pnfs.c 45.3 KB
Newer Older
R
Ricardo Labiaga 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
/*
 *  pNFS functions to call and manage layout drivers.
 *
 *  Copyright (c) 2002 [year of first publication]
 *  The Regents of the University of Michigan
 *  All Rights Reserved
 *
 *  Dean Hildebrand <dhildebz@umich.edu>
 *
 *  Permission is granted to use, copy, create derivative works, and
 *  redistribute this software and such derivative works for any purpose,
 *  so long as the name of the University of Michigan is not used in
 *  any advertising or publicity pertaining to the use or distribution
 *  of this software without specific, written prior authorization. If
 *  the above copyright notice or any other identification of the
 *  University of Michigan is included in any copy of any portion of
 *  this software, then the disclaimer below must also be included.
 *
 *  This software is provided as is, without representation or warranty
 *  of any kind either express or implied, including without limitation
 *  the implied warranties of merchantability, fitness for a particular
 *  purpose, or noninfringement.  The Regents of the University of
 *  Michigan shall not be liable for any damages, including special,
 *  indirect, incidental, or consequential damages, with respect to any
 *  claim arising out of or in connection with the use of the software,
 *  even if it has been or is hereafter advised of the possibility of
 *  such damages.
 */

#include <linux/nfs_fs.h>
31
#include <linux/nfs_page.h>
32
#include <linux/module.h>
33
#include "internal.h"
R
Ricardo Labiaga 已提交
34
#include "pnfs.h"
A
Andy Adamson 已提交
35
#include "iostat.h"
R
Ricardo Labiaga 已提交
36 37 38

#define NFSDBG_FACILITY		NFSDBG_PNFS

39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
/* Locking:
 *
 * pnfs_spinlock:
 *      protects pnfs_modules_tbl.
 */
static DEFINE_SPINLOCK(pnfs_spinlock);

/*
 * pnfs_modules_tbl holds all pnfs modules
 */
static LIST_HEAD(pnfs_modules_tbl);

/* Return the registered pnfs layout driver module matching given id */
static struct pnfs_layoutdriver_type *
find_pnfs_driver_locked(u32 id)
{
	struct pnfs_layoutdriver_type *local;

	list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
		if (local->id == id)
			goto out;
	local = NULL;
out:
	dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
	return local;
}

R
Ricardo Labiaga 已提交
66 67 68
static struct pnfs_layoutdriver_type *
find_pnfs_driver(u32 id)
{
69 70 71 72 73 74
	struct pnfs_layoutdriver_type *local;

	spin_lock(&pnfs_spinlock);
	local = find_pnfs_driver_locked(id);
	spin_unlock(&pnfs_spinlock);
	return local;
R
Ricardo Labiaga 已提交
75 76 77 78 79
}

void
unset_pnfs_layoutdriver(struct nfs_server *nfss)
{
80 81 82
	if (nfss->pnfs_curr_ld) {
		if (nfss->pnfs_curr_ld->clear_layoutdriver)
			nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
83
		module_put(nfss->pnfs_curr_ld->owner);
84
	}
R
Ricardo Labiaga 已提交
85 86 87 88 89 90 91 92 93 94
	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
95 96
set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
		      u32 id)
R
Ricardo Labiaga 已提交
97 98 99 100 101 102 103
{
	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))) {
104 105
		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
106 107 108 109 110 111 112 113 114 115 116 117
		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;
		}
	}
118 119 120 121
	if (!try_module_get(ld_type->owner)) {
		dprintk("%s: Could not grab reference on module\n", __func__);
		goto out_no_driver;
	}
R
Ricardo Labiaga 已提交
122
	server->pnfs_curr_ld = ld_type;
123 124
	if (ld_type->set_layoutdriver
	    && ld_type->set_layoutdriver(server, mntfh)) {
125 126
		printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
			"driver %u.\n", __func__, id);
127 128 129
		module_put(ld_type->owner);
		goto out_no_driver;
	}
130

R
Ricardo Labiaga 已提交
131 132 133 134 135 136 137
	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;
}
138 139 140 141 142 143 144 145

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

	if (ld_type->id == 0) {
146
		printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
147 148
		return status;
	}
149
	if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
150
		printk(KERN_ERR "NFS: %s Layout driver must provide "
151 152 153
		       "alloc_lseg and free_lseg.\n", __func__);
		return status;
	}
154 155 156 157 158 159 160 161 162

	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 {
163
		printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
			__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);
181

182 183 184 185
/*
 * pNFS client layout cache
 */

186
/* Need to hold i_lock if caller does not already hold reference */
F
Fred Isaman 已提交
187
void
188
get_layout_hdr(struct pnfs_layout_hdr *lo)
189
{
190
	atomic_inc(&lo->plh_refcount);
191 192
}

193 194 195 196 197 198 199 200 201 202 203 204
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;
	return ld->alloc_layout_hdr ? ld->alloc_layout_hdr(ino, gfp_flags) :
		kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags);
}

static void
pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
{
	struct pnfs_layoutdriver_type *ld = NFS_SERVER(lo->plh_inode)->pnfs_curr_ld;
205
	put_rpccred(lo->plh_lc_cred);
206 207 208
	return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo);
}

209
static void
210
destroy_layout_hdr(struct pnfs_layout_hdr *lo)
211
{
212 213 214
	dprintk("%s: freeing layout cache %p\n", __func__, lo);
	BUG_ON(!list_empty(&lo->plh_layouts));
	NFS_I(lo->plh_inode)->layout = NULL;
215
	pnfs_free_layout_hdr(lo);
216
}
217

218 219 220 221 222
static void
put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
{
	if (atomic_dec_and_test(&lo->plh_refcount))
		destroy_layout_hdr(lo);
223 224
}

225
void
226
put_layout_hdr(struct pnfs_layout_hdr *lo)
227
{
228 229 230 231 232 233
	struct inode *inode = lo->plh_inode;

	if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
		destroy_layout_hdr(lo);
		spin_unlock(&inode->i_lock);
	}
234 235 236 237 238
}

static void
init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
{
239
	INIT_LIST_HEAD(&lseg->pls_list);
240
	INIT_LIST_HEAD(&lseg->pls_lc_list);
241 242 243
	atomic_set(&lseg->pls_refcount, 1);
	smp_mb();
	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
244
	lseg->pls_layout = lo;
245 246
}

247
static void free_lseg(struct pnfs_layout_segment *lseg)
248
{
249
	struct inode *ino = lseg->pls_layout->plh_inode;
250

251
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
252
	/* Matched by get_layout_hdr in pnfs_insert_layout */
253
	put_layout_hdr(NFS_I(ino)->layout);
254 255
}

F
Fred Isaman 已提交
256 257 258 259 260
static void
put_lseg_common(struct pnfs_layout_segment *lseg)
{
	struct inode *inode = lseg->pls_layout->plh_inode;

261
	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
F
Fred Isaman 已提交
262 263 264 265 266 267 268 269 270
	list_del_init(&lseg->pls_list);
	if (list_empty(&lseg->pls_layout->plh_segs)) {
		set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
		/* Matched by initial refcount set in alloc_init_layout_hdr */
		put_layout_hdr_locked(lseg->pls_layout);
	}
	rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
}

271
void
F
Fred Isaman 已提交
272
put_lseg(struct pnfs_layout_segment *lseg)
273
{
F
Fred Isaman 已提交
274 275 276 277 278
	struct inode *inode;

	if (!lseg)
		return;

279 280 281
	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
		atomic_read(&lseg->pls_refcount),
		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
F
Fred Isaman 已提交
282 283 284
	inode = lseg->pls_layout->plh_inode;
	if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
		LIST_HEAD(free_me);
285

F
Fred Isaman 已提交
286 287 288 289
		put_lseg_common(lseg);
		list_add(&lseg->pls_list, &free_me);
		spin_unlock(&inode->i_lock);
		pnfs_free_lseg_list(&free_me);
290 291
	}
}
292
EXPORT_SYMBOL_GPL(put_lseg);
293

294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
static inline u64
end_offset(u64 start, u64 len)
{
	u64 end;

	end = start + len;
	return end >= start ? end : NFS4_MAX_UINT64;
}

/* last octet in a range */
static inline u64
last_byte_offset(u64 start, u64 len)
{
	u64 end;

	BUG_ON(!len);
	end = start + len;
	return end > start ? end - 1 : NFS4_MAX_UINT64;
}

/*
 * is l2 fully contained in l1?
 *   start1                             end1
 *   [----------------------------------)
 *           start2           end2
 *           [----------------)
 */
static inline int
lo_seg_contained(struct pnfs_layout_range *l1,
		 struct pnfs_layout_range *l2)
{
	u64 start1 = l1->offset;
	u64 end1 = end_offset(start1, l1->length);
	u64 start2 = l2->offset;
	u64 end2 = end_offset(start2, l2->length);

	return (start1 <= start2) && (end1 >= end2);
}

/*
 * is l1 and l2 intersecting?
 *   start1                             end1
 *   [----------------------------------)
 *                              start2           end2
 *                              [----------------)
 */
static inline int
lo_seg_intersecting(struct pnfs_layout_range *l1,
		    struct pnfs_layout_range *l2)
{
	u64 start1 = l1->offset;
	u64 end1 = end_offset(start1, l1->length);
	u64 start2 = l2->offset;
	u64 end2 = end_offset(start2, l2->length);

	return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
	       (end2 == NFS4_MAX_UINT64 || end2 > start1);
}

353
static bool
354 355
should_free_lseg(struct pnfs_layout_range *lseg_range,
		 struct pnfs_layout_range *recall_range)
356
{
357 358 359
	return (recall_range->iomode == IOMODE_ANY ||
		lseg_range->iomode == recall_range->iomode) &&
	       lo_seg_intersecting(lseg_range, recall_range);
360 361
}

362 363 364 365 366 367 368 369 370 371 372
/* 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
373 374 375 376 377 378 379
		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
			atomic_read(&lseg->pls_refcount));
		if (atomic_dec_and_test(&lseg->pls_refcount)) {
			put_lseg_common(lseg);
			list_add(&lseg->pls_list, tmp_list);
			rv = 1;
		}
380 381 382 383 384 385 386
	}
	return rv;
}

/* Returns count of number of matching invalid lsegs remaining in list
 * after call.
 */
F
Fred Isaman 已提交
387
int
388 389
mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
			    struct list_head *tmp_list,
390
			    struct pnfs_layout_range *recall_range)
391 392
{
	struct pnfs_layout_segment *lseg, *next;
393
	int invalid = 0, removed = 0;
394 395 396

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

397
	if (list_empty(&lo->plh_segs)) {
398 399 400
		/* Reset MDS Threshold I/O counters */
		NFS_I(lo->plh_inode)->write_io = 0;
		NFS_I(lo->plh_inode)->read_io = 0;
401 402 403 404
		if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
			put_layout_hdr_locked(lo);
		return 0;
	}
405
	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
406 407
		if (!recall_range ||
		    should_free_lseg(&lseg->pls_range, recall_range)) {
408 409 410 411 412 413 414 415 416
			dprintk("%s: freeing lseg %p iomode %d "
				"offset %llu length %llu\n", __func__,
				lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
				lseg->pls_range.length);
			invalid++;
			removed += mark_lseg_invalid(lseg, tmp_list);
		}
	dprintk("%s:Return %i\n", __func__, invalid - removed);
	return invalid - removed;
417 418
}

419
/* note free_me must contain lsegs from a single layout_hdr */
F
Fred Isaman 已提交
420
void
421
pnfs_free_lseg_list(struct list_head *free_me)
422
{
423
	struct pnfs_layout_segment *lseg, *tmp;
424 425 426 427 428 429 430
	struct pnfs_layout_hdr *lo;

	if (list_empty(free_me))
		return;

	lo = list_first_entry(free_me, struct pnfs_layout_segment,
			      pls_list)->pls_layout;
431

432 433 434 435 436 437 438 439
	if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
		struct nfs_client *clp;

		clp = NFS_SERVER(lo->plh_inode)->nfs_client;
		spin_lock(&clp->cl_lock);
		list_del_init(&lo->plh_layouts);
		spin_unlock(&clp->cl_lock);
	}
440
	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
441
		list_del(&lseg->pls_list);
442
		free_lseg(lseg);
443 444 445
	}
}

446 447 448 449
void
pnfs_destroy_layout(struct nfs_inode *nfsi)
{
	struct pnfs_layout_hdr *lo;
450
	LIST_HEAD(tmp_list);
451 452 453 454

	spin_lock(&nfsi->vfs_inode.i_lock);
	lo = nfsi->layout;
	if (lo) {
455
		lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
456
		mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
457 458
	}
	spin_unlock(&nfsi->vfs_inode.i_lock);
459 460
	pnfs_free_lseg_list(&tmp_list);
}
461
EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
462 463 464 465 466 467 468 469

/*
 * Called by the state manger to remove all layouts established under an
 * expired lease.
 */
void
pnfs_destroy_all_layouts(struct nfs_client *clp)
{
470
	struct nfs_server *server;
471 472 473
	struct pnfs_layout_hdr *lo;
	LIST_HEAD(tmp_list);

474 475 476
	nfs4_deviceid_mark_client_invalid(clp);
	nfs4_deviceid_purge_client(clp);

477
	spin_lock(&clp->cl_lock);
478 479 480 481 482 483
	rcu_read_lock();
	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
		if (!list_empty(&server->layouts))
			list_splice_init(&server->layouts, &tmp_list);
	}
	rcu_read_unlock();
484 485 486 487
	spin_unlock(&clp->cl_lock);

	while (!list_empty(&tmp_list)) {
		lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
488
				plh_layouts);
489
		dprintk("%s freeing layout for inode %lu\n", __func__,
490
			lo->plh_inode->i_ino);
491
		list_del_init(&lo->plh_layouts);
492
		pnfs_destroy_layout(NFS_I(lo->plh_inode));
493
	}
494 495
}

496
/* update lo->plh_stateid with new if is more recent */
F
Fred Isaman 已提交
497 498 499
void
pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
			bool update_barrier)
500
{
501
	u32 oldseq, newseq;
502

503 504
	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
	newseq = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
505
	if ((int)(newseq - oldseq) > 0) {
506
		nfs4_stateid_copy(&lo->plh_stateid, new);
F
Fred Isaman 已提交
507
		if (update_barrier) {
508
			u32 new_barrier = be32_to_cpu(new->seqid);
F
Fred Isaman 已提交
509 510 511 512 513 514 515 516 517 518 519 520 521 522 523

			if ((int)(new_barrier - lo->plh_barrier))
				lo->plh_barrier = new_barrier;
		} else {
			/* Because of wraparound, we want to keep the barrier
			 * "close" to the current seqids.  It needs to be
			 * within 2**31 to count as "behind", so if it
			 * gets too near that limit, give us a litle leeway
			 * and bring it to within 2**30.
			 * NOTE - and yes, this is all unsigned arithmetic.
			 */
			if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
				lo->plh_barrier = newseq - (1 << 30);
		}
	}
524 525
}

526 527
/* lget is set to 1 if called from inside send_layoutget call chain */
static bool
F
Fred Isaman 已提交
528 529 530 531
pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
			int lget)
{
	if ((stateid) &&
532
	    (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0)
F
Fred Isaman 已提交
533
		return true;
F
Fred Isaman 已提交
534
	return lo->plh_block_lgets ||
535
		test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
F
Fred Isaman 已提交
536
		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
F
Fred Isaman 已提交
537
		(list_empty(&lo->plh_segs) &&
538 539 540
		 (atomic_read(&lo->plh_outstanding) > lget));
}

541 542 543
int
pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
			      struct nfs4_state *open_state)
544
{
545
	int status = 0;
546

547
	dprintk("--> %s\n", __func__);
548
	spin_lock(&lo->plh_inode->i_lock);
F
Fred Isaman 已提交
549
	if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
550 551
		status = -EAGAIN;
	} else if (list_empty(&lo->plh_segs)) {
552 553 554 555
		int seq;

		do {
			seq = read_seqbegin(&open_state->seqlock);
556
			nfs4_stateid_copy(dst, &open_state->stateid);
557 558
		} while (read_seqretry(&open_state->seqlock, seq));
	} else
559
		nfs4_stateid_copy(dst, &lo->plh_stateid);
560
	spin_unlock(&lo->plh_inode->i_lock);
561
	dprintk("<-- %s\n", __func__);
562
	return status;
563 564 565 566 567 568 569 570
}

/*
* Get layout from server.
*    for now, assume that whole file layouts are requested.
*    arg->offset: 0
*    arg->length: all ones
*/
571 572 573
static struct pnfs_layout_segment *
send_layoutget(struct pnfs_layout_hdr *lo,
	   struct nfs_open_context *ctx,
574
	   struct pnfs_layout_range *range,
575
	   gfp_t gfp_flags)
576
{
577
	struct inode *ino = lo->plh_inode;
578 579 580
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs4_layoutget *lgp;
	struct pnfs_layout_segment *lseg = NULL;
581 582 583
	struct page **pages = NULL;
	int i;
	u32 max_resp_sz, max_pages;
584 585

	dprintk("--> %s\n", __func__);
586

587
	BUG_ON(ctx == NULL);
588
	lgp = kzalloc(sizeof(*lgp), gfp_flags);
589
	if (lgp == NULL)
590
		return NULL;
591 592 593

	/* allocate pages for xdr post processing */
	max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
594
	max_pages = nfs_page_array_len(0, max_resp_sz);
595

596
	pages = kcalloc(max_pages, sizeof(struct page *), gfp_flags);
597 598 599 600
	if (!pages)
		goto out_err_free;

	for (i = 0; i < max_pages; i++) {
601
		pages[i] = alloc_page(gfp_flags);
602 603 604 605
		if (!pages[i])
			goto out_err_free;
	}

606 607 608
	lgp->args.minlength = PAGE_CACHE_SIZE;
	if (lgp->args.minlength > range->length)
		lgp->args.minlength = range->length;
609
	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
610
	lgp->args.range = *range;
611 612 613
	lgp->args.type = server->pnfs_curr_ld->id;
	lgp->args.inode = ino;
	lgp->args.ctx = get_nfs_open_context(ctx);
614 615
	lgp->args.layout.pages = pages;
	lgp->args.layout.pglen = max_pages * PAGE_SIZE;
616
	lgp->lsegpp = &lseg;
617
	lgp->gfp_flags = gfp_flags;
618 619 620 621 622

	/* Synchronously retrieve layout information from server and
	 * store in lseg.
	 */
	nfs4_proc_layoutget(lgp);
623
	if (!lseg) {
624
		/* remember that LAYOUTGET failed and suspend trying */
625
		set_bit(lo_fail_bit(range->iomode), &lo->plh_flags);
626
	}
627 628 629 630 631 632

	/* free xdr pages */
	for (i = 0; i < max_pages; i++)
		__free_page(pages[i]);
	kfree(pages);

633
	return lseg;
634 635 636 637 638 639 640 641 642 643 644 645 646

out_err_free:
	/* free any allocated xdr pages, lgp as it's not used */
	if (pages) {
		for (i = 0; i < max_pages; i++) {
			if (!pages[i])
				break;
			__free_page(pages[i]);
		}
		kfree(pages);
	}
	kfree(lgp);
	return NULL;
647 648
}

B
Benny Halevy 已提交
649 650 651 652 653 654 655 656 657 658 659 660 661 662 663
/* Initiates a LAYOUTRETURN(FILE) */
int
_pnfs_return_layout(struct inode *ino)
{
	struct pnfs_layout_hdr *lo = NULL;
	struct nfs_inode *nfsi = NFS_I(ino);
	LIST_HEAD(tmp_list);
	struct nfs4_layoutreturn *lrp;
	nfs4_stateid stateid;
	int status = 0;

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

	spin_lock(&ino->i_lock);
	lo = nfsi->layout;
664
	if (!lo) {
B
Benny Halevy 已提交
665
		spin_unlock(&ino->i_lock);
666 667
		dprintk("%s: no layout to return\n", __func__);
		return status;
B
Benny Halevy 已提交
668 669 670 671
	}
	stateid = nfsi->layout->plh_stateid;
	/* Reference matched in nfs4_layoutreturn_release */
	get_layout_hdr(lo);
672 673
	mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
	lo->plh_block_lgets++;
B
Benny Halevy 已提交
674 675 676 677 678 679 680 681
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);

	WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags));

	lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
	if (unlikely(lrp == NULL)) {
		status = -ENOMEM;
682 683
		set_bit(NFS_LAYOUT_RW_FAILED, &lo->plh_flags);
		set_bit(NFS_LAYOUT_RO_FAILED, &lo->plh_flags);
684
		put_layout_hdr(lo);
B
Benny Halevy 已提交
685 686 687 688 689 690
		goto out;
	}

	lrp->args.stateid = stateid;
	lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
	lrp->args.inode = ino;
T
Trond Myklebust 已提交
691
	lrp->args.layout = lo;
B
Benny Halevy 已提交
692 693 694 695 696 697 698
	lrp->clp = NFS_SERVER(ino)->nfs_client;

	status = nfs4_proc_layoutreturn(lrp);
out:
	dprintk("<-- %s status: %d\n", __func__, status);
	return status;
}
699
EXPORT_SYMBOL_GPL(_pnfs_return_layout);
B
Benny Halevy 已提交
700

F
Fred Isaman 已提交
701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766
bool pnfs_roc(struct inode *ino)
{
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg, *tmp;
	LIST_HEAD(tmp_list);
	bool found = false;

	spin_lock(&ino->i_lock);
	lo = NFS_I(ino)->layout;
	if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
	    test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
		goto out_nolayout;
	list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
		if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
			mark_lseg_invalid(lseg, &tmp_list);
			found = true;
		}
	if (!found)
		goto out_nolayout;
	lo->plh_block_lgets++;
	get_layout_hdr(lo); /* matched in pnfs_roc_release */
	spin_unlock(&ino->i_lock);
	pnfs_free_lseg_list(&tmp_list);
	return true;

out_nolayout:
	spin_unlock(&ino->i_lock);
	return false;
}

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

	spin_lock(&ino->i_lock);
	lo = NFS_I(ino)->layout;
	lo->plh_block_lgets--;
	put_layout_hdr_locked(lo);
	spin_unlock(&ino->i_lock);
}

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;
	if ((int)(barrier - lo->plh_barrier) > 0)
		lo->plh_barrier = barrier;
	spin_unlock(&ino->i_lock);
}

bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
{
	struct nfs_inode *nfsi = NFS_I(ino);
	struct pnfs_layout_segment *lseg;
	bool found = false;

	spin_lock(&ino->i_lock);
	list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
		if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
			found = true;
			break;
		}
	if (!found) {
		struct pnfs_layout_hdr *lo = nfsi->layout;
767
		u32 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
F
Fred Isaman 已提交
768 769 770 771 772 773 774 775 776 777

		/* 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);
	}
	spin_unlock(&ino->i_lock);
	return found;
}

778 779 780 781 782 783
/*
 * 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
784 785
cmp_layout(struct pnfs_layout_range *l1,
	   struct pnfs_layout_range *l2)
786
{
787 788 789 790 791 792 793 794 795 796 797 798
	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;

799
	/* read > read/write */
800
	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
801 802
}

803 804 805 806
static void
pnfs_insert_layout(struct pnfs_layout_hdr *lo,
		   struct pnfs_layout_segment *lseg)
{
807 808
	struct pnfs_layout_segment *lp;

809 810
	dprintk("%s:Begin\n", __func__);

811 812
	assert_spin_locked(&lo->plh_inode->i_lock);
	list_for_each_entry(lp, &lo->plh_segs, pls_list) {
813
		if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
814
			continue;
815
		list_add_tail(&lseg->pls_list, &lp->pls_list);
816 817 818
		dprintk("%s: inserted lseg %p "
			"iomode %d offset %llu length %llu before "
			"lp %p iomode %d offset %llu length %llu\n",
819 820 821 822
			__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);
823
		goto out;
824
	}
825 826 827 828 829 830
	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:
831
	get_layout_hdr(lo);
832 833

	dprintk("%s:Return\n", __func__);
834 835 836
}

static struct pnfs_layout_hdr *
837 838 839
alloc_init_layout_hdr(struct inode *ino,
		      struct nfs_open_context *ctx,
		      gfp_t gfp_flags)
840 841 842
{
	struct pnfs_layout_hdr *lo;

843
	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
844 845
	if (!lo)
		return NULL;
846
	atomic_set(&lo->plh_refcount, 1);
847 848
	INIT_LIST_HEAD(&lo->plh_layouts);
	INIT_LIST_HEAD(&lo->plh_segs);
F
Fred Isaman 已提交
849
	INIT_LIST_HEAD(&lo->plh_bulk_recall);
850
	lo->plh_inode = ino;
851
	lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
852 853 854 855
	return lo;
}

static struct pnfs_layout_hdr *
856 857 858
pnfs_find_alloc_layout(struct inode *ino,
		       struct nfs_open_context *ctx,
		       gfp_t gfp_flags)
859 860 861 862 863 864 865
{
	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);

	assert_spin_locked(&ino->i_lock);
866 867 868 869 870 871
	if (nfsi->layout) {
		if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
			return NULL;
		else
			return nfsi->layout;
	}
872
	spin_unlock(&ino->i_lock);
873
	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
874 875 876 877 878
	spin_lock(&ino->i_lock);

	if (likely(nfsi->layout == NULL))	/* Won the race? */
		nfsi->layout = new;
	else
879
		pnfs_free_layout_hdr(new);
880 881 882
	return nfsi->layout;
}

883 884 885 886 887 888 889 890 891 892 893 894
/*
 * iomode matching rules:
 * iomode	lseg	match
 * -----	-----	-----
 * ANY		READ	true
 * ANY		RW	true
 * RW		READ	false
 * RW		RW	true
 * READ		READ	true
 * READ		RW	true
 */
static int
895 896
is_matching_lseg(struct pnfs_layout_range *ls_range,
		 struct pnfs_layout_range *range)
897
{
898 899 900 901 902 903 904 905 906 907 908
	struct pnfs_layout_range range1;

	if ((range->iomode == IOMODE_RW &&
	     ls_range->iomode != IOMODE_RW) ||
	    !lo_seg_intersecting(ls_range, range))
		return 0;

	/* range1 covers only the first byte in the range */
	range1 = *range;
	range1.length = 1;
	return lo_seg_contained(ls_range, &range1);
909 910 911 912 913
}

/*
 * lookup range in layout
 */
914
static struct pnfs_layout_segment *
915 916
pnfs_find_lseg(struct pnfs_layout_hdr *lo,
		struct pnfs_layout_range *range)
917
{
918 919 920 921
	struct pnfs_layout_segment *lseg, *ret = NULL;

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

922 923
	assert_spin_locked(&lo->plh_inode->i_lock);
	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
924
		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
925
		    is_matching_lseg(&lseg->pls_range, range)) {
F
Fred Isaman 已提交
926
			ret = get_lseg(lseg);
927 928
			break;
		}
929
		if (lseg->pls_range.offset > range->offset)
930 931 932 933
			break;
	}

	dprintk("%s:Return lseg %p ref %d\n",
934
		__func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
935
	return ret;
936 937
}

938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012
/*
 * 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;
}

1013 1014 1015 1016
/*
 * Layout segment is retreived from the server if not cached.
 * The appropriate layout segment is referenced and returned to the caller.
 */
1017
struct pnfs_layout_segment *
1018 1019
pnfs_update_layout(struct inode *ino,
		   struct nfs_open_context *ctx,
1020 1021
		   loff_t pos,
		   u64 count,
1022 1023
		   enum pnfs_iomode iomode,
		   gfp_t gfp_flags)
1024
{
1025 1026 1027 1028 1029
	struct pnfs_layout_range arg = {
		.iomode = iomode,
		.offset = pos,
		.length = count,
	};
1030
	unsigned pg_offset;
1031
	struct nfs_inode *nfsi = NFS_I(ino);
1032 1033
	struct nfs_server *server = NFS_SERVER(ino);
	struct nfs_client *clp = server->nfs_client;
1034 1035
	struct pnfs_layout_hdr *lo;
	struct pnfs_layout_segment *lseg = NULL;
1036
	bool first = false;
1037 1038 1039

	if (!pnfs_enabled_sb(NFS_SERVER(ino)))
		return NULL;
1040 1041 1042 1043

	if (pnfs_within_mdsthreshold(ctx, ino, iomode))
		return NULL;

1044
	spin_lock(&ino->i_lock);
1045
	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1046 1047 1048 1049 1050
	if (lo == NULL) {
		dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
		goto out_unlock;
	}

F
Fred Isaman 已提交
1051
	/* Do we even need to bother with this? */
1052
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1053
		dprintk("%s matches recall, use MDS\n", __func__);
1054 1055 1056 1057
		goto out_unlock;
	}

	/* if LAYOUTGET already failed once we don't try again */
1058
	if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
1059 1060
		goto out_unlock;

1061
	/* Check to see if the layout for the given range already exists */
1062
	lseg = pnfs_find_lseg(lo, &arg);
1063 1064 1065
	if (lseg)
		goto out_unlock;

F
Fred Isaman 已提交
1066
	if (pnfs_layoutgets_blocked(lo, NULL, 0))
1067 1068 1069
		goto out_unlock;
	atomic_inc(&lo->plh_outstanding);

1070
	get_layout_hdr(lo);
1071 1072 1073 1074
	if (list_empty(&lo->plh_segs))
		first = true;
	spin_unlock(&ino->i_lock);
	if (first) {
1075 1076 1077 1078 1079
		/* 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);
		BUG_ON(!list_empty(&lo->plh_layouts));
1080
		list_add_tail(&lo->plh_layouts, &server->layouts);
1081 1082
		spin_unlock(&clp->cl_lock);
	}
1083

1084 1085 1086 1087 1088
	pg_offset = arg.offset & ~PAGE_CACHE_MASK;
	if (pg_offset) {
		arg.offset -= pg_offset;
		arg.length += pg_offset;
	}
1089 1090
	if (arg.length != NFS4_MAX_UINT64)
		arg.length = PAGE_CACHE_ALIGN(arg.length);
1091

1092
	lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1093 1094 1095 1096
	if (!lseg && first) {
		spin_lock(&clp->cl_lock);
		list_del_init(&lo->plh_layouts);
		spin_unlock(&clp->cl_lock);
1097
	}
1098
	atomic_dec(&lo->plh_outstanding);
1099
	put_layout_hdr(lo);
1100 1101
out:
	dprintk("%s end, state 0x%lx lseg %p\n", __func__,
1102
		nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
1103 1104 1105 1106 1107
	return lseg;
out_unlock:
	spin_unlock(&ino->i_lock);
	goto out;
}
1108
EXPORT_SYMBOL_GPL(pnfs_update_layout);
1109 1110 1111 1112 1113 1114 1115

int
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;
1116
	struct inode *ino = lo->plh_inode;
1117 1118 1119
	int status = 0;

	/* Inject layout blob into I/O device driver */
1120
	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131
	if (!lseg || IS_ERR(lseg)) {
		if (!lseg)
			status = -ENOMEM;
		else
			status = PTR_ERR(lseg);
		dprintk("%s: Could not allocate layout: error %d\n",
		       __func__, status);
		goto out;
	}

	spin_lock(&ino->i_lock);
1132
	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
F
Fred Isaman 已提交
1133 1134 1135 1136 1137 1138 1139 1140
		dprintk("%s forget reply due to recall\n", __func__);
		goto out_forget_reply;
	}

	if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
		dprintk("%s forget reply due to state\n", __func__);
		goto out_forget_reply;
	}
1141
	init_lseg(lo, lseg);
1142
	lseg->pls_range = res->range;
F
Fred Isaman 已提交
1143
	*lgp->lsegpp = get_lseg(lseg);
1144 1145
	pnfs_insert_layout(lo, lseg);

F
Fred Isaman 已提交
1146 1147 1148 1149 1150
	if (res->return_on_close) {
		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
		set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
	}

1151
	/* Done processing layoutget. Set the layout stateid */
F
Fred Isaman 已提交
1152
	pnfs_set_layout_stateid(lo, &res->stateid, false);
1153 1154 1155
	spin_unlock(&ino->i_lock);
out:
	return status;
F
Fred Isaman 已提交
1156 1157 1158 1159 1160 1161

out_forget_reply:
	spin_unlock(&ino->i_lock);
	lseg->pls_layout = lo;
	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
	goto out;
1162 1163
}

1164 1165 1166 1167 1168
void
pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
	BUG_ON(pgio->pg_lseg != NULL);

1169 1170 1171 1172
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_read_mds(pgio);
		return;
	}
1173 1174 1175 1176 1177 1178
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
					   req->wb_bytes,
					   IOMODE_READ,
					   GFP_KERNEL);
1179 1180
	/* If no lseg, fall back to read through mds */
	if (pgio->pg_lseg == NULL)
1181
		nfs_pageio_reset_read_mds(pgio);
1182

1183 1184 1185 1186 1187 1188 1189 1190
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);

void
pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
{
	BUG_ON(pgio->pg_lseg != NULL);

1191 1192 1193 1194
	if (req->wb_offset != req->wb_pgbase) {
		nfs_pageio_reset_write_mds(pgio);
		return;
	}
1195 1196 1197 1198 1199 1200
	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
					   req->wb_context,
					   req_offset(req),
					   req->wb_bytes,
					   IOMODE_RW,
					   GFP_NOFS);
1201 1202
	/* If no lseg, fall back to write through mds */
	if (pgio->pg_lseg == NULL)
1203
		nfs_pageio_reset_write_mds(pgio);
1204 1205 1206
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);

1207
bool
1208 1209
pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode,
		      const struct nfs_pgio_completion_ops *compl_ops)
1210 1211 1212 1213 1214 1215
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (ld == NULL)
		return false;
1216 1217
	nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops,
			server->rsize, 0);
1218 1219 1220 1221
	return true;
}

bool
1222 1223 1224
pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode,
		       int ioflags,
		       const struct nfs_pgio_completion_ops *compl_ops)
1225 1226 1227 1228 1229 1230
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;

	if (ld == NULL)
		return false;
1231 1232
	nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops,
			server->wsize, ioflags);
1233 1234 1235
	return true;
}

1236
bool
1237 1238
pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
		     struct nfs_page *req)
1239
{
1240 1241
	if (pgio->pg_lseg == NULL)
		return nfs_generic_pg_test(pgio, prev, req);
1242

1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258
	/*
	 * Test if a nfs_page is fully contained in the pnfs_layout_range.
	 * Note that this test makes several assumptions:
	 * - that the previous nfs_page in the struct nfs_pageio_descriptor
	 *   is known to lie within the range.
	 *   - that the nfs_page being tested is known to be contiguous with the
	 *   previous nfs_page.
	 *   - Layout ranges are page aligned, so we only have to test the
	 *   start offset of the request.
	 *
	 * Please also note that 'end_offset' is actually the offset of the
	 * first byte that lies outside the pnfs_layout_range. FIXME?
	 *
	 */
	return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
					 pgio->pg_lseg->pls_range.length);
1259
}
1260
EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1261

1262
int pnfs_write_done_resend_to_mds(struct inode *inode,
1263 1264
				struct list_head *head,
				const struct nfs_pgio_completion_ops *compl_ops)
1265 1266 1267 1268 1269
{
	struct nfs_pageio_descriptor pgio;
	LIST_HEAD(failed);

	/* Resend all requests through the MDS */
1270
	nfs_pageio_init_write_mds(&pgio, inode, FLUSH_STABLE, compl_ops);
1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);

		nfs_list_remove_request(req);
		if (!nfs_pageio_add_request(&pgio, req))
			nfs_list_add_request(req, &failed);
	}
	nfs_pageio_complete(&pgio);

	if (!list_empty(&failed)) {
		/* For some reason our attempt to resend pages. Mark the
		 * overall send request as having failed, and let
		 * nfs_writeback_release_full deal with the error.
		 */
		list_move(&failed, head);
		return -EIO;
	}
	return 0;
}
1290
EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1291

1292 1293
static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
{
1294 1295 1296 1297
	struct nfs_pgio_header *hdr = data->header;

	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1298
	    PNFS_LAYOUTRET_ON_ERROR) {
1299 1300
		clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
		pnfs_return_layout(hdr->inode);
1301
	}
1302 1303
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
1304 1305
							&hdr->pages,
							hdr->completion_ops);
1306 1307
}

1308 1309 1310
/*
 * Called by non rpc-based layout drivers
 */
1311
void pnfs_ld_write_done(struct nfs_write_data *data)
1312
{
1313 1314 1315
	struct nfs_pgio_header *hdr = data->header;

	if (!hdr->pnfs_error) {
1316
		pnfs_set_layoutcommit(data);
1317
		hdr->mds_ops->rpc_call_done(&data->task, data);
1318 1319
	} else
		pnfs_ld_handle_write_error(data);
1320
	hdr->mds_ops->rpc_release(data);
1321
}
1322
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1323

1324 1325 1326 1327
static void
pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
		struct nfs_write_data *data)
{
1328 1329
	struct nfs_pgio_header *hdr = data->header;

1330 1331 1332 1333 1334
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
		list_splice_tail_init(&hdr->pages, &desc->pg_list);
		nfs_pageio_reset_write_mds(desc);
		desc->pg_recoalesce = 1;
	}
1335 1336 1337 1338
	nfs_writedata_release(data);
}

static enum pnfs_try_status
1339
pnfs_try_to_write_data(struct nfs_write_data *wdata,
1340 1341 1342
			const struct rpc_call_ops *call_ops,
			struct pnfs_layout_segment *lseg,
			int how)
1343
{
1344 1345
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
1346 1347 1348
	enum pnfs_try_status trypnfs;
	struct nfs_server *nfss = NFS_SERVER(inode);

1349
	hdr->mds_ops = call_ops;
1350 1351 1352 1353

	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
		inode->i_ino, wdata->args.count, wdata->args.offset, how);
	trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
1354
	if (trypnfs != PNFS_NOT_ATTEMPTED)
1355 1356 1357 1358 1359
		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}

1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370
static void
pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
{
	struct nfs_write_data *data;
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;

	desc->pg_lseg = NULL;
	while (!list_empty(head)) {
		enum pnfs_try_status trypnfs;

1371
		data = list_first_entry(head, struct nfs_write_data, list);
1372 1373 1374 1375 1376 1377 1378 1379 1380
		list_del_init(&data->list);

		trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
		if (trypnfs == PNFS_NOT_ATTEMPTED)
			pnfs_write_through_mds(desc, data);
	}
	put_lseg(lseg);
}

1381 1382 1383 1384 1385 1386
static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
{
	put_lseg(hdr->lseg);
	nfs_writehdr_free(hdr);
}

1387 1388 1389
int
pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
{
1390 1391
	struct nfs_write_header *whdr;
	struct nfs_pgio_header *hdr;
1392 1393
	int ret;

1394 1395
	whdr = nfs_writehdr_alloc();
	if (!whdr) {
1396
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1397 1398
		put_lseg(desc->pg_lseg);
		desc->pg_lseg = NULL;
1399
		return -ENOMEM;
1400
	}
1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411
	hdr = &whdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
	hdr->lseg = get_lseg(desc->pg_lseg);
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_flush(desc, hdr);
	if (ret != 0) {
		put_lseg(desc->pg_lseg);
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
	if (atomic_dec_and_test(&hdr->refcnt))
1412
		hdr->completion_ops->completion(hdr);
1413
	return ret;
1414 1415 1416
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);

1417
int pnfs_read_done_resend_to_mds(struct inode *inode,
1418 1419
				struct list_head *head,
				const struct nfs_pgio_completion_ops *compl_ops)
1420 1421
{
	struct nfs_pageio_descriptor pgio;
1422
	LIST_HEAD(failed);
1423

1424
	/* Resend all requests through the MDS */
1425
	nfs_pageio_init_read_mds(&pgio, inode, compl_ops);
1426 1427
	while (!list_empty(head)) {
		struct nfs_page *req = nfs_list_entry(head->next);
1428 1429

		nfs_list_remove_request(req);
1430 1431
		if (!nfs_pageio_add_request(&pgio, req))
			nfs_list_add_request(req, &failed);
1432 1433
	}
	nfs_pageio_complete(&pgio);
1434 1435 1436 1437 1438 1439 1440

	if (!list_empty(&failed)) {
		list_move(&failed, head);
		return -EIO;
	}
	return 0;
}
1441
EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1442 1443 1444

static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
{
1445 1446 1447 1448
	struct nfs_pgio_header *hdr = data->header;

	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1449
	    PNFS_LAYOUTRET_ON_ERROR) {
1450 1451
		clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
		pnfs_return_layout(hdr->inode);
1452
	}
1453 1454
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
		data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
1455 1456
							&hdr->pages,
							hdr->completion_ops);
1457 1458
}

1459 1460 1461
/*
 * Called by non rpc-based layout drivers
 */
1462
void pnfs_ld_read_done(struct nfs_read_data *data)
1463
{
1464 1465 1466
	struct nfs_pgio_header *hdr = data->header;

	if (likely(!hdr->pnfs_error)) {
1467
		__nfs4_read_done_cb(data);
1468
		hdr->mds_ops->rpc_call_done(&data->task, data);
1469 1470
	} else
		pnfs_ld_handle_read_error(data);
1471
	hdr->mds_ops->rpc_release(data);
1472 1473 1474
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

1475 1476 1477 1478
static void
pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
		struct nfs_read_data *data)
{
1479 1480
	struct nfs_pgio_header *hdr = data->header;

1481 1482 1483 1484 1485
	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
		list_splice_tail_init(&hdr->pages, &desc->pg_list);
		nfs_pageio_reset_read_mds(desc);
		desc->pg_recoalesce = 1;
	}
1486 1487 1488
	nfs_readdata_release(data);
}

A
Andy Adamson 已提交
1489 1490 1491
/*
 * Call the appropriate parallel I/O subsystem read function.
 */
1492
static enum pnfs_try_status
A
Andy Adamson 已提交
1493
pnfs_try_to_read_data(struct nfs_read_data *rdata,
1494 1495
		       const struct rpc_call_ops *call_ops,
		       struct pnfs_layout_segment *lseg)
A
Andy Adamson 已提交
1496
{
1497 1498
	struct nfs_pgio_header *hdr = rdata->header;
	struct inode *inode = hdr->inode;
A
Andy Adamson 已提交
1499 1500 1501
	struct nfs_server *nfss = NFS_SERVER(inode);
	enum pnfs_try_status trypnfs;

1502
	hdr->mds_ops = call_ops;
A
Andy Adamson 已提交
1503 1504 1505 1506 1507

	dprintk("%s: Reading ino:%lu %u@%llu\n",
		__func__, inode->i_ino, rdata->args.count, rdata->args.offset);

	trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
1508
	if (trypnfs != PNFS_NOT_ATTEMPTED)
A
Andy Adamson 已提交
1509 1510 1511 1512
		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
	return trypnfs;
}
A
Andy Adamson 已提交
1513

1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524
static void
pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
{
	struct nfs_read_data *data;
	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
	struct pnfs_layout_segment *lseg = desc->pg_lseg;

	desc->pg_lseg = NULL;
	while (!list_empty(head)) {
		enum pnfs_try_status trypnfs;

1525
		data = list_first_entry(head, struct nfs_read_data, list);
1526 1527 1528 1529 1530 1531 1532 1533 1534
		list_del_init(&data->list);

		trypnfs = pnfs_try_to_read_data(data, call_ops, lseg);
		if (trypnfs == PNFS_NOT_ATTEMPTED)
			pnfs_read_through_mds(desc, data);
	}
	put_lseg(lseg);
}

1535 1536 1537 1538 1539 1540
static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
{
	put_lseg(hdr->lseg);
	nfs_readhdr_free(hdr);
}

1541 1542 1543
int
pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
{
1544 1545
	struct nfs_read_header *rhdr;
	struct nfs_pgio_header *hdr;
1546 1547
	int ret;

1548 1549
	rhdr = nfs_readhdr_alloc();
	if (!rhdr) {
1550
		desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1551
		ret = -ENOMEM;
1552 1553 1554 1555
		put_lseg(desc->pg_lseg);
		desc->pg_lseg = NULL;
		return ret;
	}
1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566
	hdr = &rhdr->header;
	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
	hdr->lseg = get_lseg(desc->pg_lseg);
	atomic_inc(&hdr->refcnt);
	ret = nfs_generic_pagein(desc, hdr);
	if (ret != 0) {
		put_lseg(desc->pg_lseg);
		desc->pg_lseg = NULL;
	} else
		pnfs_do_multiple_reads(desc, &hdr->rpc_list);
	if (atomic_dec_and_test(&hdr->refcnt))
1567
		hdr->completion_ops->completion(hdr);
1568
	return ret;
1569 1570 1571
}
EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);

A
Andy Adamson 已提交
1572
/*
1573
 * There can be multiple RW segments.
A
Andy Adamson 已提交
1574
 */
1575
static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
A
Andy Adamson 已提交
1576
{
1577
	struct pnfs_layout_segment *lseg;
A
Andy Adamson 已提交
1578

1579 1580 1581 1582 1583
	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
		if (lseg->pls_range.iomode == IOMODE_RW &&
		    test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
			list_add(&lseg->pls_lc_list, listp);
	}
A
Andy Adamson 已提交
1584 1585
}

P
Peng Tao 已提交
1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597
void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
{
	if (lseg->pls_range.iomode == IOMODE_RW) {
		dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__);
		set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
	} else {
		dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__);
		set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
	}
}
EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);

A
Andy Adamson 已提交
1598 1599 1600
void
pnfs_set_layoutcommit(struct nfs_write_data *wdata)
{
1601 1602 1603
	struct nfs_pgio_header *hdr = wdata->header;
	struct inode *inode = hdr->inode;
	struct nfs_inode *nfsi = NFS_I(inode);
1604
	loff_t end_pos = wdata->mds_offset + wdata->res.count;
1605
	bool mark_as_dirty = false;
A
Andy Adamson 已提交
1606

1607
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1608
	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1609
		mark_as_dirty = true;
A
Andy Adamson 已提交
1610
		dprintk("%s: Set layoutcommit for inode %lu ",
1611
			__func__, inode->i_ino);
A
Andy Adamson 已提交
1612
	}
1613
	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
1614
		/* references matched in nfs4_layoutcommit_release */
1615
		get_lseg(hdr->lseg);
1616
	}
1617 1618
	if (end_pos > nfsi->layout->plh_lwb)
		nfsi->layout->plh_lwb = end_pos;
1619
	spin_unlock(&inode->i_lock);
1620
	dprintk("%s: lseg %p end_pos %llu\n",
1621
		__func__, hdr->lseg, nfsi->layout->plh_lwb);
1622 1623 1624 1625

	/* 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)
1626
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1627 1628 1629
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

A
Andy Adamson 已提交
1630 1631 1632 1633 1634 1635 1636 1637
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);
}

1638 1639 1640 1641 1642 1643 1644 1645
/*
 * 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
1646
int
1647
pnfs_layoutcommit_inode(struct inode *inode, bool sync)
A
Andy Adamson 已提交
1648 1649 1650 1651 1652 1653 1654 1655
{
	struct nfs4_layoutcommit_data *data;
	struct nfs_inode *nfsi = NFS_I(inode);
	loff_t end_pos;
	int status = 0;

	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);

1656 1657 1658
	if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		return 0;

A
Andy Adamson 已提交
1659 1660
	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
	data = kzalloc(sizeof(*data), GFP_NOFS);
1661 1662 1663 1664
	if (!data) {
		status = -ENOMEM;
		goto out;
	}
A
Andy Adamson 已提交
1665

P
Peng Tao 已提交
1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679
	if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
		goto out_free;

	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
		if (!sync) {
			status = -EAGAIN;
			goto out_free;
		}
		status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING,
					nfs_wait_bit_killable, TASK_KILLABLE);
		if (status)
			goto out_free;
	}

1680
	INIT_LIST_HEAD(&data->lseg_list);
1681
	spin_lock(&inode->i_lock);
A
Andy Adamson 已提交
1682
	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
P
Peng Tao 已提交
1683
		clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
A
Andy Adamson 已提交
1684
		spin_unlock(&inode->i_lock);
P
Peng Tao 已提交
1685 1686
		wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
		goto out_free;
A
Andy Adamson 已提交
1687
	}
1688 1689

	pnfs_list_write_lseg(inode, &data->lseg_list);
A
Andy Adamson 已提交
1690

1691 1692
	end_pos = nfsi->layout->plh_lwb;
	nfsi->layout->plh_lwb = 0;
A
Andy Adamson 已提交
1693

1694
	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
A
Andy Adamson 已提交
1695 1696 1697
	spin_unlock(&inode->i_lock);

	data->args.inode = inode;
1698
	data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
A
Andy Adamson 已提交
1699 1700 1701 1702 1703 1704 1705 1706
	nfs_fattr_init(&data->fattr);
	data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
	data->res.fattr = &data->fattr;
	data->args.lastbytewritten = end_pos - 1;
	data->res.server = NFS_SERVER(inode);

	status = nfs4_proc_layoutcommit(data, sync);
out:
P
Peng Tao 已提交
1707 1708
	if (status)
		mark_inode_dirty_sync(inode);
A
Andy Adamson 已提交
1709 1710
	dprintk("<-- %s status %d\n", __func__, status);
	return status;
P
Peng Tao 已提交
1711 1712 1713
out_free:
	kfree(data);
	goto out;
A
Andy Adamson 已提交
1714
}
1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726

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;
}