projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'multipath_tcp'
[linux-block.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
95void
96unset_pnfs_layoutdriver(struct nfs_server *nfss)
97{
738fd0f3
BH		 98 if (nfss->pnfs_curr_ld) {
99 if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 104 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 105 }
85e174ba
RL		 106 nfss->pnfs_curr_ld = NULL;
107}
108
ca440c38
JL		 109/*
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
112 *
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
115 */
116static const u32 ld_prefs[] = {
117 LAYOUT_SCSI,
118 LAYOUT_BLOCK_VOLUME,
119 LAYOUT_OSD2_OBJECTS,
120 LAYOUT_FLEX_FILES,
121 LAYOUT_NFSV4_1_FILES,
122 0
123};
124
125static int
126ld_cmp(const void *e1, const void *e2)
127{
128 u32 ld1 = *((u32 *)e1);
129 u32 ld2 = *((u32 *)e2);
130 int i;
131
132 for (i = 0; ld_prefs[i] != 0; i++) {
133 if (ld1 == ld_prefs[i])
134 return -1;
135
136 if (ld2 == ld_prefs[i])
137 return 1;
138 }
139 return 0;
140}
141
85e174ba
RL		 142/*
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
145 *
3132e49e 146 * @ids array of layout types supported by MDS.
85e174ba
RL		 147 */
148void
738fd0f3 149set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 150 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 151{
152 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 153 u32 id;
ca440c38 154 int i;
85e174ba 155
19274716
AS		 156 if (fsinfo->nlayouttypes == 0)
157 goto out_no_driver;
85e174ba
RL		 158 if (!(server->nfs_client->cl_exchange_flags &
159 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 160 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 162 goto out_no_driver;
163 }
3132e49e 164
ca440c38
JL		 165 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 167
ca440c38
JL		 168 for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 id = fsinfo->layouttype[i];
85e174ba 170 ld_type = find_pnfs_driver(id);
ca440c38
JL		 171 if (!ld_type) {
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 id);
174 ld_type = find_pnfs_driver(id);
175 }
176 if (ld_type)
177 break;
85e174ba 178 }
3132e49e
JL		 179
180 if (!ld_type) {
ca440c38 181 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 182 goto out_no_driver;
183 }
184
85e174ba 185 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 186 if (ld_type->set_layoutdriver
187 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 188 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__, id);
738fd0f3
BH		 190 module_put(ld_type->owner);
191 goto out_no_driver;
192 }
2a4c8994
TM		 193 /* Bump the MDS count */
194 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 195
85e174ba
RL		 196 dprintk("%s: pNFS module for %u set\n", __func__, id);
197 return;
198
199out_no_driver:
200 dprintk("%s: Using NFSv4 I/O\n", __func__);
201 server->pnfs_curr_ld = NULL;
202}
02c35fca
FI		 203
204int
205pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206{
207 int status = -EINVAL;
208 struct pnfs_layoutdriver_type *tmp;
209
210 if (ld_type->id == 0) {
a030889a 211 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 212 return status;
213 }
b1f69b75 214 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 215 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 216 "alloc_lseg and free_lseg.\n", __func__);
217 return status;
218 }
02c35fca
FI		 219
220 spin_lock(&pnfs_spinlock);
221 tmp = find_pnfs_driver_locked(ld_type->id);
222 if (!tmp) {
223 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 status = 0;
225 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 ld_type->name);
227 } else {
a030889a 228 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 229 __func__, ld_type->id);
230 }
231 spin_unlock(&pnfs_spinlock);
232
233 return status;
234}
235EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236
237void
238pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239{
240 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 spin_lock(&pnfs_spinlock);
242 list_del(&ld_type->pnfs_tblid);
243 spin_unlock(&pnfs_spinlock);
244}
245EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 246
b1f69b75
AA		 247/*
248 * pNFS client layout cache
249 */
250
cc6e5340 251/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 252void
70c3bd2b 253pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 254{
2b28a7be 255 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 256}
257
636fb9c8
BH		 258static struct pnfs_layout_hdr *
259pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260{
261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 262 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 263}
264
265static void
266pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267{
9c626381
TM		 268 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270
271 if (!list_empty(&lo->plh_layouts)) {
272 struct nfs_client *clp = server->nfs_client;
273
274 spin_lock(&clp->cl_lock);
275 list_del_init(&lo->plh_layouts);
276 spin_unlock(&clp->cl_lock);
277 }
a52458b4 278 put_cred(lo->plh_lc_cred);
57934278 279 return ld->free_layout_hdr(lo);
636fb9c8
BH		 280}
281
e5e94017 282static void
6622c3ea 283pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 284{
bb346f63 285 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 286 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 287 nfsi->layout = NULL;
288 /* Reset MDS Threshold I/O counters */
289 nfsi->write_io = 0;
290 nfsi->read_io = 0;
e5e94017
BH		 291}
292
b1f69b75 293void
70c3bd2b 294pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 295{
9c6376eb 296 struct inode *inode;
cc6e5340 297
9c6376eb
TM		 298 if (!lo)
299 return;
300 inode = lo->plh_inode;
13c13a6a
TM		 301 pnfs_layoutreturn_before_put_layout_hdr(lo);
302
2b28a7be 303 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 304 if (!list_empty(&lo->plh_segs))
305 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 306 pnfs_detach_layout_hdr(lo);
cc6e5340 307 spin_unlock(&inode->i_lock);
6622c3ea 308 pnfs_free_layout_hdr(lo);
cc6e5340 309 }
974cec8c
AA		 310}
311
4aab9732
TM		 312static void
313pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
314 u32 seq)
315{
316 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
317 iomode = IOMODE_ANY;
318 lo->plh_return_iomode = iomode;
319 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
320 if (seq != 0) {
321 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
322 lo->plh_return_seq = seq;
323 }
324}
325
ae5a459d
TM		 326static void
327pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
328{
5466d214 329 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 330 lo->plh_return_iomode = 0;
331 lo->plh_return_seq = 0;
332 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 333 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
334 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
335 continue;
336 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
337 }
ae5a459d
TM		 338}
339
362fb578
TM		 340static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
341{
342 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
343 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
344 smp_mb__after_atomic();
345 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
346 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
347}
348
fe1cf946
TM		 349static void
350pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
351 struct list_head *free_me)
352{
353 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
354 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
355 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
356 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
357 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
358 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
359}
360
7380020e
TM		 361/*
362 * Update the seqid of a layout stateid
363 */
ecf84026
TM		 364bool nfs4_layoutreturn_refresh_stateid(nfs4_stateid *dst,
365 struct pnfs_layout_range *dst_range,
366 struct inode *inode)
7380020e
TM		 367{
368 struct pnfs_layout_hdr *lo;
c16467dc
TM		 369 struct pnfs_layout_range range = {
370 .iomode = IOMODE_ANY,
371 .offset = 0,
372 .length = NFS4_MAX_UINT64,
373 };
7380020e 374 bool ret = false;
c16467dc
TM		 375 LIST_HEAD(head);
376 int err;
7380020e
TM		 377
378 spin_lock(&inode->i_lock);
379 lo = NFS_I(inode)->layout;
380 if (lo && nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
c16467dc
TM		 381 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
382 if (err != -EBUSY) {
383 dst->seqid = lo->plh_stateid.seqid;
ecf84026 384 *dst_range = range;
c16467dc
TM		 385 ret = true;
386 }
7380020e
TM		 387 }
388 spin_unlock(&inode->i_lock);
c16467dc 389 pnfs_free_lseg_list(&head);
7380020e
TM		 390 return ret;
391}
392
2454dfea
TM		 393/*
394 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
395 *
396 * In order to continue using the pnfs_layout_hdr, a full recovery
397 * is required.
398 * Note that caller must hold inode->i_lock.
399 */
5f46be04 400int
2454dfea
TM		 401pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
402 struct list_head *lseg_list)
403{
404 struct pnfs_layout_range range = {
405 .iomode = IOMODE_ANY,
406 .offset = 0,
407 .length = NFS4_MAX_UINT64,
408 };
fe1cf946 409 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 410
411 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 412 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
413 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 414 pnfs_clear_layoutreturn_info(lo);
68f74479 415 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 416 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
417 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
418 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 419 return !list_empty(&lo->plh_segs);
2454dfea
TM		 420}
421
b9e028fd
TM		 422static int
423pnfs_iomode_to_fail_bit(u32 iomode)
424{
425 return iomode == IOMODE_RW ?
426 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
427}
428
429static void
3e621214 430pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 431{
25c75333 432 lo->plh_retry_timestamp = jiffies;
39e88fcf 433 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 434 refcount_inc(&lo->plh_refcount);
3e621214
TM		 435}
436
437static void
438pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
439{
440 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 441 refcount_dec(&lo->plh_refcount);
3e621214
TM		 442}
443
444static void
445pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
446{
447 struct inode *inode = lo->plh_inode;
115ce575
TM		 448 struct pnfs_layout_range range = {
449 .iomode = iomode,
450 .offset = 0,
451 .length = NFS4_MAX_UINT64,
452 };
453 LIST_HEAD(head);
3e621214
TM		 454
455 spin_lock(&inode->i_lock);
456 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 457 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 458 spin_unlock(&inode->i_lock);
115ce575 459 pnfs_free_lseg_list(&head);
b9e028fd
TM		 460 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
461 iomode == IOMODE_RW ? "RW" : "READ");
462}
463
464static bool
465pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
466{
25c75333 467 unsigned long start, end;
3e621214
TM		 468 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
469
470 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 471 return false;
472 end = jiffies;
473 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
474 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
475 /* It is time to retry the failed layoutgets */
3e621214 476 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 477 return false;
478 }
479 return true;
b9e028fd
TM		 480}
481
974cec8c 482static void
119cef97
TM		 483pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
484 const struct pnfs_layout_range *range,
485 const nfs4_stateid *stateid)
974cec8c 486{
566052c5 487 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 488 INIT_LIST_HEAD(&lseg->pls_lc_list);
eba6dd69 489 refcount_set(&lseg->pls_refcount, 1);
4541d16c 490 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 491 lseg->pls_layout = lo;
119cef97
TM		 492 lseg->pls_range = *range;
493 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 494}
495
905ca191 496static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 497{
68f74479
TM		 498 if (lseg != NULL) {
499 struct inode *inode = lseg->pls_layout->plh_inode;
500 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
501 }
974cec8c
AA		 502}
503
d684d2ae 504static void
57036a37
TM		 505pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
506 struct pnfs_layout_segment *lseg)
d684d2ae 507{
d20581aa 508 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 509 list_del_init(&lseg->pls_list);
8f0d27dc 510 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 511 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 512 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
513 return;
7b650994
TM		 514 if (list_empty(&lo->plh_segs) &&
515 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
516 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 517 if (atomic_read(&lo->plh_outstanding) == 0)
518 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 519 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 520 }
d684d2ae
FI		 521}
522
68f74479
TM		 523static bool
524pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
525 struct pnfs_layout_segment *lseg)
526{
527 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
528 pnfs_layout_is_valid(lo)) {
4aab9732 529 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 530 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
531 return true;
532 }
533 return false;
534}
535
bae724ef 536void
9369a431 537pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 538{
57036a37 539 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 540 struct inode *inode;
541
542 if (!lseg)
543 return;
544
4541d16c 545 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 546 refcount_read(&lseg->pls_refcount),
4541d16c 547 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 548
57036a37
TM		 549 lo = lseg->pls_layout;
550 inode = lo->plh_inode;
4ef2e4f8 551
eba6dd69 552 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 553 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
554 spin_unlock(&inode->i_lock);
555 return;
556 }
8f0d27dc 557 pnfs_get_layout_hdr(lo);
4ef2e4f8 558 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 559 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
560 lseg = NULL;
4ef2e4f8
TM		 561 spin_unlock(&inode->i_lock);
562 pnfs_free_lseg(lseg);
563 pnfs_put_layout_hdr(lo);
4541d16c 564 }
4541d16c 565}
9369a431 566EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 567
fb3296eb
BH		 568/*
569 * is l2 fully contained in l1?
570 * start1 end1
571 * [----------------------------------)
572 * start2 end2
573 * [----------------)
574 */
3cb2df17 575static bool
7dc0ac70 576pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 577 const struct pnfs_layout_range *l2)
fb3296eb
BH		 578{
579 u64 start1 = l1->offset;
17822b20 580 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 581 u64 start2 = l2->offset;
17822b20 582 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 583
584 return (start1 <= start2) && (end1 >= end2);
585}
586
24956804
TM		 587static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
588 struct list_head *tmp_list)
589{
eba6dd69 590 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 591 return false;
592 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
593 list_add(&lseg->pls_list, tmp_list);
594 return true;
595}
596
4541d16c
FI		 597/* Returns 1 if lseg is removed from list, 0 otherwise */
598static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
599 struct list_head *tmp_list)
600{
601 int rv = 0;
602
603 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
604 /* Remove the reference keeping the lseg in the
605 * list. It will now be removed when all
606 * outstanding io is finished.
607 */
d684d2ae 608 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 609 refcount_read(&lseg->pls_refcount));
24956804 610 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 611 rv = 1;
4541d16c
FI		 612 }
613 return rv;
614}
615
6d597e17
JL		 616/*
617 * Compare 2 layout stateid sequence ids, to see which is newer,
618 * taking into account wraparound issues.
619 */
620static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
621{
622 return (s32)(s1 - s2) > 0;
623}
624
e036f464
TM		 625static bool
626pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
627 const struct pnfs_layout_range *recall_range)
628{
629 return (recall_range->iomode == IOMODE_ANY ||
630 lseg_range->iomode == recall_range->iomode) &&
631 pnfs_lseg_range_intersecting(lseg_range, recall_range);
632}
633
634static bool
635pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
636 const struct pnfs_layout_range *recall_range,
637 u32 seq)
638{
639 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
640 return false;
641 if (recall_range == NULL)
642 return true;
643 return pnfs_should_free_range(&lseg->pls_range, recall_range);
644}
645
6d597e17
JL		 646/**
647 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
648 * @lo: layout header containing the lsegs
649 * @tmp_list: list head where doomed lsegs should go
650 * @recall_range: optional recall range argument to match (may be NULL)
651 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
652 *
653 * Walk the list of lsegs in the layout header, and tear down any that should
654 * be destroyed. If "recall_range" is specified then the segment must match
655 * that range. If "seq" is non-zero, then only match segments that were handed
656 * out at or before that sequence.
657 *
658 * Returns number of matching invalid lsegs remaining in list after scanning
659 * it and purging them.
4541d16c 660 */
43f1b3da 661int
49a85061 662pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 663 struct list_head *tmp_list,
6d597e17
JL		 664 const struct pnfs_layout_range *recall_range,
665 u32 seq)
974cec8c
AA		 666{
667 struct pnfs_layout_segment *lseg, *next;
71b39854 668 int remaining = 0;
974cec8c
AA		 669
670 dprintk("%s:Begin lo %p\n", __func__, lo);
671
8006bfba 672 if (list_empty(&lo->plh_segs))
38511722 673 return 0;
4541d16c 674 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 675 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 676 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 677 "offset %llu length %llu\n", __func__,
6d597e17
JL		 678 lseg, lseg->pls_range.iomode, lseg->pls_seq,
679 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 680 if (!mark_lseg_invalid(lseg, tmp_list))
681 remaining++;
4541d16c 682 }
71b39854
TM		 683 dprintk("%s:Return %i\n", __func__, remaining);
684 return remaining;
974cec8c
AA		 685}
686
68f74479
TM		 687static void
688pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
689 struct list_head *free_me,
690 const struct pnfs_layout_range *range,
691 u32 seq)
692{
693 struct pnfs_layout_segment *lseg, *next;
694
695 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
696 if (pnfs_match_lseg_recall(lseg, range, seq))
697 list_move_tail(&lseg->pls_list, free_me);
698 }
699}
700
f49f9baa 701/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 702void
4541d16c 703pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 704{
4541d16c 705 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 706
707 if (list_empty(free_me))
708 return;
709
4541d16c 710 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 711 list_del(&lseg->pls_list);
905ca191 712 pnfs_free_lseg(lseg);
974cec8c
AA		 713 }
714}
715
e5e94017
BH		 716void
717pnfs_destroy_layout(struct nfs_inode *nfsi)
718{
719 struct pnfs_layout_hdr *lo;
974cec8c 720 LIST_HEAD(tmp_list);
e5e94017
BH		 721
722 spin_lock(&nfsi->vfs_inode.i_lock);
723 lo = nfsi->layout;
724 if (lo) {
3e621214 725 pnfs_get_layout_hdr(lo);
2454dfea 726 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 727 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
728 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
729 spin_unlock(&nfsi->vfs_inode.i_lock);
730 pnfs_free_lseg_list(&tmp_list);
1f18b82c 731 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 732 pnfs_put_layout_hdr(lo);
733 } else
734 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 735}
041245c8 736EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 737
fd9a8d71
TM		 738static bool
739pnfs_layout_add_bulk_destroy_list(struct inode *inode,
740 struct list_head *layout_list)
974cec8c
AA		 741{
742 struct pnfs_layout_hdr *lo;
fd9a8d71 743 bool ret = false;
974cec8c 744
fd9a8d71
TM		 745 spin_lock(&inode->i_lock);
746 lo = NFS_I(inode)->layout;
747 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
748 pnfs_get_layout_hdr(lo);
749 list_add(&lo->plh_bulk_destroy, layout_list);
750 ret = true;
751 }
752 spin_unlock(&inode->i_lock);
753 return ret;
754}
755
756/* Caller must hold rcu_read_lock and clp->cl_lock */
757static int
758pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
759 struct nfs_server *server,
760 struct list_head *layout_list)
5085607d
TM		 761 __must_hold(&clp->cl_lock)
762 __must_hold(RCU)
fd9a8d71
TM		 763{
764 struct pnfs_layout_hdr *lo, *next;
765 struct inode *inode;
766
767 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
5085607d
TM		 768 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
769 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
770 !list_empty(&lo->plh_bulk_destroy))
b85f5620 771 continue;
5085607d
TM		 772 /* If the sb is being destroyed, just bail */
773 if (!nfs_sb_active(server->super))
774 break;
fd9a8d71 775 inode = igrab(lo->plh_inode);
5085607d
TM		 776 if (inode != NULL) {
777 list_del_init(&lo->plh_layouts);
778 if (pnfs_layout_add_bulk_destroy_list(inode,
779 layout_list))
780 continue;
781 rcu_read_unlock();
782 spin_unlock(&clp->cl_lock);
783 iput(inode);
784 } else {
785 rcu_read_unlock();
786 spin_unlock(&clp->cl_lock);
787 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
788 }
789 nfs_sb_deactive(server->super);
fd9a8d71
TM		 790 spin_lock(&clp->cl_lock);
791 rcu_read_lock();
792 return -EAGAIN;
793 }
794 return 0;
795}
796
797static int
798pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
799 bool is_bulk_recall)
800{
801 struct pnfs_layout_hdr *lo;
802 struct inode *inode;
fd9a8d71
TM		 803 LIST_HEAD(lseg_list);
804 int ret = 0;
805
806 while (!list_empty(layout_list)) {
807 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
808 plh_bulk_destroy);
809 dprintk("%s freeing layout for inode %lu\n", __func__,
810 lo->plh_inode->i_ino);
811 inode = lo->plh_inode;
7c5d1875
CH		 812
813 pnfs_layoutcommit_inode(inode, false);
814
fd9a8d71
TM		 815 spin_lock(&inode->i_lock);
816 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 817 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
818 if (is_bulk_recall)
819 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 820 ret = -EAGAIN;
9fd4b9fc 821 }
fd9a8d71
TM		 822 spin_unlock(&inode->i_lock);
823 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 824 /* Free all lsegs that are attached to commit buckets */
825 nfs_commit_inode(inode, 0);
fd9a8d71 826 pnfs_put_layout_hdr(lo);
5085607d 827 nfs_iput_and_deactive(inode);
fd9a8d71
TM		 828 }
829 return ret;
830}
831
832int
833pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
834 struct nfs_fsid *fsid,
835 bool is_recall)
836{
837 struct nfs_server *server;
838 LIST_HEAD(layout_list);
c47abcf8 839
974cec8c 840 spin_lock(&clp->cl_lock);
6382a441 841 rcu_read_lock();
fd9a8d71 842restart:
6382a441 843 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 844 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
845 continue;
846 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
847 server,
848 &layout_list) != 0)
849 goto restart;
6382a441
WAA		 850 }
851 rcu_read_unlock();
974cec8c
AA		 852 spin_unlock(&clp->cl_lock);
853
fd9a8d71
TM		 854 if (list_empty(&layout_list))
855 return 0;
856 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
857}
858
859int
860pnfs_destroy_layouts_byclid(struct nfs_client *clp,
861 bool is_recall)
862{
863 struct nfs_server *server;
864 LIST_HEAD(layout_list);
865
866 spin_lock(&clp->cl_lock);
867 rcu_read_lock();
868restart:
869 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
870 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
871 server,
872 &layout_list) != 0)
873 goto restart;
974cec8c 874 }
fd9a8d71
TM		 875 rcu_read_unlock();
876 spin_unlock(&clp->cl_lock);
877
878 if (list_empty(&layout_list))
879 return 0;
880 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
881}
882
883/*
884 * Called by the state manger to remove all layouts established under an
885 * expired lease.
886 */
887void
888pnfs_destroy_all_layouts(struct nfs_client *clp)
889{
890 nfs4_deviceid_mark_client_invalid(clp);
891 nfs4_deviceid_purge_client(clp);
892
893 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 894}
895
fd6002e9 896/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 897void
898pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
899 bool update_barrier)
b1f69b75 900{
ecebb80b 901 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 902
2d2f24ad
TM		 903 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
904 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 905
906 if (!pnfs_layout_is_valid(lo)) {
907 nfs4_stateid_copy(&lo->plh_stateid, new);
908 lo->plh_barrier = newseq;
909 pnfs_clear_layoutreturn_info(lo);
910 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
911 return;
912 }
913 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 914 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 915 /*
916 * Because of wraparound, we want to keep the barrier
917 * "close" to the current seqids.
918 */
919 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 920 }
ecebb80b
TM		 921 if (update_barrier)
922 new_barrier = be32_to_cpu(new->seqid);
923 else if (new_barrier == 0)
924 return;
2a59a041 925 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 926 lo->plh_barrier = new_barrier;
b1f69b75
AA		 927}
928
cf7d63f1 929static bool
19c54aba
TM		 930pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
931 const nfs4_stateid *stateid)
43f1b3da 932{
19c54aba 933 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 934
19c54aba
TM		 935 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
936}
937
938/* lget is set to 1 if called from inside send_layoutget call chain */
939static bool
e1c06f80 940pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 941{
f7e8917a 942 return lo->plh_block_lgets ||
e1c06f80 943 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 944}
945
5e36e2a9
FI		 946static struct nfs_server *
947pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
948{
949 struct nfs_server *server;
950
78746a38 951 if (inode) {
5e36e2a9 952 server = NFS_SERVER(inode);
78746a38 953 } else {
5e36e2a9
FI		 954 struct dentry *parent_dir = dget_parent(ctx->dentry);
955 server = NFS_SERVER(parent_dir->d_inode);
956 dput(parent_dir);
957 }
958 return server;
959}
960
29a8bfe5
TM		 961static void nfs4_free_pages(struct page **pages, size_t size)
962{
963 int i;
964
965 if (!pages)
966 return;
967
968 for (i = 0; i < size; i++) {
969 if (!pages[i])
970 break;
971 __free_page(pages[i]);
972 }
973 kfree(pages);
974}
975
976static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
977{
978 struct page **pages;
979 int i;
980
a2791d3a 981 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 982 if (!pages) {
983 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
984 return NULL;
985 }
986
987 for (i = 0; i < size; i++) {
988 pages[i] = alloc_page(gfp_flags);
989 if (!pages[i]) {
990 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 991 nfs4_free_pages(pages, i);
29a8bfe5
TM		 992 return NULL;
993 }
994 }
995
996 return pages;
997}
998
587f03de 999static struct nfs4_layoutget *
5e36e2a9 1000pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 1001 struct nfs_open_context *ctx,
2409a976 1002 const nfs4_stateid *stateid,
e144e539 1003 const struct pnfs_layout_range *range,
587f03de 1004 gfp_t gfp_flags)
e5e94017 1005{
5e36e2a9 1006 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 1007 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 1008 size_t max_pages = max_response_pages(server);
b1f69b75 1009 struct nfs4_layoutget *lgp;
b1f69b75
AA		 1010
1011 dprintk("--> %s\n", __func__);
e5e94017 1012
83026d80
JL		 1013 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1014 if (lgp == NULL)
587f03de 1015 return NULL;
83026d80 1016
28ced9a8
TM		 1017 if (max_reply_sz) {
1018 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1019 if (npages < max_pages)
1020 max_pages = npages;
1021 }
1022
dacb452d
FI		 1023 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1024 if (!lgp->args.layout.pages) {
1025 kfree(lgp);
1026 return NULL;
1027 }
1028 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1029 lgp->res.layoutp = &lgp->args.layout;
1030
d49e0d5b
TM		 1031 /* Don't confuse uninitialised result and success */
1032 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1033
1034 lgp->args.minlength = PAGE_SIZE;
1035 if (lgp->args.minlength > range->length)
1036 lgp->args.minlength = range->length;
5e36e2a9
FI		 1037 if (ino) {
1038 loff_t i_size = i_size_read(ino);
1039
1040 if (range->iomode == IOMODE_READ) {
1041 if (range->offset >= i_size)
1042 lgp->args.minlength = 0;
1043 else if (i_size - range->offset < lgp->args.minlength)
1044 lgp->args.minlength = i_size - range->offset;
1045 }
d03ab29d 1046 }
83026d80
JL		 1047 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1048 pnfs_copy_range(&lgp->args.range, range);
1049 lgp->args.type = server->pnfs_curr_ld->id;
1050 lgp->args.inode = ino;
1051 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1052 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1053 lgp->gfp_flags = gfp_flags;
a52458b4 1054 lgp->cred = get_cred(ctx->cred);
587f03de 1055 return lgp;
974cec8c
AA		 1056}
1057
29a8bfe5
TM		 1058void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1059{
1060 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1061
1062 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1063 if (lgp->args.inode)
1064 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
a52458b4 1065 put_cred(lgp->cred);
29a8bfe5
TM		 1066 put_nfs_open_context(lgp->args.ctx);
1067 kfree(lgp);
1068}
1069
24956804
TM		 1070static void pnfs_clear_layoutcommit(struct inode *inode,
1071 struct list_head *head)
1072{
1073 struct nfs_inode *nfsi = NFS_I(inode);
1074 struct pnfs_layout_segment *lseg, *tmp;
1075
1076 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1077 return;
1078 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1079 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1080 continue;
1081 pnfs_lseg_dec_and_remove_zero(lseg, head);
1082 }
1083}
1084
68f74479 1085void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1086 const nfs4_stateid *arg_stateid,
68f74479 1087 const struct pnfs_layout_range *range,
68f74479
TM		 1088 const nfs4_stateid *stateid)
1089{
1090 struct inode *inode = lo->plh_inode;
1091 LIST_HEAD(freeme);
1092
1093 spin_lock(&inode->i_lock);
2a974425
TM		 1094 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1095 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1096 goto out_unlock;
68f74479 1097 if (stateid) {
2a974425
TM		 1098 u32 seq = be32_to_cpu(arg_stateid->seqid);
1099
68f74479
TM		 1100 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1101 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1102 pnfs_set_layout_stateid(lo, stateid, true);
1103 } else
1104 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1105out_unlock:
68f74479
TM		 1106 pnfs_clear_layoutreturn_waitbit(lo);
1107 spin_unlock(&inode->i_lock);
1108 pnfs_free_lseg_list(&freeme);
1109
1110}
1111
13c13a6a 1112static bool
e5fd1904
TM		 1113pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1114 nfs4_stateid *stateid,
1115 enum pnfs_iomode *iomode)
13c13a6a 1116{
bf0291dd
TM		 1117 /* Serialise LAYOUTGET/LAYOUTRETURN */
1118 if (atomic_read(&lo->plh_outstanding) != 0)
1119 return false;
6604b203 1120 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1121 return false;
6604b203 1122 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1123 pnfs_get_layout_hdr(lo);
e5fd1904
TM		 1124 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1125 if (stateid != NULL) {
1126 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1127 if (lo->plh_return_seq != 0)
1128 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1129 }
1130 if (iomode != NULL)
1131 *iomode = lo->plh_return_iomode;
1132 pnfs_clear_layoutreturn_info(lo);
1133 return true;
1134 }
1135 if (stateid != NULL)
1136 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1137 if (iomode != NULL)
1138 *iomode = IOMODE_ANY;
13c13a6a
TM		 1139 return true;
1140}
1141
828ed9ec
TM		 1142static void
1143pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1144 struct pnfs_layout_hdr *lo,
1145 const nfs4_stateid *stateid,
1146 enum pnfs_iomode iomode)
1147{
1148 struct inode *inode = lo->plh_inode;
1149
1150 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1151 args->inode = inode;
1152 args->range.iomode = iomode;
1153 args->range.offset = 0;
1154 args->range.length = NFS4_MAX_UINT64;
1155 args->layout = lo;
1156 nfs4_stateid_copy(&args->stateid, stateid);
1157}
1158
f40eb5d0 1159static int
ed429d6b 1160pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
6c16605d 1161 enum pnfs_iomode iomode, bool sync)
f40eb5d0
PT		 1162{
1163 struct inode *ino = lo->plh_inode;
287bd3e9 1164 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0
PT		 1165 struct nfs4_layoutreturn *lrp;
1166 int status = 0;
1167
e4af440a 1168 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1169 if (unlikely(lrp == NULL)) {
1170 status = -ENOMEM;
1171 spin_lock(&ino->i_lock);
d67ae825 1172 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0
PT		 1173 spin_unlock(&ino->i_lock);
1174 pnfs_put_layout_hdr(lo);
1175 goto out;
1176 }
1177
828ed9ec 1178 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1179 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0
PT		 1180 lrp->clp = NFS_SERVER(ino)->nfs_client;
1181 lrp->cred = lo->plh_lc_cred;
287bd3e9
TM		 1182 if (ld->prepare_layoutreturn)
1183 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1184
6c16605d 1185 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1186out:
1187 dprintk("<-- %s status: %d\n", __func__, status);
1188 return status;
1189}
1190
13c13a6a
TM		 1191/* Return true if layoutreturn is needed */
1192static bool
1193pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1194{
1195 struct pnfs_layout_segment *s;
f0b42981
TM		 1196 enum pnfs_iomode iomode;
1197 u32 seq;
13c13a6a 1198
2370abda 1199 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1200 return false;
1201
f0b42981
TM		 1202 seq = lo->plh_return_seq;
1203 iomode = lo->plh_return_iomode;
1204
1205 /* Defer layoutreturn until all recalled lsegs are done */
13c13a6a 1206 list_for_each_entry(s, &lo->plh_segs, pls_list) {
f0b42981
TM		 1207 if (seq && pnfs_seqid_is_newer(s->pls_seq, seq))
1208 continue;
1209 if (iomode != IOMODE_ANY && s->pls_range.iomode != iomode)
1210 continue;
13c13a6a
TM		 1211 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1212 return false;
1213 }
1214
1215 return true;
1216}
1217
1218static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1219{
1220 struct inode *inode= lo->plh_inode;
1221
2370abda 1222 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1223 return;
1224 spin_lock(&inode->i_lock);
1225 if (pnfs_layout_need_return(lo)) {
1226 nfs4_stateid stateid;
1227 enum pnfs_iomode iomode;
1228 bool send;
1229
e5fd1904 1230 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
13c13a6a
TM		 1231 spin_unlock(&inode->i_lock);
1232 if (send) {
1233 /* Send an async layoutreturn so we dont deadlock */
1234 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1235 }
1236 } else
1237 spin_unlock(&inode->i_lock);
1238}
1239
293b3b06
AA		 1240/*
1241 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1242 * when the layout segment list is empty.
1243 *
1244 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1245 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1246 * deviceid is marked invalid.
1247 */
cbe82603
BH		 1248int
1249_pnfs_return_layout(struct inode *ino)
1250{
1251 struct pnfs_layout_hdr *lo = NULL;
1252 struct nfs_inode *nfsi = NFS_I(ino);
1253 LIST_HEAD(tmp_list);
cbe82603 1254 nfs4_stateid stateid;
24408f52 1255 int status = 0;
93b7f7ad 1256 bool send, valid_layout;
cbe82603 1257
366d5052 1258 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1259
1260 spin_lock(&ino->i_lock);
1261 lo = nfsi->layout;
e5929f3c 1262 if (!lo) {
cbe82603 1263 spin_unlock(&ino->i_lock);
293b3b06
AA		 1264 dprintk("NFS: %s no layout to return\n", __func__);
1265 goto out;
cbe82603 1266 }
cbe82603 1267 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1268 pnfs_get_layout_hdr(lo);
24408f52
TM		 1269 /* Is there an outstanding layoutreturn ? */
1270 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1271 spin_unlock(&ino->i_lock);
1272 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1273 TASK_UNINTERRUPTIBLE))
1274 goto out_put_layout_hdr;
1275 spin_lock(&ino->i_lock);
1276 }
93b7f7ad 1277 valid_layout = pnfs_layout_is_valid(lo);
24956804 1278 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1279 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1280
1281 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1282 struct pnfs_layout_range range = {
1283 .iomode = IOMODE_ANY,
1284 .offset = 0,
1285 .length = NFS4_MAX_UINT64,
1286 };
1287 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1288 }
1289
293b3b06 1290 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1291 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1292 !valid_layout) {
293b3b06 1293 spin_unlock(&ino->i_lock);
293b3b06 1294 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1295 goto out_put_layout_hdr;
293b3b06 1296 }
47abadef 1297
e5fd1904 1298 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603 1299 spin_unlock(&ino->i_lock);
7f27392c 1300 if (send)
ed429d6b 1301 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c 1302out_put_layout_hdr:
ee6625a9 1303 pnfs_free_lseg_list(&tmp_list);
7f27392c 1304 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1305out:
1306 dprintk("<-- %s status: %d\n", __func__, status);
1307 return status;
1308}
1309
24028672
TM		 1310int
1311pnfs_commit_and_return_layout(struct inode *inode)
1312{
1313 struct pnfs_layout_hdr *lo;
1314 int ret;
1315
1316 spin_lock(&inode->i_lock);
1317 lo = NFS_I(inode)->layout;
1318 if (lo == NULL) {
1319 spin_unlock(&inode->i_lock);
1320 return 0;
1321 }
1322 pnfs_get_layout_hdr(lo);
1323 /* Block new layoutgets and read/write to ds */
1324 lo->plh_block_lgets++;
1325 spin_unlock(&inode->i_lock);
1326 filemap_fdatawait(inode->i_mapping);
1327 ret = pnfs_layoutcommit_inode(inode, true);
1328 if (ret == 0)
1329 ret = _pnfs_return_layout(inode);
1330 spin_lock(&inode->i_lock);
1331 lo->plh_block_lgets--;
1332 spin_unlock(&inode->i_lock);
1333 pnfs_put_layout_hdr(lo);
1334 return ret;
1335}
1336
1c5bd76d
TM		 1337bool pnfs_roc(struct inode *ino,
1338 struct nfs4_layoutreturn_args *args,
1339 struct nfs4_layoutreturn_res *res,
a52458b4 1340 const struct cred *cred)
f7e8917a 1341{
40dd4b7a
TM		 1342 struct nfs_inode *nfsi = NFS_I(ino);
1343 struct nfs_open_context *ctx;
1344 struct nfs4_state *state;
f7e8917a 1345 struct pnfs_layout_hdr *lo;
1c5bd76d 1346 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1347 nfs4_stateid stateid;
1c5bd76d
TM		 1348 enum pnfs_iomode iomode = 0;
1349 bool layoutreturn = false, roc = false;
e71708d4 1350 bool skip_read = false;
f7e8917a 1351
1c5bd76d
TM		 1352 if (!nfs_have_layout(ino))
1353 return false;
29ade5db 1354retry:
0de43976 1355 rcu_read_lock();
f7e8917a 1356 spin_lock(&ino->i_lock);
40dd4b7a 1357 lo = nfsi->layout;
0cdc329e 1358 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1359 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1360 lo = NULL;
40dd4b7a 1361 goto out_noroc;
9c6376eb
TM		 1362 }
1363 pnfs_get_layout_hdr(lo);
29ade5db 1364 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1365 spin_unlock(&ino->i_lock);
0de43976 1366 rcu_read_unlock();
29ade5db
TM		 1367 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1368 TASK_UNINTERRUPTIBLE);
1369 pnfs_put_layout_hdr(lo);
1370 goto retry;
1371 }
40dd4b7a 1372
e755d638 1373 /* no roc if we hold a delegation */
e71708d4
TM		 1374 if (nfs4_check_delegation(ino, FMODE_READ)) {
1375 if (nfs4_check_delegation(ino, FMODE_WRITE))
1376 goto out_noroc;
1377 skip_read = true;
1378 }
40dd4b7a 1379
0de43976 1380 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1381 state = ctx->state;
e71708d4
TM		 1382 if (state == NULL)
1383 continue;
40dd4b7a 1384 /* Don't return layout if there is open file state */
e71708d4 1385 if (state->state & FMODE_WRITE)
40dd4b7a 1386 goto out_noroc;
e71708d4
TM		 1387 if (state->state & FMODE_READ)
1388 skip_read = true;
40dd4b7a
TM		 1389 }
1390
e755d638 1391
1c5bd76d 1392 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1393 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1394 continue;
e755d638 1395 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1396 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1397 continue;
1398 /*
1399 * Note: mark lseg for return so pnfs_layout_remove_lseg
1400 * doesn't invalidate the layout for us.
1401 */
1402 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1403 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1404 continue;
1405 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1406 }
1407
1c5bd76d
TM		 1408 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1409 goto out_noroc;
69820d22 1410
500d701f 1411 /* ROC in two conditions:
e755d638
PT		 1412 * 1. there are ROC lsegs
1413 * 2. we don't send layoutreturn
e755d638 1414 */
1c5bd76d
TM		 1415 /* lo ref dropped in pnfs_roc_release() */
1416 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1417 /* If the creds don't match, we can't compound the layoutreturn */
1418 if (!layoutreturn || cred != lo->plh_lc_cred)
1419 goto out_noroc;
1420
1421 roc = layoutreturn;
1422 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1423 res->lrs_present = 0;
1424 layoutreturn = false;
f7e8917a 1425
40dd4b7a 1426out_noroc:
f7e8917a 1427 spin_unlock(&ino->i_lock);
0de43976 1428 rcu_read_unlock();
e755d638 1429 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1430 if (roc) {
1431 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1432 if (ld->prepare_layoutreturn)
1433 ld->prepare_layoutreturn(args);
9c6376eb 1434 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1435 return true;
1436 }
e755d638 1437 if (layoutreturn)
1c5bd76d 1438 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
9c6376eb 1439 pnfs_put_layout_hdr(lo);
287bd3e9 1440 return false;
f7e8917a
FI		 1441}
1442
1c5bd76d
TM		 1443void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1444 struct nfs4_layoutreturn_res *res,
1445 int ret)
f7e8917a 1446{
1c5bd76d
TM		 1447 struct pnfs_layout_hdr *lo = args->layout;
1448 const nfs4_stateid *arg_stateid = NULL;
1449 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1450 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1451
1c5bd76d
TM		 1452 if (ret == 0) {
1453 arg_stateid = &args->stateid;
1454 if (res->lrs_present)
1455 res_stateid = &res->stateid;
1456 }
1457 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1458 res_stateid);
287bd3e9
TM		 1459 if (ld_private && ld_private->ops && ld_private->ops->free)
1460 ld_private->ops->free(ld_private);
1c5bd76d
TM		 1461 pnfs_put_layout_hdr(lo);
1462 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1463}
1464
500d701f
PT		 1465bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1466{
1467 struct nfs_inode *nfsi = NFS_I(ino);
1468 struct pnfs_layout_hdr *lo;
1469 bool sleep = false;
1470
1471 /* we might not have grabbed lo reference. so need to check under
1472 * i_lock */
1473 spin_lock(&ino->i_lock);
1474 lo = nfsi->layout;
ee284e35
TM		 1475 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1476 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1477 sleep = true;
ee284e35 1478 }
500d701f 1479 spin_unlock(&ino->i_lock);
500d701f
PT		 1480 return sleep;
1481}
1482
b1f69b75
AA		 1483/*
1484 * Compare two layout segments for sorting into layout cache.
1485 * We want to preferentially return RW over RO layouts, so ensure those
1486 * are seen first.
1487 */
1488static s64
7dc0ac70 1489pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1490 const struct pnfs_layout_range *l2)
b1f69b75 1491{
fb3296eb
BH		 1492 s64 d;
1493
1494 /* high offset > low offset */
1495 d = l1->offset - l2->offset;
1496 if (d)
1497 return d;
1498
1499 /* short length > long length */
1500 d = l2->length - l1->length;
1501 if (d)
1502 return d;
1503
b1f69b75 1504 /* read > read/write */
fb3296eb 1505 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1506}
1507
03772d2f
TM		 1508static bool
1509pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1510 const struct pnfs_layout_range *l2)
1511{
1512 return pnfs_lseg_range_cmp(l1, l2) > 0;
1513}
1514
1515static bool
1516pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1517 struct pnfs_layout_segment *old)
1518{
1519 return false;
1520}
1521
1522void
1523pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1524 struct pnfs_layout_segment *lseg,
1525 bool (*is_after)(const struct pnfs_layout_range *,
1526 const struct pnfs_layout_range *),
1527 bool (*do_merge)(struct pnfs_layout_segment *,
1528 struct pnfs_layout_segment *),
1529 struct list_head *free_me)
974cec8c 1530{
03772d2f 1531 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1532
974cec8c
AA		 1533 dprintk("%s:Begin\n", __func__);
1534
03772d2f
TM		 1535 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1536 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1537 continue;
1538 if (do_merge(lseg, lp)) {
1539 mark_lseg_invalid(lp, free_me);
1540 continue;
1541 }
1542 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1543 continue;
566052c5 1544 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1545 dprintk("%s: inserted lseg %p "
1546 "iomode %d offset %llu length %llu before "
1547 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1548 __func__, lseg, lseg->pls_range.iomode,
1549 lseg->pls_range.offset, lseg->pls_range.length,
1550 lp, lp->pls_range.iomode, lp->pls_range.offset,
1551 lp->pls_range.length);
fb3296eb 1552 goto out;
974cec8c 1553 }
fb3296eb
BH		 1554 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1555 dprintk("%s: inserted lseg %p "
1556 "iomode %d offset %llu length %llu at tail\n",
1557 __func__, lseg, lseg->pls_range.iomode,
1558 lseg->pls_range.offset, lseg->pls_range.length);
1559out:
70c3bd2b 1560 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1561
1562 dprintk("%s:Return\n", __func__);
e5e94017 1563}
03772d2f
TM		 1564EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1565
1566static void
1567pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1568 struct pnfs_layout_segment *lseg,
1569 struct list_head *free_me)
1570{
1571 struct inode *inode = lo->plh_inode;
1572 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1573
1574 if (ld->add_lseg != NULL)
1575 ld->add_lseg(lo, lseg, free_me);
1576 else
1577 pnfs_generic_layout_insert_lseg(lo, lseg,
1578 pnfs_lseg_range_is_after,
1579 pnfs_lseg_no_merge,
1580 free_me);
1581}
e5e94017
BH		 1582
1583static struct pnfs_layout_hdr *
9fa40758
PT		 1584alloc_init_layout_hdr(struct inode *ino,
1585 struct nfs_open_context *ctx,
1586 gfp_t gfp_flags)
e5e94017
BH		 1587{
1588 struct pnfs_layout_hdr *lo;
1589
636fb9c8 1590 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1591 if (!lo)
1592 return NULL;
2b28a7be 1593 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1594 INIT_LIST_HEAD(&lo->plh_layouts);
1595 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1596 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1597 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1598 lo->plh_inode = ino;
a52458b4 1599 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1600 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1601 return lo;
1602}
1603
1604static struct pnfs_layout_hdr *
9fa40758
PT		 1605pnfs_find_alloc_layout(struct inode *ino,
1606 struct nfs_open_context *ctx,
1607 gfp_t gfp_flags)
e5241e43
TM		 1608 __releases(&ino->i_lock)
1609 __acquires(&ino->i_lock)
e5e94017
BH		 1610{
1611 struct nfs_inode *nfsi = NFS_I(ino);
1612 struct pnfs_layout_hdr *new = NULL;
1613
1614 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1615
251ec410
TM		 1616 if (nfsi->layout != NULL)
1617 goto out_existing;
e5e94017 1618 spin_unlock(&ino->i_lock);
9fa40758 1619 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1620 spin_lock(&ino->i_lock);
1621
251ec410 1622 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1623 nfsi->layout = new;
251ec410 1624 return new;
7175fe90
YN		 1625 } else if (new != NULL)
1626 pnfs_free_layout_hdr(new);
251ec410
TM		 1627out_existing:
1628 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1629 return nfsi->layout;
1630}
1631
b1f69b75
AA		 1632/*
1633 * iomode matching rules:
c7d73af2
TH		 1634 * iomode lseg strict match
1635 * iomode
1636 * ----- ----- ------ -----
1637 * ANY READ N/A true
1638 * ANY RW N/A true
1639 * RW READ N/A false
1640 * RW RW N/A true
1641 * READ READ N/A true
1642 * READ RW true false
1643 * READ RW false true
b1f69b75 1644 */
3cb2df17 1645static bool
7dc0ac70 1646pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1647 const struct pnfs_layout_range *range,
1648 bool strict_iomode)
b1f69b75 1649{
fb3296eb
BH		 1650 struct pnfs_layout_range range1;
1651
1652 if ((range->iomode == IOMODE_RW &&
1653 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1654 (range->iomode != ls_range->iomode &&
6089dd0d 1655 strict_iomode) ||
7dc0ac70 1656 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1657 return false;
fb3296eb
BH		 1658
1659 /* range1 covers only the first byte in the range */
1660 range1 = *range;
1661 range1.length = 1;
7dc0ac70 1662 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1663}
1664
1665/*
1666 * lookup range in layout
1667 */
e5e94017 1668static struct pnfs_layout_segment *
fb3296eb 1669pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1670 struct pnfs_layout_range *range,
1671 bool strict_iomode)
e5e94017 1672{
b1f69b75
AA		 1673 struct pnfs_layout_segment *lseg, *ret = NULL;
1674
1675 dprintk("%s:Begin\n", __func__);
1676
b7edfaa1 1677 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1678 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1679 pnfs_lseg_range_match(&lseg->pls_range, range,
1680 strict_iomode)) {
9369a431 1681 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1682 break;
1683 }
b1f69b75
AA		 1684 }
1685
1686 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1687 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1688 return ret;
e5e94017
BH		 1689}
1690
d23d61c8
AA		 1691/*
1692 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1693 * to the MDS or over pNFS
1694 *
1695 * The nfs_inode read_io and write_io fields are cumulative counters reset
1696 * when there are no layout segments. Note that in pnfs_update_layout iomode
1697 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1698 * WRITE request.
1699 *
1700 * A return of true means use MDS I/O.
1701 *
1702 * From rfc 5661:
1703 * If a file's size is smaller than the file size threshold, data accesses
1704 * SHOULD be sent to the metadata server. If an I/O request has a length that
1705 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1706 * server. If both file size and I/O size are provided, the client SHOULD
1707 * reach or exceed both thresholds before sending its read or write
1708 * requests to the data server.
1709 */
1710static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1711 struct inode *ino, int iomode)
1712{
1713 struct nfs4_threshold *t = ctx->mdsthreshold;
1714 struct nfs_inode *nfsi = NFS_I(ino);
1715 loff_t fsize = i_size_read(ino);
1716 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1717
1718 if (t == NULL)
1719 return ret;
1720
1721 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1722 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1723
1724 switch (iomode) {
1725 case IOMODE_READ:
1726 if (t->bm & THRESHOLD_RD) {
1727 dprintk("%s fsize %llu\n", __func__, fsize);
1728 size_set = true;
1729 if (fsize < t->rd_sz)
1730 size = true;
1731 }
1732 if (t->bm & THRESHOLD_RD_IO) {
1733 dprintk("%s nfsi->read_io %llu\n", __func__,
1734 nfsi->read_io);
1735 io_set = true;
1736 if (nfsi->read_io < t->rd_io_sz)
1737 io = true;
1738 }
1739 break;
1740 case IOMODE_RW:
1741 if (t->bm & THRESHOLD_WR) {
1742 dprintk("%s fsize %llu\n", __func__, fsize);
1743 size_set = true;
1744 if (fsize < t->wr_sz)
1745 size = true;
1746 }
1747 if (t->bm & THRESHOLD_WR_IO) {
1748 dprintk("%s nfsi->write_io %llu\n", __func__,
1749 nfsi->write_io);
1750 io_set = true;
1751 if (nfsi->write_io < t->wr_io_sz)
1752 io = true;
1753 }
1754 break;
1755 }
1756 if (size_set && io_set) {
1757 if (size && io)
1758 ret = true;
1759 } else if (size || io)
1760 ret = true;
1761
1762 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1763 return ret;
1764}
1765
d03360aa 1766static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1767{
1768 /*
1769 * send layoutcommit as it can hold up layoutreturn due to lseg
1770 * reference
1771 */
1772 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1773 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1774 nfs_wait_bit_killable,
d03360aa 1775 TASK_KILLABLE);
aa8a45ee
PT		 1776}
1777
411ae722
TM		 1778static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1779{
1780 atomic_inc(&lo->plh_outstanding);
1781}
1782
1783static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1784{
1785 if (atomic_dec_and_test(&lo->plh_outstanding))
1786 wake_up_var(&lo->plh_outstanding);
1787}
1788
d67ae825
TH		 1789static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1790{
1791 unsigned long *bitlock = &lo->plh_flags;
1792
1793 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1794 smp_mb__after_atomic();
1795 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1796}
1797
78746a38
FI		 1798static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1799 struct nfs_server *server)
1800{
1801 if (list_empty(&lo->plh_layouts)) {
1802 struct nfs_client *clp = server->nfs_client;
1803
1804 /* The lo must be on the clp list if there is any
1805 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1806 */
1807 spin_lock(&clp->cl_lock);
1808 if (list_empty(&lo->plh_layouts))
1809 list_add_tail(&lo->plh_layouts, &server->layouts);
1810 spin_unlock(&clp->cl_lock);
1811 }
1812}
1813
e5e94017
BH		 1814/*
1815 * Layout segment is retreived from the server if not cached.
1816 * The appropriate layout segment is referenced and returned to the caller.
1817 */
7c24d948 1818struct pnfs_layout_segment *
e5e94017
BH		 1819pnfs_update_layout(struct inode *ino,
1820 struct nfs_open_context *ctx,
fb3296eb
BH		 1821 loff_t pos,
1822 u64 count,
a75b9df9 1823 enum pnfs_iomode iomode,
c7d73af2 1824 bool strict_iomode,
a75b9df9 1825 gfp_t gfp_flags)
e5e94017 1826{
fb3296eb
BH		 1827 struct pnfs_layout_range arg = {
1828 .iomode = iomode,
1829 .offset = pos,
1830 .length = count,
1831 };
70d2f7b1 1832 unsigned pg_offset;
6382a441
WAA		 1833 struct nfs_server *server = NFS_SERVER(ino);
1834 struct nfs_client *clp = server->nfs_client;
183d9e7b 1835 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1836 struct pnfs_layout_segment *lseg = NULL;
587f03de 1837 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1838 nfs4_stateid stateid;
1839 long timeout = 0;
66b53f32 1840 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1841 bool first;
e5e94017 1842
9a4bf31d 1843 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1844 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1845 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1846 goto out;
9a4bf31d 1847 }
d23d61c8 1848
9a4bf31d 1849 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1850 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1851 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1852 goto out;
9a4bf31d 1853 }
d23d61c8 1854
9bf87482 1855lookup_again:
d03360aa
TM		 1856 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1857 if (IS_ERR(lseg))
1858 goto out;
9bf87482 1859 first = false;
e5e94017 1860 spin_lock(&ino->i_lock);
9fa40758 1861 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1862 if (lo == NULL) {
1863 spin_unlock(&ino->i_lock);
183d9e7b 1864 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1865 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1866 goto out;
1867 }
e5e94017 1868
43f1b3da 1869 /* Do we even need to bother with this? */
a59c30ac 1870 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1871 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1872 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1873 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1874 goto out_unlock;
1875 }
1876
1877 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1878 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1879 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1880 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1881 goto out_unlock;
9a4bf31d 1882 }
e5e94017 1883
411ae722
TM		 1884 /*
1885 * If the layout segment list is empty, but there are outstanding
1886 * layoutget calls, then they might be subject to a layoutrecall.
1887 */
1888 if (list_empty(&lo->plh_segs) &&
1889 atomic_read(&lo->plh_outstanding) != 0) {
1890 spin_unlock(&ino->i_lock);
d03360aa 1891 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
400417b0 1892 !atomic_read(&lo->plh_outstanding)));
d03360aa 1893 if (IS_ERR(lseg) || !list_empty(&lo->plh_segs))
411ae722
TM		 1894 goto out_put_layout_hdr;
1895 pnfs_put_layout_hdr(lo);
1896 goto lookup_again;
1897 }
1898
c7d73af2 1899 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1900 if (lseg) {
1901 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1902 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1903 goto out_unlock;
1904 }
1905
1906 if (!nfs4_valid_open_stateid(ctx->state)) {
1907 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1908 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1909 goto out_unlock;
1910 }
1911
1912 /*
1913 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1914 * stateid, or it has been invalidated, then we must use the open
1915 * stateid.
1916 */
67a3b721 1917 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
183d9e7b
JL		 1918
1919 /*
1920 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1921 * RFC 5661 Errata 3208.
1922 */
1923 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1924 &lo->plh_flags)) {
1925 spin_unlock(&ino->i_lock);
d03360aa
TM		 1926 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
1927 NFS_LAYOUT_FIRST_LAYOUTGET,
1928 TASK_KILLABLE));
1929 if (IS_ERR(lseg))
1930 goto out_put_layout_hdr;
9bf87482 1931 pnfs_put_layout_hdr(lo);
183d9e7b 1932 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1933 goto lookup_again;
1934 }
183d9e7b
JL		 1935
1936 first = true;
70d2f7b1
TM		 1937 if (nfs4_select_rw_stateid(ctx->state,
1938 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1939 NULL, &stateid, NULL) != 0) {
1940 trace_pnfs_update_layout(ino, pos, count,
1941 iomode, lo, lseg,
1942 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1943 goto out_unlock;
1944 }
9bf87482 1945 } else {
183d9e7b 1946 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 1947 }
568e8c49 1948
aa8a45ee
PT		 1949 /*
1950 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1951 * for LAYOUTRETURN even if first is true.
1952 */
8f70f53a 1953 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 1954 spin_unlock(&ino->i_lock);
1955 dprintk("%s wait for layoutreturn\n", __func__);
d03360aa
TM		 1956 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
1957 if (!IS_ERR(lseg)) {
d67ae825
TH		 1958 if (first)
1959 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 1960 pnfs_put_layout_hdr(lo);
1961 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 1962 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1963 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 1964 goto lookup_again;
1965 }
183d9e7b 1966 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1967 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 1968 goto out_put_layout_hdr;
1969 }
1970
9a4bf31d 1971 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 1972 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1973 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 1974 goto out_unlock;
9a4bf31d 1975 }
411ae722 1976 nfs_layoutget_begin(lo);
f49f9baa 1977 spin_unlock(&ino->i_lock);
30005121 1978
78746a38 1979 _add_to_server_list(lo, server);
e5e94017 1980
09cbfeaf 1981 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 1982 if (pg_offset) {
1983 arg.offset -= pg_offset;
1984 arg.length += pg_offset;
1985 }
7c24d948 1986 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 1987 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 1988
5e36e2a9 1989 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 1990 if (!lgp) {
1991 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
1992 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 1993 nfs_layoutget_end(lo);
587f03de
FI		 1994 goto out_put_layout_hdr;
1995 }
1996
dacb452d 1997 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 1998 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1999 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 2000 nfs_layoutget_end(lo);
83026d80 2001 if (IS_ERR(lseg)) {
183d9e7b 2002 switch(PTR_ERR(lseg)) {
e85d7ee4 2003 case -EBUSY:
183d9e7b
JL		 2004 if (time_after(jiffies, giveup))
2005 lseg = NULL;
66b53f32
TM		 2006 break;
2007 case -ERECALLCONFLICT:
183d9e7b 2008 case -EAGAIN:
56b38a1f 2009 break;
183d9e7b
JL		 2010 default:
2011 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2012 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2013 lseg = NULL;
2014 }
56b38a1f
TM		 2015 goto out_put_layout_hdr;
2016 }
2017 if (lseg) {
2018 if (first)
2019 pnfs_clear_first_layoutget(lo);
2020 trace_pnfs_update_layout(ino, pos, count,
2021 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2022 pnfs_put_layout_hdr(lo);
2023 goto lookup_again;
83026d80
JL		 2024 }
2025 } else {
2026 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2027 }
2028
830ffb56 2029out_put_layout_hdr:
d67ae825
TH		 2030 if (first)
2031 pnfs_clear_first_layoutget(lo);
70c3bd2b 2032 pnfs_put_layout_hdr(lo);
e5e94017 2033out:
f86bbcf8
TM		 2034 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2035 "(%s, offset: %llu, length: %llu)\n",
2036 __func__, ino->i_sb->s_id,
2037 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2038 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2039 iomode==IOMODE_RW ? "read/write" : "read-only",
2040 (unsigned long long)pos,
2041 (unsigned long long)count);
e5e94017
BH		 2042 return lseg;
2043out_unlock:
2044 spin_unlock(&ino->i_lock);
830ffb56 2045 goto out_put_layout_hdr;
e5e94017 2046}
7c24d948 2047EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2048
540d9864
TM		 2049static bool
2050pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2051{
2052 switch (range->iomode) {
2053 case IOMODE_READ:
2054 case IOMODE_RW:
2055 break;
2056 default:
2057 return false;
2058 }
2059 if (range->offset == NFS4_MAX_UINT64)
2060 return false;
2061 if (range->length == 0)
2062 return false;
2063 if (range->length != NFS4_MAX_UINT64 &&
2064 range->length > NFS4_MAX_UINT64 - range->offset)
2065 return false;
2066 return true;
2067}
2068
78746a38
FI		 2069static struct pnfs_layout_hdr *
2070_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2071{
2072 struct pnfs_layout_hdr *lo;
2073
2074 spin_lock(&ino->i_lock);
2075 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2076 if (!lo)
2077 goto out_unlock;
2078 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2079 goto out_unlock;
2080 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2081 goto out_unlock;
2082 if (pnfs_layoutgets_blocked(lo))
2083 goto out_unlock;
2084 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2085 goto out_unlock;
411ae722 2086 nfs_layoutget_begin(lo);
78746a38
FI		 2087 spin_unlock(&ino->i_lock);
2088 _add_to_server_list(lo, NFS_SERVER(ino));
2089 return lo;
2090
2091out_unlock:
2092 spin_unlock(&ino->i_lock);
2093 pnfs_put_layout_hdr(lo);
2094 return NULL;
2095}
2096
2409a976
FI		 2097extern const nfs4_stateid current_stateid;
2098
2099static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2100 struct nfs_open_context *ctx)
2101{
78746a38
FI		 2102 struct inode *ino = data->dentry->d_inode;
2103 struct pnfs_layout_range rng = {
2104 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2105 IOMODE_RW: IOMODE_READ,
2106 .offset = 0,
2107 .length = NFS4_MAX_UINT64,
2108 };
2109 struct nfs4_layoutget *lgp;
2110 struct pnfs_layout_hdr *lo;
2111
64294b08
TM		 2112 /* Heuristic: don't send layoutget if we have cached data */
2113 if (rng.iomode == IOMODE_READ &&
2114 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2115 return;
2116
78746a38
FI		 2117 lo = _pnfs_grab_empty_layout(ino, ctx);
2118 if (!lo)
2119 return;
2120 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2121 &rng, GFP_KERNEL);
2122 if (!lgp) {
2123 pnfs_clear_first_layoutget(lo);
2124 pnfs_put_layout_hdr(lo);
2125 return;
2126 }
2127 data->lgp = lgp;
2128 data->o_arg.lg_args = &lgp->args;
2129 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2130}
2131
2132static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2133 struct nfs_open_context *ctx)
2134{
2135 struct pnfs_layout_range rng = {
2136 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2137 IOMODE_RW: IOMODE_READ,
2138 .offset = 0,
2139 .length = NFS4_MAX_UINT64,
2140 };
2141 struct nfs4_layoutget *lgp;
2142
2143 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2144 &rng, GFP_KERNEL);
2145 if (!lgp)
2146 return;
2147 data->lgp = lgp;
2148 data->o_arg.lg_args = &lgp->args;
2149 data->o_res.lg_res = &lgp->res;
2150}
2151
2152void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2153 struct nfs_open_context *ctx)
2154{
2155 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2156
2157 if (!(pnfs_enabled_sb(server) &&
2158 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2159 return;
2160 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2161 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2162 return;
2409a976
FI		 2163 if (data->state)
2164 _lgopen_prepare_attached(data, ctx);
2165 else
2166 _lgopen_prepare_floating(data, ctx);
2167}
2168
2169void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2170 struct nfs_open_context *ctx)
2171{
2172 struct pnfs_layout_hdr *lo;
2173 struct pnfs_layout_segment *lseg;
c49b5209 2174 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2175 u32 iomode;
2176
6e01260c 2177 if (!lgp)
2409a976 2178 return;
6e01260c
FI		 2179 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2180 if (lgp->res.status) {
2181 switch (lgp->res.status) {
6e01260c 2182 default:
8dc96566
TM		 2183 break;
2184 /*
2185 * Halt lgopen attempts if the server doesn't recognise
2186 * the "current stateid" value, the layout type, or the
2187 * layoutget operation as being valid.
2188 * Also if it complains about too many ops in the compound
2189 * or of the request/reply being too big.
2190 */
2191 case -NFS4ERR_BAD_STATEID:
2192 case -NFS4ERR_NOTSUPP:
2193 case -NFS4ERR_REP_TOO_BIG:
2194 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2195 case -NFS4ERR_REQ_TOO_BIG:
2196 case -NFS4ERR_TOO_MANY_OPS:
2197 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2198 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2199 }
2200 return;
2201 }
2409a976 2202 if (!lgp->args.inode) {
78746a38
FI		 2203 lo = _pnfs_grab_empty_layout(ino, ctx);
2204 if (!lo)
2205 return;
2409a976
FI		 2206 lgp->args.inode = ino;
2207 } else
2208 lo = NFS_I(lgp->args.inode)->layout;
2409a976
FI		 2209
2210 lseg = pnfs_layout_process(lgp);
32f1c28f 2211 if (!IS_ERR(lseg)) {
2409a976
FI		 2212 iomode = lgp->args.range.iomode;
2213 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2214 pnfs_put_lseg(lseg);
2215 }
30ae2412
FI		 2216}
2217
2218void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2219{
2220 if (lgp != NULL) {
2221 struct inode *inode = lgp->args.inode;
2222 if (inode) {
2223 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
30ae2412 2224 pnfs_clear_first_layoutget(lo);
411ae722 2225 nfs_layoutget_end(lo);
30ae2412
FI		 2226 }
2227 pnfs_layoutget_free(lgp);
2228 }
2409a976
FI		 2229}
2230
a0b0a6e3 2231struct pnfs_layout_segment *
b1f69b75
AA		 2232pnfs_layout_process(struct nfs4_layoutget *lgp)
2233{
2234 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2235 struct nfs4_layoutget_res *res = &lgp->res;
2236 struct pnfs_layout_segment *lseg;
b7edfaa1 2237 struct inode *ino = lo->plh_inode;
78096cca 2238 LIST_HEAD(free_me);
540d9864
TM		 2239
2240 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2241 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2242
2243 /* Inject layout blob into I/O device driver */
a75b9df9 2244 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2245 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2246 if (!lseg)
1b3c6d07
JL		 2247 lseg = ERR_PTR(-ENOMEM);
2248
2249 dprintk("%s: Could not allocate layout: error %ld\n",
2250 __func__, PTR_ERR(lseg));
2251 return lseg;
b1f69b75
AA		 2252 }
2253
119cef97 2254 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2255
b1f69b75 2256 spin_lock(&ino->i_lock);
e1c06f80 2257 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2258 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2259 goto out_forget;
43f1b3da 2260 }
038d6493 2261
9888d837
TM		 2262 if (!pnfs_layout_is_valid(lo)) {
2263 /* We have a completely new layout */
2264 pnfs_set_layout_stateid(lo, &res->stateid, true);
2265 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2266 /* existing state ID, make sure the sequence number matches. */
2267 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2268 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2269 goto out_forget;
362f7474
CH		 2270 }
2271 pnfs_set_layout_stateid(lo, &res->stateid, false);
2272 } else {
2273 /*
2274 * We got an entirely new state ID. Mark all segments for the
9888d837 2275 * inode invalid, and retry the layoutget
362f7474 2276 */
d9b61708 2277 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 2278 goto out_forget;
362f7474 2279 }
038d6493 2280
9369a431 2281 pnfs_get_lseg(lseg);
03772d2f 2282 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2283
b1f69b75 2284
3976143b 2285 if (res->return_on_close)
f7e8917a 2286 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2287
b1f69b75 2288 spin_unlock(&ino->i_lock);
78096cca 2289 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2290 return lseg;
43f1b3da 2291
1b3c6d07 2292out_forget:
43f1b3da
FI		 2293 spin_unlock(&ino->i_lock);
2294 lseg->pls_layout = lo;
2295 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2296 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2297}
2298
e0b7d420
TM		 2299static int
2300mark_lseg_invalid_or_return(struct pnfs_layout_segment *lseg,
2301 struct list_head *tmp_list)
2302{
2303 if (!mark_lseg_invalid(lseg, tmp_list))
2304 return 0;
2305 pnfs_cache_lseg_for_layoutreturn(lseg->pls_layout, lseg);
2306 return 1;
2307}
2308
2f215968
TM		 2309/**
2310 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2311 * @lo: pointer to layout header
2312 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2313 * @return_range: describe layout segment ranges to be returned
e0b7d420 2314 * @seq: stateid seqid to match
2f215968
TM		 2315 *
2316 * This function is mainly intended for use by layoutrecall. It attempts
2317 * to free the layout segment immediately, or else to mark it for return
2318 * as soon as its reference count drops to zero.
e0b7d420
TM		 2319 *
2320 * Returns
2321 * - 0: a layoutreturn needs to be scheduled.
2322 * - EBUSY: there are layout segment that are still in use.
2323 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2324 */
10335556 2325int
016256df
PT		 2326pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2327 struct list_head *tmp_list,
6d597e17
JL		 2328 const struct pnfs_layout_range *return_range,
2329 u32 seq)
016256df
PT		 2330{
2331 struct pnfs_layout_segment *lseg, *next;
10335556 2332 int remaining = 0;
016256df
PT		 2333
2334 dprintk("%s:Begin lo %p\n", __func__, lo);
2335
fc7ff367 2336 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2337
2338 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2339 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2340 dprintk("%s: marking lseg %p iomode %d "
2341 "offset %llu length %llu\n", __func__,
2342 lseg, lseg->pls_range.iomode,
2343 lseg->pls_range.offset,
2344 lseg->pls_range.length);
e0b7d420 2345 if (mark_lseg_invalid_or_return(lseg, tmp_list))
2f215968
TM		 2346 continue;
2347 remaining++;
016256df 2348 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2349 }
6d597e17 2350
e0b7d420 2351 if (remaining) {
6d597e17 2352 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2353 return -EBUSY;
2354 }
6d597e17 2355
e0b7d420
TM		 2356 if (!list_empty(&lo->plh_return_segs)) {
2357 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2358 return 0;
2359 }
2360
2361 return -ENOENT;
016256df
PT		 2362}
2363
2364void pnfs_error_mark_layout_for_return(struct inode *inode,
2365 struct pnfs_layout_segment *lseg)
2366{
2367 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2368 struct pnfs_layout_range range = {
2369 .iomode = lseg->pls_range.iomode,
2370 .offset = 0,
2371 .length = NFS4_MAX_UINT64,
2372 };
10335556 2373 bool return_now = false;
016256df
PT		 2374
2375 spin_lock(&inode->i_lock);
bdebfccd
TM		 2376 if (!pnfs_layout_is_valid(lo)) {
2377 spin_unlock(&inode->i_lock);
2378 return;
2379 }
2d6cf5ab 2380 pnfs_set_plh_return_info(lo, range.iomode, 0);
016256df
PT		 2381 /*
2382 * mark all matching lsegs so that we are sure to have no live
2383 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2384 * for how it works.
2385 */
e0b7d420 2386 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
10335556 2387 nfs4_stateid stateid;
e5fd1904 2388 enum pnfs_iomode iomode;
10335556 2389
e5fd1904 2390 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2391 spin_unlock(&inode->i_lock);
2392 if (return_now)
2393 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2394 } else {
2395 spin_unlock(&inode->i_lock);
2396 nfs_commit_inode(inode, 0);
2397 }
016256df
PT		 2398}
2399EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2400
b3230e80
TM		 2401void
2402pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2403{
2404 if (pgio->pg_lseg == NULL ||
2405 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2406 return;
2407 pnfs_put_lseg(pgio->pg_lseg);
2408 pgio->pg_lseg = NULL;
2409}
2410EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2411
08cb5b0f
BC		 2412/*
2413 * Check for any intersection between the request and the pgio->pg_lseg,
2414 * and if none, put this pgio->pg_lseg away.
2415 */
2416static void
2417pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2418{
2419 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2420 pnfs_put_lseg(pgio->pg_lseg);
2421 pgio->pg_lseg = NULL;
2422 }
2423}
2424
d8007d4d
TM		 2425void
2426pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2427{
1fd937bd
PT		 2428 u64 rd_size = req->wb_bytes;
2429
b3230e80 2430 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2431 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2432 if (pgio->pg_lseg == NULL) {
2433 if (pgio->pg_dreq == NULL)
2434 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2435 else
2436 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2437
2438 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2439 nfs_req_openctx(req),
cb5d04bc
PT		 2440 req_offset(req),
2441 rd_size,
2442 IOMODE_READ,
c7d73af2 2443 false,
cb5d04bc 2444 GFP_KERNEL);
d600ad1f
PT		 2445 if (IS_ERR(pgio->pg_lseg)) {
2446 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2447 pgio->pg_lseg = NULL;
2448 return;
2449 }
cb5d04bc 2450 }
e885de1a
TM		 2451 /* If no lseg, fall back to read through mds */
2452 if (pgio->pg_lseg == NULL)
1f945357 2453 nfs_pageio_reset_read_mds(pgio);
e885de1a 2454
d8007d4d
TM		 2455}
2456EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2457
2458void
6296556f
PT		 2459pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2460 struct nfs_page *req, u64 wb_size)
d8007d4d 2461{
b3230e80 2462 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2463 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2464 if (pgio->pg_lseg == NULL) {
cb5d04bc 2465 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2466 nfs_req_openctx(req),
cb5d04bc
PT		 2467 req_offset(req),
2468 wb_size,
2469 IOMODE_RW,
c7d73af2 2470 false,
cb5d04bc 2471 GFP_NOFS);
d600ad1f
PT		 2472 if (IS_ERR(pgio->pg_lseg)) {
2473 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2474 pgio->pg_lseg = NULL;
2475 return;
2476 }
2477 }
e885de1a
TM		 2478 /* If no lseg, fall back to write through mds */
2479 if (pgio->pg_lseg == NULL)
1f945357 2480 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2481}
2482EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2483
180bb5ec
WAA		 2484void
2485pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2486{
2487 if (desc->pg_lseg) {
2488 pnfs_put_lseg(desc->pg_lseg);
2489 desc->pg_lseg = NULL;
2490 }
2491}
2492EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2493
b4fdac1a
WAA		 2494/*
2495 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2496 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2497 */
2498size_t
a7d42ddb
WAA		 2499pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2500 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2501{
0f9c429e 2502 unsigned int size;
c5e20cb7 2503 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2504
2505 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2506 if (!size)
2507 return 0;
94ad1c80 2508
19982ba8 2509 /*
c5e20cb7
WAA		 2510 * 'size' contains the number of bytes left in the current page (up
2511 * to the original size asked for in @req->wb_bytes).
2512 *
2513 * Calculate how many bytes are left in the layout segment
2514 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2515 *
2516 * Please also note that 'end_offset' is actually the offset of the
2517 * first byte that lies outside the pnfs_layout_range. FIXME?
2518 *
2519 */
19b54848 2520 if (pgio->pg_lseg) {
17822b20 2521 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2522 pgio->pg_lseg->pls_range.length);
2523 req_start = req_offset(req);
08cb5b0f 2524
c5e20cb7 2525 /* start of request is past the last byte of this segment */
08cb5b0f 2526 if (req_start >= seg_end)
19b54848 2527 return 0;
c5e20cb7
WAA		 2528
2529 /* adjust 'size' iff there are fewer bytes left in the
2530 * segment than what nfs_generic_pg_test returned */
2531 seg_left = seg_end - req_start;
2532 if (seg_left < size)
2533 size = (unsigned int)seg_left;
19b54848 2534 }
0f9c429e 2535
19b54848 2536 return size;
94ad1c80 2537}
89a58e32 2538EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2539
53113ad3 2540int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2541{
2542 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2543
2544 /* Resend all requests through the MDS */
53113ad3
WAA		 2545 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2546 hdr->completion_ops);
c7070113 2547 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2548 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2549}
e7dd79af 2550EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2551
d45f60c6 2552static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2553{
cd841605
FI		 2554
2555 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2556 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2557 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2558 pnfs_return_layout(hdr->inode);
1acbbb4e 2559 }
6c75dc0d 2560 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2561 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2562}
2563
d20581aa
BH		 2564/*
2565 * Called by non rpc-based layout drivers
2566 */
d45f60c6 2567void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2568{
f8417b48 2569 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2570 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2571 hdr->mds_offset + hdr->res.count);
d45f60c6 2572 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2573 }
2574 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2575 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2576 pnfs_ld_handle_write_error(hdr);
2577 hdr->mds_ops->rpc_release(hdr);
44b83799 2578}
d20581aa 2579EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2580
dce81290
TM		 2581static void
2582pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2583 struct nfs_pgio_header *hdr)
dce81290 2584{
48d635f1 2585 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2586
6c75dc0d 2587 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2588 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2589 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2590 mirror->pg_recoalesce = 1;
6c75dc0d 2591 }
ba4a76f7 2592 hdr->completion_ops->completion(hdr);
dce81290
TM		 2593}
2594
2595static enum pnfs_try_status
d45f60c6 2596pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2597 const struct rpc_call_ops *call_ops,
2598 struct pnfs_layout_segment *lseg,
2599 int how)
0382b744 2600{
cd841605 2601 struct inode *inode = hdr->inode;
0382b744
AA		 2602 enum pnfs_try_status trypnfs;
2603 struct nfs_server *nfss = NFS_SERVER(inode);
2604
cd841605 2605 hdr->mds_ops = call_ops;
0382b744
AA		 2606
2607 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2608 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2609 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2610 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2611 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2612 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2613 return trypnfs;
2614}
2615
dce81290 2616static void
7f714720
WAA		 2617pnfs_do_write(struct nfs_pageio_descriptor *desc,
2618 struct nfs_pgio_header *hdr, int how)
dce81290 2619{
dce81290
TM		 2620 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2621 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2622 enum pnfs_try_status trypnfs;
dce81290 2623
d45f60c6 2624 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2625 switch (trypnfs) {
2626 case PNFS_NOT_ATTEMPTED:
d45f60c6 2627 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2628 case PNFS_ATTEMPTED:
2629 break;
2630 case PNFS_TRY_AGAIN:
2631 /* cleanup hdr and prepare to redo pnfs */
2632 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2633 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2634 list_splice_init(&hdr->pages, &mirror->pg_list);
2635 mirror->pg_recoalesce = 1;
2636 }
2637 hdr->mds_ops->rpc_release(hdr);
2638 }
dce81290
TM		 2639}
2640
6c75dc0d
FI		 2641static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2642{
9369a431 2643 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2644 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2645}
2646
dce81290
TM		 2647int
2648pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2649{
6c75dc0d 2650 struct nfs_pgio_header *hdr;
dce81290
TM		 2651 int ret;
2652
1e7f3a48
WAA		 2653 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2654 if (!hdr) {
2bff2288
PT		 2655 desc->pg_error = -ENOMEM;
2656 return desc->pg_error;
dce81290 2657 }
6c75dc0d 2658 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2659
9369a431 2660 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2661 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2662 if (!ret)
7f714720 2663 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2664
6c75dc0d 2665 return ret;
dce81290
TM		 2666}
2667EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2668
53113ad3 2669int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2670{
2671 struct nfs_pageio_descriptor pgio;
2672
1acbbb4e 2673 /* Resend all requests through the MDS */
53113ad3
WAA		 2674 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2675 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2676}
e7dd79af 2677EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2678
d45f60c6 2679static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2680{
cd841605
FI		 2681 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2682 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2683 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2684 pnfs_return_layout(hdr->inode);
1acbbb4e 2685 }
4db6e0b7 2686 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2687 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2688}
2689
d20581aa
BH		 2690/*
2691 * Called by non rpc-based layout drivers
2692 */
d45f60c6 2693void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2694{
bfc505de 2695 if (likely(!hdr->pnfs_error))
d45f60c6 2696 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2697 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2698 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2699 pnfs_ld_handle_read_error(hdr);
2700 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2701}
2702EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2703
493292dd
TM		 2704static void
2705pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2706 struct nfs_pgio_header *hdr)
493292dd 2707{
48d635f1 2708 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2709
4db6e0b7 2710 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2711 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2712 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2713 mirror->pg_recoalesce = 1;
4db6e0b7 2714 }
ba4a76f7 2715 hdr->completion_ops->completion(hdr);
493292dd
TM		 2716}
2717
64419a9b
AA		 2718/*
2719 * Call the appropriate parallel I/O subsystem read function.
2720 */
493292dd 2721static enum pnfs_try_status
d45f60c6 2722pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2723 const struct rpc_call_ops *call_ops,
2724 struct pnfs_layout_segment *lseg)
64419a9b 2725{
cd841605 2726 struct inode *inode = hdr->inode;
64419a9b
AA		 2727 struct nfs_server *nfss = NFS_SERVER(inode);
2728 enum pnfs_try_status trypnfs;
2729
cd841605 2730 hdr->mds_ops = call_ops;
64419a9b
AA		 2731
2732 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2733 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2734
d45f60c6 2735 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2736 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2737 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2738 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2739 return trypnfs;
2740}
863a3c6c 2741
ceb11e13 2742/* Resend all requests through pnfs. */
1b1bc66b 2743void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2744{
2745 struct nfs_pageio_descriptor pgio;
2746
1b1bc66b 2747 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2748 /* Prevent deadlocks with layoutreturn! */
2749 pnfs_put_lseg(hdr->lseg);
2750 hdr->lseg = NULL;
2751
1b1bc66b
WAA		 2752 nfs_pageio_init_read(&pgio, hdr->inode, false,
2753 hdr->completion_ops);
2754 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2755 }
ceb11e13
PT		 2756}
2757EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2758
493292dd 2759static void
7f714720 2760pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2761{
493292dd
TM		 2762 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2763 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2764 enum pnfs_try_status trypnfs;
493292dd 2765
d45f60c6 2766 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2767 switch (trypnfs) {
2768 case PNFS_NOT_ATTEMPTED:
d45f60c6 2769 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2770 case PNFS_ATTEMPTED:
2771 break;
2772 case PNFS_TRY_AGAIN:
2773 /* cleanup hdr and prepare to redo pnfs */
2774 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2775 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2776 list_splice_init(&hdr->pages, &mirror->pg_list);
2777 mirror->pg_recoalesce = 1;
2778 }
2779 hdr->mds_ops->rpc_release(hdr);
2780 }
493292dd
TM		 2781}
2782
4db6e0b7
FI		 2783static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2784{
9369a431 2785 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2786 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2787}
2788
493292dd
TM		 2789int
2790pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2791{
4db6e0b7 2792 struct nfs_pgio_header *hdr;
493292dd
TM		 2793 int ret;
2794
1e7f3a48
WAA		 2795 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2796 if (!hdr) {
2bff2288
PT		 2797 desc->pg_error = -ENOMEM;
2798 return desc->pg_error;
493292dd 2799 }
4db6e0b7 2800 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2801 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2802 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2803 if (!ret)
7f714720 2804 pnfs_do_read(desc, hdr);
4db6e0b7 2805 return ret;
493292dd
TM		 2806}
2807EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2808
71244d9b
TM		 2809static void pnfs_clear_layoutcommitting(struct inode *inode)
2810{
2811 unsigned long *bitlock = &NFS_I(inode)->flags;
2812
2813 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2814 smp_mb__after_atomic();
71244d9b
TM		 2815 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2816}
2817
863a3c6c 2818/*
a9bae566 2819 * There can be multiple RW segments.
863a3c6c 2820 */
a9bae566 2821static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2822{
a9bae566 2823 struct pnfs_layout_segment *lseg;
863a3c6c 2824
a9bae566
PT		 2825 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2826 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2827 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2828 list_add(&lseg->pls_lc_list, listp);
2829 }
863a3c6c
AA		 2830}
2831
a073dbff
TM		 2832static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2833{
2834 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2835
2836 /* Matched by references in pnfs_set_layoutcommit */
2837 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2838 list_del_init(&lseg->pls_lc_list);
2839 pnfs_put_lseg(lseg);
2840 }
2841
71244d9b 2842 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2843}
2844
1b0ae068
PT		 2845void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2846{
b9e028fd 2847 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2848}
2849EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2850
863a3c6c 2851void
67af7611
TM		 2852pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2853 loff_t end_pos)
863a3c6c 2854{
cd841605 2855 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2856 bool mark_as_dirty = false;
863a3c6c 2857
cd841605 2858 spin_lock(&inode->i_lock);
863a3c6c 2859 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2860 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2861 mark_as_dirty = true;
863a3c6c 2862 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2863 __func__, inode->i_ino);
29559b11
TM		 2864 } else if (end_pos > nfsi->layout->plh_lwb)
2865 nfsi->layout->plh_lwb = end_pos;
67af7611 2866 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2867 /* references matched in nfs4_layoutcommit_release */
67af7611 2868 pnfs_get_lseg(lseg);
a9bae566 2869 }
cd841605 2870 spin_unlock(&inode->i_lock);
acff5880 2871 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2872 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2873
2874 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2875 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2876 if (mark_as_dirty)
cd841605 2877 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2878}
2879EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2880
db29c089
AA		 2881void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2882{
2883 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2884
2885 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2886 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2887 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2888}
2889
de4b15c7
AA		 2890/*
2891 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2892 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2893 * data to disk to allow the server to recover the data if it crashes.
2894 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2895 * is off, and a COMMIT is sent to a data server, or
2896 * if WRITEs to a data server return NFS_DATA_SYNC.
2897 */
863a3c6c 2898int
ef311537 2899pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2900{
5f919c9f 2901 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2902 struct nfs4_layoutcommit_data *data;
2903 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2904 loff_t end_pos;
71244d9b 2905 int status;
863a3c6c 2906
71244d9b 2907 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2908 return 0;
2909
71244d9b 2910 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2911
71244d9b 2912 status = -EAGAIN;
92407e75 2913 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2914 if (!sync)
2915 goto out;
74316201 2916 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2917 NFS_INO_LAYOUTCOMMITTING,
2918 nfs_wait_bit_killable,
2919 TASK_KILLABLE);
92407e75 2920 if (status)
71244d9b 2921 goto out;
92407e75
PT		 2922 }
2923
71244d9b
TM		 2924 status = -ENOMEM;
2925 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2926 data = kzalloc(sizeof(*data), GFP_NOFS);
2927 if (!data)
2928 goto clear_layoutcommitting;
2929
2930 status = 0;
de4b15c7 2931 spin_lock(&inode->i_lock);
71244d9b
TM		 2932 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2933 goto out_unlock;
a9bae566 2934
71244d9b 2935 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 2936 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 2937
acff5880 2938 end_pos = nfsi->layout->plh_lwb;
863a3c6c 2939
f597c537 2940 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 2941 spin_unlock(&inode->i_lock);
2942
2943 data->args.inode = inode;
a52458b4 2944 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 2945 nfs_fattr_init(&data->fattr);
2946 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2947 data->res.fattr = &data->fattr;
2e18d4d8
TM		 2948 if (end_pos != 0)
2949 data->args.lastbytewritten = end_pos - 1;
2950 else
2951 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 2952 data->res.server = NFS_SERVER(inode);
2953
5f919c9f
CH		 2954 if (ld->prepare_layoutcommit) {
2955 status = ld->prepare_layoutcommit(&data->args);
2956 if (status) {
a52458b4 2957 put_cred(data->cred);
5f919c9f 2958 spin_lock(&inode->i_lock);
29559b11
TM		 2959 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2960 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 2961 nfsi->layout->plh_lwb = end_pos;
3471648a 2962 goto out_unlock;
5f919c9f
CH		 2963 }
2964 }
2965
2966
863a3c6c
AA		 2967 status = nfs4_proc_layoutcommit(data, sync);
2968out:
92407e75
PT		 2969 if (status)
2970 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2971 dprintk("<-- %s status %d\n", __func__, status);
2972 return status;
71244d9b
TM		 2973out_unlock:
2974 spin_unlock(&inode->i_lock);
92407e75 2975 kfree(data);
71244d9b
TM		 2976clear_layoutcommitting:
2977 pnfs_clear_layoutcommitting(inode);
92407e75 2978 goto out;
863a3c6c 2979}
72cff449 2980EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 2981
5bb89b47
TM		 2982int
2983pnfs_generic_sync(struct inode *inode, bool datasync)
2984{
2985 return pnfs_layoutcommit_inode(inode, true);
2986}
2987EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2988
82be417a
AA		 2989struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2990{
2991 struct nfs4_threshold *thp;
2992
2993 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2994 if (!thp) {
2995 dprintk("%s mdsthreshold allocation failed\n", __func__);
2996 return NULL;
2997 }
2998 return thp;
2999}
8733408d 3000
865a7ecb 3001#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 3002int
c8ad8894 3003pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 3004{
3005 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3006 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 3007 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 3008 struct nfs42_layoutstat_data *data;
3009 struct pnfs_layout_hdr *hdr;
3010 int status = 0;
3011
3012 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3013 goto out;
3014
6c5a0d89
TM		 3015 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3016 goto out;
3017
1bfe3b25
PT		 3018 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3019 goto out;
3020
8733408d
PT		 3021 spin_lock(&inode->i_lock);
3022 if (!NFS_I(inode)->layout) {
3023 spin_unlock(&inode->i_lock);
f538d0ba 3024 goto out_clear_layoutstats;
8733408d
PT		 3025 }
3026 hdr = NFS_I(inode)->layout;
3027 pnfs_get_layout_hdr(hdr);
3028 spin_unlock(&inode->i_lock);
3029
c8ad8894 3030 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3031 if (!data) {
3032 status = -ENOMEM;
3033 goto out_put;
3034 }
3035
3036 data->args.fh = NFS_FH(inode);
3037 data->args.inode = inode;
8733408d
PT		 3038 status = ld->prepare_layoutstats(&data->args);
3039 if (status)
3040 goto out_free;
3041
3042 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3043
3044out:
3045 dprintk("%s returns %d\n", __func__, status);
3046 return status;
3047
3048out_free:
3049 kfree(data);
3050out_put:
3051 pnfs_put_layout_hdr(hdr);
f538d0ba 3052out_clear_layoutstats:
1bfe3b25
PT		 3053 smp_mb__before_atomic();
3054 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3055 smp_mb__after_atomic();
8733408d
PT		 3056 goto out;
3057}
3058EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3059#endif
bbf58bf3
TM		 3060
3061unsigned int layoutstats_timer;
3062module_param(layoutstats_timer, uint, 0644);
3063EXPORT_SYMBOL_GPL(layoutstats_timer);
Linux 6.x block layer and io_uring tree(s)