pnfs.c 64.5 KB
Newer Older
R
Ricardo Labiaga 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
/*
 *  pNFS functions to call and manage layout drivers.
 *
 *  Copyright (c) 2002 [year of first publication]
 *  The Regents of the University of Michigan
 *  All Rights Reserved
 *
 *  Dean Hildebrand <dhildebz@umich.edu>
 *
 *  Permission is granted to use, copy, create derivative works, and
 *  redistribute this software and such derivative works for any purpose,
 *  so long as the name of the University of Michigan is not used in
 *  any advertising or publicity pertaining to the use or distribution
 *  of this software without specific, written prior authorization. If
 *  the above copyright notice or any other identification of the
 *  University of Michigan is included in any copy of any portion of
 *  this software, then the disclaimer below must also be included.
 *
 *  This software is provided as is, without representation or warranty
 *  of any kind either express or implied, including without limitation
 *  the implied warranties of merchantability, fitness for a particular
 *  purpose, or noninfringement.  The Regents of the University of
 *  Michigan shall not be liable for any damages, including special,
 *  indirect, incidental, or consequential damages, with respect to any
 *  claim arising out of or in connection with the use of the software,
 *  even if it has been or is hereafter advised of the possibility of
 *  such damages.
 */

#include <linux/nfs_fs.h>
31
#include <linux/nfs_page.h>
32
#include <linux/module.h>
33
#include "internal.h"
R
Ricardo Labiaga 已提交
34
#include "pnfs.h"
A
Andy Adamson 已提交
35
#include "iostat.h"
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);
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);
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))
		clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
F
Fred Isaman 已提交
366 367 368
	rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
}

369
void
370
pnfs_put_lseg(struct pnfs_layout_segment *lseg)
371
{
372
	struct pnfs_layout_hdr *lo;
F
Fred Isaman 已提交
373 374 375 376 377
	struct inode *inode;

	if (!lseg)
		return;

378 379 380
	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
		atomic_read(&lseg->pls_refcount),
		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
381

382 383
	lo = lseg->pls_layout;
	inode = lo->plh_inode;
384

F
Fred Isaman 已提交
385
	if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
386 387 388 389
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
			spin_unlock(&inode->i_lock);
			return;
		}
390
		pnfs_get_layout_hdr(lo);
391 392 393 394
		pnfs_layout_remove_lseg(lo, lseg);
		spin_unlock(&inode->i_lock);
		pnfs_free_lseg(lseg);
		pnfs_put_layout_hdr(lo);
395 396
	}
}
397
EXPORT_SYMBOL_GPL(pnfs_put_lseg);
398

399
static void pnfs_free_lseg_async_work(struct work_struct *work)
400 401
{
	struct pnfs_layout_segment *lseg;
402
	struct pnfs_layout_hdr *lo;
403 404

	lseg = container_of(work, struct pnfs_layout_segment, pls_work);
405
	lo = lseg->pls_layout;
406

407 408
	pnfs_free_lseg(lseg);
	pnfs_put_layout_hdr(lo);
409 410
}

411
static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
412
{
413
	INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
414 415
	schedule_work(&lseg->pls_work);
}
416 417 418 419 420 421 422 423 424 425 426 427 428 429

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;
430 431
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
			return;
432 433 434 435 436 437
		pnfs_get_layout_hdr(lo);
		pnfs_layout_remove_lseg(lo, lseg);
		pnfs_free_lseg_async(lseg);
	}
}
EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
438

439
static u64
440 441 442 443 444 445 446 447 448 449 450 451 452 453 454
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
 *           [----------------)
 */
455
static bool
456
pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
457
		 const struct pnfs_layout_range *l2)
458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473
{
	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
 *                              [----------------)
 */
474
static bool
475
pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
476
		    const struct pnfs_layout_range *l2)
477 478 479 480 481 482 483 484 485 486
{
	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);
}

487
static bool
488 489
should_free_lseg(const struct pnfs_layout_range *lseg_range,
		 const struct pnfs_layout_range *recall_range)
490
{
491 492
	return (recall_range->iomode == IOMODE_ANY ||
		lseg_range->iomode == recall_range->iomode) &&
493
	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
494 495
}

496 497 498 499 500 501 502 503 504 505
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;
}

506 507 508 509 510 511 512 513 514 515 516
/* 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
517 518
		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
			atomic_read(&lseg->pls_refcount));
519
		if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
F
Fred Isaman 已提交
520
			rv = 1;
521 522 523 524 525 526 527
	}
	return rv;
}

/* Returns count of number of matching invalid lsegs remaining in list
 * after call.
 */
F
Fred Isaman 已提交
528
int
529
pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
530
			    struct list_head *tmp_list,
531
			    const struct pnfs_layout_range *recall_range)
532 533
{
	struct pnfs_layout_segment *lseg, *next;
534
	int remaining = 0;
535 536 537

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

538
	if (list_empty(&lo->plh_segs))
539
		return 0;
540
	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
541 542
		if (!recall_range ||
		    should_free_lseg(&lseg->pls_range, recall_range)) {
543 544 545 546
			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);
547 548
			if (!mark_lseg_invalid(lseg, tmp_list))
				remaining++;
549
		}
550 551
	dprintk("%s:Return %i\n", __func__, remaining);
	return remaining;
552 553
}

554
/* note free_me must contain lsegs from a single layout_hdr */
F
Fred Isaman 已提交
555
void
556
pnfs_free_lseg_list(struct list_head *free_me)
557
{
558
	struct pnfs_layout_segment *lseg, *tmp;
559 560 561 562

	if (list_empty(free_me))
		return;

563
	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
564
		list_del(&lseg->pls_list);
565
		pnfs_free_lseg(lseg);
566 567 568
	}
}

569 570 571 572
void
pnfs_destroy_layout(struct nfs_inode *nfsi)
{
	struct pnfs_layout_hdr *lo;
573
	LIST_HEAD(tmp_list);
574 575 576 577

	spin_lock(&nfsi->vfs_inode.i_lock);
	lo = nfsi->layout;
	if (lo) {
578
		pnfs_get_layout_hdr(lo);
579
		pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
580 581 582 583 584 585 586
		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);
587
}
588
EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
589

590 591 592
static bool
pnfs_layout_add_bulk_destroy_list(struct inode *inode,
		struct list_head *layout_list)
593 594
{
	struct pnfs_layout_hdr *lo;
595
	bool ret = false;
596

597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648
	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;
649 650 651

		pnfs_layoutcommit_inode(inode, false);

652 653
		spin_lock(&inode->i_lock);
		list_del_init(&lo->plh_bulk_destroy);
654 655 656
		if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
			if (is_bulk_recall)
				set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
657
			ret = -EAGAIN;
658
		}
659 660
		spin_unlock(&inode->i_lock);
		pnfs_free_lseg_list(&lseg_list);
661 662
		/* Free all lsegs that are attached to commit buckets */
		nfs_commit_inode(inode, 0);
663 664 665 666 667 668 669 670 671 672 673 674 675
		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);
676

677
	spin_lock(&clp->cl_lock);
678
	rcu_read_lock();
679
restart:
680
	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
681 682 683 684 685 686
		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
			continue;
		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
				server,
				&layout_list) != 0)
			goto restart;
687 688
	}
	rcu_read_unlock();
689 690
	spin_unlock(&clp->cl_lock);

691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710
	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;
711
	}
712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730
	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);
731 732
}

733 734 735 736 737 738
/*
 * 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)
{
739
	return (s32)(s1 - s2) > 0;
740 741
}

742
/* update lo->plh_stateid with new if is more recent */
F
Fred Isaman 已提交
743 744 745
void
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
			bool update_barrier)
746
{
747 748
	u32 oldseq, newseq, new_barrier;
	int empty = list_empty(&lo->plh_segs);
749

750 751
	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
	newseq = be32_to_cpu(new->seqid);
752
	if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
753
		nfs4_stateid_copy(&lo->plh_stateid, new);
F
Fred Isaman 已提交
754
		if (update_barrier) {
755
			new_barrier = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
756 757
		} else {
			/* Because of wraparound, we want to keep the barrier
758
			 * "close" to the current seqids.
F
Fred Isaman 已提交
759
			 */
760
			new_barrier = newseq - atomic_read(&lo->plh_outstanding);
F
Fred Isaman 已提交
761
		}
762 763
		if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
			lo->plh_barrier = new_barrier;
F
Fred Isaman 已提交
764
	}
765 766
}

767
static bool
768 769
pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
		const nfs4_stateid *stateid)
F
Fred Isaman 已提交
770
{
771
	u32 seqid = be32_to_cpu(stateid->seqid);
772

773 774 775 776 777
	return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
}

/* lget is set to 1 if called from inside send_layoutget call chain */
static bool
778
pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
779
{
F
Fred Isaman 已提交
780
	return lo->plh_block_lgets ||
781
		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
782 783
}

784 785
int
pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
786
			      const struct pnfs_layout_range *range,
787
			      struct nfs4_state *open_state)
788
{
789
	int status = 0;
790

791
	dprintk("--> %s\n", __func__);
792
	spin_lock(&lo->plh_inode->i_lock);
793
	if (pnfs_layoutgets_blocked(lo)) {
794
		status = -EAGAIN;
795 796
	} else if (!nfs4_valid_open_stateid(open_state)) {
		status = -EBADF;
797 798
	} else if (list_empty(&lo->plh_segs) ||
		   test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
799 800 801 802
		int seq;

		do {
			seq = read_seqbegin(&open_state->seqlock);
803
			nfs4_stateid_copy(dst, &open_state->stateid);
804 805
		} while (read_seqretry(&open_state->seqlock, seq));
	} else
806
		nfs4_stateid_copy(dst, &lo->plh_stateid);
807
	spin_unlock(&lo->plh_inode->i_lock);
808
	dprintk("<-- %s\n", __func__);
809
	return status;
810 811 812 813 814 815 816 817
}

/*
* Get layout from server.
*    for now, assume that whole file layouts are requested.
*    arg->offset: 0
*    arg->length: all ones
*/
818 819 820
static struct pnfs_layout_segment *
send_layoutget(struct pnfs_layout_hdr *lo,
	   struct nfs_open_context *ctx,
821
	   const struct pnfs_layout_range *range,
822
	   gfp_t gfp_flags)
823
{
824
	struct inode *ino = lo->plh_inode;
825 826
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs4_layoutget *lgp;
827
	struct pnfs_layout_segment *lseg;
828
	loff_t i_size;
829 830

	dprintk("--> %s\n", __func__);
831

832 833 834 835 836 837 838 839 840 841 842 843
	/*
	 * Synchronously retrieve layout information from server and
	 * store in lseg. If we race with a concurrent seqid morphing
	 * op, then re-send the LAYOUTGET.
	 */
	do {
		lgp = kzalloc(sizeof(*lgp), gfp_flags);
		if (lgp == NULL)
			return NULL;

		i_size = i_size_read(ino);

844
		lgp->args.minlength = PAGE_SIZE;
845 846 847 848 849 850 851 852 853
		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;
		}
		lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
854
		pnfs_copy_range(&lgp->args.range, range);
855 856 857 858 859
		lgp->args.type = server->pnfs_curr_ld->id;
		lgp->args.inode = ino;
		lgp->args.ctx = get_nfs_open_context(ctx);
		lgp->gfp_flags = gfp_flags;
		lgp->cred = lo->plh_lc_cred;
860

861 862
		lseg = nfs4_proc_layoutget(lgp, gfp_flags);
	} while (lseg == ERR_PTR(-EAGAIN));
863

864 865 866
	if (IS_ERR(lseg) && !nfs_error_is_fatal(PTR_ERR(lseg)))
		lseg = NULL;
	else
867 868
		pnfs_layout_clear_fail_bit(lo,
				pnfs_iomode_to_fail_bit(range->iomode));
869

870 871 872
	return lseg;
}

873 874 875 876 877 878 879 880 881 882 883 884 885 886 887
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);
	}
}

888 889 890 891 892
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);
893
	rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
894 895
}

896 897 898 899 900 901
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;
902
	lo->plh_return_seq = 0;
903
	pnfs_get_layout_hdr(lo);
904
	clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
905 906 907
	return true;
}

908
static int
909
pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
910
		       enum pnfs_iomode iomode, bool sync)
911 912 913 914 915
{
	struct inode *ino = lo->plh_inode;
	struct nfs4_layoutreturn *lrp;
	int status = 0;

916
	lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
917 918 919
	if (unlikely(lrp == NULL)) {
		status = -ENOMEM;
		spin_lock(&ino->i_lock);
920
		pnfs_clear_layoutreturn_waitbit(lo);
921 922 923 924 925
		spin_unlock(&ino->i_lock);
		pnfs_put_layout_hdr(lo);
		goto out;
	}

926
	nfs4_stateid_copy(&lrp->args.stateid, stateid);
927 928
	lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
	lrp->args.inode = ino;
P
Peng Tao 已提交
929 930 931
	lrp->args.range.iomode = iomode;
	lrp->args.range.offset = 0;
	lrp->args.range.length = NFS4_MAX_UINT64;
932 933 934 935
	lrp->args.layout = lo;
	lrp->clp = NFS_SERVER(ino)->nfs_client;
	lrp->cred = lo->plh_lc_cred;

936
	status = nfs4_proc_layoutreturn(lrp, sync);
937 938 939 940 941
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}

942 943 944 945 946 947
/* Return true if layoutreturn is needed */
static bool
pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
{
	struct pnfs_layout_segment *s;

948
	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
949 950 951 952 953 954 955 956 957 958 959 960 961 962 963
		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;

964
	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
965 966 967 968 969 970 971 972
		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);
973
		stateid.seqid = cpu_to_be32(lo->plh_return_seq);
974 975 976 977 978 979 980 981 982 983 984
		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);
}

985 986 987 988 989 990 991 992
/*
 * 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
993 994 995 996 997 998 999
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;
1000
	int status = 0, empty;
1001
	bool send;
B
Benny Halevy 已提交
1002

1003
	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
B
Benny Halevy 已提交
1004 1005 1006

	spin_lock(&ino->i_lock);
	lo = nfsi->layout;
1007
	if (!lo) {
B
Benny Halevy 已提交
1008
		spin_unlock(&ino->i_lock);
1009 1010
		dprintk("NFS: %s no layout to return\n", __func__);
		goto out;
B
Benny Halevy 已提交
1011
	}
1012
	nfs4_stateid_copy(&stateid, &nfsi->layout->plh_stateid);
B
Benny Halevy 已提交
1013
	/* Reference matched in nfs4_layoutreturn_release */
1014
	pnfs_get_layout_hdr(lo);
1015
	empty = list_empty(&lo->plh_segs);
1016
	pnfs_clear_layoutcommit(ino, &tmp_list);
1017
	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
1018 1019 1020 1021 1022 1023 1024 1025 1026 1027

	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);
	}

1028 1029 1030 1031
	/* 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__);
1032
		goto out_put_layout_hdr;
1033
	}
1034 1035

	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1036
	send = pnfs_prepare_layoutreturn(lo);
B
Benny Halevy 已提交
1037 1038
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);
1039
	if (send)
1040
		status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1041 1042
out_put_layout_hdr:
	pnfs_put_layout_hdr(lo);
B
Benny Halevy 已提交
1043 1044 1045 1046
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}
1047
EXPORT_SYMBOL_GPL(_pnfs_return_layout);
B
Benny Halevy 已提交
1048

1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
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
1076 1077
bool pnfs_roc(struct inode *ino)
{
1078 1079 1080
	struct nfs_inode *nfsi = NFS_I(ino);
	struct nfs_open_context *ctx;
	struct nfs4_state *state;
F
Fred Isaman 已提交
1081 1082
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg, *tmp;
1083
	nfs4_stateid stateid;
F
Fred Isaman 已提交
1084
	LIST_HEAD(tmp_list);
1085
	bool found = false, layoutreturn = false, roc = false;
F
Fred Isaman 已提交
1086 1087

	spin_lock(&ino->i_lock);
1088
	lo = nfsi->layout;
P
Peng Tao 已提交
1089
	if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
1090 1091
		goto out_noroc;

1092
	/* no roc if we hold a delegation */
1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
	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;
	}

1103
	nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1104
	/* always send layoutreturn if being marked so */
1105
	if (test_and_clear_bit(NFS_LAYOUT_RETURN_REQUESTED,
1106 1107 1108
				   &lo->plh_flags))
		layoutreturn = pnfs_prepare_layoutreturn(lo);

F
Fred Isaman 已提交
1109
	list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
1110 1111
		/* 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
1112 1113 1114
			mark_lseg_invalid(lseg, &tmp_list);
			found = true;
		}
1115
	/* ROC in two conditions:
1116 1117 1118
	 * 1. there are ROC lsegs
	 * 2. we don't send layoutreturn
	 */
1119 1120 1121
	if (found && !layoutreturn) {
		/* lo ref dropped in pnfs_roc_release() */
		pnfs_get_layout_hdr(lo);
1122
		roc = true;
1123
	}
F
Fred Isaman 已提交
1124

1125
out_noroc:
F
Fred Isaman 已提交
1126
	spin_unlock(&ino->i_lock);
1127 1128 1129
	pnfs_free_lseg_list(&tmp_list);
	pnfs_layoutcommit_inode(ino, true);
	if (layoutreturn)
1130
		pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1131
	return roc;
F
Fred Isaman 已提交
1132 1133 1134 1135 1136 1137 1138 1139
}

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

	spin_lock(&ino->i_lock);
	lo = NFS_I(ino)->layout;
1140
	pnfs_clear_layoutreturn_waitbit(lo);
1141 1142 1143 1144 1145 1146
	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
1147 1148 1149 1150 1151 1152 1153 1154
}

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;
1155
	pnfs_mark_layout_returned_if_empty(lo);
1156
	if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
F
Fred Isaman 已提交
1157 1158
		lo->plh_barrier = barrier;
	spin_unlock(&ino->i_lock);
1159
	trace_nfs4_layoutreturn_on_close(ino, 0);
F
Fred Isaman 已提交
1160 1161
}

1162
void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier)
F
Fred Isaman 已提交
1163 1164
{
	struct nfs_inode *nfsi = NFS_I(ino);
1165 1166
	struct pnfs_layout_hdr *lo;
	u32 current_seqid;
F
Fred Isaman 已提交
1167 1168

	spin_lock(&ino->i_lock);
1169 1170
	lo = nfsi->layout;
	current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
F
Fred Isaman 已提交
1171

1172 1173 1174 1175
	/* 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
1176 1177 1178
	spin_unlock(&ino->i_lock);
}

1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198
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;
}

1199 1200 1201 1202 1203 1204
/*
 * 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
1205
pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1206
	   const struct pnfs_layout_range *l2)
1207
{
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219
	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;

1220
	/* read > read/write */
1221
	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1222 1223
}

1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245
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)
1246
{
1247
	struct pnfs_layout_segment *lp, *tmp;
1248

1249 1250
	dprintk("%s:Begin\n", __func__);

1251 1252 1253 1254 1255 1256 1257 1258
	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))
1259
			continue;
1260
		list_add_tail(&lseg->pls_list, &lp->pls_list);
1261 1262 1263
		dprintk("%s: inserted lseg %p "
			"iomode %d offset %llu length %llu before "
			"lp %p iomode %d offset %llu length %llu\n",
1264 1265 1266 1267
			__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);
1268
		goto out;
1269
	}
1270 1271 1272 1273 1274 1275
	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:
1276
	pnfs_get_layout_hdr(lo);
1277 1278

	dprintk("%s:Return\n", __func__);
1279
}
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
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);
}
1298 1299

static struct pnfs_layout_hdr *
1300 1301 1302
alloc_init_layout_hdr(struct inode *ino,
		      struct nfs_open_context *ctx,
		      gfp_t gfp_flags)
1303 1304 1305
{
	struct pnfs_layout_hdr *lo;

1306
	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1307 1308
	if (!lo)
		return NULL;
1309
	atomic_set(&lo->plh_refcount, 1);
1310 1311
	INIT_LIST_HEAD(&lo->plh_layouts);
	INIT_LIST_HEAD(&lo->plh_segs);
1312
	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1313
	lo->plh_inode = ino;
1314
	lo->plh_lc_cred = get_rpccred(ctx->cred);
1315 1316 1317 1318
	return lo;
}

static struct pnfs_layout_hdr *
1319 1320 1321
pnfs_find_alloc_layout(struct inode *ino,
		       struct nfs_open_context *ctx,
		       gfp_t gfp_flags)
1322 1323 1324 1325 1326 1327
{
	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);

1328 1329
	if (nfsi->layout != NULL)
		goto out_existing;
1330
	spin_unlock(&ino->i_lock);
1331
	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1332 1333
	spin_lock(&ino->i_lock);

1334
	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1335
		nfsi->layout = new;
1336
		return new;
1337 1338
	} else if (new != NULL)
		pnfs_free_layout_hdr(new);
1339 1340
out_existing:
	pnfs_get_layout_hdr(nfsi->layout);
1341 1342 1343
	return nfsi->layout;
}

1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354
/*
 * iomode matching rules:
 * iomode	lseg	match
 * -----	-----	-----
 * ANY		READ	true
 * ANY		RW	true
 * RW		READ	false
 * RW		RW	true
 * READ		READ	true
 * READ		RW	true
 */
1355
static bool
1356
pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1357
		 const struct pnfs_layout_range *range)
1358
{
1359 1360 1361 1362
	struct pnfs_layout_range range1;

	if ((range->iomode == IOMODE_RW &&
	     ls_range->iomode != IOMODE_RW) ||
1363
	    !pnfs_lseg_range_intersecting(ls_range, range))
1364 1365 1366 1367 1368
		return 0;

	/* range1 covers only the first byte in the range */
	range1 = *range;
	range1.length = 1;
1369
	return pnfs_lseg_range_contained(ls_range, &range1);
1370 1371 1372 1373 1374
}

/*
 * lookup range in layout
 */
1375
static struct pnfs_layout_segment *
1376 1377
pnfs_find_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_range *range)
1378
{
1379 1380 1381 1382
	struct pnfs_layout_segment *lseg, *ret = NULL;

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

1383
	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1384
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1385
		    !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
1386
		    pnfs_lseg_range_match(&lseg->pls_range, range)) {
1387
			ret = pnfs_get_lseg(lseg);
1388 1389 1390 1391 1392
			break;
		}
	}

	dprintk("%s:Return lseg %p ref %d\n",
1393
		__func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1394
	return ret;
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 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471
/*
 * 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;
}

1472 1473 1474 1475 1476 1477 1478 1479
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,
1480
				   nfs_wait_bit_killable,
1481 1482 1483
				   TASK_UNINTERRUPTIBLE);
}

1484 1485 1486 1487 1488 1489 1490 1491 1492
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);
}

1493 1494 1495 1496
/*
 * Layout segment is retreived from the server if not cached.
 * The appropriate layout segment is referenced and returned to the caller.
 */
1497
struct pnfs_layout_segment *
1498 1499
pnfs_update_layout(struct inode *ino,
		   struct nfs_open_context *ctx,
1500 1501
		   loff_t pos,
		   u64 count,
1502 1503
		   enum pnfs_iomode iomode,
		   gfp_t gfp_flags)
1504
{
1505 1506 1507 1508 1509
	struct pnfs_layout_range arg = {
		.iomode = iomode,
		.offset = pos,
		.length = count,
	};
1510
	unsigned pg_offset;
1511 1512
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs_client *clp = server->nfs_client;
1513 1514
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg = NULL;
1515
	bool first;
1516

1517
	if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1518
		trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1519
				 PNFS_UPDATE_LAYOUT_NO_PNFS);
1520
		goto out;
1521
	}
1522

1523
	if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
1524
		trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1525
				 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
1526
		goto out;
1527
	}
1528

1529
	if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1530
		trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1531
				 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1532
		goto out;
1533
	}
1534

1535 1536
lookup_again:
	first = false;
1537
	spin_lock(&ino->i_lock);
1538
	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1539 1540
	if (lo == NULL) {
		spin_unlock(&ino->i_lock);
1541
		trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1542
				 PNFS_UPDATE_LAYOUT_NOMEM);
1543 1544
		goto out;
	}
1545

F
Fred Isaman 已提交
1546
	/* Do we even need to bother with this? */
1547
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1548
		trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1549
				 PNFS_UPDATE_LAYOUT_BULK_RECALL);
F
Fred Isaman 已提交
1550
		dprintk("%s matches recall, use MDS\n", __func__);
1551 1552 1553 1554
		goto out_unlock;
	}

	/* if LAYOUTGET already failed once we don't try again */
1555
	if (pnfs_layout_io_test_failed(lo, iomode)) {
1556
		trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1557
				 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1558
		goto out_unlock;
1559
	}
1560

1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578
	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);
1579
		if (lseg) {
1580
			trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1581
					PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1582
			goto out_unlock;
1583
		}
1584
	}
1585

1586 1587 1588 1589
	/*
	 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
	 * for LAYOUTRETURN even if first is true.
	 */
1590
	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1591 1592 1593
		spin_unlock(&ino->i_lock);
		dprintk("%s wait for layoutreturn\n", __func__);
		if (pnfs_prepare_to_retry_layoutget(lo)) {
1594 1595
			if (first)
				pnfs_clear_first_layoutget(lo);
1596 1597 1598 1599
			pnfs_put_layout_hdr(lo);
			dprintk("%s retrying\n", __func__);
			goto lookup_again;
		}
1600
		trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1601
				PNFS_UPDATE_LAYOUT_RETURN);
1602 1603 1604
		goto out_put_layout_hdr;
	}

1605
	if (pnfs_layoutgets_blocked(lo)) {
1606
		trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1607
				PNFS_UPDATE_LAYOUT_BLOCKED);
1608
		goto out_unlock;
1609
	}
1610
	atomic_inc(&lo->plh_outstanding);
1611
	spin_unlock(&ino->i_lock);
1612

1613
	if (list_empty(&lo->plh_layouts)) {
1614 1615 1616 1617
		/* 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);
1618 1619
		if (list_empty(&lo->plh_layouts))
			list_add_tail(&lo->plh_layouts, &server->layouts);
1620 1621
		spin_unlock(&clp->cl_lock);
	}
1622

1623
	pg_offset = arg.offset & ~PAGE_MASK;
1624 1625 1626 1627
	if (pg_offset) {
		arg.offset -= pg_offset;
		arg.length += pg_offset;
	}
1628
	if (arg.length != NFS4_MAX_UINT64)
1629
		arg.length = PAGE_ALIGN(arg.length);
1630

1631
	lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1632
	atomic_dec(&lo->plh_outstanding);
1633
	trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1634
				 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
1635
out_put_layout_hdr:
1636 1637
	if (first)
		pnfs_clear_first_layoutget(lo);
1638
	pnfs_put_layout_hdr(lo);
1639
out:
1640 1641 1642 1643
	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),
1644
			IS_ERR_OR_NULL(lseg) ? "not found" : "found",
1645 1646 1647
			iomode==IOMODE_RW ?  "read/write" : "read-only",
			(unsigned long long)pos,
			(unsigned long long)count);
1648 1649 1650
	return lseg;
out_unlock:
	spin_unlock(&ino->i_lock);
1651
	goto out_put_layout_hdr;
1652
}
1653
EXPORT_SYMBOL_GPL(pnfs_update_layout);
1654

1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674
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;
}

1675
struct pnfs_layout_segment *
1676 1677 1678 1679 1680
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;
1681
	struct inode *ino = lo->plh_inode;
1682
	LIST_HEAD(free_me);
1683 1684 1685 1686
	int status = -EINVAL;

	if (!pnfs_sanity_check_layout_range(&res->range))
		goto out;
1687 1688

	/* Inject layout blob into I/O device driver */
1689
	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1690 1691 1692 1693 1694 1695 1696 1697 1698 1699
	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;
	}

1700 1701
	init_lseg(lo, lseg);
	lseg->pls_range = res->range;
1702
	lseg->pls_seq = be32_to_cpu(res->stateid.seqid);
1703

1704
	spin_lock(&ino->i_lock);
1705
	if (pnfs_layoutgets_blocked(lo)) {
F
Fred Isaman 已提交
1706 1707 1708
		dprintk("%s forget reply due to state\n", __func__);
		goto out_forget_reply;
	}
1709

1710 1711 1712 1713
	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__);
1714
			status = -EAGAIN;
1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728
			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);
	}
1729

1730 1731
	clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);

1732
	pnfs_get_lseg(lseg);
1733
	pnfs_layout_insert_lseg(lo, lseg, &free_me);
1734

P
Peng Tao 已提交
1735
	if (res->return_on_close)
F
Fred Isaman 已提交
1736 1737
		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);

1738
	spin_unlock(&ino->i_lock);
1739
	pnfs_free_lseg_list(&free_me);
1740
	return lseg;
1741
out:
1742
	return ERR_PTR(status);
F
Fred Isaman 已提交
1743 1744 1745 1746 1747 1748

out_forget_reply:
	spin_unlock(&ino->i_lock);
	lseg->pls_layout = lo;
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
	goto out;
1749 1750
}

1751
static void
1752 1753
pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
			 u32 seq)
1754 1755 1756 1757 1758 1759
{
	if (lo->plh_return_iomode == iomode)
		return;
	if (lo->plh_return_iomode != 0)
		iomode = IOMODE_ANY;
	lo->plh_return_iomode = iomode;
1760
	set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
1761 1762
	if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq))
		lo->plh_return_seq = seq;
1763 1764
}

1765 1766 1767 1768 1769 1770 1771 1772 1773 1774
/**
 * 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.
 */
1775
int
1776 1777
pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
				struct list_head *tmp_list,
1778
				const struct pnfs_layout_range *return_range)
1779 1780
{
	struct pnfs_layout_segment *lseg, *next;
1781
	int remaining = 0;
1782 1783 1784 1785

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

	if (list_empty(&lo->plh_segs))
1786
		return 0;
1787

1788
	assert_spin_locked(&lo->plh_inode->i_lock);
1789 1790 1791 1792 1793 1794 1795 1796

	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);
1797 1798 1799
			if (mark_lseg_invalid(lseg, tmp_list))
				continue;
			remaining++;
1800
			set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1801
			pnfs_set_plh_return_info(lo, return_range->iomode, lseg->pls_seq);
1802
		}
1803
	return remaining;
1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815
}

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);
1816
	bool return_now = false;
1817 1818

	spin_lock(&inode->i_lock);
1819
	pnfs_set_plh_return_info(lo, range.iomode, lseg->pls_seq);
1820 1821 1822 1823 1824
	/*
	 * 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.
	 */
1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837
	if (!pnfs_mark_matching_lsegs_return(lo, &free_me, &range)) {
		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);
	}
1838 1839 1840 1841
	pnfs_free_lseg_list(&free_me);
}
EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);

1842 1843 1844
void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
P
Peng Tao 已提交
1845 1846
	u64 rd_size = req->wb_bytes;

1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858
	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,
						   GFP_KERNEL);
1859 1860 1861 1862 1863
		if (IS_ERR(pgio->pg_lseg)) {
			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
			pgio->pg_lseg = NULL;
			return;
		}
1864
	}
1865 1866
	/* If no lseg, fall back to read through mds */
	if (pgio->pg_lseg == NULL)
1867
		nfs_pageio_reset_read_mds(pgio);
1868

1869 1870 1871 1872
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);

void
1873 1874
pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
			   struct nfs_page *req, u64 wb_size)
1875
{
1876
	if (pgio->pg_lseg == NULL) {
1877 1878 1879 1880 1881 1882
		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
						   req->wb_context,
						   req_offset(req),
						   wb_size,
						   IOMODE_RW,
						   GFP_NOFS);
1883 1884 1885 1886 1887 1888
		if (IS_ERR(pgio->pg_lseg)) {
			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
			pgio->pg_lseg = NULL;
			return;
		}
	}
1889 1890
	/* If no lseg, fall back to write through mds */
	if (pgio->pg_lseg == NULL)
1891
		nfs_pageio_reset_write_mds(pgio);
1892 1893 1894
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);

1895 1896 1897 1898 1899 1900 1901 1902 1903 1904
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);

1905 1906 1907 1908 1909
/*
 * 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
1910 1911
pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
		     struct nfs_page *prev, struct nfs_page *req)
1912
{
1913
	unsigned int size;
1914
	u64 seg_end, req_start, seg_left;
1915 1916 1917 1918

	size = nfs_generic_pg_test(pgio, prev, req);
	if (!size)
		return 0;
1919

1920
	/*
1921 1922 1923 1924 1925
	 * '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.
1926 1927 1928 1929 1930
	 *
	 * Please also note that 'end_offset' is actually the offset of the
	 * first byte that lies outside the pnfs_layout_range. FIXME?
	 *
	 */
1931
	if (pgio->pg_lseg) {
1932 1933 1934
		seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
				     pgio->pg_lseg->pls_range.length);
		req_start = req_offset(req);
1935
		WARN_ON_ONCE(req_start >= seg_end);
1936
		/* start of request is past the last byte of this segment */
1937 1938 1939 1940 1941 1942
		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);
1943
			return 0;
1944
		}
1945 1946 1947 1948 1949 1950

		/* 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;
1951
	}
1952

1953
	return size;
1954
}
1955
EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1956

1957
int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
1958 1959 1960 1961
{
	struct nfs_pageio_descriptor pgio;

	/* Resend all requests through the MDS */
1962 1963
	nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
			      hdr->completion_ops);
1964
	set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
1965
	return nfs_pageio_resend(&pgio, hdr);
1966
}
1967
EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1968

1969
static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1970
{
1971 1972 1973

	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1974
	    PNFS_LAYOUTRET_ON_ERROR) {
1975
		pnfs_return_layout(hdr->inode);
1976
	}
1977
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1978
		hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1979 1980
}

1981 1982 1983
/*
 * Called by non rpc-based layout drivers
 */
1984
void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
1985
{
1986
	if (likely(!hdr->pnfs_error)) {
1987 1988
		pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
				hdr->mds_offset + hdr->res.count);
1989
		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1990 1991 1992
	}
	trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
	if (unlikely(hdr->pnfs_error))
1993 1994
		pnfs_ld_handle_write_error(hdr);
	hdr->mds_ops->rpc_release(hdr);
1995
}
1996
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1997

1998 1999
static void
pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
2000
		struct nfs_pgio_header *hdr)
2001
{
2002
	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2003

2004
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2005
		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2006
		nfs_pageio_reset_write_mds(desc);
2007
		mirror->pg_recoalesce = 1;
2008
	}
2009
	nfs_pgio_data_destroy(hdr);
2010
	hdr->release(hdr);
2011 2012 2013
}

static enum pnfs_try_status
2014
pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
2015 2016 2017
			const struct rpc_call_ops *call_ops,
			struct pnfs_layout_segment *lseg,
			int how)
2018
{
2019
	struct inode *inode = hdr->inode;
2020 2021 2022
	enum pnfs_try_status trypnfs;
	struct nfs_server *nfss = NFS_SERVER(inode);

2023
	hdr->mds_ops = call_ops;
2024 2025

	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
2026 2027
		inode->i_ino, hdr->args.count, hdr->args.offset, how);
	trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
2028
	if (trypnfs != PNFS_NOT_ATTEMPTED)
2029 2030 2031 2032 2033
		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}

2034
static void
2035 2036
pnfs_do_write(struct nfs_pageio_descriptor *desc,
	      struct nfs_pgio_header *hdr, int how)
2037 2038 2039
{
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2040
	enum pnfs_try_status trypnfs;
2041

2042
	trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2043
	if (trypnfs == PNFS_NOT_ATTEMPTED)
2044
		pnfs_write_through_mds(desc, hdr);
2045 2046
}

2047 2048
static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
{
2049
	pnfs_put_lseg(hdr->lseg);
2050
	nfs_pgio_header_free(hdr);
2051 2052
}

2053 2054 2055
int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
{
2056
	struct nfs_pgio_header *hdr;
2057 2058
	int ret;

2059 2060
	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
	if (!hdr) {
P
Peng Tao 已提交
2061 2062
		desc->pg_error = -ENOMEM;
		return desc->pg_error;
2063
	}
2064
	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2065

2066
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2067
	ret = nfs_generic_pgio(desc, hdr);
2068
	if (!ret)
2069
		pnfs_do_write(desc, hdr, desc->pg_ioflags);
2070

2071
	return ret;
2072 2073 2074
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);

2075
int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2076 2077 2078
{
	struct nfs_pageio_descriptor pgio;

2079
	/* Resend all requests through the MDS */
2080 2081
	nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
	return nfs_pageio_resend(&pgio, hdr);
2082
}
2083
EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2084

2085
static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2086
{
2087 2088
	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2089
	    PNFS_LAYOUTRET_ON_ERROR) {
2090
		pnfs_return_layout(hdr->inode);
2091
	}
2092
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2093
		hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2094 2095
}

2096 2097 2098
/*
 * Called by non rpc-based layout drivers
 */
2099
void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2100
{
2101
	if (likely(!hdr->pnfs_error)) {
2102 2103
		__nfs4_read_done_cb(hdr);
		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2104 2105 2106
	}
	trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
	if (unlikely(hdr->pnfs_error))
2107 2108
		pnfs_ld_handle_read_error(hdr);
	hdr->mds_ops->rpc_release(hdr);
2109 2110 2111
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

2112 2113
static void
pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2114
		struct nfs_pgio_header *hdr)
2115
{
2116
	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2117

2118
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2119
		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2120
		nfs_pageio_reset_read_mds(desc);
2121
		mirror->pg_recoalesce = 1;
2122
	}
2123
	nfs_pgio_data_destroy(hdr);
2124
	hdr->release(hdr);
2125 2126
}

A
Andy Adamson 已提交
2127 2128 2129
/*
 * Call the appropriate parallel I/O subsystem read function.
 */
2130
static enum pnfs_try_status
2131
pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2132 2133
		       const struct rpc_call_ops *call_ops,
		       struct pnfs_layout_segment *lseg)
A
Andy Adamson 已提交
2134
{
2135
	struct inode *inode = hdr->inode;
A
Andy Adamson 已提交
2136 2137 2138
	struct nfs_server *nfss = NFS_SERVER(inode);
	enum pnfs_try_status trypnfs;

2139
	hdr->mds_ops = call_ops;
A
Andy Adamson 已提交
2140 2141

	dprintk("%s: Reading ino:%lu %u@%llu\n",
2142
		__func__, inode->i_ino, hdr->args.count, hdr->args.offset);
A
Andy Adamson 已提交
2143

2144
	trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2145
	if (trypnfs != PNFS_NOT_ATTEMPTED)
A
Andy Adamson 已提交
2146 2147 2148 2149
		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}
A
Andy Adamson 已提交
2150

2151
/* Resend all requests through pnfs. */
2152
void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2153 2154 2155
{
	struct nfs_pageio_descriptor pgio;

2156 2157 2158 2159 2160
	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);
	}
2161 2162 2163
}
EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);

2164
static void
2165
pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2166 2167 2168
{
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2169
	enum pnfs_try_status trypnfs;
2170

2171
	trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2172
	if (trypnfs == PNFS_TRY_AGAIN)
2173 2174
		pnfs_read_resend_pnfs(hdr);
	if (trypnfs == PNFS_NOT_ATTEMPTED || hdr->task.tk_status)
2175
		pnfs_read_through_mds(desc, hdr);
2176 2177
}

2178 2179
static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
{
2180
	pnfs_put_lseg(hdr->lseg);
2181
	nfs_pgio_header_free(hdr);
2182 2183
}

2184 2185 2186
int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
{
2187
	struct nfs_pgio_header *hdr;
2188 2189
	int ret;

2190 2191
	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
	if (!hdr) {
P
Peng Tao 已提交
2192 2193
		desc->pg_error = -ENOMEM;
		return desc->pg_error;
2194
	}
2195
	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2196
	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2197
	ret = nfs_generic_pgio(desc, hdr);
2198
	if (!ret)
2199
		pnfs_do_read(desc, hdr);
2200
	return ret;
2201 2202 2203
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);

2204 2205 2206 2207 2208
static void pnfs_clear_layoutcommitting(struct inode *inode)
{
	unsigned long *bitlock = &NFS_I(inode)->flags;

	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2209
	smp_mb__after_atomic();
2210 2211 2212
	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
}

A
Andy Adamson 已提交
2213
/*
2214
 * There can be multiple RW segments.
A
Andy Adamson 已提交
2215
 */
2216
static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
A
Andy Adamson 已提交
2217
{
2218
	struct pnfs_layout_segment *lseg;
A
Andy Adamson 已提交
2219

2220 2221
	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
		if (lseg->pls_range.iomode == IOMODE_RW &&
2222
		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2223 2224
			list_add(&lseg->pls_lc_list, listp);
	}
A
Andy Adamson 已提交
2225 2226
}

2227 2228 2229 2230 2231 2232 2233 2234 2235 2236
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);
	}

2237
	pnfs_clear_layoutcommitting(inode);
2238 2239
}

P
Peng Tao 已提交
2240 2241
void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
{
2242
	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
P
Peng Tao 已提交
2243 2244 2245
}
EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);

A
Andy Adamson 已提交
2246
void
2247 2248
pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
		loff_t end_pos)
A
Andy Adamson 已提交
2249
{
2250
	struct nfs_inode *nfsi = NFS_I(inode);
2251
	bool mark_as_dirty = false;
A
Andy Adamson 已提交
2252

2253
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
2254
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2255
		nfsi->layout->plh_lwb = end_pos;
2256
		mark_as_dirty = true;
A
Andy Adamson 已提交
2257
		dprintk("%s: Set layoutcommit for inode %lu ",
2258
			__func__, inode->i_ino);
2259 2260
	} else if (end_pos > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = end_pos;
2261
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2262
		/* references matched in nfs4_layoutcommit_release */
2263
		pnfs_get_lseg(lseg);
2264
	}
2265
	spin_unlock(&inode->i_lock);
2266
	dprintk("%s: lseg %p end_pos %llu\n",
2267
		__func__, lseg, nfsi->layout->plh_lwb);
2268 2269 2270 2271

	/* 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)
2272
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
2273 2274 2275
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

A
Andy Adamson 已提交
2276 2277 2278 2279 2280 2281
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);
2282
	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
A
Andy Adamson 已提交
2283 2284
}

2285 2286 2287 2288 2289 2290 2291 2292
/*
 * 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
2293
int
2294
pnfs_layoutcommit_inode(struct inode *inode, bool sync)
A
Andy Adamson 已提交
2295
{
2296
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
A
Andy Adamson 已提交
2297 2298 2299
	struct nfs4_layoutcommit_data *data;
	struct nfs_inode *nfsi = NFS_I(inode);
	loff_t end_pos;
2300
	int status;
A
Andy Adamson 已提交
2301

2302
	if (!pnfs_layoutcommit_outstanding(inode))
2303 2304
		return 0;

2305
	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
P
Peng Tao 已提交
2306

2307
	status = -EAGAIN;
P
Peng Tao 已提交
2308
	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2309 2310
		if (!sync)
			goto out;
2311
		status = wait_on_bit_lock_action(&nfsi->flags,
2312 2313 2314
				NFS_INO_LAYOUTCOMMITTING,
				nfs_wait_bit_killable,
				TASK_KILLABLE);
P
Peng Tao 已提交
2315
		if (status)
2316
			goto out;
P
Peng Tao 已提交
2317 2318
	}

2319 2320 2321 2322 2323 2324 2325
	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;
2326
	spin_lock(&inode->i_lock);
2327 2328
	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		goto out_unlock;
2329

2330
	INIT_LIST_HEAD(&data->lseg_list);
2331
	pnfs_list_write_lseg(inode, &data->lseg_list);
A
Andy Adamson 已提交
2332

2333
	end_pos = nfsi->layout->plh_lwb;
A
Andy Adamson 已提交
2334

2335
	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
A
Andy Adamson 已提交
2336 2337 2338
	spin_unlock(&inode->i_lock);

	data->args.inode = inode;
2339
	data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
A
Andy Adamson 已提交
2340 2341 2342 2343 2344 2345
	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);

2346 2347 2348
	if (ld->prepare_layoutcommit) {
		status = ld->prepare_layoutcommit(&data->args);
		if (status) {
2349
			put_rpccred(data->cred);
2350
			spin_lock(&inode->i_lock);
2351 2352
			set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
			if (end_pos > nfsi->layout->plh_lwb)
2353
				nfsi->layout->plh_lwb = end_pos;
2354
			goto out_unlock;
2355 2356 2357 2358
		}
	}


A
Andy Adamson 已提交
2359 2360
	status = nfs4_proc_layoutcommit(data, sync);
out:
P
Peng Tao 已提交
2361 2362
	if (status)
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
2363 2364
	dprintk("<-- %s status %d\n", __func__, status);
	return status;
2365 2366
out_unlock:
	spin_unlock(&inode->i_lock);
P
Peng Tao 已提交
2367
	kfree(data);
2368 2369
clear_layoutcommitting:
	pnfs_clear_layoutcommitting(inode);
P
Peng Tao 已提交
2370
	goto out;
A
Andy Adamson 已提交
2371
}
2372
EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2373

2374 2375 2376 2377 2378 2379 2380
int
pnfs_generic_sync(struct inode *inode, bool datasync)
{
	return pnfs_layoutcommit_inode(inode, true);
}
EXPORT_SYMBOL_GPL(pnfs_generic_sync);

2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391
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;
}
2392

2393
#if IS_ENABLED(CONFIG_NFS_V4_2)
2394
int
2395
pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
2396 2397 2398
{
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
	struct nfs_server *server = NFS_SERVER(inode);
2399
	struct nfs_inode *nfsi = NFS_I(inode);
2400 2401 2402 2403 2404 2405 2406
	struct nfs42_layoutstat_data *data;
	struct pnfs_layout_hdr *hdr;
	int status = 0;

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

2407 2408 2409
	if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
		goto out;

2410 2411 2412
	if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
		goto out;

2413 2414 2415
	spin_lock(&inode->i_lock);
	if (!NFS_I(inode)->layout) {
		spin_unlock(&inode->i_lock);
2416
		goto out_clear_layoutstats;
2417 2418 2419 2420 2421
	}
	hdr = NFS_I(inode)->layout;
	pnfs_get_layout_hdr(hdr);
	spin_unlock(&inode->i_lock);

2422
	data = kzalloc(sizeof(*data), gfp_flags);
2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444
	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);
2445
out_clear_layoutstats:
2446 2447 2448
	smp_mb__before_atomic();
	clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
	smp_mb__after_atomic();
2449 2450 2451
	goto out;
}
EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
2452
#endif
2453 2454 2455 2456

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