xref: /openbmc/linux/fs/nfs/pnfs.c (revision 5a170e9e)
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 {
762 	struct pnfs_layout_hdr *lo, *next;
763 	struct inode *inode;
764 
765 	list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
766 		if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
767 			continue;
768 		inode = igrab(lo->plh_inode);
769 		if (inode == NULL)
770 			continue;
771 		list_del_init(&lo->plh_layouts);
772 		if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
773 			continue;
774 		rcu_read_unlock();
775 		spin_unlock(&clp->cl_lock);
776 		iput(inode);
777 		spin_lock(&clp->cl_lock);
778 		rcu_read_lock();
779 		return -EAGAIN;
780 	}
781 	return 0;
782 }
783 
784 static int
785 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
786 		bool is_bulk_recall)
787 {
788 	struct pnfs_layout_hdr *lo;
789 	struct inode *inode;
790 	LIST_HEAD(lseg_list);
791 	int ret = 0;
792 
793 	while (!list_empty(layout_list)) {
794 		lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
795 				plh_bulk_destroy);
796 		dprintk("%s freeing layout for inode %lu\n", __func__,
797 			lo->plh_inode->i_ino);
798 		inode = lo->plh_inode;
799 
800 		pnfs_layoutcommit_inode(inode, false);
801 
802 		spin_lock(&inode->i_lock);
803 		list_del_init(&lo->plh_bulk_destroy);
804 		if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
805 			if (is_bulk_recall)
806 				set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
807 			ret = -EAGAIN;
808 		}
809 		spin_unlock(&inode->i_lock);
810 		pnfs_free_lseg_list(&lseg_list);
811 		/* Free all lsegs that are attached to commit buckets */
812 		nfs_commit_inode(inode, 0);
813 		pnfs_put_layout_hdr(lo);
814 		iput(inode);
815 	}
816 	return ret;
817 }
818 
819 int
820 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
821 		struct nfs_fsid *fsid,
822 		bool is_recall)
823 {
824 	struct nfs_server *server;
825 	LIST_HEAD(layout_list);
826 
827 	spin_lock(&clp->cl_lock);
828 	rcu_read_lock();
829 restart:
830 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
831 		if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
832 			continue;
833 		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
834 				server,
835 				&layout_list) != 0)
836 			goto restart;
837 	}
838 	rcu_read_unlock();
839 	spin_unlock(&clp->cl_lock);
840 
841 	if (list_empty(&layout_list))
842 		return 0;
843 	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
844 }
845 
846 int
847 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
848 		bool is_recall)
849 {
850 	struct nfs_server *server;
851 	LIST_HEAD(layout_list);
852 
853 	spin_lock(&clp->cl_lock);
854 	rcu_read_lock();
855 restart:
856 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
857 		if (pnfs_layout_bulk_destroy_byserver_locked(clp,
858 					server,
859 					&layout_list) != 0)
860 			goto restart;
861 	}
862 	rcu_read_unlock();
863 	spin_unlock(&clp->cl_lock);
864 
865 	if (list_empty(&layout_list))
866 		return 0;
867 	return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
868 }
869 
870 /*
871  * Called by the state manger to remove all layouts established under an
872  * expired lease.
873  */
874 void
875 pnfs_destroy_all_layouts(struct nfs_client *clp)
876 {
877 	nfs4_deviceid_mark_client_invalid(clp);
878 	nfs4_deviceid_purge_client(clp);
879 
880 	pnfs_destroy_layouts_byclid(clp, false);
881 }
882 
883 /* update lo->plh_stateid with new if is more recent */
884 void
885 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
886 			bool update_barrier)
887 {
888 	u32 oldseq, newseq, new_barrier = 0;
889 
890 	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
891 	newseq = be32_to_cpu(new->seqid);
892 
893 	if (!pnfs_layout_is_valid(lo)) {
894 		nfs4_stateid_copy(&lo->plh_stateid, new);
895 		lo->plh_barrier = newseq;
896 		pnfs_clear_layoutreturn_info(lo);
897 		clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
898 		return;
899 	}
900 	if (pnfs_seqid_is_newer(newseq, oldseq)) {
901 		nfs4_stateid_copy(&lo->plh_stateid, new);
902 		/*
903 		 * Because of wraparound, we want to keep the barrier
904 		 * "close" to the current seqids.
905 		 */
906 		new_barrier = newseq - atomic_read(&lo->plh_outstanding);
907 	}
908 	if (update_barrier)
909 		new_barrier = be32_to_cpu(new->seqid);
910 	else if (new_barrier == 0)
911 		return;
912 	if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
913 		lo->plh_barrier = new_barrier;
914 }
915 
916 static bool
917 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
918 		const nfs4_stateid *stateid)
919 {
920 	u32 seqid = be32_to_cpu(stateid->seqid);
921 
922 	return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
923 }
924 
925 /* lget is set to 1 if called from inside send_layoutget call chain */
926 static bool
927 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
928 {
929 	return lo->plh_block_lgets ||
930 		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
931 }
932 
933 static struct nfs_server *
934 pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
935 {
936 	struct nfs_server *server;
937 
938 	if (inode) {
939 		server = NFS_SERVER(inode);
940 	} else {
941 		struct dentry *parent_dir = dget_parent(ctx->dentry);
942 		server = NFS_SERVER(parent_dir->d_inode);
943 		dput(parent_dir);
944 	}
945 	return server;
946 }
947 
948 static void nfs4_free_pages(struct page **pages, size_t size)
949 {
950 	int i;
951 
952 	if (!pages)
953 		return;
954 
955 	for (i = 0; i < size; i++) {
956 		if (!pages[i])
957 			break;
958 		__free_page(pages[i]);
959 	}
960 	kfree(pages);
961 }
962 
963 static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
964 {
965 	struct page **pages;
966 	int i;
967 
968 	pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
969 	if (!pages) {
970 		dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
971 		return NULL;
972 	}
973 
974 	for (i = 0; i < size; i++) {
975 		pages[i] = alloc_page(gfp_flags);
976 		if (!pages[i]) {
977 			dprintk("%s: failed to allocate page\n", __func__);
978 			nfs4_free_pages(pages, i);
979 			return NULL;
980 		}
981 	}
982 
983 	return pages;
984 }
985 
986 static struct nfs4_layoutget *
987 pnfs_alloc_init_layoutget_args(struct inode *ino,
988 	   struct nfs_open_context *ctx,
989 	   const nfs4_stateid *stateid,
990 	   const struct pnfs_layout_range *range,
991 	   gfp_t gfp_flags)
992 {
993 	struct nfs_server *server = pnfs_find_server(ino, ctx);
994 	size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
995 	size_t max_pages = max_response_pages(server);
996 	struct nfs4_layoutget *lgp;
997 
998 	dprintk("--> %s\n", __func__);
999 
1000 	lgp = kzalloc(sizeof(*lgp), gfp_flags);
1001 	if (lgp == NULL)
1002 		return NULL;
1003 
1004 	if (max_reply_sz) {
1005 		size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1006 		if (npages < max_pages)
1007 			max_pages = npages;
1008 	}
1009 
1010 	lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1011 	if (!lgp->args.layout.pages) {
1012 		kfree(lgp);
1013 		return NULL;
1014 	}
1015 	lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1016 	lgp->res.layoutp = &lgp->args.layout;
1017 
1018 	/* Don't confuse uninitialised result and success */
1019 	lgp->res.status = -NFS4ERR_DELAY;
1020 
1021 	lgp->args.minlength = PAGE_SIZE;
1022 	if (lgp->args.minlength > range->length)
1023 		lgp->args.minlength = range->length;
1024 	if (ino) {
1025 		loff_t i_size = i_size_read(ino);
1026 
1027 		if (range->iomode == IOMODE_READ) {
1028 			if (range->offset >= i_size)
1029 				lgp->args.minlength = 0;
1030 			else if (i_size - range->offset < lgp->args.minlength)
1031 				lgp->args.minlength = i_size - range->offset;
1032 		}
1033 	}
1034 	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1035 	pnfs_copy_range(&lgp->args.range, range);
1036 	lgp->args.type = server->pnfs_curr_ld->id;
1037 	lgp->args.inode = ino;
1038 	lgp->args.ctx = get_nfs_open_context(ctx);
1039 	nfs4_stateid_copy(&lgp->args.stateid, stateid);
1040 	lgp->gfp_flags = gfp_flags;
1041 	lgp->cred = get_cred(ctx->cred);
1042 	return lgp;
1043 }
1044 
1045 void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1046 {
1047 	size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1048 
1049 	nfs4_free_pages(lgp->args.layout.pages, max_pages);
1050 	if (lgp->args.inode)
1051 		pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
1052 	put_cred(lgp->cred);
1053 	put_nfs_open_context(lgp->args.ctx);
1054 	kfree(lgp);
1055 }
1056 
1057 static void pnfs_clear_layoutcommit(struct inode *inode,
1058 		struct list_head *head)
1059 {
1060 	struct nfs_inode *nfsi = NFS_I(inode);
1061 	struct pnfs_layout_segment *lseg, *tmp;
1062 
1063 	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1064 		return;
1065 	list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1066 		if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1067 			continue;
1068 		pnfs_lseg_dec_and_remove_zero(lseg, head);
1069 	}
1070 }
1071 
1072 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
1073 		const nfs4_stateid *arg_stateid,
1074 		const struct pnfs_layout_range *range,
1075 		const nfs4_stateid *stateid)
1076 {
1077 	struct inode *inode = lo->plh_inode;
1078 	LIST_HEAD(freeme);
1079 
1080 	spin_lock(&inode->i_lock);
1081 	if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1082 	    !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1083 		goto out_unlock;
1084 	if (stateid) {
1085 		u32 seq = be32_to_cpu(arg_stateid->seqid);
1086 
1087 		pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1088 		pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1089 		pnfs_set_layout_stateid(lo, stateid, true);
1090 	} else
1091 		pnfs_mark_layout_stateid_invalid(lo, &freeme);
1092 out_unlock:
1093 	pnfs_clear_layoutreturn_waitbit(lo);
1094 	spin_unlock(&inode->i_lock);
1095 	pnfs_free_lseg_list(&freeme);
1096 
1097 }
1098 
1099 static bool
1100 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1101 		nfs4_stateid *stateid,
1102 		enum pnfs_iomode *iomode)
1103 {
1104 	/* Serialise LAYOUTGET/LAYOUTRETURN */
1105 	if (atomic_read(&lo->plh_outstanding) != 0)
1106 		return false;
1107 	if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
1108 		return false;
1109 	set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
1110 	pnfs_get_layout_hdr(lo);
1111 	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1112 		if (stateid != NULL) {
1113 			nfs4_stateid_copy(stateid, &lo->plh_stateid);
1114 			if (lo->plh_return_seq != 0)
1115 				stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1116 		}
1117 		if (iomode != NULL)
1118 			*iomode = lo->plh_return_iomode;
1119 		pnfs_clear_layoutreturn_info(lo);
1120 		return true;
1121 	}
1122 	if (stateid != NULL)
1123 		nfs4_stateid_copy(stateid, &lo->plh_stateid);
1124 	if (iomode != NULL)
1125 		*iomode = IOMODE_ANY;
1126 	return true;
1127 }
1128 
1129 static void
1130 pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1131 		struct pnfs_layout_hdr *lo,
1132 		const nfs4_stateid *stateid,
1133 		enum pnfs_iomode iomode)
1134 {
1135 	struct inode *inode = lo->plh_inode;
1136 
1137 	args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1138 	args->inode = inode;
1139 	args->range.iomode = iomode;
1140 	args->range.offset = 0;
1141 	args->range.length = NFS4_MAX_UINT64;
1142 	args->layout = lo;
1143 	nfs4_stateid_copy(&args->stateid, stateid);
1144 }
1145 
1146 static int
1147 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
1148 		       enum pnfs_iomode iomode, bool sync)
1149 {
1150 	struct inode *ino = lo->plh_inode;
1151 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1152 	struct nfs4_layoutreturn *lrp;
1153 	int status = 0;
1154 
1155 	lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
1156 	if (unlikely(lrp == NULL)) {
1157 		status = -ENOMEM;
1158 		spin_lock(&ino->i_lock);
1159 		pnfs_clear_layoutreturn_waitbit(lo);
1160 		spin_unlock(&ino->i_lock);
1161 		pnfs_put_layout_hdr(lo);
1162 		goto out;
1163 	}
1164 
1165 	pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
1166 	lrp->args.ld_private = &lrp->ld_private;
1167 	lrp->clp = NFS_SERVER(ino)->nfs_client;
1168 	lrp->cred = lo->plh_lc_cred;
1169 	if (ld->prepare_layoutreturn)
1170 		ld->prepare_layoutreturn(&lrp->args);
1171 
1172 	status = nfs4_proc_layoutreturn(lrp, sync);
1173 out:
1174 	dprintk("<-- %s status: %d\n", __func__, status);
1175 	return status;
1176 }
1177 
1178 /* Return true if layoutreturn is needed */
1179 static bool
1180 pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1181 {
1182 	struct pnfs_layout_segment *s;
1183 	enum pnfs_iomode iomode;
1184 	u32 seq;
1185 
1186 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1187 		return false;
1188 
1189 	seq = lo->plh_return_seq;
1190 	iomode = lo->plh_return_iomode;
1191 
1192 	/* Defer layoutreturn until all recalled lsegs are done */
1193 	list_for_each_entry(s, &lo->plh_segs, pls_list) {
1194 		if (seq && pnfs_seqid_is_newer(s->pls_seq, seq))
1195 			continue;
1196 		if (iomode != IOMODE_ANY && s->pls_range.iomode != iomode)
1197 			continue;
1198 		if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1199 			return false;
1200 	}
1201 
1202 	return true;
1203 }
1204 
1205 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1206 {
1207 	struct inode *inode= lo->plh_inode;
1208 
1209 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1210 		return;
1211 	spin_lock(&inode->i_lock);
1212 	if (pnfs_layout_need_return(lo)) {
1213 		nfs4_stateid stateid;
1214 		enum pnfs_iomode iomode;
1215 		bool send;
1216 
1217 		send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1218 		spin_unlock(&inode->i_lock);
1219 		if (send) {
1220 			/* Send an async layoutreturn so we dont deadlock */
1221 			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1222 		}
1223 	} else
1224 		spin_unlock(&inode->i_lock);
1225 }
1226 
1227 /*
1228  * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1229  * when the layout segment list is empty.
1230  *
1231  * Note that a pnfs_layout_hdr can exist with an empty layout segment
1232  * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1233  * deviceid is marked invalid.
1234  */
1235 int
1236 _pnfs_return_layout(struct inode *ino)
1237 {
1238 	struct pnfs_layout_hdr *lo = NULL;
1239 	struct nfs_inode *nfsi = NFS_I(ino);
1240 	LIST_HEAD(tmp_list);
1241 	nfs4_stateid stateid;
1242 	int status = 0;
1243 	bool send, valid_layout;
1244 
1245 	dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
1246 
1247 	spin_lock(&ino->i_lock);
1248 	lo = nfsi->layout;
1249 	if (!lo) {
1250 		spin_unlock(&ino->i_lock);
1251 		dprintk("NFS: %s no layout to return\n", __func__);
1252 		goto out;
1253 	}
1254 	/* Reference matched in nfs4_layoutreturn_release */
1255 	pnfs_get_layout_hdr(lo);
1256 	/* Is there an outstanding layoutreturn ? */
1257 	if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1258 		spin_unlock(&ino->i_lock);
1259 		if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1260 					TASK_UNINTERRUPTIBLE))
1261 			goto out_put_layout_hdr;
1262 		spin_lock(&ino->i_lock);
1263 	}
1264 	valid_layout = pnfs_layout_is_valid(lo);
1265 	pnfs_clear_layoutcommit(ino, &tmp_list);
1266 	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
1267 
1268 	if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1269 		struct pnfs_layout_range range = {
1270 			.iomode		= IOMODE_ANY,
1271 			.offset		= 0,
1272 			.length		= NFS4_MAX_UINT64,
1273 		};
1274 		NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1275 	}
1276 
1277 	/* Don't send a LAYOUTRETURN if list was initially empty */
1278 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1279 			!valid_layout) {
1280 		spin_unlock(&ino->i_lock);
1281 		dprintk("NFS: %s no layout segments to return\n", __func__);
1282 		goto out_put_layout_hdr;
1283 	}
1284 
1285 	send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
1286 	spin_unlock(&ino->i_lock);
1287 	if (send)
1288 		status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1289 out_put_layout_hdr:
1290 	pnfs_free_lseg_list(&tmp_list);
1291 	pnfs_put_layout_hdr(lo);
1292 out:
1293 	dprintk("<-- %s status: %d\n", __func__, status);
1294 	return status;
1295 }
1296 
1297 int
1298 pnfs_commit_and_return_layout(struct inode *inode)
1299 {
1300 	struct pnfs_layout_hdr *lo;
1301 	int ret;
1302 
1303 	spin_lock(&inode->i_lock);
1304 	lo = NFS_I(inode)->layout;
1305 	if (lo == NULL) {
1306 		spin_unlock(&inode->i_lock);
1307 		return 0;
1308 	}
1309 	pnfs_get_layout_hdr(lo);
1310 	/* Block new layoutgets and read/write to ds */
1311 	lo->plh_block_lgets++;
1312 	spin_unlock(&inode->i_lock);
1313 	filemap_fdatawait(inode->i_mapping);
1314 	ret = pnfs_layoutcommit_inode(inode, true);
1315 	if (ret == 0)
1316 		ret = _pnfs_return_layout(inode);
1317 	spin_lock(&inode->i_lock);
1318 	lo->plh_block_lgets--;
1319 	spin_unlock(&inode->i_lock);
1320 	pnfs_put_layout_hdr(lo);
1321 	return ret;
1322 }
1323 
1324 bool pnfs_roc(struct inode *ino,
1325 		struct nfs4_layoutreturn_args *args,
1326 		struct nfs4_layoutreturn_res *res,
1327 		const struct cred *cred)
1328 {
1329 	struct nfs_inode *nfsi = NFS_I(ino);
1330 	struct nfs_open_context *ctx;
1331 	struct nfs4_state *state;
1332 	struct pnfs_layout_hdr *lo;
1333 	struct pnfs_layout_segment *lseg, *next;
1334 	nfs4_stateid stateid;
1335 	enum pnfs_iomode iomode = 0;
1336 	bool layoutreturn = false, roc = false;
1337 	bool skip_read = false;
1338 
1339 	if (!nfs_have_layout(ino))
1340 		return false;
1341 retry:
1342 	rcu_read_lock();
1343 	spin_lock(&ino->i_lock);
1344 	lo = nfsi->layout;
1345 	if (!lo || !pnfs_layout_is_valid(lo) ||
1346 	    test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1347 		lo = NULL;
1348 		goto out_noroc;
1349 	}
1350 	pnfs_get_layout_hdr(lo);
1351 	if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1352 		spin_unlock(&ino->i_lock);
1353 		rcu_read_unlock();
1354 		wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1355 				TASK_UNINTERRUPTIBLE);
1356 		pnfs_put_layout_hdr(lo);
1357 		goto retry;
1358 	}
1359 
1360 	/* no roc if we hold a delegation */
1361 	if (nfs4_check_delegation(ino, FMODE_READ)) {
1362 		if (nfs4_check_delegation(ino, FMODE_WRITE))
1363 			goto out_noroc;
1364 		skip_read = true;
1365 	}
1366 
1367 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1368 		state = ctx->state;
1369 		if (state == NULL)
1370 			continue;
1371 		/* Don't return layout if there is open file state */
1372 		if (state->state & FMODE_WRITE)
1373 			goto out_noroc;
1374 		if (state->state & FMODE_READ)
1375 			skip_read = true;
1376 	}
1377 
1378 
1379 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
1380 		if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1381 			continue;
1382 		/* If we are sending layoutreturn, invalidate all valid lsegs */
1383 		if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1384 			continue;
1385 		/*
1386 		 * Note: mark lseg for return so pnfs_layout_remove_lseg
1387 		 * doesn't invalidate the layout for us.
1388 		 */
1389 		set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1390 		if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1391 			continue;
1392 		pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
1393 	}
1394 
1395 	if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1396 		goto out_noroc;
1397 
1398 	/* ROC in two conditions:
1399 	 * 1. there are ROC lsegs
1400 	 * 2. we don't send layoutreturn
1401 	 */
1402 	/* lo ref dropped in pnfs_roc_release() */
1403 	layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1404 	/* If the creds don't match, we can't compound the layoutreturn */
1405 	if (!layoutreturn || cred != lo->plh_lc_cred)
1406 		goto out_noroc;
1407 
1408 	roc = layoutreturn;
1409 	pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1410 	res->lrs_present = 0;
1411 	layoutreturn = false;
1412 
1413 out_noroc:
1414 	spin_unlock(&ino->i_lock);
1415 	rcu_read_unlock();
1416 	pnfs_layoutcommit_inode(ino, true);
1417 	if (roc) {
1418 		struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1419 		if (ld->prepare_layoutreturn)
1420 			ld->prepare_layoutreturn(args);
1421 		pnfs_put_layout_hdr(lo);
1422 		return true;
1423 	}
1424 	if (layoutreturn)
1425 		pnfs_send_layoutreturn(lo, &stateid, iomode, true);
1426 	pnfs_put_layout_hdr(lo);
1427 	return false;
1428 }
1429 
1430 void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1431 		struct nfs4_layoutreturn_res *res,
1432 		int ret)
1433 {
1434 	struct pnfs_layout_hdr *lo = args->layout;
1435 	const nfs4_stateid *arg_stateid = NULL;
1436 	const nfs4_stateid *res_stateid = NULL;
1437 	struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
1438 
1439 	if (ret == 0) {
1440 		arg_stateid = &args->stateid;
1441 		if (res->lrs_present)
1442 			res_stateid = &res->stateid;
1443 	}
1444 	pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1445 			res_stateid);
1446 	if (ld_private && ld_private->ops && ld_private->ops->free)
1447 		ld_private->ops->free(ld_private);
1448 	pnfs_put_layout_hdr(lo);
1449 	trace_nfs4_layoutreturn_on_close(args->inode, 0);
1450 }
1451 
1452 bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1453 {
1454 	struct nfs_inode *nfsi = NFS_I(ino);
1455         struct pnfs_layout_hdr *lo;
1456         bool sleep = false;
1457 
1458 	/* we might not have grabbed lo reference. so need to check under
1459 	 * i_lock */
1460         spin_lock(&ino->i_lock);
1461         lo = nfsi->layout;
1462         if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1463                 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1464                 sleep = true;
1465 	}
1466         spin_unlock(&ino->i_lock);
1467         return sleep;
1468 }
1469 
1470 /*
1471  * Compare two layout segments for sorting into layout cache.
1472  * We want to preferentially return RW over RO layouts, so ensure those
1473  * are seen first.
1474  */
1475 static s64
1476 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1477 	   const struct pnfs_layout_range *l2)
1478 {
1479 	s64 d;
1480 
1481 	/* high offset > low offset */
1482 	d = l1->offset - l2->offset;
1483 	if (d)
1484 		return d;
1485 
1486 	/* short length > long length */
1487 	d = l2->length - l1->length;
1488 	if (d)
1489 		return d;
1490 
1491 	/* read > read/write */
1492 	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1493 }
1494 
1495 static bool
1496 pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1497 		const struct pnfs_layout_range *l2)
1498 {
1499 	return pnfs_lseg_range_cmp(l1, l2) > 0;
1500 }
1501 
1502 static bool
1503 pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1504 		struct pnfs_layout_segment *old)
1505 {
1506 	return false;
1507 }
1508 
1509 void
1510 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1511 		   struct pnfs_layout_segment *lseg,
1512 		   bool (*is_after)(const struct pnfs_layout_range *,
1513 			   const struct pnfs_layout_range *),
1514 		   bool (*do_merge)(struct pnfs_layout_segment *,
1515 			   struct pnfs_layout_segment *),
1516 		   struct list_head *free_me)
1517 {
1518 	struct pnfs_layout_segment *lp, *tmp;
1519 
1520 	dprintk("%s:Begin\n", __func__);
1521 
1522 	list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1523 		if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1524 			continue;
1525 		if (do_merge(lseg, lp)) {
1526 			mark_lseg_invalid(lp, free_me);
1527 			continue;
1528 		}
1529 		if (is_after(&lseg->pls_range, &lp->pls_range))
1530 			continue;
1531 		list_add_tail(&lseg->pls_list, &lp->pls_list);
1532 		dprintk("%s: inserted lseg %p "
1533 			"iomode %d offset %llu length %llu before "
1534 			"lp %p iomode %d offset %llu length %llu\n",
1535 			__func__, lseg, lseg->pls_range.iomode,
1536 			lseg->pls_range.offset, lseg->pls_range.length,
1537 			lp, lp->pls_range.iomode, lp->pls_range.offset,
1538 			lp->pls_range.length);
1539 		goto out;
1540 	}
1541 	list_add_tail(&lseg->pls_list, &lo->plh_segs);
1542 	dprintk("%s: inserted lseg %p "
1543 		"iomode %d offset %llu length %llu at tail\n",
1544 		__func__, lseg, lseg->pls_range.iomode,
1545 		lseg->pls_range.offset, lseg->pls_range.length);
1546 out:
1547 	pnfs_get_layout_hdr(lo);
1548 
1549 	dprintk("%s:Return\n", __func__);
1550 }
1551 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1552 
1553 static void
1554 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1555 		   struct pnfs_layout_segment *lseg,
1556 		   struct list_head *free_me)
1557 {
1558 	struct inode *inode = lo->plh_inode;
1559 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1560 
1561 	if (ld->add_lseg != NULL)
1562 		ld->add_lseg(lo, lseg, free_me);
1563 	else
1564 		pnfs_generic_layout_insert_lseg(lo, lseg,
1565 				pnfs_lseg_range_is_after,
1566 				pnfs_lseg_no_merge,
1567 				free_me);
1568 }
1569 
1570 static struct pnfs_layout_hdr *
1571 alloc_init_layout_hdr(struct inode *ino,
1572 		      struct nfs_open_context *ctx,
1573 		      gfp_t gfp_flags)
1574 {
1575 	struct pnfs_layout_hdr *lo;
1576 
1577 	lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1578 	if (!lo)
1579 		return NULL;
1580 	refcount_set(&lo->plh_refcount, 1);
1581 	INIT_LIST_HEAD(&lo->plh_layouts);
1582 	INIT_LIST_HEAD(&lo->plh_segs);
1583 	INIT_LIST_HEAD(&lo->plh_return_segs);
1584 	INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1585 	lo->plh_inode = ino;
1586 	lo->plh_lc_cred = get_cred(ctx->cred);
1587 	lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
1588 	return lo;
1589 }
1590 
1591 static struct pnfs_layout_hdr *
1592 pnfs_find_alloc_layout(struct inode *ino,
1593 		       struct nfs_open_context *ctx,
1594 		       gfp_t gfp_flags)
1595 	__releases(&ino->i_lock)
1596 	__acquires(&ino->i_lock)
1597 {
1598 	struct nfs_inode *nfsi = NFS_I(ino);
1599 	struct pnfs_layout_hdr *new = NULL;
1600 
1601 	dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1602 
1603 	if (nfsi->layout != NULL)
1604 		goto out_existing;
1605 	spin_unlock(&ino->i_lock);
1606 	new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1607 	spin_lock(&ino->i_lock);
1608 
1609 	if (likely(nfsi->layout == NULL)) {	/* Won the race? */
1610 		nfsi->layout = new;
1611 		return new;
1612 	} else if (new != NULL)
1613 		pnfs_free_layout_hdr(new);
1614 out_existing:
1615 	pnfs_get_layout_hdr(nfsi->layout);
1616 	return nfsi->layout;
1617 }
1618 
1619 /*
1620  * iomode matching rules:
1621  * iomode	lseg	strict match
1622  *                      iomode
1623  * -----	-----	------ -----
1624  * ANY		READ	N/A    true
1625  * ANY		RW	N/A    true
1626  * RW		READ	N/A    false
1627  * RW		RW	N/A    true
1628  * READ		READ	N/A    true
1629  * READ		RW	true   false
1630  * READ		RW	false  true
1631  */
1632 static bool
1633 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1634 		 const struct pnfs_layout_range *range,
1635 		 bool strict_iomode)
1636 {
1637 	struct pnfs_layout_range range1;
1638 
1639 	if ((range->iomode == IOMODE_RW &&
1640 	     ls_range->iomode != IOMODE_RW) ||
1641 	    (range->iomode != ls_range->iomode &&
1642 	     strict_iomode) ||
1643 	    !pnfs_lseg_range_intersecting(ls_range, range))
1644 		return false;
1645 
1646 	/* range1 covers only the first byte in the range */
1647 	range1 = *range;
1648 	range1.length = 1;
1649 	return pnfs_lseg_range_contained(ls_range, &range1);
1650 }
1651 
1652 /*
1653  * lookup range in layout
1654  */
1655 static struct pnfs_layout_segment *
1656 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1657 		struct pnfs_layout_range *range,
1658 		bool strict_iomode)
1659 {
1660 	struct pnfs_layout_segment *lseg, *ret = NULL;
1661 
1662 	dprintk("%s:Begin\n", __func__);
1663 
1664 	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1665 		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1666 		    pnfs_lseg_range_match(&lseg->pls_range, range,
1667 					  strict_iomode)) {
1668 			ret = pnfs_get_lseg(lseg);
1669 			break;
1670 		}
1671 	}
1672 
1673 	dprintk("%s:Return lseg %p ref %d\n",
1674 		__func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
1675 	return ret;
1676 }
1677 
1678 /*
1679  * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1680  * to the MDS or over pNFS
1681  *
1682  * The nfs_inode read_io and write_io fields are cumulative counters reset
1683  * when there are no layout segments. Note that in pnfs_update_layout iomode
1684  * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1685  * WRITE request.
1686  *
1687  * A return of true means use MDS I/O.
1688  *
1689  * From rfc 5661:
1690  * If a file's size is smaller than the file size threshold, data accesses
1691  * SHOULD be sent to the metadata server.  If an I/O request has a length that
1692  * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1693  * server.  If both file size and I/O size are provided, the client SHOULD
1694  * reach or exceed  both thresholds before sending its read or write
1695  * requests to the data server.
1696  */
1697 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1698 				     struct inode *ino, int iomode)
1699 {
1700 	struct nfs4_threshold *t = ctx->mdsthreshold;
1701 	struct nfs_inode *nfsi = NFS_I(ino);
1702 	loff_t fsize = i_size_read(ino);
1703 	bool size = false, size_set = false, io = false, io_set = false, ret = false;
1704 
1705 	if (t == NULL)
1706 		return ret;
1707 
1708 	dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1709 		__func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1710 
1711 	switch (iomode) {
1712 	case IOMODE_READ:
1713 		if (t->bm & THRESHOLD_RD) {
1714 			dprintk("%s fsize %llu\n", __func__, fsize);
1715 			size_set = true;
1716 			if (fsize < t->rd_sz)
1717 				size = true;
1718 		}
1719 		if (t->bm & THRESHOLD_RD_IO) {
1720 			dprintk("%s nfsi->read_io %llu\n", __func__,
1721 				nfsi->read_io);
1722 			io_set = true;
1723 			if (nfsi->read_io < t->rd_io_sz)
1724 				io = true;
1725 		}
1726 		break;
1727 	case IOMODE_RW:
1728 		if (t->bm & THRESHOLD_WR) {
1729 			dprintk("%s fsize %llu\n", __func__, fsize);
1730 			size_set = true;
1731 			if (fsize < t->wr_sz)
1732 				size = true;
1733 		}
1734 		if (t->bm & THRESHOLD_WR_IO) {
1735 			dprintk("%s nfsi->write_io %llu\n", __func__,
1736 				nfsi->write_io);
1737 			io_set = true;
1738 			if (nfsi->write_io < t->wr_io_sz)
1739 				io = true;
1740 		}
1741 		break;
1742 	}
1743 	if (size_set && io_set) {
1744 		if (size && io)
1745 			ret = true;
1746 	} else if (size || io)
1747 		ret = true;
1748 
1749 	dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1750 	return ret;
1751 }
1752 
1753 static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1754 {
1755 	/*
1756 	 * send layoutcommit as it can hold up layoutreturn due to lseg
1757 	 * reference
1758 	 */
1759 	pnfs_layoutcommit_inode(lo->plh_inode, false);
1760 	return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1761 				   nfs_wait_bit_killable,
1762 				   TASK_KILLABLE);
1763 }
1764 
1765 static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1766 {
1767 	atomic_inc(&lo->plh_outstanding);
1768 }
1769 
1770 static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1771 {
1772 	if (atomic_dec_and_test(&lo->plh_outstanding))
1773 		wake_up_var(&lo->plh_outstanding);
1774 }
1775 
1776 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1777 {
1778 	unsigned long *bitlock = &lo->plh_flags;
1779 
1780 	clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1781 	smp_mb__after_atomic();
1782 	wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1783 }
1784 
1785 static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1786 				struct nfs_server *server)
1787 {
1788 	if (list_empty(&lo->plh_layouts)) {
1789 		struct nfs_client *clp = server->nfs_client;
1790 
1791 		/* The lo must be on the clp list if there is any
1792 		 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1793 		 */
1794 		spin_lock(&clp->cl_lock);
1795 		if (list_empty(&lo->plh_layouts))
1796 			list_add_tail(&lo->plh_layouts, &server->layouts);
1797 		spin_unlock(&clp->cl_lock);
1798 	}
1799 }
1800 
1801 /*
1802  * Layout segment is retreived from the server if not cached.
1803  * The appropriate layout segment is referenced and returned to the caller.
1804  */
1805 struct pnfs_layout_segment *
1806 pnfs_update_layout(struct inode *ino,
1807 		   struct nfs_open_context *ctx,
1808 		   loff_t pos,
1809 		   u64 count,
1810 		   enum pnfs_iomode iomode,
1811 		   bool strict_iomode,
1812 		   gfp_t gfp_flags)
1813 {
1814 	struct pnfs_layout_range arg = {
1815 		.iomode = iomode,
1816 		.offset = pos,
1817 		.length = count,
1818 	};
1819 	unsigned pg_offset;
1820 	struct nfs_server *server = NFS_SERVER(ino);
1821 	struct nfs_client *clp = server->nfs_client;
1822 	struct pnfs_layout_hdr *lo = NULL;
1823 	struct pnfs_layout_segment *lseg = NULL;
1824 	struct nfs4_layoutget *lgp;
1825 	nfs4_stateid stateid;
1826 	long timeout = 0;
1827 	unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
1828 	bool first;
1829 
1830 	if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1831 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1832 				 PNFS_UPDATE_LAYOUT_NO_PNFS);
1833 		goto out;
1834 	}
1835 
1836 	if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1837 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1838 				 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1839 		goto out;
1840 	}
1841 
1842 lookup_again:
1843 	lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1844 	if (IS_ERR(lseg))
1845 		goto out;
1846 	first = false;
1847 	spin_lock(&ino->i_lock);
1848 	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1849 	if (lo == NULL) {
1850 		spin_unlock(&ino->i_lock);
1851 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1852 				 PNFS_UPDATE_LAYOUT_NOMEM);
1853 		goto out;
1854 	}
1855 
1856 	/* Do we even need to bother with this? */
1857 	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1858 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1859 				 PNFS_UPDATE_LAYOUT_BULK_RECALL);
1860 		dprintk("%s matches recall, use MDS\n", __func__);
1861 		goto out_unlock;
1862 	}
1863 
1864 	/* if LAYOUTGET already failed once we don't try again */
1865 	if (pnfs_layout_io_test_failed(lo, iomode)) {
1866 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1867 				 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1868 		goto out_unlock;
1869 	}
1870 
1871 	/*
1872 	 * If the layout segment list is empty, but there are outstanding
1873 	 * layoutget calls, then they might be subject to a layoutrecall.
1874 	 */
1875 	if (list_empty(&lo->plh_segs) &&
1876 	    atomic_read(&lo->plh_outstanding) != 0) {
1877 		spin_unlock(&ino->i_lock);
1878 		lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
1879 					atomic_read(&lo->plh_outstanding)));
1880 		if (IS_ERR(lseg) || !list_empty(&lo->plh_segs))
1881 			goto out_put_layout_hdr;
1882 		pnfs_put_layout_hdr(lo);
1883 		goto lookup_again;
1884 	}
1885 
1886 	lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
1887 	if (lseg) {
1888 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1889 				PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1890 		goto out_unlock;
1891 	}
1892 
1893 	if (!nfs4_valid_open_stateid(ctx->state)) {
1894 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1895 				PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1896 		goto out_unlock;
1897 	}
1898 
1899 	/*
1900 	 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1901 	 * stateid, or it has been invalidated, then we must use the open
1902 	 * stateid.
1903 	 */
1904 	if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
1905 
1906 		/*
1907 		 * The first layoutget for the file. Need to serialize per
1908 		 * RFC 5661 Errata 3208.
1909 		 */
1910 		if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1911 				     &lo->plh_flags)) {
1912 			spin_unlock(&ino->i_lock);
1913 			lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
1914 						NFS_LAYOUT_FIRST_LAYOUTGET,
1915 						TASK_KILLABLE));
1916 			if (IS_ERR(lseg))
1917 				goto out_put_layout_hdr;
1918 			pnfs_put_layout_hdr(lo);
1919 			dprintk("%s retrying\n", __func__);
1920 			goto lookup_again;
1921 		}
1922 
1923 		first = true;
1924 		if (nfs4_select_rw_stateid(ctx->state,
1925 					iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1926 					NULL, &stateid, NULL) != 0) {
1927 			trace_pnfs_update_layout(ino, pos, count,
1928 					iomode, lo, lseg,
1929 					PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1930 			goto out_unlock;
1931 		}
1932 	} else {
1933 		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1934 	}
1935 
1936 	/*
1937 	 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1938 	 * for LAYOUTRETURN even if first is true.
1939 	 */
1940 	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1941 		spin_unlock(&ino->i_lock);
1942 		dprintk("%s wait for layoutreturn\n", __func__);
1943 		lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
1944 		if (!IS_ERR(lseg)) {
1945 			if (first)
1946 				pnfs_clear_first_layoutget(lo);
1947 			pnfs_put_layout_hdr(lo);
1948 			dprintk("%s retrying\n", __func__);
1949 			trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1950 					lseg, PNFS_UPDATE_LAYOUT_RETRY);
1951 			goto lookup_again;
1952 		}
1953 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1954 				PNFS_UPDATE_LAYOUT_RETURN);
1955 		goto out_put_layout_hdr;
1956 	}
1957 
1958 	if (pnfs_layoutgets_blocked(lo)) {
1959 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1960 				PNFS_UPDATE_LAYOUT_BLOCKED);
1961 		goto out_unlock;
1962 	}
1963 	nfs_layoutget_begin(lo);
1964 	spin_unlock(&ino->i_lock);
1965 
1966 	_add_to_server_list(lo, server);
1967 
1968 	pg_offset = arg.offset & ~PAGE_MASK;
1969 	if (pg_offset) {
1970 		arg.offset -= pg_offset;
1971 		arg.length += pg_offset;
1972 	}
1973 	if (arg.length != NFS4_MAX_UINT64)
1974 		arg.length = PAGE_ALIGN(arg.length);
1975 
1976 	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
1977 	if (!lgp) {
1978 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
1979 					 PNFS_UPDATE_LAYOUT_NOMEM);
1980 		nfs_layoutget_end(lo);
1981 		goto out_put_layout_hdr;
1982 	}
1983 
1984 	lseg = nfs4_proc_layoutget(lgp, &timeout);
1985 	trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1986 				 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
1987 	nfs_layoutget_end(lo);
1988 	if (IS_ERR(lseg)) {
1989 		switch(PTR_ERR(lseg)) {
1990 		case -EBUSY:
1991 			if (time_after(jiffies, giveup))
1992 				lseg = NULL;
1993 			break;
1994 		case -ERECALLCONFLICT:
1995 		case -EAGAIN:
1996 			break;
1997 		default:
1998 			if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
1999 				pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2000 				lseg = NULL;
2001 			}
2002 			goto out_put_layout_hdr;
2003 		}
2004 		if (lseg) {
2005 			if (first)
2006 				pnfs_clear_first_layoutget(lo);
2007 			trace_pnfs_update_layout(ino, pos, count,
2008 				iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2009 			pnfs_put_layout_hdr(lo);
2010 			goto lookup_again;
2011 		}
2012 	} else {
2013 		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2014 	}
2015 
2016 out_put_layout_hdr:
2017 	if (first)
2018 		pnfs_clear_first_layoutget(lo);
2019 	pnfs_put_layout_hdr(lo);
2020 out:
2021 	dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2022 			"(%s, offset: %llu, length: %llu)\n",
2023 			__func__, ino->i_sb->s_id,
2024 			(unsigned long long)NFS_FILEID(ino),
2025 			IS_ERR_OR_NULL(lseg) ? "not found" : "found",
2026 			iomode==IOMODE_RW ?  "read/write" : "read-only",
2027 			(unsigned long long)pos,
2028 			(unsigned long long)count);
2029 	return lseg;
2030 out_unlock:
2031 	spin_unlock(&ino->i_lock);
2032 	goto out_put_layout_hdr;
2033 }
2034 EXPORT_SYMBOL_GPL(pnfs_update_layout);
2035 
2036 static bool
2037 pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2038 {
2039 	switch (range->iomode) {
2040 	case IOMODE_READ:
2041 	case IOMODE_RW:
2042 		break;
2043 	default:
2044 		return false;
2045 	}
2046 	if (range->offset == NFS4_MAX_UINT64)
2047 		return false;
2048 	if (range->length == 0)
2049 		return false;
2050 	if (range->length != NFS4_MAX_UINT64 &&
2051 	    range->length > NFS4_MAX_UINT64 - range->offset)
2052 		return false;
2053 	return true;
2054 }
2055 
2056 static struct pnfs_layout_hdr *
2057 _pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2058 {
2059 	struct pnfs_layout_hdr *lo;
2060 
2061 	spin_lock(&ino->i_lock);
2062 	lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2063 	if (!lo)
2064 		goto out_unlock;
2065 	if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2066 		goto out_unlock;
2067 	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2068 		goto out_unlock;
2069 	if (pnfs_layoutgets_blocked(lo))
2070 		goto out_unlock;
2071 	if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2072 		goto out_unlock;
2073 	nfs_layoutget_begin(lo);
2074 	spin_unlock(&ino->i_lock);
2075 	_add_to_server_list(lo, NFS_SERVER(ino));
2076 	return lo;
2077 
2078 out_unlock:
2079 	spin_unlock(&ino->i_lock);
2080 	pnfs_put_layout_hdr(lo);
2081 	return NULL;
2082 }
2083 
2084 extern const nfs4_stateid current_stateid;
2085 
2086 static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2087 				     struct nfs_open_context *ctx)
2088 {
2089 	struct inode *ino = data->dentry->d_inode;
2090 	struct pnfs_layout_range rng = {
2091 		.iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2092 			  IOMODE_RW: IOMODE_READ,
2093 		.offset = 0,
2094 		.length = NFS4_MAX_UINT64,
2095 	};
2096 	struct nfs4_layoutget *lgp;
2097 	struct pnfs_layout_hdr *lo;
2098 
2099 	/* Heuristic: don't send layoutget if we have cached data */
2100 	if (rng.iomode == IOMODE_READ &&
2101 	   (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2102 		return;
2103 
2104 	lo = _pnfs_grab_empty_layout(ino, ctx);
2105 	if (!lo)
2106 		return;
2107 	lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2108 					     &rng, GFP_KERNEL);
2109 	if (!lgp) {
2110 		pnfs_clear_first_layoutget(lo);
2111 		pnfs_put_layout_hdr(lo);
2112 		return;
2113 	}
2114 	data->lgp = lgp;
2115 	data->o_arg.lg_args = &lgp->args;
2116 	data->o_res.lg_res = &lgp->res;
2117 }
2118 
2119 static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2120 				     struct nfs_open_context *ctx)
2121 {
2122 	struct pnfs_layout_range rng = {
2123 		.iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2124 			  IOMODE_RW: IOMODE_READ,
2125 		.offset = 0,
2126 		.length = NFS4_MAX_UINT64,
2127 	};
2128 	struct nfs4_layoutget *lgp;
2129 
2130 	lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2131 					     &rng, GFP_KERNEL);
2132 	if (!lgp)
2133 		return;
2134 	data->lgp = lgp;
2135 	data->o_arg.lg_args = &lgp->args;
2136 	data->o_res.lg_res = &lgp->res;
2137 }
2138 
2139 void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2140 			 struct nfs_open_context *ctx)
2141 {
2142 	struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2143 
2144 	if (!(pnfs_enabled_sb(server) &&
2145 	      server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2146 		return;
2147 	/* Could check on max_ops, but currently hardcoded high enough */
2148 	if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2149 		return;
2150 	if (data->state)
2151 		_lgopen_prepare_attached(data, ctx);
2152 	else
2153 		_lgopen_prepare_floating(data, ctx);
2154 }
2155 
2156 void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2157 		       struct nfs_open_context *ctx)
2158 {
2159 	struct pnfs_layout_hdr *lo;
2160 	struct pnfs_layout_segment *lseg;
2161 	struct nfs_server *srv = NFS_SERVER(ino);
2162 	u32 iomode;
2163 
2164 	if (!lgp)
2165 		return;
2166 	dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2167 	if (lgp->res.status) {
2168 		switch (lgp->res.status) {
2169 		default:
2170 			break;
2171 		/*
2172 		 * Halt lgopen attempts if the server doesn't recognise
2173 		 * the "current stateid" value, the layout type, or the
2174 		 * layoutget operation as being valid.
2175 		 * Also if it complains about too many ops in the compound
2176 		 * or of the request/reply being too big.
2177 		 */
2178 		case -NFS4ERR_BAD_STATEID:
2179 		case -NFS4ERR_NOTSUPP:
2180 		case -NFS4ERR_REP_TOO_BIG:
2181 		case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2182 		case -NFS4ERR_REQ_TOO_BIG:
2183 		case -NFS4ERR_TOO_MANY_OPS:
2184 		case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
2185 			srv->caps &= ~NFS_CAP_LGOPEN;
2186 		}
2187 		return;
2188 	}
2189 	if (!lgp->args.inode) {
2190 		lo = _pnfs_grab_empty_layout(ino, ctx);
2191 		if (!lo)
2192 			return;
2193 		lgp->args.inode = ino;
2194 	} else
2195 		lo = NFS_I(lgp->args.inode)->layout;
2196 
2197 	lseg = pnfs_layout_process(lgp);
2198 	if (!IS_ERR(lseg)) {
2199 		iomode = lgp->args.range.iomode;
2200 		pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2201 		pnfs_put_lseg(lseg);
2202 	}
2203 }
2204 
2205 void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2206 {
2207 	if (lgp != NULL) {
2208 		struct inode *inode = lgp->args.inode;
2209 		if (inode) {
2210 			struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2211 			pnfs_clear_first_layoutget(lo);
2212 			nfs_layoutget_end(lo);
2213 		}
2214 		pnfs_layoutget_free(lgp);
2215 	}
2216 }
2217 
2218 struct pnfs_layout_segment *
2219 pnfs_layout_process(struct nfs4_layoutget *lgp)
2220 {
2221 	struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2222 	struct nfs4_layoutget_res *res = &lgp->res;
2223 	struct pnfs_layout_segment *lseg;
2224 	struct inode *ino = lo->plh_inode;
2225 	LIST_HEAD(free_me);
2226 
2227 	if (!pnfs_sanity_check_layout_range(&res->range))
2228 		return ERR_PTR(-EINVAL);
2229 
2230 	/* Inject layout blob into I/O device driver */
2231 	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
2232 	if (IS_ERR_OR_NULL(lseg)) {
2233 		if (!lseg)
2234 			lseg = ERR_PTR(-ENOMEM);
2235 
2236 		dprintk("%s: Could not allocate layout: error %ld\n",
2237 		       __func__, PTR_ERR(lseg));
2238 		return lseg;
2239 	}
2240 
2241 	pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
2242 
2243 	spin_lock(&ino->i_lock);
2244 	if (pnfs_layoutgets_blocked(lo)) {
2245 		dprintk("%s forget reply due to state\n", __func__);
2246 		goto out_forget;
2247 	}
2248 
2249 	if (!pnfs_layout_is_valid(lo)) {
2250 		/* We have a completely new layout */
2251 		pnfs_set_layout_stateid(lo, &res->stateid, true);
2252 	} else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
2253 		/* existing state ID, make sure the sequence number matches. */
2254 		if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2255 			dprintk("%s forget reply due to sequence\n", __func__);
2256 			goto out_forget;
2257 		}
2258 		pnfs_set_layout_stateid(lo, &res->stateid, false);
2259 	} else {
2260 		/*
2261 		 * We got an entirely new state ID.  Mark all segments for the
2262 		 * inode invalid, and retry the layoutget
2263 		 */
2264 		pnfs_mark_layout_stateid_invalid(lo, &free_me);
2265 		goto out_forget;
2266 	}
2267 
2268 	pnfs_get_lseg(lseg);
2269 	pnfs_layout_insert_lseg(lo, lseg, &free_me);
2270 
2271 
2272 	if (res->return_on_close)
2273 		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
2274 
2275 	spin_unlock(&ino->i_lock);
2276 	pnfs_free_lseg_list(&free_me);
2277 	return lseg;
2278 
2279 out_forget:
2280 	spin_unlock(&ino->i_lock);
2281 	lseg->pls_layout = lo;
2282 	NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
2283 	return ERR_PTR(-EAGAIN);
2284 }
2285 
2286 static int
2287 mark_lseg_invalid_or_return(struct pnfs_layout_segment *lseg,
2288 		struct list_head *tmp_list)
2289 {
2290 	if (!mark_lseg_invalid(lseg, tmp_list))
2291 		return 0;
2292 	pnfs_cache_lseg_for_layoutreturn(lseg->pls_layout, lseg);
2293 	return 1;
2294 }
2295 
2296 /**
2297  * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2298  * @lo: pointer to layout header
2299  * @tmp_list: list header to be used with pnfs_free_lseg_list()
2300  * @return_range: describe layout segment ranges to be returned
2301  * @seq: stateid seqid to match
2302  *
2303  * This function is mainly intended for use by layoutrecall. It attempts
2304  * to free the layout segment immediately, or else to mark it for return
2305  * as soon as its reference count drops to zero.
2306  *
2307  * Returns
2308  * - 0: a layoutreturn needs to be scheduled.
2309  * - EBUSY: there are layout segment that are still in use.
2310  * - ENOENT: there are no layout segments that need to be returned.
2311  */
2312 int
2313 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2314 				struct list_head *tmp_list,
2315 				const struct pnfs_layout_range *return_range,
2316 				u32 seq)
2317 {
2318 	struct pnfs_layout_segment *lseg, *next;
2319 	int remaining = 0;
2320 
2321 	dprintk("%s:Begin lo %p\n", __func__, lo);
2322 
2323 	assert_spin_locked(&lo->plh_inode->i_lock);
2324 
2325 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
2326 		if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
2327 			dprintk("%s: marking lseg %p iomode %d "
2328 				"offset %llu length %llu\n", __func__,
2329 				lseg, lseg->pls_range.iomode,
2330 				lseg->pls_range.offset,
2331 				lseg->pls_range.length);
2332 			if (mark_lseg_invalid_or_return(lseg, tmp_list))
2333 				continue;
2334 			remaining++;
2335 			set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
2336 		}
2337 
2338 	if (remaining) {
2339 		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2340 		return -EBUSY;
2341 	}
2342 
2343 	if (!list_empty(&lo->plh_return_segs)) {
2344 		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2345 		return 0;
2346 	}
2347 
2348 	return -ENOENT;
2349 }
2350 
2351 void pnfs_error_mark_layout_for_return(struct inode *inode,
2352 				       struct pnfs_layout_segment *lseg)
2353 {
2354 	struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2355 	struct pnfs_layout_range range = {
2356 		.iomode = lseg->pls_range.iomode,
2357 		.offset = 0,
2358 		.length = NFS4_MAX_UINT64,
2359 	};
2360 	bool return_now = false;
2361 
2362 	spin_lock(&inode->i_lock);
2363 	if (!pnfs_layout_is_valid(lo)) {
2364 		spin_unlock(&inode->i_lock);
2365 		return;
2366 	}
2367 	pnfs_set_plh_return_info(lo, range.iomode, 0);
2368 	/*
2369 	 * mark all matching lsegs so that we are sure to have no live
2370 	 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2371 	 * for how it works.
2372 	 */
2373 	if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
2374 		nfs4_stateid stateid;
2375 		enum pnfs_iomode iomode;
2376 
2377 		return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
2378 		spin_unlock(&inode->i_lock);
2379 		if (return_now)
2380 			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2381 	} else {
2382 		spin_unlock(&inode->i_lock);
2383 		nfs_commit_inode(inode, 0);
2384 	}
2385 }
2386 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2387 
2388 void
2389 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2390 {
2391 	if (pgio->pg_lseg == NULL ||
2392 	    test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2393 		return;
2394 	pnfs_put_lseg(pgio->pg_lseg);
2395 	pgio->pg_lseg = NULL;
2396 }
2397 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2398 
2399 /*
2400  * Check for any intersection between the request and the pgio->pg_lseg,
2401  * and if none, put this pgio->pg_lseg away.
2402  */
2403 static void
2404 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2405 {
2406 	if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2407 		pnfs_put_lseg(pgio->pg_lseg);
2408 		pgio->pg_lseg = NULL;
2409 	}
2410 }
2411 
2412 void
2413 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2414 {
2415 	u64 rd_size = req->wb_bytes;
2416 
2417 	pnfs_generic_pg_check_layout(pgio);
2418 	pnfs_generic_pg_check_range(pgio, req);
2419 	if (pgio->pg_lseg == NULL) {
2420 		if (pgio->pg_dreq == NULL)
2421 			rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2422 		else
2423 			rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2424 
2425 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2426 						   req->wb_context,
2427 						   req_offset(req),
2428 						   rd_size,
2429 						   IOMODE_READ,
2430 						   false,
2431 						   GFP_KERNEL);
2432 		if (IS_ERR(pgio->pg_lseg)) {
2433 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2434 			pgio->pg_lseg = NULL;
2435 			return;
2436 		}
2437 	}
2438 	/* If no lseg, fall back to read through mds */
2439 	if (pgio->pg_lseg == NULL)
2440 		nfs_pageio_reset_read_mds(pgio);
2441 
2442 }
2443 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2444 
2445 void
2446 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2447 			   struct nfs_page *req, u64 wb_size)
2448 {
2449 	pnfs_generic_pg_check_layout(pgio);
2450 	pnfs_generic_pg_check_range(pgio, req);
2451 	if (pgio->pg_lseg == NULL) {
2452 		pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2453 						   req->wb_context,
2454 						   req_offset(req),
2455 						   wb_size,
2456 						   IOMODE_RW,
2457 						   false,
2458 						   GFP_NOFS);
2459 		if (IS_ERR(pgio->pg_lseg)) {
2460 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2461 			pgio->pg_lseg = NULL;
2462 			return;
2463 		}
2464 	}
2465 	/* If no lseg, fall back to write through mds */
2466 	if (pgio->pg_lseg == NULL)
2467 		nfs_pageio_reset_write_mds(pgio);
2468 }
2469 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2470 
2471 void
2472 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2473 {
2474 	if (desc->pg_lseg) {
2475 		pnfs_put_lseg(desc->pg_lseg);
2476 		desc->pg_lseg = NULL;
2477 	}
2478 }
2479 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2480 
2481 /*
2482  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2483  * of bytes (maximum @req->wb_bytes) that can be coalesced.
2484  */
2485 size_t
2486 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2487 		     struct nfs_page *prev, struct nfs_page *req)
2488 {
2489 	unsigned int size;
2490 	u64 seg_end, req_start, seg_left;
2491 
2492 	size = nfs_generic_pg_test(pgio, prev, req);
2493 	if (!size)
2494 		return 0;
2495 
2496 	/*
2497 	 * 'size' contains the number of bytes left in the current page (up
2498 	 * to the original size asked for in @req->wb_bytes).
2499 	 *
2500 	 * Calculate how many bytes are left in the layout segment
2501 	 * and if there are less bytes than 'size', return that instead.
2502 	 *
2503 	 * Please also note that 'end_offset' is actually the offset of the
2504 	 * first byte that lies outside the pnfs_layout_range. FIXME?
2505 	 *
2506 	 */
2507 	if (pgio->pg_lseg) {
2508 		seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
2509 				     pgio->pg_lseg->pls_range.length);
2510 		req_start = req_offset(req);
2511 
2512 		/* start of request is past the last byte of this segment */
2513 		if (req_start >= seg_end)
2514 			return 0;
2515 
2516 		/* adjust 'size' iff there are fewer bytes left in the
2517 		 * segment than what nfs_generic_pg_test returned */
2518 		seg_left = seg_end - req_start;
2519 		if (seg_left < size)
2520 			size = (unsigned int)seg_left;
2521 	}
2522 
2523 	return size;
2524 }
2525 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
2526 
2527 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
2528 {
2529 	struct nfs_pageio_descriptor pgio;
2530 
2531 	/* Resend all requests through the MDS */
2532 	nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2533 			      hdr->completion_ops);
2534 	set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
2535 	return nfs_pageio_resend(&pgio, hdr);
2536 }
2537 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
2538 
2539 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
2540 {
2541 
2542 	dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2543 	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2544 	    PNFS_LAYOUTRET_ON_ERROR) {
2545 		pnfs_return_layout(hdr->inode);
2546 	}
2547 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2548 		hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
2549 }
2550 
2551 /*
2552  * Called by non rpc-based layout drivers
2553  */
2554 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
2555 {
2556 	if (likely(!hdr->pnfs_error)) {
2557 		pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2558 				hdr->mds_offset + hdr->res.count);
2559 		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2560 	}
2561 	trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2562 	if (unlikely(hdr->pnfs_error))
2563 		pnfs_ld_handle_write_error(hdr);
2564 	hdr->mds_ops->rpc_release(hdr);
2565 }
2566 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
2567 
2568 static void
2569 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
2570 		struct nfs_pgio_header *hdr)
2571 {
2572 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2573 
2574 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2575 		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2576 		nfs_pageio_reset_write_mds(desc);
2577 		mirror->pg_recoalesce = 1;
2578 	}
2579 	hdr->completion_ops->completion(hdr);
2580 }
2581 
2582 static enum pnfs_try_status
2583 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
2584 			const struct rpc_call_ops *call_ops,
2585 			struct pnfs_layout_segment *lseg,
2586 			int how)
2587 {
2588 	struct inode *inode = hdr->inode;
2589 	enum pnfs_try_status trypnfs;
2590 	struct nfs_server *nfss = NFS_SERVER(inode);
2591 
2592 	hdr->mds_ops = call_ops;
2593 
2594 	dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
2595 		inode->i_ino, hdr->args.count, hdr->args.offset, how);
2596 	trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
2597 	if (trypnfs != PNFS_NOT_ATTEMPTED)
2598 		nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
2599 	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2600 	return trypnfs;
2601 }
2602 
2603 static void
2604 pnfs_do_write(struct nfs_pageio_descriptor *desc,
2605 	      struct nfs_pgio_header *hdr, int how)
2606 {
2607 	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2608 	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2609 	enum pnfs_try_status trypnfs;
2610 
2611 	trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2612 	switch (trypnfs) {
2613 	case PNFS_NOT_ATTEMPTED:
2614 		pnfs_write_through_mds(desc, hdr);
2615 	case PNFS_ATTEMPTED:
2616 		break;
2617 	case PNFS_TRY_AGAIN:
2618 		/* cleanup hdr and prepare to redo pnfs */
2619 		if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2620 			struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2621 			list_splice_init(&hdr->pages, &mirror->pg_list);
2622 			mirror->pg_recoalesce = 1;
2623 		}
2624 		hdr->mds_ops->rpc_release(hdr);
2625 	}
2626 }
2627 
2628 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2629 {
2630 	pnfs_put_lseg(hdr->lseg);
2631 	nfs_pgio_header_free(hdr);
2632 }
2633 
2634 int
2635 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2636 {
2637 	struct nfs_pgio_header *hdr;
2638 	int ret;
2639 
2640 	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2641 	if (!hdr) {
2642 		desc->pg_error = -ENOMEM;
2643 		return desc->pg_error;
2644 	}
2645 	nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2646 
2647 	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2648 	ret = nfs_generic_pgio(desc, hdr);
2649 	if (!ret)
2650 		pnfs_do_write(desc, hdr, desc->pg_ioflags);
2651 
2652 	return ret;
2653 }
2654 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2655 
2656 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2657 {
2658 	struct nfs_pageio_descriptor pgio;
2659 
2660 	/* Resend all requests through the MDS */
2661 	nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2662 	return nfs_pageio_resend(&pgio, hdr);
2663 }
2664 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2665 
2666 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2667 {
2668 	dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2669 	if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2670 	    PNFS_LAYOUTRET_ON_ERROR) {
2671 		pnfs_return_layout(hdr->inode);
2672 	}
2673 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2674 		hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2675 }
2676 
2677 /*
2678  * Called by non rpc-based layout drivers
2679  */
2680 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2681 {
2682 	if (likely(!hdr->pnfs_error))
2683 		hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2684 	trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2685 	if (unlikely(hdr->pnfs_error))
2686 		pnfs_ld_handle_read_error(hdr);
2687 	hdr->mds_ops->rpc_release(hdr);
2688 }
2689 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2690 
2691 static void
2692 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2693 		struct nfs_pgio_header *hdr)
2694 {
2695 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2696 
2697 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2698 		list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2699 		nfs_pageio_reset_read_mds(desc);
2700 		mirror->pg_recoalesce = 1;
2701 	}
2702 	hdr->completion_ops->completion(hdr);
2703 }
2704 
2705 /*
2706  * Call the appropriate parallel I/O subsystem read function.
2707  */
2708 static enum pnfs_try_status
2709 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2710 		       const struct rpc_call_ops *call_ops,
2711 		       struct pnfs_layout_segment *lseg)
2712 {
2713 	struct inode *inode = hdr->inode;
2714 	struct nfs_server *nfss = NFS_SERVER(inode);
2715 	enum pnfs_try_status trypnfs;
2716 
2717 	hdr->mds_ops = call_ops;
2718 
2719 	dprintk("%s: Reading ino:%lu %u@%llu\n",
2720 		__func__, inode->i_ino, hdr->args.count, hdr->args.offset);
2721 
2722 	trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2723 	if (trypnfs != PNFS_NOT_ATTEMPTED)
2724 		nfs_inc_stats(inode, NFSIOS_PNFS_READ);
2725 	dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2726 	return trypnfs;
2727 }
2728 
2729 /* Resend all requests through pnfs. */
2730 void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2731 {
2732 	struct nfs_pageio_descriptor pgio;
2733 
2734 	if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2735 		/* Prevent deadlocks with layoutreturn! */
2736 		pnfs_put_lseg(hdr->lseg);
2737 		hdr->lseg = NULL;
2738 
2739 		nfs_pageio_init_read(&pgio, hdr->inode, false,
2740 					hdr->completion_ops);
2741 		hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2742 	}
2743 }
2744 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2745 
2746 static void
2747 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2748 {
2749 	const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2750 	struct pnfs_layout_segment *lseg = desc->pg_lseg;
2751 	enum pnfs_try_status trypnfs;
2752 
2753 	trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2754 	switch (trypnfs) {
2755 	case PNFS_NOT_ATTEMPTED:
2756 		pnfs_read_through_mds(desc, hdr);
2757 	case PNFS_ATTEMPTED:
2758 		break;
2759 	case PNFS_TRY_AGAIN:
2760 		/* cleanup hdr and prepare to redo pnfs */
2761 		if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2762 			struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2763 			list_splice_init(&hdr->pages, &mirror->pg_list);
2764 			mirror->pg_recoalesce = 1;
2765 		}
2766 		hdr->mds_ops->rpc_release(hdr);
2767 	}
2768 }
2769 
2770 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2771 {
2772 	pnfs_put_lseg(hdr->lseg);
2773 	nfs_pgio_header_free(hdr);
2774 }
2775 
2776 int
2777 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2778 {
2779 	struct nfs_pgio_header *hdr;
2780 	int ret;
2781 
2782 	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2783 	if (!hdr) {
2784 		desc->pg_error = -ENOMEM;
2785 		return desc->pg_error;
2786 	}
2787 	nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2788 	hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2789 	ret = nfs_generic_pgio(desc, hdr);
2790 	if (!ret)
2791 		pnfs_do_read(desc, hdr);
2792 	return ret;
2793 }
2794 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2795 
2796 static void pnfs_clear_layoutcommitting(struct inode *inode)
2797 {
2798 	unsigned long *bitlock = &NFS_I(inode)->flags;
2799 
2800 	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2801 	smp_mb__after_atomic();
2802 	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2803 }
2804 
2805 /*
2806  * There can be multiple RW segments.
2807  */
2808 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
2809 {
2810 	struct pnfs_layout_segment *lseg;
2811 
2812 	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2813 		if (lseg->pls_range.iomode == IOMODE_RW &&
2814 		    test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2815 			list_add(&lseg->pls_lc_list, listp);
2816 	}
2817 }
2818 
2819 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2820 {
2821 	struct pnfs_layout_segment *lseg, *tmp;
2822 
2823 	/* Matched by references in pnfs_set_layoutcommit */
2824 	list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2825 		list_del_init(&lseg->pls_lc_list);
2826 		pnfs_put_lseg(lseg);
2827 	}
2828 
2829 	pnfs_clear_layoutcommitting(inode);
2830 }
2831 
2832 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2833 {
2834 	pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
2835 }
2836 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2837 
2838 void
2839 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2840 		loff_t end_pos)
2841 {
2842 	struct nfs_inode *nfsi = NFS_I(inode);
2843 	bool mark_as_dirty = false;
2844 
2845 	spin_lock(&inode->i_lock);
2846 	if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2847 		nfsi->layout->plh_lwb = end_pos;
2848 		mark_as_dirty = true;
2849 		dprintk("%s: Set layoutcommit for inode %lu ",
2850 			__func__, inode->i_ino);
2851 	} else if (end_pos > nfsi->layout->plh_lwb)
2852 		nfsi->layout->plh_lwb = end_pos;
2853 	if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2854 		/* references matched in nfs4_layoutcommit_release */
2855 		pnfs_get_lseg(lseg);
2856 	}
2857 	spin_unlock(&inode->i_lock);
2858 	dprintk("%s: lseg %p end_pos %llu\n",
2859 		__func__, lseg, nfsi->layout->plh_lwb);
2860 
2861 	/* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2862 	 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2863 	if (mark_as_dirty)
2864 		mark_inode_dirty_sync(inode);
2865 }
2866 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2867 
2868 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2869 {
2870 	struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2871 
2872 	if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2873 		nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
2874 	pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
2875 }
2876 
2877 /*
2878  * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2879  * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2880  * data to disk to allow the server to recover the data if it crashes.
2881  * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2882  * is off, and a COMMIT is sent to a data server, or
2883  * if WRITEs to a data server return NFS_DATA_SYNC.
2884  */
2885 int
2886 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
2887 {
2888 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2889 	struct nfs4_layoutcommit_data *data;
2890 	struct nfs_inode *nfsi = NFS_I(inode);
2891 	loff_t end_pos;
2892 	int status;
2893 
2894 	if (!pnfs_layoutcommit_outstanding(inode))
2895 		return 0;
2896 
2897 	dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
2898 
2899 	status = -EAGAIN;
2900 	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2901 		if (!sync)
2902 			goto out;
2903 		status = wait_on_bit_lock_action(&nfsi->flags,
2904 				NFS_INO_LAYOUTCOMMITTING,
2905 				nfs_wait_bit_killable,
2906 				TASK_KILLABLE);
2907 		if (status)
2908 			goto out;
2909 	}
2910 
2911 	status = -ENOMEM;
2912 	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2913 	data = kzalloc(sizeof(*data), GFP_NOFS);
2914 	if (!data)
2915 		goto clear_layoutcommitting;
2916 
2917 	status = 0;
2918 	spin_lock(&inode->i_lock);
2919 	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2920 		goto out_unlock;
2921 
2922 	INIT_LIST_HEAD(&data->lseg_list);
2923 	pnfs_list_write_lseg(inode, &data->lseg_list);
2924 
2925 	end_pos = nfsi->layout->plh_lwb;
2926 
2927 	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
2928 	spin_unlock(&inode->i_lock);
2929 
2930 	data->args.inode = inode;
2931 	data->cred = get_cred(nfsi->layout->plh_lc_cred);
2932 	nfs_fattr_init(&data->fattr);
2933 	data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2934 	data->res.fattr = &data->fattr;
2935 	if (end_pos != 0)
2936 		data->args.lastbytewritten = end_pos - 1;
2937 	else
2938 		data->args.lastbytewritten = U64_MAX;
2939 	data->res.server = NFS_SERVER(inode);
2940 
2941 	if (ld->prepare_layoutcommit) {
2942 		status = ld->prepare_layoutcommit(&data->args);
2943 		if (status) {
2944 			put_cred(data->cred);
2945 			spin_lock(&inode->i_lock);
2946 			set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2947 			if (end_pos > nfsi->layout->plh_lwb)
2948 				nfsi->layout->plh_lwb = end_pos;
2949 			goto out_unlock;
2950 		}
2951 	}
2952 
2953 
2954 	status = nfs4_proc_layoutcommit(data, sync);
2955 out:
2956 	if (status)
2957 		mark_inode_dirty_sync(inode);
2958 	dprintk("<-- %s status %d\n", __func__, status);
2959 	return status;
2960 out_unlock:
2961 	spin_unlock(&inode->i_lock);
2962 	kfree(data);
2963 clear_layoutcommitting:
2964 	pnfs_clear_layoutcommitting(inode);
2965 	goto out;
2966 }
2967 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2968 
2969 int
2970 pnfs_generic_sync(struct inode *inode, bool datasync)
2971 {
2972 	return pnfs_layoutcommit_inode(inode, true);
2973 }
2974 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2975 
2976 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2977 {
2978 	struct nfs4_threshold *thp;
2979 
2980 	thp = kzalloc(sizeof(*thp), GFP_NOFS);
2981 	if (!thp) {
2982 		dprintk("%s mdsthreshold allocation failed\n", __func__);
2983 		return NULL;
2984 	}
2985 	return thp;
2986 }
2987 
2988 #if IS_ENABLED(CONFIG_NFS_V4_2)
2989 int
2990 pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
2991 {
2992 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2993 	struct nfs_server *server = NFS_SERVER(inode);
2994 	struct nfs_inode *nfsi = NFS_I(inode);
2995 	struct nfs42_layoutstat_data *data;
2996 	struct pnfs_layout_hdr *hdr;
2997 	int status = 0;
2998 
2999 	if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3000 		goto out;
3001 
3002 	if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3003 		goto out;
3004 
3005 	if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3006 		goto out;
3007 
3008 	spin_lock(&inode->i_lock);
3009 	if (!NFS_I(inode)->layout) {
3010 		spin_unlock(&inode->i_lock);
3011 		goto out_clear_layoutstats;
3012 	}
3013 	hdr = NFS_I(inode)->layout;
3014 	pnfs_get_layout_hdr(hdr);
3015 	spin_unlock(&inode->i_lock);
3016 
3017 	data = kzalloc(sizeof(*data), gfp_flags);
3018 	if (!data) {
3019 		status = -ENOMEM;
3020 		goto out_put;
3021 	}
3022 
3023 	data->args.fh = NFS_FH(inode);
3024 	data->args.inode = inode;
3025 	status = ld->prepare_layoutstats(&data->args);
3026 	if (status)
3027 		goto out_free;
3028 
3029 	status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3030 
3031 out:
3032 	dprintk("%s returns %d\n", __func__, status);
3033 	return status;
3034 
3035 out_free:
3036 	kfree(data);
3037 out_put:
3038 	pnfs_put_layout_hdr(hdr);
3039 out_clear_layoutstats:
3040 	smp_mb__before_atomic();
3041 	clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3042 	smp_mb__after_atomic();
3043 	goto out;
3044 }
3045 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
3046 #endif
3047 
3048 unsigned int layoutstats_timer;
3049 module_param(layoutstats_timer, uint, 0644);
3050 EXPORT_SYMBOL_GPL(layoutstats_timer);
3051