]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/nfs/pnfs.c
Merge remote-tracking branch 'hwmon-staging/hwmon-next'
[karo-tx-linux.git] / fs / nfs / pnfs.c
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 "internal.h"
34 #include "pnfs.h"
35 #include "iostat.h"
36 #include "nfs4trace.h"
37 #include "delegation.h"
38 #include "nfs42.h"
39
40 #define NFSDBG_FACILITY         NFSDBG_PNFS
41 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
42
43 /* Locking:
44  *
45  * pnfs_spinlock:
46  *      protects pnfs_modules_tbl.
47  */
48 static DEFINE_SPINLOCK(pnfs_spinlock);
49
50 /*
51  * pnfs_modules_tbl holds all pnfs modules
52  */
53 static LIST_HEAD(pnfs_modules_tbl);
54
55 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
56
57 /* Return the registered pnfs layout driver module matching given id */
58 static struct pnfs_layoutdriver_type *
59 find_pnfs_driver_locked(u32 id)
60 {
61         struct pnfs_layoutdriver_type *local;
62
63         list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
64                 if (local->id == id)
65                         goto out;
66         local = NULL;
67 out:
68         dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
69         return local;
70 }
71
72 static struct pnfs_layoutdriver_type *
73 find_pnfs_driver(u32 id)
74 {
75         struct pnfs_layoutdriver_type *local;
76
77         spin_lock(&pnfs_spinlock);
78         local = find_pnfs_driver_locked(id);
79         if (local != NULL && !try_module_get(local->owner)) {
80                 dprintk("%s: Could not grab reference on module\n", __func__);
81                 local = NULL;
82         }
83         spin_unlock(&pnfs_spinlock);
84         return local;
85 }
86
87 void
88 unset_pnfs_layoutdriver(struct nfs_server *nfss)
89 {
90         if (nfss->pnfs_curr_ld) {
91                 if (nfss->pnfs_curr_ld->clear_layoutdriver)
92                         nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
93                 /* Decrement the MDS count. Purge the deviceid cache if zero */
94                 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
95                         nfs4_deviceid_purge_client(nfss->nfs_client);
96                 module_put(nfss->pnfs_curr_ld->owner);
97         }
98         nfss->pnfs_curr_ld = NULL;
99 }
100
101 /*
102  * Try to set the server's pnfs module to the pnfs layout type specified by id.
103  * Currently only one pNFS layout driver per filesystem is supported.
104  *
105  * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
106  */
107 void
108 set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
109                       u32 id)
110 {
111         struct pnfs_layoutdriver_type *ld_type = NULL;
112
113         if (id == 0)
114                 goto out_no_driver;
115         if (!(server->nfs_client->cl_exchange_flags &
116                  (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
117                 printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
118                         __func__, id, server->nfs_client->cl_exchange_flags);
119                 goto out_no_driver;
120         }
121         ld_type = find_pnfs_driver(id);
122         if (!ld_type) {
123                 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
124                 ld_type = find_pnfs_driver(id);
125                 if (!ld_type) {
126                         dprintk("%s: No pNFS module found for %u.\n",
127                                 __func__, id);
128                         goto out_no_driver;
129                 }
130         }
131         server->pnfs_curr_ld = ld_type;
132         if (ld_type->set_layoutdriver
133             && ld_type->set_layoutdriver(server, mntfh)) {
134                 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
135                         "driver %u.\n", __func__, id);
136                 module_put(ld_type->owner);
137                 goto out_no_driver;
138         }
139         /* Bump the MDS count */
140         atomic_inc(&server->nfs_client->cl_mds_count);
141
142         dprintk("%s: pNFS module for %u set\n", __func__, id);
143         return;
144
145 out_no_driver:
146         dprintk("%s: Using NFSv4 I/O\n", __func__);
147         server->pnfs_curr_ld = NULL;
148 }
149
150 int
151 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
152 {
153         int status = -EINVAL;
154         struct pnfs_layoutdriver_type *tmp;
155
156         if (ld_type->id == 0) {
157                 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
158                 return status;
159         }
160         if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
161                 printk(KERN_ERR "NFS: %s Layout driver must provide "
162                        "alloc_lseg and free_lseg.\n", __func__);
163                 return status;
164         }
165
166         spin_lock(&pnfs_spinlock);
167         tmp = find_pnfs_driver_locked(ld_type->id);
168         if (!tmp) {
169                 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
170                 status = 0;
171                 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
172                         ld_type->name);
173         } else {
174                 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
175                         __func__, ld_type->id);
176         }
177         spin_unlock(&pnfs_spinlock);
178
179         return status;
180 }
181 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
182
183 void
184 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
185 {
186         dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
187         spin_lock(&pnfs_spinlock);
188         list_del(&ld_type->pnfs_tblid);
189         spin_unlock(&pnfs_spinlock);
190 }
191 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
192
193 /*
194  * pNFS client layout cache
195  */
196
197 /* Need to hold i_lock if caller does not already hold reference */
198 void
199 pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
200 {
201         atomic_inc(&lo->plh_refcount);
202 }
203
204 static struct pnfs_layout_hdr *
205 pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
206 {
207         struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
208         return ld->alloc_layout_hdr(ino, gfp_flags);
209 }
210
211 static void
212 pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
213 {
214         struct nfs_server *server = NFS_SERVER(lo->plh_inode);
215         struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
216
217         if (!list_empty(&lo->plh_layouts)) {
218                 struct nfs_client *clp = server->nfs_client;
219
220                 spin_lock(&clp->cl_lock);
221                 list_del_init(&lo->plh_layouts);
222                 spin_unlock(&clp->cl_lock);
223         }
224         put_rpccred(lo->plh_lc_cred);
225         return ld->free_layout_hdr(lo);
226 }
227
228 static void
229 pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
230 {
231         struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
232         dprintk("%s: freeing layout cache %p\n", __func__, lo);
233         nfsi->layout = NULL;
234         /* Reset MDS Threshold I/O counters */
235         nfsi->write_io = 0;
236         nfsi->read_io = 0;
237 }
238
239 void
240 pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
241 {
242         struct inode *inode = lo->plh_inode;
243
244         pnfs_layoutreturn_before_put_layout_hdr(lo);
245
246         if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
247                 if (!list_empty(&lo->plh_segs))
248                         WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
249                 pnfs_detach_layout_hdr(lo);
250                 spin_unlock(&inode->i_lock);
251                 pnfs_free_layout_hdr(lo);
252         }
253 }
254
255 static int
256 pnfs_iomode_to_fail_bit(u32 iomode)
257 {
258         return iomode == IOMODE_RW ?
259                 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
260 }
261
262 static void
263 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
264 {
265         lo->plh_retry_timestamp = jiffies;
266         if (!test_and_set_bit(fail_bit, &lo->plh_flags))
267                 atomic_inc(&lo->plh_refcount);
268 }
269
270 static void
271 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
272 {
273         if (test_and_clear_bit(fail_bit, &lo->plh_flags))
274                 atomic_dec(&lo->plh_refcount);
275 }
276
277 static void
278 pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
279 {
280         struct inode *inode = lo->plh_inode;
281         struct pnfs_layout_range range = {
282                 .iomode = iomode,
283                 .offset = 0,
284                 .length = NFS4_MAX_UINT64,
285         };
286         LIST_HEAD(head);
287
288         spin_lock(&inode->i_lock);
289         pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
290         pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
291         spin_unlock(&inode->i_lock);
292         pnfs_free_lseg_list(&head);
293         dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
294                         iomode == IOMODE_RW ?  "RW" : "READ");
295 }
296
297 static bool
298 pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
299 {
300         unsigned long start, end;
301         int fail_bit = pnfs_iomode_to_fail_bit(iomode);
302
303         if (test_bit(fail_bit, &lo->plh_flags) == 0)
304                 return false;
305         end = jiffies;
306         start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
307         if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
308                 /* It is time to retry the failed layoutgets */
309                 pnfs_layout_clear_fail_bit(lo, fail_bit);
310                 return false;
311         }
312         return true;
313 }
314
315 static void
316 init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
317 {
318         INIT_LIST_HEAD(&lseg->pls_list);
319         INIT_LIST_HEAD(&lseg->pls_lc_list);
320         atomic_set(&lseg->pls_refcount, 1);
321         smp_mb();
322         set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
323         lseg->pls_layout = lo;
324 }
325
326 static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
327 {
328         struct inode *ino = lseg->pls_layout->plh_inode;
329
330         NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
331 }
332
333 static void
334 pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
335                 struct pnfs_layout_segment *lseg)
336 {
337         struct inode *inode = lo->plh_inode;
338
339         WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
340         list_del_init(&lseg->pls_list);
341         /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
342         atomic_dec(&lo->plh_refcount);
343         if (list_empty(&lo->plh_segs))
344                 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
345         rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
346 }
347
348 void
349 pnfs_put_lseg(struct pnfs_layout_segment *lseg)
350 {
351         struct pnfs_layout_hdr *lo;
352         struct inode *inode;
353
354         if (!lseg)
355                 return;
356
357         dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
358                 atomic_read(&lseg->pls_refcount),
359                 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
360
361         lo = lseg->pls_layout;
362         inode = lo->plh_inode;
363
364         if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
365                 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
366                         spin_unlock(&inode->i_lock);
367                         return;
368                 }
369                 pnfs_get_layout_hdr(lo);
370                 pnfs_layout_remove_lseg(lo, lseg);
371                 spin_unlock(&inode->i_lock);
372                 pnfs_free_lseg(lseg);
373                 pnfs_put_layout_hdr(lo);
374         }
375 }
376 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
377
378 static void pnfs_free_lseg_async_work(struct work_struct *work)
379 {
380         struct pnfs_layout_segment *lseg;
381         struct pnfs_layout_hdr *lo;
382
383         lseg = container_of(work, struct pnfs_layout_segment, pls_work);
384         lo = lseg->pls_layout;
385
386         pnfs_free_lseg(lseg);
387         pnfs_put_layout_hdr(lo);
388 }
389
390 static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
391 {
392         INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
393         schedule_work(&lseg->pls_work);
394 }
395
396 void
397 pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
398 {
399         if (!lseg)
400                 return;
401
402         assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);
403
404         dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
405                 atomic_read(&lseg->pls_refcount),
406                 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
407         if (atomic_dec_and_test(&lseg->pls_refcount)) {
408                 struct pnfs_layout_hdr *lo = lseg->pls_layout;
409                 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
410                         return;
411                 pnfs_get_layout_hdr(lo);
412                 pnfs_layout_remove_lseg(lo, lseg);
413                 pnfs_free_lseg_async(lseg);
414         }
415 }
416 EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
417
418 static u64
419 end_offset(u64 start, u64 len)
420 {
421         u64 end;
422
423         end = start + len;
424         return end >= start ? end : NFS4_MAX_UINT64;
425 }
426
427 /*
428  * is l2 fully contained in l1?
429  *   start1                             end1
430  *   [----------------------------------)
431  *           start2           end2
432  *           [----------------)
433  */
434 static bool
435 pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
436                  const struct pnfs_layout_range *l2)
437 {
438         u64 start1 = l1->offset;
439         u64 end1 = end_offset(start1, l1->length);
440         u64 start2 = l2->offset;
441         u64 end2 = end_offset(start2, l2->length);
442
443         return (start1 <= start2) && (end1 >= end2);
444 }
445
446 /*
447  * is l1 and l2 intersecting?
448  *   start1                             end1
449  *   [----------------------------------)
450  *                              start2           end2
451  *                              [----------------)
452  */
453 static bool
454 pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
455                     const struct pnfs_layout_range *l2)
456 {
457         u64 start1 = l1->offset;
458         u64 end1 = end_offset(start1, l1->length);
459         u64 start2 = l2->offset;
460         u64 end2 = end_offset(start2, l2->length);
461
462         return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
463                (end2 == NFS4_MAX_UINT64 || end2 > start1);
464 }
465
466 static bool
467 should_free_lseg(const struct pnfs_layout_range *lseg_range,
468                  const struct pnfs_layout_range *recall_range)
469 {
470         return (recall_range->iomode == IOMODE_ANY ||
471                 lseg_range->iomode == recall_range->iomode) &&
472                pnfs_lseg_range_intersecting(lseg_range, recall_range);
473 }
474
475 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
476                 struct list_head *tmp_list)
477 {
478         if (!atomic_dec_and_test(&lseg->pls_refcount))
479                 return false;
480         pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
481         list_add(&lseg->pls_list, tmp_list);
482         return true;
483 }
484
485 /* Returns 1 if lseg is removed from list, 0 otherwise */
486 static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
487                              struct list_head *tmp_list)
488 {
489         int rv = 0;
490
491         if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
492                 /* Remove the reference keeping the lseg in the
493                  * list.  It will now be removed when all
494                  * outstanding io is finished.
495                  */
496                 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
497                         atomic_read(&lseg->pls_refcount));
498                 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
499                         rv = 1;
500         }
501         return rv;
502 }
503
504 /* Returns count of number of matching invalid lsegs remaining in list
505  * after call.
506  */
507 int
508 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
509                             struct list_head *tmp_list,
510                             const struct pnfs_layout_range *recall_range)
511 {
512         struct pnfs_layout_segment *lseg, *next;
513         int remaining = 0;
514
515         dprintk("%s:Begin lo %p\n", __func__, lo);
516
517         if (list_empty(&lo->plh_segs))
518                 return 0;
519         list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
520                 if (!recall_range ||
521                     should_free_lseg(&lseg->pls_range, recall_range)) {
522                         dprintk("%s: freeing lseg %p iomode %d "
523                                 "offset %llu length %llu\n", __func__,
524                                 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
525                                 lseg->pls_range.length);
526                         if (!mark_lseg_invalid(lseg, tmp_list))
527                                 remaining++;
528                 }
529         dprintk("%s:Return %i\n", __func__, remaining);
530         return remaining;
531 }
532
533 /* note free_me must contain lsegs from a single layout_hdr */
534 void
535 pnfs_free_lseg_list(struct list_head *free_me)
536 {
537         struct pnfs_layout_segment *lseg, *tmp;
538
539         if (list_empty(free_me))
540                 return;
541
542         list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
543                 list_del(&lseg->pls_list);
544                 pnfs_free_lseg(lseg);
545         }
546 }
547
548 void
549 pnfs_destroy_layout(struct nfs_inode *nfsi)
550 {
551         struct pnfs_layout_hdr *lo;
552         LIST_HEAD(tmp_list);
553
554         spin_lock(&nfsi->vfs_inode.i_lock);
555         lo = nfsi->layout;
556         if (lo) {
557                 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
558                 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
559                 pnfs_get_layout_hdr(lo);
560                 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
561                 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
562                 spin_unlock(&nfsi->vfs_inode.i_lock);
563                 pnfs_free_lseg_list(&tmp_list);
564                 pnfs_put_layout_hdr(lo);
565         } else
566                 spin_unlock(&nfsi->vfs_inode.i_lock);
567 }
568 EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
569
570 static bool
571 pnfs_layout_add_bulk_destroy_list(struct inode *inode,
572                 struct list_head *layout_list)
573 {
574         struct pnfs_layout_hdr *lo;
575         bool ret = false;
576
577         spin_lock(&inode->i_lock);
578         lo = NFS_I(inode)->layout;
579         if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
580                 pnfs_get_layout_hdr(lo);
581                 list_add(&lo->plh_bulk_destroy, layout_list);
582                 ret = true;
583         }
584         spin_unlock(&inode->i_lock);
585         return ret;
586 }
587
588 /* Caller must hold rcu_read_lock and clp->cl_lock */
589 static int
590 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
591                 struct nfs_server *server,
592                 struct list_head *layout_list)
593 {
594         struct pnfs_layout_hdr *lo, *next;
595         struct inode *inode;
596
597         list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
598                 inode = igrab(lo->plh_inode);
599                 if (inode == NULL)
600                         continue;
601                 list_del_init(&lo->plh_layouts);
602                 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
603                         continue;
604                 rcu_read_unlock();
605                 spin_unlock(&clp->cl_lock);
606                 iput(inode);
607                 spin_lock(&clp->cl_lock);
608                 rcu_read_lock();
609                 return -EAGAIN;
610         }
611         return 0;
612 }
613
614 static int
615 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
616                 bool is_bulk_recall)
617 {
618         struct pnfs_layout_hdr *lo;
619         struct inode *inode;
620         struct pnfs_layout_range range = {
621                 .iomode = IOMODE_ANY,
622                 .offset = 0,
623                 .length = NFS4_MAX_UINT64,
624         };
625         LIST_HEAD(lseg_list);
626         int ret = 0;
627
628         while (!list_empty(layout_list)) {
629                 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
630                                 plh_bulk_destroy);
631                 dprintk("%s freeing layout for inode %lu\n", __func__,
632                         lo->plh_inode->i_ino);
633                 inode = lo->plh_inode;
634
635                 pnfs_layoutcommit_inode(inode, false);
636
637                 spin_lock(&inode->i_lock);
638                 list_del_init(&lo->plh_bulk_destroy);
639                 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
640                 if (is_bulk_recall)
641                         set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
642                 if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range))
643                         ret = -EAGAIN;
644                 spin_unlock(&inode->i_lock);
645                 pnfs_free_lseg_list(&lseg_list);
646                 /* Free all lsegs that are attached to commit buckets */
647                 nfs_commit_inode(inode, 0);
648                 pnfs_put_layout_hdr(lo);
649                 iput(inode);
650         }
651         return ret;
652 }
653
654 int
655 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
656                 struct nfs_fsid *fsid,
657                 bool is_recall)
658 {
659         struct nfs_server *server;
660         LIST_HEAD(layout_list);
661
662         spin_lock(&clp->cl_lock);
663         rcu_read_lock();
664 restart:
665         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
666                 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
667                         continue;
668                 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
669                                 server,
670                                 &layout_list) != 0)
671                         goto restart;
672         }
673         rcu_read_unlock();
674         spin_unlock(&clp->cl_lock);
675
676         if (list_empty(&layout_list))
677                 return 0;
678         return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
679 }
680
681 int
682 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
683                 bool is_recall)
684 {
685         struct nfs_server *server;
686         LIST_HEAD(layout_list);
687
688         spin_lock(&clp->cl_lock);
689         rcu_read_lock();
690 restart:
691         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
692                 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
693                                         server,
694                                         &layout_list) != 0)
695                         goto restart;
696         }
697         rcu_read_unlock();
698         spin_unlock(&clp->cl_lock);
699
700         if (list_empty(&layout_list))
701                 return 0;
702         return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
703 }
704
705 /*
706  * Called by the state manger to remove all layouts established under an
707  * expired lease.
708  */
709 void
710 pnfs_destroy_all_layouts(struct nfs_client *clp)
711 {
712         nfs4_deviceid_mark_client_invalid(clp);
713         nfs4_deviceid_purge_client(clp);
714
715         pnfs_destroy_layouts_byclid(clp, false);
716 }
717
718 /*
719  * Compare 2 layout stateid sequence ids, to see which is newer,
720  * taking into account wraparound issues.
721  */
722 static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
723 {
724         return (s32)(s1 - s2) > 0;
725 }
726
727 /* update lo->plh_stateid with new if is more recent */
728 void
729 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
730                         bool update_barrier)
731 {
732         u32 oldseq, newseq, new_barrier;
733         int empty = list_empty(&lo->plh_segs);
734
735         oldseq = be32_to_cpu(lo->plh_stateid.seqid);
736         newseq = be32_to_cpu(new->seqid);
737         if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
738                 nfs4_stateid_copy(&lo->plh_stateid, new);
739                 if (update_barrier) {
740                         new_barrier = be32_to_cpu(new->seqid);
741                 } else {
742                         /* Because of wraparound, we want to keep the barrier
743                          * "close" to the current seqids.
744                          */
745                         new_barrier = newseq - atomic_read(&lo->plh_outstanding);
746                 }
747                 if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
748                         lo->plh_barrier = new_barrier;
749         }
750 }
751
752 static bool
753 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
754                 const nfs4_stateid *stateid)
755 {
756         u32 seqid = be32_to_cpu(stateid->seqid);
757
758         return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
759 }
760
761 /* lget is set to 1 if called from inside send_layoutget call chain */
762 static bool
763 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
764 {
765         return lo->plh_block_lgets ||
766                 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
767 }
768
769 int
770 pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
771                               const struct pnfs_layout_range *range,
772                               struct nfs4_state *open_state)
773 {
774         int status = 0;
775
776         dprintk("--> %s\n", __func__);
777         spin_lock(&lo->plh_inode->i_lock);
778         if (pnfs_layoutgets_blocked(lo)) {
779                 status = -EAGAIN;
780         } else if (!nfs4_valid_open_stateid(open_state)) {
781                 status = -EBADF;
782         } else if (list_empty(&lo->plh_segs) ||
783                    test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
784                 int seq;
785
786                 do {
787                         seq = read_seqbegin(&open_state->seqlock);
788                         nfs4_stateid_copy(dst, &open_state->stateid);
789                 } while (read_seqretry(&open_state->seqlock, seq));
790         } else
791                 nfs4_stateid_copy(dst, &lo->plh_stateid);
792         spin_unlock(&lo->plh_inode->i_lock);
793         dprintk("<-- %s\n", __func__);
794         return status;
795 }
796
797 /*
798 * Get layout from server.
799 *    for now, assume that whole file layouts are requested.
800 *    arg->offset: 0
801 *    arg->length: all ones
802 */
803 static struct pnfs_layout_segment *
804 send_layoutget(struct pnfs_layout_hdr *lo,
805            struct nfs_open_context *ctx,
806            const struct pnfs_layout_range *range,
807            gfp_t gfp_flags)
808 {
809         struct inode *ino = lo->plh_inode;
810         struct nfs_server *server = NFS_SERVER(ino);
811         struct nfs4_layoutget *lgp;
812         struct pnfs_layout_segment *lseg;
813         loff_t i_size;
814
815         dprintk("--> %s\n", __func__);
816
817         /*
818          * Synchronously retrieve layout information from server and
819          * store in lseg. If we race with a concurrent seqid morphing
820          * op, then re-send the LAYOUTGET.
821          */
822         do {
823                 lgp = kzalloc(sizeof(*lgp), gfp_flags);
824                 if (lgp == NULL)
825                         return NULL;
826
827                 i_size = i_size_read(ino);
828
829                 lgp->args.minlength = PAGE_CACHE_SIZE;
830                 if (lgp->args.minlength > range->length)
831                         lgp->args.minlength = range->length;
832                 if (range->iomode == IOMODE_READ) {
833                         if (range->offset >= i_size)
834                                 lgp->args.minlength = 0;
835                         else if (i_size - range->offset < lgp->args.minlength)
836                                 lgp->args.minlength = i_size - range->offset;
837                 }
838                 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
839                 pnfs_copy_range(&lgp->args.range, range);
840                 lgp->args.type = server->pnfs_curr_ld->id;
841                 lgp->args.inode = ino;
842                 lgp->args.ctx = get_nfs_open_context(ctx);
843                 lgp->gfp_flags = gfp_flags;
844                 lgp->cred = lo->plh_lc_cred;
845
846                 lseg = nfs4_proc_layoutget(lgp, gfp_flags);
847         } while (lseg == ERR_PTR(-EAGAIN));
848
849         if (IS_ERR(lseg) && !nfs_error_is_fatal(PTR_ERR(lseg)))
850                 lseg = NULL;
851         else
852                 pnfs_layout_clear_fail_bit(lo,
853                                 pnfs_iomode_to_fail_bit(range->iomode));
854
855         return lseg;
856 }
857
858 static void pnfs_clear_layoutcommit(struct inode *inode,
859                 struct list_head *head)
860 {
861         struct nfs_inode *nfsi = NFS_I(inode);
862         struct pnfs_layout_segment *lseg, *tmp;
863
864         if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
865                 return;
866         list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
867                 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
868                         continue;
869                 pnfs_lseg_dec_and_remove_zero(lseg, head);
870         }
871 }
872
873 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
874 {
875         clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
876         smp_mb__after_atomic();
877         wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
878         rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
879 }
880
881 static bool
882 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo)
883 {
884         if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
885                 return false;
886         lo->plh_return_iomode = 0;
887         pnfs_get_layout_hdr(lo);
888         clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
889         return true;
890 }
891
892 static int
893 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
894                        enum pnfs_iomode iomode, bool sync)
895 {
896         struct inode *ino = lo->plh_inode;
897         struct nfs4_layoutreturn *lrp;
898         int status = 0;
899
900         lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
901         if (unlikely(lrp == NULL)) {
902                 status = -ENOMEM;
903                 spin_lock(&ino->i_lock);
904                 pnfs_clear_layoutreturn_waitbit(lo);
905                 spin_unlock(&ino->i_lock);
906                 pnfs_put_layout_hdr(lo);
907                 goto out;
908         }
909
910         nfs4_stateid_copy(&lrp->args.stateid, stateid);
911         lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
912         lrp->args.inode = ino;
913         lrp->args.range.iomode = iomode;
914         lrp->args.range.offset = 0;
915         lrp->args.range.length = NFS4_MAX_UINT64;
916         lrp->args.layout = lo;
917         lrp->clp = NFS_SERVER(ino)->nfs_client;
918         lrp->cred = lo->plh_lc_cred;
919
920         status = nfs4_proc_layoutreturn(lrp, sync);
921 out:
922         dprintk("<-- %s status: %d\n", __func__, status);
923         return status;
924 }
925
926 /* Return true if layoutreturn is needed */
927 static bool
928 pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
929 {
930         struct pnfs_layout_segment *s;
931
932         if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
933                 return false;
934
935         /* Defer layoutreturn until all lsegs are done */
936         list_for_each_entry(s, &lo->plh_segs, pls_list) {
937                 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
938                         return false;
939         }
940
941         return true;
942 }
943
944 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
945 {
946         struct inode *inode= lo->plh_inode;
947
948         if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
949                 return;
950         spin_lock(&inode->i_lock);
951         if (pnfs_layout_need_return(lo)) {
952                 nfs4_stateid stateid;
953                 enum pnfs_iomode iomode;
954                 bool send;
955
956                 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
957                 iomode = lo->plh_return_iomode;
958                 send = pnfs_prepare_layoutreturn(lo);
959                 spin_unlock(&inode->i_lock);
960                 if (send) {
961                         /* Send an async layoutreturn so we dont deadlock */
962                         pnfs_send_layoutreturn(lo, &stateid, iomode, false);
963                 }
964         } else
965                 spin_unlock(&inode->i_lock);
966 }
967
968 /*
969  * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
970  * when the layout segment list is empty.
971  *
972  * Note that a pnfs_layout_hdr can exist with an empty layout segment
973  * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
974  * deviceid is marked invalid.
975  */
976 int
977 _pnfs_return_layout(struct inode *ino)
978 {
979         struct pnfs_layout_hdr *lo = NULL;
980         struct nfs_inode *nfsi = NFS_I(ino);
981         LIST_HEAD(tmp_list);
982         nfs4_stateid stateid;
983         int status = 0, empty;
984         bool send;
985
986         dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
987
988         spin_lock(&ino->i_lock);
989         lo = nfsi->layout;
990         if (!lo) {
991                 spin_unlock(&ino->i_lock);
992                 dprintk("NFS: %s no layout to return\n", __func__);
993                 goto out;
994         }
995         nfs4_stateid_copy(&stateid, &nfsi->layout->plh_stateid);
996         /* Reference matched in nfs4_layoutreturn_release */
997         pnfs_get_layout_hdr(lo);
998         empty = list_empty(&lo->plh_segs);
999         pnfs_clear_layoutcommit(ino, &tmp_list);
1000         pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
1001
1002         if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1003                 struct pnfs_layout_range range = {
1004                         .iomode         = IOMODE_ANY,
1005                         .offset         = 0,
1006                         .length         = NFS4_MAX_UINT64,
1007                 };
1008                 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1009         }
1010
1011         /* Don't send a LAYOUTRETURN if list was initially empty */
1012         if (empty) {
1013                 spin_unlock(&ino->i_lock);
1014                 dprintk("NFS: %s no layout segments to return\n", __func__);
1015                 goto out_put_layout_hdr;
1016         }
1017
1018         set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1019         send = pnfs_prepare_layoutreturn(lo);
1020         spin_unlock(&ino->i_lock);
1021         pnfs_free_lseg_list(&tmp_list);
1022         if (send)
1023                 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1024 out_put_layout_hdr:
1025         pnfs_put_layout_hdr(lo);
1026 out:
1027         dprintk("<-- %s status: %d\n", __func__, status);
1028         return status;
1029 }
1030 EXPORT_SYMBOL_GPL(_pnfs_return_layout);
1031
1032 int
1033 pnfs_commit_and_return_layout(struct inode *inode)
1034 {
1035         struct pnfs_layout_hdr *lo;
1036         int ret;
1037
1038         spin_lock(&inode->i_lock);
1039         lo = NFS_I(inode)->layout;
1040         if (lo == NULL) {
1041                 spin_unlock(&inode->i_lock);
1042                 return 0;
1043         }
1044         pnfs_get_layout_hdr(lo);
1045         /* Block new layoutgets and read/write to ds */
1046         lo->plh_block_lgets++;
1047         spin_unlock(&inode->i_lock);
1048         filemap_fdatawait(inode->i_mapping);
1049         ret = pnfs_layoutcommit_inode(inode, true);
1050         if (ret == 0)
1051                 ret = _pnfs_return_layout(inode);
1052         spin_lock(&inode->i_lock);
1053         lo->plh_block_lgets--;
1054         spin_unlock(&inode->i_lock);
1055         pnfs_put_layout_hdr(lo);
1056         return ret;
1057 }
1058
1059 bool pnfs_roc(struct inode *ino)
1060 {
1061         struct nfs_inode *nfsi = NFS_I(ino);
1062         struct nfs_open_context *ctx;
1063         struct nfs4_state *state;
1064         struct pnfs_layout_hdr *lo;
1065         struct pnfs_layout_segment *lseg, *tmp;
1066         nfs4_stateid stateid;
1067         LIST_HEAD(tmp_list);
1068         bool found = false, layoutreturn = false, roc = false;
1069
1070         spin_lock(&ino->i_lock);
1071         lo = nfsi->layout;
1072         if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
1073                 goto out_noroc;
1074
1075         /* no roc if we hold a delegation */
1076         if (nfs4_check_delegation(ino, FMODE_READ))
1077                 goto out_noroc;
1078
1079         list_for_each_entry(ctx, &nfsi->open_files, list) {
1080                 state = ctx->state;
1081                 /* Don't return layout if there is open file state */
1082                 if (state != NULL && state->state != 0)
1083                         goto out_noroc;
1084         }
1085
1086         nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1087         /* always send layoutreturn if being marked so */
1088         if (test_and_clear_bit(NFS_LAYOUT_RETURN_REQUESTED,
1089                                    &lo->plh_flags))
1090                 layoutreturn = pnfs_prepare_layoutreturn(lo);
1091
1092         list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
1093                 /* If we are sending layoutreturn, invalidate all valid lsegs */
1094                 if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
1095                         mark_lseg_invalid(lseg, &tmp_list);
1096                         found = true;
1097                 }
1098         /* ROC in two conditions:
1099          * 1. there are ROC lsegs
1100          * 2. we don't send layoutreturn
1101          */
1102         if (found && !layoutreturn) {
1103                 /* lo ref dropped in pnfs_roc_release() */
1104                 pnfs_get_layout_hdr(lo);
1105                 roc = true;
1106         }
1107
1108 out_noroc:
1109         spin_unlock(&ino->i_lock);
1110         pnfs_free_lseg_list(&tmp_list);
1111         pnfs_layoutcommit_inode(ino, true);
1112         if (layoutreturn)
1113                 pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1114         return roc;
1115 }
1116
1117 void pnfs_roc_release(struct inode *ino)
1118 {
1119         struct pnfs_layout_hdr *lo;
1120
1121         spin_lock(&ino->i_lock);
1122         lo = NFS_I(ino)->layout;
1123         pnfs_clear_layoutreturn_waitbit(lo);
1124         if (atomic_dec_and_test(&lo->plh_refcount)) {
1125                 pnfs_detach_layout_hdr(lo);
1126                 spin_unlock(&ino->i_lock);
1127                 pnfs_free_layout_hdr(lo);
1128         } else
1129                 spin_unlock(&ino->i_lock);
1130 }
1131
1132 void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
1133 {
1134         struct pnfs_layout_hdr *lo;
1135
1136         spin_lock(&ino->i_lock);
1137         lo = NFS_I(ino)->layout;
1138         pnfs_mark_layout_returned_if_empty(lo);
1139         if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
1140                 lo->plh_barrier = barrier;
1141         spin_unlock(&ino->i_lock);
1142         trace_nfs4_layoutreturn_on_close(ino, 0);
1143 }
1144
1145 void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier)
1146 {
1147         struct nfs_inode *nfsi = NFS_I(ino);
1148         struct pnfs_layout_hdr *lo;
1149         u32 current_seqid;
1150
1151         spin_lock(&ino->i_lock);
1152         lo = nfsi->layout;
1153         current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
1154
1155         /* Since close does not return a layout stateid for use as
1156          * a barrier, we choose the worst-case barrier.
1157          */
1158         *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
1159         spin_unlock(&ino->i_lock);
1160 }
1161
1162 bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1163 {
1164         struct nfs_inode *nfsi = NFS_I(ino);
1165         struct pnfs_layout_hdr *lo;
1166         bool sleep = false;
1167
1168         /* we might not have grabbed lo reference. so need to check under
1169          * i_lock */
1170         spin_lock(&ino->i_lock);
1171         lo = nfsi->layout;
1172         if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
1173                 sleep = true;
1174         spin_unlock(&ino->i_lock);
1175
1176         if (sleep)
1177                 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1178
1179         return sleep;
1180 }
1181
1182 /*
1183  * Compare two layout segments for sorting into layout cache.
1184  * We want to preferentially return RW over RO layouts, so ensure those
1185  * are seen first.
1186  */
1187 static s64
1188 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1189            const struct pnfs_layout_range *l2)
1190 {
1191         s64 d;
1192
1193         /* high offset > low offset */
1194         d = l1->offset - l2->offset;
1195         if (d)
1196                 return d;
1197
1198         /* short length > long length */
1199         d = l2->length - l1->length;
1200         if (d)
1201                 return d;
1202
1203         /* read > read/write */
1204         return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1205 }
1206
1207 static bool
1208 pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1209                 const struct pnfs_layout_range *l2)
1210 {
1211         return pnfs_lseg_range_cmp(l1, l2) > 0;
1212 }
1213
1214 static bool
1215 pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1216                 struct pnfs_layout_segment *old)
1217 {
1218         return false;
1219 }
1220
1221 void
1222 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1223                    struct pnfs_layout_segment *lseg,
1224                    bool (*is_after)(const struct pnfs_layout_range *,
1225                            const struct pnfs_layout_range *),
1226                    bool (*do_merge)(struct pnfs_layout_segment *,
1227                            struct pnfs_layout_segment *),
1228                    struct list_head *free_me)
1229 {
1230         struct pnfs_layout_segment *lp, *tmp;
1231
1232         dprintk("%s:Begin\n", __func__);
1233
1234         list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1235                 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1236                         continue;
1237                 if (do_merge(lseg, lp)) {
1238                         mark_lseg_invalid(lp, free_me);
1239                         continue;
1240                 }
1241                 if (is_after(&lseg->pls_range, &lp->pls_range))
1242                         continue;
1243                 list_add_tail(&lseg->pls_list, &lp->pls_list);
1244                 dprintk("%s: inserted lseg %p "
1245                         "iomode %d offset %llu length %llu before "
1246                         "lp %p iomode %d offset %llu length %llu\n",
1247                         __func__, lseg, lseg->pls_range.iomode,
1248                         lseg->pls_range.offset, lseg->pls_range.length,
1249                         lp, lp->pls_range.iomode, lp->pls_range.offset,
1250                         lp->pls_range.length);
1251                 goto out;
1252         }
1253         list_add_tail(&lseg->pls_list, &lo->plh_segs);
1254         dprintk("%s: inserted lseg %p "
1255                 "iomode %d offset %llu length %llu at tail\n",
1256                 __func__, lseg, lseg->pls_range.iomode,
1257                 lseg->pls_range.offset, lseg->pls_range.length);
1258 out:
1259         pnfs_get_layout_hdr(lo);
1260
1261         dprintk("%s:Return\n", __func__);
1262 }
1263 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1264
1265 static void
1266 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1267                    struct pnfs_layout_segment *lseg,
1268                    struct list_head *free_me)
1269 {
1270         struct inode *inode = lo->plh_inode;
1271         struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1272
1273         if (ld->add_lseg != NULL)
1274                 ld->add_lseg(lo, lseg, free_me);
1275         else
1276                 pnfs_generic_layout_insert_lseg(lo, lseg,
1277                                 pnfs_lseg_range_is_after,
1278                                 pnfs_lseg_no_merge,
1279                                 free_me);
1280 }
1281
1282 static struct pnfs_layout_hdr *
1283 alloc_init_layout_hdr(struct inode *ino,
1284                       struct nfs_open_context *ctx,
1285                       gfp_t gfp_flags)
1286 {
1287         struct pnfs_layout_hdr *lo;
1288
1289         lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1290         if (!lo)
1291                 return NULL;
1292         atomic_set(&lo->plh_refcount, 1);
1293         INIT_LIST_HEAD(&lo->plh_layouts);
1294         INIT_LIST_HEAD(&lo->plh_segs);
1295         INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1296         lo->plh_inode = ino;
1297         lo->plh_lc_cred = get_rpccred(ctx->cred);
1298         return lo;
1299 }
1300
1301 static struct pnfs_layout_hdr *
1302 pnfs_find_alloc_layout(struct inode *ino,
1303                        struct nfs_open_context *ctx,
1304                        gfp_t gfp_flags)
1305 {
1306         struct nfs_inode *nfsi = NFS_I(ino);
1307         struct pnfs_layout_hdr *new = NULL;
1308
1309         dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1310
1311         if (nfsi->layout != NULL)
1312                 goto out_existing;
1313         spin_unlock(&ino->i_lock);
1314         new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1315         spin_lock(&ino->i_lock);
1316
1317         if (likely(nfsi->layout == NULL)) {     /* Won the race? */
1318                 nfsi->layout = new;
1319                 return new;
1320         } else if (new != NULL)
1321                 pnfs_free_layout_hdr(new);
1322 out_existing:
1323         pnfs_get_layout_hdr(nfsi->layout);
1324         return nfsi->layout;
1325 }
1326
1327 /*
1328  * iomode matching rules:
1329  * iomode       lseg    match
1330  * -----        -----   -----
1331  * ANY          READ    true
1332  * ANY          RW      true
1333  * RW           READ    false
1334  * RW           RW      true
1335  * READ         READ    true
1336  * READ         RW      true
1337  */
1338 static bool
1339 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1340                  const struct pnfs_layout_range *range)
1341 {
1342         struct pnfs_layout_range range1;
1343
1344         if ((range->iomode == IOMODE_RW &&
1345              ls_range->iomode != IOMODE_RW) ||
1346             !pnfs_lseg_range_intersecting(ls_range, range))
1347                 return 0;
1348
1349         /* range1 covers only the first byte in the range */
1350         range1 = *range;
1351         range1.length = 1;
1352         return pnfs_lseg_range_contained(ls_range, &range1);
1353 }
1354
1355 /*
1356  * lookup range in layout
1357  */
1358 static struct pnfs_layout_segment *
1359 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1360                 struct pnfs_layout_range *range)
1361 {
1362         struct pnfs_layout_segment *lseg, *ret = NULL;
1363
1364         dprintk("%s:Begin\n", __func__);
1365
1366         list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1367                 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1368                     !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
1369                     pnfs_lseg_range_match(&lseg->pls_range, range)) {
1370                         ret = pnfs_get_lseg(lseg);
1371                         break;
1372                 }
1373         }
1374
1375         dprintk("%s:Return lseg %p ref %d\n",
1376                 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1377         return ret;
1378 }
1379
1380 /*
1381  * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1382  * to the MDS or over pNFS
1383  *
1384  * The nfs_inode read_io and write_io fields are cumulative counters reset
1385  * when there are no layout segments. Note that in pnfs_update_layout iomode
1386  * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1387  * WRITE request.
1388  *
1389  * A return of true means use MDS I/O.
1390  *
1391  * From rfc 5661:
1392  * If a file's size is smaller than the file size threshold, data accesses
1393  * SHOULD be sent to the metadata server.  If an I/O request has a length that
1394  * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1395  * server.  If both file size and I/O size are provided, the client SHOULD
1396  * reach or exceed  both thresholds before sending its read or write
1397  * requests to the data server.
1398  */
1399 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1400                                      struct inode *ino, int iomode)
1401 {
1402         struct nfs4_threshold *t = ctx->mdsthreshold;
1403         struct nfs_inode *nfsi = NFS_I(ino);
1404         loff_t fsize = i_size_read(ino);
1405         bool size = false, size_set = false, io = false, io_set = false, ret = false;
1406
1407         if (t == NULL)
1408                 return ret;
1409
1410         dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1411                 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1412
1413         switch (iomode) {
1414         case IOMODE_READ:
1415                 if (t->bm & THRESHOLD_RD) {
1416                         dprintk("%s fsize %llu\n", __func__, fsize);
1417                         size_set = true;
1418                         if (fsize < t->rd_sz)
1419                                 size = true;
1420                 }
1421                 if (t->bm & THRESHOLD_RD_IO) {
1422                         dprintk("%s nfsi->read_io %llu\n", __func__,
1423                                 nfsi->read_io);
1424                         io_set = true;
1425                         if (nfsi->read_io < t->rd_io_sz)
1426                                 io = true;
1427                 }
1428                 break;
1429         case IOMODE_RW:
1430                 if (t->bm & THRESHOLD_WR) {
1431                         dprintk("%s fsize %llu\n", __func__, fsize);
1432                         size_set = true;
1433                         if (fsize < t->wr_sz)
1434                                 size = true;
1435                 }
1436                 if (t->bm & THRESHOLD_WR_IO) {
1437                         dprintk("%s nfsi->write_io %llu\n", __func__,
1438                                 nfsi->write_io);
1439                         io_set = true;
1440                         if (nfsi->write_io < t->wr_io_sz)
1441                                 io = true;
1442                 }
1443                 break;
1444         }
1445         if (size_set && io_set) {
1446                 if (size && io)
1447                         ret = true;
1448         } else if (size || io)
1449                 ret = true;
1450
1451         dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1452         return ret;
1453 }
1454
1455 static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1456 {
1457         /*
1458          * send layoutcommit as it can hold up layoutreturn due to lseg
1459          * reference
1460          */
1461         pnfs_layoutcommit_inode(lo->plh_inode, false);
1462         return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1463                                    nfs_wait_bit_killable,
1464                                    TASK_UNINTERRUPTIBLE);
1465 }
1466
1467 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1468 {
1469         unsigned long *bitlock = &lo->plh_flags;
1470
1471         clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1472         smp_mb__after_atomic();
1473         wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1474 }
1475
1476 /*
1477  * Layout segment is retreived from the server if not cached.
1478  * The appropriate layout segment is referenced and returned to the caller.
1479  */
1480 struct pnfs_layout_segment *
1481 pnfs_update_layout(struct inode *ino,
1482                    struct nfs_open_context *ctx,
1483                    loff_t pos,
1484                    u64 count,
1485                    enum pnfs_iomode iomode,
1486                    gfp_t gfp_flags)
1487 {
1488         struct pnfs_layout_range arg = {
1489                 .iomode = iomode,
1490                 .offset = pos,
1491                 .length = count,
1492         };
1493         unsigned pg_offset;
1494         struct nfs_server *server = NFS_SERVER(ino);
1495         struct nfs_client *clp = server->nfs_client;
1496         struct pnfs_layout_hdr *lo;
1497         struct pnfs_layout_segment *lseg = NULL;
1498         bool first;
1499
1500         if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1501                 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1502                                  PNFS_UPDATE_LAYOUT_NO_PNFS);
1503                 goto out;
1504         }
1505
1506         if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
1507                 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1508                                  PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
1509                 goto out;
1510         }
1511
1512         if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1513                 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1514                                  PNFS_UPDATE_LAYOUT_MDSTHRESH);
1515                 goto out;
1516         }
1517
1518 lookup_again:
1519         first = false;
1520         spin_lock(&ino->i_lock);
1521         lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1522         if (lo == NULL) {
1523                 spin_unlock(&ino->i_lock);
1524                 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1525                                  PNFS_UPDATE_LAYOUT_NOMEM);
1526                 goto out;
1527         }
1528
1529         /* Do we even need to bother with this? */
1530         if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1531                 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1532                                  PNFS_UPDATE_LAYOUT_BULK_RECALL);
1533                 dprintk("%s matches recall, use MDS\n", __func__);
1534                 goto out_unlock;
1535         }
1536
1537         /* if LAYOUTGET already failed once we don't try again */
1538         if (pnfs_layout_io_test_failed(lo, iomode)) {
1539                 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1540                                  PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1541                 goto out_unlock;
1542         }
1543
1544         first = list_empty(&lo->plh_segs);
1545         if (first) {
1546                 /* The first layoutget for the file. Need to serialize per
1547                  * RFC 5661 Errata 3208.
1548                  */
1549                 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1550                                      &lo->plh_flags)) {
1551                         spin_unlock(&ino->i_lock);
1552                         wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1553                                     TASK_UNINTERRUPTIBLE);
1554                         pnfs_put_layout_hdr(lo);
1555                         goto lookup_again;
1556                 }
1557         } else {
1558                 /* Check to see if the layout for the given range
1559                  * already exists
1560                  */
1561                 lseg = pnfs_find_lseg(lo, &arg);
1562                 if (lseg) {
1563                         trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1564                                         PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1565                         goto out_unlock;
1566                 }
1567         }
1568
1569         /*
1570          * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1571          * for LAYOUTRETURN even if first is true.
1572          */
1573         if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1574                 spin_unlock(&ino->i_lock);
1575                 dprintk("%s wait for layoutreturn\n", __func__);
1576                 if (pnfs_prepare_to_retry_layoutget(lo)) {
1577                         if (first)
1578                                 pnfs_clear_first_layoutget(lo);
1579                         pnfs_put_layout_hdr(lo);
1580                         dprintk("%s retrying\n", __func__);
1581                         goto lookup_again;
1582                 }
1583                 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1584                                 PNFS_UPDATE_LAYOUT_RETURN);
1585                 goto out_put_layout_hdr;
1586         }
1587
1588         if (pnfs_layoutgets_blocked(lo)) {
1589                 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1590                                 PNFS_UPDATE_LAYOUT_BLOCKED);
1591                 goto out_unlock;
1592         }
1593         atomic_inc(&lo->plh_outstanding);
1594         spin_unlock(&ino->i_lock);
1595
1596         if (list_empty(&lo->plh_layouts)) {
1597                 /* The lo must be on the clp list if there is any
1598                  * chance of a CB_LAYOUTRECALL(FILE) coming in.
1599                  */
1600                 spin_lock(&clp->cl_lock);
1601                 if (list_empty(&lo->plh_layouts))
1602                         list_add_tail(&lo->plh_layouts, &server->layouts);
1603                 spin_unlock(&clp->cl_lock);
1604         }
1605
1606         pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1607         if (pg_offset) {
1608                 arg.offset -= pg_offset;
1609                 arg.length += pg_offset;
1610         }
1611         if (arg.length != NFS4_MAX_UINT64)
1612                 arg.length = PAGE_CACHE_ALIGN(arg.length);
1613
1614         lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1615         atomic_dec(&lo->plh_outstanding);
1616         trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1617                                  PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
1618 out_put_layout_hdr:
1619         if (first)
1620                 pnfs_clear_first_layoutget(lo);
1621         pnfs_put_layout_hdr(lo);
1622 out:
1623         dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1624                         "(%s, offset: %llu, length: %llu)\n",
1625                         __func__, ino->i_sb->s_id,
1626                         (unsigned long long)NFS_FILEID(ino),
1627                         IS_ERR_OR_NULL(lseg) ? "not found" : "found",
1628                         iomode==IOMODE_RW ?  "read/write" : "read-only",
1629                         (unsigned long long)pos,
1630                         (unsigned long long)count);
1631         return lseg;
1632 out_unlock:
1633         spin_unlock(&ino->i_lock);
1634         goto out_put_layout_hdr;
1635 }
1636 EXPORT_SYMBOL_GPL(pnfs_update_layout);
1637
1638 static bool
1639 pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
1640 {
1641         switch (range->iomode) {
1642         case IOMODE_READ:
1643         case IOMODE_RW:
1644                 break;
1645         default:
1646                 return false;
1647         }
1648         if (range->offset == NFS4_MAX_UINT64)
1649                 return false;
1650         if (range->length == 0)
1651                 return false;
1652         if (range->length != NFS4_MAX_UINT64 &&
1653             range->length > NFS4_MAX_UINT64 - range->offset)
1654                 return false;
1655         return true;
1656 }
1657
1658 struct pnfs_layout_segment *
1659 pnfs_layout_process(struct nfs4_layoutget *lgp)
1660 {
1661         struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1662         struct nfs4_layoutget_res *res = &lgp->res;
1663         struct pnfs_layout_segment *lseg;
1664         struct inode *ino = lo->plh_inode;
1665         LIST_HEAD(free_me);
1666         int status = -EINVAL;
1667
1668         if (!pnfs_sanity_check_layout_range(&res->range))
1669                 goto out;
1670
1671         /* Inject layout blob into I/O device driver */
1672         lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1673         if (!lseg || IS_ERR(lseg)) {
1674                 if (!lseg)
1675                         status = -ENOMEM;
1676                 else
1677                         status = PTR_ERR(lseg);
1678                 dprintk("%s: Could not allocate layout: error %d\n",
1679                        __func__, status);
1680                 goto out;
1681         }
1682
1683         init_lseg(lo, lseg);
1684         lseg->pls_range = res->range;
1685
1686         spin_lock(&ino->i_lock);
1687         if (pnfs_layoutgets_blocked(lo)) {
1688                 dprintk("%s forget reply due to state\n", __func__);
1689                 goto out_forget_reply;
1690         }
1691
1692         if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
1693                 /* existing state ID, make sure the sequence number matches. */
1694                 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
1695                         dprintk("%s forget reply due to sequence\n", __func__);
1696                         status = -EAGAIN;
1697                         goto out_forget_reply;
1698                 }
1699                 pnfs_set_layout_stateid(lo, &res->stateid, false);
1700         } else {
1701                 /*
1702                  * We got an entirely new state ID.  Mark all segments for the
1703                  * inode invalid, and don't bother validating the stateid
1704                  * sequence number.
1705                  */
1706                 pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL);
1707
1708                 nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
1709                 lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
1710         }
1711
1712         clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1713
1714         pnfs_get_lseg(lseg);
1715         pnfs_layout_insert_lseg(lo, lseg, &free_me);
1716
1717         if (res->return_on_close)
1718                 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1719
1720         spin_unlock(&ino->i_lock);
1721         pnfs_free_lseg_list(&free_me);
1722         return lseg;
1723 out:
1724         return ERR_PTR(status);
1725
1726 out_forget_reply:
1727         spin_unlock(&ino->i_lock);
1728         lseg->pls_layout = lo;
1729         NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1730         goto out;
1731 }
1732
1733 static void
1734 pnfs_set_plh_return_iomode(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode)
1735 {
1736         if (lo->plh_return_iomode == iomode)
1737                 return;
1738         if (lo->plh_return_iomode != 0)
1739                 iomode = IOMODE_ANY;
1740         lo->plh_return_iomode = iomode;
1741 }
1742
1743 int
1744 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
1745                                 struct list_head *tmp_list,
1746                                 const struct pnfs_layout_range *return_range)
1747 {
1748         struct pnfs_layout_segment *lseg, *next;
1749         int remaining = 0;
1750
1751         dprintk("%s:Begin lo %p\n", __func__, lo);
1752
1753         if (list_empty(&lo->plh_segs))
1754                 return 0;
1755
1756         assert_spin_locked(&lo->plh_inode->i_lock);
1757
1758         list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
1759                 if (should_free_lseg(&lseg->pls_range, return_range)) {
1760                         dprintk("%s: marking lseg %p iomode %d "
1761                                 "offset %llu length %llu\n", __func__,
1762                                 lseg, lseg->pls_range.iomode,
1763                                 lseg->pls_range.offset,
1764                                 lseg->pls_range.length);
1765                         set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1766                         pnfs_set_plh_return_iomode(lo, return_range->iomode);
1767                         if (!mark_lseg_invalid(lseg, tmp_list))
1768                                 remaining++;
1769                         set_bit(NFS_LAYOUT_RETURN_REQUESTED,
1770                                         &lo->plh_flags);
1771                 }
1772         return remaining;
1773 }
1774
1775 void pnfs_error_mark_layout_for_return(struct inode *inode,
1776                                        struct pnfs_layout_segment *lseg)
1777 {
1778         struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
1779         struct pnfs_layout_range range = {
1780                 .iomode = lseg->pls_range.iomode,
1781                 .offset = 0,
1782                 .length = NFS4_MAX_UINT64,
1783         };
1784         LIST_HEAD(free_me);
1785         bool return_now = false;
1786
1787         spin_lock(&inode->i_lock);
1788         pnfs_set_plh_return_iomode(lo, range.iomode);
1789         /*
1790          * mark all matching lsegs so that we are sure to have no live
1791          * segments at hand when sending layoutreturn. See pnfs_put_lseg()
1792          * for how it works.
1793          */
1794         if (!pnfs_mark_matching_lsegs_return(lo, &free_me, &range)) {
1795                 nfs4_stateid stateid;
1796                 enum pnfs_iomode iomode = lo->plh_return_iomode;
1797
1798                 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1799                 return_now = pnfs_prepare_layoutreturn(lo);
1800                 spin_unlock(&inode->i_lock);
1801                 if (return_now)
1802                         pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1803         } else {
1804                 spin_unlock(&inode->i_lock);
1805                 nfs_commit_inode(inode, 0);
1806         }
1807         pnfs_free_lseg_list(&free_me);
1808 }
1809 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
1810
1811 void
1812 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1813 {
1814         u64 rd_size = req->wb_bytes;
1815
1816         if (pgio->pg_lseg == NULL) {
1817                 if (pgio->pg_dreq == NULL)
1818                         rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
1819                 else
1820                         rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1821
1822                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1823                                                    req->wb_context,
1824                                                    req_offset(req),
1825                                                    rd_size,
1826                                                    IOMODE_READ,
1827                                                    GFP_KERNEL);
1828                 if (IS_ERR(pgio->pg_lseg)) {
1829                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1830                         pgio->pg_lseg = NULL;
1831                         return;
1832                 }
1833         }
1834         /* If no lseg, fall back to read through mds */
1835         if (pgio->pg_lseg == NULL)
1836                 nfs_pageio_reset_read_mds(pgio);
1837
1838 }
1839 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1840
1841 void
1842 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
1843                            struct nfs_page *req, u64 wb_size)
1844 {
1845         if (pgio->pg_lseg == NULL) {
1846                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1847                                                    req->wb_context,
1848                                                    req_offset(req),
1849                                                    wb_size,
1850                                                    IOMODE_RW,
1851                                                    GFP_NOFS);
1852                 if (IS_ERR(pgio->pg_lseg)) {
1853                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1854                         pgio->pg_lseg = NULL;
1855                         return;
1856                 }
1857         }
1858         /* If no lseg, fall back to write through mds */
1859         if (pgio->pg_lseg == NULL)
1860                 nfs_pageio_reset_write_mds(pgio);
1861 }
1862 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1863
1864 void
1865 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
1866 {
1867         if (desc->pg_lseg) {
1868                 pnfs_put_lseg(desc->pg_lseg);
1869                 desc->pg_lseg = NULL;
1870         }
1871 }
1872 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
1873
1874 /*
1875  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1876  * of bytes (maximum @req->wb_bytes) that can be coalesced.
1877  */
1878 size_t
1879 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
1880                      struct nfs_page *prev, struct nfs_page *req)
1881 {
1882         unsigned int size;
1883         u64 seg_end, req_start, seg_left;
1884
1885         size = nfs_generic_pg_test(pgio, prev, req);
1886         if (!size)
1887                 return 0;
1888
1889         /*
1890          * 'size' contains the number of bytes left in the current page (up
1891          * to the original size asked for in @req->wb_bytes).
1892          *
1893          * Calculate how many bytes are left in the layout segment
1894          * and if there are less bytes than 'size', return that instead.
1895          *
1896          * Please also note that 'end_offset' is actually the offset of the
1897          * first byte that lies outside the pnfs_layout_range. FIXME?
1898          *
1899          */
1900         if (pgio->pg_lseg) {
1901                 seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
1902                                      pgio->pg_lseg->pls_range.length);
1903                 req_start = req_offset(req);
1904                 WARN_ON_ONCE(req_start >= seg_end);
1905                 /* start of request is past the last byte of this segment */
1906                 if (req_start >= seg_end) {
1907                         /* reference the new lseg */
1908                         if (pgio->pg_ops->pg_cleanup)
1909                                 pgio->pg_ops->pg_cleanup(pgio);
1910                         if (pgio->pg_ops->pg_init)
1911                                 pgio->pg_ops->pg_init(pgio, req);
1912                         return 0;
1913                 }
1914
1915                 /* adjust 'size' iff there are fewer bytes left in the
1916                  * segment than what nfs_generic_pg_test returned */
1917                 seg_left = seg_end - req_start;
1918                 if (seg_left < size)
1919                         size = (unsigned int)seg_left;
1920         }
1921
1922         return size;
1923 }
1924 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1925
1926 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
1927 {
1928         struct nfs_pageio_descriptor pgio;
1929
1930         /* Resend all requests through the MDS */
1931         nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
1932                               hdr->completion_ops);
1933         set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
1934         return nfs_pageio_resend(&pgio, hdr);
1935 }
1936 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1937
1938 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1939 {
1940
1941         dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1942         if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1943             PNFS_LAYOUTRET_ON_ERROR) {
1944                 pnfs_return_layout(hdr->inode);
1945         }
1946         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1947                 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1948 }
1949
1950 /*
1951  * Called by non rpc-based layout drivers
1952  */
1953 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
1954 {
1955         if (likely(!hdr->pnfs_error)) {
1956                 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1957                                 hdr->mds_offset + hdr->res.count);
1958                 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1959         }
1960         trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
1961         if (unlikely(hdr->pnfs_error))
1962                 pnfs_ld_handle_write_error(hdr);
1963         hdr->mds_ops->rpc_release(hdr);
1964 }
1965 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1966
1967 static void
1968 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1969                 struct nfs_pgio_header *hdr)
1970 {
1971         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1972
1973         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1974                 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
1975                 nfs_pageio_reset_write_mds(desc);
1976                 mirror->pg_recoalesce = 1;
1977         }
1978         nfs_pgio_data_destroy(hdr);
1979         hdr->release(hdr);
1980 }
1981
1982 static enum pnfs_try_status
1983 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
1984                         const struct rpc_call_ops *call_ops,
1985                         struct pnfs_layout_segment *lseg,
1986                         int how)
1987 {
1988         struct inode *inode = hdr->inode;
1989         enum pnfs_try_status trypnfs;
1990         struct nfs_server *nfss = NFS_SERVER(inode);
1991
1992         hdr->mds_ops = call_ops;
1993
1994         dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1995                 inode->i_ino, hdr->args.count, hdr->args.offset, how);
1996         trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
1997         if (trypnfs != PNFS_NOT_ATTEMPTED)
1998                 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
1999         dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2000         return trypnfs;
2001 }
2002
2003 static void
2004 pnfs_do_write(struct nfs_pageio_descriptor *desc,
2005               struct nfs_pgio_header *hdr, int how)
2006 {
2007         const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2008         struct pnfs_layout_segment *lseg = desc->pg_lseg;
2009         enum pnfs_try_status trypnfs;
2010
2011         trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2012         if (trypnfs == PNFS_NOT_ATTEMPTED)
2013                 pnfs_write_through_mds(desc, hdr);
2014 }
2015
2016 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2017 {
2018         pnfs_put_lseg(hdr->lseg);
2019         nfs_pgio_header_free(hdr);
2020 }
2021
2022 int
2023 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2024 {
2025         struct nfs_pgio_header *hdr;
2026         int ret;
2027
2028         hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2029         if (!hdr) {
2030                 desc->pg_error = -ENOMEM;
2031                 return desc->pg_error;
2032         }
2033         nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2034
2035         hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2036         ret = nfs_generic_pgio(desc, hdr);
2037         if (!ret)
2038                 pnfs_do_write(desc, hdr, desc->pg_ioflags);
2039
2040         return ret;
2041 }
2042 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2043
2044 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2045 {
2046         struct nfs_pageio_descriptor pgio;
2047
2048         /* Resend all requests through the MDS */
2049         nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2050         return nfs_pageio_resend(&pgio, hdr);
2051 }
2052 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2053
2054 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2055 {
2056         dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2057         if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2058             PNFS_LAYOUTRET_ON_ERROR) {
2059                 pnfs_return_layout(hdr->inode);
2060         }
2061         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2062                 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2063 }
2064
2065 /*
2066  * Called by non rpc-based layout drivers
2067  */
2068 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2069 {
2070         if (likely(!hdr->pnfs_error)) {
2071                 __nfs4_read_done_cb(hdr);
2072                 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2073         }
2074         trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2075         if (unlikely(hdr->pnfs_error))
2076                 pnfs_ld_handle_read_error(hdr);
2077         hdr->mds_ops->rpc_release(hdr);
2078 }
2079 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2080
2081 static void
2082 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2083                 struct nfs_pgio_header *hdr)
2084 {
2085         struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2086
2087         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2088                 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2089                 nfs_pageio_reset_read_mds(desc);
2090                 mirror->pg_recoalesce = 1;
2091         }
2092         nfs_pgio_data_destroy(hdr);
2093         hdr->release(hdr);
2094 }
2095
2096 /*
2097  * Call the appropriate parallel I/O subsystem read function.
2098  */
2099 static enum pnfs_try_status
2100 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2101                        const struct rpc_call_ops *call_ops,
2102                        struct pnfs_layout_segment *lseg)
2103 {
2104         struct inode *inode = hdr->inode;
2105         struct nfs_server *nfss = NFS_SERVER(inode);
2106         enum pnfs_try_status trypnfs;
2107
2108         hdr->mds_ops = call_ops;
2109
2110         dprintk("%s: Reading ino:%lu %u@%llu\n",
2111                 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
2112
2113         trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2114         if (trypnfs != PNFS_NOT_ATTEMPTED)
2115                 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
2116         dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2117         return trypnfs;
2118 }
2119
2120 /* Resend all requests through pnfs. */
2121 int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2122 {
2123         struct nfs_pageio_descriptor pgio;
2124
2125         nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops);
2126         return nfs_pageio_resend(&pgio, hdr);
2127 }
2128 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2129
2130 static void
2131 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2132 {
2133         const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2134         struct pnfs_layout_segment *lseg = desc->pg_lseg;
2135         enum pnfs_try_status trypnfs;
2136         int err = 0;
2137
2138         trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2139         if (trypnfs == PNFS_TRY_AGAIN)
2140                 err = pnfs_read_resend_pnfs(hdr);
2141         if (trypnfs == PNFS_NOT_ATTEMPTED || err)
2142                 pnfs_read_through_mds(desc, hdr);
2143 }
2144
2145 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2146 {
2147         pnfs_put_lseg(hdr->lseg);
2148         nfs_pgio_header_free(hdr);
2149 }
2150
2151 int
2152 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2153 {
2154         struct nfs_pgio_header *hdr;
2155         int ret;
2156
2157         hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2158         if (!hdr) {
2159                 desc->pg_error = -ENOMEM;
2160                 return desc->pg_error;
2161         }
2162         nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2163         hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2164         ret = nfs_generic_pgio(desc, hdr);
2165         if (!ret)
2166                 pnfs_do_read(desc, hdr);
2167         return ret;
2168 }
2169 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2170
2171 static void pnfs_clear_layoutcommitting(struct inode *inode)
2172 {
2173         unsigned long *bitlock = &NFS_I(inode)->flags;
2174
2175         clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2176         smp_mb__after_atomic();
2177         wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2178 }
2179
2180 /*
2181  * There can be multiple RW segments.
2182  */
2183 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
2184 {
2185         struct pnfs_layout_segment *lseg;
2186
2187         list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2188                 if (lseg->pls_range.iomode == IOMODE_RW &&
2189                     test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2190                         list_add(&lseg->pls_lc_list, listp);
2191         }
2192 }
2193
2194 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2195 {
2196         struct pnfs_layout_segment *lseg, *tmp;
2197
2198         /* Matched by references in pnfs_set_layoutcommit */
2199         list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2200                 list_del_init(&lseg->pls_lc_list);
2201                 pnfs_put_lseg(lseg);
2202         }
2203
2204         pnfs_clear_layoutcommitting(inode);
2205 }
2206
2207 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2208 {
2209         pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
2210 }
2211 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2212
2213 void
2214 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2215                 loff_t end_pos)
2216 {
2217         struct nfs_inode *nfsi = NFS_I(inode);
2218         bool mark_as_dirty = false;
2219
2220         spin_lock(&inode->i_lock);
2221         if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2222                 nfsi->layout->plh_lwb = end_pos;
2223                 mark_as_dirty = true;
2224                 dprintk("%s: Set layoutcommit for inode %lu ",
2225                         __func__, inode->i_ino);
2226         } else if (end_pos > nfsi->layout->plh_lwb)
2227                 nfsi->layout->plh_lwb = end_pos;
2228         if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2229                 /* references matched in nfs4_layoutcommit_release */
2230                 pnfs_get_lseg(lseg);
2231         }
2232         spin_unlock(&inode->i_lock);
2233         dprintk("%s: lseg %p end_pos %llu\n",
2234                 __func__, lseg, nfsi->layout->plh_lwb);
2235
2236         /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2237          * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2238         if (mark_as_dirty)
2239                 mark_inode_dirty_sync(inode);
2240 }
2241 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2242
2243 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2244 {
2245         struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2246
2247         if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2248                 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
2249         pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
2250 }
2251
2252 /*
2253  * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2254  * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2255  * data to disk to allow the server to recover the data if it crashes.
2256  * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2257  * is off, and a COMMIT is sent to a data server, or
2258  * if WRITEs to a data server return NFS_DATA_SYNC.
2259  */
2260 int
2261 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
2262 {
2263         struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2264         struct nfs4_layoutcommit_data *data;
2265         struct nfs_inode *nfsi = NFS_I(inode);
2266         loff_t end_pos;
2267         int status;
2268
2269         if (!pnfs_layoutcommit_outstanding(inode))
2270                 return 0;
2271
2272         dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
2273
2274         status = -EAGAIN;
2275         if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2276                 if (!sync)
2277                         goto out;
2278                 status = wait_on_bit_lock_action(&nfsi->flags,
2279                                 NFS_INO_LAYOUTCOMMITTING,
2280                                 nfs_wait_bit_killable,
2281                                 TASK_KILLABLE);
2282                 if (status)
2283                         goto out;
2284         }
2285
2286         status = -ENOMEM;
2287         /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2288         data = kzalloc(sizeof(*data), GFP_NOFS);
2289         if (!data)
2290                 goto clear_layoutcommitting;
2291
2292         status = 0;
2293         spin_lock(&inode->i_lock);
2294         if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2295                 goto out_unlock;
2296
2297         INIT_LIST_HEAD(&data->lseg_list);
2298         pnfs_list_write_lseg(inode, &data->lseg_list);
2299
2300         end_pos = nfsi->layout->plh_lwb;
2301
2302         nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
2303         spin_unlock(&inode->i_lock);
2304
2305         data->args.inode = inode;
2306         data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
2307         nfs_fattr_init(&data->fattr);
2308         data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2309         data->res.fattr = &data->fattr;
2310         data->args.lastbytewritten = end_pos - 1;
2311         data->res.server = NFS_SERVER(inode);
2312
2313         if (ld->prepare_layoutcommit) {
2314                 status = ld->prepare_layoutcommit(&data->args);
2315                 if (status) {
2316                         put_rpccred(data->cred);
2317                         spin_lock(&inode->i_lock);
2318                         set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2319                         if (end_pos > nfsi->layout->plh_lwb)
2320                                 nfsi->layout->plh_lwb = end_pos;
2321                         goto out_unlock;
2322                 }
2323         }
2324
2325
2326         status = nfs4_proc_layoutcommit(data, sync);
2327 out:
2328         if (status)
2329                 mark_inode_dirty_sync(inode);
2330         dprintk("<-- %s status %d\n", __func__, status);
2331         return status;
2332 out_unlock:
2333         spin_unlock(&inode->i_lock);
2334         kfree(data);
2335 clear_layoutcommitting:
2336         pnfs_clear_layoutcommitting(inode);
2337         goto out;
2338 }
2339 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2340
2341 int
2342 pnfs_generic_sync(struct inode *inode, bool datasync)
2343 {
2344         return pnfs_layoutcommit_inode(inode, true);
2345 }
2346 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2347
2348 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2349 {
2350         struct nfs4_threshold *thp;
2351
2352         thp = kzalloc(sizeof(*thp), GFP_NOFS);
2353         if (!thp) {
2354                 dprintk("%s mdsthreshold allocation failed\n", __func__);
2355                 return NULL;
2356         }
2357         return thp;
2358 }
2359
2360 #if IS_ENABLED(CONFIG_NFS_V4_2)
2361 int
2362 pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
2363 {
2364         struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2365         struct nfs_server *server = NFS_SERVER(inode);
2366         struct nfs_inode *nfsi = NFS_I(inode);
2367         struct nfs42_layoutstat_data *data;
2368         struct pnfs_layout_hdr *hdr;
2369         int status = 0;
2370
2371         if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2372                 goto out;
2373
2374         if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2375                 goto out;
2376
2377         if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2378                 goto out;
2379
2380         spin_lock(&inode->i_lock);
2381         if (!NFS_I(inode)->layout) {
2382                 spin_unlock(&inode->i_lock);
2383                 goto out;
2384         }
2385         hdr = NFS_I(inode)->layout;
2386         pnfs_get_layout_hdr(hdr);
2387         spin_unlock(&inode->i_lock);
2388
2389         data = kzalloc(sizeof(*data), gfp_flags);
2390         if (!data) {
2391                 status = -ENOMEM;
2392                 goto out_put;
2393         }
2394
2395         data->args.fh = NFS_FH(inode);
2396         data->args.inode = inode;
2397         nfs4_stateid_copy(&data->args.stateid, &hdr->plh_stateid);
2398         status = ld->prepare_layoutstats(&data->args);
2399         if (status)
2400                 goto out_free;
2401
2402         status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2403
2404 out:
2405         dprintk("%s returns %d\n", __func__, status);
2406         return status;
2407
2408 out_free:
2409         kfree(data);
2410 out_put:
2411         pnfs_put_layout_hdr(hdr);
2412         smp_mb__before_atomic();
2413         clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2414         smp_mb__after_atomic();
2415         goto out;
2416 }
2417 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
2418 #endif
2419
2420 unsigned int layoutstats_timer;
2421 module_param(layoutstats_timer, uint, 0644);
2422 EXPORT_SYMBOL_GPL(layoutstats_timer);