xref: /openbmc/linux/fs/nfs/pnfs.c (revision cb1aaebe)
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
363  */
364 bool nfs4_layoutreturn_refresh_stateid(nfs4_stateid *dst,
365 		struct pnfs_layout_range *dst_range,
366 		struct inode *inode)
367 {
368 	struct pnfs_layout_hdr *lo;
369 	struct pnfs_layout_range range = {
370 		.iomode = IOMODE_ANY,
371 		.offset = 0,
372 		.length = NFS4_MAX_UINT64,
373 	};
374 	bool ret = false;
375 	LIST_HEAD(head);
376 	int err;
377 
378 	spin_lock(&inode->i_lock);
379 	lo = NFS_I(inode)->layout;
380 	if (lo && nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
381 		err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
382 		if (err != -EBUSY) {
383 			dst->seqid = lo->plh_stateid.seqid;
384 			*dst_range = range;
385 			ret = true;
386 		}
387 	}
388 	spin_unlock(&inode->i_lock);
389 	pnfs_free_lseg_list(&head);
390 	return ret;
391 }
392 
393 /*
394  * Mark a pnfs_layout_hdr and all associated layout segments as invalid
395  *
396  * In order to continue using the pnfs_layout_hdr, a full recovery
397  * is required.
398  * Note that caller must hold inode->i_lock.
399  */
400 int
401 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
402 		struct list_head *lseg_list)
403 {
404 	struct pnfs_layout_range range = {
405 		.iomode = IOMODE_ANY,
406 		.offset = 0,
407 		.length = NFS4_MAX_UINT64,
408 	};
409 	struct pnfs_layout_segment *lseg, *next;
410 
411 	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
412 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
413 		pnfs_clear_lseg_state(lseg, lseg_list);
414 	pnfs_clear_layoutreturn_info(lo);
415 	pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
416 	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
417 	    !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
418 		pnfs_clear_layoutreturn_waitbit(lo);
419 	return !list_empty(&lo->plh_segs);
420 }
421 
422 static int
423 pnfs_iomode_to_fail_bit(u32 iomode)
424 {
425 	return iomode == IOMODE_RW ?
426 		NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
427 }
428 
429 static void
430 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
431 {
432 	lo->plh_retry_timestamp = jiffies;
433 	if (!test_and_set_bit(fail_bit, &lo->plh_flags))
434 		refcount_inc(&lo->plh_refcount);
435 }
436 
437 static void
438 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
439 {
440 	if (test_and_clear_bit(fail_bit, &lo->plh_flags))
441 		refcount_dec(&lo->plh_refcount);
442 }
443 
444 static void
445 pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
446 {
447 	struct inode *inode = lo->plh_inode;
448 	struct pnfs_layout_range range = {
449 		.iomode = iomode,
450 		.offset = 0,
451 		.length = NFS4_MAX_UINT64,
452 	};
453 	LIST_HEAD(head);
454 
455 	spin_lock(&inode->i_lock);
456 	pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
457 	pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
458 	spin_unlock(&inode->i_lock);
459 	pnfs_free_lseg_list(&head);
460 	dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
461 			iomode == IOMODE_RW ?  "RW" : "READ");
462 }
463 
464 static bool
465 pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
466 {
467 	unsigned long start, end;
468 	int fail_bit = pnfs_iomode_to_fail_bit(iomode);
469 
470 	if (test_bit(fail_bit, &lo->plh_flags) == 0)
471 		return false;
472 	end = jiffies;
473 	start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
474 	if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
475 		/* It is time to retry the failed layoutgets */
476 		pnfs_layout_clear_fail_bit(lo, fail_bit);
477 		return false;
478 	}
479 	return true;
480 }
481 
482 static void
483 pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
484 		const struct pnfs_layout_range *range,
485 		const nfs4_stateid *stateid)
486 {
487 	INIT_LIST_HEAD(&lseg->pls_list);
488 	INIT_LIST_HEAD(&lseg->pls_lc_list);
489 	refcount_set(&lseg->pls_refcount, 1);
490 	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
491 	lseg->pls_layout = lo;
492 	lseg->pls_range = *range;
493 	lseg->pls_seq = be32_to_cpu(stateid->seqid);
494 }
495 
496 static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
497 {
498 	if (lseg != NULL) {
499 		struct inode *inode = lseg->pls_layout->plh_inode;
500 		NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
501 	}
502 }
503 
504 static void
505 pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
506 		struct pnfs_layout_segment *lseg)
507 {
508 	WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
509 	list_del_init(&lseg->pls_list);
510 	/* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
511 	refcount_dec(&lo->plh_refcount);
512 	if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
513 		return;
514 	if (list_empty(&lo->plh_segs) &&
515 	    !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
516 	    !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
517 		if (atomic_read(&lo->plh_outstanding) == 0)
518 			set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
519 		clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
520 	}
521 }
522 
523 static bool
524 pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
525 		struct pnfs_layout_segment *lseg)
526 {
527 	if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
528 	    pnfs_layout_is_valid(lo)) {
529 		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
530 		list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
531 		return true;
532 	}
533 	return false;
534 }
535 
536 void
537 pnfs_put_lseg(struct pnfs_layout_segment *lseg)
538 {
539 	struct pnfs_layout_hdr *lo;
540 	struct inode *inode;
541 
542 	if (!lseg)
543 		return;
544 
545 	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
546 		refcount_read(&lseg->pls_refcount),
547 		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
548 
549 	lo = lseg->pls_layout;
550 	inode = lo->plh_inode;
551 
552 	if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
553 		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
554 			spin_unlock(&inode->i_lock);
555 			return;
556 		}
557 		pnfs_get_layout_hdr(lo);
558 		pnfs_layout_remove_lseg(lo, lseg);
559 		if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
560 			lseg = NULL;
561 		spin_unlock(&inode->i_lock);
562 		pnfs_free_lseg(lseg);
563 		pnfs_put_layout_hdr(lo);
564 	}
565 }
566 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
567 
568 /*
569  * is l2 fully contained in l1?
570  *   start1                             end1
571  *   [----------------------------------)
572  *           start2           end2
573  *           [----------------)
574  */
575 static bool
576 pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
577 		 const struct pnfs_layout_range *l2)
578 {
579 	u64 start1 = l1->offset;
580 	u64 end1 = pnfs_end_offset(start1, l1->length);
581 	u64 start2 = l2->offset;
582 	u64 end2 = pnfs_end_offset(start2, l2->length);
583 
584 	return (start1 <= start2) && (end1 >= end2);
585 }
586 
587 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
588 		struct list_head *tmp_list)
589 {
590 	if (!refcount_dec_and_test(&lseg->pls_refcount))
591 		return false;
592 	pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
593 	list_add(&lseg->pls_list, tmp_list);
594 	return true;
595 }
596 
597 /* Returns 1 if lseg is removed from list, 0 otherwise */
598 static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
599 			     struct list_head *tmp_list)
600 {
601 	int rv = 0;
602 
603 	if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
604 		/* Remove the reference keeping the lseg in the
605 		 * list.  It will now be removed when all
606 		 * outstanding io is finished.
607 		 */
608 		dprintk("%s: lseg %p ref %d\n", __func__, lseg,
609 			refcount_read(&lseg->pls_refcount));
610 		if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
611 			rv = 1;
612 	}
613 	return rv;
614 }
615 
616 /*
617  * Compare 2 layout stateid sequence ids, to see which is newer,
618  * taking into account wraparound issues.
619  */
620 static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
621 {
622 	return (s32)(s1 - s2) > 0;
623 }
624 
625 static bool
626 pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
627 		 const struct pnfs_layout_range *recall_range)
628 {
629 	return (recall_range->iomode == IOMODE_ANY ||
630 		lseg_range->iomode == recall_range->iomode) &&
631 	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
632 }
633 
634 static bool
635 pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
636 		const struct pnfs_layout_range *recall_range,
637 		u32 seq)
638 {
639 	if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
640 		return false;
641 	if (recall_range == NULL)
642 		return true;
643 	return pnfs_should_free_range(&lseg->pls_range, recall_range);
644 }
645 
646 /**
647  * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
648  * @lo: layout header containing the lsegs
649  * @tmp_list: list head where doomed lsegs should go
650  * @recall_range: optional recall range argument to match (may be NULL)
651  * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
652  *
653  * Walk the list of lsegs in the layout header, and tear down any that should
654  * be destroyed. If "recall_range" is specified then the segment must match
655  * that range. If "seq" is non-zero, then only match segments that were handed
656  * out at or before that sequence.
657  *
658  * Returns number of matching invalid lsegs remaining in list after scanning
659  * it and purging them.
660  */
661 int
662 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
663 			    struct list_head *tmp_list,
664 			    const struct pnfs_layout_range *recall_range,
665 			    u32 seq)
666 {
667 	struct pnfs_layout_segment *lseg, *next;
668 	int remaining = 0;
669 
670 	dprintk("%s:Begin lo %p\n", __func__, lo);
671 
672 	if (list_empty(&lo->plh_segs))
673 		return 0;
674 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
675 		if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
676 			dprintk("%s: freeing lseg %p iomode %d seq %u "
677 				"offset %llu length %llu\n", __func__,
678 				lseg, lseg->pls_range.iomode, lseg->pls_seq,
679 				lseg->pls_range.offset, lseg->pls_range.length);
680 			if (!mark_lseg_invalid(lseg, tmp_list))
681 				remaining++;
682 		}
683 	dprintk("%s:Return %i\n", __func__, remaining);
684 	return remaining;
685 }
686 
687 static void
688 pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
689 		struct list_head *free_me,
690 		const struct pnfs_layout_range *range,
691 		u32 seq)
692 {
693 	struct pnfs_layout_segment *lseg, *next;
694 
695 	list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
696 		if (pnfs_match_lseg_recall(lseg, range, seq))
697 			list_move_tail(&lseg->pls_list, free_me);
698 	}
699 }
700 
701 /* note free_me must contain lsegs from a single layout_hdr */
702 void
703 pnfs_free_lseg_list(struct list_head *free_me)
704 {
705 	struct pnfs_layout_segment *lseg, *tmp;
706 
707 	if (list_empty(free_me))
708 		return;
709 
710 	list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
711 		list_del(&lseg->pls_list);
712 		pnfs_free_lseg(lseg);
713 	}
714 }
715 
716 void
717 pnfs_destroy_layout(struct nfs_inode *nfsi)
718 {
719 	struct pnfs_layout_hdr *lo;
720 	LIST_HEAD(tmp_list);
721 
722 	spin_lock(&nfsi->vfs_inode.i_lock);
723 	lo = nfsi->layout;
724 	if (lo) {
725 		pnfs_get_layout_hdr(lo);
726 		pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
727 		pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
728 		pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
729 		spin_unlock(&nfsi->vfs_inode.i_lock);
730 		pnfs_free_lseg_list(&tmp_list);
731 		nfs_commit_inode(&nfsi->vfs_inode, 0);
732 		pnfs_put_layout_hdr(lo);
733 	} else
734 		spin_unlock(&nfsi->vfs_inode.i_lock);
735 }
736 EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
737 
738 static bool
739 pnfs_layout_add_bulk_destroy_list(struct inode *inode,
740 		struct list_head *layout_list)
741 {
742 	struct pnfs_layout_hdr *lo;
743 	bool ret = false;
744 
745 	spin_lock(&inode->i_lock);
746 	lo = NFS_I(inode)->layout;
747 	if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
748 		pnfs_get_layout_hdr(lo);
749 		list_add(&lo->plh_bulk_destroy, layout_list);
750 		ret = true;
751 	}
752 	spin_unlock(&inode->i_lock);
753 	return ret;
754 }
755 
756 /* Caller must hold rcu_read_lock and clp->cl_lock */
757 static int
758 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
759 		struct nfs_server *server,
760 		struct list_head *layout_list)
761 	__must_hold(&clp->cl_lock)
762 	__must_hold(RCU)
763 {
764 	struct pnfs_layout_hdr *lo, *next;
765 	struct inode *inode;
766 
767 	list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
768 		if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
769 		    test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
770 		    !list_empty(&lo->plh_bulk_destroy))
771 			continue;
772 		/* If the sb is being destroyed, just bail */
773 		if (!nfs_sb_active(server->super))
774 			break;
775 		inode = igrab(lo->plh_inode);
776 		if (inode != NULL) {
777 			list_del_init(&lo->plh_layouts);
778 			if (pnfs_layout_add_bulk_destroy_list(inode,
779 						layout_list))
780 				continue;
781 			rcu_read_unlock();
782 			spin_unlock(&clp->cl_lock);
783 			iput(inode);
784 		} else {
785 			rcu_read_unlock();
786 			spin_unlock(&clp->cl_lock);
787 			set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
788 		}
789 		nfs_sb_deactive(server->super);
790 		spin_lock(&clp->cl_lock);
791 		rcu_read_lock();
792 		return -EAGAIN;
793 	}
794 	return 0;
795 }
796 
797 static int
798 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
799 		bool is_bulk_recall)
800 {
801 	struct pnfs_layout_hdr *lo;
802 	struct inode *inode;
803 	LIST_HEAD(lseg_list);
804 	int ret = 0;
805 
806 	while (!list_empty(layout_list)) {
807 		lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
808 				plh_bulk_destroy);
809 		dprintk("%s freeing layout for inode %lu\n", __func__,
810 			lo->plh_inode->i_ino);
811 		inode = lo->plh_inode;
812 
813 		pnfs_layoutcommit_inode(inode, false);
814 
815 		spin_lock(&inode->i_lock);
816 		list_del_init(&lo->plh_bulk_destroy);
817 		if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
818 			if (is_bulk_recall)
819 				set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
820 			ret = -EAGAIN;
821 		}
822 		spin_unlock(&inode->i_lock);
823 		pnfs_free_lseg_list(&lseg_list);
824 		/* Free all lsegs that are attached to commit buckets */
825 		nfs_commit_inode(inode, 0);
826 		pnfs_put_layout_hdr(lo);
827 		nfs_iput_and_deactive(inode);
828 	}
829 	return ret;
830 }
831 
832 int
833 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
834 		struct nfs_fsid *fsid,
835 		bool is_recall)
836 {
837 	struct nfs_server *server;
838 	LIST_HEAD(layout_list);
839 
840 	spin_lock(&clp->cl_lock);
841 	rcu_read_lock();
842 restart:
843 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
844 		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
845 			continue;
846 		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
847 				server,
848 				&layout_list) != 0)
849 			goto restart;
850 	}
851 	rcu_read_unlock();
852 	spin_unlock(&clp->cl_lock);
853 
854 	if (list_empty(&layout_list))
855 		return 0;
856 	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
857 }
858 
859 int
860 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
861 		bool is_recall)
862 {
863 	struct nfs_server *server;
864 	LIST_HEAD(layout_list);
865 
866 	spin_lock(&clp->cl_lock);
867 	rcu_read_lock();
868 restart:
869 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
870 		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
871 					server,
872 					&layout_list) != 0)
873 			goto restart;
874 	}
875 	rcu_read_unlock();
876 	spin_unlock(&clp->cl_lock);
877 
878 	if (list_empty(&layout_list))
879 		return 0;
880 	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
881 }
882 
883 /*
884  * Called by the state manger to remove all layouts established under an
885  * expired lease.
886  */
887 void
888 pnfs_destroy_all_layouts(struct nfs_client *clp)
889 {
890 	nfs4_deviceid_mark_client_invalid(clp);
891 	nfs4_deviceid_purge_client(clp);
892 
893 	pnfs_destroy_layouts_byclid(clp, false);
894 }
895 
896 /* update lo->plh_stateid with new if is more recent */
897 void
898 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
899 			bool update_barrier)
900 {
901 	u32 oldseq, newseq, new_barrier = 0;
902 
903 	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
904 	newseq = be32_to_cpu(new->seqid);
905 
906 	if (!pnfs_layout_is_valid(lo)) {
907 		nfs4_stateid_copy(&lo->plh_stateid, new);
908 		lo->plh_barrier = newseq;
909 		pnfs_clear_layoutreturn_info(lo);
910 		clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
911 		return;
912 	}
913 	if (pnfs_seqid_is_newer(newseq, oldseq)) {
914 		nfs4_stateid_copy(&lo->plh_stateid, new);
915 		/*
916 		 * Because of wraparound, we want to keep the barrier
917 		 * "close" to the current seqids.
918 		 */
919 		new_barrier = newseq - atomic_read(&lo->plh_outstanding);
920 	}
921 	if (update_barrier)
922 		new_barrier = be32_to_cpu(new->seqid);
923 	else if (new_barrier == 0)
924 		return;
925 	if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
926 		lo->plh_barrier = new_barrier;
927 }
928 
929 static bool
930 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
931 		const nfs4_stateid *stateid)
932 {
933 	u32 seqid = be32_to_cpu(stateid->seqid);
934 
935 	return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
936 }
937 
938 /* lget is set to 1 if called from inside send_layoutget call chain */
939 static bool
940 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
941 {
942 	return lo->plh_block_lgets ||
943 		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
944 }
945 
946 static struct nfs_server *
947 pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
948 {
949 	struct nfs_server *server;
950 
951 	if (inode) {
952 		server = NFS_SERVER(inode);
953 	} else {
954 		struct dentry *parent_dir = dget_parent(ctx->dentry);
955 		server = NFS_SERVER(parent_dir->d_inode);
956 		dput(parent_dir);
957 	}
958 	return server;
959 }
960 
961 static void nfs4_free_pages(struct page **pages, size_t size)
962 {
963 	int i;
964 
965 	if (!pages)
966 		return;
967 
968 	for (i = 0; i < size; i++) {
969 		if (!pages[i])
970 			break;
971 		__free_page(pages[i]);
972 	}
973 	kfree(pages);
974 }
975 
976 static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
977 {
978 	struct page **pages;
979 	int i;
980 
981 	pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
982 	if (!pages) {
983 		dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
984 		return NULL;
985 	}
986 
987 	for (i = 0; i < size; i++) {
988 		pages[i] = alloc_page(gfp_flags);
989 		if (!pages[i]) {
990 			dprintk("%s: failed to allocate page\n", __func__);
991 			nfs4_free_pages(pages, i);
992 			return NULL;
993 		}
994 	}
995 
996 	return pages;
997 }
998 
999 static struct nfs4_layoutget *
1000 pnfs_alloc_init_layoutget_args(struct inode *ino,
1001 	   struct nfs_open_context *ctx,
1002 	   const nfs4_stateid *stateid,
1003 	   const struct pnfs_layout_range *range,
1004 	   gfp_t gfp_flags)
1005 {
1006 	struct nfs_server *server = pnfs_find_server(ino, ctx);
1007 	size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
1008 	size_t max_pages = max_response_pages(server);
1009 	struct nfs4_layoutget *lgp;
1010 
1011 	dprintk("--> %s\n", __func__);
1012 
1013 	lgp = kzalloc(sizeof(*lgp), gfp_flags);
1014 	if (lgp == NULL)
1015 		return NULL;
1016 
1017 	if (max_reply_sz) {
1018 		size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1019 		if (npages < max_pages)
1020 			max_pages = npages;
1021 	}
1022 
1023 	lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1024 	if (!lgp->args.layout.pages) {
1025 		kfree(lgp);
1026 		return NULL;
1027 	}
1028 	lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1029 	lgp->res.layoutp = &lgp->args.layout;
1030 
1031 	/* Don't confuse uninitialised result and success */
1032 	lgp->res.status = -NFS4ERR_DELAY;
1033 
1034 	lgp->args.minlength = PAGE_SIZE;
1035 	if (lgp->args.minlength > range->length)
1036 		lgp->args.minlength = range->length;
1037 	if (ino) {
1038 		loff_t i_size = i_size_read(ino);
1039 
1040 		if (range->iomode == IOMODE_READ) {
1041 			if (range->offset >= i_size)
1042 				lgp->args.minlength = 0;
1043 			else if (i_size - range->offset < lgp->args.minlength)
1044 				lgp->args.minlength = i_size - range->offset;
1045 		}
1046 	}
1047 	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1048 	pnfs_copy_range(&lgp->args.range, range);
1049 	lgp->args.type = server->pnfs_curr_ld->id;
1050 	lgp->args.inode = ino;
1051 	lgp->args.ctx = get_nfs_open_context(ctx);
1052 	nfs4_stateid_copy(&lgp->args.stateid, stateid);
1053 	lgp->gfp_flags = gfp_flags;
1054 	lgp->cred = get_cred(ctx->cred);
1055 	return lgp;
1056 }
1057 
1058 void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1059 {
1060 	size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1061 
1062 	nfs4_free_pages(lgp->args.layout.pages, max_pages);
1063 	if (lgp->args.inode)
1064 		pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
1065 	put_cred(lgp->cred);
1066 	put_nfs_open_context(lgp->args.ctx);
1067 	kfree(lgp);
1068 }
1069 
1070 static void pnfs_clear_layoutcommit(struct inode *inode,
1071 		struct list_head *head)
1072 {
1073 	struct nfs_inode *nfsi = NFS_I(inode);
1074 	struct pnfs_layout_segment *lseg, *tmp;
1075 
1076 	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1077 		return;
1078 	list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1079 		if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1080 			continue;
1081 		pnfs_lseg_dec_and_remove_zero(lseg, head);
1082 	}
1083 }
1084 
1085 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
1086 		const nfs4_stateid *arg_stateid,
1087 		const struct pnfs_layout_range *range,
1088 		const nfs4_stateid *stateid)
1089 {
1090 	struct inode *inode = lo->plh_inode;
1091 	LIST_HEAD(freeme);
1092 
1093 	spin_lock(&inode->i_lock);
1094 	if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1095 	    !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1096 		goto out_unlock;
1097 	if (stateid) {
1098 		u32 seq = be32_to_cpu(arg_stateid->seqid);
1099 
1100 		pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1101 		pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1102 		pnfs_set_layout_stateid(lo, stateid, true);
1103 	} else
1104 		pnfs_mark_layout_stateid_invalid(lo, &freeme);
1105 out_unlock:
1106 	pnfs_clear_layoutreturn_waitbit(lo);
1107 	spin_unlock(&inode->i_lock);
1108 	pnfs_free_lseg_list(&freeme);
1109 
1110 }
1111 
1112 static bool
1113 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1114 		nfs4_stateid *stateid,
1115 		enum pnfs_iomode *iomode)
1116 {
1117 	/* Serialise LAYOUTGET/LAYOUTRETURN */
1118 	if (atomic_read(&lo->plh_outstanding) != 0)
1119 		return false;
1120 	if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
1121 		return false;
1122 	set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
1123 	pnfs_get_layout_hdr(lo);
1124 	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1125 		if (stateid != NULL) {
1126 			nfs4_stateid_copy(stateid, &lo->plh_stateid);
1127 			if (lo->plh_return_seq != 0)
1128 				stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1129 		}
1130 		if (iomode != NULL)
1131 			*iomode = lo->plh_return_iomode;
1132 		pnfs_clear_layoutreturn_info(lo);
1133 		return true;
1134 	}
1135 	if (stateid != NULL)
1136 		nfs4_stateid_copy(stateid, &lo->plh_stateid);
1137 	if (iomode != NULL)
1138 		*iomode = IOMODE_ANY;
1139 	return true;
1140 }
1141 
1142 static void
1143 pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1144 		struct pnfs_layout_hdr *lo,
1145 		const nfs4_stateid *stateid,
1146 		enum pnfs_iomode iomode)
1147 {
1148 	struct inode *inode = lo->plh_inode;
1149 
1150 	args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1151 	args->inode = inode;
1152 	args->range.iomode = iomode;
1153 	args->range.offset = 0;
1154 	args->range.length = NFS4_MAX_UINT64;
1155 	args->layout = lo;
1156 	nfs4_stateid_copy(&args->stateid, stateid);
1157 }
1158 
1159 static int
1160 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
1161 		       enum pnfs_iomode iomode, bool sync)
1162 {
1163 	struct inode *ino = lo->plh_inode;
1164 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1165 	struct nfs4_layoutreturn *lrp;
1166 	int status = 0;
1167 
1168 	lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
1169 	if (unlikely(lrp == NULL)) {
1170 		status = -ENOMEM;
1171 		spin_lock(&ino->i_lock);
1172 		pnfs_clear_layoutreturn_waitbit(lo);
1173 		spin_unlock(&ino->i_lock);
1174 		pnfs_put_layout_hdr(lo);
1175 		goto out;
1176 	}
1177 
1178 	pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
1179 	lrp->args.ld_private = &lrp->ld_private;
1180 	lrp->clp = NFS_SERVER(ino)->nfs_client;
1181 	lrp->cred = lo->plh_lc_cred;
1182 	if (ld->prepare_layoutreturn)
1183 		ld->prepare_layoutreturn(&lrp->args);
1184 
1185 	status = nfs4_proc_layoutreturn(lrp, sync);
1186 out:
1187 	dprintk("<-- %s status: %d\n", __func__, status);
1188 	return status;
1189 }
1190 
1191 /* Return true if layoutreturn is needed */
1192 static bool
1193 pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1194 {
1195 	struct pnfs_layout_segment *s;
1196 	enum pnfs_iomode iomode;
1197 	u32 seq;
1198 
1199 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1200 		return false;
1201 
1202 	seq = lo->plh_return_seq;
1203 	iomode = lo->plh_return_iomode;
1204 
1205 	/* Defer layoutreturn until all recalled lsegs are done */
1206 	list_for_each_entry(s, &lo->plh_segs, pls_list) {
1207 		if (seq && pnfs_seqid_is_newer(s->pls_seq, seq))
1208 			continue;
1209 		if (iomode != IOMODE_ANY && s->pls_range.iomode != iomode)
1210 			continue;
1211 		if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1212 			return false;
1213 	}
1214 
1215 	return true;
1216 }
1217 
1218 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1219 {
1220 	struct inode *inode= lo->plh_inode;
1221 
1222 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1223 		return;
1224 	spin_lock(&inode->i_lock);
1225 	if (pnfs_layout_need_return(lo)) {
1226 		nfs4_stateid stateid;
1227 		enum pnfs_iomode iomode;
1228 		bool send;
1229 
1230 		send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1231 		spin_unlock(&inode->i_lock);
1232 		if (send) {
1233 			/* Send an async layoutreturn so we dont deadlock */
1234 			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1235 		}
1236 	} else
1237 		spin_unlock(&inode->i_lock);
1238 }
1239 
1240 /*
1241  * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1242  * when the layout segment list is empty.
1243  *
1244  * Note that a pnfs_layout_hdr can exist with an empty layout segment
1245  * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1246  * deviceid is marked invalid.
1247  */
1248 int
1249 _pnfs_return_layout(struct inode *ino)
1250 {
1251 	struct pnfs_layout_hdr *lo = NULL;
1252 	struct nfs_inode *nfsi = NFS_I(ino);
1253 	LIST_HEAD(tmp_list);
1254 	nfs4_stateid stateid;
1255 	int status = 0;
1256 	bool send, valid_layout;
1257 
1258 	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
1259 
1260 	spin_lock(&ino->i_lock);
1261 	lo = nfsi->layout;
1262 	if (!lo) {
1263 		spin_unlock(&ino->i_lock);
1264 		dprintk("NFS: %s no layout to return\n", __func__);
1265 		goto out;
1266 	}
1267 	/* Reference matched in nfs4_layoutreturn_release */
1268 	pnfs_get_layout_hdr(lo);
1269 	/* Is there an outstanding layoutreturn ? */
1270 	if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1271 		spin_unlock(&ino->i_lock);
1272 		if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1273 					TASK_UNINTERRUPTIBLE))
1274 			goto out_put_layout_hdr;
1275 		spin_lock(&ino->i_lock);
1276 	}
1277 	valid_layout = pnfs_layout_is_valid(lo);
1278 	pnfs_clear_layoutcommit(ino, &tmp_list);
1279 	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
1280 
1281 	if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1282 		struct pnfs_layout_range range = {
1283 			.iomode		= IOMODE_ANY,
1284 			.offset		= 0,
1285 			.length		= NFS4_MAX_UINT64,
1286 		};
1287 		NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1288 	}
1289 
1290 	/* Don't send a LAYOUTRETURN if list was initially empty */
1291 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1292 			!valid_layout) {
1293 		spin_unlock(&ino->i_lock);
1294 		dprintk("NFS: %s no layout segments to return\n", __func__);
1295 		goto out_put_layout_hdr;
1296 	}
1297 
1298 	send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
1299 	spin_unlock(&ino->i_lock);
1300 	if (send)
1301 		status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1302 out_put_layout_hdr:
1303 	pnfs_free_lseg_list(&tmp_list);
1304 	pnfs_put_layout_hdr(lo);
1305 out:
1306 	dprintk("<-- %s status: %d\n", __func__, status);
1307 	return status;
1308 }
1309 
1310 int
1311 pnfs_commit_and_return_layout(struct inode *inode)
1312 {
1313 	struct pnfs_layout_hdr *lo;
1314 	int ret;
1315 
1316 	spin_lock(&inode->i_lock);
1317 	lo = NFS_I(inode)->layout;
1318 	if (lo == NULL) {
1319 		spin_unlock(&inode->i_lock);
1320 		return 0;
1321 	}
1322 	pnfs_get_layout_hdr(lo);
1323 	/* Block new layoutgets and read/write to ds */
1324 	lo->plh_block_lgets++;
1325 	spin_unlock(&inode->i_lock);
1326 	filemap_fdatawait(inode->i_mapping);
1327 	ret = pnfs_layoutcommit_inode(inode, true);
1328 	if (ret == 0)
1329 		ret = _pnfs_return_layout(inode);
1330 	spin_lock(&inode->i_lock);
1331 	lo->plh_block_lgets--;
1332 	spin_unlock(&inode->i_lock);
1333 	pnfs_put_layout_hdr(lo);
1334 	return ret;
1335 }
1336 
1337 bool pnfs_roc(struct inode *ino,
1338 		struct nfs4_layoutreturn_args *args,
1339 		struct nfs4_layoutreturn_res *res,
1340 		const struct cred *cred)
1341 {
1342 	struct nfs_inode *nfsi = NFS_I(ino);
1343 	struct nfs_open_context *ctx;
1344 	struct nfs4_state *state;
1345 	struct pnfs_layout_hdr *lo;
1346 	struct pnfs_layout_segment *lseg, *next;
1347 	nfs4_stateid stateid;
1348 	enum pnfs_iomode iomode = 0;
1349 	bool layoutreturn = false, roc = false;
1350 	bool skip_read = false;
1351 
1352 	if (!nfs_have_layout(ino))
1353 		return false;
1354 retry:
1355 	rcu_read_lock();
1356 	spin_lock(&ino->i_lock);
1357 	lo = nfsi->layout;
1358 	if (!lo || !pnfs_layout_is_valid(lo) ||
1359 	    test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1360 		lo = NULL;
1361 		goto out_noroc;
1362 	}
1363 	pnfs_get_layout_hdr(lo);
1364 	if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1365 		spin_unlock(&ino->i_lock);
1366 		rcu_read_unlock();
1367 		wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1368 				TASK_UNINTERRUPTIBLE);
1369 		pnfs_put_layout_hdr(lo);
1370 		goto retry;
1371 	}
1372 
1373 	/* no roc if we hold a delegation */
1374 	if (nfs4_check_delegation(ino, FMODE_READ)) {
1375 		if (nfs4_check_delegation(ino, FMODE_WRITE))
1376 			goto out_noroc;
1377 		skip_read = true;
1378 	}
1379 
1380 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1381 		state = ctx->state;
1382 		if (state == NULL)
1383 			continue;
1384 		/* Don't return layout if there is open file state */
1385 		if (state->state & FMODE_WRITE)
1386 			goto out_noroc;
1387 		if (state->state & FMODE_READ)
1388 			skip_read = true;
1389 	}
1390 
1391 
1392 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
1393 		if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1394 			continue;
1395 		/* If we are sending layoutreturn, invalidate all valid lsegs */
1396 		if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1397 			continue;
1398 		/*
1399 		 * Note: mark lseg for return so pnfs_layout_remove_lseg
1400 		 * doesn't invalidate the layout for us.
1401 		 */
1402 		set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1403 		if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1404 			continue;
1405 		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
1406 	}
1407 
1408 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1409 		goto out_noroc;
1410 
1411 	/* ROC in two conditions:
1412 	 * 1. there are ROC lsegs
1413 	 * 2. we don't send layoutreturn
1414 	 */
1415 	/* lo ref dropped in pnfs_roc_release() */
1416 	layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1417 	/* If the creds don't match, we can't compound the layoutreturn */
1418 	if (!layoutreturn || cred != lo->plh_lc_cred)
1419 		goto out_noroc;
1420 
1421 	roc = layoutreturn;
1422 	pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1423 	res->lrs_present = 0;
1424 	layoutreturn = false;
1425 
1426 out_noroc:
1427 	spin_unlock(&ino->i_lock);
1428 	rcu_read_unlock();
1429 	pnfs_layoutcommit_inode(ino, true);
1430 	if (roc) {
1431 		struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1432 		if (ld->prepare_layoutreturn)
1433 			ld->prepare_layoutreturn(args);
1434 		pnfs_put_layout_hdr(lo);
1435 		return true;
1436 	}
1437 	if (layoutreturn)
1438 		pnfs_send_layoutreturn(lo, &stateid, iomode, true);
1439 	pnfs_put_layout_hdr(lo);
1440 	return false;
1441 }
1442 
1443 void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1444 		struct nfs4_layoutreturn_res *res,
1445 		int ret)
1446 {
1447 	struct pnfs_layout_hdr *lo = args->layout;
1448 	const nfs4_stateid *arg_stateid = NULL;
1449 	const nfs4_stateid *res_stateid = NULL;
1450 	struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
1451 
1452 	if (ret == 0) {
1453 		arg_stateid = &args->stateid;
1454 		if (res->lrs_present)
1455 			res_stateid = &res->stateid;
1456 	}
1457 	pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1458 			res_stateid);
1459 	if (ld_private && ld_private->ops && ld_private->ops->free)
1460 		ld_private->ops->free(ld_private);
1461 	pnfs_put_layout_hdr(lo);
1462 	trace_nfs4_layoutreturn_on_close(args->inode, 0);
1463 }
1464 
1465 bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1466 {
1467 	struct nfs_inode *nfsi = NFS_I(ino);
1468         struct pnfs_layout_hdr *lo;
1469         bool sleep = false;
1470 
1471 	/* we might not have grabbed lo reference. so need to check under
1472 	 * i_lock */
1473         spin_lock(&ino->i_lock);
1474         lo = nfsi->layout;
1475         if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1476                 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1477                 sleep = true;
1478 	}
1479         spin_unlock(&ino->i_lock);
1480         return sleep;
1481 }
1482 
1483 /*
1484  * Compare two layout segments for sorting into layout cache.
1485  * We want to preferentially return RW over RO layouts, so ensure those
1486  * are seen first.
1487  */
1488 static s64
1489 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1490 	   const struct pnfs_layout_range *l2)
1491 {
1492 	s64 d;
1493 
1494 	/* high offset > low offset */
1495 	d = l1->offset - l2->offset;
1496 	if (d)
1497 		return d;
1498 
1499 	/* short length > long length */
1500 	d = l2->length - l1->length;
1501 	if (d)
1502 		return d;
1503 
1504 	/* read > read/write */
1505 	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1506 }
1507 
1508 static bool
1509 pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1510 		const struct pnfs_layout_range *l2)
1511 {
1512 	return pnfs_lseg_range_cmp(l1, l2) > 0;
1513 }
1514 
1515 static bool
1516 pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1517 		struct pnfs_layout_segment *old)
1518 {
1519 	return false;
1520 }
1521 
1522 void
1523 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1524 		   struct pnfs_layout_segment *lseg,
1525 		   bool (*is_after)(const struct pnfs_layout_range *,
1526 			   const struct pnfs_layout_range *),
1527 		   bool (*do_merge)(struct pnfs_layout_segment *,
1528 			   struct pnfs_layout_segment *),
1529 		   struct list_head *free_me)
1530 {
1531 	struct pnfs_layout_segment *lp, *tmp;
1532 
1533 	dprintk("%s:Begin\n", __func__);
1534 
1535 	list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1536 		if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1537 			continue;
1538 		if (do_merge(lseg, lp)) {
1539 			mark_lseg_invalid(lp, free_me);
1540 			continue;
1541 		}
1542 		if (is_after(&lseg->pls_range, &lp->pls_range))
1543 			continue;
1544 		list_add_tail(&lseg->pls_list, &lp->pls_list);
1545 		dprintk("%s: inserted lseg %p "
1546 			"iomode %d offset %llu length %llu before "
1547 			"lp %p iomode %d offset %llu length %llu\n",
1548 			__func__, lseg, lseg->pls_range.iomode,
1549 			lseg->pls_range.offset, lseg->pls_range.length,
1550 			lp, lp->pls_range.iomode, lp->pls_range.offset,
1551 			lp->pls_range.length);
1552 		goto out;
1553 	}
1554 	list_add_tail(&lseg->pls_list, &lo->plh_segs);
1555 	dprintk("%s: inserted lseg %p "
1556 		"iomode %d offset %llu length %llu at tail\n",
1557 		__func__, lseg, lseg->pls_range.iomode,
1558 		lseg->pls_range.offset, lseg->pls_range.length);
1559 out:
1560 	pnfs_get_layout_hdr(lo);
1561 
1562 	dprintk("%s:Return\n", __func__);
1563 }
1564 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1565 
1566 static void
1567 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1568 		   struct pnfs_layout_segment *lseg,
1569 		   struct list_head *free_me)
1570 {
1571 	struct inode *inode = lo->plh_inode;
1572 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1573 
1574 	if (ld->add_lseg != NULL)
1575 		ld->add_lseg(lo, lseg, free_me);
1576 	else
1577 		pnfs_generic_layout_insert_lseg(lo, lseg,
1578 				pnfs_lseg_range_is_after,
1579 				pnfs_lseg_no_merge,
1580 				free_me);
1581 }
1582 
1583 static struct pnfs_layout_hdr *
1584 alloc_init_layout_hdr(struct inode *ino,
1585 		      struct nfs_open_context *ctx,
1586 		      gfp_t gfp_flags)
1587 {
1588 	struct pnfs_layout_hdr *lo;
1589 
1590 	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1591 	if (!lo)
1592 		return NULL;
1593 	refcount_set(&lo->plh_refcount, 1);
1594 	INIT_LIST_HEAD(&lo->plh_layouts);
1595 	INIT_LIST_HEAD(&lo->plh_segs);
1596 	INIT_LIST_HEAD(&lo->plh_return_segs);
1597 	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1598 	lo->plh_inode = ino;
1599 	lo->plh_lc_cred = get_cred(ctx->cred);
1600 	lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
1601 	return lo;
1602 }
1603 
1604 static struct pnfs_layout_hdr *
1605 pnfs_find_alloc_layout(struct inode *ino,
1606 		       struct nfs_open_context *ctx,
1607 		       gfp_t gfp_flags)
1608 	__releases(&ino->i_lock)
1609 	__acquires(&ino->i_lock)
1610 {
1611 	struct nfs_inode *nfsi = NFS_I(ino);
1612 	struct pnfs_layout_hdr *new = NULL;
1613 
1614 	dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1615 
1616 	if (nfsi->layout != NULL)
1617 		goto out_existing;
1618 	spin_unlock(&ino->i_lock);
1619 	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1620 	spin_lock(&ino->i_lock);
1621 
1622 	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1623 		nfsi->layout = new;
1624 		return new;
1625 	} else if (new != NULL)
1626 		pnfs_free_layout_hdr(new);
1627 out_existing:
1628 	pnfs_get_layout_hdr(nfsi->layout);
1629 	return nfsi->layout;
1630 }
1631 
1632 /*
1633  * iomode matching rules:
1634  * iomode	lseg	strict match
1635  *                      iomode
1636  * -----	-----	------ -----
1637  * ANY		READ	N/A    true
1638  * ANY		RW	N/A    true
1639  * RW		READ	N/A    false
1640  * RW		RW	N/A    true
1641  * READ		READ	N/A    true
1642  * READ		RW	true   false
1643  * READ		RW	false  true
1644  */
1645 static bool
1646 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1647 		 const struct pnfs_layout_range *range,
1648 		 bool strict_iomode)
1649 {
1650 	struct pnfs_layout_range range1;
1651 
1652 	if ((range->iomode == IOMODE_RW &&
1653 	     ls_range->iomode != IOMODE_RW) ||
1654 	    (range->iomode != ls_range->iomode &&
1655 	     strict_iomode) ||
1656 	    !pnfs_lseg_range_intersecting(ls_range, range))
1657 		return false;
1658 
1659 	/* range1 covers only the first byte in the range */
1660 	range1 = *range;
1661 	range1.length = 1;
1662 	return pnfs_lseg_range_contained(ls_range, &range1);
1663 }
1664 
1665 /*
1666  * lookup range in layout
1667  */
1668 static struct pnfs_layout_segment *
1669 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1670 		struct pnfs_layout_range *range,
1671 		bool strict_iomode)
1672 {
1673 	struct pnfs_layout_segment *lseg, *ret = NULL;
1674 
1675 	dprintk("%s:Begin\n", __func__);
1676 
1677 	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1678 		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1679 		    pnfs_lseg_range_match(&lseg->pls_range, range,
1680 					  strict_iomode)) {
1681 			ret = pnfs_get_lseg(lseg);
1682 			break;
1683 		}
1684 	}
1685 
1686 	dprintk("%s:Return lseg %p ref %d\n",
1687 		__func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
1688 	return ret;
1689 }
1690 
1691 /*
1692  * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1693  * to the MDS or over pNFS
1694  *
1695  * The nfs_inode read_io and write_io fields are cumulative counters reset
1696  * when there are no layout segments. Note that in pnfs_update_layout iomode
1697  * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1698  * WRITE request.
1699  *
1700  * A return of true means use MDS I/O.
1701  *
1702  * From rfc 5661:
1703  * If a file's size is smaller than the file size threshold, data accesses
1704  * SHOULD be sent to the metadata server.  If an I/O request has a length that
1705  * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1706  * server.  If both file size and I/O size are provided, the client SHOULD
1707  * reach or exceed  both thresholds before sending its read or write
1708  * requests to the data server.
1709  */
1710 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1711 				     struct inode *ino, int iomode)
1712 {
1713 	struct nfs4_threshold *t = ctx->mdsthreshold;
1714 	struct nfs_inode *nfsi = NFS_I(ino);
1715 	loff_t fsize = i_size_read(ino);
1716 	bool size = false, size_set = false, io = false, io_set = false, ret = false;
1717 
1718 	if (t == NULL)
1719 		return ret;
1720 
1721 	dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1722 		__func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1723 
1724 	switch (iomode) {
1725 	case IOMODE_READ:
1726 		if (t->bm & THRESHOLD_RD) {
1727 			dprintk("%s fsize %llu\n", __func__, fsize);
1728 			size_set = true;
1729 			if (fsize < t->rd_sz)
1730 				size = true;
1731 		}
1732 		if (t->bm & THRESHOLD_RD_IO) {
1733 			dprintk("%s nfsi->read_io %llu\n", __func__,
1734 				nfsi->read_io);
1735 			io_set = true;
1736 			if (nfsi->read_io < t->rd_io_sz)
1737 				io = true;
1738 		}
1739 		break;
1740 	case IOMODE_RW:
1741 		if (t->bm & THRESHOLD_WR) {
1742 			dprintk("%s fsize %llu\n", __func__, fsize);
1743 			size_set = true;
1744 			if (fsize < t->wr_sz)
1745 				size = true;
1746 		}
1747 		if (t->bm & THRESHOLD_WR_IO) {
1748 			dprintk("%s nfsi->write_io %llu\n", __func__,
1749 				nfsi->write_io);
1750 			io_set = true;
1751 			if (nfsi->write_io < t->wr_io_sz)
1752 				io = true;
1753 		}
1754 		break;
1755 	}
1756 	if (size_set && io_set) {
1757 		if (size && io)
1758 			ret = true;
1759 	} else if (size || io)
1760 		ret = true;
1761 
1762 	dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1763 	return ret;
1764 }
1765 
1766 static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1767 {
1768 	/*
1769 	 * send layoutcommit as it can hold up layoutreturn due to lseg
1770 	 * reference
1771 	 */
1772 	pnfs_layoutcommit_inode(lo->plh_inode, false);
1773 	return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1774 				   nfs_wait_bit_killable,
1775 				   TASK_KILLABLE);
1776 }
1777 
1778 static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1779 {
1780 	atomic_inc(&lo->plh_outstanding);
1781 }
1782 
1783 static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1784 {
1785 	if (atomic_dec_and_test(&lo->plh_outstanding))
1786 		wake_up_var(&lo->plh_outstanding);
1787 }
1788 
1789 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1790 {
1791 	unsigned long *bitlock = &lo->plh_flags;
1792 
1793 	clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1794 	smp_mb__after_atomic();
1795 	wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1796 }
1797 
1798 static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1799 				struct nfs_server *server)
1800 {
1801 	if (list_empty(&lo->plh_layouts)) {
1802 		struct nfs_client *clp = server->nfs_client;
1803 
1804 		/* The lo must be on the clp list if there is any
1805 		 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1806 		 */
1807 		spin_lock(&clp->cl_lock);
1808 		if (list_empty(&lo->plh_layouts))
1809 			list_add_tail(&lo->plh_layouts, &server->layouts);
1810 		spin_unlock(&clp->cl_lock);
1811 	}
1812 }
1813 
1814 /*
1815  * Layout segment is retreived from the server if not cached.
1816  * The appropriate layout segment is referenced and returned to the caller.
1817  */
1818 struct pnfs_layout_segment *
1819 pnfs_update_layout(struct inode *ino,
1820 		   struct nfs_open_context *ctx,
1821 		   loff_t pos,
1822 		   u64 count,
1823 		   enum pnfs_iomode iomode,
1824 		   bool strict_iomode,
1825 		   gfp_t gfp_flags)
1826 {
1827 	struct pnfs_layout_range arg = {
1828 		.iomode = iomode,
1829 		.offset = pos,
1830 		.length = count,
1831 	};
1832 	unsigned pg_offset;
1833 	struct nfs_server *server = NFS_SERVER(ino);
1834 	struct nfs_client *clp = server->nfs_client;
1835 	struct pnfs_layout_hdr *lo = NULL;
1836 	struct pnfs_layout_segment *lseg = NULL;
1837 	struct nfs4_layoutget *lgp;
1838 	nfs4_stateid stateid;
1839 	long timeout = 0;
1840 	unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
1841 	bool first;
1842 
1843 	if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1844 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1845 				 PNFS_UPDATE_LAYOUT_NO_PNFS);
1846 		goto out;
1847 	}
1848 
1849 	if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1850 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1851 				 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1852 		goto out;
1853 	}
1854 
1855 lookup_again:
1856 	lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1857 	if (IS_ERR(lseg))
1858 		goto out;
1859 	first = false;
1860 	spin_lock(&ino->i_lock);
1861 	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1862 	if (lo == NULL) {
1863 		spin_unlock(&ino->i_lock);
1864 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1865 				 PNFS_UPDATE_LAYOUT_NOMEM);
1866 		goto out;
1867 	}
1868 
1869 	/* Do we even need to bother with this? */
1870 	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1871 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1872 				 PNFS_UPDATE_LAYOUT_BULK_RECALL);
1873 		dprintk("%s matches recall, use MDS\n", __func__);
1874 		goto out_unlock;
1875 	}
1876 
1877 	/* if LAYOUTGET already failed once we don't try again */
1878 	if (pnfs_layout_io_test_failed(lo, iomode)) {
1879 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1880 				 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1881 		goto out_unlock;
1882 	}
1883 
1884 	/*
1885 	 * If the layout segment list is empty, but there are outstanding
1886 	 * layoutget calls, then they might be subject to a layoutrecall.
1887 	 */
1888 	if (list_empty(&lo->plh_segs) &&
1889 	    atomic_read(&lo->plh_outstanding) != 0) {
1890 		spin_unlock(&ino->i_lock);
1891 		lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
1892 					!atomic_read(&lo->plh_outstanding)));
1893 		if (IS_ERR(lseg) || !list_empty(&lo->plh_segs))
1894 			goto out_put_layout_hdr;
1895 		pnfs_put_layout_hdr(lo);
1896 		goto lookup_again;
1897 	}
1898 
1899 	lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
1900 	if (lseg) {
1901 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1902 				PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1903 		goto out_unlock;
1904 	}
1905 
1906 	if (!nfs4_valid_open_stateid(ctx->state)) {
1907 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1908 				PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1909 		goto out_unlock;
1910 	}
1911 
1912 	/*
1913 	 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1914 	 * stateid, or it has been invalidated, then we must use the open
1915 	 * stateid.
1916 	 */
1917 	if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
1918 
1919 		/*
1920 		 * The first layoutget for the file. Need to serialize per
1921 		 * RFC 5661 Errata 3208.
1922 		 */
1923 		if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1924 				     &lo->plh_flags)) {
1925 			spin_unlock(&ino->i_lock);
1926 			lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
1927 						NFS_LAYOUT_FIRST_LAYOUTGET,
1928 						TASK_KILLABLE));
1929 			if (IS_ERR(lseg))
1930 				goto out_put_layout_hdr;
1931 			pnfs_put_layout_hdr(lo);
1932 			dprintk("%s retrying\n", __func__);
1933 			goto lookup_again;
1934 		}
1935 
1936 		first = true;
1937 		if (nfs4_select_rw_stateid(ctx->state,
1938 					iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1939 					NULL, &stateid, NULL) != 0) {
1940 			trace_pnfs_update_layout(ino, pos, count,
1941 					iomode, lo, lseg,
1942 					PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1943 			goto out_unlock;
1944 		}
1945 	} else {
1946 		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1947 	}
1948 
1949 	/*
1950 	 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1951 	 * for LAYOUTRETURN even if first is true.
1952 	 */
1953 	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1954 		spin_unlock(&ino->i_lock);
1955 		dprintk("%s wait for layoutreturn\n", __func__);
1956 		lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
1957 		if (!IS_ERR(lseg)) {
1958 			if (first)
1959 				pnfs_clear_first_layoutget(lo);
1960 			pnfs_put_layout_hdr(lo);
1961 			dprintk("%s retrying\n", __func__);
1962 			trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1963 					lseg, PNFS_UPDATE_LAYOUT_RETRY);
1964 			goto lookup_again;
1965 		}
1966 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1967 				PNFS_UPDATE_LAYOUT_RETURN);
1968 		goto out_put_layout_hdr;
1969 	}
1970 
1971 	if (pnfs_layoutgets_blocked(lo)) {
1972 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1973 				PNFS_UPDATE_LAYOUT_BLOCKED);
1974 		goto out_unlock;
1975 	}
1976 	nfs_layoutget_begin(lo);
1977 	spin_unlock(&ino->i_lock);
1978 
1979 	_add_to_server_list(lo, server);
1980 
1981 	pg_offset = arg.offset & ~PAGE_MASK;
1982 	if (pg_offset) {
1983 		arg.offset -= pg_offset;
1984 		arg.length += pg_offset;
1985 	}
1986 	if (arg.length != NFS4_MAX_UINT64)
1987 		arg.length = PAGE_ALIGN(arg.length);
1988 
1989 	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
1990 	if (!lgp) {
1991 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
1992 					 PNFS_UPDATE_LAYOUT_NOMEM);
1993 		nfs_layoutget_end(lo);
1994 		goto out_put_layout_hdr;
1995 	}
1996 
1997 	lseg = nfs4_proc_layoutget(lgp, &timeout);
1998 	trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1999 				 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
2000 	nfs_layoutget_end(lo);
2001 	if (IS_ERR(lseg)) {
2002 		switch(PTR_ERR(lseg)) {
2003 		case -EBUSY:
2004 			if (time_after(jiffies, giveup))
2005 				lseg = NULL;
2006 			break;
2007 		case -ERECALLCONFLICT:
2008 		case -EAGAIN:
2009 			break;
2010 		default:
2011 			if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2012 				pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2013 				lseg = NULL;
2014 			}
2015 			goto out_put_layout_hdr;
2016 		}
2017 		if (lseg) {
2018 			if (first)
2019 				pnfs_clear_first_layoutget(lo);
2020 			trace_pnfs_update_layout(ino, pos, count,
2021 				iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2022 			pnfs_put_layout_hdr(lo);
2023 			goto lookup_again;
2024 		}
2025 	} else {
2026 		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2027 	}
2028 
2029 out_put_layout_hdr:
2030 	if (first)
2031 		pnfs_clear_first_layoutget(lo);
2032 	pnfs_put_layout_hdr(lo);
2033 out:
2034 	dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2035 			"(%s, offset: %llu, length: %llu)\n",
2036 			__func__, ino->i_sb->s_id,
2037 			(unsigned long long)NFS_FILEID(ino),
2038 			IS_ERR_OR_NULL(lseg) ? "not found" : "found",
2039 			iomode==IOMODE_RW ?  "read/write" : "read-only",
2040 			(unsigned long long)pos,
2041 			(unsigned long long)count);
2042 	return lseg;
2043 out_unlock:
2044 	spin_unlock(&ino->i_lock);
2045 	goto out_put_layout_hdr;
2046 }
2047 EXPORT_SYMBOL_GPL(pnfs_update_layout);
2048 
2049 static bool
2050 pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2051 {
2052 	switch (range->iomode) {
2053 	case IOMODE_READ:
2054 	case IOMODE_RW:
2055 		break;
2056 	default:
2057 		return false;
2058 	}
2059 	if (range->offset == NFS4_MAX_UINT64)
2060 		return false;
2061 	if (range->length == 0)
2062 		return false;
2063 	if (range->length != NFS4_MAX_UINT64 &&
2064 	    range->length > NFS4_MAX_UINT64 - range->offset)
2065 		return false;
2066 	return true;
2067 }
2068 
2069 static struct pnfs_layout_hdr *
2070 _pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2071 {
2072 	struct pnfs_layout_hdr *lo;
2073 
2074 	spin_lock(&ino->i_lock);
2075 	lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2076 	if (!lo)
2077 		goto out_unlock;
2078 	if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2079 		goto out_unlock;
2080 	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2081 		goto out_unlock;
2082 	if (pnfs_layoutgets_blocked(lo))
2083 		goto out_unlock;
2084 	if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2085 		goto out_unlock;
2086 	nfs_layoutget_begin(lo);
2087 	spin_unlock(&ino->i_lock);
2088 	_add_to_server_list(lo, NFS_SERVER(ino));
2089 	return lo;
2090 
2091 out_unlock:
2092 	spin_unlock(&ino->i_lock);
2093 	pnfs_put_layout_hdr(lo);
2094 	return NULL;
2095 }
2096 
2097 extern const nfs4_stateid current_stateid;
2098 
2099 static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2100 				     struct nfs_open_context *ctx)
2101 {
2102 	struct inode *ino = data->dentry->d_inode;
2103 	struct pnfs_layout_range rng = {
2104 		.iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2105 			  IOMODE_RW: IOMODE_READ,
2106 		.offset = 0,
2107 		.length = NFS4_MAX_UINT64,
2108 	};
2109 	struct nfs4_layoutget *lgp;
2110 	struct pnfs_layout_hdr *lo;
2111 
2112 	/* Heuristic: don't send layoutget if we have cached data */
2113 	if (rng.iomode == IOMODE_READ &&
2114 	   (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2115 		return;
2116 
2117 	lo = _pnfs_grab_empty_layout(ino, ctx);
2118 	if (!lo)
2119 		return;
2120 	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2121 					     &rng, GFP_KERNEL);
2122 	if (!lgp) {
2123 		pnfs_clear_first_layoutget(lo);
2124 		pnfs_put_layout_hdr(lo);
2125 		return;
2126 	}
2127 	data->lgp = lgp;
2128 	data->o_arg.lg_args = &lgp->args;
2129 	data->o_res.lg_res = &lgp->res;
2130 }
2131 
2132 static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2133 				     struct nfs_open_context *ctx)
2134 {
2135 	struct pnfs_layout_range rng = {
2136 		.iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2137 			  IOMODE_RW: IOMODE_READ,
2138 		.offset = 0,
2139 		.length = NFS4_MAX_UINT64,
2140 	};
2141 	struct nfs4_layoutget *lgp;
2142 
2143 	lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2144 					     &rng, GFP_KERNEL);
2145 	if (!lgp)
2146 		return;
2147 	data->lgp = lgp;
2148 	data->o_arg.lg_args = &lgp->args;
2149 	data->o_res.lg_res = &lgp->res;
2150 }
2151 
2152 void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2153 			 struct nfs_open_context *ctx)
2154 {
2155 	struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2156 
2157 	if (!(pnfs_enabled_sb(server) &&
2158 	      server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2159 		return;
2160 	/* Could check on max_ops, but currently hardcoded high enough */
2161 	if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2162 		return;
2163 	if (data->state)
2164 		_lgopen_prepare_attached(data, ctx);
2165 	else
2166 		_lgopen_prepare_floating(data, ctx);
2167 }
2168 
2169 void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2170 		       struct nfs_open_context *ctx)
2171 {
2172 	struct pnfs_layout_hdr *lo;
2173 	struct pnfs_layout_segment *lseg;
2174 	struct nfs_server *srv = NFS_SERVER(ino);
2175 	u32 iomode;
2176 
2177 	if (!lgp)
2178 		return;
2179 	dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2180 	if (lgp->res.status) {
2181 		switch (lgp->res.status) {
2182 		default:
2183 			break;
2184 		/*
2185 		 * Halt lgopen attempts if the server doesn't recognise
2186 		 * the "current stateid" value, the layout type, or the
2187 		 * layoutget operation as being valid.
2188 		 * Also if it complains about too many ops in the compound
2189 		 * or of the request/reply being too big.
2190 		 */
2191 		case -NFS4ERR_BAD_STATEID:
2192 		case -NFS4ERR_NOTSUPP:
2193 		case -NFS4ERR_REP_TOO_BIG:
2194 		case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2195 		case -NFS4ERR_REQ_TOO_BIG:
2196 		case -NFS4ERR_TOO_MANY_OPS:
2197 		case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
2198 			srv->caps &= ~NFS_CAP_LGOPEN;
2199 		}
2200 		return;
2201 	}
2202 	if (!lgp->args.inode) {
2203 		lo = _pnfs_grab_empty_layout(ino, ctx);
2204 		if (!lo)
2205 			return;
2206 		lgp->args.inode = ino;
2207 	} else
2208 		lo = NFS_I(lgp->args.inode)->layout;
2209 
2210 	lseg = pnfs_layout_process(lgp);
2211 	if (!IS_ERR(lseg)) {
2212 		iomode = lgp->args.range.iomode;
2213 		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2214 		pnfs_put_lseg(lseg);
2215 	}
2216 }
2217 
2218 void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2219 {
2220 	if (lgp != NULL) {
2221 		struct inode *inode = lgp->args.inode;
2222 		if (inode) {
2223 			struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2224 			pnfs_clear_first_layoutget(lo);
2225 			nfs_layoutget_end(lo);
2226 		}
2227 		pnfs_layoutget_free(lgp);
2228 	}
2229 }
2230 
2231 struct pnfs_layout_segment *
2232 pnfs_layout_process(struct nfs4_layoutget *lgp)
2233 {
2234 	struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2235 	struct nfs4_layoutget_res *res = &lgp->res;
2236 	struct pnfs_layout_segment *lseg;
2237 	struct inode *ino = lo->plh_inode;
2238 	LIST_HEAD(free_me);
2239 
2240 	if (!pnfs_sanity_check_layout_range(&res->range))
2241 		return ERR_PTR(-EINVAL);
2242 
2243 	/* Inject layout blob into I/O device driver */
2244 	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
2245 	if (IS_ERR_OR_NULL(lseg)) {
2246 		if (!lseg)
2247 			lseg = ERR_PTR(-ENOMEM);
2248 
2249 		dprintk("%s: Could not allocate layout: error %ld\n",
2250 		       __func__, PTR_ERR(lseg));
2251 		return lseg;
2252 	}
2253 
2254 	pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
2255 
2256 	spin_lock(&ino->i_lock);
2257 	if (pnfs_layoutgets_blocked(lo)) {
2258 		dprintk("%s forget reply due to state\n", __func__);
2259 		goto out_forget;
2260 	}
2261 
2262 	if (!pnfs_layout_is_valid(lo)) {
2263 		/* We have a completely new layout */
2264 		pnfs_set_layout_stateid(lo, &res->stateid, true);
2265 	} else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
2266 		/* existing state ID, make sure the sequence number matches. */
2267 		if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2268 			dprintk("%s forget reply due to sequence\n", __func__);
2269 			goto out_forget;
2270 		}
2271 		pnfs_set_layout_stateid(lo, &res->stateid, false);
2272 	} else {
2273 		/*
2274 		 * We got an entirely new state ID.  Mark all segments for the
2275 		 * inode invalid, and retry the layoutget
2276 		 */
2277 		pnfs_mark_layout_stateid_invalid(lo, &free_me);
2278 		goto out_forget;
2279 	}
2280 
2281 	pnfs_get_lseg(lseg);
2282 	pnfs_layout_insert_lseg(lo, lseg, &free_me);
2283 
2284 
2285 	if (res->return_on_close)
2286 		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
2287 
2288 	spin_unlock(&ino->i_lock);
2289 	pnfs_free_lseg_list(&free_me);
2290 	return lseg;
2291 
2292 out_forget:
2293 	spin_unlock(&ino->i_lock);
2294 	lseg->pls_layout = lo;
2295 	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
2296 	return ERR_PTR(-EAGAIN);
2297 }
2298 
2299 static int
2300 mark_lseg_invalid_or_return(struct pnfs_layout_segment *lseg,
2301 		struct list_head *tmp_list)
2302 {
2303 	if (!mark_lseg_invalid(lseg, tmp_list))
2304 		return 0;
2305 	pnfs_cache_lseg_for_layoutreturn(lseg->pls_layout, lseg);
2306 	return 1;
2307 }
2308 
2309 /**
2310  * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2311  * @lo: pointer to layout header
2312  * @tmp_list: list header to be used with pnfs_free_lseg_list()
2313  * @return_range: describe layout segment ranges to be returned
2314  * @seq: stateid seqid to match
2315  *
2316  * This function is mainly intended for use by layoutrecall. It attempts
2317  * to free the layout segment immediately, or else to mark it for return
2318  * as soon as its reference count drops to zero.
2319  *
2320  * Returns
2321  * - 0: a layoutreturn needs to be scheduled.
2322  * - EBUSY: there are layout segment that are still in use.
2323  * - ENOENT: there are no layout segments that need to be returned.
2324  */
2325 int
2326 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2327 				struct list_head *tmp_list,
2328 				const struct pnfs_layout_range *return_range,
2329 				u32 seq)
2330 {
2331 	struct pnfs_layout_segment *lseg, *next;
2332 	int remaining = 0;
2333 
2334 	dprintk("%s:Begin lo %p\n", __func__, lo);
2335 
2336 	assert_spin_locked(&lo->plh_inode->i_lock);
2337 
2338 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
2339 		if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
2340 			dprintk("%s: marking lseg %p iomode %d "
2341 				"offset %llu length %llu\n", __func__,
2342 				lseg, lseg->pls_range.iomode,
2343 				lseg->pls_range.offset,
2344 				lseg->pls_range.length);
2345 			if (mark_lseg_invalid_or_return(lseg, tmp_list))
2346 				continue;
2347 			remaining++;
2348 			set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
2349 		}
2350 
2351 	if (remaining) {
2352 		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2353 		return -EBUSY;
2354 	}
2355 
2356 	if (!list_empty(&lo->plh_return_segs)) {
2357 		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2358 		return 0;
2359 	}
2360 
2361 	return -ENOENT;
2362 }
2363 
2364 void pnfs_error_mark_layout_for_return(struct inode *inode,
2365 				       struct pnfs_layout_segment *lseg)
2366 {
2367 	struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2368 	struct pnfs_layout_range range = {
2369 		.iomode = lseg->pls_range.iomode,
2370 		.offset = 0,
2371 		.length = NFS4_MAX_UINT64,
2372 	};
2373 	bool return_now = false;
2374 
2375 	spin_lock(&inode->i_lock);
2376 	if (!pnfs_layout_is_valid(lo)) {
2377 		spin_unlock(&inode->i_lock);
2378 		return;
2379 	}
2380 	pnfs_set_plh_return_info(lo, range.iomode, 0);
2381 	/*
2382 	 * mark all matching lsegs so that we are sure to have no live
2383 	 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2384 	 * for how it works.
2385 	 */
2386 	if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
2387 		nfs4_stateid stateid;
2388 		enum pnfs_iomode iomode;
2389 
2390 		return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
2391 		spin_unlock(&inode->i_lock);
2392 		if (return_now)
2393 			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2394 	} else {
2395 		spin_unlock(&inode->i_lock);
2396 		nfs_commit_inode(inode, 0);
2397 	}
2398 }
2399 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2400 
2401 void
2402 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2403 {
2404 	if (pgio->pg_lseg == NULL ||
2405 	    test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2406 		return;
2407 	pnfs_put_lseg(pgio->pg_lseg);
2408 	pgio->pg_lseg = NULL;
2409 }
2410 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2411 
2412 /*
2413  * Check for any intersection between the request and the pgio->pg_lseg,
2414  * and if none, put this pgio->pg_lseg away.
2415  */
2416 static void
2417 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2418 {
2419 	if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2420 		pnfs_put_lseg(pgio->pg_lseg);
2421 		pgio->pg_lseg = NULL;
2422 	}
2423 }
2424 
2425 void
2426 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2427 {
2428 	u64 rd_size = req->wb_bytes;
2429 
2430 	pnfs_generic_pg_check_layout(pgio);
2431 	pnfs_generic_pg_check_range(pgio, req);
2432 	if (pgio->pg_lseg == NULL) {
2433 		if (pgio->pg_dreq == NULL)
2434 			rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2435 		else
2436 			rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2437 
2438 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2439 						   nfs_req_openctx(req),
2440 						   req_offset(req),
2441 						   rd_size,
2442 						   IOMODE_READ,
2443 						   false,
2444 						   GFP_KERNEL);
2445 		if (IS_ERR(pgio->pg_lseg)) {
2446 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2447 			pgio->pg_lseg = NULL;
2448 			return;
2449 		}
2450 	}
2451 	/* If no lseg, fall back to read through mds */
2452 	if (pgio->pg_lseg == NULL)
2453 		nfs_pageio_reset_read_mds(pgio);
2454 
2455 }
2456 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2457 
2458 void
2459 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2460 			   struct nfs_page *req, u64 wb_size)
2461 {
2462 	pnfs_generic_pg_check_layout(pgio);
2463 	pnfs_generic_pg_check_range(pgio, req);
2464 	if (pgio->pg_lseg == NULL) {
2465 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2466 						   nfs_req_openctx(req),
2467 						   req_offset(req),
2468 						   wb_size,
2469 						   IOMODE_RW,
2470 						   false,
2471 						   GFP_NOFS);
2472 		if (IS_ERR(pgio->pg_lseg)) {
2473 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2474 			pgio->pg_lseg = NULL;
2475 			return;
2476 		}
2477 	}
2478 	/* If no lseg, fall back to write through mds */
2479 	if (pgio->pg_lseg == NULL)
2480 		nfs_pageio_reset_write_mds(pgio);
2481 }
2482 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2483 
2484 void
2485 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2486 {
2487 	if (desc->pg_lseg) {
2488 		pnfs_put_lseg(desc->pg_lseg);
2489 		desc->pg_lseg = NULL;
2490 	}
2491 }
2492 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2493 
2494 /*
2495  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2496  * of bytes (maximum @req->wb_bytes) that can be coalesced.
2497  */
2498 size_t
2499 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2500 		     struct nfs_page *prev, struct nfs_page *req)
2501 {
2502 	unsigned int size;
2503 	u64 seg_end, req_start, seg_left;
2504 
2505 	size = nfs_generic_pg_test(pgio, prev, req);
2506 	if (!size)
2507 		return 0;
2508 
2509 	/*
2510 	 * 'size' contains the number of bytes left in the current page (up
2511 	 * to the original size asked for in @req->wb_bytes).
2512 	 *
2513 	 * Calculate how many bytes are left in the layout segment
2514 	 * and if there are less bytes than 'size', return that instead.
2515 	 *
2516 	 * Please also note that 'end_offset' is actually the offset of the
2517 	 * first byte that lies outside the pnfs_layout_range. FIXME?
2518 	 *
2519 	 */
2520 	if (pgio->pg_lseg) {
2521 		seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
2522 				     pgio->pg_lseg->pls_range.length);
2523 		req_start = req_offset(req);
2524 
2525 		/* start of request is past the last byte of this segment */
2526 		if (req_start >= seg_end)
2527 			return 0;
2528 
2529 		/* adjust 'size' iff there are fewer bytes left in the
2530 		 * segment than what nfs_generic_pg_test returned */
2531 		seg_left = seg_end - req_start;
2532 		if (seg_left < size)
2533 			size = (unsigned int)seg_left;
2534 	}
2535 
2536 	return size;
2537 }
2538 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
2539 
2540 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
2541 {
2542 	struct nfs_pageio_descriptor pgio;
2543 
2544 	/* Resend all requests through the MDS */
2545 	nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2546 			      hdr->completion_ops);
2547 	set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
2548 	return nfs_pageio_resend(&pgio, hdr);
2549 }
2550 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
2551 
2552 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
2553 {
2554 
2555 	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2556 	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2557 	    PNFS_LAYOUTRET_ON_ERROR) {
2558 		pnfs_return_layout(hdr->inode);
2559 	}
2560 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2561 		hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
2562 }
2563 
2564 /*
2565  * Called by non rpc-based layout drivers
2566  */
2567 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
2568 {
2569 	if (likely(!hdr->pnfs_error)) {
2570 		pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2571 				hdr->mds_offset + hdr->res.count);
2572 		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2573 	}
2574 	trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2575 	if (unlikely(hdr->pnfs_error))
2576 		pnfs_ld_handle_write_error(hdr);
2577 	hdr->mds_ops->rpc_release(hdr);
2578 }
2579 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
2580 
2581 static void
2582 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
2583 		struct nfs_pgio_header *hdr)
2584 {
2585 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2586 
2587 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2588 		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2589 		nfs_pageio_reset_write_mds(desc);
2590 		mirror->pg_recoalesce = 1;
2591 	}
2592 	hdr->completion_ops->completion(hdr);
2593 }
2594 
2595 static enum pnfs_try_status
2596 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
2597 			const struct rpc_call_ops *call_ops,
2598 			struct pnfs_layout_segment *lseg,
2599 			int how)
2600 {
2601 	struct inode *inode = hdr->inode;
2602 	enum pnfs_try_status trypnfs;
2603 	struct nfs_server *nfss = NFS_SERVER(inode);
2604 
2605 	hdr->mds_ops = call_ops;
2606 
2607 	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
2608 		inode->i_ino, hdr->args.count, hdr->args.offset, how);
2609 	trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
2610 	if (trypnfs != PNFS_NOT_ATTEMPTED)
2611 		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
2612 	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2613 	return trypnfs;
2614 }
2615 
2616 static void
2617 pnfs_do_write(struct nfs_pageio_descriptor *desc,
2618 	      struct nfs_pgio_header *hdr, int how)
2619 {
2620 	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2621 	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2622 	enum pnfs_try_status trypnfs;
2623 
2624 	trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2625 	switch (trypnfs) {
2626 	case PNFS_NOT_ATTEMPTED:
2627 		pnfs_write_through_mds(desc, hdr);
2628 	case PNFS_ATTEMPTED:
2629 		break;
2630 	case PNFS_TRY_AGAIN:
2631 		/* cleanup hdr and prepare to redo pnfs */
2632 		if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2633 			struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2634 			list_splice_init(&hdr->pages, &mirror->pg_list);
2635 			mirror->pg_recoalesce = 1;
2636 		}
2637 		hdr->mds_ops->rpc_release(hdr);
2638 	}
2639 }
2640 
2641 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2642 {
2643 	pnfs_put_lseg(hdr->lseg);
2644 	nfs_pgio_header_free(hdr);
2645 }
2646 
2647 int
2648 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2649 {
2650 	struct nfs_pgio_header *hdr;
2651 	int ret;
2652 
2653 	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2654 	if (!hdr) {
2655 		desc->pg_error = -ENOMEM;
2656 		return desc->pg_error;
2657 	}
2658 	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2659 
2660 	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2661 	ret = nfs_generic_pgio(desc, hdr);
2662 	if (!ret)
2663 		pnfs_do_write(desc, hdr, desc->pg_ioflags);
2664 
2665 	return ret;
2666 }
2667 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2668 
2669 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2670 {
2671 	struct nfs_pageio_descriptor pgio;
2672 
2673 	/* Resend all requests through the MDS */
2674 	nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2675 	return nfs_pageio_resend(&pgio, hdr);
2676 }
2677 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2678 
2679 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2680 {
2681 	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2682 	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2683 	    PNFS_LAYOUTRET_ON_ERROR) {
2684 		pnfs_return_layout(hdr->inode);
2685 	}
2686 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2687 		hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2688 }
2689 
2690 /*
2691  * Called by non rpc-based layout drivers
2692  */
2693 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2694 {
2695 	if (likely(!hdr->pnfs_error))
2696 		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2697 	trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2698 	if (unlikely(hdr->pnfs_error))
2699 		pnfs_ld_handle_read_error(hdr);
2700 	hdr->mds_ops->rpc_release(hdr);
2701 }
2702 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2703 
2704 static void
2705 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2706 		struct nfs_pgio_header *hdr)
2707 {
2708 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2709 
2710 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2711 		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2712 		nfs_pageio_reset_read_mds(desc);
2713 		mirror->pg_recoalesce = 1;
2714 	}
2715 	hdr->completion_ops->completion(hdr);
2716 }
2717 
2718 /*
2719  * Call the appropriate parallel I/O subsystem read function.
2720  */
2721 static enum pnfs_try_status
2722 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2723 		       const struct rpc_call_ops *call_ops,
2724 		       struct pnfs_layout_segment *lseg)
2725 {
2726 	struct inode *inode = hdr->inode;
2727 	struct nfs_server *nfss = NFS_SERVER(inode);
2728 	enum pnfs_try_status trypnfs;
2729 
2730 	hdr->mds_ops = call_ops;
2731 
2732 	dprintk("%s: Reading ino:%lu %u@%llu\n",
2733 		__func__, inode->i_ino, hdr->args.count, hdr->args.offset);
2734 
2735 	trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2736 	if (trypnfs != PNFS_NOT_ATTEMPTED)
2737 		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
2738 	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2739 	return trypnfs;
2740 }
2741 
2742 /* Resend all requests through pnfs. */
2743 void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2744 {
2745 	struct nfs_pageio_descriptor pgio;
2746 
2747 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2748 		/* Prevent deadlocks with layoutreturn! */
2749 		pnfs_put_lseg(hdr->lseg);
2750 		hdr->lseg = NULL;
2751 
2752 		nfs_pageio_init_read(&pgio, hdr->inode, false,
2753 					hdr->completion_ops);
2754 		hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2755 	}
2756 }
2757 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2758 
2759 static void
2760 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2761 {
2762 	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2763 	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2764 	enum pnfs_try_status trypnfs;
2765 
2766 	trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2767 	switch (trypnfs) {
2768 	case PNFS_NOT_ATTEMPTED:
2769 		pnfs_read_through_mds(desc, hdr);
2770 	case PNFS_ATTEMPTED:
2771 		break;
2772 	case PNFS_TRY_AGAIN:
2773 		/* cleanup hdr and prepare to redo pnfs */
2774 		if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2775 			struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2776 			list_splice_init(&hdr->pages, &mirror->pg_list);
2777 			mirror->pg_recoalesce = 1;
2778 		}
2779 		hdr->mds_ops->rpc_release(hdr);
2780 	}
2781 }
2782 
2783 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2784 {
2785 	pnfs_put_lseg(hdr->lseg);
2786 	nfs_pgio_header_free(hdr);
2787 }
2788 
2789 int
2790 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2791 {
2792 	struct nfs_pgio_header *hdr;
2793 	int ret;
2794 
2795 	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2796 	if (!hdr) {
2797 		desc->pg_error = -ENOMEM;
2798 		return desc->pg_error;
2799 	}
2800 	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2801 	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2802 	ret = nfs_generic_pgio(desc, hdr);
2803 	if (!ret)
2804 		pnfs_do_read(desc, hdr);
2805 	return ret;
2806 }
2807 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2808 
2809 static void pnfs_clear_layoutcommitting(struct inode *inode)
2810 {
2811 	unsigned long *bitlock = &NFS_I(inode)->flags;
2812 
2813 	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2814 	smp_mb__after_atomic();
2815 	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2816 }
2817 
2818 /*
2819  * There can be multiple RW segments.
2820  */
2821 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
2822 {
2823 	struct pnfs_layout_segment *lseg;
2824 
2825 	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2826 		if (lseg->pls_range.iomode == IOMODE_RW &&
2827 		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2828 			list_add(&lseg->pls_lc_list, listp);
2829 	}
2830 }
2831 
2832 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2833 {
2834 	struct pnfs_layout_segment *lseg, *tmp;
2835 
2836 	/* Matched by references in pnfs_set_layoutcommit */
2837 	list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2838 		list_del_init(&lseg->pls_lc_list);
2839 		pnfs_put_lseg(lseg);
2840 	}
2841 
2842 	pnfs_clear_layoutcommitting(inode);
2843 }
2844 
2845 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2846 {
2847 	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
2848 }
2849 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2850 
2851 void
2852 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2853 		loff_t end_pos)
2854 {
2855 	struct nfs_inode *nfsi = NFS_I(inode);
2856 	bool mark_as_dirty = false;
2857 
2858 	spin_lock(&inode->i_lock);
2859 	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2860 		nfsi->layout->plh_lwb = end_pos;
2861 		mark_as_dirty = true;
2862 		dprintk("%s: Set layoutcommit for inode %lu ",
2863 			__func__, inode->i_ino);
2864 	} else if (end_pos > nfsi->layout->plh_lwb)
2865 		nfsi->layout->plh_lwb = end_pos;
2866 	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2867 		/* references matched in nfs4_layoutcommit_release */
2868 		pnfs_get_lseg(lseg);
2869 	}
2870 	spin_unlock(&inode->i_lock);
2871 	dprintk("%s: lseg %p end_pos %llu\n",
2872 		__func__, lseg, nfsi->layout->plh_lwb);
2873 
2874 	/* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2875 	 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2876 	if (mark_as_dirty)
2877 		mark_inode_dirty_sync(inode);
2878 }
2879 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2880 
2881 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2882 {
2883 	struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2884 
2885 	if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2886 		nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
2887 	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
2888 }
2889 
2890 /*
2891  * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2892  * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2893  * data to disk to allow the server to recover the data if it crashes.
2894  * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2895  * is off, and a COMMIT is sent to a data server, or
2896  * if WRITEs to a data server return NFS_DATA_SYNC.
2897  */
2898 int
2899 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
2900 {
2901 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2902 	struct nfs4_layoutcommit_data *data;
2903 	struct nfs_inode *nfsi = NFS_I(inode);
2904 	loff_t end_pos;
2905 	int status;
2906 
2907 	if (!pnfs_layoutcommit_outstanding(inode))
2908 		return 0;
2909 
2910 	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
2911 
2912 	status = -EAGAIN;
2913 	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2914 		if (!sync)
2915 			goto out;
2916 		status = wait_on_bit_lock_action(&nfsi->flags,
2917 				NFS_INO_LAYOUTCOMMITTING,
2918 				nfs_wait_bit_killable,
2919 				TASK_KILLABLE);
2920 		if (status)
2921 			goto out;
2922 	}
2923 
2924 	status = -ENOMEM;
2925 	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2926 	data = kzalloc(sizeof(*data), GFP_NOFS);
2927 	if (!data)
2928 		goto clear_layoutcommitting;
2929 
2930 	status = 0;
2931 	spin_lock(&inode->i_lock);
2932 	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2933 		goto out_unlock;
2934 
2935 	INIT_LIST_HEAD(&data->lseg_list);
2936 	pnfs_list_write_lseg(inode, &data->lseg_list);
2937 
2938 	end_pos = nfsi->layout->plh_lwb;
2939 
2940 	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
2941 	spin_unlock(&inode->i_lock);
2942 
2943 	data->args.inode = inode;
2944 	data->cred = get_cred(nfsi->layout->plh_lc_cred);
2945 	nfs_fattr_init(&data->fattr);
2946 	data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2947 	data->res.fattr = &data->fattr;
2948 	if (end_pos != 0)
2949 		data->args.lastbytewritten = end_pos - 1;
2950 	else
2951 		data->args.lastbytewritten = U64_MAX;
2952 	data->res.server = NFS_SERVER(inode);
2953 
2954 	if (ld->prepare_layoutcommit) {
2955 		status = ld->prepare_layoutcommit(&data->args);
2956 		if (status) {
2957 			put_cred(data->cred);
2958 			spin_lock(&inode->i_lock);
2959 			set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2960 			if (end_pos > nfsi->layout->plh_lwb)
2961 				nfsi->layout->plh_lwb = end_pos;
2962 			goto out_unlock;
2963 		}
2964 	}
2965 
2966 
2967 	status = nfs4_proc_layoutcommit(data, sync);
2968 out:
2969 	if (status)
2970 		mark_inode_dirty_sync(inode);
2971 	dprintk("<-- %s status %d\n", __func__, status);
2972 	return status;
2973 out_unlock:
2974 	spin_unlock(&inode->i_lock);
2975 	kfree(data);
2976 clear_layoutcommitting:
2977 	pnfs_clear_layoutcommitting(inode);
2978 	goto out;
2979 }
2980 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2981 
2982 int
2983 pnfs_generic_sync(struct inode *inode, bool datasync)
2984 {
2985 	return pnfs_layoutcommit_inode(inode, true);
2986 }
2987 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2988 
2989 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2990 {
2991 	struct nfs4_threshold *thp;
2992 
2993 	thp = kzalloc(sizeof(*thp), GFP_NOFS);
2994 	if (!thp) {
2995 		dprintk("%s mdsthreshold allocation failed\n", __func__);
2996 		return NULL;
2997 	}
2998 	return thp;
2999 }
3000 
3001 #if IS_ENABLED(CONFIG_NFS_V4_2)
3002 int
3003 pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
3004 {
3005 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3006 	struct nfs_server *server = NFS_SERVER(inode);
3007 	struct nfs_inode *nfsi = NFS_I(inode);
3008 	struct nfs42_layoutstat_data *data;
3009 	struct pnfs_layout_hdr *hdr;
3010 	int status = 0;
3011 
3012 	if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3013 		goto out;
3014 
3015 	if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3016 		goto out;
3017 
3018 	if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3019 		goto out;
3020 
3021 	spin_lock(&inode->i_lock);
3022 	if (!NFS_I(inode)->layout) {
3023 		spin_unlock(&inode->i_lock);
3024 		goto out_clear_layoutstats;
3025 	}
3026 	hdr = NFS_I(inode)->layout;
3027 	pnfs_get_layout_hdr(hdr);
3028 	spin_unlock(&inode->i_lock);
3029 
3030 	data = kzalloc(sizeof(*data), gfp_flags);
3031 	if (!data) {
3032 		status = -ENOMEM;
3033 		goto out_put;
3034 	}
3035 
3036 	data->args.fh = NFS_FH(inode);
3037 	data->args.inode = inode;
3038 	status = ld->prepare_layoutstats(&data->args);
3039 	if (status)
3040 		goto out_free;
3041 
3042 	status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3043 
3044 out:
3045 	dprintk("%s returns %d\n", __func__, status);
3046 	return status;
3047 
3048 out_free:
3049 	kfree(data);
3050 out_put:
3051 	pnfs_put_layout_hdr(hdr);
3052 out_clear_layoutstats:
3053 	smp_mb__before_atomic();
3054 	clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3055 	smp_mb__after_atomic();
3056 	goto out;
3057 }
3058 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
3059 #endif
3060 
3061 unsigned int layoutstats_timer;
3062 module_param(layoutstats_timer, uint, 0644);
3063 EXPORT_SYMBOL_GPL(layoutstats_timer);
3064