]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/nfs/pnfs.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
NFSv4: Add a generic structure for managing layout-private information
[mirror_ubuntu-bionic-kernel.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
85e174ba
RL		 40
41#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 42#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 43
02c35fca
FI		 44/* Locking:
45 *
46 * pnfs_spinlock:
47 * protects pnfs_modules_tbl.
48 */
49static DEFINE_SPINLOCK(pnfs_spinlock);
50
51/*
52 * pnfs_modules_tbl holds all pnfs modules
53 */
54static LIST_HEAD(pnfs_modules_tbl);
55
13c13a6a 56static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 57static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
58 struct list_head *free_me,
59 const struct pnfs_layout_range *range,
60 u32 seq);
fe1cf946
TM		 61static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
62 struct list_head *tmp_list);
aa1e0e3a 63
02c35fca
FI		 64/* Return the registered pnfs layout driver module matching given id */
65static struct pnfs_layoutdriver_type *
66find_pnfs_driver_locked(u32 id)
67{
68 struct pnfs_layoutdriver_type *local;
69
70 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
71 if (local->id == id)
72 goto out;
73 local = NULL;
74out:
75 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
76 return local;
77}
78
85e174ba
RL		 79static struct pnfs_layoutdriver_type *
80find_pnfs_driver(u32 id)
81{
02c35fca
FI		 82 struct pnfs_layoutdriver_type *local;
83
84 spin_lock(&pnfs_spinlock);
85 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 86 if (local != NULL && !try_module_get(local->owner)) {
87 dprintk("%s: Could not grab reference on module\n", __func__);
88 local = NULL;
89 }
02c35fca
FI		 90 spin_unlock(&pnfs_spinlock);
91 return local;
85e174ba
RL		 92}
93
94void
95unset_pnfs_layoutdriver(struct nfs_server *nfss)
96{
738fd0f3
BH		 97 if (nfss->pnfs_curr_ld) {
98 if (nfss->pnfs_curr_ld->clear_layoutdriver)
99 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 100 /* Decrement the MDS count. Purge the deviceid cache if zero */
101 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
102 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 103 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 104 }
85e174ba
RL		 105 nfss->pnfs_curr_ld = NULL;
106}
107
ca440c38
JL		 108/*
109 * When the server sends a list of layout types, we choose one in the order
110 * given in the list below.
111 *
112 * FIXME: should this list be configurable in some fashion? module param?
113 * mount option? something else?
114 */
115static const u32 ld_prefs[] = {
116 LAYOUT_SCSI,
117 LAYOUT_BLOCK_VOLUME,
118 LAYOUT_OSD2_OBJECTS,
119 LAYOUT_FLEX_FILES,
120 LAYOUT_NFSV4_1_FILES,
121 0
122};
123
124static int
125ld_cmp(const void *e1, const void *e2)
126{
127 u32 ld1 = *((u32 *)e1);
128 u32 ld2 = *((u32 *)e2);
129 int i;
130
131 for (i = 0; ld_prefs[i] != 0; i++) {
132 if (ld1 == ld_prefs[i])
133 return -1;
134
135 if (ld2 == ld_prefs[i])
136 return 1;
137 }
138 return 0;
139}
140
85e174ba
RL		 141/*
142 * Try to set the server's pnfs module to the pnfs layout type specified by id.
143 * Currently only one pNFS layout driver per filesystem is supported.
144 *
3132e49e 145 * @ids array of layout types supported by MDS.
85e174ba
RL		 146 */
147void
738fd0f3 148set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 149 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 150{
151 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 152 u32 id;
ca440c38 153 int i;
85e174ba 154
19274716
AS		 155 if (fsinfo->nlayouttypes == 0)
156 goto out_no_driver;
85e174ba
RL		 157 if (!(server->nfs_client->cl_exchange_flags &
158 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 159 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
160 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 161 goto out_no_driver;
162 }
3132e49e 163
ca440c38
JL		 164 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
165 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 166
ca440c38
JL		 167 for (i = 0; i < fsinfo->nlayouttypes; i++) {
168 id = fsinfo->layouttype[i];
85e174ba 169 ld_type = find_pnfs_driver(id);
ca440c38
JL		 170 if (!ld_type) {
171 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
172 id);
173 ld_type = find_pnfs_driver(id);
174 }
175 if (ld_type)
176 break;
85e174ba 177 }
3132e49e
JL		 178
179 if (!ld_type) {
ca440c38 180 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 181 goto out_no_driver;
182 }
183
85e174ba 184 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 185 if (ld_type->set_layoutdriver
186 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 187 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
188 "driver %u.\n", __func__, id);
738fd0f3
BH		 189 module_put(ld_type->owner);
190 goto out_no_driver;
191 }
2a4c8994
TM		 192 /* Bump the MDS count */
193 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 194
85e174ba
RL		 195 dprintk("%s: pNFS module for %u set\n", __func__, id);
196 return;
197
198out_no_driver:
199 dprintk("%s: Using NFSv4 I/O\n", __func__);
200 server->pnfs_curr_ld = NULL;
201}
02c35fca
FI		 202
203int
204pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
205{
206 int status = -EINVAL;
207 struct pnfs_layoutdriver_type *tmp;
208
209 if (ld_type->id == 0) {
a030889a 210 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 211 return status;
212 }
b1f69b75 213 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 214 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 215 "alloc_lseg and free_lseg.\n", __func__);
216 return status;
217 }
02c35fca
FI		 218
219 spin_lock(&pnfs_spinlock);
220 tmp = find_pnfs_driver_locked(ld_type->id);
221 if (!tmp) {
222 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
223 status = 0;
224 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
225 ld_type->name);
226 } else {
a030889a 227 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 228 __func__, ld_type->id);
229 }
230 spin_unlock(&pnfs_spinlock);
231
232 return status;
233}
234EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
235
236void
237pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
238{
239 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
240 spin_lock(&pnfs_spinlock);
241 list_del(&ld_type->pnfs_tblid);
242 spin_unlock(&pnfs_spinlock);
243}
244EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 245
b1f69b75
AA		 246/*
247 * pNFS client layout cache
248 */
249
cc6e5340 250/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 251void
70c3bd2b 252pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 253{
cc6e5340 254 atomic_inc(&lo->plh_refcount);
e5e94017
BH		 255}
256
636fb9c8
BH		 257static struct pnfs_layout_hdr *
258pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
259{
260 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 261 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 262}
263
264static void
265pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
266{
9c626381
TM		 267 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
268 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
269
270 if (!list_empty(&lo->plh_layouts)) {
271 struct nfs_client *clp = server->nfs_client;
272
273 spin_lock(&clp->cl_lock);
274 list_del_init(&lo->plh_layouts);
275 spin_unlock(&clp->cl_lock);
276 }
9fa40758 277 put_rpccred(lo->plh_lc_cred);
57934278 278 return ld->free_layout_hdr(lo);
636fb9c8
BH		 279}
280
e5e94017 281static void
6622c3ea 282pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 283{
bb346f63 284 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 285 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 286 nfsi->layout = NULL;
287 /* Reset MDS Threshold I/O counters */
288 nfsi->write_io = 0;
289 nfsi->read_io = 0;
e5e94017
BH		 290}
291
b1f69b75 292void
70c3bd2b 293pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 294{
cc6e5340
FI		 295 struct inode *inode = lo->plh_inode;
296
13c13a6a
TM		 297 pnfs_layoutreturn_before_put_layout_hdr(lo);
298
cc6e5340 299 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 300 if (!list_empty(&lo->plh_segs))
301 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 302 pnfs_detach_layout_hdr(lo);
cc6e5340 303 spin_unlock(&inode->i_lock);
6622c3ea 304 pnfs_free_layout_hdr(lo);
cc6e5340 305 }
974cec8c
AA		 306}
307
4aab9732
TM		 308static void
309pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
310 u32 seq)
311{
312 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
313 iomode = IOMODE_ANY;
314 lo->plh_return_iomode = iomode;
315 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
316 if (seq != 0) {
317 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
318 lo->plh_return_seq = seq;
319 }
320}
321
ae5a459d
TM		 322static void
323pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
324{
325 lo->plh_return_iomode = 0;
326 lo->plh_return_seq = 0;
327 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
328}
329
fe1cf946
TM		 330static void
331pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
332 struct list_head *free_me)
333{
334 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
335 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
336 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
337 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
338 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
339 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
340}
341
2454dfea
TM		 342/*
343 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
344 *
345 * In order to continue using the pnfs_layout_hdr, a full recovery
346 * is required.
347 * Note that caller must hold inode->i_lock.
348 */
5f46be04 349int
2454dfea
TM		 350pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
351 struct list_head *lseg_list)
352{
353 struct pnfs_layout_range range = {
354 .iomode = IOMODE_ANY,
355 .offset = 0,
356 .length = NFS4_MAX_UINT64,
357 };
fe1cf946 358 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 359
360 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
ae5a459d 361 pnfs_clear_layoutreturn_info(lo);
fe1cf946
TM		 362 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
363 pnfs_clear_lseg_state(lseg, lseg_list);
68f74479 364 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
fe1cf946 365 return !list_empty(&lo->plh_segs);
2454dfea
TM		 366}
367
b9e028fd
TM		 368static int
369pnfs_iomode_to_fail_bit(u32 iomode)
370{
371 return iomode == IOMODE_RW ?
372 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
373}
374
375static void
3e621214 376pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 377{
25c75333 378 lo->plh_retry_timestamp = jiffies;
39e88fcf 379 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
3e621214
TM		 380 atomic_inc(&lo->plh_refcount);
381}
382
383static void
384pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
385{
386 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
387 atomic_dec(&lo->plh_refcount);
388}
389
390static void
391pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
392{
393 struct inode *inode = lo->plh_inode;
115ce575
TM		 394 struct pnfs_layout_range range = {
395 .iomode = iomode,
396 .offset = 0,
397 .length = NFS4_MAX_UINT64,
398 };
399 LIST_HEAD(head);
3e621214
TM		 400
401 spin_lock(&inode->i_lock);
402 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 403 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 404 spin_unlock(&inode->i_lock);
115ce575 405 pnfs_free_lseg_list(&head);
b9e028fd
TM		 406 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
407 iomode == IOMODE_RW ? "RW" : "READ");
408}
409
410static bool
411pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
412{
25c75333 413 unsigned long start, end;
3e621214
TM		 414 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
415
416 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 417 return false;
418 end = jiffies;
419 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
420 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
421 /* It is time to retry the failed layoutgets */
3e621214 422 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 423 return false;
424 }
425 return true;
b9e028fd
TM		 426}
427
974cec8c 428static void
119cef97
TM		 429pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
430 const struct pnfs_layout_range *range,
431 const nfs4_stateid *stateid)
974cec8c 432{
566052c5 433 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 434 INIT_LIST_HEAD(&lseg->pls_lc_list);
4541d16c 435 atomic_set(&lseg->pls_refcount, 1);
4541d16c 436 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 437 lseg->pls_layout = lo;
119cef97
TM		 438 lseg->pls_range = *range;
439 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 440}
441
905ca191 442static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 443{
68f74479
TM		 444 if (lseg != NULL) {
445 struct inode *inode = lseg->pls_layout->plh_inode;
446 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
447 }
974cec8c
AA		 448}
449
d684d2ae 450static void
57036a37
TM		 451pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
452 struct pnfs_layout_segment *lseg)
d684d2ae 453{
d20581aa 454 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 455 list_del_init(&lseg->pls_list);
8f0d27dc
TM		 456 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
457 atomic_dec(&lo->plh_refcount);
abb3e1c8
TM		 458 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
459 return;
7b650994
TM		 460 if (list_empty(&lo->plh_segs) &&
461 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
462 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 463 if (atomic_read(&lo->plh_outstanding) == 0)
464 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 465 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 466 }
d684d2ae
FI		 467}
468
68f74479
TM		 469static bool
470pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
471 struct pnfs_layout_segment *lseg)
472{
473 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
474 pnfs_layout_is_valid(lo)) {
4aab9732 475 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 476 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
477 return true;
478 }
479 return false;
480}
481
bae724ef 482void
9369a431 483pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 484{
57036a37 485 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 486 struct inode *inode;
487
488 if (!lseg)
489 return;
490
4541d16c
FI		 491 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
492 atomic_read(&lseg->pls_refcount),
493 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 494
57036a37
TM		 495 lo = lseg->pls_layout;
496 inode = lo->plh_inode;
4ef2e4f8 497
d684d2ae 498 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 499 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
500 spin_unlock(&inode->i_lock);
501 return;
502 }
8f0d27dc 503 pnfs_get_layout_hdr(lo);
4ef2e4f8 504 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 505 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
506 lseg = NULL;
4ef2e4f8
TM		 507 spin_unlock(&inode->i_lock);
508 pnfs_free_lseg(lseg);
509 pnfs_put_layout_hdr(lo);
4541d16c 510 }
4541d16c 511}
9369a431 512EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 513
6543f803 514static void pnfs_free_lseg_async_work(struct work_struct *work)
e6cf82d1
WAA		 515{
516 struct pnfs_layout_segment *lseg;
6543f803 517 struct pnfs_layout_hdr *lo;
e6cf82d1
WAA		 518
519 lseg = container_of(work, struct pnfs_layout_segment, pls_work);
6543f803 520 lo = lseg->pls_layout;
e6cf82d1 521
6543f803
TM		 522 pnfs_free_lseg(lseg);
523 pnfs_put_layout_hdr(lo);
e6cf82d1
WAA		 524}
525
6543f803 526static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
e6cf82d1 527{
6543f803 528 INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
e6cf82d1
WAA		 529 schedule_work(&lseg->pls_work);
530}
6543f803
TM		 531
532void
533pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
534{
535 if (!lseg)
536 return;
537
538 assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);
539
540 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
541 atomic_read(&lseg->pls_refcount),
542 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
543 if (atomic_dec_and_test(&lseg->pls_refcount)) {
544 struct pnfs_layout_hdr *lo = lseg->pls_layout;
faa4a54f
TM		 545 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
546 return;
6543f803 547 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 548 if (!pnfs_cache_lseg_for_layoutreturn(lo, lseg)) {
549 pnfs_get_layout_hdr(lo);
550 pnfs_free_lseg_async(lseg);
551 }
6543f803
TM		 552 }
553}
554EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
e6cf82d1 555
fb3296eb
BH		 556/*
557 * is l2 fully contained in l1?
558 * start1 end1
559 * [----------------------------------)
560 * start2 end2
561 * [----------------)
562 */
3cb2df17 563static bool
7dc0ac70 564pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 565 const struct pnfs_layout_range *l2)
fb3296eb
BH		 566{
567 u64 start1 = l1->offset;
17822b20 568 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 569 u64 start2 = l2->offset;
17822b20 570 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 571
572 return (start1 <= start2) && (end1 >= end2);
573}
574
24956804
TM		 575static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
576 struct list_head *tmp_list)
577{
578 if (!atomic_dec_and_test(&lseg->pls_refcount))
579 return false;
580 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
581 list_add(&lseg->pls_list, tmp_list);
582 return true;
583}
584
4541d16c
FI		 585/* Returns 1 if lseg is removed from list, 0 otherwise */
586static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
587 struct list_head *tmp_list)
588{
589 int rv = 0;
590
591 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
592 /* Remove the reference keeping the lseg in the
593 * list. It will now be removed when all
594 * outstanding io is finished.
595 */
d684d2ae
FI		 596 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
597 atomic_read(&lseg->pls_refcount));
24956804 598 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 599 rv = 1;
4541d16c
FI		 600 }
601 return rv;
602}
603
6d597e17
JL		 604/*
605 * Compare 2 layout stateid sequence ids, to see which is newer,
606 * taking into account wraparound issues.
607 */
608static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
609{
610 return (s32)(s1 - s2) > 0;
611}
612
e036f464
TM		 613static bool
614pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
615 const struct pnfs_layout_range *recall_range)
616{
617 return (recall_range->iomode == IOMODE_ANY ||
618 lseg_range->iomode == recall_range->iomode) &&
619 pnfs_lseg_range_intersecting(lseg_range, recall_range);
620}
621
622static bool
623pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
624 const struct pnfs_layout_range *recall_range,
625 u32 seq)
626{
627 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
628 return false;
629 if (recall_range == NULL)
630 return true;
631 return pnfs_should_free_range(&lseg->pls_range, recall_range);
632}
633
6d597e17
JL		 634/**
635 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
636 * @lo: layout header containing the lsegs
637 * @tmp_list: list head where doomed lsegs should go
638 * @recall_range: optional recall range argument to match (may be NULL)
639 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
640 *
641 * Walk the list of lsegs in the layout header, and tear down any that should
642 * be destroyed. If "recall_range" is specified then the segment must match
643 * that range. If "seq" is non-zero, then only match segments that were handed
644 * out at or before that sequence.
645 *
646 * Returns number of matching invalid lsegs remaining in list after scanning
647 * it and purging them.
4541d16c 648 */
43f1b3da 649int
49a85061 650pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 651 struct list_head *tmp_list,
6d597e17
JL		 652 const struct pnfs_layout_range *recall_range,
653 u32 seq)
974cec8c
AA		 654{
655 struct pnfs_layout_segment *lseg, *next;
71b39854 656 int remaining = 0;
974cec8c
AA		 657
658 dprintk("%s:Begin lo %p\n", __func__, lo);
659
8006bfba 660 if (list_empty(&lo->plh_segs))
38511722 661 return 0;
4541d16c 662 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 663 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
6d597e17 664 dprintk("%s: freeing lseg %p iomode %d seq %u"
4541d16c 665 "offset %llu length %llu\n", __func__,
6d597e17
JL		 666 lseg, lseg->pls_range.iomode, lseg->pls_seq,
667 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 668 if (!mark_lseg_invalid(lseg, tmp_list))
669 remaining++;
4541d16c 670 }
71b39854
TM		 671 dprintk("%s:Return %i\n", __func__, remaining);
672 return remaining;
974cec8c
AA		 673}
674
68f74479
TM		 675static void
676pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
677 struct list_head *free_me,
678 const struct pnfs_layout_range *range,
679 u32 seq)
680{
681 struct pnfs_layout_segment *lseg, *next;
682
683 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
684 if (pnfs_match_lseg_recall(lseg, range, seq))
685 list_move_tail(&lseg->pls_list, free_me);
686 }
687}
688
f49f9baa 689/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 690void
4541d16c 691pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 692{
4541d16c 693 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 694
695 if (list_empty(free_me))
696 return;
697
4541d16c 698 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 699 list_del(&lseg->pls_list);
905ca191 700 pnfs_free_lseg(lseg);
974cec8c
AA		 701 }
702}
703
e5e94017
BH		 704void
705pnfs_destroy_layout(struct nfs_inode *nfsi)
706{
707 struct pnfs_layout_hdr *lo;
974cec8c 708 LIST_HEAD(tmp_list);
e5e94017
BH		 709
710 spin_lock(&nfsi->vfs_inode.i_lock);
711 lo = nfsi->layout;
712 if (lo) {
3e621214 713 pnfs_get_layout_hdr(lo);
2454dfea 714 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 715 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
716 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
717 spin_unlock(&nfsi->vfs_inode.i_lock);
718 pnfs_free_lseg_list(&tmp_list);
719 pnfs_put_layout_hdr(lo);
720 } else
721 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 722}
041245c8 723EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 724
fd9a8d71
TM		 725static bool
726pnfs_layout_add_bulk_destroy_list(struct inode *inode,
727 struct list_head *layout_list)
974cec8c
AA		 728{
729 struct pnfs_layout_hdr *lo;
fd9a8d71 730 bool ret = false;
974cec8c 731
fd9a8d71
TM		 732 spin_lock(&inode->i_lock);
733 lo = NFS_I(inode)->layout;
734 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
735 pnfs_get_layout_hdr(lo);
736 list_add(&lo->plh_bulk_destroy, layout_list);
737 ret = true;
738 }
739 spin_unlock(&inode->i_lock);
740 return ret;
741}
742
743/* Caller must hold rcu_read_lock and clp->cl_lock */
744static int
745pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
746 struct nfs_server *server,
747 struct list_head *layout_list)
748{
749 struct pnfs_layout_hdr *lo, *next;
750 struct inode *inode;
751
752 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
b85f5620
TM		 753 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
754 continue;
fd9a8d71
TM		 755 inode = igrab(lo->plh_inode);
756 if (inode == NULL)
757 continue;
758 list_del_init(&lo->plh_layouts);
759 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
760 continue;
761 rcu_read_unlock();
762 spin_unlock(&clp->cl_lock);
763 iput(inode);
764 spin_lock(&clp->cl_lock);
765 rcu_read_lock();
766 return -EAGAIN;
767 }
768 return 0;
769}
770
771static int
772pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
773 bool is_bulk_recall)
774{
775 struct pnfs_layout_hdr *lo;
776 struct inode *inode;
fd9a8d71
TM		 777 LIST_HEAD(lseg_list);
778 int ret = 0;
779
780 while (!list_empty(layout_list)) {
781 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
782 plh_bulk_destroy);
783 dprintk("%s freeing layout for inode %lu\n", __func__,
784 lo->plh_inode->i_ino);
785 inode = lo->plh_inode;
7c5d1875
CH		 786
787 pnfs_layoutcommit_inode(inode, false);
788
fd9a8d71
TM		 789 spin_lock(&inode->i_lock);
790 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 791 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
792 if (is_bulk_recall)
793 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 794 ret = -EAGAIN;
9fd4b9fc 795 }
fd9a8d71
TM		 796 spin_unlock(&inode->i_lock);
797 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 798 /* Free all lsegs that are attached to commit buckets */
799 nfs_commit_inode(inode, 0);
fd9a8d71
TM		 800 pnfs_put_layout_hdr(lo);
801 iput(inode);
802 }
803 return ret;
804}
805
806int
807pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
808 struct nfs_fsid *fsid,
809 bool is_recall)
810{
811 struct nfs_server *server;
812 LIST_HEAD(layout_list);
c47abcf8 813
974cec8c 814 spin_lock(&clp->cl_lock);
6382a441 815 rcu_read_lock();
fd9a8d71 816restart:
6382a441 817 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 818 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
819 continue;
820 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
821 server,
822 &layout_list) != 0)
823 goto restart;
6382a441
WAA		 824 }
825 rcu_read_unlock();
974cec8c
AA		 826 spin_unlock(&clp->cl_lock);
827
fd9a8d71
TM		 828 if (list_empty(&layout_list))
829 return 0;
830 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
831}
832
833int
834pnfs_destroy_layouts_byclid(struct nfs_client *clp,
835 bool is_recall)
836{
837 struct nfs_server *server;
838 LIST_HEAD(layout_list);
839
840 spin_lock(&clp->cl_lock);
841 rcu_read_lock();
842restart:
843 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
844 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
845 server,
846 &layout_list) != 0)
847 goto restart;
974cec8c 848 }
fd9a8d71
TM		 849 rcu_read_unlock();
850 spin_unlock(&clp->cl_lock);
851
852 if (list_empty(&layout_list))
853 return 0;
854 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
855}
856
857/*
858 * Called by the state manger to remove all layouts established under an
859 * expired lease.
860 */
861void
862pnfs_destroy_all_layouts(struct nfs_client *clp)
863{
864 nfs4_deviceid_mark_client_invalid(clp);
865 nfs4_deviceid_purge_client(clp);
866
867 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 868}
869
fd6002e9 870/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 871void
872pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
873 bool update_barrier)
b1f69b75 874{
ecebb80b 875 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 876
2d2f24ad
TM		 877 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
878 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 879
880 if (!pnfs_layout_is_valid(lo)) {
881 nfs4_stateid_copy(&lo->plh_stateid, new);
882 lo->plh_barrier = newseq;
883 pnfs_clear_layoutreturn_info(lo);
884 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
885 return;
886 }
887 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 888 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 889 /*
890 * Because of wraparound, we want to keep the barrier
891 * "close" to the current seqids.
892 */
893 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 894 }
ecebb80b
TM		 895 if (update_barrier)
896 new_barrier = be32_to_cpu(new->seqid);
897 else if (new_barrier == 0)
898 return;
2a59a041 899 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 900 lo->plh_barrier = new_barrier;
b1f69b75
AA		 901}
902
cf7d63f1 903static bool
19c54aba
TM		 904pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
905 const nfs4_stateid *stateid)
43f1b3da 906{
19c54aba 907 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 908
19c54aba
TM		 909 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
910}
911
912/* lget is set to 1 if called from inside send_layoutget call chain */
913static bool
e1c06f80 914pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 915{
f7e8917a 916 return lo->plh_block_lgets ||
e1c06f80 917 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 918}
919
b1f69b75 920/*
183d9e7b
JL		 921 * Get layout from server.
922 * for now, assume that whole file layouts are requested.
923 * arg->offset: 0
924 * arg->length: all ones
925 */
e5e94017
BH		 926static struct pnfs_layout_segment *
927send_layoutget(struct pnfs_layout_hdr *lo,
928 struct nfs_open_context *ctx,
183d9e7b 929 nfs4_stateid *stateid,
e144e539 930 const struct pnfs_layout_range *range,
183d9e7b 931 long *timeout, gfp_t gfp_flags)
e5e94017 932{
b7edfaa1 933 struct inode *ino = lo->plh_inode;
b1f69b75
AA		 934 struct nfs_server *server = NFS_SERVER(ino);
935 struct nfs4_layoutget *lgp;
2d89a1d3 936 loff_t i_size;
b1f69b75
AA		 937
938 dprintk("--> %s\n", __func__);
e5e94017 939
4f2e9dce
JL		 940 /*
941 * Synchronously retrieve layout information from server and
942 * store in lseg. If we race with a concurrent seqid morphing
943 * op, then re-send the LAYOUTGET.
944 */
83026d80
JL		 945 lgp = kzalloc(sizeof(*lgp), gfp_flags);
946 if (lgp == NULL)
947 return ERR_PTR(-ENOMEM);
948
949 i_size = i_size_read(ino);
950
951 lgp->args.minlength = PAGE_SIZE;
952 if (lgp->args.minlength > range->length)
953 lgp->args.minlength = range->length;
954 if (range->iomode == IOMODE_READ) {
955 if (range->offset >= i_size)
956 lgp->args.minlength = 0;
957 else if (i_size - range->offset < lgp->args.minlength)
958 lgp->args.minlength = i_size - range->offset;
d03ab29d 959 }
83026d80
JL		 960 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
961 pnfs_copy_range(&lgp->args.range, range);
962 lgp->args.type = server->pnfs_curr_ld->id;
963 lgp->args.inode = ino;
964 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 965 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80
JL		 966 lgp->gfp_flags = gfp_flags;
967 lgp->cred = lo->plh_lc_cred;
35124a09 968
183d9e7b 969 return nfs4_proc_layoutget(lgp, timeout, gfp_flags);
974cec8c
AA		 970}
971
24956804
TM		 972static void pnfs_clear_layoutcommit(struct inode *inode,
973 struct list_head *head)
974{
975 struct nfs_inode *nfsi = NFS_I(inode);
976 struct pnfs_layout_segment *lseg, *tmp;
977
978 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
979 return;
980 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
981 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
982 continue;
983 pnfs_lseg_dec_and_remove_zero(lseg, head);
984 }
985}
986
68f74479 987static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
d67ae825
TH		 988{
989 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
6604b203 990 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
d67ae825
TH		 991 smp_mb__after_atomic();
992 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
7f27392c 993 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
d67ae825
TH		 994}
995
68f74479 996void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 997 const nfs4_stateid *arg_stateid,
68f74479 998 const struct pnfs_layout_range *range,
68f74479
TM		 999 const nfs4_stateid *stateid)
1000{
1001 struct inode *inode = lo->plh_inode;
1002 LIST_HEAD(freeme);
1003
1004 spin_lock(&inode->i_lock);
2a974425
TM		 1005 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1006 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1007 goto out_unlock;
68f74479 1008 if (stateid) {
2a974425
TM		 1009 u32 seq = be32_to_cpu(arg_stateid->seqid);
1010
68f74479
TM		 1011 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1012 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1013 pnfs_set_layout_stateid(lo, stateid, true);
1014 } else
1015 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1016out_unlock:
68f74479
TM		 1017 pnfs_clear_layoutreturn_waitbit(lo);
1018 spin_unlock(&inode->i_lock);
1019 pnfs_free_lseg_list(&freeme);
1020
1021}
1022
13c13a6a 1023static bool
e5fd1904
TM		 1024pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1025 nfs4_stateid *stateid,
1026 enum pnfs_iomode *iomode)
13c13a6a 1027{
bf0291dd
TM		 1028 /* Serialise LAYOUTGET/LAYOUTRETURN */
1029 if (atomic_read(&lo->plh_outstanding) != 0)
1030 return false;
6604b203 1031 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1032 return false;
6604b203 1033 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1034 pnfs_get_layout_hdr(lo);
e5fd1904
TM		 1035 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1036 if (stateid != NULL) {
1037 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1038 if (lo->plh_return_seq != 0)
1039 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1040 }
1041 if (iomode != NULL)
1042 *iomode = lo->plh_return_iomode;
1043 pnfs_clear_layoutreturn_info(lo);
1044 return true;
1045 }
1046 if (stateid != NULL)
1047 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1048 if (iomode != NULL)
1049 *iomode = IOMODE_ANY;
13c13a6a
TM		 1050 return true;
1051}
1052
828ed9ec
TM		 1053static void
1054pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1055 struct pnfs_layout_hdr *lo,
1056 const nfs4_stateid *stateid,
1057 enum pnfs_iomode iomode)
1058{
1059 struct inode *inode = lo->plh_inode;
1060
1061 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1062 args->inode = inode;
1063 args->range.iomode = iomode;
1064 args->range.offset = 0;
1065 args->range.length = NFS4_MAX_UINT64;
1066 args->layout = lo;
1067 nfs4_stateid_copy(&args->stateid, stateid);
1068}
1069
f40eb5d0 1070static int
ed429d6b 1071pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
6c16605d 1072 enum pnfs_iomode iomode, bool sync)
f40eb5d0
PT		 1073{
1074 struct inode *ino = lo->plh_inode;
1075 struct nfs4_layoutreturn *lrp;
1076 int status = 0;
1077
e4af440a 1078 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1079 if (unlikely(lrp == NULL)) {
1080 status = -ENOMEM;
1081 spin_lock(&ino->i_lock);
d67ae825 1082 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0
PT		 1083 spin_unlock(&ino->i_lock);
1084 pnfs_put_layout_hdr(lo);
1085 goto out;
1086 }
1087
828ed9ec 1088 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
f40eb5d0
PT		 1089 lrp->clp = NFS_SERVER(ino)->nfs_client;
1090 lrp->cred = lo->plh_lc_cred;
1091
6c16605d 1092 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1093out:
1094 dprintk("<-- %s status: %d\n", __func__, status);
1095 return status;
1096}
1097
13c13a6a
TM		 1098/* Return true if layoutreturn is needed */
1099static bool
1100pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1101{
1102 struct pnfs_layout_segment *s;
1103
2370abda 1104 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1105 return false;
1106
1107 /* Defer layoutreturn until all lsegs are done */
1108 list_for_each_entry(s, &lo->plh_segs, pls_list) {
1109 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1110 return false;
1111 }
1112
1113 return true;
1114}
1115
1116static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1117{
1118 struct inode *inode= lo->plh_inode;
1119
2370abda 1120 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1121 return;
1122 spin_lock(&inode->i_lock);
1123 if (pnfs_layout_need_return(lo)) {
1124 nfs4_stateid stateid;
1125 enum pnfs_iomode iomode;
1126 bool send;
1127
e5fd1904 1128 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
13c13a6a
TM		 1129 spin_unlock(&inode->i_lock);
1130 if (send) {
1131 /* Send an async layoutreturn so we dont deadlock */
1132 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1133 }
1134 } else
1135 spin_unlock(&inode->i_lock);
1136}
1137
293b3b06
AA		 1138/*
1139 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1140 * when the layout segment list is empty.
1141 *
1142 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1143 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1144 * deviceid is marked invalid.
1145 */
cbe82603
BH		 1146int
1147_pnfs_return_layout(struct inode *ino)
1148{
1149 struct pnfs_layout_hdr *lo = NULL;
1150 struct nfs_inode *nfsi = NFS_I(ino);
1151 LIST_HEAD(tmp_list);
cbe82603 1152 nfs4_stateid stateid;
24408f52 1153 int status = 0;
7f27392c 1154 bool send;
cbe82603 1155
366d5052 1156 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1157
1158 spin_lock(&ino->i_lock);
1159 lo = nfsi->layout;
e5929f3c 1160 if (!lo) {
cbe82603 1161 spin_unlock(&ino->i_lock);
293b3b06
AA		 1162 dprintk("NFS: %s no layout to return\n", __func__);
1163 goto out;
cbe82603 1164 }
cbe82603 1165 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1166 pnfs_get_layout_hdr(lo);
24408f52
TM		 1167 /* Is there an outstanding layoutreturn ? */
1168 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1169 spin_unlock(&ino->i_lock);
1170 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1171 TASK_UNINTERRUPTIBLE))
1172 goto out_put_layout_hdr;
1173 spin_lock(&ino->i_lock);
1174 }
24956804 1175 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1176 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1177
1178 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1179 struct pnfs_layout_range range = {
1180 .iomode = IOMODE_ANY,
1181 .offset = 0,
1182 .length = NFS4_MAX_UINT64,
1183 };
1184 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1185 }
1186
293b3b06 1187 /* Don't send a LAYOUTRETURN if list was initially empty */
24408f52 1188 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
293b3b06 1189 spin_unlock(&ino->i_lock);
293b3b06 1190 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1191 goto out_put_layout_hdr;
293b3b06 1192 }
47abadef 1193
e5fd1904 1194 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603
BH		 1195 spin_unlock(&ino->i_lock);
1196 pnfs_free_lseg_list(&tmp_list);
7f27392c 1197 if (send)
ed429d6b 1198 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c
TM		 1199out_put_layout_hdr:
1200 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1201out:
1202 dprintk("<-- %s status: %d\n", __func__, status);
1203 return status;
1204}
0a57cdac 1205EXPORT_SYMBOL_GPL(_pnfs_return_layout);
cbe82603 1206
24028672
TM		 1207int
1208pnfs_commit_and_return_layout(struct inode *inode)
1209{
1210 struct pnfs_layout_hdr *lo;
1211 int ret;
1212
1213 spin_lock(&inode->i_lock);
1214 lo = NFS_I(inode)->layout;
1215 if (lo == NULL) {
1216 spin_unlock(&inode->i_lock);
1217 return 0;
1218 }
1219 pnfs_get_layout_hdr(lo);
1220 /* Block new layoutgets and read/write to ds */
1221 lo->plh_block_lgets++;
1222 spin_unlock(&inode->i_lock);
1223 filemap_fdatawait(inode->i_mapping);
1224 ret = pnfs_layoutcommit_inode(inode, true);
1225 if (ret == 0)
1226 ret = _pnfs_return_layout(inode);
1227 spin_lock(&inode->i_lock);
1228 lo->plh_block_lgets--;
1229 spin_unlock(&inode->i_lock);
1230 pnfs_put_layout_hdr(lo);
1231 return ret;
1232}
1233
1c5bd76d
TM		 1234bool pnfs_roc(struct inode *ino,
1235 struct nfs4_layoutreturn_args *args,
1236 struct nfs4_layoutreturn_res *res,
1237 const struct rpc_cred *cred)
f7e8917a 1238{
40dd4b7a
TM		 1239 struct nfs_inode *nfsi = NFS_I(ino);
1240 struct nfs_open_context *ctx;
1241 struct nfs4_state *state;
f7e8917a 1242 struct pnfs_layout_hdr *lo;
1c5bd76d 1243 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1244 nfs4_stateid stateid;
1c5bd76d
TM		 1245 enum pnfs_iomode iomode = 0;
1246 bool layoutreturn = false, roc = false;
f7e8917a 1247
1c5bd76d
TM		 1248 if (!nfs_have_layout(ino))
1249 return false;
29ade5db 1250retry:
f7e8917a 1251 spin_lock(&ino->i_lock);
40dd4b7a 1252 lo = nfsi->layout;
0cdc329e
TM		 1253 if (!lo || !pnfs_layout_is_valid(lo) ||
1254 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
40dd4b7a 1255 goto out_noroc;
29ade5db
TM		 1256 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1257 pnfs_get_layout_hdr(lo);
1258 spin_unlock(&ino->i_lock);
1259 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1260 TASK_UNINTERRUPTIBLE);
1261 pnfs_put_layout_hdr(lo);
1262 goto retry;
1263 }
40dd4b7a 1264
e755d638 1265 /* no roc if we hold a delegation */
40dd4b7a
TM		 1266 if (nfs4_check_delegation(ino, FMODE_READ))
1267 goto out_noroc;
1268
1269 list_for_each_entry(ctx, &nfsi->open_files, list) {
1270 state = ctx->state;
1271 /* Don't return layout if there is open file state */
1272 if (state != NULL && state->state != 0)
1273 goto out_noroc;
1274 }
1275
e755d638 1276
1c5bd76d 1277 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e755d638 1278 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1279 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1280 continue;
1281 /*
1282 * Note: mark lseg for return so pnfs_layout_remove_lseg
1283 * doesn't invalidate the layout for us.
1284 */
1285 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1286 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1287 continue;
1288 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1289 }
1290
1c5bd76d
TM		 1291 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1292 goto out_noroc;
69820d22 1293
500d701f 1294 /* ROC in two conditions:
e755d638
PT		 1295 * 1. there are ROC lsegs
1296 * 2. we don't send layoutreturn
e755d638 1297 */
1c5bd76d
TM		 1298 /* lo ref dropped in pnfs_roc_release() */
1299 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1300 /* If the creds don't match, we can't compound the layoutreturn */
1301 if (!layoutreturn || cred != lo->plh_lc_cred)
1302 goto out_noroc;
1303
1304 roc = layoutreturn;
1305 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1306 res->lrs_present = 0;
1307 layoutreturn = false;
f7e8917a 1308
40dd4b7a 1309out_noroc:
f7e8917a 1310 spin_unlock(&ino->i_lock);
e755d638
PT		 1311 pnfs_layoutcommit_inode(ino, true);
1312 if (layoutreturn)
1c5bd76d 1313 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
e755d638 1314 return roc;
f7e8917a
FI		 1315}
1316
1c5bd76d
TM		 1317void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1318 struct nfs4_layoutreturn_res *res,
1319 int ret)
f7e8917a 1320{
1c5bd76d
TM		 1321 struct pnfs_layout_hdr *lo = args->layout;
1322 const nfs4_stateid *arg_stateid = NULL;
1323 const nfs4_stateid *res_stateid = NULL;
f7e8917a 1324
1c5bd76d
TM		 1325 if (ret == 0) {
1326 arg_stateid = &args->stateid;
1327 if (res->lrs_present)
1328 res_stateid = &res->stateid;
1329 }
1330 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1331 res_stateid);
1332 pnfs_put_layout_hdr(lo);
1333 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1334}
1335
500d701f
PT		 1336bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1337{
1338 struct nfs_inode *nfsi = NFS_I(ino);
1339 struct pnfs_layout_hdr *lo;
1340 bool sleep = false;
1341
1342 /* we might not have grabbed lo reference. so need to check under
1343 * i_lock */
1344 spin_lock(&ino->i_lock);
1345 lo = nfsi->layout;
ee284e35
TM		 1346 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1347 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1348 sleep = true;
ee284e35 1349 }
500d701f 1350 spin_unlock(&ino->i_lock);
500d701f
PT		 1351 return sleep;
1352}
1353
b1f69b75
AA		 1354/*
1355 * Compare two layout segments for sorting into layout cache.
1356 * We want to preferentially return RW over RO layouts, so ensure those
1357 * are seen first.
1358 */
1359static s64
7dc0ac70 1360pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1361 const struct pnfs_layout_range *l2)
b1f69b75 1362{
fb3296eb
BH		 1363 s64 d;
1364
1365 /* high offset > low offset */
1366 d = l1->offset - l2->offset;
1367 if (d)
1368 return d;
1369
1370 /* short length > long length */
1371 d = l2->length - l1->length;
1372 if (d)
1373 return d;
1374
b1f69b75 1375 /* read > read/write */
fb3296eb 1376 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1377}
1378
03772d2f
TM		 1379static bool
1380pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1381 const struct pnfs_layout_range *l2)
1382{
1383 return pnfs_lseg_range_cmp(l1, l2) > 0;
1384}
1385
1386static bool
1387pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1388 struct pnfs_layout_segment *old)
1389{
1390 return false;
1391}
1392
1393void
1394pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1395 struct pnfs_layout_segment *lseg,
1396 bool (*is_after)(const struct pnfs_layout_range *,
1397 const struct pnfs_layout_range *),
1398 bool (*do_merge)(struct pnfs_layout_segment *,
1399 struct pnfs_layout_segment *),
1400 struct list_head *free_me)
974cec8c 1401{
03772d2f 1402 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1403
974cec8c
AA		 1404 dprintk("%s:Begin\n", __func__);
1405
03772d2f
TM		 1406 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1407 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1408 continue;
1409 if (do_merge(lseg, lp)) {
1410 mark_lseg_invalid(lp, free_me);
1411 continue;
1412 }
1413 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1414 continue;
566052c5 1415 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1416 dprintk("%s: inserted lseg %p "
1417 "iomode %d offset %llu length %llu before "
1418 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1419 __func__, lseg, lseg->pls_range.iomode,
1420 lseg->pls_range.offset, lseg->pls_range.length,
1421 lp, lp->pls_range.iomode, lp->pls_range.offset,
1422 lp->pls_range.length);
fb3296eb 1423 goto out;
974cec8c 1424 }
fb3296eb
BH		 1425 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1426 dprintk("%s: inserted lseg %p "
1427 "iomode %d offset %llu length %llu at tail\n",
1428 __func__, lseg, lseg->pls_range.iomode,
1429 lseg->pls_range.offset, lseg->pls_range.length);
1430out:
70c3bd2b 1431 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1432
1433 dprintk("%s:Return\n", __func__);
e5e94017 1434}
03772d2f
TM		 1435EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1436
1437static void
1438pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1439 struct pnfs_layout_segment *lseg,
1440 struct list_head *free_me)
1441{
1442 struct inode *inode = lo->plh_inode;
1443 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1444
1445 if (ld->add_lseg != NULL)
1446 ld->add_lseg(lo, lseg, free_me);
1447 else
1448 pnfs_generic_layout_insert_lseg(lo, lseg,
1449 pnfs_lseg_range_is_after,
1450 pnfs_lseg_no_merge,
1451 free_me);
1452}
e5e94017
BH		 1453
1454static struct pnfs_layout_hdr *
9fa40758
PT		 1455alloc_init_layout_hdr(struct inode *ino,
1456 struct nfs_open_context *ctx,
1457 gfp_t gfp_flags)
e5e94017
BH		 1458{
1459 struct pnfs_layout_hdr *lo;
1460
636fb9c8 1461 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1462 if (!lo)
1463 return NULL;
cc6e5340 1464 atomic_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1465 INIT_LIST_HEAD(&lo->plh_layouts);
1466 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1467 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1468 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1469 lo->plh_inode = ino;
5cc2216d 1470 lo->plh_lc_cred = get_rpccred(ctx->cred);
67a3b721 1471 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1472 return lo;
1473}
1474
1475static struct pnfs_layout_hdr *
9fa40758
PT		 1476pnfs_find_alloc_layout(struct inode *ino,
1477 struct nfs_open_context *ctx,
1478 gfp_t gfp_flags)
e5241e43
TM		 1479 __releases(&ino->i_lock)
1480 __acquires(&ino->i_lock)
e5e94017
BH		 1481{
1482 struct nfs_inode *nfsi = NFS_I(ino);
1483 struct pnfs_layout_hdr *new = NULL;
1484
1485 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1486
251ec410
TM		 1487 if (nfsi->layout != NULL)
1488 goto out_existing;
e5e94017 1489 spin_unlock(&ino->i_lock);
9fa40758 1490 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1491 spin_lock(&ino->i_lock);
1492
251ec410 1493 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1494 nfsi->layout = new;
251ec410 1495 return new;
7175fe90
YN		 1496 } else if (new != NULL)
1497 pnfs_free_layout_hdr(new);
251ec410
TM		 1498out_existing:
1499 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1500 return nfsi->layout;
1501}
1502
b1f69b75
AA		 1503/*
1504 * iomode matching rules:
c7d73af2
TH		 1505 * iomode lseg strict match
1506 * iomode
1507 * ----- ----- ------ -----
1508 * ANY READ N/A true
1509 * ANY RW N/A true
1510 * RW READ N/A false
1511 * RW RW N/A true
1512 * READ READ N/A true
1513 * READ RW true false
1514 * READ RW false true
b1f69b75 1515 */
3cb2df17 1516static bool
7dc0ac70 1517pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1518 const struct pnfs_layout_range *range,
1519 bool strict_iomode)
b1f69b75 1520{
fb3296eb
BH		 1521 struct pnfs_layout_range range1;
1522
1523 if ((range->iomode == IOMODE_RW &&
1524 ls_range->iomode != IOMODE_RW) ||
c7d73af2
TH		 1525 (range->iomode != ls_range->iomode &&
1526 strict_iomode == true) ||
7dc0ac70 1527 !pnfs_lseg_range_intersecting(ls_range, range))
fb3296eb
BH		 1528 return 0;
1529
1530 /* range1 covers only the first byte in the range */
1531 range1 = *range;
1532 range1.length = 1;
7dc0ac70 1533 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1534}
1535
1536/*
1537 * lookup range in layout
1538 */
e5e94017 1539static struct pnfs_layout_segment *
fb3296eb 1540pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1541 struct pnfs_layout_range *range,
1542 bool strict_iomode)
e5e94017 1543{
b1f69b75
AA		 1544 struct pnfs_layout_segment *lseg, *ret = NULL;
1545
1546 dprintk("%s:Begin\n", __func__);
1547
b7edfaa1 1548 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1549 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
ce6ab4f2 1550 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
c7d73af2
TH		 1551 pnfs_lseg_range_match(&lseg->pls_range, range,
1552 strict_iomode)) {
9369a431 1553 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1554 break;
1555 }
b1f69b75
AA		 1556 }
1557
1558 dprintk("%s:Return lseg %p ref %d\n",
4541d16c 1559 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
b1f69b75 1560 return ret;
e5e94017
BH		 1561}
1562
d23d61c8
AA		 1563/*
1564 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1565 * to the MDS or over pNFS
1566 *
1567 * The nfs_inode read_io and write_io fields are cumulative counters reset
1568 * when there are no layout segments. Note that in pnfs_update_layout iomode
1569 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1570 * WRITE request.
1571 *
1572 * A return of true means use MDS I/O.
1573 *
1574 * From rfc 5661:
1575 * If a file's size is smaller than the file size threshold, data accesses
1576 * SHOULD be sent to the metadata server. If an I/O request has a length that
1577 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1578 * server. If both file size and I/O size are provided, the client SHOULD
1579 * reach or exceed both thresholds before sending its read or write
1580 * requests to the data server.
1581 */
1582static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1583 struct inode *ino, int iomode)
1584{
1585 struct nfs4_threshold *t = ctx->mdsthreshold;
1586 struct nfs_inode *nfsi = NFS_I(ino);
1587 loff_t fsize = i_size_read(ino);
1588 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1589
1590 if (t == NULL)
1591 return ret;
1592
1593 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1594 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1595
1596 switch (iomode) {
1597 case IOMODE_READ:
1598 if (t->bm & THRESHOLD_RD) {
1599 dprintk("%s fsize %llu\n", __func__, fsize);
1600 size_set = true;
1601 if (fsize < t->rd_sz)
1602 size = true;
1603 }
1604 if (t->bm & THRESHOLD_RD_IO) {
1605 dprintk("%s nfsi->read_io %llu\n", __func__,
1606 nfsi->read_io);
1607 io_set = true;
1608 if (nfsi->read_io < t->rd_io_sz)
1609 io = true;
1610 }
1611 break;
1612 case IOMODE_RW:
1613 if (t->bm & THRESHOLD_WR) {
1614 dprintk("%s fsize %llu\n", __func__, fsize);
1615 size_set = true;
1616 if (fsize < t->wr_sz)
1617 size = true;
1618 }
1619 if (t->bm & THRESHOLD_WR_IO) {
1620 dprintk("%s nfsi->write_io %llu\n", __func__,
1621 nfsi->write_io);
1622 io_set = true;
1623 if (nfsi->write_io < t->wr_io_sz)
1624 io = true;
1625 }
1626 break;
1627 }
1628 if (size_set && io_set) {
1629 if (size && io)
1630 ret = true;
1631 } else if (size || io)
1632 ret = true;
1633
1634 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1635 return ret;
1636}
1637
aa8a45ee
PT		 1638static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1639{
1640 /*
1641 * send layoutcommit as it can hold up layoutreturn due to lseg
1642 * reference
1643 */
1644 pnfs_layoutcommit_inode(lo->plh_inode, false);
1645 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1646 nfs_wait_bit_killable,
aa8a45ee
PT		 1647 TASK_UNINTERRUPTIBLE);
1648}
1649
d67ae825
TH		 1650static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1651{
1652 unsigned long *bitlock = &lo->plh_flags;
1653
1654 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1655 smp_mb__after_atomic();
1656 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1657}
1658
e5e94017
BH		 1659/*
1660 * Layout segment is retreived from the server if not cached.
1661 * The appropriate layout segment is referenced and returned to the caller.
1662 */
7c24d948 1663struct pnfs_layout_segment *
e5e94017
BH		 1664pnfs_update_layout(struct inode *ino,
1665 struct nfs_open_context *ctx,
fb3296eb
BH		 1666 loff_t pos,
1667 u64 count,
a75b9df9 1668 enum pnfs_iomode iomode,
c7d73af2 1669 bool strict_iomode,
a75b9df9 1670 gfp_t gfp_flags)
e5e94017 1671{
fb3296eb
BH		 1672 struct pnfs_layout_range arg = {
1673 .iomode = iomode,
1674 .offset = pos,
1675 .length = count,
1676 };
183d9e7b 1677 unsigned pg_offset, seq;
6382a441
WAA		 1678 struct nfs_server *server = NFS_SERVER(ino);
1679 struct nfs_client *clp = server->nfs_client;
183d9e7b 1680 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1681 struct pnfs_layout_segment *lseg = NULL;
183d9e7b
JL		 1682 nfs4_stateid stateid;
1683 long timeout = 0;
66b53f32 1684 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1685 bool first;
e5e94017 1686
9a4bf31d 1687 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1688 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1689 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1690 goto out;
9a4bf31d 1691 }
d23d61c8 1692
9a4bf31d 1693 if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
183d9e7b 1694 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1695 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
4ae93560 1696 goto out;
9a4bf31d 1697 }
4ae93560 1698
9a4bf31d 1699 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1700 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1701 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1702 goto out;
9a4bf31d 1703 }
d23d61c8 1704
9bf87482 1705lookup_again:
b88fa69e 1706 nfs4_client_recover_expired_lease(clp);
9bf87482 1707 first = false;
e5e94017 1708 spin_lock(&ino->i_lock);
9fa40758 1709 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1710 if (lo == NULL) {
1711 spin_unlock(&ino->i_lock);
183d9e7b 1712 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1713 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1714 goto out;
1715 }
e5e94017 1716
43f1b3da 1717 /* Do we even need to bother with this? */
a59c30ac 1718 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1719 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1720 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1721 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1722 goto out_unlock;
1723 }
1724
1725 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1726 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1727 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1728 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1729 goto out_unlock;
9a4bf31d 1730 }
e5e94017 1731
c7d73af2 1732 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1733 if (lseg) {
1734 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1735 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1736 goto out_unlock;
1737 }
1738
1739 if (!nfs4_valid_open_stateid(ctx->state)) {
1740 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1741 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1742 goto out_unlock;
1743 }
1744
1745 /*
1746 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1747 * stateid, or it has been invalidated, then we must use the open
1748 * stateid.
1749 */
67a3b721 1750 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
183d9e7b
JL		 1751
1752 /*
1753 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1754 * RFC 5661 Errata 3208.
1755 */
1756 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1757 &lo->plh_flags)) {
1758 spin_unlock(&ino->i_lock);
1759 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1760 TASK_UNINTERRUPTIBLE);
1761 pnfs_put_layout_hdr(lo);
183d9e7b 1762 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1763 goto lookup_again;
1764 }
183d9e7b
JL		 1765
1766 first = true;
1767 do {
1768 seq = read_seqbegin(&ctx->state->seqlock);
1769 nfs4_stateid_copy(&stateid, &ctx->state->stateid);
1770 } while (read_seqretry(&ctx->state->seqlock, seq));
9bf87482 1771 } else {
183d9e7b 1772 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 1773 }
568e8c49 1774
aa8a45ee
PT		 1775 /*
1776 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1777 * for LAYOUTRETURN even if first is true.
1778 */
8f70f53a 1779 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 1780 spin_unlock(&ino->i_lock);
1781 dprintk("%s wait for layoutreturn\n", __func__);
1782 if (pnfs_prepare_to_retry_layoutget(lo)) {
d67ae825
TH		 1783 if (first)
1784 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 1785 pnfs_put_layout_hdr(lo);
1786 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 1787 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1788 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 1789 goto lookup_again;
1790 }
183d9e7b 1791 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1792 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 1793 goto out_put_layout_hdr;
1794 }
1795
9a4bf31d 1796 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 1797 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1798 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 1799 goto out_unlock;
9a4bf31d 1800 }
cf7d63f1 1801 atomic_inc(&lo->plh_outstanding);
f49f9baa 1802 spin_unlock(&ino->i_lock);
30005121 1803
abb9a007 1804 if (list_empty(&lo->plh_layouts)) {
2130ff66
FI		 1805 /* The lo must be on the clp list if there is any
1806 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1807 */
1808 spin_lock(&clp->cl_lock);
abb9a007
PT		 1809 if (list_empty(&lo->plh_layouts))
1810 list_add_tail(&lo->plh_layouts, &server->layouts);
2130ff66
FI		 1811 spin_unlock(&clp->cl_lock);
1812 }
e5e94017 1813
09cbfeaf 1814 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 1815 if (pg_offset) {
1816 arg.offset -= pg_offset;
1817 arg.length += pg_offset;
1818 }
7c24d948 1819 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 1820 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 1821
183d9e7b
JL		 1822 lseg = send_layoutget(lo, ctx, &stateid, &arg, &timeout, gfp_flags);
1823 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1824 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
56b38a1f 1825 atomic_dec(&lo->plh_outstanding);
83026d80 1826 if (IS_ERR(lseg)) {
183d9e7b 1827 switch(PTR_ERR(lseg)) {
e85d7ee4 1828 case -EBUSY:
183d9e7b
JL		 1829 if (time_after(jiffies, giveup))
1830 lseg = NULL;
66b53f32
TM		 1831 break;
1832 case -ERECALLCONFLICT:
1833 /* Huh? We hold no layouts, how is there a recall? */
1834 if (first) {
1835 lseg = NULL;
1836 break;
1837 }
1838 /* Destroy the existing layout and start over */
1839 if (time_after(jiffies, giveup))
1840 pnfs_destroy_layout(NFS_I(ino));
183d9e7b
JL		 1841 /* Fallthrough */
1842 case -EAGAIN:
56b38a1f 1843 break;
183d9e7b
JL		 1844 default:
1845 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
1846 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1847 lseg = NULL;
1848 }
56b38a1f
TM		 1849 goto out_put_layout_hdr;
1850 }
1851 if (lseg) {
1852 if (first)
1853 pnfs_clear_first_layoutget(lo);
1854 trace_pnfs_update_layout(ino, pos, count,
1855 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
1856 pnfs_put_layout_hdr(lo);
1857 goto lookup_again;
83026d80
JL		 1858 }
1859 } else {
1860 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1861 }
1862
830ffb56 1863out_put_layout_hdr:
d67ae825
TH		 1864 if (first)
1865 pnfs_clear_first_layoutget(lo);
70c3bd2b 1866 pnfs_put_layout_hdr(lo);
e5e94017 1867out:
f86bbcf8
TM		 1868 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1869 "(%s, offset: %llu, length: %llu)\n",
1870 __func__, ino->i_sb->s_id,
1871 (unsigned long long)NFS_FILEID(ino),
d600ad1f 1872 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 1873 iomode==IOMODE_RW ? "read/write" : "read-only",
1874 (unsigned long long)pos,
1875 (unsigned long long)count);
e5e94017
BH		 1876 return lseg;
1877out_unlock:
1878 spin_unlock(&ino->i_lock);
830ffb56 1879 goto out_put_layout_hdr;
e5e94017 1880}
7c24d948 1881EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 1882
540d9864
TM		 1883static bool
1884pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
1885{
1886 switch (range->iomode) {
1887 case IOMODE_READ:
1888 case IOMODE_RW:
1889 break;
1890 default:
1891 return false;
1892 }
1893 if (range->offset == NFS4_MAX_UINT64)
1894 return false;
1895 if (range->length == 0)
1896 return false;
1897 if (range->length != NFS4_MAX_UINT64 &&
1898 range->length > NFS4_MAX_UINT64 - range->offset)
1899 return false;
1900 return true;
1901}
1902
a0b0a6e3 1903struct pnfs_layout_segment *
b1f69b75
AA		 1904pnfs_layout_process(struct nfs4_layoutget *lgp)
1905{
1906 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1907 struct nfs4_layoutget_res *res = &lgp->res;
1908 struct pnfs_layout_segment *lseg;
b7edfaa1 1909 struct inode *ino = lo->plh_inode;
78096cca 1910 LIST_HEAD(free_me);
540d9864
TM		 1911
1912 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 1913 return ERR_PTR(-EINVAL);
b1f69b75
AA		 1914
1915 /* Inject layout blob into I/O device driver */
a75b9df9 1916 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 1917 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 1918 if (!lseg)
1b3c6d07
JL		 1919 lseg = ERR_PTR(-ENOMEM);
1920
1921 dprintk("%s: Could not allocate layout: error %ld\n",
1922 __func__, PTR_ERR(lseg));
1923 return lseg;
b1f69b75
AA		 1924 }
1925
119cef97 1926 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 1927
b1f69b75 1928 spin_lock(&ino->i_lock);
e1c06f80 1929 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 1930 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 1931 goto out_forget;
43f1b3da 1932 }
038d6493 1933
9888d837
TM		 1934 if (!pnfs_layout_is_valid(lo)) {
1935 /* We have a completely new layout */
1936 pnfs_set_layout_stateid(lo, &res->stateid, true);
1937 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 1938 /* existing state ID, make sure the sequence number matches. */
1939 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
1940 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 1941 goto out_forget;
362f7474
CH		 1942 }
1943 pnfs_set_layout_stateid(lo, &res->stateid, false);
1944 } else {
1945 /*
1946 * We got an entirely new state ID. Mark all segments for the
9888d837 1947 * inode invalid, and retry the layoutget
362f7474 1948 */
d9b61708 1949 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 1950 goto out_forget;
362f7474 1951 }
038d6493 1952
9369a431 1953 pnfs_get_lseg(lseg);
03772d2f 1954 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 1955
b1f69b75 1956
3976143b 1957 if (res->return_on_close)
f7e8917a 1958 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 1959
b1f69b75 1960 spin_unlock(&ino->i_lock);
78096cca 1961 pnfs_free_lseg_list(&free_me);
a0b0a6e3 1962 return lseg;
43f1b3da 1963
1b3c6d07 1964out_forget:
43f1b3da
FI		 1965 spin_unlock(&ino->i_lock);
1966 lseg->pls_layout = lo;
1967 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 1968 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 1969}
1970
2f215968
TM		 1971/**
1972 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
1973 * @lo: pointer to layout header
1974 * @tmp_list: list header to be used with pnfs_free_lseg_list()
1975 * @return_range: describe layout segment ranges to be returned
1976 *
1977 * This function is mainly intended for use by layoutrecall. It attempts
1978 * to free the layout segment immediately, or else to mark it for return
1979 * as soon as its reference count drops to zero.
1980 */
10335556 1981int
016256df
PT		 1982pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
1983 struct list_head *tmp_list,
6d597e17
JL		 1984 const struct pnfs_layout_range *return_range,
1985 u32 seq)
016256df
PT		 1986{
1987 struct pnfs_layout_segment *lseg, *next;
10335556 1988 int remaining = 0;
016256df
PT		 1989
1990 dprintk("%s:Begin lo %p\n", __func__, lo);
1991
1992 if (list_empty(&lo->plh_segs))
10335556 1993 return 0;
016256df 1994
fc7ff367 1995 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 1996
1997 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 1998 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 1999 dprintk("%s: marking lseg %p iomode %d "
2000 "offset %llu length %llu\n", __func__,
2001 lseg, lseg->pls_range.iomode,
2002 lseg->pls_range.offset,
2003 lseg->pls_range.length);
2f215968
TM		 2004 if (mark_lseg_invalid(lseg, tmp_list))
2005 continue;
2006 remaining++;
016256df 2007 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2008 }
6d597e17
JL		 2009
2010 if (remaining)
2011 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2012
10335556 2013 return remaining;
016256df
PT		 2014}
2015
2016void pnfs_error_mark_layout_for_return(struct inode *inode,
2017 struct pnfs_layout_segment *lseg)
2018{
2019 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2020 struct pnfs_layout_range range = {
2021 .iomode = lseg->pls_range.iomode,
2022 .offset = 0,
2023 .length = NFS4_MAX_UINT64,
2024 };
10335556 2025 bool return_now = false;
016256df
PT		 2026
2027 spin_lock(&inode->i_lock);
2d6cf5ab 2028 pnfs_set_plh_return_info(lo, range.iomode, 0);
6604b203
TM		 2029 /* Block LAYOUTGET */
2030 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
016256df
PT		 2031 /*
2032 * mark all matching lsegs so that we are sure to have no live
2033 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2034 * for how it works.
2035 */
68f74479 2036 if (!pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0)) {
10335556 2037 nfs4_stateid stateid;
e5fd1904 2038 enum pnfs_iomode iomode;
10335556 2039
e5fd1904 2040 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2041 spin_unlock(&inode->i_lock);
2042 if (return_now)
2043 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2044 } else {
2045 spin_unlock(&inode->i_lock);
2046 nfs_commit_inode(inode, 0);
2047 }
016256df
PT		 2048}
2049EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2050
d8007d4d
TM		 2051void
2052pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2053{
1fd937bd
PT		 2054 u64 rd_size = req->wb_bytes;
2055
cb5d04bc
PT		 2056 if (pgio->pg_lseg == NULL) {
2057 if (pgio->pg_dreq == NULL)
2058 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2059 else
2060 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2061
2062 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2063 req->wb_context,
2064 req_offset(req),
2065 rd_size,
2066 IOMODE_READ,
c7d73af2 2067 false,
cb5d04bc 2068 GFP_KERNEL);
d600ad1f
PT		 2069 if (IS_ERR(pgio->pg_lseg)) {
2070 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2071 pgio->pg_lseg = NULL;
2072 return;
2073 }
cb5d04bc 2074 }
e885de1a
TM		 2075 /* If no lseg, fall back to read through mds */
2076 if (pgio->pg_lseg == NULL)
1f945357 2077 nfs_pageio_reset_read_mds(pgio);
e885de1a 2078
d8007d4d
TM		 2079}
2080EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2081
2082void
6296556f
PT		 2083pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2084 struct nfs_page *req, u64 wb_size)
d8007d4d 2085{
d600ad1f 2086 if (pgio->pg_lseg == NULL) {
cb5d04bc
PT		 2087 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2088 req->wb_context,
2089 req_offset(req),
2090 wb_size,
2091 IOMODE_RW,
c7d73af2 2092 false,
cb5d04bc 2093 GFP_NOFS);
d600ad1f
PT		 2094 if (IS_ERR(pgio->pg_lseg)) {
2095 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2096 pgio->pg_lseg = NULL;
2097 return;
2098 }
2099 }
e885de1a
TM		 2100 /* If no lseg, fall back to write through mds */
2101 if (pgio->pg_lseg == NULL)
1f945357 2102 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2103}
2104EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2105
180bb5ec
WAA		 2106void
2107pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2108{
2109 if (desc->pg_lseg) {
2110 pnfs_put_lseg(desc->pg_lseg);
2111 desc->pg_lseg = NULL;
2112 }
2113}
2114EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2115
b4fdac1a
WAA		 2116/*
2117 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2118 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2119 */
2120size_t
a7d42ddb
WAA		 2121pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2122 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2123{
0f9c429e 2124 unsigned int size;
c5e20cb7 2125 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2126
2127 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2128 if (!size)
2129 return 0;
94ad1c80 2130
19982ba8 2131 /*
c5e20cb7
WAA		 2132 * 'size' contains the number of bytes left in the current page (up
2133 * to the original size asked for in @req->wb_bytes).
2134 *
2135 * Calculate how many bytes are left in the layout segment
2136 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2137 *
2138 * Please also note that 'end_offset' is actually the offset of the
2139 * first byte that lies outside the pnfs_layout_range. FIXME?
2140 *
2141 */
19b54848 2142 if (pgio->pg_lseg) {
17822b20 2143 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2144 pgio->pg_lseg->pls_range.length);
2145 req_start = req_offset(req);
7c13789e 2146 WARN_ON_ONCE(req_start >= seg_end);
c5e20cb7 2147 /* start of request is past the last byte of this segment */
7c13789e
WAA		 2148 if (req_start >= seg_end) {
2149 /* reference the new lseg */
2150 if (pgio->pg_ops->pg_cleanup)
2151 pgio->pg_ops->pg_cleanup(pgio);
2152 if (pgio->pg_ops->pg_init)
2153 pgio->pg_ops->pg_init(pgio, req);
19b54848 2154 return 0;
7c13789e 2155 }
c5e20cb7
WAA		 2156
2157 /* adjust 'size' iff there are fewer bytes left in the
2158 * segment than what nfs_generic_pg_test returned */
2159 seg_left = seg_end - req_start;
2160 if (seg_left < size)
2161 size = (unsigned int)seg_left;
19b54848 2162 }
0f9c429e 2163
19b54848 2164 return size;
94ad1c80 2165}
89a58e32 2166EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2167
53113ad3 2168int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2169{
2170 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2171
2172 /* Resend all requests through the MDS */
53113ad3
WAA		 2173 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2174 hdr->completion_ops);
c7070113 2175 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2176 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2177}
e7dd79af 2178EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2179
d45f60c6 2180static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2181{
cd841605
FI		 2182
2183 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2184 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2185 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2186 pnfs_return_layout(hdr->inode);
1acbbb4e 2187 }
6c75dc0d 2188 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2189 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2190}
2191
d20581aa
BH		 2192/*
2193 * Called by non rpc-based layout drivers
2194 */
d45f60c6 2195void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2196{
f8417b48 2197 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2198 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2199 hdr->mds_offset + hdr->res.count);
d45f60c6 2200 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2201 }
2202 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2203 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2204 pnfs_ld_handle_write_error(hdr);
2205 hdr->mds_ops->rpc_release(hdr);
44b83799 2206}
d20581aa 2207EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2208
dce81290
TM		 2209static void
2210pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2211 struct nfs_pgio_header *hdr)
dce81290 2212{
48d635f1 2213 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2214
6c75dc0d 2215 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2216 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2217 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2218 mirror->pg_recoalesce = 1;
6c75dc0d 2219 }
d45f60c6 2220 nfs_pgio_data_destroy(hdr);
1ca018d2 2221 hdr->release(hdr);
dce81290
TM		 2222}
2223
2224static enum pnfs_try_status
d45f60c6 2225pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2226 const struct rpc_call_ops *call_ops,
2227 struct pnfs_layout_segment *lseg,
2228 int how)
0382b744 2229{
cd841605 2230 struct inode *inode = hdr->inode;
0382b744
AA		 2231 enum pnfs_try_status trypnfs;
2232 struct nfs_server *nfss = NFS_SERVER(inode);
2233
cd841605 2234 hdr->mds_ops = call_ops;
0382b744
AA		 2235
2236 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2237 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2238 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2239 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2240 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2241 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2242 return trypnfs;
2243}
2244
dce81290 2245static void
7f714720
WAA		 2246pnfs_do_write(struct nfs_pageio_descriptor *desc,
2247 struct nfs_pgio_header *hdr, int how)
dce81290 2248{
dce81290
TM		 2249 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2250 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2251 enum pnfs_try_status trypnfs;
dce81290 2252
d45f60c6 2253 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
7f714720 2254 if (trypnfs == PNFS_NOT_ATTEMPTED)
d45f60c6 2255 pnfs_write_through_mds(desc, hdr);
dce81290
TM		 2256}
2257
6c75dc0d
FI		 2258static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2259{
9369a431 2260 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2261 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2262}
2263
dce81290
TM		 2264int
2265pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2266{
6c75dc0d 2267 struct nfs_pgio_header *hdr;
dce81290
TM		 2268 int ret;
2269
1e7f3a48
WAA		 2270 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2271 if (!hdr) {
2bff2288
PT		 2272 desc->pg_error = -ENOMEM;
2273 return desc->pg_error;
dce81290 2274 }
6c75dc0d 2275 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2276
9369a431 2277 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2278 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2279 if (!ret)
7f714720 2280 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2281
6c75dc0d 2282 return ret;
dce81290
TM		 2283}
2284EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2285
53113ad3 2286int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2287{
2288 struct nfs_pageio_descriptor pgio;
2289
1acbbb4e 2290 /* Resend all requests through the MDS */
53113ad3
WAA		 2291 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2292 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2293}
e7dd79af 2294EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2295
d45f60c6 2296static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2297{
cd841605
FI		 2298 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2299 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2300 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2301 pnfs_return_layout(hdr->inode);
1acbbb4e 2302 }
4db6e0b7 2303 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2304 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2305}
2306
d20581aa
BH		 2307/*
2308 * Called by non rpc-based layout drivers
2309 */
d45f60c6 2310void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2311{
bfc505de 2312 if (likely(!hdr->pnfs_error))
d45f60c6 2313 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2314 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2315 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2316 pnfs_ld_handle_read_error(hdr);
2317 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2318}
2319EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2320
493292dd
TM		 2321static void
2322pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2323 struct nfs_pgio_header *hdr)
493292dd 2324{
48d635f1 2325 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2326
4db6e0b7 2327 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2328 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2329 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2330 mirror->pg_recoalesce = 1;
4db6e0b7 2331 }
d45f60c6 2332 nfs_pgio_data_destroy(hdr);
1ca018d2 2333 hdr->release(hdr);
493292dd
TM		 2334}
2335
64419a9b
AA		 2336/*
2337 * Call the appropriate parallel I/O subsystem read function.
2338 */
493292dd 2339static enum pnfs_try_status
d45f60c6 2340pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2341 const struct rpc_call_ops *call_ops,
2342 struct pnfs_layout_segment *lseg)
64419a9b 2343{
cd841605 2344 struct inode *inode = hdr->inode;
64419a9b
AA		 2345 struct nfs_server *nfss = NFS_SERVER(inode);
2346 enum pnfs_try_status trypnfs;
2347
cd841605 2348 hdr->mds_ops = call_ops;
64419a9b
AA		 2349
2350 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2351 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2352
d45f60c6 2353 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2354 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2355 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2356 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2357 return trypnfs;
2358}
863a3c6c 2359
ceb11e13 2360/* Resend all requests through pnfs. */
1b1bc66b 2361void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2362{
2363 struct nfs_pageio_descriptor pgio;
2364
1b1bc66b 2365 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2366 /* Prevent deadlocks with layoutreturn! */
2367 pnfs_put_lseg(hdr->lseg);
2368 hdr->lseg = NULL;
2369
1b1bc66b
WAA		 2370 nfs_pageio_init_read(&pgio, hdr->inode, false,
2371 hdr->completion_ops);
2372 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2373 }
ceb11e13
PT		 2374}
2375EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2376
493292dd 2377static void
7f714720 2378pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2379{
493292dd
TM		 2380 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2381 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2382 enum pnfs_try_status trypnfs;
493292dd 2383
d45f60c6 2384 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
ceb11e13 2385 if (trypnfs == PNFS_TRY_AGAIN)
1b1bc66b
WAA		 2386 pnfs_read_resend_pnfs(hdr);
2387 if (trypnfs == PNFS_NOT_ATTEMPTED || hdr->task.tk_status)
d45f60c6 2388 pnfs_read_through_mds(desc, hdr);
493292dd
TM		 2389}
2390
4db6e0b7
FI		 2391static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2392{
9369a431 2393 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2394 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2395}
2396
493292dd
TM		 2397int
2398pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2399{
4db6e0b7 2400 struct nfs_pgio_header *hdr;
493292dd
TM		 2401 int ret;
2402
1e7f3a48
WAA		 2403 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2404 if (!hdr) {
2bff2288
PT		 2405 desc->pg_error = -ENOMEM;
2406 return desc->pg_error;
493292dd 2407 }
4db6e0b7 2408 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2409 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2410 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2411 if (!ret)
7f714720 2412 pnfs_do_read(desc, hdr);
4db6e0b7 2413 return ret;
493292dd
TM		 2414}
2415EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2416
71244d9b
TM		 2417static void pnfs_clear_layoutcommitting(struct inode *inode)
2418{
2419 unsigned long *bitlock = &NFS_I(inode)->flags;
2420
2421 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2422 smp_mb__after_atomic();
71244d9b
TM		 2423 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2424}
2425
863a3c6c 2426/*
a9bae566 2427 * There can be multiple RW segments.
863a3c6c 2428 */
a9bae566 2429static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2430{
a9bae566 2431 struct pnfs_layout_segment *lseg;
863a3c6c 2432
a9bae566
PT		 2433 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2434 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2435 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2436 list_add(&lseg->pls_lc_list, listp);
2437 }
863a3c6c
AA		 2438}
2439
a073dbff
TM		 2440static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2441{
2442 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2443
2444 /* Matched by references in pnfs_set_layoutcommit */
2445 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2446 list_del_init(&lseg->pls_lc_list);
2447 pnfs_put_lseg(lseg);
2448 }
2449
71244d9b 2450 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2451}
2452
1b0ae068
PT		 2453void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2454{
b9e028fd 2455 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2456}
2457EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2458
863a3c6c 2459void
67af7611
TM		 2460pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2461 loff_t end_pos)
863a3c6c 2462{
cd841605 2463 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2464 bool mark_as_dirty = false;
863a3c6c 2465
cd841605 2466 spin_lock(&inode->i_lock);
863a3c6c 2467 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2468 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2469 mark_as_dirty = true;
863a3c6c 2470 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2471 __func__, inode->i_ino);
29559b11
TM		 2472 } else if (end_pos > nfsi->layout->plh_lwb)
2473 nfsi->layout->plh_lwb = end_pos;
67af7611 2474 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2475 /* references matched in nfs4_layoutcommit_release */
67af7611 2476 pnfs_get_lseg(lseg);
a9bae566 2477 }
cd841605 2478 spin_unlock(&inode->i_lock);
acff5880 2479 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2480 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2481
2482 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2483 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2484 if (mark_as_dirty)
cd841605 2485 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2486}
2487EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2488
db29c089
AA		 2489void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2490{
2491 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2492
2493 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2494 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2495 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2496}
2497
de4b15c7
AA		 2498/*
2499 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2500 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2501 * data to disk to allow the server to recover the data if it crashes.
2502 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2503 * is off, and a COMMIT is sent to a data server, or
2504 * if WRITEs to a data server return NFS_DATA_SYNC.
2505 */
863a3c6c 2506int
ef311537 2507pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2508{
5f919c9f 2509 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2510 struct nfs4_layoutcommit_data *data;
2511 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2512 loff_t end_pos;
71244d9b 2513 int status;
863a3c6c 2514
71244d9b 2515 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2516 return 0;
2517
71244d9b 2518 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2519
71244d9b 2520 status = -EAGAIN;
92407e75 2521 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2522 if (!sync)
2523 goto out;
74316201 2524 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2525 NFS_INO_LAYOUTCOMMITTING,
2526 nfs_wait_bit_killable,
2527 TASK_KILLABLE);
92407e75 2528 if (status)
71244d9b 2529 goto out;
92407e75
PT		 2530 }
2531
71244d9b
TM		 2532 status = -ENOMEM;
2533 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2534 data = kzalloc(sizeof(*data), GFP_NOFS);
2535 if (!data)
2536 goto clear_layoutcommitting;
2537
2538 status = 0;
de4b15c7 2539 spin_lock(&inode->i_lock);
71244d9b
TM		 2540 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2541 goto out_unlock;
a9bae566 2542
71244d9b 2543 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 2544 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 2545
acff5880 2546 end_pos = nfsi->layout->plh_lwb;
863a3c6c 2547
f597c537 2548 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 2549 spin_unlock(&inode->i_lock);
2550
2551 data->args.inode = inode;
9fa40758 2552 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 2553 nfs_fattr_init(&data->fattr);
2554 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2555 data->res.fattr = &data->fattr;
2e18d4d8
TM		 2556 if (end_pos != 0)
2557 data->args.lastbytewritten = end_pos - 1;
2558 else
2559 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 2560 data->res.server = NFS_SERVER(inode);
2561
5f919c9f
CH		 2562 if (ld->prepare_layoutcommit) {
2563 status = ld->prepare_layoutcommit(&data->args);
2564 if (status) {
3471648a 2565 put_rpccred(data->cred);
5f919c9f 2566 spin_lock(&inode->i_lock);
29559b11
TM		 2567 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2568 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 2569 nfsi->layout->plh_lwb = end_pos;
3471648a 2570 goto out_unlock;
5f919c9f
CH		 2571 }
2572 }
2573
2574
863a3c6c
AA		 2575 status = nfs4_proc_layoutcommit(data, sync);
2576out:
92407e75
PT		 2577 if (status)
2578 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2579 dprintk("<-- %s status %d\n", __func__, status);
2580 return status;
71244d9b
TM		 2581out_unlock:
2582 spin_unlock(&inode->i_lock);
92407e75 2583 kfree(data);
71244d9b
TM		 2584clear_layoutcommitting:
2585 pnfs_clear_layoutcommitting(inode);
92407e75 2586 goto out;
863a3c6c 2587}
72cff449 2588EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 2589
5bb89b47
TM		 2590int
2591pnfs_generic_sync(struct inode *inode, bool datasync)
2592{
2593 return pnfs_layoutcommit_inode(inode, true);
2594}
2595EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2596
82be417a
AA		 2597struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2598{
2599 struct nfs4_threshold *thp;
2600
2601 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2602 if (!thp) {
2603 dprintk("%s mdsthreshold allocation failed\n", __func__);
2604 return NULL;
2605 }
2606 return thp;
2607}
8733408d 2608
865a7ecb 2609#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 2610int
c8ad8894 2611pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 2612{
2613 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2614 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 2615 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 2616 struct nfs42_layoutstat_data *data;
2617 struct pnfs_layout_hdr *hdr;
2618 int status = 0;
2619
2620 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2621 goto out;
2622
6c5a0d89
TM		 2623 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2624 goto out;
2625
1bfe3b25
PT		 2626 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2627 goto out;
2628
8733408d
PT		 2629 spin_lock(&inode->i_lock);
2630 if (!NFS_I(inode)->layout) {
2631 spin_unlock(&inode->i_lock);
f538d0ba 2632 goto out_clear_layoutstats;
8733408d
PT		 2633 }
2634 hdr = NFS_I(inode)->layout;
2635 pnfs_get_layout_hdr(hdr);
2636 spin_unlock(&inode->i_lock);
2637
c8ad8894 2638 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 2639 if (!data) {
2640 status = -ENOMEM;
2641 goto out_put;
2642 }
2643
2644 data->args.fh = NFS_FH(inode);
2645 data->args.inode = inode;
8733408d
PT		 2646 status = ld->prepare_layoutstats(&data->args);
2647 if (status)
2648 goto out_free;
2649
2650 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2651
2652out:
2653 dprintk("%s returns %d\n", __func__, status);
2654 return status;
2655
2656out_free:
2657 kfree(data);
2658out_put:
2659 pnfs_put_layout_hdr(hdr);
f538d0ba 2660out_clear_layoutstats:
1bfe3b25
PT		 2661 smp_mb__before_atomic();
2662 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2663 smp_mb__after_atomic();
8733408d
PT		 2664 goto out;
2665}
2666EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 2667#endif
bbf58bf3
TM		 2668
2669unsigned int layoutstats_timer;
2670module_param(layoutstats_timer, uint, 0644);
2671EXPORT_SYMBOL_GPL(layoutstats_timer);
ubuntu-bionic-kernel mirror