projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
treewide: Use fallthrough pseudo-keyword
[linux-block.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
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"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
95void
96unset_pnfs_layoutdriver(struct nfs_server *nfss)
97{
738fd0f3
BH		 98 if (nfss->pnfs_curr_ld) {
99 if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 104 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 105 }
85e174ba
RL		 106 nfss->pnfs_curr_ld = NULL;
107}
108
ca440c38
JL		 109/*
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
112 *
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
115 */
116static const u32 ld_prefs[] = {
117 LAYOUT_SCSI,
118 LAYOUT_BLOCK_VOLUME,
119 LAYOUT_OSD2_OBJECTS,
120 LAYOUT_FLEX_FILES,
121 LAYOUT_NFSV4_1_FILES,
122 0
123};
124
125static int
126ld_cmp(const void *e1, const void *e2)
127{
128 u32 ld1 = *((u32 *)e1);
129 u32 ld2 = *((u32 *)e2);
130 int i;
131
132 for (i = 0; ld_prefs[i] != 0; i++) {
133 if (ld1 == ld_prefs[i])
134 return -1;
135
136 if (ld2 == ld_prefs[i])
137 return 1;
138 }
139 return 0;
140}
141
85e174ba
RL		 142/*
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
145 *
3132e49e 146 * @ids array of layout types supported by MDS.
85e174ba
RL		 147 */
148void
738fd0f3 149set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 150 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 151{
152 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 153 u32 id;
ca440c38 154 int i;
85e174ba 155
19274716
AS		 156 if (fsinfo->nlayouttypes == 0)
157 goto out_no_driver;
85e174ba
RL		 158 if (!(server->nfs_client->cl_exchange_flags &
159 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 160 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 162 goto out_no_driver;
163 }
3132e49e 164
ca440c38
JL		 165 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 167
ca440c38
JL		 168 for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 id = fsinfo->layouttype[i];
85e174ba 170 ld_type = find_pnfs_driver(id);
ca440c38
JL		 171 if (!ld_type) {
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 id);
174 ld_type = find_pnfs_driver(id);
175 }
176 if (ld_type)
177 break;
85e174ba 178 }
3132e49e
JL		 179
180 if (!ld_type) {
ca440c38 181 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 182 goto out_no_driver;
183 }
184
85e174ba 185 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 186 if (ld_type->set_layoutdriver
187 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 188 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__, id);
738fd0f3
BH		 190 module_put(ld_type->owner);
191 goto out_no_driver;
192 }
2a4c8994
TM		 193 /* Bump the MDS count */
194 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 195
85e174ba
RL		 196 dprintk("%s: pNFS module for %u set\n", __func__, id);
197 return;
198
199out_no_driver:
200 dprintk("%s: Using NFSv4 I/O\n", __func__);
201 server->pnfs_curr_ld = NULL;
202}
02c35fca
FI		 203
204int
205pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206{
207 int status = -EINVAL;
208 struct pnfs_layoutdriver_type *tmp;
209
210 if (ld_type->id == 0) {
a030889a 211 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 212 return status;
213 }
b1f69b75 214 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 215 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 216 "alloc_lseg and free_lseg.\n", __func__);
217 return status;
218 }
02c35fca
FI		 219
220 spin_lock(&pnfs_spinlock);
221 tmp = find_pnfs_driver_locked(ld_type->id);
222 if (!tmp) {
223 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 status = 0;
225 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 ld_type->name);
227 } else {
a030889a 228 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 229 __func__, ld_type->id);
230 }
231 spin_unlock(&pnfs_spinlock);
232
233 return status;
234}
235EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236
237void
238pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239{
240 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 spin_lock(&pnfs_spinlock);
242 list_del(&ld_type->pnfs_tblid);
243 spin_unlock(&pnfs_spinlock);
244}
245EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 246
b1f69b75
AA		 247/*
248 * pNFS client layout cache
249 */
250
cc6e5340 251/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 252void
70c3bd2b 253pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 254{
2b28a7be 255 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 256}
257
636fb9c8
BH		 258static struct pnfs_layout_hdr *
259pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260{
261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 262 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 263}
264
265static void
266pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267{
9c626381
TM		 268 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270
cf6605d1 271 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
9c626381
TM		 272 struct nfs_client *clp = server->nfs_client;
273
274 spin_lock(&clp->cl_lock);
cf6605d1 275 list_del_rcu(&lo->plh_layouts);
9c626381
TM		 276 spin_unlock(&clp->cl_lock);
277 }
a52458b4 278 put_cred(lo->plh_lc_cred);
57934278 279 return ld->free_layout_hdr(lo);
636fb9c8
BH		 280}
281
e5e94017 282static void
6622c3ea 283pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 284{
bb346f63 285 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 286 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 287 nfsi->layout = NULL;
288 /* Reset MDS Threshold I/O counters */
289 nfsi->write_io = 0;
290 nfsi->read_io = 0;
e5e94017
BH		 291}
292
b1f69b75 293void
70c3bd2b 294pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 295{
9c6376eb 296 struct inode *inode;
cc6e5340 297
9c6376eb
TM		 298 if (!lo)
299 return;
300 inode = lo->plh_inode;
13c13a6a
TM		 301 pnfs_layoutreturn_before_put_layout_hdr(lo);
302
2b28a7be 303 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 304 if (!list_empty(&lo->plh_segs))
305 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 306 pnfs_detach_layout_hdr(lo);
cc6e5340 307 spin_unlock(&inode->i_lock);
6622c3ea 308 pnfs_free_layout_hdr(lo);
cc6e5340 309 }
974cec8c
AA		 310}
311
b5fdf841
TM		 312static struct inode *
313pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
314{
315 struct inode *inode = igrab(lo->plh_inode);
316 if (inode)
317 return inode;
318 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
319 return NULL;
320}
321
4aab9732
TM		 322static void
323pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
324 u32 seq)
325{
326 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
327 iomode = IOMODE_ANY;
328 lo->plh_return_iomode = iomode;
329 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
330 if (seq != 0) {
331 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
332 lo->plh_return_seq = seq;
333 }
334}
335
ae5a459d
TM		 336static void
337pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
338{
5466d214 339 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 340 lo->plh_return_iomode = 0;
341 lo->plh_return_seq = 0;
342 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 343 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
344 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
345 continue;
346 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
347 }
ae5a459d
TM		 348}
349
362fb578
TM		 350static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
351{
352 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
353 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
354 smp_mb__after_atomic();
355 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
356 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
357}
358
fe1cf946
TM		 359static void
360pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
361 struct list_head *free_me)
362{
363 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
364 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
365 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
366 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
367 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
368 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
369}
370
7380020e 371/*
30cb3ee2
TM		 372 * Update the seqid of a layout stateid after receiving
373 * NFS4ERR_OLD_STATEID
7380020e 374 */
30cb3ee2 375bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
ecf84026
TM		 376 struct pnfs_layout_range *dst_range,
377 struct inode *inode)
7380020e
TM		 378{
379 struct pnfs_layout_hdr *lo;
c16467dc
TM		 380 struct pnfs_layout_range range = {
381 .iomode = IOMODE_ANY,
382 .offset = 0,
383 .length = NFS4_MAX_UINT64,
384 };
7380020e 385 bool ret = false;
c16467dc
TM		 386 LIST_HEAD(head);
387 int err;
7380020e
TM		 388
389 spin_lock(&inode->i_lock);
390 lo = NFS_I(inode)->layout;
30cb3ee2
TM		 391 if (lo && pnfs_layout_is_valid(lo) &&
392 nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
393 /* Is our call using the most recent seqid? If so, bump it */
394 if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
395 nfs4_stateid_seqid_inc(dst);
396 ret = true;
397 goto out;
398 }
399 /* Try to update the seqid to the most recent */
c16467dc
TM		 400 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
401 if (err != -EBUSY) {
402 dst->seqid = lo->plh_stateid.seqid;
ecf84026 403 *dst_range = range;
c16467dc
TM		 404 ret = true;
405 }
7380020e 406 }
30cb3ee2 407out:
7380020e 408 spin_unlock(&inode->i_lock);
c16467dc 409 pnfs_free_lseg_list(&head);
7380020e
TM		 410 return ret;
411}
412
2454dfea
TM		 413/*
414 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
415 *
416 * In order to continue using the pnfs_layout_hdr, a full recovery
417 * is required.
418 * Note that caller must hold inode->i_lock.
419 */
5f46be04 420int
2454dfea
TM		 421pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
422 struct list_head *lseg_list)
423{
424 struct pnfs_layout_range range = {
425 .iomode = IOMODE_ANY,
426 .offset = 0,
427 .length = NFS4_MAX_UINT64,
428 };
fe1cf946 429 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 430
431 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 432 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
433 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 434 pnfs_clear_layoutreturn_info(lo);
68f74479 435 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 436 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
437 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
438 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 439 return !list_empty(&lo->plh_segs);
2454dfea
TM		 440}
441
b9e028fd
TM		 442static int
443pnfs_iomode_to_fail_bit(u32 iomode)
444{
445 return iomode == IOMODE_RW ?
446 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
447}
448
449static void
3e621214 450pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 451{
25c75333 452 lo->plh_retry_timestamp = jiffies;
39e88fcf 453 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 454 refcount_inc(&lo->plh_refcount);
3e621214
TM		 455}
456
457static void
458pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
459{
460 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 461 refcount_dec(&lo->plh_refcount);
3e621214
TM		 462}
463
464static void
465pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
466{
467 struct inode *inode = lo->plh_inode;
115ce575
TM		 468 struct pnfs_layout_range range = {
469 .iomode = iomode,
470 .offset = 0,
471 .length = NFS4_MAX_UINT64,
472 };
473 LIST_HEAD(head);
3e621214
TM		 474
475 spin_lock(&inode->i_lock);
476 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 477 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 478 spin_unlock(&inode->i_lock);
115ce575 479 pnfs_free_lseg_list(&head);
b9e028fd
TM		 480 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
481 iomode == IOMODE_RW ? "RW" : "READ");
482}
483
484static bool
485pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
486{
25c75333 487 unsigned long start, end;
3e621214
TM		 488 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
489
490 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 491 return false;
492 end = jiffies;
493 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
494 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
495 /* It is time to retry the failed layoutgets */
3e621214 496 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 497 return false;
498 }
499 return true;
b9e028fd
TM		 500}
501
974cec8c 502static void
119cef97
TM		 503pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
504 const struct pnfs_layout_range *range,
505 const nfs4_stateid *stateid)
974cec8c 506{
566052c5 507 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 508 INIT_LIST_HEAD(&lseg->pls_lc_list);
a9901899 509 INIT_LIST_HEAD(&lseg->pls_commits);
eba6dd69 510 refcount_set(&lseg->pls_refcount, 1);
4541d16c 511 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 512 lseg->pls_layout = lo;
119cef97
TM		 513 lseg->pls_range = *range;
514 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 515}
516
905ca191 517static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 518{
68f74479
TM		 519 if (lseg != NULL) {
520 struct inode *inode = lseg->pls_layout->plh_inode;
521 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
522 }
974cec8c
AA		 523}
524
d684d2ae 525static void
57036a37
TM		 526pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
527 struct pnfs_layout_segment *lseg)
d684d2ae 528{
d20581aa 529 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 530 list_del_init(&lseg->pls_list);
8f0d27dc 531 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 532 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 533 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
534 return;
7b650994
TM		 535 if (list_empty(&lo->plh_segs) &&
536 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
537 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 538 if (atomic_read(&lo->plh_outstanding) == 0)
539 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 540 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 541 }
d684d2ae
FI		 542}
543
68f74479
TM		 544static bool
545pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
546 struct pnfs_layout_segment *lseg)
547{
548 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
549 pnfs_layout_is_valid(lo)) {
4aab9732 550 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 551 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
552 return true;
553 }
554 return false;
555}
556
bae724ef 557void
9369a431 558pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 559{
57036a37 560 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 561 struct inode *inode;
562
563 if (!lseg)
564 return;
565
4541d16c 566 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 567 refcount_read(&lseg->pls_refcount),
4541d16c 568 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 569
57036a37
TM		 570 lo = lseg->pls_layout;
571 inode = lo->plh_inode;
4ef2e4f8 572
eba6dd69 573 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 574 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
575 spin_unlock(&inode->i_lock);
576 return;
577 }
8f0d27dc 578 pnfs_get_layout_hdr(lo);
4ef2e4f8 579 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 580 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
581 lseg = NULL;
4ef2e4f8
TM		 582 spin_unlock(&inode->i_lock);
583 pnfs_free_lseg(lseg);
584 pnfs_put_layout_hdr(lo);
4541d16c 585 }
4541d16c 586}
9369a431 587EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 588
fb3296eb
BH		 589/*
590 * is l2 fully contained in l1?
591 * start1 end1
592 * [----------------------------------)
593 * start2 end2
594 * [----------------)
595 */
3cb2df17 596static bool
7dc0ac70 597pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 598 const struct pnfs_layout_range *l2)
fb3296eb
BH		 599{
600 u64 start1 = l1->offset;
17822b20 601 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 602 u64 start2 = l2->offset;
17822b20 603 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 604
605 return (start1 <= start2) && (end1 >= end2);
606}
607
24956804
TM		 608static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
609 struct list_head *tmp_list)
610{
eba6dd69 611 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 612 return false;
613 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
614 list_add(&lseg->pls_list, tmp_list);
615 return true;
616}
617
4541d16c
FI		 618/* Returns 1 if lseg is removed from list, 0 otherwise */
619static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
620 struct list_head *tmp_list)
621{
622 int rv = 0;
623
624 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
625 /* Remove the reference keeping the lseg in the
626 * list. It will now be removed when all
627 * outstanding io is finished.
628 */
d684d2ae 629 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 630 refcount_read(&lseg->pls_refcount));
24956804 631 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 632 rv = 1;
4541d16c
FI		 633 }
634 return rv;
635}
636
6d597e17
JL		 637/*
638 * Compare 2 layout stateid sequence ids, to see which is newer,
639 * taking into account wraparound issues.
640 */
641static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
642{
643 return (s32)(s1 - s2) > 0;
644}
645
e036f464
TM		 646static bool
647pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
648 const struct pnfs_layout_range *recall_range)
649{
650 return (recall_range->iomode == IOMODE_ANY ||
651 lseg_range->iomode == recall_range->iomode) &&
652 pnfs_lseg_range_intersecting(lseg_range, recall_range);
653}
654
655static bool
656pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
657 const struct pnfs_layout_range *recall_range,
658 u32 seq)
659{
660 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
661 return false;
662 if (recall_range == NULL)
663 return true;
664 return pnfs_should_free_range(&lseg->pls_range, recall_range);
665}
666
6d597e17
JL		 667/**
668 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
669 * @lo: layout header containing the lsegs
670 * @tmp_list: list head where doomed lsegs should go
671 * @recall_range: optional recall range argument to match (may be NULL)
672 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
673 *
674 * Walk the list of lsegs in the layout header, and tear down any that should
675 * be destroyed. If "recall_range" is specified then the segment must match
676 * that range. If "seq" is non-zero, then only match segments that were handed
677 * out at or before that sequence.
678 *
679 * Returns number of matching invalid lsegs remaining in list after scanning
680 * it and purging them.
4541d16c 681 */
43f1b3da 682int
49a85061 683pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 684 struct list_head *tmp_list,
6d597e17
JL		 685 const struct pnfs_layout_range *recall_range,
686 u32 seq)
974cec8c
AA		 687{
688 struct pnfs_layout_segment *lseg, *next;
71b39854 689 int remaining = 0;
974cec8c
AA		 690
691 dprintk("%s:Begin lo %p\n", __func__, lo);
692
8006bfba 693 if (list_empty(&lo->plh_segs))
38511722 694 return 0;
4541d16c 695 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 696 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 697 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 698 "offset %llu length %llu\n", __func__,
6d597e17
JL		 699 lseg, lseg->pls_range.iomode, lseg->pls_seq,
700 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 701 if (!mark_lseg_invalid(lseg, tmp_list))
702 remaining++;
4541d16c 703 }
71b39854
TM		 704 dprintk("%s:Return %i\n", __func__, remaining);
705 return remaining;
974cec8c
AA		 706}
707
68f74479
TM		 708static void
709pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
710 struct list_head *free_me,
711 const struct pnfs_layout_range *range,
712 u32 seq)
713{
714 struct pnfs_layout_segment *lseg, *next;
715
716 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
717 if (pnfs_match_lseg_recall(lseg, range, seq))
718 list_move_tail(&lseg->pls_list, free_me);
719 }
720}
721
f49f9baa 722/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 723void
4541d16c 724pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 725{
4541d16c 726 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 727
728 if (list_empty(free_me))
729 return;
730
4541d16c 731 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 732 list_del(&lseg->pls_list);
905ca191 733 pnfs_free_lseg(lseg);
974cec8c
AA		 734 }
735}
736
e5e94017
BH		 737void
738pnfs_destroy_layout(struct nfs_inode *nfsi)
739{
740 struct pnfs_layout_hdr *lo;
974cec8c 741 LIST_HEAD(tmp_list);
e5e94017
BH		 742
743 spin_lock(&nfsi->vfs_inode.i_lock);
744 lo = nfsi->layout;
745 if (lo) {
3e621214 746 pnfs_get_layout_hdr(lo);
2454dfea 747 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 748 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
749 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
750 spin_unlock(&nfsi->vfs_inode.i_lock);
751 pnfs_free_lseg_list(&tmp_list);
1f18b82c 752 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 753 pnfs_put_layout_hdr(lo);
754 } else
755 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 756}
041245c8 757EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 758
fd9a8d71
TM		 759static bool
760pnfs_layout_add_bulk_destroy_list(struct inode *inode,
761 struct list_head *layout_list)
974cec8c
AA		 762{
763 struct pnfs_layout_hdr *lo;
fd9a8d71 764 bool ret = false;
974cec8c 765
fd9a8d71
TM		 766 spin_lock(&inode->i_lock);
767 lo = NFS_I(inode)->layout;
768 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
769 pnfs_get_layout_hdr(lo);
770 list_add(&lo->plh_bulk_destroy, layout_list);
771 ret = true;
772 }
773 spin_unlock(&inode->i_lock);
774 return ret;
775}
776
777/* Caller must hold rcu_read_lock and clp->cl_lock */
778static int
779pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
780 struct nfs_server *server,
781 struct list_head *layout_list)
5085607d
TM		 782 __must_hold(&clp->cl_lock)
783 __must_hold(RCU)
fd9a8d71
TM		 784{
785 struct pnfs_layout_hdr *lo, *next;
786 struct inode *inode;
787
788 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
5085607d
TM		 789 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
790 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
791 !list_empty(&lo->plh_bulk_destroy))
b85f5620 792 continue;
5085607d
TM		 793 /* If the sb is being destroyed, just bail */
794 if (!nfs_sb_active(server->super))
795 break;
b5fdf841 796 inode = pnfs_grab_inode_layout_hdr(lo);
5085607d 797 if (inode != NULL) {
cf6605d1
TM		 798 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
799 list_del_rcu(&lo->plh_layouts);
5085607d
TM		 800 if (pnfs_layout_add_bulk_destroy_list(inode,
801 layout_list))
802 continue;
803 rcu_read_unlock();
804 spin_unlock(&clp->cl_lock);
805 iput(inode);
806 } else {
807 rcu_read_unlock();
808 spin_unlock(&clp->cl_lock);
5085607d
TM		 809 }
810 nfs_sb_deactive(server->super);
fd9a8d71
TM		 811 spin_lock(&clp->cl_lock);
812 rcu_read_lock();
813 return -EAGAIN;
814 }
815 return 0;
816}
817
818static int
819pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
820 bool is_bulk_recall)
821{
822 struct pnfs_layout_hdr *lo;
823 struct inode *inode;
fd9a8d71
TM		 824 LIST_HEAD(lseg_list);
825 int ret = 0;
826
827 while (!list_empty(layout_list)) {
828 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
829 plh_bulk_destroy);
830 dprintk("%s freeing layout for inode %lu\n", __func__,
831 lo->plh_inode->i_ino);
832 inode = lo->plh_inode;
7c5d1875
CH		 833
834 pnfs_layoutcommit_inode(inode, false);
835
fd9a8d71
TM		 836 spin_lock(&inode->i_lock);
837 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 838 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
839 if (is_bulk_recall)
840 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 841 ret = -EAGAIN;
9fd4b9fc 842 }
fd9a8d71
TM		 843 spin_unlock(&inode->i_lock);
844 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 845 /* Free all lsegs that are attached to commit buckets */
846 nfs_commit_inode(inode, 0);
fd9a8d71 847 pnfs_put_layout_hdr(lo);
5085607d 848 nfs_iput_and_deactive(inode);
fd9a8d71
TM		 849 }
850 return ret;
851}
852
853int
854pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
855 struct nfs_fsid *fsid,
856 bool is_recall)
857{
858 struct nfs_server *server;
859 LIST_HEAD(layout_list);
c47abcf8 860
974cec8c 861 spin_lock(&clp->cl_lock);
6382a441 862 rcu_read_lock();
fd9a8d71 863restart:
6382a441 864 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 865 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
866 continue;
867 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
868 server,
869 &layout_list) != 0)
870 goto restart;
6382a441
WAA		 871 }
872 rcu_read_unlock();
974cec8c
AA		 873 spin_unlock(&clp->cl_lock);
874
fd9a8d71
TM		 875 if (list_empty(&layout_list))
876 return 0;
877 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
878}
879
880int
881pnfs_destroy_layouts_byclid(struct nfs_client *clp,
882 bool is_recall)
883{
884 struct nfs_server *server;
885 LIST_HEAD(layout_list);
886
887 spin_lock(&clp->cl_lock);
888 rcu_read_lock();
889restart:
890 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
891 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
892 server,
893 &layout_list) != 0)
894 goto restart;
974cec8c 895 }
fd9a8d71
TM		 896 rcu_read_unlock();
897 spin_unlock(&clp->cl_lock);
898
899 if (list_empty(&layout_list))
900 return 0;
901 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
902}
903
904/*
905 * Called by the state manger to remove all layouts established under an
906 * expired lease.
907 */
908void
909pnfs_destroy_all_layouts(struct nfs_client *clp)
910{
911 nfs4_deviceid_mark_client_invalid(clp);
912 nfs4_deviceid_purge_client(clp);
913
914 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 915}
916
59b56394
TM		 917static void
918pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
919{
920 const struct cred *old;
921
922 if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
923 old = xchg(&lo->plh_lc_cred, get_cred(cred));
924 put_cred(old);
925 }
926}
927
fd6002e9 928/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 929void
930pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
59b56394 931 const struct cred *cred, bool update_barrier)
b1f69b75 932{
ecebb80b 933 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 934
2d2f24ad
TM		 935 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
936 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 937
938 if (!pnfs_layout_is_valid(lo)) {
59b56394 939 pnfs_set_layout_cred(lo, cred);
2a59a041
TM		 940 nfs4_stateid_copy(&lo->plh_stateid, new);
941 lo->plh_barrier = newseq;
942 pnfs_clear_layoutreturn_info(lo);
943 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
944 return;
945 }
946 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 947 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 948 /*
949 * Because of wraparound, we want to keep the barrier
950 * "close" to the current seqids.
951 */
952 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 953 }
ecebb80b
TM		 954 if (update_barrier)
955 new_barrier = be32_to_cpu(new->seqid);
956 else if (new_barrier == 0)
957 return;
2a59a041 958 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 959 lo->plh_barrier = new_barrier;
b1f69b75
AA		 960}
961
cf7d63f1 962static bool
19c54aba
TM		 963pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
964 const nfs4_stateid *stateid)
43f1b3da 965{
19c54aba 966 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 967
19c54aba
TM		 968 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
969}
970
971/* lget is set to 1 if called from inside send_layoutget call chain */
972static bool
e1c06f80 973pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 974{
f7e8917a 975 return lo->plh_block_lgets ||
e1c06f80 976 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 977}
978
5e36e2a9
FI		 979static struct nfs_server *
980pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
981{
982 struct nfs_server *server;
983
78746a38 984 if (inode) {
5e36e2a9 985 server = NFS_SERVER(inode);
78746a38 986 } else {
5e36e2a9
FI		 987 struct dentry *parent_dir = dget_parent(ctx->dentry);
988 server = NFS_SERVER(parent_dir->d_inode);
989 dput(parent_dir);
990 }
991 return server;
992}
993
29a8bfe5
TM		 994static void nfs4_free_pages(struct page **pages, size_t size)
995{
996 int i;
997
998 if (!pages)
999 return;
1000
1001 for (i = 0; i < size; i++) {
1002 if (!pages[i])
1003 break;
1004 __free_page(pages[i]);
1005 }
1006 kfree(pages);
1007}
1008
1009static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1010{
1011 struct page **pages;
1012 int i;
1013
a2791d3a 1014 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 1015 if (!pages) {
1016 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1017 return NULL;
1018 }
1019
1020 for (i = 0; i < size; i++) {
1021 pages[i] = alloc_page(gfp_flags);
1022 if (!pages[i]) {
1023 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 1024 nfs4_free_pages(pages, i);
29a8bfe5
TM		 1025 return NULL;
1026 }
1027 }
1028
1029 return pages;
1030}
1031
587f03de 1032static struct nfs4_layoutget *
5e36e2a9 1033pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 1034 struct nfs_open_context *ctx,
2409a976 1035 const nfs4_stateid *stateid,
e144e539 1036 const struct pnfs_layout_range *range,
587f03de 1037 gfp_t gfp_flags)
e5e94017 1038{
5e36e2a9 1039 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 1040 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 1041 size_t max_pages = max_response_pages(server);
b1f69b75 1042 struct nfs4_layoutget *lgp;
b1f69b75
AA		 1043
1044 dprintk("--> %s\n", __func__);
e5e94017 1045
83026d80
JL		 1046 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1047 if (lgp == NULL)
587f03de 1048 return NULL;
83026d80 1049
28ced9a8
TM		 1050 if (max_reply_sz) {
1051 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1052 if (npages < max_pages)
1053 max_pages = npages;
1054 }
1055
dacb452d
FI		 1056 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1057 if (!lgp->args.layout.pages) {
1058 kfree(lgp);
1059 return NULL;
1060 }
1061 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1062 lgp->res.layoutp = &lgp->args.layout;
1063
d49e0d5b
TM		 1064 /* Don't confuse uninitialised result and success */
1065 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1066
1067 lgp->args.minlength = PAGE_SIZE;
1068 if (lgp->args.minlength > range->length)
1069 lgp->args.minlength = range->length;
5e36e2a9
FI		 1070 if (ino) {
1071 loff_t i_size = i_size_read(ino);
1072
1073 if (range->iomode == IOMODE_READ) {
1074 if (range->offset >= i_size)
1075 lgp->args.minlength = 0;
1076 else if (i_size - range->offset < lgp->args.minlength)
1077 lgp->args.minlength = i_size - range->offset;
1078 }
d03ab29d 1079 }
83026d80
JL		 1080 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1081 pnfs_copy_range(&lgp->args.range, range);
1082 lgp->args.type = server->pnfs_curr_ld->id;
1083 lgp->args.inode = ino;
1084 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1085 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1086 lgp->gfp_flags = gfp_flags;
63ec2b69 1087 lgp->cred = ctx->cred;
587f03de 1088 return lgp;
974cec8c
AA		 1089}
1090
29a8bfe5
TM		 1091void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1092{
1093 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1094
1095 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1096 if (lgp->args.inode)
1097 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
29a8bfe5
TM		 1098 put_nfs_open_context(lgp->args.ctx);
1099 kfree(lgp);
1100}
1101
24956804
TM		 1102static void pnfs_clear_layoutcommit(struct inode *inode,
1103 struct list_head *head)
1104{
1105 struct nfs_inode *nfsi = NFS_I(inode);
1106 struct pnfs_layout_segment *lseg, *tmp;
1107
1108 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1109 return;
1110 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1111 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1112 continue;
1113 pnfs_lseg_dec_and_remove_zero(lseg, head);
1114 }
1115}
1116
68f74479 1117void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1118 const nfs4_stateid *arg_stateid,
68f74479 1119 const struct pnfs_layout_range *range,
68f74479
TM		 1120 const nfs4_stateid *stateid)
1121{
1122 struct inode *inode = lo->plh_inode;
1123 LIST_HEAD(freeme);
1124
1125 spin_lock(&inode->i_lock);
2a974425
TM		 1126 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1127 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1128 goto out_unlock;
68f74479 1129 if (stateid) {
2a974425
TM		 1130 u32 seq = be32_to_cpu(arg_stateid->seqid);
1131
68f74479
TM		 1132 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1133 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
59b56394 1134 pnfs_set_layout_stateid(lo, stateid, NULL, true);
68f74479
TM		 1135 } else
1136 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1137out_unlock:
68f74479
TM		 1138 pnfs_clear_layoutreturn_waitbit(lo);
1139 spin_unlock(&inode->i_lock);
1140 pnfs_free_lseg_list(&freeme);
1141
1142}
1143
13c13a6a 1144static bool
e5fd1904
TM		 1145pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1146 nfs4_stateid *stateid,
44ea8dfc 1147 const struct cred **cred,
e5fd1904 1148 enum pnfs_iomode *iomode)
13c13a6a 1149{
bf0291dd
TM		 1150 /* Serialise LAYOUTGET/LAYOUTRETURN */
1151 if (atomic_read(&lo->plh_outstanding) != 0)
1152 return false;
6604b203 1153 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1154 return false;
6604b203 1155 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1156 pnfs_get_layout_hdr(lo);
e5fd1904 1157 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
44ea8dfc
TM		 1158 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1159 *cred = get_cred(lo->plh_lc_cred);
1160 if (lo->plh_return_seq != 0)
1161 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
e5fd1904
TM		 1162 if (iomode != NULL)
1163 *iomode = lo->plh_return_iomode;
1164 pnfs_clear_layoutreturn_info(lo);
1165 return true;
1166 }
44ea8dfc
TM		 1167 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1168 *cred = get_cred(lo->plh_lc_cred);
e5fd1904
TM		 1169 if (iomode != NULL)
1170 *iomode = IOMODE_ANY;
13c13a6a
TM		 1171 return true;
1172}
1173
828ed9ec
TM		 1174static void
1175pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1176 struct pnfs_layout_hdr *lo,
1177 const nfs4_stateid *stateid,
1178 enum pnfs_iomode iomode)
1179{
1180 struct inode *inode = lo->plh_inode;
1181
1182 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1183 args->inode = inode;
1184 args->range.iomode = iomode;
1185 args->range.offset = 0;
1186 args->range.length = NFS4_MAX_UINT64;
1187 args->layout = lo;
1188 nfs4_stateid_copy(&args->stateid, stateid);
1189}
1190
f40eb5d0 1191static int
44ea8dfc
TM		 1192pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
1193 const nfs4_stateid *stateid,
1194 const struct cred **pcred,
1195 enum pnfs_iomode iomode,
1196 bool sync)
f40eb5d0
PT		 1197{
1198 struct inode *ino = lo->plh_inode;
287bd3e9 1199 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0 1200 struct nfs4_layoutreturn *lrp;
44ea8dfc 1201 const struct cred *cred = *pcred;
f40eb5d0
PT		 1202 int status = 0;
1203
44ea8dfc 1204 *pcred = NULL;
e4af440a 1205 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1206 if (unlikely(lrp == NULL)) {
1207 status = -ENOMEM;
1208 spin_lock(&ino->i_lock);
d67ae825 1209 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0 1210 spin_unlock(&ino->i_lock);
44ea8dfc 1211 put_cred(cred);
f40eb5d0
PT		 1212 pnfs_put_layout_hdr(lo);
1213 goto out;
1214 }
1215
828ed9ec 1216 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1217 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0 1218 lrp->clp = NFS_SERVER(ino)->nfs_client;
44ea8dfc 1219 lrp->cred = cred;
287bd3e9
TM		 1220 if (ld->prepare_layoutreturn)
1221 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1222
6c16605d 1223 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1224out:
1225 dprintk("<-- %s status: %d\n", __func__, status);
1226 return status;
1227}
1228
d474f961
TM		 1229static bool
1230pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1231 enum pnfs_iomode iomode,
1232 u32 seq)
1233{
1234 struct pnfs_layout_range recall_range = {
1235 .length = NFS4_MAX_UINT64,
1236 .iomode = iomode,
1237 };
1238 return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1239 &recall_range, seq) != -EBUSY;
1240}
1241
13c13a6a
TM		 1242/* Return true if layoutreturn is needed */
1243static bool
1244pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1245{
2370abda 1246 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a 1247 return false;
d474f961
TM		 1248 return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1249 lo->plh_return_seq);
13c13a6a
TM		 1250}
1251
1252static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1253{
1254 struct inode *inode= lo->plh_inode;
1255
2370abda 1256 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1257 return;
1258 spin_lock(&inode->i_lock);
1259 if (pnfs_layout_need_return(lo)) {
44ea8dfc 1260 const struct cred *cred;
13c13a6a
TM		 1261 nfs4_stateid stateid;
1262 enum pnfs_iomode iomode;
1263 bool send;
1264
44ea8dfc 1265 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
13c13a6a
TM		 1266 spin_unlock(&inode->i_lock);
1267 if (send) {
1268 /* Send an async layoutreturn so we dont deadlock */
44ea8dfc 1269 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
13c13a6a
TM		 1270 }
1271 } else
1272 spin_unlock(&inode->i_lock);
1273}
1274
293b3b06
AA		 1275/*
1276 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1277 * when the layout segment list is empty.
1278 *
1279 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1280 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1281 * deviceid is marked invalid.
1282 */
cbe82603
BH		 1283int
1284_pnfs_return_layout(struct inode *ino)
1285{
1286 struct pnfs_layout_hdr *lo = NULL;
1287 struct nfs_inode *nfsi = NFS_I(ino);
1288 LIST_HEAD(tmp_list);
44ea8dfc 1289 const struct cred *cred;
cbe82603 1290 nfs4_stateid stateid;
24408f52 1291 int status = 0;
93b7f7ad 1292 bool send, valid_layout;
cbe82603 1293
366d5052 1294 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1295
1296 spin_lock(&ino->i_lock);
1297 lo = nfsi->layout;
e5929f3c 1298 if (!lo) {
cbe82603 1299 spin_unlock(&ino->i_lock);
293b3b06
AA		 1300 dprintk("NFS: %s no layout to return\n", __func__);
1301 goto out;
cbe82603 1302 }
cbe82603 1303 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1304 pnfs_get_layout_hdr(lo);
24408f52
TM		 1305 /* Is there an outstanding layoutreturn ? */
1306 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1307 spin_unlock(&ino->i_lock);
1308 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1309 TASK_UNINTERRUPTIBLE))
1310 goto out_put_layout_hdr;
1311 spin_lock(&ino->i_lock);
1312 }
93b7f7ad 1313 valid_layout = pnfs_layout_is_valid(lo);
24956804 1314 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1315 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1316
1317 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1318 struct pnfs_layout_range range = {
1319 .iomode = IOMODE_ANY,
1320 .offset = 0,
1321 .length = NFS4_MAX_UINT64,
1322 };
1323 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1324 }
1325
293b3b06 1326 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1327 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1328 !valid_layout) {
293b3b06 1329 spin_unlock(&ino->i_lock);
293b3b06 1330 dprintk("NFS: %s no layout segments to return\n", __func__);
7bcc1058 1331 goto out_wait_layoutreturn;
293b3b06 1332 }
47abadef 1333
44ea8dfc 1334 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
cbe82603 1335 spin_unlock(&ino->i_lock);
7f27392c 1336 if (send)
44ea8dfc 1337 status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
7bcc1058
TM		 1338out_wait_layoutreturn:
1339 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE);
7f27392c 1340out_put_layout_hdr:
ee6625a9 1341 pnfs_free_lseg_list(&tmp_list);
7f27392c 1342 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1343out:
1344 dprintk("<-- %s status: %d\n", __func__, status);
1345 return status;
1346}
1347
24028672
TM		 1348int
1349pnfs_commit_and_return_layout(struct inode *inode)
1350{
1351 struct pnfs_layout_hdr *lo;
1352 int ret;
1353
1354 spin_lock(&inode->i_lock);
1355 lo = NFS_I(inode)->layout;
1356 if (lo == NULL) {
1357 spin_unlock(&inode->i_lock);
1358 return 0;
1359 }
1360 pnfs_get_layout_hdr(lo);
1361 /* Block new layoutgets and read/write to ds */
1362 lo->plh_block_lgets++;
1363 spin_unlock(&inode->i_lock);
1364 filemap_fdatawait(inode->i_mapping);
1365 ret = pnfs_layoutcommit_inode(inode, true);
1366 if (ret == 0)
1367 ret = _pnfs_return_layout(inode);
1368 spin_lock(&inode->i_lock);
1369 lo->plh_block_lgets--;
1370 spin_unlock(&inode->i_lock);
1371 pnfs_put_layout_hdr(lo);
1372 return ret;
1373}
1374
1c5bd76d
TM		 1375bool pnfs_roc(struct inode *ino,
1376 struct nfs4_layoutreturn_args *args,
1377 struct nfs4_layoutreturn_res *res,
a52458b4 1378 const struct cred *cred)
f7e8917a 1379{
40dd4b7a
TM		 1380 struct nfs_inode *nfsi = NFS_I(ino);
1381 struct nfs_open_context *ctx;
1382 struct nfs4_state *state;
f7e8917a 1383 struct pnfs_layout_hdr *lo;
1c5bd76d 1384 struct pnfs_layout_segment *lseg, *next;
44ea8dfc 1385 const struct cred *lc_cred;
193e3aa2 1386 nfs4_stateid stateid;
1c5bd76d
TM		 1387 enum pnfs_iomode iomode = 0;
1388 bool layoutreturn = false, roc = false;
e71708d4 1389 bool skip_read = false;
f7e8917a 1390
1c5bd76d
TM		 1391 if (!nfs_have_layout(ino))
1392 return false;
29ade5db 1393retry:
0de43976 1394 rcu_read_lock();
f7e8917a 1395 spin_lock(&ino->i_lock);
40dd4b7a 1396 lo = nfsi->layout;
0cdc329e 1397 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1398 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1399 lo = NULL;
40dd4b7a 1400 goto out_noroc;
9c6376eb
TM		 1401 }
1402 pnfs_get_layout_hdr(lo);
29ade5db 1403 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1404 spin_unlock(&ino->i_lock);
0de43976 1405 rcu_read_unlock();
29ade5db
TM		 1406 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1407 TASK_UNINTERRUPTIBLE);
1408 pnfs_put_layout_hdr(lo);
1409 goto retry;
1410 }
40dd4b7a 1411
e755d638 1412 /* no roc if we hold a delegation */
e71708d4
TM		 1413 if (nfs4_check_delegation(ino, FMODE_READ)) {
1414 if (nfs4_check_delegation(ino, FMODE_WRITE))
1415 goto out_noroc;
1416 skip_read = true;
1417 }
40dd4b7a 1418
0de43976 1419 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1420 state = ctx->state;
e71708d4
TM		 1421 if (state == NULL)
1422 continue;
40dd4b7a 1423 /* Don't return layout if there is open file state */
e71708d4 1424 if (state->state & FMODE_WRITE)
40dd4b7a 1425 goto out_noroc;
e71708d4
TM		 1426 if (state->state & FMODE_READ)
1427 skip_read = true;
40dd4b7a
TM		 1428 }
1429
e755d638 1430
1c5bd76d 1431 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1432 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1433 continue;
e755d638 1434 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1435 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1436 continue;
1437 /*
1438 * Note: mark lseg for return so pnfs_layout_remove_lseg
1439 * doesn't invalidate the layout for us.
1440 */
1441 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1442 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1443 continue;
1444 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1445 }
1446
1c5bd76d
TM		 1447 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1448 goto out_noroc;
69820d22 1449
500d701f 1450 /* ROC in two conditions:
e755d638
PT		 1451 * 1. there are ROC lsegs
1452 * 2. we don't send layoutreturn
e755d638 1453 */
1c5bd76d 1454 /* lo ref dropped in pnfs_roc_release() */
44ea8dfc 1455 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
1c5bd76d 1456 /* If the creds don't match, we can't compound the layoutreturn */
4d8948c7 1457 if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0)
1c5bd76d
TM		 1458 goto out_noroc;
1459
1460 roc = layoutreturn;
1461 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1462 res->lrs_present = 0;
1463 layoutreturn = false;
44ea8dfc 1464 put_cred(lc_cred);
4d8948c7 1465
40dd4b7a 1466out_noroc:
f7e8917a 1467 spin_unlock(&ino->i_lock);
0de43976 1468 rcu_read_unlock();
e755d638 1469 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1470 if (roc) {
1471 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1472 if (ld->prepare_layoutreturn)
1473 ld->prepare_layoutreturn(args);
9c6376eb 1474 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1475 return true;
1476 }
e755d638 1477 if (layoutreturn)
44ea8dfc 1478 pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
9c6376eb 1479 pnfs_put_layout_hdr(lo);
287bd3e9 1480 return false;
f7e8917a
FI		 1481}
1482
287a9c55
TM		 1483int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
1484 struct nfs4_layoutreturn_args **argpp,
1485 struct nfs4_layoutreturn_res **respp,
1486 int *ret)
1487{
1488 struct nfs4_layoutreturn_args *arg = *argpp;
1489 int retval = -EAGAIN;
1490
1491 if (!arg)
1492 return 0;
1493 /* Handle Layoutreturn errors */
1494 switch (*ret) {
1495 case 0:
1496 retval = 0;
1497 break;
6109bcf7
TM		 1498 case -NFS4ERR_NOMATCHING_LAYOUT:
1499 /* Was there an RPC level error? If not, retry */
1500 if (task->tk_rpc_status == 0)
1501 break;
1502 /* If the call was not sent, let caller handle it */
1503 if (!RPC_WAS_SENT(task))
1504 return 0;
1505 /*
1506 * Otherwise, assume the call succeeded and
1507 * that we need to release the layout
1508 */
1509 *ret = 0;
1510 (*respp)->lrs_present = 0;
1511 retval = 0;
1512 break;
078a432d
TM		 1513 case -NFS4ERR_DELAY:
1514 /* Let the caller handle the retry */
1515 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1516 return 0;
287a9c55 1517 case -NFS4ERR_OLD_STATEID:
30cb3ee2 1518 if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
287a9c55
TM		 1519 &arg->range, inode))
1520 break;
1521 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1522 return -EAGAIN;
1523 }
1524 *argpp = NULL;
1525 *respp = NULL;
1526 return retval;
1527}
1528
1c5bd76d
TM		 1529void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1530 struct nfs4_layoutreturn_res *res,
1531 int ret)
f7e8917a 1532{
1c5bd76d
TM		 1533 struct pnfs_layout_hdr *lo = args->layout;
1534 const nfs4_stateid *arg_stateid = NULL;
1535 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1536 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1537
9c47b18c
TM		 1538 switch (ret) {
1539 case -NFS4ERR_NOMATCHING_LAYOUT:
1540 break;
1541 case 0:
1c5bd76d
TM		 1542 if (res->lrs_present)
1543 res_stateid = &res->stateid;
df561f66 1544 fallthrough;
9c47b18c
TM		 1545 default:
1546 arg_stateid = &args->stateid;
1c5bd76d 1547 }
a19b4785 1548 trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
1c5bd76d
TM		 1549 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1550 res_stateid);
287bd3e9
TM		 1551 if (ld_private && ld_private->ops && ld_private->ops->free)
1552 ld_private->ops->free(ld_private);
1c5bd76d 1553 pnfs_put_layout_hdr(lo);
f7e8917a
FI		 1554}
1555
500d701f
PT		 1556bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1557{
1558 struct nfs_inode *nfsi = NFS_I(ino);
1559 struct pnfs_layout_hdr *lo;
1560 bool sleep = false;
1561
1562 /* we might not have grabbed lo reference. so need to check under
1563 * i_lock */
1564 spin_lock(&ino->i_lock);
1565 lo = nfsi->layout;
ee284e35
TM		 1566 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1567 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1568 sleep = true;
ee284e35 1569 }
500d701f 1570 spin_unlock(&ino->i_lock);
500d701f
PT		 1571 return sleep;
1572}
1573
b1f69b75
AA		 1574/*
1575 * Compare two layout segments for sorting into layout cache.
1576 * We want to preferentially return RW over RO layouts, so ensure those
1577 * are seen first.
1578 */
1579static s64
7dc0ac70 1580pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1581 const struct pnfs_layout_range *l2)
b1f69b75 1582{
fb3296eb
BH		 1583 s64 d;
1584
1585 /* high offset > low offset */
1586 d = l1->offset - l2->offset;
1587 if (d)
1588 return d;
1589
1590 /* short length > long length */
1591 d = l2->length - l1->length;
1592 if (d)
1593 return d;
1594
b1f69b75 1595 /* read > read/write */
fb3296eb 1596 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1597}
1598
03772d2f
TM		 1599static bool
1600pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1601 const struct pnfs_layout_range *l2)
1602{
1603 return pnfs_lseg_range_cmp(l1, l2) > 0;
1604}
1605
1606static bool
1607pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1608 struct pnfs_layout_segment *old)
1609{
1610 return false;
1611}
1612
1613void
1614pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1615 struct pnfs_layout_segment *lseg,
1616 bool (*is_after)(const struct pnfs_layout_range *,
1617 const struct pnfs_layout_range *),
1618 bool (*do_merge)(struct pnfs_layout_segment *,
1619 struct pnfs_layout_segment *),
1620 struct list_head *free_me)
974cec8c 1621{
03772d2f 1622 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1623
974cec8c
AA		 1624 dprintk("%s:Begin\n", __func__);
1625
03772d2f
TM		 1626 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1627 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1628 continue;
1629 if (do_merge(lseg, lp)) {
1630 mark_lseg_invalid(lp, free_me);
1631 continue;
1632 }
1633 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1634 continue;
566052c5 1635 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1636 dprintk("%s: inserted lseg %p "
1637 "iomode %d offset %llu length %llu before "
1638 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1639 __func__, lseg, lseg->pls_range.iomode,
1640 lseg->pls_range.offset, lseg->pls_range.length,
1641 lp, lp->pls_range.iomode, lp->pls_range.offset,
1642 lp->pls_range.length);
fb3296eb 1643 goto out;
974cec8c 1644 }
fb3296eb
BH		 1645 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1646 dprintk("%s: inserted lseg %p "
1647 "iomode %d offset %llu length %llu at tail\n",
1648 __func__, lseg, lseg->pls_range.iomode,
1649 lseg->pls_range.offset, lseg->pls_range.length);
1650out:
70c3bd2b 1651 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1652
1653 dprintk("%s:Return\n", __func__);
e5e94017 1654}
03772d2f
TM		 1655EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1656
1657static void
1658pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1659 struct pnfs_layout_segment *lseg,
1660 struct list_head *free_me)
1661{
1662 struct inode *inode = lo->plh_inode;
1663 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1664
1665 if (ld->add_lseg != NULL)
1666 ld->add_lseg(lo, lseg, free_me);
1667 else
1668 pnfs_generic_layout_insert_lseg(lo, lseg,
1669 pnfs_lseg_range_is_after,
1670 pnfs_lseg_no_merge,
1671 free_me);
1672}
e5e94017
BH		 1673
1674static struct pnfs_layout_hdr *
9fa40758
PT		 1675alloc_init_layout_hdr(struct inode *ino,
1676 struct nfs_open_context *ctx,
1677 gfp_t gfp_flags)
e5e94017
BH		 1678{
1679 struct pnfs_layout_hdr *lo;
1680
636fb9c8 1681 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1682 if (!lo)
1683 return NULL;
2b28a7be 1684 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1685 INIT_LIST_HEAD(&lo->plh_layouts);
1686 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1687 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1688 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1689 lo->plh_inode = ino;
a52458b4 1690 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1691 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1692 return lo;
1693}
1694
1695static struct pnfs_layout_hdr *
9fa40758
PT		 1696pnfs_find_alloc_layout(struct inode *ino,
1697 struct nfs_open_context *ctx,
1698 gfp_t gfp_flags)
e5241e43
TM		 1699 __releases(&ino->i_lock)
1700 __acquires(&ino->i_lock)
e5e94017
BH		 1701{
1702 struct nfs_inode *nfsi = NFS_I(ino);
1703 struct pnfs_layout_hdr *new = NULL;
1704
1705 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1706
251ec410
TM		 1707 if (nfsi->layout != NULL)
1708 goto out_existing;
e5e94017 1709 spin_unlock(&ino->i_lock);
9fa40758 1710 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1711 spin_lock(&ino->i_lock);
1712
251ec410 1713 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1714 nfsi->layout = new;
251ec410 1715 return new;
7175fe90
YN		 1716 } else if (new != NULL)
1717 pnfs_free_layout_hdr(new);
251ec410
TM		 1718out_existing:
1719 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1720 return nfsi->layout;
1721}
1722
b1f69b75
AA		 1723/*
1724 * iomode matching rules:
c7d73af2
TH		 1725 * iomode lseg strict match
1726 * iomode
1727 * ----- ----- ------ -----
1728 * ANY READ N/A true
1729 * ANY RW N/A true
1730 * RW READ N/A false
1731 * RW RW N/A true
1732 * READ READ N/A true
1733 * READ RW true false
1734 * READ RW false true
b1f69b75 1735 */
3cb2df17 1736static bool
7dc0ac70 1737pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1738 const struct pnfs_layout_range *range,
1739 bool strict_iomode)
b1f69b75 1740{
fb3296eb
BH		 1741 struct pnfs_layout_range range1;
1742
1743 if ((range->iomode == IOMODE_RW &&
1744 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1745 (range->iomode != ls_range->iomode &&
6089dd0d 1746 strict_iomode) ||
7dc0ac70 1747 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1748 return false;
fb3296eb
BH		 1749
1750 /* range1 covers only the first byte in the range */
1751 range1 = *range;
1752 range1.length = 1;
7dc0ac70 1753 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1754}
1755
1756/*
1757 * lookup range in layout
1758 */
e5e94017 1759static struct pnfs_layout_segment *
fb3296eb 1760pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1761 struct pnfs_layout_range *range,
1762 bool strict_iomode)
e5e94017 1763{
b1f69b75
AA		 1764 struct pnfs_layout_segment *lseg, *ret = NULL;
1765
1766 dprintk("%s:Begin\n", __func__);
1767
b7edfaa1 1768 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1769 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1770 pnfs_lseg_range_match(&lseg->pls_range, range,
1771 strict_iomode)) {
9369a431 1772 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1773 break;
1774 }
b1f69b75
AA		 1775 }
1776
1777 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1778 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1779 return ret;
e5e94017
BH		 1780}
1781
d23d61c8
AA		 1782/*
1783 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1784 * to the MDS or over pNFS
1785 *
1786 * The nfs_inode read_io and write_io fields are cumulative counters reset
1787 * when there are no layout segments. Note that in pnfs_update_layout iomode
1788 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1789 * WRITE request.
1790 *
1791 * A return of true means use MDS I/O.
1792 *
1793 * From rfc 5661:
1794 * If a file's size is smaller than the file size threshold, data accesses
1795 * SHOULD be sent to the metadata server. If an I/O request has a length that
1796 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1797 * server. If both file size and I/O size are provided, the client SHOULD
1798 * reach or exceed both thresholds before sending its read or write
1799 * requests to the data server.
1800 */
1801static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1802 struct inode *ino, int iomode)
1803{
1804 struct nfs4_threshold *t = ctx->mdsthreshold;
1805 struct nfs_inode *nfsi = NFS_I(ino);
1806 loff_t fsize = i_size_read(ino);
1807 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1808
1809 if (t == NULL)
1810 return ret;
1811
1812 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1813 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1814
1815 switch (iomode) {
1816 case IOMODE_READ:
1817 if (t->bm & THRESHOLD_RD) {
1818 dprintk("%s fsize %llu\n", __func__, fsize);
1819 size_set = true;
1820 if (fsize < t->rd_sz)
1821 size = true;
1822 }
1823 if (t->bm & THRESHOLD_RD_IO) {
1824 dprintk("%s nfsi->read_io %llu\n", __func__,
1825 nfsi->read_io);
1826 io_set = true;
1827 if (nfsi->read_io < t->rd_io_sz)
1828 io = true;
1829 }
1830 break;
1831 case IOMODE_RW:
1832 if (t->bm & THRESHOLD_WR) {
1833 dprintk("%s fsize %llu\n", __func__, fsize);
1834 size_set = true;
1835 if (fsize < t->wr_sz)
1836 size = true;
1837 }
1838 if (t->bm & THRESHOLD_WR_IO) {
1839 dprintk("%s nfsi->write_io %llu\n", __func__,
1840 nfsi->write_io);
1841 io_set = true;
1842 if (nfsi->write_io < t->wr_io_sz)
1843 io = true;
1844 }
1845 break;
1846 }
1847 if (size_set && io_set) {
1848 if (size && io)
1849 ret = true;
1850 } else if (size || io)
1851 ret = true;
1852
1853 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1854 return ret;
1855}
1856
d03360aa 1857static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1858{
1859 /*
1860 * send layoutcommit as it can hold up layoutreturn due to lseg
1861 * reference
1862 */
1863 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1864 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1865 nfs_wait_bit_killable,
d03360aa 1866 TASK_KILLABLE);
aa8a45ee
PT		 1867}
1868
411ae722
TM		 1869static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1870{
1871 atomic_inc(&lo->plh_outstanding);
1872}
1873
1874static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1875{
1876 if (atomic_dec_and_test(&lo->plh_outstanding))
1877 wake_up_var(&lo->plh_outstanding);
1878}
1879
d67ae825
TH		 1880static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1881{
1882 unsigned long *bitlock = &lo->plh_flags;
1883
1884 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1885 smp_mb__after_atomic();
1886 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1887}
1888
78746a38
FI		 1889static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1890 struct nfs_server *server)
1891{
cf6605d1 1892 if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
78746a38
FI		 1893 struct nfs_client *clp = server->nfs_client;
1894
1895 /* The lo must be on the clp list if there is any
1896 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1897 */
1898 spin_lock(&clp->cl_lock);
cf6605d1 1899 list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
78746a38
FI		 1900 spin_unlock(&clp->cl_lock);
1901 }
1902}
1903
e5e94017
BH		 1904/*
1905 * Layout segment is retreived from the server if not cached.
1906 * The appropriate layout segment is referenced and returned to the caller.
1907 */
7c24d948 1908struct pnfs_layout_segment *
e5e94017
BH		 1909pnfs_update_layout(struct inode *ino,
1910 struct nfs_open_context *ctx,
fb3296eb
BH		 1911 loff_t pos,
1912 u64 count,
a75b9df9 1913 enum pnfs_iomode iomode,
c7d73af2 1914 bool strict_iomode,
a75b9df9 1915 gfp_t gfp_flags)
e5e94017 1916{
fb3296eb
BH		 1917 struct pnfs_layout_range arg = {
1918 .iomode = iomode,
1919 .offset = pos,
1920 .length = count,
1921 };
70d2f7b1 1922 unsigned pg_offset;
6382a441
WAA		 1923 struct nfs_server *server = NFS_SERVER(ino);
1924 struct nfs_client *clp = server->nfs_client;
183d9e7b 1925 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1926 struct pnfs_layout_segment *lseg = NULL;
587f03de 1927 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1928 nfs4_stateid stateid;
1929 long timeout = 0;
66b53f32 1930 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1931 bool first;
e5e94017 1932
9a4bf31d 1933 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1934 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1935 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1936 goto out;
9a4bf31d 1937 }
d23d61c8 1938
9a4bf31d 1939 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1940 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1941 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1942 goto out;
9a4bf31d 1943 }
d23d61c8 1944
9bf87482 1945lookup_again:
d03360aa
TM		 1946 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1947 if (IS_ERR(lseg))
1948 goto out;
9bf87482 1949 first = false;
e5e94017 1950 spin_lock(&ino->i_lock);
9fa40758 1951 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1952 if (lo == NULL) {
1953 spin_unlock(&ino->i_lock);
183d9e7b 1954 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1955 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1956 goto out;
1957 }
e5e94017 1958
43f1b3da 1959 /* Do we even need to bother with this? */
a59c30ac 1960 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1961 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1962 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1963 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1964 goto out_unlock;
1965 }
1966
1967 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1968 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1969 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1970 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1971 goto out_unlock;
9a4bf31d 1972 }
e5e94017 1973
411ae722
TM		 1974 /*
1975 * If the layout segment list is empty, but there are outstanding
1976 * layoutget calls, then they might be subject to a layoutrecall.
1977 */
1978 if (list_empty(&lo->plh_segs) &&
1979 atomic_read(&lo->plh_outstanding) != 0) {
1980 spin_unlock(&ino->i_lock);
d03360aa 1981 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
400417b0 1982 !atomic_read(&lo->plh_outstanding)));
58bbeab4 1983 if (IS_ERR(lseg))
411ae722
TM		 1984 goto out_put_layout_hdr;
1985 pnfs_put_layout_hdr(lo);
1986 goto lookup_again;
1987 }
1988
c7d73af2 1989 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1990 if (lseg) {
1991 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1992 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1993 goto out_unlock;
1994 }
1995
183d9e7b
JL		 1996 /*
1997 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1998 * stateid, or it has been invalidated, then we must use the open
1999 * stateid.
2000 */
67a3b721 2001 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
d9aba2b4 2002 int status;
183d9e7b
JL		 2003
2004 /*
2005 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 2006 * RFC 5661 Errata 3208.
2007 */
2008 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
2009 &lo->plh_flags)) {
2010 spin_unlock(&ino->i_lock);
d03360aa
TM		 2011 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
2012 NFS_LAYOUT_FIRST_LAYOUTGET,
2013 TASK_KILLABLE));
2014 if (IS_ERR(lseg))
2015 goto out_put_layout_hdr;
9bf87482 2016 pnfs_put_layout_hdr(lo);
183d9e7b 2017 dprintk("%s retrying\n", __func__);
9bf87482
PT		 2018 goto lookup_again;
2019 }
183d9e7b 2020
fbf4bcc9 2021 spin_unlock(&ino->i_lock);
183d9e7b 2022 first = true;
d9aba2b4 2023 status = nfs4_select_rw_stateid(ctx->state,
70d2f7b1 2024 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
d9aba2b4
TM		 2025 NULL, &stateid, NULL);
2026 if (status != 0) {
731c74dd 2027 lseg = ERR_PTR(status);
70d2f7b1
TM		 2028 trace_pnfs_update_layout(ino, pos, count,
2029 iomode, lo, lseg,
2030 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
d9aba2b4
TM		 2031 nfs4_schedule_stateid_recovery(server, ctx->state);
2032 pnfs_clear_first_layoutget(lo);
2033 pnfs_put_layout_hdr(lo);
2034 goto lookup_again;
70d2f7b1 2035 }
fbf4bcc9 2036 spin_lock(&ino->i_lock);
9bf87482 2037 } else {
183d9e7b 2038 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 2039 }
568e8c49 2040
aa8a45ee
PT		 2041 /*
2042 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
2043 * for LAYOUTRETURN even if first is true.
2044 */
8f70f53a 2045 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 2046 spin_unlock(&ino->i_lock);
2047 dprintk("%s wait for layoutreturn\n", __func__);
d03360aa
TM		 2048 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2049 if (!IS_ERR(lseg)) {
d67ae825
TH		 2050 if (first)
2051 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 2052 pnfs_put_layout_hdr(lo);
2053 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 2054 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2055 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 2056 goto lookup_again;
2057 }
183d9e7b 2058 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2059 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 2060 goto out_put_layout_hdr;
2061 }
2062
9a4bf31d 2063 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 2064 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2065 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 2066 goto out_unlock;
9a4bf31d 2067 }
411ae722 2068 nfs_layoutget_begin(lo);
f49f9baa 2069 spin_unlock(&ino->i_lock);
30005121 2070
78746a38 2071 _add_to_server_list(lo, server);
e5e94017 2072
09cbfeaf 2073 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 2074 if (pg_offset) {
2075 arg.offset -= pg_offset;
2076 arg.length += pg_offset;
2077 }
7c24d948 2078 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 2079 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 2080
5e36e2a9 2081 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 2082 if (!lgp) {
2083 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2084 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 2085 nfs_layoutget_end(lo);
587f03de
FI		 2086 goto out_put_layout_hdr;
2087 }
2088
dacb452d 2089 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 2090 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2091 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 2092 nfs_layoutget_end(lo);
83026d80 2093 if (IS_ERR(lseg)) {
183d9e7b 2094 switch(PTR_ERR(lseg)) {
e85d7ee4 2095 case -EBUSY:
183d9e7b
JL		 2096 if (time_after(jiffies, giveup))
2097 lseg = NULL;
66b53f32
TM		 2098 break;
2099 case -ERECALLCONFLICT:
183d9e7b 2100 case -EAGAIN:
56b38a1f 2101 break;
183d9e7b
JL		 2102 default:
2103 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2104 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2105 lseg = NULL;
2106 }
56b38a1f
TM		 2107 goto out_put_layout_hdr;
2108 }
2109 if (lseg) {
2110 if (first)
2111 pnfs_clear_first_layoutget(lo);
2112 trace_pnfs_update_layout(ino, pos, count,
2113 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2114 pnfs_put_layout_hdr(lo);
2115 goto lookup_again;
83026d80
JL		 2116 }
2117 } else {
2118 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2119 }
2120
830ffb56 2121out_put_layout_hdr:
d67ae825
TH		 2122 if (first)
2123 pnfs_clear_first_layoutget(lo);
d5b9216f
TM		 2124 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2125 PNFS_UPDATE_LAYOUT_EXIT);
70c3bd2b 2126 pnfs_put_layout_hdr(lo);
e5e94017 2127out:
f86bbcf8
TM		 2128 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2129 "(%s, offset: %llu, length: %llu)\n",
2130 __func__, ino->i_sb->s_id,
2131 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2132 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2133 iomode==IOMODE_RW ? "read/write" : "read-only",
2134 (unsigned long long)pos,
2135 (unsigned long long)count);
e5e94017
BH		 2136 return lseg;
2137out_unlock:
2138 spin_unlock(&ino->i_lock);
830ffb56 2139 goto out_put_layout_hdr;
e5e94017 2140}
7c24d948 2141EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2142
540d9864
TM		 2143static bool
2144pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2145{
2146 switch (range->iomode) {
2147 case IOMODE_READ:
2148 case IOMODE_RW:
2149 break;
2150 default:
2151 return false;
2152 }
2153 if (range->offset == NFS4_MAX_UINT64)
2154 return false;
2155 if (range->length == 0)
2156 return false;
2157 if (range->length != NFS4_MAX_UINT64 &&
2158 range->length > NFS4_MAX_UINT64 - range->offset)
2159 return false;
2160 return true;
2161}
2162
78746a38
FI		 2163static struct pnfs_layout_hdr *
2164_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2165{
2166 struct pnfs_layout_hdr *lo;
2167
2168 spin_lock(&ino->i_lock);
2169 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2170 if (!lo)
2171 goto out_unlock;
2172 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2173 goto out_unlock;
2174 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2175 goto out_unlock;
2176 if (pnfs_layoutgets_blocked(lo))
2177 goto out_unlock;
2178 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2179 goto out_unlock;
411ae722 2180 nfs_layoutget_begin(lo);
78746a38
FI		 2181 spin_unlock(&ino->i_lock);
2182 _add_to_server_list(lo, NFS_SERVER(ino));
2183 return lo;
2184
2185out_unlock:
2186 spin_unlock(&ino->i_lock);
2187 pnfs_put_layout_hdr(lo);
2188 return NULL;
2189}
2190
2409a976
FI		 2191static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2192 struct nfs_open_context *ctx)
2193{
78746a38
FI		 2194 struct inode *ino = data->dentry->d_inode;
2195 struct pnfs_layout_range rng = {
2196 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2197 IOMODE_RW: IOMODE_READ,
2198 .offset = 0,
2199 .length = NFS4_MAX_UINT64,
2200 };
2201 struct nfs4_layoutget *lgp;
2202 struct pnfs_layout_hdr *lo;
2203
64294b08
TM		 2204 /* Heuristic: don't send layoutget if we have cached data */
2205 if (rng.iomode == IOMODE_READ &&
2206 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2207 return;
2208
78746a38
FI		 2209 lo = _pnfs_grab_empty_layout(ino, ctx);
2210 if (!lo)
2211 return;
2212 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2213 &rng, GFP_KERNEL);
2214 if (!lgp) {
2215 pnfs_clear_first_layoutget(lo);
2216 pnfs_put_layout_hdr(lo);
2217 return;
2218 }
2219 data->lgp = lgp;
2220 data->o_arg.lg_args = &lgp->args;
2221 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2222}
2223
2224static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2225 struct nfs_open_context *ctx)
2226{
2227 struct pnfs_layout_range rng = {
2228 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2229 IOMODE_RW: IOMODE_READ,
2230 .offset = 0,
2231 .length = NFS4_MAX_UINT64,
2232 };
2233 struct nfs4_layoutget *lgp;
2234
2235 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2236 &rng, GFP_KERNEL);
2237 if (!lgp)
2238 return;
2239 data->lgp = lgp;
2240 data->o_arg.lg_args = &lgp->args;
2241 data->o_res.lg_res = &lgp->res;
2242}
2243
2244void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2245 struct nfs_open_context *ctx)
2246{
2247 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2248
2249 if (!(pnfs_enabled_sb(server) &&
2250 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2251 return;
2252 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2253 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2254 return;
2409a976
FI		 2255 if (data->state)
2256 _lgopen_prepare_attached(data, ctx);
2257 else
2258 _lgopen_prepare_floating(data, ctx);
2259}
2260
2261void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2262 struct nfs_open_context *ctx)
2263{
2264 struct pnfs_layout_hdr *lo;
2265 struct pnfs_layout_segment *lseg;
c49b5209 2266 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2267 u32 iomode;
2268
6e01260c 2269 if (!lgp)
2409a976 2270 return;
6e01260c
FI		 2271 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2272 if (lgp->res.status) {
2273 switch (lgp->res.status) {
6e01260c 2274 default:
8dc96566
TM		 2275 break;
2276 /*
2277 * Halt lgopen attempts if the server doesn't recognise
2278 * the "current stateid" value, the layout type, or the
2279 * layoutget operation as being valid.
2280 * Also if it complains about too many ops in the compound
2281 * or of the request/reply being too big.
2282 */
2283 case -NFS4ERR_BAD_STATEID:
2284 case -NFS4ERR_NOTSUPP:
2285 case -NFS4ERR_REP_TOO_BIG:
2286 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2287 case -NFS4ERR_REQ_TOO_BIG:
2288 case -NFS4ERR_TOO_MANY_OPS:
2289 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2290 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2291 }
2292 return;
2293 }
2409a976 2294 if (!lgp->args.inode) {
78746a38
FI		 2295 lo = _pnfs_grab_empty_layout(ino, ctx);
2296 if (!lo)
2297 return;
2409a976
FI		 2298 lgp->args.inode = ino;
2299 } else
2300 lo = NFS_I(lgp->args.inode)->layout;
2409a976
FI		 2301
2302 lseg = pnfs_layout_process(lgp);
32f1c28f 2303 if (!IS_ERR(lseg)) {
2409a976
FI		 2304 iomode = lgp->args.range.iomode;
2305 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2306 pnfs_put_lseg(lseg);
2307 }
30ae2412
FI		 2308}
2309
2310void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2311{
2312 if (lgp != NULL) {
2313 struct inode *inode = lgp->args.inode;
2314 if (inode) {
2315 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
30ae2412 2316 pnfs_clear_first_layoutget(lo);
411ae722 2317 nfs_layoutget_end(lo);
30ae2412
FI		 2318 }
2319 pnfs_layoutget_free(lgp);
2320 }
2409a976
FI		 2321}
2322
a0b0a6e3 2323struct pnfs_layout_segment *
b1f69b75
AA		 2324pnfs_layout_process(struct nfs4_layoutget *lgp)
2325{
2326 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2327 struct nfs4_layoutget_res *res = &lgp->res;
2328 struct pnfs_layout_segment *lseg;
b7edfaa1 2329 struct inode *ino = lo->plh_inode;
78096cca 2330 LIST_HEAD(free_me);
540d9864
TM		 2331
2332 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2333 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2334
2335 /* Inject layout blob into I/O device driver */
a75b9df9 2336 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2337 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2338 if (!lseg)
1b3c6d07
JL		 2339 lseg = ERR_PTR(-ENOMEM);
2340
2341 dprintk("%s: Could not allocate layout: error %ld\n",
2342 __func__, PTR_ERR(lseg));
2343 return lseg;
b1f69b75
AA		 2344 }
2345
119cef97 2346 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2347
b1f69b75 2348 spin_lock(&ino->i_lock);
e1c06f80 2349 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2350 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2351 goto out_forget;
43f1b3da 2352 }
038d6493 2353
9888d837
TM		 2354 if (!pnfs_layout_is_valid(lo)) {
2355 /* We have a completely new layout */
59b56394 2356 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
9888d837 2357 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2358 /* existing state ID, make sure the sequence number matches. */
2359 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2360 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2361 goto out_forget;
362f7474 2362 }
59b56394 2363 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
362f7474
CH		 2364 } else {
2365 /*
2366 * We got an entirely new state ID. Mark all segments for the
9888d837 2367 * inode invalid, and retry the layoutget
362f7474 2368 */
d9b61708 2369 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 2370 goto out_forget;
362f7474 2371 }
038d6493 2372
9369a431 2373 pnfs_get_lseg(lseg);
03772d2f 2374 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2375
b1f69b75 2376
3976143b 2377 if (res->return_on_close)
f7e8917a 2378 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2379
b1f69b75 2380 spin_unlock(&ino->i_lock);
78096cca 2381 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2382 return lseg;
43f1b3da 2383
1b3c6d07 2384out_forget:
43f1b3da
FI		 2385 spin_unlock(&ino->i_lock);
2386 lseg->pls_layout = lo;
2387 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2388 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2389}
2390
2f215968
TM		 2391/**
2392 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2393 * @lo: pointer to layout header
2394 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2395 * @return_range: describe layout segment ranges to be returned
e0b7d420 2396 * @seq: stateid seqid to match
2f215968
TM		 2397 *
2398 * This function is mainly intended for use by layoutrecall. It attempts
2399 * to free the layout segment immediately, or else to mark it for return
2400 * as soon as its reference count drops to zero.
e0b7d420
TM		 2401 *
2402 * Returns
2403 * - 0: a layoutreturn needs to be scheduled.
2404 * - EBUSY: there are layout segment that are still in use.
2405 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2406 */
10335556 2407int
016256df
PT		 2408pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2409 struct list_head *tmp_list,
6d597e17
JL		 2410 const struct pnfs_layout_range *return_range,
2411 u32 seq)
016256df
PT		 2412{
2413 struct pnfs_layout_segment *lseg, *next;
10335556 2414 int remaining = 0;
016256df
PT		 2415
2416 dprintk("%s:Begin lo %p\n", __func__, lo);
2417
fc7ff367 2418 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2419
2420 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2421 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2422 dprintk("%s: marking lseg %p iomode %d "
2423 "offset %llu length %llu\n", __func__,
2424 lseg, lseg->pls_range.iomode,
2425 lseg->pls_range.offset,
2426 lseg->pls_range.length);
ff041727 2427 if (mark_lseg_invalid(lseg, tmp_list))
2f215968
TM		 2428 continue;
2429 remaining++;
016256df 2430 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2431 }
6d597e17 2432
e0b7d420 2433 if (remaining) {
6d597e17 2434 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2435 return -EBUSY;
2436 }
6d597e17 2437
e0b7d420
TM		 2438 if (!list_empty(&lo->plh_return_segs)) {
2439 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2440 return 0;
2441 }
2442
2443 return -ENOENT;
016256df
PT		 2444}
2445
b5fdf841
TM		 2446static void
2447pnfs_mark_layout_for_return(struct inode *inode,
2448 const struct pnfs_layout_range *range)
016256df 2449{
b5fdf841 2450 struct pnfs_layout_hdr *lo;
10335556 2451 bool return_now = false;
016256df
PT		 2452
2453 spin_lock(&inode->i_lock);
b5fdf841 2454 lo = NFS_I(inode)->layout;
bdebfccd
TM		 2455 if (!pnfs_layout_is_valid(lo)) {
2456 spin_unlock(&inode->i_lock);
2457 return;
2458 }
b5fdf841 2459 pnfs_set_plh_return_info(lo, range->iomode, 0);
016256df
PT		 2460 /*
2461 * mark all matching lsegs so that we are sure to have no live
2462 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2463 * for how it works.
2464 */
b5fdf841 2465 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
44ea8dfc 2466 const struct cred *cred;
10335556 2467 nfs4_stateid stateid;
e5fd1904 2468 enum pnfs_iomode iomode;
10335556 2469
44ea8dfc 2470 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
10335556
TM		 2471 spin_unlock(&inode->i_lock);
2472 if (return_now)
44ea8dfc 2473 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
10335556
TM		 2474 } else {
2475 spin_unlock(&inode->i_lock);
2476 nfs_commit_inode(inode, 0);
2477 }
016256df 2478}
b5fdf841
TM		 2479
2480void pnfs_error_mark_layout_for_return(struct inode *inode,
2481 struct pnfs_layout_segment *lseg)
2482{
2483 struct pnfs_layout_range range = {
2484 .iomode = lseg->pls_range.iomode,
2485 .offset = 0,
2486 .length = NFS4_MAX_UINT64,
2487 };
2488
2489 pnfs_mark_layout_for_return(inode, &range);
2490}
016256df
PT		 2491EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2492
b5fdf841
TM		 2493static bool
2494pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
2495{
2496 return pnfs_layout_is_valid(lo) &&
2497 !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
2498 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
2499}
2500
2501static struct pnfs_layout_segment *
2502pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
2503 const struct pnfs_layout_range *range,
2504 enum pnfs_iomode iomode)
2505{
2506 struct pnfs_layout_segment *lseg;
2507
2508 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
2509 if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
2510 continue;
2511 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2512 continue;
2513 if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
2514 continue;
2515 if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
2516 return lseg;
2517 }
2518 return NULL;
2519}
2520
2521/* Find open file states whose mode matches that of the range */
2522static bool
2523pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
2524 const struct pnfs_layout_range *range)
2525{
2526 struct list_head *head;
2527 struct nfs_open_context *ctx;
2528 fmode_t mode = 0;
2529
2530 if (!pnfs_layout_can_be_returned(lo) ||
2531 !pnfs_find_first_lseg(lo, range, range->iomode))
2532 return false;
2533
2534 head = &NFS_I(lo->plh_inode)->open_files;
2535 list_for_each_entry_rcu(ctx, head, list) {
2536 if (ctx->state)
2537 mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
2538 }
2539
2540 switch (range->iomode) {
2541 default:
2542 break;
2543 case IOMODE_READ:
2544 mode &= ~FMODE_WRITE;
2545 break;
2546 case IOMODE_RW:
2547 if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
2548 mode &= ~FMODE_READ;
2549 }
2550 return mode == 0;
2551}
2552
2553static int
2554pnfs_layout_return_unused_byserver(struct nfs_server *server, void *data)
2555{
2556 const struct pnfs_layout_range *range = data;
2557 struct pnfs_layout_hdr *lo;
2558 struct inode *inode;
2559restart:
2560 rcu_read_lock();
2561 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
2562 if (!pnfs_layout_can_be_returned(lo) ||
2563 test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2564 continue;
2565 inode = lo->plh_inode;
2566 spin_lock(&inode->i_lock);
2567 if (!pnfs_should_return_unused_layout(lo, range)) {
2568 spin_unlock(&inode->i_lock);
2569 continue;
2570 }
2571 spin_unlock(&inode->i_lock);
2572 inode = pnfs_grab_inode_layout_hdr(lo);
2573 if (!inode)
2574 continue;
2575 rcu_read_unlock();
2576 pnfs_mark_layout_for_return(inode, range);
2577 iput(inode);
2578 cond_resched();
2579 goto restart;
2580 }
2581 rcu_read_unlock();
2582 return 0;
2583}
2584
2585void
2586pnfs_layout_return_unused_byclid(struct nfs_client *clp,
2587 enum pnfs_iomode iomode)
2588{
2589 struct pnfs_layout_range range = {
2590 .iomode = iomode,
2591 .offset = 0,
2592 .length = NFS4_MAX_UINT64,
2593 };
2594
2595 nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
2596 &range);
2597}
2598
b3230e80
TM		 2599void
2600pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2601{
2602 if (pgio->pg_lseg == NULL ||
2603 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2604 return;
2605 pnfs_put_lseg(pgio->pg_lseg);
2606 pgio->pg_lseg = NULL;
2607}
2608EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2609
08cb5b0f
BC		 2610/*
2611 * Check for any intersection between the request and the pgio->pg_lseg,
2612 * and if none, put this pgio->pg_lseg away.
2613 */
e1e54ab7 2614void
08cb5b0f
BC		 2615pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2616{
2617 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2618 pnfs_put_lseg(pgio->pg_lseg);
2619 pgio->pg_lseg = NULL;
2620 }
2621}
e1e54ab7 2622EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
08cb5b0f 2623
d8007d4d
TM		 2624void
2625pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2626{
1fd937bd
PT		 2627 u64 rd_size = req->wb_bytes;
2628
b3230e80 2629 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2630 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2631 if (pgio->pg_lseg == NULL) {
2632 if (pgio->pg_dreq == NULL)
2633 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2634 else
2635 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2636
2637 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2638 nfs_req_openctx(req),
cb5d04bc
PT		 2639 req_offset(req),
2640 rd_size,
2641 IOMODE_READ,
c7d73af2 2642 false,
cb5d04bc 2643 GFP_KERNEL);
d600ad1f
PT		 2644 if (IS_ERR(pgio->pg_lseg)) {
2645 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2646 pgio->pg_lseg = NULL;
2647 return;
2648 }
cb5d04bc 2649 }
e885de1a
TM		 2650 /* If no lseg, fall back to read through mds */
2651 if (pgio->pg_lseg == NULL)
1f945357 2652 nfs_pageio_reset_read_mds(pgio);
e885de1a 2653
d8007d4d
TM		 2654}
2655EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2656
2657void
6296556f
PT		 2658pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2659 struct nfs_page *req, u64 wb_size)
d8007d4d 2660{
b3230e80 2661 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2662 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2663 if (pgio->pg_lseg == NULL) {
cb5d04bc 2664 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2665 nfs_req_openctx(req),
cb5d04bc
PT		 2666 req_offset(req),
2667 wb_size,
2668 IOMODE_RW,
c7d73af2 2669 false,
2b17d725 2670 GFP_KERNEL);
d600ad1f
PT		 2671 if (IS_ERR(pgio->pg_lseg)) {
2672 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2673 pgio->pg_lseg = NULL;
2674 return;
2675 }
2676 }
e885de1a
TM		 2677 /* If no lseg, fall back to write through mds */
2678 if (pgio->pg_lseg == NULL)
1f945357 2679 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2680}
2681EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2682
180bb5ec
WAA		 2683void
2684pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2685{
2686 if (desc->pg_lseg) {
2687 pnfs_put_lseg(desc->pg_lseg);
2688 desc->pg_lseg = NULL;
2689 }
2690}
2691EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2692
b4fdac1a
WAA		 2693/*
2694 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2695 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2696 */
2697size_t
a7d42ddb
WAA		 2698pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2699 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2700{
0f9c429e 2701 unsigned int size;
c5e20cb7 2702 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2703
2704 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2705 if (!size)
2706 return 0;
94ad1c80 2707
19982ba8 2708 /*
c5e20cb7
WAA		 2709 * 'size' contains the number of bytes left in the current page (up
2710 * to the original size asked for in @req->wb_bytes).
2711 *
2712 * Calculate how many bytes are left in the layout segment
2713 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2714 *
2715 * Please also note that 'end_offset' is actually the offset of the
2716 * first byte that lies outside the pnfs_layout_range. FIXME?
2717 *
2718 */
19b54848 2719 if (pgio->pg_lseg) {
17822b20 2720 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2721 pgio->pg_lseg->pls_range.length);
2722 req_start = req_offset(req);
08cb5b0f 2723
c5e20cb7 2724 /* start of request is past the last byte of this segment */
08cb5b0f 2725 if (req_start >= seg_end)
19b54848 2726 return 0;
c5e20cb7
WAA		 2727
2728 /* adjust 'size' iff there are fewer bytes left in the
2729 * segment than what nfs_generic_pg_test returned */
2730 seg_left = seg_end - req_start;
2731 if (seg_left < size)
2732 size = (unsigned int)seg_left;
19b54848 2733 }
0f9c429e 2734
19b54848 2735 return size;
94ad1c80 2736}
89a58e32 2737EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2738
53113ad3 2739int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2740{
2741 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2742
2743 /* Resend all requests through the MDS */
53113ad3
WAA		 2744 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2745 hdr->completion_ops);
c7070113 2746 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2747 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2748}
e7dd79af 2749EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2750
d45f60c6 2751static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2752{
cd841605
FI		 2753
2754 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2755 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2756 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2757 pnfs_return_layout(hdr->inode);
1acbbb4e 2758 }
6c75dc0d 2759 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2760 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2761}
2762
d20581aa
BH		 2763/*
2764 * Called by non rpc-based layout drivers
2765 */
d45f60c6 2766void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2767{
f8417b48 2768 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2769 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2770 hdr->mds_offset + hdr->res.count);
d45f60c6 2771 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2772 }
2773 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2774 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2775 pnfs_ld_handle_write_error(hdr);
2776 hdr->mds_ops->rpc_release(hdr);
44b83799 2777}
d20581aa 2778EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2779
dce81290
TM		 2780static void
2781pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2782 struct nfs_pgio_header *hdr)
dce81290 2783{
48d635f1 2784 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2785
6c75dc0d 2786 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2787 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2788 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2789 mirror->pg_recoalesce = 1;
6c75dc0d 2790 }
ba4a76f7 2791 hdr->completion_ops->completion(hdr);
dce81290
TM		 2792}
2793
2794static enum pnfs_try_status
d45f60c6 2795pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2796 const struct rpc_call_ops *call_ops,
2797 struct pnfs_layout_segment *lseg,
2798 int how)
0382b744 2799{
cd841605 2800 struct inode *inode = hdr->inode;
0382b744
AA		 2801 enum pnfs_try_status trypnfs;
2802 struct nfs_server *nfss = NFS_SERVER(inode);
2803
cd841605 2804 hdr->mds_ops = call_ops;
0382b744
AA		 2805
2806 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2807 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2808 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2809 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2810 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2811 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2812 return trypnfs;
2813}
2814
dce81290 2815static void
7f714720
WAA		 2816pnfs_do_write(struct nfs_pageio_descriptor *desc,
2817 struct nfs_pgio_header *hdr, int how)
dce81290 2818{
dce81290
TM		 2819 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2820 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2821 enum pnfs_try_status trypnfs;
dce81290 2822
d45f60c6 2823 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2824 switch (trypnfs) {
2825 case PNFS_NOT_ATTEMPTED:
d45f60c6 2826 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2827 case PNFS_ATTEMPTED:
2828 break;
2829 case PNFS_TRY_AGAIN:
2830 /* cleanup hdr and prepare to redo pnfs */
2831 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2832 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2833 list_splice_init(&hdr->pages, &mirror->pg_list);
2834 mirror->pg_recoalesce = 1;
2835 }
2836 hdr->mds_ops->rpc_release(hdr);
2837 }
dce81290
TM		 2838}
2839
6c75dc0d
FI		 2840static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2841{
9369a431 2842 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2843 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2844}
2845
dce81290
TM		 2846int
2847pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2848{
6c75dc0d 2849 struct nfs_pgio_header *hdr;
dce81290
TM		 2850 int ret;
2851
1e7f3a48
WAA		 2852 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2853 if (!hdr) {
2bff2288
PT		 2854 desc->pg_error = -ENOMEM;
2855 return desc->pg_error;
dce81290 2856 }
6c75dc0d 2857 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2858
9369a431 2859 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2860 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2861 if (!ret)
7f714720 2862 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2863
6c75dc0d 2864 return ret;
dce81290
TM		 2865}
2866EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2867
53113ad3 2868int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2869{
2870 struct nfs_pageio_descriptor pgio;
2871
1acbbb4e 2872 /* Resend all requests through the MDS */
53113ad3
WAA		 2873 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2874 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2875}
e7dd79af 2876EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2877
d45f60c6 2878static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2879{
cd841605
FI		 2880 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2881 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2882 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2883 pnfs_return_layout(hdr->inode);
1acbbb4e 2884 }
4db6e0b7 2885 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2886 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2887}
2888
d20581aa
BH		 2889/*
2890 * Called by non rpc-based layout drivers
2891 */
d45f60c6 2892void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2893{
bfc505de 2894 if (likely(!hdr->pnfs_error))
d45f60c6 2895 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2896 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2897 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2898 pnfs_ld_handle_read_error(hdr);
2899 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2900}
2901EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2902
493292dd
TM		 2903static void
2904pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2905 struct nfs_pgio_header *hdr)
493292dd 2906{
48d635f1 2907 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2908
4db6e0b7 2909 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2910 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2911 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2912 mirror->pg_recoalesce = 1;
4db6e0b7 2913 }
ba4a76f7 2914 hdr->completion_ops->completion(hdr);
493292dd
TM		 2915}
2916
64419a9b
AA		 2917/*
2918 * Call the appropriate parallel I/O subsystem read function.
2919 */
493292dd 2920static enum pnfs_try_status
d45f60c6 2921pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2922 const struct rpc_call_ops *call_ops,
2923 struct pnfs_layout_segment *lseg)
64419a9b 2924{
cd841605 2925 struct inode *inode = hdr->inode;
64419a9b
AA		 2926 struct nfs_server *nfss = NFS_SERVER(inode);
2927 enum pnfs_try_status trypnfs;
2928
cd841605 2929 hdr->mds_ops = call_ops;
64419a9b
AA		 2930
2931 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2932 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2933
d45f60c6 2934 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2935 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2936 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2937 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2938 return trypnfs;
2939}
863a3c6c 2940
ceb11e13 2941/* Resend all requests through pnfs. */
563c53e7
TM		 2942void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr,
2943 unsigned int mirror_idx)
ceb11e13
PT		 2944{
2945 struct nfs_pageio_descriptor pgio;
2946
1b1bc66b 2947 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2948 /* Prevent deadlocks with layoutreturn! */
2949 pnfs_put_lseg(hdr->lseg);
2950 hdr->lseg = NULL;
2951
1b1bc66b
WAA		 2952 nfs_pageio_init_read(&pgio, hdr->inode, false,
2953 hdr->completion_ops);
563c53e7 2954 pgio.pg_mirror_idx = mirror_idx;
1b1bc66b
WAA		 2955 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2956 }
ceb11e13
PT		 2957}
2958EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2959
493292dd 2960static void
7f714720 2961pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2962{
493292dd
TM		 2963 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2964 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2965 enum pnfs_try_status trypnfs;
493292dd 2966
d45f60c6 2967 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2968 switch (trypnfs) {
2969 case PNFS_NOT_ATTEMPTED:
d45f60c6 2970 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2971 case PNFS_ATTEMPTED:
2972 break;
2973 case PNFS_TRY_AGAIN:
2974 /* cleanup hdr and prepare to redo pnfs */
2975 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2976 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2977 list_splice_init(&hdr->pages, &mirror->pg_list);
2978 mirror->pg_recoalesce = 1;
2979 }
2980 hdr->mds_ops->rpc_release(hdr);
2981 }
493292dd
TM		 2982}
2983
4db6e0b7
FI		 2984static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2985{
9369a431 2986 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2987 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2988}
2989
493292dd
TM		 2990int
2991pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2992{
4db6e0b7 2993 struct nfs_pgio_header *hdr;
493292dd
TM		 2994 int ret;
2995
1e7f3a48
WAA		 2996 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2997 if (!hdr) {
2bff2288
PT		 2998 desc->pg_error = -ENOMEM;
2999 return desc->pg_error;
493292dd 3000 }
4db6e0b7 3001 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 3002 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 3003 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 3004 if (!ret)
7f714720 3005 pnfs_do_read(desc, hdr);
4db6e0b7 3006 return ret;
493292dd
TM		 3007}
3008EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
3009
71244d9b
TM		 3010static void pnfs_clear_layoutcommitting(struct inode *inode)
3011{
3012 unsigned long *bitlock = &NFS_I(inode)->flags;
3013
3014 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 3015 smp_mb__after_atomic();
71244d9b
TM		 3016 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
3017}
3018
863a3c6c 3019/*
a9bae566 3020 * There can be multiple RW segments.
863a3c6c 3021 */
a9bae566 3022static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 3023{
a9bae566 3024 struct pnfs_layout_segment *lseg;
863a3c6c 3025
a9bae566
PT		 3026 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
3027 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 3028 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 3029 list_add(&lseg->pls_lc_list, listp);
3030 }
863a3c6c
AA		 3031}
3032
a073dbff
TM		 3033static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
3034{
3035 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 3036
3037 /* Matched by references in pnfs_set_layoutcommit */
3038 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
3039 list_del_init(&lseg->pls_lc_list);
3040 pnfs_put_lseg(lseg);
3041 }
3042
71244d9b 3043 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 3044}
3045
1b0ae068
PT		 3046void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
3047{
b9e028fd 3048 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 3049}
3050EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
3051
863a3c6c 3052void
67af7611
TM		 3053pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
3054 loff_t end_pos)
863a3c6c 3055{
cd841605 3056 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 3057 bool mark_as_dirty = false;
863a3c6c 3058
cd841605 3059 spin_lock(&inode->i_lock);
863a3c6c 3060 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 3061 nfsi->layout->plh_lwb = end_pos;
79a48a1f 3062 mark_as_dirty = true;
863a3c6c 3063 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 3064 __func__, inode->i_ino);
29559b11
TM		 3065 } else if (end_pos > nfsi->layout->plh_lwb)
3066 nfsi->layout->plh_lwb = end_pos;
67af7611 3067 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 3068 /* references matched in nfs4_layoutcommit_release */
67af7611 3069 pnfs_get_lseg(lseg);
a9bae566 3070 }
cd841605 3071 spin_unlock(&inode->i_lock);
acff5880 3072 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 3073 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 3074
3075 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3076 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3077 if (mark_as_dirty)
cd841605 3078 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3079}
3080EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
3081
db29c089
AA		 3082void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
3083{
3084 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
3085
3086 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
3087 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 3088 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 3089}
3090
de4b15c7
AA		 3091/*
3092 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3093 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3094 * data to disk to allow the server to recover the data if it crashes.
3095 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3096 * is off, and a COMMIT is sent to a data server, or
3097 * if WRITEs to a data server return NFS_DATA_SYNC.
3098 */
863a3c6c 3099int
ef311537 3100pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 3101{
5f919c9f 3102 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 3103 struct nfs4_layoutcommit_data *data;
3104 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 3105 loff_t end_pos;
71244d9b 3106 int status;
863a3c6c 3107
71244d9b 3108 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 3109 return 0;
3110
71244d9b 3111 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 3112
71244d9b 3113 status = -EAGAIN;
92407e75 3114 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 3115 if (!sync)
3116 goto out;
74316201 3117 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 3118 NFS_INO_LAYOUTCOMMITTING,
3119 nfs_wait_bit_killable,
3120 TASK_KILLABLE);
92407e75 3121 if (status)
71244d9b 3122 goto out;
92407e75
PT		 3123 }
3124
71244d9b
TM		 3125 status = -ENOMEM;
3126 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
3127 data = kzalloc(sizeof(*data), GFP_NOFS);
3128 if (!data)
3129 goto clear_layoutcommitting;
3130
3131 status = 0;
de4b15c7 3132 spin_lock(&inode->i_lock);
71244d9b
TM		 3133 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
3134 goto out_unlock;
a9bae566 3135
71244d9b 3136 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 3137 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 3138
acff5880 3139 end_pos = nfsi->layout->plh_lwb;
863a3c6c 3140
f597c537 3141 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
97a728f5 3142 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 3143 spin_unlock(&inode->i_lock);
3144
3145 data->args.inode = inode;
863a3c6c
AA		 3146 nfs_fattr_init(&data->fattr);
3147 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3148 data->res.fattr = &data->fattr;
2e18d4d8
TM		 3149 if (end_pos != 0)
3150 data->args.lastbytewritten = end_pos - 1;
3151 else
3152 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 3153 data->res.server = NFS_SERVER(inode);
3154
5f919c9f
CH		 3155 if (ld->prepare_layoutcommit) {
3156 status = ld->prepare_layoutcommit(&data->args);
3157 if (status) {
a52458b4 3158 put_cred(data->cred);
5f919c9f 3159 spin_lock(&inode->i_lock);
29559b11
TM		 3160 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3161 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 3162 nfsi->layout->plh_lwb = end_pos;
3471648a 3163 goto out_unlock;
5f919c9f
CH		 3164 }
3165 }
3166
3167
863a3c6c
AA		 3168 status = nfs4_proc_layoutcommit(data, sync);
3169out:
92407e75
PT		 3170 if (status)
3171 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3172 dprintk("<-- %s status %d\n", __func__, status);
3173 return status;
71244d9b
TM		 3174out_unlock:
3175 spin_unlock(&inode->i_lock);
92407e75 3176 kfree(data);
71244d9b
TM		 3177clear_layoutcommitting:
3178 pnfs_clear_layoutcommitting(inode);
92407e75 3179 goto out;
863a3c6c 3180}
72cff449 3181EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 3182
5bb89b47
TM		 3183int
3184pnfs_generic_sync(struct inode *inode, bool datasync)
3185{
3186 return pnfs_layoutcommit_inode(inode, true);
3187}
3188EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3189
82be417a
AA		 3190struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3191{
3192 struct nfs4_threshold *thp;
3193
3194 thp = kzalloc(sizeof(*thp), GFP_NOFS);
3195 if (!thp) {
3196 dprintk("%s mdsthreshold allocation failed\n", __func__);
3197 return NULL;
3198 }
3199 return thp;
3200}
8733408d 3201
865a7ecb 3202#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 3203int
c8ad8894 3204pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 3205{
3206 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3207 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 3208 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 3209 struct nfs42_layoutstat_data *data;
3210 struct pnfs_layout_hdr *hdr;
3211 int status = 0;
3212
3213 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3214 goto out;
3215
6c5a0d89
TM		 3216 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3217 goto out;
3218
1bfe3b25
PT		 3219 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3220 goto out;
3221
8733408d
PT		 3222 spin_lock(&inode->i_lock);
3223 if (!NFS_I(inode)->layout) {
3224 spin_unlock(&inode->i_lock);
f538d0ba 3225 goto out_clear_layoutstats;
8733408d
PT		 3226 }
3227 hdr = NFS_I(inode)->layout;
3228 pnfs_get_layout_hdr(hdr);
3229 spin_unlock(&inode->i_lock);
3230
c8ad8894 3231 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3232 if (!data) {
3233 status = -ENOMEM;
3234 goto out_put;
3235 }
3236
3237 data->args.fh = NFS_FH(inode);
3238 data->args.inode = inode;
8733408d
PT		 3239 status = ld->prepare_layoutstats(&data->args);
3240 if (status)
3241 goto out_free;
3242
3243 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3244
3245out:
3246 dprintk("%s returns %d\n", __func__, status);
3247 return status;
3248
3249out_free:
3250 kfree(data);
3251out_put:
3252 pnfs_put_layout_hdr(hdr);
f538d0ba 3253out_clear_layoutstats:
1bfe3b25
PT		 3254 smp_mb__before_atomic();
3255 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3256 smp_mb__after_atomic();
8733408d
PT		 3257 goto out;
3258}
3259EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3260#endif
bbf58bf3
TM		 3261
3262unsigned int layoutstats_timer;
3263module_param(layoutstats_timer, uint, 0644);
3264EXPORT_SYMBOL_GPL(layoutstats_timer);
Linux 6.x block layer and io_uring tree(s)