xref: /linux/fs/nfs/pnfs.c (revision 4eb7ae7a301d3586c3351e81d5c3cfe2304a1a6a)
1 /*
2  *  pNFS functions to call and manage layout drivers.
3  *
4  *  Copyright (c) 2002 [year of first publication]
5  *  The Regents of the University of Michigan
6  *  All Rights Reserved
7  *
8  *  Dean Hildebrand <dhildebz@umich.edu>
9  *
10  *  Permission is granted to use, copy, create derivative works, and
11  *  redistribute this software and such derivative works for any purpose,
12  *  so long as the name of the University of Michigan is not used in
13  *  any advertising or publicity pertaining to the use or distribution
14  *  of this software without specific, written prior authorization. If
15  *  the above copyright notice or any other identification of the
16  *  University of Michigan is included in any copy of any portion of
17  *  this software, then the disclaimer below must also be included.
18  *
19  *  This software is provided as is, without representation or warranty
20  *  of any kind either express or implied, including without limitation
21  *  the implied warranties of merchantability, fitness for a particular
22  *  purpose, or noninfringement.  The Regents of the University of
23  *  Michigan shall not be liable for any damages, including special,
24  *  indirect, incidental, or consequential damages, with respect to any
25  *  claim arising out of or in connection with the use of the software,
26  *  even if it has been or is hereafter advised of the possibility of
27  *  such damages.
28  */
29 
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_page.h>
32 #include <linux/module.h>
33 #include <linux/sort.h>
34 #include "internal.h"
35 #include "pnfs.h"
36 #include "iostat.h"
37 #include "nfs4trace.h"
38 #include "delegation.h"
39 #include "nfs42.h"
40 #include "nfs4_fs.h"
41 
42 #define NFSDBG_FACILITY		NFSDBG_PNFS
43 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
44 
45 /* Locking:
46  *
47  * pnfs_spinlock:
48  *      protects pnfs_modules_tbl.
49  */
50 static DEFINE_SPINLOCK(pnfs_spinlock);
51 
52 /*
53  * pnfs_modules_tbl holds all pnfs modules
54  */
55 static LIST_HEAD(pnfs_modules_tbl);
56 
57 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
58 static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 		struct list_head *free_me,
60 		const struct pnfs_layout_range *range,
61 		u32 seq);
62 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 		                struct list_head *tmp_list);
64 
65 /* Return the registered pnfs layout driver module matching given id */
66 static struct pnfs_layoutdriver_type *
67 find_pnfs_driver_locked(u32 id)
68 {
69 	struct pnfs_layoutdriver_type *local;
70 
71 	list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 		if (local->id == id)
73 			goto out;
74 	local = NULL;
75 out:
76 	dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 	return local;
78 }
79 
80 static struct pnfs_layoutdriver_type *
81 find_pnfs_driver(u32 id)
82 {
83 	struct pnfs_layoutdriver_type *local;
84 
85 	spin_lock(&pnfs_spinlock);
86 	local = find_pnfs_driver_locked(id);
87 	if (local != NULL && !try_module_get(local->owner)) {
88 		dprintk("%s: Could not grab reference on module\n", __func__);
89 		local = NULL;
90 	}
91 	spin_unlock(&pnfs_spinlock);
92 	return local;
93 }
94 
95 void
96 unset_pnfs_layoutdriver(struct nfs_server *nfss)
97 {
98 	if (nfss->pnfs_curr_ld) {
99 		if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 			nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
101 		/* Decrement the MDS count. Purge the deviceid cache if zero */
102 		if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 			nfs4_deviceid_purge_client(nfss->nfs_client);
104 		module_put(nfss->pnfs_curr_ld->owner);
105 	}
106 	nfss->pnfs_curr_ld = NULL;
107 }
108 
109 /*
110  * When the server sends a list of layout types, we choose one in the order
111  * given in the list below.
112  *
113  * FIXME: should this list be configurable in some fashion? module param?
114  * 	  mount option? something else?
115  */
116 static const u32 ld_prefs[] = {
117 	LAYOUT_SCSI,
118 	LAYOUT_BLOCK_VOLUME,
119 	LAYOUT_OSD2_OBJECTS,
120 	LAYOUT_FLEX_FILES,
121 	LAYOUT_NFSV4_1_FILES,
122 	0
123 };
124 
125 static int
126 ld_cmp(const void *e1, const void *e2)
127 {
128 	u32 ld1 = *((u32 *)e1);
129 	u32 ld2 = *((u32 *)e2);
130 	int i;
131 
132 	for (i = 0; ld_prefs[i] != 0; i++) {
133 		if (ld1 == ld_prefs[i])
134 			return -1;
135 
136 		if (ld2 == ld_prefs[i])
137 			return 1;
138 	}
139 	return 0;
140 }
141 
142 /*
143  * Try to set the server's pnfs module to the pnfs layout type specified by id.
144  * Currently only one pNFS layout driver per filesystem is supported.
145  *
146  * @ids array of layout types supported by MDS.
147  */
148 void
149 set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
150 		      struct nfs_fsinfo *fsinfo)
151 {
152 	struct pnfs_layoutdriver_type *ld_type = NULL;
153 	u32 id;
154 	int i;
155 
156 	if (fsinfo->nlayouttypes == 0)
157 		goto out_no_driver;
158 	if (!(server->nfs_client->cl_exchange_flags &
159 		 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
160 		printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 			__func__, server->nfs_client->cl_exchange_flags);
162 		goto out_no_driver;
163 	}
164 
165 	sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 		sizeof(*fsinfo->layouttype), ld_cmp, NULL);
167 
168 	for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 		id = fsinfo->layouttype[i];
170 		ld_type = find_pnfs_driver(id);
171 		if (!ld_type) {
172 			request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 					id);
174 			ld_type = find_pnfs_driver(id);
175 		}
176 		if (ld_type)
177 			break;
178 	}
179 
180 	if (!ld_type) {
181 		dprintk("%s: No pNFS module found!\n", __func__);
182 		goto out_no_driver;
183 	}
184 
185 	server->pnfs_curr_ld = ld_type;
186 	if (ld_type->set_layoutdriver
187 	    && ld_type->set_layoutdriver(server, mntfh)) {
188 		printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 			"driver %u.\n", __func__, id);
190 		module_put(ld_type->owner);
191 		goto out_no_driver;
192 	}
193 	/* Bump the MDS count */
194 	atomic_inc(&server->nfs_client->cl_mds_count);
195 
196 	dprintk("%s: pNFS module for %u set\n", __func__, id);
197 	return;
198 
199 out_no_driver:
200 	dprintk("%s: Using NFSv4 I/O\n", __func__);
201 	server->pnfs_curr_ld = NULL;
202 }
203 
204 int
205 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206 {
207 	int status = -EINVAL;
208 	struct pnfs_layoutdriver_type *tmp;
209 
210 	if (ld_type->id == 0) {
211 		printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
212 		return status;
213 	}
214 	if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
215 		printk(KERN_ERR "NFS: %s Layout driver must provide "
216 		       "alloc_lseg and free_lseg.\n", __func__);
217 		return status;
218 	}
219 
220 	spin_lock(&pnfs_spinlock);
221 	tmp = find_pnfs_driver_locked(ld_type->id);
222 	if (!tmp) {
223 		list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 		status = 0;
225 		dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 			ld_type->name);
227 	} else {
228 		printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
229 			__func__, ld_type->id);
230 	}
231 	spin_unlock(&pnfs_spinlock);
232 
233 	return status;
234 }
235 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236 
237 void
238 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239 {
240 	dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 	spin_lock(&pnfs_spinlock);
242 	list_del(&ld_type->pnfs_tblid);
243 	spin_unlock(&pnfs_spinlock);
244 }
245 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
246 
247 /*
248  * pNFS client layout cache
249  */
250 
251 /* Need to hold i_lock if caller does not already hold reference */
252 void
253 pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
254 {
255 	refcount_inc(&lo->plh_refcount);
256 }
257 
258 static struct pnfs_layout_hdr *
259 pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260 {
261 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
262 	return ld->alloc_layout_hdr(ino, gfp_flags);
263 }
264 
265 static void
266 pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267 {
268 	struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270 
271 	if (!list_empty(&lo->plh_layouts)) {
272 		struct nfs_client *clp = server->nfs_client;
273 
274 		spin_lock(&clp->cl_lock);
275 		list_del_init(&lo->plh_layouts);
276 		spin_unlock(&clp->cl_lock);
277 	}
278 	put_cred(lo->plh_lc_cred);
279 	return ld->free_layout_hdr(lo);
280 }
281 
282 static void
283 pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
284 {
285 	struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
286 	dprintk("%s: freeing layout cache %p\n", __func__, lo);
287 	nfsi->layout = NULL;
288 	/* Reset MDS Threshold I/O counters */
289 	nfsi->write_io = 0;
290 	nfsi->read_io = 0;
291 }
292 
293 void
294 pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
295 {
296 	struct inode *inode;
297 
298 	if (!lo)
299 		return;
300 	inode = lo->plh_inode;
301 	pnfs_layoutreturn_before_put_layout_hdr(lo);
302 
303 	if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
304 		if (!list_empty(&lo->plh_segs))
305 			WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
306 		pnfs_detach_layout_hdr(lo);
307 		spin_unlock(&inode->i_lock);
308 		pnfs_free_layout_hdr(lo);
309 	}
310 }
311 
312 static void
313 pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
314 			 u32 seq)
315 {
316 	if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
317 		iomode = IOMODE_ANY;
318 	lo->plh_return_iomode = iomode;
319 	set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
320 	if (seq != 0) {
321 		WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
322 		lo->plh_return_seq = seq;
323 	}
324 }
325 
326 static void
327 pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
328 {
329 	struct pnfs_layout_segment *lseg;
330 	lo->plh_return_iomode = 0;
331 	lo->plh_return_seq = 0;
332 	clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
333 	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
334 		if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
335 			continue;
336 		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
337 	}
338 }
339 
340 static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
341 {
342 	clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
343 	clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
344 	smp_mb__after_atomic();
345 	wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
346 	rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
347 }
348 
349 static void
350 pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
351 		struct list_head *free_me)
352 {
353 	clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
354 	clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
355 	if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
356 		pnfs_lseg_dec_and_remove_zero(lseg, free_me);
357 	if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
358 		pnfs_lseg_dec_and_remove_zero(lseg, free_me);
359 }
360 
361 /*
362  * Update the seqid of a layout stateid after receiving
363  * NFS4ERR_OLD_STATEID
364  */
365 bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
366 		struct pnfs_layout_range *dst_range,
367 		struct inode *inode)
368 {
369 	struct pnfs_layout_hdr *lo;
370 	struct pnfs_layout_range range = {
371 		.iomode = IOMODE_ANY,
372 		.offset = 0,
373 		.length = NFS4_MAX_UINT64,
374 	};
375 	bool ret = false;
376 	LIST_HEAD(head);
377 	int err;
378 
379 	spin_lock(&inode->i_lock);
380 	lo = NFS_I(inode)->layout;
381 	if (lo &&  pnfs_layout_is_valid(lo) &&
382 	    nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
383 		/* Is our call using the most recent seqid? If so, bump it */
384 		if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
385 			nfs4_stateid_seqid_inc(dst);
386 			ret = true;
387 			goto out;
388 		}
389 		/* Try to update the seqid to the most recent */
390 		err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
391 		if (err != -EBUSY) {
392 			dst->seqid = lo->plh_stateid.seqid;
393 			*dst_range = range;
394 			ret = true;
395 		}
396 	}
397 out:
398 	spin_unlock(&inode->i_lock);
399 	pnfs_free_lseg_list(&head);
400 	return ret;
401 }
402 
403 /*
404  * Mark a pnfs_layout_hdr and all associated layout segments as invalid
405  *
406  * In order to continue using the pnfs_layout_hdr, a full recovery
407  * is required.
408  * Note that caller must hold inode->i_lock.
409  */
410 int
411 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
412 		struct list_head *lseg_list)
413 {
414 	struct pnfs_layout_range range = {
415 		.iomode = IOMODE_ANY,
416 		.offset = 0,
417 		.length = NFS4_MAX_UINT64,
418 	};
419 	struct pnfs_layout_segment *lseg, *next;
420 
421 	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
422 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
423 		pnfs_clear_lseg_state(lseg, lseg_list);
424 	pnfs_clear_layoutreturn_info(lo);
425 	pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
426 	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
427 	    !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
428 		pnfs_clear_layoutreturn_waitbit(lo);
429 	return !list_empty(&lo->plh_segs);
430 }
431 
432 static int
433 pnfs_iomode_to_fail_bit(u32 iomode)
434 {
435 	return iomode == IOMODE_RW ?
436 		NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
437 }
438 
439 static void
440 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
441 {
442 	lo->plh_retry_timestamp = jiffies;
443 	if (!test_and_set_bit(fail_bit, &lo->plh_flags))
444 		refcount_inc(&lo->plh_refcount);
445 }
446 
447 static void
448 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
449 {
450 	if (test_and_clear_bit(fail_bit, &lo->plh_flags))
451 		refcount_dec(&lo->plh_refcount);
452 }
453 
454 static void
455 pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
456 {
457 	struct inode *inode = lo->plh_inode;
458 	struct pnfs_layout_range range = {
459 		.iomode = iomode,
460 		.offset = 0,
461 		.length = NFS4_MAX_UINT64,
462 	};
463 	LIST_HEAD(head);
464 
465 	spin_lock(&inode->i_lock);
466 	pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
467 	pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
468 	spin_unlock(&inode->i_lock);
469 	pnfs_free_lseg_list(&head);
470 	dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
471 			iomode == IOMODE_RW ?  "RW" : "READ");
472 }
473 
474 static bool
475 pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
476 {
477 	unsigned long start, end;
478 	int fail_bit = pnfs_iomode_to_fail_bit(iomode);
479 
480 	if (test_bit(fail_bit, &lo->plh_flags) == 0)
481 		return false;
482 	end = jiffies;
483 	start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
484 	if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
485 		/* It is time to retry the failed layoutgets */
486 		pnfs_layout_clear_fail_bit(lo, fail_bit);
487 		return false;
488 	}
489 	return true;
490 }
491 
492 static void
493 pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
494 		const struct pnfs_layout_range *range,
495 		const nfs4_stateid *stateid)
496 {
497 	INIT_LIST_HEAD(&lseg->pls_list);
498 	INIT_LIST_HEAD(&lseg->pls_lc_list);
499 	refcount_set(&lseg->pls_refcount, 1);
500 	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
501 	lseg->pls_layout = lo;
502 	lseg->pls_range = *range;
503 	lseg->pls_seq = be32_to_cpu(stateid->seqid);
504 }
505 
506 static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
507 {
508 	if (lseg != NULL) {
509 		struct inode *inode = lseg->pls_layout->plh_inode;
510 		NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
511 	}
512 }
513 
514 static void
515 pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
516 		struct pnfs_layout_segment *lseg)
517 {
518 	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
519 	list_del_init(&lseg->pls_list);
520 	/* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
521 	refcount_dec(&lo->plh_refcount);
522 	if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
523 		return;
524 	if (list_empty(&lo->plh_segs) &&
525 	    !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
526 	    !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
527 		if (atomic_read(&lo->plh_outstanding) == 0)
528 			set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
529 		clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
530 	}
531 }
532 
533 static bool
534 pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
535 		struct pnfs_layout_segment *lseg)
536 {
537 	if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
538 	    pnfs_layout_is_valid(lo)) {
539 		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
540 		list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
541 		return true;
542 	}
543 	return false;
544 }
545 
546 void
547 pnfs_put_lseg(struct pnfs_layout_segment *lseg)
548 {
549 	struct pnfs_layout_hdr *lo;
550 	struct inode *inode;
551 
552 	if (!lseg)
553 		return;
554 
555 	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
556 		refcount_read(&lseg->pls_refcount),
557 		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
558 
559 	lo = lseg->pls_layout;
560 	inode = lo->plh_inode;
561 
562 	if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
563 		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
564 			spin_unlock(&inode->i_lock);
565 			return;
566 		}
567 		pnfs_get_layout_hdr(lo);
568 		pnfs_layout_remove_lseg(lo, lseg);
569 		if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
570 			lseg = NULL;
571 		spin_unlock(&inode->i_lock);
572 		pnfs_free_lseg(lseg);
573 		pnfs_put_layout_hdr(lo);
574 	}
575 }
576 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
577 
578 /*
579  * is l2 fully contained in l1?
580  *   start1                             end1
581  *   [----------------------------------)
582  *           start2           end2
583  *           [----------------)
584  */
585 static bool
586 pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
587 		 const struct pnfs_layout_range *l2)
588 {
589 	u64 start1 = l1->offset;
590 	u64 end1 = pnfs_end_offset(start1, l1->length);
591 	u64 start2 = l2->offset;
592 	u64 end2 = pnfs_end_offset(start2, l2->length);
593 
594 	return (start1 <= start2) && (end1 >= end2);
595 }
596 
597 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
598 		struct list_head *tmp_list)
599 {
600 	if (!refcount_dec_and_test(&lseg->pls_refcount))
601 		return false;
602 	pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
603 	list_add(&lseg->pls_list, tmp_list);
604 	return true;
605 }
606 
607 /* Returns 1 if lseg is removed from list, 0 otherwise */
608 static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
609 			     struct list_head *tmp_list)
610 {
611 	int rv = 0;
612 
613 	if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
614 		/* Remove the reference keeping the lseg in the
615 		 * list.  It will now be removed when all
616 		 * outstanding io is finished.
617 		 */
618 		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
619 			refcount_read(&lseg->pls_refcount));
620 		if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
621 			rv = 1;
622 	}
623 	return rv;
624 }
625 
626 /*
627  * Compare 2 layout stateid sequence ids, to see which is newer,
628  * taking into account wraparound issues.
629  */
630 static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
631 {
632 	return (s32)(s1 - s2) > 0;
633 }
634 
635 static bool
636 pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
637 		 const struct pnfs_layout_range *recall_range)
638 {
639 	return (recall_range->iomode == IOMODE_ANY ||
640 		lseg_range->iomode == recall_range->iomode) &&
641 	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
642 }
643 
644 static bool
645 pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
646 		const struct pnfs_layout_range *recall_range,
647 		u32 seq)
648 {
649 	if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
650 		return false;
651 	if (recall_range == NULL)
652 		return true;
653 	return pnfs_should_free_range(&lseg->pls_range, recall_range);
654 }
655 
656 /**
657  * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
658  * @lo: layout header containing the lsegs
659  * @tmp_list: list head where doomed lsegs should go
660  * @recall_range: optional recall range argument to match (may be NULL)
661  * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
662  *
663  * Walk the list of lsegs in the layout header, and tear down any that should
664  * be destroyed. If "recall_range" is specified then the segment must match
665  * that range. If "seq" is non-zero, then only match segments that were handed
666  * out at or before that sequence.
667  *
668  * Returns number of matching invalid lsegs remaining in list after scanning
669  * it and purging them.
670  */
671 int
672 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
673 			    struct list_head *tmp_list,
674 			    const struct pnfs_layout_range *recall_range,
675 			    u32 seq)
676 {
677 	struct pnfs_layout_segment *lseg, *next;
678 	int remaining = 0;
679 
680 	dprintk("%s:Begin lo %p\n", __func__, lo);
681 
682 	if (list_empty(&lo->plh_segs))
683 		return 0;
684 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
685 		if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
686 			dprintk("%s: freeing lseg %p iomode %d seq %u "
687 				"offset %llu length %llu\n", __func__,
688 				lseg, lseg->pls_range.iomode, lseg->pls_seq,
689 				lseg->pls_range.offset, lseg->pls_range.length);
690 			if (!mark_lseg_invalid(lseg, tmp_list))
691 				remaining++;
692 		}
693 	dprintk("%s:Return %i\n", __func__, remaining);
694 	return remaining;
695 }
696 
697 static void
698 pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
699 		struct list_head *free_me,
700 		const struct pnfs_layout_range *range,
701 		u32 seq)
702 {
703 	struct pnfs_layout_segment *lseg, *next;
704 
705 	list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
706 		if (pnfs_match_lseg_recall(lseg, range, seq))
707 			list_move_tail(&lseg->pls_list, free_me);
708 	}
709 }
710 
711 /* note free_me must contain lsegs from a single layout_hdr */
712 void
713 pnfs_free_lseg_list(struct list_head *free_me)
714 {
715 	struct pnfs_layout_segment *lseg, *tmp;
716 
717 	if (list_empty(free_me))
718 		return;
719 
720 	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
721 		list_del(&lseg->pls_list);
722 		pnfs_free_lseg(lseg);
723 	}
724 }
725 
726 void
727 pnfs_destroy_layout(struct nfs_inode *nfsi)
728 {
729 	struct pnfs_layout_hdr *lo;
730 	LIST_HEAD(tmp_list);
731 
732 	spin_lock(&nfsi->vfs_inode.i_lock);
733 	lo = nfsi->layout;
734 	if (lo) {
735 		pnfs_get_layout_hdr(lo);
736 		pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
737 		pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
738 		pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
739 		spin_unlock(&nfsi->vfs_inode.i_lock);
740 		pnfs_free_lseg_list(&tmp_list);
741 		nfs_commit_inode(&nfsi->vfs_inode, 0);
742 		pnfs_put_layout_hdr(lo);
743 	} else
744 		spin_unlock(&nfsi->vfs_inode.i_lock);
745 }
746 EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
747 
748 static bool
749 pnfs_layout_add_bulk_destroy_list(struct inode *inode,
750 		struct list_head *layout_list)
751 {
752 	struct pnfs_layout_hdr *lo;
753 	bool ret = false;
754 
755 	spin_lock(&inode->i_lock);
756 	lo = NFS_I(inode)->layout;
757 	if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
758 		pnfs_get_layout_hdr(lo);
759 		list_add(&lo->plh_bulk_destroy, layout_list);
760 		ret = true;
761 	}
762 	spin_unlock(&inode->i_lock);
763 	return ret;
764 }
765 
766 /* Caller must hold rcu_read_lock and clp->cl_lock */
767 static int
768 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
769 		struct nfs_server *server,
770 		struct list_head *layout_list)
771 	__must_hold(&clp->cl_lock)
772 	__must_hold(RCU)
773 {
774 	struct pnfs_layout_hdr *lo, *next;
775 	struct inode *inode;
776 
777 	list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
778 		if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
779 		    test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
780 		    !list_empty(&lo->plh_bulk_destroy))
781 			continue;
782 		/* If the sb is being destroyed, just bail */
783 		if (!nfs_sb_active(server->super))
784 			break;
785 		inode = igrab(lo->plh_inode);
786 		if (inode != NULL) {
787 			list_del_init(&lo->plh_layouts);
788 			if (pnfs_layout_add_bulk_destroy_list(inode,
789 						layout_list))
790 				continue;
791 			rcu_read_unlock();
792 			spin_unlock(&clp->cl_lock);
793 			iput(inode);
794 		} else {
795 			rcu_read_unlock();
796 			spin_unlock(&clp->cl_lock);
797 			set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
798 		}
799 		nfs_sb_deactive(server->super);
800 		spin_lock(&clp->cl_lock);
801 		rcu_read_lock();
802 		return -EAGAIN;
803 	}
804 	return 0;
805 }
806 
807 static int
808 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
809 		bool is_bulk_recall)
810 {
811 	struct pnfs_layout_hdr *lo;
812 	struct inode *inode;
813 	LIST_HEAD(lseg_list);
814 	int ret = 0;
815 
816 	while (!list_empty(layout_list)) {
817 		lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
818 				plh_bulk_destroy);
819 		dprintk("%s freeing layout for inode %lu\n", __func__,
820 			lo->plh_inode->i_ino);
821 		inode = lo->plh_inode;
822 
823 		pnfs_layoutcommit_inode(inode, false);
824 
825 		spin_lock(&inode->i_lock);
826 		list_del_init(&lo->plh_bulk_destroy);
827 		if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
828 			if (is_bulk_recall)
829 				set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
830 			ret = -EAGAIN;
831 		}
832 		spin_unlock(&inode->i_lock);
833 		pnfs_free_lseg_list(&lseg_list);
834 		/* Free all lsegs that are attached to commit buckets */
835 		nfs_commit_inode(inode, 0);
836 		pnfs_put_layout_hdr(lo);
837 		nfs_iput_and_deactive(inode);
838 	}
839 	return ret;
840 }
841 
842 int
843 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
844 		struct nfs_fsid *fsid,
845 		bool is_recall)
846 {
847 	struct nfs_server *server;
848 	LIST_HEAD(layout_list);
849 
850 	spin_lock(&clp->cl_lock);
851 	rcu_read_lock();
852 restart:
853 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
854 		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
855 			continue;
856 		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
857 				server,
858 				&layout_list) != 0)
859 			goto restart;
860 	}
861 	rcu_read_unlock();
862 	spin_unlock(&clp->cl_lock);
863 
864 	if (list_empty(&layout_list))
865 		return 0;
866 	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
867 }
868 
869 int
870 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
871 		bool is_recall)
872 {
873 	struct nfs_server *server;
874 	LIST_HEAD(layout_list);
875 
876 	spin_lock(&clp->cl_lock);
877 	rcu_read_lock();
878 restart:
879 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
880 		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
881 					server,
882 					&layout_list) != 0)
883 			goto restart;
884 	}
885 	rcu_read_unlock();
886 	spin_unlock(&clp->cl_lock);
887 
888 	if (list_empty(&layout_list))
889 		return 0;
890 	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
891 }
892 
893 /*
894  * Called by the state manger to remove all layouts established under an
895  * expired lease.
896  */
897 void
898 pnfs_destroy_all_layouts(struct nfs_client *clp)
899 {
900 	nfs4_deviceid_mark_client_invalid(clp);
901 	nfs4_deviceid_purge_client(clp);
902 
903 	pnfs_destroy_layouts_byclid(clp, false);
904 }
905 
906 /* update lo->plh_stateid with new if is more recent */
907 void
908 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
909 			bool update_barrier)
910 {
911 	u32 oldseq, newseq, new_barrier = 0;
912 
913 	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
914 	newseq = be32_to_cpu(new->seqid);
915 
916 	if (!pnfs_layout_is_valid(lo)) {
917 		nfs4_stateid_copy(&lo->plh_stateid, new);
918 		lo->plh_barrier = newseq;
919 		pnfs_clear_layoutreturn_info(lo);
920 		clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
921 		return;
922 	}
923 	if (pnfs_seqid_is_newer(newseq, oldseq)) {
924 		nfs4_stateid_copy(&lo->plh_stateid, new);
925 		/*
926 		 * Because of wraparound, we want to keep the barrier
927 		 * "close" to the current seqids.
928 		 */
929 		new_barrier = newseq - atomic_read(&lo->plh_outstanding);
930 	}
931 	if (update_barrier)
932 		new_barrier = be32_to_cpu(new->seqid);
933 	else if (new_barrier == 0)
934 		return;
935 	if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
936 		lo->plh_barrier = new_barrier;
937 }
938 
939 static bool
940 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
941 		const nfs4_stateid *stateid)
942 {
943 	u32 seqid = be32_to_cpu(stateid->seqid);
944 
945 	return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
946 }
947 
948 /* lget is set to 1 if called from inside send_layoutget call chain */
949 static bool
950 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
951 {
952 	return lo->plh_block_lgets ||
953 		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
954 }
955 
956 static struct nfs_server *
957 pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
958 {
959 	struct nfs_server *server;
960 
961 	if (inode) {
962 		server = NFS_SERVER(inode);
963 	} else {
964 		struct dentry *parent_dir = dget_parent(ctx->dentry);
965 		server = NFS_SERVER(parent_dir->d_inode);
966 		dput(parent_dir);
967 	}
968 	return server;
969 }
970 
971 static void nfs4_free_pages(struct page **pages, size_t size)
972 {
973 	int i;
974 
975 	if (!pages)
976 		return;
977 
978 	for (i = 0; i < size; i++) {
979 		if (!pages[i])
980 			break;
981 		__free_page(pages[i]);
982 	}
983 	kfree(pages);
984 }
985 
986 static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
987 {
988 	struct page **pages;
989 	int i;
990 
991 	pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
992 	if (!pages) {
993 		dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
994 		return NULL;
995 	}
996 
997 	for (i = 0; i < size; i++) {
998 		pages[i] = alloc_page(gfp_flags);
999 		if (!pages[i]) {
1000 			dprintk("%s: failed to allocate page\n", __func__);
1001 			nfs4_free_pages(pages, i);
1002 			return NULL;
1003 		}
1004 	}
1005 
1006 	return pages;
1007 }
1008 
1009 static struct nfs4_layoutget *
1010 pnfs_alloc_init_layoutget_args(struct inode *ino,
1011 	   struct nfs_open_context *ctx,
1012 	   const nfs4_stateid *stateid,
1013 	   const struct pnfs_layout_range *range,
1014 	   gfp_t gfp_flags)
1015 {
1016 	struct nfs_server *server = pnfs_find_server(ino, ctx);
1017 	size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
1018 	size_t max_pages = max_response_pages(server);
1019 	struct nfs4_layoutget *lgp;
1020 
1021 	dprintk("--> %s\n", __func__);
1022 
1023 	lgp = kzalloc(sizeof(*lgp), gfp_flags);
1024 	if (lgp == NULL)
1025 		return NULL;
1026 
1027 	if (max_reply_sz) {
1028 		size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1029 		if (npages < max_pages)
1030 			max_pages = npages;
1031 	}
1032 
1033 	lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1034 	if (!lgp->args.layout.pages) {
1035 		kfree(lgp);
1036 		return NULL;
1037 	}
1038 	lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1039 	lgp->res.layoutp = &lgp->args.layout;
1040 
1041 	/* Don't confuse uninitialised result and success */
1042 	lgp->res.status = -NFS4ERR_DELAY;
1043 
1044 	lgp->args.minlength = PAGE_SIZE;
1045 	if (lgp->args.minlength > range->length)
1046 		lgp->args.minlength = range->length;
1047 	if (ino) {
1048 		loff_t i_size = i_size_read(ino);
1049 
1050 		if (range->iomode == IOMODE_READ) {
1051 			if (range->offset >= i_size)
1052 				lgp->args.minlength = 0;
1053 			else if (i_size - range->offset < lgp->args.minlength)
1054 				lgp->args.minlength = i_size - range->offset;
1055 		}
1056 	}
1057 	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1058 	pnfs_copy_range(&lgp->args.range, range);
1059 	lgp->args.type = server->pnfs_curr_ld->id;
1060 	lgp->args.inode = ino;
1061 	lgp->args.ctx = get_nfs_open_context(ctx);
1062 	nfs4_stateid_copy(&lgp->args.stateid, stateid);
1063 	lgp->gfp_flags = gfp_flags;
1064 	lgp->cred = get_cred(ctx->cred);
1065 	return lgp;
1066 }
1067 
1068 void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1069 {
1070 	size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1071 
1072 	nfs4_free_pages(lgp->args.layout.pages, max_pages);
1073 	if (lgp->args.inode)
1074 		pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
1075 	put_cred(lgp->cred);
1076 	put_nfs_open_context(lgp->args.ctx);
1077 	kfree(lgp);
1078 }
1079 
1080 static void pnfs_clear_layoutcommit(struct inode *inode,
1081 		struct list_head *head)
1082 {
1083 	struct nfs_inode *nfsi = NFS_I(inode);
1084 	struct pnfs_layout_segment *lseg, *tmp;
1085 
1086 	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1087 		return;
1088 	list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1089 		if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1090 			continue;
1091 		pnfs_lseg_dec_and_remove_zero(lseg, head);
1092 	}
1093 }
1094 
1095 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
1096 		const nfs4_stateid *arg_stateid,
1097 		const struct pnfs_layout_range *range,
1098 		const nfs4_stateid *stateid)
1099 {
1100 	struct inode *inode = lo->plh_inode;
1101 	LIST_HEAD(freeme);
1102 
1103 	spin_lock(&inode->i_lock);
1104 	if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1105 	    !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1106 		goto out_unlock;
1107 	if (stateid) {
1108 		u32 seq = be32_to_cpu(arg_stateid->seqid);
1109 
1110 		pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1111 		pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1112 		pnfs_set_layout_stateid(lo, stateid, true);
1113 	} else
1114 		pnfs_mark_layout_stateid_invalid(lo, &freeme);
1115 out_unlock:
1116 	pnfs_clear_layoutreturn_waitbit(lo);
1117 	spin_unlock(&inode->i_lock);
1118 	pnfs_free_lseg_list(&freeme);
1119 
1120 }
1121 
1122 static bool
1123 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1124 		nfs4_stateid *stateid,
1125 		enum pnfs_iomode *iomode)
1126 {
1127 	/* Serialise LAYOUTGET/LAYOUTRETURN */
1128 	if (atomic_read(&lo->plh_outstanding) != 0)
1129 		return false;
1130 	if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
1131 		return false;
1132 	set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
1133 	pnfs_get_layout_hdr(lo);
1134 	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1135 		if (stateid != NULL) {
1136 			nfs4_stateid_copy(stateid, &lo->plh_stateid);
1137 			if (lo->plh_return_seq != 0)
1138 				stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1139 		}
1140 		if (iomode != NULL)
1141 			*iomode = lo->plh_return_iomode;
1142 		pnfs_clear_layoutreturn_info(lo);
1143 		return true;
1144 	}
1145 	if (stateid != NULL)
1146 		nfs4_stateid_copy(stateid, &lo->plh_stateid);
1147 	if (iomode != NULL)
1148 		*iomode = IOMODE_ANY;
1149 	return true;
1150 }
1151 
1152 static void
1153 pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1154 		struct pnfs_layout_hdr *lo,
1155 		const nfs4_stateid *stateid,
1156 		enum pnfs_iomode iomode)
1157 {
1158 	struct inode *inode = lo->plh_inode;
1159 
1160 	args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1161 	args->inode = inode;
1162 	args->range.iomode = iomode;
1163 	args->range.offset = 0;
1164 	args->range.length = NFS4_MAX_UINT64;
1165 	args->layout = lo;
1166 	nfs4_stateid_copy(&args->stateid, stateid);
1167 }
1168 
1169 static int
1170 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
1171 		       enum pnfs_iomode iomode, bool sync)
1172 {
1173 	struct inode *ino = lo->plh_inode;
1174 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1175 	struct nfs4_layoutreturn *lrp;
1176 	int status = 0;
1177 
1178 	lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
1179 	if (unlikely(lrp == NULL)) {
1180 		status = -ENOMEM;
1181 		spin_lock(&ino->i_lock);
1182 		pnfs_clear_layoutreturn_waitbit(lo);
1183 		spin_unlock(&ino->i_lock);
1184 		pnfs_put_layout_hdr(lo);
1185 		goto out;
1186 	}
1187 
1188 	pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
1189 	lrp->args.ld_private = &lrp->ld_private;
1190 	lrp->clp = NFS_SERVER(ino)->nfs_client;
1191 	lrp->cred = lo->plh_lc_cred;
1192 	if (ld->prepare_layoutreturn)
1193 		ld->prepare_layoutreturn(&lrp->args);
1194 
1195 	status = nfs4_proc_layoutreturn(lrp, sync);
1196 out:
1197 	dprintk("<-- %s status: %d\n", __func__, status);
1198 	return status;
1199 }
1200 
1201 /* Return true if layoutreturn is needed */
1202 static bool
1203 pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1204 {
1205 	struct pnfs_layout_segment *s;
1206 	enum pnfs_iomode iomode;
1207 	u32 seq;
1208 
1209 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1210 		return false;
1211 
1212 	seq = lo->plh_return_seq;
1213 	iomode = lo->plh_return_iomode;
1214 
1215 	/* Defer layoutreturn until all recalled lsegs are done */
1216 	list_for_each_entry(s, &lo->plh_segs, pls_list) {
1217 		if (seq && pnfs_seqid_is_newer(s->pls_seq, seq))
1218 			continue;
1219 		if (iomode != IOMODE_ANY && s->pls_range.iomode != iomode)
1220 			continue;
1221 		if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1222 			return false;
1223 	}
1224 
1225 	return true;
1226 }
1227 
1228 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1229 {
1230 	struct inode *inode= lo->plh_inode;
1231 
1232 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1233 		return;
1234 	spin_lock(&inode->i_lock);
1235 	if (pnfs_layout_need_return(lo)) {
1236 		nfs4_stateid stateid;
1237 		enum pnfs_iomode iomode;
1238 		bool send;
1239 
1240 		send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1241 		spin_unlock(&inode->i_lock);
1242 		if (send) {
1243 			/* Send an async layoutreturn so we dont deadlock */
1244 			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1245 		}
1246 	} else
1247 		spin_unlock(&inode->i_lock);
1248 }
1249 
1250 /*
1251  * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1252  * when the layout segment list is empty.
1253  *
1254  * Note that a pnfs_layout_hdr can exist with an empty layout segment
1255  * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1256  * deviceid is marked invalid.
1257  */
1258 int
1259 _pnfs_return_layout(struct inode *ino)
1260 {
1261 	struct pnfs_layout_hdr *lo = NULL;
1262 	struct nfs_inode *nfsi = NFS_I(ino);
1263 	LIST_HEAD(tmp_list);
1264 	nfs4_stateid stateid;
1265 	int status = 0;
1266 	bool send, valid_layout;
1267 
1268 	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
1269 
1270 	spin_lock(&ino->i_lock);
1271 	lo = nfsi->layout;
1272 	if (!lo) {
1273 		spin_unlock(&ino->i_lock);
1274 		dprintk("NFS: %s no layout to return\n", __func__);
1275 		goto out;
1276 	}
1277 	/* Reference matched in nfs4_layoutreturn_release */
1278 	pnfs_get_layout_hdr(lo);
1279 	/* Is there an outstanding layoutreturn ? */
1280 	if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1281 		spin_unlock(&ino->i_lock);
1282 		if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1283 					TASK_UNINTERRUPTIBLE))
1284 			goto out_put_layout_hdr;
1285 		spin_lock(&ino->i_lock);
1286 	}
1287 	valid_layout = pnfs_layout_is_valid(lo);
1288 	pnfs_clear_layoutcommit(ino, &tmp_list);
1289 	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
1290 
1291 	if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1292 		struct pnfs_layout_range range = {
1293 			.iomode		= IOMODE_ANY,
1294 			.offset		= 0,
1295 			.length		= NFS4_MAX_UINT64,
1296 		};
1297 		NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1298 	}
1299 
1300 	/* Don't send a LAYOUTRETURN if list was initially empty */
1301 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1302 			!valid_layout) {
1303 		spin_unlock(&ino->i_lock);
1304 		dprintk("NFS: %s no layout segments to return\n", __func__);
1305 		goto out_put_layout_hdr;
1306 	}
1307 
1308 	send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
1309 	spin_unlock(&ino->i_lock);
1310 	if (send)
1311 		status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1312 out_put_layout_hdr:
1313 	pnfs_free_lseg_list(&tmp_list);
1314 	pnfs_put_layout_hdr(lo);
1315 out:
1316 	dprintk("<-- %s status: %d\n", __func__, status);
1317 	return status;
1318 }
1319 
1320 int
1321 pnfs_commit_and_return_layout(struct inode *inode)
1322 {
1323 	struct pnfs_layout_hdr *lo;
1324 	int ret;
1325 
1326 	spin_lock(&inode->i_lock);
1327 	lo = NFS_I(inode)->layout;
1328 	if (lo == NULL) {
1329 		spin_unlock(&inode->i_lock);
1330 		return 0;
1331 	}
1332 	pnfs_get_layout_hdr(lo);
1333 	/* Block new layoutgets and read/write to ds */
1334 	lo->plh_block_lgets++;
1335 	spin_unlock(&inode->i_lock);
1336 	filemap_fdatawait(inode->i_mapping);
1337 	ret = pnfs_layoutcommit_inode(inode, true);
1338 	if (ret == 0)
1339 		ret = _pnfs_return_layout(inode);
1340 	spin_lock(&inode->i_lock);
1341 	lo->plh_block_lgets--;
1342 	spin_unlock(&inode->i_lock);
1343 	pnfs_put_layout_hdr(lo);
1344 	return ret;
1345 }
1346 
1347 bool pnfs_roc(struct inode *ino,
1348 		struct nfs4_layoutreturn_args *args,
1349 		struct nfs4_layoutreturn_res *res,
1350 		const struct cred *cred)
1351 {
1352 	struct nfs_inode *nfsi = NFS_I(ino);
1353 	struct nfs_open_context *ctx;
1354 	struct nfs4_state *state;
1355 	struct pnfs_layout_hdr *lo;
1356 	struct pnfs_layout_segment *lseg, *next;
1357 	nfs4_stateid stateid;
1358 	enum pnfs_iomode iomode = 0;
1359 	bool layoutreturn = false, roc = false;
1360 	bool skip_read = false;
1361 
1362 	if (!nfs_have_layout(ino))
1363 		return false;
1364 retry:
1365 	rcu_read_lock();
1366 	spin_lock(&ino->i_lock);
1367 	lo = nfsi->layout;
1368 	if (!lo || !pnfs_layout_is_valid(lo) ||
1369 	    test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1370 		lo = NULL;
1371 		goto out_noroc;
1372 	}
1373 	pnfs_get_layout_hdr(lo);
1374 	if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1375 		spin_unlock(&ino->i_lock);
1376 		rcu_read_unlock();
1377 		wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1378 				TASK_UNINTERRUPTIBLE);
1379 		pnfs_put_layout_hdr(lo);
1380 		goto retry;
1381 	}
1382 
1383 	/* no roc if we hold a delegation */
1384 	if (nfs4_check_delegation(ino, FMODE_READ)) {
1385 		if (nfs4_check_delegation(ino, FMODE_WRITE))
1386 			goto out_noroc;
1387 		skip_read = true;
1388 	}
1389 
1390 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1391 		state = ctx->state;
1392 		if (state == NULL)
1393 			continue;
1394 		/* Don't return layout if there is open file state */
1395 		if (state->state & FMODE_WRITE)
1396 			goto out_noroc;
1397 		if (state->state & FMODE_READ)
1398 			skip_read = true;
1399 	}
1400 
1401 
1402 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
1403 		if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1404 			continue;
1405 		/* If we are sending layoutreturn, invalidate all valid lsegs */
1406 		if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1407 			continue;
1408 		/*
1409 		 * Note: mark lseg for return so pnfs_layout_remove_lseg
1410 		 * doesn't invalidate the layout for us.
1411 		 */
1412 		set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1413 		if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1414 			continue;
1415 		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
1416 	}
1417 
1418 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1419 		goto out_noroc;
1420 
1421 	/* ROC in two conditions:
1422 	 * 1. there are ROC lsegs
1423 	 * 2. we don't send layoutreturn
1424 	 */
1425 	/* lo ref dropped in pnfs_roc_release() */
1426 	layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1427 	/* If the creds don't match, we can't compound the layoutreturn */
1428 	if (!layoutreturn || cred != lo->plh_lc_cred)
1429 		goto out_noroc;
1430 
1431 	roc = layoutreturn;
1432 	pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1433 	res->lrs_present = 0;
1434 	layoutreturn = false;
1435 
1436 out_noroc:
1437 	spin_unlock(&ino->i_lock);
1438 	rcu_read_unlock();
1439 	pnfs_layoutcommit_inode(ino, true);
1440 	if (roc) {
1441 		struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1442 		if (ld->prepare_layoutreturn)
1443 			ld->prepare_layoutreturn(args);
1444 		pnfs_put_layout_hdr(lo);
1445 		return true;
1446 	}
1447 	if (layoutreturn)
1448 		pnfs_send_layoutreturn(lo, &stateid, iomode, true);
1449 	pnfs_put_layout_hdr(lo);
1450 	return false;
1451 }
1452 
1453 int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
1454 		struct nfs4_layoutreturn_args **argpp,
1455 		struct nfs4_layoutreturn_res **respp,
1456 		int *ret)
1457 {
1458 	struct nfs4_layoutreturn_args *arg = *argpp;
1459 	int retval = -EAGAIN;
1460 
1461 	if (!arg)
1462 		return 0;
1463 	/* Handle Layoutreturn errors */
1464 	switch (*ret) {
1465 	case 0:
1466 		retval = 0;
1467 		break;
1468 	case -NFS4ERR_NOMATCHING_LAYOUT:
1469 		/* Was there an RPC level error? If not, retry */
1470 		if (task->tk_rpc_status == 0)
1471 			break;
1472 		/* If the call was not sent, let caller handle it */
1473 		if (!RPC_WAS_SENT(task))
1474 			return 0;
1475 		/*
1476 		 * Otherwise, assume the call succeeded and
1477 		 * that we need to release the layout
1478 		 */
1479 		*ret = 0;
1480 		(*respp)->lrs_present = 0;
1481 		retval = 0;
1482 		break;
1483 	case -NFS4ERR_DELAY:
1484 		/* Let the caller handle the retry */
1485 		*ret = -NFS4ERR_NOMATCHING_LAYOUT;
1486 		return 0;
1487 	case -NFS4ERR_OLD_STATEID:
1488 		if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
1489 					&arg->range, inode))
1490 			break;
1491 		*ret = -NFS4ERR_NOMATCHING_LAYOUT;
1492 		return -EAGAIN;
1493 	}
1494 	*argpp = NULL;
1495 	*respp = NULL;
1496 	return retval;
1497 }
1498 
1499 void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1500 		struct nfs4_layoutreturn_res *res,
1501 		int ret)
1502 {
1503 	struct pnfs_layout_hdr *lo = args->layout;
1504 	const nfs4_stateid *arg_stateid = NULL;
1505 	const nfs4_stateid *res_stateid = NULL;
1506 	struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
1507 
1508 	switch (ret) {
1509 	case -NFS4ERR_NOMATCHING_LAYOUT:
1510 		break;
1511 	case 0:
1512 		if (res->lrs_present)
1513 			res_stateid = &res->stateid;
1514 		/* Fallthrough */
1515 	default:
1516 		arg_stateid = &args->stateid;
1517 	}
1518 	pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1519 			res_stateid);
1520 	if (ld_private && ld_private->ops && ld_private->ops->free)
1521 		ld_private->ops->free(ld_private);
1522 	pnfs_put_layout_hdr(lo);
1523 	trace_nfs4_layoutreturn_on_close(args->inode, 0);
1524 }
1525 
1526 bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1527 {
1528 	struct nfs_inode *nfsi = NFS_I(ino);
1529         struct pnfs_layout_hdr *lo;
1530         bool sleep = false;
1531 
1532 	/* we might not have grabbed lo reference. so need to check under
1533 	 * i_lock */
1534         spin_lock(&ino->i_lock);
1535         lo = nfsi->layout;
1536         if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1537                 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1538                 sleep = true;
1539 	}
1540         spin_unlock(&ino->i_lock);
1541         return sleep;
1542 }
1543 
1544 /*
1545  * Compare two layout segments for sorting into layout cache.
1546  * We want to preferentially return RW over RO layouts, so ensure those
1547  * are seen first.
1548  */
1549 static s64
1550 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1551 	   const struct pnfs_layout_range *l2)
1552 {
1553 	s64 d;
1554 
1555 	/* high offset > low offset */
1556 	d = l1->offset - l2->offset;
1557 	if (d)
1558 		return d;
1559 
1560 	/* short length > long length */
1561 	d = l2->length - l1->length;
1562 	if (d)
1563 		return d;
1564 
1565 	/* read > read/write */
1566 	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1567 }
1568 
1569 static bool
1570 pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1571 		const struct pnfs_layout_range *l2)
1572 {
1573 	return pnfs_lseg_range_cmp(l1, l2) > 0;
1574 }
1575 
1576 static bool
1577 pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1578 		struct pnfs_layout_segment *old)
1579 {
1580 	return false;
1581 }
1582 
1583 void
1584 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1585 		   struct pnfs_layout_segment *lseg,
1586 		   bool (*is_after)(const struct pnfs_layout_range *,
1587 			   const struct pnfs_layout_range *),
1588 		   bool (*do_merge)(struct pnfs_layout_segment *,
1589 			   struct pnfs_layout_segment *),
1590 		   struct list_head *free_me)
1591 {
1592 	struct pnfs_layout_segment *lp, *tmp;
1593 
1594 	dprintk("%s:Begin\n", __func__);
1595 
1596 	list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1597 		if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1598 			continue;
1599 		if (do_merge(lseg, lp)) {
1600 			mark_lseg_invalid(lp, free_me);
1601 			continue;
1602 		}
1603 		if (is_after(&lseg->pls_range, &lp->pls_range))
1604 			continue;
1605 		list_add_tail(&lseg->pls_list, &lp->pls_list);
1606 		dprintk("%s: inserted lseg %p "
1607 			"iomode %d offset %llu length %llu before "
1608 			"lp %p iomode %d offset %llu length %llu\n",
1609 			__func__, lseg, lseg->pls_range.iomode,
1610 			lseg->pls_range.offset, lseg->pls_range.length,
1611 			lp, lp->pls_range.iomode, lp->pls_range.offset,
1612 			lp->pls_range.length);
1613 		goto out;
1614 	}
1615 	list_add_tail(&lseg->pls_list, &lo->plh_segs);
1616 	dprintk("%s: inserted lseg %p "
1617 		"iomode %d offset %llu length %llu at tail\n",
1618 		__func__, lseg, lseg->pls_range.iomode,
1619 		lseg->pls_range.offset, lseg->pls_range.length);
1620 out:
1621 	pnfs_get_layout_hdr(lo);
1622 
1623 	dprintk("%s:Return\n", __func__);
1624 }
1625 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1626 
1627 static void
1628 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1629 		   struct pnfs_layout_segment *lseg,
1630 		   struct list_head *free_me)
1631 {
1632 	struct inode *inode = lo->plh_inode;
1633 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1634 
1635 	if (ld->add_lseg != NULL)
1636 		ld->add_lseg(lo, lseg, free_me);
1637 	else
1638 		pnfs_generic_layout_insert_lseg(lo, lseg,
1639 				pnfs_lseg_range_is_after,
1640 				pnfs_lseg_no_merge,
1641 				free_me);
1642 }
1643 
1644 static struct pnfs_layout_hdr *
1645 alloc_init_layout_hdr(struct inode *ino,
1646 		      struct nfs_open_context *ctx,
1647 		      gfp_t gfp_flags)
1648 {
1649 	struct pnfs_layout_hdr *lo;
1650 
1651 	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1652 	if (!lo)
1653 		return NULL;
1654 	refcount_set(&lo->plh_refcount, 1);
1655 	INIT_LIST_HEAD(&lo->plh_layouts);
1656 	INIT_LIST_HEAD(&lo->plh_segs);
1657 	INIT_LIST_HEAD(&lo->plh_return_segs);
1658 	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1659 	lo->plh_inode = ino;
1660 	lo->plh_lc_cred = get_cred(ctx->cred);
1661 	lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
1662 	return lo;
1663 }
1664 
1665 static struct pnfs_layout_hdr *
1666 pnfs_find_alloc_layout(struct inode *ino,
1667 		       struct nfs_open_context *ctx,
1668 		       gfp_t gfp_flags)
1669 	__releases(&ino->i_lock)
1670 	__acquires(&ino->i_lock)
1671 {
1672 	struct nfs_inode *nfsi = NFS_I(ino);
1673 	struct pnfs_layout_hdr *new = NULL;
1674 
1675 	dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1676 
1677 	if (nfsi->layout != NULL)
1678 		goto out_existing;
1679 	spin_unlock(&ino->i_lock);
1680 	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1681 	spin_lock(&ino->i_lock);
1682 
1683 	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1684 		nfsi->layout = new;
1685 		return new;
1686 	} else if (new != NULL)
1687 		pnfs_free_layout_hdr(new);
1688 out_existing:
1689 	pnfs_get_layout_hdr(nfsi->layout);
1690 	return nfsi->layout;
1691 }
1692 
1693 /*
1694  * iomode matching rules:
1695  * iomode	lseg	strict match
1696  *                      iomode
1697  * -----	-----	------ -----
1698  * ANY		READ	N/A    true
1699  * ANY		RW	N/A    true
1700  * RW		READ	N/A    false
1701  * RW		RW	N/A    true
1702  * READ		READ	N/A    true
1703  * READ		RW	true   false
1704  * READ		RW	false  true
1705  */
1706 static bool
1707 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1708 		 const struct pnfs_layout_range *range,
1709 		 bool strict_iomode)
1710 {
1711 	struct pnfs_layout_range range1;
1712 
1713 	if ((range->iomode == IOMODE_RW &&
1714 	     ls_range->iomode != IOMODE_RW) ||
1715 	    (range->iomode != ls_range->iomode &&
1716 	     strict_iomode) ||
1717 	    !pnfs_lseg_range_intersecting(ls_range, range))
1718 		return false;
1719 
1720 	/* range1 covers only the first byte in the range */
1721 	range1 = *range;
1722 	range1.length = 1;
1723 	return pnfs_lseg_range_contained(ls_range, &range1);
1724 }
1725 
1726 /*
1727  * lookup range in layout
1728  */
1729 static struct pnfs_layout_segment *
1730 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1731 		struct pnfs_layout_range *range,
1732 		bool strict_iomode)
1733 {
1734 	struct pnfs_layout_segment *lseg, *ret = NULL;
1735 
1736 	dprintk("%s:Begin\n", __func__);
1737 
1738 	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1739 		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1740 		    pnfs_lseg_range_match(&lseg->pls_range, range,
1741 					  strict_iomode)) {
1742 			ret = pnfs_get_lseg(lseg);
1743 			break;
1744 		}
1745 	}
1746 
1747 	dprintk("%s:Return lseg %p ref %d\n",
1748 		__func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
1749 	return ret;
1750 }
1751 
1752 /*
1753  * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1754  * to the MDS or over pNFS
1755  *
1756  * The nfs_inode read_io and write_io fields are cumulative counters reset
1757  * when there are no layout segments. Note that in pnfs_update_layout iomode
1758  * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1759  * WRITE request.
1760  *
1761  * A return of true means use MDS I/O.
1762  *
1763  * From rfc 5661:
1764  * If a file's size is smaller than the file size threshold, data accesses
1765  * SHOULD be sent to the metadata server.  If an I/O request has a length that
1766  * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1767  * server.  If both file size and I/O size are provided, the client SHOULD
1768  * reach or exceed  both thresholds before sending its read or write
1769  * requests to the data server.
1770  */
1771 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1772 				     struct inode *ino, int iomode)
1773 {
1774 	struct nfs4_threshold *t = ctx->mdsthreshold;
1775 	struct nfs_inode *nfsi = NFS_I(ino);
1776 	loff_t fsize = i_size_read(ino);
1777 	bool size = false, size_set = false, io = false, io_set = false, ret = false;
1778 
1779 	if (t == NULL)
1780 		return ret;
1781 
1782 	dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1783 		__func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1784 
1785 	switch (iomode) {
1786 	case IOMODE_READ:
1787 		if (t->bm & THRESHOLD_RD) {
1788 			dprintk("%s fsize %llu\n", __func__, fsize);
1789 			size_set = true;
1790 			if (fsize < t->rd_sz)
1791 				size = true;
1792 		}
1793 		if (t->bm & THRESHOLD_RD_IO) {
1794 			dprintk("%s nfsi->read_io %llu\n", __func__,
1795 				nfsi->read_io);
1796 			io_set = true;
1797 			if (nfsi->read_io < t->rd_io_sz)
1798 				io = true;
1799 		}
1800 		break;
1801 	case IOMODE_RW:
1802 		if (t->bm & THRESHOLD_WR) {
1803 			dprintk("%s fsize %llu\n", __func__, fsize);
1804 			size_set = true;
1805 			if (fsize < t->wr_sz)
1806 				size = true;
1807 		}
1808 		if (t->bm & THRESHOLD_WR_IO) {
1809 			dprintk("%s nfsi->write_io %llu\n", __func__,
1810 				nfsi->write_io);
1811 			io_set = true;
1812 			if (nfsi->write_io < t->wr_io_sz)
1813 				io = true;
1814 		}
1815 		break;
1816 	}
1817 	if (size_set && io_set) {
1818 		if (size && io)
1819 			ret = true;
1820 	} else if (size || io)
1821 		ret = true;
1822 
1823 	dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1824 	return ret;
1825 }
1826 
1827 static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1828 {
1829 	/*
1830 	 * send layoutcommit as it can hold up layoutreturn due to lseg
1831 	 * reference
1832 	 */
1833 	pnfs_layoutcommit_inode(lo->plh_inode, false);
1834 	return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1835 				   nfs_wait_bit_killable,
1836 				   TASK_KILLABLE);
1837 }
1838 
1839 static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1840 {
1841 	atomic_inc(&lo->plh_outstanding);
1842 }
1843 
1844 static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1845 {
1846 	if (atomic_dec_and_test(&lo->plh_outstanding))
1847 		wake_up_var(&lo->plh_outstanding);
1848 }
1849 
1850 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1851 {
1852 	unsigned long *bitlock = &lo->plh_flags;
1853 
1854 	clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1855 	smp_mb__after_atomic();
1856 	wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1857 }
1858 
1859 static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1860 				struct nfs_server *server)
1861 {
1862 	if (list_empty(&lo->plh_layouts)) {
1863 		struct nfs_client *clp = server->nfs_client;
1864 
1865 		/* The lo must be on the clp list if there is any
1866 		 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1867 		 */
1868 		spin_lock(&clp->cl_lock);
1869 		if (list_empty(&lo->plh_layouts))
1870 			list_add_tail(&lo->plh_layouts, &server->layouts);
1871 		spin_unlock(&clp->cl_lock);
1872 	}
1873 }
1874 
1875 /*
1876  * Layout segment is retreived from the server if not cached.
1877  * The appropriate layout segment is referenced and returned to the caller.
1878  */
1879 struct pnfs_layout_segment *
1880 pnfs_update_layout(struct inode *ino,
1881 		   struct nfs_open_context *ctx,
1882 		   loff_t pos,
1883 		   u64 count,
1884 		   enum pnfs_iomode iomode,
1885 		   bool strict_iomode,
1886 		   gfp_t gfp_flags)
1887 {
1888 	struct pnfs_layout_range arg = {
1889 		.iomode = iomode,
1890 		.offset = pos,
1891 		.length = count,
1892 	};
1893 	unsigned pg_offset;
1894 	struct nfs_server *server = NFS_SERVER(ino);
1895 	struct nfs_client *clp = server->nfs_client;
1896 	struct pnfs_layout_hdr *lo = NULL;
1897 	struct pnfs_layout_segment *lseg = NULL;
1898 	struct nfs4_layoutget *lgp;
1899 	nfs4_stateid stateid;
1900 	long timeout = 0;
1901 	unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
1902 	bool first;
1903 
1904 	if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1905 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1906 				 PNFS_UPDATE_LAYOUT_NO_PNFS);
1907 		goto out;
1908 	}
1909 
1910 	if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1911 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1912 				 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1913 		goto out;
1914 	}
1915 
1916 lookup_again:
1917 	lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1918 	if (IS_ERR(lseg))
1919 		goto out;
1920 	first = false;
1921 	spin_lock(&ino->i_lock);
1922 	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1923 	if (lo == NULL) {
1924 		spin_unlock(&ino->i_lock);
1925 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1926 				 PNFS_UPDATE_LAYOUT_NOMEM);
1927 		goto out;
1928 	}
1929 
1930 	/* Do we even need to bother with this? */
1931 	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1932 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1933 				 PNFS_UPDATE_LAYOUT_BULK_RECALL);
1934 		dprintk("%s matches recall, use MDS\n", __func__);
1935 		goto out_unlock;
1936 	}
1937 
1938 	/* if LAYOUTGET already failed once we don't try again */
1939 	if (pnfs_layout_io_test_failed(lo, iomode)) {
1940 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1941 				 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1942 		goto out_unlock;
1943 	}
1944 
1945 	/*
1946 	 * If the layout segment list is empty, but there are outstanding
1947 	 * layoutget calls, then they might be subject to a layoutrecall.
1948 	 */
1949 	if (list_empty(&lo->plh_segs) &&
1950 	    atomic_read(&lo->plh_outstanding) != 0) {
1951 		spin_unlock(&ino->i_lock);
1952 		lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
1953 					!atomic_read(&lo->plh_outstanding)));
1954 		if (IS_ERR(lseg))
1955 			goto out_put_layout_hdr;
1956 		pnfs_put_layout_hdr(lo);
1957 		goto lookup_again;
1958 	}
1959 
1960 	lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
1961 	if (lseg) {
1962 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1963 				PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1964 		goto out_unlock;
1965 	}
1966 
1967 	/*
1968 	 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1969 	 * stateid, or it has been invalidated, then we must use the open
1970 	 * stateid.
1971 	 */
1972 	if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
1973 		int status;
1974 
1975 		/*
1976 		 * The first layoutget for the file. Need to serialize per
1977 		 * RFC 5661 Errata 3208.
1978 		 */
1979 		if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1980 				     &lo->plh_flags)) {
1981 			spin_unlock(&ino->i_lock);
1982 			lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
1983 						NFS_LAYOUT_FIRST_LAYOUTGET,
1984 						TASK_KILLABLE));
1985 			if (IS_ERR(lseg))
1986 				goto out_put_layout_hdr;
1987 			pnfs_put_layout_hdr(lo);
1988 			dprintk("%s retrying\n", __func__);
1989 			goto lookup_again;
1990 		}
1991 
1992 		first = true;
1993 		status = nfs4_select_rw_stateid(ctx->state,
1994 					iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1995 					NULL, &stateid, NULL);
1996 		if (status != 0) {
1997 			lseg = ERR_PTR(status);
1998 			trace_pnfs_update_layout(ino, pos, count,
1999 					iomode, lo, lseg,
2000 					PNFS_UPDATE_LAYOUT_INVALID_OPEN);
2001 			if (status != -EAGAIN)
2002 				goto out_unlock;
2003 			spin_unlock(&ino->i_lock);
2004 			nfs4_schedule_stateid_recovery(server, ctx->state);
2005 			pnfs_clear_first_layoutget(lo);
2006 			pnfs_put_layout_hdr(lo);
2007 			goto lookup_again;
2008 		}
2009 	} else {
2010 		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
2011 	}
2012 
2013 	/*
2014 	 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
2015 	 * for LAYOUTRETURN even if first is true.
2016 	 */
2017 	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
2018 		spin_unlock(&ino->i_lock);
2019 		dprintk("%s wait for layoutreturn\n", __func__);
2020 		lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2021 		if (!IS_ERR(lseg)) {
2022 			if (first)
2023 				pnfs_clear_first_layoutget(lo);
2024 			pnfs_put_layout_hdr(lo);
2025 			dprintk("%s retrying\n", __func__);
2026 			trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2027 					lseg, PNFS_UPDATE_LAYOUT_RETRY);
2028 			goto lookup_again;
2029 		}
2030 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2031 				PNFS_UPDATE_LAYOUT_RETURN);
2032 		goto out_put_layout_hdr;
2033 	}
2034 
2035 	if (pnfs_layoutgets_blocked(lo)) {
2036 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2037 				PNFS_UPDATE_LAYOUT_BLOCKED);
2038 		goto out_unlock;
2039 	}
2040 	nfs_layoutget_begin(lo);
2041 	spin_unlock(&ino->i_lock);
2042 
2043 	_add_to_server_list(lo, server);
2044 
2045 	pg_offset = arg.offset & ~PAGE_MASK;
2046 	if (pg_offset) {
2047 		arg.offset -= pg_offset;
2048 		arg.length += pg_offset;
2049 	}
2050 	if (arg.length != NFS4_MAX_UINT64)
2051 		arg.length = PAGE_ALIGN(arg.length);
2052 
2053 	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
2054 	if (!lgp) {
2055 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2056 					 PNFS_UPDATE_LAYOUT_NOMEM);
2057 		nfs_layoutget_end(lo);
2058 		goto out_put_layout_hdr;
2059 	}
2060 
2061 	lseg = nfs4_proc_layoutget(lgp, &timeout);
2062 	trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2063 				 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
2064 	nfs_layoutget_end(lo);
2065 	if (IS_ERR(lseg)) {
2066 		switch(PTR_ERR(lseg)) {
2067 		case -EBUSY:
2068 			if (time_after(jiffies, giveup))
2069 				lseg = NULL;
2070 			break;
2071 		case -ERECALLCONFLICT:
2072 		case -EAGAIN:
2073 			break;
2074 		default:
2075 			if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2076 				pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2077 				lseg = NULL;
2078 			}
2079 			goto out_put_layout_hdr;
2080 		}
2081 		if (lseg) {
2082 			if (first)
2083 				pnfs_clear_first_layoutget(lo);
2084 			trace_pnfs_update_layout(ino, pos, count,
2085 				iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2086 			pnfs_put_layout_hdr(lo);
2087 			goto lookup_again;
2088 		}
2089 	} else {
2090 		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2091 	}
2092 
2093 out_put_layout_hdr:
2094 	if (first)
2095 		pnfs_clear_first_layoutget(lo);
2096 	trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2097 				 PNFS_UPDATE_LAYOUT_EXIT);
2098 	pnfs_put_layout_hdr(lo);
2099 out:
2100 	dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2101 			"(%s, offset: %llu, length: %llu)\n",
2102 			__func__, ino->i_sb->s_id,
2103 			(unsigned long long)NFS_FILEID(ino),
2104 			IS_ERR_OR_NULL(lseg) ? "not found" : "found",
2105 			iomode==IOMODE_RW ?  "read/write" : "read-only",
2106 			(unsigned long long)pos,
2107 			(unsigned long long)count);
2108 	return lseg;
2109 out_unlock:
2110 	spin_unlock(&ino->i_lock);
2111 	goto out_put_layout_hdr;
2112 }
2113 EXPORT_SYMBOL_GPL(pnfs_update_layout);
2114 
2115 static bool
2116 pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2117 {
2118 	switch (range->iomode) {
2119 	case IOMODE_READ:
2120 	case IOMODE_RW:
2121 		break;
2122 	default:
2123 		return false;
2124 	}
2125 	if (range->offset == NFS4_MAX_UINT64)
2126 		return false;
2127 	if (range->length == 0)
2128 		return false;
2129 	if (range->length != NFS4_MAX_UINT64 &&
2130 	    range->length > NFS4_MAX_UINT64 - range->offset)
2131 		return false;
2132 	return true;
2133 }
2134 
2135 static struct pnfs_layout_hdr *
2136 _pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2137 {
2138 	struct pnfs_layout_hdr *lo;
2139 
2140 	spin_lock(&ino->i_lock);
2141 	lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2142 	if (!lo)
2143 		goto out_unlock;
2144 	if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2145 		goto out_unlock;
2146 	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2147 		goto out_unlock;
2148 	if (pnfs_layoutgets_blocked(lo))
2149 		goto out_unlock;
2150 	if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2151 		goto out_unlock;
2152 	nfs_layoutget_begin(lo);
2153 	spin_unlock(&ino->i_lock);
2154 	_add_to_server_list(lo, NFS_SERVER(ino));
2155 	return lo;
2156 
2157 out_unlock:
2158 	spin_unlock(&ino->i_lock);
2159 	pnfs_put_layout_hdr(lo);
2160 	return NULL;
2161 }
2162 
2163 static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2164 				     struct nfs_open_context *ctx)
2165 {
2166 	struct inode *ino = data->dentry->d_inode;
2167 	struct pnfs_layout_range rng = {
2168 		.iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2169 			  IOMODE_RW: IOMODE_READ,
2170 		.offset = 0,
2171 		.length = NFS4_MAX_UINT64,
2172 	};
2173 	struct nfs4_layoutget *lgp;
2174 	struct pnfs_layout_hdr *lo;
2175 
2176 	/* Heuristic: don't send layoutget if we have cached data */
2177 	if (rng.iomode == IOMODE_READ &&
2178 	   (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2179 		return;
2180 
2181 	lo = _pnfs_grab_empty_layout(ino, ctx);
2182 	if (!lo)
2183 		return;
2184 	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2185 					     &rng, GFP_KERNEL);
2186 	if (!lgp) {
2187 		pnfs_clear_first_layoutget(lo);
2188 		pnfs_put_layout_hdr(lo);
2189 		return;
2190 	}
2191 	data->lgp = lgp;
2192 	data->o_arg.lg_args = &lgp->args;
2193 	data->o_res.lg_res = &lgp->res;
2194 }
2195 
2196 static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2197 				     struct nfs_open_context *ctx)
2198 {
2199 	struct pnfs_layout_range rng = {
2200 		.iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2201 			  IOMODE_RW: IOMODE_READ,
2202 		.offset = 0,
2203 		.length = NFS4_MAX_UINT64,
2204 	};
2205 	struct nfs4_layoutget *lgp;
2206 
2207 	lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2208 					     &rng, GFP_KERNEL);
2209 	if (!lgp)
2210 		return;
2211 	data->lgp = lgp;
2212 	data->o_arg.lg_args = &lgp->args;
2213 	data->o_res.lg_res = &lgp->res;
2214 }
2215 
2216 void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2217 			 struct nfs_open_context *ctx)
2218 {
2219 	struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2220 
2221 	if (!(pnfs_enabled_sb(server) &&
2222 	      server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2223 		return;
2224 	/* Could check on max_ops, but currently hardcoded high enough */
2225 	if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2226 		return;
2227 	if (data->state)
2228 		_lgopen_prepare_attached(data, ctx);
2229 	else
2230 		_lgopen_prepare_floating(data, ctx);
2231 }
2232 
2233 void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2234 		       struct nfs_open_context *ctx)
2235 {
2236 	struct pnfs_layout_hdr *lo;
2237 	struct pnfs_layout_segment *lseg;
2238 	struct nfs_server *srv = NFS_SERVER(ino);
2239 	u32 iomode;
2240 
2241 	if (!lgp)
2242 		return;
2243 	dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2244 	if (lgp->res.status) {
2245 		switch (lgp->res.status) {
2246 		default:
2247 			break;
2248 		/*
2249 		 * Halt lgopen attempts if the server doesn't recognise
2250 		 * the "current stateid" value, the layout type, or the
2251 		 * layoutget operation as being valid.
2252 		 * Also if it complains about too many ops in the compound
2253 		 * or of the request/reply being too big.
2254 		 */
2255 		case -NFS4ERR_BAD_STATEID:
2256 		case -NFS4ERR_NOTSUPP:
2257 		case -NFS4ERR_REP_TOO_BIG:
2258 		case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2259 		case -NFS4ERR_REQ_TOO_BIG:
2260 		case -NFS4ERR_TOO_MANY_OPS:
2261 		case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
2262 			srv->caps &= ~NFS_CAP_LGOPEN;
2263 		}
2264 		return;
2265 	}
2266 	if (!lgp->args.inode) {
2267 		lo = _pnfs_grab_empty_layout(ino, ctx);
2268 		if (!lo)
2269 			return;
2270 		lgp->args.inode = ino;
2271 	} else
2272 		lo = NFS_I(lgp->args.inode)->layout;
2273 
2274 	lseg = pnfs_layout_process(lgp);
2275 	if (!IS_ERR(lseg)) {
2276 		iomode = lgp->args.range.iomode;
2277 		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2278 		pnfs_put_lseg(lseg);
2279 	}
2280 }
2281 
2282 void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2283 {
2284 	if (lgp != NULL) {
2285 		struct inode *inode = lgp->args.inode;
2286 		if (inode) {
2287 			struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2288 			pnfs_clear_first_layoutget(lo);
2289 			nfs_layoutget_end(lo);
2290 		}
2291 		pnfs_layoutget_free(lgp);
2292 	}
2293 }
2294 
2295 struct pnfs_layout_segment *
2296 pnfs_layout_process(struct nfs4_layoutget *lgp)
2297 {
2298 	struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2299 	struct nfs4_layoutget_res *res = &lgp->res;
2300 	struct pnfs_layout_segment *lseg;
2301 	struct inode *ino = lo->plh_inode;
2302 	LIST_HEAD(free_me);
2303 
2304 	if (!pnfs_sanity_check_layout_range(&res->range))
2305 		return ERR_PTR(-EINVAL);
2306 
2307 	/* Inject layout blob into I/O device driver */
2308 	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
2309 	if (IS_ERR_OR_NULL(lseg)) {
2310 		if (!lseg)
2311 			lseg = ERR_PTR(-ENOMEM);
2312 
2313 		dprintk("%s: Could not allocate layout: error %ld\n",
2314 		       __func__, PTR_ERR(lseg));
2315 		return lseg;
2316 	}
2317 
2318 	pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
2319 
2320 	spin_lock(&ino->i_lock);
2321 	if (pnfs_layoutgets_blocked(lo)) {
2322 		dprintk("%s forget reply due to state\n", __func__);
2323 		goto out_forget;
2324 	}
2325 
2326 	if (!pnfs_layout_is_valid(lo)) {
2327 		/* We have a completely new layout */
2328 		pnfs_set_layout_stateid(lo, &res->stateid, true);
2329 	} else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
2330 		/* existing state ID, make sure the sequence number matches. */
2331 		if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2332 			dprintk("%s forget reply due to sequence\n", __func__);
2333 			goto out_forget;
2334 		}
2335 		pnfs_set_layout_stateid(lo, &res->stateid, false);
2336 	} else {
2337 		/*
2338 		 * We got an entirely new state ID.  Mark all segments for the
2339 		 * inode invalid, and retry the layoutget
2340 		 */
2341 		pnfs_mark_layout_stateid_invalid(lo, &free_me);
2342 		goto out_forget;
2343 	}
2344 
2345 	pnfs_get_lseg(lseg);
2346 	pnfs_layout_insert_lseg(lo, lseg, &free_me);
2347 
2348 
2349 	if (res->return_on_close)
2350 		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
2351 
2352 	spin_unlock(&ino->i_lock);
2353 	pnfs_free_lseg_list(&free_me);
2354 	return lseg;
2355 
2356 out_forget:
2357 	spin_unlock(&ino->i_lock);
2358 	lseg->pls_layout = lo;
2359 	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
2360 	return ERR_PTR(-EAGAIN);
2361 }
2362 
2363 static int
2364 mark_lseg_invalid_or_return(struct pnfs_layout_segment *lseg,
2365 		struct list_head *tmp_list)
2366 {
2367 	if (!mark_lseg_invalid(lseg, tmp_list))
2368 		return 0;
2369 	pnfs_cache_lseg_for_layoutreturn(lseg->pls_layout, lseg);
2370 	return 1;
2371 }
2372 
2373 /**
2374  * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2375  * @lo: pointer to layout header
2376  * @tmp_list: list header to be used with pnfs_free_lseg_list()
2377  * @return_range: describe layout segment ranges to be returned
2378  * @seq: stateid seqid to match
2379  *
2380  * This function is mainly intended for use by layoutrecall. It attempts
2381  * to free the layout segment immediately, or else to mark it for return
2382  * as soon as its reference count drops to zero.
2383  *
2384  * Returns
2385  * - 0: a layoutreturn needs to be scheduled.
2386  * - EBUSY: there are layout segment that are still in use.
2387  * - ENOENT: there are no layout segments that need to be returned.
2388  */
2389 int
2390 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2391 				struct list_head *tmp_list,
2392 				const struct pnfs_layout_range *return_range,
2393 				u32 seq)
2394 {
2395 	struct pnfs_layout_segment *lseg, *next;
2396 	int remaining = 0;
2397 
2398 	dprintk("%s:Begin lo %p\n", __func__, lo);
2399 
2400 	assert_spin_locked(&lo->plh_inode->i_lock);
2401 
2402 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
2403 		if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
2404 			dprintk("%s: marking lseg %p iomode %d "
2405 				"offset %llu length %llu\n", __func__,
2406 				lseg, lseg->pls_range.iomode,
2407 				lseg->pls_range.offset,
2408 				lseg->pls_range.length);
2409 			if (mark_lseg_invalid_or_return(lseg, tmp_list))
2410 				continue;
2411 			remaining++;
2412 			set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
2413 		}
2414 
2415 	if (remaining) {
2416 		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2417 		return -EBUSY;
2418 	}
2419 
2420 	if (!list_empty(&lo->plh_return_segs)) {
2421 		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2422 		return 0;
2423 	}
2424 
2425 	return -ENOENT;
2426 }
2427 
2428 void pnfs_error_mark_layout_for_return(struct inode *inode,
2429 				       struct pnfs_layout_segment *lseg)
2430 {
2431 	struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2432 	struct pnfs_layout_range range = {
2433 		.iomode = lseg->pls_range.iomode,
2434 		.offset = 0,
2435 		.length = NFS4_MAX_UINT64,
2436 	};
2437 	bool return_now = false;
2438 
2439 	spin_lock(&inode->i_lock);
2440 	if (!pnfs_layout_is_valid(lo)) {
2441 		spin_unlock(&inode->i_lock);
2442 		return;
2443 	}
2444 	pnfs_set_plh_return_info(lo, range.iomode, 0);
2445 	/*
2446 	 * mark all matching lsegs so that we are sure to have no live
2447 	 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2448 	 * for how it works.
2449 	 */
2450 	if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
2451 		nfs4_stateid stateid;
2452 		enum pnfs_iomode iomode;
2453 
2454 		return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
2455 		spin_unlock(&inode->i_lock);
2456 		if (return_now)
2457 			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2458 	} else {
2459 		spin_unlock(&inode->i_lock);
2460 		nfs_commit_inode(inode, 0);
2461 	}
2462 }
2463 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2464 
2465 void
2466 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2467 {
2468 	if (pgio->pg_lseg == NULL ||
2469 	    test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2470 		return;
2471 	pnfs_put_lseg(pgio->pg_lseg);
2472 	pgio->pg_lseg = NULL;
2473 }
2474 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2475 
2476 /*
2477  * Check for any intersection between the request and the pgio->pg_lseg,
2478  * and if none, put this pgio->pg_lseg away.
2479  */
2480 static void
2481 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2482 {
2483 	if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2484 		pnfs_put_lseg(pgio->pg_lseg);
2485 		pgio->pg_lseg = NULL;
2486 	}
2487 }
2488 
2489 void
2490 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2491 {
2492 	u64 rd_size = req->wb_bytes;
2493 
2494 	pnfs_generic_pg_check_layout(pgio);
2495 	pnfs_generic_pg_check_range(pgio, req);
2496 	if (pgio->pg_lseg == NULL) {
2497 		if (pgio->pg_dreq == NULL)
2498 			rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2499 		else
2500 			rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2501 
2502 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2503 						   nfs_req_openctx(req),
2504 						   req_offset(req),
2505 						   rd_size,
2506 						   IOMODE_READ,
2507 						   false,
2508 						   GFP_KERNEL);
2509 		if (IS_ERR(pgio->pg_lseg)) {
2510 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2511 			pgio->pg_lseg = NULL;
2512 			return;
2513 		}
2514 	}
2515 	/* If no lseg, fall back to read through mds */
2516 	if (pgio->pg_lseg == NULL)
2517 		nfs_pageio_reset_read_mds(pgio);
2518 
2519 }
2520 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2521 
2522 void
2523 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2524 			   struct nfs_page *req, u64 wb_size)
2525 {
2526 	pnfs_generic_pg_check_layout(pgio);
2527 	pnfs_generic_pg_check_range(pgio, req);
2528 	if (pgio->pg_lseg == NULL) {
2529 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2530 						   nfs_req_openctx(req),
2531 						   req_offset(req),
2532 						   wb_size,
2533 						   IOMODE_RW,
2534 						   false,
2535 						   GFP_KERNEL);
2536 		if (IS_ERR(pgio->pg_lseg)) {
2537 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2538 			pgio->pg_lseg = NULL;
2539 			return;
2540 		}
2541 	}
2542 	/* If no lseg, fall back to write through mds */
2543 	if (pgio->pg_lseg == NULL)
2544 		nfs_pageio_reset_write_mds(pgio);
2545 }
2546 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2547 
2548 void
2549 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2550 {
2551 	if (desc->pg_lseg) {
2552 		pnfs_put_lseg(desc->pg_lseg);
2553 		desc->pg_lseg = NULL;
2554 	}
2555 }
2556 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2557 
2558 /*
2559  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2560  * of bytes (maximum @req->wb_bytes) that can be coalesced.
2561  */
2562 size_t
2563 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2564 		     struct nfs_page *prev, struct nfs_page *req)
2565 {
2566 	unsigned int size;
2567 	u64 seg_end, req_start, seg_left;
2568 
2569 	size = nfs_generic_pg_test(pgio, prev, req);
2570 	if (!size)
2571 		return 0;
2572 
2573 	/*
2574 	 * 'size' contains the number of bytes left in the current page (up
2575 	 * to the original size asked for in @req->wb_bytes).
2576 	 *
2577 	 * Calculate how many bytes are left in the layout segment
2578 	 * and if there are less bytes than 'size', return that instead.
2579 	 *
2580 	 * Please also note that 'end_offset' is actually the offset of the
2581 	 * first byte that lies outside the pnfs_layout_range. FIXME?
2582 	 *
2583 	 */
2584 	if (pgio->pg_lseg) {
2585 		seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
2586 				     pgio->pg_lseg->pls_range.length);
2587 		req_start = req_offset(req);
2588 
2589 		/* start of request is past the last byte of this segment */
2590 		if (req_start >= seg_end)
2591 			return 0;
2592 
2593 		/* adjust 'size' iff there are fewer bytes left in the
2594 		 * segment than what nfs_generic_pg_test returned */
2595 		seg_left = seg_end - req_start;
2596 		if (seg_left < size)
2597 			size = (unsigned int)seg_left;
2598 	}
2599 
2600 	return size;
2601 }
2602 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
2603 
2604 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
2605 {
2606 	struct nfs_pageio_descriptor pgio;
2607 
2608 	/* Resend all requests through the MDS */
2609 	nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2610 			      hdr->completion_ops);
2611 	set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
2612 	return nfs_pageio_resend(&pgio, hdr);
2613 }
2614 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
2615 
2616 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
2617 {
2618 
2619 	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2620 	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2621 	    PNFS_LAYOUTRET_ON_ERROR) {
2622 		pnfs_return_layout(hdr->inode);
2623 	}
2624 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2625 		hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
2626 }
2627 
2628 /*
2629  * Called by non rpc-based layout drivers
2630  */
2631 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
2632 {
2633 	if (likely(!hdr->pnfs_error)) {
2634 		pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2635 				hdr->mds_offset + hdr->res.count);
2636 		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2637 	}
2638 	trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2639 	if (unlikely(hdr->pnfs_error))
2640 		pnfs_ld_handle_write_error(hdr);
2641 	hdr->mds_ops->rpc_release(hdr);
2642 }
2643 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
2644 
2645 static void
2646 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
2647 		struct nfs_pgio_header *hdr)
2648 {
2649 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2650 
2651 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2652 		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2653 		nfs_pageio_reset_write_mds(desc);
2654 		mirror->pg_recoalesce = 1;
2655 	}
2656 	hdr->completion_ops->completion(hdr);
2657 }
2658 
2659 static enum pnfs_try_status
2660 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
2661 			const struct rpc_call_ops *call_ops,
2662 			struct pnfs_layout_segment *lseg,
2663 			int how)
2664 {
2665 	struct inode *inode = hdr->inode;
2666 	enum pnfs_try_status trypnfs;
2667 	struct nfs_server *nfss = NFS_SERVER(inode);
2668 
2669 	hdr->mds_ops = call_ops;
2670 
2671 	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
2672 		inode->i_ino, hdr->args.count, hdr->args.offset, how);
2673 	trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
2674 	if (trypnfs != PNFS_NOT_ATTEMPTED)
2675 		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
2676 	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2677 	return trypnfs;
2678 }
2679 
2680 static void
2681 pnfs_do_write(struct nfs_pageio_descriptor *desc,
2682 	      struct nfs_pgio_header *hdr, int how)
2683 {
2684 	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2685 	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2686 	enum pnfs_try_status trypnfs;
2687 
2688 	trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2689 	switch (trypnfs) {
2690 	case PNFS_NOT_ATTEMPTED:
2691 		pnfs_write_through_mds(desc, hdr);
2692 	case PNFS_ATTEMPTED:
2693 		break;
2694 	case PNFS_TRY_AGAIN:
2695 		/* cleanup hdr and prepare to redo pnfs */
2696 		if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2697 			struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2698 			list_splice_init(&hdr->pages, &mirror->pg_list);
2699 			mirror->pg_recoalesce = 1;
2700 		}
2701 		hdr->mds_ops->rpc_release(hdr);
2702 	}
2703 }
2704 
2705 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2706 {
2707 	pnfs_put_lseg(hdr->lseg);
2708 	nfs_pgio_header_free(hdr);
2709 }
2710 
2711 int
2712 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2713 {
2714 	struct nfs_pgio_header *hdr;
2715 	int ret;
2716 
2717 	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2718 	if (!hdr) {
2719 		desc->pg_error = -ENOMEM;
2720 		return desc->pg_error;
2721 	}
2722 	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2723 
2724 	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2725 	ret = nfs_generic_pgio(desc, hdr);
2726 	if (!ret)
2727 		pnfs_do_write(desc, hdr, desc->pg_ioflags);
2728 
2729 	return ret;
2730 }
2731 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2732 
2733 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2734 {
2735 	struct nfs_pageio_descriptor pgio;
2736 
2737 	/* Resend all requests through the MDS */
2738 	nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2739 	return nfs_pageio_resend(&pgio, hdr);
2740 }
2741 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2742 
2743 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2744 {
2745 	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2746 	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2747 	    PNFS_LAYOUTRET_ON_ERROR) {
2748 		pnfs_return_layout(hdr->inode);
2749 	}
2750 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2751 		hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2752 }
2753 
2754 /*
2755  * Called by non rpc-based layout drivers
2756  */
2757 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2758 {
2759 	if (likely(!hdr->pnfs_error))
2760 		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2761 	trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2762 	if (unlikely(hdr->pnfs_error))
2763 		pnfs_ld_handle_read_error(hdr);
2764 	hdr->mds_ops->rpc_release(hdr);
2765 }
2766 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2767 
2768 static void
2769 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2770 		struct nfs_pgio_header *hdr)
2771 {
2772 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2773 
2774 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2775 		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2776 		nfs_pageio_reset_read_mds(desc);
2777 		mirror->pg_recoalesce = 1;
2778 	}
2779 	hdr->completion_ops->completion(hdr);
2780 }
2781 
2782 /*
2783  * Call the appropriate parallel I/O subsystem read function.
2784  */
2785 static enum pnfs_try_status
2786 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2787 		       const struct rpc_call_ops *call_ops,
2788 		       struct pnfs_layout_segment *lseg)
2789 {
2790 	struct inode *inode = hdr->inode;
2791 	struct nfs_server *nfss = NFS_SERVER(inode);
2792 	enum pnfs_try_status trypnfs;
2793 
2794 	hdr->mds_ops = call_ops;
2795 
2796 	dprintk("%s: Reading ino:%lu %u@%llu\n",
2797 		__func__, inode->i_ino, hdr->args.count, hdr->args.offset);
2798 
2799 	trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2800 	if (trypnfs != PNFS_NOT_ATTEMPTED)
2801 		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
2802 	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2803 	return trypnfs;
2804 }
2805 
2806 /* Resend all requests through pnfs. */
2807 void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2808 {
2809 	struct nfs_pageio_descriptor pgio;
2810 
2811 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2812 		/* Prevent deadlocks with layoutreturn! */
2813 		pnfs_put_lseg(hdr->lseg);
2814 		hdr->lseg = NULL;
2815 
2816 		nfs_pageio_init_read(&pgio, hdr->inode, false,
2817 					hdr->completion_ops);
2818 		hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2819 	}
2820 }
2821 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2822 
2823 static void
2824 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2825 {
2826 	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2827 	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2828 	enum pnfs_try_status trypnfs;
2829 
2830 	trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2831 	switch (trypnfs) {
2832 	case PNFS_NOT_ATTEMPTED:
2833 		pnfs_read_through_mds(desc, hdr);
2834 	case PNFS_ATTEMPTED:
2835 		break;
2836 	case PNFS_TRY_AGAIN:
2837 		/* cleanup hdr and prepare to redo pnfs */
2838 		if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2839 			struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2840 			list_splice_init(&hdr->pages, &mirror->pg_list);
2841 			mirror->pg_recoalesce = 1;
2842 		}
2843 		hdr->mds_ops->rpc_release(hdr);
2844 	}
2845 }
2846 
2847 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2848 {
2849 	pnfs_put_lseg(hdr->lseg);
2850 	nfs_pgio_header_free(hdr);
2851 }
2852 
2853 int
2854 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2855 {
2856 	struct nfs_pgio_header *hdr;
2857 	int ret;
2858 
2859 	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2860 	if (!hdr) {
2861 		desc->pg_error = -ENOMEM;
2862 		return desc->pg_error;
2863 	}
2864 	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2865 	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2866 	ret = nfs_generic_pgio(desc, hdr);
2867 	if (!ret)
2868 		pnfs_do_read(desc, hdr);
2869 	return ret;
2870 }
2871 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2872 
2873 static void pnfs_clear_layoutcommitting(struct inode *inode)
2874 {
2875 	unsigned long *bitlock = &NFS_I(inode)->flags;
2876 
2877 	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2878 	smp_mb__after_atomic();
2879 	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2880 }
2881 
2882 /*
2883  * There can be multiple RW segments.
2884  */
2885 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
2886 {
2887 	struct pnfs_layout_segment *lseg;
2888 
2889 	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2890 		if (lseg->pls_range.iomode == IOMODE_RW &&
2891 		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2892 			list_add(&lseg->pls_lc_list, listp);
2893 	}
2894 }
2895 
2896 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2897 {
2898 	struct pnfs_layout_segment *lseg, *tmp;
2899 
2900 	/* Matched by references in pnfs_set_layoutcommit */
2901 	list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2902 		list_del_init(&lseg->pls_lc_list);
2903 		pnfs_put_lseg(lseg);
2904 	}
2905 
2906 	pnfs_clear_layoutcommitting(inode);
2907 }
2908 
2909 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2910 {
2911 	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
2912 }
2913 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2914 
2915 void
2916 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2917 		loff_t end_pos)
2918 {
2919 	struct nfs_inode *nfsi = NFS_I(inode);
2920 	bool mark_as_dirty = false;
2921 
2922 	spin_lock(&inode->i_lock);
2923 	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2924 		nfsi->layout->plh_lwb = end_pos;
2925 		mark_as_dirty = true;
2926 		dprintk("%s: Set layoutcommit for inode %lu ",
2927 			__func__, inode->i_ino);
2928 	} else if (end_pos > nfsi->layout->plh_lwb)
2929 		nfsi->layout->plh_lwb = end_pos;
2930 	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2931 		/* references matched in nfs4_layoutcommit_release */
2932 		pnfs_get_lseg(lseg);
2933 	}
2934 	spin_unlock(&inode->i_lock);
2935 	dprintk("%s: lseg %p end_pos %llu\n",
2936 		__func__, lseg, nfsi->layout->plh_lwb);
2937 
2938 	/* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2939 	 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2940 	if (mark_as_dirty)
2941 		mark_inode_dirty_sync(inode);
2942 }
2943 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2944 
2945 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2946 {
2947 	struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2948 
2949 	if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2950 		nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
2951 	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
2952 }
2953 
2954 /*
2955  * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2956  * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2957  * data to disk to allow the server to recover the data if it crashes.
2958  * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2959  * is off, and a COMMIT is sent to a data server, or
2960  * if WRITEs to a data server return NFS_DATA_SYNC.
2961  */
2962 int
2963 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
2964 {
2965 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2966 	struct nfs4_layoutcommit_data *data;
2967 	struct nfs_inode *nfsi = NFS_I(inode);
2968 	loff_t end_pos;
2969 	int status;
2970 
2971 	if (!pnfs_layoutcommit_outstanding(inode))
2972 		return 0;
2973 
2974 	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
2975 
2976 	status = -EAGAIN;
2977 	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2978 		if (!sync)
2979 			goto out;
2980 		status = wait_on_bit_lock_action(&nfsi->flags,
2981 				NFS_INO_LAYOUTCOMMITTING,
2982 				nfs_wait_bit_killable,
2983 				TASK_KILLABLE);
2984 		if (status)
2985 			goto out;
2986 	}
2987 
2988 	status = -ENOMEM;
2989 	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2990 	data = kzalloc(sizeof(*data), GFP_NOFS);
2991 	if (!data)
2992 		goto clear_layoutcommitting;
2993 
2994 	status = 0;
2995 	spin_lock(&inode->i_lock);
2996 	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2997 		goto out_unlock;
2998 
2999 	INIT_LIST_HEAD(&data->lseg_list);
3000 	pnfs_list_write_lseg(inode, &data->lseg_list);
3001 
3002 	end_pos = nfsi->layout->plh_lwb;
3003 
3004 	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
3005 	spin_unlock(&inode->i_lock);
3006 
3007 	data->args.inode = inode;
3008 	data->cred = get_cred(nfsi->layout->plh_lc_cred);
3009 	nfs_fattr_init(&data->fattr);
3010 	data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3011 	data->res.fattr = &data->fattr;
3012 	if (end_pos != 0)
3013 		data->args.lastbytewritten = end_pos - 1;
3014 	else
3015 		data->args.lastbytewritten = U64_MAX;
3016 	data->res.server = NFS_SERVER(inode);
3017 
3018 	if (ld->prepare_layoutcommit) {
3019 		status = ld->prepare_layoutcommit(&data->args);
3020 		if (status) {
3021 			put_cred(data->cred);
3022 			spin_lock(&inode->i_lock);
3023 			set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3024 			if (end_pos > nfsi->layout->plh_lwb)
3025 				nfsi->layout->plh_lwb = end_pos;
3026 			goto out_unlock;
3027 		}
3028 	}
3029 
3030 
3031 	status = nfs4_proc_layoutcommit(data, sync);
3032 out:
3033 	if (status)
3034 		mark_inode_dirty_sync(inode);
3035 	dprintk("<-- %s status %d\n", __func__, status);
3036 	return status;
3037 out_unlock:
3038 	spin_unlock(&inode->i_lock);
3039 	kfree(data);
3040 clear_layoutcommitting:
3041 	pnfs_clear_layoutcommitting(inode);
3042 	goto out;
3043 }
3044 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
3045 
3046 int
3047 pnfs_generic_sync(struct inode *inode, bool datasync)
3048 {
3049 	return pnfs_layoutcommit_inode(inode, true);
3050 }
3051 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3052 
3053 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3054 {
3055 	struct nfs4_threshold *thp;
3056 
3057 	thp = kzalloc(sizeof(*thp), GFP_NOFS);
3058 	if (!thp) {
3059 		dprintk("%s mdsthreshold allocation failed\n", __func__);
3060 		return NULL;
3061 	}
3062 	return thp;
3063 }
3064 
3065 #if IS_ENABLED(CONFIG_NFS_V4_2)
3066 int
3067 pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
3068 {
3069 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3070 	struct nfs_server *server = NFS_SERVER(inode);
3071 	struct nfs_inode *nfsi = NFS_I(inode);
3072 	struct nfs42_layoutstat_data *data;
3073 	struct pnfs_layout_hdr *hdr;
3074 	int status = 0;
3075 
3076 	if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3077 		goto out;
3078 
3079 	if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3080 		goto out;
3081 
3082 	if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3083 		goto out;
3084 
3085 	spin_lock(&inode->i_lock);
3086 	if (!NFS_I(inode)->layout) {
3087 		spin_unlock(&inode->i_lock);
3088 		goto out_clear_layoutstats;
3089 	}
3090 	hdr = NFS_I(inode)->layout;
3091 	pnfs_get_layout_hdr(hdr);
3092 	spin_unlock(&inode->i_lock);
3093 
3094 	data = kzalloc(sizeof(*data), gfp_flags);
3095 	if (!data) {
3096 		status = -ENOMEM;
3097 		goto out_put;
3098 	}
3099 
3100 	data->args.fh = NFS_FH(inode);
3101 	data->args.inode = inode;
3102 	status = ld->prepare_layoutstats(&data->args);
3103 	if (status)
3104 		goto out_free;
3105 
3106 	status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3107 
3108 out:
3109 	dprintk("%s returns %d\n", __func__, status);
3110 	return status;
3111 
3112 out_free:
3113 	kfree(data);
3114 out_put:
3115 	pnfs_put_layout_hdr(hdr);
3116 out_clear_layoutstats:
3117 	smp_mb__before_atomic();
3118 	clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3119 	smp_mb__after_atomic();
3120 	goto out;
3121 }
3122 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
3123 #endif
3124 
3125 unsigned int layoutstats_timer;
3126 module_param(layoutstats_timer, uint, 0644);
3127 EXPORT_SYMBOL_GPL(layoutstats_timer);
3128