]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/nfs/pnfs.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
UBUNTU: Ubuntu-4.13.0-45.50
[mirror_ubuntu-artful-kernel.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{
cc6e5340 254 atomic_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
cc6e5340 299 if (atomic_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
2454dfea
TM		 357/*
358 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
359 *
360 * In order to continue using the pnfs_layout_hdr, a full recovery
361 * is required.
362 * Note that caller must hold inode->i_lock.
363 */
5f46be04 364int
2454dfea
TM		 365pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
366 struct list_head *lseg_list)
367{
368 struct pnfs_layout_range range = {
369 .iomode = IOMODE_ANY,
370 .offset = 0,
371 .length = NFS4_MAX_UINT64,
372 };
fe1cf946 373 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 374
375 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 376 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
377 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 378 pnfs_clear_layoutreturn_info(lo);
68f74479 379 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 380 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
381 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
382 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 383 return !list_empty(&lo->plh_segs);
2454dfea
TM		 384}
385
b9e028fd
TM		 386static int
387pnfs_iomode_to_fail_bit(u32 iomode)
388{
389 return iomode == IOMODE_RW ?
390 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
391}
392
393static void
3e621214 394pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 395{
25c75333 396 lo->plh_retry_timestamp = jiffies;
39e88fcf 397 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
3e621214
TM		 398 atomic_inc(&lo->plh_refcount);
399}
400
401static void
402pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
403{
404 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
405 atomic_dec(&lo->plh_refcount);
406}
407
408static void
409pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
410{
411 struct inode *inode = lo->plh_inode;
115ce575
TM		 412 struct pnfs_layout_range range = {
413 .iomode = iomode,
414 .offset = 0,
415 .length = NFS4_MAX_UINT64,
416 };
417 LIST_HEAD(head);
3e621214
TM		 418
419 spin_lock(&inode->i_lock);
420 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 421 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 422 spin_unlock(&inode->i_lock);
115ce575 423 pnfs_free_lseg_list(&head);
b9e028fd
TM		 424 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
425 iomode == IOMODE_RW ? "RW" : "READ");
426}
427
428static bool
429pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
430{
25c75333 431 unsigned long start, end;
3e621214
TM		 432 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
433
434 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 435 return false;
436 end = jiffies;
437 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
438 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
439 /* It is time to retry the failed layoutgets */
3e621214 440 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 441 return false;
442 }
443 return true;
b9e028fd
TM		 444}
445
974cec8c 446static void
119cef97
TM		 447pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
448 const struct pnfs_layout_range *range,
449 const nfs4_stateid *stateid)
974cec8c 450{
566052c5 451 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 452 INIT_LIST_HEAD(&lseg->pls_lc_list);
4541d16c 453 atomic_set(&lseg->pls_refcount, 1);
4541d16c 454 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 455 lseg->pls_layout = lo;
119cef97
TM		 456 lseg->pls_range = *range;
457 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 458}
459
905ca191 460static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 461{
68f74479
TM		 462 if (lseg != NULL) {
463 struct inode *inode = lseg->pls_layout->plh_inode;
464 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
465 }
974cec8c
AA		 466}
467
d684d2ae 468static void
57036a37
TM		 469pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
470 struct pnfs_layout_segment *lseg)
d684d2ae 471{
d20581aa 472 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 473 list_del_init(&lseg->pls_list);
8f0d27dc
TM		 474 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
475 atomic_dec(&lo->plh_refcount);
abb3e1c8
TM		 476 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
477 return;
7b650994
TM		 478 if (list_empty(&lo->plh_segs) &&
479 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
480 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 481 if (atomic_read(&lo->plh_outstanding) == 0)
482 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 483 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 484 }
d684d2ae
FI		 485}
486
68f74479
TM		 487static bool
488pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
489 struct pnfs_layout_segment *lseg)
490{
491 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
492 pnfs_layout_is_valid(lo)) {
4aab9732 493 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 494 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
495 return true;
496 }
497 return false;
498}
499
bae724ef 500void
9369a431 501pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 502{
57036a37 503 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 504 struct inode *inode;
505
506 if (!lseg)
507 return;
508
4541d16c
FI		 509 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
510 atomic_read(&lseg->pls_refcount),
511 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 512
57036a37
TM		 513 lo = lseg->pls_layout;
514 inode = lo->plh_inode;
4ef2e4f8 515
d684d2ae 516 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 517 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
518 spin_unlock(&inode->i_lock);
519 return;
520 }
8f0d27dc 521 pnfs_get_layout_hdr(lo);
4ef2e4f8 522 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 523 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
524 lseg = NULL;
4ef2e4f8
TM		 525 spin_unlock(&inode->i_lock);
526 pnfs_free_lseg(lseg);
527 pnfs_put_layout_hdr(lo);
4541d16c 528 }
4541d16c 529}
9369a431 530EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 531
6543f803 532static void pnfs_free_lseg_async_work(struct work_struct *work)
e6cf82d1
WAA		 533{
534 struct pnfs_layout_segment *lseg;
6543f803 535 struct pnfs_layout_hdr *lo;
e6cf82d1
WAA		 536
537 lseg = container_of(work, struct pnfs_layout_segment, pls_work);
6543f803 538 lo = lseg->pls_layout;
e6cf82d1 539
6543f803
TM		 540 pnfs_free_lseg(lseg);
541 pnfs_put_layout_hdr(lo);
e6cf82d1
WAA		 542}
543
6543f803 544static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
e6cf82d1 545{
6543f803 546 INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
e6cf82d1
WAA		 547 schedule_work(&lseg->pls_work);
548}
6543f803
TM		 549
550void
551pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
552{
553 if (!lseg)
554 return;
555
556 assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);
557
558 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
559 atomic_read(&lseg->pls_refcount),
560 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
561 if (atomic_dec_and_test(&lseg->pls_refcount)) {
562 struct pnfs_layout_hdr *lo = lseg->pls_layout;
faa4a54f
TM		 563 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
564 return;
6543f803 565 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 566 if (!pnfs_cache_lseg_for_layoutreturn(lo, lseg)) {
567 pnfs_get_layout_hdr(lo);
568 pnfs_free_lseg_async(lseg);
569 }
6543f803
TM		 570 }
571}
e6cf82d1 572
fb3296eb
BH		 573/*
574 * is l2 fully contained in l1?
575 * start1 end1
576 * [----------------------------------)
577 * start2 end2
578 * [----------------)
579 */
3cb2df17 580static bool
7dc0ac70 581pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 582 const struct pnfs_layout_range *l2)
fb3296eb
BH		 583{
584 u64 start1 = l1->offset;
17822b20 585 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 586 u64 start2 = l2->offset;
17822b20 587 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 588
589 return (start1 <= start2) && (end1 >= end2);
590}
591
24956804
TM		 592static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
593 struct list_head *tmp_list)
594{
595 if (!atomic_dec_and_test(&lseg->pls_refcount))
596 return false;
597 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
598 list_add(&lseg->pls_list, tmp_list);
599 return true;
600}
601
4541d16c
FI		 602/* Returns 1 if lseg is removed from list, 0 otherwise */
603static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
604 struct list_head *tmp_list)
605{
606 int rv = 0;
607
608 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
609 /* Remove the reference keeping the lseg in the
610 * list. It will now be removed when all
611 * outstanding io is finished.
612 */
d684d2ae
FI		 613 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
614 atomic_read(&lseg->pls_refcount));
24956804 615 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 616 rv = 1;
4541d16c
FI		 617 }
618 return rv;
619}
620
6d597e17
JL		 621/*
622 * Compare 2 layout stateid sequence ids, to see which is newer,
623 * taking into account wraparound issues.
624 */
625static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
626{
627 return (s32)(s1 - s2) > 0;
628}
629
e036f464
TM		 630static bool
631pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
632 const struct pnfs_layout_range *recall_range)
633{
634 return (recall_range->iomode == IOMODE_ANY ||
635 lseg_range->iomode == recall_range->iomode) &&
636 pnfs_lseg_range_intersecting(lseg_range, recall_range);
637}
638
639static bool
640pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
641 const struct pnfs_layout_range *recall_range,
642 u32 seq)
643{
644 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
645 return false;
646 if (recall_range == NULL)
647 return true;
648 return pnfs_should_free_range(&lseg->pls_range, recall_range);
649}
650
6d597e17
JL		 651/**
652 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
653 * @lo: layout header containing the lsegs
654 * @tmp_list: list head where doomed lsegs should go
655 * @recall_range: optional recall range argument to match (may be NULL)
656 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
657 *
658 * Walk the list of lsegs in the layout header, and tear down any that should
659 * be destroyed. If "recall_range" is specified then the segment must match
660 * that range. If "seq" is non-zero, then only match segments that were handed
661 * out at or before that sequence.
662 *
663 * Returns number of matching invalid lsegs remaining in list after scanning
664 * it and purging them.
4541d16c 665 */
43f1b3da 666int
49a85061 667pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 668 struct list_head *tmp_list,
6d597e17
JL		 669 const struct pnfs_layout_range *recall_range,
670 u32 seq)
974cec8c
AA		 671{
672 struct pnfs_layout_segment *lseg, *next;
71b39854 673 int remaining = 0;
974cec8c
AA		 674
675 dprintk("%s:Begin lo %p\n", __func__, lo);
676
8006bfba 677 if (list_empty(&lo->plh_segs))
38511722 678 return 0;
4541d16c 679 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 680 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
6d597e17 681 dprintk("%s: freeing lseg %p iomode %d seq %u"
4541d16c 682 "offset %llu length %llu\n", __func__,
6d597e17
JL		 683 lseg, lseg->pls_range.iomode, lseg->pls_seq,
684 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 685 if (!mark_lseg_invalid(lseg, tmp_list))
686 remaining++;
4541d16c 687 }
71b39854
TM		 688 dprintk("%s:Return %i\n", __func__, remaining);
689 return remaining;
974cec8c
AA		 690}
691
68f74479
TM		 692static void
693pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
694 struct list_head *free_me,
695 const struct pnfs_layout_range *range,
696 u32 seq)
697{
698 struct pnfs_layout_segment *lseg, *next;
699
700 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
701 if (pnfs_match_lseg_recall(lseg, range, seq))
702 list_move_tail(&lseg->pls_list, free_me);
703 }
704}
705
f49f9baa 706/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 707void
4541d16c 708pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 709{
4541d16c 710 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 711
712 if (list_empty(free_me))
713 return;
714
4541d16c 715 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 716 list_del(&lseg->pls_list);
905ca191 717 pnfs_free_lseg(lseg);
974cec8c
AA		 718 }
719}
720
e5e94017
BH		 721void
722pnfs_destroy_layout(struct nfs_inode *nfsi)
723{
724 struct pnfs_layout_hdr *lo;
974cec8c 725 LIST_HEAD(tmp_list);
e5e94017
BH		 726
727 spin_lock(&nfsi->vfs_inode.i_lock);
728 lo = nfsi->layout;
729 if (lo) {
3e621214 730 pnfs_get_layout_hdr(lo);
2454dfea 731 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 732 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
733 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
734 spin_unlock(&nfsi->vfs_inode.i_lock);
735 pnfs_free_lseg_list(&tmp_list);
1f18b82c 736 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 737 pnfs_put_layout_hdr(lo);
738 } else
739 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 740}
041245c8 741EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 742
fd9a8d71
TM		 743static bool
744pnfs_layout_add_bulk_destroy_list(struct inode *inode,
745 struct list_head *layout_list)
974cec8c
AA		 746{
747 struct pnfs_layout_hdr *lo;
fd9a8d71 748 bool ret = false;
974cec8c 749
fd9a8d71
TM		 750 spin_lock(&inode->i_lock);
751 lo = NFS_I(inode)->layout;
752 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
753 pnfs_get_layout_hdr(lo);
754 list_add(&lo->plh_bulk_destroy, layout_list);
755 ret = true;
756 }
757 spin_unlock(&inode->i_lock);
758 return ret;
759}
760
761/* Caller must hold rcu_read_lock and clp->cl_lock */
762static int
763pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
764 struct nfs_server *server,
765 struct list_head *layout_list)
766{
767 struct pnfs_layout_hdr *lo, *next;
768 struct inode *inode;
769
770 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
b85f5620
TM		 771 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
772 continue;
fd9a8d71
TM		 773 inode = igrab(lo->plh_inode);
774 if (inode == NULL)
775 continue;
776 list_del_init(&lo->plh_layouts);
777 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
778 continue;
779 rcu_read_unlock();
780 spin_unlock(&clp->cl_lock);
781 iput(inode);
782 spin_lock(&clp->cl_lock);
783 rcu_read_lock();
784 return -EAGAIN;
785 }
786 return 0;
787}
788
789static int
790pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
791 bool is_bulk_recall)
792{
793 struct pnfs_layout_hdr *lo;
794 struct inode *inode;
fd9a8d71
TM		 795 LIST_HEAD(lseg_list);
796 int ret = 0;
797
798 while (!list_empty(layout_list)) {
799 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
800 plh_bulk_destroy);
801 dprintk("%s freeing layout for inode %lu\n", __func__,
802 lo->plh_inode->i_ino);
803 inode = lo->plh_inode;
7c5d1875
CH		 804
805 pnfs_layoutcommit_inode(inode, false);
806
fd9a8d71
TM		 807 spin_lock(&inode->i_lock);
808 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 809 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
810 if (is_bulk_recall)
811 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 812 ret = -EAGAIN;
9fd4b9fc 813 }
fd9a8d71
TM		 814 spin_unlock(&inode->i_lock);
815 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 816 /* Free all lsegs that are attached to commit buckets */
817 nfs_commit_inode(inode, 0);
fd9a8d71
TM		 818 pnfs_put_layout_hdr(lo);
819 iput(inode);
820 }
821 return ret;
822}
823
824int
825pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
826 struct nfs_fsid *fsid,
827 bool is_recall)
828{
829 struct nfs_server *server;
830 LIST_HEAD(layout_list);
c47abcf8 831
974cec8c 832 spin_lock(&clp->cl_lock);
6382a441 833 rcu_read_lock();
fd9a8d71 834restart:
6382a441 835 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 836 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
837 continue;
838 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
839 server,
840 &layout_list) != 0)
841 goto restart;
6382a441
WAA		 842 }
843 rcu_read_unlock();
974cec8c
AA		 844 spin_unlock(&clp->cl_lock);
845
fd9a8d71
TM		 846 if (list_empty(&layout_list))
847 return 0;
848 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
849}
850
851int
852pnfs_destroy_layouts_byclid(struct nfs_client *clp,
853 bool is_recall)
854{
855 struct nfs_server *server;
856 LIST_HEAD(layout_list);
857
858 spin_lock(&clp->cl_lock);
859 rcu_read_lock();
860restart:
861 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
862 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
863 server,
864 &layout_list) != 0)
865 goto restart;
974cec8c 866 }
fd9a8d71
TM		 867 rcu_read_unlock();
868 spin_unlock(&clp->cl_lock);
869
870 if (list_empty(&layout_list))
871 return 0;
872 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
873}
874
875/*
876 * Called by the state manger to remove all layouts established under an
877 * expired lease.
878 */
879void
880pnfs_destroy_all_layouts(struct nfs_client *clp)
881{
882 nfs4_deviceid_mark_client_invalid(clp);
883 nfs4_deviceid_purge_client(clp);
884
885 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 886}
887
fd6002e9 888/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 889void
890pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
891 bool update_barrier)
b1f69b75 892{
ecebb80b 893 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 894
2d2f24ad
TM		 895 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
896 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 897
898 if (!pnfs_layout_is_valid(lo)) {
899 nfs4_stateid_copy(&lo->plh_stateid, new);
900 lo->plh_barrier = newseq;
901 pnfs_clear_layoutreturn_info(lo);
902 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
903 return;
904 }
905 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 906 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 907 /*
908 * Because of wraparound, we want to keep the barrier
909 * "close" to the current seqids.
910 */
911 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 912 }
ecebb80b
TM		 913 if (update_barrier)
914 new_barrier = be32_to_cpu(new->seqid);
915 else if (new_barrier == 0)
916 return;
2a59a041 917 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 918 lo->plh_barrier = new_barrier;
b1f69b75
AA		 919}
920
cf7d63f1 921static bool
19c54aba
TM		 922pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
923 const nfs4_stateid *stateid)
43f1b3da 924{
19c54aba 925 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 926
19c54aba
TM		 927 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
928}
929
930/* lget is set to 1 if called from inside send_layoutget call chain */
931static bool
e1c06f80 932pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 933{
f7e8917a 934 return lo->plh_block_lgets ||
e1c06f80 935 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 936}
937
b1f69b75 938/*
183d9e7b
JL		 939 * Get layout from server.
940 * for now, assume that whole file layouts are requested.
941 * arg->offset: 0
942 * arg->length: all ones
943 */
e5e94017
BH		 944static struct pnfs_layout_segment *
945send_layoutget(struct pnfs_layout_hdr *lo,
946 struct nfs_open_context *ctx,
183d9e7b 947 nfs4_stateid *stateid,
e144e539 948 const struct pnfs_layout_range *range,
183d9e7b 949 long *timeout, gfp_t gfp_flags)
e5e94017 950{
b7edfaa1 951 struct inode *ino = lo->plh_inode;
b1f69b75
AA		 952 struct nfs_server *server = NFS_SERVER(ino);
953 struct nfs4_layoutget *lgp;
2d89a1d3 954 loff_t i_size;
b1f69b75
AA		 955
956 dprintk("--> %s\n", __func__);
e5e94017 957
4f2e9dce
JL		 958 /*
959 * Synchronously retrieve layout information from server and
960 * store in lseg. If we race with a concurrent seqid morphing
961 * op, then re-send the LAYOUTGET.
962 */
83026d80
JL		 963 lgp = kzalloc(sizeof(*lgp), gfp_flags);
964 if (lgp == NULL)
965 return ERR_PTR(-ENOMEM);
966
967 i_size = i_size_read(ino);
968
969 lgp->args.minlength = PAGE_SIZE;
970 if (lgp->args.minlength > range->length)
971 lgp->args.minlength = range->length;
972 if (range->iomode == IOMODE_READ) {
973 if (range->offset >= i_size)
974 lgp->args.minlength = 0;
975 else if (i_size - range->offset < lgp->args.minlength)
976 lgp->args.minlength = i_size - range->offset;
d03ab29d 977 }
83026d80
JL		 978 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
979 pnfs_copy_range(&lgp->args.range, range);
980 lgp->args.type = server->pnfs_curr_ld->id;
981 lgp->args.inode = ino;
982 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 983 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80
JL		 984 lgp->gfp_flags = gfp_flags;
985 lgp->cred = lo->plh_lc_cred;
35124a09 986
183d9e7b 987 return nfs4_proc_layoutget(lgp, timeout, gfp_flags);
974cec8c
AA		 988}
989
24956804
TM		 990static void pnfs_clear_layoutcommit(struct inode *inode,
991 struct list_head *head)
992{
993 struct nfs_inode *nfsi = NFS_I(inode);
994 struct pnfs_layout_segment *lseg, *tmp;
995
996 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
997 return;
998 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
999 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1000 continue;
1001 pnfs_lseg_dec_and_remove_zero(lseg, head);
1002 }
1003}
1004
68f74479 1005void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1006 const nfs4_stateid *arg_stateid,
68f74479 1007 const struct pnfs_layout_range *range,
68f74479
TM		 1008 const nfs4_stateid *stateid)
1009{
1010 struct inode *inode = lo->plh_inode;
1011 LIST_HEAD(freeme);
1012
1013 spin_lock(&inode->i_lock);
2a974425
TM		 1014 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1015 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1016 goto out_unlock;
68f74479 1017 if (stateid) {
2a974425
TM		 1018 u32 seq = be32_to_cpu(arg_stateid->seqid);
1019
68f74479
TM		 1020 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1021 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1022 pnfs_set_layout_stateid(lo, stateid, true);
1023 } else
1024 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1025out_unlock:
68f74479
TM		 1026 pnfs_clear_layoutreturn_waitbit(lo);
1027 spin_unlock(&inode->i_lock);
1028 pnfs_free_lseg_list(&freeme);
1029
1030}
1031
13c13a6a 1032static bool
e5fd1904
TM		 1033pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1034 nfs4_stateid *stateid,
1035 enum pnfs_iomode *iomode)
13c13a6a 1036{
bf0291dd
TM		 1037 /* Serialise LAYOUTGET/LAYOUTRETURN */
1038 if (atomic_read(&lo->plh_outstanding) != 0)
1039 return false;
6604b203 1040 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1041 return false;
6604b203 1042 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1043 pnfs_get_layout_hdr(lo);
e5fd1904
TM		 1044 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1045 if (stateid != NULL) {
1046 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1047 if (lo->plh_return_seq != 0)
1048 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1049 }
1050 if (iomode != NULL)
1051 *iomode = lo->plh_return_iomode;
1052 pnfs_clear_layoutreturn_info(lo);
1053 return true;
1054 }
1055 if (stateid != NULL)
1056 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1057 if (iomode != NULL)
1058 *iomode = IOMODE_ANY;
13c13a6a
TM		 1059 return true;
1060}
1061
828ed9ec
TM		 1062static void
1063pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1064 struct pnfs_layout_hdr *lo,
1065 const nfs4_stateid *stateid,
1066 enum pnfs_iomode iomode)
1067{
1068 struct inode *inode = lo->plh_inode;
1069
1070 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1071 args->inode = inode;
1072 args->range.iomode = iomode;
1073 args->range.offset = 0;
1074 args->range.length = NFS4_MAX_UINT64;
1075 args->layout = lo;
1076 nfs4_stateid_copy(&args->stateid, stateid);
1077}
1078
f40eb5d0 1079static int
ed429d6b 1080pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
6c16605d 1081 enum pnfs_iomode iomode, bool sync)
f40eb5d0
PT		 1082{
1083 struct inode *ino = lo->plh_inode;
287bd3e9 1084 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0
PT		 1085 struct nfs4_layoutreturn *lrp;
1086 int status = 0;
1087
e4af440a 1088 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1089 if (unlikely(lrp == NULL)) {
1090 status = -ENOMEM;
1091 spin_lock(&ino->i_lock);
d67ae825 1092 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0
PT		 1093 spin_unlock(&ino->i_lock);
1094 pnfs_put_layout_hdr(lo);
1095 goto out;
1096 }
1097
828ed9ec 1098 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1099 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0
PT		 1100 lrp->clp = NFS_SERVER(ino)->nfs_client;
1101 lrp->cred = lo->plh_lc_cred;
287bd3e9
TM		 1102 if (ld->prepare_layoutreturn)
1103 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1104
6c16605d 1105 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1106out:
1107 dprintk("<-- %s status: %d\n", __func__, status);
1108 return status;
1109}
1110
13c13a6a
TM		 1111/* Return true if layoutreturn is needed */
1112static bool
1113pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1114{
1115 struct pnfs_layout_segment *s;
1116
2370abda 1117 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1118 return false;
1119
1120 /* Defer layoutreturn until all lsegs are done */
1121 list_for_each_entry(s, &lo->plh_segs, pls_list) {
1122 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1123 return false;
1124 }
1125
1126 return true;
1127}
1128
1129static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1130{
1131 struct inode *inode= lo->plh_inode;
1132
2370abda 1133 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1134 return;
1135 spin_lock(&inode->i_lock);
1136 if (pnfs_layout_need_return(lo)) {
1137 nfs4_stateid stateid;
1138 enum pnfs_iomode iomode;
1139 bool send;
1140
e5fd1904 1141 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
13c13a6a
TM		 1142 spin_unlock(&inode->i_lock);
1143 if (send) {
1144 /* Send an async layoutreturn so we dont deadlock */
1145 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1146 }
1147 } else
1148 spin_unlock(&inode->i_lock);
1149}
1150
293b3b06
AA		 1151/*
1152 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1153 * when the layout segment list is empty.
1154 *
1155 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1156 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1157 * deviceid is marked invalid.
1158 */
cbe82603
BH		 1159int
1160_pnfs_return_layout(struct inode *ino)
1161{
1162 struct pnfs_layout_hdr *lo = NULL;
1163 struct nfs_inode *nfsi = NFS_I(ino);
1164 LIST_HEAD(tmp_list);
cbe82603 1165 nfs4_stateid stateid;
24408f52 1166 int status = 0;
7f27392c 1167 bool send;
cbe82603 1168
366d5052 1169 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1170
1171 spin_lock(&ino->i_lock);
1172 lo = nfsi->layout;
e5929f3c 1173 if (!lo) {
cbe82603 1174 spin_unlock(&ino->i_lock);
293b3b06
AA		 1175 dprintk("NFS: %s no layout to return\n", __func__);
1176 goto out;
cbe82603 1177 }
cbe82603 1178 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1179 pnfs_get_layout_hdr(lo);
24408f52
TM		 1180 /* Is there an outstanding layoutreturn ? */
1181 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1182 spin_unlock(&ino->i_lock);
1183 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1184 TASK_UNINTERRUPTIBLE))
1185 goto out_put_layout_hdr;
1186 spin_lock(&ino->i_lock);
1187 }
24956804 1188 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1189 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1190
1191 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1192 struct pnfs_layout_range range = {
1193 .iomode = IOMODE_ANY,
1194 .offset = 0,
1195 .length = NFS4_MAX_UINT64,
1196 };
1197 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1198 }
1199
293b3b06 1200 /* Don't send a LAYOUTRETURN if list was initially empty */
24408f52 1201 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
293b3b06 1202 spin_unlock(&ino->i_lock);
293b3b06 1203 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1204 goto out_put_layout_hdr;
293b3b06 1205 }
47abadef 1206
e5fd1904 1207 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603 1208 spin_unlock(&ino->i_lock);
7f27392c 1209 if (send)
ed429d6b 1210 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c 1211out_put_layout_hdr:
ee6625a9 1212 pnfs_free_lseg_list(&tmp_list);
7f27392c 1213 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1214out:
1215 dprintk("<-- %s status: %d\n", __func__, status);
1216 return status;
1217}
1218
24028672
TM		 1219int
1220pnfs_commit_and_return_layout(struct inode *inode)
1221{
1222 struct pnfs_layout_hdr *lo;
1223 int ret;
1224
1225 spin_lock(&inode->i_lock);
1226 lo = NFS_I(inode)->layout;
1227 if (lo == NULL) {
1228 spin_unlock(&inode->i_lock);
1229 return 0;
1230 }
1231 pnfs_get_layout_hdr(lo);
1232 /* Block new layoutgets and read/write to ds */
1233 lo->plh_block_lgets++;
1234 spin_unlock(&inode->i_lock);
1235 filemap_fdatawait(inode->i_mapping);
1236 ret = pnfs_layoutcommit_inode(inode, true);
1237 if (ret == 0)
1238 ret = _pnfs_return_layout(inode);
1239 spin_lock(&inode->i_lock);
1240 lo->plh_block_lgets--;
1241 spin_unlock(&inode->i_lock);
1242 pnfs_put_layout_hdr(lo);
1243 return ret;
1244}
1245
1c5bd76d
TM		 1246bool pnfs_roc(struct inode *ino,
1247 struct nfs4_layoutreturn_args *args,
1248 struct nfs4_layoutreturn_res *res,
1249 const struct rpc_cred *cred)
f7e8917a 1250{
40dd4b7a
TM		 1251 struct nfs_inode *nfsi = NFS_I(ino);
1252 struct nfs_open_context *ctx;
1253 struct nfs4_state *state;
f7e8917a 1254 struct pnfs_layout_hdr *lo;
1c5bd76d 1255 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1256 nfs4_stateid stateid;
1c5bd76d
TM		 1257 enum pnfs_iomode iomode = 0;
1258 bool layoutreturn = false, roc = false;
e71708d4 1259 bool skip_read = false;
f7e8917a 1260
1c5bd76d
TM		 1261 if (!nfs_have_layout(ino))
1262 return false;
29ade5db 1263retry:
f7e8917a 1264 spin_lock(&ino->i_lock);
40dd4b7a 1265 lo = nfsi->layout;
0cdc329e
TM		 1266 if (!lo || !pnfs_layout_is_valid(lo) ||
1267 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
40dd4b7a 1268 goto out_noroc;
29ade5db
TM		 1269 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1270 pnfs_get_layout_hdr(lo);
1271 spin_unlock(&ino->i_lock);
1272 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1273 TASK_UNINTERRUPTIBLE);
1274 pnfs_put_layout_hdr(lo);
1275 goto retry;
1276 }
40dd4b7a 1277
e755d638 1278 /* no roc if we hold a delegation */
e71708d4
TM		 1279 if (nfs4_check_delegation(ino, FMODE_READ)) {
1280 if (nfs4_check_delegation(ino, FMODE_WRITE))
1281 goto out_noroc;
1282 skip_read = true;
1283 }
40dd4b7a
TM		 1284
1285 list_for_each_entry(ctx, &nfsi->open_files, list) {
1286 state = ctx->state;
e71708d4
TM		 1287 if (state == NULL)
1288 continue;
40dd4b7a 1289 /* Don't return layout if there is open file state */
e71708d4 1290 if (state->state & FMODE_WRITE)
40dd4b7a 1291 goto out_noroc;
e71708d4
TM		 1292 if (state->state & FMODE_READ)
1293 skip_read = true;
40dd4b7a
TM		 1294 }
1295
e755d638 1296
1c5bd76d 1297 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1298 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1299 continue;
e755d638 1300 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1301 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1302 continue;
1303 /*
1304 * Note: mark lseg for return so pnfs_layout_remove_lseg
1305 * doesn't invalidate the layout for us.
1306 */
1307 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1308 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1309 continue;
1310 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1311 }
1312
1c5bd76d
TM		 1313 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1314 goto out_noroc;
69820d22 1315
500d701f 1316 /* ROC in two conditions:
e755d638
PT		 1317 * 1. there are ROC lsegs
1318 * 2. we don't send layoutreturn
e755d638 1319 */
1c5bd76d
TM		 1320 /* lo ref dropped in pnfs_roc_release() */
1321 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1322 /* If the creds don't match, we can't compound the layoutreturn */
1323 if (!layoutreturn || cred != lo->plh_lc_cred)
1324 goto out_noroc;
1325
1326 roc = layoutreturn;
1327 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1328 res->lrs_present = 0;
1329 layoutreturn = false;
f7e8917a 1330
40dd4b7a 1331out_noroc:
f7e8917a 1332 spin_unlock(&ino->i_lock);
e755d638 1333 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1334 if (roc) {
1335 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1336 if (ld->prepare_layoutreturn)
1337 ld->prepare_layoutreturn(args);
1338 return true;
1339 }
e755d638 1340 if (layoutreturn)
1c5bd76d 1341 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
287bd3e9 1342 return false;
f7e8917a
FI		 1343}
1344
1c5bd76d
TM		 1345void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1346 struct nfs4_layoutreturn_res *res,
1347 int ret)
f7e8917a 1348{
1c5bd76d
TM		 1349 struct pnfs_layout_hdr *lo = args->layout;
1350 const nfs4_stateid *arg_stateid = NULL;
1351 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1352 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1353
1c5bd76d
TM		 1354 if (ret == 0) {
1355 arg_stateid = &args->stateid;
1356 if (res->lrs_present)
1357 res_stateid = &res->stateid;
1358 }
1359 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1360 res_stateid);
287bd3e9
TM		 1361 if (ld_private && ld_private->ops && ld_private->ops->free)
1362 ld_private->ops->free(ld_private);
1c5bd76d
TM		 1363 pnfs_put_layout_hdr(lo);
1364 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1365}
1366
500d701f
PT		 1367bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1368{
1369 struct nfs_inode *nfsi = NFS_I(ino);
1370 struct pnfs_layout_hdr *lo;
1371 bool sleep = false;
1372
1373 /* we might not have grabbed lo reference. so need to check under
1374 * i_lock */
1375 spin_lock(&ino->i_lock);
1376 lo = nfsi->layout;
ee284e35
TM		 1377 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1378 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1379 sleep = true;
ee284e35 1380 }
500d701f 1381 spin_unlock(&ino->i_lock);
500d701f
PT		 1382 return sleep;
1383}
1384
b1f69b75
AA		 1385/*
1386 * Compare two layout segments for sorting into layout cache.
1387 * We want to preferentially return RW over RO layouts, so ensure those
1388 * are seen first.
1389 */
1390static s64
7dc0ac70 1391pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1392 const struct pnfs_layout_range *l2)
b1f69b75 1393{
fb3296eb
BH		 1394 s64 d;
1395
1396 /* high offset > low offset */
1397 d = l1->offset - l2->offset;
1398 if (d)
1399 return d;
1400
1401 /* short length > long length */
1402 d = l2->length - l1->length;
1403 if (d)
1404 return d;
1405
b1f69b75 1406 /* read > read/write */
fb3296eb 1407 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1408}
1409
03772d2f
TM		 1410static bool
1411pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1412 const struct pnfs_layout_range *l2)
1413{
1414 return pnfs_lseg_range_cmp(l1, l2) > 0;
1415}
1416
1417static bool
1418pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1419 struct pnfs_layout_segment *old)
1420{
1421 return false;
1422}
1423
1424void
1425pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1426 struct pnfs_layout_segment *lseg,
1427 bool (*is_after)(const struct pnfs_layout_range *,
1428 const struct pnfs_layout_range *),
1429 bool (*do_merge)(struct pnfs_layout_segment *,
1430 struct pnfs_layout_segment *),
1431 struct list_head *free_me)
974cec8c 1432{
03772d2f 1433 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1434
974cec8c
AA		 1435 dprintk("%s:Begin\n", __func__);
1436
03772d2f
TM		 1437 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1438 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1439 continue;
1440 if (do_merge(lseg, lp)) {
1441 mark_lseg_invalid(lp, free_me);
1442 continue;
1443 }
1444 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1445 continue;
566052c5 1446 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1447 dprintk("%s: inserted lseg %p "
1448 "iomode %d offset %llu length %llu before "
1449 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1450 __func__, lseg, lseg->pls_range.iomode,
1451 lseg->pls_range.offset, lseg->pls_range.length,
1452 lp, lp->pls_range.iomode, lp->pls_range.offset,
1453 lp->pls_range.length);
fb3296eb 1454 goto out;
974cec8c 1455 }
fb3296eb
BH		 1456 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1457 dprintk("%s: inserted lseg %p "
1458 "iomode %d offset %llu length %llu at tail\n",
1459 __func__, lseg, lseg->pls_range.iomode,
1460 lseg->pls_range.offset, lseg->pls_range.length);
1461out:
70c3bd2b 1462 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1463
1464 dprintk("%s:Return\n", __func__);
e5e94017 1465}
03772d2f
TM		 1466EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1467
1468static void
1469pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1470 struct pnfs_layout_segment *lseg,
1471 struct list_head *free_me)
1472{
1473 struct inode *inode = lo->plh_inode;
1474 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1475
1476 if (ld->add_lseg != NULL)
1477 ld->add_lseg(lo, lseg, free_me);
1478 else
1479 pnfs_generic_layout_insert_lseg(lo, lseg,
1480 pnfs_lseg_range_is_after,
1481 pnfs_lseg_no_merge,
1482 free_me);
1483}
e5e94017
BH		 1484
1485static struct pnfs_layout_hdr *
9fa40758
PT		 1486alloc_init_layout_hdr(struct inode *ino,
1487 struct nfs_open_context *ctx,
1488 gfp_t gfp_flags)
e5e94017
BH		 1489{
1490 struct pnfs_layout_hdr *lo;
1491
636fb9c8 1492 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1493 if (!lo)
1494 return NULL;
cc6e5340 1495 atomic_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1496 INIT_LIST_HEAD(&lo->plh_layouts);
1497 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1498 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1499 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1500 lo->plh_inode = ino;
5cc2216d 1501 lo->plh_lc_cred = get_rpccred(ctx->cred);
67a3b721 1502 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1503 return lo;
1504}
1505
1506static struct pnfs_layout_hdr *
9fa40758
PT		 1507pnfs_find_alloc_layout(struct inode *ino,
1508 struct nfs_open_context *ctx,
1509 gfp_t gfp_flags)
e5241e43
TM		 1510 __releases(&ino->i_lock)
1511 __acquires(&ino->i_lock)
e5e94017
BH		 1512{
1513 struct nfs_inode *nfsi = NFS_I(ino);
1514 struct pnfs_layout_hdr *new = NULL;
1515
1516 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1517
251ec410
TM		 1518 if (nfsi->layout != NULL)
1519 goto out_existing;
e5e94017 1520 spin_unlock(&ino->i_lock);
9fa40758 1521 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1522 spin_lock(&ino->i_lock);
1523
251ec410 1524 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1525 nfsi->layout = new;
251ec410 1526 return new;
7175fe90
YN		 1527 } else if (new != NULL)
1528 pnfs_free_layout_hdr(new);
251ec410
TM		 1529out_existing:
1530 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1531 return nfsi->layout;
1532}
1533
b1f69b75
AA		 1534/*
1535 * iomode matching rules:
c7d73af2
TH		 1536 * iomode lseg strict match
1537 * iomode
1538 * ----- ----- ------ -----
1539 * ANY READ N/A true
1540 * ANY RW N/A true
1541 * RW READ N/A false
1542 * RW RW N/A true
1543 * READ READ N/A true
1544 * READ RW true false
1545 * READ RW false true
b1f69b75 1546 */
3cb2df17 1547static bool
7dc0ac70 1548pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1549 const struct pnfs_layout_range *range,
1550 bool strict_iomode)
b1f69b75 1551{
fb3296eb
BH		 1552 struct pnfs_layout_range range1;
1553
1554 if ((range->iomode == IOMODE_RW &&
1555 ls_range->iomode != IOMODE_RW) ||
c7d73af2
TH		 1556 (range->iomode != ls_range->iomode &&
1557 strict_iomode == true) ||
7dc0ac70 1558 !pnfs_lseg_range_intersecting(ls_range, range))
fb3296eb
BH		 1559 return 0;
1560
1561 /* range1 covers only the first byte in the range */
1562 range1 = *range;
1563 range1.length = 1;
7dc0ac70 1564 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1565}
1566
1567/*
1568 * lookup range in layout
1569 */
e5e94017 1570static struct pnfs_layout_segment *
fb3296eb 1571pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1572 struct pnfs_layout_range *range,
1573 bool strict_iomode)
e5e94017 1574{
b1f69b75
AA		 1575 struct pnfs_layout_segment *lseg, *ret = NULL;
1576
1577 dprintk("%s:Begin\n", __func__);
1578
b7edfaa1 1579 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1580 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
ce6ab4f2 1581 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
c7d73af2
TH		 1582 pnfs_lseg_range_match(&lseg->pls_range, range,
1583 strict_iomode)) {
9369a431 1584 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1585 break;
1586 }
b1f69b75
AA		 1587 }
1588
1589 dprintk("%s:Return lseg %p ref %d\n",
4541d16c 1590 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
b1f69b75 1591 return ret;
e5e94017
BH		 1592}
1593
d23d61c8
AA		 1594/*
1595 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1596 * to the MDS or over pNFS
1597 *
1598 * The nfs_inode read_io and write_io fields are cumulative counters reset
1599 * when there are no layout segments. Note that in pnfs_update_layout iomode
1600 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1601 * WRITE request.
1602 *
1603 * A return of true means use MDS I/O.
1604 *
1605 * From rfc 5661:
1606 * If a file's size is smaller than the file size threshold, data accesses
1607 * SHOULD be sent to the metadata server. If an I/O request has a length that
1608 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1609 * server. If both file size and I/O size are provided, the client SHOULD
1610 * reach or exceed both thresholds before sending its read or write
1611 * requests to the data server.
1612 */
1613static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1614 struct inode *ino, int iomode)
1615{
1616 struct nfs4_threshold *t = ctx->mdsthreshold;
1617 struct nfs_inode *nfsi = NFS_I(ino);
1618 loff_t fsize = i_size_read(ino);
1619 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1620
1621 if (t == NULL)
1622 return ret;
1623
1624 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1625 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1626
1627 switch (iomode) {
1628 case IOMODE_READ:
1629 if (t->bm & THRESHOLD_RD) {
1630 dprintk("%s fsize %llu\n", __func__, fsize);
1631 size_set = true;
1632 if (fsize < t->rd_sz)
1633 size = true;
1634 }
1635 if (t->bm & THRESHOLD_RD_IO) {
1636 dprintk("%s nfsi->read_io %llu\n", __func__,
1637 nfsi->read_io);
1638 io_set = true;
1639 if (nfsi->read_io < t->rd_io_sz)
1640 io = true;
1641 }
1642 break;
1643 case IOMODE_RW:
1644 if (t->bm & THRESHOLD_WR) {
1645 dprintk("%s fsize %llu\n", __func__, fsize);
1646 size_set = true;
1647 if (fsize < t->wr_sz)
1648 size = true;
1649 }
1650 if (t->bm & THRESHOLD_WR_IO) {
1651 dprintk("%s nfsi->write_io %llu\n", __func__,
1652 nfsi->write_io);
1653 io_set = true;
1654 if (nfsi->write_io < t->wr_io_sz)
1655 io = true;
1656 }
1657 break;
1658 }
1659 if (size_set && io_set) {
1660 if (size && io)
1661 ret = true;
1662 } else if (size || io)
1663 ret = true;
1664
1665 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1666 return ret;
1667}
1668
aa8a45ee
PT		 1669static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1670{
1671 /*
1672 * send layoutcommit as it can hold up layoutreturn due to lseg
1673 * reference
1674 */
1675 pnfs_layoutcommit_inode(lo->plh_inode, false);
1676 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1677 nfs_wait_bit_killable,
aa8a45ee
PT		 1678 TASK_UNINTERRUPTIBLE);
1679}
1680
d67ae825
TH		 1681static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1682{
1683 unsigned long *bitlock = &lo->plh_flags;
1684
1685 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1686 smp_mb__after_atomic();
1687 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1688}
1689
e5e94017
BH		 1690/*
1691 * Layout segment is retreived from the server if not cached.
1692 * The appropriate layout segment is referenced and returned to the caller.
1693 */
7c24d948 1694struct pnfs_layout_segment *
e5e94017
BH		 1695pnfs_update_layout(struct inode *ino,
1696 struct nfs_open_context *ctx,
fb3296eb
BH		 1697 loff_t pos,
1698 u64 count,
a75b9df9 1699 enum pnfs_iomode iomode,
c7d73af2 1700 bool strict_iomode,
a75b9df9 1701 gfp_t gfp_flags)
e5e94017 1702{
fb3296eb
BH		 1703 struct pnfs_layout_range arg = {
1704 .iomode = iomode,
1705 .offset = pos,
1706 .length = count,
1707 };
183d9e7b 1708 unsigned pg_offset, seq;
6382a441
WAA		 1709 struct nfs_server *server = NFS_SERVER(ino);
1710 struct nfs_client *clp = server->nfs_client;
183d9e7b 1711 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1712 struct pnfs_layout_segment *lseg = NULL;
183d9e7b
JL		 1713 nfs4_stateid stateid;
1714 long timeout = 0;
66b53f32 1715 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1716 bool first;
e5e94017 1717
9a4bf31d 1718 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1719 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1720 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1721 goto out;
9a4bf31d 1722 }
d23d61c8 1723
9a4bf31d 1724 if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
183d9e7b 1725 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1726 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
4ae93560 1727 goto out;
9a4bf31d 1728 }
4ae93560 1729
9a4bf31d 1730 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1731 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1732 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1733 goto out;
9a4bf31d 1734 }
d23d61c8 1735
9bf87482 1736lookup_again:
b88fa69e 1737 nfs4_client_recover_expired_lease(clp);
9bf87482 1738 first = false;
e5e94017 1739 spin_lock(&ino->i_lock);
9fa40758 1740 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1741 if (lo == NULL) {
1742 spin_unlock(&ino->i_lock);
183d9e7b 1743 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1744 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1745 goto out;
1746 }
e5e94017 1747
43f1b3da 1748 /* Do we even need to bother with this? */
a59c30ac 1749 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1750 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1751 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1752 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1753 goto out_unlock;
1754 }
1755
1756 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1757 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1758 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1759 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1760 goto out_unlock;
9a4bf31d 1761 }
e5e94017 1762
c7d73af2 1763 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1764 if (lseg) {
1765 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1766 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1767 goto out_unlock;
1768 }
1769
1770 if (!nfs4_valid_open_stateid(ctx->state)) {
1771 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1772 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1773 goto out_unlock;
1774 }
1775
1776 /*
1777 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1778 * stateid, or it has been invalidated, then we must use the open
1779 * stateid.
1780 */
67a3b721 1781 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
183d9e7b
JL		 1782
1783 /*
1784 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1785 * RFC 5661 Errata 3208.
1786 */
1787 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1788 &lo->plh_flags)) {
1789 spin_unlock(&ino->i_lock);
1790 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1791 TASK_UNINTERRUPTIBLE);
1792 pnfs_put_layout_hdr(lo);
183d9e7b 1793 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1794 goto lookup_again;
1795 }
183d9e7b
JL		 1796
1797 first = true;
1798 do {
1799 seq = read_seqbegin(&ctx->state->seqlock);
1800 nfs4_stateid_copy(&stateid, &ctx->state->stateid);
1801 } while (read_seqretry(&ctx->state->seqlock, seq));
9bf87482 1802 } else {
183d9e7b 1803 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 1804 }
568e8c49 1805
aa8a45ee
PT		 1806 /*
1807 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1808 * for LAYOUTRETURN even if first is true.
1809 */
8f70f53a 1810 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 1811 spin_unlock(&ino->i_lock);
1812 dprintk("%s wait for layoutreturn\n", __func__);
1813 if (pnfs_prepare_to_retry_layoutget(lo)) {
d67ae825
TH		 1814 if (first)
1815 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 1816 pnfs_put_layout_hdr(lo);
1817 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 1818 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1819 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 1820 goto lookup_again;
1821 }
183d9e7b 1822 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1823 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 1824 goto out_put_layout_hdr;
1825 }
1826
9a4bf31d 1827 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 1828 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1829 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 1830 goto out_unlock;
9a4bf31d 1831 }
cf7d63f1 1832 atomic_inc(&lo->plh_outstanding);
f49f9baa 1833 spin_unlock(&ino->i_lock);
30005121 1834
abb9a007 1835 if (list_empty(&lo->plh_layouts)) {
2130ff66
FI		 1836 /* The lo must be on the clp list if there is any
1837 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1838 */
1839 spin_lock(&clp->cl_lock);
abb9a007
PT		 1840 if (list_empty(&lo->plh_layouts))
1841 list_add_tail(&lo->plh_layouts, &server->layouts);
2130ff66
FI		 1842 spin_unlock(&clp->cl_lock);
1843 }
e5e94017 1844
09cbfeaf 1845 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 1846 if (pg_offset) {
1847 arg.offset -= pg_offset;
1848 arg.length += pg_offset;
1849 }
7c24d948 1850 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 1851 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 1852
183d9e7b
JL		 1853 lseg = send_layoutget(lo, ctx, &stateid, &arg, &timeout, gfp_flags);
1854 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1855 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
56b38a1f 1856 atomic_dec(&lo->plh_outstanding);
83026d80 1857 if (IS_ERR(lseg)) {
183d9e7b 1858 switch(PTR_ERR(lseg)) {
e85d7ee4 1859 case -EBUSY:
183d9e7b
JL		 1860 if (time_after(jiffies, giveup))
1861 lseg = NULL;
66b53f32
TM		 1862 break;
1863 case -ERECALLCONFLICT:
1864 /* Huh? We hold no layouts, how is there a recall? */
1865 if (first) {
1866 lseg = NULL;
1867 break;
1868 }
1869 /* Destroy the existing layout and start over */
1870 if (time_after(jiffies, giveup))
1871 pnfs_destroy_layout(NFS_I(ino));
183d9e7b
JL		 1872 /* Fallthrough */
1873 case -EAGAIN:
56b38a1f 1874 break;
183d9e7b
JL		 1875 default:
1876 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
1877 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1878 lseg = NULL;
1879 }
56b38a1f
TM		 1880 goto out_put_layout_hdr;
1881 }
1882 if (lseg) {
1883 if (first)
1884 pnfs_clear_first_layoutget(lo);
1885 trace_pnfs_update_layout(ino, pos, count,
1886 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
1887 pnfs_put_layout_hdr(lo);
1888 goto lookup_again;
83026d80
JL		 1889 }
1890 } else {
1891 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1892 }
1893
830ffb56 1894out_put_layout_hdr:
d67ae825
TH		 1895 if (first)
1896 pnfs_clear_first_layoutget(lo);
70c3bd2b 1897 pnfs_put_layout_hdr(lo);
e5e94017 1898out:
f86bbcf8
TM		 1899 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1900 "(%s, offset: %llu, length: %llu)\n",
1901 __func__, ino->i_sb->s_id,
1902 (unsigned long long)NFS_FILEID(ino),
d600ad1f 1903 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 1904 iomode==IOMODE_RW ? "read/write" : "read-only",
1905 (unsigned long long)pos,
1906 (unsigned long long)count);
e5e94017
BH		 1907 return lseg;
1908out_unlock:
1909 spin_unlock(&ino->i_lock);
830ffb56 1910 goto out_put_layout_hdr;
e5e94017 1911}
7c24d948 1912EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 1913
540d9864
TM		 1914static bool
1915pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
1916{
1917 switch (range->iomode) {
1918 case IOMODE_READ:
1919 case IOMODE_RW:
1920 break;
1921 default:
1922 return false;
1923 }
1924 if (range->offset == NFS4_MAX_UINT64)
1925 return false;
1926 if (range->length == 0)
1927 return false;
1928 if (range->length != NFS4_MAX_UINT64 &&
1929 range->length > NFS4_MAX_UINT64 - range->offset)
1930 return false;
1931 return true;
1932}
1933
a0b0a6e3 1934struct pnfs_layout_segment *
b1f69b75
AA		 1935pnfs_layout_process(struct nfs4_layoutget *lgp)
1936{
1937 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1938 struct nfs4_layoutget_res *res = &lgp->res;
1939 struct pnfs_layout_segment *lseg;
b7edfaa1 1940 struct inode *ino = lo->plh_inode;
78096cca 1941 LIST_HEAD(free_me);
540d9864
TM		 1942
1943 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 1944 return ERR_PTR(-EINVAL);
b1f69b75
AA		 1945
1946 /* Inject layout blob into I/O device driver */
a75b9df9 1947 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 1948 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 1949 if (!lseg)
1b3c6d07
JL		 1950 lseg = ERR_PTR(-ENOMEM);
1951
1952 dprintk("%s: Could not allocate layout: error %ld\n",
1953 __func__, PTR_ERR(lseg));
1954 return lseg;
b1f69b75
AA		 1955 }
1956
119cef97 1957 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 1958
b1f69b75 1959 spin_lock(&ino->i_lock);
e1c06f80 1960 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 1961 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 1962 goto out_forget;
43f1b3da 1963 }
038d6493 1964
9888d837
TM		 1965 if (!pnfs_layout_is_valid(lo)) {
1966 /* We have a completely new layout */
1967 pnfs_set_layout_stateid(lo, &res->stateid, true);
1968 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 1969 /* existing state ID, make sure the sequence number matches. */
1970 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
1971 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 1972 goto out_forget;
362f7474
CH		 1973 }
1974 pnfs_set_layout_stateid(lo, &res->stateid, false);
1975 } else {
1976 /*
1977 * We got an entirely new state ID. Mark all segments for the
9888d837 1978 * inode invalid, and retry the layoutget
362f7474 1979 */
d9b61708 1980 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 1981 goto out_forget;
362f7474 1982 }
038d6493 1983
9369a431 1984 pnfs_get_lseg(lseg);
03772d2f 1985 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 1986
b1f69b75 1987
3976143b 1988 if (res->return_on_close)
f7e8917a 1989 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 1990
b1f69b75 1991 spin_unlock(&ino->i_lock);
78096cca 1992 pnfs_free_lseg_list(&free_me);
a0b0a6e3 1993 return lseg;
43f1b3da 1994
1b3c6d07 1995out_forget:
43f1b3da
FI		 1996 spin_unlock(&ino->i_lock);
1997 lseg->pls_layout = lo;
1998 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1f18b82c
TM		 1999 if (!pnfs_layout_is_valid(lo))
2000 nfs_commit_inode(ino, 0);
1b3c6d07 2001 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2002}
2003
2f215968
TM		 2004/**
2005 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2006 * @lo: pointer to layout header
2007 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2008 * @return_range: describe layout segment ranges to be returned
2009 *
2010 * This function is mainly intended for use by layoutrecall. It attempts
2011 * to free the layout segment immediately, or else to mark it for return
2012 * as soon as its reference count drops to zero.
2013 */
10335556 2014int
016256df
PT		 2015pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2016 struct list_head *tmp_list,
6d597e17
JL		 2017 const struct pnfs_layout_range *return_range,
2018 u32 seq)
016256df
PT		 2019{
2020 struct pnfs_layout_segment *lseg, *next;
10335556 2021 int remaining = 0;
016256df
PT		 2022
2023 dprintk("%s:Begin lo %p\n", __func__, lo);
2024
2025 if (list_empty(&lo->plh_segs))
10335556 2026 return 0;
016256df 2027
fc7ff367 2028 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2029
2030 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2031 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2032 dprintk("%s: marking lseg %p iomode %d "
2033 "offset %llu length %llu\n", __func__,
2034 lseg, lseg->pls_range.iomode,
2035 lseg->pls_range.offset,
2036 lseg->pls_range.length);
2f215968
TM		 2037 if (mark_lseg_invalid(lseg, tmp_list))
2038 continue;
2039 remaining++;
016256df 2040 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2041 }
6d597e17
JL		 2042
2043 if (remaining)
2044 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2045
10335556 2046 return remaining;
016256df
PT		 2047}
2048
2049void pnfs_error_mark_layout_for_return(struct inode *inode,
2050 struct pnfs_layout_segment *lseg)
2051{
2052 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2053 struct pnfs_layout_range range = {
2054 .iomode = lseg->pls_range.iomode,
2055 .offset = 0,
2056 .length = NFS4_MAX_UINT64,
2057 };
10335556 2058 bool return_now = false;
016256df
PT		 2059
2060 spin_lock(&inode->i_lock);
bdebfccd
TM		 2061 if (!pnfs_layout_is_valid(lo)) {
2062 spin_unlock(&inode->i_lock);
2063 return;
2064 }
2d6cf5ab 2065 pnfs_set_plh_return_info(lo, range.iomode, 0);
016256df
PT		 2066 /*
2067 * mark all matching lsegs so that we are sure to have no live
2068 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2069 * for how it works.
2070 */
68f74479 2071 if (!pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0)) {
10335556 2072 nfs4_stateid stateid;
e5fd1904 2073 enum pnfs_iomode iomode;
10335556 2074
e5fd1904 2075 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2076 spin_unlock(&inode->i_lock);
2077 if (return_now)
2078 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2079 } else {
2080 spin_unlock(&inode->i_lock);
2081 nfs_commit_inode(inode, 0);
2082 }
016256df
PT		 2083}
2084EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2085
b3230e80
TM		 2086void
2087pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2088{
2089 if (pgio->pg_lseg == NULL ||
2090 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2091 return;
2092 pnfs_put_lseg(pgio->pg_lseg);
2093 pgio->pg_lseg = NULL;
2094}
2095EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2096
08cb5b0f
BC		 2097/*
2098 * Check for any intersection between the request and the pgio->pg_lseg,
2099 * and if none, put this pgio->pg_lseg away.
2100 */
2101static void
2102pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2103{
2104 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2105 pnfs_put_lseg(pgio->pg_lseg);
2106 pgio->pg_lseg = NULL;
2107 }
2108}
2109
d8007d4d
TM		 2110void
2111pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2112{
1fd937bd
PT		 2113 u64 rd_size = req->wb_bytes;
2114
b3230e80 2115 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2116 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2117 if (pgio->pg_lseg == NULL) {
2118 if (pgio->pg_dreq == NULL)
2119 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2120 else
2121 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2122
2123 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2124 req->wb_context,
2125 req_offset(req),
2126 rd_size,
2127 IOMODE_READ,
c7d73af2 2128 false,
cb5d04bc 2129 GFP_KERNEL);
d600ad1f
PT		 2130 if (IS_ERR(pgio->pg_lseg)) {
2131 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2132 pgio->pg_lseg = NULL;
2133 return;
2134 }
cb5d04bc 2135 }
e885de1a
TM		 2136 /* If no lseg, fall back to read through mds */
2137 if (pgio->pg_lseg == NULL)
1f945357 2138 nfs_pageio_reset_read_mds(pgio);
e885de1a 2139
d8007d4d
TM		 2140}
2141EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2142
2143void
6296556f
PT		 2144pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2145 struct nfs_page *req, u64 wb_size)
d8007d4d 2146{
b3230e80 2147 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2148 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2149 if (pgio->pg_lseg == NULL) {
cb5d04bc
PT		 2150 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2151 req->wb_context,
2152 req_offset(req),
2153 wb_size,
2154 IOMODE_RW,
c7d73af2 2155 false,
cb5d04bc 2156 GFP_NOFS);
d600ad1f
PT		 2157 if (IS_ERR(pgio->pg_lseg)) {
2158 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2159 pgio->pg_lseg = NULL;
2160 return;
2161 }
2162 }
e885de1a
TM		 2163 /* If no lseg, fall back to write through mds */
2164 if (pgio->pg_lseg == NULL)
1f945357 2165 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2166}
2167EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2168
180bb5ec
WAA		 2169void
2170pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2171{
2172 if (desc->pg_lseg) {
2173 pnfs_put_lseg(desc->pg_lseg);
2174 desc->pg_lseg = NULL;
2175 }
2176}
2177EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2178
b4fdac1a
WAA		 2179/*
2180 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2181 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2182 */
2183size_t
a7d42ddb
WAA		 2184pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2185 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2186{
0f9c429e 2187 unsigned int size;
c5e20cb7 2188 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2189
2190 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2191 if (!size)
2192 return 0;
94ad1c80 2193
19982ba8 2194 /*
c5e20cb7
WAA		 2195 * 'size' contains the number of bytes left in the current page (up
2196 * to the original size asked for in @req->wb_bytes).
2197 *
2198 * Calculate how many bytes are left in the layout segment
2199 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2200 *
2201 * Please also note that 'end_offset' is actually the offset of the
2202 * first byte that lies outside the pnfs_layout_range. FIXME?
2203 *
2204 */
19b54848 2205 if (pgio->pg_lseg) {
17822b20 2206 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2207 pgio->pg_lseg->pls_range.length);
2208 req_start = req_offset(req);
08cb5b0f 2209
c5e20cb7 2210 /* start of request is past the last byte of this segment */
08cb5b0f 2211 if (req_start >= seg_end)
19b54848 2212 return 0;
c5e20cb7
WAA		 2213
2214 /* adjust 'size' iff there are fewer bytes left in the
2215 * segment than what nfs_generic_pg_test returned */
2216 seg_left = seg_end - req_start;
2217 if (seg_left < size)
2218 size = (unsigned int)seg_left;
19b54848 2219 }
0f9c429e 2220
19b54848 2221 return size;
94ad1c80 2222}
89a58e32 2223EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2224
53113ad3 2225int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2226{
2227 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2228
2229 /* Resend all requests through the MDS */
53113ad3
WAA		 2230 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2231 hdr->completion_ops);
c7070113 2232 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2233 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2234}
e7dd79af 2235EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2236
d45f60c6 2237static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2238{
cd841605
FI		 2239
2240 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2241 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2242 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2243 pnfs_return_layout(hdr->inode);
1acbbb4e 2244 }
6c75dc0d 2245 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2246 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2247}
2248
d20581aa
BH		 2249/*
2250 * Called by non rpc-based layout drivers
2251 */
d45f60c6 2252void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2253{
f8417b48 2254 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2255 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2256 hdr->mds_offset + hdr->res.count);
d45f60c6 2257 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2258 }
2259 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2260 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2261 pnfs_ld_handle_write_error(hdr);
2262 hdr->mds_ops->rpc_release(hdr);
44b83799 2263}
d20581aa 2264EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2265
dce81290
TM		 2266static void
2267pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2268 struct nfs_pgio_header *hdr)
dce81290 2269{
48d635f1 2270 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2271
6c75dc0d 2272 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2273 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2274 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2275 mirror->pg_recoalesce = 1;
6c75dc0d 2276 }
1ca018d2 2277 hdr->release(hdr);
dce81290
TM		 2278}
2279
2280static enum pnfs_try_status
d45f60c6 2281pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2282 const struct rpc_call_ops *call_ops,
2283 struct pnfs_layout_segment *lseg,
2284 int how)
0382b744 2285{
cd841605 2286 struct inode *inode = hdr->inode;
0382b744
AA		 2287 enum pnfs_try_status trypnfs;
2288 struct nfs_server *nfss = NFS_SERVER(inode);
2289
cd841605 2290 hdr->mds_ops = call_ops;
0382b744
AA		 2291
2292 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2293 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2294 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2295 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2296 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2297 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2298 return trypnfs;
2299}
2300
dce81290 2301static void
7f714720
WAA		 2302pnfs_do_write(struct nfs_pageio_descriptor *desc,
2303 struct nfs_pgio_header *hdr, int how)
dce81290 2304{
dce81290
TM		 2305 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2306 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2307 enum pnfs_try_status trypnfs;
dce81290 2308
d45f60c6 2309 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2310 switch (trypnfs) {
2311 case PNFS_NOT_ATTEMPTED:
d45f60c6 2312 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2313 case PNFS_ATTEMPTED:
2314 break;
2315 case PNFS_TRY_AGAIN:
2316 /* cleanup hdr and prepare to redo pnfs */
2317 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2318 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2319 list_splice_init(&hdr->pages, &mirror->pg_list);
2320 mirror->pg_recoalesce = 1;
2321 }
2322 hdr->mds_ops->rpc_release(hdr);
2323 }
dce81290
TM		 2324}
2325
6c75dc0d
FI		 2326static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2327{
9369a431 2328 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2329 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2330}
2331
dce81290
TM		 2332int
2333pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2334{
6c75dc0d 2335 struct nfs_pgio_header *hdr;
dce81290
TM		 2336 int ret;
2337
1e7f3a48
WAA		 2338 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2339 if (!hdr) {
2bff2288
PT		 2340 desc->pg_error = -ENOMEM;
2341 return desc->pg_error;
dce81290 2342 }
6c75dc0d 2343 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2344
9369a431 2345 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2346 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2347 if (!ret)
7f714720 2348 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2349
6c75dc0d 2350 return ret;
dce81290
TM		 2351}
2352EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2353
53113ad3 2354int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2355{
2356 struct nfs_pageio_descriptor pgio;
2357
1acbbb4e 2358 /* Resend all requests through the MDS */
53113ad3
WAA		 2359 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2360 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2361}
e7dd79af 2362EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2363
d45f60c6 2364static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2365{
cd841605
FI		 2366 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2367 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2368 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2369 pnfs_return_layout(hdr->inode);
1acbbb4e 2370 }
4db6e0b7 2371 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2372 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2373}
2374
d20581aa
BH		 2375/*
2376 * Called by non rpc-based layout drivers
2377 */
d45f60c6 2378void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2379{
bfc505de 2380 if (likely(!hdr->pnfs_error))
d45f60c6 2381 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2382 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2383 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2384 pnfs_ld_handle_read_error(hdr);
2385 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2386}
2387EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2388
493292dd
TM		 2389static void
2390pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2391 struct nfs_pgio_header *hdr)
493292dd 2392{
48d635f1 2393 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2394
4db6e0b7 2395 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2396 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2397 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2398 mirror->pg_recoalesce = 1;
4db6e0b7 2399 }
1ca018d2 2400 hdr->release(hdr);
493292dd
TM		 2401}
2402
64419a9b
AA		 2403/*
2404 * Call the appropriate parallel I/O subsystem read function.
2405 */
493292dd 2406static enum pnfs_try_status
d45f60c6 2407pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2408 const struct rpc_call_ops *call_ops,
2409 struct pnfs_layout_segment *lseg)
64419a9b 2410{
cd841605 2411 struct inode *inode = hdr->inode;
64419a9b
AA		 2412 struct nfs_server *nfss = NFS_SERVER(inode);
2413 enum pnfs_try_status trypnfs;
2414
cd841605 2415 hdr->mds_ops = call_ops;
64419a9b
AA		 2416
2417 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2418 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2419
d45f60c6 2420 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2421 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2422 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2423 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2424 return trypnfs;
2425}
863a3c6c 2426
ceb11e13 2427/* Resend all requests through pnfs. */
1b1bc66b 2428void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2429{
2430 struct nfs_pageio_descriptor pgio;
2431
1b1bc66b 2432 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2433 /* Prevent deadlocks with layoutreturn! */
2434 pnfs_put_lseg(hdr->lseg);
2435 hdr->lseg = NULL;
2436
1b1bc66b
WAA		 2437 nfs_pageio_init_read(&pgio, hdr->inode, false,
2438 hdr->completion_ops);
2439 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2440 }
ceb11e13
PT		 2441}
2442EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2443
493292dd 2444static void
7f714720 2445pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2446{
493292dd
TM		 2447 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2448 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2449 enum pnfs_try_status trypnfs;
493292dd 2450
d45f60c6 2451 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2452 switch (trypnfs) {
2453 case PNFS_NOT_ATTEMPTED:
d45f60c6 2454 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2455 case PNFS_ATTEMPTED:
2456 break;
2457 case PNFS_TRY_AGAIN:
2458 /* cleanup hdr and prepare to redo pnfs */
2459 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2460 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2461 list_splice_init(&hdr->pages, &mirror->pg_list);
2462 mirror->pg_recoalesce = 1;
2463 }
2464 hdr->mds_ops->rpc_release(hdr);
2465 }
493292dd
TM		 2466}
2467
4db6e0b7
FI		 2468static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2469{
9369a431 2470 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2471 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2472}
2473
493292dd
TM		 2474int
2475pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2476{
4db6e0b7 2477 struct nfs_pgio_header *hdr;
493292dd
TM		 2478 int ret;
2479
1e7f3a48
WAA		 2480 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2481 if (!hdr) {
2bff2288
PT		 2482 desc->pg_error = -ENOMEM;
2483 return desc->pg_error;
493292dd 2484 }
4db6e0b7 2485 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2486 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2487 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2488 if (!ret)
7f714720 2489 pnfs_do_read(desc, hdr);
4db6e0b7 2490 return ret;
493292dd
TM		 2491}
2492EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2493
71244d9b
TM		 2494static void pnfs_clear_layoutcommitting(struct inode *inode)
2495{
2496 unsigned long *bitlock = &NFS_I(inode)->flags;
2497
2498 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2499 smp_mb__after_atomic();
71244d9b
TM		 2500 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2501}
2502
863a3c6c 2503/*
a9bae566 2504 * There can be multiple RW segments.
863a3c6c 2505 */
a9bae566 2506static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2507{
a9bae566 2508 struct pnfs_layout_segment *lseg;
863a3c6c 2509
a9bae566
PT		 2510 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2511 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2512 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2513 list_add(&lseg->pls_lc_list, listp);
2514 }
863a3c6c
AA		 2515}
2516
a073dbff
TM		 2517static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2518{
2519 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2520
2521 /* Matched by references in pnfs_set_layoutcommit */
2522 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2523 list_del_init(&lseg->pls_lc_list);
2524 pnfs_put_lseg(lseg);
2525 }
2526
71244d9b 2527 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2528}
2529
1b0ae068
PT		 2530void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2531{
b9e028fd 2532 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2533}
2534EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2535
863a3c6c 2536void
67af7611
TM		 2537pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2538 loff_t end_pos)
863a3c6c 2539{
cd841605 2540 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2541 bool mark_as_dirty = false;
863a3c6c 2542
cd841605 2543 spin_lock(&inode->i_lock);
863a3c6c 2544 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2545 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2546 mark_as_dirty = true;
863a3c6c 2547 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2548 __func__, inode->i_ino);
29559b11
TM		 2549 } else if (end_pos > nfsi->layout->plh_lwb)
2550 nfsi->layout->plh_lwb = end_pos;
67af7611 2551 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2552 /* references matched in nfs4_layoutcommit_release */
67af7611 2553 pnfs_get_lseg(lseg);
a9bae566 2554 }
cd841605 2555 spin_unlock(&inode->i_lock);
acff5880 2556 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2557 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2558
2559 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2560 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2561 if (mark_as_dirty)
cd841605 2562 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2563}
2564EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2565
db29c089
AA		 2566void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2567{
2568 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2569
2570 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2571 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2572 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2573}
2574
de4b15c7
AA		 2575/*
2576 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2577 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2578 * data to disk to allow the server to recover the data if it crashes.
2579 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2580 * is off, and a COMMIT is sent to a data server, or
2581 * if WRITEs to a data server return NFS_DATA_SYNC.
2582 */
863a3c6c 2583int
ef311537 2584pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2585{
5f919c9f 2586 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2587 struct nfs4_layoutcommit_data *data;
2588 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2589 loff_t end_pos;
71244d9b 2590 int status;
863a3c6c 2591
71244d9b 2592 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2593 return 0;
2594
71244d9b 2595 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2596
71244d9b 2597 status = -EAGAIN;
92407e75 2598 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2599 if (!sync)
2600 goto out;
74316201 2601 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2602 NFS_INO_LAYOUTCOMMITTING,
2603 nfs_wait_bit_killable,
2604 TASK_KILLABLE);
92407e75 2605 if (status)
71244d9b 2606 goto out;
92407e75
PT		 2607 }
2608
71244d9b
TM		 2609 status = -ENOMEM;
2610 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2611 data = kzalloc(sizeof(*data), GFP_NOFS);
2612 if (!data)
2613 goto clear_layoutcommitting;
2614
2615 status = 0;
de4b15c7 2616 spin_lock(&inode->i_lock);
71244d9b
TM		 2617 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2618 goto out_unlock;
a9bae566 2619
71244d9b 2620 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 2621 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 2622
acff5880 2623 end_pos = nfsi->layout->plh_lwb;
863a3c6c 2624
f597c537 2625 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 2626 spin_unlock(&inode->i_lock);
2627
2628 data->args.inode = inode;
9fa40758 2629 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 2630 nfs_fattr_init(&data->fattr);
2631 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2632 data->res.fattr = &data->fattr;
2e18d4d8
TM		 2633 if (end_pos != 0)
2634 data->args.lastbytewritten = end_pos - 1;
2635 else
2636 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 2637 data->res.server = NFS_SERVER(inode);
2638
5f919c9f
CH		 2639 if (ld->prepare_layoutcommit) {
2640 status = ld->prepare_layoutcommit(&data->args);
2641 if (status) {
3471648a 2642 put_rpccred(data->cred);
5f919c9f 2643 spin_lock(&inode->i_lock);
29559b11
TM		 2644 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2645 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 2646 nfsi->layout->plh_lwb = end_pos;
3471648a 2647 goto out_unlock;
5f919c9f
CH		 2648 }
2649 }
2650
2651
863a3c6c
AA		 2652 status = nfs4_proc_layoutcommit(data, sync);
2653out:
92407e75
PT		 2654 if (status)
2655 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2656 dprintk("<-- %s status %d\n", __func__, status);
2657 return status;
71244d9b
TM		 2658out_unlock:
2659 spin_unlock(&inode->i_lock);
92407e75 2660 kfree(data);
71244d9b
TM		 2661clear_layoutcommitting:
2662 pnfs_clear_layoutcommitting(inode);
92407e75 2663 goto out;
863a3c6c 2664}
72cff449 2665EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 2666
5bb89b47
TM		 2667int
2668pnfs_generic_sync(struct inode *inode, bool datasync)
2669{
2670 return pnfs_layoutcommit_inode(inode, true);
2671}
2672EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2673
82be417a
AA		 2674struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2675{
2676 struct nfs4_threshold *thp;
2677
2678 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2679 if (!thp) {
2680 dprintk("%s mdsthreshold allocation failed\n", __func__);
2681 return NULL;
2682 }
2683 return thp;
2684}
8733408d 2685
865a7ecb 2686#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 2687int
c8ad8894 2688pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 2689{
2690 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2691 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 2692 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 2693 struct nfs42_layoutstat_data *data;
2694 struct pnfs_layout_hdr *hdr;
2695 int status = 0;
2696
2697 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2698 goto out;
2699
6c5a0d89
TM		 2700 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2701 goto out;
2702
1bfe3b25
PT		 2703 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2704 goto out;
2705
8733408d
PT		 2706 spin_lock(&inode->i_lock);
2707 if (!NFS_I(inode)->layout) {
2708 spin_unlock(&inode->i_lock);
f538d0ba 2709 goto out_clear_layoutstats;
8733408d
PT		 2710 }
2711 hdr = NFS_I(inode)->layout;
2712 pnfs_get_layout_hdr(hdr);
2713 spin_unlock(&inode->i_lock);
2714
c8ad8894 2715 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 2716 if (!data) {
2717 status = -ENOMEM;
2718 goto out_put;
2719 }
2720
2721 data->args.fh = NFS_FH(inode);
2722 data->args.inode = inode;
8733408d
PT		 2723 status = ld->prepare_layoutstats(&data->args);
2724 if (status)
2725 goto out_free;
2726
2727 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2728
2729out:
2730 dprintk("%s returns %d\n", __func__, status);
2731 return status;
2732
2733out_free:
2734 kfree(data);
2735out_put:
2736 pnfs_put_layout_hdr(hdr);
f538d0ba 2737out_clear_layoutstats:
1bfe3b25
PT		 2738 smp_mb__before_atomic();
2739 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2740 smp_mb__after_atomic();
8733408d
PT		 2741 goto out;
2742}
2743EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 2744#endif
bbf58bf3
TM		 2745
2746unsigned int layoutstats_timer;
2747module_param(layoutstats_timer, uint, 0644);
2748EXPORT_SYMBOL_GPL(layoutstats_timer);
Ubuntu Artful kernel mirror