]> Git Repo - linux.git/blame - fs/nfs/pnfs.c
projects / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
pNFS: Ensure we do clear the return-on-close layout stateid on fatal errors
[linux.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 <[email protected]>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
95void
96unset_pnfs_layoutdriver(struct nfs_server *nfss)
97{
738fd0f3
BH		 98 if (nfss->pnfs_curr_ld) {
99 if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 104 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 105 }
85e174ba
RL		 106 nfss->pnfs_curr_ld = NULL;
107}
108
ca440c38
JL		 109/*
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
112 *
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
115 */
116static const u32 ld_prefs[] = {
117 LAYOUT_SCSI,
118 LAYOUT_BLOCK_VOLUME,
119 LAYOUT_OSD2_OBJECTS,
120 LAYOUT_FLEX_FILES,
121 LAYOUT_NFSV4_1_FILES,
122 0
123};
124
125static int
126ld_cmp(const void *e1, const void *e2)
127{
128 u32 ld1 = *((u32 *)e1);
129 u32 ld2 = *((u32 *)e2);
130 int i;
131
132 for (i = 0; ld_prefs[i] != 0; i++) {
133 if (ld1 == ld_prefs[i])
134 return -1;
135
136 if (ld2 == ld_prefs[i])
137 return 1;
138 }
139 return 0;
140}
141
85e174ba
RL		 142/*
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
145 *
3132e49e 146 * @ids array of layout types supported by MDS.
85e174ba
RL		 147 */
148void
738fd0f3 149set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 150 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 151{
152 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 153 u32 id;
ca440c38 154 int i;
85e174ba 155
19274716
AS		 156 if (fsinfo->nlayouttypes == 0)
157 goto out_no_driver;
85e174ba
RL		 158 if (!(server->nfs_client->cl_exchange_flags &
159 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 160 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 162 goto out_no_driver;
163 }
3132e49e 164
ca440c38
JL		 165 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 167
ca440c38
JL		 168 for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 id = fsinfo->layouttype[i];
85e174ba 170 ld_type = find_pnfs_driver(id);
ca440c38
JL		 171 if (!ld_type) {
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 id);
174 ld_type = find_pnfs_driver(id);
175 }
176 if (ld_type)
177 break;
85e174ba 178 }
3132e49e
JL		 179
180 if (!ld_type) {
ca440c38 181 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 182 goto out_no_driver;
183 }
184
85e174ba 185 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 186 if (ld_type->set_layoutdriver
187 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 188 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__, id);
738fd0f3
BH		 190 module_put(ld_type->owner);
191 goto out_no_driver;
192 }
2a4c8994
TM		 193 /* Bump the MDS count */
194 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 195
85e174ba
RL		 196 dprintk("%s: pNFS module for %u set\n", __func__, id);
197 return;
198
199out_no_driver:
200 dprintk("%s: Using NFSv4 I/O\n", __func__);
201 server->pnfs_curr_ld = NULL;
202}
02c35fca
FI		 203
204int
205pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206{
207 int status = -EINVAL;
208 struct pnfs_layoutdriver_type *tmp;
209
210 if (ld_type->id == 0) {
a030889a 211 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 212 return status;
213 }
b1f69b75 214 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 215 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 216 "alloc_lseg and free_lseg.\n", __func__);
217 return status;
218 }
02c35fca
FI		 219
220 spin_lock(&pnfs_spinlock);
221 tmp = find_pnfs_driver_locked(ld_type->id);
222 if (!tmp) {
223 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 status = 0;
225 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 ld_type->name);
227 } else {
a030889a 228 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 229 __func__, ld_type->id);
230 }
231 spin_unlock(&pnfs_spinlock);
232
233 return status;
234}
235EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236
237void
238pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239{
240 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 spin_lock(&pnfs_spinlock);
242 list_del(&ld_type->pnfs_tblid);
243 spin_unlock(&pnfs_spinlock);
244}
245EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 246
b1f69b75
AA		 247/*
248 * pNFS client layout cache
249 */
250
cc6e5340 251/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 252void
70c3bd2b 253pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 254{
2b28a7be 255 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 256}
257
636fb9c8
BH		 258static struct pnfs_layout_hdr *
259pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260{
261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 262 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 263}
264
265static void
266pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267{
9c626381
TM		 268 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270
271 if (!list_empty(&lo->plh_layouts)) {
272 struct nfs_client *clp = server->nfs_client;
273
274 spin_lock(&clp->cl_lock);
275 list_del_init(&lo->plh_layouts);
276 spin_unlock(&clp->cl_lock);
277 }
a52458b4 278 put_cred(lo->plh_lc_cred);
57934278 279 return ld->free_layout_hdr(lo);
636fb9c8
BH		 280}
281
e5e94017 282static void
6622c3ea 283pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 284{
bb346f63 285 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 286 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 287 nfsi->layout = NULL;
288 /* Reset MDS Threshold I/O counters */
289 nfsi->write_io = 0;
290 nfsi->read_io = 0;
e5e94017
BH		 291}
292
b1f69b75 293void
70c3bd2b 294pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 295{
9c6376eb 296 struct inode *inode;
cc6e5340 297
9c6376eb
TM		 298 if (!lo)
299 return;
300 inode = lo->plh_inode;
13c13a6a
TM		 301 pnfs_layoutreturn_before_put_layout_hdr(lo);
302
2b28a7be 303 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 304 if (!list_empty(&lo->plh_segs))
305 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 306 pnfs_detach_layout_hdr(lo);
cc6e5340 307 spin_unlock(&inode->i_lock);
6622c3ea 308 pnfs_free_layout_hdr(lo);
cc6e5340 309 }
974cec8c
AA		 310}
311
4aab9732
TM		 312static void
313pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
314 u32 seq)
315{
316 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
317 iomode = IOMODE_ANY;
318 lo->plh_return_iomode = iomode;
319 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
320 if (seq != 0) {
321 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
322 lo->plh_return_seq = seq;
323 }
324}
325
ae5a459d
TM		 326static void
327pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
328{
5466d214 329 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 330 lo->plh_return_iomode = 0;
331 lo->plh_return_seq = 0;
332 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 333 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
334 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
335 continue;
336 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
337 }
ae5a459d
TM		 338}
339
362fb578
TM		 340static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
341{
342 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
343 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
344 smp_mb__after_atomic();
345 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
346 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
347}
348
fe1cf946
TM		 349static void
350pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
351 struct list_head *free_me)
352{
353 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
354 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
355 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
356 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
357 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
358 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
359}
360
7380020e
TM		 361/*
362 * Update the seqid of a layout stateid
363 */
ecf84026
TM		 364bool nfs4_layoutreturn_refresh_stateid(nfs4_stateid *dst,
365 struct pnfs_layout_range *dst_range,
366 struct inode *inode)
7380020e
TM		 367{
368 struct pnfs_layout_hdr *lo;
c16467dc
TM		 369 struct pnfs_layout_range range = {
370 .iomode = IOMODE_ANY,
371 .offset = 0,
372 .length = NFS4_MAX_UINT64,
373 };
7380020e 374 bool ret = false;
c16467dc
TM		 375 LIST_HEAD(head);
376 int err;
7380020e
TM		 377
378 spin_lock(&inode->i_lock);
379 lo = NFS_I(inode)->layout;
380 if (lo && nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
c16467dc
TM		 381 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
382 if (err != -EBUSY) {
383 dst->seqid = lo->plh_stateid.seqid;
ecf84026 384 *dst_range = range;
c16467dc
TM		 385 ret = true;
386 }
7380020e
TM		 387 }
388 spin_unlock(&inode->i_lock);
c16467dc 389 pnfs_free_lseg_list(&head);
7380020e
TM		 390 return ret;
391}
392
2454dfea
TM		 393/*
394 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
395 *
396 * In order to continue using the pnfs_layout_hdr, a full recovery
397 * is required.
398 * Note that caller must hold inode->i_lock.
399 */
5f46be04 400int
2454dfea
TM		 401pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
402 struct list_head *lseg_list)
403{
404 struct pnfs_layout_range range = {
405 .iomode = IOMODE_ANY,
406 .offset = 0,
407 .length = NFS4_MAX_UINT64,
408 };
fe1cf946 409 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 410
411 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 412 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
413 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 414 pnfs_clear_layoutreturn_info(lo);
68f74479 415 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 416 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
417 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
418 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 419 return !list_empty(&lo->plh_segs);
2454dfea
TM		 420}
421
b9e028fd
TM		 422static int
423pnfs_iomode_to_fail_bit(u32 iomode)
424{
425 return iomode == IOMODE_RW ?
426 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
427}
428
429static void
3e621214 430pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 431{
25c75333 432 lo->plh_retry_timestamp = jiffies;
39e88fcf 433 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 434 refcount_inc(&lo->plh_refcount);
3e621214
TM		 435}
436
437static void
438pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
439{
440 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 441 refcount_dec(&lo->plh_refcount);
3e621214
TM		 442}
443
444static void
445pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
446{
447 struct inode *inode = lo->plh_inode;
115ce575
TM		 448 struct pnfs_layout_range range = {
449 .iomode = iomode,
450 .offset = 0,
451 .length = NFS4_MAX_UINT64,
452 };
453 LIST_HEAD(head);
3e621214
TM		 454
455 spin_lock(&inode->i_lock);
456 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 457 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 458 spin_unlock(&inode->i_lock);
115ce575 459 pnfs_free_lseg_list(&head);
b9e028fd
TM		 460 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
461 iomode == IOMODE_RW ? "RW" : "READ");
462}
463
464static bool
465pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
466{
25c75333 467 unsigned long start, end;
3e621214
TM		 468 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
469
470 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 471 return false;
472 end = jiffies;
473 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
474 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
475 /* It is time to retry the failed layoutgets */
3e621214 476 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 477 return false;
478 }
479 return true;
b9e028fd
TM		 480}
481
974cec8c 482static void
119cef97
TM		 483pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
484 const struct pnfs_layout_range *range,
485 const nfs4_stateid *stateid)
974cec8c 486{
566052c5 487 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 488 INIT_LIST_HEAD(&lseg->pls_lc_list);
eba6dd69 489 refcount_set(&lseg->pls_refcount, 1);
4541d16c 490 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 491 lseg->pls_layout = lo;
119cef97
TM		 492 lseg->pls_range = *range;
493 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 494}
495
905ca191 496static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 497{
68f74479
TM		 498 if (lseg != NULL) {
499 struct inode *inode = lseg->pls_layout->plh_inode;
500 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
501 }
974cec8c
AA		 502}
503
d684d2ae 504static void
57036a37
TM		 505pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
506 struct pnfs_layout_segment *lseg)
d684d2ae 507{
d20581aa 508 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 509 list_del_init(&lseg->pls_list);
8f0d27dc 510 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 511 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 512 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
513 return;
7b650994
TM		 514 if (list_empty(&lo->plh_segs) &&
515 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
516 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 517 if (atomic_read(&lo->plh_outstanding) == 0)
518 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 519 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 520 }
d684d2ae
FI		 521}
522
68f74479
TM		 523static bool
524pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
525 struct pnfs_layout_segment *lseg)
526{
527 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
528 pnfs_layout_is_valid(lo)) {
4aab9732 529 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 530 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
531 return true;
532 }
533 return false;
534}
535
bae724ef 536void
9369a431 537pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 538{
57036a37 539 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 540 struct inode *inode;
541
542 if (!lseg)
543 return;
544
4541d16c 545 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 546 refcount_read(&lseg->pls_refcount),
4541d16c 547 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 548
57036a37
TM		 549 lo = lseg->pls_layout;
550 inode = lo->plh_inode;
4ef2e4f8 551
eba6dd69 552 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 553 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
554 spin_unlock(&inode->i_lock);
555 return;
556 }
8f0d27dc 557 pnfs_get_layout_hdr(lo);
4ef2e4f8 558 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 559 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
560 lseg = NULL;
4ef2e4f8
TM		 561 spin_unlock(&inode->i_lock);
562 pnfs_free_lseg(lseg);
563 pnfs_put_layout_hdr(lo);
4541d16c 564 }
4541d16c 565}
9369a431 566EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 567
fb3296eb
BH		 568/*
569 * is l2 fully contained in l1?
570 * start1 end1
571 * [----------------------------------)
572 * start2 end2
573 * [----------------)
574 */
3cb2df17 575static bool
7dc0ac70 576pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 577 const struct pnfs_layout_range *l2)
fb3296eb
BH		 578{
579 u64 start1 = l1->offset;
17822b20 580 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 581 u64 start2 = l2->offset;
17822b20 582 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 583
584 return (start1 <= start2) && (end1 >= end2);
585}
586
24956804
TM		 587static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
588 struct list_head *tmp_list)
589{
eba6dd69 590 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 591 return false;
592 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
593 list_add(&lseg->pls_list, tmp_list);
594 return true;
595}
596
4541d16c
FI		 597/* Returns 1 if lseg is removed from list, 0 otherwise */
598static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
599 struct list_head *tmp_list)
600{
601 int rv = 0;
602
603 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
604 /* Remove the reference keeping the lseg in the
605 * list. It will now be removed when all
606 * outstanding io is finished.
607 */
d684d2ae 608 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 609 refcount_read(&lseg->pls_refcount));
24956804 610 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 611 rv = 1;
4541d16c
FI		 612 }
613 return rv;
614}
615
6d597e17
JL		 616/*
617 * Compare 2 layout stateid sequence ids, to see which is newer,
618 * taking into account wraparound issues.
619 */
620static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
621{
622 return (s32)(s1 - s2) > 0;
623}
624
e036f464
TM		 625static bool
626pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
627 const struct pnfs_layout_range *recall_range)
628{
629 return (recall_range->iomode == IOMODE_ANY ||
630 lseg_range->iomode == recall_range->iomode) &&
631 pnfs_lseg_range_intersecting(lseg_range, recall_range);
632}
633
634static bool
635pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
636 const struct pnfs_layout_range *recall_range,
637 u32 seq)
638{
639 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
640 return false;
641 if (recall_range == NULL)
642 return true;
643 return pnfs_should_free_range(&lseg->pls_range, recall_range);
644}
645
6d597e17
JL		 646/**
647 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
648 * @lo: layout header containing the lsegs
649 * @tmp_list: list head where doomed lsegs should go
650 * @recall_range: optional recall range argument to match (may be NULL)
651 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
652 *
653 * Walk the list of lsegs in the layout header, and tear down any that should
654 * be destroyed. If "recall_range" is specified then the segment must match
655 * that range. If "seq" is non-zero, then only match segments that were handed
656 * out at or before that sequence.
657 *
658 * Returns number of matching invalid lsegs remaining in list after scanning
659 * it and purging them.
4541d16c 660 */
43f1b3da 661int
49a85061 662pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 663 struct list_head *tmp_list,
6d597e17
JL		 664 const struct pnfs_layout_range *recall_range,
665 u32 seq)
974cec8c
AA		 666{
667 struct pnfs_layout_segment *lseg, *next;
71b39854 668 int remaining = 0;
974cec8c
AA		 669
670 dprintk("%s:Begin lo %p\n", __func__, lo);
671
8006bfba 672 if (list_empty(&lo->plh_segs))
38511722 673 return 0;
4541d16c 674 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 675 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 676 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 677 "offset %llu length %llu\n", __func__,
6d597e17
JL		 678 lseg, lseg->pls_range.iomode, lseg->pls_seq,
679 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 680 if (!mark_lseg_invalid(lseg, tmp_list))
681 remaining++;
4541d16c 682 }
71b39854
TM		 683 dprintk("%s:Return %i\n", __func__, remaining);
684 return remaining;
974cec8c
AA		 685}
686
68f74479
TM		 687static void
688pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
689 struct list_head *free_me,
690 const struct pnfs_layout_range *range,
691 u32 seq)
692{
693 struct pnfs_layout_segment *lseg, *next;
694
695 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
696 if (pnfs_match_lseg_recall(lseg, range, seq))
697 list_move_tail(&lseg->pls_list, free_me);
698 }
699}
700
f49f9baa 701/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 702void
4541d16c 703pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 704{
4541d16c 705 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 706
707 if (list_empty(free_me))
708 return;
709
4541d16c 710 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 711 list_del(&lseg->pls_list);
905ca191 712 pnfs_free_lseg(lseg);
974cec8c
AA		 713 }
714}
715
e5e94017
BH		 716void
717pnfs_destroy_layout(struct nfs_inode *nfsi)
718{
719 struct pnfs_layout_hdr *lo;
974cec8c 720 LIST_HEAD(tmp_list);
e5e94017
BH		 721
722 spin_lock(&nfsi->vfs_inode.i_lock);
723 lo = nfsi->layout;
724 if (lo) {
3e621214 725 pnfs_get_layout_hdr(lo);
2454dfea 726 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 727 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
728 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
729 spin_unlock(&nfsi->vfs_inode.i_lock);
730 pnfs_free_lseg_list(&tmp_list);
1f18b82c 731 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 732 pnfs_put_layout_hdr(lo);
733 } else
734 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 735}
041245c8 736EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 737
fd9a8d71
TM		 738static bool
739pnfs_layout_add_bulk_destroy_list(struct inode *inode,
740 struct list_head *layout_list)
974cec8c
AA		 741{
742 struct pnfs_layout_hdr *lo;
fd9a8d71 743 bool ret = false;
974cec8c 744
fd9a8d71
TM		 745 spin_lock(&inode->i_lock);
746 lo = NFS_I(inode)->layout;
747 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
748 pnfs_get_layout_hdr(lo);
749 list_add(&lo->plh_bulk_destroy, layout_list);
750 ret = true;
751 }
752 spin_unlock(&inode->i_lock);
753 return ret;
754}
755
756/* Caller must hold rcu_read_lock and clp->cl_lock */
757static int
758pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
759 struct nfs_server *server,
760 struct list_head *layout_list)
5085607d
TM		 761 __must_hold(&clp->cl_lock)
762 __must_hold(RCU)
fd9a8d71
TM		 763{
764 struct pnfs_layout_hdr *lo, *next;
765 struct inode *inode;
766
767 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
5085607d
TM		 768 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
769 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
770 !list_empty(&lo->plh_bulk_destroy))
b85f5620 771 continue;
5085607d
TM		 772 /* If the sb is being destroyed, just bail */
773 if (!nfs_sb_active(server->super))
774 break;
fd9a8d71 775 inode = igrab(lo->plh_inode);
5085607d
TM		 776 if (inode != NULL) {
777 list_del_init(&lo->plh_layouts);
778 if (pnfs_layout_add_bulk_destroy_list(inode,
779 layout_list))
780 continue;
781 rcu_read_unlock();
782 spin_unlock(&clp->cl_lock);
783 iput(inode);
784 } else {
785 rcu_read_unlock();
786 spin_unlock(&clp->cl_lock);
787 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
788 }
789 nfs_sb_deactive(server->super);
fd9a8d71
TM		 790 spin_lock(&clp->cl_lock);
791 rcu_read_lock();
792 return -EAGAIN;
793 }
794 return 0;
795}
796
797static int
798pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
799 bool is_bulk_recall)
800{
801 struct pnfs_layout_hdr *lo;
802 struct inode *inode;
fd9a8d71
TM		 803 LIST_HEAD(lseg_list);
804 int ret = 0;
805
806 while (!list_empty(layout_list)) {
807 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
808 plh_bulk_destroy);
809 dprintk("%s freeing layout for inode %lu\n", __func__,
810 lo->plh_inode->i_ino);
811 inode = lo->plh_inode;
7c5d1875
CH		 812
813 pnfs_layoutcommit_inode(inode, false);
814
fd9a8d71
TM		 815 spin_lock(&inode->i_lock);
816 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 817 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
818 if (is_bulk_recall)
819 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 820 ret = -EAGAIN;
9fd4b9fc 821 }
fd9a8d71
TM		 822 spin_unlock(&inode->i_lock);
823 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 824 /* Free all lsegs that are attached to commit buckets */
825 nfs_commit_inode(inode, 0);
fd9a8d71 826 pnfs_put_layout_hdr(lo);
5085607d 827 nfs_iput_and_deactive(inode);
fd9a8d71
TM		 828 }
829 return ret;
830}
831
832int
833pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
834 struct nfs_fsid *fsid,
835 bool is_recall)
836{
837 struct nfs_server *server;
838 LIST_HEAD(layout_list);
c47abcf8 839
974cec8c 840 spin_lock(&clp->cl_lock);
6382a441 841 rcu_read_lock();
fd9a8d71 842restart:
6382a441 843 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 844 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
845 continue;
846 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
847 server,
848 &layout_list) != 0)
849 goto restart;
6382a441
WAA		 850 }
851 rcu_read_unlock();
974cec8c
AA		 852 spin_unlock(&clp->cl_lock);
853
fd9a8d71
TM		 854 if (list_empty(&layout_list))
855 return 0;
856 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
857}
858
859int
860pnfs_destroy_layouts_byclid(struct nfs_client *clp,
861 bool is_recall)
862{
863 struct nfs_server *server;
864 LIST_HEAD(layout_list);
865
866 spin_lock(&clp->cl_lock);
867 rcu_read_lock();
868restart:
869 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
870 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
871 server,
872 &layout_list) != 0)
873 goto restart;
974cec8c 874 }
fd9a8d71
TM		 875 rcu_read_unlock();
876 spin_unlock(&clp->cl_lock);
877
878 if (list_empty(&layout_list))
879 return 0;
880 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
881}
882
883/*
884 * Called by the state manger to remove all layouts established under an
885 * expired lease.
886 */
887void
888pnfs_destroy_all_layouts(struct nfs_client *clp)
889{
890 nfs4_deviceid_mark_client_invalid(clp);
891 nfs4_deviceid_purge_client(clp);
892
893 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 894}
895
fd6002e9 896/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 897void
898pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
899 bool update_barrier)
b1f69b75 900{
ecebb80b 901 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 902
2d2f24ad
TM		 903 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
904 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 905
906 if (!pnfs_layout_is_valid(lo)) {
907 nfs4_stateid_copy(&lo->plh_stateid, new);
908 lo->plh_barrier = newseq;
909 pnfs_clear_layoutreturn_info(lo);
910 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
911 return;
912 }
913 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 914 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 915 /*
916 * Because of wraparound, we want to keep the barrier
917 * "close" to the current seqids.
918 */
919 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 920 }
ecebb80b
TM		 921 if (update_barrier)
922 new_barrier = be32_to_cpu(new->seqid);
923 else if (new_barrier == 0)
924 return;
2a59a041 925 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 926 lo->plh_barrier = new_barrier;
b1f69b75
AA		 927}
928
cf7d63f1 929static bool
19c54aba
TM		 930pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
931 const nfs4_stateid *stateid)
43f1b3da 932{
19c54aba 933 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 934
19c54aba
TM		 935 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
936}
937
938/* lget is set to 1 if called from inside send_layoutget call chain */
939static bool
e1c06f80 940pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 941{
f7e8917a 942 return lo->plh_block_lgets ||
e1c06f80 943 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 944}
945
5e36e2a9
FI		 946static struct nfs_server *
947pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
948{
949 struct nfs_server *server;
950
78746a38 951 if (inode) {
5e36e2a9 952 server = NFS_SERVER(inode);
78746a38 953 } else {
5e36e2a9
FI		 954 struct dentry *parent_dir = dget_parent(ctx->dentry);
955 server = NFS_SERVER(parent_dir->d_inode);
956 dput(parent_dir);
957 }
958 return server;
959}
960
29a8bfe5
TM		 961static void nfs4_free_pages(struct page **pages, size_t size)
962{
963 int i;
964
965 if (!pages)
966 return;
967
968 for (i = 0; i < size; i++) {
969 if (!pages[i])
970 break;
971 __free_page(pages[i]);
972 }
973 kfree(pages);
974}
975
976static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
977{
978 struct page **pages;
979 int i;
980
a2791d3a 981 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 982 if (!pages) {
983 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
984 return NULL;
985 }
986
987 for (i = 0; i < size; i++) {
988 pages[i] = alloc_page(gfp_flags);
989 if (!pages[i]) {
990 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 991 nfs4_free_pages(pages, i);
29a8bfe5
TM		 992 return NULL;
993 }
994 }
995
996 return pages;
997}
998
587f03de 999static struct nfs4_layoutget *
5e36e2a9 1000pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 1001 struct nfs_open_context *ctx,
2409a976 1002 const nfs4_stateid *stateid,
e144e539 1003 const struct pnfs_layout_range *range,
587f03de 1004 gfp_t gfp_flags)
e5e94017 1005{
5e36e2a9 1006 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 1007 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 1008 size_t max_pages = max_response_pages(server);
b1f69b75 1009 struct nfs4_layoutget *lgp;
b1f69b75
AA		 1010
1011 dprintk("--> %s\n", __func__);
e5e94017 1012
83026d80
JL		 1013 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1014 if (lgp == NULL)
587f03de 1015 return NULL;
83026d80 1016
28ced9a8
TM		 1017 if (max_reply_sz) {
1018 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1019 if (npages < max_pages)
1020 max_pages = npages;
1021 }
1022
dacb452d
FI		 1023 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1024 if (!lgp->args.layout.pages) {
1025 kfree(lgp);
1026 return NULL;
1027 }
1028 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1029 lgp->res.layoutp = &lgp->args.layout;
1030
d49e0d5b
TM		 1031 /* Don't confuse uninitialised result and success */
1032 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1033
1034 lgp->args.minlength = PAGE_SIZE;
1035 if (lgp->args.minlength > range->length)
1036 lgp->args.minlength = range->length;
5e36e2a9
FI		 1037 if (ino) {
1038 loff_t i_size = i_size_read(ino);
1039
1040 if (range->iomode == IOMODE_READ) {
1041 if (range->offset >= i_size)
1042 lgp->args.minlength = 0;
1043 else if (i_size - range->offset < lgp->args.minlength)
1044 lgp->args.minlength = i_size - range->offset;
1045 }
d03ab29d 1046 }
83026d80
JL		 1047 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1048 pnfs_copy_range(&lgp->args.range, range);
1049 lgp->args.type = server->pnfs_curr_ld->id;
1050 lgp->args.inode = ino;
1051 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1052 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1053 lgp->gfp_flags = gfp_flags;
a52458b4 1054 lgp->cred = get_cred(ctx->cred);
587f03de 1055 return lgp;
974cec8c
AA		 1056}
1057
29a8bfe5
TM		 1058void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1059{
1060 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1061
1062 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1063 if (lgp->args.inode)
1064 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
a52458b4 1065 put_cred(lgp->cred);
29a8bfe5
TM		 1066 put_nfs_open_context(lgp->args.ctx);
1067 kfree(lgp);
1068}
1069
24956804
TM		 1070static void pnfs_clear_layoutcommit(struct inode *inode,
1071 struct list_head *head)
1072{
1073 struct nfs_inode *nfsi = NFS_I(inode);
1074 struct pnfs_layout_segment *lseg, *tmp;
1075
1076 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1077 return;
1078 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1079 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1080 continue;
1081 pnfs_lseg_dec_and_remove_zero(lseg, head);
1082 }
1083}
1084
68f74479 1085void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1086 const nfs4_stateid *arg_stateid,
68f74479 1087 const struct pnfs_layout_range *range,
68f74479
TM		 1088 const nfs4_stateid *stateid)
1089{
1090 struct inode *inode = lo->plh_inode;
1091 LIST_HEAD(freeme);
1092
1093 spin_lock(&inode->i_lock);
2a974425
TM		 1094 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1095 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1096 goto out_unlock;
68f74479 1097 if (stateid) {
2a974425
TM		 1098 u32 seq = be32_to_cpu(arg_stateid->seqid);
1099
68f74479
TM		 1100 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1101 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1102 pnfs_set_layout_stateid(lo, stateid, true);
1103 } else
1104 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1105out_unlock:
68f74479
TM		 1106 pnfs_clear_layoutreturn_waitbit(lo);
1107 spin_unlock(&inode->i_lock);
1108 pnfs_free_lseg_list(&freeme);
1109
1110}
1111
13c13a6a 1112static bool
e5fd1904
TM		 1113pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1114 nfs4_stateid *stateid,
1115 enum pnfs_iomode *iomode)
13c13a6a 1116{
bf0291dd
TM		 1117 /* Serialise LAYOUTGET/LAYOUTRETURN */
1118 if (atomic_read(&lo->plh_outstanding) != 0)
1119 return false;
6604b203 1120 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1121 return false;
6604b203 1122 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1123 pnfs_get_layout_hdr(lo);
e5fd1904
TM		 1124 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1125 if (stateid != NULL) {
1126 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1127 if (lo->plh_return_seq != 0)
1128 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1129 }
1130 if (iomode != NULL)
1131 *iomode = lo->plh_return_iomode;
1132 pnfs_clear_layoutreturn_info(lo);
1133 return true;
1134 }
1135 if (stateid != NULL)
1136 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1137 if (iomode != NULL)
1138 *iomode = IOMODE_ANY;
13c13a6a
TM		 1139 return true;
1140}
1141
828ed9ec
TM		 1142static void
1143pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1144 struct pnfs_layout_hdr *lo,
1145 const nfs4_stateid *stateid,
1146 enum pnfs_iomode iomode)
1147{
1148 struct inode *inode = lo->plh_inode;
1149
1150 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1151 args->inode = inode;
1152 args->range.iomode = iomode;
1153 args->range.offset = 0;
1154 args->range.length = NFS4_MAX_UINT64;
1155 args->layout = lo;
1156 nfs4_stateid_copy(&args->stateid, stateid);
1157}
1158
f40eb5d0 1159static int
ed429d6b 1160pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
6c16605d 1161 enum pnfs_iomode iomode, bool sync)
f40eb5d0
PT		 1162{
1163 struct inode *ino = lo->plh_inode;
287bd3e9 1164 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0
PT		 1165 struct nfs4_layoutreturn *lrp;
1166 int status = 0;
1167
e4af440a 1168 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1169 if (unlikely(lrp == NULL)) {
1170 status = -ENOMEM;
1171 spin_lock(&ino->i_lock);
d67ae825 1172 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0
PT		 1173 spin_unlock(&ino->i_lock);
1174 pnfs_put_layout_hdr(lo);
1175 goto out;
1176 }
1177
828ed9ec 1178 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1179 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0
PT		 1180 lrp->clp = NFS_SERVER(ino)->nfs_client;
1181 lrp->cred = lo->plh_lc_cred;
287bd3e9
TM		 1182 if (ld->prepare_layoutreturn)
1183 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1184
6c16605d 1185 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1186out:
1187 dprintk("<-- %s status: %d\n", __func__, status);
1188 return status;
1189}
1190
13c13a6a
TM		 1191/* Return true if layoutreturn is needed */
1192static bool
1193pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1194{
1195 struct pnfs_layout_segment *s;
f0b42981
TM		 1196 enum pnfs_iomode iomode;
1197 u32 seq;
13c13a6a 1198
2370abda 1199 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1200 return false;
1201
f0b42981
TM		 1202 seq = lo->plh_return_seq;
1203 iomode = lo->plh_return_iomode;
1204
1205 /* Defer layoutreturn until all recalled lsegs are done */
13c13a6a 1206 list_for_each_entry(s, &lo->plh_segs, pls_list) {
f0b42981
TM		 1207 if (seq && pnfs_seqid_is_newer(s->pls_seq, seq))
1208 continue;
1209 if (iomode != IOMODE_ANY && s->pls_range.iomode != iomode)
1210 continue;
13c13a6a
TM		 1211 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1212 return false;
1213 }
1214
1215 return true;
1216}
1217
1218static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1219{
1220 struct inode *inode= lo->plh_inode;
1221
2370abda 1222 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1223 return;
1224 spin_lock(&inode->i_lock);
1225 if (pnfs_layout_need_return(lo)) {
1226 nfs4_stateid stateid;
1227 enum pnfs_iomode iomode;
1228 bool send;
1229
e5fd1904 1230 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
13c13a6a
TM		 1231 spin_unlock(&inode->i_lock);
1232 if (send) {
1233 /* Send an async layoutreturn so we dont deadlock */
1234 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1235 }
1236 } else
1237 spin_unlock(&inode->i_lock);
1238}
1239
293b3b06
AA		 1240/*
1241 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1242 * when the layout segment list is empty.
1243 *
1244 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1245 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1246 * deviceid is marked invalid.
1247 */
cbe82603
BH		 1248int
1249_pnfs_return_layout(struct inode *ino)
1250{
1251 struct pnfs_layout_hdr *lo = NULL;
1252 struct nfs_inode *nfsi = NFS_I(ino);
1253 LIST_HEAD(tmp_list);
cbe82603 1254 nfs4_stateid stateid;
24408f52 1255 int status = 0;
93b7f7ad 1256 bool send, valid_layout;
cbe82603 1257
366d5052 1258 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1259
1260 spin_lock(&ino->i_lock);
1261 lo = nfsi->layout;
e5929f3c 1262 if (!lo) {
cbe82603 1263 spin_unlock(&ino->i_lock);
293b3b06
AA		 1264 dprintk("NFS: %s no layout to return\n", __func__);
1265 goto out;
cbe82603 1266 }
cbe82603 1267 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1268 pnfs_get_layout_hdr(lo);
24408f52
TM		 1269 /* Is there an outstanding layoutreturn ? */
1270 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1271 spin_unlock(&ino->i_lock);
1272 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1273 TASK_UNINTERRUPTIBLE))
1274 goto out_put_layout_hdr;
1275 spin_lock(&ino->i_lock);
1276 }
93b7f7ad 1277 valid_layout = pnfs_layout_is_valid(lo);
24956804 1278 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1279 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1280
1281 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1282 struct pnfs_layout_range range = {
1283 .iomode = IOMODE_ANY,
1284 .offset = 0,
1285 .length = NFS4_MAX_UINT64,
1286 };
1287 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1288 }
1289
293b3b06 1290 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1291 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1292 !valid_layout) {
293b3b06 1293 spin_unlock(&ino->i_lock);
293b3b06 1294 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1295 goto out_put_layout_hdr;
293b3b06 1296 }
47abadef 1297
e5fd1904 1298 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603 1299 spin_unlock(&ino->i_lock);
7f27392c 1300 if (send)
ed429d6b 1301 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c 1302out_put_layout_hdr:
ee6625a9 1303 pnfs_free_lseg_list(&tmp_list);
7f27392c 1304 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1305out:
1306 dprintk("<-- %s status: %d\n", __func__, status);
1307 return status;
1308}
1309
24028672
TM		 1310int
1311pnfs_commit_and_return_layout(struct inode *inode)
1312{
1313 struct pnfs_layout_hdr *lo;
1314 int ret;
1315
1316 spin_lock(&inode->i_lock);
1317 lo = NFS_I(inode)->layout;
1318 if (lo == NULL) {
1319 spin_unlock(&inode->i_lock);
1320 return 0;
1321 }
1322 pnfs_get_layout_hdr(lo);
1323 /* Block new layoutgets and read/write to ds */
1324 lo->plh_block_lgets++;
1325 spin_unlock(&inode->i_lock);
1326 filemap_fdatawait(inode->i_mapping);
1327 ret = pnfs_layoutcommit_inode(inode, true);
1328 if (ret == 0)
1329 ret = _pnfs_return_layout(inode);
1330 spin_lock(&inode->i_lock);
1331 lo->plh_block_lgets--;
1332 spin_unlock(&inode->i_lock);
1333 pnfs_put_layout_hdr(lo);
1334 return ret;
1335}
1336
1c5bd76d
TM		 1337bool pnfs_roc(struct inode *ino,
1338 struct nfs4_layoutreturn_args *args,
1339 struct nfs4_layoutreturn_res *res,
a52458b4 1340 const struct cred *cred)
f7e8917a 1341{
40dd4b7a
TM		 1342 struct nfs_inode *nfsi = NFS_I(ino);
1343 struct nfs_open_context *ctx;
1344 struct nfs4_state *state;
f7e8917a 1345 struct pnfs_layout_hdr *lo;
1c5bd76d 1346 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1347 nfs4_stateid stateid;
1c5bd76d
TM		 1348 enum pnfs_iomode iomode = 0;
1349 bool layoutreturn = false, roc = false;
e71708d4 1350 bool skip_read = false;
f7e8917a 1351
1c5bd76d
TM		 1352 if (!nfs_have_layout(ino))
1353 return false;
29ade5db 1354retry:
0de43976 1355 rcu_read_lock();
f7e8917a 1356 spin_lock(&ino->i_lock);
40dd4b7a 1357 lo = nfsi->layout;
0cdc329e 1358 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1359 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1360 lo = NULL;
40dd4b7a 1361 goto out_noroc;
9c6376eb
TM		 1362 }
1363 pnfs_get_layout_hdr(lo);
29ade5db 1364 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1365 spin_unlock(&ino->i_lock);
0de43976 1366 rcu_read_unlock();
29ade5db
TM		 1367 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1368 TASK_UNINTERRUPTIBLE);
1369 pnfs_put_layout_hdr(lo);
1370 goto retry;
1371 }
40dd4b7a 1372
e755d638 1373 /* no roc if we hold a delegation */
e71708d4
TM		 1374 if (nfs4_check_delegation(ino, FMODE_READ)) {
1375 if (nfs4_check_delegation(ino, FMODE_WRITE))
1376 goto out_noroc;
1377 skip_read = true;
1378 }
40dd4b7a 1379
0de43976 1380 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1381 state = ctx->state;
e71708d4
TM		 1382 if (state == NULL)
1383 continue;
40dd4b7a 1384 /* Don't return layout if there is open file state */
e71708d4 1385 if (state->state & FMODE_WRITE)
40dd4b7a 1386 goto out_noroc;
e71708d4
TM		 1387 if (state->state & FMODE_READ)
1388 skip_read = true;
40dd4b7a
TM		 1389 }
1390
e755d638 1391
1c5bd76d 1392 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1393 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1394 continue;
e755d638 1395 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1396 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1397 continue;
1398 /*
1399 * Note: mark lseg for return so pnfs_layout_remove_lseg
1400 * doesn't invalidate the layout for us.
1401 */
1402 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1403 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1404 continue;
1405 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1406 }
1407
1c5bd76d
TM		 1408 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1409 goto out_noroc;
69820d22 1410
500d701f 1411 /* ROC in two conditions:
e755d638
PT		 1412 * 1. there are ROC lsegs
1413 * 2. we don't send layoutreturn
e755d638 1414 */
1c5bd76d
TM		 1415 /* lo ref dropped in pnfs_roc_release() */
1416 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1417 /* If the creds don't match, we can't compound the layoutreturn */
1418 if (!layoutreturn || cred != lo->plh_lc_cred)
1419 goto out_noroc;
1420
1421 roc = layoutreturn;
1422 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1423 res->lrs_present = 0;
1424 layoutreturn = false;
f7e8917a 1425
40dd4b7a 1426out_noroc:
f7e8917a 1427 spin_unlock(&ino->i_lock);
0de43976 1428 rcu_read_unlock();
e755d638 1429 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1430 if (roc) {
1431 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1432 if (ld->prepare_layoutreturn)
1433 ld->prepare_layoutreturn(args);
9c6376eb 1434 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1435 return true;
1436 }
e755d638 1437 if (layoutreturn)
1c5bd76d 1438 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
9c6376eb 1439 pnfs_put_layout_hdr(lo);
287bd3e9 1440 return false;
f7e8917a
FI		 1441}
1442
1c5bd76d
TM		 1443void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1444 struct nfs4_layoutreturn_res *res,
1445 int ret)
f7e8917a 1446{
1c5bd76d
TM		 1447 struct pnfs_layout_hdr *lo = args->layout;
1448 const nfs4_stateid *arg_stateid = NULL;
1449 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1450 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1451
9c47b18c
TM		 1452 switch (ret) {
1453 case -NFS4ERR_NOMATCHING_LAYOUT:
1454 break;
1455 case 0:
1c5bd76d
TM		 1456 if (res->lrs_present)
1457 res_stateid = &res->stateid;
9c47b18c
TM		 1458 /* Fallthrough */
1459 default:
1460 arg_stateid = &args->stateid;
1c5bd76d
TM		 1461 }
1462 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1463 res_stateid);
287bd3e9
TM		 1464 if (ld_private && ld_private->ops && ld_private->ops->free)
1465 ld_private->ops->free(ld_private);
1c5bd76d
TM		 1466 pnfs_put_layout_hdr(lo);
1467 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1468}
1469
500d701f
PT		 1470bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1471{
1472 struct nfs_inode *nfsi = NFS_I(ino);
1473 struct pnfs_layout_hdr *lo;
1474 bool sleep = false;
1475
1476 /* we might not have grabbed lo reference. so need to check under
1477 * i_lock */
1478 spin_lock(&ino->i_lock);
1479 lo = nfsi->layout;
ee284e35
TM		 1480 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1481 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1482 sleep = true;
ee284e35 1483 }
500d701f 1484 spin_unlock(&ino->i_lock);
500d701f
PT		 1485 return sleep;
1486}
1487
b1f69b75
AA		 1488/*
1489 * Compare two layout segments for sorting into layout cache.
1490 * We want to preferentially return RW over RO layouts, so ensure those
1491 * are seen first.
1492 */
1493static s64
7dc0ac70 1494pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1495 const struct pnfs_layout_range *l2)
b1f69b75 1496{
fb3296eb
BH		 1497 s64 d;
1498
1499 /* high offset > low offset */
1500 d = l1->offset - l2->offset;
1501 if (d)
1502 return d;
1503
1504 /* short length > long length */
1505 d = l2->length - l1->length;
1506 if (d)
1507 return d;
1508
b1f69b75 1509 /* read > read/write */
fb3296eb 1510 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1511}
1512
03772d2f
TM		 1513static bool
1514pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1515 const struct pnfs_layout_range *l2)
1516{
1517 return pnfs_lseg_range_cmp(l1, l2) > 0;
1518}
1519
1520static bool
1521pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1522 struct pnfs_layout_segment *old)
1523{
1524 return false;
1525}
1526
1527void
1528pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1529 struct pnfs_layout_segment *lseg,
1530 bool (*is_after)(const struct pnfs_layout_range *,
1531 const struct pnfs_layout_range *),
1532 bool (*do_merge)(struct pnfs_layout_segment *,
1533 struct pnfs_layout_segment *),
1534 struct list_head *free_me)
974cec8c 1535{
03772d2f 1536 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1537
974cec8c
AA		 1538 dprintk("%s:Begin\n", __func__);
1539
03772d2f
TM		 1540 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1541 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1542 continue;
1543 if (do_merge(lseg, lp)) {
1544 mark_lseg_invalid(lp, free_me);
1545 continue;
1546 }
1547 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1548 continue;
566052c5 1549 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1550 dprintk("%s: inserted lseg %p "
1551 "iomode %d offset %llu length %llu before "
1552 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1553 __func__, lseg, lseg->pls_range.iomode,
1554 lseg->pls_range.offset, lseg->pls_range.length,
1555 lp, lp->pls_range.iomode, lp->pls_range.offset,
1556 lp->pls_range.length);
fb3296eb 1557 goto out;
974cec8c 1558 }
fb3296eb
BH		 1559 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1560 dprintk("%s: inserted lseg %p "
1561 "iomode %d offset %llu length %llu at tail\n",
1562 __func__, lseg, lseg->pls_range.iomode,
1563 lseg->pls_range.offset, lseg->pls_range.length);
1564out:
70c3bd2b 1565 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1566
1567 dprintk("%s:Return\n", __func__);
e5e94017 1568}
03772d2f
TM		 1569EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1570
1571static void
1572pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1573 struct pnfs_layout_segment *lseg,
1574 struct list_head *free_me)
1575{
1576 struct inode *inode = lo->plh_inode;
1577 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1578
1579 if (ld->add_lseg != NULL)
1580 ld->add_lseg(lo, lseg, free_me);
1581 else
1582 pnfs_generic_layout_insert_lseg(lo, lseg,
1583 pnfs_lseg_range_is_after,
1584 pnfs_lseg_no_merge,
1585 free_me);
1586}
e5e94017
BH		 1587
1588static struct pnfs_layout_hdr *
9fa40758
PT		 1589alloc_init_layout_hdr(struct inode *ino,
1590 struct nfs_open_context *ctx,
1591 gfp_t gfp_flags)
e5e94017
BH		 1592{
1593 struct pnfs_layout_hdr *lo;
1594
636fb9c8 1595 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1596 if (!lo)
1597 return NULL;
2b28a7be 1598 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1599 INIT_LIST_HEAD(&lo->plh_layouts);
1600 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1601 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1602 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1603 lo->plh_inode = ino;
a52458b4 1604 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1605 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1606 return lo;
1607}
1608
1609static struct pnfs_layout_hdr *
9fa40758
PT		 1610pnfs_find_alloc_layout(struct inode *ino,
1611 struct nfs_open_context *ctx,
1612 gfp_t gfp_flags)
e5241e43
TM		 1613 __releases(&ino->i_lock)
1614 __acquires(&ino->i_lock)
e5e94017
BH		 1615{
1616 struct nfs_inode *nfsi = NFS_I(ino);
1617 struct pnfs_layout_hdr *new = NULL;
1618
1619 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1620
251ec410
TM		 1621 if (nfsi->layout != NULL)
1622 goto out_existing;
e5e94017 1623 spin_unlock(&ino->i_lock);
9fa40758 1624 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1625 spin_lock(&ino->i_lock);
1626
251ec410 1627 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1628 nfsi->layout = new;
251ec410 1629 return new;
7175fe90
YN		 1630 } else if (new != NULL)
1631 pnfs_free_layout_hdr(new);
251ec410
TM		 1632out_existing:
1633 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1634 return nfsi->layout;
1635}
1636
b1f69b75
AA		 1637/*
1638 * iomode matching rules:
c7d73af2
TH		 1639 * iomode lseg strict match
1640 * iomode
1641 * ----- ----- ------ -----
1642 * ANY READ N/A true
1643 * ANY RW N/A true
1644 * RW READ N/A false
1645 * RW RW N/A true
1646 * READ READ N/A true
1647 * READ RW true false
1648 * READ RW false true
b1f69b75 1649 */
3cb2df17 1650static bool
7dc0ac70 1651pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1652 const struct pnfs_layout_range *range,
1653 bool strict_iomode)
b1f69b75 1654{
fb3296eb
BH		 1655 struct pnfs_layout_range range1;
1656
1657 if ((range->iomode == IOMODE_RW &&
1658 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1659 (range->iomode != ls_range->iomode &&
6089dd0d 1660 strict_iomode) ||
7dc0ac70 1661 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1662 return false;
fb3296eb
BH		 1663
1664 /* range1 covers only the first byte in the range */
1665 range1 = *range;
1666 range1.length = 1;
7dc0ac70 1667 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1668}
1669
1670/*
1671 * lookup range in layout
1672 */
e5e94017 1673static struct pnfs_layout_segment *
fb3296eb 1674pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1675 struct pnfs_layout_range *range,
1676 bool strict_iomode)
e5e94017 1677{
b1f69b75
AA		 1678 struct pnfs_layout_segment *lseg, *ret = NULL;
1679
1680 dprintk("%s:Begin\n", __func__);
1681
b7edfaa1 1682 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1683 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1684 pnfs_lseg_range_match(&lseg->pls_range, range,
1685 strict_iomode)) {
9369a431 1686 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1687 break;
1688 }
b1f69b75
AA		 1689 }
1690
1691 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1692 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1693 return ret;
e5e94017
BH		 1694}
1695
d23d61c8
AA		 1696/*
1697 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1698 * to the MDS or over pNFS
1699 *
1700 * The nfs_inode read_io and write_io fields are cumulative counters reset
1701 * when there are no layout segments. Note that in pnfs_update_layout iomode
1702 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1703 * WRITE request.
1704 *
1705 * A return of true means use MDS I/O.
1706 *
1707 * From rfc 5661:
1708 * If a file's size is smaller than the file size threshold, data accesses
1709 * SHOULD be sent to the metadata server. If an I/O request has a length that
1710 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1711 * server. If both file size and I/O size are provided, the client SHOULD
1712 * reach or exceed both thresholds before sending its read or write
1713 * requests to the data server.
1714 */
1715static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1716 struct inode *ino, int iomode)
1717{
1718 struct nfs4_threshold *t = ctx->mdsthreshold;
1719 struct nfs_inode *nfsi = NFS_I(ino);
1720 loff_t fsize = i_size_read(ino);
1721 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1722
1723 if (t == NULL)
1724 return ret;
1725
1726 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1727 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1728
1729 switch (iomode) {
1730 case IOMODE_READ:
1731 if (t->bm & THRESHOLD_RD) {
1732 dprintk("%s fsize %llu\n", __func__, fsize);
1733 size_set = true;
1734 if (fsize < t->rd_sz)
1735 size = true;
1736 }
1737 if (t->bm & THRESHOLD_RD_IO) {
1738 dprintk("%s nfsi->read_io %llu\n", __func__,
1739 nfsi->read_io);
1740 io_set = true;
1741 if (nfsi->read_io < t->rd_io_sz)
1742 io = true;
1743 }
1744 break;
1745 case IOMODE_RW:
1746 if (t->bm & THRESHOLD_WR) {
1747 dprintk("%s fsize %llu\n", __func__, fsize);
1748 size_set = true;
1749 if (fsize < t->wr_sz)
1750 size = true;
1751 }
1752 if (t->bm & THRESHOLD_WR_IO) {
1753 dprintk("%s nfsi->write_io %llu\n", __func__,
1754 nfsi->write_io);
1755 io_set = true;
1756 if (nfsi->write_io < t->wr_io_sz)
1757 io = true;
1758 }
1759 break;
1760 }
1761 if (size_set && io_set) {
1762 if (size && io)
1763 ret = true;
1764 } else if (size || io)
1765 ret = true;
1766
1767 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1768 return ret;
1769}
1770
d03360aa 1771static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1772{
1773 /*
1774 * send layoutcommit as it can hold up layoutreturn due to lseg
1775 * reference
1776 */
1777 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1778 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1779 nfs_wait_bit_killable,
d03360aa 1780 TASK_KILLABLE);
aa8a45ee
PT		 1781}
1782
411ae722
TM		 1783static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1784{
1785 atomic_inc(&lo->plh_outstanding);
1786}
1787
1788static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1789{
1790 if (atomic_dec_and_test(&lo->plh_outstanding))
1791 wake_up_var(&lo->plh_outstanding);
1792}
1793
d67ae825
TH		 1794static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1795{
1796 unsigned long *bitlock = &lo->plh_flags;
1797
1798 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1799 smp_mb__after_atomic();
1800 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1801}
1802
78746a38
FI		 1803static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1804 struct nfs_server *server)
1805{
1806 if (list_empty(&lo->plh_layouts)) {
1807 struct nfs_client *clp = server->nfs_client;
1808
1809 /* The lo must be on the clp list if there is any
1810 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1811 */
1812 spin_lock(&clp->cl_lock);
1813 if (list_empty(&lo->plh_layouts))
1814 list_add_tail(&lo->plh_layouts, &server->layouts);
1815 spin_unlock(&clp->cl_lock);
1816 }
1817}
1818
e5e94017
BH		 1819/*
1820 * Layout segment is retreived from the server if not cached.
1821 * The appropriate layout segment is referenced and returned to the caller.
1822 */
7c24d948 1823struct pnfs_layout_segment *
e5e94017
BH		 1824pnfs_update_layout(struct inode *ino,
1825 struct nfs_open_context *ctx,
fb3296eb
BH		 1826 loff_t pos,
1827 u64 count,
a75b9df9 1828 enum pnfs_iomode iomode,
c7d73af2 1829 bool strict_iomode,
a75b9df9 1830 gfp_t gfp_flags)
e5e94017 1831{
fb3296eb
BH		 1832 struct pnfs_layout_range arg = {
1833 .iomode = iomode,
1834 .offset = pos,
1835 .length = count,
1836 };
70d2f7b1 1837 unsigned pg_offset;
6382a441
WAA		 1838 struct nfs_server *server = NFS_SERVER(ino);
1839 struct nfs_client *clp = server->nfs_client;
183d9e7b 1840 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1841 struct pnfs_layout_segment *lseg = NULL;
587f03de 1842 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1843 nfs4_stateid stateid;
1844 long timeout = 0;
66b53f32 1845 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1846 bool first;
e5e94017 1847
9a4bf31d 1848 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1849 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1850 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1851 goto out;
9a4bf31d 1852 }
d23d61c8 1853
9a4bf31d 1854 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1855 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1856 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1857 goto out;
9a4bf31d 1858 }
d23d61c8 1859
9bf87482 1860lookup_again:
d03360aa
TM		 1861 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1862 if (IS_ERR(lseg))
1863 goto out;
9bf87482 1864 first = false;
e5e94017 1865 spin_lock(&ino->i_lock);
9fa40758 1866 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1867 if (lo == NULL) {
1868 spin_unlock(&ino->i_lock);
183d9e7b 1869 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1870 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1871 goto out;
1872 }
e5e94017 1873
43f1b3da 1874 /* Do we even need to bother with this? */
a59c30ac 1875 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1876 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1877 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1878 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1879 goto out_unlock;
1880 }
1881
1882 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1883 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1884 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1885 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1886 goto out_unlock;
9a4bf31d 1887 }
e5e94017 1888
411ae722
TM		 1889 /*
1890 * If the layout segment list is empty, but there are outstanding
1891 * layoutget calls, then they might be subject to a layoutrecall.
1892 */
1893 if (list_empty(&lo->plh_segs) &&
1894 atomic_read(&lo->plh_outstanding) != 0) {
1895 spin_unlock(&ino->i_lock);
d03360aa 1896 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
400417b0 1897 !atomic_read(&lo->plh_outstanding)));
58bbeab4 1898 if (IS_ERR(lseg))
411ae722
TM		 1899 goto out_put_layout_hdr;
1900 pnfs_put_layout_hdr(lo);
1901 goto lookup_again;
1902 }
1903
c7d73af2 1904 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1905 if (lseg) {
1906 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1907 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1908 goto out_unlock;
1909 }
1910
183d9e7b
JL		 1911 /*
1912 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1913 * stateid, or it has been invalidated, then we must use the open
1914 * stateid.
1915 */
67a3b721 1916 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
d9aba2b4 1917 int status;
183d9e7b
JL		 1918
1919 /*
1920 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1921 * RFC 5661 Errata 3208.
1922 */
1923 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1924 &lo->plh_flags)) {
1925 spin_unlock(&ino->i_lock);
d03360aa
TM		 1926 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
1927 NFS_LAYOUT_FIRST_LAYOUTGET,
1928 TASK_KILLABLE));
1929 if (IS_ERR(lseg))
1930 goto out_put_layout_hdr;
9bf87482 1931 pnfs_put_layout_hdr(lo);
183d9e7b 1932 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1933 goto lookup_again;
1934 }
183d9e7b
JL		 1935
1936 first = true;
d9aba2b4 1937 status = nfs4_select_rw_stateid(ctx->state,
70d2f7b1 1938 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
d9aba2b4
TM		 1939 NULL, &stateid, NULL);
1940 if (status != 0) {
731c74dd 1941 lseg = ERR_PTR(status);
70d2f7b1
TM		 1942 trace_pnfs_update_layout(ino, pos, count,
1943 iomode, lo, lseg,
1944 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
d9aba2b4
TM		 1945 if (status != -EAGAIN)
1946 goto out_unlock;
1947 spin_unlock(&ino->i_lock);
1948 nfs4_schedule_stateid_recovery(server, ctx->state);
1949 pnfs_clear_first_layoutget(lo);
1950 pnfs_put_layout_hdr(lo);
1951 goto lookup_again;
70d2f7b1 1952 }
9bf87482 1953 } else {
183d9e7b 1954 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 1955 }
568e8c49 1956
aa8a45ee
PT		 1957 /*
1958 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1959 * for LAYOUTRETURN even if first is true.
1960 */
8f70f53a 1961 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 1962 spin_unlock(&ino->i_lock);
1963 dprintk("%s wait for layoutreturn\n", __func__);
d03360aa
TM		 1964 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
1965 if (!IS_ERR(lseg)) {
d67ae825
TH		 1966 if (first)
1967 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 1968 pnfs_put_layout_hdr(lo);
1969 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 1970 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1971 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 1972 goto lookup_again;
1973 }
183d9e7b 1974 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1975 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 1976 goto out_put_layout_hdr;
1977 }
1978
9a4bf31d 1979 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 1980 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1981 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 1982 goto out_unlock;
9a4bf31d 1983 }
411ae722 1984 nfs_layoutget_begin(lo);
f49f9baa 1985 spin_unlock(&ino->i_lock);
30005121 1986
78746a38 1987 _add_to_server_list(lo, server);
e5e94017 1988
09cbfeaf 1989 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 1990 if (pg_offset) {
1991 arg.offset -= pg_offset;
1992 arg.length += pg_offset;
1993 }
7c24d948 1994 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 1995 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 1996
5e36e2a9 1997 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 1998 if (!lgp) {
1999 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2000 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 2001 nfs_layoutget_end(lo);
587f03de
FI		 2002 goto out_put_layout_hdr;
2003 }
2004
dacb452d 2005 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 2006 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2007 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 2008 nfs_layoutget_end(lo);
83026d80 2009 if (IS_ERR(lseg)) {
183d9e7b 2010 switch(PTR_ERR(lseg)) {
e85d7ee4 2011 case -EBUSY:
183d9e7b
JL		 2012 if (time_after(jiffies, giveup))
2013 lseg = NULL;
66b53f32
TM		 2014 break;
2015 case -ERECALLCONFLICT:
183d9e7b 2016 case -EAGAIN:
56b38a1f 2017 break;
183d9e7b
JL		 2018 default:
2019 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2020 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2021 lseg = NULL;
2022 }
56b38a1f
TM		 2023 goto out_put_layout_hdr;
2024 }
2025 if (lseg) {
2026 if (first)
2027 pnfs_clear_first_layoutget(lo);
2028 trace_pnfs_update_layout(ino, pos, count,
2029 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2030 pnfs_put_layout_hdr(lo);
2031 goto lookup_again;
83026d80
JL		 2032 }
2033 } else {
2034 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2035 }
2036
830ffb56 2037out_put_layout_hdr:
d67ae825
TH		 2038 if (first)
2039 pnfs_clear_first_layoutget(lo);
d5b9216f
TM		 2040 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2041 PNFS_UPDATE_LAYOUT_EXIT);
70c3bd2b 2042 pnfs_put_layout_hdr(lo);
e5e94017 2043out:
f86bbcf8
TM		 2044 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2045 "(%s, offset: %llu, length: %llu)\n",
2046 __func__, ino->i_sb->s_id,
2047 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2048 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2049 iomode==IOMODE_RW ? "read/write" : "read-only",
2050 (unsigned long long)pos,
2051 (unsigned long long)count);
e5e94017
BH		 2052 return lseg;
2053out_unlock:
2054 spin_unlock(&ino->i_lock);
830ffb56 2055 goto out_put_layout_hdr;
e5e94017 2056}
7c24d948 2057EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2058
540d9864
TM		 2059static bool
2060pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2061{
2062 switch (range->iomode) {
2063 case IOMODE_READ:
2064 case IOMODE_RW:
2065 break;
2066 default:
2067 return false;
2068 }
2069 if (range->offset == NFS4_MAX_UINT64)
2070 return false;
2071 if (range->length == 0)
2072 return false;
2073 if (range->length != NFS4_MAX_UINT64 &&
2074 range->length > NFS4_MAX_UINT64 - range->offset)
2075 return false;
2076 return true;
2077}
2078
78746a38
FI		 2079static struct pnfs_layout_hdr *
2080_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2081{
2082 struct pnfs_layout_hdr *lo;
2083
2084 spin_lock(&ino->i_lock);
2085 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2086 if (!lo)
2087 goto out_unlock;
2088 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2089 goto out_unlock;
2090 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2091 goto out_unlock;
2092 if (pnfs_layoutgets_blocked(lo))
2093 goto out_unlock;
2094 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2095 goto out_unlock;
411ae722 2096 nfs_layoutget_begin(lo);
78746a38
FI		 2097 spin_unlock(&ino->i_lock);
2098 _add_to_server_list(lo, NFS_SERVER(ino));
2099 return lo;
2100
2101out_unlock:
2102 spin_unlock(&ino->i_lock);
2103 pnfs_put_layout_hdr(lo);
2104 return NULL;
2105}
2106
2409a976
FI		 2107extern const nfs4_stateid current_stateid;
2108
2109static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2110 struct nfs_open_context *ctx)
2111{
78746a38
FI		 2112 struct inode *ino = data->dentry->d_inode;
2113 struct pnfs_layout_range rng = {
2114 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2115 IOMODE_RW: IOMODE_READ,
2116 .offset = 0,
2117 .length = NFS4_MAX_UINT64,
2118 };
2119 struct nfs4_layoutget *lgp;
2120 struct pnfs_layout_hdr *lo;
2121
64294b08
TM		 2122 /* Heuristic: don't send layoutget if we have cached data */
2123 if (rng.iomode == IOMODE_READ &&
2124 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2125 return;
2126
78746a38
FI		 2127 lo = _pnfs_grab_empty_layout(ino, ctx);
2128 if (!lo)
2129 return;
2130 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2131 &rng, GFP_KERNEL);
2132 if (!lgp) {
2133 pnfs_clear_first_layoutget(lo);
2134 pnfs_put_layout_hdr(lo);
2135 return;
2136 }
2137 data->lgp = lgp;
2138 data->o_arg.lg_args = &lgp->args;
2139 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2140}
2141
2142static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2143 struct nfs_open_context *ctx)
2144{
2145 struct pnfs_layout_range rng = {
2146 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2147 IOMODE_RW: IOMODE_READ,
2148 .offset = 0,
2149 .length = NFS4_MAX_UINT64,
2150 };
2151 struct nfs4_layoutget *lgp;
2152
2153 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2154 &rng, GFP_KERNEL);
2155 if (!lgp)
2156 return;
2157 data->lgp = lgp;
2158 data->o_arg.lg_args = &lgp->args;
2159 data->o_res.lg_res = &lgp->res;
2160}
2161
2162void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2163 struct nfs_open_context *ctx)
2164{
2165 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2166
2167 if (!(pnfs_enabled_sb(server) &&
2168 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2169 return;
2170 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2171 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2172 return;
2409a976
FI		 2173 if (data->state)
2174 _lgopen_prepare_attached(data, ctx);
2175 else
2176 _lgopen_prepare_floating(data, ctx);
2177}
2178
2179void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2180 struct nfs_open_context *ctx)
2181{
2182 struct pnfs_layout_hdr *lo;
2183 struct pnfs_layout_segment *lseg;
c49b5209 2184 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2185 u32 iomode;
2186
6e01260c 2187 if (!lgp)
2409a976 2188 return;
6e01260c
FI		 2189 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2190 if (lgp->res.status) {
2191 switch (lgp->res.status) {
6e01260c 2192 default:
8dc96566
TM		 2193 break;
2194 /*
2195 * Halt lgopen attempts if the server doesn't recognise
2196 * the "current stateid" value, the layout type, or the
2197 * layoutget operation as being valid.
2198 * Also if it complains about too many ops in the compound
2199 * or of the request/reply being too big.
2200 */
2201 case -NFS4ERR_BAD_STATEID:
2202 case -NFS4ERR_NOTSUPP:
2203 case -NFS4ERR_REP_TOO_BIG:
2204 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2205 case -NFS4ERR_REQ_TOO_BIG:
2206 case -NFS4ERR_TOO_MANY_OPS:
2207 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2208 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2209 }
2210 return;
2211 }
2409a976 2212 if (!lgp->args.inode) {
78746a38
FI		 2213 lo = _pnfs_grab_empty_layout(ino, ctx);
2214 if (!lo)
2215 return;
2409a976
FI		 2216 lgp->args.inode = ino;
2217 } else
2218 lo = NFS_I(lgp->args.inode)->layout;
2409a976
FI		 2219
2220 lseg = pnfs_layout_process(lgp);
32f1c28f 2221 if (!IS_ERR(lseg)) {
2409a976
FI		 2222 iomode = lgp->args.range.iomode;
2223 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2224 pnfs_put_lseg(lseg);
2225 }
30ae2412
FI		 2226}
2227
2228void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2229{
2230 if (lgp != NULL) {
2231 struct inode *inode = lgp->args.inode;
2232 if (inode) {
2233 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
30ae2412 2234 pnfs_clear_first_layoutget(lo);
411ae722 2235 nfs_layoutget_end(lo);
30ae2412
FI		 2236 }
2237 pnfs_layoutget_free(lgp);
2238 }
2409a976
FI		 2239}
2240
a0b0a6e3 2241struct pnfs_layout_segment *
b1f69b75
AA		 2242pnfs_layout_process(struct nfs4_layoutget *lgp)
2243{
2244 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2245 struct nfs4_layoutget_res *res = &lgp->res;
2246 struct pnfs_layout_segment *lseg;
b7edfaa1 2247 struct inode *ino = lo->plh_inode;
78096cca 2248 LIST_HEAD(free_me);
540d9864
TM		 2249
2250 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2251 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2252
2253 /* Inject layout blob into I/O device driver */
a75b9df9 2254 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2255 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2256 if (!lseg)
1b3c6d07
JL		 2257 lseg = ERR_PTR(-ENOMEM);
2258
2259 dprintk("%s: Could not allocate layout: error %ld\n",
2260 __func__, PTR_ERR(lseg));
2261 return lseg;
b1f69b75
AA		 2262 }
2263
119cef97 2264 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2265
b1f69b75 2266 spin_lock(&ino->i_lock);
e1c06f80 2267 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2268 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2269 goto out_forget;
43f1b3da 2270 }
038d6493 2271
9888d837
TM		 2272 if (!pnfs_layout_is_valid(lo)) {
2273 /* We have a completely new layout */
2274 pnfs_set_layout_stateid(lo, &res->stateid, true);
2275 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2276 /* existing state ID, make sure the sequence number matches. */
2277 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2278 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2279 goto out_forget;
362f7474
CH		 2280 }
2281 pnfs_set_layout_stateid(lo, &res->stateid, false);
2282 } else {
2283 /*
2284 * We got an entirely new state ID. Mark all segments for the
9888d837 2285 * inode invalid, and retry the layoutget
362f7474 2286 */
d9b61708 2287 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 2288 goto out_forget;
362f7474 2289 }
038d6493 2290
9369a431 2291 pnfs_get_lseg(lseg);
03772d2f 2292 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2293
b1f69b75 2294
3976143b 2295 if (res->return_on_close)
f7e8917a 2296 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2297
b1f69b75 2298 spin_unlock(&ino->i_lock);
78096cca 2299 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2300 return lseg;
43f1b3da 2301
1b3c6d07 2302out_forget:
43f1b3da
FI		 2303 spin_unlock(&ino->i_lock);
2304 lseg->pls_layout = lo;
2305 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2306 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2307}
2308
e0b7d420
TM		 2309static int
2310mark_lseg_invalid_or_return(struct pnfs_layout_segment *lseg,
2311 struct list_head *tmp_list)
2312{
2313 if (!mark_lseg_invalid(lseg, tmp_list))
2314 return 0;
2315 pnfs_cache_lseg_for_layoutreturn(lseg->pls_layout, lseg);
2316 return 1;
2317}
2318
2f215968
TM		 2319/**
2320 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2321 * @lo: pointer to layout header
2322 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2323 * @return_range: describe layout segment ranges to be returned
e0b7d420 2324 * @seq: stateid seqid to match
2f215968
TM		 2325 *
2326 * This function is mainly intended for use by layoutrecall. It attempts
2327 * to free the layout segment immediately, or else to mark it for return
2328 * as soon as its reference count drops to zero.
e0b7d420
TM		 2329 *
2330 * Returns
2331 * - 0: a layoutreturn needs to be scheduled.
2332 * - EBUSY: there are layout segment that are still in use.
2333 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2334 */
10335556 2335int
016256df
PT		 2336pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2337 struct list_head *tmp_list,
6d597e17
JL		 2338 const struct pnfs_layout_range *return_range,
2339 u32 seq)
016256df
PT		 2340{
2341 struct pnfs_layout_segment *lseg, *next;
10335556 2342 int remaining = 0;
016256df
PT		 2343
2344 dprintk("%s:Begin lo %p\n", __func__, lo);
2345
fc7ff367 2346 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2347
2348 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2349 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2350 dprintk("%s: marking lseg %p iomode %d "
2351 "offset %llu length %llu\n", __func__,
2352 lseg, lseg->pls_range.iomode,
2353 lseg->pls_range.offset,
2354 lseg->pls_range.length);
e0b7d420 2355 if (mark_lseg_invalid_or_return(lseg, tmp_list))
2f215968
TM		 2356 continue;
2357 remaining++;
016256df 2358 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2359 }
6d597e17 2360
e0b7d420 2361 if (remaining) {
6d597e17 2362 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2363 return -EBUSY;
2364 }
6d597e17 2365
e0b7d420
TM		 2366 if (!list_empty(&lo->plh_return_segs)) {
2367 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2368 return 0;
2369 }
2370
2371 return -ENOENT;
016256df
PT		 2372}
2373
2374void pnfs_error_mark_layout_for_return(struct inode *inode,
2375 struct pnfs_layout_segment *lseg)
2376{
2377 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2378 struct pnfs_layout_range range = {
2379 .iomode = lseg->pls_range.iomode,
2380 .offset = 0,
2381 .length = NFS4_MAX_UINT64,
2382 };
10335556 2383 bool return_now = false;
016256df
PT		 2384
2385 spin_lock(&inode->i_lock);
bdebfccd
TM		 2386 if (!pnfs_layout_is_valid(lo)) {
2387 spin_unlock(&inode->i_lock);
2388 return;
2389 }
2d6cf5ab 2390 pnfs_set_plh_return_info(lo, range.iomode, 0);
016256df
PT		 2391 /*
2392 * mark all matching lsegs so that we are sure to have no live
2393 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2394 * for how it works.
2395 */
e0b7d420 2396 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
10335556 2397 nfs4_stateid stateid;
e5fd1904 2398 enum pnfs_iomode iomode;
10335556 2399
e5fd1904 2400 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2401 spin_unlock(&inode->i_lock);
2402 if (return_now)
2403 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2404 } else {
2405 spin_unlock(&inode->i_lock);
2406 nfs_commit_inode(inode, 0);
2407 }
016256df
PT		 2408}
2409EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2410
b3230e80
TM		 2411void
2412pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2413{
2414 if (pgio->pg_lseg == NULL ||
2415 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2416 return;
2417 pnfs_put_lseg(pgio->pg_lseg);
2418 pgio->pg_lseg = NULL;
2419}
2420EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2421
08cb5b0f
BC		 2422/*
2423 * Check for any intersection between the request and the pgio->pg_lseg,
2424 * and if none, put this pgio->pg_lseg away.
2425 */
2426static void
2427pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2428{
2429 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2430 pnfs_put_lseg(pgio->pg_lseg);
2431 pgio->pg_lseg = NULL;
2432 }
2433}
2434
d8007d4d
TM		 2435void
2436pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2437{
1fd937bd
PT		 2438 u64 rd_size = req->wb_bytes;
2439
b3230e80 2440 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2441 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2442 if (pgio->pg_lseg == NULL) {
2443 if (pgio->pg_dreq == NULL)
2444 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2445 else
2446 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2447
2448 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2449 nfs_req_openctx(req),
cb5d04bc
PT		 2450 req_offset(req),
2451 rd_size,
2452 IOMODE_READ,
c7d73af2 2453 false,
cb5d04bc 2454 GFP_KERNEL);
d600ad1f
PT		 2455 if (IS_ERR(pgio->pg_lseg)) {
2456 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2457 pgio->pg_lseg = NULL;
2458 return;
2459 }
cb5d04bc 2460 }
e885de1a
TM		 2461 /* If no lseg, fall back to read through mds */
2462 if (pgio->pg_lseg == NULL)
1f945357 2463 nfs_pageio_reset_read_mds(pgio);
e885de1a 2464
d8007d4d
TM		 2465}
2466EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2467
2468void
6296556f
PT		 2469pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2470 struct nfs_page *req, u64 wb_size)
d8007d4d 2471{
b3230e80 2472 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2473 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2474 if (pgio->pg_lseg == NULL) {
cb5d04bc 2475 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2476 nfs_req_openctx(req),
cb5d04bc
PT		 2477 req_offset(req),
2478 wb_size,
2479 IOMODE_RW,
c7d73af2 2480 false,
2b17d725 2481 GFP_KERNEL);
d600ad1f
PT		 2482 if (IS_ERR(pgio->pg_lseg)) {
2483 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2484 pgio->pg_lseg = NULL;
2485 return;
2486 }
2487 }
e885de1a
TM		 2488 /* If no lseg, fall back to write through mds */
2489 if (pgio->pg_lseg == NULL)
1f945357 2490 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2491}
2492EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2493
180bb5ec
WAA		 2494void
2495pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2496{
2497 if (desc->pg_lseg) {
2498 pnfs_put_lseg(desc->pg_lseg);
2499 desc->pg_lseg = NULL;
2500 }
2501}
2502EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2503
b4fdac1a
WAA		 2504/*
2505 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2506 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2507 */
2508size_t
a7d42ddb
WAA		 2509pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2510 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2511{
0f9c429e 2512 unsigned int size;
c5e20cb7 2513 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2514
2515 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2516 if (!size)
2517 return 0;
94ad1c80 2518
19982ba8 2519 /*
c5e20cb7
WAA		 2520 * 'size' contains the number of bytes left in the current page (up
2521 * to the original size asked for in @req->wb_bytes).
2522 *
2523 * Calculate how many bytes are left in the layout segment
2524 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2525 *
2526 * Please also note that 'end_offset' is actually the offset of the
2527 * first byte that lies outside the pnfs_layout_range. FIXME?
2528 *
2529 */
19b54848 2530 if (pgio->pg_lseg) {
17822b20 2531 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2532 pgio->pg_lseg->pls_range.length);
2533 req_start = req_offset(req);
08cb5b0f 2534
c5e20cb7 2535 /* start of request is past the last byte of this segment */
08cb5b0f 2536 if (req_start >= seg_end)
19b54848 2537 return 0;
c5e20cb7
WAA		 2538
2539 /* adjust 'size' iff there are fewer bytes left in the
2540 * segment than what nfs_generic_pg_test returned */
2541 seg_left = seg_end - req_start;
2542 if (seg_left < size)
2543 size = (unsigned int)seg_left;
19b54848 2544 }
0f9c429e 2545
19b54848 2546 return size;
94ad1c80 2547}
89a58e32 2548EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2549
53113ad3 2550int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2551{
2552 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2553
2554 /* Resend all requests through the MDS */
53113ad3
WAA		 2555 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2556 hdr->completion_ops);
c7070113 2557 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2558 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2559}
e7dd79af 2560EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2561
d45f60c6 2562static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2563{
cd841605
FI		 2564
2565 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2566 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2567 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2568 pnfs_return_layout(hdr->inode);
1acbbb4e 2569 }
6c75dc0d 2570 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2571 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2572}
2573
d20581aa
BH		 2574/*
2575 * Called by non rpc-based layout drivers
2576 */
d45f60c6 2577void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2578{
f8417b48 2579 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2580 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2581 hdr->mds_offset + hdr->res.count);
d45f60c6 2582 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2583 }
2584 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2585 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2586 pnfs_ld_handle_write_error(hdr);
2587 hdr->mds_ops->rpc_release(hdr);
44b83799 2588}
d20581aa 2589EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2590
dce81290
TM		 2591static void
2592pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2593 struct nfs_pgio_header *hdr)
dce81290 2594{
48d635f1 2595 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2596
6c75dc0d 2597 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2598 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2599 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2600 mirror->pg_recoalesce = 1;
6c75dc0d 2601 }
ba4a76f7 2602 hdr->completion_ops->completion(hdr);
dce81290
TM		 2603}
2604
2605static enum pnfs_try_status
d45f60c6 2606pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2607 const struct rpc_call_ops *call_ops,
2608 struct pnfs_layout_segment *lseg,
2609 int how)
0382b744 2610{
cd841605 2611 struct inode *inode = hdr->inode;
0382b744
AA		 2612 enum pnfs_try_status trypnfs;
2613 struct nfs_server *nfss = NFS_SERVER(inode);
2614
cd841605 2615 hdr->mds_ops = call_ops;
0382b744
AA		 2616
2617 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2618 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2619 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2620 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2621 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2622 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2623 return trypnfs;
2624}
2625
dce81290 2626static void
7f714720
WAA		 2627pnfs_do_write(struct nfs_pageio_descriptor *desc,
2628 struct nfs_pgio_header *hdr, int how)
dce81290 2629{
dce81290
TM		 2630 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2631 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2632 enum pnfs_try_status trypnfs;
dce81290 2633
d45f60c6 2634 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2635 switch (trypnfs) {
2636 case PNFS_NOT_ATTEMPTED:
d45f60c6 2637 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2638 case PNFS_ATTEMPTED:
2639 break;
2640 case PNFS_TRY_AGAIN:
2641 /* cleanup hdr and prepare to redo pnfs */
2642 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2643 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2644 list_splice_init(&hdr->pages, &mirror->pg_list);
2645 mirror->pg_recoalesce = 1;
2646 }
2647 hdr->mds_ops->rpc_release(hdr);
2648 }
dce81290
TM		 2649}
2650
6c75dc0d
FI		 2651static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2652{
9369a431 2653 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2654 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2655}
2656
dce81290
TM		 2657int
2658pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2659{
6c75dc0d 2660 struct nfs_pgio_header *hdr;
dce81290
TM		 2661 int ret;
2662
1e7f3a48
WAA		 2663 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2664 if (!hdr) {
2bff2288
PT		 2665 desc->pg_error = -ENOMEM;
2666 return desc->pg_error;
dce81290 2667 }
6c75dc0d 2668 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2669
9369a431 2670 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2671 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2672 if (!ret)
7f714720 2673 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2674
6c75dc0d 2675 return ret;
dce81290
TM		 2676}
2677EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2678
53113ad3 2679int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2680{
2681 struct nfs_pageio_descriptor pgio;
2682
1acbbb4e 2683 /* Resend all requests through the MDS */
53113ad3
WAA		 2684 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2685 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2686}
e7dd79af 2687EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2688
d45f60c6 2689static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2690{
cd841605
FI		 2691 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2692 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2693 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2694 pnfs_return_layout(hdr->inode);
1acbbb4e 2695 }
4db6e0b7 2696 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2697 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2698}
2699
d20581aa
BH		 2700/*
2701 * Called by non rpc-based layout drivers
2702 */
d45f60c6 2703void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2704{
bfc505de 2705 if (likely(!hdr->pnfs_error))
d45f60c6 2706 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2707 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2708 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2709 pnfs_ld_handle_read_error(hdr);
2710 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2711}
2712EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2713
493292dd
TM		 2714static void
2715pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2716 struct nfs_pgio_header *hdr)
493292dd 2717{
48d635f1 2718 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2719
4db6e0b7 2720 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2721 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2722 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2723 mirror->pg_recoalesce = 1;
4db6e0b7 2724 }
ba4a76f7 2725 hdr->completion_ops->completion(hdr);
493292dd
TM		 2726}
2727
64419a9b
AA		 2728/*
2729 * Call the appropriate parallel I/O subsystem read function.
2730 */
493292dd 2731static enum pnfs_try_status
d45f60c6 2732pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2733 const struct rpc_call_ops *call_ops,
2734 struct pnfs_layout_segment *lseg)
64419a9b 2735{
cd841605 2736 struct inode *inode = hdr->inode;
64419a9b
AA		 2737 struct nfs_server *nfss = NFS_SERVER(inode);
2738 enum pnfs_try_status trypnfs;
2739
cd841605 2740 hdr->mds_ops = call_ops;
64419a9b
AA		 2741
2742 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2743 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2744
d45f60c6 2745 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2746 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2747 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2748 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2749 return trypnfs;
2750}
863a3c6c 2751
ceb11e13 2752/* Resend all requests through pnfs. */
1b1bc66b 2753void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2754{
2755 struct nfs_pageio_descriptor pgio;
2756
1b1bc66b 2757 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2758 /* Prevent deadlocks with layoutreturn! */
2759 pnfs_put_lseg(hdr->lseg);
2760 hdr->lseg = NULL;
2761
1b1bc66b
WAA		 2762 nfs_pageio_init_read(&pgio, hdr->inode, false,
2763 hdr->completion_ops);
2764 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2765 }
ceb11e13
PT		 2766}
2767EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2768
493292dd 2769static void
7f714720 2770pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2771{
493292dd
TM		 2772 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2773 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2774 enum pnfs_try_status trypnfs;
493292dd 2775
d45f60c6 2776 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2777 switch (trypnfs) {
2778 case PNFS_NOT_ATTEMPTED:
d45f60c6 2779 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2780 case PNFS_ATTEMPTED:
2781 break;
2782 case PNFS_TRY_AGAIN:
2783 /* cleanup hdr and prepare to redo pnfs */
2784 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2785 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2786 list_splice_init(&hdr->pages, &mirror->pg_list);
2787 mirror->pg_recoalesce = 1;
2788 }
2789 hdr->mds_ops->rpc_release(hdr);
2790 }
493292dd
TM		 2791}
2792
4db6e0b7
FI		 2793static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2794{
9369a431 2795 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2796 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2797}
2798
493292dd
TM		 2799int
2800pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2801{
4db6e0b7 2802 struct nfs_pgio_header *hdr;
493292dd
TM		 2803 int ret;
2804
1e7f3a48
WAA		 2805 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2806 if (!hdr) {
2bff2288
PT		 2807 desc->pg_error = -ENOMEM;
2808 return desc->pg_error;
493292dd 2809 }
4db6e0b7 2810 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2811 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2812 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2813 if (!ret)
7f714720 2814 pnfs_do_read(desc, hdr);
4db6e0b7 2815 return ret;
493292dd
TM		 2816}
2817EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2818
71244d9b
TM		 2819static void pnfs_clear_layoutcommitting(struct inode *inode)
2820{
2821 unsigned long *bitlock = &NFS_I(inode)->flags;
2822
2823 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2824 smp_mb__after_atomic();
71244d9b
TM		 2825 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2826}
2827
863a3c6c 2828/*
a9bae566 2829 * There can be multiple RW segments.
863a3c6c 2830 */
a9bae566 2831static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2832{
a9bae566 2833 struct pnfs_layout_segment *lseg;
863a3c6c 2834
a9bae566
PT		 2835 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2836 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2837 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2838 list_add(&lseg->pls_lc_list, listp);
2839 }
863a3c6c
AA		 2840}
2841
a073dbff
TM		 2842static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2843{
2844 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2845
2846 /* Matched by references in pnfs_set_layoutcommit */
2847 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2848 list_del_init(&lseg->pls_lc_list);
2849 pnfs_put_lseg(lseg);
2850 }
2851
71244d9b 2852 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2853}
2854
1b0ae068
PT		 2855void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2856{
b9e028fd 2857 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2858}
2859EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2860
863a3c6c 2861void
67af7611
TM		 2862pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2863 loff_t end_pos)
863a3c6c 2864{
cd841605 2865 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2866 bool mark_as_dirty = false;
863a3c6c 2867
cd841605 2868 spin_lock(&inode->i_lock);
863a3c6c 2869 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2870 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2871 mark_as_dirty = true;
863a3c6c 2872 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2873 __func__, inode->i_ino);
29559b11
TM		 2874 } else if (end_pos > nfsi->layout->plh_lwb)
2875 nfsi->layout->plh_lwb = end_pos;
67af7611 2876 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2877 /* references matched in nfs4_layoutcommit_release */
67af7611 2878 pnfs_get_lseg(lseg);
a9bae566 2879 }
cd841605 2880 spin_unlock(&inode->i_lock);
acff5880 2881 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2882 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2883
2884 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2885 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2886 if (mark_as_dirty)
cd841605 2887 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2888}
2889EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2890
db29c089
AA		 2891void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2892{
2893 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2894
2895 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2896 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2897 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2898}
2899
de4b15c7
AA		 2900/*
2901 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2902 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2903 * data to disk to allow the server to recover the data if it crashes.
2904 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2905 * is off, and a COMMIT is sent to a data server, or
2906 * if WRITEs to a data server return NFS_DATA_SYNC.
2907 */
863a3c6c 2908int
ef311537 2909pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2910{
5f919c9f 2911 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2912 struct nfs4_layoutcommit_data *data;
2913 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2914 loff_t end_pos;
71244d9b 2915 int status;
863a3c6c 2916
71244d9b 2917 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2918 return 0;
2919
71244d9b 2920 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2921
71244d9b 2922 status = -EAGAIN;
92407e75 2923 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2924 if (!sync)
2925 goto out;
74316201 2926 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2927 NFS_INO_LAYOUTCOMMITTING,
2928 nfs_wait_bit_killable,
2929 TASK_KILLABLE);
92407e75 2930 if (status)
71244d9b 2931 goto out;
92407e75
PT		 2932 }
2933
71244d9b
TM		 2934 status = -ENOMEM;
2935 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2936 data = kzalloc(sizeof(*data), GFP_NOFS);
2937 if (!data)
2938 goto clear_layoutcommitting;
2939
2940 status = 0;
de4b15c7 2941 spin_lock(&inode->i_lock);
71244d9b
TM		 2942 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2943 goto out_unlock;
a9bae566 2944
71244d9b 2945 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 2946 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 2947
acff5880 2948 end_pos = nfsi->layout->plh_lwb;
863a3c6c 2949
f597c537 2950 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 2951 spin_unlock(&inode->i_lock);
2952
2953 data->args.inode = inode;
a52458b4 2954 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 2955 nfs_fattr_init(&data->fattr);
2956 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2957 data->res.fattr = &data->fattr;
2e18d4d8
TM		 2958 if (end_pos != 0)
2959 data->args.lastbytewritten = end_pos - 1;
2960 else
2961 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 2962 data->res.server = NFS_SERVER(inode);
2963
5f919c9f
CH		 2964 if (ld->prepare_layoutcommit) {
2965 status = ld->prepare_layoutcommit(&data->args);
2966 if (status) {
a52458b4 2967 put_cred(data->cred);
5f919c9f 2968 spin_lock(&inode->i_lock);
29559b11
TM		 2969 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2970 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 2971 nfsi->layout->plh_lwb = end_pos;
3471648a 2972 goto out_unlock;
5f919c9f
CH		 2973 }
2974 }
2975
2976
863a3c6c
AA		 2977 status = nfs4_proc_layoutcommit(data, sync);
2978out:
92407e75
PT		 2979 if (status)
2980 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2981 dprintk("<-- %s status %d\n", __func__, status);
2982 return status;
71244d9b
TM		 2983out_unlock:
2984 spin_unlock(&inode->i_lock);
92407e75 2985 kfree(data);
71244d9b
TM		 2986clear_layoutcommitting:
2987 pnfs_clear_layoutcommitting(inode);
92407e75 2988 goto out;
863a3c6c 2989}
72cff449 2990EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 2991
5bb89b47
TM		 2992int
2993pnfs_generic_sync(struct inode *inode, bool datasync)
2994{
2995 return pnfs_layoutcommit_inode(inode, true);
2996}
2997EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2998
82be417a
AA		 2999struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3000{
3001 struct nfs4_threshold *thp;
3002
3003 thp = kzalloc(sizeof(*thp), GFP_NOFS);
3004 if (!thp) {
3005 dprintk("%s mdsthreshold allocation failed\n", __func__);
3006 return NULL;
3007 }
3008 return thp;
3009}
8733408d 3010
865a7ecb 3011#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 3012int
c8ad8894 3013pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 3014{
3015 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3016 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 3017 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 3018 struct nfs42_layoutstat_data *data;
3019 struct pnfs_layout_hdr *hdr;
3020 int status = 0;
3021
3022 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3023 goto out;
3024
6c5a0d89
TM		 3025 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3026 goto out;
3027
1bfe3b25
PT		 3028 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3029 goto out;
3030
8733408d
PT		 3031 spin_lock(&inode->i_lock);
3032 if (!NFS_I(inode)->layout) {
3033 spin_unlock(&inode->i_lock);
f538d0ba 3034 goto out_clear_layoutstats;
8733408d
PT		 3035 }
3036 hdr = NFS_I(inode)->layout;
3037 pnfs_get_layout_hdr(hdr);
3038 spin_unlock(&inode->i_lock);
3039
c8ad8894 3040 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3041 if (!data) {
3042 status = -ENOMEM;
3043 goto out_put;
3044 }
3045
3046 data->args.fh = NFS_FH(inode);
3047 data->args.inode = inode;
8733408d
PT		 3048 status = ld->prepare_layoutstats(&data->args);
3049 if (status)
3050 goto out_free;
3051
3052 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3053
3054out:
3055 dprintk("%s returns %d\n", __func__, status);
3056 return status;
3057
3058out_free:
3059 kfree(data);
3060out_put:
3061 pnfs_put_layout_hdr(hdr);
f538d0ba 3062out_clear_layoutstats:
1bfe3b25
PT		 3063 smp_mb__before_atomic();
3064 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3065 smp_mb__after_atomic();
8733408d
PT		 3066 goto out;
3067}
3068EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3069#endif
bbf58bf3
TM		 3070
3071unsigned int layoutstats_timer;
3072module_param(layoutstats_timer, uint, 0644);
3073EXPORT_SYMBOL_GPL(layoutstats_timer);
Empty description
This page took 0.991156 seconds and 4 git commands to generate.