projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6-block.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
95void
96unset_pnfs_layoutdriver(struct nfs_server *nfss)
97{
738fd0f3
BH		 98 if (nfss->pnfs_curr_ld) {
99 if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 104 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 105 }
85e174ba
RL		 106 nfss->pnfs_curr_ld = NULL;
107}
108
ca440c38
JL		 109/*
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
112 *
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
115 */
116static const u32 ld_prefs[] = {
117 LAYOUT_SCSI,
118 LAYOUT_BLOCK_VOLUME,
119 LAYOUT_OSD2_OBJECTS,
120 LAYOUT_FLEX_FILES,
121 LAYOUT_NFSV4_1_FILES,
122 0
123};
124
125static int
126ld_cmp(const void *e1, const void *e2)
127{
128 u32 ld1 = *((u32 *)e1);
129 u32 ld2 = *((u32 *)e2);
130 int i;
131
132 for (i = 0; ld_prefs[i] != 0; i++) {
133 if (ld1 == ld_prefs[i])
134 return -1;
135
136 if (ld2 == ld_prefs[i])
137 return 1;
138 }
139 return 0;
140}
141
85e174ba
RL		 142/*
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
145 *
3132e49e 146 * @ids array of layout types supported by MDS.
85e174ba
RL		 147 */
148void
738fd0f3 149set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 150 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 151{
152 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 153 u32 id;
ca440c38 154 int i;
85e174ba 155
19274716
AS		 156 if (fsinfo->nlayouttypes == 0)
157 goto out_no_driver;
85e174ba
RL		 158 if (!(server->nfs_client->cl_exchange_flags &
159 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 160 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 162 goto out_no_driver;
163 }
3132e49e 164
ca440c38
JL		 165 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 167
ca440c38
JL		 168 for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 id = fsinfo->layouttype[i];
85e174ba 170 ld_type = find_pnfs_driver(id);
ca440c38
JL		 171 if (!ld_type) {
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 id);
174 ld_type = find_pnfs_driver(id);
175 }
176 if (ld_type)
177 break;
85e174ba 178 }
3132e49e
JL		 179
180 if (!ld_type) {
ca440c38 181 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 182 goto out_no_driver;
183 }
184
85e174ba 185 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 186 if (ld_type->set_layoutdriver
187 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 188 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__, id);
738fd0f3
BH		 190 module_put(ld_type->owner);
191 goto out_no_driver;
192 }
2a4c8994
TM		 193 /* Bump the MDS count */
194 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 195
85e174ba
RL		 196 dprintk("%s: pNFS module for %u set\n", __func__, id);
197 return;
198
199out_no_driver:
200 dprintk("%s: Using NFSv4 I/O\n", __func__);
201 server->pnfs_curr_ld = NULL;
202}
02c35fca
FI		 203
204int
205pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206{
207 int status = -EINVAL;
208 struct pnfs_layoutdriver_type *tmp;
209
210 if (ld_type->id == 0) {
a030889a 211 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 212 return status;
213 }
b1f69b75 214 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 215 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 216 "alloc_lseg and free_lseg.\n", __func__);
217 return status;
218 }
02c35fca
FI		 219
220 spin_lock(&pnfs_spinlock);
221 tmp = find_pnfs_driver_locked(ld_type->id);
222 if (!tmp) {
223 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 status = 0;
225 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 ld_type->name);
227 } else {
a030889a 228 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 229 __func__, ld_type->id);
230 }
231 spin_unlock(&pnfs_spinlock);
232
233 return status;
234}
235EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236
237void
238pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239{
240 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 spin_lock(&pnfs_spinlock);
242 list_del(&ld_type->pnfs_tblid);
243 spin_unlock(&pnfs_spinlock);
244}
245EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 246
b1f69b75
AA		 247/*
248 * pNFS client layout cache
249 */
250
cc6e5340 251/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 252void
70c3bd2b 253pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 254{
2b28a7be 255 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 256}
257
636fb9c8
BH		 258static struct pnfs_layout_hdr *
259pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260{
261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 262 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 263}
264
265static void
266pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267{
9c626381
TM		 268 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270
271 if (!list_empty(&lo->plh_layouts)) {
272 struct nfs_client *clp = server->nfs_client;
273
274 spin_lock(&clp->cl_lock);
275 list_del_init(&lo->plh_layouts);
276 spin_unlock(&clp->cl_lock);
277 }
a52458b4 278 put_cred(lo->plh_lc_cred);
57934278 279 return ld->free_layout_hdr(lo);
636fb9c8
BH		 280}
281
e5e94017 282static void
6622c3ea 283pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 284{
bb346f63 285 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 286 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 287 nfsi->layout = NULL;
288 /* Reset MDS Threshold I/O counters */
289 nfsi->write_io = 0;
290 nfsi->read_io = 0;
e5e94017
BH		 291}
292
b1f69b75 293void
70c3bd2b 294pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 295{
9c6376eb 296 struct inode *inode;
cc6e5340 297
9c6376eb
TM		 298 if (!lo)
299 return;
300 inode = lo->plh_inode;
13c13a6a
TM		 301 pnfs_layoutreturn_before_put_layout_hdr(lo);
302
2b28a7be 303 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 304 if (!list_empty(&lo->plh_segs))
305 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 306 pnfs_detach_layout_hdr(lo);
cc6e5340 307 spin_unlock(&inode->i_lock);
6622c3ea 308 pnfs_free_layout_hdr(lo);
cc6e5340 309 }
974cec8c
AA		 310}
311
4aab9732
TM		 312static void
313pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
314 u32 seq)
315{
316 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
317 iomode = IOMODE_ANY;
318 lo->plh_return_iomode = iomode;
319 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
320 if (seq != 0) {
321 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
322 lo->plh_return_seq = seq;
323 }
324}
325
ae5a459d
TM		 326static void
327pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
328{
5466d214 329 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 330 lo->plh_return_iomode = 0;
331 lo->plh_return_seq = 0;
332 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 333 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
334 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
335 continue;
336 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
337 }
ae5a459d
TM		 338}
339
362fb578
TM		 340static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
341{
342 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
343 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
344 smp_mb__after_atomic();
345 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
346 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
347}
348
fe1cf946
TM		 349static void
350pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
351 struct list_head *free_me)
352{
353 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
354 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
355 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
356 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
357 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
358 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
359}
360
7380020e
TM		 361/*
362 * Update the seqid of a layout stateid
363 */
ecf84026
TM		 364bool nfs4_layoutreturn_refresh_stateid(nfs4_stateid *dst,
365 struct pnfs_layout_range *dst_range,
366 struct inode *inode)
7380020e
TM		 367{
368 struct pnfs_layout_hdr *lo;
c16467dc
TM		 369 struct pnfs_layout_range range = {
370 .iomode = IOMODE_ANY,
371 .offset = 0,
372 .length = NFS4_MAX_UINT64,
373 };
7380020e 374 bool ret = false;
c16467dc
TM		 375 LIST_HEAD(head);
376 int err;
7380020e
TM		 377
378 spin_lock(&inode->i_lock);
379 lo = NFS_I(inode)->layout;
380 if (lo && nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
c16467dc
TM		 381 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
382 if (err != -EBUSY) {
383 dst->seqid = lo->plh_stateid.seqid;
ecf84026 384 *dst_range = range;
c16467dc
TM		 385 ret = true;
386 }
7380020e
TM		 387 }
388 spin_unlock(&inode->i_lock);
c16467dc 389 pnfs_free_lseg_list(&head);
7380020e
TM		 390 return ret;
391}
392
2454dfea
TM		 393/*
394 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
395 *
396 * In order to continue using the pnfs_layout_hdr, a full recovery
397 * is required.
398 * Note that caller must hold inode->i_lock.
399 */
5f46be04 400int
2454dfea
TM		 401pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
402 struct list_head *lseg_list)
403{
404 struct pnfs_layout_range range = {
405 .iomode = IOMODE_ANY,
406 .offset = 0,
407 .length = NFS4_MAX_UINT64,
408 };
fe1cf946 409 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 410
411 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 412 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
413 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 414 pnfs_clear_layoutreturn_info(lo);
68f74479 415 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 416 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
417 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
418 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 419 return !list_empty(&lo->plh_segs);
2454dfea
TM		 420}
421
b9e028fd
TM		 422static int
423pnfs_iomode_to_fail_bit(u32 iomode)
424{
425 return iomode == IOMODE_RW ?
426 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
427}
428
429static void
3e621214 430pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 431{
25c75333 432 lo->plh_retry_timestamp = jiffies;
39e88fcf 433 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 434 refcount_inc(&lo->plh_refcount);
3e621214
TM		 435}
436
437static void
438pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
439{
440 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 441 refcount_dec(&lo->plh_refcount);
3e621214
TM		 442}
443
444static void
445pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
446{
447 struct inode *inode = lo->plh_inode;
115ce575
TM		 448 struct pnfs_layout_range range = {
449 .iomode = iomode,
450 .offset = 0,
451 .length = NFS4_MAX_UINT64,
452 };
453 LIST_HEAD(head);
3e621214
TM		 454
455 spin_lock(&inode->i_lock);
456 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 457 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 458 spin_unlock(&inode->i_lock);
115ce575 459 pnfs_free_lseg_list(&head);
b9e028fd
TM		 460 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
461 iomode == IOMODE_RW ? "RW" : "READ");
462}
463
464static bool
465pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
466{
25c75333 467 unsigned long start, end;
3e621214
TM		 468 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
469
470 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 471 return false;
472 end = jiffies;
473 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
474 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
475 /* It is time to retry the failed layoutgets */
3e621214 476 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 477 return false;
478 }
479 return true;
b9e028fd
TM		 480}
481
974cec8c 482static void
119cef97
TM		 483pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
484 const struct pnfs_layout_range *range,
485 const nfs4_stateid *stateid)
974cec8c 486{
566052c5 487 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 488 INIT_LIST_HEAD(&lseg->pls_lc_list);
eba6dd69 489 refcount_set(&lseg->pls_refcount, 1);
4541d16c 490 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 491 lseg->pls_layout = lo;
119cef97
TM		 492 lseg->pls_range = *range;
493 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 494}
495
905ca191 496static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 497{
68f74479
TM		 498 if (lseg != NULL) {
499 struct inode *inode = lseg->pls_layout->plh_inode;
500 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
501 }
974cec8c
AA		 502}
503
d684d2ae 504static void
57036a37
TM		 505pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
506 struct pnfs_layout_segment *lseg)
d684d2ae 507{
d20581aa 508 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 509 list_del_init(&lseg->pls_list);
8f0d27dc 510 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 511 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 512 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
513 return;
7b650994
TM		 514 if (list_empty(&lo->plh_segs) &&
515 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
516 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 517 if (atomic_read(&lo->plh_outstanding) == 0)
518 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 519 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 520 }
d684d2ae
FI		 521}
522
68f74479
TM		 523static bool
524pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
525 struct pnfs_layout_segment *lseg)
526{
527 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
528 pnfs_layout_is_valid(lo)) {
4aab9732 529 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 530 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
531 return true;
532 }
533 return false;
534}
535
bae724ef 536void
9369a431 537pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 538{
57036a37 539 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 540 struct inode *inode;
541
542 if (!lseg)
543 return;
544
4541d16c 545 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 546 refcount_read(&lseg->pls_refcount),
4541d16c 547 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 548
57036a37
TM		 549 lo = lseg->pls_layout;
550 inode = lo->plh_inode;
4ef2e4f8 551
eba6dd69 552 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 553 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
554 spin_unlock(&inode->i_lock);
555 return;
556 }
8f0d27dc 557 pnfs_get_layout_hdr(lo);
4ef2e4f8 558 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 559 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
560 lseg = NULL;
4ef2e4f8
TM		 561 spin_unlock(&inode->i_lock);
562 pnfs_free_lseg(lseg);
563 pnfs_put_layout_hdr(lo);
4541d16c 564 }
4541d16c 565}
9369a431 566EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 567
fb3296eb
BH		 568/*
569 * is l2 fully contained in l1?
570 * start1 end1
571 * [----------------------------------)
572 * start2 end2
573 * [----------------)
574 */
3cb2df17 575static bool
7dc0ac70 576pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 577 const struct pnfs_layout_range *l2)
fb3296eb
BH		 578{
579 u64 start1 = l1->offset;
17822b20 580 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 581 u64 start2 = l2->offset;
17822b20 582 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 583
584 return (start1 <= start2) && (end1 >= end2);
585}
586
24956804
TM		 587static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
588 struct list_head *tmp_list)
589{
eba6dd69 590 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 591 return false;
592 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
593 list_add(&lseg->pls_list, tmp_list);
594 return true;
595}
596
4541d16c
FI		 597/* Returns 1 if lseg is removed from list, 0 otherwise */
598static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
599 struct list_head *tmp_list)
600{
601 int rv = 0;
602
603 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
604 /* Remove the reference keeping the lseg in the
605 * list. It will now be removed when all
606 * outstanding io is finished.
607 */
d684d2ae 608 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 609 refcount_read(&lseg->pls_refcount));
24956804 610 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 611 rv = 1;
4541d16c
FI		 612 }
613 return rv;
614}
615
6d597e17
JL		 616/*
617 * Compare 2 layout stateid sequence ids, to see which is newer,
618 * taking into account wraparound issues.
619 */
620static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
621{
622 return (s32)(s1 - s2) > 0;
623}
624
e036f464
TM		 625static bool
626pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
627 const struct pnfs_layout_range *recall_range)
628{
629 return (recall_range->iomode == IOMODE_ANY ||
630 lseg_range->iomode == recall_range->iomode) &&
631 pnfs_lseg_range_intersecting(lseg_range, recall_range);
632}
633
634static bool
635pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
636 const struct pnfs_layout_range *recall_range,
637 u32 seq)
638{
639 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
640 return false;
641 if (recall_range == NULL)
642 return true;
643 return pnfs_should_free_range(&lseg->pls_range, recall_range);
644}
645
6d597e17
JL		 646/**
647 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
648 * @lo: layout header containing the lsegs
649 * @tmp_list: list head where doomed lsegs should go
650 * @recall_range: optional recall range argument to match (may be NULL)
651 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
652 *
653 * Walk the list of lsegs in the layout header, and tear down any that should
654 * be destroyed. If "recall_range" is specified then the segment must match
655 * that range. If "seq" is non-zero, then only match segments that were handed
656 * out at or before that sequence.
657 *
658 * Returns number of matching invalid lsegs remaining in list after scanning
659 * it and purging them.
4541d16c 660 */
43f1b3da 661int
49a85061 662pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 663 struct list_head *tmp_list,
6d597e17
JL		 664 const struct pnfs_layout_range *recall_range,
665 u32 seq)
974cec8c
AA		 666{
667 struct pnfs_layout_segment *lseg, *next;
71b39854 668 int remaining = 0;
974cec8c
AA		 669
670 dprintk("%s:Begin lo %p\n", __func__, lo);
671
8006bfba 672 if (list_empty(&lo->plh_segs))
38511722 673 return 0;
4541d16c 674 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 675 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 676 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 677 "offset %llu length %llu\n", __func__,
6d597e17
JL		 678 lseg, lseg->pls_range.iomode, lseg->pls_seq,
679 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 680 if (!mark_lseg_invalid(lseg, tmp_list))
681 remaining++;
4541d16c 682 }
71b39854
TM		 683 dprintk("%s:Return %i\n", __func__, remaining);
684 return remaining;
974cec8c
AA		 685}
686
68f74479
TM		 687static void
688pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
689 struct list_head *free_me,
690 const struct pnfs_layout_range *range,
691 u32 seq)
692{
693 struct pnfs_layout_segment *lseg, *next;
694
695 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
696 if (pnfs_match_lseg_recall(lseg, range, seq))
697 list_move_tail(&lseg->pls_list, free_me);
698 }
699}
700
f49f9baa 701/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 702void
4541d16c 703pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 704{
4541d16c 705 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 706
707 if (list_empty(free_me))
708 return;
709
4541d16c 710 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 711 list_del(&lseg->pls_list);
905ca191 712 pnfs_free_lseg(lseg);
974cec8c
AA		 713 }
714}
715
e5e94017
BH		 716void
717pnfs_destroy_layout(struct nfs_inode *nfsi)
718{
719 struct pnfs_layout_hdr *lo;
974cec8c 720 LIST_HEAD(tmp_list);
e5e94017
BH		 721
722 spin_lock(&nfsi->vfs_inode.i_lock);
723 lo = nfsi->layout;
724 if (lo) {
3e621214 725 pnfs_get_layout_hdr(lo);
2454dfea 726 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 727 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
728 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
729 spin_unlock(&nfsi->vfs_inode.i_lock);
730 pnfs_free_lseg_list(&tmp_list);
1f18b82c 731 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 732 pnfs_put_layout_hdr(lo);
733 } else
734 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 735}
041245c8 736EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 737
fd9a8d71
TM		 738static bool
739pnfs_layout_add_bulk_destroy_list(struct inode *inode,
740 struct list_head *layout_list)
974cec8c
AA		 741{
742 struct pnfs_layout_hdr *lo;
fd9a8d71 743 bool ret = false;
974cec8c 744
fd9a8d71
TM		 745 spin_lock(&inode->i_lock);
746 lo = NFS_I(inode)->layout;
747 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
748 pnfs_get_layout_hdr(lo);
749 list_add(&lo->plh_bulk_destroy, layout_list);
750 ret = true;
751 }
752 spin_unlock(&inode->i_lock);
753 return ret;
754}
755
756/* Caller must hold rcu_read_lock and clp->cl_lock */
757static int
758pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
759 struct nfs_server *server,
760 struct list_head *layout_list)
761{
762 struct pnfs_layout_hdr *lo, *next;
763 struct inode *inode;
764
765 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
b85f5620
TM		 766 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
767 continue;
fd9a8d71
TM		 768 inode = igrab(lo->plh_inode);
769 if (inode == NULL)
770 continue;
771 list_del_init(&lo->plh_layouts);
772 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
773 continue;
774 rcu_read_unlock();
775 spin_unlock(&clp->cl_lock);
776 iput(inode);
777 spin_lock(&clp->cl_lock);
778 rcu_read_lock();
779 return -EAGAIN;
780 }
781 return 0;
782}
783
784static int
785pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
786 bool is_bulk_recall)
787{
788 struct pnfs_layout_hdr *lo;
789 struct inode *inode;
fd9a8d71
TM		 790 LIST_HEAD(lseg_list);
791 int ret = 0;
792
793 while (!list_empty(layout_list)) {
794 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
795 plh_bulk_destroy);
796 dprintk("%s freeing layout for inode %lu\n", __func__,
797 lo->plh_inode->i_ino);
798 inode = lo->plh_inode;
7c5d1875
CH		 799
800 pnfs_layoutcommit_inode(inode, false);
801
fd9a8d71
TM		 802 spin_lock(&inode->i_lock);
803 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 804 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
805 if (is_bulk_recall)
806 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 807 ret = -EAGAIN;
9fd4b9fc 808 }
fd9a8d71
TM		 809 spin_unlock(&inode->i_lock);
810 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 811 /* Free all lsegs that are attached to commit buckets */
812 nfs_commit_inode(inode, 0);
fd9a8d71
TM		 813 pnfs_put_layout_hdr(lo);
814 iput(inode);
815 }
816 return ret;
817}
818
819int
820pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
821 struct nfs_fsid *fsid,
822 bool is_recall)
823{
824 struct nfs_server *server;
825 LIST_HEAD(layout_list);
c47abcf8 826
974cec8c 827 spin_lock(&clp->cl_lock);
6382a441 828 rcu_read_lock();
fd9a8d71 829restart:
6382a441 830 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 831 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
832 continue;
833 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
834 server,
835 &layout_list) != 0)
836 goto restart;
6382a441
WAA		 837 }
838 rcu_read_unlock();
974cec8c
AA		 839 spin_unlock(&clp->cl_lock);
840
fd9a8d71
TM		 841 if (list_empty(&layout_list))
842 return 0;
843 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
844}
845
846int
847pnfs_destroy_layouts_byclid(struct nfs_client *clp,
848 bool is_recall)
849{
850 struct nfs_server *server;
851 LIST_HEAD(layout_list);
852
853 spin_lock(&clp->cl_lock);
854 rcu_read_lock();
855restart:
856 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
857 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
858 server,
859 &layout_list) != 0)
860 goto restart;
974cec8c 861 }
fd9a8d71
TM		 862 rcu_read_unlock();
863 spin_unlock(&clp->cl_lock);
864
865 if (list_empty(&layout_list))
866 return 0;
867 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
868}
869
870/*
871 * Called by the state manger to remove all layouts established under an
872 * expired lease.
873 */
874void
875pnfs_destroy_all_layouts(struct nfs_client *clp)
876{
877 nfs4_deviceid_mark_client_invalid(clp);
878 nfs4_deviceid_purge_client(clp);
879
880 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 881}
882
fd6002e9 883/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 884void
885pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
886 bool update_barrier)
b1f69b75 887{
ecebb80b 888 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 889
2d2f24ad
TM		 890 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
891 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 892
893 if (!pnfs_layout_is_valid(lo)) {
894 nfs4_stateid_copy(&lo->plh_stateid, new);
895 lo->plh_barrier = newseq;
896 pnfs_clear_layoutreturn_info(lo);
897 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
898 return;
899 }
900 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 901 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 902 /*
903 * Because of wraparound, we want to keep the barrier
904 * "close" to the current seqids.
905 */
906 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 907 }
ecebb80b
TM		 908 if (update_barrier)
909 new_barrier = be32_to_cpu(new->seqid);
910 else if (new_barrier == 0)
911 return;
2a59a041 912 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 913 lo->plh_barrier = new_barrier;
b1f69b75
AA		 914}
915
cf7d63f1 916static bool
19c54aba
TM		 917pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
918 const nfs4_stateid *stateid)
43f1b3da 919{
19c54aba 920 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 921
19c54aba
TM		 922 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
923}
924
925/* lget is set to 1 if called from inside send_layoutget call chain */
926static bool
e1c06f80 927pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 928{
f7e8917a 929 return lo->plh_block_lgets ||
e1c06f80 930 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 931}
932
5e36e2a9
FI		 933static struct nfs_server *
934pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
935{
936 struct nfs_server *server;
937
78746a38 938 if (inode) {
5e36e2a9 939 server = NFS_SERVER(inode);
78746a38 940 } else {
5e36e2a9
FI		 941 struct dentry *parent_dir = dget_parent(ctx->dentry);
942 server = NFS_SERVER(parent_dir->d_inode);
943 dput(parent_dir);
944 }
945 return server;
946}
947
29a8bfe5
TM		 948static void nfs4_free_pages(struct page **pages, size_t size)
949{
950 int i;
951
952 if (!pages)
953 return;
954
955 for (i = 0; i < size; i++) {
956 if (!pages[i])
957 break;
958 __free_page(pages[i]);
959 }
960 kfree(pages);
961}
962
963static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
964{
965 struct page **pages;
966 int i;
967
a2791d3a 968 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 969 if (!pages) {
970 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
971 return NULL;
972 }
973
974 for (i = 0; i < size; i++) {
975 pages[i] = alloc_page(gfp_flags);
976 if (!pages[i]) {
977 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 978 nfs4_free_pages(pages, i);
29a8bfe5
TM		 979 return NULL;
980 }
981 }
982
983 return pages;
984}
985
587f03de 986static struct nfs4_layoutget *
5e36e2a9 987pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 988 struct nfs_open_context *ctx,
2409a976 989 const nfs4_stateid *stateid,
e144e539 990 const struct pnfs_layout_range *range,
587f03de 991 gfp_t gfp_flags)
e5e94017 992{
5e36e2a9 993 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 994 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 995 size_t max_pages = max_response_pages(server);
b1f69b75 996 struct nfs4_layoutget *lgp;
b1f69b75
AA		 997
998 dprintk("--> %s\n", __func__);
e5e94017 999
83026d80
JL		 1000 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1001 if (lgp == NULL)
587f03de 1002 return NULL;
83026d80 1003
28ced9a8
TM		 1004 if (max_reply_sz) {
1005 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1006 if (npages < max_pages)
1007 max_pages = npages;
1008 }
1009
dacb452d
FI		 1010 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1011 if (!lgp->args.layout.pages) {
1012 kfree(lgp);
1013 return NULL;
1014 }
1015 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1016 lgp->res.layoutp = &lgp->args.layout;
1017
d49e0d5b
TM		 1018 /* Don't confuse uninitialised result and success */
1019 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1020
1021 lgp->args.minlength = PAGE_SIZE;
1022 if (lgp->args.minlength > range->length)
1023 lgp->args.minlength = range->length;
5e36e2a9
FI		 1024 if (ino) {
1025 loff_t i_size = i_size_read(ino);
1026
1027 if (range->iomode == IOMODE_READ) {
1028 if (range->offset >= i_size)
1029 lgp->args.minlength = 0;
1030 else if (i_size - range->offset < lgp->args.minlength)
1031 lgp->args.minlength = i_size - range->offset;
1032 }
d03ab29d 1033 }
83026d80
JL		 1034 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1035 pnfs_copy_range(&lgp->args.range, range);
1036 lgp->args.type = server->pnfs_curr_ld->id;
1037 lgp->args.inode = ino;
1038 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1039 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1040 lgp->gfp_flags = gfp_flags;
a52458b4 1041 lgp->cred = get_cred(ctx->cred);
587f03de 1042 return lgp;
974cec8c
AA		 1043}
1044
29a8bfe5
TM		 1045void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1046{
1047 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1048
1049 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1050 if (lgp->args.inode)
1051 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
a52458b4 1052 put_cred(lgp->cred);
29a8bfe5
TM		 1053 put_nfs_open_context(lgp->args.ctx);
1054 kfree(lgp);
1055}
1056
24956804
TM		 1057static void pnfs_clear_layoutcommit(struct inode *inode,
1058 struct list_head *head)
1059{
1060 struct nfs_inode *nfsi = NFS_I(inode);
1061 struct pnfs_layout_segment *lseg, *tmp;
1062
1063 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1064 return;
1065 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1066 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1067 continue;
1068 pnfs_lseg_dec_and_remove_zero(lseg, head);
1069 }
1070}
1071
68f74479 1072void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1073 const nfs4_stateid *arg_stateid,
68f74479 1074 const struct pnfs_layout_range *range,
68f74479
TM		 1075 const nfs4_stateid *stateid)
1076{
1077 struct inode *inode = lo->plh_inode;
1078 LIST_HEAD(freeme);
1079
1080 spin_lock(&inode->i_lock);
2a974425
TM		 1081 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1082 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1083 goto out_unlock;
68f74479 1084 if (stateid) {
2a974425
TM		 1085 u32 seq = be32_to_cpu(arg_stateid->seqid);
1086
68f74479
TM		 1087 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1088 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1089 pnfs_set_layout_stateid(lo, stateid, true);
1090 } else
1091 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1092out_unlock:
68f74479
TM		 1093 pnfs_clear_layoutreturn_waitbit(lo);
1094 spin_unlock(&inode->i_lock);
1095 pnfs_free_lseg_list(&freeme);
1096
1097}
1098
13c13a6a 1099static bool
e5fd1904
TM		 1100pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1101 nfs4_stateid *stateid,
1102 enum pnfs_iomode *iomode)
13c13a6a 1103{
bf0291dd
TM		 1104 /* Serialise LAYOUTGET/LAYOUTRETURN */
1105 if (atomic_read(&lo->plh_outstanding) != 0)
1106 return false;
6604b203 1107 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1108 return false;
6604b203 1109 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1110 pnfs_get_layout_hdr(lo);
e5fd1904
TM		 1111 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1112 if (stateid != NULL) {
1113 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1114 if (lo->plh_return_seq != 0)
1115 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1116 }
1117 if (iomode != NULL)
1118 *iomode = lo->plh_return_iomode;
1119 pnfs_clear_layoutreturn_info(lo);
1120 return true;
1121 }
1122 if (stateid != NULL)
1123 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1124 if (iomode != NULL)
1125 *iomode = IOMODE_ANY;
13c13a6a
TM		 1126 return true;
1127}
1128
828ed9ec
TM		 1129static void
1130pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1131 struct pnfs_layout_hdr *lo,
1132 const nfs4_stateid *stateid,
1133 enum pnfs_iomode iomode)
1134{
1135 struct inode *inode = lo->plh_inode;
1136
1137 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1138 args->inode = inode;
1139 args->range.iomode = iomode;
1140 args->range.offset = 0;
1141 args->range.length = NFS4_MAX_UINT64;
1142 args->layout = lo;
1143 nfs4_stateid_copy(&args->stateid, stateid);
1144}
1145
f40eb5d0 1146static int
ed429d6b 1147pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
6c16605d 1148 enum pnfs_iomode iomode, bool sync)
f40eb5d0
PT		 1149{
1150 struct inode *ino = lo->plh_inode;
287bd3e9 1151 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0
PT		 1152 struct nfs4_layoutreturn *lrp;
1153 int status = 0;
1154
e4af440a 1155 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1156 if (unlikely(lrp == NULL)) {
1157 status = -ENOMEM;
1158 spin_lock(&ino->i_lock);
d67ae825 1159 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0
PT		 1160 spin_unlock(&ino->i_lock);
1161 pnfs_put_layout_hdr(lo);
1162 goto out;
1163 }
1164
828ed9ec 1165 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1166 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0
PT		 1167 lrp->clp = NFS_SERVER(ino)->nfs_client;
1168 lrp->cred = lo->plh_lc_cred;
287bd3e9
TM		 1169 if (ld->prepare_layoutreturn)
1170 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1171
6c16605d 1172 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1173out:
1174 dprintk("<-- %s status: %d\n", __func__, status);
1175 return status;
1176}
1177
13c13a6a
TM		 1178/* Return true if layoutreturn is needed */
1179static bool
1180pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1181{
1182 struct pnfs_layout_segment *s;
f0b42981
TM		 1183 enum pnfs_iomode iomode;
1184 u32 seq;
13c13a6a 1185
2370abda 1186 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1187 return false;
1188
f0b42981
TM		 1189 seq = lo->plh_return_seq;
1190 iomode = lo->plh_return_iomode;
1191
1192 /* Defer layoutreturn until all recalled lsegs are done */
13c13a6a 1193 list_for_each_entry(s, &lo->plh_segs, pls_list) {
f0b42981
TM		 1194 if (seq && pnfs_seqid_is_newer(s->pls_seq, seq))
1195 continue;
1196 if (iomode != IOMODE_ANY && s->pls_range.iomode != iomode)
1197 continue;
13c13a6a
TM		 1198 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1199 return false;
1200 }
1201
1202 return true;
1203}
1204
1205static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1206{
1207 struct inode *inode= lo->plh_inode;
1208
2370abda 1209 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1210 return;
1211 spin_lock(&inode->i_lock);
1212 if (pnfs_layout_need_return(lo)) {
1213 nfs4_stateid stateid;
1214 enum pnfs_iomode iomode;
1215 bool send;
1216
e5fd1904 1217 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
13c13a6a
TM		 1218 spin_unlock(&inode->i_lock);
1219 if (send) {
1220 /* Send an async layoutreturn so we dont deadlock */
1221 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1222 }
1223 } else
1224 spin_unlock(&inode->i_lock);
1225}
1226
293b3b06
AA		 1227/*
1228 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1229 * when the layout segment list is empty.
1230 *
1231 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1232 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1233 * deviceid is marked invalid.
1234 */
cbe82603
BH		 1235int
1236_pnfs_return_layout(struct inode *ino)
1237{
1238 struct pnfs_layout_hdr *lo = NULL;
1239 struct nfs_inode *nfsi = NFS_I(ino);
1240 LIST_HEAD(tmp_list);
cbe82603 1241 nfs4_stateid stateid;
24408f52 1242 int status = 0;
93b7f7ad 1243 bool send, valid_layout;
cbe82603 1244
366d5052 1245 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1246
1247 spin_lock(&ino->i_lock);
1248 lo = nfsi->layout;
e5929f3c 1249 if (!lo) {
cbe82603 1250 spin_unlock(&ino->i_lock);
293b3b06
AA		 1251 dprintk("NFS: %s no layout to return\n", __func__);
1252 goto out;
cbe82603 1253 }
cbe82603 1254 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1255 pnfs_get_layout_hdr(lo);
24408f52
TM		 1256 /* Is there an outstanding layoutreturn ? */
1257 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1258 spin_unlock(&ino->i_lock);
1259 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1260 TASK_UNINTERRUPTIBLE))
1261 goto out_put_layout_hdr;
1262 spin_lock(&ino->i_lock);
1263 }
93b7f7ad 1264 valid_layout = pnfs_layout_is_valid(lo);
24956804 1265 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1266 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1267
1268 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1269 struct pnfs_layout_range range = {
1270 .iomode = IOMODE_ANY,
1271 .offset = 0,
1272 .length = NFS4_MAX_UINT64,
1273 };
1274 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1275 }
1276
293b3b06 1277 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1278 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1279 !valid_layout) {
293b3b06 1280 spin_unlock(&ino->i_lock);
293b3b06 1281 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1282 goto out_put_layout_hdr;
293b3b06 1283 }
47abadef 1284
e5fd1904 1285 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603 1286 spin_unlock(&ino->i_lock);
7f27392c 1287 if (send)
ed429d6b 1288 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c 1289out_put_layout_hdr:
ee6625a9 1290 pnfs_free_lseg_list(&tmp_list);
7f27392c 1291 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1292out:
1293 dprintk("<-- %s status: %d\n", __func__, status);
1294 return status;
1295}
1296
24028672
TM		 1297int
1298pnfs_commit_and_return_layout(struct inode *inode)
1299{
1300 struct pnfs_layout_hdr *lo;
1301 int ret;
1302
1303 spin_lock(&inode->i_lock);
1304 lo = NFS_I(inode)->layout;
1305 if (lo == NULL) {
1306 spin_unlock(&inode->i_lock);
1307 return 0;
1308 }
1309 pnfs_get_layout_hdr(lo);
1310 /* Block new layoutgets and read/write to ds */
1311 lo->plh_block_lgets++;
1312 spin_unlock(&inode->i_lock);
1313 filemap_fdatawait(inode->i_mapping);
1314 ret = pnfs_layoutcommit_inode(inode, true);
1315 if (ret == 0)
1316 ret = _pnfs_return_layout(inode);
1317 spin_lock(&inode->i_lock);
1318 lo->plh_block_lgets--;
1319 spin_unlock(&inode->i_lock);
1320 pnfs_put_layout_hdr(lo);
1321 return ret;
1322}
1323
1c5bd76d
TM		 1324bool pnfs_roc(struct inode *ino,
1325 struct nfs4_layoutreturn_args *args,
1326 struct nfs4_layoutreturn_res *res,
a52458b4 1327 const struct cred *cred)
f7e8917a 1328{
40dd4b7a
TM		 1329 struct nfs_inode *nfsi = NFS_I(ino);
1330 struct nfs_open_context *ctx;
1331 struct nfs4_state *state;
f7e8917a 1332 struct pnfs_layout_hdr *lo;
1c5bd76d 1333 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1334 nfs4_stateid stateid;
1c5bd76d
TM		 1335 enum pnfs_iomode iomode = 0;
1336 bool layoutreturn = false, roc = false;
e71708d4 1337 bool skip_read = false;
f7e8917a 1338
1c5bd76d
TM		 1339 if (!nfs_have_layout(ino))
1340 return false;
29ade5db 1341retry:
0de43976 1342 rcu_read_lock();
f7e8917a 1343 spin_lock(&ino->i_lock);
40dd4b7a 1344 lo = nfsi->layout;
0cdc329e 1345 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1346 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1347 lo = NULL;
40dd4b7a 1348 goto out_noroc;
9c6376eb
TM		 1349 }
1350 pnfs_get_layout_hdr(lo);
29ade5db 1351 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1352 spin_unlock(&ino->i_lock);
0de43976 1353 rcu_read_unlock();
29ade5db
TM		 1354 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1355 TASK_UNINTERRUPTIBLE);
1356 pnfs_put_layout_hdr(lo);
1357 goto retry;
1358 }
40dd4b7a 1359
e755d638 1360 /* no roc if we hold a delegation */
e71708d4
TM		 1361 if (nfs4_check_delegation(ino, FMODE_READ)) {
1362 if (nfs4_check_delegation(ino, FMODE_WRITE))
1363 goto out_noroc;
1364 skip_read = true;
1365 }
40dd4b7a 1366
0de43976 1367 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1368 state = ctx->state;
e71708d4
TM		 1369 if (state == NULL)
1370 continue;
40dd4b7a 1371 /* Don't return layout if there is open file state */
e71708d4 1372 if (state->state & FMODE_WRITE)
40dd4b7a 1373 goto out_noroc;
e71708d4
TM		 1374 if (state->state & FMODE_READ)
1375 skip_read = true;
40dd4b7a
TM		 1376 }
1377
e755d638 1378
1c5bd76d 1379 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1380 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1381 continue;
e755d638 1382 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1383 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1384 continue;
1385 /*
1386 * Note: mark lseg for return so pnfs_layout_remove_lseg
1387 * doesn't invalidate the layout for us.
1388 */
1389 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1390 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1391 continue;
1392 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1393 }
1394
1c5bd76d
TM		 1395 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1396 goto out_noroc;
69820d22 1397
500d701f 1398 /* ROC in two conditions:
e755d638
PT		 1399 * 1. there are ROC lsegs
1400 * 2. we don't send layoutreturn
e755d638 1401 */
1c5bd76d
TM		 1402 /* lo ref dropped in pnfs_roc_release() */
1403 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1404 /* If the creds don't match, we can't compound the layoutreturn */
1405 if (!layoutreturn || cred != lo->plh_lc_cred)
1406 goto out_noroc;
1407
1408 roc = layoutreturn;
1409 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1410 res->lrs_present = 0;
1411 layoutreturn = false;
f7e8917a 1412
40dd4b7a 1413out_noroc:
f7e8917a 1414 spin_unlock(&ino->i_lock);
0de43976 1415 rcu_read_unlock();
e755d638 1416 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1417 if (roc) {
1418 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1419 if (ld->prepare_layoutreturn)
1420 ld->prepare_layoutreturn(args);
9c6376eb 1421 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1422 return true;
1423 }
e755d638 1424 if (layoutreturn)
1c5bd76d 1425 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
9c6376eb 1426 pnfs_put_layout_hdr(lo);
287bd3e9 1427 return false;
f7e8917a
FI		 1428}
1429
1c5bd76d
TM		 1430void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1431 struct nfs4_layoutreturn_res *res,
1432 int ret)
f7e8917a 1433{
1c5bd76d
TM		 1434 struct pnfs_layout_hdr *lo = args->layout;
1435 const nfs4_stateid *arg_stateid = NULL;
1436 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1437 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1438
1c5bd76d
TM		 1439 if (ret == 0) {
1440 arg_stateid = &args->stateid;
1441 if (res->lrs_present)
1442 res_stateid = &res->stateid;
1443 }
1444 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1445 res_stateid);
287bd3e9
TM		 1446 if (ld_private && ld_private->ops && ld_private->ops->free)
1447 ld_private->ops->free(ld_private);
1c5bd76d
TM		 1448 pnfs_put_layout_hdr(lo);
1449 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1450}
1451
500d701f
PT		 1452bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1453{
1454 struct nfs_inode *nfsi = NFS_I(ino);
1455 struct pnfs_layout_hdr *lo;
1456 bool sleep = false;
1457
1458 /* we might not have grabbed lo reference. so need to check under
1459 * i_lock */
1460 spin_lock(&ino->i_lock);
1461 lo = nfsi->layout;
ee284e35
TM		 1462 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1463 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1464 sleep = true;
ee284e35 1465 }
500d701f 1466 spin_unlock(&ino->i_lock);
500d701f
PT		 1467 return sleep;
1468}
1469
b1f69b75
AA		 1470/*
1471 * Compare two layout segments for sorting into layout cache.
1472 * We want to preferentially return RW over RO layouts, so ensure those
1473 * are seen first.
1474 */
1475static s64
7dc0ac70 1476pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1477 const struct pnfs_layout_range *l2)
b1f69b75 1478{
fb3296eb
BH		 1479 s64 d;
1480
1481 /* high offset > low offset */
1482 d = l1->offset - l2->offset;
1483 if (d)
1484 return d;
1485
1486 /* short length > long length */
1487 d = l2->length - l1->length;
1488 if (d)
1489 return d;
1490
b1f69b75 1491 /* read > read/write */
fb3296eb 1492 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1493}
1494
03772d2f
TM		 1495static bool
1496pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1497 const struct pnfs_layout_range *l2)
1498{
1499 return pnfs_lseg_range_cmp(l1, l2) > 0;
1500}
1501
1502static bool
1503pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1504 struct pnfs_layout_segment *old)
1505{
1506 return false;
1507}
1508
1509void
1510pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1511 struct pnfs_layout_segment *lseg,
1512 bool (*is_after)(const struct pnfs_layout_range *,
1513 const struct pnfs_layout_range *),
1514 bool (*do_merge)(struct pnfs_layout_segment *,
1515 struct pnfs_layout_segment *),
1516 struct list_head *free_me)
974cec8c 1517{
03772d2f 1518 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1519
974cec8c
AA		 1520 dprintk("%s:Begin\n", __func__);
1521
03772d2f
TM		 1522 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1523 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1524 continue;
1525 if (do_merge(lseg, lp)) {
1526 mark_lseg_invalid(lp, free_me);
1527 continue;
1528 }
1529 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1530 continue;
566052c5 1531 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1532 dprintk("%s: inserted lseg %p "
1533 "iomode %d offset %llu length %llu before "
1534 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1535 __func__, lseg, lseg->pls_range.iomode,
1536 lseg->pls_range.offset, lseg->pls_range.length,
1537 lp, lp->pls_range.iomode, lp->pls_range.offset,
1538 lp->pls_range.length);
fb3296eb 1539 goto out;
974cec8c 1540 }
fb3296eb
BH		 1541 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1542 dprintk("%s: inserted lseg %p "
1543 "iomode %d offset %llu length %llu at tail\n",
1544 __func__, lseg, lseg->pls_range.iomode,
1545 lseg->pls_range.offset, lseg->pls_range.length);
1546out:
70c3bd2b 1547 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1548
1549 dprintk("%s:Return\n", __func__);
e5e94017 1550}
03772d2f
TM		 1551EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1552
1553static void
1554pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1555 struct pnfs_layout_segment *lseg,
1556 struct list_head *free_me)
1557{
1558 struct inode *inode = lo->plh_inode;
1559 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1560
1561 if (ld->add_lseg != NULL)
1562 ld->add_lseg(lo, lseg, free_me);
1563 else
1564 pnfs_generic_layout_insert_lseg(lo, lseg,
1565 pnfs_lseg_range_is_after,
1566 pnfs_lseg_no_merge,
1567 free_me);
1568}
e5e94017
BH		 1569
1570static struct pnfs_layout_hdr *
9fa40758
PT		 1571alloc_init_layout_hdr(struct inode *ino,
1572 struct nfs_open_context *ctx,
1573 gfp_t gfp_flags)
e5e94017
BH		 1574{
1575 struct pnfs_layout_hdr *lo;
1576
636fb9c8 1577 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1578 if (!lo)
1579 return NULL;
2b28a7be 1580 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1581 INIT_LIST_HEAD(&lo->plh_layouts);
1582 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1583 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1584 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1585 lo->plh_inode = ino;
a52458b4 1586 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1587 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1588 return lo;
1589}
1590
1591static struct pnfs_layout_hdr *
9fa40758
PT		 1592pnfs_find_alloc_layout(struct inode *ino,
1593 struct nfs_open_context *ctx,
1594 gfp_t gfp_flags)
e5241e43
TM		 1595 __releases(&ino->i_lock)
1596 __acquires(&ino->i_lock)
e5e94017
BH		 1597{
1598 struct nfs_inode *nfsi = NFS_I(ino);
1599 struct pnfs_layout_hdr *new = NULL;
1600
1601 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1602
251ec410
TM		 1603 if (nfsi->layout != NULL)
1604 goto out_existing;
e5e94017 1605 spin_unlock(&ino->i_lock);
9fa40758 1606 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1607 spin_lock(&ino->i_lock);
1608
251ec410 1609 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1610 nfsi->layout = new;
251ec410 1611 return new;
7175fe90
YN		 1612 } else if (new != NULL)
1613 pnfs_free_layout_hdr(new);
251ec410
TM		 1614out_existing:
1615 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1616 return nfsi->layout;
1617}
1618
b1f69b75
AA		 1619/*
1620 * iomode matching rules:
c7d73af2
TH		 1621 * iomode lseg strict match
1622 * iomode
1623 * ----- ----- ------ -----
1624 * ANY READ N/A true
1625 * ANY RW N/A true
1626 * RW READ N/A false
1627 * RW RW N/A true
1628 * READ READ N/A true
1629 * READ RW true false
1630 * READ RW false true
b1f69b75 1631 */
3cb2df17 1632static bool
7dc0ac70 1633pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1634 const struct pnfs_layout_range *range,
1635 bool strict_iomode)
b1f69b75 1636{
fb3296eb
BH		 1637 struct pnfs_layout_range range1;
1638
1639 if ((range->iomode == IOMODE_RW &&
1640 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1641 (range->iomode != ls_range->iomode &&
6089dd0d 1642 strict_iomode) ||
7dc0ac70 1643 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1644 return false;
fb3296eb
BH		 1645
1646 /* range1 covers only the first byte in the range */
1647 range1 = *range;
1648 range1.length = 1;
7dc0ac70 1649 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1650}
1651
1652/*
1653 * lookup range in layout
1654 */
e5e94017 1655static struct pnfs_layout_segment *
fb3296eb 1656pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1657 struct pnfs_layout_range *range,
1658 bool strict_iomode)
e5e94017 1659{
b1f69b75
AA		 1660 struct pnfs_layout_segment *lseg, *ret = NULL;
1661
1662 dprintk("%s:Begin\n", __func__);
1663
b7edfaa1 1664 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1665 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1666 pnfs_lseg_range_match(&lseg->pls_range, range,
1667 strict_iomode)) {
9369a431 1668 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1669 break;
1670 }
b1f69b75
AA		 1671 }
1672
1673 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1674 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1675 return ret;
e5e94017
BH		 1676}
1677
d23d61c8
AA		 1678/*
1679 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1680 * to the MDS or over pNFS
1681 *
1682 * The nfs_inode read_io and write_io fields are cumulative counters reset
1683 * when there are no layout segments. Note that in pnfs_update_layout iomode
1684 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1685 * WRITE request.
1686 *
1687 * A return of true means use MDS I/O.
1688 *
1689 * From rfc 5661:
1690 * If a file's size is smaller than the file size threshold, data accesses
1691 * SHOULD be sent to the metadata server. If an I/O request has a length that
1692 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1693 * server. If both file size and I/O size are provided, the client SHOULD
1694 * reach or exceed both thresholds before sending its read or write
1695 * requests to the data server.
1696 */
1697static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1698 struct inode *ino, int iomode)
1699{
1700 struct nfs4_threshold *t = ctx->mdsthreshold;
1701 struct nfs_inode *nfsi = NFS_I(ino);
1702 loff_t fsize = i_size_read(ino);
1703 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1704
1705 if (t == NULL)
1706 return ret;
1707
1708 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1709 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1710
1711 switch (iomode) {
1712 case IOMODE_READ:
1713 if (t->bm & THRESHOLD_RD) {
1714 dprintk("%s fsize %llu\n", __func__, fsize);
1715 size_set = true;
1716 if (fsize < t->rd_sz)
1717 size = true;
1718 }
1719 if (t->bm & THRESHOLD_RD_IO) {
1720 dprintk("%s nfsi->read_io %llu\n", __func__,
1721 nfsi->read_io);
1722 io_set = true;
1723 if (nfsi->read_io < t->rd_io_sz)
1724 io = true;
1725 }
1726 break;
1727 case IOMODE_RW:
1728 if (t->bm & THRESHOLD_WR) {
1729 dprintk("%s fsize %llu\n", __func__, fsize);
1730 size_set = true;
1731 if (fsize < t->wr_sz)
1732 size = true;
1733 }
1734 if (t->bm & THRESHOLD_WR_IO) {
1735 dprintk("%s nfsi->write_io %llu\n", __func__,
1736 nfsi->write_io);
1737 io_set = true;
1738 if (nfsi->write_io < t->wr_io_sz)
1739 io = true;
1740 }
1741 break;
1742 }
1743 if (size_set && io_set) {
1744 if (size && io)
1745 ret = true;
1746 } else if (size || io)
1747 ret = true;
1748
1749 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1750 return ret;
1751}
1752
d03360aa 1753static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1754{
1755 /*
1756 * send layoutcommit as it can hold up layoutreturn due to lseg
1757 * reference
1758 */
1759 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1760 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1761 nfs_wait_bit_killable,
d03360aa 1762 TASK_KILLABLE);
aa8a45ee
PT		 1763}
1764
411ae722
TM		 1765static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1766{
1767 atomic_inc(&lo->plh_outstanding);
1768}
1769
1770static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1771{
1772 if (atomic_dec_and_test(&lo->plh_outstanding))
1773 wake_up_var(&lo->plh_outstanding);
1774}
1775
d67ae825
TH		 1776static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1777{
1778 unsigned long *bitlock = &lo->plh_flags;
1779
1780 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1781 smp_mb__after_atomic();
1782 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1783}
1784
78746a38
FI		 1785static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1786 struct nfs_server *server)
1787{
1788 if (list_empty(&lo->plh_layouts)) {
1789 struct nfs_client *clp = server->nfs_client;
1790
1791 /* The lo must be on the clp list if there is any
1792 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1793 */
1794 spin_lock(&clp->cl_lock);
1795 if (list_empty(&lo->plh_layouts))
1796 list_add_tail(&lo->plh_layouts, &server->layouts);
1797 spin_unlock(&clp->cl_lock);
1798 }
1799}
1800
e5e94017
BH		 1801/*
1802 * Layout segment is retreived from the server if not cached.
1803 * The appropriate layout segment is referenced and returned to the caller.
1804 */
7c24d948 1805struct pnfs_layout_segment *
e5e94017
BH		 1806pnfs_update_layout(struct inode *ino,
1807 struct nfs_open_context *ctx,
fb3296eb
BH		 1808 loff_t pos,
1809 u64 count,
a75b9df9 1810 enum pnfs_iomode iomode,
c7d73af2 1811 bool strict_iomode,
a75b9df9 1812 gfp_t gfp_flags)
e5e94017 1813{
fb3296eb
BH		 1814 struct pnfs_layout_range arg = {
1815 .iomode = iomode,
1816 .offset = pos,
1817 .length = count,
1818 };
70d2f7b1 1819 unsigned pg_offset;
6382a441
WAA		 1820 struct nfs_server *server = NFS_SERVER(ino);
1821 struct nfs_client *clp = server->nfs_client;
183d9e7b 1822 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1823 struct pnfs_layout_segment *lseg = NULL;
587f03de 1824 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1825 nfs4_stateid stateid;
1826 long timeout = 0;
66b53f32 1827 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1828 bool first;
e5e94017 1829
9a4bf31d 1830 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1831 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1832 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1833 goto out;
9a4bf31d 1834 }
d23d61c8 1835
9a4bf31d 1836 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1837 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1838 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1839 goto out;
9a4bf31d 1840 }
d23d61c8 1841
9bf87482 1842lookup_again:
d03360aa
TM		 1843 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1844 if (IS_ERR(lseg))
1845 goto out;
9bf87482 1846 first = false;
e5e94017 1847 spin_lock(&ino->i_lock);
9fa40758 1848 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1849 if (lo == NULL) {
1850 spin_unlock(&ino->i_lock);
183d9e7b 1851 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1852 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1853 goto out;
1854 }
e5e94017 1855
43f1b3da 1856 /* Do we even need to bother with this? */
a59c30ac 1857 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1858 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1859 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1860 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1861 goto out_unlock;
1862 }
1863
1864 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1865 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1866 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1867 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1868 goto out_unlock;
9a4bf31d 1869 }
e5e94017 1870
411ae722
TM		 1871 /*
1872 * If the layout segment list is empty, but there are outstanding
1873 * layoutget calls, then they might be subject to a layoutrecall.
1874 */
1875 if (list_empty(&lo->plh_segs) &&
1876 atomic_read(&lo->plh_outstanding) != 0) {
1877 spin_unlock(&ino->i_lock);
d03360aa
TM		 1878 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
1879 atomic_read(&lo->plh_outstanding)));
1880 if (IS_ERR(lseg) || !list_empty(&lo->plh_segs))
411ae722
TM		 1881 goto out_put_layout_hdr;
1882 pnfs_put_layout_hdr(lo);
1883 goto lookup_again;
1884 }
1885
c7d73af2 1886 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1887 if (lseg) {
1888 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1889 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1890 goto out_unlock;
1891 }
1892
1893 if (!nfs4_valid_open_stateid(ctx->state)) {
1894 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1895 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1896 goto out_unlock;
1897 }
1898
1899 /*
1900 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1901 * stateid, or it has been invalidated, then we must use the open
1902 * stateid.
1903 */
67a3b721 1904 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
183d9e7b
JL		 1905
1906 /*
1907 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1908 * RFC 5661 Errata 3208.
1909 */
1910 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1911 &lo->plh_flags)) {
1912 spin_unlock(&ino->i_lock);
d03360aa
TM		 1913 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
1914 NFS_LAYOUT_FIRST_LAYOUTGET,
1915 TASK_KILLABLE));
1916 if (IS_ERR(lseg))
1917 goto out_put_layout_hdr;
9bf87482 1918 pnfs_put_layout_hdr(lo);
183d9e7b 1919 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1920 goto lookup_again;
1921 }
183d9e7b
JL		 1922
1923 first = true;
70d2f7b1
TM		 1924 if (nfs4_select_rw_stateid(ctx->state,
1925 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1926 NULL, &stateid, NULL) != 0) {
1927 trace_pnfs_update_layout(ino, pos, count,
1928 iomode, lo, lseg,
1929 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1930 goto out_unlock;
1931 }
9bf87482 1932 } else {
183d9e7b 1933 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 1934 }
568e8c49 1935
aa8a45ee
PT		 1936 /*
1937 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1938 * for LAYOUTRETURN even if first is true.
1939 */
8f70f53a 1940 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 1941 spin_unlock(&ino->i_lock);
1942 dprintk("%s wait for layoutreturn\n", __func__);
d03360aa
TM		 1943 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
1944 if (!IS_ERR(lseg)) {
d67ae825
TH		 1945 if (first)
1946 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 1947 pnfs_put_layout_hdr(lo);
1948 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 1949 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1950 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 1951 goto lookup_again;
1952 }
183d9e7b 1953 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1954 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 1955 goto out_put_layout_hdr;
1956 }
1957
9a4bf31d 1958 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 1959 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1960 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 1961 goto out_unlock;
9a4bf31d 1962 }
411ae722 1963 nfs_layoutget_begin(lo);
f49f9baa 1964 spin_unlock(&ino->i_lock);
30005121 1965
78746a38 1966 _add_to_server_list(lo, server);
e5e94017 1967
09cbfeaf 1968 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 1969 if (pg_offset) {
1970 arg.offset -= pg_offset;
1971 arg.length += pg_offset;
1972 }
7c24d948 1973 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 1974 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 1975
5e36e2a9 1976 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 1977 if (!lgp) {
1978 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
1979 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 1980 nfs_layoutget_end(lo);
587f03de
FI		 1981 goto out_put_layout_hdr;
1982 }
1983
dacb452d 1984 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 1985 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1986 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 1987 nfs_layoutget_end(lo);
83026d80 1988 if (IS_ERR(lseg)) {
183d9e7b 1989 switch(PTR_ERR(lseg)) {
e85d7ee4 1990 case -EBUSY:
183d9e7b
JL		 1991 if (time_after(jiffies, giveup))
1992 lseg = NULL;
66b53f32
TM		 1993 break;
1994 case -ERECALLCONFLICT:
183d9e7b 1995 case -EAGAIN:
56b38a1f 1996 break;
183d9e7b
JL		 1997 default:
1998 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
1999 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2000 lseg = NULL;
2001 }
56b38a1f
TM		 2002 goto out_put_layout_hdr;
2003 }
2004 if (lseg) {
2005 if (first)
2006 pnfs_clear_first_layoutget(lo);
2007 trace_pnfs_update_layout(ino, pos, count,
2008 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2009 pnfs_put_layout_hdr(lo);
2010 goto lookup_again;
83026d80
JL		 2011 }
2012 } else {
2013 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2014 }
2015
830ffb56 2016out_put_layout_hdr:
d67ae825
TH		 2017 if (first)
2018 pnfs_clear_first_layoutget(lo);
70c3bd2b 2019 pnfs_put_layout_hdr(lo);
e5e94017 2020out:
f86bbcf8
TM		 2021 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2022 "(%s, offset: %llu, length: %llu)\n",
2023 __func__, ino->i_sb->s_id,
2024 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2025 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2026 iomode==IOMODE_RW ? "read/write" : "read-only",
2027 (unsigned long long)pos,
2028 (unsigned long long)count);
e5e94017
BH		 2029 return lseg;
2030out_unlock:
2031 spin_unlock(&ino->i_lock);
830ffb56 2032 goto out_put_layout_hdr;
e5e94017 2033}
7c24d948 2034EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2035
540d9864
TM		 2036static bool
2037pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2038{
2039 switch (range->iomode) {
2040 case IOMODE_READ:
2041 case IOMODE_RW:
2042 break;
2043 default:
2044 return false;
2045 }
2046 if (range->offset == NFS4_MAX_UINT64)
2047 return false;
2048 if (range->length == 0)
2049 return false;
2050 if (range->length != NFS4_MAX_UINT64 &&
2051 range->length > NFS4_MAX_UINT64 - range->offset)
2052 return false;
2053 return true;
2054}
2055
78746a38
FI		 2056static struct pnfs_layout_hdr *
2057_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2058{
2059 struct pnfs_layout_hdr *lo;
2060
2061 spin_lock(&ino->i_lock);
2062 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2063 if (!lo)
2064 goto out_unlock;
2065 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2066 goto out_unlock;
2067 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2068 goto out_unlock;
2069 if (pnfs_layoutgets_blocked(lo))
2070 goto out_unlock;
2071 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2072 goto out_unlock;
411ae722 2073 nfs_layoutget_begin(lo);
78746a38
FI		 2074 spin_unlock(&ino->i_lock);
2075 _add_to_server_list(lo, NFS_SERVER(ino));
2076 return lo;
2077
2078out_unlock:
2079 spin_unlock(&ino->i_lock);
2080 pnfs_put_layout_hdr(lo);
2081 return NULL;
2082}
2083
2409a976
FI		 2084extern const nfs4_stateid current_stateid;
2085
2086static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2087 struct nfs_open_context *ctx)
2088{
78746a38
FI		 2089 struct inode *ino = data->dentry->d_inode;
2090 struct pnfs_layout_range rng = {
2091 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2092 IOMODE_RW: IOMODE_READ,
2093 .offset = 0,
2094 .length = NFS4_MAX_UINT64,
2095 };
2096 struct nfs4_layoutget *lgp;
2097 struct pnfs_layout_hdr *lo;
2098
64294b08
TM		 2099 /* Heuristic: don't send layoutget if we have cached data */
2100 if (rng.iomode == IOMODE_READ &&
2101 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2102 return;
2103
78746a38
FI		 2104 lo = _pnfs_grab_empty_layout(ino, ctx);
2105 if (!lo)
2106 return;
2107 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2108 &rng, GFP_KERNEL);
2109 if (!lgp) {
2110 pnfs_clear_first_layoutget(lo);
2111 pnfs_put_layout_hdr(lo);
2112 return;
2113 }
2114 data->lgp = lgp;
2115 data->o_arg.lg_args = &lgp->args;
2116 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2117}
2118
2119static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2120 struct nfs_open_context *ctx)
2121{
2122 struct pnfs_layout_range rng = {
2123 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2124 IOMODE_RW: IOMODE_READ,
2125 .offset = 0,
2126 .length = NFS4_MAX_UINT64,
2127 };
2128 struct nfs4_layoutget *lgp;
2129
2130 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2131 &rng, GFP_KERNEL);
2132 if (!lgp)
2133 return;
2134 data->lgp = lgp;
2135 data->o_arg.lg_args = &lgp->args;
2136 data->o_res.lg_res = &lgp->res;
2137}
2138
2139void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2140 struct nfs_open_context *ctx)
2141{
2142 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2143
2144 if (!(pnfs_enabled_sb(server) &&
2145 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2146 return;
2147 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2148 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2149 return;
2409a976
FI		 2150 if (data->state)
2151 _lgopen_prepare_attached(data, ctx);
2152 else
2153 _lgopen_prepare_floating(data, ctx);
2154}
2155
2156void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2157 struct nfs_open_context *ctx)
2158{
2159 struct pnfs_layout_hdr *lo;
2160 struct pnfs_layout_segment *lseg;
c49b5209 2161 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2162 u32 iomode;
2163
6e01260c 2164 if (!lgp)
2409a976 2165 return;
6e01260c
FI		 2166 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2167 if (lgp->res.status) {
2168 switch (lgp->res.status) {
6e01260c 2169 default:
8dc96566
TM		 2170 break;
2171 /*
2172 * Halt lgopen attempts if the server doesn't recognise
2173 * the "current stateid" value, the layout type, or the
2174 * layoutget operation as being valid.
2175 * Also if it complains about too many ops in the compound
2176 * or of the request/reply being too big.
2177 */
2178 case -NFS4ERR_BAD_STATEID:
2179 case -NFS4ERR_NOTSUPP:
2180 case -NFS4ERR_REP_TOO_BIG:
2181 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2182 case -NFS4ERR_REQ_TOO_BIG:
2183 case -NFS4ERR_TOO_MANY_OPS:
2184 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2185 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2186 }
2187 return;
2188 }
2409a976 2189 if (!lgp->args.inode) {
78746a38
FI		 2190 lo = _pnfs_grab_empty_layout(ino, ctx);
2191 if (!lo)
2192 return;
2409a976
FI		 2193 lgp->args.inode = ino;
2194 } else
2195 lo = NFS_I(lgp->args.inode)->layout;
2409a976
FI		 2196
2197 lseg = pnfs_layout_process(lgp);
32f1c28f 2198 if (!IS_ERR(lseg)) {
2409a976
FI		 2199 iomode = lgp->args.range.iomode;
2200 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2201 pnfs_put_lseg(lseg);
2202 }
30ae2412
FI		 2203}
2204
2205void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2206{
2207 if (lgp != NULL) {
2208 struct inode *inode = lgp->args.inode;
2209 if (inode) {
2210 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
30ae2412 2211 pnfs_clear_first_layoutget(lo);
411ae722 2212 nfs_layoutget_end(lo);
30ae2412
FI		 2213 }
2214 pnfs_layoutget_free(lgp);
2215 }
2409a976
FI		 2216}
2217
a0b0a6e3 2218struct pnfs_layout_segment *
b1f69b75
AA		 2219pnfs_layout_process(struct nfs4_layoutget *lgp)
2220{
2221 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2222 struct nfs4_layoutget_res *res = &lgp->res;
2223 struct pnfs_layout_segment *lseg;
b7edfaa1 2224 struct inode *ino = lo->plh_inode;
78096cca 2225 LIST_HEAD(free_me);
540d9864
TM		 2226
2227 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2228 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2229
2230 /* Inject layout blob into I/O device driver */
a75b9df9 2231 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2232 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2233 if (!lseg)
1b3c6d07
JL		 2234 lseg = ERR_PTR(-ENOMEM);
2235
2236 dprintk("%s: Could not allocate layout: error %ld\n",
2237 __func__, PTR_ERR(lseg));
2238 return lseg;
b1f69b75
AA		 2239 }
2240
119cef97 2241 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2242
b1f69b75 2243 spin_lock(&ino->i_lock);
e1c06f80 2244 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2245 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2246 goto out_forget;
43f1b3da 2247 }
038d6493 2248
9888d837
TM		 2249 if (!pnfs_layout_is_valid(lo)) {
2250 /* We have a completely new layout */
2251 pnfs_set_layout_stateid(lo, &res->stateid, true);
2252 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2253 /* existing state ID, make sure the sequence number matches. */
2254 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2255 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2256 goto out_forget;
362f7474
CH		 2257 }
2258 pnfs_set_layout_stateid(lo, &res->stateid, false);
2259 } else {
2260 /*
2261 * We got an entirely new state ID. Mark all segments for the
9888d837 2262 * inode invalid, and retry the layoutget
362f7474 2263 */
d9b61708 2264 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 2265 goto out_forget;
362f7474 2266 }
038d6493 2267
9369a431 2268 pnfs_get_lseg(lseg);
03772d2f 2269 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2270
b1f69b75 2271
3976143b 2272 if (res->return_on_close)
f7e8917a 2273 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2274
b1f69b75 2275 spin_unlock(&ino->i_lock);
78096cca 2276 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2277 return lseg;
43f1b3da 2278
1b3c6d07 2279out_forget:
43f1b3da
FI		 2280 spin_unlock(&ino->i_lock);
2281 lseg->pls_layout = lo;
2282 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2283 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2284}
2285
e0b7d420
TM		 2286static int
2287mark_lseg_invalid_or_return(struct pnfs_layout_segment *lseg,
2288 struct list_head *tmp_list)
2289{
2290 if (!mark_lseg_invalid(lseg, tmp_list))
2291 return 0;
2292 pnfs_cache_lseg_for_layoutreturn(lseg->pls_layout, lseg);
2293 return 1;
2294}
2295
2f215968
TM		 2296/**
2297 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2298 * @lo: pointer to layout header
2299 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2300 * @return_range: describe layout segment ranges to be returned
e0b7d420 2301 * @seq: stateid seqid to match
2f215968
TM		 2302 *
2303 * This function is mainly intended for use by layoutrecall. It attempts
2304 * to free the layout segment immediately, or else to mark it for return
2305 * as soon as its reference count drops to zero.
e0b7d420
TM		 2306 *
2307 * Returns
2308 * - 0: a layoutreturn needs to be scheduled.
2309 * - EBUSY: there are layout segment that are still in use.
2310 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2311 */
10335556 2312int
016256df
PT		 2313pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2314 struct list_head *tmp_list,
6d597e17
JL		 2315 const struct pnfs_layout_range *return_range,
2316 u32 seq)
016256df
PT		 2317{
2318 struct pnfs_layout_segment *lseg, *next;
10335556 2319 int remaining = 0;
016256df
PT		 2320
2321 dprintk("%s:Begin lo %p\n", __func__, lo);
2322
fc7ff367 2323 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2324
2325 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2326 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2327 dprintk("%s: marking lseg %p iomode %d "
2328 "offset %llu length %llu\n", __func__,
2329 lseg, lseg->pls_range.iomode,
2330 lseg->pls_range.offset,
2331 lseg->pls_range.length);
e0b7d420 2332 if (mark_lseg_invalid_or_return(lseg, tmp_list))
2f215968
TM		 2333 continue;
2334 remaining++;
016256df 2335 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2336 }
6d597e17 2337
e0b7d420 2338 if (remaining) {
6d597e17 2339 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2340 return -EBUSY;
2341 }
6d597e17 2342
e0b7d420
TM		 2343 if (!list_empty(&lo->plh_return_segs)) {
2344 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2345 return 0;
2346 }
2347
2348 return -ENOENT;
016256df
PT		 2349}
2350
2351void pnfs_error_mark_layout_for_return(struct inode *inode,
2352 struct pnfs_layout_segment *lseg)
2353{
2354 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2355 struct pnfs_layout_range range = {
2356 .iomode = lseg->pls_range.iomode,
2357 .offset = 0,
2358 .length = NFS4_MAX_UINT64,
2359 };
10335556 2360 bool return_now = false;
016256df
PT		 2361
2362 spin_lock(&inode->i_lock);
bdebfccd
TM		 2363 if (!pnfs_layout_is_valid(lo)) {
2364 spin_unlock(&inode->i_lock);
2365 return;
2366 }
2d6cf5ab 2367 pnfs_set_plh_return_info(lo, range.iomode, 0);
016256df
PT		 2368 /*
2369 * mark all matching lsegs so that we are sure to have no live
2370 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2371 * for how it works.
2372 */
e0b7d420 2373 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
10335556 2374 nfs4_stateid stateid;
e5fd1904 2375 enum pnfs_iomode iomode;
10335556 2376
e5fd1904 2377 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2378 spin_unlock(&inode->i_lock);
2379 if (return_now)
2380 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2381 } else {
2382 spin_unlock(&inode->i_lock);
2383 nfs_commit_inode(inode, 0);
2384 }
016256df
PT		 2385}
2386EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2387
b3230e80
TM		 2388void
2389pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2390{
2391 if (pgio->pg_lseg == NULL ||
2392 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2393 return;
2394 pnfs_put_lseg(pgio->pg_lseg);
2395 pgio->pg_lseg = NULL;
2396}
2397EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2398
08cb5b0f
BC		 2399/*
2400 * Check for any intersection between the request and the pgio->pg_lseg,
2401 * and if none, put this pgio->pg_lseg away.
2402 */
2403static void
2404pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2405{
2406 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2407 pnfs_put_lseg(pgio->pg_lseg);
2408 pgio->pg_lseg = NULL;
2409 }
2410}
2411
d8007d4d
TM		 2412void
2413pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2414{
1fd937bd
PT		 2415 u64 rd_size = req->wb_bytes;
2416
b3230e80 2417 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2418 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2419 if (pgio->pg_lseg == NULL) {
2420 if (pgio->pg_dreq == NULL)
2421 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2422 else
2423 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2424
2425 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2426 req->wb_context,
2427 req_offset(req),
2428 rd_size,
2429 IOMODE_READ,
c7d73af2 2430 false,
cb5d04bc 2431 GFP_KERNEL);
d600ad1f
PT		 2432 if (IS_ERR(pgio->pg_lseg)) {
2433 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2434 pgio->pg_lseg = NULL;
2435 return;
2436 }
cb5d04bc 2437 }
e885de1a
TM		 2438 /* If no lseg, fall back to read through mds */
2439 if (pgio->pg_lseg == NULL)
1f945357 2440 nfs_pageio_reset_read_mds(pgio);
e885de1a 2441
d8007d4d
TM		 2442}
2443EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2444
2445void
6296556f
PT		 2446pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2447 struct nfs_page *req, u64 wb_size)
d8007d4d 2448{
b3230e80 2449 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2450 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2451 if (pgio->pg_lseg == NULL) {
cb5d04bc
PT		 2452 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2453 req->wb_context,
2454 req_offset(req),
2455 wb_size,
2456 IOMODE_RW,
c7d73af2 2457 false,
cb5d04bc 2458 GFP_NOFS);
d600ad1f
PT		 2459 if (IS_ERR(pgio->pg_lseg)) {
2460 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2461 pgio->pg_lseg = NULL;
2462 return;
2463 }
2464 }
e885de1a
TM		 2465 /* If no lseg, fall back to write through mds */
2466 if (pgio->pg_lseg == NULL)
1f945357 2467 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2468}
2469EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2470
180bb5ec
WAA		 2471void
2472pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2473{
2474 if (desc->pg_lseg) {
2475 pnfs_put_lseg(desc->pg_lseg);
2476 desc->pg_lseg = NULL;
2477 }
2478}
2479EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2480
b4fdac1a
WAA		 2481/*
2482 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2483 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2484 */
2485size_t
a7d42ddb
WAA		 2486pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2487 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2488{
0f9c429e 2489 unsigned int size;
c5e20cb7 2490 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2491
2492 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2493 if (!size)
2494 return 0;
94ad1c80 2495
19982ba8 2496 /*
c5e20cb7
WAA		 2497 * 'size' contains the number of bytes left in the current page (up
2498 * to the original size asked for in @req->wb_bytes).
2499 *
2500 * Calculate how many bytes are left in the layout segment
2501 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2502 *
2503 * Please also note that 'end_offset' is actually the offset of the
2504 * first byte that lies outside the pnfs_layout_range. FIXME?
2505 *
2506 */
19b54848 2507 if (pgio->pg_lseg) {
17822b20 2508 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2509 pgio->pg_lseg->pls_range.length);
2510 req_start = req_offset(req);
08cb5b0f 2511
c5e20cb7 2512 /* start of request is past the last byte of this segment */
08cb5b0f 2513 if (req_start >= seg_end)
19b54848 2514 return 0;
c5e20cb7
WAA		 2515
2516 /* adjust 'size' iff there are fewer bytes left in the
2517 * segment than what nfs_generic_pg_test returned */
2518 seg_left = seg_end - req_start;
2519 if (seg_left < size)
2520 size = (unsigned int)seg_left;
19b54848 2521 }
0f9c429e 2522
19b54848 2523 return size;
94ad1c80 2524}
89a58e32 2525EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2526
53113ad3 2527int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2528{
2529 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2530
2531 /* Resend all requests through the MDS */
53113ad3
WAA		 2532 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2533 hdr->completion_ops);
c7070113 2534 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2535 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2536}
e7dd79af 2537EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2538
d45f60c6 2539static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2540{
cd841605
FI		 2541
2542 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2543 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2544 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2545 pnfs_return_layout(hdr->inode);
1acbbb4e 2546 }
6c75dc0d 2547 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2548 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2549}
2550
d20581aa
BH		 2551/*
2552 * Called by non rpc-based layout drivers
2553 */
d45f60c6 2554void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2555{
f8417b48 2556 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2557 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2558 hdr->mds_offset + hdr->res.count);
d45f60c6 2559 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2560 }
2561 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2562 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2563 pnfs_ld_handle_write_error(hdr);
2564 hdr->mds_ops->rpc_release(hdr);
44b83799 2565}
d20581aa 2566EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2567
dce81290
TM		 2568static void
2569pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2570 struct nfs_pgio_header *hdr)
dce81290 2571{
48d635f1 2572 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2573
6c75dc0d 2574 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2575 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2576 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2577 mirror->pg_recoalesce = 1;
6c75dc0d 2578 }
ba4a76f7 2579 hdr->completion_ops->completion(hdr);
dce81290
TM		 2580}
2581
2582static enum pnfs_try_status
d45f60c6 2583pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2584 const struct rpc_call_ops *call_ops,
2585 struct pnfs_layout_segment *lseg,
2586 int how)
0382b744 2587{
cd841605 2588 struct inode *inode = hdr->inode;
0382b744
AA		 2589 enum pnfs_try_status trypnfs;
2590 struct nfs_server *nfss = NFS_SERVER(inode);
2591
cd841605 2592 hdr->mds_ops = call_ops;
0382b744
AA		 2593
2594 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2595 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2596 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2597 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2598 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2599 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2600 return trypnfs;
2601}
2602
dce81290 2603static void
7f714720
WAA		 2604pnfs_do_write(struct nfs_pageio_descriptor *desc,
2605 struct nfs_pgio_header *hdr, int how)
dce81290 2606{
dce81290
TM		 2607 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2608 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2609 enum pnfs_try_status trypnfs;
dce81290 2610
d45f60c6 2611 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2612 switch (trypnfs) {
2613 case PNFS_NOT_ATTEMPTED:
d45f60c6 2614 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2615 case PNFS_ATTEMPTED:
2616 break;
2617 case PNFS_TRY_AGAIN:
2618 /* cleanup hdr and prepare to redo pnfs */
2619 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2620 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2621 list_splice_init(&hdr->pages, &mirror->pg_list);
2622 mirror->pg_recoalesce = 1;
2623 }
2624 hdr->mds_ops->rpc_release(hdr);
2625 }
dce81290
TM		 2626}
2627
6c75dc0d
FI		 2628static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2629{
9369a431 2630 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2631 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2632}
2633
dce81290
TM		 2634int
2635pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2636{
6c75dc0d 2637 struct nfs_pgio_header *hdr;
dce81290
TM		 2638 int ret;
2639
1e7f3a48
WAA		 2640 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2641 if (!hdr) {
2bff2288
PT		 2642 desc->pg_error = -ENOMEM;
2643 return desc->pg_error;
dce81290 2644 }
6c75dc0d 2645 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2646
9369a431 2647 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2648 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2649 if (!ret)
7f714720 2650 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2651
6c75dc0d 2652 return ret;
dce81290
TM		 2653}
2654EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2655
53113ad3 2656int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2657{
2658 struct nfs_pageio_descriptor pgio;
2659
1acbbb4e 2660 /* Resend all requests through the MDS */
53113ad3
WAA		 2661 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2662 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2663}
e7dd79af 2664EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2665
d45f60c6 2666static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2667{
cd841605
FI		 2668 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2669 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2670 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2671 pnfs_return_layout(hdr->inode);
1acbbb4e 2672 }
4db6e0b7 2673 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2674 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2675}
2676
d20581aa
BH		 2677/*
2678 * Called by non rpc-based layout drivers
2679 */
d45f60c6 2680void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2681{
bfc505de 2682 if (likely(!hdr->pnfs_error))
d45f60c6 2683 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2684 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2685 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2686 pnfs_ld_handle_read_error(hdr);
2687 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2688}
2689EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2690
493292dd
TM		 2691static void
2692pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2693 struct nfs_pgio_header *hdr)
493292dd 2694{
48d635f1 2695 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2696
4db6e0b7 2697 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2698 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2699 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2700 mirror->pg_recoalesce = 1;
4db6e0b7 2701 }
ba4a76f7 2702 hdr->completion_ops->completion(hdr);
493292dd
TM		 2703}
2704
64419a9b
AA		 2705/*
2706 * Call the appropriate parallel I/O subsystem read function.
2707 */
493292dd 2708static enum pnfs_try_status
d45f60c6 2709pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2710 const struct rpc_call_ops *call_ops,
2711 struct pnfs_layout_segment *lseg)
64419a9b 2712{
cd841605 2713 struct inode *inode = hdr->inode;
64419a9b
AA		 2714 struct nfs_server *nfss = NFS_SERVER(inode);
2715 enum pnfs_try_status trypnfs;
2716
cd841605 2717 hdr->mds_ops = call_ops;
64419a9b
AA		 2718
2719 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2720 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2721
d45f60c6 2722 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2723 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2724 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2725 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2726 return trypnfs;
2727}
863a3c6c 2728
ceb11e13 2729/* Resend all requests through pnfs. */
1b1bc66b 2730void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2731{
2732 struct nfs_pageio_descriptor pgio;
2733
1b1bc66b 2734 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2735 /* Prevent deadlocks with layoutreturn! */
2736 pnfs_put_lseg(hdr->lseg);
2737 hdr->lseg = NULL;
2738
1b1bc66b
WAA		 2739 nfs_pageio_init_read(&pgio, hdr->inode, false,
2740 hdr->completion_ops);
2741 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2742 }
ceb11e13
PT		 2743}
2744EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2745
493292dd 2746static void
7f714720 2747pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2748{
493292dd
TM		 2749 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2750 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2751 enum pnfs_try_status trypnfs;
493292dd 2752
d45f60c6 2753 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2754 switch (trypnfs) {
2755 case PNFS_NOT_ATTEMPTED:
d45f60c6 2756 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2757 case PNFS_ATTEMPTED:
2758 break;
2759 case PNFS_TRY_AGAIN:
2760 /* cleanup hdr and prepare to redo pnfs */
2761 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2762 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2763 list_splice_init(&hdr->pages, &mirror->pg_list);
2764 mirror->pg_recoalesce = 1;
2765 }
2766 hdr->mds_ops->rpc_release(hdr);
2767 }
493292dd
TM		 2768}
2769
4db6e0b7
FI		 2770static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2771{
9369a431 2772 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2773 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2774}
2775
493292dd
TM		 2776int
2777pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2778{
4db6e0b7 2779 struct nfs_pgio_header *hdr;
493292dd
TM		 2780 int ret;
2781
1e7f3a48
WAA		 2782 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2783 if (!hdr) {
2bff2288
PT		 2784 desc->pg_error = -ENOMEM;
2785 return desc->pg_error;
493292dd 2786 }
4db6e0b7 2787 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2788 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2789 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2790 if (!ret)
7f714720 2791 pnfs_do_read(desc, hdr);
4db6e0b7 2792 return ret;
493292dd
TM		 2793}
2794EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2795
71244d9b
TM		 2796static void pnfs_clear_layoutcommitting(struct inode *inode)
2797{
2798 unsigned long *bitlock = &NFS_I(inode)->flags;
2799
2800 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2801 smp_mb__after_atomic();
71244d9b
TM		 2802 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2803}
2804
863a3c6c 2805/*
a9bae566 2806 * There can be multiple RW segments.
863a3c6c 2807 */
a9bae566 2808static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2809{
a9bae566 2810 struct pnfs_layout_segment *lseg;
863a3c6c 2811
a9bae566
PT		 2812 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2813 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2814 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2815 list_add(&lseg->pls_lc_list, listp);
2816 }
863a3c6c
AA		 2817}
2818
a073dbff
TM		 2819static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2820{
2821 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2822
2823 /* Matched by references in pnfs_set_layoutcommit */
2824 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2825 list_del_init(&lseg->pls_lc_list);
2826 pnfs_put_lseg(lseg);
2827 }
2828
71244d9b 2829 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2830}
2831
1b0ae068
PT		 2832void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2833{
b9e028fd 2834 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2835}
2836EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2837
863a3c6c 2838void
67af7611
TM		 2839pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2840 loff_t end_pos)
863a3c6c 2841{
cd841605 2842 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2843 bool mark_as_dirty = false;
863a3c6c 2844
cd841605 2845 spin_lock(&inode->i_lock);
863a3c6c 2846 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2847 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2848 mark_as_dirty = true;
863a3c6c 2849 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2850 __func__, inode->i_ino);
29559b11
TM		 2851 } else if (end_pos > nfsi->layout->plh_lwb)
2852 nfsi->layout->plh_lwb = end_pos;
67af7611 2853 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2854 /* references matched in nfs4_layoutcommit_release */
67af7611 2855 pnfs_get_lseg(lseg);
a9bae566 2856 }
cd841605 2857 spin_unlock(&inode->i_lock);
acff5880 2858 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2859 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2860
2861 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2862 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2863 if (mark_as_dirty)
cd841605 2864 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2865}
2866EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2867
db29c089
AA		 2868void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2869{
2870 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2871
2872 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2873 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2874 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2875}
2876
de4b15c7
AA		 2877/*
2878 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2879 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2880 * data to disk to allow the server to recover the data if it crashes.
2881 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2882 * is off, and a COMMIT is sent to a data server, or
2883 * if WRITEs to a data server return NFS_DATA_SYNC.
2884 */
863a3c6c 2885int
ef311537 2886pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2887{
5f919c9f 2888 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2889 struct nfs4_layoutcommit_data *data;
2890 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2891 loff_t end_pos;
71244d9b 2892 int status;
863a3c6c 2893
71244d9b 2894 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2895 return 0;
2896
71244d9b 2897 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2898
71244d9b 2899 status = -EAGAIN;
92407e75 2900 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2901 if (!sync)
2902 goto out;
74316201 2903 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2904 NFS_INO_LAYOUTCOMMITTING,
2905 nfs_wait_bit_killable,
2906 TASK_KILLABLE);
92407e75 2907 if (status)
71244d9b 2908 goto out;
92407e75
PT		 2909 }
2910
71244d9b
TM		 2911 status = -ENOMEM;
2912 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2913 data = kzalloc(sizeof(*data), GFP_NOFS);
2914 if (!data)
2915 goto clear_layoutcommitting;
2916
2917 status = 0;
de4b15c7 2918 spin_lock(&inode->i_lock);
71244d9b
TM		 2919 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2920 goto out_unlock;
a9bae566 2921
71244d9b 2922 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 2923 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 2924
acff5880 2925 end_pos = nfsi->layout->plh_lwb;
863a3c6c 2926
f597c537 2927 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 2928 spin_unlock(&inode->i_lock);
2929
2930 data->args.inode = inode;
a52458b4 2931 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 2932 nfs_fattr_init(&data->fattr);
2933 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2934 data->res.fattr = &data->fattr;
2e18d4d8
TM		 2935 if (end_pos != 0)
2936 data->args.lastbytewritten = end_pos - 1;
2937 else
2938 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 2939 data->res.server = NFS_SERVER(inode);
2940
5f919c9f
CH		 2941 if (ld->prepare_layoutcommit) {
2942 status = ld->prepare_layoutcommit(&data->args);
2943 if (status) {
a52458b4 2944 put_cred(data->cred);
5f919c9f 2945 spin_lock(&inode->i_lock);
29559b11
TM		 2946 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2947 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 2948 nfsi->layout->plh_lwb = end_pos;
3471648a 2949 goto out_unlock;
5f919c9f
CH		 2950 }
2951 }
2952
2953
863a3c6c
AA		 2954 status = nfs4_proc_layoutcommit(data, sync);
2955out:
92407e75
PT		 2956 if (status)
2957 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2958 dprintk("<-- %s status %d\n", __func__, status);
2959 return status;
71244d9b
TM		 2960out_unlock:
2961 spin_unlock(&inode->i_lock);
92407e75 2962 kfree(data);
71244d9b
TM		 2963clear_layoutcommitting:
2964 pnfs_clear_layoutcommitting(inode);
92407e75 2965 goto out;
863a3c6c 2966}
72cff449 2967EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 2968
5bb89b47
TM		 2969int
2970pnfs_generic_sync(struct inode *inode, bool datasync)
2971{
2972 return pnfs_layoutcommit_inode(inode, true);
2973}
2974EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2975
82be417a
AA		 2976struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2977{
2978 struct nfs4_threshold *thp;
2979
2980 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2981 if (!thp) {
2982 dprintk("%s mdsthreshold allocation failed\n", __func__);
2983 return NULL;
2984 }
2985 return thp;
2986}
8733408d 2987
865a7ecb 2988#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 2989int
c8ad8894 2990pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 2991{
2992 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2993 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 2994 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 2995 struct nfs42_layoutstat_data *data;
2996 struct pnfs_layout_hdr *hdr;
2997 int status = 0;
2998
2999 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3000 goto out;
3001
6c5a0d89
TM		 3002 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3003 goto out;
3004
1bfe3b25
PT		 3005 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3006 goto out;
3007
8733408d
PT		 3008 spin_lock(&inode->i_lock);
3009 if (!NFS_I(inode)->layout) {
3010 spin_unlock(&inode->i_lock);
f538d0ba 3011 goto out_clear_layoutstats;
8733408d
PT		 3012 }
3013 hdr = NFS_I(inode)->layout;
3014 pnfs_get_layout_hdr(hdr);
3015 spin_unlock(&inode->i_lock);
3016
c8ad8894 3017 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3018 if (!data) {
3019 status = -ENOMEM;
3020 goto out_put;
3021 }
3022
3023 data->args.fh = NFS_FH(inode);
3024 data->args.inode = inode;
8733408d
PT		 3025 status = ld->prepare_layoutstats(&data->args);
3026 if (status)
3027 goto out_free;
3028
3029 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3030
3031out:
3032 dprintk("%s returns %d\n", __func__, status);
3033 return status;
3034
3035out_free:
3036 kfree(data);
3037out_put:
3038 pnfs_put_layout_hdr(hdr);
f538d0ba 3039out_clear_layoutstats:
1bfe3b25
PT		 3040 smp_mb__before_atomic();
3041 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3042 smp_mb__after_atomic();
8733408d
PT		 3043 goto out;
3044}
3045EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3046#endif
bbf58bf3
TM		 3047
3048unsigned int layoutstats_timer;
3049module_param(layoutstats_timer, uint, 0644);
3050EXPORT_SYMBOL_GPL(layoutstats_timer);
Linux 6.x block layer and io_uring tree(s)