]> Git Repo - linux.git/blame - fs/nfs/pnfs.c
projects / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mm: hugetlb_vmemmap: cleanup CONFIG_HUGETLB_PAGE_FREE_VMEMMAP*
[linux.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <[email protected]>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
7c9d845f
TM		 95const struct pnfs_layoutdriver_type *pnfs_find_layoutdriver(u32 id)
96{
97 return find_pnfs_driver(id);
98}
99
100void pnfs_put_layoutdriver(const struct pnfs_layoutdriver_type *ld)
101{
102 if (ld)
103 module_put(ld->owner);
104}
105
85e174ba
RL		 106void
107unset_pnfs_layoutdriver(struct nfs_server *nfss)
108{
738fd0f3
BH		 109 if (nfss->pnfs_curr_ld) {
110 if (nfss->pnfs_curr_ld->clear_layoutdriver)
111 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 112 /* Decrement the MDS count. Purge the deviceid cache if zero */
113 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
114 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 115 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 116 }
85e174ba
RL		 117 nfss->pnfs_curr_ld = NULL;
118}
119
ca440c38
JL		 120/*
121 * When the server sends a list of layout types, we choose one in the order
122 * given in the list below.
123 *
124 * FIXME: should this list be configurable in some fashion? module param?
125 * mount option? something else?
126 */
127static const u32 ld_prefs[] = {
128 LAYOUT_SCSI,
129 LAYOUT_BLOCK_VOLUME,
130 LAYOUT_OSD2_OBJECTS,
131 LAYOUT_FLEX_FILES,
132 LAYOUT_NFSV4_1_FILES,
133 0
134};
135
136static int
137ld_cmp(const void *e1, const void *e2)
138{
139 u32 ld1 = *((u32 *)e1);
140 u32 ld2 = *((u32 *)e2);
141 int i;
142
143 for (i = 0; ld_prefs[i] != 0; i++) {
144 if (ld1 == ld_prefs[i])
145 return -1;
146
147 if (ld2 == ld_prefs[i])
148 return 1;
149 }
150 return 0;
151}
152
85e174ba
RL		 153/*
154 * Try to set the server's pnfs module to the pnfs layout type specified by id.
155 * Currently only one pNFS layout driver per filesystem is supported.
156 *
3132e49e 157 * @ids array of layout types supported by MDS.
85e174ba
RL		 158 */
159void
738fd0f3 160set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 161 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 162{
163 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 164 u32 id;
ca440c38 165 int i;
85e174ba 166
19274716
AS		 167 if (fsinfo->nlayouttypes == 0)
168 goto out_no_driver;
85e174ba
RL		 169 if (!(server->nfs_client->cl_exchange_flags &
170 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 171 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
172 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 173 goto out_no_driver;
174 }
3132e49e 175
ca440c38
JL		 176 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
177 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 178
ca440c38
JL		 179 for (i = 0; i < fsinfo->nlayouttypes; i++) {
180 id = fsinfo->layouttype[i];
85e174ba 181 ld_type = find_pnfs_driver(id);
ca440c38
JL		 182 if (!ld_type) {
183 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
184 id);
185 ld_type = find_pnfs_driver(id);
186 }
187 if (ld_type)
188 break;
85e174ba 189 }
3132e49e
JL		 190
191 if (!ld_type) {
ca440c38 192 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 193 goto out_no_driver;
194 }
195
85e174ba 196 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 197 if (ld_type->set_layoutdriver
198 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 199 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
200 "driver %u.\n", __func__, id);
738fd0f3
BH		 201 module_put(ld_type->owner);
202 goto out_no_driver;
203 }
2a4c8994
TM		 204 /* Bump the MDS count */
205 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 206
85e174ba
RL		 207 dprintk("%s: pNFS module for %u set\n", __func__, id);
208 return;
209
210out_no_driver:
211 dprintk("%s: Using NFSv4 I/O\n", __func__);
212 server->pnfs_curr_ld = NULL;
213}
02c35fca
FI		 214
215int
216pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
217{
218 int status = -EINVAL;
219 struct pnfs_layoutdriver_type *tmp;
220
221 if (ld_type->id == 0) {
a030889a 222 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 223 return status;
224 }
b1f69b75 225 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 226 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 227 "alloc_lseg and free_lseg.\n", __func__);
228 return status;
229 }
02c35fca
FI		 230
231 spin_lock(&pnfs_spinlock);
232 tmp = find_pnfs_driver_locked(ld_type->id);
233 if (!tmp) {
234 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
235 status = 0;
236 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
237 ld_type->name);
238 } else {
a030889a 239 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 240 __func__, ld_type->id);
241 }
242 spin_unlock(&pnfs_spinlock);
243
244 return status;
245}
246EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
247
248void
249pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
250{
251 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
252 spin_lock(&pnfs_spinlock);
253 list_del(&ld_type->pnfs_tblid);
254 spin_unlock(&pnfs_spinlock);
255}
256EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 257
b1f69b75
AA		 258/*
259 * pNFS client layout cache
260 */
261
cc6e5340 262/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 263void
70c3bd2b 264pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 265{
2b28a7be 266 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 267}
268
636fb9c8
BH		 269static struct pnfs_layout_hdr *
270pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
271{
272 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 273 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 274}
275
276static void
277pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
278{
9c626381
TM		 279 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
280 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
281
cf6605d1 282 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
9c626381
TM		 283 struct nfs_client *clp = server->nfs_client;
284
285 spin_lock(&clp->cl_lock);
cf6605d1 286 list_del_rcu(&lo->plh_layouts);
9c626381
TM		 287 spin_unlock(&clp->cl_lock);
288 }
a52458b4 289 put_cred(lo->plh_lc_cred);
57934278 290 return ld->free_layout_hdr(lo);
636fb9c8
BH		 291}
292
e5e94017 293static void
6622c3ea 294pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 295{
bb346f63 296 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 297 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 298 nfsi->layout = NULL;
299 /* Reset MDS Threshold I/O counters */
300 nfsi->write_io = 0;
301 nfsi->read_io = 0;
e5e94017
BH		 302}
303
b1f69b75 304void
70c3bd2b 305pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 306{
9c6376eb 307 struct inode *inode;
b6d49ecd 308 unsigned long i_state;
cc6e5340 309
9c6376eb
TM		 310 if (!lo)
311 return;
312 inode = lo->plh_inode;
13c13a6a
TM		 313 pnfs_layoutreturn_before_put_layout_hdr(lo);
314
2b28a7be 315 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 316 if (!list_empty(&lo->plh_segs))
317 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 318 pnfs_detach_layout_hdr(lo);
b6d49ecd 319 i_state = inode->i_state;
cc6e5340 320 spin_unlock(&inode->i_lock);
6622c3ea 321 pnfs_free_layout_hdr(lo);
b6d49ecd
TM		 322 /* Notify pnfs_destroy_layout_final() that we're done */
323 if (i_state & (I_FREEING | I_CLEAR))
324 wake_up_var(lo);
cc6e5340 325 }
974cec8c
AA		 326}
327
b5fdf841
TM		 328static struct inode *
329pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
330{
331 struct inode *inode = igrab(lo->plh_inode);
332 if (inode)
333 return inode;
334 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
335 return NULL;
336}
337
1bcf34fd
TM		 338/*
339 * Compare 2 layout stateid sequence ids, to see which is newer,
340 * taking into account wraparound issues.
341 */
342static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
343{
344 return (s32)(s1 - s2) > 0;
345}
346
347static void pnfs_barrier_update(struct pnfs_layout_hdr *lo, u32 newseq)
348{
45baadaa 349 if (pnfs_seqid_is_newer(newseq, lo->plh_barrier) || !lo->plh_barrier)
1bcf34fd
TM		 350 lo->plh_barrier = newseq;
351}
352
4aab9732
TM		 353static void
354pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
355 u32 seq)
356{
357 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
358 iomode = IOMODE_ANY;
359 lo->plh_return_iomode = iomode;
360 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
e20772cb
TM		 361 /*
362 * We must set lo->plh_return_seq to avoid livelocks with
363 * pnfs_layout_need_return()
364 */
365 if (seq == 0)
366 seq = be32_to_cpu(lo->plh_stateid.seqid);
367 if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq))
4aab9732 368 lo->plh_return_seq = seq;
e20772cb 369 pnfs_barrier_update(lo, seq);
4aab9732
TM		 370}
371
ae5a459d
TM		 372static void
373pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
374{
5466d214 375 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 376 lo->plh_return_iomode = 0;
377 lo->plh_return_seq = 0;
378 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 379 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
380 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
381 continue;
382 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
383 }
ae5a459d
TM		 384}
385
362fb578
TM		 386static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
387{
388 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
389 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
390 smp_mb__after_atomic();
391 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
392 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
393}
394
fe1cf946
TM		 395static void
396pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
397 struct list_head *free_me)
398{
399 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
400 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
401 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
402 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
403 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
404 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
405}
406
7380020e 407/*
30cb3ee2
TM		 408 * Update the seqid of a layout stateid after receiving
409 * NFS4ERR_OLD_STATEID
7380020e 410 */
30cb3ee2 411bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
ecf84026
TM		 412 struct pnfs_layout_range *dst_range,
413 struct inode *inode)
7380020e
TM		 414{
415 struct pnfs_layout_hdr *lo;
c16467dc
TM		 416 struct pnfs_layout_range range = {
417 .iomode = IOMODE_ANY,
418 .offset = 0,
419 .length = NFS4_MAX_UINT64,
420 };
7380020e 421 bool ret = false;
c16467dc
TM		 422 LIST_HEAD(head);
423 int err;
7380020e
TM		 424
425 spin_lock(&inode->i_lock);
426 lo = NFS_I(inode)->layout;
30cb3ee2
TM		 427 if (lo && pnfs_layout_is_valid(lo) &&
428 nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
429 /* Is our call using the most recent seqid? If so, bump it */
430 if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
431 nfs4_stateid_seqid_inc(dst);
432 ret = true;
433 goto out;
434 }
435 /* Try to update the seqid to the most recent */
c16467dc
TM		 436 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
437 if (err != -EBUSY) {
438 dst->seqid = lo->plh_stateid.seqid;
ecf84026 439 *dst_range = range;
c16467dc
TM		 440 ret = true;
441 }
7380020e 442 }
30cb3ee2 443out:
7380020e 444 spin_unlock(&inode->i_lock);
c16467dc 445 pnfs_free_lseg_list(&head);
7380020e
TM		 446 return ret;
447}
448
2454dfea
TM		 449/*
450 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
451 *
452 * In order to continue using the pnfs_layout_hdr, a full recovery
453 * is required.
454 * Note that caller must hold inode->i_lock.
455 */
5f46be04 456int
2454dfea
TM		 457pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
458 struct list_head *lseg_list)
459{
460 struct pnfs_layout_range range = {
461 .iomode = IOMODE_ANY,
462 .offset = 0,
463 .length = NFS4_MAX_UINT64,
464 };
fe1cf946 465 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 466
467 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 468 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
469 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 470 pnfs_clear_layoutreturn_info(lo);
68f74479 471 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 472 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
473 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
474 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 475 return !list_empty(&lo->plh_segs);
2454dfea
TM		 476}
477
b9e028fd
TM		 478static int
479pnfs_iomode_to_fail_bit(u32 iomode)
480{
481 return iomode == IOMODE_RW ?
482 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
483}
484
485static void
3e621214 486pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 487{
25c75333 488 lo->plh_retry_timestamp = jiffies;
39e88fcf 489 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 490 refcount_inc(&lo->plh_refcount);
3e621214
TM		 491}
492
493static void
494pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
495{
496 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 497 refcount_dec(&lo->plh_refcount);
3e621214
TM		 498}
499
500static void
501pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
502{
503 struct inode *inode = lo->plh_inode;
115ce575
TM		 504 struct pnfs_layout_range range = {
505 .iomode = iomode,
506 .offset = 0,
507 .length = NFS4_MAX_UINT64,
508 };
509 LIST_HEAD(head);
3e621214
TM		 510
511 spin_lock(&inode->i_lock);
512 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 513 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 514 spin_unlock(&inode->i_lock);
115ce575 515 pnfs_free_lseg_list(&head);
b9e028fd
TM		 516 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
517 iomode == IOMODE_RW ? "RW" : "READ");
518}
519
520static bool
521pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
522{
25c75333 523 unsigned long start, end;
3e621214
TM		 524 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
525
526 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 527 return false;
528 end = jiffies;
529 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
530 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
531 /* It is time to retry the failed layoutgets */
3e621214 532 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 533 return false;
534 }
535 return true;
b9e028fd
TM		 536}
537
974cec8c 538static void
119cef97
TM		 539pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
540 const struct pnfs_layout_range *range,
541 const nfs4_stateid *stateid)
974cec8c 542{
566052c5 543 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 544 INIT_LIST_HEAD(&lseg->pls_lc_list);
a9901899 545 INIT_LIST_HEAD(&lseg->pls_commits);
eba6dd69 546 refcount_set(&lseg->pls_refcount, 1);
4541d16c 547 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 548 lseg->pls_layout = lo;
119cef97
TM		 549 lseg->pls_range = *range;
550 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 551}
552
905ca191 553static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 554{
68f74479
TM		 555 if (lseg != NULL) {
556 struct inode *inode = lseg->pls_layout->plh_inode;
557 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
558 }
974cec8c
AA		 559}
560
d684d2ae 561static void
57036a37
TM		 562pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
563 struct pnfs_layout_segment *lseg)
d684d2ae 564{
d20581aa 565 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 566 list_del_init(&lseg->pls_list);
8f0d27dc 567 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 568 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 569 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
570 return;
7b650994
TM		 571 if (list_empty(&lo->plh_segs) &&
572 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
573 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 574 if (atomic_read(&lo->plh_outstanding) == 0)
575 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 576 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 577 }
d684d2ae
FI		 578}
579
68f74479
TM		 580static bool
581pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
582 struct pnfs_layout_segment *lseg)
583{
584 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
585 pnfs_layout_is_valid(lo)) {
4aab9732 586 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 587 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
588 return true;
589 }
590 return false;
591}
592
bae724ef 593void
9369a431 594pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 595{
57036a37 596 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 597 struct inode *inode;
598
599 if (!lseg)
600 return;
601
4541d16c 602 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 603 refcount_read(&lseg->pls_refcount),
4541d16c 604 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 605
57036a37
TM		 606 lo = lseg->pls_layout;
607 inode = lo->plh_inode;
4ef2e4f8 608
eba6dd69 609 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
8f0d27dc 610 pnfs_get_layout_hdr(lo);
4ef2e4f8 611 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 612 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
613 lseg = NULL;
4ef2e4f8
TM		 614 spin_unlock(&inode->i_lock);
615 pnfs_free_lseg(lseg);
616 pnfs_put_layout_hdr(lo);
4541d16c 617 }
4541d16c 618}
9369a431 619EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 620
fb3296eb
BH		 621/*
622 * is l2 fully contained in l1?
623 * start1 end1
624 * [----------------------------------)
625 * start2 end2
626 * [----------------)
627 */
3cb2df17 628static bool
7dc0ac70 629pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 630 const struct pnfs_layout_range *l2)
fb3296eb
BH		 631{
632 u64 start1 = l1->offset;
17822b20 633 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 634 u64 start2 = l2->offset;
17822b20 635 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 636
637 return (start1 <= start2) && (end1 >= end2);
638}
639
24956804
TM		 640static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
641 struct list_head *tmp_list)
642{
eba6dd69 643 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 644 return false;
645 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
646 list_add(&lseg->pls_list, tmp_list);
647 return true;
648}
649
4541d16c
FI		 650/* Returns 1 if lseg is removed from list, 0 otherwise */
651static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
652 struct list_head *tmp_list)
653{
654 int rv = 0;
655
656 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
657 /* Remove the reference keeping the lseg in the
658 * list. It will now be removed when all
659 * outstanding io is finished.
660 */
d684d2ae 661 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 662 refcount_read(&lseg->pls_refcount));
24956804 663 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 664 rv = 1;
4541d16c
FI		 665 }
666 return rv;
667}
668
e036f464
TM		 669static bool
670pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
671 const struct pnfs_layout_range *recall_range)
672{
673 return (recall_range->iomode == IOMODE_ANY ||
674 lseg_range->iomode == recall_range->iomode) &&
675 pnfs_lseg_range_intersecting(lseg_range, recall_range);
676}
677
678static bool
679pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
680 const struct pnfs_layout_range *recall_range,
681 u32 seq)
682{
683 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
684 return false;
685 if (recall_range == NULL)
686 return true;
687 return pnfs_should_free_range(&lseg->pls_range, recall_range);
688}
689
6d597e17
JL		 690/**
691 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
692 * @lo: layout header containing the lsegs
693 * @tmp_list: list head where doomed lsegs should go
694 * @recall_range: optional recall range argument to match (may be NULL)
695 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
696 *
697 * Walk the list of lsegs in the layout header, and tear down any that should
698 * be destroyed. If "recall_range" is specified then the segment must match
699 * that range. If "seq" is non-zero, then only match segments that were handed
700 * out at or before that sequence.
701 *
702 * Returns number of matching invalid lsegs remaining in list after scanning
703 * it and purging them.
4541d16c 704 */
43f1b3da 705int
49a85061 706pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 707 struct list_head *tmp_list,
6d597e17
JL		 708 const struct pnfs_layout_range *recall_range,
709 u32 seq)
974cec8c
AA		 710{
711 struct pnfs_layout_segment *lseg, *next;
71b39854 712 int remaining = 0;
974cec8c
AA		 713
714 dprintk("%s:Begin lo %p\n", __func__, lo);
715
8006bfba 716 if (list_empty(&lo->plh_segs))
38511722 717 return 0;
4541d16c 718 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 719 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 720 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 721 "offset %llu length %llu\n", __func__,
6d597e17
JL		 722 lseg, lseg->pls_range.iomode, lseg->pls_seq,
723 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 724 if (!mark_lseg_invalid(lseg, tmp_list))
725 remaining++;
4541d16c 726 }
71b39854
TM		 727 dprintk("%s:Return %i\n", __func__, remaining);
728 return remaining;
974cec8c
AA		 729}
730
68f74479
TM		 731static void
732pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
733 struct list_head *free_me,
734 const struct pnfs_layout_range *range,
735 u32 seq)
736{
737 struct pnfs_layout_segment *lseg, *next;
738
739 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
740 if (pnfs_match_lseg_recall(lseg, range, seq))
741 list_move_tail(&lseg->pls_list, free_me);
742 }
743}
744
f49f9baa 745/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 746void
4541d16c 747pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 748{
4541d16c 749 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 750
751 if (list_empty(free_me))
752 return;
753
4541d16c 754 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 755 list_del(&lseg->pls_list);
905ca191 756 pnfs_free_lseg(lseg);
974cec8c
AA		 757 }
758}
759
b6d49ecd 760static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi)
e5e94017
BH		 761{
762 struct pnfs_layout_hdr *lo;
974cec8c 763 LIST_HEAD(tmp_list);
e5e94017
BH		 764
765 spin_lock(&nfsi->vfs_inode.i_lock);
766 lo = nfsi->layout;
767 if (lo) {
3e621214 768 pnfs_get_layout_hdr(lo);
2454dfea 769 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 770 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
771 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
772 spin_unlock(&nfsi->vfs_inode.i_lock);
773 pnfs_free_lseg_list(&tmp_list);
1f18b82c 774 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 775 pnfs_put_layout_hdr(lo);
776 } else
777 spin_unlock(&nfsi->vfs_inode.i_lock);
b6d49ecd
TM		 778 return lo;
779}
780
781void pnfs_destroy_layout(struct nfs_inode *nfsi)
782{
783 __pnfs_destroy_layout(nfsi);
974cec8c 784}
041245c8 785EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 786
b6d49ecd
TM		 787static bool pnfs_layout_removed(struct nfs_inode *nfsi,
788 struct pnfs_layout_hdr *lo)
789{
790 bool ret;
791
792 spin_lock(&nfsi->vfs_inode.i_lock);
793 ret = nfsi->layout != lo;
794 spin_unlock(&nfsi->vfs_inode.i_lock);
795 return ret;
796}
797
798void pnfs_destroy_layout_final(struct nfs_inode *nfsi)
799{
800 struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi);
801
802 if (lo)
803 wait_var_event(lo, pnfs_layout_removed(nfsi, lo));
804}
805
fd9a8d71
TM		 806static bool
807pnfs_layout_add_bulk_destroy_list(struct inode *inode,
808 struct list_head *layout_list)
974cec8c
AA		 809{
810 struct pnfs_layout_hdr *lo;
fd9a8d71 811 bool ret = false;
974cec8c 812
fd9a8d71
TM		 813 spin_lock(&inode->i_lock);
814 lo = NFS_I(inode)->layout;
815 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
816 pnfs_get_layout_hdr(lo);
817 list_add(&lo->plh_bulk_destroy, layout_list);
818 ret = true;
819 }
820 spin_unlock(&inode->i_lock);
821 return ret;
822}
823
824/* Caller must hold rcu_read_lock and clp->cl_lock */
825static int
826pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
827 struct nfs_server *server,
828 struct list_head *layout_list)
5085607d
TM		 829 __must_hold(&clp->cl_lock)
830 __must_hold(RCU)
fd9a8d71
TM		 831{
832 struct pnfs_layout_hdr *lo, *next;
833 struct inode *inode;
834
835 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
5085607d
TM		 836 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
837 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
838 !list_empty(&lo->plh_bulk_destroy))
b85f5620 839 continue;
5085607d
TM		 840 /* If the sb is being destroyed, just bail */
841 if (!nfs_sb_active(server->super))
842 break;
b5fdf841 843 inode = pnfs_grab_inode_layout_hdr(lo);
5085607d 844 if (inode != NULL) {
cf6605d1
TM		 845 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
846 list_del_rcu(&lo->plh_layouts);
5085607d
TM		 847 if (pnfs_layout_add_bulk_destroy_list(inode,
848 layout_list))
849 continue;
850 rcu_read_unlock();
851 spin_unlock(&clp->cl_lock);
852 iput(inode);
853 } else {
854 rcu_read_unlock();
855 spin_unlock(&clp->cl_lock);
5085607d
TM		 856 }
857 nfs_sb_deactive(server->super);
fd9a8d71
TM		 858 spin_lock(&clp->cl_lock);
859 rcu_read_lock();
860 return -EAGAIN;
861 }
862 return 0;
863}
864
865static int
866pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
867 bool is_bulk_recall)
868{
869 struct pnfs_layout_hdr *lo;
870 struct inode *inode;
fd9a8d71
TM		 871 LIST_HEAD(lseg_list);
872 int ret = 0;
873
874 while (!list_empty(layout_list)) {
875 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
876 plh_bulk_destroy);
877 dprintk("%s freeing layout for inode %lu\n", __func__,
878 lo->plh_inode->i_ino);
879 inode = lo->plh_inode;
7c5d1875
CH		 880
881 pnfs_layoutcommit_inode(inode, false);
882
fd9a8d71
TM		 883 spin_lock(&inode->i_lock);
884 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 885 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
886 if (is_bulk_recall)
887 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 888 ret = -EAGAIN;
9fd4b9fc 889 }
fd9a8d71
TM		 890 spin_unlock(&inode->i_lock);
891 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 892 /* Free all lsegs that are attached to commit buckets */
893 nfs_commit_inode(inode, 0);
fd9a8d71 894 pnfs_put_layout_hdr(lo);
5085607d 895 nfs_iput_and_deactive(inode);
fd9a8d71
TM		 896 }
897 return ret;
898}
899
900int
901pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
902 struct nfs_fsid *fsid,
903 bool is_recall)
904{
905 struct nfs_server *server;
906 LIST_HEAD(layout_list);
c47abcf8 907
974cec8c 908 spin_lock(&clp->cl_lock);
6382a441 909 rcu_read_lock();
fd9a8d71 910restart:
6382a441 911 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 912 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
913 continue;
914 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
915 server,
916 &layout_list) != 0)
917 goto restart;
6382a441
WAA		 918 }
919 rcu_read_unlock();
974cec8c
AA		 920 spin_unlock(&clp->cl_lock);
921
fd9a8d71
TM		 922 if (list_empty(&layout_list))
923 return 0;
924 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
925}
926
927int
928pnfs_destroy_layouts_byclid(struct nfs_client *clp,
929 bool is_recall)
930{
931 struct nfs_server *server;
932 LIST_HEAD(layout_list);
933
934 spin_lock(&clp->cl_lock);
935 rcu_read_lock();
936restart:
937 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
938 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
939 server,
940 &layout_list) != 0)
941 goto restart;
974cec8c 942 }
fd9a8d71
TM		 943 rcu_read_unlock();
944 spin_unlock(&clp->cl_lock);
945
946 if (list_empty(&layout_list))
947 return 0;
948 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
949}
950
951/*
9f266451 952 * Called by the state manager to remove all layouts established under an
fd9a8d71
TM		 953 * expired lease.
954 */
955void
956pnfs_destroy_all_layouts(struct nfs_client *clp)
957{
958 nfs4_deviceid_mark_client_invalid(clp);
959 nfs4_deviceid_purge_client(clp);
960
961 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 962}
963
59b56394
TM		 964static void
965pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
966{
967 const struct cred *old;
968
969 if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
970 old = xchg(&lo->plh_lc_cred, get_cred(cred));
971 put_cred(old);
972 }
973}
974
fd6002e9 975/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 976void
977pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
59b56394 978 const struct cred *cred, bool update_barrier)
b1f69b75 979{
aa95edf3
TM		 980 u32 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
981 u32 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 982
983 if (!pnfs_layout_is_valid(lo)) {
59b56394 984 pnfs_set_layout_cred(lo, cred);
2a59a041
TM		 985 nfs4_stateid_copy(&lo->plh_stateid, new);
986 lo->plh_barrier = newseq;
987 pnfs_clear_layoutreturn_info(lo);
988 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
989 return;
990 }
aa95edf3
TM		 991
992 if (pnfs_seqid_is_newer(newseq, oldseq))
f597c537 993 nfs4_stateid_copy(&lo->plh_stateid, new);
aa95edf3
TM		 994
995 if (update_barrier) {
996 pnfs_barrier_update(lo, newseq);
ecebb80b 997 return;
aa95edf3
TM		 998 }
999 /*
1000 * Because of wraparound, we want to keep the barrier
1001 * "close" to the current seqids. We really only want to
1002 * get here from a layoutget call.
1003 */
1004 if (atomic_read(&lo->plh_outstanding) == 1)
1005 pnfs_barrier_update(lo, be32_to_cpu(lo->plh_stateid.seqid));
b1f69b75
AA		 1006}
1007
cf7d63f1 1008static bool
19c54aba
TM		 1009pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
1010 const nfs4_stateid *stateid)
43f1b3da 1011{
19c54aba 1012 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 1013
d6236a98 1014 return lo->plh_barrier && pnfs_seqid_is_newer(lo->plh_barrier, seqid);
19c54aba
TM		 1015}
1016
1017/* lget is set to 1 if called from inside send_layoutget call chain */
1018static bool
e1c06f80 1019pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 1020{
f7e8917a 1021 return lo->plh_block_lgets ||
e1c06f80 1022 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 1023}
1024
5e36e2a9
FI		 1025static struct nfs_server *
1026pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
1027{
1028 struct nfs_server *server;
1029
78746a38 1030 if (inode) {
5e36e2a9 1031 server = NFS_SERVER(inode);
78746a38 1032 } else {
5e36e2a9
FI		 1033 struct dentry *parent_dir = dget_parent(ctx->dentry);
1034 server = NFS_SERVER(parent_dir->d_inode);
1035 dput(parent_dir);
1036 }
1037 return server;
1038}
1039
29a8bfe5
TM		 1040static void nfs4_free_pages(struct page **pages, size_t size)
1041{
1042 int i;
1043
1044 if (!pages)
1045 return;
1046
1047 for (i = 0; i < size; i++) {
1048 if (!pages[i])
1049 break;
1050 __free_page(pages[i]);
1051 }
1052 kfree(pages);
1053}
1054
1055static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1056{
1057 struct page **pages;
1058 int i;
1059
a2791d3a 1060 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 1061 if (!pages) {
1062 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1063 return NULL;
1064 }
1065
1066 for (i = 0; i < size; i++) {
1067 pages[i] = alloc_page(gfp_flags);
1068 if (!pages[i]) {
1069 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 1070 nfs4_free_pages(pages, i);
29a8bfe5
TM		 1071 return NULL;
1072 }
1073 }
1074
1075 return pages;
1076}
1077
587f03de 1078static struct nfs4_layoutget *
5e36e2a9 1079pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 1080 struct nfs_open_context *ctx,
2409a976 1081 const nfs4_stateid *stateid,
e144e539 1082 const struct pnfs_layout_range *range,
587f03de 1083 gfp_t gfp_flags)
e5e94017 1084{
5e36e2a9 1085 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 1086 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 1087 size_t max_pages = max_response_pages(server);
b1f69b75 1088 struct nfs4_layoutget *lgp;
b1f69b75
AA		 1089
1090 dprintk("--> %s\n", __func__);
e5e94017 1091
83026d80
JL		 1092 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1093 if (lgp == NULL)
587f03de 1094 return NULL;
83026d80 1095
28ced9a8
TM		 1096 if (max_reply_sz) {
1097 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1098 if (npages < max_pages)
1099 max_pages = npages;
1100 }
1101
dacb452d
FI		 1102 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1103 if (!lgp->args.layout.pages) {
1104 kfree(lgp);
1105 return NULL;
1106 }
1107 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1108 lgp->res.layoutp = &lgp->args.layout;
1109
d49e0d5b
TM		 1110 /* Don't confuse uninitialised result and success */
1111 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1112
1113 lgp->args.minlength = PAGE_SIZE;
1114 if (lgp->args.minlength > range->length)
1115 lgp->args.minlength = range->length;
5e36e2a9
FI		 1116 if (ino) {
1117 loff_t i_size = i_size_read(ino);
1118
1119 if (range->iomode == IOMODE_READ) {
1120 if (range->offset >= i_size)
1121 lgp->args.minlength = 0;
1122 else if (i_size - range->offset < lgp->args.minlength)
1123 lgp->args.minlength = i_size - range->offset;
1124 }
d03ab29d 1125 }
83026d80
JL		 1126 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1127 pnfs_copy_range(&lgp->args.range, range);
1128 lgp->args.type = server->pnfs_curr_ld->id;
1129 lgp->args.inode = ino;
1130 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1131 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1132 lgp->gfp_flags = gfp_flags;
63ec2b69 1133 lgp->cred = ctx->cred;
587f03de 1134 return lgp;
974cec8c
AA		 1135}
1136
29a8bfe5
TM		 1137void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1138{
1139 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1140
1141 nfs4_free_pages(lgp->args.layout.pages, max_pages);
b4e89bcb 1142 pnfs_put_layout_hdr(lgp->lo);
29a8bfe5
TM		 1143 put_nfs_open_context(lgp->args.ctx);
1144 kfree(lgp);
1145}
1146
24956804
TM		 1147static void pnfs_clear_layoutcommit(struct inode *inode,
1148 struct list_head *head)
1149{
1150 struct nfs_inode *nfsi = NFS_I(inode);
1151 struct pnfs_layout_segment *lseg, *tmp;
1152
1153 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1154 return;
1155 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1156 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1157 continue;
1158 pnfs_lseg_dec_and_remove_zero(lseg, head);
1159 }
1160}
1161
68f74479 1162void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1163 const nfs4_stateid *arg_stateid,
68f74479 1164 const struct pnfs_layout_range *range,
68f74479
TM		 1165 const nfs4_stateid *stateid)
1166{
1167 struct inode *inode = lo->plh_inode;
1168 LIST_HEAD(freeme);
1169
1170 spin_lock(&inode->i_lock);
c18d1e17 1171 if (!pnfs_layout_is_valid(lo) ||
2a974425
TM		 1172 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1173 goto out_unlock;
68f74479 1174 if (stateid) {
2a974425
TM		 1175 u32 seq = be32_to_cpu(arg_stateid->seqid);
1176
68f74479
TM		 1177 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1178 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
59b56394 1179 pnfs_set_layout_stateid(lo, stateid, NULL, true);
68f74479
TM		 1180 } else
1181 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1182out_unlock:
68f74479
TM		 1183 pnfs_clear_layoutreturn_waitbit(lo);
1184 spin_unlock(&inode->i_lock);
1185 pnfs_free_lseg_list(&freeme);
1186
1187}
1188
13c13a6a 1189static bool
e5fd1904
TM		 1190pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1191 nfs4_stateid *stateid,
44ea8dfc 1192 const struct cred **cred,
e5fd1904 1193 enum pnfs_iomode *iomode)
13c13a6a 1194{
bf0291dd
TM		 1195 /* Serialise LAYOUTGET/LAYOUTRETURN */
1196 if (atomic_read(&lo->plh_outstanding) != 0)
1197 return false;
6604b203 1198 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1199 return false;
6604b203 1200 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1201 pnfs_get_layout_hdr(lo);
1bcf34fd
TM		 1202 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1203 *cred = get_cred(lo->plh_lc_cred);
e5fd1904 1204 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
44ea8dfc
TM		 1205 if (lo->plh_return_seq != 0)
1206 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
e5fd1904
TM		 1207 if (iomode != NULL)
1208 *iomode = lo->plh_return_iomode;
1209 pnfs_clear_layoutreturn_info(lo);
1bcf34fd 1210 } else if (iomode != NULL)
e5fd1904 1211 *iomode = IOMODE_ANY;
1bcf34fd 1212 pnfs_barrier_update(lo, be32_to_cpu(stateid->seqid));
13c13a6a
TM		 1213 return true;
1214}
1215
828ed9ec
TM		 1216static void
1217pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1218 struct pnfs_layout_hdr *lo,
1219 const nfs4_stateid *stateid,
1220 enum pnfs_iomode iomode)
1221{
1222 struct inode *inode = lo->plh_inode;
1223
1224 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1225 args->inode = inode;
1226 args->range.iomode = iomode;
1227 args->range.offset = 0;
1228 args->range.length = NFS4_MAX_UINT64;
1229 args->layout = lo;
1230 nfs4_stateid_copy(&args->stateid, stateid);
1231}
1232
f40eb5d0 1233static int
44ea8dfc
TM		 1234pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
1235 const nfs4_stateid *stateid,
1236 const struct cred **pcred,
1237 enum pnfs_iomode iomode,
1238 bool sync)
f40eb5d0
PT		 1239{
1240 struct inode *ino = lo->plh_inode;
287bd3e9 1241 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0 1242 struct nfs4_layoutreturn *lrp;
44ea8dfc 1243 const struct cred *cred = *pcred;
f40eb5d0
PT		 1244 int status = 0;
1245
44ea8dfc 1246 *pcred = NULL;
63d8a41b 1247 lrp = kzalloc(sizeof(*lrp), nfs_io_gfp_mask());
f40eb5d0
PT		 1248 if (unlikely(lrp == NULL)) {
1249 status = -ENOMEM;
1250 spin_lock(&ino->i_lock);
d67ae825 1251 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0 1252 spin_unlock(&ino->i_lock);
44ea8dfc 1253 put_cred(cred);
f40eb5d0
PT		 1254 pnfs_put_layout_hdr(lo);
1255 goto out;
1256 }
1257
828ed9ec 1258 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1259 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0 1260 lrp->clp = NFS_SERVER(ino)->nfs_client;
44ea8dfc 1261 lrp->cred = cred;
287bd3e9
TM		 1262 if (ld->prepare_layoutreturn)
1263 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1264
6c16605d 1265 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1266out:
1267 dprintk("<-- %s status: %d\n", __func__, status);
1268 return status;
1269}
1270
d474f961
TM		 1271static bool
1272pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1273 enum pnfs_iomode iomode,
1274 u32 seq)
1275{
1276 struct pnfs_layout_range recall_range = {
1277 .length = NFS4_MAX_UINT64,
1278 .iomode = iomode,
1279 };
1280 return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1281 &recall_range, seq) != -EBUSY;
1282}
1283
13c13a6a
TM		 1284/* Return true if layoutreturn is needed */
1285static bool
1286pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1287{
2370abda 1288 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a 1289 return false;
d474f961
TM		 1290 return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1291 lo->plh_return_seq);
13c13a6a
TM		 1292}
1293
1294static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1295{
1296 struct inode *inode= lo->plh_inode;
1297
2370abda 1298 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1299 return;
1300 spin_lock(&inode->i_lock);
1301 if (pnfs_layout_need_return(lo)) {
44ea8dfc 1302 const struct cred *cred;
13c13a6a
TM		 1303 nfs4_stateid stateid;
1304 enum pnfs_iomode iomode;
1305 bool send;
1306
44ea8dfc 1307 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
13c13a6a
TM		 1308 spin_unlock(&inode->i_lock);
1309 if (send) {
1310 /* Send an async layoutreturn so we dont deadlock */
44ea8dfc 1311 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
13c13a6a
TM		 1312 }
1313 } else
1314 spin_unlock(&inode->i_lock);
1315}
1316
293b3b06
AA		 1317/*
1318 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1319 * when the layout segment list is empty.
1320 *
1321 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1322 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1323 * deviceid is marked invalid.
1324 */
cbe82603
BH		 1325int
1326_pnfs_return_layout(struct inode *ino)
1327{
1328 struct pnfs_layout_hdr *lo = NULL;
1329 struct nfs_inode *nfsi = NFS_I(ino);
a421d218
AS		 1330 struct pnfs_layout_range range = {
1331 .iomode = IOMODE_ANY,
1332 .offset = 0,
1333 .length = NFS4_MAX_UINT64,
1334 };
cbe82603 1335 LIST_HEAD(tmp_list);
44ea8dfc 1336 const struct cred *cred;
cbe82603 1337 nfs4_stateid stateid;
24408f52 1338 int status = 0;
93b7f7ad 1339 bool send, valid_layout;
cbe82603 1340
366d5052 1341 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1342
1343 spin_lock(&ino->i_lock);
1344 lo = nfsi->layout;
e5929f3c 1345 if (!lo) {
cbe82603 1346 spin_unlock(&ino->i_lock);
293b3b06
AA		 1347 dprintk("NFS: %s no layout to return\n", __func__);
1348 goto out;
cbe82603 1349 }
cbe82603 1350 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1351 pnfs_get_layout_hdr(lo);
24408f52
TM		 1352 /* Is there an outstanding layoutreturn ? */
1353 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1354 spin_unlock(&ino->i_lock);
1355 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1356 TASK_UNINTERRUPTIBLE))
1357 goto out_put_layout_hdr;
1358 spin_lock(&ino->i_lock);
1359 }
93b7f7ad 1360 valid_layout = pnfs_layout_is_valid(lo);
24956804 1361 pnfs_clear_layoutcommit(ino, &tmp_list);
a421d218
AS		 1362 pnfs_mark_matching_lsegs_return(lo, &tmp_list, &range, 0);
1363
1364 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range)
c88953d8 1365 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
c88953d8 1366
293b3b06 1367 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1368 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1369 !valid_layout) {
293b3b06 1370 spin_unlock(&ino->i_lock);
293b3b06 1371 dprintk("NFS: %s no layout segments to return\n", __func__);
7bcc1058 1372 goto out_wait_layoutreturn;
293b3b06 1373 }
47abadef 1374
44ea8dfc 1375 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
cbe82603 1376 spin_unlock(&ino->i_lock);
7f27392c 1377 if (send)
44ea8dfc 1378 status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
7bcc1058
TM		 1379out_wait_layoutreturn:
1380 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE);
7f27392c 1381out_put_layout_hdr:
ee6625a9 1382 pnfs_free_lseg_list(&tmp_list);
7f27392c 1383 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1384out:
1385 dprintk("<-- %s status: %d\n", __func__, status);
1386 return status;
1387}
1388
24028672
TM		 1389int
1390pnfs_commit_and_return_layout(struct inode *inode)
1391{
1392 struct pnfs_layout_hdr *lo;
1393 int ret;
1394
1395 spin_lock(&inode->i_lock);
1396 lo = NFS_I(inode)->layout;
1397 if (lo == NULL) {
1398 spin_unlock(&inode->i_lock);
1399 return 0;
1400 }
1401 pnfs_get_layout_hdr(lo);
1402 /* Block new layoutgets and read/write to ds */
1403 lo->plh_block_lgets++;
1404 spin_unlock(&inode->i_lock);
1405 filemap_fdatawait(inode->i_mapping);
1406 ret = pnfs_layoutcommit_inode(inode, true);
1407 if (ret == 0)
1408 ret = _pnfs_return_layout(inode);
1409 spin_lock(&inode->i_lock);
1410 lo->plh_block_lgets--;
1411 spin_unlock(&inode->i_lock);
1412 pnfs_put_layout_hdr(lo);
1413 return ret;
1414}
1415
1c5bd76d
TM		 1416bool pnfs_roc(struct inode *ino,
1417 struct nfs4_layoutreturn_args *args,
1418 struct nfs4_layoutreturn_res *res,
a52458b4 1419 const struct cred *cred)
f7e8917a 1420{
40dd4b7a
TM		 1421 struct nfs_inode *nfsi = NFS_I(ino);
1422 struct nfs_open_context *ctx;
1423 struct nfs4_state *state;
f7e8917a 1424 struct pnfs_layout_hdr *lo;
1c5bd76d 1425 struct pnfs_layout_segment *lseg, *next;
44ea8dfc 1426 const struct cred *lc_cred;
193e3aa2 1427 nfs4_stateid stateid;
1c5bd76d
TM		 1428 enum pnfs_iomode iomode = 0;
1429 bool layoutreturn = false, roc = false;
e71708d4 1430 bool skip_read = false;
f7e8917a 1431
1c5bd76d
TM		 1432 if (!nfs_have_layout(ino))
1433 return false;
29ade5db 1434retry:
0de43976 1435 rcu_read_lock();
f7e8917a 1436 spin_lock(&ino->i_lock);
40dd4b7a 1437 lo = nfsi->layout;
0cdc329e 1438 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1439 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1440 lo = NULL;
40dd4b7a 1441 goto out_noroc;
9c6376eb
TM		 1442 }
1443 pnfs_get_layout_hdr(lo);
29ade5db 1444 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1445 spin_unlock(&ino->i_lock);
0de43976 1446 rcu_read_unlock();
29ade5db
TM		 1447 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1448 TASK_UNINTERRUPTIBLE);
1449 pnfs_put_layout_hdr(lo);
1450 goto retry;
1451 }
40dd4b7a 1452
e755d638 1453 /* no roc if we hold a delegation */
e71708d4
TM		 1454 if (nfs4_check_delegation(ino, FMODE_READ)) {
1455 if (nfs4_check_delegation(ino, FMODE_WRITE))
1456 goto out_noroc;
1457 skip_read = true;
1458 }
40dd4b7a 1459
0de43976 1460 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1461 state = ctx->state;
e71708d4
TM		 1462 if (state == NULL)
1463 continue;
40dd4b7a 1464 /* Don't return layout if there is open file state */
e71708d4 1465 if (state->state & FMODE_WRITE)
40dd4b7a 1466 goto out_noroc;
e71708d4
TM		 1467 if (state->state & FMODE_READ)
1468 skip_read = true;
40dd4b7a
TM		 1469 }
1470
e755d638 1471
1c5bd76d 1472 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1473 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1474 continue;
e755d638 1475 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1476 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1477 continue;
1478 /*
1479 * Note: mark lseg for return so pnfs_layout_remove_lseg
1480 * doesn't invalidate the layout for us.
1481 */
1482 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1483 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1484 continue;
1485 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1486 }
1487
1c5bd76d
TM		 1488 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1489 goto out_noroc;
69820d22 1490
500d701f 1491 /* ROC in two conditions:
e755d638
PT		 1492 * 1. there are ROC lsegs
1493 * 2. we don't send layoutreturn
e755d638 1494 */
1c5bd76d 1495 /* lo ref dropped in pnfs_roc_release() */
44ea8dfc 1496 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
1c5bd76d 1497 /* If the creds don't match, we can't compound the layoutreturn */
4d8948c7 1498 if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0)
1c5bd76d
TM		 1499 goto out_noroc;
1500
1501 roc = layoutreturn;
1502 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1503 res->lrs_present = 0;
1504 layoutreturn = false;
44ea8dfc 1505 put_cred(lc_cred);
4d8948c7 1506
40dd4b7a 1507out_noroc:
f7e8917a 1508 spin_unlock(&ino->i_lock);
0de43976 1509 rcu_read_unlock();
e755d638 1510 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1511 if (roc) {
1512 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1513 if (ld->prepare_layoutreturn)
1514 ld->prepare_layoutreturn(args);
9c6376eb 1515 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1516 return true;
1517 }
e755d638 1518 if (layoutreturn)
44ea8dfc 1519 pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
9c6376eb 1520 pnfs_put_layout_hdr(lo);
287bd3e9 1521 return false;
f7e8917a
FI		 1522}
1523
078000d0
TM		 1524int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
1525 struct nfs4_layoutreturn_res **respp, int *ret)
287a9c55
TM		 1526{
1527 struct nfs4_layoutreturn_args *arg = *argpp;
1528 int retval = -EAGAIN;
1529
1530 if (!arg)
1531 return 0;
1532 /* Handle Layoutreturn errors */
1533 switch (*ret) {
1534 case 0:
1535 retval = 0;
1536 break;
6109bcf7
TM		 1537 case -NFS4ERR_NOMATCHING_LAYOUT:
1538 /* Was there an RPC level error? If not, retry */
1539 if (task->tk_rpc_status == 0)
1540 break;
1541 /* If the call was not sent, let caller handle it */
1542 if (!RPC_WAS_SENT(task))
1543 return 0;
1544 /*
1545 * Otherwise, assume the call succeeded and
1546 * that we need to release the layout
1547 */
1548 *ret = 0;
1549 (*respp)->lrs_present = 0;
1550 retval = 0;
1551 break;
078a432d
TM		 1552 case -NFS4ERR_DELAY:
1553 /* Let the caller handle the retry */
1554 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1555 return 0;
287a9c55 1556 case -NFS4ERR_OLD_STATEID:
30cb3ee2 1557 if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
078000d0 1558 &arg->range, arg->inode))
287a9c55
TM		 1559 break;
1560 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1561 return -EAGAIN;
1562 }
1563 *argpp = NULL;
1564 *respp = NULL;
1565 return retval;
1566}
1567
1c5bd76d
TM		 1568void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1569 struct nfs4_layoutreturn_res *res,
1570 int ret)
f7e8917a 1571{
1c5bd76d 1572 struct pnfs_layout_hdr *lo = args->layout;
67bbceed 1573 struct inode *inode = args->inode;
1c5bd76d 1574 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1575 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1576
9c47b18c
TM		 1577 switch (ret) {
1578 case -NFS4ERR_NOMATCHING_LAYOUT:
67bbceed
TM		 1579 spin_lock(&inode->i_lock);
1580 if (pnfs_layout_is_valid(lo) &&
1581 nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
1582 pnfs_set_plh_return_info(lo, args->range.iomode, 0);
c18d1e17 1583 pnfs_clear_layoutreturn_waitbit(lo);
67bbceed 1584 spin_unlock(&inode->i_lock);
9c47b18c
TM		 1585 break;
1586 case 0:
1c5bd76d
TM		 1587 if (res->lrs_present)
1588 res_stateid = &res->stateid;
df561f66 1589 fallthrough;
9c47b18c 1590 default:
c18d1e17
TM		 1591 pnfs_layoutreturn_free_lsegs(lo, &args->stateid, &args->range,
1592 res_stateid);
1c5bd76d 1593 }
a19b4785 1594 trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
287bd3e9
TM		 1595 if (ld_private && ld_private->ops && ld_private->ops->free)
1596 ld_private->ops->free(ld_private);
1c5bd76d 1597 pnfs_put_layout_hdr(lo);
f7e8917a
FI		 1598}
1599
500d701f
PT		 1600bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1601{
1602 struct nfs_inode *nfsi = NFS_I(ino);
1603 struct pnfs_layout_hdr *lo;
1604 bool sleep = false;
1605
1606 /* we might not have grabbed lo reference. so need to check under
1607 * i_lock */
1608 spin_lock(&ino->i_lock);
1609 lo = nfsi->layout;
ee284e35
TM		 1610 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1611 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1612 sleep = true;
ee284e35 1613 }
500d701f 1614 spin_unlock(&ino->i_lock);
500d701f
PT		 1615 return sleep;
1616}
1617
b1f69b75
AA		 1618/*
1619 * Compare two layout segments for sorting into layout cache.
1620 * We want to preferentially return RW over RO layouts, so ensure those
1621 * are seen first.
1622 */
1623static s64
7dc0ac70 1624pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1625 const struct pnfs_layout_range *l2)
b1f69b75 1626{
fb3296eb
BH		 1627 s64 d;
1628
1629 /* high offset > low offset */
1630 d = l1->offset - l2->offset;
1631 if (d)
1632 return d;
1633
1634 /* short length > long length */
1635 d = l2->length - l1->length;
1636 if (d)
1637 return d;
1638
b1f69b75 1639 /* read > read/write */
fb3296eb 1640 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1641}
1642
03772d2f
TM		 1643static bool
1644pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1645 const struct pnfs_layout_range *l2)
1646{
1647 return pnfs_lseg_range_cmp(l1, l2) > 0;
1648}
1649
1650static bool
1651pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1652 struct pnfs_layout_segment *old)
1653{
1654 return false;
1655}
1656
1657void
1658pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1659 struct pnfs_layout_segment *lseg,
1660 bool (*is_after)(const struct pnfs_layout_range *,
1661 const struct pnfs_layout_range *),
1662 bool (*do_merge)(struct pnfs_layout_segment *,
1663 struct pnfs_layout_segment *),
1664 struct list_head *free_me)
974cec8c 1665{
03772d2f 1666 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1667
974cec8c
AA		 1668 dprintk("%s:Begin\n", __func__);
1669
03772d2f
TM		 1670 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1671 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1672 continue;
1673 if (do_merge(lseg, lp)) {
1674 mark_lseg_invalid(lp, free_me);
1675 continue;
1676 }
1677 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1678 continue;
566052c5 1679 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1680 dprintk("%s: inserted lseg %p "
1681 "iomode %d offset %llu length %llu before "
1682 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1683 __func__, lseg, lseg->pls_range.iomode,
1684 lseg->pls_range.offset, lseg->pls_range.length,
1685 lp, lp->pls_range.iomode, lp->pls_range.offset,
1686 lp->pls_range.length);
fb3296eb 1687 goto out;
974cec8c 1688 }
fb3296eb
BH		 1689 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1690 dprintk("%s: inserted lseg %p "
1691 "iomode %d offset %llu length %llu at tail\n",
1692 __func__, lseg, lseg->pls_range.iomode,
1693 lseg->pls_range.offset, lseg->pls_range.length);
1694out:
70c3bd2b 1695 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1696
1697 dprintk("%s:Return\n", __func__);
e5e94017 1698}
03772d2f
TM		 1699EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1700
1701static void
1702pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1703 struct pnfs_layout_segment *lseg,
1704 struct list_head *free_me)
1705{
1706 struct inode *inode = lo->plh_inode;
1707 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1708
1709 if (ld->add_lseg != NULL)
1710 ld->add_lseg(lo, lseg, free_me);
1711 else
1712 pnfs_generic_layout_insert_lseg(lo, lseg,
1713 pnfs_lseg_range_is_after,
1714 pnfs_lseg_no_merge,
1715 free_me);
1716}
e5e94017
BH		 1717
1718static struct pnfs_layout_hdr *
9fa40758
PT		 1719alloc_init_layout_hdr(struct inode *ino,
1720 struct nfs_open_context *ctx,
1721 gfp_t gfp_flags)
e5e94017
BH		 1722{
1723 struct pnfs_layout_hdr *lo;
1724
636fb9c8 1725 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1726 if (!lo)
1727 return NULL;
2b28a7be 1728 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1729 INIT_LIST_HEAD(&lo->plh_layouts);
1730 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1731 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1732 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1733 lo->plh_inode = ino;
a52458b4 1734 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1735 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1736 return lo;
1737}
1738
1739static struct pnfs_layout_hdr *
9fa40758
PT		 1740pnfs_find_alloc_layout(struct inode *ino,
1741 struct nfs_open_context *ctx,
1742 gfp_t gfp_flags)
e5241e43
TM		 1743 __releases(&ino->i_lock)
1744 __acquires(&ino->i_lock)
e5e94017
BH		 1745{
1746 struct nfs_inode *nfsi = NFS_I(ino);
1747 struct pnfs_layout_hdr *new = NULL;
1748
1749 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1750
251ec410
TM		 1751 if (nfsi->layout != NULL)
1752 goto out_existing;
e5e94017 1753 spin_unlock(&ino->i_lock);
9fa40758 1754 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1755 spin_lock(&ino->i_lock);
1756
251ec410 1757 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1758 nfsi->layout = new;
251ec410 1759 return new;
7175fe90
YN		 1760 } else if (new != NULL)
1761 pnfs_free_layout_hdr(new);
251ec410
TM		 1762out_existing:
1763 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1764 return nfsi->layout;
1765}
1766
b1f69b75
AA		 1767/*
1768 * iomode matching rules:
c7d73af2
TH		 1769 * iomode lseg strict match
1770 * iomode
1771 * ----- ----- ------ -----
1772 * ANY READ N/A true
1773 * ANY RW N/A true
1774 * RW READ N/A false
1775 * RW RW N/A true
1776 * READ READ N/A true
1777 * READ RW true false
1778 * READ RW false true
b1f69b75 1779 */
3cb2df17 1780static bool
7dc0ac70 1781pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1782 const struct pnfs_layout_range *range,
1783 bool strict_iomode)
b1f69b75 1784{
fb3296eb
BH		 1785 struct pnfs_layout_range range1;
1786
1787 if ((range->iomode == IOMODE_RW &&
1788 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1789 (range->iomode != ls_range->iomode &&
6089dd0d 1790 strict_iomode) ||
7dc0ac70 1791 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1792 return false;
fb3296eb
BH		 1793
1794 /* range1 covers only the first byte in the range */
1795 range1 = *range;
1796 range1.length = 1;
7dc0ac70 1797 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1798}
1799
1800/*
1801 * lookup range in layout
1802 */
e5e94017 1803static struct pnfs_layout_segment *
fb3296eb 1804pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1805 struct pnfs_layout_range *range,
1806 bool strict_iomode)
e5e94017 1807{
b1f69b75
AA		 1808 struct pnfs_layout_segment *lseg, *ret = NULL;
1809
1810 dprintk("%s:Begin\n", __func__);
1811
b7edfaa1 1812 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1813 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1814 pnfs_lseg_range_match(&lseg->pls_range, range,
1815 strict_iomode)) {
9369a431 1816 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1817 break;
1818 }
b1f69b75
AA		 1819 }
1820
1821 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1822 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1823 return ret;
e5e94017
BH		 1824}
1825
d23d61c8
AA		 1826/*
1827 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1828 * to the MDS or over pNFS
1829 *
1830 * The nfs_inode read_io and write_io fields are cumulative counters reset
1831 * when there are no layout segments. Note that in pnfs_update_layout iomode
1832 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1833 * WRITE request.
1834 *
1835 * A return of true means use MDS I/O.
1836 *
1837 * From rfc 5661:
1838 * If a file's size is smaller than the file size threshold, data accesses
1839 * SHOULD be sent to the metadata server. If an I/O request has a length that
1840 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1841 * server. If both file size and I/O size are provided, the client SHOULD
1842 * reach or exceed both thresholds before sending its read or write
1843 * requests to the data server.
1844 */
1845static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1846 struct inode *ino, int iomode)
1847{
1848 struct nfs4_threshold *t = ctx->mdsthreshold;
1849 struct nfs_inode *nfsi = NFS_I(ino);
1850 loff_t fsize = i_size_read(ino);
1851 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1852
1853 if (t == NULL)
1854 return ret;
1855
1856 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1857 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1858
1859 switch (iomode) {
1860 case IOMODE_READ:
1861 if (t->bm & THRESHOLD_RD) {
1862 dprintk("%s fsize %llu\n", __func__, fsize);
1863 size_set = true;
1864 if (fsize < t->rd_sz)
1865 size = true;
1866 }
1867 if (t->bm & THRESHOLD_RD_IO) {
1868 dprintk("%s nfsi->read_io %llu\n", __func__,
1869 nfsi->read_io);
1870 io_set = true;
1871 if (nfsi->read_io < t->rd_io_sz)
1872 io = true;
1873 }
1874 break;
1875 case IOMODE_RW:
1876 if (t->bm & THRESHOLD_WR) {
1877 dprintk("%s fsize %llu\n", __func__, fsize);
1878 size_set = true;
1879 if (fsize < t->wr_sz)
1880 size = true;
1881 }
1882 if (t->bm & THRESHOLD_WR_IO) {
1883 dprintk("%s nfsi->write_io %llu\n", __func__,
1884 nfsi->write_io);
1885 io_set = true;
1886 if (nfsi->write_io < t->wr_io_sz)
1887 io = true;
1888 }
1889 break;
1890 }
1891 if (size_set && io_set) {
1892 if (size && io)
1893 ret = true;
1894 } else if (size || io)
1895 ret = true;
1896
1897 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1898 return ret;
1899}
1900
d03360aa 1901static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1902{
1903 /*
1904 * send layoutcommit as it can hold up layoutreturn due to lseg
1905 * reference
1906 */
1907 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1908 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1909 nfs_wait_bit_killable,
d03360aa 1910 TASK_KILLABLE);
aa8a45ee
PT		 1911}
1912
411ae722
TM		 1913static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1914{
1915 atomic_inc(&lo->plh_outstanding);
1916}
1917
1918static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1919{
1920 if (atomic_dec_and_test(&lo->plh_outstanding))
1921 wake_up_var(&lo->plh_outstanding);
1922}
1923
d29b468d
TM		 1924static bool pnfs_is_first_layoutget(struct pnfs_layout_hdr *lo)
1925{
1926 return test_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags);
1927}
1928
d67ae825
TH		 1929static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1930{
1931 unsigned long *bitlock = &lo->plh_flags;
1932
1933 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1934 smp_mb__after_atomic();
1935 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1936}
1937
78746a38
FI		 1938static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1939 struct nfs_server *server)
1940{
cf6605d1 1941 if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
78746a38
FI		 1942 struct nfs_client *clp = server->nfs_client;
1943
1944 /* The lo must be on the clp list if there is any
1945 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1946 */
1947 spin_lock(&clp->cl_lock);
cf6605d1 1948 list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
78746a38
FI		 1949 spin_unlock(&clp->cl_lock);
1950 }
1951}
1952
e5e94017
BH		 1953/*
1954 * Layout segment is retreived from the server if not cached.
1955 * The appropriate layout segment is referenced and returned to the caller.
1956 */
7c24d948 1957struct pnfs_layout_segment *
e5e94017
BH		 1958pnfs_update_layout(struct inode *ino,
1959 struct nfs_open_context *ctx,
fb3296eb
BH		 1960 loff_t pos,
1961 u64 count,
a75b9df9 1962 enum pnfs_iomode iomode,
c7d73af2 1963 bool strict_iomode,
a75b9df9 1964 gfp_t gfp_flags)
e5e94017 1965{
fb3296eb
BH		 1966 struct pnfs_layout_range arg = {
1967 .iomode = iomode,
1968 .offset = pos,
1969 .length = count,
1970 };
70d2f7b1 1971 unsigned pg_offset;
6382a441
WAA		 1972 struct nfs_server *server = NFS_SERVER(ino);
1973 struct nfs_client *clp = server->nfs_client;
183d9e7b 1974 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1975 struct pnfs_layout_segment *lseg = NULL;
587f03de 1976 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1977 nfs4_stateid stateid;
1978 long timeout = 0;
66b53f32 1979 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1980 bool first;
e5e94017 1981
9a4bf31d 1982 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1983 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1984 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1985 goto out;
9a4bf31d 1986 }
d23d61c8 1987
9a4bf31d 1988 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1989 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1990 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1991 goto out;
9a4bf31d 1992 }
d23d61c8 1993
9bf87482 1994lookup_again:
d03360aa
TM		 1995 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1996 if (IS_ERR(lseg))
1997 goto out;
9bf87482 1998 first = false;
e5e94017 1999 spin_lock(&ino->i_lock);
9fa40758 2000 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 2001 if (lo == NULL) {
2002 spin_unlock(&ino->i_lock);
183d9e7b 2003 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2004 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 2005 goto out;
2006 }
e5e94017 2007
43f1b3da 2008 /* Do we even need to bother with this? */
a59c30ac 2009 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 2010 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2011 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 2012 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 2013 goto out_unlock;
2014 }
2015
2016 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 2017 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 2018 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2019 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 2020 goto out_unlock;
9a4bf31d 2021 }
e5e94017 2022
411ae722
TM		 2023 /*
2024 * If the layout segment list is empty, but there are outstanding
2025 * layoutget calls, then they might be subject to a layoutrecall.
2026 */
0b77f97a 2027 if ((list_empty(&lo->plh_segs) || !pnfs_layout_is_valid(lo)) &&
411ae722
TM		 2028 atomic_read(&lo->plh_outstanding) != 0) {
2029 spin_unlock(&ino->i_lock);
d03360aa 2030 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
400417b0 2031 !atomic_read(&lo->plh_outstanding)));
58bbeab4 2032 if (IS_ERR(lseg))
411ae722
TM		 2033 goto out_put_layout_hdr;
2034 pnfs_put_layout_hdr(lo);
2035 goto lookup_again;
2036 }
2037
2c8d5fc3
TM		 2038 /*
2039 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
2040 * for LAYOUTRETURN.
2041 */
2042 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
2043 spin_unlock(&ino->i_lock);
2044 dprintk("%s wait for layoutreturn\n", __func__);
2045 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2046 if (!IS_ERR(lseg)) {
2047 pnfs_put_layout_hdr(lo);
2048 dprintk("%s retrying\n", __func__);
2049 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2050 lseg,
2051 PNFS_UPDATE_LAYOUT_RETRY);
2052 goto lookup_again;
2053 }
2054 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2055 PNFS_UPDATE_LAYOUT_RETURN);
2056 goto out_put_layout_hdr;
2057 }
2058
c7d73af2 2059 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 2060 if (lseg) {
2061 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2062 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
2063 goto out_unlock;
2064 }
2065
183d9e7b
JL		 2066 /*
2067 * Choose a stateid for the LAYOUTGET. If we don't have a layout
2068 * stateid, or it has been invalidated, then we must use the open
2069 * stateid.
2070 */
67a3b721 2071 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
d9aba2b4 2072 int status;
183d9e7b
JL		 2073
2074 /*
2075 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 2076 * RFC 5661 Errata 3208.
2077 */
2078 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
2079 &lo->plh_flags)) {
2080 spin_unlock(&ino->i_lock);
d03360aa
TM		 2081 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
2082 NFS_LAYOUT_FIRST_LAYOUTGET,
2083 TASK_KILLABLE));
2084 if (IS_ERR(lseg))
2085 goto out_put_layout_hdr;
9bf87482 2086 pnfs_put_layout_hdr(lo);
183d9e7b 2087 dprintk("%s retrying\n", __func__);
9bf87482
PT		 2088 goto lookup_again;
2089 }
183d9e7b 2090
fbf4bcc9 2091 spin_unlock(&ino->i_lock);
183d9e7b 2092 first = true;
d9aba2b4 2093 status = nfs4_select_rw_stateid(ctx->state,
70d2f7b1 2094 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
d9aba2b4
TM		 2095 NULL, &stateid, NULL);
2096 if (status != 0) {
731c74dd 2097 lseg = ERR_PTR(status);
70d2f7b1
TM		 2098 trace_pnfs_update_layout(ino, pos, count,
2099 iomode, lo, lseg,
2100 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
d9aba2b4
TM		 2101 nfs4_schedule_stateid_recovery(server, ctx->state);
2102 pnfs_clear_first_layoutget(lo);
2103 pnfs_put_layout_hdr(lo);
2104 goto lookup_again;
70d2f7b1 2105 }
fbf4bcc9 2106 spin_lock(&ino->i_lock);
9bf87482 2107 } else {
183d9e7b 2108 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 2109 }
568e8c49 2110
9a4bf31d 2111 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 2112 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2113 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 2114 goto out_unlock;
9a4bf31d 2115 }
411ae722 2116 nfs_layoutget_begin(lo);
f49f9baa 2117 spin_unlock(&ino->i_lock);
30005121 2118
78746a38 2119 _add_to_server_list(lo, server);
e5e94017 2120
09cbfeaf 2121 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 2122 if (pg_offset) {
2123 arg.offset -= pg_offset;
2124 arg.length += pg_offset;
2125 }
7c24d948 2126 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 2127 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 2128
5e36e2a9 2129 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 2130 if (!lgp) {
2131 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2132 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 2133 nfs_layoutget_end(lo);
587f03de
FI		 2134 goto out_put_layout_hdr;
2135 }
2136
b4e89bcb
TM		 2137 lgp->lo = lo;
2138 pnfs_get_layout_hdr(lo);
2139
dacb452d 2140 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 2141 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2142 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 2143 nfs_layoutget_end(lo);
83026d80 2144 if (IS_ERR(lseg)) {
183d9e7b 2145 switch(PTR_ERR(lseg)) {
e85d7ee4 2146 case -EBUSY:
183d9e7b
JL		 2147 if (time_after(jiffies, giveup))
2148 lseg = NULL;
66b53f32
TM		 2149 break;
2150 case -ERECALLCONFLICT:
183d9e7b 2151 case -EAGAIN:
56b38a1f 2152 break;
183d9e7b
JL		 2153 default:
2154 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2155 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2156 lseg = NULL;
2157 }
56b38a1f
TM		 2158 goto out_put_layout_hdr;
2159 }
2160 if (lseg) {
2161 if (first)
2162 pnfs_clear_first_layoutget(lo);
2163 trace_pnfs_update_layout(ino, pos, count,
2164 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2165 pnfs_put_layout_hdr(lo);
2166 goto lookup_again;
83026d80
JL		 2167 }
2168 } else {
2169 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2170 }
2171
830ffb56 2172out_put_layout_hdr:
d67ae825
TH		 2173 if (first)
2174 pnfs_clear_first_layoutget(lo);
d5b9216f
TM		 2175 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2176 PNFS_UPDATE_LAYOUT_EXIT);
70c3bd2b 2177 pnfs_put_layout_hdr(lo);
e5e94017 2178out:
f86bbcf8
TM		 2179 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2180 "(%s, offset: %llu, length: %llu)\n",
2181 __func__, ino->i_sb->s_id,
2182 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2183 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2184 iomode==IOMODE_RW ? "read/write" : "read-only",
2185 (unsigned long long)pos,
2186 (unsigned long long)count);
e5e94017
BH		 2187 return lseg;
2188out_unlock:
2189 spin_unlock(&ino->i_lock);
830ffb56 2190 goto out_put_layout_hdr;
e5e94017 2191}
7c24d948 2192EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2193
540d9864
TM		 2194static bool
2195pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2196{
2197 switch (range->iomode) {
2198 case IOMODE_READ:
2199 case IOMODE_RW:
2200 break;
2201 default:
2202 return false;
2203 }
2204 if (range->offset == NFS4_MAX_UINT64)
2205 return false;
2206 if (range->length == 0)
2207 return false;
2208 if (range->length != NFS4_MAX_UINT64 &&
2209 range->length > NFS4_MAX_UINT64 - range->offset)
2210 return false;
2211 return true;
2212}
2213
78746a38
FI		 2214static struct pnfs_layout_hdr *
2215_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2216{
2217 struct pnfs_layout_hdr *lo;
2218
2219 spin_lock(&ino->i_lock);
63d8a41b 2220 lo = pnfs_find_alloc_layout(ino, ctx, nfs_io_gfp_mask());
78746a38
FI		 2221 if (!lo)
2222 goto out_unlock;
2223 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2224 goto out_unlock;
2225 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2226 goto out_unlock;
2227 if (pnfs_layoutgets_blocked(lo))
2228 goto out_unlock;
2229 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2230 goto out_unlock;
411ae722 2231 nfs_layoutget_begin(lo);
78746a38
FI		 2232 spin_unlock(&ino->i_lock);
2233 _add_to_server_list(lo, NFS_SERVER(ino));
2234 return lo;
2235
2236out_unlock:
2237 spin_unlock(&ino->i_lock);
2238 pnfs_put_layout_hdr(lo);
2239 return NULL;
2240}
2241
2409a976
FI		 2242static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2243 struct nfs_open_context *ctx)
2244{
78746a38
FI		 2245 struct inode *ino = data->dentry->d_inode;
2246 struct pnfs_layout_range rng = {
2247 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2248 IOMODE_RW: IOMODE_READ,
2249 .offset = 0,
2250 .length = NFS4_MAX_UINT64,
2251 };
2252 struct nfs4_layoutget *lgp;
2253 struct pnfs_layout_hdr *lo;
2254
64294b08
TM		 2255 /* Heuristic: don't send layoutget if we have cached data */
2256 if (rng.iomode == IOMODE_READ &&
2257 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2258 return;
2259
78746a38
FI		 2260 lo = _pnfs_grab_empty_layout(ino, ctx);
2261 if (!lo)
2262 return;
63d8a41b
TM		 2263 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid, &rng,
2264 nfs_io_gfp_mask());
78746a38
FI		 2265 if (!lgp) {
2266 pnfs_clear_first_layoutget(lo);
cb2856c5 2267 nfs_layoutget_end(lo);
78746a38
FI		 2268 pnfs_put_layout_hdr(lo);
2269 return;
2270 }
b4e89bcb 2271 lgp->lo = lo;
78746a38
FI		 2272 data->lgp = lgp;
2273 data->o_arg.lg_args = &lgp->args;
2274 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2275}
2276
2277static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2278 struct nfs_open_context *ctx)
2279{
b4e89bcb 2280 struct inode *ino = data->dentry->d_inode;
2409a976
FI		 2281 struct pnfs_layout_range rng = {
2282 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2283 IOMODE_RW: IOMODE_READ,
2284 .offset = 0,
2285 .length = NFS4_MAX_UINT64,
2286 };
2287 struct nfs4_layoutget *lgp;
2288
63d8a41b
TM		 2289 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid, &rng,
2290 nfs_io_gfp_mask());
2409a976
FI		 2291 if (!lgp)
2292 return;
2293 data->lgp = lgp;
2294 data->o_arg.lg_args = &lgp->args;
2295 data->o_res.lg_res = &lgp->res;
2296}
2297
2298void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2299 struct nfs_open_context *ctx)
2300{
2301 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2302
2303 if (!(pnfs_enabled_sb(server) &&
2304 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2305 return;
2306 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2307 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2308 return;
b4e89bcb
TM		 2309 if (data->lgp)
2310 return;
2409a976
FI		 2311 if (data->state)
2312 _lgopen_prepare_attached(data, ctx);
2313 else
2314 _lgopen_prepare_floating(data, ctx);
2315}
2316
2317void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2318 struct nfs_open_context *ctx)
2319{
2320 struct pnfs_layout_hdr *lo;
2321 struct pnfs_layout_segment *lseg;
c49b5209 2322 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2323 u32 iomode;
2324
6e01260c 2325 if (!lgp)
2409a976 2326 return;
6e01260c
FI		 2327 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2328 if (lgp->res.status) {
2329 switch (lgp->res.status) {
6e01260c 2330 default:
8dc96566
TM		 2331 break;
2332 /*
2333 * Halt lgopen attempts if the server doesn't recognise
2334 * the "current stateid" value, the layout type, or the
2335 * layoutget operation as being valid.
2336 * Also if it complains about too many ops in the compound
2337 * or of the request/reply being too big.
2338 */
2339 case -NFS4ERR_BAD_STATEID:
2340 case -NFS4ERR_NOTSUPP:
2341 case -NFS4ERR_REP_TOO_BIG:
2342 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2343 case -NFS4ERR_REQ_TOO_BIG:
2344 case -NFS4ERR_TOO_MANY_OPS:
2345 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2346 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2347 }
2348 return;
2349 }
b4e89bcb 2350 if (!lgp->lo) {
78746a38
FI		 2351 lo = _pnfs_grab_empty_layout(ino, ctx);
2352 if (!lo)
2353 return;
b4e89bcb 2354 lgp->lo = lo;
2409a976 2355 } else
b4e89bcb 2356 lo = lgp->lo;
2409a976
FI		 2357
2358 lseg = pnfs_layout_process(lgp);
32f1c28f 2359 if (!IS_ERR(lseg)) {
2409a976
FI		 2360 iomode = lgp->args.range.iomode;
2361 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2362 pnfs_put_lseg(lseg);
2363 }
30ae2412
FI		 2364}
2365
2366void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2367{
2368 if (lgp != NULL) {
b4e89bcb
TM		 2369 if (lgp->lo) {
2370 pnfs_clear_first_layoutget(lgp->lo);
2371 nfs_layoutget_end(lgp->lo);
30ae2412
FI		 2372 }
2373 pnfs_layoutget_free(lgp);
2374 }
2409a976
FI		 2375}
2376
a0b0a6e3 2377struct pnfs_layout_segment *
b1f69b75
AA		 2378pnfs_layout_process(struct nfs4_layoutget *lgp)
2379{
b4e89bcb 2380 struct pnfs_layout_hdr *lo = lgp->lo;
b1f69b75
AA		 2381 struct nfs4_layoutget_res *res = &lgp->res;
2382 struct pnfs_layout_segment *lseg;
b7edfaa1 2383 struct inode *ino = lo->plh_inode;
78096cca 2384 LIST_HEAD(free_me);
540d9864
TM		 2385
2386 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2387 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2388
2389 /* Inject layout blob into I/O device driver */
a75b9df9 2390 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2391 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2392 if (!lseg)
1b3c6d07
JL		 2393 lseg = ERR_PTR(-ENOMEM);
2394
2395 dprintk("%s: Could not allocate layout: error %ld\n",
2396 __func__, PTR_ERR(lseg));
2397 return lseg;
b1f69b75
AA		 2398 }
2399
119cef97 2400 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2401
b1f69b75 2402 spin_lock(&ino->i_lock);
e1c06f80 2403 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2404 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2405 goto out_forget;
43f1b3da 2406 }
038d6493 2407
0b77f97a
TM		 2408 if (!pnfs_layout_is_valid(lo) && !pnfs_is_first_layoutget(lo))
2409 goto out_forget;
2410
d29b468d 2411 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2412 /* existing state ID, make sure the sequence number matches. */
2413 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
0b77f97a 2414 if (!pnfs_layout_is_valid(lo))
d29b468d 2415 lo->plh_barrier = 0;
362f7474 2416 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2417 goto out_forget;
362f7474 2418 }
59b56394 2419 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
d29b468d 2420 } else if (pnfs_layout_is_valid(lo)) {
362f7474
CH		 2421 /*
2422 * We got an entirely new state ID. Mark all segments for the
9888d837 2423 * inode invalid, and retry the layoutget
362f7474 2424 */
08bd8dbe
TM		 2425 struct pnfs_layout_range range = {
2426 .iomode = IOMODE_ANY,
2427 .length = NFS4_MAX_UINT64,
2428 };
fb700ef0 2429 pnfs_mark_matching_lsegs_return(lo, &free_me, &range, 0);
9888d837 2430 goto out_forget;
d29b468d
TM		 2431 } else {
2432 /* We have a completely new layout */
d29b468d 2433 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
362f7474 2434 }
038d6493 2435
9369a431 2436 pnfs_get_lseg(lseg);
03772d2f 2437 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2438
b1f69b75 2439
3976143b 2440 if (res->return_on_close)
f7e8917a 2441 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2442
b1f69b75 2443 spin_unlock(&ino->i_lock);
78096cca 2444 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2445 return lseg;
43f1b3da 2446
1b3c6d07 2447out_forget:
43f1b3da
FI		 2448 spin_unlock(&ino->i_lock);
2449 lseg->pls_layout = lo;
2450 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2451 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2452}
2453
2f215968
TM		 2454/**
2455 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2456 * @lo: pointer to layout header
2457 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2458 * @return_range: describe layout segment ranges to be returned
e0b7d420 2459 * @seq: stateid seqid to match
2f215968
TM		 2460 *
2461 * This function is mainly intended for use by layoutrecall. It attempts
2462 * to free the layout segment immediately, or else to mark it for return
2463 * as soon as its reference count drops to zero.
e0b7d420
TM		 2464 *
2465 * Returns
2466 * - 0: a layoutreturn needs to be scheduled.
2467 * - EBUSY: there are layout segment that are still in use.
2468 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2469 */
10335556 2470int
016256df
PT		 2471pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2472 struct list_head *tmp_list,
6d597e17
JL		 2473 const struct pnfs_layout_range *return_range,
2474 u32 seq)
016256df
PT		 2475{
2476 struct pnfs_layout_segment *lseg, *next;
10335556 2477 int remaining = 0;
016256df
PT		 2478
2479 dprintk("%s:Begin lo %p\n", __func__, lo);
2480
fc7ff367 2481 assert_spin_locked(&lo->plh_inode->i_lock);
016256df 2482
39fd0186
TM		 2483 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2484 tmp_list = &lo->plh_return_segs;
2485
016256df 2486 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2487 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2488 dprintk("%s: marking lseg %p iomode %d "
2489 "offset %llu length %llu\n", __func__,
2490 lseg, lseg->pls_range.iomode,
2491 lseg->pls_range.offset,
2492 lseg->pls_range.length);
39fd0186
TM		 2493 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2494 tmp_list = &lo->plh_return_segs;
ff041727 2495 if (mark_lseg_invalid(lseg, tmp_list))
2f215968
TM		 2496 continue;
2497 remaining++;
016256df 2498 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2499 }
6d597e17 2500
e0b7d420 2501 if (remaining) {
6d597e17 2502 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2503 return -EBUSY;
2504 }
6d597e17 2505
e0b7d420
TM		 2506 if (!list_empty(&lo->plh_return_segs)) {
2507 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2508 return 0;
2509 }
2510
2511 return -ENOENT;
016256df
PT		 2512}
2513
b5fdf841
TM		 2514static void
2515pnfs_mark_layout_for_return(struct inode *inode,
2516 const struct pnfs_layout_range *range)
016256df 2517{
b5fdf841 2518 struct pnfs_layout_hdr *lo;
10335556 2519 bool return_now = false;
016256df
PT		 2520
2521 spin_lock(&inode->i_lock);
b5fdf841 2522 lo = NFS_I(inode)->layout;
bdebfccd
TM		 2523 if (!pnfs_layout_is_valid(lo)) {
2524 spin_unlock(&inode->i_lock);
2525 return;
2526 }
b5fdf841 2527 pnfs_set_plh_return_info(lo, range->iomode, 0);
016256df
PT		 2528 /*
2529 * mark all matching lsegs so that we are sure to have no live
2530 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2531 * for how it works.
2532 */
b5fdf841 2533 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
44ea8dfc 2534 const struct cred *cred;
10335556 2535 nfs4_stateid stateid;
e5fd1904 2536 enum pnfs_iomode iomode;
10335556 2537
44ea8dfc 2538 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
10335556
TM		 2539 spin_unlock(&inode->i_lock);
2540 if (return_now)
44ea8dfc 2541 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
10335556
TM		 2542 } else {
2543 spin_unlock(&inode->i_lock);
2544 nfs_commit_inode(inode, 0);
2545 }
016256df 2546}
b5fdf841
TM		 2547
2548void pnfs_error_mark_layout_for_return(struct inode *inode,
2549 struct pnfs_layout_segment *lseg)
2550{
2551 struct pnfs_layout_range range = {
2552 .iomode = lseg->pls_range.iomode,
2553 .offset = 0,
2554 .length = NFS4_MAX_UINT64,
2555 };
2556
2557 pnfs_mark_layout_for_return(inode, &range);
2558}
016256df
PT		 2559EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2560
b5fdf841
TM		 2561static bool
2562pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
2563{
2564 return pnfs_layout_is_valid(lo) &&
2565 !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
2566 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
2567}
2568
2569static struct pnfs_layout_segment *
2570pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
2571 const struct pnfs_layout_range *range,
2572 enum pnfs_iomode iomode)
2573{
2574 struct pnfs_layout_segment *lseg;
2575
2576 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
2577 if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
2578 continue;
2579 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2580 continue;
2581 if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
2582 continue;
2583 if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
2584 return lseg;
2585 }
2586 return NULL;
2587}
2588
2589/* Find open file states whose mode matches that of the range */
2590static bool
2591pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
2592 const struct pnfs_layout_range *range)
2593{
2594 struct list_head *head;
2595 struct nfs_open_context *ctx;
2596 fmode_t mode = 0;
2597
2598 if (!pnfs_layout_can_be_returned(lo) ||
2599 !pnfs_find_first_lseg(lo, range, range->iomode))
2600 return false;
2601
2602 head = &NFS_I(lo->plh_inode)->open_files;
2603 list_for_each_entry_rcu(ctx, head, list) {
2604 if (ctx->state)
2605 mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
2606 }
2607
2608 switch (range->iomode) {
2609 default:
2610 break;
2611 case IOMODE_READ:
2612 mode &= ~FMODE_WRITE;
2613 break;
2614 case IOMODE_RW:
2615 if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
2616 mode &= ~FMODE_READ;
2617 }
2618 return mode == 0;
2619}
2620
2621static int
2622pnfs_layout_return_unused_byserver(struct nfs_server *server, void *data)
2623{
2624 const struct pnfs_layout_range *range = data;
2625 struct pnfs_layout_hdr *lo;
2626 struct inode *inode;
2627restart:
2628 rcu_read_lock();
2629 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
2630 if (!pnfs_layout_can_be_returned(lo) ||
2631 test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2632 continue;
2633 inode = lo->plh_inode;
2634 spin_lock(&inode->i_lock);
2635 if (!pnfs_should_return_unused_layout(lo, range)) {
2636 spin_unlock(&inode->i_lock);
2637 continue;
2638 }
2639 spin_unlock(&inode->i_lock);
2640 inode = pnfs_grab_inode_layout_hdr(lo);
2641 if (!inode)
2642 continue;
2643 rcu_read_unlock();
2644 pnfs_mark_layout_for_return(inode, range);
2645 iput(inode);
2646 cond_resched();
2647 goto restart;
2648 }
2649 rcu_read_unlock();
2650 return 0;
2651}
2652
2653void
2654pnfs_layout_return_unused_byclid(struct nfs_client *clp,
2655 enum pnfs_iomode iomode)
2656{
2657 struct pnfs_layout_range range = {
2658 .iomode = iomode,
2659 .offset = 0,
2660 .length = NFS4_MAX_UINT64,
2661 };
2662
2663 nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
2664 &range);
2665}
2666
b3230e80
TM		 2667void
2668pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2669{
2670 if (pgio->pg_lseg == NULL ||
2671 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2672 return;
2673 pnfs_put_lseg(pgio->pg_lseg);
2674 pgio->pg_lseg = NULL;
2675}
2676EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2677
08cb5b0f
BC		 2678/*
2679 * Check for any intersection between the request and the pgio->pg_lseg,
2680 * and if none, put this pgio->pg_lseg away.
2681 */
e1e54ab7 2682void
08cb5b0f
BC		 2683pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2684{
2685 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2686 pnfs_put_lseg(pgio->pg_lseg);
2687 pgio->pg_lseg = NULL;
2688 }
2689}
e1e54ab7 2690EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
08cb5b0f 2691
d8007d4d
TM		 2692void
2693pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2694{
d1d97395 2695 u64 rd_size;
1fd937bd 2696
b3230e80 2697 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2698 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2699 if (pgio->pg_lseg == NULL) {
2700 if (pgio->pg_dreq == NULL)
2701 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2702 else
2703 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2704
63d8a41b
TM		 2705 pgio->pg_lseg =
2706 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
2707 req_offset(req), rd_size,
2708 IOMODE_READ, false,
2709 nfs_io_gfp_mask());
d600ad1f
PT		 2710 if (IS_ERR(pgio->pg_lseg)) {
2711 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2712 pgio->pg_lseg = NULL;
2713 return;
2714 }
cb5d04bc 2715 }
e885de1a
TM		 2716 /* If no lseg, fall back to read through mds */
2717 if (pgio->pg_lseg == NULL)
1f945357 2718 nfs_pageio_reset_read_mds(pgio);
e885de1a 2719
d8007d4d
TM		 2720}
2721EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2722
2723void
6296556f
PT		 2724pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2725 struct nfs_page *req, u64 wb_size)
d8007d4d 2726{
b3230e80 2727 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2728 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2729 if (pgio->pg_lseg == NULL) {
63d8a41b
TM		 2730 pgio->pg_lseg =
2731 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
2732 req_offset(req), wb_size, IOMODE_RW,
2733 false, nfs_io_gfp_mask());
d600ad1f
PT		 2734 if (IS_ERR(pgio->pg_lseg)) {
2735 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2736 pgio->pg_lseg = NULL;
2737 return;
2738 }
2739 }
e885de1a
TM		 2740 /* If no lseg, fall back to write through mds */
2741 if (pgio->pg_lseg == NULL)
1f945357 2742 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2743}
2744EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2745
180bb5ec
WAA		 2746void
2747pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2748{
2749 if (desc->pg_lseg) {
2750 pnfs_put_lseg(desc->pg_lseg);
2751 desc->pg_lseg = NULL;
2752 }
2753}
2754EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2755
b4fdac1a
WAA		 2756/*
2757 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2758 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2759 */
2760size_t
a7d42ddb
WAA		 2761pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2762 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2763{
0f9c429e 2764 unsigned int size;
c5e20cb7 2765 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2766
2767 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2768 if (!size)
2769 return 0;
94ad1c80 2770
19982ba8 2771 /*
c5e20cb7
WAA		 2772 * 'size' contains the number of bytes left in the current page (up
2773 * to the original size asked for in @req->wb_bytes).
2774 *
2775 * Calculate how many bytes are left in the layout segment
2776 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2777 *
2778 * Please also note that 'end_offset' is actually the offset of the
2779 * first byte that lies outside the pnfs_layout_range. FIXME?
2780 *
2781 */
19b54848 2782 if (pgio->pg_lseg) {
17822b20 2783 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2784 pgio->pg_lseg->pls_range.length);
2785 req_start = req_offset(req);
08cb5b0f 2786
c5e20cb7 2787 /* start of request is past the last byte of this segment */
08cb5b0f 2788 if (req_start >= seg_end)
19b54848 2789 return 0;
c5e20cb7
WAA		 2790
2791 /* adjust 'size' iff there are fewer bytes left in the
2792 * segment than what nfs_generic_pg_test returned */
2793 seg_left = seg_end - req_start;
2794 if (seg_left < size)
2795 size = (unsigned int)seg_left;
19b54848 2796 }
0f9c429e 2797
19b54848 2798 return size;
94ad1c80 2799}
89a58e32 2800EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2801
53113ad3 2802int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2803{
2804 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2805
2806 /* Resend all requests through the MDS */
53113ad3
WAA		 2807 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2808 hdr->completion_ops);
c7070113 2809 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2810 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2811}
e7dd79af 2812EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2813
d45f60c6 2814static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2815{
cd841605
FI		 2816
2817 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2818 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2819 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2820 pnfs_return_layout(hdr->inode);
1acbbb4e 2821 }
6c75dc0d 2822 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2823 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2824}
2825
d20581aa
BH		 2826/*
2827 * Called by non rpc-based layout drivers
2828 */
d45f60c6 2829void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2830{
f8417b48 2831 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2832 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2833 hdr->mds_offset + hdr->res.count);
d45f60c6 2834 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2835 }
2836 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2837 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2838 pnfs_ld_handle_write_error(hdr);
2839 hdr->mds_ops->rpc_release(hdr);
44b83799 2840}
d20581aa 2841EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2842
dce81290
TM		 2843static void
2844pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2845 struct nfs_pgio_header *hdr)
dce81290 2846{
48d635f1 2847 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2848
6c75dc0d 2849 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2850 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2851 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2852 mirror->pg_recoalesce = 1;
6c75dc0d 2853 }
ba4a76f7 2854 hdr->completion_ops->completion(hdr);
dce81290
TM		 2855}
2856
2857static enum pnfs_try_status
d45f60c6 2858pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2859 const struct rpc_call_ops *call_ops,
2860 struct pnfs_layout_segment *lseg,
2861 int how)
0382b744 2862{
cd841605 2863 struct inode *inode = hdr->inode;
0382b744
AA		 2864 enum pnfs_try_status trypnfs;
2865 struct nfs_server *nfss = NFS_SERVER(inode);
2866
cd841605 2867 hdr->mds_ops = call_ops;
0382b744
AA		 2868
2869 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2870 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2871 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2872 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2873 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2874 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2875 return trypnfs;
2876}
2877
dce81290 2878static void
7f714720
WAA		 2879pnfs_do_write(struct nfs_pageio_descriptor *desc,
2880 struct nfs_pgio_header *hdr, int how)
dce81290 2881{
dce81290
TM		 2882 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2883 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2884 enum pnfs_try_status trypnfs;
dce81290 2885
d45f60c6 2886 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2887 switch (trypnfs) {
2888 case PNFS_NOT_ATTEMPTED:
d45f60c6 2889 pnfs_write_through_mds(desc, hdr);
ffb81717 2890 break;
37f8aa16
TM		 2891 case PNFS_ATTEMPTED:
2892 break;
2893 case PNFS_TRY_AGAIN:
2894 /* cleanup hdr and prepare to redo pnfs */
2895 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2896 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2897 list_splice_init(&hdr->pages, &mirror->pg_list);
2898 mirror->pg_recoalesce = 1;
2899 }
2900 hdr->mds_ops->rpc_release(hdr);
2901 }
dce81290
TM		 2902}
2903
6c75dc0d
FI		 2904static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2905{
9369a431 2906 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2907 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2908}
2909
dce81290
TM		 2910int
2911pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2912{
6c75dc0d 2913 struct nfs_pgio_header *hdr;
dce81290
TM		 2914 int ret;
2915
1e7f3a48
WAA		 2916 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2917 if (!hdr) {
2bff2288
PT		 2918 desc->pg_error = -ENOMEM;
2919 return desc->pg_error;
dce81290 2920 }
6c75dc0d 2921 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2922
9369a431 2923 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2924 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2925 if (!ret)
7f714720 2926 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2927
6c75dc0d 2928 return ret;
dce81290
TM		 2929}
2930EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2931
53113ad3 2932int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2933{
2934 struct nfs_pageio_descriptor pgio;
2935
1acbbb4e 2936 /* Resend all requests through the MDS */
53113ad3
WAA		 2937 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2938 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2939}
e7dd79af 2940EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2941
d45f60c6 2942static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2943{
cd841605
FI		 2944 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2945 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2946 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2947 pnfs_return_layout(hdr->inode);
1acbbb4e 2948 }
4db6e0b7 2949 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2950 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2951}
2952
d20581aa
BH		 2953/*
2954 * Called by non rpc-based layout drivers
2955 */
d45f60c6 2956void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2957{
bfc505de 2958 if (likely(!hdr->pnfs_error))
d45f60c6 2959 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2960 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2961 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2962 pnfs_ld_handle_read_error(hdr);
2963 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2964}
2965EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2966
493292dd
TM		 2967static void
2968pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2969 struct nfs_pgio_header *hdr)
493292dd 2970{
48d635f1 2971 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2972
4db6e0b7 2973 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2974 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2975 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2976 mirror->pg_recoalesce = 1;
4db6e0b7 2977 }
ba4a76f7 2978 hdr->completion_ops->completion(hdr);
493292dd
TM		 2979}
2980
64419a9b
AA		 2981/*
2982 * Call the appropriate parallel I/O subsystem read function.
2983 */
493292dd 2984static enum pnfs_try_status
d45f60c6 2985pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2986 const struct rpc_call_ops *call_ops,
2987 struct pnfs_layout_segment *lseg)
64419a9b 2988{
cd841605 2989 struct inode *inode = hdr->inode;
64419a9b
AA		 2990 struct nfs_server *nfss = NFS_SERVER(inode);
2991 enum pnfs_try_status trypnfs;
2992
cd841605 2993 hdr->mds_ops = call_ops;
64419a9b
AA		 2994
2995 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2996 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2997
d45f60c6 2998 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2999 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 3000 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 3001 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
3002 return trypnfs;
3003}
863a3c6c 3004
ceb11e13 3005/* Resend all requests through pnfs. */
563c53e7
TM		 3006void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr,
3007 unsigned int mirror_idx)
ceb11e13
PT		 3008{
3009 struct nfs_pageio_descriptor pgio;
3010
1b1bc66b 3011 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 3012 /* Prevent deadlocks with layoutreturn! */
3013 pnfs_put_lseg(hdr->lseg);
3014 hdr->lseg = NULL;
3015
1b1bc66b
WAA		 3016 nfs_pageio_init_read(&pgio, hdr->inode, false,
3017 hdr->completion_ops);
563c53e7 3018 pgio.pg_mirror_idx = mirror_idx;
1b1bc66b
WAA		 3019 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
3020 }
ceb11e13
PT		 3021}
3022EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
3023
493292dd 3024static void
7f714720 3025pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 3026{
493292dd
TM		 3027 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
3028 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 3029 enum pnfs_try_status trypnfs;
493292dd 3030
d45f60c6 3031 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 3032 switch (trypnfs) {
3033 case PNFS_NOT_ATTEMPTED:
d45f60c6 3034 pnfs_read_through_mds(desc, hdr);
ffb81717 3035 break;
6aeafd05
TM		 3036 case PNFS_ATTEMPTED:
3037 break;
3038 case PNFS_TRY_AGAIN:
3039 /* cleanup hdr and prepare to redo pnfs */
3040 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3041 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3042 list_splice_init(&hdr->pages, &mirror->pg_list);
3043 mirror->pg_recoalesce = 1;
3044 }
3045 hdr->mds_ops->rpc_release(hdr);
3046 }
493292dd
TM		 3047}
3048
4db6e0b7
FI		 3049static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
3050{
9369a431 3051 pnfs_put_lseg(hdr->lseg);
1e7f3a48 3052 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 3053}
3054
493292dd
TM		 3055int
3056pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
3057{
4db6e0b7 3058 struct nfs_pgio_header *hdr;
493292dd
TM		 3059 int ret;
3060
1e7f3a48
WAA		 3061 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
3062 if (!hdr) {
2bff2288
PT		 3063 desc->pg_error = -ENOMEM;
3064 return desc->pg_error;
493292dd 3065 }
4db6e0b7 3066 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 3067 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 3068 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 3069 if (!ret)
7f714720 3070 pnfs_do_read(desc, hdr);
4db6e0b7 3071 return ret;
493292dd
TM		 3072}
3073EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
3074
71244d9b
TM		 3075static void pnfs_clear_layoutcommitting(struct inode *inode)
3076{
3077 unsigned long *bitlock = &NFS_I(inode)->flags;
3078
3079 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 3080 smp_mb__after_atomic();
71244d9b
TM		 3081 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
3082}
3083
863a3c6c 3084/*
a9bae566 3085 * There can be multiple RW segments.
863a3c6c 3086 */
a9bae566 3087static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 3088{
a9bae566 3089 struct pnfs_layout_segment *lseg;
863a3c6c 3090
a9bae566
PT		 3091 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
3092 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 3093 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 3094 list_add(&lseg->pls_lc_list, listp);
3095 }
863a3c6c
AA		 3096}
3097
a073dbff
TM		 3098static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
3099{
3100 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 3101
3102 /* Matched by references in pnfs_set_layoutcommit */
3103 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
3104 list_del_init(&lseg->pls_lc_list);
3105 pnfs_put_lseg(lseg);
3106 }
3107
71244d9b 3108 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 3109}
3110
1b0ae068
PT		 3111void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
3112{
b9e028fd 3113 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 3114}
3115EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
3116
863a3c6c 3117void
67af7611
TM		 3118pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
3119 loff_t end_pos)
863a3c6c 3120{
cd841605 3121 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 3122 bool mark_as_dirty = false;
863a3c6c 3123
cd841605 3124 spin_lock(&inode->i_lock);
863a3c6c 3125 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 3126 nfsi->layout->plh_lwb = end_pos;
79a48a1f 3127 mark_as_dirty = true;
863a3c6c 3128 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 3129 __func__, inode->i_ino);
29559b11
TM		 3130 } else if (end_pos > nfsi->layout->plh_lwb)
3131 nfsi->layout->plh_lwb = end_pos;
67af7611 3132 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 3133 /* references matched in nfs4_layoutcommit_release */
67af7611 3134 pnfs_get_lseg(lseg);
a9bae566 3135 }
cd841605 3136 spin_unlock(&inode->i_lock);
acff5880 3137 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 3138 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 3139
3140 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3141 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3142 if (mark_as_dirty)
cd841605 3143 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3144}
3145EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
3146
db29c089
AA		 3147void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
3148{
3149 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
3150
3151 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
3152 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 3153 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 3154}
3155
de4b15c7
AA		 3156/*
3157 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3158 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3159 * data to disk to allow the server to recover the data if it crashes.
3160 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3161 * is off, and a COMMIT is sent to a data server, or
3162 * if WRITEs to a data server return NFS_DATA_SYNC.
3163 */
863a3c6c 3164int
ef311537 3165pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 3166{
5f919c9f 3167 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 3168 struct nfs4_layoutcommit_data *data;
3169 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 3170 loff_t end_pos;
71244d9b 3171 int status;
863a3c6c 3172
71244d9b 3173 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 3174 return 0;
3175
71244d9b 3176 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 3177
71244d9b 3178 status = -EAGAIN;
92407e75 3179 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 3180 if (!sync)
3181 goto out;
74316201 3182 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 3183 NFS_INO_LAYOUTCOMMITTING,
3184 nfs_wait_bit_killable,
3185 TASK_KILLABLE);
92407e75 3186 if (status)
71244d9b 3187 goto out;
92407e75
PT		 3188 }
3189
71244d9b
TM		 3190 status = -ENOMEM;
3191 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
63d8a41b 3192 data = kzalloc(sizeof(*data), nfs_io_gfp_mask());
71244d9b
TM		 3193 if (!data)
3194 goto clear_layoutcommitting;
3195
3196 status = 0;
de4b15c7 3197 spin_lock(&inode->i_lock);
71244d9b
TM		 3198 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
3199 goto out_unlock;
a9bae566 3200
71244d9b 3201 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 3202 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 3203
acff5880 3204 end_pos = nfsi->layout->plh_lwb;
863a3c6c 3205
f597c537 3206 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
97a728f5 3207 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 3208 spin_unlock(&inode->i_lock);
3209
3210 data->args.inode = inode;
863a3c6c
AA		 3211 nfs_fattr_init(&data->fattr);
3212 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3213 data->res.fattr = &data->fattr;
2e18d4d8
TM		 3214 if (end_pos != 0)
3215 data->args.lastbytewritten = end_pos - 1;
3216 else
3217 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 3218 data->res.server = NFS_SERVER(inode);
3219
5f919c9f
CH		 3220 if (ld->prepare_layoutcommit) {
3221 status = ld->prepare_layoutcommit(&data->args);
3222 if (status) {
a52458b4 3223 put_cred(data->cred);
5f919c9f 3224 spin_lock(&inode->i_lock);
29559b11
TM		 3225 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3226 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 3227 nfsi->layout->plh_lwb = end_pos;
3471648a 3228 goto out_unlock;
5f919c9f
CH		 3229 }
3230 }
3231
3232
863a3c6c
AA		 3233 status = nfs4_proc_layoutcommit(data, sync);
3234out:
92407e75
PT		 3235 if (status)
3236 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3237 dprintk("<-- %s status %d\n", __func__, status);
3238 return status;
71244d9b
TM		 3239out_unlock:
3240 spin_unlock(&inode->i_lock);
92407e75 3241 kfree(data);
71244d9b
TM		 3242clear_layoutcommitting:
3243 pnfs_clear_layoutcommitting(inode);
92407e75 3244 goto out;
863a3c6c 3245}
72cff449 3246EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 3247
5bb89b47
TM		 3248int
3249pnfs_generic_sync(struct inode *inode, bool datasync)
3250{
3251 return pnfs_layoutcommit_inode(inode, true);
3252}
3253EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3254
82be417a
AA		 3255struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3256{
3257 struct nfs4_threshold *thp;
3258
63d8a41b 3259 thp = kzalloc(sizeof(*thp), nfs_io_gfp_mask());
82be417a
AA		 3260 if (!thp) {
3261 dprintk("%s mdsthreshold allocation failed\n", __func__);
3262 return NULL;
3263 }
3264 return thp;
3265}
8733408d 3266
865a7ecb 3267#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 3268int
c8ad8894 3269pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 3270{
3271 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3272 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 3273 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 3274 struct nfs42_layoutstat_data *data;
3275 struct pnfs_layout_hdr *hdr;
3276 int status = 0;
3277
3278 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3279 goto out;
3280
6c5a0d89
TM		 3281 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3282 goto out;
3283
1bfe3b25
PT		 3284 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3285 goto out;
3286
8733408d
PT		 3287 spin_lock(&inode->i_lock);
3288 if (!NFS_I(inode)->layout) {
3289 spin_unlock(&inode->i_lock);
f538d0ba 3290 goto out_clear_layoutstats;
8733408d
PT		 3291 }
3292 hdr = NFS_I(inode)->layout;
3293 pnfs_get_layout_hdr(hdr);
3294 spin_unlock(&inode->i_lock);
3295
c8ad8894 3296 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3297 if (!data) {
3298 status = -ENOMEM;
3299 goto out_put;
3300 }
3301
3302 data->args.fh = NFS_FH(inode);
3303 data->args.inode = inode;
8733408d
PT		 3304 status = ld->prepare_layoutstats(&data->args);
3305 if (status)
3306 goto out_free;
3307
3308 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3309
3310out:
3311 dprintk("%s returns %d\n", __func__, status);
3312 return status;
3313
3314out_free:
3315 kfree(data);
3316out_put:
3317 pnfs_put_layout_hdr(hdr);
f538d0ba 3318out_clear_layoutstats:
1bfe3b25
PT		 3319 smp_mb__before_atomic();
3320 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3321 smp_mb__after_atomic();
8733408d
PT		 3322 goto out;
3323}
3324EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3325#endif
bbf58bf3
TM		 3326
3327unsigned int layoutstats_timer;
3328module_param(layoutstats_timer, uint, 0644);
3329EXPORT_SYMBOL_GPL(layoutstats_timer);
Empty description
This page took 1.151581 seconds and 4 git commands to generate.