projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'nfs-for-4.15-1' of git://git.linux-nfs.org/projects/anna/linux-nfs
[linux-2.6-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>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
85e174ba
RL		 40
41#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 42#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 43
02c35fca
FI		 44/* Locking:
45 *
46 * pnfs_spinlock:
47 * protects pnfs_modules_tbl.
48 */
49static DEFINE_SPINLOCK(pnfs_spinlock);
50
51/*
52 * pnfs_modules_tbl holds all pnfs modules
53 */
54static LIST_HEAD(pnfs_modules_tbl);
55
13c13a6a 56static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 57static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
58 struct list_head *free_me,
59 const struct pnfs_layout_range *range,
60 u32 seq);
fe1cf946
TM		 61static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
62 struct list_head *tmp_list);
aa1e0e3a 63
02c35fca
FI		 64/* Return the registered pnfs layout driver module matching given id */
65static struct pnfs_layoutdriver_type *
66find_pnfs_driver_locked(u32 id)
67{
68 struct pnfs_layoutdriver_type *local;
69
70 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
71 if (local->id == id)
72 goto out;
73 local = NULL;
74out:
75 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
76 return local;
77}
78
85e174ba
RL		 79static struct pnfs_layoutdriver_type *
80find_pnfs_driver(u32 id)
81{
02c35fca
FI		 82 struct pnfs_layoutdriver_type *local;
83
84 spin_lock(&pnfs_spinlock);
85 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 86 if (local != NULL && !try_module_get(local->owner)) {
87 dprintk("%s: Could not grab reference on module\n", __func__);
88 local = NULL;
89 }
02c35fca
FI		 90 spin_unlock(&pnfs_spinlock);
91 return local;
85e174ba
RL		 92}
93
94void
95unset_pnfs_layoutdriver(struct nfs_server *nfss)
96{
738fd0f3
BH		 97 if (nfss->pnfs_curr_ld) {
98 if (nfss->pnfs_curr_ld->clear_layoutdriver)
99 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 100 /* Decrement the MDS count. Purge the deviceid cache if zero */
101 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
102 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 103 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 104 }
85e174ba
RL		 105 nfss->pnfs_curr_ld = NULL;
106}
107
ca440c38
JL		 108/*
109 * When the server sends a list of layout types, we choose one in the order
110 * given in the list below.
111 *
112 * FIXME: should this list be configurable in some fashion? module param?
113 * mount option? something else?
114 */
115static const u32 ld_prefs[] = {
116 LAYOUT_SCSI,
117 LAYOUT_BLOCK_VOLUME,
118 LAYOUT_OSD2_OBJECTS,
119 LAYOUT_FLEX_FILES,
120 LAYOUT_NFSV4_1_FILES,
121 0
122};
123
124static int
125ld_cmp(const void *e1, const void *e2)
126{
127 u32 ld1 = *((u32 *)e1);
128 u32 ld2 = *((u32 *)e2);
129 int i;
130
131 for (i = 0; ld_prefs[i] != 0; i++) {
132 if (ld1 == ld_prefs[i])
133 return -1;
134
135 if (ld2 == ld_prefs[i])
136 return 1;
137 }
138 return 0;
139}
140
85e174ba
RL		 141/*
142 * Try to set the server's pnfs module to the pnfs layout type specified by id.
143 * Currently only one pNFS layout driver per filesystem is supported.
144 *
3132e49e 145 * @ids array of layout types supported by MDS.
85e174ba
RL		 146 */
147void
738fd0f3 148set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 149 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 150{
151 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 152 u32 id;
ca440c38 153 int i;
85e174ba 154
19274716
AS		 155 if (fsinfo->nlayouttypes == 0)
156 goto out_no_driver;
85e174ba
RL		 157 if (!(server->nfs_client->cl_exchange_flags &
158 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 159 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
160 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 161 goto out_no_driver;
162 }
3132e49e 163
ca440c38
JL		 164 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
165 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 166
ca440c38
JL		 167 for (i = 0; i < fsinfo->nlayouttypes; i++) {
168 id = fsinfo->layouttype[i];
85e174ba 169 ld_type = find_pnfs_driver(id);
ca440c38
JL		 170 if (!ld_type) {
171 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
172 id);
173 ld_type = find_pnfs_driver(id);
174 }
175 if (ld_type)
176 break;
85e174ba 177 }
3132e49e
JL		 178
179 if (!ld_type) {
ca440c38 180 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 181 goto out_no_driver;
182 }
183
85e174ba 184 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 185 if (ld_type->set_layoutdriver
186 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 187 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
188 "driver %u.\n", __func__, id);
738fd0f3
BH		 189 module_put(ld_type->owner);
190 goto out_no_driver;
191 }
2a4c8994
TM		 192 /* Bump the MDS count */
193 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 194
85e174ba
RL		 195 dprintk("%s: pNFS module for %u set\n", __func__, id);
196 return;
197
198out_no_driver:
199 dprintk("%s: Using NFSv4 I/O\n", __func__);
200 server->pnfs_curr_ld = NULL;
201}
02c35fca
FI		 202
203int
204pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
205{
206 int status = -EINVAL;
207 struct pnfs_layoutdriver_type *tmp;
208
209 if (ld_type->id == 0) {
a030889a 210 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 211 return status;
212 }
b1f69b75 213 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 214 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 215 "alloc_lseg and free_lseg.\n", __func__);
216 return status;
217 }
02c35fca
FI		 218
219 spin_lock(&pnfs_spinlock);
220 tmp = find_pnfs_driver_locked(ld_type->id);
221 if (!tmp) {
222 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
223 status = 0;
224 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
225 ld_type->name);
226 } else {
a030889a 227 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 228 __func__, ld_type->id);
229 }
230 spin_unlock(&pnfs_spinlock);
231
232 return status;
233}
234EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
235
236void
237pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
238{
239 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
240 spin_lock(&pnfs_spinlock);
241 list_del(&ld_type->pnfs_tblid);
242 spin_unlock(&pnfs_spinlock);
243}
244EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 245
b1f69b75
AA		 246/*
247 * pNFS client layout cache
248 */
249
cc6e5340 250/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 251void
70c3bd2b 252pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 253{
2b28a7be 254 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 255}
256
636fb9c8
BH		 257static struct pnfs_layout_hdr *
258pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
259{
260 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 261 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 262}
263
264static void
265pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
266{
9c626381
TM		 267 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
268 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
269
270 if (!list_empty(&lo->plh_layouts)) {
271 struct nfs_client *clp = server->nfs_client;
272
273 spin_lock(&clp->cl_lock);
274 list_del_init(&lo->plh_layouts);
275 spin_unlock(&clp->cl_lock);
276 }
9fa40758 277 put_rpccred(lo->plh_lc_cred);
57934278 278 return ld->free_layout_hdr(lo);
636fb9c8
BH		 279}
280
e5e94017 281static void
6622c3ea 282pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 283{
bb346f63 284 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 285 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 286 nfsi->layout = NULL;
287 /* Reset MDS Threshold I/O counters */
288 nfsi->write_io = 0;
289 nfsi->read_io = 0;
e5e94017
BH		 290}
291
b1f69b75 292void
70c3bd2b 293pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 294{
cc6e5340
FI		 295 struct inode *inode = lo->plh_inode;
296
13c13a6a
TM		 297 pnfs_layoutreturn_before_put_layout_hdr(lo);
298
2b28a7be 299 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 300 if (!list_empty(&lo->plh_segs))
301 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 302 pnfs_detach_layout_hdr(lo);
cc6e5340 303 spin_unlock(&inode->i_lock);
6622c3ea 304 pnfs_free_layout_hdr(lo);
cc6e5340 305 }
974cec8c
AA		 306}
307
4aab9732
TM		 308static void
309pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
310 u32 seq)
311{
312 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
313 iomode = IOMODE_ANY;
314 lo->plh_return_iomode = iomode;
315 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
316 if (seq != 0) {
317 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
318 lo->plh_return_seq = seq;
319 }
320}
321
ae5a459d
TM		 322static void
323pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
324{
5466d214 325 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 326 lo->plh_return_iomode = 0;
327 lo->plh_return_seq = 0;
328 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 329 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
330 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
331 continue;
332 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
333 }
ae5a459d
TM		 334}
335
362fb578
TM		 336static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
337{
338 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
339 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
340 smp_mb__after_atomic();
341 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
342 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
343}
344
fe1cf946
TM		 345static void
346pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
347 struct list_head *free_me)
348{
349 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
350 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
351 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
352 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
353 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
354 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
355}
356
7380020e
TM		 357/*
358 * Update the seqid of a layout stateid
359 */
360bool nfs4_refresh_layout_stateid(nfs4_stateid *dst, struct inode *inode)
361{
362 struct pnfs_layout_hdr *lo;
363 bool ret = false;
364
365 spin_lock(&inode->i_lock);
366 lo = NFS_I(inode)->layout;
367 if (lo && nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
368 dst->seqid = lo->plh_stateid.seqid;
369 ret = true;
370 }
371 spin_unlock(&inode->i_lock);
372 return ret;
373}
374
2454dfea
TM		 375/*
376 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
377 *
378 * In order to continue using the pnfs_layout_hdr, a full recovery
379 * is required.
380 * Note that caller must hold inode->i_lock.
381 */
5f46be04 382int
2454dfea
TM		 383pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
384 struct list_head *lseg_list)
385{
386 struct pnfs_layout_range range = {
387 .iomode = IOMODE_ANY,
388 .offset = 0,
389 .length = NFS4_MAX_UINT64,
390 };
fe1cf946 391 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 392
393 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 394 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
395 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 396 pnfs_clear_layoutreturn_info(lo);
68f74479 397 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 398 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
399 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
400 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 401 return !list_empty(&lo->plh_segs);
2454dfea
TM		 402}
403
b9e028fd
TM		 404static int
405pnfs_iomode_to_fail_bit(u32 iomode)
406{
407 return iomode == IOMODE_RW ?
408 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
409}
410
411static void
3e621214 412pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 413{
25c75333 414 lo->plh_retry_timestamp = jiffies;
39e88fcf 415 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 416 refcount_inc(&lo->plh_refcount);
3e621214
TM		 417}
418
419static void
420pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
421{
422 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 423 refcount_dec(&lo->plh_refcount);
3e621214
TM		 424}
425
426static void
427pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
428{
429 struct inode *inode = lo->plh_inode;
115ce575
TM		 430 struct pnfs_layout_range range = {
431 .iomode = iomode,
432 .offset = 0,
433 .length = NFS4_MAX_UINT64,
434 };
435 LIST_HEAD(head);
3e621214
TM		 436
437 spin_lock(&inode->i_lock);
438 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 439 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 440 spin_unlock(&inode->i_lock);
115ce575 441 pnfs_free_lseg_list(&head);
b9e028fd
TM		 442 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
443 iomode == IOMODE_RW ? "RW" : "READ");
444}
445
446static bool
447pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
448{
25c75333 449 unsigned long start, end;
3e621214
TM		 450 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
451
452 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 453 return false;
454 end = jiffies;
455 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
456 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
457 /* It is time to retry the failed layoutgets */
3e621214 458 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 459 return false;
460 }
461 return true;
b9e028fd
TM		 462}
463
974cec8c 464static void
119cef97
TM		 465pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
466 const struct pnfs_layout_range *range,
467 const nfs4_stateid *stateid)
974cec8c 468{
566052c5 469 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 470 INIT_LIST_HEAD(&lseg->pls_lc_list);
eba6dd69 471 refcount_set(&lseg->pls_refcount, 1);
4541d16c 472 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 473 lseg->pls_layout = lo;
119cef97
TM		 474 lseg->pls_range = *range;
475 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 476}
477
905ca191 478static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 479{
68f74479
TM		 480 if (lseg != NULL) {
481 struct inode *inode = lseg->pls_layout->plh_inode;
482 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
483 }
974cec8c
AA		 484}
485
d684d2ae 486static void
57036a37
TM		 487pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
488 struct pnfs_layout_segment *lseg)
d684d2ae 489{
d20581aa 490 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 491 list_del_init(&lseg->pls_list);
8f0d27dc 492 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 493 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 494 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
495 return;
7b650994
TM		 496 if (list_empty(&lo->plh_segs) &&
497 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
498 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 499 if (atomic_read(&lo->plh_outstanding) == 0)
500 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 501 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 502 }
d684d2ae
FI		 503}
504
68f74479
TM		 505static bool
506pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
507 struct pnfs_layout_segment *lseg)
508{
509 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
510 pnfs_layout_is_valid(lo)) {
4aab9732 511 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 512 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
513 return true;
514 }
515 return false;
516}
517
bae724ef 518void
9369a431 519pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 520{
57036a37 521 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 522 struct inode *inode;
523
524 if (!lseg)
525 return;
526
4541d16c 527 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 528 refcount_read(&lseg->pls_refcount),
4541d16c 529 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 530
57036a37
TM		 531 lo = lseg->pls_layout;
532 inode = lo->plh_inode;
4ef2e4f8 533
eba6dd69 534 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 535 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
536 spin_unlock(&inode->i_lock);
537 return;
538 }
8f0d27dc 539 pnfs_get_layout_hdr(lo);
4ef2e4f8 540 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 541 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
542 lseg = NULL;
4ef2e4f8
TM		 543 spin_unlock(&inode->i_lock);
544 pnfs_free_lseg(lseg);
545 pnfs_put_layout_hdr(lo);
4541d16c 546 }
4541d16c 547}
9369a431 548EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 549
fb3296eb
BH		 550/*
551 * is l2 fully contained in l1?
552 * start1 end1
553 * [----------------------------------)
554 * start2 end2
555 * [----------------)
556 */
3cb2df17 557static bool
7dc0ac70 558pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 559 const struct pnfs_layout_range *l2)
fb3296eb
BH		 560{
561 u64 start1 = l1->offset;
17822b20 562 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 563 u64 start2 = l2->offset;
17822b20 564 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 565
566 return (start1 <= start2) && (end1 >= end2);
567}
568
24956804
TM		 569static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
570 struct list_head *tmp_list)
571{
eba6dd69 572 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 573 return false;
574 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
575 list_add(&lseg->pls_list, tmp_list);
576 return true;
577}
578
4541d16c
FI		 579/* Returns 1 if lseg is removed from list, 0 otherwise */
580static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
581 struct list_head *tmp_list)
582{
583 int rv = 0;
584
585 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
586 /* Remove the reference keeping the lseg in the
587 * list. It will now be removed when all
588 * outstanding io is finished.
589 */
d684d2ae 590 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 591 refcount_read(&lseg->pls_refcount));
24956804 592 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 593 rv = 1;
4541d16c
FI		 594 }
595 return rv;
596}
597
6d597e17
JL		 598/*
599 * Compare 2 layout stateid sequence ids, to see which is newer,
600 * taking into account wraparound issues.
601 */
602static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
603{
604 return (s32)(s1 - s2) > 0;
605}
606
e036f464
TM		 607static bool
608pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
609 const struct pnfs_layout_range *recall_range)
610{
611 return (recall_range->iomode == IOMODE_ANY ||
612 lseg_range->iomode == recall_range->iomode) &&
613 pnfs_lseg_range_intersecting(lseg_range, recall_range);
614}
615
616static bool
617pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
618 const struct pnfs_layout_range *recall_range,
619 u32 seq)
620{
621 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
622 return false;
623 if (recall_range == NULL)
624 return true;
625 return pnfs_should_free_range(&lseg->pls_range, recall_range);
626}
627
6d597e17
JL		 628/**
629 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
630 * @lo: layout header containing the lsegs
631 * @tmp_list: list head where doomed lsegs should go
632 * @recall_range: optional recall range argument to match (may be NULL)
633 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
634 *
635 * Walk the list of lsegs in the layout header, and tear down any that should
636 * be destroyed. If "recall_range" is specified then the segment must match
637 * that range. If "seq" is non-zero, then only match segments that were handed
638 * out at or before that sequence.
639 *
640 * Returns number of matching invalid lsegs remaining in list after scanning
641 * it and purging them.
4541d16c 642 */
43f1b3da 643int
49a85061 644pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 645 struct list_head *tmp_list,
6d597e17
JL		 646 const struct pnfs_layout_range *recall_range,
647 u32 seq)
974cec8c
AA		 648{
649 struct pnfs_layout_segment *lseg, *next;
71b39854 650 int remaining = 0;
974cec8c
AA		 651
652 dprintk("%s:Begin lo %p\n", __func__, lo);
653
8006bfba 654 if (list_empty(&lo->plh_segs))
38511722 655 return 0;
4541d16c 656 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 657 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
6d597e17 658 dprintk("%s: freeing lseg %p iomode %d seq %u"
4541d16c 659 "offset %llu length %llu\n", __func__,
6d597e17
JL		 660 lseg, lseg->pls_range.iomode, lseg->pls_seq,
661 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 662 if (!mark_lseg_invalid(lseg, tmp_list))
663 remaining++;
4541d16c 664 }
71b39854
TM		 665 dprintk("%s:Return %i\n", __func__, remaining);
666 return remaining;
974cec8c
AA		 667}
668
68f74479
TM		 669static void
670pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
671 struct list_head *free_me,
672 const struct pnfs_layout_range *range,
673 u32 seq)
674{
675 struct pnfs_layout_segment *lseg, *next;
676
677 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
678 if (pnfs_match_lseg_recall(lseg, range, seq))
679 list_move_tail(&lseg->pls_list, free_me);
680 }
681}
682
f49f9baa 683/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 684void
4541d16c 685pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 686{
4541d16c 687 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 688
689 if (list_empty(free_me))
690 return;
691
4541d16c 692 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 693 list_del(&lseg->pls_list);
905ca191 694 pnfs_free_lseg(lseg);
974cec8c
AA		 695 }
696}
697
e5e94017
BH		 698void
699pnfs_destroy_layout(struct nfs_inode *nfsi)
700{
701 struct pnfs_layout_hdr *lo;
974cec8c 702 LIST_HEAD(tmp_list);
e5e94017
BH		 703
704 spin_lock(&nfsi->vfs_inode.i_lock);
705 lo = nfsi->layout;
706 if (lo) {
3e621214 707 pnfs_get_layout_hdr(lo);
2454dfea 708 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 709 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
710 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
711 spin_unlock(&nfsi->vfs_inode.i_lock);
712 pnfs_free_lseg_list(&tmp_list);
1f18b82c 713 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 714 pnfs_put_layout_hdr(lo);
715 } else
716 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 717}
041245c8 718EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 719
fd9a8d71
TM		 720static bool
721pnfs_layout_add_bulk_destroy_list(struct inode *inode,
722 struct list_head *layout_list)
974cec8c
AA		 723{
724 struct pnfs_layout_hdr *lo;
fd9a8d71 725 bool ret = false;
974cec8c 726
fd9a8d71
TM		 727 spin_lock(&inode->i_lock);
728 lo = NFS_I(inode)->layout;
729 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
730 pnfs_get_layout_hdr(lo);
731 list_add(&lo->plh_bulk_destroy, layout_list);
732 ret = true;
733 }
734 spin_unlock(&inode->i_lock);
735 return ret;
736}
737
738/* Caller must hold rcu_read_lock and clp->cl_lock */
739static int
740pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
741 struct nfs_server *server,
742 struct list_head *layout_list)
743{
744 struct pnfs_layout_hdr *lo, *next;
745 struct inode *inode;
746
747 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
b85f5620
TM		 748 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
749 continue;
fd9a8d71
TM		 750 inode = igrab(lo->plh_inode);
751 if (inode == NULL)
752 continue;
753 list_del_init(&lo->plh_layouts);
754 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
755 continue;
756 rcu_read_unlock();
757 spin_unlock(&clp->cl_lock);
758 iput(inode);
759 spin_lock(&clp->cl_lock);
760 rcu_read_lock();
761 return -EAGAIN;
762 }
763 return 0;
764}
765
766static int
767pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
768 bool is_bulk_recall)
769{
770 struct pnfs_layout_hdr *lo;
771 struct inode *inode;
fd9a8d71
TM		 772 LIST_HEAD(lseg_list);
773 int ret = 0;
774
775 while (!list_empty(layout_list)) {
776 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
777 plh_bulk_destroy);
778 dprintk("%s freeing layout for inode %lu\n", __func__,
779 lo->plh_inode->i_ino);
780 inode = lo->plh_inode;
7c5d1875
CH		 781
782 pnfs_layoutcommit_inode(inode, false);
783
fd9a8d71
TM		 784 spin_lock(&inode->i_lock);
785 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 786 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
787 if (is_bulk_recall)
788 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 789 ret = -EAGAIN;
9fd4b9fc 790 }
fd9a8d71
TM		 791 spin_unlock(&inode->i_lock);
792 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 793 /* Free all lsegs that are attached to commit buckets */
794 nfs_commit_inode(inode, 0);
fd9a8d71
TM		 795 pnfs_put_layout_hdr(lo);
796 iput(inode);
797 }
798 return ret;
799}
800
801int
802pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
803 struct nfs_fsid *fsid,
804 bool is_recall)
805{
806 struct nfs_server *server;
807 LIST_HEAD(layout_list);
c47abcf8 808
974cec8c 809 spin_lock(&clp->cl_lock);
6382a441 810 rcu_read_lock();
fd9a8d71 811restart:
6382a441 812 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 813 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
814 continue;
815 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
816 server,
817 &layout_list) != 0)
818 goto restart;
6382a441
WAA		 819 }
820 rcu_read_unlock();
974cec8c
AA		 821 spin_unlock(&clp->cl_lock);
822
fd9a8d71
TM		 823 if (list_empty(&layout_list))
824 return 0;
825 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
826}
827
828int
829pnfs_destroy_layouts_byclid(struct nfs_client *clp,
830 bool is_recall)
831{
832 struct nfs_server *server;
833 LIST_HEAD(layout_list);
834
835 spin_lock(&clp->cl_lock);
836 rcu_read_lock();
837restart:
838 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
839 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
840 server,
841 &layout_list) != 0)
842 goto restart;
974cec8c 843 }
fd9a8d71
TM		 844 rcu_read_unlock();
845 spin_unlock(&clp->cl_lock);
846
847 if (list_empty(&layout_list))
848 return 0;
849 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
850}
851
852/*
853 * Called by the state manger to remove all layouts established under an
854 * expired lease.
855 */
856void
857pnfs_destroy_all_layouts(struct nfs_client *clp)
858{
859 nfs4_deviceid_mark_client_invalid(clp);
860 nfs4_deviceid_purge_client(clp);
861
862 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 863}
864
fd6002e9 865/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 866void
867pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
868 bool update_barrier)
b1f69b75 869{
ecebb80b 870 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 871
2d2f24ad
TM		 872 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
873 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 874
875 if (!pnfs_layout_is_valid(lo)) {
876 nfs4_stateid_copy(&lo->plh_stateid, new);
877 lo->plh_barrier = newseq;
878 pnfs_clear_layoutreturn_info(lo);
879 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
880 return;
881 }
882 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 883 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 884 /*
885 * Because of wraparound, we want to keep the barrier
886 * "close" to the current seqids.
887 */
888 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 889 }
ecebb80b
TM		 890 if (update_barrier)
891 new_barrier = be32_to_cpu(new->seqid);
892 else if (new_barrier == 0)
893 return;
2a59a041 894 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 895 lo->plh_barrier = new_barrier;
b1f69b75
AA		 896}
897
cf7d63f1 898static bool
19c54aba
TM		 899pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
900 const nfs4_stateid *stateid)
43f1b3da 901{
19c54aba 902 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 903
19c54aba
TM		 904 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
905}
906
907/* lget is set to 1 if called from inside send_layoutget call chain */
908static bool
e1c06f80 909pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 910{
f7e8917a 911 return lo->plh_block_lgets ||
e1c06f80 912 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 913}
914
b1f69b75 915/*
183d9e7b
JL		 916 * Get layout from server.
917 * for now, assume that whole file layouts are requested.
918 * arg->offset: 0
919 * arg->length: all ones
920 */
e5e94017
BH		 921static struct pnfs_layout_segment *
922send_layoutget(struct pnfs_layout_hdr *lo,
923 struct nfs_open_context *ctx,
183d9e7b 924 nfs4_stateid *stateid,
e144e539 925 const struct pnfs_layout_range *range,
183d9e7b 926 long *timeout, gfp_t gfp_flags)
e5e94017 927{
b7edfaa1 928 struct inode *ino = lo->plh_inode;
b1f69b75
AA		 929 struct nfs_server *server = NFS_SERVER(ino);
930 struct nfs4_layoutget *lgp;
2d89a1d3 931 loff_t i_size;
b1f69b75
AA		 932
933 dprintk("--> %s\n", __func__);
e5e94017 934
4f2e9dce
JL		 935 /*
936 * Synchronously retrieve layout information from server and
937 * store in lseg. If we race with a concurrent seqid morphing
938 * op, then re-send the LAYOUTGET.
939 */
83026d80
JL		 940 lgp = kzalloc(sizeof(*lgp), gfp_flags);
941 if (lgp == NULL)
942 return ERR_PTR(-ENOMEM);
943
944 i_size = i_size_read(ino);
945
946 lgp->args.minlength = PAGE_SIZE;
947 if (lgp->args.minlength > range->length)
948 lgp->args.minlength = range->length;
949 if (range->iomode == IOMODE_READ) {
950 if (range->offset >= i_size)
951 lgp->args.minlength = 0;
952 else if (i_size - range->offset < lgp->args.minlength)
953 lgp->args.minlength = i_size - range->offset;
d03ab29d 954 }
83026d80
JL		 955 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
956 pnfs_copy_range(&lgp->args.range, range);
957 lgp->args.type = server->pnfs_curr_ld->id;
958 lgp->args.inode = ino;
959 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 960 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80
JL		 961 lgp->gfp_flags = gfp_flags;
962 lgp->cred = lo->plh_lc_cred;
35124a09 963
183d9e7b 964 return nfs4_proc_layoutget(lgp, timeout, gfp_flags);
974cec8c
AA		 965}
966
24956804
TM		 967static void pnfs_clear_layoutcommit(struct inode *inode,
968 struct list_head *head)
969{
970 struct nfs_inode *nfsi = NFS_I(inode);
971 struct pnfs_layout_segment *lseg, *tmp;
972
973 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
974 return;
975 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
976 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
977 continue;
978 pnfs_lseg_dec_and_remove_zero(lseg, head);
979 }
980}
981
68f74479 982void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 983 const nfs4_stateid *arg_stateid,
68f74479 984 const struct pnfs_layout_range *range,
68f74479
TM		 985 const nfs4_stateid *stateid)
986{
987 struct inode *inode = lo->plh_inode;
988 LIST_HEAD(freeme);
989
990 spin_lock(&inode->i_lock);
2a974425
TM		 991 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
992 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
993 goto out_unlock;
68f74479 994 if (stateid) {
2a974425
TM		 995 u32 seq = be32_to_cpu(arg_stateid->seqid);
996
68f74479
TM		 997 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
998 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
999 pnfs_set_layout_stateid(lo, stateid, true);
1000 } else
1001 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1002out_unlock:
68f74479
TM		 1003 pnfs_clear_layoutreturn_waitbit(lo);
1004 spin_unlock(&inode->i_lock);
1005 pnfs_free_lseg_list(&freeme);
1006
1007}
1008
13c13a6a 1009static bool
e5fd1904
TM		 1010pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1011 nfs4_stateid *stateid,
1012 enum pnfs_iomode *iomode)
13c13a6a 1013{
bf0291dd
TM		 1014 /* Serialise LAYOUTGET/LAYOUTRETURN */
1015 if (atomic_read(&lo->plh_outstanding) != 0)
1016 return false;
6604b203 1017 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1018 return false;
6604b203 1019 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1020 pnfs_get_layout_hdr(lo);
e5fd1904
TM		 1021 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1022 if (stateid != NULL) {
1023 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1024 if (lo->plh_return_seq != 0)
1025 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1026 }
1027 if (iomode != NULL)
1028 *iomode = lo->plh_return_iomode;
1029 pnfs_clear_layoutreturn_info(lo);
1030 return true;
1031 }
1032 if (stateid != NULL)
1033 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1034 if (iomode != NULL)
1035 *iomode = IOMODE_ANY;
13c13a6a
TM		 1036 return true;
1037}
1038
828ed9ec
TM		 1039static void
1040pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1041 struct pnfs_layout_hdr *lo,
1042 const nfs4_stateid *stateid,
1043 enum pnfs_iomode iomode)
1044{
1045 struct inode *inode = lo->plh_inode;
1046
1047 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1048 args->inode = inode;
1049 args->range.iomode = iomode;
1050 args->range.offset = 0;
1051 args->range.length = NFS4_MAX_UINT64;
1052 args->layout = lo;
1053 nfs4_stateid_copy(&args->stateid, stateid);
1054}
1055
f40eb5d0 1056static int
ed429d6b 1057pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
6c16605d 1058 enum pnfs_iomode iomode, bool sync)
f40eb5d0
PT		 1059{
1060 struct inode *ino = lo->plh_inode;
287bd3e9 1061 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0
PT		 1062 struct nfs4_layoutreturn *lrp;
1063 int status = 0;
1064
e4af440a 1065 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1066 if (unlikely(lrp == NULL)) {
1067 status = -ENOMEM;
1068 spin_lock(&ino->i_lock);
d67ae825 1069 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0
PT		 1070 spin_unlock(&ino->i_lock);
1071 pnfs_put_layout_hdr(lo);
1072 goto out;
1073 }
1074
828ed9ec 1075 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1076 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0
PT		 1077 lrp->clp = NFS_SERVER(ino)->nfs_client;
1078 lrp->cred = lo->plh_lc_cred;
287bd3e9
TM		 1079 if (ld->prepare_layoutreturn)
1080 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1081
6c16605d 1082 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1083out:
1084 dprintk("<-- %s status: %d\n", __func__, status);
1085 return status;
1086}
1087
13c13a6a
TM		 1088/* Return true if layoutreturn is needed */
1089static bool
1090pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1091{
1092 struct pnfs_layout_segment *s;
1093
2370abda 1094 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1095 return false;
1096
1097 /* Defer layoutreturn until all lsegs are done */
1098 list_for_each_entry(s, &lo->plh_segs, pls_list) {
1099 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1100 return false;
1101 }
1102
1103 return true;
1104}
1105
1106static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1107{
1108 struct inode *inode= lo->plh_inode;
1109
2370abda 1110 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1111 return;
1112 spin_lock(&inode->i_lock);
1113 if (pnfs_layout_need_return(lo)) {
1114 nfs4_stateid stateid;
1115 enum pnfs_iomode iomode;
1116 bool send;
1117
e5fd1904 1118 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
13c13a6a
TM		 1119 spin_unlock(&inode->i_lock);
1120 if (send) {
1121 /* Send an async layoutreturn so we dont deadlock */
1122 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1123 }
1124 } else
1125 spin_unlock(&inode->i_lock);
1126}
1127
293b3b06
AA		 1128/*
1129 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1130 * when the layout segment list is empty.
1131 *
1132 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1133 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1134 * deviceid is marked invalid.
1135 */
cbe82603
BH		 1136int
1137_pnfs_return_layout(struct inode *ino)
1138{
1139 struct pnfs_layout_hdr *lo = NULL;
1140 struct nfs_inode *nfsi = NFS_I(ino);
1141 LIST_HEAD(tmp_list);
cbe82603 1142 nfs4_stateid stateid;
24408f52 1143 int status = 0;
7f27392c 1144 bool send;
cbe82603 1145
366d5052 1146 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1147
1148 spin_lock(&ino->i_lock);
1149 lo = nfsi->layout;
e5929f3c 1150 if (!lo) {
cbe82603 1151 spin_unlock(&ino->i_lock);
293b3b06
AA		 1152 dprintk("NFS: %s no layout to return\n", __func__);
1153 goto out;
cbe82603 1154 }
cbe82603 1155 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1156 pnfs_get_layout_hdr(lo);
24408f52
TM		 1157 /* Is there an outstanding layoutreturn ? */
1158 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1159 spin_unlock(&ino->i_lock);
1160 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1161 TASK_UNINTERRUPTIBLE))
1162 goto out_put_layout_hdr;
1163 spin_lock(&ino->i_lock);
1164 }
24956804 1165 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1166 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1167
1168 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1169 struct pnfs_layout_range range = {
1170 .iomode = IOMODE_ANY,
1171 .offset = 0,
1172 .length = NFS4_MAX_UINT64,
1173 };
1174 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1175 }
1176
293b3b06 1177 /* Don't send a LAYOUTRETURN if list was initially empty */
24408f52 1178 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
293b3b06 1179 spin_unlock(&ino->i_lock);
293b3b06 1180 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1181 goto out_put_layout_hdr;
293b3b06 1182 }
47abadef 1183
e5fd1904 1184 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603 1185 spin_unlock(&ino->i_lock);
7f27392c 1186 if (send)
ed429d6b 1187 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c 1188out_put_layout_hdr:
ee6625a9 1189 pnfs_free_lseg_list(&tmp_list);
7f27392c 1190 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1191out:
1192 dprintk("<-- %s status: %d\n", __func__, status);
1193 return status;
1194}
1195
24028672
TM		 1196int
1197pnfs_commit_and_return_layout(struct inode *inode)
1198{
1199 struct pnfs_layout_hdr *lo;
1200 int ret;
1201
1202 spin_lock(&inode->i_lock);
1203 lo = NFS_I(inode)->layout;
1204 if (lo == NULL) {
1205 spin_unlock(&inode->i_lock);
1206 return 0;
1207 }
1208 pnfs_get_layout_hdr(lo);
1209 /* Block new layoutgets and read/write to ds */
1210 lo->plh_block_lgets++;
1211 spin_unlock(&inode->i_lock);
1212 filemap_fdatawait(inode->i_mapping);
1213 ret = pnfs_layoutcommit_inode(inode, true);
1214 if (ret == 0)
1215 ret = _pnfs_return_layout(inode);
1216 spin_lock(&inode->i_lock);
1217 lo->plh_block_lgets--;
1218 spin_unlock(&inode->i_lock);
1219 pnfs_put_layout_hdr(lo);
1220 return ret;
1221}
1222
1c5bd76d
TM		 1223bool pnfs_roc(struct inode *ino,
1224 struct nfs4_layoutreturn_args *args,
1225 struct nfs4_layoutreturn_res *res,
1226 const struct rpc_cred *cred)
f7e8917a 1227{
40dd4b7a
TM		 1228 struct nfs_inode *nfsi = NFS_I(ino);
1229 struct nfs_open_context *ctx;
1230 struct nfs4_state *state;
f7e8917a 1231 struct pnfs_layout_hdr *lo;
1c5bd76d 1232 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1233 nfs4_stateid stateid;
1c5bd76d
TM		 1234 enum pnfs_iomode iomode = 0;
1235 bool layoutreturn = false, roc = false;
e71708d4 1236 bool skip_read = false;
f7e8917a 1237
1c5bd76d
TM		 1238 if (!nfs_have_layout(ino))
1239 return false;
29ade5db 1240retry:
f7e8917a 1241 spin_lock(&ino->i_lock);
40dd4b7a 1242 lo = nfsi->layout;
0cdc329e
TM		 1243 if (!lo || !pnfs_layout_is_valid(lo) ||
1244 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
40dd4b7a 1245 goto out_noroc;
29ade5db
TM		 1246 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1247 pnfs_get_layout_hdr(lo);
1248 spin_unlock(&ino->i_lock);
1249 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1250 TASK_UNINTERRUPTIBLE);
1251 pnfs_put_layout_hdr(lo);
1252 goto retry;
1253 }
40dd4b7a 1254
e755d638 1255 /* no roc if we hold a delegation */
e71708d4
TM		 1256 if (nfs4_check_delegation(ino, FMODE_READ)) {
1257 if (nfs4_check_delegation(ino, FMODE_WRITE))
1258 goto out_noroc;
1259 skip_read = true;
1260 }
40dd4b7a
TM		 1261
1262 list_for_each_entry(ctx, &nfsi->open_files, list) {
1263 state = ctx->state;
e71708d4
TM		 1264 if (state == NULL)
1265 continue;
40dd4b7a 1266 /* Don't return layout if there is open file state */
e71708d4 1267 if (state->state & FMODE_WRITE)
40dd4b7a 1268 goto out_noroc;
e71708d4
TM		 1269 if (state->state & FMODE_READ)
1270 skip_read = true;
40dd4b7a
TM		 1271 }
1272
e755d638 1273
1c5bd76d 1274 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1275 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1276 continue;
e755d638 1277 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1278 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1279 continue;
1280 /*
1281 * Note: mark lseg for return so pnfs_layout_remove_lseg
1282 * doesn't invalidate the layout for us.
1283 */
1284 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1285 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1286 continue;
1287 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1288 }
1289
1c5bd76d
TM		 1290 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1291 goto out_noroc;
69820d22 1292
500d701f 1293 /* ROC in two conditions:
e755d638
PT		 1294 * 1. there are ROC lsegs
1295 * 2. we don't send layoutreturn
e755d638 1296 */
1c5bd76d
TM		 1297 /* lo ref dropped in pnfs_roc_release() */
1298 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1299 /* If the creds don't match, we can't compound the layoutreturn */
1300 if (!layoutreturn || cred != lo->plh_lc_cred)
1301 goto out_noroc;
1302
1303 roc = layoutreturn;
1304 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1305 res->lrs_present = 0;
1306 layoutreturn = false;
f7e8917a 1307
40dd4b7a 1308out_noroc:
f7e8917a 1309 spin_unlock(&ino->i_lock);
e755d638 1310 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1311 if (roc) {
1312 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1313 if (ld->prepare_layoutreturn)
1314 ld->prepare_layoutreturn(args);
1315 return true;
1316 }
e755d638 1317 if (layoutreturn)
1c5bd76d 1318 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
287bd3e9 1319 return false;
f7e8917a
FI		 1320}
1321
1c5bd76d
TM		 1322void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1323 struct nfs4_layoutreturn_res *res,
1324 int ret)
f7e8917a 1325{
1c5bd76d
TM		 1326 struct pnfs_layout_hdr *lo = args->layout;
1327 const nfs4_stateid *arg_stateid = NULL;
1328 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1329 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1330
1c5bd76d
TM		 1331 if (ret == 0) {
1332 arg_stateid = &args->stateid;
1333 if (res->lrs_present)
1334 res_stateid = &res->stateid;
1335 }
1336 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1337 res_stateid);
287bd3e9
TM		 1338 if (ld_private && ld_private->ops && ld_private->ops->free)
1339 ld_private->ops->free(ld_private);
1c5bd76d
TM		 1340 pnfs_put_layout_hdr(lo);
1341 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1342}
1343
500d701f
PT		 1344bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1345{
1346 struct nfs_inode *nfsi = NFS_I(ino);
1347 struct pnfs_layout_hdr *lo;
1348 bool sleep = false;
1349
1350 /* we might not have grabbed lo reference. so need to check under
1351 * i_lock */
1352 spin_lock(&ino->i_lock);
1353 lo = nfsi->layout;
ee284e35
TM		 1354 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1355 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1356 sleep = true;
ee284e35 1357 }
500d701f 1358 spin_unlock(&ino->i_lock);
500d701f
PT		 1359 return sleep;
1360}
1361
b1f69b75
AA		 1362/*
1363 * Compare two layout segments for sorting into layout cache.
1364 * We want to preferentially return RW over RO layouts, so ensure those
1365 * are seen first.
1366 */
1367static s64
7dc0ac70 1368pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1369 const struct pnfs_layout_range *l2)
b1f69b75 1370{
fb3296eb
BH		 1371 s64 d;
1372
1373 /* high offset > low offset */
1374 d = l1->offset - l2->offset;
1375 if (d)
1376 return d;
1377
1378 /* short length > long length */
1379 d = l2->length - l1->length;
1380 if (d)
1381 return d;
1382
b1f69b75 1383 /* read > read/write */
fb3296eb 1384 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1385}
1386
03772d2f
TM		 1387static bool
1388pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1389 const struct pnfs_layout_range *l2)
1390{
1391 return pnfs_lseg_range_cmp(l1, l2) > 0;
1392}
1393
1394static bool
1395pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1396 struct pnfs_layout_segment *old)
1397{
1398 return false;
1399}
1400
1401void
1402pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1403 struct pnfs_layout_segment *lseg,
1404 bool (*is_after)(const struct pnfs_layout_range *,
1405 const struct pnfs_layout_range *),
1406 bool (*do_merge)(struct pnfs_layout_segment *,
1407 struct pnfs_layout_segment *),
1408 struct list_head *free_me)
974cec8c 1409{
03772d2f 1410 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1411
974cec8c
AA		 1412 dprintk("%s:Begin\n", __func__);
1413
03772d2f
TM		 1414 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1415 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1416 continue;
1417 if (do_merge(lseg, lp)) {
1418 mark_lseg_invalid(lp, free_me);
1419 continue;
1420 }
1421 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1422 continue;
566052c5 1423 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1424 dprintk("%s: inserted lseg %p "
1425 "iomode %d offset %llu length %llu before "
1426 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1427 __func__, lseg, lseg->pls_range.iomode,
1428 lseg->pls_range.offset, lseg->pls_range.length,
1429 lp, lp->pls_range.iomode, lp->pls_range.offset,
1430 lp->pls_range.length);
fb3296eb 1431 goto out;
974cec8c 1432 }
fb3296eb
BH		 1433 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1434 dprintk("%s: inserted lseg %p "
1435 "iomode %d offset %llu length %llu at tail\n",
1436 __func__, lseg, lseg->pls_range.iomode,
1437 lseg->pls_range.offset, lseg->pls_range.length);
1438out:
70c3bd2b 1439 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1440
1441 dprintk("%s:Return\n", __func__);
e5e94017 1442}
03772d2f
TM		 1443EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1444
1445static void
1446pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1447 struct pnfs_layout_segment *lseg,
1448 struct list_head *free_me)
1449{
1450 struct inode *inode = lo->plh_inode;
1451 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1452
1453 if (ld->add_lseg != NULL)
1454 ld->add_lseg(lo, lseg, free_me);
1455 else
1456 pnfs_generic_layout_insert_lseg(lo, lseg,
1457 pnfs_lseg_range_is_after,
1458 pnfs_lseg_no_merge,
1459 free_me);
1460}
e5e94017
BH		 1461
1462static struct pnfs_layout_hdr *
9fa40758
PT		 1463alloc_init_layout_hdr(struct inode *ino,
1464 struct nfs_open_context *ctx,
1465 gfp_t gfp_flags)
e5e94017
BH		 1466{
1467 struct pnfs_layout_hdr *lo;
1468
636fb9c8 1469 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1470 if (!lo)
1471 return NULL;
2b28a7be 1472 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1473 INIT_LIST_HEAD(&lo->plh_layouts);
1474 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1475 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1476 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1477 lo->plh_inode = ino;
5cc2216d 1478 lo->plh_lc_cred = get_rpccred(ctx->cred);
67a3b721 1479 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1480 return lo;
1481}
1482
1483static struct pnfs_layout_hdr *
9fa40758
PT		 1484pnfs_find_alloc_layout(struct inode *ino,
1485 struct nfs_open_context *ctx,
1486 gfp_t gfp_flags)
e5241e43
TM		 1487 __releases(&ino->i_lock)
1488 __acquires(&ino->i_lock)
e5e94017
BH		 1489{
1490 struct nfs_inode *nfsi = NFS_I(ino);
1491 struct pnfs_layout_hdr *new = NULL;
1492
1493 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1494
251ec410
TM		 1495 if (nfsi->layout != NULL)
1496 goto out_existing;
e5e94017 1497 spin_unlock(&ino->i_lock);
9fa40758 1498 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1499 spin_lock(&ino->i_lock);
1500
251ec410 1501 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1502 nfsi->layout = new;
251ec410 1503 return new;
7175fe90
YN		 1504 } else if (new != NULL)
1505 pnfs_free_layout_hdr(new);
251ec410
TM		 1506out_existing:
1507 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1508 return nfsi->layout;
1509}
1510
b1f69b75
AA		 1511/*
1512 * iomode matching rules:
c7d73af2
TH		 1513 * iomode lseg strict match
1514 * iomode
1515 * ----- ----- ------ -----
1516 * ANY READ N/A true
1517 * ANY RW N/A true
1518 * RW READ N/A false
1519 * RW RW N/A true
1520 * READ READ N/A true
1521 * READ RW true false
1522 * READ RW false true
b1f69b75 1523 */
3cb2df17 1524static bool
7dc0ac70 1525pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1526 const struct pnfs_layout_range *range,
1527 bool strict_iomode)
b1f69b75 1528{
fb3296eb
BH		 1529 struct pnfs_layout_range range1;
1530
1531 if ((range->iomode == IOMODE_RW &&
1532 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1533 (range->iomode != ls_range->iomode &&
6089dd0d 1534 strict_iomode) ||
7dc0ac70 1535 !pnfs_lseg_range_intersecting(ls_range, range))
fb3296eb
BH		 1536 return 0;
1537
1538 /* range1 covers only the first byte in the range */
1539 range1 = *range;
1540 range1.length = 1;
7dc0ac70 1541 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1542}
1543
1544/*
1545 * lookup range in layout
1546 */
e5e94017 1547static struct pnfs_layout_segment *
fb3296eb 1548pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1549 struct pnfs_layout_range *range,
1550 bool strict_iomode)
e5e94017 1551{
b1f69b75
AA		 1552 struct pnfs_layout_segment *lseg, *ret = NULL;
1553
1554 dprintk("%s:Begin\n", __func__);
1555
b7edfaa1 1556 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1557 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
ce6ab4f2 1558 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
c7d73af2
TH		 1559 pnfs_lseg_range_match(&lseg->pls_range, range,
1560 strict_iomode)) {
9369a431 1561 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1562 break;
1563 }
b1f69b75
AA		 1564 }
1565
1566 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1567 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1568 return ret;
e5e94017
BH		 1569}
1570
d23d61c8
AA		 1571/*
1572 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1573 * to the MDS or over pNFS
1574 *
1575 * The nfs_inode read_io and write_io fields are cumulative counters reset
1576 * when there are no layout segments. Note that in pnfs_update_layout iomode
1577 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1578 * WRITE request.
1579 *
1580 * A return of true means use MDS I/O.
1581 *
1582 * From rfc 5661:
1583 * If a file's size is smaller than the file size threshold, data accesses
1584 * SHOULD be sent to the metadata server. If an I/O request has a length that
1585 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1586 * server. If both file size and I/O size are provided, the client SHOULD
1587 * reach or exceed both thresholds before sending its read or write
1588 * requests to the data server.
1589 */
1590static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1591 struct inode *ino, int iomode)
1592{
1593 struct nfs4_threshold *t = ctx->mdsthreshold;
1594 struct nfs_inode *nfsi = NFS_I(ino);
1595 loff_t fsize = i_size_read(ino);
1596 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1597
1598 if (t == NULL)
1599 return ret;
1600
1601 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1602 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1603
1604 switch (iomode) {
1605 case IOMODE_READ:
1606 if (t->bm & THRESHOLD_RD) {
1607 dprintk("%s fsize %llu\n", __func__, fsize);
1608 size_set = true;
1609 if (fsize < t->rd_sz)
1610 size = true;
1611 }
1612 if (t->bm & THRESHOLD_RD_IO) {
1613 dprintk("%s nfsi->read_io %llu\n", __func__,
1614 nfsi->read_io);
1615 io_set = true;
1616 if (nfsi->read_io < t->rd_io_sz)
1617 io = true;
1618 }
1619 break;
1620 case IOMODE_RW:
1621 if (t->bm & THRESHOLD_WR) {
1622 dprintk("%s fsize %llu\n", __func__, fsize);
1623 size_set = true;
1624 if (fsize < t->wr_sz)
1625 size = true;
1626 }
1627 if (t->bm & THRESHOLD_WR_IO) {
1628 dprintk("%s nfsi->write_io %llu\n", __func__,
1629 nfsi->write_io);
1630 io_set = true;
1631 if (nfsi->write_io < t->wr_io_sz)
1632 io = true;
1633 }
1634 break;
1635 }
1636 if (size_set && io_set) {
1637 if (size && io)
1638 ret = true;
1639 } else if (size || io)
1640 ret = true;
1641
1642 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1643 return ret;
1644}
1645
aa8a45ee
PT		 1646static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1647{
1648 /*
1649 * send layoutcommit as it can hold up layoutreturn due to lseg
1650 * reference
1651 */
1652 pnfs_layoutcommit_inode(lo->plh_inode, false);
1653 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1654 nfs_wait_bit_killable,
aa8a45ee
PT		 1655 TASK_UNINTERRUPTIBLE);
1656}
1657
d67ae825
TH		 1658static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1659{
1660 unsigned long *bitlock = &lo->plh_flags;
1661
1662 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1663 smp_mb__after_atomic();
1664 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1665}
1666
e5e94017
BH		 1667/*
1668 * Layout segment is retreived from the server if not cached.
1669 * The appropriate layout segment is referenced and returned to the caller.
1670 */
7c24d948 1671struct pnfs_layout_segment *
e5e94017
BH		 1672pnfs_update_layout(struct inode *ino,
1673 struct nfs_open_context *ctx,
fb3296eb
BH		 1674 loff_t pos,
1675 u64 count,
a75b9df9 1676 enum pnfs_iomode iomode,
c7d73af2 1677 bool strict_iomode,
a75b9df9 1678 gfp_t gfp_flags)
e5e94017 1679{
fb3296eb
BH		 1680 struct pnfs_layout_range arg = {
1681 .iomode = iomode,
1682 .offset = pos,
1683 .length = count,
1684 };
70d2f7b1 1685 unsigned pg_offset;
6382a441
WAA		 1686 struct nfs_server *server = NFS_SERVER(ino);
1687 struct nfs_client *clp = server->nfs_client;
183d9e7b 1688 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1689 struct pnfs_layout_segment *lseg = NULL;
183d9e7b
JL		 1690 nfs4_stateid stateid;
1691 long timeout = 0;
66b53f32 1692 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1693 bool first;
e5e94017 1694
9a4bf31d 1695 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1696 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1697 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1698 goto out;
9a4bf31d 1699 }
d23d61c8 1700
9a4bf31d 1701 if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
183d9e7b 1702 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1703 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
4ae93560 1704 goto out;
9a4bf31d 1705 }
4ae93560 1706
9a4bf31d 1707 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1708 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1709 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1710 goto out;
9a4bf31d 1711 }
d23d61c8 1712
9bf87482 1713lookup_again:
b88fa69e 1714 nfs4_client_recover_expired_lease(clp);
9bf87482 1715 first = false;
e5e94017 1716 spin_lock(&ino->i_lock);
9fa40758 1717 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1718 if (lo == NULL) {
1719 spin_unlock(&ino->i_lock);
183d9e7b 1720 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1721 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1722 goto out;
1723 }
e5e94017 1724
43f1b3da 1725 /* Do we even need to bother with this? */
a59c30ac 1726 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1727 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1728 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1729 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1730 goto out_unlock;
1731 }
1732
1733 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1734 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1735 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1736 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1737 goto out_unlock;
9a4bf31d 1738 }
e5e94017 1739
c7d73af2 1740 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1741 if (lseg) {
1742 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1743 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1744 goto out_unlock;
1745 }
1746
1747 if (!nfs4_valid_open_stateid(ctx->state)) {
1748 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1749 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1750 goto out_unlock;
1751 }
1752
1753 /*
1754 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1755 * stateid, or it has been invalidated, then we must use the open
1756 * stateid.
1757 */
67a3b721 1758 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
183d9e7b
JL		 1759
1760 /*
1761 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1762 * RFC 5661 Errata 3208.
1763 */
1764 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1765 &lo->plh_flags)) {
1766 spin_unlock(&ino->i_lock);
1767 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1768 TASK_UNINTERRUPTIBLE);
1769 pnfs_put_layout_hdr(lo);
183d9e7b 1770 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1771 goto lookup_again;
1772 }
183d9e7b
JL		 1773
1774 first = true;
70d2f7b1
TM		 1775 if (nfs4_select_rw_stateid(ctx->state,
1776 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1777 NULL, &stateid, NULL) != 0) {
1778 trace_pnfs_update_layout(ino, pos, count,
1779 iomode, lo, lseg,
1780 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1781 goto out_unlock;
1782 }
9bf87482 1783 } else {
183d9e7b 1784 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 1785 }
568e8c49 1786
aa8a45ee
PT		 1787 /*
1788 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1789 * for LAYOUTRETURN even if first is true.
1790 */
8f70f53a 1791 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 1792 spin_unlock(&ino->i_lock);
1793 dprintk("%s wait for layoutreturn\n", __func__);
1794 if (pnfs_prepare_to_retry_layoutget(lo)) {
d67ae825
TH		 1795 if (first)
1796 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 1797 pnfs_put_layout_hdr(lo);
1798 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 1799 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1800 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 1801 goto lookup_again;
1802 }
183d9e7b 1803 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1804 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 1805 goto out_put_layout_hdr;
1806 }
1807
9a4bf31d 1808 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 1809 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1810 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 1811 goto out_unlock;
9a4bf31d 1812 }
cf7d63f1 1813 atomic_inc(&lo->plh_outstanding);
f49f9baa 1814 spin_unlock(&ino->i_lock);
30005121 1815
abb9a007 1816 if (list_empty(&lo->plh_layouts)) {
2130ff66
FI		 1817 /* The lo must be on the clp list if there is any
1818 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1819 */
1820 spin_lock(&clp->cl_lock);
abb9a007
PT		 1821 if (list_empty(&lo->plh_layouts))
1822 list_add_tail(&lo->plh_layouts, &server->layouts);
2130ff66
FI		 1823 spin_unlock(&clp->cl_lock);
1824 }
e5e94017 1825
09cbfeaf 1826 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 1827 if (pg_offset) {
1828 arg.offset -= pg_offset;
1829 arg.length += pg_offset;
1830 }
7c24d948 1831 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 1832 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 1833
183d9e7b
JL		 1834 lseg = send_layoutget(lo, ctx, &stateid, &arg, &timeout, gfp_flags);
1835 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1836 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
56b38a1f 1837 atomic_dec(&lo->plh_outstanding);
83026d80 1838 if (IS_ERR(lseg)) {
183d9e7b 1839 switch(PTR_ERR(lseg)) {
e85d7ee4 1840 case -EBUSY:
183d9e7b
JL		 1841 if (time_after(jiffies, giveup))
1842 lseg = NULL;
66b53f32
TM		 1843 break;
1844 case -ERECALLCONFLICT:
1845 /* Huh? We hold no layouts, how is there a recall? */
1846 if (first) {
1847 lseg = NULL;
1848 break;
1849 }
1850 /* Destroy the existing layout and start over */
1851 if (time_after(jiffies, giveup))
1852 pnfs_destroy_layout(NFS_I(ino));
183d9e7b
JL		 1853 /* Fallthrough */
1854 case -EAGAIN:
56b38a1f 1855 break;
183d9e7b
JL		 1856 default:
1857 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
1858 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1859 lseg = NULL;
1860 }
56b38a1f
TM		 1861 goto out_put_layout_hdr;
1862 }
1863 if (lseg) {
1864 if (first)
1865 pnfs_clear_first_layoutget(lo);
1866 trace_pnfs_update_layout(ino, pos, count,
1867 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
1868 pnfs_put_layout_hdr(lo);
1869 goto lookup_again;
83026d80
JL		 1870 }
1871 } else {
1872 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1873 }
1874
830ffb56 1875out_put_layout_hdr:
d67ae825
TH		 1876 if (first)
1877 pnfs_clear_first_layoutget(lo);
70c3bd2b 1878 pnfs_put_layout_hdr(lo);
e5e94017 1879out:
f86bbcf8
TM		 1880 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1881 "(%s, offset: %llu, length: %llu)\n",
1882 __func__, ino->i_sb->s_id,
1883 (unsigned long long)NFS_FILEID(ino),
d600ad1f 1884 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 1885 iomode==IOMODE_RW ? "read/write" : "read-only",
1886 (unsigned long long)pos,
1887 (unsigned long long)count);
e5e94017
BH		 1888 return lseg;
1889out_unlock:
1890 spin_unlock(&ino->i_lock);
830ffb56 1891 goto out_put_layout_hdr;
e5e94017 1892}
7c24d948 1893EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 1894
540d9864
TM		 1895static bool
1896pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
1897{
1898 switch (range->iomode) {
1899 case IOMODE_READ:
1900 case IOMODE_RW:
1901 break;
1902 default:
1903 return false;
1904 }
1905 if (range->offset == NFS4_MAX_UINT64)
1906 return false;
1907 if (range->length == 0)
1908 return false;
1909 if (range->length != NFS4_MAX_UINT64 &&
1910 range->length > NFS4_MAX_UINT64 - range->offset)
1911 return false;
1912 return true;
1913}
1914
a0b0a6e3 1915struct pnfs_layout_segment *
b1f69b75
AA		 1916pnfs_layout_process(struct nfs4_layoutget *lgp)
1917{
1918 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1919 struct nfs4_layoutget_res *res = &lgp->res;
1920 struct pnfs_layout_segment *lseg;
b7edfaa1 1921 struct inode *ino = lo->plh_inode;
78096cca 1922 LIST_HEAD(free_me);
540d9864
TM		 1923
1924 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 1925 return ERR_PTR(-EINVAL);
b1f69b75
AA		 1926
1927 /* Inject layout blob into I/O device driver */
a75b9df9 1928 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 1929 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 1930 if (!lseg)
1b3c6d07
JL		 1931 lseg = ERR_PTR(-ENOMEM);
1932
1933 dprintk("%s: Could not allocate layout: error %ld\n",
1934 __func__, PTR_ERR(lseg));
1935 return lseg;
b1f69b75
AA		 1936 }
1937
119cef97 1938 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 1939
b1f69b75 1940 spin_lock(&ino->i_lock);
e1c06f80 1941 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 1942 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 1943 goto out_forget;
43f1b3da 1944 }
038d6493 1945
9888d837
TM		 1946 if (!pnfs_layout_is_valid(lo)) {
1947 /* We have a completely new layout */
1948 pnfs_set_layout_stateid(lo, &res->stateid, true);
1949 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 1950 /* existing state ID, make sure the sequence number matches. */
1951 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
1952 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 1953 goto out_forget;
362f7474
CH		 1954 }
1955 pnfs_set_layout_stateid(lo, &res->stateid, false);
1956 } else {
1957 /*
1958 * We got an entirely new state ID. Mark all segments for the
9888d837 1959 * inode invalid, and retry the layoutget
362f7474 1960 */
d9b61708 1961 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 1962 goto out_forget;
362f7474 1963 }
038d6493 1964
9369a431 1965 pnfs_get_lseg(lseg);
03772d2f 1966 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 1967
b1f69b75 1968
3976143b 1969 if (res->return_on_close)
f7e8917a 1970 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 1971
b1f69b75 1972 spin_unlock(&ino->i_lock);
78096cca 1973 pnfs_free_lseg_list(&free_me);
a0b0a6e3 1974 return lseg;
43f1b3da 1975
1b3c6d07 1976out_forget:
43f1b3da
FI		 1977 spin_unlock(&ino->i_lock);
1978 lseg->pls_layout = lo;
1979 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1f18b82c
TM		 1980 if (!pnfs_layout_is_valid(lo))
1981 nfs_commit_inode(ino, 0);
1b3c6d07 1982 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 1983}
1984
2f215968
TM		 1985/**
1986 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
1987 * @lo: pointer to layout header
1988 * @tmp_list: list header to be used with pnfs_free_lseg_list()
1989 * @return_range: describe layout segment ranges to be returned
1990 *
1991 * This function is mainly intended for use by layoutrecall. It attempts
1992 * to free the layout segment immediately, or else to mark it for return
1993 * as soon as its reference count drops to zero.
1994 */
10335556 1995int
016256df
PT		 1996pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
1997 struct list_head *tmp_list,
6d597e17
JL		 1998 const struct pnfs_layout_range *return_range,
1999 u32 seq)
016256df
PT		 2000{
2001 struct pnfs_layout_segment *lseg, *next;
10335556 2002 int remaining = 0;
016256df
PT		 2003
2004 dprintk("%s:Begin lo %p\n", __func__, lo);
2005
2006 if (list_empty(&lo->plh_segs))
10335556 2007 return 0;
016256df 2008
fc7ff367 2009 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2010
2011 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2012 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2013 dprintk("%s: marking lseg %p iomode %d "
2014 "offset %llu length %llu\n", __func__,
2015 lseg, lseg->pls_range.iomode,
2016 lseg->pls_range.offset,
2017 lseg->pls_range.length);
2f215968
TM		 2018 if (mark_lseg_invalid(lseg, tmp_list))
2019 continue;
2020 remaining++;
016256df 2021 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2022 }
6d597e17
JL		 2023
2024 if (remaining)
2025 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2026
10335556 2027 return remaining;
016256df
PT		 2028}
2029
2030void pnfs_error_mark_layout_for_return(struct inode *inode,
2031 struct pnfs_layout_segment *lseg)
2032{
2033 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2034 struct pnfs_layout_range range = {
2035 .iomode = lseg->pls_range.iomode,
2036 .offset = 0,
2037 .length = NFS4_MAX_UINT64,
2038 };
10335556 2039 bool return_now = false;
016256df
PT		 2040
2041 spin_lock(&inode->i_lock);
bdebfccd
TM		 2042 if (!pnfs_layout_is_valid(lo)) {
2043 spin_unlock(&inode->i_lock);
2044 return;
2045 }
2d6cf5ab 2046 pnfs_set_plh_return_info(lo, range.iomode, 0);
016256df
PT		 2047 /*
2048 * mark all matching lsegs so that we are sure to have no live
2049 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2050 * for how it works.
2051 */
68f74479 2052 if (!pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0)) {
10335556 2053 nfs4_stateid stateid;
e5fd1904 2054 enum pnfs_iomode iomode;
10335556 2055
e5fd1904 2056 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2057 spin_unlock(&inode->i_lock);
2058 if (return_now)
2059 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2060 } else {
2061 spin_unlock(&inode->i_lock);
2062 nfs_commit_inode(inode, 0);
2063 }
016256df
PT		 2064}
2065EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2066
b3230e80
TM		 2067void
2068pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2069{
2070 if (pgio->pg_lseg == NULL ||
2071 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2072 return;
2073 pnfs_put_lseg(pgio->pg_lseg);
2074 pgio->pg_lseg = NULL;
2075}
2076EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2077
08cb5b0f
BC		 2078/*
2079 * Check for any intersection between the request and the pgio->pg_lseg,
2080 * and if none, put this pgio->pg_lseg away.
2081 */
2082static void
2083pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2084{
2085 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2086 pnfs_put_lseg(pgio->pg_lseg);
2087 pgio->pg_lseg = NULL;
2088 }
2089}
2090
d8007d4d
TM		 2091void
2092pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2093{
1fd937bd
PT		 2094 u64 rd_size = req->wb_bytes;
2095
b3230e80 2096 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2097 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2098 if (pgio->pg_lseg == NULL) {
2099 if (pgio->pg_dreq == NULL)
2100 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2101 else
2102 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2103
2104 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2105 req->wb_context,
2106 req_offset(req),
2107 rd_size,
2108 IOMODE_READ,
c7d73af2 2109 false,
cb5d04bc 2110 GFP_KERNEL);
d600ad1f
PT		 2111 if (IS_ERR(pgio->pg_lseg)) {
2112 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2113 pgio->pg_lseg = NULL;
2114 return;
2115 }
cb5d04bc 2116 }
e885de1a
TM		 2117 /* If no lseg, fall back to read through mds */
2118 if (pgio->pg_lseg == NULL)
1f945357 2119 nfs_pageio_reset_read_mds(pgio);
e885de1a 2120
d8007d4d
TM		 2121}
2122EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2123
2124void
6296556f
PT		 2125pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2126 struct nfs_page *req, u64 wb_size)
d8007d4d 2127{
b3230e80 2128 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2129 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2130 if (pgio->pg_lseg == NULL) {
cb5d04bc
PT		 2131 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2132 req->wb_context,
2133 req_offset(req),
2134 wb_size,
2135 IOMODE_RW,
c7d73af2 2136 false,
cb5d04bc 2137 GFP_NOFS);
d600ad1f
PT		 2138 if (IS_ERR(pgio->pg_lseg)) {
2139 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2140 pgio->pg_lseg = NULL;
2141 return;
2142 }
2143 }
e885de1a
TM		 2144 /* If no lseg, fall back to write through mds */
2145 if (pgio->pg_lseg == NULL)
1f945357 2146 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2147}
2148EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2149
180bb5ec
WAA		 2150void
2151pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2152{
2153 if (desc->pg_lseg) {
2154 pnfs_put_lseg(desc->pg_lseg);
2155 desc->pg_lseg = NULL;
2156 }
2157}
2158EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2159
b4fdac1a
WAA		 2160/*
2161 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2162 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2163 */
2164size_t
a7d42ddb
WAA		 2165pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2166 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2167{
0f9c429e 2168 unsigned int size;
c5e20cb7 2169 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2170
2171 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2172 if (!size)
2173 return 0;
94ad1c80 2174
19982ba8 2175 /*
c5e20cb7
WAA		 2176 * 'size' contains the number of bytes left in the current page (up
2177 * to the original size asked for in @req->wb_bytes).
2178 *
2179 * Calculate how many bytes are left in the layout segment
2180 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2181 *
2182 * Please also note that 'end_offset' is actually the offset of the
2183 * first byte that lies outside the pnfs_layout_range. FIXME?
2184 *
2185 */
19b54848 2186 if (pgio->pg_lseg) {
17822b20 2187 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2188 pgio->pg_lseg->pls_range.length);
2189 req_start = req_offset(req);
08cb5b0f 2190
c5e20cb7 2191 /* start of request is past the last byte of this segment */
08cb5b0f 2192 if (req_start >= seg_end)
19b54848 2193 return 0;
c5e20cb7
WAA		 2194
2195 /* adjust 'size' iff there are fewer bytes left in the
2196 * segment than what nfs_generic_pg_test returned */
2197 seg_left = seg_end - req_start;
2198 if (seg_left < size)
2199 size = (unsigned int)seg_left;
19b54848 2200 }
0f9c429e 2201
19b54848 2202 return size;
94ad1c80 2203}
89a58e32 2204EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2205
53113ad3 2206int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2207{
2208 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2209
2210 /* Resend all requests through the MDS */
53113ad3
WAA		 2211 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2212 hdr->completion_ops);
c7070113 2213 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2214 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2215}
e7dd79af 2216EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2217
d45f60c6 2218static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2219{
cd841605
FI		 2220
2221 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2222 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2223 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2224 pnfs_return_layout(hdr->inode);
1acbbb4e 2225 }
6c75dc0d 2226 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2227 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2228}
2229
d20581aa
BH		 2230/*
2231 * Called by non rpc-based layout drivers
2232 */
d45f60c6 2233void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2234{
f8417b48 2235 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2236 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2237 hdr->mds_offset + hdr->res.count);
d45f60c6 2238 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2239 }
2240 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2241 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2242 pnfs_ld_handle_write_error(hdr);
2243 hdr->mds_ops->rpc_release(hdr);
44b83799 2244}
d20581aa 2245EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2246
dce81290
TM		 2247static void
2248pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2249 struct nfs_pgio_header *hdr)
dce81290 2250{
48d635f1 2251 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2252
6c75dc0d 2253 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2254 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2255 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2256 mirror->pg_recoalesce = 1;
6c75dc0d 2257 }
1ca018d2 2258 hdr->release(hdr);
dce81290
TM		 2259}
2260
2261static enum pnfs_try_status
d45f60c6 2262pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2263 const struct rpc_call_ops *call_ops,
2264 struct pnfs_layout_segment *lseg,
2265 int how)
0382b744 2266{
cd841605 2267 struct inode *inode = hdr->inode;
0382b744
AA		 2268 enum pnfs_try_status trypnfs;
2269 struct nfs_server *nfss = NFS_SERVER(inode);
2270
cd841605 2271 hdr->mds_ops = call_ops;
0382b744
AA		 2272
2273 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2274 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2275 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2276 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2277 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2278 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2279 return trypnfs;
2280}
2281
dce81290 2282static void
7f714720
WAA		 2283pnfs_do_write(struct nfs_pageio_descriptor *desc,
2284 struct nfs_pgio_header *hdr, int how)
dce81290 2285{
dce81290
TM		 2286 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2287 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2288 enum pnfs_try_status trypnfs;
dce81290 2289
d45f60c6 2290 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2291 switch (trypnfs) {
2292 case PNFS_NOT_ATTEMPTED:
d45f60c6 2293 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2294 case PNFS_ATTEMPTED:
2295 break;
2296 case PNFS_TRY_AGAIN:
2297 /* cleanup hdr and prepare to redo pnfs */
2298 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2299 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2300 list_splice_init(&hdr->pages, &mirror->pg_list);
2301 mirror->pg_recoalesce = 1;
2302 }
2303 hdr->mds_ops->rpc_release(hdr);
2304 }
dce81290
TM		 2305}
2306
6c75dc0d
FI		 2307static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2308{
9369a431 2309 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2310 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2311}
2312
dce81290
TM		 2313int
2314pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2315{
6c75dc0d 2316 struct nfs_pgio_header *hdr;
dce81290
TM		 2317 int ret;
2318
1e7f3a48
WAA		 2319 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2320 if (!hdr) {
2bff2288
PT		 2321 desc->pg_error = -ENOMEM;
2322 return desc->pg_error;
dce81290 2323 }
6c75dc0d 2324 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2325
9369a431 2326 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2327 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2328 if (!ret)
7f714720 2329 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2330
6c75dc0d 2331 return ret;
dce81290
TM		 2332}
2333EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2334
53113ad3 2335int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2336{
2337 struct nfs_pageio_descriptor pgio;
2338
1acbbb4e 2339 /* Resend all requests through the MDS */
53113ad3
WAA		 2340 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2341 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2342}
e7dd79af 2343EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2344
d45f60c6 2345static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2346{
cd841605
FI		 2347 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2348 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2349 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2350 pnfs_return_layout(hdr->inode);
1acbbb4e 2351 }
4db6e0b7 2352 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2353 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2354}
2355
d20581aa
BH		 2356/*
2357 * Called by non rpc-based layout drivers
2358 */
d45f60c6 2359void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2360{
bfc505de 2361 if (likely(!hdr->pnfs_error))
d45f60c6 2362 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2363 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2364 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2365 pnfs_ld_handle_read_error(hdr);
2366 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2367}
2368EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2369
493292dd
TM		 2370static void
2371pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2372 struct nfs_pgio_header *hdr)
493292dd 2373{
48d635f1 2374 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2375
4db6e0b7 2376 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2377 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2378 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2379 mirror->pg_recoalesce = 1;
4db6e0b7 2380 }
1ca018d2 2381 hdr->release(hdr);
493292dd
TM		 2382}
2383
64419a9b
AA		 2384/*
2385 * Call the appropriate parallel I/O subsystem read function.
2386 */
493292dd 2387static enum pnfs_try_status
d45f60c6 2388pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2389 const struct rpc_call_ops *call_ops,
2390 struct pnfs_layout_segment *lseg)
64419a9b 2391{
cd841605 2392 struct inode *inode = hdr->inode;
64419a9b
AA		 2393 struct nfs_server *nfss = NFS_SERVER(inode);
2394 enum pnfs_try_status trypnfs;
2395
cd841605 2396 hdr->mds_ops = call_ops;
64419a9b
AA		 2397
2398 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2399 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2400
d45f60c6 2401 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2402 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2403 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2404 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2405 return trypnfs;
2406}
863a3c6c 2407
ceb11e13 2408/* Resend all requests through pnfs. */
1b1bc66b 2409void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2410{
2411 struct nfs_pageio_descriptor pgio;
2412
1b1bc66b 2413 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2414 /* Prevent deadlocks with layoutreturn! */
2415 pnfs_put_lseg(hdr->lseg);
2416 hdr->lseg = NULL;
2417
1b1bc66b
WAA		 2418 nfs_pageio_init_read(&pgio, hdr->inode, false,
2419 hdr->completion_ops);
2420 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2421 }
ceb11e13
PT		 2422}
2423EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2424
493292dd 2425static void
7f714720 2426pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2427{
493292dd
TM		 2428 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2429 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2430 enum pnfs_try_status trypnfs;
493292dd 2431
d45f60c6 2432 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2433 switch (trypnfs) {
2434 case PNFS_NOT_ATTEMPTED:
d45f60c6 2435 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2436 case PNFS_ATTEMPTED:
2437 break;
2438 case PNFS_TRY_AGAIN:
2439 /* cleanup hdr and prepare to redo pnfs */
2440 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2441 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2442 list_splice_init(&hdr->pages, &mirror->pg_list);
2443 mirror->pg_recoalesce = 1;
2444 }
2445 hdr->mds_ops->rpc_release(hdr);
2446 }
493292dd
TM		 2447}
2448
4db6e0b7
FI		 2449static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2450{
9369a431 2451 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2452 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2453}
2454
493292dd
TM		 2455int
2456pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2457{
4db6e0b7 2458 struct nfs_pgio_header *hdr;
493292dd
TM		 2459 int ret;
2460
1e7f3a48
WAA		 2461 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2462 if (!hdr) {
2bff2288
PT		 2463 desc->pg_error = -ENOMEM;
2464 return desc->pg_error;
493292dd 2465 }
4db6e0b7 2466 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2467 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2468 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2469 if (!ret)
7f714720 2470 pnfs_do_read(desc, hdr);
4db6e0b7 2471 return ret;
493292dd
TM		 2472}
2473EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2474
71244d9b
TM		 2475static void pnfs_clear_layoutcommitting(struct inode *inode)
2476{
2477 unsigned long *bitlock = &NFS_I(inode)->flags;
2478
2479 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2480 smp_mb__after_atomic();
71244d9b
TM		 2481 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2482}
2483
863a3c6c 2484/*
a9bae566 2485 * There can be multiple RW segments.
863a3c6c 2486 */
a9bae566 2487static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2488{
a9bae566 2489 struct pnfs_layout_segment *lseg;
863a3c6c 2490
a9bae566
PT		 2491 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2492 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2493 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2494 list_add(&lseg->pls_lc_list, listp);
2495 }
863a3c6c
AA		 2496}
2497
a073dbff
TM		 2498static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2499{
2500 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2501
2502 /* Matched by references in pnfs_set_layoutcommit */
2503 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2504 list_del_init(&lseg->pls_lc_list);
2505 pnfs_put_lseg(lseg);
2506 }
2507
71244d9b 2508 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2509}
2510
1b0ae068
PT		 2511void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2512{
b9e028fd 2513 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2514}
2515EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2516
863a3c6c 2517void
67af7611
TM		 2518pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2519 loff_t end_pos)
863a3c6c 2520{
cd841605 2521 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2522 bool mark_as_dirty = false;
863a3c6c 2523
cd841605 2524 spin_lock(&inode->i_lock);
863a3c6c 2525 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2526 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2527 mark_as_dirty = true;
863a3c6c 2528 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2529 __func__, inode->i_ino);
29559b11
TM		 2530 } else if (end_pos > nfsi->layout->plh_lwb)
2531 nfsi->layout->plh_lwb = end_pos;
67af7611 2532 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2533 /* references matched in nfs4_layoutcommit_release */
67af7611 2534 pnfs_get_lseg(lseg);
a9bae566 2535 }
cd841605 2536 spin_unlock(&inode->i_lock);
acff5880 2537 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2538 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2539
2540 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2541 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2542 if (mark_as_dirty)
cd841605 2543 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2544}
2545EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2546
db29c089
AA		 2547void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2548{
2549 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2550
2551 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2552 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2553 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2554}
2555
de4b15c7
AA		 2556/*
2557 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2558 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2559 * data to disk to allow the server to recover the data if it crashes.
2560 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2561 * is off, and a COMMIT is sent to a data server, or
2562 * if WRITEs to a data server return NFS_DATA_SYNC.
2563 */
863a3c6c 2564int
ef311537 2565pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2566{
5f919c9f 2567 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2568 struct nfs4_layoutcommit_data *data;
2569 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2570 loff_t end_pos;
71244d9b 2571 int status;
863a3c6c 2572
71244d9b 2573 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2574 return 0;
2575
71244d9b 2576 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2577
71244d9b 2578 status = -EAGAIN;
92407e75 2579 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2580 if (!sync)
2581 goto out;
74316201 2582 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2583 NFS_INO_LAYOUTCOMMITTING,
2584 nfs_wait_bit_killable,
2585 TASK_KILLABLE);
92407e75 2586 if (status)
71244d9b 2587 goto out;
92407e75
PT		 2588 }
2589
71244d9b
TM		 2590 status = -ENOMEM;
2591 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2592 data = kzalloc(sizeof(*data), GFP_NOFS);
2593 if (!data)
2594 goto clear_layoutcommitting;
2595
2596 status = 0;
de4b15c7 2597 spin_lock(&inode->i_lock);
71244d9b
TM		 2598 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2599 goto out_unlock;
a9bae566 2600
71244d9b 2601 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 2602 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 2603
acff5880 2604 end_pos = nfsi->layout->plh_lwb;
863a3c6c 2605
f597c537 2606 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 2607 spin_unlock(&inode->i_lock);
2608
2609 data->args.inode = inode;
9fa40758 2610 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 2611 nfs_fattr_init(&data->fattr);
2612 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2613 data->res.fattr = &data->fattr;
2e18d4d8
TM		 2614 if (end_pos != 0)
2615 data->args.lastbytewritten = end_pos - 1;
2616 else
2617 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 2618 data->res.server = NFS_SERVER(inode);
2619
5f919c9f
CH		 2620 if (ld->prepare_layoutcommit) {
2621 status = ld->prepare_layoutcommit(&data->args);
2622 if (status) {
3471648a 2623 put_rpccred(data->cred);
5f919c9f 2624 spin_lock(&inode->i_lock);
29559b11
TM		 2625 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2626 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 2627 nfsi->layout->plh_lwb = end_pos;
3471648a 2628 goto out_unlock;
5f919c9f
CH		 2629 }
2630 }
2631
2632
863a3c6c
AA		 2633 status = nfs4_proc_layoutcommit(data, sync);
2634out:
92407e75
PT		 2635 if (status)
2636 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2637 dprintk("<-- %s status %d\n", __func__, status);
2638 return status;
71244d9b
TM		 2639out_unlock:
2640 spin_unlock(&inode->i_lock);
92407e75 2641 kfree(data);
71244d9b
TM		 2642clear_layoutcommitting:
2643 pnfs_clear_layoutcommitting(inode);
92407e75 2644 goto out;
863a3c6c 2645}
72cff449 2646EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 2647
5bb89b47
TM		 2648int
2649pnfs_generic_sync(struct inode *inode, bool datasync)
2650{
2651 return pnfs_layoutcommit_inode(inode, true);
2652}
2653EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2654
82be417a
AA		 2655struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2656{
2657 struct nfs4_threshold *thp;
2658
2659 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2660 if (!thp) {
2661 dprintk("%s mdsthreshold allocation failed\n", __func__);
2662 return NULL;
2663 }
2664 return thp;
2665}
8733408d 2666
865a7ecb 2667#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 2668int
c8ad8894 2669pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 2670{
2671 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2672 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 2673 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 2674 struct nfs42_layoutstat_data *data;
2675 struct pnfs_layout_hdr *hdr;
2676 int status = 0;
2677
2678 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2679 goto out;
2680
6c5a0d89
TM		 2681 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2682 goto out;
2683
1bfe3b25
PT		 2684 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2685 goto out;
2686
8733408d
PT		 2687 spin_lock(&inode->i_lock);
2688 if (!NFS_I(inode)->layout) {
2689 spin_unlock(&inode->i_lock);
f538d0ba 2690 goto out_clear_layoutstats;
8733408d
PT		 2691 }
2692 hdr = NFS_I(inode)->layout;
2693 pnfs_get_layout_hdr(hdr);
2694 spin_unlock(&inode->i_lock);
2695
c8ad8894 2696 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 2697 if (!data) {
2698 status = -ENOMEM;
2699 goto out_put;
2700 }
2701
2702 data->args.fh = NFS_FH(inode);
2703 data->args.inode = inode;
8733408d
PT		 2704 status = ld->prepare_layoutstats(&data->args);
2705 if (status)
2706 goto out_free;
2707
2708 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2709
2710out:
2711 dprintk("%s returns %d\n", __func__, status);
2712 return status;
2713
2714out_free:
2715 kfree(data);
2716out_put:
2717 pnfs_put_layout_hdr(hdr);
f538d0ba 2718out_clear_layoutstats:
1bfe3b25
PT		 2719 smp_mb__before_atomic();
2720 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2721 smp_mb__after_atomic();
8733408d
PT		 2722 goto out;
2723}
2724EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 2725#endif
bbf58bf3
TM		 2726
2727unsigned int layoutstats_timer;
2728module_param(layoutstats_timer, uint, 0644);
2729EXPORT_SYMBOL_GPL(layoutstats_timer);
Linux 6.x block layer and io_uring tree(s)