pnfs.c 66.1 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"
37
#include "delegation.h"
38
#include "nfs42.h"
R
Ricardo Labiaga 已提交
39 40

#define NFSDBG_FACILITY		NFSDBG_PNFS
41
#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
R
Ricardo Labiaga 已提交
42

43 44 45 46 47 48 49 50 51 52 53 54
/* 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);

55
static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
56

57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
/* 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
72 73 74
static struct pnfs_layoutdriver_type *
find_pnfs_driver(u32 id)
{
75 76 77 78
	struct pnfs_layoutdriver_type *local;

	spin_lock(&pnfs_spinlock);
	local = find_pnfs_driver_locked(id);
79 80 81 82
	if (local != NULL && !try_module_get(local->owner)) {
		dprintk("%s: Could not grab reference on module\n", __func__);
		local = NULL;
	}
83 84
	spin_unlock(&pnfs_spinlock);
	return local;
R
Ricardo Labiaga 已提交
85 86 87 88 89
}

void
unset_pnfs_layoutdriver(struct nfs_server *nfss)
{
90 91 92
	if (nfss->pnfs_curr_ld) {
		if (nfss->pnfs_curr_ld->clear_layoutdriver)
			nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
93 94 95
		/* 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);
96
		module_put(nfss->pnfs_curr_ld->owner);
97
	}
R
Ricardo Labiaga 已提交
98 99 100 101 102 103 104 105 106 107
	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
108 109
set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
		      u32 id)
R
Ricardo Labiaga 已提交
110 111 112 113 114 115 116
{
	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))) {
117 118
		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
119 120 121 122 123 124 125 126 127 128 129 130 131
		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;
132 133
	if (ld_type->set_layoutdriver
	    && ld_type->set_layoutdriver(server, mntfh)) {
134 135
		printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
			"driver %u.\n", __func__, id);
136 137 138
		module_put(ld_type->owner);
		goto out_no_driver;
	}
139 140
	/* Bump the MDS count */
	atomic_inc(&server->nfs_client->cl_mds_count);
141

R
Ricardo Labiaga 已提交
142 143 144 145 146 147 148
	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;
}
149 150 151 152 153 154 155 156

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

	if (ld_type->id == 0) {
157
		printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
158 159
		return status;
	}
160
	if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
161
		printk(KERN_ERR "NFS: %s Layout driver must provide "
162 163 164
		       "alloc_lseg and free_lseg.\n", __func__);
		return status;
	}
165 166 167 168 169 170 171 172 173

	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 {
174
		printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191
			__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);
192

193 194 195 196
/*
 * pNFS client layout cache
 */

197
/* Need to hold i_lock if caller does not already hold reference */
F
Fred Isaman 已提交
198
void
199
pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
200
{
201
	atomic_inc(&lo->plh_refcount);
202 203
}

204 205 206 207
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;
208
	return ld->alloc_layout_hdr(ino, gfp_flags);
209 210 211 212 213
}

static void
pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
{
214 215 216 217 218 219 220 221 222 223
	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);
	}
224
	put_rpccred(lo->plh_lc_cred);
225
	return ld->free_layout_hdr(lo);
226 227
}

228
static void
229
pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
230
{
231
	struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
232
	dprintk("%s: freeing layout cache %p\n", __func__, lo);
233 234 235 236
	nfsi->layout = NULL;
	/* Reset MDS Threshold I/O counters */
	nfsi->write_io = 0;
	nfsi->read_io = 0;
237 238
}

239
void
240
pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
241
{
242 243
	struct inode *inode = lo->plh_inode;

244 245
	pnfs_layoutreturn_before_put_layout_hdr(lo);

246
	if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
247 248
		if (!list_empty(&lo->plh_segs))
			WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
249
		pnfs_detach_layout_hdr(lo);
250
		spin_unlock(&inode->i_lock);
251
		pnfs_free_layout_hdr(lo);
252
	}
253 254
}

255 256 257 258 259 260 261
/*
 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
 *
 * In order to continue using the pnfs_layout_hdr, a full recovery
 * is required.
 * Note that caller must hold inode->i_lock.
 */
262
static int
263 264 265 266 267 268 269 270 271 272
pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
		struct list_head *lseg_list)
{
	struct pnfs_layout_range range = {
		.iomode = IOMODE_ANY,
		.offset = 0,
		.length = NFS4_MAX_UINT64,
	};

	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
273
	return pnfs_mark_matching_lsegs_invalid(lo, lseg_list, &range, 0);
274 275
}

276 277 278 279 280 281 282 283
static int
pnfs_iomode_to_fail_bit(u32 iomode)
{
	return iomode == IOMODE_RW ?
		NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
}

static void
284
pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
285
{
286
	lo->plh_retry_timestamp = jiffies;
287
	if (!test_and_set_bit(fail_bit, &lo->plh_flags))
288 289 290 291 292 293 294 295 296 297 298 299 300 301
		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;
302 303 304 305 306 307
	struct pnfs_layout_range range = {
		.iomode = iomode,
		.offset = 0,
		.length = NFS4_MAX_UINT64,
	};
	LIST_HEAD(head);
308 309 310

	spin_lock(&inode->i_lock);
	pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
311
	pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
312
	spin_unlock(&inode->i_lock);
313
	pnfs_free_lseg_list(&head);
314 315 316 317 318 319 320
	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)
{
321
	unsigned long start, end;
322 323 324
	int fail_bit = pnfs_iomode_to_fail_bit(iomode);

	if (test_bit(fail_bit, &lo->plh_flags) == 0)
325 326 327 328 329
		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 */
330
		pnfs_layout_clear_fail_bit(lo, fail_bit);
331 332 333
		return false;
	}
	return true;
334 335
}

336 337 338
static void
init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
{
339
	INIT_LIST_HEAD(&lseg->pls_list);
340
	INIT_LIST_HEAD(&lseg->pls_lc_list);
341 342 343
	atomic_set(&lseg->pls_refcount, 1);
	smp_mb();
	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
344
	lseg->pls_layout = lo;
345 346
}

347
static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
348
{
349
	struct inode *ino = lseg->pls_layout->plh_inode;
350

351
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
352 353
}

F
Fred Isaman 已提交
354
static void
355 356
pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_segment *lseg)
F
Fred Isaman 已提交
357
{
358
	struct inode *inode = lo->plh_inode;
F
Fred Isaman 已提交
359

360
	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
F
Fred Isaman 已提交
361
	list_del_init(&lseg->pls_list);
362 363
	/* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
	atomic_dec(&lo->plh_refcount);
364 365
	if (list_empty(&lo->plh_segs)) {
		set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
366
		clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
367
	}
F
Fred Isaman 已提交
368 369 370
	rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
}

371
void
372
pnfs_put_lseg(struct pnfs_layout_segment *lseg)
373
{
374
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
375 376 377 378 379
	struct inode *inode;

	if (!lseg)
		return;

380 381 382
	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
		atomic_read(&lseg->pls_refcount),
		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
383

384 385
	lo = lseg->pls_layout;
	inode = lo->plh_inode;
386

F
Fred Isaman 已提交
387
	if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
388 389 390 391
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
			spin_unlock(&inode->i_lock);
			return;
		}
392
		pnfs_get_layout_hdr(lo);
393 394 395 396
		pnfs_layout_remove_lseg(lo, lseg);
		spin_unlock(&inode->i_lock);
		pnfs_free_lseg(lseg);
		pnfs_put_layout_hdr(lo);
397 398
	}
}
399
EXPORT_SYMBOL_GPL(pnfs_put_lseg);
400

401
static void pnfs_free_lseg_async_work(struct work_struct *work)
402 403
{
	struct pnfs_layout_segment *lseg;
404
	struct pnfs_layout_hdr *lo;
405 406

	lseg = container_of(work, struct pnfs_layout_segment, pls_work);
407
	lo = lseg->pls_layout;
408

409 410
	pnfs_free_lseg(lseg);
	pnfs_put_layout_hdr(lo);
411 412
}

413
static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
414
{
415
	INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
416 417
	schedule_work(&lseg->pls_work);
}
418 419 420 421 422 423 424 425 426 427 428 429 430 431

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;
432 433
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
			return;
434 435 436 437 438 439
		pnfs_get_layout_hdr(lo);
		pnfs_layout_remove_lseg(lo, lseg);
		pnfs_free_lseg_async(lseg);
	}
}
EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
440

441
static u64
442 443 444 445 446 447 448 449 450 451 452 453 454 455 456
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
 *           [----------------)
 */
457
static bool
458
pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
459
		 const struct pnfs_layout_range *l2)
460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475
{
	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
 *                              [----------------)
 */
476
static bool
477
pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
478
		    const struct pnfs_layout_range *l2)
479 480 481 482 483 484 485 486 487 488
{
	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);
}

489
static bool
490 491
should_free_lseg(const struct pnfs_layout_range *lseg_range,
		 const struct pnfs_layout_range *recall_range)
492
{
493 494
	return (recall_range->iomode == IOMODE_ANY ||
		lseg_range->iomode == recall_range->iomode) &&
495
	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
496 497
}

498 499 500 501 502 503 504 505 506 507
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;
}

508 509 510 511 512 513 514 515 516 517 518
/* 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
519 520
		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
			atomic_read(&lseg->pls_refcount));
521
		if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
F
Fred Isaman 已提交
522
			rv = 1;
523 524 525 526
	}
	return rv;
}

527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549
/*
 * Compare 2 layout stateid sequence ids, to see which is newer,
 * taking into account wraparound issues.
 */
static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
{
	return (s32)(s1 - s2) > 0;
}

/**
 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
 * @lo: layout header containing the lsegs
 * @tmp_list: list head where doomed lsegs should go
 * @recall_range: optional recall range argument to match (may be NULL)
 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
 *
 * Walk the list of lsegs in the layout header, and tear down any that should
 * be destroyed. If "recall_range" is specified then the segment must match
 * that range. If "seq" is non-zero, then only match segments that were handed
 * out at or before that sequence.
 *
 * Returns number of matching invalid lsegs remaining in list after scanning
 * it and purging them.
550
 */
F
Fred Isaman 已提交
551
int
552
pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
553
			    struct list_head *tmp_list,
554 555
			    const struct pnfs_layout_range *recall_range,
			    u32 seq)
556 557
{
	struct pnfs_layout_segment *lseg, *next;
558
	int remaining = 0;
559 560 561

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

562
	if (list_empty(&lo->plh_segs))
563
		return 0;
564
	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
565 566
		if (!recall_range ||
		    should_free_lseg(&lseg->pls_range, recall_range)) {
567 568 569
			if (seq && pnfs_seqid_is_newer(lseg->pls_seq, seq))
				continue;
			dprintk("%s: freeing lseg %p iomode %d seq %u"
570
				"offset %llu length %llu\n", __func__,
571 572
				lseg, lseg->pls_range.iomode, lseg->pls_seq,
				lseg->pls_range.offset, lseg->pls_range.length);
573 574
			if (!mark_lseg_invalid(lseg, tmp_list))
				remaining++;
575
		}
576 577
	dprintk("%s:Return %i\n", __func__, remaining);
	return remaining;
578 579
}

580
/* note free_me must contain lsegs from a single layout_hdr */
F
Fred Isaman 已提交
581
void
582
pnfs_free_lseg_list(struct list_head *free_me)
583
{
584
	struct pnfs_layout_segment *lseg, *tmp;
585 586 587 588

	if (list_empty(free_me))
		return;

589
	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
590
		list_del(&lseg->pls_list);
591
		pnfs_free_lseg(lseg);
592 593 594
	}
}

595 596 597 598
void
pnfs_destroy_layout(struct nfs_inode *nfsi)
{
	struct pnfs_layout_hdr *lo;
599
	LIST_HEAD(tmp_list);
600 601 602 603

	spin_lock(&nfsi->vfs_inode.i_lock);
	lo = nfsi->layout;
	if (lo) {
604
		pnfs_get_layout_hdr(lo);
605
		pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
606 607 608 609 610 611 612
		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);
613
}
614
EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
615

616 617 618
static bool
pnfs_layout_add_bulk_destroy_list(struct inode *inode,
		struct list_head *layout_list)
619 620
{
	struct pnfs_layout_hdr *lo;
621
	bool ret = false;
622

623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674
	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;
	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;
675 676 677

		pnfs_layoutcommit_inode(inode, false);

678 679
		spin_lock(&inode->i_lock);
		list_del_init(&lo->plh_bulk_destroy);
680 681 682
		if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
			if (is_bulk_recall)
				set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
683
			ret = -EAGAIN;
684
		}
685 686
		spin_unlock(&inode->i_lock);
		pnfs_free_lseg_list(&lseg_list);
687 688
		/* Free all lsegs that are attached to commit buckets */
		nfs_commit_inode(inode, 0);
689 690 691 692 693 694 695 696 697 698 699 700 701
		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);
702

703
	spin_lock(&clp->cl_lock);
704
	rcu_read_lock();
705
restart:
706
	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
707 708 709 710 711 712
		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
			continue;
		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
				server,
				&layout_list) != 0)
			goto restart;
713 714
	}
	rcu_read_unlock();
715 716
	spin_unlock(&clp->cl_lock);

717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736
	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;
737
	}
738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756
	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);
757 758
}

759
/* update lo->plh_stateid with new if is more recent */
F
Fred Isaman 已提交
760 761 762
void
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
			bool update_barrier)
763
{
764 765
	u32 oldseq, newseq, new_barrier;
	int empty = list_empty(&lo->plh_segs);
766

767 768
	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
	newseq = be32_to_cpu(new->seqid);
769
	if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
770
		nfs4_stateid_copy(&lo->plh_stateid, new);
F
Fred Isaman 已提交
771
		if (update_barrier) {
772
			new_barrier = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
773 774
		} else {
			/* Because of wraparound, we want to keep the barrier
775
			 * "close" to the current seqids.
F
Fred Isaman 已提交
776
			 */
777
			new_barrier = newseq - atomic_read(&lo->plh_outstanding);
F
Fred Isaman 已提交
778
		}
779 780
		if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
			lo->plh_barrier = new_barrier;
F
Fred Isaman 已提交
781
	}
782 783
}

784
static bool
785 786
pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
		const nfs4_stateid *stateid)
F
Fred Isaman 已提交
787
{
788
	u32 seqid = be32_to_cpu(stateid->seqid);
789

790 791 792 793 794
	return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
}

/* lget is set to 1 if called from inside send_layoutget call chain */
static bool
795
pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
796
{
F
Fred Isaman 已提交
797
	return lo->plh_block_lgets ||
798
		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
799 800
}

801
/*
802 803 804 805 806
 * Get layout from server.
 *    for now, assume that whole file layouts are requested.
 *    arg->offset: 0
 *    arg->length: all ones
 */
807 808 809
static struct pnfs_layout_segment *
send_layoutget(struct pnfs_layout_hdr *lo,
	   struct nfs_open_context *ctx,
810
	   nfs4_stateid *stateid,
811
	   const struct pnfs_layout_range *range,
812
	   long *timeout, gfp_t gfp_flags)
813
{
814
	struct inode *ino = lo->plh_inode;
815 816
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs4_layoutget *lgp;
817
	loff_t i_size;
818 819

	dprintk("--> %s\n", __func__);
820

821 822 823 824 825
	/*
	 * Synchronously retrieve layout information from server and
	 * store in lseg. If we race with a concurrent seqid morphing
	 * op, then re-send the LAYOUTGET.
	 */
826 827 828 829 830 831 832 833 834 835 836 837 838 839
	lgp = kzalloc(sizeof(*lgp), gfp_flags);
	if (lgp == NULL)
		return ERR_PTR(-ENOMEM);

	i_size = i_size_read(ino);

	lgp->args.minlength = PAGE_SIZE;
	if (lgp->args.minlength > range->length)
		lgp->args.minlength = range->length;
	if (range->iomode == IOMODE_READ) {
		if (range->offset >= i_size)
			lgp->args.minlength = 0;
		else if (i_size - range->offset < lgp->args.minlength)
			lgp->args.minlength = i_size - range->offset;
840
	}
841 842 843 844 845
	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
	pnfs_copy_range(&lgp->args.range, range);
	lgp->args.type = server->pnfs_curr_ld->id;
	lgp->args.inode = ino;
	lgp->args.ctx = get_nfs_open_context(ctx);
846
	nfs4_stateid_copy(&lgp->args.stateid, stateid);
847 848
	lgp->gfp_flags = gfp_flags;
	lgp->cred = lo->plh_lc_cred;
849

850
	return nfs4_proc_layoutget(lgp, timeout, gfp_flags);
851 852
}

853 854 855 856 857 858 859 860 861 862 863 864 865 866 867
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);
	}
}

868 869 870 871 872
void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
{
	clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
	smp_mb__after_atomic();
	wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
873
	rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
874 875
}

876 877 878 879 880 881
static bool
pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo)
{
	if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
		return false;
	lo->plh_return_iomode = 0;
882
	lo->plh_return_seq = 0;
883
	pnfs_get_layout_hdr(lo);
884
	clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
885 886 887
	return true;
}

888
static int
889
pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
890
		       enum pnfs_iomode iomode, bool sync)
891 892 893 894 895
{
	struct inode *ino = lo->plh_inode;
	struct nfs4_layoutreturn *lrp;
	int status = 0;

896
	lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
897 898 899
	if (unlikely(lrp == NULL)) {
		status = -ENOMEM;
		spin_lock(&ino->i_lock);
900
		pnfs_clear_layoutreturn_waitbit(lo);
901 902 903 904 905
		spin_unlock(&ino->i_lock);
		pnfs_put_layout_hdr(lo);
		goto out;
	}

906
	nfs4_stateid_copy(&lrp->args.stateid, stateid);
907 908
	lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
	lrp->args.inode = ino;
P
Peng Tao 已提交
909 910 911
	lrp->args.range.iomode = iomode;
	lrp->args.range.offset = 0;
	lrp->args.range.length = NFS4_MAX_UINT64;
912 913 914 915
	lrp->args.layout = lo;
	lrp->clp = NFS_SERVER(ino)->nfs_client;
	lrp->cred = lo->plh_lc_cred;

916
	status = nfs4_proc_layoutreturn(lrp, sync);
917 918 919 920 921
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}

922 923 924 925 926 927
/* Return true if layoutreturn is needed */
static bool
pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
{
	struct pnfs_layout_segment *s;

928
	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
929 930 931 932 933 934 935 936 937 938 939 940 941 942 943
		return false;

	/* Defer layoutreturn until all lsegs are done */
	list_for_each_entry(s, &lo->plh_segs, pls_list) {
		if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
			return false;
	}

	return true;
}

static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
{
	struct inode *inode= lo->plh_inode;

944
	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
945 946 947 948 949 950 951 952
		return;
	spin_lock(&inode->i_lock);
	if (pnfs_layout_need_return(lo)) {
		nfs4_stateid stateid;
		enum pnfs_iomode iomode;
		bool send;

		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
953
		stateid.seqid = cpu_to_be32(lo->plh_return_seq);
954 955 956 957 958 959 960 961 962 963 964
		iomode = lo->plh_return_iomode;
		send = pnfs_prepare_layoutreturn(lo);
		spin_unlock(&inode->i_lock);
		if (send) {
			/* Send an async layoutreturn so we dont deadlock */
			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
		}
	} else
		spin_unlock(&inode->i_lock);
}

965 966 967 968 969 970 971 972
/*
 * 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
973 974 975 976 977 978 979
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;
980
	int status = 0, empty;
981
	bool send;
B
Benny Halevy 已提交
982

983
	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
B
Benny Halevy 已提交
984 985 986

	spin_lock(&ino->i_lock);
	lo = nfsi->layout;
987
	if (!lo) {
B
Benny Halevy 已提交
988
		spin_unlock(&ino->i_lock);
989 990
		dprintk("NFS: %s no layout to return\n", __func__);
		goto out;
B
Benny Halevy 已提交
991
	}
992
	nfs4_stateid_copy(&stateid, &nfsi->layout->plh_stateid);
B
Benny Halevy 已提交
993
	/* Reference matched in nfs4_layoutreturn_release */
994
	pnfs_get_layout_hdr(lo);
995
	empty = list_empty(&lo->plh_segs);
996
	pnfs_clear_layoutcommit(ino, &tmp_list);
997
	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
998 999 1000 1001 1002 1003 1004 1005 1006 1007

	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);
	}

1008 1009 1010 1011
	/* Don't send a LAYOUTRETURN if list was initially empty */
	if (empty) {
		spin_unlock(&ino->i_lock);
		dprintk("NFS: %s no layout segments to return\n", __func__);
1012
		goto out_put_layout_hdr;
1013
	}
1014 1015

	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1016
	send = pnfs_prepare_layoutreturn(lo);
B
Benny Halevy 已提交
1017 1018
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);
1019
	if (send)
1020
		status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1021 1022
out_put_layout_hdr:
	pnfs_put_layout_hdr(lo);
B
Benny Halevy 已提交
1023 1024 1025 1026
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}
1027
EXPORT_SYMBOL_GPL(_pnfs_return_layout);
B
Benny Halevy 已提交
1028

1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055
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
1056 1057
bool pnfs_roc(struct inode *ino)
{
1058 1059 1060
	struct nfs_inode *nfsi = NFS_I(ino);
	struct nfs_open_context *ctx;
	struct nfs4_state *state;
F
Fred Isaman 已提交
1061 1062
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg, *tmp;
1063
	nfs4_stateid stateid;
F
Fred Isaman 已提交
1064
	LIST_HEAD(tmp_list);
1065
	bool found = false, layoutreturn = false, roc = false;
F
Fred Isaman 已提交
1066 1067

	spin_lock(&ino->i_lock);
1068
	lo = nfsi->layout;
P
Peng Tao 已提交
1069
	if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
1070 1071
		goto out_noroc;

1072
	/* no roc if we hold a delegation */
1073 1074 1075 1076 1077 1078 1079 1080 1081 1082
	if (nfs4_check_delegation(ino, FMODE_READ))
		goto out_noroc;

	list_for_each_entry(ctx, &nfsi->open_files, list) {
		state = ctx->state;
		/* Don't return layout if there is open file state */
		if (state != NULL && state->state != 0)
			goto out_noroc;
	}

1083
	nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1084
	/* always send layoutreturn if being marked so */
1085
	if (test_and_clear_bit(NFS_LAYOUT_RETURN_REQUESTED,
1086 1087 1088
				   &lo->plh_flags))
		layoutreturn = pnfs_prepare_layoutreturn(lo);

F
Fred Isaman 已提交
1089
	list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
1090 1091
		/* If we are sending layoutreturn, invalidate all valid lsegs */
		if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
F
Fred Isaman 已提交
1092 1093 1094
			mark_lseg_invalid(lseg, &tmp_list);
			found = true;
		}
1095
	/* ROC in two conditions:
1096 1097 1098
	 * 1. there are ROC lsegs
	 * 2. we don't send layoutreturn
	 */
1099 1100 1101
	if (found && !layoutreturn) {
		/* lo ref dropped in pnfs_roc_release() */
		pnfs_get_layout_hdr(lo);
1102
		roc = true;
1103
	}
F
Fred Isaman 已提交
1104

1105
out_noroc:
F
Fred Isaman 已提交
1106
	spin_unlock(&ino->i_lock);
1107 1108 1109
	pnfs_free_lseg_list(&tmp_list);
	pnfs_layoutcommit_inode(ino, true);
	if (layoutreturn)
1110
		pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1111
	return roc;
F
Fred Isaman 已提交
1112 1113 1114 1115 1116 1117 1118 1119
}

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

	spin_lock(&ino->i_lock);
	lo = NFS_I(ino)->layout;
1120
	pnfs_clear_layoutreturn_waitbit(lo);
1121 1122 1123 1124 1125 1126
	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
1127 1128 1129 1130 1131 1132 1133 1134
}

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;
1135
	pnfs_mark_layout_returned_if_empty(lo);
1136
	if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
F
Fred Isaman 已提交
1137 1138
		lo->plh_barrier = barrier;
	spin_unlock(&ino->i_lock);
1139
	trace_nfs4_layoutreturn_on_close(ino, 0);
F
Fred Isaman 已提交
1140 1141
}

1142
void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier)
F
Fred Isaman 已提交
1143 1144
{
	struct nfs_inode *nfsi = NFS_I(ino);
1145 1146
	struct pnfs_layout_hdr *lo;
	u32 current_seqid;
F
Fred Isaman 已提交
1147 1148

	spin_lock(&ino->i_lock);
1149 1150
	lo = nfsi->layout;
	current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
F
Fred Isaman 已提交
1151

1152 1153 1154 1155
	/* 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);
F
Fred Isaman 已提交
1156 1157 1158
	spin_unlock(&ino->i_lock);
}

1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178
bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
{
	struct nfs_inode *nfsi = NFS_I(ino);
        struct pnfs_layout_hdr *lo;
        bool sleep = false;

	/* we might not have grabbed lo reference. so need to check under
	 * i_lock */
        spin_lock(&ino->i_lock);
        lo = nfsi->layout;
        if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
                sleep = true;
        spin_unlock(&ino->i_lock);

        if (sleep)
                rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);

        return sleep;
}

1179 1180 1181 1182 1183 1184
/*
 * 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
1185
pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1186
	   const struct pnfs_layout_range *l2)
1187
{
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199
	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;

1200
	/* read > read/write */
1201
	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1202 1203
}

1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225
static bool
pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
		const struct pnfs_layout_range *l2)
{
	return pnfs_lseg_range_cmp(l1, l2) > 0;
}

static bool
pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
		struct pnfs_layout_segment *old)
{
	return false;
}

void
pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
		   struct pnfs_layout_segment *lseg,
		   bool (*is_after)(const struct pnfs_layout_range *,
			   const struct pnfs_layout_range *),
		   bool (*do_merge)(struct pnfs_layout_segment *,
			   struct pnfs_layout_segment *),
		   struct list_head *free_me)
1226
{
1227
	struct pnfs_layout_segment *lp, *tmp;
1228

1229 1230
	dprintk("%s:Begin\n", __func__);

1231 1232 1233 1234 1235 1236 1237 1238
	list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
		if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
			continue;
		if (do_merge(lseg, lp)) {
			mark_lseg_invalid(lp, free_me);
			continue;
		}
		if (is_after(&lseg->pls_range, &lp->pls_range))
1239
			continue;
1240
		list_add_tail(&lseg->pls_list, &lp->pls_list);
1241 1242 1243
		dprintk("%s: inserted lseg %p "
			"iomode %d offset %llu length %llu before "
			"lp %p iomode %d offset %llu length %llu\n",
1244 1245 1246 1247
			__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);
1248
		goto out;
1249
	}
1250 1251 1252 1253 1254 1255
	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:
1256
	pnfs_get_layout_hdr(lo);
1257 1258

	dprintk("%s:Return\n", __func__);
1259
}
1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277
EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);

static void
pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
		   struct pnfs_layout_segment *lseg,
		   struct list_head *free_me)
{
	struct inode *inode = lo->plh_inode;
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;

	if (ld->add_lseg != NULL)
		ld->add_lseg(lo, lseg, free_me);
	else
		pnfs_generic_layout_insert_lseg(lo, lseg,
				pnfs_lseg_range_is_after,
				pnfs_lseg_no_merge,
				free_me);
}
1278 1279

static struct pnfs_layout_hdr *
1280 1281 1282
alloc_init_layout_hdr(struct inode *ino,
		      struct nfs_open_context *ctx,
		      gfp_t gfp_flags)
1283 1284 1285
{
	struct pnfs_layout_hdr *lo;

1286
	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1287 1288
	if (!lo)
		return NULL;
1289
	atomic_set(&lo->plh_refcount, 1);
1290 1291
	INIT_LIST_HEAD(&lo->plh_layouts);
	INIT_LIST_HEAD(&lo->plh_segs);
1292
	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1293
	lo->plh_inode = ino;
1294
	lo->plh_lc_cred = get_rpccred(ctx->cred);
1295
	lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
1296 1297 1298 1299
	return lo;
}

static struct pnfs_layout_hdr *
1300 1301 1302
pnfs_find_alloc_layout(struct inode *ino,
		       struct nfs_open_context *ctx,
		       gfp_t gfp_flags)
1303 1304
	__releases(&ino->i_lock)
	__acquires(&ino->i_lock)
1305 1306 1307 1308 1309 1310
{
	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);

1311 1312
	if (nfsi->layout != NULL)
		goto out_existing;
1313
	spin_unlock(&ino->i_lock);
1314
	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1315 1316
	spin_lock(&ino->i_lock);

1317
	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1318
		nfsi->layout = new;
1319
		return new;
1320 1321
	} else if (new != NULL)
		pnfs_free_layout_hdr(new);
1322 1323
out_existing:
	pnfs_get_layout_hdr(nfsi->layout);
1324 1325 1326
	return nfsi->layout;
}

1327 1328
/*
 * iomode matching rules:
1329 1330 1331 1332 1333 1334 1335 1336 1337 1338
 * iomode	lseg	strict match
 *                      iomode
 * -----	-----	------ -----
 * ANY		READ	N/A    true
 * ANY		RW	N/A    true
 * RW		READ	N/A    false
 * RW		RW	N/A    true
 * READ		READ	N/A    true
 * READ		RW	true   false
 * READ		RW	false  true
1339
 */
1340
static bool
1341
pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1342 1343
		 const struct pnfs_layout_range *range,
		 bool strict_iomode)
1344
{
1345 1346 1347 1348
	struct pnfs_layout_range range1;

	if ((range->iomode == IOMODE_RW &&
	     ls_range->iomode != IOMODE_RW) ||
1349 1350
	    (range->iomode != ls_range->iomode &&
	     strict_iomode == true) ||
1351
	    !pnfs_lseg_range_intersecting(ls_range, range))
1352 1353 1354 1355 1356
		return 0;

	/* range1 covers only the first byte in the range */
	range1 = *range;
	range1.length = 1;
1357
	return pnfs_lseg_range_contained(ls_range, &range1);
1358 1359 1360 1361 1362
}

/*
 * lookup range in layout
 */
1363
static struct pnfs_layout_segment *
1364
pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1365 1366
		struct pnfs_layout_range *range,
		bool strict_iomode)
1367
{
1368 1369 1370 1371
	struct pnfs_layout_segment *lseg, *ret = NULL;

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

1372
	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1373
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1374
		    !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
1375 1376
		    pnfs_lseg_range_match(&lseg->pls_range, range,
					  strict_iomode)) {
1377
			ret = pnfs_get_lseg(lseg);
1378 1379 1380 1381 1382
			break;
		}
	}

	dprintk("%s:Return lseg %p ref %d\n",
1383
		__func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1384
	return ret;
1385 1386
}

1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461
/*
 * 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;
}

1462 1463 1464 1465 1466 1467 1468 1469
static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
{
	/*
	 * send layoutcommit as it can hold up layoutreturn due to lseg
	 * reference
	 */
	pnfs_layoutcommit_inode(lo->plh_inode, false);
	return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1470
				   nfs_wait_bit_killable,
1471 1472 1473
				   TASK_UNINTERRUPTIBLE);
}

1474 1475 1476 1477 1478 1479 1480 1481 1482
static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
{
	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);
}

1483 1484 1485 1486
/*
 * Layout segment is retreived from the server if not cached.
 * The appropriate layout segment is referenced and returned to the caller.
 */
1487
struct pnfs_layout_segment *
1488 1489
pnfs_update_layout(struct inode *ino,
		   struct nfs_open_context *ctx,
1490 1491
		   loff_t pos,
		   u64 count,
1492
		   enum pnfs_iomode iomode,
1493
		   bool strict_iomode,
1494
		   gfp_t gfp_flags)
1495
{
1496 1497 1498 1499 1500
	struct pnfs_layout_range arg = {
		.iomode = iomode,
		.offset = pos,
		.length = count,
	};
1501
	unsigned pg_offset, seq;
1502 1503
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs_client *clp = server->nfs_client;
1504
	struct pnfs_layout_hdr *lo = NULL;
1505
	struct pnfs_layout_segment *lseg = NULL;
1506 1507 1508
	nfs4_stateid stateid;
	long timeout = 0;
	unsigned long giveup = jiffies + rpc_get_timeout(server->client);
1509
	bool first;
1510

1511
	if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1512
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1513
				 PNFS_UPDATE_LAYOUT_NO_PNFS);
1514
		goto out;
1515
	}
1516

1517
	if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
1518
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1519
				 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
1520
		goto out;
1521
	}
1522

1523
	if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1524
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1525
				 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1526
		goto out;
1527
	}
1528

1529 1530
lookup_again:
	first = false;
1531
	spin_lock(&ino->i_lock);
1532
	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1533 1534
	if (lo == NULL) {
		spin_unlock(&ino->i_lock);
1535
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1536
				 PNFS_UPDATE_LAYOUT_NOMEM);
1537 1538
		goto out;
	}
1539

F
Fred Isaman 已提交
1540
	/* Do we even need to bother with this? */
1541
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1542
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1543
				 PNFS_UPDATE_LAYOUT_BULK_RECALL);
F
Fred Isaman 已提交
1544
		dprintk("%s matches recall, use MDS\n", __func__);
1545 1546 1547 1548
		goto out_unlock;
	}

	/* if LAYOUTGET already failed once we don't try again */
1549
	if (pnfs_layout_io_test_failed(lo, iomode)) {
1550
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1551
				 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1552
		goto out_unlock;
1553
	}
1554

1555
	lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572
	if (lseg) {
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
				PNFS_UPDATE_LAYOUT_FOUND_CACHED);
		goto out_unlock;
	}

	if (!nfs4_valid_open_stateid(ctx->state)) {
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
				PNFS_UPDATE_LAYOUT_INVALID_OPEN);
		goto out_unlock;
	}

	/*
	 * Choose a stateid for the LAYOUTGET. If we don't have a layout
	 * stateid, or it has been invalidated, then we must use the open
	 * stateid.
	 */
1573
	if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
1574 1575 1576

		/*
		 * The first layoutget for the file. Need to serialize per
1577 1578 1579 1580 1581 1582 1583 1584
		 * 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);
1585
			dprintk("%s retrying\n", __func__);
1586 1587
			goto lookup_again;
		}
1588 1589 1590 1591 1592 1593

		first = true;
		do {
			seq = read_seqbegin(&ctx->state->seqlock);
			nfs4_stateid_copy(&stateid, &ctx->state->stateid);
		} while (read_seqretry(&ctx->state->seqlock, seq));
1594
	} else {
1595
		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1596
	}
1597

1598 1599 1600 1601
	/*
	 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
	 * for LAYOUTRETURN even if first is true.
	 */
1602
	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1603 1604 1605
		spin_unlock(&ino->i_lock);
		dprintk("%s wait for layoutreturn\n", __func__);
		if (pnfs_prepare_to_retry_layoutget(lo)) {
1606 1607
			if (first)
				pnfs_clear_first_layoutget(lo);
1608 1609
			pnfs_put_layout_hdr(lo);
			dprintk("%s retrying\n", __func__);
1610 1611
			trace_pnfs_update_layout(ino, pos, count, iomode, lo,
					lseg, PNFS_UPDATE_LAYOUT_RETRY);
1612 1613
			goto lookup_again;
		}
1614
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1615
				PNFS_UPDATE_LAYOUT_RETURN);
1616 1617 1618
		goto out_put_layout_hdr;
	}

1619
	if (pnfs_layoutgets_blocked(lo)) {
1620
		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1621
				PNFS_UPDATE_LAYOUT_BLOCKED);
1622
		goto out_unlock;
1623
	}
1624
	atomic_inc(&lo->plh_outstanding);
1625
	spin_unlock(&ino->i_lock);
1626

1627
	if (list_empty(&lo->plh_layouts)) {
1628 1629 1630 1631
		/* 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);
1632 1633
		if (list_empty(&lo->plh_layouts))
			list_add_tail(&lo->plh_layouts, &server->layouts);
1634 1635
		spin_unlock(&clp->cl_lock);
	}
1636

1637
	pg_offset = arg.offset & ~PAGE_MASK;
1638 1639 1640 1641
	if (pg_offset) {
		arg.offset -= pg_offset;
		arg.length += pg_offset;
	}
1642
	if (arg.length != NFS4_MAX_UINT64)
1643
		arg.length = PAGE_ALIGN(arg.length);
1644

1645 1646 1647
	lseg = send_layoutget(lo, ctx, &stateid, &arg, &timeout, gfp_flags);
	trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
				 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
1648
	if (IS_ERR(lseg)) {
1649 1650 1651 1652 1653 1654 1655
		switch(PTR_ERR(lseg)) {
		case -ERECALLCONFLICT:
			if (time_after(jiffies, giveup))
				lseg = NULL;
			/* Fallthrough */
		case -EAGAIN:
			pnfs_put_layout_hdr(lo);
1656 1657
			if (first)
				pnfs_clear_first_layoutget(lo);
1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668
			if (lseg) {
				trace_pnfs_update_layout(ino, pos, count,
					iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
				goto lookup_again;
			}
			/* Fallthrough */
		default:
			if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
				pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
				lseg = NULL;
			}
1669 1670 1671 1672 1673
		}
	} else {
		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
	}

1674
	atomic_dec(&lo->plh_outstanding);
1675
out_put_layout_hdr:
1676 1677
	if (first)
		pnfs_clear_first_layoutget(lo);
1678
	pnfs_put_layout_hdr(lo);
1679
out:
1680 1681 1682 1683
	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),
1684
			IS_ERR_OR_NULL(lseg) ? "not found" : "found",
1685 1686 1687
			iomode==IOMODE_RW ?  "read/write" : "read-only",
			(unsigned long long)pos,
			(unsigned long long)count);
1688 1689 1690
	return lseg;
out_unlock:
	spin_unlock(&ino->i_lock);
1691
	goto out_put_layout_hdr;
1692
}
1693
EXPORT_SYMBOL_GPL(pnfs_update_layout);
1694

1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714
static bool
pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
{
	switch (range->iomode) {
	case IOMODE_READ:
	case IOMODE_RW:
		break;
	default:
		return false;
	}
	if (range->offset == NFS4_MAX_UINT64)
		return false;
	if (range->length == 0)
		return false;
	if (range->length != NFS4_MAX_UINT64 &&
	    range->length > NFS4_MAX_UINT64 - range->offset)
		return false;
	return true;
}

1715
struct pnfs_layout_segment *
1716 1717 1718 1719 1720
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;
1721
	struct inode *ino = lo->plh_inode;
1722
	LIST_HEAD(free_me);
1723 1724

	if (!pnfs_sanity_check_layout_range(&res->range))
1725
		return ERR_PTR(-EINVAL);
1726 1727

	/* Inject layout blob into I/O device driver */
1728
	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1729
	if (IS_ERR_OR_NULL(lseg)) {
1730
		if (!lseg)
1731 1732 1733 1734 1735
			lseg = ERR_PTR(-ENOMEM);

		dprintk("%s: Could not allocate layout: error %ld\n",
		       __func__, PTR_ERR(lseg));
		return lseg;
1736 1737
	}

1738 1739
	init_lseg(lo, lseg);
	lseg->pls_range = res->range;
1740
	lseg->pls_seq = be32_to_cpu(res->stateid.seqid);
1741

1742
	spin_lock(&ino->i_lock);
1743
	if (pnfs_layoutgets_blocked(lo)) {
F
Fred Isaman 已提交
1744
		dprintk("%s forget reply due to state\n", __func__);
1745
		goto out_forget;
F
Fred Isaman 已提交
1746
	}
1747

1748 1749 1750 1751
	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__);
1752
			goto out_forget;
1753 1754 1755 1756 1757 1758 1759 1760
		}
		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.
		 */
1761
		pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL, 0);
1762 1763 1764 1765

		nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
		lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
	}
1766

1767 1768
	clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);

1769
	pnfs_get_lseg(lseg);
1770
	pnfs_layout_insert_lseg(lo, lseg, &free_me);
1771

P
Peng Tao 已提交
1772
	if (res->return_on_close)
F
Fred Isaman 已提交
1773 1774
		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);

1775
	spin_unlock(&ino->i_lock);
1776
	pnfs_free_lseg_list(&free_me);
1777
	return lseg;
F
Fred Isaman 已提交
1778

1779
out_forget:
F
Fred Isaman 已提交
1780 1781 1782
	spin_unlock(&ino->i_lock);
	lseg->pls_layout = lo;
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1783
	return ERR_PTR(-EAGAIN);
1784 1785
}

1786
static void
1787 1788
pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
			 u32 seq)
1789 1790 1791 1792 1793 1794
{
	if (lo->plh_return_iomode == iomode)
		return;
	if (lo->plh_return_iomode != 0)
		iomode = IOMODE_ANY;
	lo->plh_return_iomode = iomode;
1795
	set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
1796 1797
	if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq))
		lo->plh_return_seq = seq;
1798 1799
}

1800 1801 1802 1803 1804 1805 1806 1807 1808 1809
/**
 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
 * @lo: pointer to layout header
 * @tmp_list: list header to be used with pnfs_free_lseg_list()
 * @return_range: describe layout segment ranges to be returned
 *
 * This function is mainly intended for use by layoutrecall. It attempts
 * to free the layout segment immediately, or else to mark it for return
 * as soon as its reference count drops to zero.
 */
1810
int
1811 1812
pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
				struct list_head *tmp_list,
1813 1814
				const struct pnfs_layout_range *return_range,
				u32 seq)
1815 1816
{
	struct pnfs_layout_segment *lseg, *next;
1817
	int remaining = 0;
1818 1819 1820 1821

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

	if (list_empty(&lo->plh_segs))
1822
		return 0;
1823

1824
	assert_spin_locked(&lo->plh_inode->i_lock);
1825 1826 1827 1828 1829 1830 1831 1832

	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
		if (should_free_lseg(&lseg->pls_range, return_range)) {
			dprintk("%s: marking lseg %p iomode %d "
				"offset %llu length %llu\n", __func__,
				lseg, lseg->pls_range.iomode,
				lseg->pls_range.offset,
				lseg->pls_range.length);
1833 1834 1835
			if (mark_lseg_invalid(lseg, tmp_list))
				continue;
			remaining++;
1836 1837
			set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
		}
1838 1839 1840 1841

	if (remaining)
		pnfs_set_plh_return_info(lo, return_range->iomode, seq);

1842
	return remaining;
1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854
}

void pnfs_error_mark_layout_for_return(struct inode *inode,
				       struct pnfs_layout_segment *lseg)
{
	struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
	struct pnfs_layout_range range = {
		.iomode = lseg->pls_range.iomode,
		.offset = 0,
		.length = NFS4_MAX_UINT64,
	};
	LIST_HEAD(free_me);
1855
	bool return_now = false;
1856 1857

	spin_lock(&inode->i_lock);
1858
	pnfs_set_plh_return_info(lo, range.iomode, lseg->pls_seq);
1859 1860 1861 1862 1863
	/*
	 * mark all matching lsegs so that we are sure to have no live
	 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
	 * for how it works.
	 */
1864 1865
	if (!pnfs_mark_matching_lsegs_return(lo, &free_me,
						&range, lseg->pls_seq)) {
1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877
		nfs4_stateid stateid;
		enum pnfs_iomode iomode = lo->plh_return_iomode;

		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
		return_now = pnfs_prepare_layoutreturn(lo);
		spin_unlock(&inode->i_lock);
		if (return_now)
			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
	} else {
		spin_unlock(&inode->i_lock);
		nfs_commit_inode(inode, 0);
	}
1878 1879 1880 1881
	pnfs_free_lseg_list(&free_me);
}
EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);

1882 1883 1884
void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
P
Peng Tao 已提交
1885 1886
	u64 rd_size = req->wb_bytes;

1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897
	if (pgio->pg_lseg == NULL) {
		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);

		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
						   req->wb_context,
						   req_offset(req),
						   rd_size,
						   IOMODE_READ,
1898
						   false,
1899
						   GFP_KERNEL);
1900 1901 1902 1903 1904
		if (IS_ERR(pgio->pg_lseg)) {
			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
			pgio->pg_lseg = NULL;
			return;
		}
1905
	}
1906 1907
	/* If no lseg, fall back to read through mds */
	if (pgio->pg_lseg == NULL)
1908
		nfs_pageio_reset_read_mds(pgio);
1909

1910 1911 1912 1913
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);

void
1914 1915
pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
			   struct nfs_page *req, u64 wb_size)
1916
{
1917
	if (pgio->pg_lseg == NULL) {
1918 1919 1920 1921 1922
		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
						   req->wb_context,
						   req_offset(req),
						   wb_size,
						   IOMODE_RW,
1923
						   false,
1924
						   GFP_NOFS);
1925 1926 1927 1928 1929 1930
		if (IS_ERR(pgio->pg_lseg)) {
			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
			pgio->pg_lseg = NULL;
			return;
		}
	}
1931 1932
	/* If no lseg, fall back to write through mds */
	if (pgio->pg_lseg == NULL)
1933
		nfs_pageio_reset_write_mds(pgio);
1934 1935 1936
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);

1937 1938 1939 1940 1941 1942 1943 1944 1945 1946
void
pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
{
	if (desc->pg_lseg) {
		pnfs_put_lseg(desc->pg_lseg);
		desc->pg_lseg = NULL;
	}
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);

1947 1948 1949 1950 1951
/*
 * 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
1952 1953
pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
		     struct nfs_page *prev, struct nfs_page *req)
1954
{
1955
	unsigned int size;
1956
	u64 seg_end, req_start, seg_left;
1957 1958 1959 1960

	size = nfs_generic_pg_test(pgio, prev, req);
	if (!size)
		return 0;
1961

1962
	/*
1963 1964 1965 1966 1967
	 * '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.
1968 1969 1970 1971 1972
	 *
	 * Please also note that 'end_offset' is actually the offset of the
	 * first byte that lies outside the pnfs_layout_range. FIXME?
	 *
	 */
1973
	if (pgio->pg_lseg) {
1974 1975 1976
		seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
				     pgio->pg_lseg->pls_range.length);
		req_start = req_offset(req);
1977
		WARN_ON_ONCE(req_start >= seg_end);
1978
		/* start of request is past the last byte of this segment */
1979 1980 1981 1982 1983 1984
		if (req_start >= seg_end) {
			/* reference the new lseg */
			if (pgio->pg_ops->pg_cleanup)
				pgio->pg_ops->pg_cleanup(pgio);
			if (pgio->pg_ops->pg_init)
				pgio->pg_ops->pg_init(pgio, req);
1985
			return 0;
1986
		}
1987 1988 1989 1990 1991 1992

		/* 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;
1993
	}
1994

1995
	return size;
1996
}
1997
EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1998

1999
int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
2000 2001 2002 2003
{
	struct nfs_pageio_descriptor pgio;

	/* Resend all requests through the MDS */
2004 2005
	nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
			      hdr->completion_ops);
2006
	set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
2007
	return nfs_pageio_resend(&pgio, hdr);
2008
}
2009
EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
2010

2011
static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
2012
{
2013 2014 2015

	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2016
	    PNFS_LAYOUTRET_ON_ERROR) {
2017
		pnfs_return_layout(hdr->inode);
2018
	}
2019
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2020
		hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
2021 2022
}

2023 2024 2025
/*
 * Called by non rpc-based layout drivers
 */
2026
void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
2027
{
2028
	if (likely(!hdr->pnfs_error)) {
2029 2030
		pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
				hdr->mds_offset + hdr->res.count);
2031
		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2032 2033 2034
	}
	trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
	if (unlikely(hdr->pnfs_error))
2035 2036
		pnfs_ld_handle_write_error(hdr);
	hdr->mds_ops->rpc_release(hdr);
2037
}
2038
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
2039

2040 2041
static void
pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
2042
		struct nfs_pgio_header *hdr)
2043
{
2044
	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2045

2046
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2047
		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2048
		nfs_pageio_reset_write_mds(desc);
2049
		mirror->pg_recoalesce = 1;
2050
	}
2051
	nfs_pgio_data_destroy(hdr);
2052
	hdr->release(hdr);
2053 2054 2055
}

static enum pnfs_try_status
2056
pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
2057 2058 2059
			const struct rpc_call_ops *call_ops,
			struct pnfs_layout_segment *lseg,
			int how)
2060
{
2061
	struct inode *inode = hdr->inode;
2062 2063 2064
	enum pnfs_try_status trypnfs;
	struct nfs_server *nfss = NFS_SERVER(inode);

2065
	hdr->mds_ops = call_ops;
2066 2067

	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
2068 2069
		inode->i_ino, hdr->args.count, hdr->args.offset, how);
	trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
2070
	if (trypnfs != PNFS_NOT_ATTEMPTED)
2071 2072 2073 2074 2075
		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}

2076
static void
2077 2078
pnfs_do_write(struct nfs_pageio_descriptor *desc,
	      struct nfs_pgio_header *hdr, int how)
2079 2080 2081
{
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2082
	enum pnfs_try_status trypnfs;
2083

2084
	trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2085
	if (trypnfs == PNFS_NOT_ATTEMPTED)
2086
		pnfs_write_through_mds(desc, hdr);
2087 2088
}

2089 2090
static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
{
2091
	pnfs_put_lseg(hdr->lseg);
2092
	nfs_pgio_header_free(hdr);
2093 2094
}

2095 2096 2097
int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
{
2098
	struct nfs_pgio_header *hdr;
2099 2100
	int ret;

2101 2102
	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
	if (!hdr) {
P
Peng Tao 已提交
2103 2104
		desc->pg_error = -ENOMEM;
		return desc->pg_error;
2105
	}
2106
	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2107

2108
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2109
	ret = nfs_generic_pgio(desc, hdr);
2110
	if (!ret)
2111
		pnfs_do_write(desc, hdr, desc->pg_ioflags);
2112

2113
	return ret;
2114 2115 2116
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);

2117
int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2118 2119 2120
{
	struct nfs_pageio_descriptor pgio;

2121
	/* Resend all requests through the MDS */
2122 2123
	nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
	return nfs_pageio_resend(&pgio, hdr);
2124
}
2125
EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2126

2127
static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2128
{
2129 2130
	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2131
	    PNFS_LAYOUTRET_ON_ERROR) {
2132
		pnfs_return_layout(hdr->inode);
2133
	}
2134
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2135
		hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2136 2137
}

2138 2139 2140
/*
 * Called by non rpc-based layout drivers
 */
2141
void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2142
{
2143
	if (likely(!hdr->pnfs_error)) {
2144 2145
		__nfs4_read_done_cb(hdr);
		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2146 2147 2148
	}
	trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
	if (unlikely(hdr->pnfs_error))
2149 2150
		pnfs_ld_handle_read_error(hdr);
	hdr->mds_ops->rpc_release(hdr);
2151 2152 2153
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

2154 2155
static void
pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2156
		struct nfs_pgio_header *hdr)
2157
{
2158
	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2159

2160
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2161
		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2162
		nfs_pageio_reset_read_mds(desc);
2163
		mirror->pg_recoalesce = 1;
2164
	}
2165
	nfs_pgio_data_destroy(hdr);
2166
	hdr->release(hdr);
2167 2168
}

A
Andy Adamson 已提交
2169 2170 2171
/*
 * Call the appropriate parallel I/O subsystem read function.
 */
2172
static enum pnfs_try_status
2173
pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2174 2175
		       const struct rpc_call_ops *call_ops,
		       struct pnfs_layout_segment *lseg)
A
Andy Adamson 已提交
2176
{
2177
	struct inode *inode = hdr->inode;
A
Andy Adamson 已提交
2178 2179 2180
	struct nfs_server *nfss = NFS_SERVER(inode);
	enum pnfs_try_status trypnfs;

2181
	hdr->mds_ops = call_ops;
A
Andy Adamson 已提交
2182 2183

	dprintk("%s: Reading ino:%lu %u@%llu\n",
2184
		__func__, inode->i_ino, hdr->args.count, hdr->args.offset);
A
Andy Adamson 已提交
2185

2186
	trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2187
	if (trypnfs != PNFS_NOT_ATTEMPTED)
A
Andy Adamson 已提交
2188 2189 2190 2191
		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}
A
Andy Adamson 已提交
2192

2193
/* Resend all requests through pnfs. */
2194
void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2195 2196 2197
{
	struct nfs_pageio_descriptor pgio;

2198 2199 2200 2201 2202
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
		nfs_pageio_init_read(&pgio, hdr->inode, false,
					hdr->completion_ops);
		hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
	}
2203 2204 2205
}
EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);

2206
static void
2207
pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2208 2209 2210
{
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2211
	enum pnfs_try_status trypnfs;
2212

2213
	trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2214
	if (trypnfs == PNFS_TRY_AGAIN)
2215 2216
		pnfs_read_resend_pnfs(hdr);
	if (trypnfs == PNFS_NOT_ATTEMPTED || hdr->task.tk_status)
2217
		pnfs_read_through_mds(desc, hdr);
2218 2219
}

2220 2221
static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
{
2222
	pnfs_put_lseg(hdr->lseg);
2223
	nfs_pgio_header_free(hdr);
2224 2225
}

2226 2227 2228
int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
{
2229
	struct nfs_pgio_header *hdr;
2230 2231
	int ret;

2232 2233
	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
	if (!hdr) {
P
Peng Tao 已提交
2234 2235
		desc->pg_error = -ENOMEM;
		return desc->pg_error;
2236
	}
2237
	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2238
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2239
	ret = nfs_generic_pgio(desc, hdr);
2240
	if (!ret)
2241
		pnfs_do_read(desc, hdr);
2242
	return ret;
2243 2244 2245
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);

2246 2247 2248 2249 2250
static void pnfs_clear_layoutcommitting(struct inode *inode)
{
	unsigned long *bitlock = &NFS_I(inode)->flags;

	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2251
	smp_mb__after_atomic();
2252 2253 2254
	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
}

A
Andy Adamson 已提交
2255
/*
2256
 * There can be multiple RW segments.
A
Andy Adamson 已提交
2257
 */
2258
static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
A
Andy Adamson 已提交
2259
{
2260
	struct pnfs_layout_segment *lseg;
A
Andy Adamson 已提交
2261

2262 2263
	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
		if (lseg->pls_range.iomode == IOMODE_RW &&
2264
		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2265 2266
			list_add(&lseg->pls_lc_list, listp);
	}
A
Andy Adamson 已提交
2267 2268
}

2269 2270 2271 2272 2273 2274 2275 2276 2277 2278
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);
	}

2279
	pnfs_clear_layoutcommitting(inode);
2280 2281
}

P
Peng Tao 已提交
2282 2283
void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
{
2284
	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
P
Peng Tao 已提交
2285 2286 2287
}
EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);

A
Andy Adamson 已提交
2288
void
2289 2290
pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
		loff_t end_pos)
A
Andy Adamson 已提交
2291
{
2292
	struct nfs_inode *nfsi = NFS_I(inode);
2293
	bool mark_as_dirty = false;
A
Andy Adamson 已提交
2294

2295
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
2296
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2297
		nfsi->layout->plh_lwb = end_pos;
2298
		mark_as_dirty = true;
A
Andy Adamson 已提交
2299
		dprintk("%s: Set layoutcommit for inode %lu ",
2300
			__func__, inode->i_ino);
2301 2302
	} else if (end_pos > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = end_pos;
2303
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2304
		/* references matched in nfs4_layoutcommit_release */
2305
		pnfs_get_lseg(lseg);
2306
	}
2307
	spin_unlock(&inode->i_lock);
2308
	dprintk("%s: lseg %p end_pos %llu\n",
2309
		__func__, lseg, nfsi->layout->plh_lwb);
2310 2311 2312 2313

	/* 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)
2314
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
2315 2316 2317
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

A
Andy Adamson 已提交
2318 2319 2320 2321 2322 2323
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);
2324
	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
A
Andy Adamson 已提交
2325 2326
}

2327 2328 2329 2330 2331 2332 2333 2334
/*
 * 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
2335
int
2336
pnfs_layoutcommit_inode(struct inode *inode, bool sync)
A
Andy Adamson 已提交
2337
{
2338
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
A
Andy Adamson 已提交
2339 2340 2341
	struct nfs4_layoutcommit_data *data;
	struct nfs_inode *nfsi = NFS_I(inode);
	loff_t end_pos;
2342
	int status;
A
Andy Adamson 已提交
2343

2344
	if (!pnfs_layoutcommit_outstanding(inode))
2345 2346
		return 0;

2347
	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
P
Peng Tao 已提交
2348

2349
	status = -EAGAIN;
P
Peng Tao 已提交
2350
	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2351 2352
		if (!sync)
			goto out;
2353
		status = wait_on_bit_lock_action(&nfsi->flags,
2354 2355 2356
				NFS_INO_LAYOUTCOMMITTING,
				nfs_wait_bit_killable,
				TASK_KILLABLE);
P
Peng Tao 已提交
2357
		if (status)
2358
			goto out;
P
Peng Tao 已提交
2359 2360
	}

2361 2362 2363 2364 2365 2366 2367
	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;
2368
	spin_lock(&inode->i_lock);
2369 2370
	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		goto out_unlock;
2371

2372
	INIT_LIST_HEAD(&data->lseg_list);
2373
	pnfs_list_write_lseg(inode, &data->lseg_list);
A
Andy Adamson 已提交
2374

2375
	end_pos = nfsi->layout->plh_lwb;
A
Andy Adamson 已提交
2376

2377
	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
A
Andy Adamson 已提交
2378 2379 2380
	spin_unlock(&inode->i_lock);

	data->args.inode = inode;
2381
	data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
A
Andy Adamson 已提交
2382 2383 2384 2385 2386 2387
	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);

2388 2389 2390
	if (ld->prepare_layoutcommit) {
		status = ld->prepare_layoutcommit(&data->args);
		if (status) {
2391
			put_rpccred(data->cred);
2392
			spin_lock(&inode->i_lock);
2393 2394
			set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
			if (end_pos > nfsi->layout->plh_lwb)
2395
				nfsi->layout->plh_lwb = end_pos;
2396
			goto out_unlock;
2397 2398 2399 2400
		}
	}


A
Andy Adamson 已提交
2401 2402
	status = nfs4_proc_layoutcommit(data, sync);
out:
P
Peng Tao 已提交
2403 2404
	if (status)
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
2405 2406
	dprintk("<-- %s status %d\n", __func__, status);
	return status;
2407 2408
out_unlock:
	spin_unlock(&inode->i_lock);
P
Peng Tao 已提交
2409
	kfree(data);
2410 2411
clear_layoutcommitting:
	pnfs_clear_layoutcommitting(inode);
P
Peng Tao 已提交
2412
	goto out;
A
Andy Adamson 已提交
2413
}
2414
EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2415

2416 2417 2418 2419 2420 2421 2422
int
pnfs_generic_sync(struct inode *inode, bool datasync)
{
	return pnfs_layoutcommit_inode(inode, true);
}
EXPORT_SYMBOL_GPL(pnfs_generic_sync);

2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433
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;
}
2434

2435
#if IS_ENABLED(CONFIG_NFS_V4_2)
2436
int
2437
pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
2438 2439 2440
{
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
	struct nfs_server *server = NFS_SERVER(inode);
2441
	struct nfs_inode *nfsi = NFS_I(inode);
2442 2443 2444 2445 2446 2447 2448
	struct nfs42_layoutstat_data *data;
	struct pnfs_layout_hdr *hdr;
	int status = 0;

	if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
		goto out;

2449 2450 2451
	if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
		goto out;

2452 2453 2454
	if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
		goto out;

2455 2456 2457
	spin_lock(&inode->i_lock);
	if (!NFS_I(inode)->layout) {
		spin_unlock(&inode->i_lock);
2458
		goto out_clear_layoutstats;
2459 2460 2461 2462 2463
	}
	hdr = NFS_I(inode)->layout;
	pnfs_get_layout_hdr(hdr);
	spin_unlock(&inode->i_lock);

2464
	data = kzalloc(sizeof(*data), gfp_flags);
2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486
	if (!data) {
		status = -ENOMEM;
		goto out_put;
	}

	data->args.fh = NFS_FH(inode);
	data->args.inode = inode;
	nfs4_stateid_copy(&data->args.stateid, &hdr->plh_stateid);
	status = ld->prepare_layoutstats(&data->args);
	if (status)
		goto out_free;

	status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);

out:
	dprintk("%s returns %d\n", __func__, status);
	return status;

out_free:
	kfree(data);
out_put:
	pnfs_put_layout_hdr(hdr);
2487
out_clear_layoutstats:
2488 2489 2490
	smp_mb__before_atomic();
	clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
	smp_mb__after_atomic();
2491 2492 2493
	goto out;
}
EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
2494
#endif
2495 2496 2497 2498

unsigned int layoutstats_timer;
module_param(layoutstats_timer, uint, 0644);
EXPORT_SYMBOL_GPL(layoutstats_timer);