pnfs.c 52.8 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 void pnfs_free_lseg_async_work(struct work_struct *work)
365 366
{
	struct pnfs_layout_segment *lseg;
367
	struct pnfs_layout_hdr *lo;
368 369

	lseg = container_of(work, struct pnfs_layout_segment, pls_work);
370
	lo = lseg->pls_layout;
371

372 373
	pnfs_free_lseg(lseg);
	pnfs_put_layout_hdr(lo);
374 375
}

376
static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
377
{
378
	INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
379 380
	schedule_work(&lseg->pls_work);
}
381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400

void
pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
{
	if (!lseg)
		return;

	assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);

	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
		atomic_read(&lseg->pls_refcount),
		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
	if (atomic_dec_and_test(&lseg->pls_refcount)) {
		struct pnfs_layout_hdr *lo = lseg->pls_layout;
		pnfs_get_layout_hdr(lo);
		pnfs_layout_remove_lseg(lo, lseg);
		pnfs_free_lseg_async(lseg);
	}
}
EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
401

402
static u64
403 404 405 406 407 408 409 410 411 412 413 414 415 416 417
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
 *           [----------------)
 */
418
static bool
419
pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
420
		 const struct pnfs_layout_range *l2)
421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436
{
	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
 *                              [----------------)
 */
437
static bool
438
pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
439
		    const struct pnfs_layout_range *l2)
440 441 442 443 444 445 446 447 448 449
{
	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);
}

450
static bool
451 452
should_free_lseg(const struct pnfs_layout_range *lseg_range,
		 const struct pnfs_layout_range *recall_range)
453
{
454 455
	return (recall_range->iomode == IOMODE_ANY ||
		lseg_range->iomode == recall_range->iomode) &&
456
	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
457 458
}

459 460 461 462 463 464 465 466 467 468
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;
}

469 470 471 472 473 474 475 476 477 478 479
/* 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
480 481
		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
			atomic_read(&lseg->pls_refcount));
482
		if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
F
Fred Isaman 已提交
483
			rv = 1;
484 485 486 487 488 489 490
	}
	return rv;
}

/* Returns count of number of matching invalid lsegs remaining in list
 * after call.
 */
F
Fred Isaman 已提交
491
int
492
pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
493
			    struct list_head *tmp_list,
494
			    struct pnfs_layout_range *recall_range)
495 496
{
	struct pnfs_layout_segment *lseg, *next;
497
	int invalid = 0, removed = 0;
498 499 500

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

501
	if (list_empty(&lo->plh_segs))
502
		return 0;
503
	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
504 505
		if (!recall_range ||
		    should_free_lseg(&lseg->pls_range, recall_range)) {
506 507 508 509 510 511 512 513 514
			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;
515 516
}

517
/* note free_me must contain lsegs from a single layout_hdr */
F
Fred Isaman 已提交
518
void
519
pnfs_free_lseg_list(struct list_head *free_me)
520
{
521
	struct pnfs_layout_segment *lseg, *tmp;
522 523 524 525

	if (list_empty(free_me))
		return;

526
	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
527
		list_del(&lseg->pls_list);
528
		pnfs_free_lseg(lseg);
529 530 531
	}
}

532 533 534 535
void
pnfs_destroy_layout(struct nfs_inode *nfsi)
{
	struct pnfs_layout_hdr *lo;
536
	LIST_HEAD(tmp_list);
537 538 539 540

	spin_lock(&nfsi->vfs_inode.i_lock);
	lo = nfsi->layout;
	if (lo) {
541
		lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
542
		pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
543 544 545 546 547 548 549 550
		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);
551
}
552
EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
553

554 555 556
static bool
pnfs_layout_add_bulk_destroy_list(struct inode *inode,
		struct list_head *layout_list)
557 558
{
	struct pnfs_layout_hdr *lo;
559
	bool ret = false;
560

561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617
	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;
618 619 620

		pnfs_layoutcommit_inode(inode, false);

621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642
		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);
643

644
	spin_lock(&clp->cl_lock);
645
	rcu_read_lock();
646
restart:
647
	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
648 649 650 651 652 653
		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
			continue;
		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
				server,
				&layout_list) != 0)
			goto restart;
654 655
	}
	rcu_read_unlock();
656 657
	spin_unlock(&clp->cl_lock);

658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677
	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;
678
	}
679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697
	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);
698 699
}

700 701 702 703 704 705
/*
 * 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)
{
706
	return (s32)(s1 - s2) > 0;
707 708
}

709
/* update lo->plh_stateid with new if is more recent */
F
Fred Isaman 已提交
710 711 712
void
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
			bool update_barrier)
713
{
714 715
	u32 oldseq, newseq, new_barrier;
	int empty = list_empty(&lo->plh_segs);
716

717 718
	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
	newseq = be32_to_cpu(new->seqid);
719
	if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
720
		nfs4_stateid_copy(&lo->plh_stateid, new);
F
Fred Isaman 已提交
721
		if (update_barrier) {
722
			new_barrier = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
723 724
		} else {
			/* Because of wraparound, we want to keep the barrier
725
			 * "close" to the current seqids.
F
Fred Isaman 已提交
726
			 */
727
			new_barrier = newseq - atomic_read(&lo->plh_outstanding);
F
Fred Isaman 已提交
728
		}
729 730
		if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
			lo->plh_barrier = new_barrier;
F
Fred Isaman 已提交
731
	}
732 733
}

734
static bool
735 736
pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
		const nfs4_stateid *stateid)
F
Fred Isaman 已提交
737
{
738
	u32 seqid = be32_to_cpu(stateid->seqid);
739

740 741 742 743 744 745 746
	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
747 748
	return lo->plh_block_lgets ||
		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
F
Fred Isaman 已提交
749
		(list_empty(&lo->plh_segs) &&
750 751 752
		 (atomic_read(&lo->plh_outstanding) > lget));
}

753 754 755
int
pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
			      struct nfs4_state *open_state)
756
{
757
	int status = 0;
758

759
	dprintk("--> %s\n", __func__);
760
	spin_lock(&lo->plh_inode->i_lock);
761
	if (pnfs_layoutgets_blocked(lo, 1)) {
762
		status = -EAGAIN;
763 764
	} else if (!nfs4_valid_open_stateid(open_state)) {
		status = -EBADF;
765 766
	} else if (list_empty(&lo->plh_segs) ||
		   test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
767 768 769 770
		int seq;

		do {
			seq = read_seqbegin(&open_state->seqlock);
771
			nfs4_stateid_copy(dst, &open_state->stateid);
772 773
		} while (read_seqretry(&open_state->seqlock, seq));
	} else
774
		nfs4_stateid_copy(dst, &lo->plh_stateid);
775
	spin_unlock(&lo->plh_inode->i_lock);
776
	dprintk("<-- %s\n", __func__);
777
	return status;
778 779 780 781 782 783 784 785
}

/*
* Get layout from server.
*    for now, assume that whole file layouts are requested.
*    arg->offset: 0
*    arg->length: all ones
*/
786 787 788
static struct pnfs_layout_segment *
send_layoutget(struct pnfs_layout_hdr *lo,
	   struct nfs_open_context *ctx,
789
	   struct pnfs_layout_range *range,
790
	   gfp_t gfp_flags)
791
{
792
	struct inode *ino = lo->plh_inode;
793 794
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs4_layoutget *lgp;
795
	struct pnfs_layout_segment *lseg;
796 797

	dprintk("--> %s\n", __func__);
798

799
	lgp = kzalloc(sizeof(*lgp), gfp_flags);
800
	if (lgp == NULL)
801
		return NULL;
802

803 804 805
	lgp->args.minlength = PAGE_CACHE_SIZE;
	if (lgp->args.minlength > range->length)
		lgp->args.minlength = range->length;
806
	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
807
	lgp->args.range = *range;
808 809 810
	lgp->args.type = server->pnfs_curr_ld->id;
	lgp->args.inode = ino;
	lgp->args.ctx = get_nfs_open_context(ctx);
811
	lgp->gfp_flags = gfp_flags;
812
	lgp->cred = lo->plh_lc_cred;
813 814 815 816

	/* Synchronously retrieve layout information from server and
	 * store in lseg.
	 */
817 818 819 820 821 822 823 824
	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 */
825
			pnfs_layout_io_set_failed(lo, range->iomode);
826 827
		}
		return NULL;
828
	}
829

830 831 832
	return lseg;
}

833 834 835 836 837 838 839 840 841 842 843 844 845 846 847
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);
	}
}

848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879
static int
pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
		       enum pnfs_iomode iomode)
{
	struct inode *ino = lo->plh_inode;
	struct nfs4_layoutreturn *lrp;
	int status = 0;

	lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
	if (unlikely(lrp == NULL)) {
		status = -ENOMEM;
		spin_lock(&ino->i_lock);
		lo->plh_block_lgets--;
		spin_unlock(&ino->i_lock);
		pnfs_put_layout_hdr(lo);
		goto out;
	}

	lrp->args.stateid = stateid;
	lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
	lrp->args.inode = ino;
	lrp->args.iomode = iomode;
	lrp->args.layout = lo;
	lrp->clp = NFS_SERVER(ino)->nfs_client;
	lrp->cred = lo->plh_lc_cred;

	status = nfs4_proc_layoutreturn(lrp);
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}

880 881 882 883 884 885 886 887
/*
 * 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
888 889 890 891 892 893 894
int
_pnfs_return_layout(struct inode *ino)
{
	struct pnfs_layout_hdr *lo = NULL;
	struct nfs_inode *nfsi = NFS_I(ino);
	LIST_HEAD(tmp_list);
	nfs4_stateid stateid;
895
	int status = 0, empty;
B
Benny Halevy 已提交
896

897
	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
B
Benny Halevy 已提交
898 899 900

	spin_lock(&ino->i_lock);
	lo = nfsi->layout;
901
	if (!lo) {
B
Benny Halevy 已提交
902
		spin_unlock(&ino->i_lock);
903 904
		dprintk("NFS: %s no layout to return\n", __func__);
		goto out;
B
Benny Halevy 已提交
905 906 907
	}
	stateid = nfsi->layout->plh_stateid;
	/* Reference matched in nfs4_layoutreturn_release */
908
	pnfs_get_layout_hdr(lo);
909
	empty = list_empty(&lo->plh_segs);
910
	pnfs_clear_layoutcommit(ino, &tmp_list);
911
	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
912 913 914 915 916 917 918 919 920 921

	if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
		struct pnfs_layout_range range = {
			.iomode		= IOMODE_ANY,
			.offset		= 0,
			.length		= NFS4_MAX_UINT64,
		};
		NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
	}

922 923 924
	/* Don't send a LAYOUTRETURN if list was initially empty */
	if (empty) {
		spin_unlock(&ino->i_lock);
925
		pnfs_put_layout_hdr(lo);
926 927 928
		dprintk("NFS: %s no layout segments to return\n", __func__);
		goto out;
	}
929 930

	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
931
	lo->plh_block_lgets++;
B
Benny Halevy 已提交
932 933 934
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);

935
	status = pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY);
B
Benny Halevy 已提交
936 937 938 939
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}
940
EXPORT_SYMBOL_GPL(_pnfs_return_layout);
B
Benny Halevy 已提交
941

942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968
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
969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988
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++;
989
	pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
F
Fred Isaman 已提交
990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005
	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--;
1006 1007 1008 1009 1010 1011
	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
1012 1013 1014 1015 1016 1017 1018 1019
}

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;
1020
	if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
F
Fred Isaman 已提交
1021 1022 1023 1024
		lo->plh_barrier = barrier;
	spin_unlock(&ino->i_lock);
}

1025
bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
F
Fred Isaman 已提交
1026 1027
{
	struct nfs_inode *nfsi = NFS_I(ino);
1028
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
1029
	struct pnfs_layout_segment *lseg;
1030
	u32 current_seqid;
F
Fred Isaman 已提交
1031 1032 1033 1034 1035
	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)) {
1036
			rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
F
Fred Isaman 已提交
1037
			found = true;
1038
			goto out;
F
Fred Isaman 已提交
1039
		}
1040 1041
	lo = nfsi->layout;
	current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
F
Fred Isaman 已提交
1042

1043 1044 1045 1046 1047
	/* 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
1048 1049 1050 1051
	spin_unlock(&ino->i_lock);
	return found;
}

1052 1053 1054 1055 1056 1057
/*
 * 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
1058
pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1059
	   const struct pnfs_layout_range *l2)
1060
{
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072
	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;

1073
	/* read > read/write */
1074
	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1075 1076
}

1077
static void
1078
pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1079 1080
		   struct pnfs_layout_segment *lseg)
{
1081 1082
	struct pnfs_layout_segment *lp;

1083 1084
	dprintk("%s:Begin\n", __func__);

1085
	list_for_each_entry(lp, &lo->plh_segs, pls_list) {
1086
		if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0)
1087
			continue;
1088
		list_add_tail(&lseg->pls_list, &lp->pls_list);
1089 1090 1091
		dprintk("%s: inserted lseg %p "
			"iomode %d offset %llu length %llu before "
			"lp %p iomode %d offset %llu length %llu\n",
1092 1093 1094 1095
			__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);
1096
		goto out;
1097
	}
1098 1099 1100 1101 1102 1103
	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:
1104
	pnfs_get_layout_hdr(lo);
1105 1106

	dprintk("%s:Return\n", __func__);
1107 1108 1109
}

static struct pnfs_layout_hdr *
1110 1111 1112
alloc_init_layout_hdr(struct inode *ino,
		      struct nfs_open_context *ctx,
		      gfp_t gfp_flags)
1113 1114 1115
{
	struct pnfs_layout_hdr *lo;

1116
	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1117 1118
	if (!lo)
		return NULL;
1119
	atomic_set(&lo->plh_refcount, 1);
1120 1121
	INIT_LIST_HEAD(&lo->plh_layouts);
	INIT_LIST_HEAD(&lo->plh_segs);
1122
	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1123
	lo->plh_inode = ino;
1124
	lo->plh_lc_cred = get_rpccred(ctx->cred);
1125 1126 1127 1128
	return lo;
}

static struct pnfs_layout_hdr *
1129 1130 1131
pnfs_find_alloc_layout(struct inode *ino,
		       struct nfs_open_context *ctx,
		       gfp_t gfp_flags)
1132 1133 1134 1135 1136 1137
{
	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);

1138 1139
	if (nfsi->layout != NULL)
		goto out_existing;
1140
	spin_unlock(&ino->i_lock);
1141
	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1142 1143
	spin_lock(&ino->i_lock);

1144
	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1145
		nfsi->layout = new;
1146
		return new;
1147 1148
	} else if (new != NULL)
		pnfs_free_layout_hdr(new);
1149 1150
out_existing:
	pnfs_get_layout_hdr(nfsi->layout);
1151 1152 1153
	return nfsi->layout;
}

1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164
/*
 * iomode matching rules:
 * iomode	lseg	match
 * -----	-----	-----
 * ANY		READ	true
 * ANY		RW	true
 * RW		READ	false
 * RW		RW	true
 * READ		READ	true
 * READ		RW	true
 */
1165
static bool
1166
pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1167
		 const struct pnfs_layout_range *range)
1168
{
1169 1170 1171 1172
	struct pnfs_layout_range range1;

	if ((range->iomode == IOMODE_RW &&
	     ls_range->iomode != IOMODE_RW) ||
1173
	    !pnfs_lseg_range_intersecting(ls_range, range))
1174 1175 1176 1177 1178
		return 0;

	/* range1 covers only the first byte in the range */
	range1 = *range;
	range1.length = 1;
1179
	return pnfs_lseg_range_contained(ls_range, &range1);
1180 1181 1182 1183 1184
}

/*
 * lookup range in layout
 */
1185
static struct pnfs_layout_segment *
1186 1187
pnfs_find_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_range *range)
1188
{
1189 1190 1191 1192
	struct pnfs_layout_segment *lseg, *ret = NULL;

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

1193
	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1194
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1195
		    pnfs_lseg_range_match(&lseg->pls_range, range)) {
1196
			ret = pnfs_get_lseg(lseg);
1197 1198
			break;
		}
1199
		if (lseg->pls_range.offset > range->offset)
1200 1201 1202 1203
			break;
	}

	dprintk("%s:Return lseg %p ref %d\n",
1204
		__func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1205
	return ret;
1206 1207
}

1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282
/*
 * 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;
}

1283 1284 1285 1286
/*
 * Layout segment is retreived from the server if not cached.
 * The appropriate layout segment is referenced and returned to the caller.
 */
1287
struct pnfs_layout_segment *
1288 1289
pnfs_update_layout(struct inode *ino,
		   struct nfs_open_context *ctx,
1290 1291
		   loff_t pos,
		   u64 count,
1292 1293
		   enum pnfs_iomode iomode,
		   gfp_t gfp_flags)
1294
{
1295 1296 1297 1298 1299
	struct pnfs_layout_range arg = {
		.iomode = iomode,
		.offset = pos,
		.length = count,
	};
1300
	unsigned pg_offset;
1301 1302
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs_client *clp = server->nfs_client;
1303 1304
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg = NULL;
1305
	bool first;
1306 1307

	if (!pnfs_enabled_sb(NFS_SERVER(ino)))
1308
		goto out;
1309 1310

	if (pnfs_within_mdsthreshold(ctx, ino, iomode))
1311
		goto out;
1312

1313 1314
lookup_again:
	first = false;
1315
	spin_lock(&ino->i_lock);
1316
	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1317 1318 1319 1320
	if (lo == NULL) {
		spin_unlock(&ino->i_lock);
		goto out;
	}
1321

F
Fred Isaman 已提交
1322
	/* Do we even need to bother with this? */
1323
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1324
		dprintk("%s matches recall, use MDS\n", __func__);
1325 1326 1327 1328
		goto out_unlock;
	}

	/* if LAYOUTGET already failed once we don't try again */
1329
	if (pnfs_layout_io_test_failed(lo, iomode))
1330 1331
		goto out_unlock;

1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352
	first = list_empty(&lo->plh_segs);
	if (first) {
		/* The first layoutget for the file. Need to serialize per
		 * RFC 5661 Errata 3208.
		 */
		if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
				     &lo->plh_flags)) {
			spin_unlock(&ino->i_lock);
			wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
				    TASK_UNINTERRUPTIBLE);
			pnfs_put_layout_hdr(lo);
			goto lookup_again;
		}
	} else {
		/* Check to see if the layout for the given range
		 * already exists
		 */
		lseg = pnfs_find_lseg(lo, &arg);
		if (lseg)
			goto out_unlock;
	}
1353

1354
	if (pnfs_layoutgets_blocked(lo, 0))
1355 1356
		goto out_unlock;
	atomic_inc(&lo->plh_outstanding);
1357
	spin_unlock(&ino->i_lock);
1358

1359
	if (list_empty(&lo->plh_layouts)) {
1360 1361 1362 1363
		/* 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);
1364 1365
		if (list_empty(&lo->plh_layouts))
			list_add_tail(&lo->plh_layouts, &server->layouts);
1366 1367
		spin_unlock(&clp->cl_lock);
	}
1368

1369 1370 1371 1372 1373
	pg_offset = arg.offset & ~PAGE_CACHE_MASK;
	if (pg_offset) {
		arg.offset -= pg_offset;
		arg.length += pg_offset;
	}
1374 1375
	if (arg.length != NFS4_MAX_UINT64)
		arg.length = PAGE_CACHE_ALIGN(arg.length);
1376

1377
	lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1378
	atomic_dec(&lo->plh_outstanding);
1379
out_put_layout_hdr:
1380 1381 1382 1383 1384 1385 1386
	if (first) {
		unsigned long *bitlock = &lo->plh_flags;

		clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
		smp_mb__after_atomic();
		wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
	}
1387
	pnfs_put_layout_hdr(lo);
1388
out:
1389 1390 1391 1392 1393 1394 1395 1396
	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);
1397 1398 1399
	return lseg;
out_unlock:
	spin_unlock(&ino->i_lock);
1400
	goto out_put_layout_hdr;
1401
}
1402
EXPORT_SYMBOL_GPL(pnfs_update_layout);
1403

1404
struct pnfs_layout_segment *
1405 1406 1407 1408 1409
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;
1410
	struct inode *ino = lo->plh_inode;
1411
	LIST_HEAD(free_me);
1412 1413 1414
	int status = 0;

	/* Inject layout blob into I/O device driver */
1415
	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1416 1417 1418 1419 1420 1421 1422 1423 1424 1425
	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;
	}

1426 1427 1428
	init_lseg(lo, lseg);
	lseg->pls_range = res->range;

1429
	spin_lock(&ino->i_lock);
1430
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1431 1432 1433 1434
		dprintk("%s forget reply due to recall\n", __func__);
		goto out_forget_reply;
	}

1435
	if (pnfs_layoutgets_blocked(lo, 1)) {
F
Fred Isaman 已提交
1436 1437 1438
		dprintk("%s forget reply due to state\n", __func__);
		goto out_forget_reply;
	}
1439

1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457
	if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
		/* existing state ID, make sure the sequence number matches. */
		if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
			dprintk("%s forget reply due to sequence\n", __func__);
			goto out_forget_reply;
		}
		pnfs_set_layout_stateid(lo, &res->stateid, false);
	} else {
		/*
		 * We got an entirely new state ID.  Mark all segments for the
		 * inode invalid, and don't bother validating the stateid
		 * sequence number.
		 */
		pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL);

		nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
		lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
	}
1458

1459 1460
	clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);

1461
	pnfs_get_lseg(lseg);
1462
	pnfs_layout_insert_lseg(lo, lseg);
1463

F
Fred Isaman 已提交
1464 1465 1466 1467 1468
	if (res->return_on_close) {
		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
		set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
	}

1469
	spin_unlock(&ino->i_lock);
1470
	pnfs_free_lseg_list(&free_me);
1471
	return lseg;
1472
out:
1473
	return ERR_PTR(status);
F
Fred Isaman 已提交
1474 1475 1476 1477 1478 1479

out_forget_reply:
	spin_unlock(&ino->i_lock);
	lseg->pls_layout = lo;
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
	goto out;
1480 1481
}

1482 1483 1484
void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
P
Peng Tao 已提交
1485 1486
	u64 rd_size = req->wb_bytes;

1487
	WARN_ON_ONCE(pgio->pg_lseg != NULL);
1488

P
Peng Tao 已提交
1489 1490 1491 1492 1493
	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);

1494 1495 1496
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
P
Peng Tao 已提交
1497
					   rd_size,
1498 1499
					   IOMODE_READ,
					   GFP_KERNEL);
1500 1501
	/* If no lseg, fall back to read through mds */
	if (pgio->pg_lseg == NULL)
1502
		nfs_pageio_reset_read_mds(pgio);
1503

1504 1505 1506 1507
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);

void
1508 1509
pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
			   struct nfs_page *req, u64 wb_size)
1510
{
1511
	WARN_ON_ONCE(pgio->pg_lseg != NULL);
1512 1513 1514 1515

	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
1516
					   wb_size,
1517 1518
					   IOMODE_RW,
					   GFP_NOFS);
1519 1520
	/* If no lseg, fall back to write through mds */
	if (pgio->pg_lseg == NULL)
1521
		nfs_pageio_reset_write_mds(pgio);
1522 1523 1524
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);

1525 1526 1527 1528 1529
/*
 * 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
1530 1531
pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
		     struct nfs_page *req)
1532
{
1533
	unsigned int size;
1534
	u64 seg_end, req_start, seg_left;
1535 1536 1537 1538

	size = nfs_generic_pg_test(pgio, prev, req);
	if (!size)
		return 0;
1539

1540
	/*
1541 1542 1543 1544 1545
	 * 'size' contains the number of bytes left in the current page (up
	 * to the original size asked for in @req->wb_bytes).
	 *
	 * Calculate how many bytes are left in the layout segment
	 * and if there are less bytes than 'size', return that instead.
1546 1547 1548 1549 1550
	 *
	 * Please also note that 'end_offset' is actually the offset of the
	 * first byte that lies outside the pnfs_layout_range. FIXME?
	 *
	 */
1551
	if (pgio->pg_lseg) {
1552 1553 1554 1555 1556 1557
		seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
				     pgio->pg_lseg->pls_range.length);
		req_start = req_offset(req);
		WARN_ON_ONCE(req_start > seg_end);
		/* start of request is past the last byte of this segment */
		if (req_start >= seg_end)
1558
			return 0;
1559 1560 1561 1562 1563 1564

		/* adjust 'size' iff there are fewer bytes left in the
		 * segment than what nfs_generic_pg_test returned */
		seg_left = seg_end - req_start;
		if (seg_left < size)
			size = (unsigned int)seg_left;
1565
	}
1566

1567
	return size;
1568
}
1569
EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1570

1571
int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
1572 1573 1574 1575
{
	struct nfs_pageio_descriptor pgio;

	/* Resend all requests through the MDS */
1576 1577 1578
	nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
			      hdr->completion_ops);
	return nfs_pageio_resend(&pgio, hdr);
1579
}
1580
EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1581

1582
static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1583
{
1584 1585 1586

	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1587
	    PNFS_LAYOUTRET_ON_ERROR) {
1588
		pnfs_return_layout(hdr->inode);
1589
	}
1590
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1591
		hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1592 1593
}

1594 1595 1596
/*
 * Called by non rpc-based layout drivers
 */
1597
void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
1598
{
1599
	trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
1600
	if (!hdr->pnfs_error) {
1601 1602
		pnfs_set_layoutcommit(hdr);
		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1603
	} else
1604 1605
		pnfs_ld_handle_write_error(hdr);
	hdr->mds_ops->rpc_release(hdr);
1606
}
1607
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1608

1609 1610
static void
pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1611
		struct nfs_pgio_header *hdr)
1612
{
1613 1614 1615 1616 1617
	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;
	}
1618
	nfs_pgio_data_destroy(hdr);
1619 1620 1621
}

static enum pnfs_try_status
1622
pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
1623 1624 1625
			const struct rpc_call_ops *call_ops,
			struct pnfs_layout_segment *lseg,
			int how)
1626
{
1627
	struct inode *inode = hdr->inode;
1628 1629 1630
	enum pnfs_try_status trypnfs;
	struct nfs_server *nfss = NFS_SERVER(inode);

1631
	hdr->mds_ops = call_ops;
1632 1633

	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1634 1635
		inode->i_ino, hdr->args.count, hdr->args.offset, how);
	trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
1636
	if (trypnfs != PNFS_NOT_ATTEMPTED)
1637 1638 1639 1640 1641
		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}

1642
static void
1643 1644
pnfs_do_write(struct nfs_pageio_descriptor *desc,
	      struct nfs_pgio_header *hdr, int how)
1645 1646 1647
{
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;
1648
	enum pnfs_try_status trypnfs;
1649 1650

	desc->pg_lseg = NULL;
1651
	trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
1652
	if (trypnfs == PNFS_NOT_ATTEMPTED)
1653
		pnfs_write_through_mds(desc, hdr);
1654
	pnfs_put_lseg(lseg);
1655 1656
}

1657 1658
static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
{
1659
	pnfs_put_lseg(hdr->lseg);
1660
	nfs_pgio_header_free(hdr);
1661
}
B
Bryan Schumaker 已提交
1662
EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
1663

1664 1665 1666
int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
{
1667
	struct nfs_pgio_header *hdr;
1668 1669
	int ret;

1670 1671
	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
	if (!hdr) {
1672
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1673
		pnfs_put_lseg(desc->pg_lseg);
1674
		desc->pg_lseg = NULL;
1675
		return -ENOMEM;
1676
	}
1677
	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1678
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1679
	ret = nfs_generic_pgio(desc, hdr);
1680
	if (ret != 0) {
1681
		pnfs_put_lseg(desc->pg_lseg);
1682 1683
		desc->pg_lseg = NULL;
	} else
1684
		pnfs_do_write(desc, hdr, desc->pg_ioflags);
1685
	return ret;
1686 1687 1688
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);

1689
int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
1690 1691 1692
{
	struct nfs_pageio_descriptor pgio;

1693
	/* Resend all requests through the MDS */
1694 1695
	nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
	return nfs_pageio_resend(&pgio, hdr);
1696
}
1697
EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1698

1699
static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1700
{
1701 1702
	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1703
	    PNFS_LAYOUTRET_ON_ERROR) {
1704
		pnfs_return_layout(hdr->inode);
1705
	}
1706
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1707
		hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
1708 1709
}

1710 1711 1712
/*
 * Called by non rpc-based layout drivers
 */
1713
void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
1714
{
1715
	trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
1716
	if (likely(!hdr->pnfs_error)) {
1717 1718
		__nfs4_read_done_cb(hdr);
		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1719
	} else
1720 1721
		pnfs_ld_handle_read_error(hdr);
	hdr->mds_ops->rpc_release(hdr);
1722 1723 1724
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

1725 1726
static void
pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1727
		struct nfs_pgio_header *hdr)
1728
{
1729 1730 1731 1732 1733
	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;
	}
1734
	nfs_pgio_data_destroy(hdr);
1735 1736
}

A
Andy Adamson 已提交
1737 1738 1739
/*
 * Call the appropriate parallel I/O subsystem read function.
 */
1740
static enum pnfs_try_status
1741
pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
1742 1743
		       const struct rpc_call_ops *call_ops,
		       struct pnfs_layout_segment *lseg)
A
Andy Adamson 已提交
1744
{
1745
	struct inode *inode = hdr->inode;
A
Andy Adamson 已提交
1746 1747 1748
	struct nfs_server *nfss = NFS_SERVER(inode);
	enum pnfs_try_status trypnfs;

1749
	hdr->mds_ops = call_ops;
A
Andy Adamson 已提交
1750 1751

	dprintk("%s: Reading ino:%lu %u@%llu\n",
1752
		__func__, inode->i_ino, hdr->args.count, hdr->args.offset);
A
Andy Adamson 已提交
1753

1754
	trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
1755
	if (trypnfs != PNFS_NOT_ATTEMPTED)
A
Andy Adamson 已提交
1756 1757 1758 1759
		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}
A
Andy Adamson 已提交
1760

1761
static void
1762
pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
1763 1764 1765
{
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;
1766
	enum pnfs_try_status trypnfs;
1767 1768

	desc->pg_lseg = NULL;
1769
	trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
1770
	if (trypnfs == PNFS_NOT_ATTEMPTED)
1771
		pnfs_read_through_mds(desc, hdr);
1772
	pnfs_put_lseg(lseg);
1773 1774
}

1775 1776
static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
{
1777
	pnfs_put_lseg(hdr->lseg);
1778
	nfs_pgio_header_free(hdr);
1779
}
B
Bryan Schumaker 已提交
1780
EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
1781

1782 1783 1784
int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
{
1785
	struct nfs_pgio_header *hdr;
1786 1787
	int ret;

1788 1789
	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
	if (!hdr) {
1790
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1791
		ret = -ENOMEM;
1792
		pnfs_put_lseg(desc->pg_lseg);
1793 1794 1795
		desc->pg_lseg = NULL;
		return ret;
	}
1796
	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1797
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1798
	ret = nfs_generic_pgio(desc, hdr);
1799
	if (ret != 0) {
1800
		pnfs_put_lseg(desc->pg_lseg);
1801 1802
		desc->pg_lseg = NULL;
	} else
1803
		pnfs_do_read(desc, hdr);
1804
	return ret;
1805 1806 1807
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);

1808 1809 1810 1811 1812
static void pnfs_clear_layoutcommitting(struct inode *inode)
{
	unsigned long *bitlock = &NFS_I(inode)->flags;

	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
1813
	smp_mb__after_atomic();
1814 1815 1816
	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
}

A
Andy Adamson 已提交
1817
/*
1818
 * There can be multiple RW segments.
A
Andy Adamson 已提交
1819
 */
1820
static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
A
Andy Adamson 已提交
1821
{
1822
	struct pnfs_layout_segment *lseg;
A
Andy Adamson 已提交
1823

1824 1825
	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
		if (lseg->pls_range.iomode == IOMODE_RW &&
1826
		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1827 1828
			list_add(&lseg->pls_lc_list, listp);
	}
A
Andy Adamson 已提交
1829 1830
}

1831 1832 1833 1834 1835 1836 1837 1838 1839 1840
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);
	}

1841
	pnfs_clear_layoutcommitting(inode);
1842 1843
}

P
Peng Tao 已提交
1844 1845
void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
{
1846
	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
P
Peng Tao 已提交
1847 1848 1849
}
EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);

A
Andy Adamson 已提交
1850
void
1851
pnfs_set_layoutcommit(struct nfs_pgio_header *hdr)
A
Andy Adamson 已提交
1852
{
1853 1854
	struct inode *inode = hdr->inode;
	struct nfs_inode *nfsi = NFS_I(inode);
1855
	loff_t end_pos = hdr->mds_offset + hdr->res.count;
1856
	bool mark_as_dirty = false;
A
Andy Adamson 已提交
1857

1858
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1859
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1860
		mark_as_dirty = true;
A
Andy Adamson 已提交
1861
		dprintk("%s: Set layoutcommit for inode %lu ",
1862
			__func__, inode->i_ino);
A
Andy Adamson 已提交
1863
	}
1864
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
1865
		/* references matched in nfs4_layoutcommit_release */
1866
		pnfs_get_lseg(hdr->lseg);
1867
	}
1868 1869
	if (end_pos > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = end_pos;
1870
	spin_unlock(&inode->i_lock);
1871
	dprintk("%s: lseg %p end_pos %llu\n",
1872
		__func__, hdr->lseg, nfsi->layout->plh_lwb);
1873 1874 1875 1876

	/* 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)
1877
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1878 1879 1880
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909
void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data)
{
	struct inode *inode = data->inode;
	struct nfs_inode *nfsi = NFS_I(inode);
	bool mark_as_dirty = false;

	spin_lock(&inode->i_lock);
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
		mark_as_dirty = true;
		dprintk("%s: Set layoutcommit for inode %lu ",
			__func__, inode->i_ino);
	}
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) {
		/* references matched in nfs4_layoutcommit_release */
		pnfs_get_lseg(data->lseg);
	}
	if (data->lwb > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = data->lwb;
	spin_unlock(&inode->i_lock);
	dprintk("%s: lseg %p end_pos %llu\n",
		__func__, data->lseg, nfsi->layout->plh_lwb);

	/* 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)
		mark_inode_dirty_sync(inode);
}
EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit);

A
Andy Adamson 已提交
1910 1911 1912 1913 1914 1915
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);
1916
	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
A
Andy Adamson 已提交
1917 1918
}

1919 1920 1921 1922 1923 1924 1925 1926
/*
 * 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
1927
int
1928
pnfs_layoutcommit_inode(struct inode *inode, bool sync)
A
Andy Adamson 已提交
1929
{
1930
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
A
Andy Adamson 已提交
1931 1932 1933
	struct nfs4_layoutcommit_data *data;
	struct nfs_inode *nfsi = NFS_I(inode);
	loff_t end_pos;
1934
	int status;
A
Andy Adamson 已提交
1935

1936
	if (!pnfs_layoutcommit_outstanding(inode))
1937 1938
		return 0;

1939
	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
P
Peng Tao 已提交
1940

1941
	status = -EAGAIN;
P
Peng Tao 已提交
1942
	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
1943 1944
		if (!sync)
			goto out;
1945
		status = wait_on_bit_lock_action(&nfsi->flags,
1946 1947 1948
				NFS_INO_LAYOUTCOMMITTING,
				nfs_wait_bit_killable,
				TASK_KILLABLE);
P
Peng Tao 已提交
1949
		if (status)
1950
			goto out;
P
Peng Tao 已提交
1951 1952
	}

1953 1954 1955 1956 1957 1958 1959
	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;
1960
	spin_lock(&inode->i_lock);
1961 1962
	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		goto out_unlock;
1963

1964
	INIT_LIST_HEAD(&data->lseg_list);
1965
	pnfs_list_write_lseg(inode, &data->lseg_list);
A
Andy Adamson 已提交
1966

1967 1968
	end_pos = nfsi->layout->plh_lwb;
	nfsi->layout->plh_lwb = 0;
A
Andy Adamson 已提交
1969

1970
	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
A
Andy Adamson 已提交
1971 1972 1973
	spin_unlock(&inode->i_lock);

	data->args.inode = inode;
1974
	data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
A
Andy Adamson 已提交
1975 1976 1977 1978 1979 1980
	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);

1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994
	if (ld->prepare_layoutcommit) {
		status = ld->prepare_layoutcommit(&data->args);
		if (status) {
			spin_lock(&inode->i_lock);
			if (end_pos < nfsi->layout->plh_lwb)
				nfsi->layout->plh_lwb = end_pos;
			spin_unlock(&inode->i_lock);
			put_rpccred(data->cred);
			set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
			goto clear_layoutcommitting;
		}
	}


A
Andy Adamson 已提交
1995 1996
	status = nfs4_proc_layoutcommit(data, sync);
out:
P
Peng Tao 已提交
1997 1998
	if (status)
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1999 2000
	dprintk("<-- %s status %d\n", __func__, status);
	return status;
2001 2002
out_unlock:
	spin_unlock(&inode->i_lock);
P
Peng Tao 已提交
2003
	kfree(data);
2004 2005
clear_layoutcommitting:
	pnfs_clear_layoutcommitting(inode);
P
Peng Tao 已提交
2006
	goto out;
A
Andy Adamson 已提交
2007
}
2008
EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020

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;
}