NFSv4.1 data server timeo and retrans module parameters
[linux-block.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL
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>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
974cec8c 33#include "internal.h"
85e174ba 34#include "pnfs.h"
64419a9b 35#include "iostat.h"
85e174ba
RL
36
37#define NFSDBG_FACILITY NFSDBG_PNFS
38
02c35fca
FI
39/* Locking:
40 *
41 * pnfs_spinlock:
42 * protects pnfs_modules_tbl.
43 */
44static DEFINE_SPINLOCK(pnfs_spinlock);
45
46/*
47 * pnfs_modules_tbl holds all pnfs modules
48 */
49static LIST_HEAD(pnfs_modules_tbl);
50
51/* Return the registered pnfs layout driver module matching given id */
52static struct pnfs_layoutdriver_type *
53find_pnfs_driver_locked(u32 id)
54{
55 struct pnfs_layoutdriver_type *local;
56
57 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
58 if (local->id == id)
59 goto out;
60 local = NULL;
61out:
62 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
63 return local;
64}
65
85e174ba
RL
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver(u32 id)
68{
02c35fca
FI
69 struct pnfs_layoutdriver_type *local;
70
71 spin_lock(&pnfs_spinlock);
72 local = find_pnfs_driver_locked(id);
73 spin_unlock(&pnfs_spinlock);
74 return local;
85e174ba
RL
75}
76
77void
78unset_pnfs_layoutdriver(struct nfs_server *nfss)
79{
738fd0f3
BH
80 if (nfss->pnfs_curr_ld) {
81 if (nfss->pnfs_curr_ld->clear_layoutdriver)
82 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
02c35fca 83 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 84 }
85e174ba
RL
85 nfss->pnfs_curr_ld = NULL;
86}
87
88/*
89 * Try to set the server's pnfs module to the pnfs layout type specified by id.
90 * Currently only one pNFS layout driver per filesystem is supported.
91 *
92 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
93 */
94void
738fd0f3
BH
95set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
96 u32 id)
85e174ba
RL
97{
98 struct pnfs_layoutdriver_type *ld_type = NULL;
99
100 if (id == 0)
101 goto out_no_driver;
102 if (!(server->nfs_client->cl_exchange_flags &
103 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
a030889a
WAA
104 printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
105 __func__, id, server->nfs_client->cl_exchange_flags);
85e174ba
RL
106 goto out_no_driver;
107 }
108 ld_type = find_pnfs_driver(id);
109 if (!ld_type) {
110 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
111 ld_type = find_pnfs_driver(id);
112 if (!ld_type) {
113 dprintk("%s: No pNFS module found for %u.\n",
114 __func__, id);
115 goto out_no_driver;
116 }
117 }
02c35fca
FI
118 if (!try_module_get(ld_type->owner)) {
119 dprintk("%s: Could not grab reference on module\n", __func__);
120 goto out_no_driver;
121 }
85e174ba 122 server->pnfs_curr_ld = ld_type;
738fd0f3
BH
123 if (ld_type->set_layoutdriver
124 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA
125 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
126 "driver %u.\n", __func__, id);
738fd0f3
BH
127 module_put(ld_type->owner);
128 goto out_no_driver;
129 }
ea8eecdd 130
85e174ba
RL
131 dprintk("%s: pNFS module for %u set\n", __func__, id);
132 return;
133
134out_no_driver:
135 dprintk("%s: Using NFSv4 I/O\n", __func__);
136 server->pnfs_curr_ld = NULL;
137}
02c35fca
FI
138
139int
140pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
141{
142 int status = -EINVAL;
143 struct pnfs_layoutdriver_type *tmp;
144
145 if (ld_type->id == 0) {
a030889a 146 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI
147 return status;
148 }
b1f69b75 149 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 150 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA
151 "alloc_lseg and free_lseg.\n", __func__);
152 return status;
153 }
02c35fca
FI
154
155 spin_lock(&pnfs_spinlock);
156 tmp = find_pnfs_driver_locked(ld_type->id);
157 if (!tmp) {
158 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
159 status = 0;
160 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
161 ld_type->name);
162 } else {
a030889a 163 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI
164 __func__, ld_type->id);
165 }
166 spin_unlock(&pnfs_spinlock);
167
168 return status;
169}
170EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
171
172void
173pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
174{
175 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
176 spin_lock(&pnfs_spinlock);
177 list_del(&ld_type->pnfs_tblid);
178 spin_unlock(&pnfs_spinlock);
179}
180EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 181
b1f69b75
AA
182/*
183 * pNFS client layout cache
184 */
185
cc6e5340 186/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 187void
cc6e5340 188get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 189{
cc6e5340 190 atomic_inc(&lo->plh_refcount);
e5e94017
BH
191}
192
636fb9c8
BH
193static struct pnfs_layout_hdr *
194pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
195{
196 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
197 return ld->alloc_layout_hdr ? ld->alloc_layout_hdr(ino, gfp_flags) :
198 kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags);
199}
200
201static void
202pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
203{
204 struct pnfs_layoutdriver_type *ld = NFS_SERVER(lo->plh_inode)->pnfs_curr_ld;
9fa40758 205 put_rpccred(lo->plh_lc_cred);
636fb9c8
BH
206 return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo);
207}
208
e5e94017 209static void
cc6e5340 210destroy_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 211{
cc6e5340
FI
212 dprintk("%s: freeing layout cache %p\n", __func__, lo);
213 BUG_ON(!list_empty(&lo->plh_layouts));
214 NFS_I(lo->plh_inode)->layout = NULL;
636fb9c8 215 pnfs_free_layout_hdr(lo);
cc6e5340 216}
e5e94017 217
cc6e5340
FI
218static void
219put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
220{
221 if (atomic_dec_and_test(&lo->plh_refcount))
222 destroy_layout_hdr(lo);
e5e94017
BH
223}
224
b1f69b75 225void
cc6e5340 226put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 227{
cc6e5340
FI
228 struct inode *inode = lo->plh_inode;
229
230 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
231 destroy_layout_hdr(lo);
232 spin_unlock(&inode->i_lock);
233 }
974cec8c
AA
234}
235
236static void
237init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
238{
566052c5 239 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 240 INIT_LIST_HEAD(&lseg->pls_lc_list);
4541d16c
FI
241 atomic_set(&lseg->pls_refcount, 1);
242 smp_mb();
243 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 244 lseg->pls_layout = lo;
974cec8c
AA
245}
246
4541d16c 247static void free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 248{
b7edfaa1 249 struct inode *ino = lseg->pls_layout->plh_inode;
974cec8c 250
b1f69b75 251 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
52fabd73 252 /* Matched by get_layout_hdr in pnfs_insert_layout */
cc6e5340 253 put_layout_hdr(NFS_I(ino)->layout);
974cec8c
AA
254}
255
d684d2ae
FI
256static void
257put_lseg_common(struct pnfs_layout_segment *lseg)
258{
259 struct inode *inode = lseg->pls_layout->plh_inode;
260
d20581aa 261 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae
FI
262 list_del_init(&lseg->pls_list);
263 if (list_empty(&lseg->pls_layout->plh_segs)) {
264 set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
265 /* Matched by initial refcount set in alloc_init_layout_hdr */
266 put_layout_hdr_locked(lseg->pls_layout);
267 }
268 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
269}
270
bae724ef 271void
d684d2ae 272put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 273{
d684d2ae
FI
274 struct inode *inode;
275
276 if (!lseg)
277 return;
278
4541d16c
FI
279 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
280 atomic_read(&lseg->pls_refcount),
281 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae
FI
282 inode = lseg->pls_layout->plh_inode;
283 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
284 LIST_HEAD(free_me);
4541d16c 285
d684d2ae
FI
286 put_lseg_common(lseg);
287 list_add(&lseg->pls_list, &free_me);
288 spin_unlock(&inode->i_lock);
289 pnfs_free_lseg_list(&free_me);
4541d16c 290 }
4541d16c 291}
e0c2b380 292EXPORT_SYMBOL_GPL(put_lseg);
974cec8c 293
fb3296eb
BH
294static inline u64
295end_offset(u64 start, u64 len)
296{
297 u64 end;
298
299 end = start + len;
300 return end >= start ? end : NFS4_MAX_UINT64;
301}
302
303/* last octet in a range */
304static inline u64
305last_byte_offset(u64 start, u64 len)
306{
307 u64 end;
308
309 BUG_ON(!len);
310 end = start + len;
311 return end > start ? end - 1 : NFS4_MAX_UINT64;
312}
313
314/*
315 * is l2 fully contained in l1?
316 * start1 end1
317 * [----------------------------------)
318 * start2 end2
319 * [----------------)
320 */
321static inline int
322lo_seg_contained(struct pnfs_layout_range *l1,
323 struct pnfs_layout_range *l2)
324{
325 u64 start1 = l1->offset;
326 u64 end1 = end_offset(start1, l1->length);
327 u64 start2 = l2->offset;
328 u64 end2 = end_offset(start2, l2->length);
329
330 return (start1 <= start2) && (end1 >= end2);
331}
332
333/*
334 * is l1 and l2 intersecting?
335 * start1 end1
336 * [----------------------------------)
337 * start2 end2
338 * [----------------)
339 */
340static inline int
341lo_seg_intersecting(struct pnfs_layout_range *l1,
342 struct pnfs_layout_range *l2)
343{
344 u64 start1 = l1->offset;
345 u64 end1 = end_offset(start1, l1->length);
346 u64 start2 = l2->offset;
347 u64 end2 = end_offset(start2, l2->length);
348
349 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
350 (end2 == NFS4_MAX_UINT64 || end2 > start1);
351}
352
4541d16c 353static bool
778b5502
BH
354should_free_lseg(struct pnfs_layout_range *lseg_range,
355 struct pnfs_layout_range *recall_range)
4541d16c 356{
778b5502
BH
357 return (recall_range->iomode == IOMODE_ANY ||
358 lseg_range->iomode == recall_range->iomode) &&
359 lo_seg_intersecting(lseg_range, recall_range);
974cec8c
AA
360}
361
4541d16c
FI
362/* Returns 1 if lseg is removed from list, 0 otherwise */
363static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
364 struct list_head *tmp_list)
365{
366 int rv = 0;
367
368 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
369 /* Remove the reference keeping the lseg in the
370 * list. It will now be removed when all
371 * outstanding io is finished.
372 */
d684d2ae
FI
373 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
374 atomic_read(&lseg->pls_refcount));
375 if (atomic_dec_and_test(&lseg->pls_refcount)) {
376 put_lseg_common(lseg);
377 list_add(&lseg->pls_list, tmp_list);
378 rv = 1;
379 }
4541d16c
FI
380 }
381 return rv;
382}
383
384/* Returns count of number of matching invalid lsegs remaining in list
385 * after call.
386 */
43f1b3da 387int
4541d16c
FI
388mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
389 struct list_head *tmp_list,
778b5502 390 struct pnfs_layout_range *recall_range)
974cec8c
AA
391{
392 struct pnfs_layout_segment *lseg, *next;
4541d16c 393 int invalid = 0, removed = 0;
974cec8c
AA
394
395 dprintk("%s:Begin lo %p\n", __func__, lo);
396
38511722
FI
397 if (list_empty(&lo->plh_segs)) {
398 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
399 put_layout_hdr_locked(lo);
400 return 0;
401 }
4541d16c 402 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
778b5502
BH
403 if (!recall_range ||
404 should_free_lseg(&lseg->pls_range, recall_range)) {
4541d16c
FI
405 dprintk("%s: freeing lseg %p iomode %d "
406 "offset %llu length %llu\n", __func__,
407 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
408 lseg->pls_range.length);
409 invalid++;
410 removed += mark_lseg_invalid(lseg, tmp_list);
411 }
412 dprintk("%s:Return %i\n", __func__, invalid - removed);
413 return invalid - removed;
974cec8c
AA
414}
415
f49f9baa 416/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 417void
4541d16c 418pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 419{
4541d16c 420 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI
421 struct pnfs_layout_hdr *lo;
422
423 if (list_empty(free_me))
424 return;
425
426 lo = list_first_entry(free_me, struct pnfs_layout_segment,
427 pls_list)->pls_layout;
974cec8c 428
f49f9baa
FI
429 if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
430 struct nfs_client *clp;
431
432 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
433 spin_lock(&clp->cl_lock);
434 list_del_init(&lo->plh_layouts);
435 spin_unlock(&clp->cl_lock);
436 }
4541d16c 437 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 438 list_del(&lseg->pls_list);
4541d16c 439 free_lseg(lseg);
974cec8c
AA
440 }
441}
442
e5e94017
BH
443void
444pnfs_destroy_layout(struct nfs_inode *nfsi)
445{
446 struct pnfs_layout_hdr *lo;
974cec8c 447 LIST_HEAD(tmp_list);
e5e94017
BH
448
449 spin_lock(&nfsi->vfs_inode.i_lock);
450 lo = nfsi->layout;
451 if (lo) {
38511722 452 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
778b5502 453 mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
e5e94017
BH
454 }
455 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c
AA
456 pnfs_free_lseg_list(&tmp_list);
457}
458
459/*
460 * Called by the state manger to remove all layouts established under an
461 * expired lease.
462 */
463void
464pnfs_destroy_all_layouts(struct nfs_client *clp)
465{
6382a441 466 struct nfs_server *server;
974cec8c
AA
467 struct pnfs_layout_hdr *lo;
468 LIST_HEAD(tmp_list);
469
c47abcf8
AA
470 nfs4_deviceid_mark_client_invalid(clp);
471 nfs4_deviceid_purge_client(clp);
472
974cec8c 473 spin_lock(&clp->cl_lock);
6382a441
WAA
474 rcu_read_lock();
475 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
476 if (!list_empty(&server->layouts))
477 list_splice_init(&server->layouts, &tmp_list);
478 }
479 rcu_read_unlock();
974cec8c
AA
480 spin_unlock(&clp->cl_lock);
481
482 while (!list_empty(&tmp_list)) {
483 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
b7edfaa1 484 plh_layouts);
974cec8c 485 dprintk("%s freeing layout for inode %lu\n", __func__,
b7edfaa1 486 lo->plh_inode->i_ino);
2887fe45 487 list_del_init(&lo->plh_layouts);
b7edfaa1 488 pnfs_destroy_layout(NFS_I(lo->plh_inode));
974cec8c 489 }
e5e94017
BH
490}
491
fd6002e9 492/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI
493void
494pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
495 bool update_barrier)
b1f69b75 496{
fd6002e9 497 u32 oldseq, newseq;
b1f69b75 498
2d2f24ad
TM
499 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
500 newseq = be32_to_cpu(new->seqid);
43f1b3da 501 if ((int)(newseq - oldseq) > 0) {
f597c537 502 nfs4_stateid_copy(&lo->plh_stateid, new);
43f1b3da 503 if (update_barrier) {
2d2f24ad 504 u32 new_barrier = be32_to_cpu(new->seqid);
43f1b3da
FI
505
506 if ((int)(new_barrier - lo->plh_barrier))
507 lo->plh_barrier = new_barrier;
508 } else {
509 /* Because of wraparound, we want to keep the barrier
510 * "close" to the current seqids. It needs to be
511 * within 2**31 to count as "behind", so if it
512 * gets too near that limit, give us a litle leeway
513 * and bring it to within 2**30.
514 * NOTE - and yes, this is all unsigned arithmetic.
515 */
516 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
517 lo->plh_barrier = newseq - (1 << 30);
518 }
519 }
b1f69b75
AA
520}
521
cf7d63f1
FI
522/* lget is set to 1 if called from inside send_layoutget call chain */
523static bool
43f1b3da
FI
524pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
525 int lget)
526{
527 if ((stateid) &&
2d2f24ad 528 (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0)
43f1b3da 529 return true;
f7e8917a 530 return lo->plh_block_lgets ||
38511722 531 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
f7e8917a 532 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
43f1b3da 533 (list_empty(&lo->plh_segs) &&
cf7d63f1
FI
534 (atomic_read(&lo->plh_outstanding) > lget));
535}
536
fd6002e9
FI
537int
538pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
539 struct nfs4_state *open_state)
b1f69b75 540{
fd6002e9 541 int status = 0;
974cec8c 542
b1f69b75 543 dprintk("--> %s\n", __func__);
fd6002e9 544 spin_lock(&lo->plh_inode->i_lock);
43f1b3da 545 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
cf7d63f1
FI
546 status = -EAGAIN;
547 } else if (list_empty(&lo->plh_segs)) {
fd6002e9
FI
548 int seq;
549
550 do {
551 seq = read_seqbegin(&open_state->seqlock);
f597c537 552 nfs4_stateid_copy(dst, &open_state->stateid);
fd6002e9
FI
553 } while (read_seqretry(&open_state->seqlock, seq));
554 } else
f597c537 555 nfs4_stateid_copy(dst, &lo->plh_stateid);
fd6002e9 556 spin_unlock(&lo->plh_inode->i_lock);
b1f69b75 557 dprintk("<-- %s\n", __func__);
fd6002e9 558 return status;
b1f69b75
AA
559}
560
561/*
562* Get layout from server.
563* for now, assume that whole file layouts are requested.
564* arg->offset: 0
565* arg->length: all ones
566*/
e5e94017
BH
567static struct pnfs_layout_segment *
568send_layoutget(struct pnfs_layout_hdr *lo,
569 struct nfs_open_context *ctx,
fb3296eb 570 struct pnfs_layout_range *range,
a75b9df9 571 gfp_t gfp_flags)
e5e94017 572{
b7edfaa1 573 struct inode *ino = lo->plh_inode;
b1f69b75
AA
574 struct nfs_server *server = NFS_SERVER(ino);
575 struct nfs4_layoutget *lgp;
576 struct pnfs_layout_segment *lseg = NULL;
35124a09
WAA
577 struct page **pages = NULL;
578 int i;
579 u32 max_resp_sz, max_pages;
b1f69b75
AA
580
581 dprintk("--> %s\n", __func__);
e5e94017 582
b1f69b75 583 BUG_ON(ctx == NULL);
a75b9df9 584 lgp = kzalloc(sizeof(*lgp), gfp_flags);
cf7d63f1 585 if (lgp == NULL)
b1f69b75 586 return NULL;
35124a09
WAA
587
588 /* allocate pages for xdr post processing */
589 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
e5265a0c 590 max_pages = nfs_page_array_len(0, max_resp_sz);
35124a09 591
7d9dea91 592 pages = kcalloc(max_pages, sizeof(struct page *), gfp_flags);
35124a09
WAA
593 if (!pages)
594 goto out_err_free;
595
596 for (i = 0; i < max_pages; i++) {
a75b9df9 597 pages[i] = alloc_page(gfp_flags);
35124a09
WAA
598 if (!pages[i])
599 goto out_err_free;
600 }
601
fb3296eb
BH
602 lgp->args.minlength = PAGE_CACHE_SIZE;
603 if (lgp->args.minlength > range->length)
604 lgp->args.minlength = range->length;
b1f69b75 605 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
fb3296eb 606 lgp->args.range = *range;
b1f69b75
AA
607 lgp->args.type = server->pnfs_curr_ld->id;
608 lgp->args.inode = ino;
609 lgp->args.ctx = get_nfs_open_context(ctx);
35124a09
WAA
610 lgp->args.layout.pages = pages;
611 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
b1f69b75 612 lgp->lsegpp = &lseg;
a75b9df9 613 lgp->gfp_flags = gfp_flags;
b1f69b75
AA
614
615 /* Synchronously retrieve layout information from server and
616 * store in lseg.
617 */
618 nfs4_proc_layoutget(lgp);
974cec8c 619 if (!lseg) {
b1f69b75 620 /* remember that LAYOUTGET failed and suspend trying */
fb3296eb 621 set_bit(lo_fail_bit(range->iomode), &lo->plh_flags);
974cec8c 622 }
35124a09
WAA
623
624 /* free xdr pages */
625 for (i = 0; i < max_pages; i++)
626 __free_page(pages[i]);
627 kfree(pages);
628
974cec8c 629 return lseg;
35124a09
WAA
630
631out_err_free:
632 /* free any allocated xdr pages, lgp as it's not used */
633 if (pages) {
634 for (i = 0; i < max_pages; i++) {
635 if (!pages[i])
636 break;
637 __free_page(pages[i]);
638 }
639 kfree(pages);
640 }
641 kfree(lgp);
642 return NULL;
974cec8c
AA
643}
644
cbe82603
BH
645/* Initiates a LAYOUTRETURN(FILE) */
646int
647_pnfs_return_layout(struct inode *ino)
648{
649 struct pnfs_layout_hdr *lo = NULL;
650 struct nfs_inode *nfsi = NFS_I(ino);
651 LIST_HEAD(tmp_list);
652 struct nfs4_layoutreturn *lrp;
653 nfs4_stateid stateid;
654 int status = 0;
655
656 dprintk("--> %s\n", __func__);
657
658 spin_lock(&ino->i_lock);
659 lo = nfsi->layout;
a2e1d4f2 660 if (!lo) {
cbe82603 661 spin_unlock(&ino->i_lock);
a2e1d4f2
FI
662 dprintk("%s: no layout to return\n", __func__);
663 return status;
cbe82603
BH
664 }
665 stateid = nfsi->layout->plh_stateid;
666 /* Reference matched in nfs4_layoutreturn_release */
667 get_layout_hdr(lo);
ea0ded74
FI
668 mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
669 lo->plh_block_lgets++;
cbe82603
BH
670 spin_unlock(&ino->i_lock);
671 pnfs_free_lseg_list(&tmp_list);
672
673 WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags));
674
675 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
676 if (unlikely(lrp == NULL)) {
677 status = -ENOMEM;
9e2dfdb3
FI
678 set_bit(NFS_LAYOUT_RW_FAILED, &lo->plh_flags);
679 set_bit(NFS_LAYOUT_RO_FAILED, &lo->plh_flags);
1ed3a853 680 put_layout_hdr(lo);
cbe82603
BH
681 goto out;
682 }
683
684 lrp->args.stateid = stateid;
685 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
686 lrp->args.inode = ino;
a56aaa02 687 lrp->args.layout = lo;
cbe82603
BH
688 lrp->clp = NFS_SERVER(ino)->nfs_client;
689
690 status = nfs4_proc_layoutreturn(lrp);
691out:
692 dprintk("<-- %s status: %d\n", __func__, status);
693 return status;
694}
695
f7e8917a
FI
696bool pnfs_roc(struct inode *ino)
697{
698 struct pnfs_layout_hdr *lo;
699 struct pnfs_layout_segment *lseg, *tmp;
700 LIST_HEAD(tmp_list);
701 bool found = false;
702
703 spin_lock(&ino->i_lock);
704 lo = NFS_I(ino)->layout;
705 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
706 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
707 goto out_nolayout;
708 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
709 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
710 mark_lseg_invalid(lseg, &tmp_list);
711 found = true;
712 }
713 if (!found)
714 goto out_nolayout;
715 lo->plh_block_lgets++;
716 get_layout_hdr(lo); /* matched in pnfs_roc_release */
717 spin_unlock(&ino->i_lock);
718 pnfs_free_lseg_list(&tmp_list);
719 return true;
720
721out_nolayout:
722 spin_unlock(&ino->i_lock);
723 return false;
724}
725
726void pnfs_roc_release(struct inode *ino)
727{
728 struct pnfs_layout_hdr *lo;
729
730 spin_lock(&ino->i_lock);
731 lo = NFS_I(ino)->layout;
732 lo->plh_block_lgets--;
733 put_layout_hdr_locked(lo);
734 spin_unlock(&ino->i_lock);
735}
736
737void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
738{
739 struct pnfs_layout_hdr *lo;
740
741 spin_lock(&ino->i_lock);
742 lo = NFS_I(ino)->layout;
743 if ((int)(barrier - lo->plh_barrier) > 0)
744 lo->plh_barrier = barrier;
745 spin_unlock(&ino->i_lock);
746}
747
748bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
749{
750 struct nfs_inode *nfsi = NFS_I(ino);
751 struct pnfs_layout_segment *lseg;
752 bool found = false;
753
754 spin_lock(&ino->i_lock);
755 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
756 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
757 found = true;
758 break;
759 }
760 if (!found) {
761 struct pnfs_layout_hdr *lo = nfsi->layout;
2d2f24ad 762 u32 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
f7e8917a
FI
763
764 /* Since close does not return a layout stateid for use as
765 * a barrier, we choose the worst-case barrier.
766 */
767 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
768 }
769 spin_unlock(&ino->i_lock);
770 return found;
771}
772
b1f69b75
AA
773/*
774 * Compare two layout segments for sorting into layout cache.
775 * We want to preferentially return RW over RO layouts, so ensure those
776 * are seen first.
777 */
778static s64
fb3296eb
BH
779cmp_layout(struct pnfs_layout_range *l1,
780 struct pnfs_layout_range *l2)
b1f69b75 781{
fb3296eb
BH
782 s64 d;
783
784 /* high offset > low offset */
785 d = l1->offset - l2->offset;
786 if (d)
787 return d;
788
789 /* short length > long length */
790 d = l2->length - l1->length;
791 if (d)
792 return d;
793
b1f69b75 794 /* read > read/write */
fb3296eb 795 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA
796}
797
974cec8c
AA
798static void
799pnfs_insert_layout(struct pnfs_layout_hdr *lo,
800 struct pnfs_layout_segment *lseg)
801{
b1f69b75 802 struct pnfs_layout_segment *lp;
b1f69b75 803
974cec8c
AA
804 dprintk("%s:Begin\n", __func__);
805
b7edfaa1 806 assert_spin_locked(&lo->plh_inode->i_lock);
b7edfaa1 807 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
fb3296eb 808 if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
b1f69b75 809 continue;
566052c5 810 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA
811 dprintk("%s: inserted lseg %p "
812 "iomode %d offset %llu length %llu before "
813 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI
814 __func__, lseg, lseg->pls_range.iomode,
815 lseg->pls_range.offset, lseg->pls_range.length,
816 lp, lp->pls_range.iomode, lp->pls_range.offset,
817 lp->pls_range.length);
fb3296eb 818 goto out;
974cec8c 819 }
fb3296eb
BH
820 list_add_tail(&lseg->pls_list, &lo->plh_segs);
821 dprintk("%s: inserted lseg %p "
822 "iomode %d offset %llu length %llu at tail\n",
823 __func__, lseg, lseg->pls_range.iomode,
824 lseg->pls_range.offset, lseg->pls_range.length);
825out:
cc6e5340 826 get_layout_hdr(lo);
974cec8c
AA
827
828 dprintk("%s:Return\n", __func__);
e5e94017
BH
829}
830
831static struct pnfs_layout_hdr *
9fa40758
PT
832alloc_init_layout_hdr(struct inode *ino,
833 struct nfs_open_context *ctx,
834 gfp_t gfp_flags)
e5e94017
BH
835{
836 struct pnfs_layout_hdr *lo;
837
636fb9c8 838 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH
839 if (!lo)
840 return NULL;
cc6e5340 841 atomic_set(&lo->plh_refcount, 1);
b7edfaa1
FI
842 INIT_LIST_HEAD(&lo->plh_layouts);
843 INIT_LIST_HEAD(&lo->plh_segs);
43f1b3da 844 INIT_LIST_HEAD(&lo->plh_bulk_recall);
b7edfaa1 845 lo->plh_inode = ino;
9fa40758 846 lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
e5e94017
BH
847 return lo;
848}
849
850static struct pnfs_layout_hdr *
9fa40758
PT
851pnfs_find_alloc_layout(struct inode *ino,
852 struct nfs_open_context *ctx,
853 gfp_t gfp_flags)
e5e94017
BH
854{
855 struct nfs_inode *nfsi = NFS_I(ino);
856 struct pnfs_layout_hdr *new = NULL;
857
858 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
859
860 assert_spin_locked(&ino->i_lock);
4541d16c
FI
861 if (nfsi->layout) {
862 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
863 return NULL;
864 else
865 return nfsi->layout;
866 }
e5e94017 867 spin_unlock(&ino->i_lock);
9fa40758 868 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH
869 spin_lock(&ino->i_lock);
870
871 if (likely(nfsi->layout == NULL)) /* Won the race? */
872 nfsi->layout = new;
873 else
636fb9c8 874 pnfs_free_layout_hdr(new);
e5e94017
BH
875 return nfsi->layout;
876}
877
b1f69b75
AA
878/*
879 * iomode matching rules:
880 * iomode lseg match
881 * ----- ----- -----
882 * ANY READ true
883 * ANY RW true
884 * RW READ false
885 * RW RW true
886 * READ READ true
887 * READ RW true
888 */
889static int
fb3296eb
BH
890is_matching_lseg(struct pnfs_layout_range *ls_range,
891 struct pnfs_layout_range *range)
b1f69b75 892{
fb3296eb
BH
893 struct pnfs_layout_range range1;
894
895 if ((range->iomode == IOMODE_RW &&
896 ls_range->iomode != IOMODE_RW) ||
897 !lo_seg_intersecting(ls_range, range))
898 return 0;
899
900 /* range1 covers only the first byte in the range */
901 range1 = *range;
902 range1.length = 1;
903 return lo_seg_contained(ls_range, &range1);
b1f69b75
AA
904}
905
906/*
907 * lookup range in layout
908 */
e5e94017 909static struct pnfs_layout_segment *
fb3296eb
BH
910pnfs_find_lseg(struct pnfs_layout_hdr *lo,
911 struct pnfs_layout_range *range)
e5e94017 912{
b1f69b75
AA
913 struct pnfs_layout_segment *lseg, *ret = NULL;
914
915 dprintk("%s:Begin\n", __func__);
916
b7edfaa1
FI
917 assert_spin_locked(&lo->plh_inode->i_lock);
918 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 919 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
fb3296eb 920 is_matching_lseg(&lseg->pls_range, range)) {
d684d2ae 921 ret = get_lseg(lseg);
b1f69b75
AA
922 break;
923 }
d771e3a4 924 if (lseg->pls_range.offset > range->offset)
b1f69b75
AA
925 break;
926 }
927
928 dprintk("%s:Return lseg %p ref %d\n",
4541d16c 929 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
b1f69b75 930 return ret;
e5e94017
BH
931}
932
933/*
934 * Layout segment is retreived from the server if not cached.
935 * The appropriate layout segment is referenced and returned to the caller.
936 */
7c24d948 937struct pnfs_layout_segment *
e5e94017
BH
938pnfs_update_layout(struct inode *ino,
939 struct nfs_open_context *ctx,
fb3296eb
BH
940 loff_t pos,
941 u64 count,
a75b9df9
TM
942 enum pnfs_iomode iomode,
943 gfp_t gfp_flags)
e5e94017 944{
fb3296eb
BH
945 struct pnfs_layout_range arg = {
946 .iomode = iomode,
947 .offset = pos,
948 .length = count,
949 };
707ed5fd 950 unsigned pg_offset;
e5e94017 951 struct nfs_inode *nfsi = NFS_I(ino);
6382a441
WAA
952 struct nfs_server *server = NFS_SERVER(ino);
953 struct nfs_client *clp = server->nfs_client;
e5e94017
BH
954 struct pnfs_layout_hdr *lo;
955 struct pnfs_layout_segment *lseg = NULL;
f49f9baa 956 bool first = false;
e5e94017
BH
957
958 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
959 return NULL;
960 spin_lock(&ino->i_lock);
9fa40758 961 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
e5e94017
BH
962 if (lo == NULL) {
963 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
964 goto out_unlock;
965 }
966
43f1b3da 967 /* Do we even need to bother with this? */
a59c30ac 968 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
43f1b3da 969 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH
970 goto out_unlock;
971 }
972
973 /* if LAYOUTGET already failed once we don't try again */
566052c5 974 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
e5e94017
BH
975 goto out_unlock;
976
568e8c49 977 /* Check to see if the layout for the given range already exists */
fb3296eb 978 lseg = pnfs_find_lseg(lo, &arg);
568e8c49
AA
979 if (lseg)
980 goto out_unlock;
981
43f1b3da 982 if (pnfs_layoutgets_blocked(lo, NULL, 0))
cf7d63f1
FI
983 goto out_unlock;
984 atomic_inc(&lo->plh_outstanding);
985
cc6e5340 986 get_layout_hdr(lo);
f49f9baa
FI
987 if (list_empty(&lo->plh_segs))
988 first = true;
989 spin_unlock(&ino->i_lock);
990 if (first) {
2130ff66
FI
991 /* The lo must be on the clp list if there is any
992 * chance of a CB_LAYOUTRECALL(FILE) coming in.
993 */
994 spin_lock(&clp->cl_lock);
995 BUG_ON(!list_empty(&lo->plh_layouts));
6382a441 996 list_add_tail(&lo->plh_layouts, &server->layouts);
2130ff66
FI
997 spin_unlock(&clp->cl_lock);
998 }
e5e94017 999
707ed5fd
BH
1000 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1001 if (pg_offset) {
1002 arg.offset -= pg_offset;
1003 arg.length += pg_offset;
1004 }
7c24d948
AA
1005 if (arg.length != NFS4_MAX_UINT64)
1006 arg.length = PAGE_CACHE_ALIGN(arg.length);
707ed5fd 1007
fb3296eb 1008 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
f49f9baa
FI
1009 if (!lseg && first) {
1010 spin_lock(&clp->cl_lock);
1011 list_del_init(&lo->plh_layouts);
1012 spin_unlock(&clp->cl_lock);
2130ff66 1013 }
cf7d63f1 1014 atomic_dec(&lo->plh_outstanding);
cc6e5340 1015 put_layout_hdr(lo);
e5e94017
BH
1016out:
1017 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
bf9c1387 1018 nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
e5e94017
BH
1019 return lseg;
1020out_unlock:
1021 spin_unlock(&ino->i_lock);
1022 goto out;
1023}
7c24d948 1024EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75
AA
1025
1026int
1027pnfs_layout_process(struct nfs4_layoutget *lgp)
1028{
1029 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1030 struct nfs4_layoutget_res *res = &lgp->res;
1031 struct pnfs_layout_segment *lseg;
b7edfaa1 1032 struct inode *ino = lo->plh_inode;
b1f69b75
AA
1033 int status = 0;
1034
1035 /* Inject layout blob into I/O device driver */
a75b9df9 1036 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
b1f69b75
AA
1037 if (!lseg || IS_ERR(lseg)) {
1038 if (!lseg)
1039 status = -ENOMEM;
1040 else
1041 status = PTR_ERR(lseg);
1042 dprintk("%s: Could not allocate layout: error %d\n",
1043 __func__, status);
1044 goto out;
1045 }
1046
1047 spin_lock(&ino->i_lock);
a59c30ac 1048 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
43f1b3da
FI
1049 dprintk("%s forget reply due to recall\n", __func__);
1050 goto out_forget_reply;
1051 }
1052
1053 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
1054 dprintk("%s forget reply due to state\n", __func__);
1055 goto out_forget_reply;
1056 }
b1f69b75 1057 init_lseg(lo, lseg);
566052c5 1058 lseg->pls_range = res->range;
d684d2ae 1059 *lgp->lsegpp = get_lseg(lseg);
b1f69b75
AA
1060 pnfs_insert_layout(lo, lseg);
1061
f7e8917a
FI
1062 if (res->return_on_close) {
1063 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1064 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
1065 }
1066
b1f69b75 1067 /* Done processing layoutget. Set the layout stateid */
43f1b3da 1068 pnfs_set_layout_stateid(lo, &res->stateid, false);
b1f69b75
AA
1069 spin_unlock(&ino->i_lock);
1070out:
1071 return status;
43f1b3da
FI
1072
1073out_forget_reply:
1074 spin_unlock(&ino->i_lock);
1075 lseg->pls_layout = lo;
1076 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1077 goto out;
b1f69b75
AA
1078}
1079
d8007d4d
TM
1080void
1081pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1082{
1083 BUG_ON(pgio->pg_lseg != NULL);
1084
1825a0d0
FI
1085 if (req->wb_offset != req->wb_pgbase) {
1086 nfs_pageio_reset_read_mds(pgio);
1087 return;
1088 }
d8007d4d
TM
1089 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1090 req->wb_context,
1091 req_offset(req),
1092 req->wb_bytes,
1093 IOMODE_READ,
1094 GFP_KERNEL);
e885de1a
TM
1095 /* If no lseg, fall back to read through mds */
1096 if (pgio->pg_lseg == NULL)
1f945357 1097 nfs_pageio_reset_read_mds(pgio);
e885de1a 1098
d8007d4d
TM
1099}
1100EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1101
1102void
1103pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1104{
1105 BUG_ON(pgio->pg_lseg != NULL);
1106
1825a0d0
FI
1107 if (req->wb_offset != req->wb_pgbase) {
1108 nfs_pageio_reset_write_mds(pgio);
1109 return;
1110 }
d8007d4d
TM
1111 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1112 req->wb_context,
1113 req_offset(req),
1114 req->wb_bytes,
1115 IOMODE_RW,
1116 GFP_NOFS);
e885de1a
TM
1117 /* If no lseg, fall back to write through mds */
1118 if (pgio->pg_lseg == NULL)
1f945357 1119 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM
1120}
1121EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1122
1751c363 1123bool
061ae2ed
FI
1124pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode,
1125 const struct nfs_pgio_completion_ops *compl_ops)
1751c363
TM
1126{
1127 struct nfs_server *server = NFS_SERVER(inode);
1128 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1129
1130 if (ld == NULL)
1131 return false;
061ae2ed
FI
1132 nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops,
1133 server->rsize, 0);
1751c363
TM
1134 return true;
1135}
1136
1137bool
061ae2ed
FI
1138pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode,
1139 int ioflags,
1140 const struct nfs_pgio_completion_ops *compl_ops)
1751c363
TM
1141{
1142 struct nfs_server *server = NFS_SERVER(inode);
1143 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1144
1145 if (ld == NULL)
1146 return false;
061ae2ed
FI
1147 nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops,
1148 server->wsize, ioflags);
1751c363
TM
1149 return true;
1150}
1151
18ad0a9f 1152bool
dfed206b
BH
1153pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1154 struct nfs_page *req)
94ad1c80 1155{
d8007d4d
TM
1156 if (pgio->pg_lseg == NULL)
1157 return nfs_generic_pg_test(pgio, prev, req);
94ad1c80 1158
19982ba8
TM
1159 /*
1160 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1161 * Note that this test makes several assumptions:
1162 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1163 * is known to lie within the range.
1164 * - that the nfs_page being tested is known to be contiguous with the
1165 * previous nfs_page.
1166 * - Layout ranges are page aligned, so we only have to test the
1167 * start offset of the request.
1168 *
1169 * Please also note that 'end_offset' is actually the offset of the
1170 * first byte that lies outside the pnfs_layout_range. FIXME?
1171 *
1172 */
1173 return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
1174 pgio->pg_lseg->pls_range.length);
94ad1c80 1175}
89a58e32 1176EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 1177
061ae2ed
FI
1178static int pnfs_write_done_resend_to_mds(struct inode *inode,
1179 struct list_head *head,
1180 const struct nfs_pgio_completion_ops *compl_ops)
e2fecb21
TM
1181{
1182 struct nfs_pageio_descriptor pgio;
1183 LIST_HEAD(failed);
1184
1185 /* Resend all requests through the MDS */
061ae2ed 1186 nfs_pageio_init_write_mds(&pgio, inode, FLUSH_STABLE, compl_ops);
e2fecb21
TM
1187 while (!list_empty(head)) {
1188 struct nfs_page *req = nfs_list_entry(head->next);
1189
1190 nfs_list_remove_request(req);
1191 if (!nfs_pageio_add_request(&pgio, req))
1192 nfs_list_add_request(req, &failed);
1193 }
1194 nfs_pageio_complete(&pgio);
1195
1196 if (!list_empty(&failed)) {
1197 /* For some reason our attempt to resend pages. Mark the
1198 * overall send request as having failed, and let
1199 * nfs_writeback_release_full deal with the error.
1200 */
1201 list_move(&failed, head);
1202 return -EIO;
1203 }
1204 return 0;
1205}
1206
1acbbb4e
FI
1207static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
1208{
cd841605
FI
1209 struct nfs_pgio_header *hdr = data->header;
1210
1211 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1212 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 1213 PNFS_LAYOUTRET_ON_ERROR) {
cd841605
FI
1214 clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
1215 pnfs_return_layout(hdr->inode);
1acbbb4e 1216 }
6c75dc0d
FI
1217 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1218 data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
061ae2ed
FI
1219 &hdr->pages,
1220 hdr->completion_ops);
1acbbb4e
FI
1221}
1222
d20581aa
BH
1223/*
1224 * Called by non rpc-based layout drivers
1225 */
8ce160c5 1226void pnfs_ld_write_done(struct nfs_write_data *data)
44b83799 1227{
cd841605
FI
1228 struct nfs_pgio_header *hdr = data->header;
1229
1230 if (!hdr->pnfs_error) {
d20581aa 1231 pnfs_set_layoutcommit(data);
cd841605 1232 hdr->mds_ops->rpc_call_done(&data->task, data);
1acbbb4e
FI
1233 } else
1234 pnfs_ld_handle_write_error(data);
cd841605 1235 hdr->mds_ops->rpc_release(data);
44b83799 1236}
d20581aa 1237EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 1238
dce81290
TM
1239static void
1240pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1241 struct nfs_write_data *data)
1242{
cd841605
FI
1243 struct nfs_pgio_header *hdr = data->header;
1244
6c75dc0d
FI
1245 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1246 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1247 nfs_pageio_reset_write_mds(desc);
1248 desc->pg_recoalesce = 1;
1249 }
dce81290
TM
1250 nfs_writedata_release(data);
1251}
1252
1253static enum pnfs_try_status
0382b744 1254pnfs_try_to_write_data(struct nfs_write_data *wdata,
dce81290
TM
1255 const struct rpc_call_ops *call_ops,
1256 struct pnfs_layout_segment *lseg,
1257 int how)
0382b744 1258{
cd841605
FI
1259 struct nfs_pgio_header *hdr = wdata->header;
1260 struct inode *inode = hdr->inode;
0382b744
AA
1261 enum pnfs_try_status trypnfs;
1262 struct nfs_server *nfss = NFS_SERVER(inode);
1263
cd841605 1264 hdr->mds_ops = call_ops;
0382b744
AA
1265
1266 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1267 inode->i_ino, wdata->args.count, wdata->args.offset, how);
0382b744 1268 trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
6c75dc0d 1269 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 1270 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA
1271 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1272 return trypnfs;
1273}
1274
dce81290
TM
1275static void
1276pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
1277{
1278 struct nfs_write_data *data;
1279 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1280 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1281
1282 desc->pg_lseg = NULL;
1283 while (!list_empty(head)) {
1284 enum pnfs_try_status trypnfs;
1285
6c75dc0d 1286 data = list_first_entry(head, struct nfs_write_data, list);
dce81290
TM
1287 list_del_init(&data->list);
1288
1289 trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
1290 if (trypnfs == PNFS_NOT_ATTEMPTED)
1291 pnfs_write_through_mds(desc, data);
1292 }
1293 put_lseg(lseg);
1294}
1295
6c75dc0d
FI
1296static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
1297{
1298 put_lseg(hdr->lseg);
1299 nfs_writehdr_free(hdr);
1300}
1301
dce81290
TM
1302int
1303pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1304{
6c75dc0d
FI
1305 struct nfs_write_header *whdr;
1306 struct nfs_pgio_header *hdr;
dce81290
TM
1307 int ret;
1308
6c75dc0d
FI
1309 whdr = nfs_writehdr_alloc();
1310 if (!whdr) {
9b5415b5 1311 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
dce81290
TM
1312 put_lseg(desc->pg_lseg);
1313 desc->pg_lseg = NULL;
6c75dc0d 1314 return -ENOMEM;
dce81290 1315 }
6c75dc0d
FI
1316 hdr = &whdr->header;
1317 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1318 hdr->lseg = get_lseg(desc->pg_lseg);
1319 atomic_inc(&hdr->refcnt);
1320 ret = nfs_generic_flush(desc, hdr);
1321 if (ret != 0) {
1322 put_lseg(desc->pg_lseg);
1323 desc->pg_lseg = NULL;
6c75dc0d
FI
1324 } else
1325 pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
1326 if (atomic_dec_and_test(&hdr->refcnt))
061ae2ed 1327 hdr->completion_ops->completion(hdr);
6c75dc0d 1328 return ret;
dce81290
TM
1329}
1330EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
1331
061ae2ed
FI
1332static int pnfs_read_done_resend_to_mds(struct inode *inode,
1333 struct list_head *head,
1334 const struct nfs_pgio_completion_ops *compl_ops)
62e4a769
TM
1335{
1336 struct nfs_pageio_descriptor pgio;
1acbbb4e 1337 LIST_HEAD(failed);
62e4a769 1338
1acbbb4e 1339 /* Resend all requests through the MDS */
061ae2ed 1340 nfs_pageio_init_read_mds(&pgio, inode, compl_ops);
1acbbb4e
FI
1341 while (!list_empty(head)) {
1342 struct nfs_page *req = nfs_list_entry(head->next);
62e4a769
TM
1343
1344 nfs_list_remove_request(req);
1acbbb4e
FI
1345 if (!nfs_pageio_add_request(&pgio, req))
1346 nfs_list_add_request(req, &failed);
62e4a769
TM
1347 }
1348 nfs_pageio_complete(&pgio);
1acbbb4e
FI
1349
1350 if (!list_empty(&failed)) {
1351 list_move(&failed, head);
1352 return -EIO;
1353 }
1354 return 0;
1355}
1356
1357static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
1358{
cd841605
FI
1359 struct nfs_pgio_header *hdr = data->header;
1360
1361 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
1362 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 1363 PNFS_LAYOUTRET_ON_ERROR) {
cd841605
FI
1364 clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
1365 pnfs_return_layout(hdr->inode);
1acbbb4e 1366 }
4db6e0b7
FI
1367 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1368 data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
061ae2ed
FI
1369 &hdr->pages,
1370 hdr->completion_ops);
62e4a769
TM
1371}
1372
d20581aa
BH
1373/*
1374 * Called by non rpc-based layout drivers
1375 */
9b7eecdc 1376void pnfs_ld_read_done(struct nfs_read_data *data)
d20581aa 1377{
cd841605
FI
1378 struct nfs_pgio_header *hdr = data->header;
1379
1380 if (likely(!hdr->pnfs_error)) {
d20581aa 1381 __nfs4_read_done_cb(data);
cd841605 1382 hdr->mds_ops->rpc_call_done(&data->task, data);
62e4a769
TM
1383 } else
1384 pnfs_ld_handle_read_error(data);
cd841605 1385 hdr->mds_ops->rpc_release(data);
d20581aa
BH
1386}
1387EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
1388
493292dd
TM
1389static void
1390pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1391 struct nfs_read_data *data)
1392{
cd841605
FI
1393 struct nfs_pgio_header *hdr = data->header;
1394
4db6e0b7
FI
1395 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1396 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1397 nfs_pageio_reset_read_mds(desc);
1398 desc->pg_recoalesce = 1;
1399 }
493292dd
TM
1400 nfs_readdata_release(data);
1401}
1402
64419a9b
AA
1403/*
1404 * Call the appropriate parallel I/O subsystem read function.
1405 */
493292dd 1406static enum pnfs_try_status
64419a9b 1407pnfs_try_to_read_data(struct nfs_read_data *rdata,
493292dd
TM
1408 const struct rpc_call_ops *call_ops,
1409 struct pnfs_layout_segment *lseg)
64419a9b 1410{
cd841605
FI
1411 struct nfs_pgio_header *hdr = rdata->header;
1412 struct inode *inode = hdr->inode;
64419a9b
AA
1413 struct nfs_server *nfss = NFS_SERVER(inode);
1414 enum pnfs_try_status trypnfs;
1415
cd841605 1416 hdr->mds_ops = call_ops;
64419a9b
AA
1417
1418 dprintk("%s: Reading ino:%lu %u@%llu\n",
1419 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
1420
1421 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
4db6e0b7 1422 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 1423 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA
1424 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1425 return trypnfs;
1426}
863a3c6c 1427
493292dd
TM
1428static void
1429pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
1430{
1431 struct nfs_read_data *data;
1432 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1433 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1434
1435 desc->pg_lseg = NULL;
1436 while (!list_empty(head)) {
1437 enum pnfs_try_status trypnfs;
1438
4db6e0b7 1439 data = list_first_entry(head, struct nfs_read_data, list);
493292dd
TM
1440 list_del_init(&data->list);
1441
1442 trypnfs = pnfs_try_to_read_data(data, call_ops, lseg);
1443 if (trypnfs == PNFS_NOT_ATTEMPTED)
1444 pnfs_read_through_mds(desc, data);
1445 }
1446 put_lseg(lseg);
1447}
1448
4db6e0b7
FI
1449static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
1450{
1451 put_lseg(hdr->lseg);
1452 nfs_readhdr_free(hdr);
1453}
1454
493292dd
TM
1455int
1456pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
1457{
4db6e0b7
FI
1458 struct nfs_read_header *rhdr;
1459 struct nfs_pgio_header *hdr;
493292dd
TM
1460 int ret;
1461
4db6e0b7
FI
1462 rhdr = nfs_readhdr_alloc();
1463 if (!rhdr) {
061ae2ed 1464 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
4db6e0b7 1465 ret = -ENOMEM;
493292dd
TM
1466 put_lseg(desc->pg_lseg);
1467 desc->pg_lseg = NULL;
1468 return ret;
1469 }
4db6e0b7
FI
1470 hdr = &rhdr->header;
1471 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1472 hdr->lseg = get_lseg(desc->pg_lseg);
1473 atomic_inc(&hdr->refcnt);
1474 ret = nfs_generic_pagein(desc, hdr);
1475 if (ret != 0) {
1476 put_lseg(desc->pg_lseg);
1477 desc->pg_lseg = NULL;
4db6e0b7
FI
1478 } else
1479 pnfs_do_multiple_reads(desc, &hdr->rpc_list);
1480 if (atomic_dec_and_test(&hdr->refcnt))
061ae2ed 1481 hdr->completion_ops->completion(hdr);
4db6e0b7 1482 return ret;
493292dd
TM
1483}
1484EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
1485
863a3c6c 1486/*
a9bae566 1487 * There can be multiple RW segments.
863a3c6c 1488 */
a9bae566 1489static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 1490{
a9bae566 1491 struct pnfs_layout_segment *lseg;
863a3c6c 1492
a9bae566
PT
1493 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
1494 if (lseg->pls_range.iomode == IOMODE_RW &&
1495 test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1496 list_add(&lseg->pls_lc_list, listp);
1497 }
863a3c6c
AA
1498}
1499
1b0ae068
PT
1500void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
1501{
1502 if (lseg->pls_range.iomode == IOMODE_RW) {
1503 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__);
1504 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
1505 } else {
1506 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__);
1507 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
1508 }
1509}
1510EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
1511
863a3c6c
AA
1512void
1513pnfs_set_layoutcommit(struct nfs_write_data *wdata)
1514{
cd841605
FI
1515 struct nfs_pgio_header *hdr = wdata->header;
1516 struct inode *inode = hdr->inode;
1517 struct nfs_inode *nfsi = NFS_I(inode);
4b8ee2b8 1518 loff_t end_pos = wdata->mds_offset + wdata->res.count;
79a48a1f 1519 bool mark_as_dirty = false;
863a3c6c 1520
cd841605 1521 spin_lock(&inode->i_lock);
863a3c6c 1522 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
79a48a1f 1523 mark_as_dirty = true;
863a3c6c 1524 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 1525 __func__, inode->i_ino);
863a3c6c 1526 }
cd841605 1527 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
a9bae566 1528 /* references matched in nfs4_layoutcommit_release */
cd841605 1529 get_lseg(hdr->lseg);
a9bae566 1530 }
acff5880
PT
1531 if (end_pos > nfsi->layout->plh_lwb)
1532 nfsi->layout->plh_lwb = end_pos;
cd841605 1533 spin_unlock(&inode->i_lock);
acff5880 1534 dprintk("%s: lseg %p end_pos %llu\n",
cd841605 1535 __func__, hdr->lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA
1536
1537 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1538 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1539 if (mark_as_dirty)
cd841605 1540 mark_inode_dirty_sync(inode);
863a3c6c
AA
1541}
1542EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1543
db29c089
AA
1544void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
1545{
1546 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
1547
1548 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
1549 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
1550}
1551
de4b15c7
AA
1552/*
1553 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1554 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1555 * data to disk to allow the server to recover the data if it crashes.
1556 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1557 * is off, and a COMMIT is sent to a data server, or
1558 * if WRITEs to a data server return NFS_DATA_SYNC.
1559 */
863a3c6c 1560int
ef311537 1561pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c
AA
1562{
1563 struct nfs4_layoutcommit_data *data;
1564 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c
AA
1565 loff_t end_pos;
1566 int status = 0;
1567
1568 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1569
de4b15c7
AA
1570 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1571 return 0;
1572
863a3c6c
AA
1573 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1574 data = kzalloc(sizeof(*data), GFP_NOFS);
de4b15c7 1575 if (!data) {
de4b15c7
AA
1576 status = -ENOMEM;
1577 goto out;
1578 }
863a3c6c 1579
92407e75
PT
1580 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1581 goto out_free;
1582
1583 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
1584 if (!sync) {
1585 status = -EAGAIN;
1586 goto out_free;
1587 }
1588 status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING,
1589 nfs_wait_bit_killable, TASK_KILLABLE);
1590 if (status)
1591 goto out_free;
1592 }
1593
a9bae566 1594 INIT_LIST_HEAD(&data->lseg_list);
de4b15c7 1595 spin_lock(&inode->i_lock);
863a3c6c 1596 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
92407e75 1597 clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
863a3c6c 1598 spin_unlock(&inode->i_lock);
92407e75
PT
1599 wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
1600 goto out_free;
863a3c6c 1601 }
a9bae566
PT
1602
1603 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 1604
acff5880 1605 end_pos = nfsi->layout->plh_lwb;
acff5880 1606 nfsi->layout->plh_lwb = 0;
863a3c6c 1607
f597c537 1608 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA
1609 spin_unlock(&inode->i_lock);
1610
1611 data->args.inode = inode;
9fa40758 1612 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
863a3c6c
AA
1613 nfs_fattr_init(&data->fattr);
1614 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1615 data->res.fattr = &data->fattr;
1616 data->args.lastbytewritten = end_pos - 1;
1617 data->res.server = NFS_SERVER(inode);
1618
1619 status = nfs4_proc_layoutcommit(data, sync);
1620out:
92407e75
PT
1621 if (status)
1622 mark_inode_dirty_sync(inode);
863a3c6c
AA
1623 dprintk("<-- %s status %d\n", __func__, status);
1624 return status;
92407e75
PT
1625out_free:
1626 kfree(data);
1627 goto out;
863a3c6c 1628}