1602
	struct nfs_pgio_header *hdr;
1603 1604
	int ret;

1605
	whdr = nfs_rw_header_alloc(desc->pg_rw_ops);
1606
	if (!whdr) {
1607
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1608
		pnfs_put_lseg(desc->pg_lseg);
1609
		desc->pg_lseg = NULL;
1610
		return -ENOMEM;
1611
	}
1612 1613
	hdr = &whdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1614
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1615
	atomic_inc(&hdr->refcnt);
1616
	ret = nfs_generic_pgio(desc, hdr);
1617
	if (ret != 0) {
1618
		pnfs_put_lseg(desc->pg_lseg);
1619 1620 1621 1622
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
	if (atomic_dec_and_test(&hdr->refcnt))
1623
		hdr->completion_ops->completion(hdr);
1624
	return ret;
1625 1626 1627
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);

1628
int pnfs_read_done_resend_to_mds(struct inode *inode,
1629
				struct list_head *head,
1630 1631
				const struct nfs_pgio_completion_ops *compl_ops,
				struct nfs_direct_req *dreq)
1632 1633
{
	struct nfs_pageio_descriptor pgio;
1634
	LIST_HEAD(failed);
1635

1636
	/* Resend all requests through the MDS */
1637
	nfs_pageio_init_read(&pgio, inode, true, compl_ops);
1638
	pgio.pg_dreq = dreq;
1639 1640
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);
1641 1642

		nfs_list_remove_request(req);
1643 1644
		if (!nfs_pageio_add_request(&pgio, req))
			nfs_list_add_request(req, &failed);
1645 1646
	}
	nfs_pageio_complete(&pgio);
1647 1648 1649 1650 1651 1652 1653

	if (!list_empty(&failed)) {
		list_move(&failed, head);
		return -EIO;
	}
	return 0;
}
1654
EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1655

1656
static void pnfs_ld_handle_read_error(struct nfs_pgio_data *data)
1657
{
1658 1659 1660 1661
	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 &
1662
	    PNFS_LAYOUTRET_ON_ERROR) {
1663
		pnfs_return_layout(hdr->inode);
1664
	}
1665 1666
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
1667
							&hdr->pages,
1668 1669
							hdr->completion_ops,
							hdr->dreq);
1670 1671
}

1672 1673 1674
/*
 * Called by non rpc-based layout drivers
 */
1675
void pnfs_ld_read_done(struct nfs_pgio_data *data)
1676
{
1677 1678
	struct nfs_pgio_header *hdr = data->header;

1679
	trace_nfs4_pnfs_read(data, hdr->pnfs_error);
1680
	if (likely(!hdr->pnfs_error)) {
1681
		__nfs4_read_done_cb(data);
1682
		hdr->mds_ops->rpc_call_done(&data->task, data);
1683 1684
	} else
		pnfs_ld_handle_read_error(data);
1685
	hdr->mds_ops->rpc_release(data);
1686 1687 1688
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

1689 1690
static void
pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1691
		struct nfs_pgio_data *data)
1692
{
1693 1694
	struct nfs_pgio_header *hdr = data->header;

1695 1696 1697 1698 1699
	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;
	}
1700
	nfs_pgio_data_release(data);
1701 1702
}

A
Andy Adamson 已提交
1703 1704 1705
/*
 * Call the appropriate parallel I/O subsystem read function.
 */
1706
static enum pnfs_try_status
1707
pnfs_try_to_read_data(struct nfs_pgio_data *rdata,
1708 1709
		       const struct rpc_call_ops *call_ops,
		       struct pnfs_layout_segment *lseg)
A
Andy Adamson 已提交
1710
{
1711 1712
	struct nfs_pgio_header *hdr = rdata->header;
	struct inode *inode = hdr->inode;
A
Andy Adamson 已提交
1713 1714 1715
	struct nfs_server *nfss = NFS_SERVER(inode);
	enum pnfs_try_status trypnfs;

1716
	hdr->mds_ops = call_ops;
A
Andy Adamson 已提交
1717 1718 1719 1720 1721

	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);
1722
	if (trypnfs != PNFS_NOT_ATTEMPTED)
A
Andy Adamson 已提交
1723 1724 1725 1726
		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}
A
Andy Adamson 已提交
1727

1728 1729 1730
static void
pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
{
1731
	struct nfs_pgio_data *data;
1732 1733 1734 1735 1736 1737 1738
	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;

1739
		data = list_first_entry(head, struct nfs_pgio_data, list);
1740 1741 1742 1743 1744 1745
		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);
	}
1746
	pnfs_put_lseg(lseg);
1747 1748
}

1749 1750
static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
{
1751
	pnfs_put_lseg(hdr->lseg);
1752
	nfs_rw_header_free(hdr);
1753
}
B
Bryan Schumaker 已提交
1754
EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
1755

1756 1757 1758
int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
{
1759
	struct nfs_rw_header *rhdr;
1760
	struct nfs_pgio_header *hdr;
1761 1762
	int ret;

1763
	rhdr = nfs_rw_header_alloc(desc->pg_rw_ops);
1764
	if (!rhdr) {
1765
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1766
		ret = -ENOMEM;
1767
		pnfs_put_lseg(desc->pg_lseg);
1768 1769 1770
		desc->pg_lseg = NULL;
		return ret;
	}
1771 1772
	hdr = &rhdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1773
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1774
	atomic_inc(&hdr->refcnt);
1775
	ret = nfs_generic_pgio(desc, hdr);
1776
	if (ret != 0) {
1777
		pnfs_put_lseg(desc->pg_lseg);
1778 1779 1780 1781
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_reads(desc, &hdr->rpc_list);
	if (atomic_dec_and_test(&hdr->refcnt))
1782
		hdr->completion_ops->completion(hdr);
1783
	return ret;
1784 1785 1786
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);

1787 1788 1789 1790 1791 1792 1793 1794 1795
static void pnfs_clear_layoutcommitting(struct inode *inode)
{
	unsigned long *bitlock = &NFS_I(inode)->flags;

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

A
Andy Adamson 已提交
1796
/*
1797
 * There can be multiple RW segments.
A
Andy Adamson 已提交
1798
 */
1799
static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
A
Andy Adamson 已提交
1800
{
1801
	struct pnfs_layout_segment *lseg;
A
Andy Adamson 已提交
1802

1803 1804
	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
		if (lseg->pls_range.iomode == IOMODE_RW &&
1805
		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1806 1807
			list_add(&lseg->pls_lc_list, listp);
	}
A
Andy Adamson 已提交
1808 1809
}

1810 1811 1812 1813 1814 1815 1816 1817 1818 1819
static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
{
	struct pnfs_layout_segment *lseg, *tmp;

	/* 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);
	}

1820
	pnfs_clear_layoutcommitting(inode);
1821 1822
}

P
Peng Tao 已提交
1823 1824
void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
{
1825
	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
P
Peng Tao 已提交
1826 1827 1828
}
EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);

A
Andy Adamson 已提交
1829
void
1830
pnfs_set_layoutcommit(struct nfs_pgio_data *wdata)
A
Andy Adamson 已提交
1831
{
1832 1833 1834
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
	struct nfs_inode *nfsi = NFS_I(inode);
1835
	loff_t end_pos = wdata->mds_offset + wdata->res.count;
1836
	bool mark_as_dirty = false;
A
Andy Adamson 已提交
1837

1838
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1839
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1840
		mark_as_dirty = true;
A
Andy Adamson 已提交
1841
		dprintk("%s: Set layoutcommit for inode %lu ",
1842
			__func__, inode->i_ino);
A
Andy Adamson 已提交
1843
	}
1844
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
1845
		/* references matched in nfs4_layoutcommit_release */
1846
		pnfs_get_lseg(hdr->lseg);
1847
	}
1848 1849
	if (end_pos > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = end_pos;
1850
	spin_unlock(&inode->i_lock);
1851
	dprintk("%s: lseg %p end_pos %llu\n",
1852
		__func__, hdr->lseg, nfsi->layout->plh_lwb);
1853 1854 1855 1856

	/* 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)
1857
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1858 1859 1860
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

A
Andy Adamson 已提交
1861 1862 1863 1864 1865 1866
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);
1867
	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
A
Andy Adamson 已提交
1868 1869
}

1870 1871 1872 1873 1874 1875 1876 1877
/*
 * 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
1878
int
1879
pnfs_layoutcommit_inode(struct inode *inode, bool sync)
A
Andy Adamson 已提交
1880 1881 1882 1883
{
	struct nfs4_layoutcommit_data *data;
	struct nfs_inode *nfsi = NFS_I(inode);
	loff_t end_pos;
1884
	int status;
A
Andy Adamson 已提交
1885

1886
	if (!pnfs_layoutcommit_outstanding(inode))
1887 1888
		return 0;

1889
	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
P
Peng Tao 已提交
1890

1891
	status = -EAGAIN;
P
Peng Tao 已提交
1892
	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
1893 1894 1895 1896 1897 1898
		if (!sync)
			goto out;
		status = wait_on_bit_lock(&nfsi->flags,
				NFS_INO_LAYOUTCOMMITTING,
				nfs_wait_bit_killable,
				TASK_KILLABLE);
P
Peng Tao 已提交
1899
		if (status)
1900
			goto out;
P
Peng Tao 已提交
1901 1902
	}

1903 1904 1905 1906 1907 1908 1909
	status = -ENOMEM;
	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
	data = kzalloc(sizeof(*data), GFP_NOFS);
	if (!data)
		goto clear_layoutcommitting;

	status = 0;
1910
	spin_lock(&inode->i_lock);
1911 1912
	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		goto out_unlock;
1913

1914
	INIT_LIST_HEAD(&data->lseg_list);
1915
	pnfs_list_write_lseg(inode, &data->lseg_list);
A
Andy Adamson 已提交
1916

1917 1918
	end_pos = nfsi->layout->plh_lwb;
	nfsi->layout->plh_lwb = 0;
A
Andy Adamson 已提交
1919

1920
	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
A
Andy Adamson 已提交
1921 1922 1923
	spin_unlock(&inode->i_lock);

	data->args.inode = inode;
1924
	data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
A
Andy Adamson 已提交
1925 1926 1927 1928 1929 1930 1931 1932
	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
1933 1934
	if (status)
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1935 1936
	dprintk("<-- %s status %d\n", __func__, status);
	return status;
1937 1938
out_unlock:
	spin_unlock(&inode->i_lock);
P
Peng Tao 已提交
1939
	kfree(data);
1940 1941
clear_layoutcommitting:
	pnfs_clear_layoutcommitting(inode);
P
Peng Tao 已提交
1942
	goto out;
A
Andy Adamson 已提交
1943
}
1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955

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;
}