projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'irq_urgent_for_v6.8_rc2' of git://git.kernel.org/pub/scm/linux/kernel...
[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
7c9d845f
TM		 95const struct pnfs_layoutdriver_type *pnfs_find_layoutdriver(u32 id)
96{
97 return find_pnfs_driver(id);
98}
99
100void pnfs_put_layoutdriver(const struct pnfs_layoutdriver_type *ld)
101{
102 if (ld)
103 module_put(ld->owner);
104}
105
85e174ba
RL		 106void
107unset_pnfs_layoutdriver(struct nfs_server *nfss)
108{
738fd0f3
BH		 109 if (nfss->pnfs_curr_ld) {
110 if (nfss->pnfs_curr_ld->clear_layoutdriver)
111 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 112 /* Decrement the MDS count. Purge the deviceid cache if zero */
113 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
114 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 115 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 116 }
85e174ba
RL		 117 nfss->pnfs_curr_ld = NULL;
118}
119
ca440c38
JL		 120/*
121 * When the server sends a list of layout types, we choose one in the order
122 * given in the list below.
123 *
124 * FIXME: should this list be configurable in some fashion? module param?
125 * mount option? something else?
126 */
127static const u32 ld_prefs[] = {
128 LAYOUT_SCSI,
129 LAYOUT_BLOCK_VOLUME,
130 LAYOUT_OSD2_OBJECTS,
131 LAYOUT_FLEX_FILES,
132 LAYOUT_NFSV4_1_FILES,
133 0
134};
135
136static int
137ld_cmp(const void *e1, const void *e2)
138{
139 u32 ld1 = *((u32 *)e1);
140 u32 ld2 = *((u32 *)e2);
141 int i;
142
143 for (i = 0; ld_prefs[i] != 0; i++) {
144 if (ld1 == ld_prefs[i])
145 return -1;
146
147 if (ld2 == ld_prefs[i])
148 return 1;
149 }
150 return 0;
151}
152
85e174ba
RL		 153/*
154 * Try to set the server's pnfs module to the pnfs layout type specified by id.
155 * Currently only one pNFS layout driver per filesystem is supported.
156 *
3132e49e 157 * @ids array of layout types supported by MDS.
85e174ba
RL		 158 */
159void
738fd0f3 160set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 161 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 162{
163 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 164 u32 id;
ca440c38 165 int i;
85e174ba 166
19274716
AS		 167 if (fsinfo->nlayouttypes == 0)
168 goto out_no_driver;
85e174ba
RL		 169 if (!(server->nfs_client->cl_exchange_flags &
170 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 171 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
172 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 173 goto out_no_driver;
174 }
3132e49e 175
ca440c38
JL		 176 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
177 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 178
ca440c38
JL		 179 for (i = 0; i < fsinfo->nlayouttypes; i++) {
180 id = fsinfo->layouttype[i];
85e174ba 181 ld_type = find_pnfs_driver(id);
ca440c38
JL		 182 if (!ld_type) {
183 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
184 id);
185 ld_type = find_pnfs_driver(id);
186 }
187 if (ld_type)
188 break;
85e174ba 189 }
3132e49e
JL		 190
191 if (!ld_type) {
ca440c38 192 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 193 goto out_no_driver;
194 }
195
85e174ba 196 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 197 if (ld_type->set_layoutdriver
198 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 199 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
200 "driver %u.\n", __func__, id);
738fd0f3
BH		 201 module_put(ld_type->owner);
202 goto out_no_driver;
203 }
2a4c8994
TM		 204 /* Bump the MDS count */
205 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 206
85e174ba
RL		 207 dprintk("%s: pNFS module for %u set\n", __func__, id);
208 return;
209
210out_no_driver:
211 dprintk("%s: Using NFSv4 I/O\n", __func__);
212 server->pnfs_curr_ld = NULL;
213}
02c35fca
FI		 214
215int
216pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
217{
218 int status = -EINVAL;
219 struct pnfs_layoutdriver_type *tmp;
220
221 if (ld_type->id == 0) {
a030889a 222 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 223 return status;
224 }
b1f69b75 225 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 226 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 227 "alloc_lseg and free_lseg.\n", __func__);
228 return status;
229 }
02c35fca
FI		 230
231 spin_lock(&pnfs_spinlock);
232 tmp = find_pnfs_driver_locked(ld_type->id);
233 if (!tmp) {
234 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
235 status = 0;
236 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
237 ld_type->name);
238 } else {
a030889a 239 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 240 __func__, ld_type->id);
241 }
242 spin_unlock(&pnfs_spinlock);
243
244 return status;
245}
246EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
247
248void
249pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
250{
251 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
252 spin_lock(&pnfs_spinlock);
253 list_del(&ld_type->pnfs_tblid);
254 spin_unlock(&pnfs_spinlock);
255}
256EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 257
b1f69b75
AA		 258/*
259 * pNFS client layout cache
260 */
261
cc6e5340 262/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 263void
70c3bd2b 264pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 265{
2b28a7be 266 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 267}
268
636fb9c8
BH		 269static struct pnfs_layout_hdr *
270pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
271{
272 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 273 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 274}
275
276static void
277pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
278{
9c626381
TM		 279 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
280 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
281
cf6605d1 282 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
9c626381
TM		 283 struct nfs_client *clp = server->nfs_client;
284
285 spin_lock(&clp->cl_lock);
cf6605d1 286 list_del_rcu(&lo->plh_layouts);
9c626381
TM		 287 spin_unlock(&clp->cl_lock);
288 }
a52458b4 289 put_cred(lo->plh_lc_cred);
57934278 290 return ld->free_layout_hdr(lo);
636fb9c8
BH		 291}
292
e5e94017 293static void
6622c3ea 294pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 295{
bb346f63 296 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 297 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 298 nfsi->layout = NULL;
299 /* Reset MDS Threshold I/O counters */
300 nfsi->write_io = 0;
301 nfsi->read_io = 0;
e5e94017
BH		 302}
303
b1f69b75 304void
70c3bd2b 305pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 306{
9c6376eb 307 struct inode *inode;
b6d49ecd 308 unsigned long i_state;
cc6e5340 309
9c6376eb
TM		 310 if (!lo)
311 return;
312 inode = lo->plh_inode;
13c13a6a
TM		 313 pnfs_layoutreturn_before_put_layout_hdr(lo);
314
2b28a7be 315 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 316 if (!list_empty(&lo->plh_segs))
317 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 318 pnfs_detach_layout_hdr(lo);
b6d49ecd 319 i_state = inode->i_state;
cc6e5340 320 spin_unlock(&inode->i_lock);
6622c3ea 321 pnfs_free_layout_hdr(lo);
b6d49ecd
TM		 322 /* Notify pnfs_destroy_layout_final() that we're done */
323 if (i_state & (I_FREEING | I_CLEAR))
324 wake_up_var(lo);
cc6e5340 325 }
974cec8c
AA		 326}
327
b5fdf841
TM		 328static struct inode *
329pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
330{
331 struct inode *inode = igrab(lo->plh_inode);
332 if (inode)
333 return inode;
334 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
335 return NULL;
336}
337
1bcf34fd
TM		 338/*
339 * Compare 2 layout stateid sequence ids, to see which is newer,
340 * taking into account wraparound issues.
341 */
342static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
343{
344 return (s32)(s1 - s2) > 0;
345}
346
347static void pnfs_barrier_update(struct pnfs_layout_hdr *lo, u32 newseq)
348{
45baadaa 349 if (pnfs_seqid_is_newer(newseq, lo->plh_barrier) || !lo->plh_barrier)
1bcf34fd
TM		 350 lo->plh_barrier = newseq;
351}
352
4aab9732
TM		 353static void
354pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
355 u32 seq)
356{
357 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
358 iomode = IOMODE_ANY;
359 lo->plh_return_iomode = iomode;
360 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
e20772cb
TM		 361 /*
362 * We must set lo->plh_return_seq to avoid livelocks with
363 * pnfs_layout_need_return()
364 */
365 if (seq == 0)
366 seq = be32_to_cpu(lo->plh_stateid.seqid);
367 if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq))
4aab9732 368 lo->plh_return_seq = seq;
e20772cb 369 pnfs_barrier_update(lo, seq);
4aab9732
TM		 370}
371
ae5a459d
TM		 372static void
373pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
374{
5466d214 375 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 376 lo->plh_return_iomode = 0;
377 lo->plh_return_seq = 0;
378 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 379 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
380 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
381 continue;
382 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
383 }
ae5a459d
TM		 384}
385
362fb578
TM		 386static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
387{
388 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
389 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
390 smp_mb__after_atomic();
391 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
392 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
393}
394
fe1cf946
TM		 395static void
396pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
397 struct list_head *free_me)
398{
399 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
400 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
401 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
402 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
403 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
404 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
405}
406
7380020e 407/*
30cb3ee2
TM		 408 * Update the seqid of a layout stateid after receiving
409 * NFS4ERR_OLD_STATEID
7380020e 410 */
30cb3ee2 411bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
ecf84026
TM		 412 struct pnfs_layout_range *dst_range,
413 struct inode *inode)
7380020e
TM		 414{
415 struct pnfs_layout_hdr *lo;
c16467dc
TM		 416 struct pnfs_layout_range range = {
417 .iomode = IOMODE_ANY,
418 .offset = 0,
419 .length = NFS4_MAX_UINT64,
420 };
7380020e 421 bool ret = false;
c16467dc
TM		 422 LIST_HEAD(head);
423 int err;
7380020e
TM		 424
425 spin_lock(&inode->i_lock);
426 lo = NFS_I(inode)->layout;
30cb3ee2
TM		 427 if (lo && pnfs_layout_is_valid(lo) &&
428 nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
429 /* Is our call using the most recent seqid? If so, bump it */
430 if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
431 nfs4_stateid_seqid_inc(dst);
432 ret = true;
433 goto out;
434 }
435 /* Try to update the seqid to the most recent */
c16467dc
TM		 436 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
437 if (err != -EBUSY) {
438 dst->seqid = lo->plh_stateid.seqid;
ecf84026 439 *dst_range = range;
c16467dc
TM		 440 ret = true;
441 }
7380020e 442 }
30cb3ee2 443out:
7380020e 444 spin_unlock(&inode->i_lock);
c16467dc 445 pnfs_free_lseg_list(&head);
7380020e
TM		 446 return ret;
447}
448
2454dfea
TM		 449/*
450 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
451 *
452 * In order to continue using the pnfs_layout_hdr, a full recovery
453 * is required.
454 * Note that caller must hold inode->i_lock.
455 */
5f46be04 456int
2454dfea
TM		 457pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
458 struct list_head *lseg_list)
459{
460 struct pnfs_layout_range range = {
461 .iomode = IOMODE_ANY,
462 .offset = 0,
463 .length = NFS4_MAX_UINT64,
464 };
fe1cf946 465 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 466
467 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 468 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
469 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 470 pnfs_clear_layoutreturn_info(lo);
68f74479 471 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
880265c7 472 set_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags);
362fb578
TM		 473 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
474 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
475 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 476 return !list_empty(&lo->plh_segs);
2454dfea
TM		 477}
478
b9e028fd
TM		 479static int
480pnfs_iomode_to_fail_bit(u32 iomode)
481{
482 return iomode == IOMODE_RW ?
483 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
484}
485
486static void
3e621214 487pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 488{
25c75333 489 lo->plh_retry_timestamp = jiffies;
39e88fcf 490 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 491 refcount_inc(&lo->plh_refcount);
3e621214
TM		 492}
493
494static void
495pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
496{
497 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 498 refcount_dec(&lo->plh_refcount);
3e621214
TM		 499}
500
501static void
502pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
503{
504 struct inode *inode = lo->plh_inode;
115ce575
TM		 505 struct pnfs_layout_range range = {
506 .iomode = iomode,
507 .offset = 0,
508 .length = NFS4_MAX_UINT64,
509 };
510 LIST_HEAD(head);
3e621214
TM		 511
512 spin_lock(&inode->i_lock);
513 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
28d4411f 514 pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
3e621214 515 spin_unlock(&inode->i_lock);
115ce575 516 pnfs_free_lseg_list(&head);
b9e028fd
TM		 517 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
518 iomode == IOMODE_RW ? "RW" : "READ");
519}
520
521static bool
522pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
523{
25c75333 524 unsigned long start, end;
3e621214
TM		 525 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
526
527 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 528 return false;
529 end = jiffies;
530 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
531 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
532 /* It is time to retry the failed layoutgets */
3e621214 533 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 534 return false;
535 }
536 return true;
b9e028fd
TM		 537}
538
974cec8c 539static void
119cef97
TM		 540pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
541 const struct pnfs_layout_range *range,
542 const nfs4_stateid *stateid)
974cec8c 543{
566052c5 544 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 545 INIT_LIST_HEAD(&lseg->pls_lc_list);
a9901899 546 INIT_LIST_HEAD(&lseg->pls_commits);
eba6dd69 547 refcount_set(&lseg->pls_refcount, 1);
4541d16c 548 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 549 lseg->pls_layout = lo;
119cef97
TM		 550 lseg->pls_range = *range;
551 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 552}
553
905ca191 554static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 555{
68f74479
TM		 556 if (lseg != NULL) {
557 struct inode *inode = lseg->pls_layout->plh_inode;
558 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
559 }
974cec8c
AA		 560}
561
d684d2ae 562static void
57036a37
TM		 563pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
564 struct pnfs_layout_segment *lseg)
d684d2ae 565{
d20581aa 566 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 567 list_del_init(&lseg->pls_list);
8f0d27dc 568 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 569 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 570 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
571 return;
7b650994
TM		 572 if (list_empty(&lo->plh_segs) &&
573 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
574 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 575 if (atomic_read(&lo->plh_outstanding) == 0)
576 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 577 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 578 }
d684d2ae
FI		 579}
580
68f74479
TM		 581static bool
582pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
583 struct pnfs_layout_segment *lseg)
584{
585 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
586 pnfs_layout_is_valid(lo)) {
4aab9732 587 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 588 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
589 return true;
590 }
591 return false;
592}
593
bae724ef 594void
9369a431 595pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 596{
57036a37 597 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 598 struct inode *inode;
599
600 if (!lseg)
601 return;
602
4541d16c 603 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 604 refcount_read(&lseg->pls_refcount),
4541d16c 605 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 606
57036a37
TM		 607 lo = lseg->pls_layout;
608 inode = lo->plh_inode;
4ef2e4f8 609
eba6dd69 610 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
8f0d27dc 611 pnfs_get_layout_hdr(lo);
4ef2e4f8 612 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 613 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
614 lseg = NULL;
4ef2e4f8
TM		 615 spin_unlock(&inode->i_lock);
616 pnfs_free_lseg(lseg);
617 pnfs_put_layout_hdr(lo);
4541d16c 618 }
4541d16c 619}
9369a431 620EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 621
fb3296eb
BH		 622/*
623 * is l2 fully contained in l1?
624 * start1 end1
625 * [----------------------------------)
626 * start2 end2
627 * [----------------)
628 */
3cb2df17 629static bool
7dc0ac70 630pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 631 const struct pnfs_layout_range *l2)
fb3296eb
BH		 632{
633 u64 start1 = l1->offset;
17822b20 634 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 635 u64 start2 = l2->offset;
17822b20 636 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 637
638 return (start1 <= start2) && (end1 >= end2);
639}
640
24956804
TM		 641static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
642 struct list_head *tmp_list)
643{
eba6dd69 644 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 645 return false;
646 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
647 list_add(&lseg->pls_list, tmp_list);
648 return true;
649}
650
4541d16c
FI		 651/* Returns 1 if lseg is removed from list, 0 otherwise */
652static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
653 struct list_head *tmp_list)
654{
655 int rv = 0;
656
657 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
658 /* Remove the reference keeping the lseg in the
659 * list. It will now be removed when all
660 * outstanding io is finished.
661 */
d684d2ae 662 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 663 refcount_read(&lseg->pls_refcount));
24956804 664 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 665 rv = 1;
4541d16c
FI		 666 }
667 return rv;
668}
669
e036f464
TM		 670static bool
671pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
672 const struct pnfs_layout_range *recall_range)
673{
674 return (recall_range->iomode == IOMODE_ANY ||
675 lseg_range->iomode == recall_range->iomode) &&
676 pnfs_lseg_range_intersecting(lseg_range, recall_range);
677}
678
679static bool
680pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
681 const struct pnfs_layout_range *recall_range,
682 u32 seq)
683{
684 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
685 return false;
686 if (recall_range == NULL)
687 return true;
688 return pnfs_should_free_range(&lseg->pls_range, recall_range);
689}
690
6d597e17
JL		 691/**
692 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
693 * @lo: layout header containing the lsegs
694 * @tmp_list: list head where doomed lsegs should go
695 * @recall_range: optional recall range argument to match (may be NULL)
696 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
697 *
698 * Walk the list of lsegs in the layout header, and tear down any that should
699 * be destroyed. If "recall_range" is specified then the segment must match
700 * that range. If "seq" is non-zero, then only match segments that were handed
701 * out at or before that sequence.
702 *
703 * Returns number of matching invalid lsegs remaining in list after scanning
704 * it and purging them.
4541d16c 705 */
43f1b3da 706int
49a85061 707pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 708 struct list_head *tmp_list,
6d597e17
JL		 709 const struct pnfs_layout_range *recall_range,
710 u32 seq)
974cec8c
AA		 711{
712 struct pnfs_layout_segment *lseg, *next;
b739a5bd 713 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
71b39854 714 int remaining = 0;
974cec8c
AA		 715
716 dprintk("%s:Begin lo %p\n", __func__, lo);
717
8006bfba 718 if (list_empty(&lo->plh_segs))
38511722 719 return 0;
4541d16c 720 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 721 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 722 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 723 "offset %llu length %llu\n", __func__,
6d597e17
JL		 724 lseg, lseg->pls_range.iomode, lseg->pls_seq,
725 lseg->pls_range.offset, lseg->pls_range.length);
b739a5bd
TM		 726 if (mark_lseg_invalid(lseg, tmp_list))
727 continue;
728 remaining++;
729 pnfs_lseg_cancel_io(server, lseg);
4541d16c 730 }
71b39854
TM		 731 dprintk("%s:Return %i\n", __func__, remaining);
732 return remaining;
974cec8c
AA		 733}
734
68f74479
TM		 735static void
736pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
737 struct list_head *free_me,
738 const struct pnfs_layout_range *range,
739 u32 seq)
740{
741 struct pnfs_layout_segment *lseg, *next;
742
743 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
744 if (pnfs_match_lseg_recall(lseg, range, seq))
745 list_move_tail(&lseg->pls_list, free_me);
746 }
747}
748
f49f9baa 749/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 750void
4541d16c 751pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 752{
4541d16c 753 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 754
755 if (list_empty(free_me))
756 return;
757
4541d16c 758 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 759 list_del(&lseg->pls_list);
905ca191 760 pnfs_free_lseg(lseg);
974cec8c
AA		 761 }
762}
763
b6d49ecd 764static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi)
e5e94017
BH		 765{
766 struct pnfs_layout_hdr *lo;
974cec8c 767 LIST_HEAD(tmp_list);
e5e94017
BH		 768
769 spin_lock(&nfsi->vfs_inode.i_lock);
770 lo = nfsi->layout;
771 if (lo) {
3e621214 772 pnfs_get_layout_hdr(lo);
2454dfea 773 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 774 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
775 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
776 spin_unlock(&nfsi->vfs_inode.i_lock);
777 pnfs_free_lseg_list(&tmp_list);
1f18b82c 778 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 779 pnfs_put_layout_hdr(lo);
780 } else
781 spin_unlock(&nfsi->vfs_inode.i_lock);
b6d49ecd
TM		 782 return lo;
783}
784
785void pnfs_destroy_layout(struct nfs_inode *nfsi)
786{
787 __pnfs_destroy_layout(nfsi);
974cec8c 788}
041245c8 789EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 790
b6d49ecd
TM		 791static bool pnfs_layout_removed(struct nfs_inode *nfsi,
792 struct pnfs_layout_hdr *lo)
793{
794 bool ret;
795
796 spin_lock(&nfsi->vfs_inode.i_lock);
797 ret = nfsi->layout != lo;
798 spin_unlock(&nfsi->vfs_inode.i_lock);
799 return ret;
800}
801
802void pnfs_destroy_layout_final(struct nfs_inode *nfsi)
803{
804 struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi);
805
806 if (lo)
807 wait_var_event(lo, pnfs_layout_removed(nfsi, lo));
808}
809
fd9a8d71
TM		 810static bool
811pnfs_layout_add_bulk_destroy_list(struct inode *inode,
812 struct list_head *layout_list)
974cec8c
AA		 813{
814 struct pnfs_layout_hdr *lo;
fd9a8d71 815 bool ret = false;
974cec8c 816
fd9a8d71
TM		 817 spin_lock(&inode->i_lock);
818 lo = NFS_I(inode)->layout;
819 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
820 pnfs_get_layout_hdr(lo);
821 list_add(&lo->plh_bulk_destroy, layout_list);
822 ret = true;
823 }
824 spin_unlock(&inode->i_lock);
825 return ret;
826}
827
828/* Caller must hold rcu_read_lock and clp->cl_lock */
829static int
830pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
831 struct nfs_server *server,
832 struct list_head *layout_list)
5085607d
TM		 833 __must_hold(&clp->cl_lock)
834 __must_hold(RCU)
fd9a8d71
TM		 835{
836 struct pnfs_layout_hdr *lo, *next;
837 struct inode *inode;
838
839 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
5085607d
TM		 840 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
841 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
842 !list_empty(&lo->plh_bulk_destroy))
b85f5620 843 continue;
5085607d
TM		 844 /* If the sb is being destroyed, just bail */
845 if (!nfs_sb_active(server->super))
846 break;
b5fdf841 847 inode = pnfs_grab_inode_layout_hdr(lo);
5085607d 848 if (inode != NULL) {
cf6605d1
TM		 849 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
850 list_del_rcu(&lo->plh_layouts);
5085607d
TM		 851 if (pnfs_layout_add_bulk_destroy_list(inode,
852 layout_list))
853 continue;
854 rcu_read_unlock();
855 spin_unlock(&clp->cl_lock);
856 iput(inode);
857 } else {
858 rcu_read_unlock();
859 spin_unlock(&clp->cl_lock);
5085607d
TM		 860 }
861 nfs_sb_deactive(server->super);
fd9a8d71
TM		 862 spin_lock(&clp->cl_lock);
863 rcu_read_lock();
864 return -EAGAIN;
865 }
866 return 0;
867}
868
869static int
870pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
871 bool is_bulk_recall)
872{
873 struct pnfs_layout_hdr *lo;
874 struct inode *inode;
fd9a8d71
TM		 875 LIST_HEAD(lseg_list);
876 int ret = 0;
877
878 while (!list_empty(layout_list)) {
879 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
880 plh_bulk_destroy);
881 dprintk("%s freeing layout for inode %lu\n", __func__,
882 lo->plh_inode->i_ino);
883 inode = lo->plh_inode;
7c5d1875
CH		 884
885 pnfs_layoutcommit_inode(inode, false);
886
fd9a8d71
TM		 887 spin_lock(&inode->i_lock);
888 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 889 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
890 if (is_bulk_recall)
891 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 892 ret = -EAGAIN;
9fd4b9fc 893 }
fd9a8d71
TM		 894 spin_unlock(&inode->i_lock);
895 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 896 /* Free all lsegs that are attached to commit buckets */
897 nfs_commit_inode(inode, 0);
fd9a8d71 898 pnfs_put_layout_hdr(lo);
5085607d 899 nfs_iput_and_deactive(inode);
fd9a8d71
TM		 900 }
901 return ret;
902}
903
904int
905pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
906 struct nfs_fsid *fsid,
907 bool is_recall)
908{
909 struct nfs_server *server;
910 LIST_HEAD(layout_list);
c47abcf8 911
974cec8c 912 spin_lock(&clp->cl_lock);
6382a441 913 rcu_read_lock();
fd9a8d71 914restart:
6382a441 915 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 916 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
917 continue;
918 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
919 server,
920 &layout_list) != 0)
921 goto restart;
6382a441
WAA		 922 }
923 rcu_read_unlock();
974cec8c
AA		 924 spin_unlock(&clp->cl_lock);
925
fd9a8d71
TM		 926 if (list_empty(&layout_list))
927 return 0;
928 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
929}
930
931int
932pnfs_destroy_layouts_byclid(struct nfs_client *clp,
933 bool is_recall)
934{
935 struct nfs_server *server;
936 LIST_HEAD(layout_list);
937
938 spin_lock(&clp->cl_lock);
939 rcu_read_lock();
940restart:
941 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
942 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
943 server,
944 &layout_list) != 0)
945 goto restart;
974cec8c 946 }
fd9a8d71
TM		 947 rcu_read_unlock();
948 spin_unlock(&clp->cl_lock);
949
950 if (list_empty(&layout_list))
951 return 0;
952 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
953}
954
955/*
9f266451 956 * Called by the state manager to remove all layouts established under an
fd9a8d71
TM		 957 * expired lease.
958 */
959void
960pnfs_destroy_all_layouts(struct nfs_client *clp)
961{
962 nfs4_deviceid_mark_client_invalid(clp);
963 nfs4_deviceid_purge_client(clp);
964
965 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 966}
967
59b56394
TM		 968static void
969pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
970{
971 const struct cred *old;
972
973 if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
974 old = xchg(&lo->plh_lc_cred, get_cred(cred));
975 put_cred(old);
976 }
977}
978
fd6002e9 979/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 980void
981pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
59b56394 982 const struct cred *cred, bool update_barrier)
b1f69b75 983{
aa95edf3
TM		 984 u32 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
985 u32 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 986
987 if (!pnfs_layout_is_valid(lo)) {
59b56394 988 pnfs_set_layout_cred(lo, cred);
2a59a041
TM		 989 nfs4_stateid_copy(&lo->plh_stateid, new);
990 lo->plh_barrier = newseq;
991 pnfs_clear_layoutreturn_info(lo);
992 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
993 return;
994 }
aa95edf3
TM		 995
996 if (pnfs_seqid_is_newer(newseq, oldseq))
f597c537 997 nfs4_stateid_copy(&lo->plh_stateid, new);
aa95edf3
TM		 998
999 if (update_barrier) {
1000 pnfs_barrier_update(lo, newseq);
ecebb80b 1001 return;
aa95edf3
TM		 1002 }
1003 /*
1004 * Because of wraparound, we want to keep the barrier
1005 * "close" to the current seqids. We really only want to
1006 * get here from a layoutget call.
1007 */
1008 if (atomic_read(&lo->plh_outstanding) == 1)
1009 pnfs_barrier_update(lo, be32_to_cpu(lo->plh_stateid.seqid));
b1f69b75
AA		 1010}
1011
cf7d63f1 1012static bool
19c54aba
TM		 1013pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
1014 const nfs4_stateid *stateid)
43f1b3da 1015{
19c54aba 1016 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 1017
d6236a98 1018 return lo->plh_barrier && pnfs_seqid_is_newer(lo->plh_barrier, seqid);
19c54aba
TM		 1019}
1020
1021/* lget is set to 1 if called from inside send_layoutget call chain */
1022static bool
e1c06f80 1023pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 1024{
f7e8917a 1025 return lo->plh_block_lgets ||
e1c06f80 1026 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 1027}
1028
5e36e2a9
FI		 1029static struct nfs_server *
1030pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
1031{
1032 struct nfs_server *server;
1033
78746a38 1034 if (inode) {
5e36e2a9 1035 server = NFS_SERVER(inode);
78746a38 1036 } else {
5e36e2a9
FI		 1037 struct dentry *parent_dir = dget_parent(ctx->dentry);
1038 server = NFS_SERVER(parent_dir->d_inode);
1039 dput(parent_dir);
1040 }
1041 return server;
1042}
1043
29a8bfe5
TM		 1044static void nfs4_free_pages(struct page **pages, size_t size)
1045{
1046 int i;
1047
1048 if (!pages)
1049 return;
1050
1051 for (i = 0; i < size; i++) {
1052 if (!pages[i])
1053 break;
1054 __free_page(pages[i]);
1055 }
1056 kfree(pages);
1057}
1058
1059static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1060{
1061 struct page **pages;
1062 int i;
1063
a2791d3a 1064 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 1065 if (!pages) {
1066 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1067 return NULL;
1068 }
1069
1070 for (i = 0; i < size; i++) {
1071 pages[i] = alloc_page(gfp_flags);
1072 if (!pages[i]) {
1073 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 1074 nfs4_free_pages(pages, i);
29a8bfe5
TM		 1075 return NULL;
1076 }
1077 }
1078
1079 return pages;
1080}
1081
587f03de 1082static struct nfs4_layoutget *
5e36e2a9 1083pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 1084 struct nfs_open_context *ctx,
2409a976 1085 const nfs4_stateid *stateid,
e144e539 1086 const struct pnfs_layout_range *range,
587f03de 1087 gfp_t gfp_flags)
e5e94017 1088{
5e36e2a9 1089 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 1090 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 1091 size_t max_pages = max_response_pages(server);
b1f69b75 1092 struct nfs4_layoutget *lgp;
b1f69b75
AA		 1093
1094 dprintk("--> %s\n", __func__);
e5e94017 1095
83026d80
JL		 1096 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1097 if (lgp == NULL)
587f03de 1098 return NULL;
83026d80 1099
28ced9a8
TM		 1100 if (max_reply_sz) {
1101 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1102 if (npages < max_pages)
1103 max_pages = npages;
1104 }
1105
dacb452d
FI		 1106 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1107 if (!lgp->args.layout.pages) {
1108 kfree(lgp);
1109 return NULL;
1110 }
1111 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1112 lgp->res.layoutp = &lgp->args.layout;
1113
d49e0d5b
TM		 1114 /* Don't confuse uninitialised result and success */
1115 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1116
1117 lgp->args.minlength = PAGE_SIZE;
1118 if (lgp->args.minlength > range->length)
1119 lgp->args.minlength = range->length;
5e36e2a9
FI		 1120 if (ino) {
1121 loff_t i_size = i_size_read(ino);
1122
1123 if (range->iomode == IOMODE_READ) {
1124 if (range->offset >= i_size)
1125 lgp->args.minlength = 0;
1126 else if (i_size - range->offset < lgp->args.minlength)
1127 lgp->args.minlength = i_size - range->offset;
1128 }
d03ab29d 1129 }
83026d80
JL		 1130 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1131 pnfs_copy_range(&lgp->args.range, range);
1132 lgp->args.type = server->pnfs_curr_ld->id;
1133 lgp->args.inode = ino;
1134 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1135 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1136 lgp->gfp_flags = gfp_flags;
63ec2b69 1137 lgp->cred = ctx->cred;
587f03de 1138 return lgp;
974cec8c
AA		 1139}
1140
29a8bfe5
TM		 1141void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1142{
1143 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1144
1145 nfs4_free_pages(lgp->args.layout.pages, max_pages);
b4e89bcb 1146 pnfs_put_layout_hdr(lgp->lo);
29a8bfe5
TM		 1147 put_nfs_open_context(lgp->args.ctx);
1148 kfree(lgp);
1149}
1150
24956804
TM		 1151static void pnfs_clear_layoutcommit(struct inode *inode,
1152 struct list_head *head)
1153{
1154 struct nfs_inode *nfsi = NFS_I(inode);
1155 struct pnfs_layout_segment *lseg, *tmp;
1156
1157 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1158 return;
1159 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1160 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1161 continue;
1162 pnfs_lseg_dec_and_remove_zero(lseg, head);
1163 }
1164}
1165
68f74479 1166void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1167 const nfs4_stateid *arg_stateid,
68f74479 1168 const struct pnfs_layout_range *range,
68f74479
TM		 1169 const nfs4_stateid *stateid)
1170{
1171 struct inode *inode = lo->plh_inode;
1172 LIST_HEAD(freeme);
1173
1174 spin_lock(&inode->i_lock);
c18d1e17 1175 if (!pnfs_layout_is_valid(lo) ||
2a974425
TM		 1176 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1177 goto out_unlock;
68f74479 1178 if (stateid) {
2a974425
TM		 1179 u32 seq = be32_to_cpu(arg_stateid->seqid);
1180
68f74479
TM		 1181 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1182 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
59b56394 1183 pnfs_set_layout_stateid(lo, stateid, NULL, true);
68f74479
TM		 1184 } else
1185 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1186out_unlock:
68f74479
TM		 1187 pnfs_clear_layoutreturn_waitbit(lo);
1188 spin_unlock(&inode->i_lock);
1189 pnfs_free_lseg_list(&freeme);
1190
1191}
1192
13c13a6a 1193static bool
e5fd1904
TM		 1194pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1195 nfs4_stateid *stateid,
44ea8dfc 1196 const struct cred **cred,
e5fd1904 1197 enum pnfs_iomode *iomode)
13c13a6a 1198{
bf0291dd
TM		 1199 /* Serialise LAYOUTGET/LAYOUTRETURN */
1200 if (atomic_read(&lo->plh_outstanding) != 0)
1201 return false;
6604b203 1202 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1203 return false;
6604b203 1204 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1205 pnfs_get_layout_hdr(lo);
1bcf34fd
TM		 1206 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1207 *cred = get_cred(lo->plh_lc_cred);
e5fd1904 1208 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
44ea8dfc
TM		 1209 if (lo->plh_return_seq != 0)
1210 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
e5fd1904
TM		 1211 if (iomode != NULL)
1212 *iomode = lo->plh_return_iomode;
1213 pnfs_clear_layoutreturn_info(lo);
1bcf34fd 1214 } else if (iomode != NULL)
e5fd1904 1215 *iomode = IOMODE_ANY;
1bcf34fd 1216 pnfs_barrier_update(lo, be32_to_cpu(stateid->seqid));
13c13a6a
TM		 1217 return true;
1218}
1219
828ed9ec
TM		 1220static void
1221pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1222 struct pnfs_layout_hdr *lo,
1223 const nfs4_stateid *stateid,
1224 enum pnfs_iomode iomode)
1225{
1226 struct inode *inode = lo->plh_inode;
1227
1228 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1229 args->inode = inode;
1230 args->range.iomode = iomode;
1231 args->range.offset = 0;
1232 args->range.length = NFS4_MAX_UINT64;
1233 args->layout = lo;
1234 nfs4_stateid_copy(&args->stateid, stateid);
1235}
1236
f40eb5d0 1237static int
44ea8dfc
TM		 1238pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
1239 const nfs4_stateid *stateid,
1240 const struct cred **pcred,
1241 enum pnfs_iomode iomode,
1242 bool sync)
f40eb5d0
PT		 1243{
1244 struct inode *ino = lo->plh_inode;
287bd3e9 1245 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0 1246 struct nfs4_layoutreturn *lrp;
44ea8dfc 1247 const struct cred *cred = *pcred;
f40eb5d0
PT		 1248 int status = 0;
1249
44ea8dfc 1250 *pcred = NULL;
63d8a41b 1251 lrp = kzalloc(sizeof(*lrp), nfs_io_gfp_mask());
f40eb5d0
PT		 1252 if (unlikely(lrp == NULL)) {
1253 status = -ENOMEM;
1254 spin_lock(&ino->i_lock);
d67ae825 1255 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0 1256 spin_unlock(&ino->i_lock);
44ea8dfc 1257 put_cred(cred);
f40eb5d0
PT		 1258 pnfs_put_layout_hdr(lo);
1259 goto out;
1260 }
1261
828ed9ec 1262 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1263 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0 1264 lrp->clp = NFS_SERVER(ino)->nfs_client;
44ea8dfc 1265 lrp->cred = cred;
287bd3e9
TM		 1266 if (ld->prepare_layoutreturn)
1267 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1268
6c16605d 1269 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1270out:
1271 dprintk("<-- %s status: %d\n", __func__, status);
1272 return status;
1273}
1274
d474f961
TM		 1275static bool
1276pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1277 enum pnfs_iomode iomode,
1278 u32 seq)
1279{
1280 struct pnfs_layout_range recall_range = {
1281 .length = NFS4_MAX_UINT64,
1282 .iomode = iomode,
1283 };
1284 return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1285 &recall_range, seq) != -EBUSY;
1286}
1287
13c13a6a
TM		 1288/* Return true if layoutreturn is needed */
1289static bool
1290pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1291{
2370abda 1292 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a 1293 return false;
d474f961
TM		 1294 return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1295 lo->plh_return_seq);
13c13a6a
TM		 1296}
1297
1298static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1299{
1300 struct inode *inode= lo->plh_inode;
1301
2370abda 1302 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1303 return;
1304 spin_lock(&inode->i_lock);
1305 if (pnfs_layout_need_return(lo)) {
44ea8dfc 1306 const struct cred *cred;
13c13a6a
TM		 1307 nfs4_stateid stateid;
1308 enum pnfs_iomode iomode;
1309 bool send;
1310
44ea8dfc 1311 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
13c13a6a
TM		 1312 spin_unlock(&inode->i_lock);
1313 if (send) {
1314 /* Send an async layoutreturn so we dont deadlock */
44ea8dfc 1315 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
13c13a6a
TM		 1316 }
1317 } else
1318 spin_unlock(&inode->i_lock);
1319}
1320
293b3b06
AA		 1321/*
1322 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1323 * when the layout segment list is empty.
1324 *
1325 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1326 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1327 * deviceid is marked invalid.
1328 */
cbe82603
BH		 1329int
1330_pnfs_return_layout(struct inode *ino)
1331{
1332 struct pnfs_layout_hdr *lo = NULL;
1333 struct nfs_inode *nfsi = NFS_I(ino);
a421d218
AS		 1334 struct pnfs_layout_range range = {
1335 .iomode = IOMODE_ANY,
1336 .offset = 0,
1337 .length = NFS4_MAX_UINT64,
1338 };
cbe82603 1339 LIST_HEAD(tmp_list);
44ea8dfc 1340 const struct cred *cred;
cbe82603 1341 nfs4_stateid stateid;
24408f52 1342 int status = 0;
93b7f7ad 1343 bool send, valid_layout;
cbe82603 1344
366d5052 1345 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1346
1347 spin_lock(&ino->i_lock);
1348 lo = nfsi->layout;
e5929f3c 1349 if (!lo) {
cbe82603 1350 spin_unlock(&ino->i_lock);
293b3b06
AA		 1351 dprintk("NFS: %s no layout to return\n", __func__);
1352 goto out;
cbe82603 1353 }
cbe82603 1354 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1355 pnfs_get_layout_hdr(lo);
24408f52
TM		 1356 /* Is there an outstanding layoutreturn ? */
1357 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1358 spin_unlock(&ino->i_lock);
1359 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1360 TASK_UNINTERRUPTIBLE))
1361 goto out_put_layout_hdr;
1362 spin_lock(&ino->i_lock);
1363 }
93b7f7ad 1364 valid_layout = pnfs_layout_is_valid(lo);
24956804 1365 pnfs_clear_layoutcommit(ino, &tmp_list);
a421d218
AS		 1366 pnfs_mark_matching_lsegs_return(lo, &tmp_list, &range, 0);
1367
1368 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range)
c88953d8 1369 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
c88953d8 1370
293b3b06 1371 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1372 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1373 !valid_layout) {
293b3b06 1374 spin_unlock(&ino->i_lock);
293b3b06 1375 dprintk("NFS: %s no layout segments to return\n", __func__);
7bcc1058 1376 goto out_wait_layoutreturn;
293b3b06 1377 }
47abadef 1378
44ea8dfc 1379 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
cbe82603 1380 spin_unlock(&ino->i_lock);
7f27392c 1381 if (send)
44ea8dfc 1382 status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
7bcc1058
TM		 1383out_wait_layoutreturn:
1384 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE);
7f27392c 1385out_put_layout_hdr:
ee6625a9 1386 pnfs_free_lseg_list(&tmp_list);
7f27392c 1387 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1388out:
1389 dprintk("<-- %s status: %d\n", __func__, status);
1390 return status;
1391}
1392
24028672
TM		 1393int
1394pnfs_commit_and_return_layout(struct inode *inode)
1395{
1396 struct pnfs_layout_hdr *lo;
1397 int ret;
1398
1399 spin_lock(&inode->i_lock);
1400 lo = NFS_I(inode)->layout;
1401 if (lo == NULL) {
1402 spin_unlock(&inode->i_lock);
1403 return 0;
1404 }
1405 pnfs_get_layout_hdr(lo);
1406 /* Block new layoutgets and read/write to ds */
1407 lo->plh_block_lgets++;
1408 spin_unlock(&inode->i_lock);
1409 filemap_fdatawait(inode->i_mapping);
1410 ret = pnfs_layoutcommit_inode(inode, true);
1411 if (ret == 0)
1412 ret = _pnfs_return_layout(inode);
1413 spin_lock(&inode->i_lock);
1414 lo->plh_block_lgets--;
1415 spin_unlock(&inode->i_lock);
1416 pnfs_put_layout_hdr(lo);
1417 return ret;
1418}
1419
1c5bd76d
TM		 1420bool pnfs_roc(struct inode *ino,
1421 struct nfs4_layoutreturn_args *args,
1422 struct nfs4_layoutreturn_res *res,
a52458b4 1423 const struct cred *cred)
f7e8917a 1424{
40dd4b7a
TM		 1425 struct nfs_inode *nfsi = NFS_I(ino);
1426 struct nfs_open_context *ctx;
1427 struct nfs4_state *state;
f7e8917a 1428 struct pnfs_layout_hdr *lo;
1c5bd76d 1429 struct pnfs_layout_segment *lseg, *next;
44ea8dfc 1430 const struct cred *lc_cred;
193e3aa2 1431 nfs4_stateid stateid;
1c5bd76d
TM		 1432 enum pnfs_iomode iomode = 0;
1433 bool layoutreturn = false, roc = false;
e71708d4 1434 bool skip_read = false;
f7e8917a 1435
1c5bd76d
TM		 1436 if (!nfs_have_layout(ino))
1437 return false;
29ade5db 1438retry:
0de43976 1439 rcu_read_lock();
f7e8917a 1440 spin_lock(&ino->i_lock);
40dd4b7a 1441 lo = nfsi->layout;
0cdc329e 1442 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1443 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1444 lo = NULL;
40dd4b7a 1445 goto out_noroc;
9c6376eb
TM		 1446 }
1447 pnfs_get_layout_hdr(lo);
29ade5db 1448 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1449 spin_unlock(&ino->i_lock);
0de43976 1450 rcu_read_unlock();
29ade5db
TM		 1451 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1452 TASK_UNINTERRUPTIBLE);
1453 pnfs_put_layout_hdr(lo);
1454 goto retry;
1455 }
40dd4b7a 1456
e755d638 1457 /* no roc if we hold a delegation */
e71708d4
TM		 1458 if (nfs4_check_delegation(ino, FMODE_READ)) {
1459 if (nfs4_check_delegation(ino, FMODE_WRITE))
1460 goto out_noroc;
1461 skip_read = true;
1462 }
40dd4b7a 1463
0de43976 1464 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1465 state = ctx->state;
e71708d4
TM		 1466 if (state == NULL)
1467 continue;
40dd4b7a 1468 /* Don't return layout if there is open file state */
e71708d4 1469 if (state->state & FMODE_WRITE)
40dd4b7a 1470 goto out_noroc;
e71708d4
TM		 1471 if (state->state & FMODE_READ)
1472 skip_read = true;
40dd4b7a
TM		 1473 }
1474
e755d638 1475
1c5bd76d 1476 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1477 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1478 continue;
e755d638 1479 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1480 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1481 continue;
1482 /*
1483 * Note: mark lseg for return so pnfs_layout_remove_lseg
1484 * doesn't invalidate the layout for us.
1485 */
1486 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1487 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1488 continue;
1489 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1490 }
1491
1c5bd76d
TM		 1492 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1493 goto out_noroc;
69820d22 1494
500d701f 1495 /* ROC in two conditions:
e755d638
PT		 1496 * 1. there are ROC lsegs
1497 * 2. we don't send layoutreturn
e755d638 1498 */
1c5bd76d 1499 /* lo ref dropped in pnfs_roc_release() */
44ea8dfc 1500 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
1c5bd76d 1501 /* If the creds don't match, we can't compound the layoutreturn */
4d8948c7 1502 if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0)
1c5bd76d
TM		 1503 goto out_noroc;
1504
1505 roc = layoutreturn;
1506 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1507 res->lrs_present = 0;
1508 layoutreturn = false;
44ea8dfc 1509 put_cred(lc_cred);
4d8948c7 1510
40dd4b7a 1511out_noroc:
f7e8917a 1512 spin_unlock(&ino->i_lock);
0de43976 1513 rcu_read_unlock();
e755d638 1514 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1515 if (roc) {
1516 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1517 if (ld->prepare_layoutreturn)
1518 ld->prepare_layoutreturn(args);
9c6376eb 1519 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1520 return true;
1521 }
e755d638 1522 if (layoutreturn)
44ea8dfc 1523 pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
9c6376eb 1524 pnfs_put_layout_hdr(lo);
287bd3e9 1525 return false;
f7e8917a
FI		 1526}
1527
078000d0
TM		 1528int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
1529 struct nfs4_layoutreturn_res **respp, int *ret)
287a9c55
TM		 1530{
1531 struct nfs4_layoutreturn_args *arg = *argpp;
1532 int retval = -EAGAIN;
1533
1534 if (!arg)
1535 return 0;
1536 /* Handle Layoutreturn errors */
1537 switch (*ret) {
1538 case 0:
1539 retval = 0;
1540 break;
6109bcf7
TM		 1541 case -NFS4ERR_NOMATCHING_LAYOUT:
1542 /* Was there an RPC level error? If not, retry */
1543 if (task->tk_rpc_status == 0)
1544 break;
1545 /* If the call was not sent, let caller handle it */
1546 if (!RPC_WAS_SENT(task))
1547 return 0;
1548 /*
1549 * Otherwise, assume the call succeeded and
1550 * that we need to release the layout
1551 */
1552 *ret = 0;
1553 (*respp)->lrs_present = 0;
1554 retval = 0;
1555 break;
078a432d
TM		 1556 case -NFS4ERR_DELAY:
1557 /* Let the caller handle the retry */
1558 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1559 return 0;
287a9c55 1560 case -NFS4ERR_OLD_STATEID:
30cb3ee2 1561 if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
078000d0 1562 &arg->range, arg->inode))
287a9c55
TM		 1563 break;
1564 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1565 return -EAGAIN;
1566 }
1567 *argpp = NULL;
1568 *respp = NULL;
1569 return retval;
1570}
1571
1c5bd76d
TM		 1572void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1573 struct nfs4_layoutreturn_res *res,
1574 int ret)
f7e8917a 1575{
1c5bd76d 1576 struct pnfs_layout_hdr *lo = args->layout;
67bbceed 1577 struct inode *inode = args->inode;
1c5bd76d 1578 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1579 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1580
9c47b18c
TM		 1581 switch (ret) {
1582 case -NFS4ERR_NOMATCHING_LAYOUT:
67bbceed
TM		 1583 spin_lock(&inode->i_lock);
1584 if (pnfs_layout_is_valid(lo) &&
1585 nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
1586 pnfs_set_plh_return_info(lo, args->range.iomode, 0);
c18d1e17 1587 pnfs_clear_layoutreturn_waitbit(lo);
67bbceed 1588 spin_unlock(&inode->i_lock);
9c47b18c
TM		 1589 break;
1590 case 0:
1c5bd76d
TM		 1591 if (res->lrs_present)
1592 res_stateid = &res->stateid;
df561f66 1593 fallthrough;
9c47b18c 1594 default:
c18d1e17
TM		 1595 pnfs_layoutreturn_free_lsegs(lo, &args->stateid, &args->range,
1596 res_stateid);
1c5bd76d 1597 }
a19b4785 1598 trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
287bd3e9
TM		 1599 if (ld_private && ld_private->ops && ld_private->ops->free)
1600 ld_private->ops->free(ld_private);
1c5bd76d 1601 pnfs_put_layout_hdr(lo);
f7e8917a
FI		 1602}
1603
500d701f
PT		 1604bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1605{
1606 struct nfs_inode *nfsi = NFS_I(ino);
1607 struct pnfs_layout_hdr *lo;
1608 bool sleep = false;
1609
1610 /* we might not have grabbed lo reference. so need to check under
1611 * i_lock */
1612 spin_lock(&ino->i_lock);
1613 lo = nfsi->layout;
ee284e35
TM		 1614 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1615 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1616 sleep = true;
ee284e35 1617 }
500d701f 1618 spin_unlock(&ino->i_lock);
500d701f
PT		 1619 return sleep;
1620}
1621
b1f69b75
AA		 1622/*
1623 * Compare two layout segments for sorting into layout cache.
1624 * We want to preferentially return RW over RO layouts, so ensure those
1625 * are seen first.
1626 */
1627static s64
7dc0ac70 1628pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1629 const struct pnfs_layout_range *l2)
b1f69b75 1630{
fb3296eb
BH		 1631 s64 d;
1632
1633 /* high offset > low offset */
1634 d = l1->offset - l2->offset;
1635 if (d)
1636 return d;
1637
1638 /* short length > long length */
1639 d = l2->length - l1->length;
1640 if (d)
1641 return d;
1642
b1f69b75 1643 /* read > read/write */
fb3296eb 1644 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1645}
1646
03772d2f
TM		 1647static bool
1648pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1649 const struct pnfs_layout_range *l2)
1650{
1651 return pnfs_lseg_range_cmp(l1, l2) > 0;
1652}
1653
1654static bool
1655pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1656 struct pnfs_layout_segment *old)
1657{
1658 return false;
1659}
1660
1661void
1662pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1663 struct pnfs_layout_segment *lseg,
1664 bool (*is_after)(const struct pnfs_layout_range *,
1665 const struct pnfs_layout_range *),
1666 bool (*do_merge)(struct pnfs_layout_segment *,
1667 struct pnfs_layout_segment *),
1668 struct list_head *free_me)
974cec8c 1669{
03772d2f 1670 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1671
974cec8c
AA		 1672 dprintk("%s:Begin\n", __func__);
1673
03772d2f
TM		 1674 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1675 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1676 continue;
1677 if (do_merge(lseg, lp)) {
1678 mark_lseg_invalid(lp, free_me);
1679 continue;
1680 }
1681 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1682 continue;
566052c5 1683 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1684 dprintk("%s: inserted lseg %p "
1685 "iomode %d offset %llu length %llu before "
1686 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1687 __func__, lseg, lseg->pls_range.iomode,
1688 lseg->pls_range.offset, lseg->pls_range.length,
1689 lp, lp->pls_range.iomode, lp->pls_range.offset,
1690 lp->pls_range.length);
fb3296eb 1691 goto out;
974cec8c 1692 }
fb3296eb
BH		 1693 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1694 dprintk("%s: inserted lseg %p "
1695 "iomode %d offset %llu length %llu at tail\n",
1696 __func__, lseg, lseg->pls_range.iomode,
1697 lseg->pls_range.offset, lseg->pls_range.length);
1698out:
70c3bd2b 1699 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1700
1701 dprintk("%s:Return\n", __func__);
e5e94017 1702}
03772d2f
TM		 1703EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1704
1705static void
1706pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1707 struct pnfs_layout_segment *lseg,
1708 struct list_head *free_me)
1709{
1710 struct inode *inode = lo->plh_inode;
1711 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1712
1713 if (ld->add_lseg != NULL)
1714 ld->add_lseg(lo, lseg, free_me);
1715 else
1716 pnfs_generic_layout_insert_lseg(lo, lseg,
1717 pnfs_lseg_range_is_after,
1718 pnfs_lseg_no_merge,
1719 free_me);
1720}
e5e94017
BH		 1721
1722static struct pnfs_layout_hdr *
9fa40758
PT		 1723alloc_init_layout_hdr(struct inode *ino,
1724 struct nfs_open_context *ctx,
1725 gfp_t gfp_flags)
e5e94017
BH		 1726{
1727 struct pnfs_layout_hdr *lo;
1728
636fb9c8 1729 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1730 if (!lo)
1731 return NULL;
2b28a7be 1732 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1733 INIT_LIST_HEAD(&lo->plh_layouts);
1734 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1735 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1736 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1737 lo->plh_inode = ino;
a52458b4 1738 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1739 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1740 return lo;
1741}
1742
1743static struct pnfs_layout_hdr *
9fa40758
PT		 1744pnfs_find_alloc_layout(struct inode *ino,
1745 struct nfs_open_context *ctx,
1746 gfp_t gfp_flags)
e5241e43
TM		 1747 __releases(&ino->i_lock)
1748 __acquires(&ino->i_lock)
e5e94017
BH		 1749{
1750 struct nfs_inode *nfsi = NFS_I(ino);
1751 struct pnfs_layout_hdr *new = NULL;
1752
1753 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1754
251ec410
TM		 1755 if (nfsi->layout != NULL)
1756 goto out_existing;
e5e94017 1757 spin_unlock(&ino->i_lock);
9fa40758 1758 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1759 spin_lock(&ino->i_lock);
1760
251ec410 1761 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1762 nfsi->layout = new;
251ec410 1763 return new;
7175fe90
YN		 1764 } else if (new != NULL)
1765 pnfs_free_layout_hdr(new);
251ec410
TM		 1766out_existing:
1767 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1768 return nfsi->layout;
1769}
1770
b1f69b75
AA		 1771/*
1772 * iomode matching rules:
c7d73af2
TH		 1773 * iomode lseg strict match
1774 * iomode
1775 * ----- ----- ------ -----
1776 * ANY READ N/A true
1777 * ANY RW N/A true
1778 * RW READ N/A false
1779 * RW RW N/A true
1780 * READ READ N/A true
1781 * READ RW true false
1782 * READ RW false true
b1f69b75 1783 */
3cb2df17 1784static bool
7dc0ac70 1785pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1786 const struct pnfs_layout_range *range,
1787 bool strict_iomode)
b1f69b75 1788{
fb3296eb
BH		 1789 struct pnfs_layout_range range1;
1790
1791 if ((range->iomode == IOMODE_RW &&
1792 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1793 (range->iomode != ls_range->iomode &&
6089dd0d 1794 strict_iomode) ||
7dc0ac70 1795 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1796 return false;
fb3296eb
BH		 1797
1798 /* range1 covers only the first byte in the range */
1799 range1 = *range;
1800 range1.length = 1;
7dc0ac70 1801 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1802}
1803
1804/*
1805 * lookup range in layout
1806 */
e5e94017 1807static struct pnfs_layout_segment *
fb3296eb 1808pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1809 struct pnfs_layout_range *range,
1810 bool strict_iomode)
e5e94017 1811{
b1f69b75
AA		 1812 struct pnfs_layout_segment *lseg, *ret = NULL;
1813
1814 dprintk("%s:Begin\n", __func__);
1815
b7edfaa1 1816 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1817 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1818 pnfs_lseg_range_match(&lseg->pls_range, range,
1819 strict_iomode)) {
9369a431 1820 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1821 break;
1822 }
b1f69b75
AA		 1823 }
1824
1825 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1826 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1827 return ret;
e5e94017
BH		 1828}
1829
d23d61c8
AA		 1830/*
1831 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1832 * to the MDS or over pNFS
1833 *
1834 * The nfs_inode read_io and write_io fields are cumulative counters reset
1835 * when there are no layout segments. Note that in pnfs_update_layout iomode
1836 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1837 * WRITE request.
1838 *
1839 * A return of true means use MDS I/O.
1840 *
1841 * From rfc 5661:
1842 * If a file's size is smaller than the file size threshold, data accesses
1843 * SHOULD be sent to the metadata server. If an I/O request has a length that
1844 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1845 * server. If both file size and I/O size are provided, the client SHOULD
1846 * reach or exceed both thresholds before sending its read or write
1847 * requests to the data server.
1848 */
1849static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1850 struct inode *ino, int iomode)
1851{
1852 struct nfs4_threshold *t = ctx->mdsthreshold;
1853 struct nfs_inode *nfsi = NFS_I(ino);
1854 loff_t fsize = i_size_read(ino);
1855 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1856
1857 if (t == NULL)
1858 return ret;
1859
1860 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1861 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1862
1863 switch (iomode) {
1864 case IOMODE_READ:
1865 if (t->bm & THRESHOLD_RD) {
1866 dprintk("%s fsize %llu\n", __func__, fsize);
1867 size_set = true;
1868 if (fsize < t->rd_sz)
1869 size = true;
1870 }
1871 if (t->bm & THRESHOLD_RD_IO) {
1872 dprintk("%s nfsi->read_io %llu\n", __func__,
1873 nfsi->read_io);
1874 io_set = true;
1875 if (nfsi->read_io < t->rd_io_sz)
1876 io = true;
1877 }
1878 break;
1879 case IOMODE_RW:
1880 if (t->bm & THRESHOLD_WR) {
1881 dprintk("%s fsize %llu\n", __func__, fsize);
1882 size_set = true;
1883 if (fsize < t->wr_sz)
1884 size = true;
1885 }
1886 if (t->bm & THRESHOLD_WR_IO) {
1887 dprintk("%s nfsi->write_io %llu\n", __func__,
1888 nfsi->write_io);
1889 io_set = true;
1890 if (nfsi->write_io < t->wr_io_sz)
1891 io = true;
1892 }
1893 break;
1894 }
1895 if (size_set && io_set) {
1896 if (size && io)
1897 ret = true;
1898 } else if (size || io)
1899 ret = true;
1900
1901 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1902 return ret;
1903}
1904
d03360aa 1905static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1906{
1907 /*
1908 * send layoutcommit as it can hold up layoutreturn due to lseg
1909 * reference
1910 */
1911 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1912 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1913 nfs_wait_bit_killable,
f5d39b02 1914 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
aa8a45ee
PT		 1915}
1916
411ae722
TM		 1917static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1918{
1919 atomic_inc(&lo->plh_outstanding);
1920}
1921
1922static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1923{
880265c7
TM		 1924 if (atomic_dec_and_test(&lo->plh_outstanding) &&
1925 test_and_clear_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags))
1926 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN);
411ae722
TM		 1927}
1928
d29b468d
TM		 1929static bool pnfs_is_first_layoutget(struct pnfs_layout_hdr *lo)
1930{
1931 return test_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags);
1932}
1933
d67ae825
TH		 1934static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1935{
1936 unsigned long *bitlock = &lo->plh_flags;
1937
1938 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1939 smp_mb__after_atomic();
1940 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1941}
1942
78746a38
FI		 1943static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1944 struct nfs_server *server)
1945{
cf6605d1 1946 if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
78746a38
FI		 1947 struct nfs_client *clp = server->nfs_client;
1948
1949 /* The lo must be on the clp list if there is any
1950 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1951 */
1952 spin_lock(&clp->cl_lock);
cf6605d1 1953 list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
78746a38
FI		 1954 spin_unlock(&clp->cl_lock);
1955 }
1956}
1957
e5e94017
BH		 1958/*
1959 * Layout segment is retreived from the server if not cached.
1960 * The appropriate layout segment is referenced and returned to the caller.
1961 */
7c24d948 1962struct pnfs_layout_segment *
e5e94017
BH		 1963pnfs_update_layout(struct inode *ino,
1964 struct nfs_open_context *ctx,
fb3296eb
BH		 1965 loff_t pos,
1966 u64 count,
a75b9df9 1967 enum pnfs_iomode iomode,
c7d73af2 1968 bool strict_iomode,
a75b9df9 1969 gfp_t gfp_flags)
e5e94017 1970{
fb3296eb
BH		 1971 struct pnfs_layout_range arg = {
1972 .iomode = iomode,
1973 .offset = pos,
1974 .length = count,
1975 };
70d2f7b1 1976 unsigned pg_offset;
6382a441
WAA		 1977 struct nfs_server *server = NFS_SERVER(ino);
1978 struct nfs_client *clp = server->nfs_client;
183d9e7b 1979 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1980 struct pnfs_layout_segment *lseg = NULL;
587f03de 1981 struct nfs4_layoutget *lgp;
183d9e7b 1982 nfs4_stateid stateid;
6e7434ab
TM		 1983 struct nfs4_exception exception = {
1984 .inode = ino,
1985 };
66b53f32 1986 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1987 bool first;
e5e94017 1988
9a4bf31d 1989 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1990 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1991 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1992 goto out;
9a4bf31d 1993 }
d23d61c8 1994
9a4bf31d 1995 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1996 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1997 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1998 goto out;
9a4bf31d 1999 }
d23d61c8 2000
9bf87482 2001lookup_again:
d03360aa
TM		 2002 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
2003 if (IS_ERR(lseg))
2004 goto out;
9bf87482 2005 first = false;
e5e94017 2006 spin_lock(&ino->i_lock);
9fa40758 2007 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 2008 if (lo == NULL) {
2009 spin_unlock(&ino->i_lock);
3764a17e 2010 lseg = ERR_PTR(-ENOMEM);
183d9e7b 2011 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2012 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 2013 goto out;
2014 }
e5e94017 2015
43f1b3da 2016 /* Do we even need to bother with this? */
a59c30ac 2017 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 2018 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2019 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 2020 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 2021 goto out_unlock;
2022 }
2023
2024 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 2025 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 2026 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2027 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 2028 goto out_unlock;
9a4bf31d 2029 }
e5e94017 2030
411ae722
TM		 2031 /*
2032 * If the layout segment list is empty, but there are outstanding
2033 * layoutget calls, then they might be subject to a layoutrecall.
2034 */
880265c7 2035 if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) &&
411ae722
TM		 2036 atomic_read(&lo->plh_outstanding) != 0) {
2037 spin_unlock(&ino->i_lock);
880265c7
TM		 2038 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN,
2039 TASK_KILLABLE));
58bbeab4 2040 if (IS_ERR(lseg))
411ae722
TM		 2041 goto out_put_layout_hdr;
2042 pnfs_put_layout_hdr(lo);
2043 goto lookup_again;
2044 }
2045
2c8d5fc3
TM		 2046 /*
2047 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
2048 * for LAYOUTRETURN.
2049 */
2050 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
2051 spin_unlock(&ino->i_lock);
2052 dprintk("%s wait for layoutreturn\n", __func__);
2053 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2054 if (!IS_ERR(lseg)) {
2055 pnfs_put_layout_hdr(lo);
2056 dprintk("%s retrying\n", __func__);
2057 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2058 lseg,
2059 PNFS_UPDATE_LAYOUT_RETRY);
2060 goto lookup_again;
2061 }
2062 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2063 PNFS_UPDATE_LAYOUT_RETURN);
2064 goto out_put_layout_hdr;
2065 }
2066
c7d73af2 2067 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 2068 if (lseg) {
2069 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2070 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
2071 goto out_unlock;
2072 }
2073
183d9e7b
JL		 2074 /*
2075 * Choose a stateid for the LAYOUTGET. If we don't have a layout
2076 * stateid, or it has been invalidated, then we must use the open
2077 * stateid.
2078 */
67a3b721 2079 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
d9aba2b4 2080 int status;
183d9e7b
JL		 2081
2082 /*
2083 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 2084 * RFC 5661 Errata 3208.
2085 */
2086 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
2087 &lo->plh_flags)) {
2088 spin_unlock(&ino->i_lock);
d03360aa
TM		 2089 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
2090 NFS_LAYOUT_FIRST_LAYOUTGET,
2091 TASK_KILLABLE));
2092 if (IS_ERR(lseg))
2093 goto out_put_layout_hdr;
9bf87482 2094 pnfs_put_layout_hdr(lo);
183d9e7b 2095 dprintk("%s retrying\n", __func__);
9bf87482
PT		 2096 goto lookup_again;
2097 }
183d9e7b 2098
fbf4bcc9 2099 spin_unlock(&ino->i_lock);
183d9e7b 2100 first = true;
d9aba2b4 2101 status = nfs4_select_rw_stateid(ctx->state,
70d2f7b1 2102 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
d9aba2b4
TM		 2103 NULL, &stateid, NULL);
2104 if (status != 0) {
731c74dd 2105 lseg = ERR_PTR(status);
70d2f7b1
TM		 2106 trace_pnfs_update_layout(ino, pos, count,
2107 iomode, lo, lseg,
2108 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
d9aba2b4
TM		 2109 nfs4_schedule_stateid_recovery(server, ctx->state);
2110 pnfs_clear_first_layoutget(lo);
2111 pnfs_put_layout_hdr(lo);
2112 goto lookup_again;
70d2f7b1 2113 }
fbf4bcc9 2114 spin_lock(&ino->i_lock);
9bf87482 2115 } else {
183d9e7b 2116 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 2117 }
568e8c49 2118
9a4bf31d 2119 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 2120 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2121 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 2122 goto out_unlock;
9a4bf31d 2123 }
411ae722 2124 nfs_layoutget_begin(lo);
f49f9baa 2125 spin_unlock(&ino->i_lock);
30005121 2126
78746a38 2127 _add_to_server_list(lo, server);
e5e94017 2128
09cbfeaf 2129 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 2130 if (pg_offset) {
2131 arg.offset -= pg_offset;
2132 arg.length += pg_offset;
2133 }
7c24d948 2134 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 2135 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 2136
5e36e2a9 2137 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de 2138 if (!lgp) {
3764a17e 2139 lseg = ERR_PTR(-ENOMEM);
587f03de
FI		 2140 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2141 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 2142 nfs_layoutget_end(lo);
587f03de
FI		 2143 goto out_put_layout_hdr;
2144 }
2145
b4e89bcb
TM		 2146 lgp->lo = lo;
2147 pnfs_get_layout_hdr(lo);
2148
6e7434ab 2149 lseg = nfs4_proc_layoutget(lgp, &exception);
183d9e7b
JL		 2150 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2151 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 2152 nfs_layoutget_end(lo);
83026d80 2153 if (IS_ERR(lseg)) {
183d9e7b 2154 switch(PTR_ERR(lseg)) {
e85d7ee4 2155 case -EBUSY:
183d9e7b
JL		 2156 if (time_after(jiffies, giveup))
2157 lseg = NULL;
66b53f32
TM		 2158 break;
2159 case -ERECALLCONFLICT:
183d9e7b 2160 case -EAGAIN:
56b38a1f 2161 break;
fe44fb23
TM		 2162 case -ENODATA:
2163 /* The server returned NFS4ERR_LAYOUTUNAVAILABLE */
2164 pnfs_layout_set_fail_bit(
2165 lo, pnfs_iomode_to_fail_bit(iomode));
2166 lseg = NULL;
2167 goto out_put_layout_hdr;
183d9e7b
JL		 2168 default:
2169 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2170 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2171 lseg = NULL;
2172 }
56b38a1f
TM		 2173 goto out_put_layout_hdr;
2174 }
2175 if (lseg) {
6e7434ab
TM		 2176 if (!exception.retry)
2177 goto out_put_layout_hdr;
56b38a1f
TM		 2178 if (first)
2179 pnfs_clear_first_layoutget(lo);
2180 trace_pnfs_update_layout(ino, pos, count,
2181 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2182 pnfs_put_layout_hdr(lo);
2183 goto lookup_again;
83026d80
JL		 2184 }
2185 } else {
2186 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2187 }
2188
830ffb56 2189out_put_layout_hdr:
d67ae825
TH		 2190 if (first)
2191 pnfs_clear_first_layoutget(lo);
d5b9216f
TM		 2192 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2193 PNFS_UPDATE_LAYOUT_EXIT);
70c3bd2b 2194 pnfs_put_layout_hdr(lo);
e5e94017 2195out:
f86bbcf8
TM		 2196 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2197 "(%s, offset: %llu, length: %llu)\n",
2198 __func__, ino->i_sb->s_id,
2199 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2200 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2201 iomode==IOMODE_RW ? "read/write" : "read-only",
2202 (unsigned long long)pos,
2203 (unsigned long long)count);
e5e94017
BH		 2204 return lseg;
2205out_unlock:
2206 spin_unlock(&ino->i_lock);
830ffb56 2207 goto out_put_layout_hdr;
e5e94017 2208}
7c24d948 2209EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2210
540d9864
TM		 2211static bool
2212pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2213{
2214 switch (range->iomode) {
2215 case IOMODE_READ:
2216 case IOMODE_RW:
2217 break;
2218 default:
2219 return false;
2220 }
2221 if (range->offset == NFS4_MAX_UINT64)
2222 return false;
2223 if (range->length == 0)
2224 return false;
2225 if (range->length != NFS4_MAX_UINT64 &&
2226 range->length > NFS4_MAX_UINT64 - range->offset)
2227 return false;
2228 return true;
2229}
2230
78746a38
FI		 2231static struct pnfs_layout_hdr *
2232_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2233{
2234 struct pnfs_layout_hdr *lo;
2235
2236 spin_lock(&ino->i_lock);
63d8a41b 2237 lo = pnfs_find_alloc_layout(ino, ctx, nfs_io_gfp_mask());
78746a38
FI		 2238 if (!lo)
2239 goto out_unlock;
2240 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2241 goto out_unlock;
2242 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2243 goto out_unlock;
2244 if (pnfs_layoutgets_blocked(lo))
2245 goto out_unlock;
2246 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2247 goto out_unlock;
411ae722 2248 nfs_layoutget_begin(lo);
78746a38
FI		 2249 spin_unlock(&ino->i_lock);
2250 _add_to_server_list(lo, NFS_SERVER(ino));
2251 return lo;
2252
2253out_unlock:
2254 spin_unlock(&ino->i_lock);
2255 pnfs_put_layout_hdr(lo);
2256 return NULL;
2257}
2258
2409a976
FI		 2259static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2260 struct nfs_open_context *ctx)
2261{
78746a38
FI		 2262 struct inode *ino = data->dentry->d_inode;
2263 struct pnfs_layout_range rng = {
2264 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2265 IOMODE_RW: IOMODE_READ,
2266 .offset = 0,
2267 .length = NFS4_MAX_UINT64,
2268 };
2269 struct nfs4_layoutget *lgp;
2270 struct pnfs_layout_hdr *lo;
2271
64294b08
TM		 2272 /* Heuristic: don't send layoutget if we have cached data */
2273 if (rng.iomode == IOMODE_READ &&
2274 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2275 return;
2276
78746a38
FI		 2277 lo = _pnfs_grab_empty_layout(ino, ctx);
2278 if (!lo)
2279 return;
63d8a41b
TM		 2280 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid, &rng,
2281 nfs_io_gfp_mask());
78746a38
FI		 2282 if (!lgp) {
2283 pnfs_clear_first_layoutget(lo);
cb2856c5 2284 nfs_layoutget_end(lo);
78746a38
FI		 2285 pnfs_put_layout_hdr(lo);
2286 return;
2287 }
b4e89bcb 2288 lgp->lo = lo;
78746a38
FI		 2289 data->lgp = lgp;
2290 data->o_arg.lg_args = &lgp->args;
2291 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2292}
2293
2294static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2295 struct nfs_open_context *ctx)
2296{
b4e89bcb 2297 struct inode *ino = data->dentry->d_inode;
2409a976
FI		 2298 struct pnfs_layout_range rng = {
2299 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2300 IOMODE_RW: IOMODE_READ,
2301 .offset = 0,
2302 .length = NFS4_MAX_UINT64,
2303 };
2304 struct nfs4_layoutget *lgp;
2305
63d8a41b
TM		 2306 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid, &rng,
2307 nfs_io_gfp_mask());
2409a976
FI		 2308 if (!lgp)
2309 return;
2310 data->lgp = lgp;
2311 data->o_arg.lg_args = &lgp->args;
2312 data->o_res.lg_res = &lgp->res;
2313}
2314
2315void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2316 struct nfs_open_context *ctx)
2317{
2318 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2319
2320 if (!(pnfs_enabled_sb(server) &&
2321 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2322 return;
2323 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2324 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2325 return;
b4e89bcb
TM		 2326 if (data->lgp)
2327 return;
2409a976
FI		 2328 if (data->state)
2329 _lgopen_prepare_attached(data, ctx);
2330 else
2331 _lgopen_prepare_floating(data, ctx);
2332}
2333
2334void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2335 struct nfs_open_context *ctx)
2336{
2337 struct pnfs_layout_hdr *lo;
2338 struct pnfs_layout_segment *lseg;
c49b5209 2339 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2340 u32 iomode;
2341
6e01260c 2342 if (!lgp)
2409a976 2343 return;
6e01260c
FI		 2344 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2345 if (lgp->res.status) {
2346 switch (lgp->res.status) {
6e01260c 2347 default:
8dc96566
TM		 2348 break;
2349 /*
2350 * Halt lgopen attempts if the server doesn't recognise
2351 * the "current stateid" value, the layout type, or the
2352 * layoutget operation as being valid.
2353 * Also if it complains about too many ops in the compound
2354 * or of the request/reply being too big.
2355 */
2356 case -NFS4ERR_BAD_STATEID:
2357 case -NFS4ERR_NOTSUPP:
2358 case -NFS4ERR_REP_TOO_BIG:
2359 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2360 case -NFS4ERR_REQ_TOO_BIG:
2361 case -NFS4ERR_TOO_MANY_OPS:
2362 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2363 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2364 }
2365 return;
2366 }
b4e89bcb 2367 if (!lgp->lo) {
78746a38
FI		 2368 lo = _pnfs_grab_empty_layout(ino, ctx);
2369 if (!lo)
2370 return;
b4e89bcb 2371 lgp->lo = lo;
2409a976 2372 } else
b4e89bcb 2373 lo = lgp->lo;
2409a976
FI		 2374
2375 lseg = pnfs_layout_process(lgp);
32f1c28f 2376 if (!IS_ERR(lseg)) {
2409a976
FI		 2377 iomode = lgp->args.range.iomode;
2378 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2379 pnfs_put_lseg(lseg);
2380 }
30ae2412
FI		 2381}
2382
2383void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2384{
2385 if (lgp != NULL) {
b4e89bcb
TM		 2386 if (lgp->lo) {
2387 pnfs_clear_first_layoutget(lgp->lo);
2388 nfs_layoutget_end(lgp->lo);
30ae2412
FI		 2389 }
2390 pnfs_layoutget_free(lgp);
2391 }
2409a976
FI		 2392}
2393
a0b0a6e3 2394struct pnfs_layout_segment *
b1f69b75
AA		 2395pnfs_layout_process(struct nfs4_layoutget *lgp)
2396{
b4e89bcb 2397 struct pnfs_layout_hdr *lo = lgp->lo;
b1f69b75
AA		 2398 struct nfs4_layoutget_res *res = &lgp->res;
2399 struct pnfs_layout_segment *lseg;
b7edfaa1 2400 struct inode *ino = lo->plh_inode;
78096cca 2401 LIST_HEAD(free_me);
540d9864
TM		 2402
2403 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2404 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2405
2406 /* Inject layout blob into I/O device driver */
a75b9df9 2407 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2408 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2409 if (!lseg)
1b3c6d07
JL		 2410 lseg = ERR_PTR(-ENOMEM);
2411
2412 dprintk("%s: Could not allocate layout: error %ld\n",
2413 __func__, PTR_ERR(lseg));
2414 return lseg;
b1f69b75
AA		 2415 }
2416
119cef97 2417 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2418
b1f69b75 2419 spin_lock(&ino->i_lock);
e1c06f80 2420 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2421 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2422 goto out_forget;
43f1b3da 2423 }
038d6493 2424
880265c7
TM		 2425 if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) &&
2426 !pnfs_is_first_layoutget(lo))
0b77f97a
TM		 2427 goto out_forget;
2428
d29b468d 2429 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2430 /* existing state ID, make sure the sequence number matches. */
2431 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
0b77f97a 2432 if (!pnfs_layout_is_valid(lo))
d29b468d 2433 lo->plh_barrier = 0;
362f7474 2434 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2435 goto out_forget;
362f7474 2436 }
59b56394 2437 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
d29b468d 2438 } else if (pnfs_layout_is_valid(lo)) {
362f7474
CH		 2439 /*
2440 * We got an entirely new state ID. Mark all segments for the
9888d837 2441 * inode invalid, and retry the layoutget
362f7474 2442 */
08bd8dbe
TM		 2443 struct pnfs_layout_range range = {
2444 .iomode = IOMODE_ANY,
2445 .length = NFS4_MAX_UINT64,
2446 };
fb700ef0 2447 pnfs_mark_matching_lsegs_return(lo, &free_me, &range, 0);
9888d837 2448 goto out_forget;
d29b468d
TM		 2449 } else {
2450 /* We have a completely new layout */
d29b468d 2451 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
362f7474 2452 }
038d6493 2453
9369a431 2454 pnfs_get_lseg(lseg);
03772d2f 2455 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2456
b1f69b75 2457
3976143b 2458 if (res->return_on_close)
f7e8917a 2459 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2460
b1f69b75 2461 spin_unlock(&ino->i_lock);
78096cca 2462 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2463 return lseg;
43f1b3da 2464
1b3c6d07 2465out_forget:
43f1b3da
FI		 2466 spin_unlock(&ino->i_lock);
2467 lseg->pls_layout = lo;
2468 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2469 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2470}
2471
2f215968
TM		 2472/**
2473 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2474 * @lo: pointer to layout header
2475 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2476 * @return_range: describe layout segment ranges to be returned
e0b7d420 2477 * @seq: stateid seqid to match
2f215968
TM		 2478 *
2479 * This function is mainly intended for use by layoutrecall. It attempts
2480 * to free the layout segment immediately, or else to mark it for return
2481 * as soon as its reference count drops to zero.
e0b7d420
TM		 2482 *
2483 * Returns
2484 * - 0: a layoutreturn needs to be scheduled.
2485 * - EBUSY: there are layout segment that are still in use.
2486 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2487 */
10335556 2488int
016256df
PT		 2489pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2490 struct list_head *tmp_list,
6d597e17
JL		 2491 const struct pnfs_layout_range *return_range,
2492 u32 seq)
016256df
PT		 2493{
2494 struct pnfs_layout_segment *lseg, *next;
b739a5bd 2495 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
10335556 2496 int remaining = 0;
016256df
PT		 2497
2498 dprintk("%s:Begin lo %p\n", __func__, lo);
2499
fc7ff367 2500 assert_spin_locked(&lo->plh_inode->i_lock);
016256df 2501
39fd0186
TM		 2502 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2503 tmp_list = &lo->plh_return_segs;
2504
016256df 2505 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2506 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2507 dprintk("%s: marking lseg %p iomode %d "
2508 "offset %llu length %llu\n", __func__,
2509 lseg, lseg->pls_range.iomode,
2510 lseg->pls_range.offset,
2511 lseg->pls_range.length);
39fd0186
TM		 2512 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2513 tmp_list = &lo->plh_return_segs;
ff041727 2514 if (mark_lseg_invalid(lseg, tmp_list))
2f215968
TM		 2515 continue;
2516 remaining++;
016256df 2517 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
b739a5bd 2518 pnfs_lseg_cancel_io(server, lseg);
016256df 2519 }
6d597e17 2520
e0b7d420 2521 if (remaining) {
6d597e17 2522 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2523 return -EBUSY;
2524 }
6d597e17 2525
e0b7d420
TM		 2526 if (!list_empty(&lo->plh_return_segs)) {
2527 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2528 return 0;
2529 }
2530
2531 return -ENOENT;
016256df
PT		 2532}
2533
b5fdf841
TM		 2534static void
2535pnfs_mark_layout_for_return(struct inode *inode,
2536 const struct pnfs_layout_range *range)
016256df 2537{
b5fdf841 2538 struct pnfs_layout_hdr *lo;
10335556 2539 bool return_now = false;
016256df
PT		 2540
2541 spin_lock(&inode->i_lock);
b5fdf841 2542 lo = NFS_I(inode)->layout;
bdebfccd
TM		 2543 if (!pnfs_layout_is_valid(lo)) {
2544 spin_unlock(&inode->i_lock);
2545 return;
2546 }
b5fdf841 2547 pnfs_set_plh_return_info(lo, range->iomode, 0);
016256df
PT		 2548 /*
2549 * mark all matching lsegs so that we are sure to have no live
2550 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2551 * for how it works.
2552 */
b5fdf841 2553 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
44ea8dfc 2554 const struct cred *cred;
10335556 2555 nfs4_stateid stateid;
e5fd1904 2556 enum pnfs_iomode iomode;
10335556 2557
44ea8dfc 2558 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
10335556
TM		 2559 spin_unlock(&inode->i_lock);
2560 if (return_now)
44ea8dfc 2561 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
10335556
TM		 2562 } else {
2563 spin_unlock(&inode->i_lock);
2564 nfs_commit_inode(inode, 0);
2565 }
016256df 2566}
b5fdf841
TM		 2567
2568void pnfs_error_mark_layout_for_return(struct inode *inode,
2569 struct pnfs_layout_segment *lseg)
2570{
2571 struct pnfs_layout_range range = {
2572 .iomode = lseg->pls_range.iomode,
2573 .offset = 0,
2574 .length = NFS4_MAX_UINT64,
2575 };
2576
2577 pnfs_mark_layout_for_return(inode, &range);
2578}
016256df
PT		 2579EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2580
b5fdf841
TM		 2581static bool
2582pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
2583{
2584 return pnfs_layout_is_valid(lo) &&
2585 !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
2586 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
2587}
2588
2589static struct pnfs_layout_segment *
2590pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
2591 const struct pnfs_layout_range *range,
2592 enum pnfs_iomode iomode)
2593{
2594 struct pnfs_layout_segment *lseg;
2595
2596 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
2597 if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
2598 continue;
2599 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2600 continue;
2601 if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
2602 continue;
2603 if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
2604 return lseg;
2605 }
2606 return NULL;
2607}
2608
2609/* Find open file states whose mode matches that of the range */
2610static bool
2611pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
2612 const struct pnfs_layout_range *range)
2613{
2614 struct list_head *head;
2615 struct nfs_open_context *ctx;
2616 fmode_t mode = 0;
2617
2618 if (!pnfs_layout_can_be_returned(lo) ||
2619 !pnfs_find_first_lseg(lo, range, range->iomode))
2620 return false;
2621
2622 head = &NFS_I(lo->plh_inode)->open_files;
2623 list_for_each_entry_rcu(ctx, head, list) {
2624 if (ctx->state)
2625 mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
2626 }
2627
2628 switch (range->iomode) {
2629 default:
2630 break;
2631 case IOMODE_READ:
2632 mode &= ~FMODE_WRITE;
2633 break;
2634 case IOMODE_RW:
2635 if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
2636 mode &= ~FMODE_READ;
2637 }
2638 return mode == 0;
2639}
2640
f6395572
TM		 2641static int pnfs_layout_return_unused_byserver(struct nfs_server *server,
2642 void *data)
b5fdf841
TM		 2643{
2644 const struct pnfs_layout_range *range = data;
f6395572 2645 const struct cred *cred;
b5fdf841
TM		 2646 struct pnfs_layout_hdr *lo;
2647 struct inode *inode;
f6395572
TM		 2648 nfs4_stateid stateid;
2649 enum pnfs_iomode iomode;
2650
b5fdf841
TM		 2651restart:
2652 rcu_read_lock();
2653 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
f6395572
TM		 2654 inode = lo->plh_inode;
2655 if (!inode || !pnfs_layout_can_be_returned(lo) ||
b5fdf841
TM		 2656 test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2657 continue;
b5fdf841 2658 spin_lock(&inode->i_lock);
f6395572
TM		 2659 if (!lo->plh_inode ||
2660 !pnfs_should_return_unused_layout(lo, range)) {
b5fdf841
TM		 2661 spin_unlock(&inode->i_lock);
2662 continue;
2663 }
f6395572
TM		 2664 pnfs_get_layout_hdr(lo);
2665 pnfs_set_plh_return_info(lo, range->iomode, 0);
2666 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
2667 range, 0) != 0 ||
2668 !pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode)) {
2669 spin_unlock(&inode->i_lock);
2670 rcu_read_unlock();
2671 pnfs_put_layout_hdr(lo);
2672 cond_resched();
2673 goto restart;
2674 }
b5fdf841 2675 spin_unlock(&inode->i_lock);
b5fdf841 2676 rcu_read_unlock();
f6395572
TM		 2677 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
2678 pnfs_put_layout_hdr(lo);
b5fdf841
TM		 2679 cond_resched();
2680 goto restart;
2681 }
2682 rcu_read_unlock();
2683 return 0;
2684}
2685
2686void
2687pnfs_layout_return_unused_byclid(struct nfs_client *clp,
2688 enum pnfs_iomode iomode)
2689{
2690 struct pnfs_layout_range range = {
2691 .iomode = iomode,
2692 .offset = 0,
2693 .length = NFS4_MAX_UINT64,
2694 };
2695
2696 nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
2697 &range);
2698}
2699
b3230e80
TM		 2700void
2701pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2702{
2703 if (pgio->pg_lseg == NULL ||
2704 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2705 return;
2706 pnfs_put_lseg(pgio->pg_lseg);
2707 pgio->pg_lseg = NULL;
2708}
2709EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2710
08cb5b0f
BC		 2711/*
2712 * Check for any intersection between the request and the pgio->pg_lseg,
2713 * and if none, put this pgio->pg_lseg away.
2714 */
e1e54ab7 2715void
08cb5b0f
BC		 2716pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2717{
2718 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2719 pnfs_put_lseg(pgio->pg_lseg);
2720 pgio->pg_lseg = NULL;
2721 }
2722}
e1e54ab7 2723EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
08cb5b0f 2724
d8007d4d
TM		 2725void
2726pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2727{
d1d97395 2728 u64 rd_size;
1fd937bd 2729
b3230e80 2730 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2731 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2732 if (pgio->pg_lseg == NULL) {
2733 if (pgio->pg_dreq == NULL)
2734 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2735 else
8a6291bf
TM		 2736 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq,
2737 req_offset(req));
cb5d04bc 2738
63d8a41b
TM		 2739 pgio->pg_lseg =
2740 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
2741 req_offset(req), rd_size,
2742 IOMODE_READ, false,
2743 nfs_io_gfp_mask());
d600ad1f
PT		 2744 if (IS_ERR(pgio->pg_lseg)) {
2745 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2746 pgio->pg_lseg = NULL;
2747 return;
2748 }
cb5d04bc 2749 }
e885de1a
TM		 2750 /* If no lseg, fall back to read through mds */
2751 if (pgio->pg_lseg == NULL)
1f945357 2752 nfs_pageio_reset_read_mds(pgio);
e885de1a 2753
d8007d4d
TM		 2754}
2755EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2756
2757void
6296556f
PT		 2758pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2759 struct nfs_page *req, u64 wb_size)
d8007d4d 2760{
b3230e80 2761 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2762 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2763 if (pgio->pg_lseg == NULL) {
63d8a41b
TM		 2764 pgio->pg_lseg =
2765 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
2766 req_offset(req), wb_size, IOMODE_RW,
2767 false, nfs_io_gfp_mask());
d600ad1f
PT		 2768 if (IS_ERR(pgio->pg_lseg)) {
2769 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2770 pgio->pg_lseg = NULL;
2771 return;
2772 }
2773 }
e885de1a
TM		 2774 /* If no lseg, fall back to write through mds */
2775 if (pgio->pg_lseg == NULL)
1f945357 2776 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2777}
2778EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2779
180bb5ec
WAA		 2780void
2781pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2782{
2783 if (desc->pg_lseg) {
2784 pnfs_put_lseg(desc->pg_lseg);
2785 desc->pg_lseg = NULL;
2786 }
2787}
2788EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2789
b4fdac1a
WAA		 2790/*
2791 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2792 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2793 */
2794size_t
a7d42ddb
WAA		 2795pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2796 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2797{
0f9c429e 2798 unsigned int size;
c5e20cb7 2799 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2800
2801 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2802 if (!size)
2803 return 0;
94ad1c80 2804
19982ba8 2805 /*
c5e20cb7
WAA		 2806 * 'size' contains the number of bytes left in the current page (up
2807 * to the original size asked for in @req->wb_bytes).
2808 *
2809 * Calculate how many bytes are left in the layout segment
2810 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2811 *
2812 * Please also note that 'end_offset' is actually the offset of the
2813 * first byte that lies outside the pnfs_layout_range. FIXME?
2814 *
2815 */
19b54848 2816 if (pgio->pg_lseg) {
17822b20 2817 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2818 pgio->pg_lseg->pls_range.length);
2819 req_start = req_offset(req);
08cb5b0f 2820
c5e20cb7 2821 /* start of request is past the last byte of this segment */
08cb5b0f 2822 if (req_start >= seg_end)
19b54848 2823 return 0;
c5e20cb7
WAA		 2824
2825 /* adjust 'size' iff there are fewer bytes left in the
2826 * segment than what nfs_generic_pg_test returned */
2827 seg_left = seg_end - req_start;
2828 if (seg_left < size)
2829 size = (unsigned int)seg_left;
19b54848 2830 }
0f9c429e 2831
19b54848 2832 return size;
94ad1c80 2833}
89a58e32 2834EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2835
53113ad3 2836int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2837{
2838 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2839
2840 /* Resend all requests through the MDS */
53113ad3
WAA		 2841 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2842 hdr->completion_ops);
2843 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2844}
e7dd79af 2845EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2846
d45f60c6 2847static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2848{
cd841605
FI		 2849
2850 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2851 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2852 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2853 pnfs_return_layout(hdr->inode);
1acbbb4e 2854 }
6c75dc0d 2855 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2856 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2857}
2858
d20581aa
BH		 2859/*
2860 * Called by non rpc-based layout drivers
2861 */
d45f60c6 2862void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2863{
f8417b48 2864 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2865 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2866 hdr->mds_offset + hdr->res.count);
d45f60c6 2867 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2868 }
2869 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2870 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2871 pnfs_ld_handle_write_error(hdr);
2872 hdr->mds_ops->rpc_release(hdr);
44b83799 2873}
d20581aa 2874EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2875
dce81290
TM		 2876static void
2877pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2878 struct nfs_pgio_header *hdr)
dce81290 2879{
48d635f1 2880 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2881
6c75dc0d 2882 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2883 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2884 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2885 mirror->pg_recoalesce = 1;
6c75dc0d 2886 }
ba4a76f7 2887 hdr->completion_ops->completion(hdr);
dce81290
TM		 2888}
2889
2890static enum pnfs_try_status
d45f60c6 2891pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2892 const struct rpc_call_ops *call_ops,
2893 struct pnfs_layout_segment *lseg,
2894 int how)
0382b744 2895{
cd841605 2896 struct inode *inode = hdr->inode;
0382b744
AA		 2897 enum pnfs_try_status trypnfs;
2898 struct nfs_server *nfss = NFS_SERVER(inode);
2899
cd841605 2900 hdr->mds_ops = call_ops;
0382b744
AA		 2901
2902 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2903 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2904 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2905 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2906 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2907 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2908 return trypnfs;
2909}
2910
dce81290 2911static void
7f714720
WAA		 2912pnfs_do_write(struct nfs_pageio_descriptor *desc,
2913 struct nfs_pgio_header *hdr, int how)
dce81290 2914{
dce81290
TM		 2915 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2916 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2917 enum pnfs_try_status trypnfs;
dce81290 2918
d45f60c6 2919 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2920 switch (trypnfs) {
2921 case PNFS_NOT_ATTEMPTED:
d45f60c6 2922 pnfs_write_through_mds(desc, hdr);
ffb81717 2923 break;
37f8aa16
TM		 2924 case PNFS_ATTEMPTED:
2925 break;
2926 case PNFS_TRY_AGAIN:
2927 /* cleanup hdr and prepare to redo pnfs */
2928 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2929 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2930 list_splice_init(&hdr->pages, &mirror->pg_list);
2931 mirror->pg_recoalesce = 1;
2932 }
2933 hdr->mds_ops->rpc_release(hdr);
2934 }
dce81290
TM		 2935}
2936
6c75dc0d
FI		 2937static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2938{
9369a431 2939 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2940 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2941}
2942
dce81290
TM		 2943int
2944pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2945{
6c75dc0d 2946 struct nfs_pgio_header *hdr;
dce81290
TM		 2947 int ret;
2948
1e7f3a48
WAA		 2949 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2950 if (!hdr) {
2bff2288
PT		 2951 desc->pg_error = -ENOMEM;
2952 return desc->pg_error;
dce81290 2953 }
6c75dc0d 2954 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2955
9369a431 2956 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2957 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2958 if (!ret)
7f714720 2959 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2960
6c75dc0d 2961 return ret;
dce81290
TM		 2962}
2963EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2964
53113ad3 2965int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2966{
2967 struct nfs_pageio_descriptor pgio;
2968
1acbbb4e 2969 /* Resend all requests through the MDS */
53113ad3
WAA		 2970 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2971 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2972}
e7dd79af 2973EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2974
d45f60c6 2975static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2976{
cd841605
FI		 2977 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2978 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2979 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2980 pnfs_return_layout(hdr->inode);
1acbbb4e 2981 }
4db6e0b7 2982 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2983 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2984}
2985
d20581aa
BH		 2986/*
2987 * Called by non rpc-based layout drivers
2988 */
d45f60c6 2989void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2990{
bfc505de 2991 if (likely(!hdr->pnfs_error))
d45f60c6 2992 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2993 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2994 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2995 pnfs_ld_handle_read_error(hdr);
2996 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2997}
2998EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2999
493292dd
TM		 3000static void
3001pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 3002 struct nfs_pgio_header *hdr)
493292dd 3003{
48d635f1 3004 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 3005
4db6e0b7 3006 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 3007 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 3008 nfs_pageio_reset_read_mds(desc);
a7d42ddb 3009 mirror->pg_recoalesce = 1;
4db6e0b7 3010 }
ba4a76f7 3011 hdr->completion_ops->completion(hdr);
493292dd
TM		 3012}
3013
64419a9b
AA		 3014/*
3015 * Call the appropriate parallel I/O subsystem read function.
3016 */
493292dd 3017static enum pnfs_try_status
d45f60c6 3018pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 3019 const struct rpc_call_ops *call_ops,
3020 struct pnfs_layout_segment *lseg)
64419a9b 3021{
cd841605 3022 struct inode *inode = hdr->inode;
64419a9b
AA		 3023 struct nfs_server *nfss = NFS_SERVER(inode);
3024 enum pnfs_try_status trypnfs;
3025
cd841605 3026 hdr->mds_ops = call_ops;
64419a9b
AA		 3027
3028 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 3029 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 3030
d45f60c6 3031 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 3032 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 3033 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 3034 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
3035 return trypnfs;
3036}
863a3c6c 3037
ceb11e13 3038/* Resend all requests through pnfs. */
563c53e7
TM		 3039void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr,
3040 unsigned int mirror_idx)
ceb11e13
PT		 3041{
3042 struct nfs_pageio_descriptor pgio;
3043
1b1bc66b 3044 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 3045 /* Prevent deadlocks with layoutreturn! */
3046 pnfs_put_lseg(hdr->lseg);
3047 hdr->lseg = NULL;
3048
1b1bc66b
WAA		 3049 nfs_pageio_init_read(&pgio, hdr->inode, false,
3050 hdr->completion_ops);
563c53e7 3051 pgio.pg_mirror_idx = mirror_idx;
1b1bc66b
WAA		 3052 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
3053 }
ceb11e13
PT		 3054}
3055EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
3056
493292dd 3057static void
7f714720 3058pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 3059{
493292dd
TM		 3060 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
3061 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 3062 enum pnfs_try_status trypnfs;
493292dd 3063
d45f60c6 3064 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 3065 switch (trypnfs) {
3066 case PNFS_NOT_ATTEMPTED:
d45f60c6 3067 pnfs_read_through_mds(desc, hdr);
ffb81717 3068 break;
6aeafd05
TM		 3069 case PNFS_ATTEMPTED:
3070 break;
3071 case PNFS_TRY_AGAIN:
3072 /* cleanup hdr and prepare to redo pnfs */
3073 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3074 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3075 list_splice_init(&hdr->pages, &mirror->pg_list);
3076 mirror->pg_recoalesce = 1;
3077 }
3078 hdr->mds_ops->rpc_release(hdr);
3079 }
493292dd
TM		 3080}
3081
4db6e0b7
FI		 3082static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
3083{
9369a431 3084 pnfs_put_lseg(hdr->lseg);
1e7f3a48 3085 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 3086}
3087
493292dd
TM		 3088int
3089pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
3090{
4db6e0b7 3091 struct nfs_pgio_header *hdr;
493292dd
TM		 3092 int ret;
3093
1e7f3a48
WAA		 3094 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
3095 if (!hdr) {
2bff2288
PT		 3096 desc->pg_error = -ENOMEM;
3097 return desc->pg_error;
493292dd 3098 }
4db6e0b7 3099 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 3100 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 3101 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 3102 if (!ret)
7f714720 3103 pnfs_do_read(desc, hdr);
4db6e0b7 3104 return ret;
493292dd
TM		 3105}
3106EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
3107
71244d9b
TM		 3108static void pnfs_clear_layoutcommitting(struct inode *inode)
3109{
3110 unsigned long *bitlock = &NFS_I(inode)->flags;
3111
3112 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 3113 smp_mb__after_atomic();
71244d9b
TM		 3114 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
3115}
3116
863a3c6c 3117/*
a9bae566 3118 * There can be multiple RW segments.
863a3c6c 3119 */
a9bae566 3120static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 3121{
a9bae566 3122 struct pnfs_layout_segment *lseg;
863a3c6c 3123
a9bae566
PT		 3124 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
3125 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 3126 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 3127 list_add(&lseg->pls_lc_list, listp);
3128 }
863a3c6c
AA		 3129}
3130
a073dbff
TM		 3131static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
3132{
3133 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 3134
3135 /* Matched by references in pnfs_set_layoutcommit */
3136 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
3137 list_del_init(&lseg->pls_lc_list);
3138 pnfs_put_lseg(lseg);
3139 }
3140
71244d9b 3141 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 3142}
3143
1b0ae068
PT		 3144void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
3145{
b9e028fd 3146 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 3147}
3148EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
3149
863a3c6c 3150void
67af7611
TM		 3151pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
3152 loff_t end_pos)
863a3c6c 3153{
cd841605 3154 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 3155 bool mark_as_dirty = false;
863a3c6c 3156
cd841605 3157 spin_lock(&inode->i_lock);
863a3c6c 3158 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 3159 nfsi->layout->plh_lwb = end_pos;
79a48a1f 3160 mark_as_dirty = true;
863a3c6c 3161 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 3162 __func__, inode->i_ino);
29559b11
TM		 3163 } else if (end_pos > nfsi->layout->plh_lwb)
3164 nfsi->layout->plh_lwb = end_pos;
67af7611 3165 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 3166 /* references matched in nfs4_layoutcommit_release */
67af7611 3167 pnfs_get_lseg(lseg);
a9bae566 3168 }
cd841605 3169 spin_unlock(&inode->i_lock);
acff5880 3170 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 3171 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 3172
3173 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3174 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3175 if (mark_as_dirty)
cd841605 3176 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3177}
3178EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
3179
db29c089
AA		 3180void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
3181{
3182 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
3183
3184 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
3185 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 3186 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 3187}
3188
de4b15c7
AA		 3189/*
3190 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3191 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3192 * data to disk to allow the server to recover the data if it crashes.
3193 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3194 * is off, and a COMMIT is sent to a data server, or
3195 * if WRITEs to a data server return NFS_DATA_SYNC.
3196 */
863a3c6c 3197int
ef311537 3198pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 3199{
5f919c9f 3200 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 3201 struct nfs4_layoutcommit_data *data;
3202 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 3203 loff_t end_pos;
71244d9b 3204 int status;
863a3c6c 3205
71244d9b 3206 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 3207 return 0;
3208
71244d9b 3209 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 3210
71244d9b 3211 status = -EAGAIN;
92407e75 3212 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 3213 if (!sync)
3214 goto out;
74316201 3215 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 3216 NFS_INO_LAYOUTCOMMITTING,
3217 nfs_wait_bit_killable,
f5d39b02 3218 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
92407e75 3219 if (status)
71244d9b 3220 goto out;
92407e75
PT		 3221 }
3222
71244d9b
TM		 3223 status = -ENOMEM;
3224 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
63d8a41b 3225 data = kzalloc(sizeof(*data), nfs_io_gfp_mask());
71244d9b
TM		 3226 if (!data)
3227 goto clear_layoutcommitting;
3228
3229 status = 0;
de4b15c7 3230 spin_lock(&inode->i_lock);
71244d9b
TM		 3231 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
3232 goto out_unlock;
a9bae566 3233
71244d9b 3234 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 3235 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 3236
acff5880 3237 end_pos = nfsi->layout->plh_lwb;
863a3c6c 3238
f597c537 3239 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
97a728f5 3240 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 3241 spin_unlock(&inode->i_lock);
3242
3243 data->args.inode = inode;
863a3c6c
AA		 3244 nfs_fattr_init(&data->fattr);
3245 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3246 data->res.fattr = &data->fattr;
2e18d4d8
TM		 3247 if (end_pos != 0)
3248 data->args.lastbytewritten = end_pos - 1;
3249 else
3250 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 3251 data->res.server = NFS_SERVER(inode);
3252
5f919c9f
CH		 3253 if (ld->prepare_layoutcommit) {
3254 status = ld->prepare_layoutcommit(&data->args);
3255 if (status) {
a52458b4 3256 put_cred(data->cred);
5f919c9f 3257 spin_lock(&inode->i_lock);
29559b11
TM		 3258 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3259 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 3260 nfsi->layout->plh_lwb = end_pos;
3471648a 3261 goto out_unlock;
5f919c9f
CH		 3262 }
3263 }
3264
3265
863a3c6c
AA		 3266 status = nfs4_proc_layoutcommit(data, sync);
3267out:
92407e75
PT		 3268 if (status)
3269 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3270 dprintk("<-- %s status %d\n", __func__, status);
3271 return status;
71244d9b
TM		 3272out_unlock:
3273 spin_unlock(&inode->i_lock);
92407e75 3274 kfree(data);
71244d9b
TM		 3275clear_layoutcommitting:
3276 pnfs_clear_layoutcommitting(inode);
92407e75 3277 goto out;
863a3c6c 3278}
72cff449 3279EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 3280
5bb89b47
TM		 3281int
3282pnfs_generic_sync(struct inode *inode, bool datasync)
3283{
3284 return pnfs_layoutcommit_inode(inode, true);
3285}
3286EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3287
82be417a
AA		 3288struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3289{
3290 struct nfs4_threshold *thp;
3291
63d8a41b 3292 thp = kzalloc(sizeof(*thp), nfs_io_gfp_mask());
82be417a
AA		 3293 if (!thp) {
3294 dprintk("%s mdsthreshold allocation failed\n", __func__);
3295 return NULL;
3296 }
3297 return thp;
3298}
8733408d 3299
865a7ecb 3300#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 3301int
c8ad8894 3302pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 3303{
3304 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3305 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 3306 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 3307 struct nfs42_layoutstat_data *data;
3308 struct pnfs_layout_hdr *hdr;
3309 int status = 0;
3310
3311 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3312 goto out;
3313
6c5a0d89
TM		 3314 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3315 goto out;
3316
1bfe3b25
PT		 3317 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3318 goto out;
3319
8733408d
PT		 3320 spin_lock(&inode->i_lock);
3321 if (!NFS_I(inode)->layout) {
3322 spin_unlock(&inode->i_lock);
f538d0ba 3323 goto out_clear_layoutstats;
8733408d
PT		 3324 }
3325 hdr = NFS_I(inode)->layout;
3326 pnfs_get_layout_hdr(hdr);
3327 spin_unlock(&inode->i_lock);
3328
c8ad8894 3329 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3330 if (!data) {
3331 status = -ENOMEM;
3332 goto out_put;
3333 }
3334
3335 data->args.fh = NFS_FH(inode);
3336 data->args.inode = inode;
8733408d
PT		 3337 status = ld->prepare_layoutstats(&data->args);
3338 if (status)
3339 goto out_free;
3340
3341 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3342
3343out:
3344 dprintk("%s returns %d\n", __func__, status);
3345 return status;
3346
3347out_free:
3348 kfree(data);
3349out_put:
3350 pnfs_put_layout_hdr(hdr);
f538d0ba 3351out_clear_layoutstats:
1bfe3b25
PT		 3352 smp_mb__before_atomic();
3353 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3354 smp_mb__after_atomic();
8733408d
PT		 3355 goto out;
3356}
3357EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3358#endif
bbf58bf3
TM		 3359
3360unsigned int layoutstats_timer;
3361module_param(layoutstats_timer, uint, 0644);
3362EXPORT_SYMBOL_GPL(layoutstats_timer);
Linux 6.x block layer and io_uring tree(s)