projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
pNFS/NFSv4: Fix a layout segment leak in pnfs_layout_process()
[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;
b6d49ecd 297 unsigned long i_state;
cc6e5340 298
9c6376eb
TM		 299 if (!lo)
300 return;
301 inode = lo->plh_inode;
13c13a6a
TM		 302 pnfs_layoutreturn_before_put_layout_hdr(lo);
303
2b28a7be 304 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 305 if (!list_empty(&lo->plh_segs))
306 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 307 pnfs_detach_layout_hdr(lo);
b6d49ecd 308 i_state = inode->i_state;
cc6e5340 309 spin_unlock(&inode->i_lock);
6622c3ea 310 pnfs_free_layout_hdr(lo);
b6d49ecd
TM		 311 /* Notify pnfs_destroy_layout_final() that we're done */
312 if (i_state & (I_FREEING | I_CLEAR))
313 wake_up_var(lo);
cc6e5340 314 }
974cec8c
AA		 315}
316
b5fdf841
TM		 317static struct inode *
318pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
319{
320 struct inode *inode = igrab(lo->plh_inode);
321 if (inode)
322 return inode;
323 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
324 return NULL;
325}
326
4aab9732
TM		 327static void
328pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
329 u32 seq)
330{
331 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
332 iomode = IOMODE_ANY;
333 lo->plh_return_iomode = iomode;
334 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
335 if (seq != 0) {
336 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
337 lo->plh_return_seq = seq;
338 }
339}
340
ae5a459d
TM		 341static void
342pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
343{
5466d214 344 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 345 lo->plh_return_iomode = 0;
346 lo->plh_return_seq = 0;
347 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 348 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
349 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
350 continue;
351 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
352 }
ae5a459d
TM		 353}
354
362fb578
TM		 355static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
356{
357 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
358 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
359 smp_mb__after_atomic();
360 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
361 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
362}
363
fe1cf946
TM		 364static void
365pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
366 struct list_head *free_me)
367{
368 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
369 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
370 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
371 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
372 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
373 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
374}
375
7380020e 376/*
30cb3ee2
TM		 377 * Update the seqid of a layout stateid after receiving
378 * NFS4ERR_OLD_STATEID
7380020e 379 */
30cb3ee2 380bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
ecf84026
TM		 381 struct pnfs_layout_range *dst_range,
382 struct inode *inode)
7380020e
TM		 383{
384 struct pnfs_layout_hdr *lo;
c16467dc
TM		 385 struct pnfs_layout_range range = {
386 .iomode = IOMODE_ANY,
387 .offset = 0,
388 .length = NFS4_MAX_UINT64,
389 };
7380020e 390 bool ret = false;
c16467dc
TM		 391 LIST_HEAD(head);
392 int err;
7380020e
TM		 393
394 spin_lock(&inode->i_lock);
395 lo = NFS_I(inode)->layout;
30cb3ee2
TM		 396 if (lo && pnfs_layout_is_valid(lo) &&
397 nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
398 /* Is our call using the most recent seqid? If so, bump it */
399 if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
400 nfs4_stateid_seqid_inc(dst);
401 ret = true;
402 goto out;
403 }
404 /* Try to update the seqid to the most recent */
c16467dc
TM		 405 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
406 if (err != -EBUSY) {
407 dst->seqid = lo->plh_stateid.seqid;
ecf84026 408 *dst_range = range;
c16467dc
TM		 409 ret = true;
410 }
7380020e 411 }
30cb3ee2 412out:
7380020e 413 spin_unlock(&inode->i_lock);
c16467dc 414 pnfs_free_lseg_list(&head);
7380020e
TM		 415 return ret;
416}
417
2454dfea
TM		 418/*
419 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
420 *
421 * In order to continue using the pnfs_layout_hdr, a full recovery
422 * is required.
423 * Note that caller must hold inode->i_lock.
424 */
5f46be04 425int
2454dfea
TM		 426pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
427 struct list_head *lseg_list)
428{
429 struct pnfs_layout_range range = {
430 .iomode = IOMODE_ANY,
431 .offset = 0,
432 .length = NFS4_MAX_UINT64,
433 };
fe1cf946 434 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 435
436 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 437 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
438 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 439 pnfs_clear_layoutreturn_info(lo);
68f74479 440 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 441 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
442 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
443 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 444 return !list_empty(&lo->plh_segs);
2454dfea
TM		 445}
446
b9e028fd
TM		 447static int
448pnfs_iomode_to_fail_bit(u32 iomode)
449{
450 return iomode == IOMODE_RW ?
451 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
452}
453
454static void
3e621214 455pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 456{
25c75333 457 lo->plh_retry_timestamp = jiffies;
39e88fcf 458 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 459 refcount_inc(&lo->plh_refcount);
3e621214
TM		 460}
461
462static void
463pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
464{
465 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 466 refcount_dec(&lo->plh_refcount);
3e621214
TM		 467}
468
469static void
470pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
471{
472 struct inode *inode = lo->plh_inode;
115ce575
TM		 473 struct pnfs_layout_range range = {
474 .iomode = iomode,
475 .offset = 0,
476 .length = NFS4_MAX_UINT64,
477 };
478 LIST_HEAD(head);
3e621214
TM		 479
480 spin_lock(&inode->i_lock);
481 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 482 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 483 spin_unlock(&inode->i_lock);
115ce575 484 pnfs_free_lseg_list(&head);
b9e028fd
TM		 485 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
486 iomode == IOMODE_RW ? "RW" : "READ");
487}
488
489static bool
490pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
491{
25c75333 492 unsigned long start, end;
3e621214
TM		 493 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
494
495 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 496 return false;
497 end = jiffies;
498 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
499 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
500 /* It is time to retry the failed layoutgets */
3e621214 501 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 502 return false;
503 }
504 return true;
b9e028fd
TM		 505}
506
974cec8c 507static void
119cef97
TM		 508pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
509 const struct pnfs_layout_range *range,
510 const nfs4_stateid *stateid)
974cec8c 511{
566052c5 512 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 513 INIT_LIST_HEAD(&lseg->pls_lc_list);
a9901899 514 INIT_LIST_HEAD(&lseg->pls_commits);
eba6dd69 515 refcount_set(&lseg->pls_refcount, 1);
4541d16c 516 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 517 lseg->pls_layout = lo;
119cef97
TM		 518 lseg->pls_range = *range;
519 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 520}
521
905ca191 522static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 523{
68f74479
TM		 524 if (lseg != NULL) {
525 struct inode *inode = lseg->pls_layout->plh_inode;
526 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
527 }
974cec8c
AA		 528}
529
d684d2ae 530static void
57036a37
TM		 531pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
532 struct pnfs_layout_segment *lseg)
d684d2ae 533{
d20581aa 534 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 535 list_del_init(&lseg->pls_list);
8f0d27dc 536 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 537 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 538 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
539 return;
7b650994
TM		 540 if (list_empty(&lo->plh_segs) &&
541 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
542 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 543 if (atomic_read(&lo->plh_outstanding) == 0)
544 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 545 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 546 }
d684d2ae
FI		 547}
548
68f74479
TM		 549static bool
550pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
551 struct pnfs_layout_segment *lseg)
552{
553 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
554 pnfs_layout_is_valid(lo)) {
4aab9732 555 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 556 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
557 return true;
558 }
559 return false;
560}
561
bae724ef 562void
9369a431 563pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 564{
57036a37 565 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 566 struct inode *inode;
567
568 if (!lseg)
569 return;
570
4541d16c 571 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 572 refcount_read(&lseg->pls_refcount),
4541d16c 573 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 574
57036a37
TM		 575 lo = lseg->pls_layout;
576 inode = lo->plh_inode;
4ef2e4f8 577
eba6dd69 578 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 579 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
580 spin_unlock(&inode->i_lock);
581 return;
582 }
8f0d27dc 583 pnfs_get_layout_hdr(lo);
4ef2e4f8 584 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 585 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
586 lseg = NULL;
4ef2e4f8
TM		 587 spin_unlock(&inode->i_lock);
588 pnfs_free_lseg(lseg);
589 pnfs_put_layout_hdr(lo);
4541d16c 590 }
4541d16c 591}
9369a431 592EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 593
fb3296eb
BH		 594/*
595 * is l2 fully contained in l1?
596 * start1 end1
597 * [----------------------------------)
598 * start2 end2
599 * [----------------)
600 */
3cb2df17 601static bool
7dc0ac70 602pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 603 const struct pnfs_layout_range *l2)
fb3296eb
BH		 604{
605 u64 start1 = l1->offset;
17822b20 606 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 607 u64 start2 = l2->offset;
17822b20 608 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 609
610 return (start1 <= start2) && (end1 >= end2);
611}
612
24956804
TM		 613static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
614 struct list_head *tmp_list)
615{
eba6dd69 616 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 617 return false;
618 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
619 list_add(&lseg->pls_list, tmp_list);
620 return true;
621}
622
4541d16c
FI		 623/* Returns 1 if lseg is removed from list, 0 otherwise */
624static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
625 struct list_head *tmp_list)
626{
627 int rv = 0;
628
629 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
630 /* Remove the reference keeping the lseg in the
631 * list. It will now be removed when all
632 * outstanding io is finished.
633 */
d684d2ae 634 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 635 refcount_read(&lseg->pls_refcount));
24956804 636 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 637 rv = 1;
4541d16c
FI		 638 }
639 return rv;
640}
641
6d597e17
JL		 642/*
643 * Compare 2 layout stateid sequence ids, to see which is newer,
644 * taking into account wraparound issues.
645 */
646static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
647{
648 return (s32)(s1 - s2) > 0;
649}
650
e036f464
TM		 651static bool
652pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
653 const struct pnfs_layout_range *recall_range)
654{
655 return (recall_range->iomode == IOMODE_ANY ||
656 lseg_range->iomode == recall_range->iomode) &&
657 pnfs_lseg_range_intersecting(lseg_range, recall_range);
658}
659
660static bool
661pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
662 const struct pnfs_layout_range *recall_range,
663 u32 seq)
664{
665 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
666 return false;
667 if (recall_range == NULL)
668 return true;
669 return pnfs_should_free_range(&lseg->pls_range, recall_range);
670}
671
6d597e17
JL		 672/**
673 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
674 * @lo: layout header containing the lsegs
675 * @tmp_list: list head where doomed lsegs should go
676 * @recall_range: optional recall range argument to match (may be NULL)
677 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
678 *
679 * Walk the list of lsegs in the layout header, and tear down any that should
680 * be destroyed. If "recall_range" is specified then the segment must match
681 * that range. If "seq" is non-zero, then only match segments that were handed
682 * out at or before that sequence.
683 *
684 * Returns number of matching invalid lsegs remaining in list after scanning
685 * it and purging them.
4541d16c 686 */
43f1b3da 687int
49a85061 688pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 689 struct list_head *tmp_list,
6d597e17
JL		 690 const struct pnfs_layout_range *recall_range,
691 u32 seq)
974cec8c
AA		 692{
693 struct pnfs_layout_segment *lseg, *next;
71b39854 694 int remaining = 0;
974cec8c
AA		 695
696 dprintk("%s:Begin lo %p\n", __func__, lo);
697
8006bfba 698 if (list_empty(&lo->plh_segs))
38511722 699 return 0;
4541d16c 700 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 701 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 702 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 703 "offset %llu length %llu\n", __func__,
6d597e17
JL		 704 lseg, lseg->pls_range.iomode, lseg->pls_seq,
705 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 706 if (!mark_lseg_invalid(lseg, tmp_list))
707 remaining++;
4541d16c 708 }
71b39854
TM		 709 dprintk("%s:Return %i\n", __func__, remaining);
710 return remaining;
974cec8c
AA		 711}
712
68f74479
TM		 713static void
714pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
715 struct list_head *free_me,
716 const struct pnfs_layout_range *range,
717 u32 seq)
718{
719 struct pnfs_layout_segment *lseg, *next;
720
721 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
722 if (pnfs_match_lseg_recall(lseg, range, seq))
723 list_move_tail(&lseg->pls_list, free_me);
724 }
725}
726
f49f9baa 727/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 728void
4541d16c 729pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 730{
4541d16c 731 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 732
733 if (list_empty(free_me))
734 return;
735
4541d16c 736 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 737 list_del(&lseg->pls_list);
905ca191 738 pnfs_free_lseg(lseg);
974cec8c
AA		 739 }
740}
741
b6d49ecd 742static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi)
e5e94017
BH		 743{
744 struct pnfs_layout_hdr *lo;
974cec8c 745 LIST_HEAD(tmp_list);
e5e94017
BH		 746
747 spin_lock(&nfsi->vfs_inode.i_lock);
748 lo = nfsi->layout;
749 if (lo) {
3e621214 750 pnfs_get_layout_hdr(lo);
2454dfea 751 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 752 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
753 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
754 spin_unlock(&nfsi->vfs_inode.i_lock);
755 pnfs_free_lseg_list(&tmp_list);
1f18b82c 756 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 757 pnfs_put_layout_hdr(lo);
758 } else
759 spin_unlock(&nfsi->vfs_inode.i_lock);
b6d49ecd
TM		 760 return lo;
761}
762
763void pnfs_destroy_layout(struct nfs_inode *nfsi)
764{
765 __pnfs_destroy_layout(nfsi);
974cec8c 766}
041245c8 767EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 768
b6d49ecd
TM		 769static bool pnfs_layout_removed(struct nfs_inode *nfsi,
770 struct pnfs_layout_hdr *lo)
771{
772 bool ret;
773
774 spin_lock(&nfsi->vfs_inode.i_lock);
775 ret = nfsi->layout != lo;
776 spin_unlock(&nfsi->vfs_inode.i_lock);
777 return ret;
778}
779
780void pnfs_destroy_layout_final(struct nfs_inode *nfsi)
781{
782 struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi);
783
784 if (lo)
785 wait_var_event(lo, pnfs_layout_removed(nfsi, lo));
786}
787
fd9a8d71
TM		 788static bool
789pnfs_layout_add_bulk_destroy_list(struct inode *inode,
790 struct list_head *layout_list)
974cec8c
AA		 791{
792 struct pnfs_layout_hdr *lo;
fd9a8d71 793 bool ret = false;
974cec8c 794
fd9a8d71
TM		 795 spin_lock(&inode->i_lock);
796 lo = NFS_I(inode)->layout;
797 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
798 pnfs_get_layout_hdr(lo);
799 list_add(&lo->plh_bulk_destroy, layout_list);
800 ret = true;
801 }
802 spin_unlock(&inode->i_lock);
803 return ret;
804}
805
806/* Caller must hold rcu_read_lock and clp->cl_lock */
807static int
808pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
809 struct nfs_server *server,
810 struct list_head *layout_list)
5085607d
TM		 811 __must_hold(&clp->cl_lock)
812 __must_hold(RCU)
fd9a8d71
TM		 813{
814 struct pnfs_layout_hdr *lo, *next;
815 struct inode *inode;
816
817 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
5085607d
TM		 818 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
819 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
820 !list_empty(&lo->plh_bulk_destroy))
b85f5620 821 continue;
5085607d
TM		 822 /* If the sb is being destroyed, just bail */
823 if (!nfs_sb_active(server->super))
824 break;
b5fdf841 825 inode = pnfs_grab_inode_layout_hdr(lo);
5085607d 826 if (inode != NULL) {
cf6605d1
TM		 827 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
828 list_del_rcu(&lo->plh_layouts);
5085607d
TM		 829 if (pnfs_layout_add_bulk_destroy_list(inode,
830 layout_list))
831 continue;
832 rcu_read_unlock();
833 spin_unlock(&clp->cl_lock);
834 iput(inode);
835 } else {
836 rcu_read_unlock();
837 spin_unlock(&clp->cl_lock);
5085607d
TM		 838 }
839 nfs_sb_deactive(server->super);
fd9a8d71
TM		 840 spin_lock(&clp->cl_lock);
841 rcu_read_lock();
842 return -EAGAIN;
843 }
844 return 0;
845}
846
847static int
848pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
849 bool is_bulk_recall)
850{
851 struct pnfs_layout_hdr *lo;
852 struct inode *inode;
fd9a8d71
TM		 853 LIST_HEAD(lseg_list);
854 int ret = 0;
855
856 while (!list_empty(layout_list)) {
857 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
858 plh_bulk_destroy);
859 dprintk("%s freeing layout for inode %lu\n", __func__,
860 lo->plh_inode->i_ino);
861 inode = lo->plh_inode;
7c5d1875
CH		 862
863 pnfs_layoutcommit_inode(inode, false);
864
fd9a8d71
TM		 865 spin_lock(&inode->i_lock);
866 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 867 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
868 if (is_bulk_recall)
869 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 870 ret = -EAGAIN;
9fd4b9fc 871 }
fd9a8d71
TM		 872 spin_unlock(&inode->i_lock);
873 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 874 /* Free all lsegs that are attached to commit buckets */
875 nfs_commit_inode(inode, 0);
fd9a8d71 876 pnfs_put_layout_hdr(lo);
5085607d 877 nfs_iput_and_deactive(inode);
fd9a8d71
TM		 878 }
879 return ret;
880}
881
882int
883pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
884 struct nfs_fsid *fsid,
885 bool is_recall)
886{
887 struct nfs_server *server;
888 LIST_HEAD(layout_list);
c47abcf8 889
974cec8c 890 spin_lock(&clp->cl_lock);
6382a441 891 rcu_read_lock();
fd9a8d71 892restart:
6382a441 893 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 894 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
895 continue;
896 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
897 server,
898 &layout_list) != 0)
899 goto restart;
6382a441
WAA		 900 }
901 rcu_read_unlock();
974cec8c
AA		 902 spin_unlock(&clp->cl_lock);
903
fd9a8d71
TM		 904 if (list_empty(&layout_list))
905 return 0;
906 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
907}
908
909int
910pnfs_destroy_layouts_byclid(struct nfs_client *clp,
911 bool is_recall)
912{
913 struct nfs_server *server;
914 LIST_HEAD(layout_list);
915
916 spin_lock(&clp->cl_lock);
917 rcu_read_lock();
918restart:
919 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
920 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
921 server,
922 &layout_list) != 0)
923 goto restart;
974cec8c 924 }
fd9a8d71
TM		 925 rcu_read_unlock();
926 spin_unlock(&clp->cl_lock);
927
928 if (list_empty(&layout_list))
929 return 0;
930 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
931}
932
933/*
9f266451 934 * Called by the state manager to remove all layouts established under an
fd9a8d71
TM		 935 * expired lease.
936 */
937void
938pnfs_destroy_all_layouts(struct nfs_client *clp)
939{
940 nfs4_deviceid_mark_client_invalid(clp);
941 nfs4_deviceid_purge_client(clp);
942
943 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 944}
945
59b56394
TM		 946static void
947pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
948{
949 const struct cred *old;
950
951 if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
952 old = xchg(&lo->plh_lc_cred, get_cred(cred));
953 put_cred(old);
954 }
955}
956
fd6002e9 957/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 958void
959pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
59b56394 960 const struct cred *cred, bool update_barrier)
b1f69b75 961{
ecebb80b 962 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 963
2d2f24ad
TM		 964 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
965 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 966
967 if (!pnfs_layout_is_valid(lo)) {
59b56394 968 pnfs_set_layout_cred(lo, cred);
2a59a041
TM		 969 nfs4_stateid_copy(&lo->plh_stateid, new);
970 lo->plh_barrier = newseq;
971 pnfs_clear_layoutreturn_info(lo);
972 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
973 return;
974 }
975 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 976 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 977 /*
978 * Because of wraparound, we want to keep the barrier
979 * "close" to the current seqids.
980 */
981 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 982 }
ecebb80b
TM		 983 if (update_barrier)
984 new_barrier = be32_to_cpu(new->seqid);
985 else if (new_barrier == 0)
986 return;
2a59a041 987 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 988 lo->plh_barrier = new_barrier;
b1f69b75
AA		 989}
990
cf7d63f1 991static bool
19c54aba
TM		 992pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
993 const nfs4_stateid *stateid)
43f1b3da 994{
19c54aba 995 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 996
19c54aba
TM		 997 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
998}
999
1000/* lget is set to 1 if called from inside send_layoutget call chain */
1001static bool
e1c06f80 1002pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 1003{
f7e8917a 1004 return lo->plh_block_lgets ||
e1c06f80 1005 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 1006}
1007
5e36e2a9
FI		 1008static struct nfs_server *
1009pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
1010{
1011 struct nfs_server *server;
1012
78746a38 1013 if (inode) {
5e36e2a9 1014 server = NFS_SERVER(inode);
78746a38 1015 } else {
5e36e2a9
FI		 1016 struct dentry *parent_dir = dget_parent(ctx->dentry);
1017 server = NFS_SERVER(parent_dir->d_inode);
1018 dput(parent_dir);
1019 }
1020 return server;
1021}
1022
29a8bfe5
TM		 1023static void nfs4_free_pages(struct page **pages, size_t size)
1024{
1025 int i;
1026
1027 if (!pages)
1028 return;
1029
1030 for (i = 0; i < size; i++) {
1031 if (!pages[i])
1032 break;
1033 __free_page(pages[i]);
1034 }
1035 kfree(pages);
1036}
1037
1038static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1039{
1040 struct page **pages;
1041 int i;
1042
a2791d3a 1043 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 1044 if (!pages) {
1045 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1046 return NULL;
1047 }
1048
1049 for (i = 0; i < size; i++) {
1050 pages[i] = alloc_page(gfp_flags);
1051 if (!pages[i]) {
1052 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 1053 nfs4_free_pages(pages, i);
29a8bfe5
TM		 1054 return NULL;
1055 }
1056 }
1057
1058 return pages;
1059}
1060
587f03de 1061static struct nfs4_layoutget *
5e36e2a9 1062pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 1063 struct nfs_open_context *ctx,
2409a976 1064 const nfs4_stateid *stateid,
e144e539 1065 const struct pnfs_layout_range *range,
587f03de 1066 gfp_t gfp_flags)
e5e94017 1067{
5e36e2a9 1068 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 1069 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 1070 size_t max_pages = max_response_pages(server);
b1f69b75 1071 struct nfs4_layoutget *lgp;
b1f69b75
AA		 1072
1073 dprintk("--> %s\n", __func__);
e5e94017 1074
83026d80
JL		 1075 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1076 if (lgp == NULL)
587f03de 1077 return NULL;
83026d80 1078
28ced9a8
TM		 1079 if (max_reply_sz) {
1080 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1081 if (npages < max_pages)
1082 max_pages = npages;
1083 }
1084
dacb452d
FI		 1085 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1086 if (!lgp->args.layout.pages) {
1087 kfree(lgp);
1088 return NULL;
1089 }
1090 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1091 lgp->res.layoutp = &lgp->args.layout;
1092
d49e0d5b
TM		 1093 /* Don't confuse uninitialised result and success */
1094 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1095
1096 lgp->args.minlength = PAGE_SIZE;
1097 if (lgp->args.minlength > range->length)
1098 lgp->args.minlength = range->length;
5e36e2a9
FI		 1099 if (ino) {
1100 loff_t i_size = i_size_read(ino);
1101
1102 if (range->iomode == IOMODE_READ) {
1103 if (range->offset >= i_size)
1104 lgp->args.minlength = 0;
1105 else if (i_size - range->offset < lgp->args.minlength)
1106 lgp->args.minlength = i_size - range->offset;
1107 }
d03ab29d 1108 }
83026d80
JL		 1109 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1110 pnfs_copy_range(&lgp->args.range, range);
1111 lgp->args.type = server->pnfs_curr_ld->id;
1112 lgp->args.inode = ino;
1113 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1114 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1115 lgp->gfp_flags = gfp_flags;
63ec2b69 1116 lgp->cred = ctx->cred;
587f03de 1117 return lgp;
974cec8c
AA		 1118}
1119
29a8bfe5
TM		 1120void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1121{
1122 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1123
1124 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1125 if (lgp->args.inode)
1126 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
29a8bfe5
TM		 1127 put_nfs_open_context(lgp->args.ctx);
1128 kfree(lgp);
1129}
1130
24956804
TM		 1131static void pnfs_clear_layoutcommit(struct inode *inode,
1132 struct list_head *head)
1133{
1134 struct nfs_inode *nfsi = NFS_I(inode);
1135 struct pnfs_layout_segment *lseg, *tmp;
1136
1137 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1138 return;
1139 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1140 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1141 continue;
1142 pnfs_lseg_dec_and_remove_zero(lseg, head);
1143 }
1144}
1145
68f74479 1146void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1147 const nfs4_stateid *arg_stateid,
68f74479 1148 const struct pnfs_layout_range *range,
68f74479
TM		 1149 const nfs4_stateid *stateid)
1150{
1151 struct inode *inode = lo->plh_inode;
1152 LIST_HEAD(freeme);
1153
1154 spin_lock(&inode->i_lock);
c18d1e17 1155 if (!pnfs_layout_is_valid(lo) ||
2a974425
TM		 1156 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1157 goto out_unlock;
68f74479 1158 if (stateid) {
2a974425
TM		 1159 u32 seq = be32_to_cpu(arg_stateid->seqid);
1160
68f74479
TM		 1161 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1162 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
59b56394 1163 pnfs_set_layout_stateid(lo, stateid, NULL, true);
68f74479
TM		 1164 } else
1165 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1166out_unlock:
68f74479
TM		 1167 pnfs_clear_layoutreturn_waitbit(lo);
1168 spin_unlock(&inode->i_lock);
1169 pnfs_free_lseg_list(&freeme);
1170
1171}
1172
13c13a6a 1173static bool
e5fd1904
TM		 1174pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1175 nfs4_stateid *stateid,
44ea8dfc 1176 const struct cred **cred,
e5fd1904 1177 enum pnfs_iomode *iomode)
13c13a6a 1178{
bf0291dd
TM		 1179 /* Serialise LAYOUTGET/LAYOUTRETURN */
1180 if (atomic_read(&lo->plh_outstanding) != 0)
1181 return false;
6604b203 1182 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1183 return false;
6604b203 1184 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1185 pnfs_get_layout_hdr(lo);
e5fd1904 1186 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
44ea8dfc
TM		 1187 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1188 *cred = get_cred(lo->plh_lc_cred);
1189 if (lo->plh_return_seq != 0)
1190 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
e5fd1904
TM		 1191 if (iomode != NULL)
1192 *iomode = lo->plh_return_iomode;
1193 pnfs_clear_layoutreturn_info(lo);
1194 return true;
1195 }
44ea8dfc
TM		 1196 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1197 *cred = get_cred(lo->plh_lc_cred);
e5fd1904
TM		 1198 if (iomode != NULL)
1199 *iomode = IOMODE_ANY;
13c13a6a
TM		 1200 return true;
1201}
1202
828ed9ec
TM		 1203static void
1204pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1205 struct pnfs_layout_hdr *lo,
1206 const nfs4_stateid *stateid,
1207 enum pnfs_iomode iomode)
1208{
1209 struct inode *inode = lo->plh_inode;
1210
1211 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1212 args->inode = inode;
1213 args->range.iomode = iomode;
1214 args->range.offset = 0;
1215 args->range.length = NFS4_MAX_UINT64;
1216 args->layout = lo;
1217 nfs4_stateid_copy(&args->stateid, stateid);
1218}
1219
f40eb5d0 1220static int
44ea8dfc
TM		 1221pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
1222 const nfs4_stateid *stateid,
1223 const struct cred **pcred,
1224 enum pnfs_iomode iomode,
1225 bool sync)
f40eb5d0
PT		 1226{
1227 struct inode *ino = lo->plh_inode;
287bd3e9 1228 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0 1229 struct nfs4_layoutreturn *lrp;
44ea8dfc 1230 const struct cred *cred = *pcred;
f40eb5d0
PT		 1231 int status = 0;
1232
44ea8dfc 1233 *pcred = NULL;
e4af440a 1234 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1235 if (unlikely(lrp == NULL)) {
1236 status = -ENOMEM;
1237 spin_lock(&ino->i_lock);
d67ae825 1238 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0 1239 spin_unlock(&ino->i_lock);
44ea8dfc 1240 put_cred(cred);
f40eb5d0
PT		 1241 pnfs_put_layout_hdr(lo);
1242 goto out;
1243 }
1244
828ed9ec 1245 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1246 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0 1247 lrp->clp = NFS_SERVER(ino)->nfs_client;
44ea8dfc 1248 lrp->cred = cred;
287bd3e9
TM		 1249 if (ld->prepare_layoutreturn)
1250 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1251
6c16605d 1252 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1253out:
1254 dprintk("<-- %s status: %d\n", __func__, status);
1255 return status;
1256}
1257
d474f961
TM		 1258static bool
1259pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1260 enum pnfs_iomode iomode,
1261 u32 seq)
1262{
1263 struct pnfs_layout_range recall_range = {
1264 .length = NFS4_MAX_UINT64,
1265 .iomode = iomode,
1266 };
1267 return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1268 &recall_range, seq) != -EBUSY;
1269}
1270
13c13a6a
TM		 1271/* Return true if layoutreturn is needed */
1272static bool
1273pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1274{
2370abda 1275 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a 1276 return false;
d474f961
TM		 1277 return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1278 lo->plh_return_seq);
13c13a6a
TM		 1279}
1280
1281static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1282{
1283 struct inode *inode= lo->plh_inode;
1284
2370abda 1285 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1286 return;
1287 spin_lock(&inode->i_lock);
1288 if (pnfs_layout_need_return(lo)) {
44ea8dfc 1289 const struct cred *cred;
13c13a6a
TM		 1290 nfs4_stateid stateid;
1291 enum pnfs_iomode iomode;
1292 bool send;
1293
44ea8dfc 1294 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
13c13a6a
TM		 1295 spin_unlock(&inode->i_lock);
1296 if (send) {
1297 /* Send an async layoutreturn so we dont deadlock */
44ea8dfc 1298 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
13c13a6a
TM		 1299 }
1300 } else
1301 spin_unlock(&inode->i_lock);
1302}
1303
293b3b06
AA		 1304/*
1305 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1306 * when the layout segment list is empty.
1307 *
1308 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1309 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1310 * deviceid is marked invalid.
1311 */
cbe82603
BH		 1312int
1313_pnfs_return_layout(struct inode *ino)
1314{
1315 struct pnfs_layout_hdr *lo = NULL;
1316 struct nfs_inode *nfsi = NFS_I(ino);
1317 LIST_HEAD(tmp_list);
44ea8dfc 1318 const struct cred *cred;
cbe82603 1319 nfs4_stateid stateid;
24408f52 1320 int status = 0;
93b7f7ad 1321 bool send, valid_layout;
cbe82603 1322
366d5052 1323 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1324
1325 spin_lock(&ino->i_lock);
1326 lo = nfsi->layout;
e5929f3c 1327 if (!lo) {
cbe82603 1328 spin_unlock(&ino->i_lock);
293b3b06
AA		 1329 dprintk("NFS: %s no layout to return\n", __func__);
1330 goto out;
cbe82603 1331 }
cbe82603 1332 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1333 pnfs_get_layout_hdr(lo);
24408f52
TM		 1334 /* Is there an outstanding layoutreturn ? */
1335 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1336 spin_unlock(&ino->i_lock);
1337 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1338 TASK_UNINTERRUPTIBLE))
1339 goto out_put_layout_hdr;
1340 spin_lock(&ino->i_lock);
1341 }
93b7f7ad 1342 valid_layout = pnfs_layout_is_valid(lo);
24956804 1343 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1344 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1345
1346 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1347 struct pnfs_layout_range range = {
1348 .iomode = IOMODE_ANY,
1349 .offset = 0,
1350 .length = NFS4_MAX_UINT64,
1351 };
1352 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1353 }
1354
293b3b06 1355 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1356 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1357 !valid_layout) {
293b3b06 1358 spin_unlock(&ino->i_lock);
293b3b06 1359 dprintk("NFS: %s no layout segments to return\n", __func__);
7bcc1058 1360 goto out_wait_layoutreturn;
293b3b06 1361 }
47abadef 1362
44ea8dfc 1363 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
cbe82603 1364 spin_unlock(&ino->i_lock);
7f27392c 1365 if (send)
44ea8dfc 1366 status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
7bcc1058
TM		 1367out_wait_layoutreturn:
1368 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE);
7f27392c 1369out_put_layout_hdr:
ee6625a9 1370 pnfs_free_lseg_list(&tmp_list);
7f27392c 1371 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1372out:
1373 dprintk("<-- %s status: %d\n", __func__, status);
1374 return status;
1375}
1376
24028672
TM		 1377int
1378pnfs_commit_and_return_layout(struct inode *inode)
1379{
1380 struct pnfs_layout_hdr *lo;
1381 int ret;
1382
1383 spin_lock(&inode->i_lock);
1384 lo = NFS_I(inode)->layout;
1385 if (lo == NULL) {
1386 spin_unlock(&inode->i_lock);
1387 return 0;
1388 }
1389 pnfs_get_layout_hdr(lo);
1390 /* Block new layoutgets and read/write to ds */
1391 lo->plh_block_lgets++;
1392 spin_unlock(&inode->i_lock);
1393 filemap_fdatawait(inode->i_mapping);
1394 ret = pnfs_layoutcommit_inode(inode, true);
1395 if (ret == 0)
1396 ret = _pnfs_return_layout(inode);
1397 spin_lock(&inode->i_lock);
1398 lo->plh_block_lgets--;
1399 spin_unlock(&inode->i_lock);
1400 pnfs_put_layout_hdr(lo);
1401 return ret;
1402}
1403
1c5bd76d
TM		 1404bool pnfs_roc(struct inode *ino,
1405 struct nfs4_layoutreturn_args *args,
1406 struct nfs4_layoutreturn_res *res,
a52458b4 1407 const struct cred *cred)
f7e8917a 1408{
40dd4b7a
TM		 1409 struct nfs_inode *nfsi = NFS_I(ino);
1410 struct nfs_open_context *ctx;
1411 struct nfs4_state *state;
f7e8917a 1412 struct pnfs_layout_hdr *lo;
1c5bd76d 1413 struct pnfs_layout_segment *lseg, *next;
44ea8dfc 1414 const struct cred *lc_cred;
193e3aa2 1415 nfs4_stateid stateid;
1c5bd76d
TM		 1416 enum pnfs_iomode iomode = 0;
1417 bool layoutreturn = false, roc = false;
e71708d4 1418 bool skip_read = false;
f7e8917a 1419
1c5bd76d
TM		 1420 if (!nfs_have_layout(ino))
1421 return false;
29ade5db 1422retry:
0de43976 1423 rcu_read_lock();
f7e8917a 1424 spin_lock(&ino->i_lock);
40dd4b7a 1425 lo = nfsi->layout;
0cdc329e 1426 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1427 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1428 lo = NULL;
40dd4b7a 1429 goto out_noroc;
9c6376eb
TM		 1430 }
1431 pnfs_get_layout_hdr(lo);
29ade5db 1432 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1433 spin_unlock(&ino->i_lock);
0de43976 1434 rcu_read_unlock();
29ade5db
TM		 1435 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1436 TASK_UNINTERRUPTIBLE);
1437 pnfs_put_layout_hdr(lo);
1438 goto retry;
1439 }
40dd4b7a 1440
e755d638 1441 /* no roc if we hold a delegation */
e71708d4
TM		 1442 if (nfs4_check_delegation(ino, FMODE_READ)) {
1443 if (nfs4_check_delegation(ino, FMODE_WRITE))
1444 goto out_noroc;
1445 skip_read = true;
1446 }
40dd4b7a 1447
0de43976 1448 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1449 state = ctx->state;
e71708d4
TM		 1450 if (state == NULL)
1451 continue;
40dd4b7a 1452 /* Don't return layout if there is open file state */
e71708d4 1453 if (state->state & FMODE_WRITE)
40dd4b7a 1454 goto out_noroc;
e71708d4
TM		 1455 if (state->state & FMODE_READ)
1456 skip_read = true;
40dd4b7a
TM		 1457 }
1458
e755d638 1459
1c5bd76d 1460 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1461 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1462 continue;
e755d638 1463 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1464 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1465 continue;
1466 /*
1467 * Note: mark lseg for return so pnfs_layout_remove_lseg
1468 * doesn't invalidate the layout for us.
1469 */
1470 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1471 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1472 continue;
1473 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1474 }
1475
1c5bd76d
TM		 1476 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1477 goto out_noroc;
69820d22 1478
500d701f 1479 /* ROC in two conditions:
e755d638
PT		 1480 * 1. there are ROC lsegs
1481 * 2. we don't send layoutreturn
e755d638 1482 */
1c5bd76d 1483 /* lo ref dropped in pnfs_roc_release() */
44ea8dfc 1484 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
1c5bd76d 1485 /* If the creds don't match, we can't compound the layoutreturn */
4d8948c7 1486 if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0)
1c5bd76d
TM		 1487 goto out_noroc;
1488
1489 roc = layoutreturn;
1490 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1491 res->lrs_present = 0;
1492 layoutreturn = false;
44ea8dfc 1493 put_cred(lc_cred);
4d8948c7 1494
40dd4b7a 1495out_noroc:
f7e8917a 1496 spin_unlock(&ino->i_lock);
0de43976 1497 rcu_read_unlock();
e755d638 1498 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1499 if (roc) {
1500 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1501 if (ld->prepare_layoutreturn)
1502 ld->prepare_layoutreturn(args);
9c6376eb 1503 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1504 return true;
1505 }
e755d638 1506 if (layoutreturn)
44ea8dfc 1507 pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
9c6376eb 1508 pnfs_put_layout_hdr(lo);
287bd3e9 1509 return false;
f7e8917a
FI		 1510}
1511
078000d0
TM		 1512int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
1513 struct nfs4_layoutreturn_res **respp, int *ret)
287a9c55
TM		 1514{
1515 struct nfs4_layoutreturn_args *arg = *argpp;
1516 int retval = -EAGAIN;
1517
1518 if (!arg)
1519 return 0;
1520 /* Handle Layoutreturn errors */
1521 switch (*ret) {
1522 case 0:
1523 retval = 0;
1524 break;
6109bcf7
TM		 1525 case -NFS4ERR_NOMATCHING_LAYOUT:
1526 /* Was there an RPC level error? If not, retry */
1527 if (task->tk_rpc_status == 0)
1528 break;
1529 /* If the call was not sent, let caller handle it */
1530 if (!RPC_WAS_SENT(task))
1531 return 0;
1532 /*
1533 * Otherwise, assume the call succeeded and
1534 * that we need to release the layout
1535 */
1536 *ret = 0;
1537 (*respp)->lrs_present = 0;
1538 retval = 0;
1539 break;
078a432d
TM		 1540 case -NFS4ERR_DELAY:
1541 /* Let the caller handle the retry */
1542 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1543 return 0;
287a9c55 1544 case -NFS4ERR_OLD_STATEID:
30cb3ee2 1545 if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
078000d0 1546 &arg->range, arg->inode))
287a9c55
TM		 1547 break;
1548 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1549 return -EAGAIN;
1550 }
1551 *argpp = NULL;
1552 *respp = NULL;
1553 return retval;
1554}
1555
1c5bd76d
TM		 1556void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1557 struct nfs4_layoutreturn_res *res,
1558 int ret)
f7e8917a 1559{
1c5bd76d 1560 struct pnfs_layout_hdr *lo = args->layout;
67bbceed 1561 struct inode *inode = args->inode;
1c5bd76d 1562 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1563 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1564
9c47b18c
TM		 1565 switch (ret) {
1566 case -NFS4ERR_NOMATCHING_LAYOUT:
67bbceed
TM		 1567 spin_lock(&inode->i_lock);
1568 if (pnfs_layout_is_valid(lo) &&
1569 nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
1570 pnfs_set_plh_return_info(lo, args->range.iomode, 0);
c18d1e17 1571 pnfs_clear_layoutreturn_waitbit(lo);
67bbceed 1572 spin_unlock(&inode->i_lock);
9c47b18c
TM		 1573 break;
1574 case 0:
1c5bd76d
TM		 1575 if (res->lrs_present)
1576 res_stateid = &res->stateid;
df561f66 1577 fallthrough;
9c47b18c 1578 default:
c18d1e17
TM		 1579 pnfs_layoutreturn_free_lsegs(lo, &args->stateid, &args->range,
1580 res_stateid);
1c5bd76d 1581 }
a19b4785 1582 trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
287bd3e9
TM		 1583 if (ld_private && ld_private->ops && ld_private->ops->free)
1584 ld_private->ops->free(ld_private);
1c5bd76d 1585 pnfs_put_layout_hdr(lo);
f7e8917a
FI		 1586}
1587
500d701f
PT		 1588bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1589{
1590 struct nfs_inode *nfsi = NFS_I(ino);
1591 struct pnfs_layout_hdr *lo;
1592 bool sleep = false;
1593
1594 /* we might not have grabbed lo reference. so need to check under
1595 * i_lock */
1596 spin_lock(&ino->i_lock);
1597 lo = nfsi->layout;
ee284e35
TM		 1598 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1599 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1600 sleep = true;
ee284e35 1601 }
500d701f 1602 spin_unlock(&ino->i_lock);
500d701f
PT		 1603 return sleep;
1604}
1605
b1f69b75
AA		 1606/*
1607 * Compare two layout segments for sorting into layout cache.
1608 * We want to preferentially return RW over RO layouts, so ensure those
1609 * are seen first.
1610 */
1611static s64
7dc0ac70 1612pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1613 const struct pnfs_layout_range *l2)
b1f69b75 1614{
fb3296eb
BH		 1615 s64 d;
1616
1617 /* high offset > low offset */
1618 d = l1->offset - l2->offset;
1619 if (d)
1620 return d;
1621
1622 /* short length > long length */
1623 d = l2->length - l1->length;
1624 if (d)
1625 return d;
1626
b1f69b75 1627 /* read > read/write */
fb3296eb 1628 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1629}
1630
03772d2f
TM		 1631static bool
1632pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1633 const struct pnfs_layout_range *l2)
1634{
1635 return pnfs_lseg_range_cmp(l1, l2) > 0;
1636}
1637
1638static bool
1639pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1640 struct pnfs_layout_segment *old)
1641{
1642 return false;
1643}
1644
1645void
1646pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1647 struct pnfs_layout_segment *lseg,
1648 bool (*is_after)(const struct pnfs_layout_range *,
1649 const struct pnfs_layout_range *),
1650 bool (*do_merge)(struct pnfs_layout_segment *,
1651 struct pnfs_layout_segment *),
1652 struct list_head *free_me)
974cec8c 1653{
03772d2f 1654 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1655
974cec8c
AA		 1656 dprintk("%s:Begin\n", __func__);
1657
03772d2f
TM		 1658 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1659 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1660 continue;
1661 if (do_merge(lseg, lp)) {
1662 mark_lseg_invalid(lp, free_me);
1663 continue;
1664 }
1665 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1666 continue;
566052c5 1667 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1668 dprintk("%s: inserted lseg %p "
1669 "iomode %d offset %llu length %llu before "
1670 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1671 __func__, lseg, lseg->pls_range.iomode,
1672 lseg->pls_range.offset, lseg->pls_range.length,
1673 lp, lp->pls_range.iomode, lp->pls_range.offset,
1674 lp->pls_range.length);
fb3296eb 1675 goto out;
974cec8c 1676 }
fb3296eb
BH		 1677 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1678 dprintk("%s: inserted lseg %p "
1679 "iomode %d offset %llu length %llu at tail\n",
1680 __func__, lseg, lseg->pls_range.iomode,
1681 lseg->pls_range.offset, lseg->pls_range.length);
1682out:
70c3bd2b 1683 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1684
1685 dprintk("%s:Return\n", __func__);
e5e94017 1686}
03772d2f
TM		 1687EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1688
1689static void
1690pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1691 struct pnfs_layout_segment *lseg,
1692 struct list_head *free_me)
1693{
1694 struct inode *inode = lo->plh_inode;
1695 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1696
1697 if (ld->add_lseg != NULL)
1698 ld->add_lseg(lo, lseg, free_me);
1699 else
1700 pnfs_generic_layout_insert_lseg(lo, lseg,
1701 pnfs_lseg_range_is_after,
1702 pnfs_lseg_no_merge,
1703 free_me);
1704}
e5e94017
BH		 1705
1706static struct pnfs_layout_hdr *
9fa40758
PT		 1707alloc_init_layout_hdr(struct inode *ino,
1708 struct nfs_open_context *ctx,
1709 gfp_t gfp_flags)
e5e94017
BH		 1710{
1711 struct pnfs_layout_hdr *lo;
1712
636fb9c8 1713 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1714 if (!lo)
1715 return NULL;
2b28a7be 1716 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1717 INIT_LIST_HEAD(&lo->plh_layouts);
1718 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1719 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1720 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1721 lo->plh_inode = ino;
a52458b4 1722 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1723 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1724 return lo;
1725}
1726
1727static struct pnfs_layout_hdr *
9fa40758
PT		 1728pnfs_find_alloc_layout(struct inode *ino,
1729 struct nfs_open_context *ctx,
1730 gfp_t gfp_flags)
e5241e43
TM		 1731 __releases(&ino->i_lock)
1732 __acquires(&ino->i_lock)
e5e94017
BH		 1733{
1734 struct nfs_inode *nfsi = NFS_I(ino);
1735 struct pnfs_layout_hdr *new = NULL;
1736
1737 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1738
251ec410
TM		 1739 if (nfsi->layout != NULL)
1740 goto out_existing;
e5e94017 1741 spin_unlock(&ino->i_lock);
9fa40758 1742 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1743 spin_lock(&ino->i_lock);
1744
251ec410 1745 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1746 nfsi->layout = new;
251ec410 1747 return new;
7175fe90
YN		 1748 } else if (new != NULL)
1749 pnfs_free_layout_hdr(new);
251ec410
TM		 1750out_existing:
1751 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1752 return nfsi->layout;
1753}
1754
b1f69b75
AA		 1755/*
1756 * iomode matching rules:
c7d73af2
TH		 1757 * iomode lseg strict match
1758 * iomode
1759 * ----- ----- ------ -----
1760 * ANY READ N/A true
1761 * ANY RW N/A true
1762 * RW READ N/A false
1763 * RW RW N/A true
1764 * READ READ N/A true
1765 * READ RW true false
1766 * READ RW false true
b1f69b75 1767 */
3cb2df17 1768static bool
7dc0ac70 1769pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1770 const struct pnfs_layout_range *range,
1771 bool strict_iomode)
b1f69b75 1772{
fb3296eb
BH		 1773 struct pnfs_layout_range range1;
1774
1775 if ((range->iomode == IOMODE_RW &&
1776 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1777 (range->iomode != ls_range->iomode &&
6089dd0d 1778 strict_iomode) ||
7dc0ac70 1779 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1780 return false;
fb3296eb
BH		 1781
1782 /* range1 covers only the first byte in the range */
1783 range1 = *range;
1784 range1.length = 1;
7dc0ac70 1785 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1786}
1787
1788/*
1789 * lookup range in layout
1790 */
e5e94017 1791static struct pnfs_layout_segment *
fb3296eb 1792pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1793 struct pnfs_layout_range *range,
1794 bool strict_iomode)
e5e94017 1795{
b1f69b75
AA		 1796 struct pnfs_layout_segment *lseg, *ret = NULL;
1797
1798 dprintk("%s:Begin\n", __func__);
1799
b7edfaa1 1800 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1801 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1802 pnfs_lseg_range_match(&lseg->pls_range, range,
1803 strict_iomode)) {
9369a431 1804 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1805 break;
1806 }
b1f69b75
AA		 1807 }
1808
1809 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1810 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1811 return ret;
e5e94017
BH		 1812}
1813
d23d61c8
AA		 1814/*
1815 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1816 * to the MDS or over pNFS
1817 *
1818 * The nfs_inode read_io and write_io fields are cumulative counters reset
1819 * when there are no layout segments. Note that in pnfs_update_layout iomode
1820 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1821 * WRITE request.
1822 *
1823 * A return of true means use MDS I/O.
1824 *
1825 * From rfc 5661:
1826 * If a file's size is smaller than the file size threshold, data accesses
1827 * SHOULD be sent to the metadata server. If an I/O request has a length that
1828 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1829 * server. If both file size and I/O size are provided, the client SHOULD
1830 * reach or exceed both thresholds before sending its read or write
1831 * requests to the data server.
1832 */
1833static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1834 struct inode *ino, int iomode)
1835{
1836 struct nfs4_threshold *t = ctx->mdsthreshold;
1837 struct nfs_inode *nfsi = NFS_I(ino);
1838 loff_t fsize = i_size_read(ino);
1839 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1840
1841 if (t == NULL)
1842 return ret;
1843
1844 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1845 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1846
1847 switch (iomode) {
1848 case IOMODE_READ:
1849 if (t->bm & THRESHOLD_RD) {
1850 dprintk("%s fsize %llu\n", __func__, fsize);
1851 size_set = true;
1852 if (fsize < t->rd_sz)
1853 size = true;
1854 }
1855 if (t->bm & THRESHOLD_RD_IO) {
1856 dprintk("%s nfsi->read_io %llu\n", __func__,
1857 nfsi->read_io);
1858 io_set = true;
1859 if (nfsi->read_io < t->rd_io_sz)
1860 io = true;
1861 }
1862 break;
1863 case IOMODE_RW:
1864 if (t->bm & THRESHOLD_WR) {
1865 dprintk("%s fsize %llu\n", __func__, fsize);
1866 size_set = true;
1867 if (fsize < t->wr_sz)
1868 size = true;
1869 }
1870 if (t->bm & THRESHOLD_WR_IO) {
1871 dprintk("%s nfsi->write_io %llu\n", __func__,
1872 nfsi->write_io);
1873 io_set = true;
1874 if (nfsi->write_io < t->wr_io_sz)
1875 io = true;
1876 }
1877 break;
1878 }
1879 if (size_set && io_set) {
1880 if (size && io)
1881 ret = true;
1882 } else if (size || io)
1883 ret = true;
1884
1885 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1886 return ret;
1887}
1888
d03360aa 1889static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1890{
1891 /*
1892 * send layoutcommit as it can hold up layoutreturn due to lseg
1893 * reference
1894 */
1895 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1896 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1897 nfs_wait_bit_killable,
d03360aa 1898 TASK_KILLABLE);
aa8a45ee
PT		 1899}
1900
411ae722
TM		 1901static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1902{
1903 atomic_inc(&lo->plh_outstanding);
1904}
1905
1906static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1907{
1908 if (atomic_dec_and_test(&lo->plh_outstanding))
1909 wake_up_var(&lo->plh_outstanding);
1910}
1911
d67ae825
TH		 1912static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1913{
1914 unsigned long *bitlock = &lo->plh_flags;
1915
1916 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1917 smp_mb__after_atomic();
1918 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1919}
1920
78746a38
FI		 1921static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1922 struct nfs_server *server)
1923{
cf6605d1 1924 if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
78746a38
FI		 1925 struct nfs_client *clp = server->nfs_client;
1926
1927 /* The lo must be on the clp list if there is any
1928 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1929 */
1930 spin_lock(&clp->cl_lock);
cf6605d1 1931 list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
78746a38
FI		 1932 spin_unlock(&clp->cl_lock);
1933 }
1934}
1935
e5e94017
BH		 1936/*
1937 * Layout segment is retreived from the server if not cached.
1938 * The appropriate layout segment is referenced and returned to the caller.
1939 */
7c24d948 1940struct pnfs_layout_segment *
e5e94017
BH		 1941pnfs_update_layout(struct inode *ino,
1942 struct nfs_open_context *ctx,
fb3296eb
BH		 1943 loff_t pos,
1944 u64 count,
a75b9df9 1945 enum pnfs_iomode iomode,
c7d73af2 1946 bool strict_iomode,
a75b9df9 1947 gfp_t gfp_flags)
e5e94017 1948{
fb3296eb
BH		 1949 struct pnfs_layout_range arg = {
1950 .iomode = iomode,
1951 .offset = pos,
1952 .length = count,
1953 };
70d2f7b1 1954 unsigned pg_offset;
6382a441
WAA		 1955 struct nfs_server *server = NFS_SERVER(ino);
1956 struct nfs_client *clp = server->nfs_client;
183d9e7b 1957 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1958 struct pnfs_layout_segment *lseg = NULL;
587f03de 1959 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1960 nfs4_stateid stateid;
1961 long timeout = 0;
66b53f32 1962 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1963 bool first;
e5e94017 1964
9a4bf31d 1965 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1966 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1967 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1968 goto out;
9a4bf31d 1969 }
d23d61c8 1970
9a4bf31d 1971 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1972 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1973 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1974 goto out;
9a4bf31d 1975 }
d23d61c8 1976
9bf87482 1977lookup_again:
d03360aa
TM		 1978 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1979 if (IS_ERR(lseg))
1980 goto out;
9bf87482 1981 first = false;
e5e94017 1982 spin_lock(&ino->i_lock);
9fa40758 1983 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1984 if (lo == NULL) {
1985 spin_unlock(&ino->i_lock);
183d9e7b 1986 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1987 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1988 goto out;
1989 }
e5e94017 1990
43f1b3da 1991 /* Do we even need to bother with this? */
a59c30ac 1992 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1993 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1994 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1995 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1996 goto out_unlock;
1997 }
1998
1999 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 2000 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 2001 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2002 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 2003 goto out_unlock;
9a4bf31d 2004 }
e5e94017 2005
411ae722
TM		 2006 /*
2007 * If the layout segment list is empty, but there are outstanding
2008 * layoutget calls, then they might be subject to a layoutrecall.
2009 */
2010 if (list_empty(&lo->plh_segs) &&
2011 atomic_read(&lo->plh_outstanding) != 0) {
2012 spin_unlock(&ino->i_lock);
d03360aa 2013 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
400417b0 2014 !atomic_read(&lo->plh_outstanding)));
58bbeab4 2015 if (IS_ERR(lseg))
411ae722
TM		 2016 goto out_put_layout_hdr;
2017 pnfs_put_layout_hdr(lo);
2018 goto lookup_again;
2019 }
2020
2c8d5fc3
TM		 2021 /*
2022 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
2023 * for LAYOUTRETURN.
2024 */
2025 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
2026 spin_unlock(&ino->i_lock);
2027 dprintk("%s wait for layoutreturn\n", __func__);
2028 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2029 if (!IS_ERR(lseg)) {
2030 pnfs_put_layout_hdr(lo);
2031 dprintk("%s retrying\n", __func__);
2032 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2033 lseg,
2034 PNFS_UPDATE_LAYOUT_RETRY);
2035 goto lookup_again;
2036 }
2037 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2038 PNFS_UPDATE_LAYOUT_RETURN);
2039 goto out_put_layout_hdr;
2040 }
2041
c7d73af2 2042 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 2043 if (lseg) {
2044 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2045 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
2046 goto out_unlock;
2047 }
2048
183d9e7b
JL		 2049 /*
2050 * Choose a stateid for the LAYOUTGET. If we don't have a layout
2051 * stateid, or it has been invalidated, then we must use the open
2052 * stateid.
2053 */
67a3b721 2054 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
d9aba2b4 2055 int status;
183d9e7b
JL		 2056
2057 /*
2058 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 2059 * RFC 5661 Errata 3208.
2060 */
2061 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
2062 &lo->plh_flags)) {
2063 spin_unlock(&ino->i_lock);
d03360aa
TM		 2064 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
2065 NFS_LAYOUT_FIRST_LAYOUTGET,
2066 TASK_KILLABLE));
2067 if (IS_ERR(lseg))
2068 goto out_put_layout_hdr;
9bf87482 2069 pnfs_put_layout_hdr(lo);
183d9e7b 2070 dprintk("%s retrying\n", __func__);
9bf87482
PT		 2071 goto lookup_again;
2072 }
183d9e7b 2073
fbf4bcc9 2074 spin_unlock(&ino->i_lock);
183d9e7b 2075 first = true;
d9aba2b4 2076 status = nfs4_select_rw_stateid(ctx->state,
70d2f7b1 2077 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
d9aba2b4
TM		 2078 NULL, &stateid, NULL);
2079 if (status != 0) {
731c74dd 2080 lseg = ERR_PTR(status);
70d2f7b1
TM		 2081 trace_pnfs_update_layout(ino, pos, count,
2082 iomode, lo, lseg,
2083 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
d9aba2b4
TM		 2084 nfs4_schedule_stateid_recovery(server, ctx->state);
2085 pnfs_clear_first_layoutget(lo);
2086 pnfs_put_layout_hdr(lo);
2087 goto lookup_again;
70d2f7b1 2088 }
fbf4bcc9 2089 spin_lock(&ino->i_lock);
9bf87482 2090 } else {
183d9e7b 2091 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 2092 }
568e8c49 2093
9a4bf31d 2094 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 2095 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2096 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 2097 goto out_unlock;
9a4bf31d 2098 }
411ae722 2099 nfs_layoutget_begin(lo);
f49f9baa 2100 spin_unlock(&ino->i_lock);
30005121 2101
78746a38 2102 _add_to_server_list(lo, server);
e5e94017 2103
09cbfeaf 2104 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 2105 if (pg_offset) {
2106 arg.offset -= pg_offset;
2107 arg.length += pg_offset;
2108 }
7c24d948 2109 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 2110 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 2111
5e36e2a9 2112 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 2113 if (!lgp) {
2114 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2115 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 2116 nfs_layoutget_end(lo);
587f03de
FI		 2117 goto out_put_layout_hdr;
2118 }
2119
dacb452d 2120 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 2121 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2122 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 2123 nfs_layoutget_end(lo);
83026d80 2124 if (IS_ERR(lseg)) {
183d9e7b 2125 switch(PTR_ERR(lseg)) {
e85d7ee4 2126 case -EBUSY:
183d9e7b
JL		 2127 if (time_after(jiffies, giveup))
2128 lseg = NULL;
66b53f32
TM		 2129 break;
2130 case -ERECALLCONFLICT:
183d9e7b 2131 case -EAGAIN:
56b38a1f 2132 break;
183d9e7b
JL		 2133 default:
2134 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2135 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2136 lseg = NULL;
2137 }
56b38a1f
TM		 2138 goto out_put_layout_hdr;
2139 }
2140 if (lseg) {
2141 if (first)
2142 pnfs_clear_first_layoutget(lo);
2143 trace_pnfs_update_layout(ino, pos, count,
2144 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2145 pnfs_put_layout_hdr(lo);
2146 goto lookup_again;
83026d80
JL		 2147 }
2148 } else {
2149 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2150 }
2151
830ffb56 2152out_put_layout_hdr:
d67ae825
TH		 2153 if (first)
2154 pnfs_clear_first_layoutget(lo);
d5b9216f
TM		 2155 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2156 PNFS_UPDATE_LAYOUT_EXIT);
70c3bd2b 2157 pnfs_put_layout_hdr(lo);
e5e94017 2158out:
f86bbcf8
TM		 2159 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2160 "(%s, offset: %llu, length: %llu)\n",
2161 __func__, ino->i_sb->s_id,
2162 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2163 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2164 iomode==IOMODE_RW ? "read/write" : "read-only",
2165 (unsigned long long)pos,
2166 (unsigned long long)count);
e5e94017
BH		 2167 return lseg;
2168out_unlock:
2169 spin_unlock(&ino->i_lock);
830ffb56 2170 goto out_put_layout_hdr;
e5e94017 2171}
7c24d948 2172EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2173
540d9864
TM		 2174static bool
2175pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2176{
2177 switch (range->iomode) {
2178 case IOMODE_READ:
2179 case IOMODE_RW:
2180 break;
2181 default:
2182 return false;
2183 }
2184 if (range->offset == NFS4_MAX_UINT64)
2185 return false;
2186 if (range->length == 0)
2187 return false;
2188 if (range->length != NFS4_MAX_UINT64 &&
2189 range->length > NFS4_MAX_UINT64 - range->offset)
2190 return false;
2191 return true;
2192}
2193
78746a38
FI		 2194static struct pnfs_layout_hdr *
2195_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2196{
2197 struct pnfs_layout_hdr *lo;
2198
2199 spin_lock(&ino->i_lock);
2200 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2201 if (!lo)
2202 goto out_unlock;
2203 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2204 goto out_unlock;
2205 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2206 goto out_unlock;
2207 if (pnfs_layoutgets_blocked(lo))
2208 goto out_unlock;
2209 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2210 goto out_unlock;
411ae722 2211 nfs_layoutget_begin(lo);
78746a38
FI		 2212 spin_unlock(&ino->i_lock);
2213 _add_to_server_list(lo, NFS_SERVER(ino));
2214 return lo;
2215
2216out_unlock:
2217 spin_unlock(&ino->i_lock);
2218 pnfs_put_layout_hdr(lo);
2219 return NULL;
2220}
2221
2409a976
FI		 2222static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2223 struct nfs_open_context *ctx)
2224{
78746a38
FI		 2225 struct inode *ino = data->dentry->d_inode;
2226 struct pnfs_layout_range rng = {
2227 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2228 IOMODE_RW: IOMODE_READ,
2229 .offset = 0,
2230 .length = NFS4_MAX_UINT64,
2231 };
2232 struct nfs4_layoutget *lgp;
2233 struct pnfs_layout_hdr *lo;
2234
64294b08
TM		 2235 /* Heuristic: don't send layoutget if we have cached data */
2236 if (rng.iomode == IOMODE_READ &&
2237 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2238 return;
2239
78746a38
FI		 2240 lo = _pnfs_grab_empty_layout(ino, ctx);
2241 if (!lo)
2242 return;
2243 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2244 &rng, GFP_KERNEL);
2245 if (!lgp) {
2246 pnfs_clear_first_layoutget(lo);
cb2856c5 2247 nfs_layoutget_end(lo);
78746a38
FI		 2248 pnfs_put_layout_hdr(lo);
2249 return;
2250 }
2251 data->lgp = lgp;
2252 data->o_arg.lg_args = &lgp->args;
2253 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2254}
2255
2256static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2257 struct nfs_open_context *ctx)
2258{
2259 struct pnfs_layout_range rng = {
2260 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2261 IOMODE_RW: IOMODE_READ,
2262 .offset = 0,
2263 .length = NFS4_MAX_UINT64,
2264 };
2265 struct nfs4_layoutget *lgp;
2266
2267 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2268 &rng, GFP_KERNEL);
2269 if (!lgp)
2270 return;
2271 data->lgp = lgp;
2272 data->o_arg.lg_args = &lgp->args;
2273 data->o_res.lg_res = &lgp->res;
2274}
2275
2276void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2277 struct nfs_open_context *ctx)
2278{
2279 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2280
2281 if (!(pnfs_enabled_sb(server) &&
2282 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2283 return;
2284 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2285 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2286 return;
2409a976
FI		 2287 if (data->state)
2288 _lgopen_prepare_attached(data, ctx);
2289 else
2290 _lgopen_prepare_floating(data, ctx);
2291}
2292
2293void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2294 struct nfs_open_context *ctx)
2295{
2296 struct pnfs_layout_hdr *lo;
2297 struct pnfs_layout_segment *lseg;
c49b5209 2298 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2299 u32 iomode;
2300
6e01260c 2301 if (!lgp)
2409a976 2302 return;
6e01260c
FI		 2303 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2304 if (lgp->res.status) {
2305 switch (lgp->res.status) {
6e01260c 2306 default:
8dc96566
TM		 2307 break;
2308 /*
2309 * Halt lgopen attempts if the server doesn't recognise
2310 * the "current stateid" value, the layout type, or the
2311 * layoutget operation as being valid.
2312 * Also if it complains about too many ops in the compound
2313 * or of the request/reply being too big.
2314 */
2315 case -NFS4ERR_BAD_STATEID:
2316 case -NFS4ERR_NOTSUPP:
2317 case -NFS4ERR_REP_TOO_BIG:
2318 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2319 case -NFS4ERR_REQ_TOO_BIG:
2320 case -NFS4ERR_TOO_MANY_OPS:
2321 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2322 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2323 }
2324 return;
2325 }
2409a976 2326 if (!lgp->args.inode) {
78746a38
FI		 2327 lo = _pnfs_grab_empty_layout(ino, ctx);
2328 if (!lo)
2329 return;
2409a976
FI		 2330 lgp->args.inode = ino;
2331 } else
2332 lo = NFS_I(lgp->args.inode)->layout;
2409a976
FI		 2333
2334 lseg = pnfs_layout_process(lgp);
32f1c28f 2335 if (!IS_ERR(lseg)) {
2409a976
FI		 2336 iomode = lgp->args.range.iomode;
2337 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2338 pnfs_put_lseg(lseg);
2339 }
30ae2412
FI		 2340}
2341
2342void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2343{
2344 if (lgp != NULL) {
2345 struct inode *inode = lgp->args.inode;
2346 if (inode) {
2347 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
30ae2412 2348 pnfs_clear_first_layoutget(lo);
411ae722 2349 nfs_layoutget_end(lo);
30ae2412
FI		 2350 }
2351 pnfs_layoutget_free(lgp);
2352 }
2409a976
FI		 2353}
2354
a0b0a6e3 2355struct pnfs_layout_segment *
b1f69b75
AA		 2356pnfs_layout_process(struct nfs4_layoutget *lgp)
2357{
2358 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2359 struct nfs4_layoutget_res *res = &lgp->res;
2360 struct pnfs_layout_segment *lseg;
b7edfaa1 2361 struct inode *ino = lo->plh_inode;
78096cca 2362 LIST_HEAD(free_me);
540d9864
TM		 2363
2364 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2365 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2366
2367 /* Inject layout blob into I/O device driver */
a75b9df9 2368 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2369 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2370 if (!lseg)
1b3c6d07
JL		 2371 lseg = ERR_PTR(-ENOMEM);
2372
2373 dprintk("%s: Could not allocate layout: error %ld\n",
2374 __func__, PTR_ERR(lseg));
2375 return lseg;
b1f69b75
AA		 2376 }
2377
119cef97 2378 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2379
b1f69b75 2380 spin_lock(&ino->i_lock);
e1c06f80 2381 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2382 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2383 goto out_forget;
43f1b3da 2384 }
038d6493 2385
9888d837
TM		 2386 if (!pnfs_layout_is_valid(lo)) {
2387 /* We have a completely new layout */
59b56394 2388 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
9888d837 2389 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2390 /* existing state ID, make sure the sequence number matches. */
2391 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2392 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2393 goto out_forget;
362f7474 2394 }
59b56394 2395 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
362f7474
CH		 2396 } else {
2397 /*
2398 * We got an entirely new state ID. Mark all segments for the
9888d837 2399 * inode invalid, and retry the layoutget
362f7474 2400 */
d9b61708 2401 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 2402 goto out_forget;
362f7474 2403 }
038d6493 2404
9369a431 2405 pnfs_get_lseg(lseg);
03772d2f 2406 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2407
b1f69b75 2408
3976143b 2409 if (res->return_on_close)
f7e8917a 2410 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2411
b1f69b75 2412 spin_unlock(&ino->i_lock);
78096cca 2413 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2414 return lseg;
43f1b3da 2415
1b3c6d07 2416out_forget:
43f1b3da
FI		 2417 spin_unlock(&ino->i_lock);
2418 lseg->pls_layout = lo;
2419 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
814b8497 2420 pnfs_free_lseg_list(&free_me);
1b3c6d07 2421 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2422}
2423
2f215968
TM		 2424/**
2425 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2426 * @lo: pointer to layout header
2427 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2428 * @return_range: describe layout segment ranges to be returned
e0b7d420 2429 * @seq: stateid seqid to match
2f215968
TM		 2430 *
2431 * This function is mainly intended for use by layoutrecall. It attempts
2432 * to free the layout segment immediately, or else to mark it for return
2433 * as soon as its reference count drops to zero.
e0b7d420
TM		 2434 *
2435 * Returns
2436 * - 0: a layoutreturn needs to be scheduled.
2437 * - EBUSY: there are layout segment that are still in use.
2438 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2439 */
10335556 2440int
016256df
PT		 2441pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2442 struct list_head *tmp_list,
6d597e17
JL		 2443 const struct pnfs_layout_range *return_range,
2444 u32 seq)
016256df
PT		 2445{
2446 struct pnfs_layout_segment *lseg, *next;
10335556 2447 int remaining = 0;
016256df
PT		 2448
2449 dprintk("%s:Begin lo %p\n", __func__, lo);
2450
fc7ff367 2451 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2452
2453 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2454 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2455 dprintk("%s: marking lseg %p iomode %d "
2456 "offset %llu length %llu\n", __func__,
2457 lseg, lseg->pls_range.iomode,
2458 lseg->pls_range.offset,
2459 lseg->pls_range.length);
ff041727 2460 if (mark_lseg_invalid(lseg, tmp_list))
2f215968
TM		 2461 continue;
2462 remaining++;
016256df 2463 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2464 }
6d597e17 2465
e0b7d420 2466 if (remaining) {
6d597e17 2467 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2468 return -EBUSY;
2469 }
6d597e17 2470
e0b7d420
TM		 2471 if (!list_empty(&lo->plh_return_segs)) {
2472 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2473 return 0;
2474 }
2475
2476 return -ENOENT;
016256df
PT		 2477}
2478
b5fdf841
TM		 2479static void
2480pnfs_mark_layout_for_return(struct inode *inode,
2481 const struct pnfs_layout_range *range)
016256df 2482{
b5fdf841 2483 struct pnfs_layout_hdr *lo;
10335556 2484 bool return_now = false;
016256df
PT		 2485
2486 spin_lock(&inode->i_lock);
b5fdf841 2487 lo = NFS_I(inode)->layout;
bdebfccd
TM		 2488 if (!pnfs_layout_is_valid(lo)) {
2489 spin_unlock(&inode->i_lock);
2490 return;
2491 }
b5fdf841 2492 pnfs_set_plh_return_info(lo, range->iomode, 0);
016256df
PT		 2493 /*
2494 * mark all matching lsegs so that we are sure to have no live
2495 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2496 * for how it works.
2497 */
b5fdf841 2498 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
44ea8dfc 2499 const struct cred *cred;
10335556 2500 nfs4_stateid stateid;
e5fd1904 2501 enum pnfs_iomode iomode;
10335556 2502
44ea8dfc 2503 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
10335556
TM		 2504 spin_unlock(&inode->i_lock);
2505 if (return_now)
44ea8dfc 2506 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
10335556
TM		 2507 } else {
2508 spin_unlock(&inode->i_lock);
2509 nfs_commit_inode(inode, 0);
2510 }
016256df 2511}
b5fdf841
TM		 2512
2513void pnfs_error_mark_layout_for_return(struct inode *inode,
2514 struct pnfs_layout_segment *lseg)
2515{
2516 struct pnfs_layout_range range = {
2517 .iomode = lseg->pls_range.iomode,
2518 .offset = 0,
2519 .length = NFS4_MAX_UINT64,
2520 };
2521
2522 pnfs_mark_layout_for_return(inode, &range);
2523}
016256df
PT		 2524EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2525
b5fdf841
TM		 2526static bool
2527pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
2528{
2529 return pnfs_layout_is_valid(lo) &&
2530 !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
2531 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
2532}
2533
2534static struct pnfs_layout_segment *
2535pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
2536 const struct pnfs_layout_range *range,
2537 enum pnfs_iomode iomode)
2538{
2539 struct pnfs_layout_segment *lseg;
2540
2541 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
2542 if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
2543 continue;
2544 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2545 continue;
2546 if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
2547 continue;
2548 if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
2549 return lseg;
2550 }
2551 return NULL;
2552}
2553
2554/* Find open file states whose mode matches that of the range */
2555static bool
2556pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
2557 const struct pnfs_layout_range *range)
2558{
2559 struct list_head *head;
2560 struct nfs_open_context *ctx;
2561 fmode_t mode = 0;
2562
2563 if (!pnfs_layout_can_be_returned(lo) ||
2564 !pnfs_find_first_lseg(lo, range, range->iomode))
2565 return false;
2566
2567 head = &NFS_I(lo->plh_inode)->open_files;
2568 list_for_each_entry_rcu(ctx, head, list) {
2569 if (ctx->state)
2570 mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
2571 }
2572
2573 switch (range->iomode) {
2574 default:
2575 break;
2576 case IOMODE_READ:
2577 mode &= ~FMODE_WRITE;
2578 break;
2579 case IOMODE_RW:
2580 if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
2581 mode &= ~FMODE_READ;
2582 }
2583 return mode == 0;
2584}
2585
2586static int
2587pnfs_layout_return_unused_byserver(struct nfs_server *server, void *data)
2588{
2589 const struct pnfs_layout_range *range = data;
2590 struct pnfs_layout_hdr *lo;
2591 struct inode *inode;
2592restart:
2593 rcu_read_lock();
2594 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
2595 if (!pnfs_layout_can_be_returned(lo) ||
2596 test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2597 continue;
2598 inode = lo->plh_inode;
2599 spin_lock(&inode->i_lock);
2600 if (!pnfs_should_return_unused_layout(lo, range)) {
2601 spin_unlock(&inode->i_lock);
2602 continue;
2603 }
2604 spin_unlock(&inode->i_lock);
2605 inode = pnfs_grab_inode_layout_hdr(lo);
2606 if (!inode)
2607 continue;
2608 rcu_read_unlock();
2609 pnfs_mark_layout_for_return(inode, range);
2610 iput(inode);
2611 cond_resched();
2612 goto restart;
2613 }
2614 rcu_read_unlock();
2615 return 0;
2616}
2617
2618void
2619pnfs_layout_return_unused_byclid(struct nfs_client *clp,
2620 enum pnfs_iomode iomode)
2621{
2622 struct pnfs_layout_range range = {
2623 .iomode = iomode,
2624 .offset = 0,
2625 .length = NFS4_MAX_UINT64,
2626 };
2627
2628 nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
2629 &range);
2630}
2631
b3230e80
TM		 2632void
2633pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2634{
2635 if (pgio->pg_lseg == NULL ||
2636 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2637 return;
2638 pnfs_put_lseg(pgio->pg_lseg);
2639 pgio->pg_lseg = NULL;
2640}
2641EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2642
08cb5b0f
BC		 2643/*
2644 * Check for any intersection between the request and the pgio->pg_lseg,
2645 * and if none, put this pgio->pg_lseg away.
2646 */
e1e54ab7 2647void
08cb5b0f
BC		 2648pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2649{
2650 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2651 pnfs_put_lseg(pgio->pg_lseg);
2652 pgio->pg_lseg = NULL;
2653 }
2654}
e1e54ab7 2655EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
08cb5b0f 2656
d8007d4d
TM		 2657void
2658pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2659{
1fd937bd
PT		 2660 u64 rd_size = req->wb_bytes;
2661
b3230e80 2662 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2663 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2664 if (pgio->pg_lseg == NULL) {
2665 if (pgio->pg_dreq == NULL)
2666 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2667 else
2668 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2669
2670 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2671 nfs_req_openctx(req),
cb5d04bc
PT		 2672 req_offset(req),
2673 rd_size,
2674 IOMODE_READ,
c7d73af2 2675 false,
cb5d04bc 2676 GFP_KERNEL);
d600ad1f
PT		 2677 if (IS_ERR(pgio->pg_lseg)) {
2678 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2679 pgio->pg_lseg = NULL;
2680 return;
2681 }
cb5d04bc 2682 }
e885de1a
TM		 2683 /* If no lseg, fall back to read through mds */
2684 if (pgio->pg_lseg == NULL)
1f945357 2685 nfs_pageio_reset_read_mds(pgio);
e885de1a 2686
d8007d4d
TM		 2687}
2688EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2689
2690void
6296556f
PT		 2691pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2692 struct nfs_page *req, u64 wb_size)
d8007d4d 2693{
b3230e80 2694 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2695 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2696 if (pgio->pg_lseg == NULL) {
cb5d04bc 2697 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2698 nfs_req_openctx(req),
cb5d04bc
PT		 2699 req_offset(req),
2700 wb_size,
2701 IOMODE_RW,
c7d73af2 2702 false,
2b17d725 2703 GFP_KERNEL);
d600ad1f
PT		 2704 if (IS_ERR(pgio->pg_lseg)) {
2705 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2706 pgio->pg_lseg = NULL;
2707 return;
2708 }
2709 }
e885de1a
TM		 2710 /* If no lseg, fall back to write through mds */
2711 if (pgio->pg_lseg == NULL)
1f945357 2712 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2713}
2714EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2715
180bb5ec
WAA		 2716void
2717pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2718{
2719 if (desc->pg_lseg) {
2720 pnfs_put_lseg(desc->pg_lseg);
2721 desc->pg_lseg = NULL;
2722 }
2723}
2724EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2725
b4fdac1a
WAA		 2726/*
2727 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2728 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2729 */
2730size_t
a7d42ddb
WAA		 2731pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2732 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2733{
0f9c429e 2734 unsigned int size;
c5e20cb7 2735 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2736
2737 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2738 if (!size)
2739 return 0;
94ad1c80 2740
19982ba8 2741 /*
c5e20cb7
WAA		 2742 * 'size' contains the number of bytes left in the current page (up
2743 * to the original size asked for in @req->wb_bytes).
2744 *
2745 * Calculate how many bytes are left in the layout segment
2746 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2747 *
2748 * Please also note that 'end_offset' is actually the offset of the
2749 * first byte that lies outside the pnfs_layout_range. FIXME?
2750 *
2751 */
19b54848 2752 if (pgio->pg_lseg) {
17822b20 2753 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2754 pgio->pg_lseg->pls_range.length);
2755 req_start = req_offset(req);
08cb5b0f 2756
c5e20cb7 2757 /* start of request is past the last byte of this segment */
08cb5b0f 2758 if (req_start >= seg_end)
19b54848 2759 return 0;
c5e20cb7
WAA		 2760
2761 /* adjust 'size' iff there are fewer bytes left in the
2762 * segment than what nfs_generic_pg_test returned */
2763 seg_left = seg_end - req_start;
2764 if (seg_left < size)
2765 size = (unsigned int)seg_left;
19b54848 2766 }
0f9c429e 2767
19b54848 2768 return size;
94ad1c80 2769}
89a58e32 2770EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2771
53113ad3 2772int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2773{
2774 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2775
2776 /* Resend all requests through the MDS */
53113ad3
WAA		 2777 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2778 hdr->completion_ops);
c7070113 2779 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2780 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2781}
e7dd79af 2782EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2783
d45f60c6 2784static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2785{
cd841605
FI		 2786
2787 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2788 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2789 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2790 pnfs_return_layout(hdr->inode);
1acbbb4e 2791 }
6c75dc0d 2792 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2793 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2794}
2795
d20581aa
BH		 2796/*
2797 * Called by non rpc-based layout drivers
2798 */
d45f60c6 2799void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2800{
f8417b48 2801 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2802 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2803 hdr->mds_offset + hdr->res.count);
d45f60c6 2804 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2805 }
2806 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2807 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2808 pnfs_ld_handle_write_error(hdr);
2809 hdr->mds_ops->rpc_release(hdr);
44b83799 2810}
d20581aa 2811EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2812
dce81290
TM		 2813static void
2814pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2815 struct nfs_pgio_header *hdr)
dce81290 2816{
48d635f1 2817 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2818
6c75dc0d 2819 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2820 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2821 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2822 mirror->pg_recoalesce = 1;
6c75dc0d 2823 }
ba4a76f7 2824 hdr->completion_ops->completion(hdr);
dce81290
TM		 2825}
2826
2827static enum pnfs_try_status
d45f60c6 2828pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2829 const struct rpc_call_ops *call_ops,
2830 struct pnfs_layout_segment *lseg,
2831 int how)
0382b744 2832{
cd841605 2833 struct inode *inode = hdr->inode;
0382b744
AA		 2834 enum pnfs_try_status trypnfs;
2835 struct nfs_server *nfss = NFS_SERVER(inode);
2836
cd841605 2837 hdr->mds_ops = call_ops;
0382b744
AA		 2838
2839 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2840 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2841 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2842 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2843 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2844 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2845 return trypnfs;
2846}
2847
dce81290 2848static void
7f714720
WAA		 2849pnfs_do_write(struct nfs_pageio_descriptor *desc,
2850 struct nfs_pgio_header *hdr, int how)
dce81290 2851{
dce81290
TM		 2852 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2853 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2854 enum pnfs_try_status trypnfs;
dce81290 2855
d45f60c6 2856 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2857 switch (trypnfs) {
2858 case PNFS_NOT_ATTEMPTED:
d45f60c6 2859 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2860 case PNFS_ATTEMPTED:
2861 break;
2862 case PNFS_TRY_AGAIN:
2863 /* cleanup hdr and prepare to redo pnfs */
2864 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2865 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2866 list_splice_init(&hdr->pages, &mirror->pg_list);
2867 mirror->pg_recoalesce = 1;
2868 }
2869 hdr->mds_ops->rpc_release(hdr);
2870 }
dce81290
TM		 2871}
2872
6c75dc0d
FI		 2873static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2874{
9369a431 2875 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2876 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2877}
2878
dce81290
TM		 2879int
2880pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2881{
6c75dc0d 2882 struct nfs_pgio_header *hdr;
dce81290
TM		 2883 int ret;
2884
1e7f3a48
WAA		 2885 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2886 if (!hdr) {
2bff2288
PT		 2887 desc->pg_error = -ENOMEM;
2888 return desc->pg_error;
dce81290 2889 }
6c75dc0d 2890 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2891
9369a431 2892 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2893 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2894 if (!ret)
7f714720 2895 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2896
6c75dc0d 2897 return ret;
dce81290
TM		 2898}
2899EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2900
53113ad3 2901int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2902{
2903 struct nfs_pageio_descriptor pgio;
2904
1acbbb4e 2905 /* Resend all requests through the MDS */
53113ad3
WAA		 2906 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2907 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2908}
e7dd79af 2909EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2910
d45f60c6 2911static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2912{
cd841605
FI		 2913 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2914 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2915 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2916 pnfs_return_layout(hdr->inode);
1acbbb4e 2917 }
4db6e0b7 2918 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2919 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2920}
2921
d20581aa
BH		 2922/*
2923 * Called by non rpc-based layout drivers
2924 */
d45f60c6 2925void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2926{
bfc505de 2927 if (likely(!hdr->pnfs_error))
d45f60c6 2928 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2929 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2930 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2931 pnfs_ld_handle_read_error(hdr);
2932 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2933}
2934EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2935
493292dd
TM		 2936static void
2937pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2938 struct nfs_pgio_header *hdr)
493292dd 2939{
48d635f1 2940 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2941
4db6e0b7 2942 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2943 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2944 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2945 mirror->pg_recoalesce = 1;
4db6e0b7 2946 }
ba4a76f7 2947 hdr->completion_ops->completion(hdr);
493292dd
TM		 2948}
2949
64419a9b
AA		 2950/*
2951 * Call the appropriate parallel I/O subsystem read function.
2952 */
493292dd 2953static enum pnfs_try_status
d45f60c6 2954pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2955 const struct rpc_call_ops *call_ops,
2956 struct pnfs_layout_segment *lseg)
64419a9b 2957{
cd841605 2958 struct inode *inode = hdr->inode;
64419a9b
AA		 2959 struct nfs_server *nfss = NFS_SERVER(inode);
2960 enum pnfs_try_status trypnfs;
2961
cd841605 2962 hdr->mds_ops = call_ops;
64419a9b
AA		 2963
2964 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2965 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2966
d45f60c6 2967 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2968 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2969 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2970 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2971 return trypnfs;
2972}
863a3c6c 2973
ceb11e13 2974/* Resend all requests through pnfs. */
563c53e7
TM		 2975void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr,
2976 unsigned int mirror_idx)
ceb11e13
PT		 2977{
2978 struct nfs_pageio_descriptor pgio;
2979
1b1bc66b 2980 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2981 /* Prevent deadlocks with layoutreturn! */
2982 pnfs_put_lseg(hdr->lseg);
2983 hdr->lseg = NULL;
2984
1b1bc66b
WAA		 2985 nfs_pageio_init_read(&pgio, hdr->inode, false,
2986 hdr->completion_ops);
563c53e7 2987 pgio.pg_mirror_idx = mirror_idx;
1b1bc66b
WAA		 2988 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2989 }
ceb11e13
PT		 2990}
2991EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2992
493292dd 2993static void
7f714720 2994pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2995{
493292dd
TM		 2996 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2997 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2998 enum pnfs_try_status trypnfs;
493292dd 2999
d45f60c6 3000 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 3001 switch (trypnfs) {
3002 case PNFS_NOT_ATTEMPTED:
d45f60c6 3003 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 3004 case PNFS_ATTEMPTED:
3005 break;
3006 case PNFS_TRY_AGAIN:
3007 /* cleanup hdr and prepare to redo pnfs */
3008 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3009 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3010 list_splice_init(&hdr->pages, &mirror->pg_list);
3011 mirror->pg_recoalesce = 1;
3012 }
3013 hdr->mds_ops->rpc_release(hdr);
3014 }
493292dd
TM		 3015}
3016
4db6e0b7
FI		 3017static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
3018{
9369a431 3019 pnfs_put_lseg(hdr->lseg);
1e7f3a48 3020 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 3021}
3022
493292dd
TM		 3023int
3024pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
3025{
4db6e0b7 3026 struct nfs_pgio_header *hdr;
493292dd
TM		 3027 int ret;
3028
1e7f3a48
WAA		 3029 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
3030 if (!hdr) {
2bff2288
PT		 3031 desc->pg_error = -ENOMEM;
3032 return desc->pg_error;
493292dd 3033 }
4db6e0b7 3034 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 3035 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 3036 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 3037 if (!ret)
7f714720 3038 pnfs_do_read(desc, hdr);
4db6e0b7 3039 return ret;
493292dd
TM		 3040}
3041EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
3042
71244d9b
TM		 3043static void pnfs_clear_layoutcommitting(struct inode *inode)
3044{
3045 unsigned long *bitlock = &NFS_I(inode)->flags;
3046
3047 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 3048 smp_mb__after_atomic();
71244d9b
TM		 3049 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
3050}
3051
863a3c6c 3052/*
a9bae566 3053 * There can be multiple RW segments.
863a3c6c 3054 */
a9bae566 3055static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 3056{
a9bae566 3057 struct pnfs_layout_segment *lseg;
863a3c6c 3058
a9bae566
PT		 3059 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
3060 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 3061 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 3062 list_add(&lseg->pls_lc_list, listp);
3063 }
863a3c6c
AA		 3064}
3065
a073dbff
TM		 3066static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
3067{
3068 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 3069
3070 /* Matched by references in pnfs_set_layoutcommit */
3071 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
3072 list_del_init(&lseg->pls_lc_list);
3073 pnfs_put_lseg(lseg);
3074 }
3075
71244d9b 3076 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 3077}
3078
1b0ae068
PT		 3079void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
3080{
b9e028fd 3081 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 3082}
3083EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
3084
863a3c6c 3085void
67af7611
TM		 3086pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
3087 loff_t end_pos)
863a3c6c 3088{
cd841605 3089 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 3090 bool mark_as_dirty = false;
863a3c6c 3091
cd841605 3092 spin_lock(&inode->i_lock);
863a3c6c 3093 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 3094 nfsi->layout->plh_lwb = end_pos;
79a48a1f 3095 mark_as_dirty = true;
863a3c6c 3096 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 3097 __func__, inode->i_ino);
29559b11
TM		 3098 } else if (end_pos > nfsi->layout->plh_lwb)
3099 nfsi->layout->plh_lwb = end_pos;
67af7611 3100 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 3101 /* references matched in nfs4_layoutcommit_release */
67af7611 3102 pnfs_get_lseg(lseg);
a9bae566 3103 }
cd841605 3104 spin_unlock(&inode->i_lock);
acff5880 3105 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 3106 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 3107
3108 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3109 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3110 if (mark_as_dirty)
cd841605 3111 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3112}
3113EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
3114
db29c089
AA		 3115void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
3116{
3117 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
3118
3119 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
3120 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 3121 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 3122}
3123
de4b15c7
AA		 3124/*
3125 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3126 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3127 * data to disk to allow the server to recover the data if it crashes.
3128 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3129 * is off, and a COMMIT is sent to a data server, or
3130 * if WRITEs to a data server return NFS_DATA_SYNC.
3131 */
863a3c6c 3132int
ef311537 3133pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 3134{
5f919c9f 3135 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 3136 struct nfs4_layoutcommit_data *data;
3137 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 3138 loff_t end_pos;
71244d9b 3139 int status;
863a3c6c 3140
71244d9b 3141 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 3142 return 0;
3143
71244d9b 3144 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 3145
71244d9b 3146 status = -EAGAIN;
92407e75 3147 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 3148 if (!sync)
3149 goto out;
74316201 3150 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 3151 NFS_INO_LAYOUTCOMMITTING,
3152 nfs_wait_bit_killable,
3153 TASK_KILLABLE);
92407e75 3154 if (status)
71244d9b 3155 goto out;
92407e75
PT		 3156 }
3157
71244d9b
TM		 3158 status = -ENOMEM;
3159 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
3160 data = kzalloc(sizeof(*data), GFP_NOFS);
3161 if (!data)
3162 goto clear_layoutcommitting;
3163
3164 status = 0;
de4b15c7 3165 spin_lock(&inode->i_lock);
71244d9b
TM		 3166 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
3167 goto out_unlock;
a9bae566 3168
71244d9b 3169 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 3170 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 3171
acff5880 3172 end_pos = nfsi->layout->plh_lwb;
863a3c6c 3173
f597c537 3174 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
97a728f5 3175 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 3176 spin_unlock(&inode->i_lock);
3177
3178 data->args.inode = inode;
863a3c6c
AA		 3179 nfs_fattr_init(&data->fattr);
3180 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3181 data->res.fattr = &data->fattr;
2e18d4d8
TM		 3182 if (end_pos != 0)
3183 data->args.lastbytewritten = end_pos - 1;
3184 else
3185 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 3186 data->res.server = NFS_SERVER(inode);
3187
5f919c9f
CH		 3188 if (ld->prepare_layoutcommit) {
3189 status = ld->prepare_layoutcommit(&data->args);
3190 if (status) {
a52458b4 3191 put_cred(data->cred);
5f919c9f 3192 spin_lock(&inode->i_lock);
29559b11
TM		 3193 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3194 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 3195 nfsi->layout->plh_lwb = end_pos;
3471648a 3196 goto out_unlock;
5f919c9f
CH		 3197 }
3198 }
3199
3200
863a3c6c
AA		 3201 status = nfs4_proc_layoutcommit(data, sync);
3202out:
92407e75
PT		 3203 if (status)
3204 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3205 dprintk("<-- %s status %d\n", __func__, status);
3206 return status;
71244d9b
TM		 3207out_unlock:
3208 spin_unlock(&inode->i_lock);
92407e75 3209 kfree(data);
71244d9b
TM		 3210clear_layoutcommitting:
3211 pnfs_clear_layoutcommitting(inode);
92407e75 3212 goto out;
863a3c6c 3213}
72cff449 3214EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 3215
5bb89b47
TM		 3216int
3217pnfs_generic_sync(struct inode *inode, bool datasync)
3218{
3219 return pnfs_layoutcommit_inode(inode, true);
3220}
3221EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3222
82be417a
AA		 3223struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3224{
3225 struct nfs4_threshold *thp;
3226
3227 thp = kzalloc(sizeof(*thp), GFP_NOFS);
3228 if (!thp) {
3229 dprintk("%s mdsthreshold allocation failed\n", __func__);
3230 return NULL;
3231 }
3232 return thp;
3233}
8733408d 3234
865a7ecb 3235#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 3236int
c8ad8894 3237pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 3238{
3239 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3240 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 3241 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 3242 struct nfs42_layoutstat_data *data;
3243 struct pnfs_layout_hdr *hdr;
3244 int status = 0;
3245
3246 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3247 goto out;
3248
6c5a0d89
TM		 3249 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3250 goto out;
3251
1bfe3b25
PT		 3252 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3253 goto out;
3254
8733408d
PT		 3255 spin_lock(&inode->i_lock);
3256 if (!NFS_I(inode)->layout) {
3257 spin_unlock(&inode->i_lock);
f538d0ba 3258 goto out_clear_layoutstats;
8733408d
PT		 3259 }
3260 hdr = NFS_I(inode)->layout;
3261 pnfs_get_layout_hdr(hdr);
3262 spin_unlock(&inode->i_lock);
3263
c8ad8894 3264 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3265 if (!data) {
3266 status = -ENOMEM;
3267 goto out_put;
3268 }
3269
3270 data->args.fh = NFS_FH(inode);
3271 data->args.inode = inode;
8733408d
PT		 3272 status = ld->prepare_layoutstats(&data->args);
3273 if (status)
3274 goto out_free;
3275
3276 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3277
3278out:
3279 dprintk("%s returns %d\n", __func__, status);
3280 return status;
3281
3282out_free:
3283 kfree(data);
3284out_put:
3285 pnfs_put_layout_hdr(hdr);
f538d0ba 3286out_clear_layoutstats:
1bfe3b25
PT		 3287 smp_mb__before_atomic();
3288 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3289 smp_mb__after_atomic();
8733408d
PT		 3290 goto out;
3291}
3292EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3293#endif
bbf58bf3
TM		 3294
3295unsigned int layoutstats_timer;
3296module_param(layoutstats_timer, uint, 0644);
3297EXPORT_SYMBOL_GPL(layoutstats_timer);
Linux 6.x block layer and io_uring tree(s)