1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30#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
38#define NFSDBG_FACILITY NFSDBG_PNFS
39#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
40
41
42
43
44
45
46static DEFINE_SPINLOCK(pnfs_spinlock);
47
48
49
50
51static LIST_HEAD(pnfs_modules_tbl);
52
53
54static struct pnfs_layoutdriver_type *
55find_pnfs_driver_locked(u32 id)
56{
57 struct pnfs_layoutdriver_type *local;
58
59 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
60 if (local->id == id)
61 goto out;
62 local = NULL;
63out:
64 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
65 return local;
66}
67
68static struct pnfs_layoutdriver_type *
69find_pnfs_driver(u32 id)
70{
71 struct pnfs_layoutdriver_type *local;
72
73 spin_lock(&pnfs_spinlock);
74 local = find_pnfs_driver_locked(id);
75 if (local != NULL && !try_module_get(local->owner)) {
76 dprintk("%s: Could not grab reference on module\n", __func__);
77 local = NULL;
78 }
79 spin_unlock(&pnfs_spinlock);
80 return local;
81}
82
83void
84unset_pnfs_layoutdriver(struct nfs_server *nfss)
85{
86 if (nfss->pnfs_curr_ld) {
87 if (nfss->pnfs_curr_ld->clear_layoutdriver)
88 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
89
90 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
91 nfs4_deviceid_purge_client(nfss->nfs_client);
92 module_put(nfss->pnfs_curr_ld->owner);
93 }
94 nfss->pnfs_curr_ld = NULL;
95}
96
97
98
99
100
101
102
103void
104set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
105 u32 id)
106{
107 struct pnfs_layoutdriver_type *ld_type = NULL;
108
109 if (id == 0)
110 goto out_no_driver;
111 if (!(server->nfs_client->cl_exchange_flags &
112 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
113 printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
114 __func__, id, server->nfs_client->cl_exchange_flags);
115 goto out_no_driver;
116 }
117 ld_type = find_pnfs_driver(id);
118 if (!ld_type) {
119 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
120 ld_type = find_pnfs_driver(id);
121 if (!ld_type) {
122 dprintk("%s: No pNFS module found for %u.\n",
123 __func__, id);
124 goto out_no_driver;
125 }
126 }
127 server->pnfs_curr_ld = ld_type;
128 if (ld_type->set_layoutdriver
129 && ld_type->set_layoutdriver(server, mntfh)) {
130 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
131 "driver %u.\n", __func__, id);
132 module_put(ld_type->owner);
133 goto out_no_driver;
134 }
135
136 atomic_inc(&server->nfs_client->cl_mds_count);
137
138 dprintk("%s: pNFS module for %u set\n", __func__, id);
139 return;
140
141out_no_driver:
142 dprintk("%s: Using NFSv4 I/O\n", __func__);
143 server->pnfs_curr_ld = NULL;
144}
145
146int
147pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
148{
149 int status = -EINVAL;
150 struct pnfs_layoutdriver_type *tmp;
151
152 if (ld_type->id == 0) {
153 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
154 return status;
155 }
156 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
157 printk(KERN_ERR "NFS: %s Layout driver must provide "
158 "alloc_lseg and free_lseg.\n", __func__);
159 return status;
160 }
161
162 spin_lock(&pnfs_spinlock);
163 tmp = find_pnfs_driver_locked(ld_type->id);
164 if (!tmp) {
165 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
166 status = 0;
167 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
168 ld_type->name);
169 } else {
170 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
171 __func__, ld_type->id);
172 }
173 spin_unlock(&pnfs_spinlock);
174
175 return status;
176}
177EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
178
179void
180pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
181{
182 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
183 spin_lock(&pnfs_spinlock);
184 list_del(&ld_type->pnfs_tblid);
185 spin_unlock(&pnfs_spinlock);
186}
187EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
188
189
190
191
192
193
194void
195pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
196{
197 atomic_inc(&lo->plh_refcount);
198}
199
200static struct pnfs_layout_hdr *
201pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
202{
203 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
204 return ld->alloc_layout_hdr(ino, gfp_flags);
205}
206
207static void
208pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
209{
210 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
211 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
212
213 if (!list_empty(&lo->plh_layouts)) {
214 struct nfs_client *clp = server->nfs_client;
215
216 spin_lock(&clp->cl_lock);
217 list_del_init(&lo->plh_layouts);
218 spin_unlock(&clp->cl_lock);
219 }
220 put_rpccred(lo->plh_lc_cred);
221 return ld->free_layout_hdr(lo);
222}
223
224static void
225pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
226{
227 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
228 dprintk("%s: freeing layout cache %p\n", __func__, lo);
229 nfsi->layout = NULL;
230
231 nfsi->write_io = 0;
232 nfsi->read_io = 0;
233}
234
235void
236pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
237{
238 struct inode *inode = lo->plh_inode;
239
240 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
241 pnfs_detach_layout_hdr(lo);
242 spin_unlock(&inode->i_lock);
243 pnfs_free_layout_hdr(lo);
244 }
245}
246
247static int
248pnfs_iomode_to_fail_bit(u32 iomode)
249{
250 return iomode == IOMODE_RW ?
251 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
252}
253
254static void
255pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
256{
257 lo->plh_retry_timestamp = jiffies;
258 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
259 atomic_inc(&lo->plh_refcount);
260}
261
262static void
263pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
264{
265 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
266 atomic_dec(&lo->plh_refcount);
267}
268
269static void
270pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
271{
272 struct inode *inode = lo->plh_inode;
273 struct pnfs_layout_range range = {
274 .iomode = iomode,
275 .offset = 0,
276 .length = NFS4_MAX_UINT64,
277 };
278 LIST_HEAD(head);
279
280 spin_lock(&inode->i_lock);
281 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
282 pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
283 spin_unlock(&inode->i_lock);
284 pnfs_free_lseg_list(&head);
285 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
286 iomode == IOMODE_RW ? "RW" : "READ");
287}
288
289static bool
290pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
291{
292 unsigned long start, end;
293 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
294
295 if (test_bit(fail_bit, &lo->plh_flags) == 0)
296 return false;
297 end = jiffies;
298 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
299 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
300
301 pnfs_layout_clear_fail_bit(lo, fail_bit);
302 return false;
303 }
304 return true;
305}
306
307static void
308init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
309{
310 INIT_LIST_HEAD(&lseg->pls_list);
311 INIT_LIST_HEAD(&lseg->pls_lc_list);
312 atomic_set(&lseg->pls_refcount, 1);
313 smp_mb();
314 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
315 lseg->pls_layout = lo;
316}
317
318static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
319{
320 struct inode *ino = lseg->pls_layout->plh_inode;
321
322 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
323}
324
325static void
326pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
327 struct pnfs_layout_segment *lseg)
328{
329 struct inode *inode = lo->plh_inode;
330
331 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
332 list_del_init(&lseg->pls_list);
333
334 atomic_dec(&lo->plh_refcount);
335 if (list_empty(&lo->plh_segs))
336 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
337 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
338}
339
340void
341pnfs_put_lseg(struct pnfs_layout_segment *lseg)
342{
343 struct pnfs_layout_hdr *lo;
344 struct inode *inode;
345
346 if (!lseg)
347 return;
348
349 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
350 atomic_read(&lseg->pls_refcount),
351 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
352 lo = lseg->pls_layout;
353 inode = lo->plh_inode;
354 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
355 pnfs_get_layout_hdr(lo);
356 pnfs_layout_remove_lseg(lo, lseg);
357 spin_unlock(&inode->i_lock);
358 pnfs_free_lseg(lseg);
359 pnfs_put_layout_hdr(lo);
360 }
361}
362EXPORT_SYMBOL_GPL(pnfs_put_lseg);
363
364static void pnfs_free_lseg_async_work(struct work_struct *work)
365{
366 struct pnfs_layout_segment *lseg;
367 struct pnfs_layout_hdr *lo;
368
369 lseg = container_of(work, struct pnfs_layout_segment, pls_work);
370 lo = lseg->pls_layout;
371
372 pnfs_free_lseg(lseg);
373 pnfs_put_layout_hdr(lo);
374}
375
376static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
377{
378 INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
379 schedule_work(&lseg->pls_work);
380}
381
382void
383pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
384{
385 if (!lseg)
386 return;
387
388 assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);
389
390 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
391 atomic_read(&lseg->pls_refcount),
392 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
393 if (atomic_dec_and_test(&lseg->pls_refcount)) {
394 struct pnfs_layout_hdr *lo = lseg->pls_layout;
395 pnfs_get_layout_hdr(lo);
396 pnfs_layout_remove_lseg(lo, lseg);
397 pnfs_free_lseg_async(lseg);
398 }
399}
400EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
401
402static u64
403end_offset(u64 start, u64 len)
404{
405 u64 end;
406
407 end = start + len;
408 return end >= start ? end : NFS4_MAX_UINT64;
409}
410
411
412
413
414
415
416
417
418static bool
419pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
420 const struct pnfs_layout_range *l2)
421{
422 u64 start1 = l1->offset;
423 u64 end1 = end_offset(start1, l1->length);
424 u64 start2 = l2->offset;
425 u64 end2 = end_offset(start2, l2->length);
426
427 return (start1 <= start2) && (end1 >= end2);
428}
429
430
431
432
433
434
435
436
437static bool
438pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
439 const struct pnfs_layout_range *l2)
440{
441 u64 start1 = l1->offset;
442 u64 end1 = end_offset(start1, l1->length);
443 u64 start2 = l2->offset;
444 u64 end2 = end_offset(start2, l2->length);
445
446 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
447 (end2 == NFS4_MAX_UINT64 || end2 > start1);
448}
449
450static bool
451should_free_lseg(const struct pnfs_layout_range *lseg_range,
452 const struct pnfs_layout_range *recall_range)
453{
454 return (recall_range->iomode == IOMODE_ANY ||
455 lseg_range->iomode == recall_range->iomode) &&
456 pnfs_lseg_range_intersecting(lseg_range, recall_range);
457}
458
459static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
460 struct list_head *tmp_list)
461{
462 if (!atomic_dec_and_test(&lseg->pls_refcount))
463 return false;
464 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
465 list_add(&lseg->pls_list, tmp_list);
466 return true;
467}
468
469
470static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
471 struct list_head *tmp_list)
472{
473 int rv = 0;
474
475 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
476
477
478
479
480 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
481 atomic_read(&lseg->pls_refcount));
482 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
483 rv = 1;
484 }
485 return rv;
486}
487
488
489
490
491int
492pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
493 struct list_head *tmp_list,
494 struct pnfs_layout_range *recall_range)
495{
496 struct pnfs_layout_segment *lseg, *next;
497 int invalid = 0, removed = 0;
498
499 dprintk("%s:Begin lo %p\n", __func__, lo);
500
501 if (list_empty(&lo->plh_segs))
502 return 0;
503 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
504 if (!recall_range ||
505 should_free_lseg(&lseg->pls_range, recall_range)) {
506 dprintk("%s: freeing lseg %p iomode %d "
507 "offset %llu length %llu\n", __func__,
508 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
509 lseg->pls_range.length);
510 invalid++;
511 removed += mark_lseg_invalid(lseg, tmp_list);
512 }
513 dprintk("%s:Return %i\n", __func__, invalid - removed);
514 return invalid - removed;
515}
516
517
518void
519pnfs_free_lseg_list(struct list_head *free_me)
520{
521 struct pnfs_layout_segment *lseg, *tmp;
522
523 if (list_empty(free_me))
524 return;
525
526 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
527 list_del(&lseg->pls_list);
528 pnfs_free_lseg(lseg);
529 }
530}
531
532void
533pnfs_destroy_layout(struct nfs_inode *nfsi)
534{
535 struct pnfs_layout_hdr *lo;
536 LIST_HEAD(tmp_list);
537
538 spin_lock(&nfsi->vfs_inode.i_lock);
539 lo = nfsi->layout;
540 if (lo) {
541 lo->plh_block_lgets++;
542 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
543 pnfs_get_layout_hdr(lo);
544 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
545 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
546 spin_unlock(&nfsi->vfs_inode.i_lock);
547 pnfs_free_lseg_list(&tmp_list);
548 pnfs_put_layout_hdr(lo);
549 } else
550 spin_unlock(&nfsi->vfs_inode.i_lock);
551}
552EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
553
554static bool
555pnfs_layout_add_bulk_destroy_list(struct inode *inode,
556 struct list_head *layout_list)
557{
558 struct pnfs_layout_hdr *lo;
559 bool ret = false;
560
561 spin_lock(&inode->i_lock);
562 lo = NFS_I(inode)->layout;
563 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
564 pnfs_get_layout_hdr(lo);
565 list_add(&lo->plh_bulk_destroy, layout_list);
566 ret = true;
567 }
568 spin_unlock(&inode->i_lock);
569 return ret;
570}
571
572
573static int
574pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
575 struct nfs_server *server,
576 struct list_head *layout_list)
577{
578 struct pnfs_layout_hdr *lo, *next;
579 struct inode *inode;
580
581 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
582 inode = igrab(lo->plh_inode);
583 if (inode == NULL)
584 continue;
585 list_del_init(&lo->plh_layouts);
586 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
587 continue;
588 rcu_read_unlock();
589 spin_unlock(&clp->cl_lock);
590 iput(inode);
591 spin_lock(&clp->cl_lock);
592 rcu_read_lock();
593 return -EAGAIN;
594 }
595 return 0;
596}
597
598static int
599pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
600 bool is_bulk_recall)
601{
602 struct pnfs_layout_hdr *lo;
603 struct inode *inode;
604 struct pnfs_layout_range range = {
605 .iomode = IOMODE_ANY,
606 .offset = 0,
607 .length = NFS4_MAX_UINT64,
608 };
609 LIST_HEAD(lseg_list);
610 int ret = 0;
611
612 while (!list_empty(layout_list)) {
613 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
614 plh_bulk_destroy);
615 dprintk("%s freeing layout for inode %lu\n", __func__,
616 lo->plh_inode->i_ino);
617 inode = lo->plh_inode;
618
619 pnfs_layoutcommit_inode(inode, false);
620
621 spin_lock(&inode->i_lock);
622 list_del_init(&lo->plh_bulk_destroy);
623 lo->plh_block_lgets++;
624 if (is_bulk_recall)
625 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
626 if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range))
627 ret = -EAGAIN;
628 spin_unlock(&inode->i_lock);
629 pnfs_free_lseg_list(&lseg_list);
630 pnfs_put_layout_hdr(lo);
631 iput(inode);
632 }
633 return ret;
634}
635
636int
637pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
638 struct nfs_fsid *fsid,
639 bool is_recall)
640{
641 struct nfs_server *server;
642 LIST_HEAD(layout_list);
643
644 spin_lock(&clp->cl_lock);
645 rcu_read_lock();
646restart:
647 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
648 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
649 continue;
650 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
651 server,
652 &layout_list) != 0)
653 goto restart;
654 }
655 rcu_read_unlock();
656 spin_unlock(&clp->cl_lock);
657
658 if (list_empty(&layout_list))
659 return 0;
660 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
661}
662
663int
664pnfs_destroy_layouts_byclid(struct nfs_client *clp,
665 bool is_recall)
666{
667 struct nfs_server *server;
668 LIST_HEAD(layout_list);
669
670 spin_lock(&clp->cl_lock);
671 rcu_read_lock();
672restart:
673 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
674 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
675 server,
676 &layout_list) != 0)
677 goto restart;
678 }
679 rcu_read_unlock();
680 spin_unlock(&clp->cl_lock);
681
682 if (list_empty(&layout_list))
683 return 0;
684 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
685}
686
687
688
689
690
691void
692pnfs_destroy_all_layouts(struct nfs_client *clp)
693{
694 nfs4_deviceid_mark_client_invalid(clp);
695 nfs4_deviceid_purge_client(clp);
696
697 pnfs_destroy_layouts_byclid(clp, false);
698}
699
700
701
702
703
704static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
705{
706 return (s32)(s1 - s2) > 0;
707}
708
709
710void
711pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
712 bool update_barrier)
713{
714 u32 oldseq, newseq, new_barrier;
715 int empty = list_empty(&lo->plh_segs);
716
717 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
718 newseq = be32_to_cpu(new->seqid);
719 if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
720 nfs4_stateid_copy(&lo->plh_stateid, new);
721 if (update_barrier) {
722 new_barrier = be32_to_cpu(new->seqid);
723 } else {
724
725
726
727 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
728 }
729 if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
730 lo->plh_barrier = new_barrier;
731 }
732}
733
734static bool
735pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
736 const nfs4_stateid *stateid)
737{
738 u32 seqid = be32_to_cpu(stateid->seqid);
739
740 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
741}
742
743
744static bool
745pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget)
746{
747 return lo->plh_block_lgets ||
748 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
749 (list_empty(&lo->plh_segs) &&
750 (atomic_read(&lo->plh_outstanding) > lget));
751}
752
753int
754pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
755 struct nfs4_state *open_state)
756{
757 int status = 0;
758
759 dprintk("--> %s\n", __func__);
760 spin_lock(&lo->plh_inode->i_lock);
761 if (pnfs_layoutgets_blocked(lo, 1)) {
762 status = -EAGAIN;
763 } else if (!nfs4_valid_open_stateid(open_state)) {
764 status = -EBADF;
765 } else if (list_empty(&lo->plh_segs) ||
766 test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
767 int seq;
768
769 do {
770 seq = read_seqbegin(&open_state->seqlock);
771 nfs4_stateid_copy(dst, &open_state->stateid);
772 } while (read_seqretry(&open_state->seqlock, seq));
773 } else
774 nfs4_stateid_copy(dst, &lo->plh_stateid);
775 spin_unlock(&lo->plh_inode->i_lock);
776 dprintk("<-- %s\n", __func__);
777 return status;
778}
779
780
781
782
783
784
785
786static struct pnfs_layout_segment *
787send_layoutget(struct pnfs_layout_hdr *lo,
788 struct nfs_open_context *ctx,
789 struct pnfs_layout_range *range,
790 gfp_t gfp_flags)
791{
792 struct inode *ino = lo->plh_inode;
793 struct nfs_server *server = NFS_SERVER(ino);
794 struct nfs4_layoutget *lgp;
795 struct pnfs_layout_segment *lseg;
796
797 dprintk("--> %s\n", __func__);
798
799 lgp = kzalloc(sizeof(*lgp), gfp_flags);
800 if (lgp == NULL)
801 return NULL;
802
803 lgp->args.minlength = PAGE_CACHE_SIZE;
804 if (lgp->args.minlength > range->length)
805 lgp->args.minlength = range->length;
806 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
807 lgp->args.range = *range;
808 lgp->args.type = server->pnfs_curr_ld->id;
809 lgp->args.inode = ino;
810 lgp->args.ctx = get_nfs_open_context(ctx);
811 lgp->gfp_flags = gfp_flags;
812 lgp->cred = lo->plh_lc_cred;
813
814
815
816
817 lseg = nfs4_proc_layoutget(lgp, gfp_flags);
818 if (IS_ERR(lseg)) {
819 switch (PTR_ERR(lseg)) {
820 case -ENOMEM:
821 case -ERESTARTSYS:
822 break;
823 default:
824
825 pnfs_layout_io_set_failed(lo, range->iomode);
826 }
827 return NULL;
828 }
829
830 return lseg;
831}
832
833static void pnfs_clear_layoutcommit(struct inode *inode,
834 struct list_head *head)
835{
836 struct nfs_inode *nfsi = NFS_I(inode);
837 struct pnfs_layout_segment *lseg, *tmp;
838
839 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
840 return;
841 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
842 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
843 continue;
844 pnfs_lseg_dec_and_remove_zero(lseg, head);
845 }
846}
847
848
849
850
851
852
853
854
855
856int
857_pnfs_return_layout(struct inode *ino)
858{
859 struct pnfs_layout_hdr *lo = NULL;
860 struct nfs_inode *nfsi = NFS_I(ino);
861 LIST_HEAD(tmp_list);
862 struct nfs4_layoutreturn *lrp;
863 nfs4_stateid stateid;
864 int status = 0, empty;
865
866 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
867
868 spin_lock(&ino->i_lock);
869 lo = nfsi->layout;
870 if (!lo) {
871 spin_unlock(&ino->i_lock);
872 dprintk("NFS: %s no layout to return\n", __func__);
873 goto out;
874 }
875 stateid = nfsi->layout->plh_stateid;
876
877 pnfs_get_layout_hdr(lo);
878 empty = list_empty(&lo->plh_segs);
879 pnfs_clear_layoutcommit(ino, &tmp_list);
880 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
881
882 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
883 struct pnfs_layout_range range = {
884 .iomode = IOMODE_ANY,
885 .offset = 0,
886 .length = NFS4_MAX_UINT64,
887 };
888 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
889 }
890
891
892 if (empty) {
893 spin_unlock(&ino->i_lock);
894 pnfs_put_layout_hdr(lo);
895 dprintk("NFS: %s no layout segments to return\n", __func__);
896 goto out;
897 }
898
899 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
900 lo->plh_block_lgets++;
901 spin_unlock(&ino->i_lock);
902 pnfs_free_lseg_list(&tmp_list);
903
904 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
905 if (unlikely(lrp == NULL)) {
906 status = -ENOMEM;
907 spin_lock(&ino->i_lock);
908 lo->plh_block_lgets--;
909 spin_unlock(&ino->i_lock);
910 pnfs_put_layout_hdr(lo);
911 goto out;
912 }
913
914 lrp->args.stateid = stateid;
915 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
916 lrp->args.inode = ino;
917 lrp->args.layout = lo;
918 lrp->clp = NFS_SERVER(ino)->nfs_client;
919 lrp->cred = lo->plh_lc_cred;
920
921 status = nfs4_proc_layoutreturn(lrp);
922out:
923 dprintk("<-- %s status: %d\n", __func__, status);
924 return status;
925}
926EXPORT_SYMBOL_GPL(_pnfs_return_layout);
927
928int
929pnfs_commit_and_return_layout(struct inode *inode)
930{
931 struct pnfs_layout_hdr *lo;
932 int ret;
933
934 spin_lock(&inode->i_lock);
935 lo = NFS_I(inode)->layout;
936 if (lo == NULL) {
937 spin_unlock(&inode->i_lock);
938 return 0;
939 }
940 pnfs_get_layout_hdr(lo);
941
942 lo->plh_block_lgets++;
943 spin_unlock(&inode->i_lock);
944 filemap_fdatawait(inode->i_mapping);
945 ret = pnfs_layoutcommit_inode(inode, true);
946 if (ret == 0)
947 ret = _pnfs_return_layout(inode);
948 spin_lock(&inode->i_lock);
949 lo->plh_block_lgets--;
950 spin_unlock(&inode->i_lock);
951 pnfs_put_layout_hdr(lo);
952 return ret;
953}
954
955bool pnfs_roc(struct inode *ino)
956{
957 struct pnfs_layout_hdr *lo;
958 struct pnfs_layout_segment *lseg, *tmp;
959 LIST_HEAD(tmp_list);
960 bool found = false;
961
962 spin_lock(&ino->i_lock);
963 lo = NFS_I(ino)->layout;
964 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
965 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
966 goto out_nolayout;
967 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
968 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
969 mark_lseg_invalid(lseg, &tmp_list);
970 found = true;
971 }
972 if (!found)
973 goto out_nolayout;
974 lo->plh_block_lgets++;
975 pnfs_get_layout_hdr(lo);
976 spin_unlock(&ino->i_lock);
977 pnfs_free_lseg_list(&tmp_list);
978 return true;
979
980out_nolayout:
981 spin_unlock(&ino->i_lock);
982 return false;
983}
984
985void pnfs_roc_release(struct inode *ino)
986{
987 struct pnfs_layout_hdr *lo;
988
989 spin_lock(&ino->i_lock);
990 lo = NFS_I(ino)->layout;
991 lo->plh_block_lgets--;
992 if (atomic_dec_and_test(&lo->plh_refcount)) {
993 pnfs_detach_layout_hdr(lo);
994 spin_unlock(&ino->i_lock);
995 pnfs_free_layout_hdr(lo);
996 } else
997 spin_unlock(&ino->i_lock);
998}
999
1000void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
1001{
1002 struct pnfs_layout_hdr *lo;
1003
1004 spin_lock(&ino->i_lock);
1005 lo = NFS_I(ino)->layout;
1006 if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
1007 lo->plh_barrier = barrier;
1008 spin_unlock(&ino->i_lock);
1009}
1010
1011bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
1012{
1013 struct nfs_inode *nfsi = NFS_I(ino);
1014 struct pnfs_layout_hdr *lo;
1015 struct pnfs_layout_segment *lseg;
1016 u32 current_seqid;
1017 bool found = false;
1018
1019 spin_lock(&ino->i_lock);
1020 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
1021 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
1022 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1023 found = true;
1024 goto out;
1025 }
1026 lo = nfsi->layout;
1027 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
1028
1029
1030
1031
1032 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
1033out:
1034 spin_unlock(&ino->i_lock);
1035 return found;
1036}
1037
1038
1039
1040
1041
1042
1043static s64
1044pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1045 const struct pnfs_layout_range *l2)
1046{
1047 s64 d;
1048
1049
1050 d = l1->offset - l2->offset;
1051 if (d)
1052 return d;
1053
1054
1055 d = l2->length - l1->length;
1056 if (d)
1057 return d;
1058
1059
1060 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1061}
1062
1063static void
1064pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1065 struct pnfs_layout_segment *lseg)
1066{
1067 struct pnfs_layout_segment *lp;
1068
1069 dprintk("%s:Begin\n", __func__);
1070
1071 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
1072 if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0)
1073 continue;
1074 list_add_tail(&lseg->pls_list, &lp->pls_list);
1075 dprintk("%s: inserted lseg %p "
1076 "iomode %d offset %llu length %llu before "
1077 "lp %p iomode %d offset %llu length %llu\n",
1078 __func__, lseg, lseg->pls_range.iomode,
1079 lseg->pls_range.offset, lseg->pls_range.length,
1080 lp, lp->pls_range.iomode, lp->pls_range.offset,
1081 lp->pls_range.length);
1082 goto out;
1083 }
1084 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1085 dprintk("%s: inserted lseg %p "
1086 "iomode %d offset %llu length %llu at tail\n",
1087 __func__, lseg, lseg->pls_range.iomode,
1088 lseg->pls_range.offset, lseg->pls_range.length);
1089out:
1090 pnfs_get_layout_hdr(lo);
1091
1092 dprintk("%s:Return\n", __func__);
1093}
1094
1095static struct pnfs_layout_hdr *
1096alloc_init_layout_hdr(struct inode *ino,
1097 struct nfs_open_context *ctx,
1098 gfp_t gfp_flags)
1099{
1100 struct pnfs_layout_hdr *lo;
1101
1102 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1103 if (!lo)
1104 return NULL;
1105 atomic_set(&lo->plh_refcount, 1);
1106 INIT_LIST_HEAD(&lo->plh_layouts);
1107 INIT_LIST_HEAD(&lo->plh_segs);
1108 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1109 lo->plh_inode = ino;
1110 lo->plh_lc_cred = get_rpccred(ctx->cred);
1111 return lo;
1112}
1113
1114static struct pnfs_layout_hdr *
1115pnfs_find_alloc_layout(struct inode *ino,
1116 struct nfs_open_context *ctx,
1117 gfp_t gfp_flags)
1118{
1119 struct nfs_inode *nfsi = NFS_I(ino);
1120 struct pnfs_layout_hdr *new = NULL;
1121
1122 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1123
1124 if (nfsi->layout != NULL)
1125 goto out_existing;
1126 spin_unlock(&ino->i_lock);
1127 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1128 spin_lock(&ino->i_lock);
1129
1130 if (likely(nfsi->layout == NULL)) {
1131 nfsi->layout = new;
1132 return new;
1133 } else if (new != NULL)
1134 pnfs_free_layout_hdr(new);
1135out_existing:
1136 pnfs_get_layout_hdr(nfsi->layout);
1137 return nfsi->layout;
1138}
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151static bool
1152pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1153 const struct pnfs_layout_range *range)
1154{
1155 struct pnfs_layout_range range1;
1156
1157 if ((range->iomode == IOMODE_RW &&
1158 ls_range->iomode != IOMODE_RW) ||
1159 !pnfs_lseg_range_intersecting(ls_range, range))
1160 return 0;
1161
1162
1163 range1 = *range;
1164 range1.length = 1;
1165 return pnfs_lseg_range_contained(ls_range, &range1);
1166}
1167
1168
1169
1170
1171static struct pnfs_layout_segment *
1172pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1173 struct pnfs_layout_range *range)
1174{
1175 struct pnfs_layout_segment *lseg, *ret = NULL;
1176
1177 dprintk("%s:Begin\n", __func__);
1178
1179 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1180 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1181 pnfs_lseg_range_match(&lseg->pls_range, range)) {
1182 ret = pnfs_get_lseg(lseg);
1183 break;
1184 }
1185 if (lseg->pls_range.offset > range->offset)
1186 break;
1187 }
1188
1189 dprintk("%s:Return lseg %p ref %d\n",
1190 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1191 return ret;
1192}
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1214 struct inode *ino, int iomode)
1215{
1216 struct nfs4_threshold *t = ctx->mdsthreshold;
1217 struct nfs_inode *nfsi = NFS_I(ino);
1218 loff_t fsize = i_size_read(ino);
1219 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1220
1221 if (t == NULL)
1222 return ret;
1223
1224 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1225 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1226
1227 switch (iomode) {
1228 case IOMODE_READ:
1229 if (t->bm & THRESHOLD_RD) {
1230 dprintk("%s fsize %llu\n", __func__, fsize);
1231 size_set = true;
1232 if (fsize < t->rd_sz)
1233 size = true;
1234 }
1235 if (t->bm & THRESHOLD_RD_IO) {
1236 dprintk("%s nfsi->read_io %llu\n", __func__,
1237 nfsi->read_io);
1238 io_set = true;
1239 if (nfsi->read_io < t->rd_io_sz)
1240 io = true;
1241 }
1242 break;
1243 case IOMODE_RW:
1244 if (t->bm & THRESHOLD_WR) {
1245 dprintk("%s fsize %llu\n", __func__, fsize);
1246 size_set = true;
1247 if (fsize < t->wr_sz)
1248 size = true;
1249 }
1250 if (t->bm & THRESHOLD_WR_IO) {
1251 dprintk("%s nfsi->write_io %llu\n", __func__,
1252 nfsi->write_io);
1253 io_set = true;
1254 if (nfsi->write_io < t->wr_io_sz)
1255 io = true;
1256 }
1257 break;
1258 }
1259 if (size_set && io_set) {
1260 if (size && io)
1261 ret = true;
1262 } else if (size || io)
1263 ret = true;
1264
1265 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1266 return ret;
1267}
1268
1269
1270
1271
1272
1273struct pnfs_layout_segment *
1274pnfs_update_layout(struct inode *ino,
1275 struct nfs_open_context *ctx,
1276 loff_t pos,
1277 u64 count,
1278 enum pnfs_iomode iomode,
1279 gfp_t gfp_flags)
1280{
1281 struct pnfs_layout_range arg = {
1282 .iomode = iomode,
1283 .offset = pos,
1284 .length = count,
1285 };
1286 unsigned pg_offset;
1287 struct nfs_server *server = NFS_SERVER(ino);
1288 struct nfs_client *clp = server->nfs_client;
1289 struct pnfs_layout_hdr *lo;
1290 struct pnfs_layout_segment *lseg = NULL;
1291 bool first;
1292
1293 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
1294 goto out;
1295
1296 if (pnfs_within_mdsthreshold(ctx, ino, iomode))
1297 goto out;
1298
1299 spin_lock(&ino->i_lock);
1300 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1301 if (lo == NULL) {
1302 spin_unlock(&ino->i_lock);
1303 goto out;
1304 }
1305
1306
1307 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1308 dprintk("%s matches recall, use MDS\n", __func__);
1309 goto out_unlock;
1310 }
1311
1312
1313 if (pnfs_layout_io_test_failed(lo, iomode))
1314 goto out_unlock;
1315
1316
1317 lseg = pnfs_find_lseg(lo, &arg);
1318 if (lseg)
1319 goto out_unlock;
1320
1321 if (pnfs_layoutgets_blocked(lo, 0))
1322 goto out_unlock;
1323 atomic_inc(&lo->plh_outstanding);
1324
1325 first = list_empty(&lo->plh_layouts) ? true : false;
1326 spin_unlock(&ino->i_lock);
1327
1328 if (first) {
1329
1330
1331
1332 spin_lock(&clp->cl_lock);
1333 list_add_tail(&lo->plh_layouts, &server->layouts);
1334 spin_unlock(&clp->cl_lock);
1335 }
1336
1337 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1338 if (pg_offset) {
1339 arg.offset -= pg_offset;
1340 arg.length += pg_offset;
1341 }
1342 if (arg.length != NFS4_MAX_UINT64)
1343 arg.length = PAGE_CACHE_ALIGN(arg.length);
1344
1345 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1346 atomic_dec(&lo->plh_outstanding);
1347out_put_layout_hdr:
1348 pnfs_put_layout_hdr(lo);
1349out:
1350 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1351 "(%s, offset: %llu, length: %llu)\n",
1352 __func__, ino->i_sb->s_id,
1353 (unsigned long long)NFS_FILEID(ino),
1354 lseg == NULL ? "not found" : "found",
1355 iomode==IOMODE_RW ? "read/write" : "read-only",
1356 (unsigned long long)pos,
1357 (unsigned long long)count);
1358 return lseg;
1359out_unlock:
1360 spin_unlock(&ino->i_lock);
1361 goto out_put_layout_hdr;
1362}
1363EXPORT_SYMBOL_GPL(pnfs_update_layout);
1364
1365struct pnfs_layout_segment *
1366pnfs_layout_process(struct nfs4_layoutget *lgp)
1367{
1368 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1369 struct nfs4_layoutget_res *res = &lgp->res;
1370 struct pnfs_layout_segment *lseg;
1371 struct inode *ino = lo->plh_inode;
1372 LIST_HEAD(free_me);
1373 int status = 0;
1374
1375
1376 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1377 if (!lseg || IS_ERR(lseg)) {
1378 if (!lseg)
1379 status = -ENOMEM;
1380 else
1381 status = PTR_ERR(lseg);
1382 dprintk("%s: Could not allocate layout: error %d\n",
1383 __func__, status);
1384 goto out;
1385 }
1386
1387 init_lseg(lo, lseg);
1388 lseg->pls_range = res->range;
1389
1390 spin_lock(&ino->i_lock);
1391 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1392 dprintk("%s forget reply due to recall\n", __func__);
1393 goto out_forget_reply;
1394 }
1395
1396 if (pnfs_layoutgets_blocked(lo, 1)) {
1397 dprintk("%s forget reply due to state\n", __func__);
1398 goto out_forget_reply;
1399 }
1400
1401 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
1402
1403 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
1404 dprintk("%s forget reply due to sequence\n", __func__);
1405 goto out_forget_reply;
1406 }
1407 pnfs_set_layout_stateid(lo, &res->stateid, false);
1408 } else {
1409
1410
1411
1412
1413
1414 pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL);
1415
1416 nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
1417 lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
1418 }
1419
1420 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1421
1422 pnfs_get_lseg(lseg);
1423 pnfs_layout_insert_lseg(lo, lseg);
1424
1425 if (res->return_on_close) {
1426 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1427 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
1428 }
1429
1430 spin_unlock(&ino->i_lock);
1431 pnfs_free_lseg_list(&free_me);
1432 return lseg;
1433out:
1434 return ERR_PTR(status);
1435
1436out_forget_reply:
1437 spin_unlock(&ino->i_lock);
1438 lseg->pls_layout = lo;
1439 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1440 goto out;
1441}
1442
1443void
1444pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1445{
1446 u64 rd_size = req->wb_bytes;
1447
1448 WARN_ON_ONCE(pgio->pg_lseg != NULL);
1449
1450 if (pgio->pg_dreq == NULL)
1451 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
1452 else
1453 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1454
1455 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1456 req->wb_context,
1457 req_offset(req),
1458 rd_size,
1459 IOMODE_READ,
1460 GFP_KERNEL);
1461
1462 if (pgio->pg_lseg == NULL)
1463 nfs_pageio_reset_read_mds(pgio);
1464
1465}
1466EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1467
1468void
1469pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
1470 struct nfs_page *req, u64 wb_size)
1471{
1472 WARN_ON_ONCE(pgio->pg_lseg != NULL);
1473
1474 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1475 req->wb_context,
1476 req_offset(req),
1477 wb_size,
1478 IOMODE_RW,
1479 GFP_NOFS);
1480
1481 if (pgio->pg_lseg == NULL)
1482 nfs_pageio_reset_write_mds(pgio);
1483}
1484EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1485
1486
1487
1488
1489
1490size_t
1491pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1492 struct nfs_page *req)
1493{
1494 unsigned int size;
1495 u64 seg_end, req_start, seg_left;
1496
1497 size = nfs_generic_pg_test(pgio, prev, req);
1498 if (!size)
1499 return 0;
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512 if (pgio->pg_lseg) {
1513 seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
1514 pgio->pg_lseg->pls_range.length);
1515 req_start = req_offset(req);
1516 WARN_ON_ONCE(req_start > seg_end);
1517
1518 if (req_start >= seg_end)
1519 return 0;
1520
1521
1522
1523 seg_left = seg_end - req_start;
1524 if (seg_left < size)
1525 size = (unsigned int)seg_left;
1526 }
1527
1528 return size;
1529}
1530EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1531
1532int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
1533{
1534 struct nfs_pageio_descriptor pgio;
1535
1536
1537 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
1538 hdr->completion_ops);
1539 return nfs_pageio_resend(&pgio, hdr);
1540}
1541EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1542
1543static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1544{
1545
1546 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1547 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1548 PNFS_LAYOUTRET_ON_ERROR) {
1549 pnfs_return_layout(hdr->inode);
1550 }
1551 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1552 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1553}
1554
1555
1556
1557
1558void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
1559{
1560 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
1561 if (!hdr->pnfs_error) {
1562 pnfs_set_layoutcommit(hdr);
1563 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1564 } else
1565 pnfs_ld_handle_write_error(hdr);
1566 hdr->mds_ops->rpc_release(hdr);
1567}
1568EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1569
1570static void
1571pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1572 struct nfs_pgio_header *hdr)
1573{
1574 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1575 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1576 nfs_pageio_reset_write_mds(desc);
1577 desc->pg_recoalesce = 1;
1578 }
1579 nfs_pgio_data_destroy(hdr);
1580}
1581
1582static enum pnfs_try_status
1583pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
1584 const struct rpc_call_ops *call_ops,
1585 struct pnfs_layout_segment *lseg,
1586 int how)
1587{
1588 struct inode *inode = hdr->inode;
1589 enum pnfs_try_status trypnfs;
1590 struct nfs_server *nfss = NFS_SERVER(inode);
1591
1592 hdr->mds_ops = call_ops;
1593
1594 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1595 inode->i_ino, hdr->args.count, hdr->args.offset, how);
1596 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
1597 if (trypnfs != PNFS_NOT_ATTEMPTED)
1598 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
1599 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1600 return trypnfs;
1601}
1602
1603static void
1604pnfs_do_write(struct nfs_pageio_descriptor *desc,
1605 struct nfs_pgio_header *hdr, int how)
1606{
1607 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1608 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1609 enum pnfs_try_status trypnfs;
1610
1611 desc->pg_lseg = NULL;
1612 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
1613 if (trypnfs == PNFS_NOT_ATTEMPTED)
1614 pnfs_write_through_mds(desc, hdr);
1615 pnfs_put_lseg(lseg);
1616}
1617
1618static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
1619{
1620 pnfs_put_lseg(hdr->lseg);
1621 nfs_pgio_header_free(hdr);
1622}
1623EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
1624
1625int
1626pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1627{
1628 struct nfs_pgio_header *hdr;
1629 int ret;
1630
1631 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
1632 if (!hdr) {
1633 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1634 pnfs_put_lseg(desc->pg_lseg);
1635 desc->pg_lseg = NULL;
1636 return -ENOMEM;
1637 }
1638 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1639 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1640 ret = nfs_generic_pgio(desc, hdr);
1641 if (ret != 0) {
1642 pnfs_put_lseg(desc->pg_lseg);
1643 desc->pg_lseg = NULL;
1644 } else
1645 pnfs_do_write(desc, hdr, desc->pg_ioflags);
1646 return ret;
1647}
1648EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
1649
1650int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
1651{
1652 struct nfs_pageio_descriptor pgio;
1653
1654
1655 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
1656 return nfs_pageio_resend(&pgio, hdr);
1657}
1658EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1659
1660static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1661{
1662 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
1663 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1664 PNFS_LAYOUTRET_ON_ERROR) {
1665 pnfs_return_layout(hdr->inode);
1666 }
1667 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1668 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
1669}
1670
1671
1672
1673
1674void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
1675{
1676 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
1677 if (likely(!hdr->pnfs_error)) {
1678 __nfs4_read_done_cb(hdr);
1679 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1680 } else
1681 pnfs_ld_handle_read_error(hdr);
1682 hdr->mds_ops->rpc_release(hdr);
1683}
1684EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
1685
1686static void
1687pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1688 struct nfs_pgio_header *hdr)
1689{
1690 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1691 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1692 nfs_pageio_reset_read_mds(desc);
1693 desc->pg_recoalesce = 1;
1694 }
1695 nfs_pgio_data_destroy(hdr);
1696}
1697
1698
1699
1700
1701static enum pnfs_try_status
1702pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
1703 const struct rpc_call_ops *call_ops,
1704 struct pnfs_layout_segment *lseg)
1705{
1706 struct inode *inode = hdr->inode;
1707 struct nfs_server *nfss = NFS_SERVER(inode);
1708 enum pnfs_try_status trypnfs;
1709
1710 hdr->mds_ops = call_ops;
1711
1712 dprintk("%s: Reading ino:%lu %u@%llu\n",
1713 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
1714
1715 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
1716 if (trypnfs != PNFS_NOT_ATTEMPTED)
1717 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
1718 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1719 return trypnfs;
1720}
1721
1722static void
1723pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
1724{
1725 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1726 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1727 enum pnfs_try_status trypnfs;
1728
1729 desc->pg_lseg = NULL;
1730 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
1731 if (trypnfs == PNFS_NOT_ATTEMPTED)
1732 pnfs_read_through_mds(desc, hdr);
1733 pnfs_put_lseg(lseg);
1734}
1735
1736static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
1737{
1738 pnfs_put_lseg(hdr->lseg);
1739 nfs_pgio_header_free(hdr);
1740}
1741EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
1742
1743int
1744pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
1745{
1746 struct nfs_pgio_header *hdr;
1747 int ret;
1748
1749 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
1750 if (!hdr) {
1751 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1752 ret = -ENOMEM;
1753 pnfs_put_lseg(desc->pg_lseg);
1754 desc->pg_lseg = NULL;
1755 return ret;
1756 }
1757 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1758 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1759 ret = nfs_generic_pgio(desc, hdr);
1760 if (ret != 0) {
1761 pnfs_put_lseg(desc->pg_lseg);
1762 desc->pg_lseg = NULL;
1763 } else
1764 pnfs_do_read(desc, hdr);
1765 return ret;
1766}
1767EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
1768
1769static void pnfs_clear_layoutcommitting(struct inode *inode)
1770{
1771 unsigned long *bitlock = &NFS_I(inode)->flags;
1772
1773 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
1774 smp_mb__after_atomic();
1775 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
1776}
1777
1778
1779
1780
1781static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
1782{
1783 struct pnfs_layout_segment *lseg;
1784
1785 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
1786 if (lseg->pls_range.iomode == IOMODE_RW &&
1787 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1788 list_add(&lseg->pls_lc_list, listp);
1789 }
1790}
1791
1792static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
1793{
1794 struct pnfs_layout_segment *lseg, *tmp;
1795
1796
1797 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
1798 list_del_init(&lseg->pls_lc_list);
1799 pnfs_put_lseg(lseg);
1800 }
1801
1802 pnfs_clear_layoutcommitting(inode);
1803}
1804
1805void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
1806{
1807 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1808}
1809EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
1810
1811void
1812pnfs_set_layoutcommit(struct nfs_pgio_header *hdr)
1813{
1814 struct inode *inode = hdr->inode;
1815 struct nfs_inode *nfsi = NFS_I(inode);
1816 loff_t end_pos = hdr->mds_offset + hdr->res.count;
1817 bool mark_as_dirty = false;
1818
1819 spin_lock(&inode->i_lock);
1820 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1821 mark_as_dirty = true;
1822 dprintk("%s: Set layoutcommit for inode %lu ",
1823 __func__, inode->i_ino);
1824 }
1825 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
1826
1827 pnfs_get_lseg(hdr->lseg);
1828 }
1829 if (end_pos > nfsi->layout->plh_lwb)
1830 nfsi->layout->plh_lwb = end_pos;
1831 spin_unlock(&inode->i_lock);
1832 dprintk("%s: lseg %p end_pos %llu\n",
1833 __func__, hdr->lseg, nfsi->layout->plh_lwb);
1834
1835
1836
1837 if (mark_as_dirty)
1838 mark_inode_dirty_sync(inode);
1839}
1840EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1841
1842void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data)
1843{
1844 struct inode *inode = data->inode;
1845 struct nfs_inode *nfsi = NFS_I(inode);
1846 bool mark_as_dirty = false;
1847
1848 spin_lock(&inode->i_lock);
1849 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1850 mark_as_dirty = true;
1851 dprintk("%s: Set layoutcommit for inode %lu ",
1852 __func__, inode->i_ino);
1853 }
1854 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) {
1855
1856 pnfs_get_lseg(data->lseg);
1857 }
1858 if (data->lwb > nfsi->layout->plh_lwb)
1859 nfsi->layout->plh_lwb = data->lwb;
1860 spin_unlock(&inode->i_lock);
1861 dprintk("%s: lseg %p end_pos %llu\n",
1862 __func__, data->lseg, nfsi->layout->plh_lwb);
1863
1864
1865
1866 if (mark_as_dirty)
1867 mark_inode_dirty_sync(inode);
1868}
1869EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit);
1870
1871void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
1872{
1873 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
1874
1875 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
1876 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
1877 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
1878}
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888int
1889pnfs_layoutcommit_inode(struct inode *inode, bool sync)
1890{
1891 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1892 struct nfs4_layoutcommit_data *data;
1893 struct nfs_inode *nfsi = NFS_I(inode);
1894 loff_t end_pos;
1895 int status;
1896
1897 if (!pnfs_layoutcommit_outstanding(inode))
1898 return 0;
1899
1900 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1901
1902 status = -EAGAIN;
1903 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
1904 if (!sync)
1905 goto out;
1906 status = wait_on_bit_lock_action(&nfsi->flags,
1907 NFS_INO_LAYOUTCOMMITTING,
1908 nfs_wait_bit_killable,
1909 TASK_KILLABLE);
1910 if (status)
1911 goto out;
1912 }
1913
1914 status = -ENOMEM;
1915
1916 data = kzalloc(sizeof(*data), GFP_NOFS);
1917 if (!data)
1918 goto clear_layoutcommitting;
1919
1920 status = 0;
1921 spin_lock(&inode->i_lock);
1922 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1923 goto out_unlock;
1924
1925 INIT_LIST_HEAD(&data->lseg_list);
1926 pnfs_list_write_lseg(inode, &data->lseg_list);
1927
1928 end_pos = nfsi->layout->plh_lwb;
1929 nfsi->layout->plh_lwb = 0;
1930
1931 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
1932 spin_unlock(&inode->i_lock);
1933
1934 data->args.inode = inode;
1935 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
1936 nfs_fattr_init(&data->fattr);
1937 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1938 data->res.fattr = &data->fattr;
1939 data->args.lastbytewritten = end_pos - 1;
1940 data->res.server = NFS_SERVER(inode);
1941
1942 if (ld->prepare_layoutcommit) {
1943 status = ld->prepare_layoutcommit(&data->args);
1944 if (status) {
1945 spin_lock(&inode->i_lock);
1946 if (end_pos < nfsi->layout->plh_lwb)
1947 nfsi->layout->plh_lwb = end_pos;
1948 spin_unlock(&inode->i_lock);
1949 put_rpccred(data->cred);
1950 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
1951 goto clear_layoutcommitting;
1952 }
1953 }
1954
1955
1956 status = nfs4_proc_layoutcommit(data, sync);
1957out:
1958 if (status)
1959 mark_inode_dirty_sync(inode);
1960 dprintk("<-- %s status %d\n", __func__, status);
1961 return status;
1962out_unlock:
1963 spin_unlock(&inode->i_lock);
1964 kfree(data);
1965clear_layoutcommitting:
1966 pnfs_clear_layoutcommitting(inode);
1967 goto out;
1968}
1969
1970struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
1971{
1972 struct nfs4_threshold *thp;
1973
1974 thp = kzalloc(sizeof(*thp), GFP_NOFS);
1975 if (!thp) {
1976 dprintk("%s mdsthreshold allocation failed\n", __func__);
1977 return NULL;
1978 }
1979 return thp;
1980}
1981