projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'trace-v5.13-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[linux-2.6-block.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
95void
96unset_pnfs_layoutdriver(struct nfs_server *nfss)
97{
738fd0f3
BH		 98 if (nfss->pnfs_curr_ld) {
99 if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 104 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 105 }
85e174ba
RL		 106 nfss->pnfs_curr_ld = NULL;
107}
108
ca440c38
JL		 109/*
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
112 *
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
115 */
116static const u32 ld_prefs[] = {
117 LAYOUT_SCSI,
118 LAYOUT_BLOCK_VOLUME,
119 LAYOUT_OSD2_OBJECTS,
120 LAYOUT_FLEX_FILES,
121 LAYOUT_NFSV4_1_FILES,
122 0
123};
124
125static int
126ld_cmp(const void *e1, const void *e2)
127{
128 u32 ld1 = *((u32 *)e1);
129 u32 ld2 = *((u32 *)e2);
130 int i;
131
132 for (i = 0; ld_prefs[i] != 0; i++) {
133 if (ld1 == ld_prefs[i])
134 return -1;
135
136 if (ld2 == ld_prefs[i])
137 return 1;
138 }
139 return 0;
140}
141
85e174ba
RL		 142/*
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
145 *
3132e49e 146 * @ids array of layout types supported by MDS.
85e174ba
RL		 147 */
148void
738fd0f3 149set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 150 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 151{
152 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 153 u32 id;
ca440c38 154 int i;
85e174ba 155
19274716
AS		 156 if (fsinfo->nlayouttypes == 0)
157 goto out_no_driver;
85e174ba
RL		 158 if (!(server->nfs_client->cl_exchange_flags &
159 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 160 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 162 goto out_no_driver;
163 }
3132e49e 164
ca440c38
JL		 165 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 167
ca440c38
JL		 168 for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 id = fsinfo->layouttype[i];
85e174ba 170 ld_type = find_pnfs_driver(id);
ca440c38
JL		 171 if (!ld_type) {
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 id);
174 ld_type = find_pnfs_driver(id);
175 }
176 if (ld_type)
177 break;
85e174ba 178 }
3132e49e
JL		 179
180 if (!ld_type) {
ca440c38 181 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 182 goto out_no_driver;
183 }
184
85e174ba 185 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 186 if (ld_type->set_layoutdriver
187 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 188 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__, id);
738fd0f3
BH		 190 module_put(ld_type->owner);
191 goto out_no_driver;
192 }
2a4c8994
TM		 193 /* Bump the MDS count */
194 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 195
85e174ba
RL		 196 dprintk("%s: pNFS module for %u set\n", __func__, id);
197 return;
198
199out_no_driver:
200 dprintk("%s: Using NFSv4 I/O\n", __func__);
201 server->pnfs_curr_ld = NULL;
202}
02c35fca
FI		 203
204int
205pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206{
207 int status = -EINVAL;
208 struct pnfs_layoutdriver_type *tmp;
209
210 if (ld_type->id == 0) {
a030889a 211 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 212 return status;
213 }
b1f69b75 214 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 215 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 216 "alloc_lseg and free_lseg.\n", __func__);
217 return status;
218 }
02c35fca
FI		 219
220 spin_lock(&pnfs_spinlock);
221 tmp = find_pnfs_driver_locked(ld_type->id);
222 if (!tmp) {
223 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 status = 0;
225 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 ld_type->name);
227 } else {
a030889a 228 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 229 __func__, ld_type->id);
230 }
231 spin_unlock(&pnfs_spinlock);
232
233 return status;
234}
235EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236
237void
238pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239{
240 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 spin_lock(&pnfs_spinlock);
242 list_del(&ld_type->pnfs_tblid);
243 spin_unlock(&pnfs_spinlock);
244}
245EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 246
b1f69b75
AA		 247/*
248 * pNFS client layout cache
249 */
250
cc6e5340 251/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 252void
70c3bd2b 253pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 254{
2b28a7be 255 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 256}
257
636fb9c8
BH		 258static struct pnfs_layout_hdr *
259pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260{
261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 262 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 263}
264
265static void
266pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267{
9c626381
TM		 268 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270
cf6605d1 271 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
9c626381
TM		 272 struct nfs_client *clp = server->nfs_client;
273
274 spin_lock(&clp->cl_lock);
cf6605d1 275 list_del_rcu(&lo->plh_layouts);
9c626381
TM		 276 spin_unlock(&clp->cl_lock);
277 }
a52458b4 278 put_cred(lo->plh_lc_cred);
57934278 279 return ld->free_layout_hdr(lo);
636fb9c8
BH		 280}
281
e5e94017 282static void
6622c3ea 283pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 284{
bb346f63 285 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 286 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 287 nfsi->layout = NULL;
288 /* Reset MDS Threshold I/O counters */
289 nfsi->write_io = 0;
290 nfsi->read_io = 0;
e5e94017
BH		 291}
292
b1f69b75 293void
70c3bd2b 294pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 295{
9c6376eb 296 struct inode *inode;
b6d49ecd 297 unsigned long i_state;
cc6e5340 298
9c6376eb
TM		 299 if (!lo)
300 return;
301 inode = lo->plh_inode;
13c13a6a
TM		 302 pnfs_layoutreturn_before_put_layout_hdr(lo);
303
2b28a7be 304 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 305 if (!list_empty(&lo->plh_segs))
306 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 307 pnfs_detach_layout_hdr(lo);
b6d49ecd 308 i_state = inode->i_state;
cc6e5340 309 spin_unlock(&inode->i_lock);
6622c3ea 310 pnfs_free_layout_hdr(lo);
b6d49ecd
TM		 311 /* Notify pnfs_destroy_layout_final() that we're done */
312 if (i_state & (I_FREEING | I_CLEAR))
313 wake_up_var(lo);
cc6e5340 314 }
974cec8c
AA		 315}
316
b5fdf841
TM		 317static struct inode *
318pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
319{
320 struct inode *inode = igrab(lo->plh_inode);
321 if (inode)
322 return inode;
323 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
324 return NULL;
325}
326
1bcf34fd
TM		 327/*
328 * Compare 2 layout stateid sequence ids, to see which is newer,
329 * taking into account wraparound issues.
330 */
331static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
332{
333 return (s32)(s1 - s2) > 0;
334}
335
336static void pnfs_barrier_update(struct pnfs_layout_hdr *lo, u32 newseq)
337{
338 if (pnfs_seqid_is_newer(newseq, lo->plh_barrier))
339 lo->plh_barrier = newseq;
340}
341
4aab9732
TM		 342static void
343pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
344 u32 seq)
345{
346 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
347 iomode = IOMODE_ANY;
348 lo->plh_return_iomode = iomode;
349 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
350 if (seq != 0) {
351 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
352 lo->plh_return_seq = seq;
1bcf34fd 353 pnfs_barrier_update(lo, seq);
4aab9732
TM		 354 }
355}
356
ae5a459d
TM		 357static void
358pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
359{
5466d214 360 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 361 lo->plh_return_iomode = 0;
362 lo->plh_return_seq = 0;
363 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 364 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
365 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
366 continue;
367 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
368 }
ae5a459d
TM		 369}
370
362fb578
TM		 371static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
372{
373 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
374 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
375 smp_mb__after_atomic();
376 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
377 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
378}
379
fe1cf946
TM		 380static void
381pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
382 struct list_head *free_me)
383{
384 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
385 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
386 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
387 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
388 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
389 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
390}
391
7380020e 392/*
30cb3ee2
TM		 393 * Update the seqid of a layout stateid after receiving
394 * NFS4ERR_OLD_STATEID
7380020e 395 */
30cb3ee2 396bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
ecf84026
TM		 397 struct pnfs_layout_range *dst_range,
398 struct inode *inode)
7380020e
TM		 399{
400 struct pnfs_layout_hdr *lo;
c16467dc
TM		 401 struct pnfs_layout_range range = {
402 .iomode = IOMODE_ANY,
403 .offset = 0,
404 .length = NFS4_MAX_UINT64,
405 };
7380020e 406 bool ret = false;
c16467dc
TM		 407 LIST_HEAD(head);
408 int err;
7380020e
TM		 409
410 spin_lock(&inode->i_lock);
411 lo = NFS_I(inode)->layout;
30cb3ee2
TM		 412 if (lo && pnfs_layout_is_valid(lo) &&
413 nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
414 /* Is our call using the most recent seqid? If so, bump it */
415 if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
416 nfs4_stateid_seqid_inc(dst);
417 ret = true;
418 goto out;
419 }
420 /* Try to update the seqid to the most recent */
c16467dc
TM		 421 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
422 if (err != -EBUSY) {
423 dst->seqid = lo->plh_stateid.seqid;
ecf84026 424 *dst_range = range;
c16467dc
TM		 425 ret = true;
426 }
7380020e 427 }
30cb3ee2 428out:
7380020e 429 spin_unlock(&inode->i_lock);
c16467dc 430 pnfs_free_lseg_list(&head);
7380020e
TM		 431 return ret;
432}
433
2454dfea
TM		 434/*
435 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
436 *
437 * In order to continue using the pnfs_layout_hdr, a full recovery
438 * is required.
439 * Note that caller must hold inode->i_lock.
440 */
5f46be04 441int
2454dfea
TM		 442pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
443 struct list_head *lseg_list)
444{
445 struct pnfs_layout_range range = {
446 .iomode = IOMODE_ANY,
447 .offset = 0,
448 .length = NFS4_MAX_UINT64,
449 };
fe1cf946 450 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 451
452 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 453 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
454 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 455 pnfs_clear_layoutreturn_info(lo);
68f74479 456 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 457 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
458 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
459 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 460 return !list_empty(&lo->plh_segs);
2454dfea
TM		 461}
462
b9e028fd
TM		 463static int
464pnfs_iomode_to_fail_bit(u32 iomode)
465{
466 return iomode == IOMODE_RW ?
467 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
468}
469
470static void
3e621214 471pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 472{
25c75333 473 lo->plh_retry_timestamp = jiffies;
39e88fcf 474 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 475 refcount_inc(&lo->plh_refcount);
3e621214
TM		 476}
477
478static void
479pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
480{
481 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 482 refcount_dec(&lo->plh_refcount);
3e621214
TM		 483}
484
485static void
486pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
487{
488 struct inode *inode = lo->plh_inode;
115ce575
TM		 489 struct pnfs_layout_range range = {
490 .iomode = iomode,
491 .offset = 0,
492 .length = NFS4_MAX_UINT64,
493 };
494 LIST_HEAD(head);
3e621214
TM		 495
496 spin_lock(&inode->i_lock);
497 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 498 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 499 spin_unlock(&inode->i_lock);
115ce575 500 pnfs_free_lseg_list(&head);
b9e028fd
TM		 501 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
502 iomode == IOMODE_RW ? "RW" : "READ");
503}
504
505static bool
506pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
507{
25c75333 508 unsigned long start, end;
3e621214
TM		 509 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
510
511 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 512 return false;
513 end = jiffies;
514 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
515 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
516 /* It is time to retry the failed layoutgets */
3e621214 517 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 518 return false;
519 }
520 return true;
b9e028fd
TM		 521}
522
974cec8c 523static void
119cef97
TM		 524pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
525 const struct pnfs_layout_range *range,
526 const nfs4_stateid *stateid)
974cec8c 527{
566052c5 528 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 529 INIT_LIST_HEAD(&lseg->pls_lc_list);
a9901899 530 INIT_LIST_HEAD(&lseg->pls_commits);
eba6dd69 531 refcount_set(&lseg->pls_refcount, 1);
4541d16c 532 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 533 lseg->pls_layout = lo;
119cef97
TM		 534 lseg->pls_range = *range;
535 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 536}
537
905ca191 538static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 539{
68f74479
TM		 540 if (lseg != NULL) {
541 struct inode *inode = lseg->pls_layout->plh_inode;
542 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
543 }
974cec8c
AA		 544}
545
d684d2ae 546static void
57036a37
TM		 547pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
548 struct pnfs_layout_segment *lseg)
d684d2ae 549{
d20581aa 550 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 551 list_del_init(&lseg->pls_list);
8f0d27dc 552 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 553 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 554 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
555 return;
7b650994
TM		 556 if (list_empty(&lo->plh_segs) &&
557 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
558 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 559 if (atomic_read(&lo->plh_outstanding) == 0)
560 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 561 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 562 }
d684d2ae
FI		 563}
564
68f74479
TM		 565static bool
566pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
567 struct pnfs_layout_segment *lseg)
568{
569 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
570 pnfs_layout_is_valid(lo)) {
4aab9732 571 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 572 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
573 return true;
574 }
575 return false;
576}
577
bae724ef 578void
9369a431 579pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 580{
57036a37 581 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 582 struct inode *inode;
583
584 if (!lseg)
585 return;
586
4541d16c 587 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 588 refcount_read(&lseg->pls_refcount),
4541d16c 589 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 590
57036a37
TM		 591 lo = lseg->pls_layout;
592 inode = lo->plh_inode;
4ef2e4f8 593
eba6dd69 594 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 595 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
596 spin_unlock(&inode->i_lock);
597 return;
598 }
8f0d27dc 599 pnfs_get_layout_hdr(lo);
4ef2e4f8 600 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 601 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
602 lseg = NULL;
4ef2e4f8
TM		 603 spin_unlock(&inode->i_lock);
604 pnfs_free_lseg(lseg);
605 pnfs_put_layout_hdr(lo);
4541d16c 606 }
4541d16c 607}
9369a431 608EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 609
fb3296eb
BH		 610/*
611 * is l2 fully contained in l1?
612 * start1 end1
613 * [----------------------------------)
614 * start2 end2
615 * [----------------)
616 */
3cb2df17 617static bool
7dc0ac70 618pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 619 const struct pnfs_layout_range *l2)
fb3296eb
BH		 620{
621 u64 start1 = l1->offset;
17822b20 622 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 623 u64 start2 = l2->offset;
17822b20 624 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 625
626 return (start1 <= start2) && (end1 >= end2);
627}
628
24956804
TM		 629static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
630 struct list_head *tmp_list)
631{
eba6dd69 632 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 633 return false;
634 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
635 list_add(&lseg->pls_list, tmp_list);
636 return true;
637}
638
4541d16c
FI		 639/* Returns 1 if lseg is removed from list, 0 otherwise */
640static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
641 struct list_head *tmp_list)
642{
643 int rv = 0;
644
645 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
646 /* Remove the reference keeping the lseg in the
647 * list. It will now be removed when all
648 * outstanding io is finished.
649 */
d684d2ae 650 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 651 refcount_read(&lseg->pls_refcount));
24956804 652 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 653 rv = 1;
4541d16c
FI		 654 }
655 return rv;
656}
657
e036f464
TM		 658static bool
659pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
660 const struct pnfs_layout_range *recall_range)
661{
662 return (recall_range->iomode == IOMODE_ANY ||
663 lseg_range->iomode == recall_range->iomode) &&
664 pnfs_lseg_range_intersecting(lseg_range, recall_range);
665}
666
667static bool
668pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
669 const struct pnfs_layout_range *recall_range,
670 u32 seq)
671{
672 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
673 return false;
674 if (recall_range == NULL)
675 return true;
676 return pnfs_should_free_range(&lseg->pls_range, recall_range);
677}
678
6d597e17
JL		 679/**
680 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
681 * @lo: layout header containing the lsegs
682 * @tmp_list: list head where doomed lsegs should go
683 * @recall_range: optional recall range argument to match (may be NULL)
684 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
685 *
686 * Walk the list of lsegs in the layout header, and tear down any that should
687 * be destroyed. If "recall_range" is specified then the segment must match
688 * that range. If "seq" is non-zero, then only match segments that were handed
689 * out at or before that sequence.
690 *
691 * Returns number of matching invalid lsegs remaining in list after scanning
692 * it and purging them.
4541d16c 693 */
43f1b3da 694int
49a85061 695pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 696 struct list_head *tmp_list,
6d597e17
JL		 697 const struct pnfs_layout_range *recall_range,
698 u32 seq)
974cec8c
AA		 699{
700 struct pnfs_layout_segment *lseg, *next;
71b39854 701 int remaining = 0;
974cec8c
AA		 702
703 dprintk("%s:Begin lo %p\n", __func__, lo);
704
8006bfba 705 if (list_empty(&lo->plh_segs))
38511722 706 return 0;
4541d16c 707 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 708 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 709 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 710 "offset %llu length %llu\n", __func__,
6d597e17
JL		 711 lseg, lseg->pls_range.iomode, lseg->pls_seq,
712 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 713 if (!mark_lseg_invalid(lseg, tmp_list))
714 remaining++;
4541d16c 715 }
71b39854
TM		 716 dprintk("%s:Return %i\n", __func__, remaining);
717 return remaining;
974cec8c
AA		 718}
719
68f74479
TM		 720static void
721pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
722 struct list_head *free_me,
723 const struct pnfs_layout_range *range,
724 u32 seq)
725{
726 struct pnfs_layout_segment *lseg, *next;
727
728 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
729 if (pnfs_match_lseg_recall(lseg, range, seq))
730 list_move_tail(&lseg->pls_list, free_me);
731 }
732}
733
f49f9baa 734/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 735void
4541d16c 736pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 737{
4541d16c 738 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 739
740 if (list_empty(free_me))
741 return;
742
4541d16c 743 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 744 list_del(&lseg->pls_list);
905ca191 745 pnfs_free_lseg(lseg);
974cec8c
AA		 746 }
747}
748
b6d49ecd 749static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi)
e5e94017
BH		 750{
751 struct pnfs_layout_hdr *lo;
974cec8c 752 LIST_HEAD(tmp_list);
e5e94017
BH		 753
754 spin_lock(&nfsi->vfs_inode.i_lock);
755 lo = nfsi->layout;
756 if (lo) {
3e621214 757 pnfs_get_layout_hdr(lo);
2454dfea 758 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 759 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
760 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
761 spin_unlock(&nfsi->vfs_inode.i_lock);
762 pnfs_free_lseg_list(&tmp_list);
1f18b82c 763 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 764 pnfs_put_layout_hdr(lo);
765 } else
766 spin_unlock(&nfsi->vfs_inode.i_lock);
b6d49ecd
TM		 767 return lo;
768}
769
770void pnfs_destroy_layout(struct nfs_inode *nfsi)
771{
772 __pnfs_destroy_layout(nfsi);
974cec8c 773}
041245c8 774EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 775
b6d49ecd
TM		 776static bool pnfs_layout_removed(struct nfs_inode *nfsi,
777 struct pnfs_layout_hdr *lo)
778{
779 bool ret;
780
781 spin_lock(&nfsi->vfs_inode.i_lock);
782 ret = nfsi->layout != lo;
783 spin_unlock(&nfsi->vfs_inode.i_lock);
784 return ret;
785}
786
787void pnfs_destroy_layout_final(struct nfs_inode *nfsi)
788{
789 struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi);
790
791 if (lo)
792 wait_var_event(lo, pnfs_layout_removed(nfsi, lo));
793}
794
fd9a8d71
TM		 795static bool
796pnfs_layout_add_bulk_destroy_list(struct inode *inode,
797 struct list_head *layout_list)
974cec8c
AA		 798{
799 struct pnfs_layout_hdr *lo;
fd9a8d71 800 bool ret = false;
974cec8c 801
fd9a8d71
TM		 802 spin_lock(&inode->i_lock);
803 lo = NFS_I(inode)->layout;
804 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
805 pnfs_get_layout_hdr(lo);
806 list_add(&lo->plh_bulk_destroy, layout_list);
807 ret = true;
808 }
809 spin_unlock(&inode->i_lock);
810 return ret;
811}
812
813/* Caller must hold rcu_read_lock and clp->cl_lock */
814static int
815pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
816 struct nfs_server *server,
817 struct list_head *layout_list)
5085607d
TM		 818 __must_hold(&clp->cl_lock)
819 __must_hold(RCU)
fd9a8d71
TM		 820{
821 struct pnfs_layout_hdr *lo, *next;
822 struct inode *inode;
823
824 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
5085607d
TM		 825 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
826 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
827 !list_empty(&lo->plh_bulk_destroy))
b85f5620 828 continue;
5085607d
TM		 829 /* If the sb is being destroyed, just bail */
830 if (!nfs_sb_active(server->super))
831 break;
b5fdf841 832 inode = pnfs_grab_inode_layout_hdr(lo);
5085607d 833 if (inode != NULL) {
cf6605d1
TM		 834 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
835 list_del_rcu(&lo->plh_layouts);
5085607d
TM		 836 if (pnfs_layout_add_bulk_destroy_list(inode,
837 layout_list))
838 continue;
839 rcu_read_unlock();
840 spin_unlock(&clp->cl_lock);
841 iput(inode);
842 } else {
843 rcu_read_unlock();
844 spin_unlock(&clp->cl_lock);
5085607d
TM		 845 }
846 nfs_sb_deactive(server->super);
fd9a8d71
TM		 847 spin_lock(&clp->cl_lock);
848 rcu_read_lock();
849 return -EAGAIN;
850 }
851 return 0;
852}
853
854static int
855pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
856 bool is_bulk_recall)
857{
858 struct pnfs_layout_hdr *lo;
859 struct inode *inode;
fd9a8d71
TM		 860 LIST_HEAD(lseg_list);
861 int ret = 0;
862
863 while (!list_empty(layout_list)) {
864 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
865 plh_bulk_destroy);
866 dprintk("%s freeing layout for inode %lu\n", __func__,
867 lo->plh_inode->i_ino);
868 inode = lo->plh_inode;
7c5d1875
CH		 869
870 pnfs_layoutcommit_inode(inode, false);
871
fd9a8d71
TM		 872 spin_lock(&inode->i_lock);
873 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 874 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
875 if (is_bulk_recall)
876 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 877 ret = -EAGAIN;
9fd4b9fc 878 }
fd9a8d71
TM		 879 spin_unlock(&inode->i_lock);
880 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 881 /* Free all lsegs that are attached to commit buckets */
882 nfs_commit_inode(inode, 0);
fd9a8d71 883 pnfs_put_layout_hdr(lo);
5085607d 884 nfs_iput_and_deactive(inode);
fd9a8d71
TM		 885 }
886 return ret;
887}
888
889int
890pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
891 struct nfs_fsid *fsid,
892 bool is_recall)
893{
894 struct nfs_server *server;
895 LIST_HEAD(layout_list);
c47abcf8 896
974cec8c 897 spin_lock(&clp->cl_lock);
6382a441 898 rcu_read_lock();
fd9a8d71 899restart:
6382a441 900 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 901 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
902 continue;
903 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
904 server,
905 &layout_list) != 0)
906 goto restart;
6382a441
WAA		 907 }
908 rcu_read_unlock();
974cec8c
AA		 909 spin_unlock(&clp->cl_lock);
910
fd9a8d71
TM		 911 if (list_empty(&layout_list))
912 return 0;
913 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
914}
915
916int
917pnfs_destroy_layouts_byclid(struct nfs_client *clp,
918 bool is_recall)
919{
920 struct nfs_server *server;
921 LIST_HEAD(layout_list);
922
923 spin_lock(&clp->cl_lock);
924 rcu_read_lock();
925restart:
926 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
927 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
928 server,
929 &layout_list) != 0)
930 goto restart;
974cec8c 931 }
fd9a8d71
TM		 932 rcu_read_unlock();
933 spin_unlock(&clp->cl_lock);
934
935 if (list_empty(&layout_list))
936 return 0;
937 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
938}
939
940/*
9f266451 941 * Called by the state manager to remove all layouts established under an
fd9a8d71
TM		 942 * expired lease.
943 */
944void
945pnfs_destroy_all_layouts(struct nfs_client *clp)
946{
947 nfs4_deviceid_mark_client_invalid(clp);
948 nfs4_deviceid_purge_client(clp);
949
950 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 951}
952
59b56394
TM		 953static void
954pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
955{
956 const struct cred *old;
957
958 if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
959 old = xchg(&lo->plh_lc_cred, get_cred(cred));
960 put_cred(old);
961 }
962}
963
fd6002e9 964/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 965void
966pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
59b56394 967 const struct cred *cred, bool update_barrier)
b1f69b75 968{
ecebb80b 969 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 970
2d2f24ad
TM		 971 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
972 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 973
974 if (!pnfs_layout_is_valid(lo)) {
59b56394 975 pnfs_set_layout_cred(lo, cred);
2a59a041
TM		 976 nfs4_stateid_copy(&lo->plh_stateid, new);
977 lo->plh_barrier = newseq;
978 pnfs_clear_layoutreturn_info(lo);
979 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
980 return;
981 }
982 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 983 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 984 /*
985 * Because of wraparound, we want to keep the barrier
986 * "close" to the current seqids.
987 */
988 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 989 }
ecebb80b
TM		 990 if (update_barrier)
991 new_barrier = be32_to_cpu(new->seqid);
992 else if (new_barrier == 0)
993 return;
1bcf34fd 994 pnfs_barrier_update(lo, new_barrier);
b1f69b75
AA		 995}
996
cf7d63f1 997static bool
19c54aba
TM		 998pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
999 const nfs4_stateid *stateid)
43f1b3da 1000{
19c54aba 1001 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 1002
d29b468d 1003 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier) && lo->plh_barrier;
19c54aba
TM		 1004}
1005
1006/* lget is set to 1 if called from inside send_layoutget call chain */
1007static bool
e1c06f80 1008pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 1009{
f7e8917a 1010 return lo->plh_block_lgets ||
e1c06f80 1011 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 1012}
1013
5e36e2a9
FI		 1014static struct nfs_server *
1015pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
1016{
1017 struct nfs_server *server;
1018
78746a38 1019 if (inode) {
5e36e2a9 1020 server = NFS_SERVER(inode);
78746a38 1021 } else {
5e36e2a9
FI		 1022 struct dentry *parent_dir = dget_parent(ctx->dentry);
1023 server = NFS_SERVER(parent_dir->d_inode);
1024 dput(parent_dir);
1025 }
1026 return server;
1027}
1028
29a8bfe5
TM		 1029static void nfs4_free_pages(struct page **pages, size_t size)
1030{
1031 int i;
1032
1033 if (!pages)
1034 return;
1035
1036 for (i = 0; i < size; i++) {
1037 if (!pages[i])
1038 break;
1039 __free_page(pages[i]);
1040 }
1041 kfree(pages);
1042}
1043
1044static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1045{
1046 struct page **pages;
1047 int i;
1048
a2791d3a 1049 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 1050 if (!pages) {
1051 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1052 return NULL;
1053 }
1054
1055 for (i = 0; i < size; i++) {
1056 pages[i] = alloc_page(gfp_flags);
1057 if (!pages[i]) {
1058 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 1059 nfs4_free_pages(pages, i);
29a8bfe5
TM		 1060 return NULL;
1061 }
1062 }
1063
1064 return pages;
1065}
1066
587f03de 1067static struct nfs4_layoutget *
5e36e2a9 1068pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 1069 struct nfs_open_context *ctx,
2409a976 1070 const nfs4_stateid *stateid,
e144e539 1071 const struct pnfs_layout_range *range,
587f03de 1072 gfp_t gfp_flags)
e5e94017 1073{
5e36e2a9 1074 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 1075 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 1076 size_t max_pages = max_response_pages(server);
b1f69b75 1077 struct nfs4_layoutget *lgp;
b1f69b75
AA		 1078
1079 dprintk("--> %s\n", __func__);
e5e94017 1080
83026d80
JL		 1081 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1082 if (lgp == NULL)
587f03de 1083 return NULL;
83026d80 1084
28ced9a8
TM		 1085 if (max_reply_sz) {
1086 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1087 if (npages < max_pages)
1088 max_pages = npages;
1089 }
1090
dacb452d
FI		 1091 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1092 if (!lgp->args.layout.pages) {
1093 kfree(lgp);
1094 return NULL;
1095 }
1096 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1097 lgp->res.layoutp = &lgp->args.layout;
1098
d49e0d5b
TM		 1099 /* Don't confuse uninitialised result and success */
1100 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1101
1102 lgp->args.minlength = PAGE_SIZE;
1103 if (lgp->args.minlength > range->length)
1104 lgp->args.minlength = range->length;
5e36e2a9
FI		 1105 if (ino) {
1106 loff_t i_size = i_size_read(ino);
1107
1108 if (range->iomode == IOMODE_READ) {
1109 if (range->offset >= i_size)
1110 lgp->args.minlength = 0;
1111 else if (i_size - range->offset < lgp->args.minlength)
1112 lgp->args.minlength = i_size - range->offset;
1113 }
d03ab29d 1114 }
83026d80
JL		 1115 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1116 pnfs_copy_range(&lgp->args.range, range);
1117 lgp->args.type = server->pnfs_curr_ld->id;
1118 lgp->args.inode = ino;
1119 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1120 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1121 lgp->gfp_flags = gfp_flags;
63ec2b69 1122 lgp->cred = ctx->cred;
587f03de 1123 return lgp;
974cec8c
AA		 1124}
1125
29a8bfe5
TM		 1126void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1127{
1128 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1129
1130 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1131 if (lgp->args.inode)
1132 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
29a8bfe5
TM		 1133 put_nfs_open_context(lgp->args.ctx);
1134 kfree(lgp);
1135}
1136
24956804
TM		 1137static void pnfs_clear_layoutcommit(struct inode *inode,
1138 struct list_head *head)
1139{
1140 struct nfs_inode *nfsi = NFS_I(inode);
1141 struct pnfs_layout_segment *lseg, *tmp;
1142
1143 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1144 return;
1145 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1146 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1147 continue;
1148 pnfs_lseg_dec_and_remove_zero(lseg, head);
1149 }
1150}
1151
68f74479 1152void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1153 const nfs4_stateid *arg_stateid,
68f74479 1154 const struct pnfs_layout_range *range,
68f74479
TM		 1155 const nfs4_stateid *stateid)
1156{
1157 struct inode *inode = lo->plh_inode;
1158 LIST_HEAD(freeme);
1159
1160 spin_lock(&inode->i_lock);
c18d1e17 1161 if (!pnfs_layout_is_valid(lo) ||
2a974425
TM		 1162 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1163 goto out_unlock;
68f74479 1164 if (stateid) {
2a974425
TM		 1165 u32 seq = be32_to_cpu(arg_stateid->seqid);
1166
68f74479
TM		 1167 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1168 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
59b56394 1169 pnfs_set_layout_stateid(lo, stateid, NULL, true);
68f74479
TM		 1170 } else
1171 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1172out_unlock:
68f74479
TM		 1173 pnfs_clear_layoutreturn_waitbit(lo);
1174 spin_unlock(&inode->i_lock);
1175 pnfs_free_lseg_list(&freeme);
1176
1177}
1178
13c13a6a 1179static bool
e5fd1904
TM		 1180pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1181 nfs4_stateid *stateid,
44ea8dfc 1182 const struct cred **cred,
e5fd1904 1183 enum pnfs_iomode *iomode)
13c13a6a 1184{
bf0291dd
TM		 1185 /* Serialise LAYOUTGET/LAYOUTRETURN */
1186 if (atomic_read(&lo->plh_outstanding) != 0)
1187 return false;
6604b203 1188 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1189 return false;
6604b203 1190 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1191 pnfs_get_layout_hdr(lo);
1bcf34fd
TM		 1192 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1193 *cred = get_cred(lo->plh_lc_cred);
e5fd1904 1194 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
44ea8dfc
TM		 1195 if (lo->plh_return_seq != 0)
1196 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
e5fd1904
TM		 1197 if (iomode != NULL)
1198 *iomode = lo->plh_return_iomode;
1199 pnfs_clear_layoutreturn_info(lo);
1bcf34fd 1200 } else if (iomode != NULL)
e5fd1904 1201 *iomode = IOMODE_ANY;
1bcf34fd 1202 pnfs_barrier_update(lo, be32_to_cpu(stateid->seqid));
13c13a6a
TM		 1203 return true;
1204}
1205
828ed9ec
TM		 1206static void
1207pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1208 struct pnfs_layout_hdr *lo,
1209 const nfs4_stateid *stateid,
1210 enum pnfs_iomode iomode)
1211{
1212 struct inode *inode = lo->plh_inode;
1213
1214 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1215 args->inode = inode;
1216 args->range.iomode = iomode;
1217 args->range.offset = 0;
1218 args->range.length = NFS4_MAX_UINT64;
1219 args->layout = lo;
1220 nfs4_stateid_copy(&args->stateid, stateid);
1221}
1222
f40eb5d0 1223static int
44ea8dfc
TM		 1224pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
1225 const nfs4_stateid *stateid,
1226 const struct cred **pcred,
1227 enum pnfs_iomode iomode,
1228 bool sync)
f40eb5d0
PT		 1229{
1230 struct inode *ino = lo->plh_inode;
287bd3e9 1231 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0 1232 struct nfs4_layoutreturn *lrp;
44ea8dfc 1233 const struct cred *cred = *pcred;
f40eb5d0
PT		 1234 int status = 0;
1235
44ea8dfc 1236 *pcred = NULL;
e4af440a 1237 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1238 if (unlikely(lrp == NULL)) {
1239 status = -ENOMEM;
1240 spin_lock(&ino->i_lock);
d67ae825 1241 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0 1242 spin_unlock(&ino->i_lock);
44ea8dfc 1243 put_cred(cred);
f40eb5d0
PT		 1244 pnfs_put_layout_hdr(lo);
1245 goto out;
1246 }
1247
828ed9ec 1248 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1249 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0 1250 lrp->clp = NFS_SERVER(ino)->nfs_client;
44ea8dfc 1251 lrp->cred = cred;
287bd3e9
TM		 1252 if (ld->prepare_layoutreturn)
1253 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1254
6c16605d 1255 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1256out:
1257 dprintk("<-- %s status: %d\n", __func__, status);
1258 return status;
1259}
1260
d474f961
TM		 1261static bool
1262pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1263 enum pnfs_iomode iomode,
1264 u32 seq)
1265{
1266 struct pnfs_layout_range recall_range = {
1267 .length = NFS4_MAX_UINT64,
1268 .iomode = iomode,
1269 };
1270 return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1271 &recall_range, seq) != -EBUSY;
1272}
1273
13c13a6a
TM		 1274/* Return true if layoutreturn is needed */
1275static bool
1276pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1277{
2370abda 1278 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a 1279 return false;
d474f961
TM		 1280 return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1281 lo->plh_return_seq);
13c13a6a
TM		 1282}
1283
1284static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1285{
1286 struct inode *inode= lo->plh_inode;
1287
2370abda 1288 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1289 return;
1290 spin_lock(&inode->i_lock);
1291 if (pnfs_layout_need_return(lo)) {
44ea8dfc 1292 const struct cred *cred;
13c13a6a
TM		 1293 nfs4_stateid stateid;
1294 enum pnfs_iomode iomode;
1295 bool send;
1296
44ea8dfc 1297 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
13c13a6a
TM		 1298 spin_unlock(&inode->i_lock);
1299 if (send) {
1300 /* Send an async layoutreturn so we dont deadlock */
44ea8dfc 1301 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
13c13a6a
TM		 1302 }
1303 } else
1304 spin_unlock(&inode->i_lock);
1305}
1306
293b3b06
AA		 1307/*
1308 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1309 * when the layout segment list is empty.
1310 *
1311 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1312 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1313 * deviceid is marked invalid.
1314 */
cbe82603
BH		 1315int
1316_pnfs_return_layout(struct inode *ino)
1317{
1318 struct pnfs_layout_hdr *lo = NULL;
1319 struct nfs_inode *nfsi = NFS_I(ino);
a421d218
AS		 1320 struct pnfs_layout_range range = {
1321 .iomode = IOMODE_ANY,
1322 .offset = 0,
1323 .length = NFS4_MAX_UINT64,
1324 };
cbe82603 1325 LIST_HEAD(tmp_list);
44ea8dfc 1326 const struct cred *cred;
cbe82603 1327 nfs4_stateid stateid;
24408f52 1328 int status = 0;
93b7f7ad 1329 bool send, valid_layout;
cbe82603 1330
366d5052 1331 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1332
1333 spin_lock(&ino->i_lock);
1334 lo = nfsi->layout;
e5929f3c 1335 if (!lo) {
cbe82603 1336 spin_unlock(&ino->i_lock);
293b3b06
AA		 1337 dprintk("NFS: %s no layout to return\n", __func__);
1338 goto out;
cbe82603 1339 }
cbe82603 1340 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1341 pnfs_get_layout_hdr(lo);
24408f52
TM		 1342 /* Is there an outstanding layoutreturn ? */
1343 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1344 spin_unlock(&ino->i_lock);
1345 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1346 TASK_UNINTERRUPTIBLE))
1347 goto out_put_layout_hdr;
1348 spin_lock(&ino->i_lock);
1349 }
93b7f7ad 1350 valid_layout = pnfs_layout_is_valid(lo);
24956804 1351 pnfs_clear_layoutcommit(ino, &tmp_list);
a421d218
AS		 1352 pnfs_mark_matching_lsegs_return(lo, &tmp_list, &range, 0);
1353
1354 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range)
c88953d8 1355 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
c88953d8 1356
293b3b06 1357 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1358 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1359 !valid_layout) {
293b3b06 1360 spin_unlock(&ino->i_lock);
293b3b06 1361 dprintk("NFS: %s no layout segments to return\n", __func__);
7bcc1058 1362 goto out_wait_layoutreturn;
293b3b06 1363 }
47abadef 1364
44ea8dfc 1365 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
cbe82603 1366 spin_unlock(&ino->i_lock);
7f27392c 1367 if (send)
44ea8dfc 1368 status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
7bcc1058
TM		 1369out_wait_layoutreturn:
1370 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE);
7f27392c 1371out_put_layout_hdr:
ee6625a9 1372 pnfs_free_lseg_list(&tmp_list);
7f27392c 1373 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1374out:
1375 dprintk("<-- %s status: %d\n", __func__, status);
1376 return status;
1377}
1378
24028672
TM		 1379int
1380pnfs_commit_and_return_layout(struct inode *inode)
1381{
1382 struct pnfs_layout_hdr *lo;
1383 int ret;
1384
1385 spin_lock(&inode->i_lock);
1386 lo = NFS_I(inode)->layout;
1387 if (lo == NULL) {
1388 spin_unlock(&inode->i_lock);
1389 return 0;
1390 }
1391 pnfs_get_layout_hdr(lo);
1392 /* Block new layoutgets and read/write to ds */
1393 lo->plh_block_lgets++;
1394 spin_unlock(&inode->i_lock);
1395 filemap_fdatawait(inode->i_mapping);
1396 ret = pnfs_layoutcommit_inode(inode, true);
1397 if (ret == 0)
1398 ret = _pnfs_return_layout(inode);
1399 spin_lock(&inode->i_lock);
1400 lo->plh_block_lgets--;
1401 spin_unlock(&inode->i_lock);
1402 pnfs_put_layout_hdr(lo);
1403 return ret;
1404}
1405
1c5bd76d
TM		 1406bool pnfs_roc(struct inode *ino,
1407 struct nfs4_layoutreturn_args *args,
1408 struct nfs4_layoutreturn_res *res,
a52458b4 1409 const struct cred *cred)
f7e8917a 1410{
40dd4b7a
TM		 1411 struct nfs_inode *nfsi = NFS_I(ino);
1412 struct nfs_open_context *ctx;
1413 struct nfs4_state *state;
f7e8917a 1414 struct pnfs_layout_hdr *lo;
1c5bd76d 1415 struct pnfs_layout_segment *lseg, *next;
44ea8dfc 1416 const struct cred *lc_cred;
193e3aa2 1417 nfs4_stateid stateid;
1c5bd76d
TM		 1418 enum pnfs_iomode iomode = 0;
1419 bool layoutreturn = false, roc = false;
e71708d4 1420 bool skip_read = false;
f7e8917a 1421
1c5bd76d
TM		 1422 if (!nfs_have_layout(ino))
1423 return false;
29ade5db 1424retry:
0de43976 1425 rcu_read_lock();
f7e8917a 1426 spin_lock(&ino->i_lock);
40dd4b7a 1427 lo = nfsi->layout;
0cdc329e 1428 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1429 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1430 lo = NULL;
40dd4b7a 1431 goto out_noroc;
9c6376eb
TM		 1432 }
1433 pnfs_get_layout_hdr(lo);
29ade5db 1434 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1435 spin_unlock(&ino->i_lock);
0de43976 1436 rcu_read_unlock();
29ade5db
TM		 1437 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1438 TASK_UNINTERRUPTIBLE);
1439 pnfs_put_layout_hdr(lo);
1440 goto retry;
1441 }
40dd4b7a 1442
e755d638 1443 /* no roc if we hold a delegation */
e71708d4
TM		 1444 if (nfs4_check_delegation(ino, FMODE_READ)) {
1445 if (nfs4_check_delegation(ino, FMODE_WRITE))
1446 goto out_noroc;
1447 skip_read = true;
1448 }
40dd4b7a 1449
0de43976 1450 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1451 state = ctx->state;
e71708d4
TM		 1452 if (state == NULL)
1453 continue;
40dd4b7a 1454 /* Don't return layout if there is open file state */
e71708d4 1455 if (state->state & FMODE_WRITE)
40dd4b7a 1456 goto out_noroc;
e71708d4
TM		 1457 if (state->state & FMODE_READ)
1458 skip_read = true;
40dd4b7a
TM		 1459 }
1460
e755d638 1461
1c5bd76d 1462 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1463 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1464 continue;
e755d638 1465 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1466 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1467 continue;
1468 /*
1469 * Note: mark lseg for return so pnfs_layout_remove_lseg
1470 * doesn't invalidate the layout for us.
1471 */
1472 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1473 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1474 continue;
1475 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1476 }
1477
1c5bd76d
TM		 1478 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1479 goto out_noroc;
69820d22 1480
500d701f 1481 /* ROC in two conditions:
e755d638
PT		 1482 * 1. there are ROC lsegs
1483 * 2. we don't send layoutreturn
e755d638 1484 */
1c5bd76d 1485 /* lo ref dropped in pnfs_roc_release() */
44ea8dfc 1486 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
1c5bd76d 1487 /* If the creds don't match, we can't compound the layoutreturn */
4d8948c7 1488 if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0)
1c5bd76d
TM		 1489 goto out_noroc;
1490
1491 roc = layoutreturn;
1492 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1493 res->lrs_present = 0;
1494 layoutreturn = false;
44ea8dfc 1495 put_cred(lc_cred);
4d8948c7 1496
40dd4b7a 1497out_noroc:
f7e8917a 1498 spin_unlock(&ino->i_lock);
0de43976 1499 rcu_read_unlock();
e755d638 1500 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1501 if (roc) {
1502 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1503 if (ld->prepare_layoutreturn)
1504 ld->prepare_layoutreturn(args);
9c6376eb 1505 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1506 return true;
1507 }
e755d638 1508 if (layoutreturn)
44ea8dfc 1509 pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
9c6376eb 1510 pnfs_put_layout_hdr(lo);
287bd3e9 1511 return false;
f7e8917a
FI		 1512}
1513
078000d0
TM		 1514int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
1515 struct nfs4_layoutreturn_res **respp, int *ret)
287a9c55
TM		 1516{
1517 struct nfs4_layoutreturn_args *arg = *argpp;
1518 int retval = -EAGAIN;
1519
1520 if (!arg)
1521 return 0;
1522 /* Handle Layoutreturn errors */
1523 switch (*ret) {
1524 case 0:
1525 retval = 0;
1526 break;
6109bcf7
TM		 1527 case -NFS4ERR_NOMATCHING_LAYOUT:
1528 /* Was there an RPC level error? If not, retry */
1529 if (task->tk_rpc_status == 0)
1530 break;
1531 /* If the call was not sent, let caller handle it */
1532 if (!RPC_WAS_SENT(task))
1533 return 0;
1534 /*
1535 * Otherwise, assume the call succeeded and
1536 * that we need to release the layout
1537 */
1538 *ret = 0;
1539 (*respp)->lrs_present = 0;
1540 retval = 0;
1541 break;
078a432d
TM		 1542 case -NFS4ERR_DELAY:
1543 /* Let the caller handle the retry */
1544 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1545 return 0;
287a9c55 1546 case -NFS4ERR_OLD_STATEID:
30cb3ee2 1547 if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
078000d0 1548 &arg->range, arg->inode))
287a9c55
TM		 1549 break;
1550 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1551 return -EAGAIN;
1552 }
1553 *argpp = NULL;
1554 *respp = NULL;
1555 return retval;
1556}
1557
1c5bd76d
TM		 1558void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1559 struct nfs4_layoutreturn_res *res,
1560 int ret)
f7e8917a 1561{
1c5bd76d 1562 struct pnfs_layout_hdr *lo = args->layout;
67bbceed 1563 struct inode *inode = args->inode;
1c5bd76d 1564 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1565 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1566
9c47b18c
TM		 1567 switch (ret) {
1568 case -NFS4ERR_NOMATCHING_LAYOUT:
67bbceed
TM		 1569 spin_lock(&inode->i_lock);
1570 if (pnfs_layout_is_valid(lo) &&
1571 nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
1572 pnfs_set_plh_return_info(lo, args->range.iomode, 0);
c18d1e17 1573 pnfs_clear_layoutreturn_waitbit(lo);
67bbceed 1574 spin_unlock(&inode->i_lock);
9c47b18c
TM		 1575 break;
1576 case 0:
1c5bd76d
TM		 1577 if (res->lrs_present)
1578 res_stateid = &res->stateid;
df561f66 1579 fallthrough;
9c47b18c 1580 default:
c18d1e17
TM		 1581 pnfs_layoutreturn_free_lsegs(lo, &args->stateid, &args->range,
1582 res_stateid);
1c5bd76d 1583 }
a19b4785 1584 trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
287bd3e9
TM		 1585 if (ld_private && ld_private->ops && ld_private->ops->free)
1586 ld_private->ops->free(ld_private);
1c5bd76d 1587 pnfs_put_layout_hdr(lo);
f7e8917a
FI		 1588}
1589
500d701f
PT		 1590bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1591{
1592 struct nfs_inode *nfsi = NFS_I(ino);
1593 struct pnfs_layout_hdr *lo;
1594 bool sleep = false;
1595
1596 /* we might not have grabbed lo reference. so need to check under
1597 * i_lock */
1598 spin_lock(&ino->i_lock);
1599 lo = nfsi->layout;
ee284e35
TM		 1600 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1601 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1602 sleep = true;
ee284e35 1603 }
500d701f 1604 spin_unlock(&ino->i_lock);
500d701f
PT		 1605 return sleep;
1606}
1607
b1f69b75
AA		 1608/*
1609 * Compare two layout segments for sorting into layout cache.
1610 * We want to preferentially return RW over RO layouts, so ensure those
1611 * are seen first.
1612 */
1613static s64
7dc0ac70 1614pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1615 const struct pnfs_layout_range *l2)
b1f69b75 1616{
fb3296eb
BH		 1617 s64 d;
1618
1619 /* high offset > low offset */
1620 d = l1->offset - l2->offset;
1621 if (d)
1622 return d;
1623
1624 /* short length > long length */
1625 d = l2->length - l1->length;
1626 if (d)
1627 return d;
1628
b1f69b75 1629 /* read > read/write */
fb3296eb 1630 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1631}
1632
03772d2f
TM		 1633static bool
1634pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1635 const struct pnfs_layout_range *l2)
1636{
1637 return pnfs_lseg_range_cmp(l1, l2) > 0;
1638}
1639
1640static bool
1641pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1642 struct pnfs_layout_segment *old)
1643{
1644 return false;
1645}
1646
1647void
1648pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1649 struct pnfs_layout_segment *lseg,
1650 bool (*is_after)(const struct pnfs_layout_range *,
1651 const struct pnfs_layout_range *),
1652 bool (*do_merge)(struct pnfs_layout_segment *,
1653 struct pnfs_layout_segment *),
1654 struct list_head *free_me)
974cec8c 1655{
03772d2f 1656 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1657
974cec8c
AA		 1658 dprintk("%s:Begin\n", __func__);
1659
03772d2f
TM		 1660 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1661 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1662 continue;
1663 if (do_merge(lseg, lp)) {
1664 mark_lseg_invalid(lp, free_me);
1665 continue;
1666 }
1667 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1668 continue;
566052c5 1669 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1670 dprintk("%s: inserted lseg %p "
1671 "iomode %d offset %llu length %llu before "
1672 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1673 __func__, lseg, lseg->pls_range.iomode,
1674 lseg->pls_range.offset, lseg->pls_range.length,
1675 lp, lp->pls_range.iomode, lp->pls_range.offset,
1676 lp->pls_range.length);
fb3296eb 1677 goto out;
974cec8c 1678 }
fb3296eb
BH		 1679 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1680 dprintk("%s: inserted lseg %p "
1681 "iomode %d offset %llu length %llu at tail\n",
1682 __func__, lseg, lseg->pls_range.iomode,
1683 lseg->pls_range.offset, lseg->pls_range.length);
1684out:
70c3bd2b 1685 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1686
1687 dprintk("%s:Return\n", __func__);
e5e94017 1688}
03772d2f
TM		 1689EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1690
1691static void
1692pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1693 struct pnfs_layout_segment *lseg,
1694 struct list_head *free_me)
1695{
1696 struct inode *inode = lo->plh_inode;
1697 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1698
1699 if (ld->add_lseg != NULL)
1700 ld->add_lseg(lo, lseg, free_me);
1701 else
1702 pnfs_generic_layout_insert_lseg(lo, lseg,
1703 pnfs_lseg_range_is_after,
1704 pnfs_lseg_no_merge,
1705 free_me);
1706}
e5e94017
BH		 1707
1708static struct pnfs_layout_hdr *
9fa40758
PT		 1709alloc_init_layout_hdr(struct inode *ino,
1710 struct nfs_open_context *ctx,
1711 gfp_t gfp_flags)
e5e94017
BH		 1712{
1713 struct pnfs_layout_hdr *lo;
1714
636fb9c8 1715 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1716 if (!lo)
1717 return NULL;
2b28a7be 1718 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1719 INIT_LIST_HEAD(&lo->plh_layouts);
1720 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1721 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1722 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1723 lo->plh_inode = ino;
a52458b4 1724 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1725 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1726 return lo;
1727}
1728
1729static struct pnfs_layout_hdr *
9fa40758
PT		 1730pnfs_find_alloc_layout(struct inode *ino,
1731 struct nfs_open_context *ctx,
1732 gfp_t gfp_flags)
e5241e43
TM		 1733 __releases(&ino->i_lock)
1734 __acquires(&ino->i_lock)
e5e94017
BH		 1735{
1736 struct nfs_inode *nfsi = NFS_I(ino);
1737 struct pnfs_layout_hdr *new = NULL;
1738
1739 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1740
251ec410
TM		 1741 if (nfsi->layout != NULL)
1742 goto out_existing;
e5e94017 1743 spin_unlock(&ino->i_lock);
9fa40758 1744 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1745 spin_lock(&ino->i_lock);
1746
251ec410 1747 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1748 nfsi->layout = new;
251ec410 1749 return new;
7175fe90
YN		 1750 } else if (new != NULL)
1751 pnfs_free_layout_hdr(new);
251ec410
TM		 1752out_existing:
1753 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1754 return nfsi->layout;
1755}
1756
b1f69b75
AA		 1757/*
1758 * iomode matching rules:
c7d73af2
TH		 1759 * iomode lseg strict match
1760 * iomode
1761 * ----- ----- ------ -----
1762 * ANY READ N/A true
1763 * ANY RW N/A true
1764 * RW READ N/A false
1765 * RW RW N/A true
1766 * READ READ N/A true
1767 * READ RW true false
1768 * READ RW false true
b1f69b75 1769 */
3cb2df17 1770static bool
7dc0ac70 1771pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1772 const struct pnfs_layout_range *range,
1773 bool strict_iomode)
b1f69b75 1774{
fb3296eb
BH		 1775 struct pnfs_layout_range range1;
1776
1777 if ((range->iomode == IOMODE_RW &&
1778 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1779 (range->iomode != ls_range->iomode &&
6089dd0d 1780 strict_iomode) ||
7dc0ac70 1781 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1782 return false;
fb3296eb
BH		 1783
1784 /* range1 covers only the first byte in the range */
1785 range1 = *range;
1786 range1.length = 1;
7dc0ac70 1787 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1788}
1789
1790/*
1791 * lookup range in layout
1792 */
e5e94017 1793static struct pnfs_layout_segment *
fb3296eb 1794pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1795 struct pnfs_layout_range *range,
1796 bool strict_iomode)
e5e94017 1797{
b1f69b75
AA		 1798 struct pnfs_layout_segment *lseg, *ret = NULL;
1799
1800 dprintk("%s:Begin\n", __func__);
1801
b7edfaa1 1802 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1803 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1804 pnfs_lseg_range_match(&lseg->pls_range, range,
1805 strict_iomode)) {
9369a431 1806 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1807 break;
1808 }
b1f69b75
AA		 1809 }
1810
1811 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1812 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1813 return ret;
e5e94017
BH		 1814}
1815
d23d61c8
AA		 1816/*
1817 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1818 * to the MDS or over pNFS
1819 *
1820 * The nfs_inode read_io and write_io fields are cumulative counters reset
1821 * when there are no layout segments. Note that in pnfs_update_layout iomode
1822 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1823 * WRITE request.
1824 *
1825 * A return of true means use MDS I/O.
1826 *
1827 * From rfc 5661:
1828 * If a file's size is smaller than the file size threshold, data accesses
1829 * SHOULD be sent to the metadata server. If an I/O request has a length that
1830 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1831 * server. If both file size and I/O size are provided, the client SHOULD
1832 * reach or exceed both thresholds before sending its read or write
1833 * requests to the data server.
1834 */
1835static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1836 struct inode *ino, int iomode)
1837{
1838 struct nfs4_threshold *t = ctx->mdsthreshold;
1839 struct nfs_inode *nfsi = NFS_I(ino);
1840 loff_t fsize = i_size_read(ino);
1841 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1842
1843 if (t == NULL)
1844 return ret;
1845
1846 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1847 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1848
1849 switch (iomode) {
1850 case IOMODE_READ:
1851 if (t->bm & THRESHOLD_RD) {
1852 dprintk("%s fsize %llu\n", __func__, fsize);
1853 size_set = true;
1854 if (fsize < t->rd_sz)
1855 size = true;
1856 }
1857 if (t->bm & THRESHOLD_RD_IO) {
1858 dprintk("%s nfsi->read_io %llu\n", __func__,
1859 nfsi->read_io);
1860 io_set = true;
1861 if (nfsi->read_io < t->rd_io_sz)
1862 io = true;
1863 }
1864 break;
1865 case IOMODE_RW:
1866 if (t->bm & THRESHOLD_WR) {
1867 dprintk("%s fsize %llu\n", __func__, fsize);
1868 size_set = true;
1869 if (fsize < t->wr_sz)
1870 size = true;
1871 }
1872 if (t->bm & THRESHOLD_WR_IO) {
1873 dprintk("%s nfsi->write_io %llu\n", __func__,
1874 nfsi->write_io);
1875 io_set = true;
1876 if (nfsi->write_io < t->wr_io_sz)
1877 io = true;
1878 }
1879 break;
1880 }
1881 if (size_set && io_set) {
1882 if (size && io)
1883 ret = true;
1884 } else if (size || io)
1885 ret = true;
1886
1887 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1888 return ret;
1889}
1890
d03360aa 1891static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1892{
1893 /*
1894 * send layoutcommit as it can hold up layoutreturn due to lseg
1895 * reference
1896 */
1897 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1898 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1899 nfs_wait_bit_killable,
d03360aa 1900 TASK_KILLABLE);
aa8a45ee
PT		 1901}
1902
411ae722
TM		 1903static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1904{
1905 atomic_inc(&lo->plh_outstanding);
1906}
1907
1908static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1909{
1910 if (atomic_dec_and_test(&lo->plh_outstanding))
1911 wake_up_var(&lo->plh_outstanding);
1912}
1913
d29b468d
TM		 1914static bool pnfs_is_first_layoutget(struct pnfs_layout_hdr *lo)
1915{
1916 return test_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags);
1917}
1918
d67ae825
TH		 1919static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1920{
1921 unsigned long *bitlock = &lo->plh_flags;
1922
1923 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1924 smp_mb__after_atomic();
1925 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1926}
1927
78746a38
FI		 1928static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1929 struct nfs_server *server)
1930{
cf6605d1 1931 if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
78746a38
FI		 1932 struct nfs_client *clp = server->nfs_client;
1933
1934 /* The lo must be on the clp list if there is any
1935 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1936 */
1937 spin_lock(&clp->cl_lock);
cf6605d1 1938 list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
78746a38
FI		 1939 spin_unlock(&clp->cl_lock);
1940 }
1941}
1942
e5e94017
BH		 1943/*
1944 * Layout segment is retreived from the server if not cached.
1945 * The appropriate layout segment is referenced and returned to the caller.
1946 */
7c24d948 1947struct pnfs_layout_segment *
e5e94017
BH		 1948pnfs_update_layout(struct inode *ino,
1949 struct nfs_open_context *ctx,
fb3296eb
BH		 1950 loff_t pos,
1951 u64 count,
a75b9df9 1952 enum pnfs_iomode iomode,
c7d73af2 1953 bool strict_iomode,
a75b9df9 1954 gfp_t gfp_flags)
e5e94017 1955{
fb3296eb
BH		 1956 struct pnfs_layout_range arg = {
1957 .iomode = iomode,
1958 .offset = pos,
1959 .length = count,
1960 };
70d2f7b1 1961 unsigned pg_offset;
6382a441
WAA		 1962 struct nfs_server *server = NFS_SERVER(ino);
1963 struct nfs_client *clp = server->nfs_client;
183d9e7b 1964 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1965 struct pnfs_layout_segment *lseg = NULL;
587f03de 1966 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1967 nfs4_stateid stateid;
1968 long timeout = 0;
66b53f32 1969 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1970 bool first;
e5e94017 1971
9a4bf31d 1972 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1973 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1974 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1975 goto out;
9a4bf31d 1976 }
d23d61c8 1977
9a4bf31d 1978 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1979 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1980 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1981 goto out;
9a4bf31d 1982 }
d23d61c8 1983
9bf87482 1984lookup_again:
d03360aa
TM		 1985 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1986 if (IS_ERR(lseg))
1987 goto out;
9bf87482 1988 first = false;
e5e94017 1989 spin_lock(&ino->i_lock);
9fa40758 1990 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1991 if (lo == NULL) {
1992 spin_unlock(&ino->i_lock);
183d9e7b 1993 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1994 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1995 goto out;
1996 }
e5e94017 1997
43f1b3da 1998 /* Do we even need to bother with this? */
a59c30ac 1999 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 2000 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2001 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 2002 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 2003 goto out_unlock;
2004 }
2005
2006 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 2007 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 2008 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2009 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 2010 goto out_unlock;
9a4bf31d 2011 }
e5e94017 2012
411ae722
TM		 2013 /*
2014 * If the layout segment list is empty, but there are outstanding
2015 * layoutget calls, then they might be subject to a layoutrecall.
2016 */
2017 if (list_empty(&lo->plh_segs) &&
2018 atomic_read(&lo->plh_outstanding) != 0) {
2019 spin_unlock(&ino->i_lock);
d03360aa 2020 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
400417b0 2021 !atomic_read(&lo->plh_outstanding)));
58bbeab4 2022 if (IS_ERR(lseg))
411ae722
TM		 2023 goto out_put_layout_hdr;
2024 pnfs_put_layout_hdr(lo);
2025 goto lookup_again;
2026 }
2027
2c8d5fc3
TM		 2028 /*
2029 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
2030 * for LAYOUTRETURN.
2031 */
2032 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
2033 spin_unlock(&ino->i_lock);
2034 dprintk("%s wait for layoutreturn\n", __func__);
2035 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2036 if (!IS_ERR(lseg)) {
2037 pnfs_put_layout_hdr(lo);
2038 dprintk("%s retrying\n", __func__);
2039 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2040 lseg,
2041 PNFS_UPDATE_LAYOUT_RETRY);
2042 goto lookup_again;
2043 }
2044 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2045 PNFS_UPDATE_LAYOUT_RETURN);
2046 goto out_put_layout_hdr;
2047 }
2048
c7d73af2 2049 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 2050 if (lseg) {
2051 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2052 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
2053 goto out_unlock;
2054 }
2055
183d9e7b
JL		 2056 /*
2057 * Choose a stateid for the LAYOUTGET. If we don't have a layout
2058 * stateid, or it has been invalidated, then we must use the open
2059 * stateid.
2060 */
67a3b721 2061 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
d9aba2b4 2062 int status;
183d9e7b
JL		 2063
2064 /*
2065 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 2066 * RFC 5661 Errata 3208.
2067 */
2068 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
2069 &lo->plh_flags)) {
2070 spin_unlock(&ino->i_lock);
d03360aa
TM		 2071 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
2072 NFS_LAYOUT_FIRST_LAYOUTGET,
2073 TASK_KILLABLE));
2074 if (IS_ERR(lseg))
2075 goto out_put_layout_hdr;
9bf87482 2076 pnfs_put_layout_hdr(lo);
183d9e7b 2077 dprintk("%s retrying\n", __func__);
9bf87482
PT		 2078 goto lookup_again;
2079 }
183d9e7b 2080
fbf4bcc9 2081 spin_unlock(&ino->i_lock);
183d9e7b 2082 first = true;
d9aba2b4 2083 status = nfs4_select_rw_stateid(ctx->state,
70d2f7b1 2084 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
d9aba2b4
TM		 2085 NULL, &stateid, NULL);
2086 if (status != 0) {
731c74dd 2087 lseg = ERR_PTR(status);
70d2f7b1
TM		 2088 trace_pnfs_update_layout(ino, pos, count,
2089 iomode, lo, lseg,
2090 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
d9aba2b4
TM		 2091 nfs4_schedule_stateid_recovery(server, ctx->state);
2092 pnfs_clear_first_layoutget(lo);
2093 pnfs_put_layout_hdr(lo);
2094 goto lookup_again;
70d2f7b1 2095 }
fbf4bcc9 2096 spin_lock(&ino->i_lock);
9bf87482 2097 } else {
183d9e7b 2098 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 2099 }
568e8c49 2100
9a4bf31d 2101 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 2102 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2103 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 2104 goto out_unlock;
9a4bf31d 2105 }
411ae722 2106 nfs_layoutget_begin(lo);
f49f9baa 2107 spin_unlock(&ino->i_lock);
30005121 2108
78746a38 2109 _add_to_server_list(lo, server);
e5e94017 2110
09cbfeaf 2111 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 2112 if (pg_offset) {
2113 arg.offset -= pg_offset;
2114 arg.length += pg_offset;
2115 }
7c24d948 2116 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 2117 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 2118
5e36e2a9 2119 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 2120 if (!lgp) {
2121 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2122 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 2123 nfs_layoutget_end(lo);
587f03de
FI		 2124 goto out_put_layout_hdr;
2125 }
2126
dacb452d 2127 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 2128 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2129 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 2130 nfs_layoutget_end(lo);
83026d80 2131 if (IS_ERR(lseg)) {
183d9e7b 2132 switch(PTR_ERR(lseg)) {
e85d7ee4 2133 case -EBUSY:
183d9e7b
JL		 2134 if (time_after(jiffies, giveup))
2135 lseg = NULL;
66b53f32
TM		 2136 break;
2137 case -ERECALLCONFLICT:
183d9e7b 2138 case -EAGAIN:
56b38a1f 2139 break;
183d9e7b
JL		 2140 default:
2141 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2142 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2143 lseg = NULL;
2144 }
56b38a1f
TM		 2145 goto out_put_layout_hdr;
2146 }
2147 if (lseg) {
2148 if (first)
2149 pnfs_clear_first_layoutget(lo);
2150 trace_pnfs_update_layout(ino, pos, count,
2151 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2152 pnfs_put_layout_hdr(lo);
2153 goto lookup_again;
83026d80
JL		 2154 }
2155 } else {
2156 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2157 }
2158
830ffb56 2159out_put_layout_hdr:
d67ae825
TH		 2160 if (first)
2161 pnfs_clear_first_layoutget(lo);
d5b9216f
TM		 2162 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2163 PNFS_UPDATE_LAYOUT_EXIT);
70c3bd2b 2164 pnfs_put_layout_hdr(lo);
e5e94017 2165out:
f86bbcf8
TM		 2166 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2167 "(%s, offset: %llu, length: %llu)\n",
2168 __func__, ino->i_sb->s_id,
2169 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2170 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2171 iomode==IOMODE_RW ? "read/write" : "read-only",
2172 (unsigned long long)pos,
2173 (unsigned long long)count);
e5e94017
BH		 2174 return lseg;
2175out_unlock:
2176 spin_unlock(&ino->i_lock);
830ffb56 2177 goto out_put_layout_hdr;
e5e94017 2178}
7c24d948 2179EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2180
540d9864
TM		 2181static bool
2182pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2183{
2184 switch (range->iomode) {
2185 case IOMODE_READ:
2186 case IOMODE_RW:
2187 break;
2188 default:
2189 return false;
2190 }
2191 if (range->offset == NFS4_MAX_UINT64)
2192 return false;
2193 if (range->length == 0)
2194 return false;
2195 if (range->length != NFS4_MAX_UINT64 &&
2196 range->length > NFS4_MAX_UINT64 - range->offset)
2197 return false;
2198 return true;
2199}
2200
78746a38
FI		 2201static struct pnfs_layout_hdr *
2202_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2203{
2204 struct pnfs_layout_hdr *lo;
2205
2206 spin_lock(&ino->i_lock);
2207 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2208 if (!lo)
2209 goto out_unlock;
2210 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2211 goto out_unlock;
2212 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2213 goto out_unlock;
2214 if (pnfs_layoutgets_blocked(lo))
2215 goto out_unlock;
2216 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2217 goto out_unlock;
411ae722 2218 nfs_layoutget_begin(lo);
78746a38
FI		 2219 spin_unlock(&ino->i_lock);
2220 _add_to_server_list(lo, NFS_SERVER(ino));
2221 return lo;
2222
2223out_unlock:
2224 spin_unlock(&ino->i_lock);
2225 pnfs_put_layout_hdr(lo);
2226 return NULL;
2227}
2228
2409a976
FI		 2229static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2230 struct nfs_open_context *ctx)
2231{
78746a38
FI		 2232 struct inode *ino = data->dentry->d_inode;
2233 struct pnfs_layout_range rng = {
2234 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2235 IOMODE_RW: IOMODE_READ,
2236 .offset = 0,
2237 .length = NFS4_MAX_UINT64,
2238 };
2239 struct nfs4_layoutget *lgp;
2240 struct pnfs_layout_hdr *lo;
2241
64294b08
TM		 2242 /* Heuristic: don't send layoutget if we have cached data */
2243 if (rng.iomode == IOMODE_READ &&
2244 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2245 return;
2246
78746a38
FI		 2247 lo = _pnfs_grab_empty_layout(ino, ctx);
2248 if (!lo)
2249 return;
2250 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2251 &rng, GFP_KERNEL);
2252 if (!lgp) {
2253 pnfs_clear_first_layoutget(lo);
cb2856c5 2254 nfs_layoutget_end(lo);
78746a38
FI		 2255 pnfs_put_layout_hdr(lo);
2256 return;
2257 }
2258 data->lgp = lgp;
2259 data->o_arg.lg_args = &lgp->args;
2260 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2261}
2262
2263static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2264 struct nfs_open_context *ctx)
2265{
2266 struct pnfs_layout_range rng = {
2267 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2268 IOMODE_RW: IOMODE_READ,
2269 .offset = 0,
2270 .length = NFS4_MAX_UINT64,
2271 };
2272 struct nfs4_layoutget *lgp;
2273
2274 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2275 &rng, GFP_KERNEL);
2276 if (!lgp)
2277 return;
2278 data->lgp = lgp;
2279 data->o_arg.lg_args = &lgp->args;
2280 data->o_res.lg_res = &lgp->res;
2281}
2282
2283void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2284 struct nfs_open_context *ctx)
2285{
2286 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2287
2288 if (!(pnfs_enabled_sb(server) &&
2289 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2290 return;
2291 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2292 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2293 return;
2409a976
FI		 2294 if (data->state)
2295 _lgopen_prepare_attached(data, ctx);
2296 else
2297 _lgopen_prepare_floating(data, ctx);
2298}
2299
2300void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2301 struct nfs_open_context *ctx)
2302{
2303 struct pnfs_layout_hdr *lo;
2304 struct pnfs_layout_segment *lseg;
c49b5209 2305 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2306 u32 iomode;
2307
6e01260c 2308 if (!lgp)
2409a976 2309 return;
6e01260c
FI		 2310 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2311 if (lgp->res.status) {
2312 switch (lgp->res.status) {
6e01260c 2313 default:
8dc96566
TM		 2314 break;
2315 /*
2316 * Halt lgopen attempts if the server doesn't recognise
2317 * the "current stateid" value, the layout type, or the
2318 * layoutget operation as being valid.
2319 * Also if it complains about too many ops in the compound
2320 * or of the request/reply being too big.
2321 */
2322 case -NFS4ERR_BAD_STATEID:
2323 case -NFS4ERR_NOTSUPP:
2324 case -NFS4ERR_REP_TOO_BIG:
2325 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2326 case -NFS4ERR_REQ_TOO_BIG:
2327 case -NFS4ERR_TOO_MANY_OPS:
2328 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2329 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2330 }
2331 return;
2332 }
2409a976 2333 if (!lgp->args.inode) {
78746a38
FI		 2334 lo = _pnfs_grab_empty_layout(ino, ctx);
2335 if (!lo)
2336 return;
2409a976
FI		 2337 lgp->args.inode = ino;
2338 } else
2339 lo = NFS_I(lgp->args.inode)->layout;
2409a976
FI		 2340
2341 lseg = pnfs_layout_process(lgp);
32f1c28f 2342 if (!IS_ERR(lseg)) {
2409a976
FI		 2343 iomode = lgp->args.range.iomode;
2344 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2345 pnfs_put_lseg(lseg);
2346 }
30ae2412
FI		 2347}
2348
2349void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2350{
2351 if (lgp != NULL) {
2352 struct inode *inode = lgp->args.inode;
2353 if (inode) {
2354 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
30ae2412 2355 pnfs_clear_first_layoutget(lo);
411ae722 2356 nfs_layoutget_end(lo);
30ae2412
FI		 2357 }
2358 pnfs_layoutget_free(lgp);
2359 }
2409a976
FI		 2360}
2361
a0b0a6e3 2362struct pnfs_layout_segment *
b1f69b75
AA		 2363pnfs_layout_process(struct nfs4_layoutget *lgp)
2364{
2365 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2366 struct nfs4_layoutget_res *res = &lgp->res;
2367 struct pnfs_layout_segment *lseg;
b7edfaa1 2368 struct inode *ino = lo->plh_inode;
78096cca 2369 LIST_HEAD(free_me);
540d9864
TM		 2370
2371 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2372 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2373
2374 /* Inject layout blob into I/O device driver */
a75b9df9 2375 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2376 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2377 if (!lseg)
1b3c6d07
JL		 2378 lseg = ERR_PTR(-ENOMEM);
2379
2380 dprintk("%s: Could not allocate layout: error %ld\n",
2381 __func__, PTR_ERR(lseg));
2382 return lseg;
b1f69b75
AA		 2383 }
2384
119cef97 2385 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2386
b1f69b75 2387 spin_lock(&ino->i_lock);
e1c06f80 2388 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2389 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2390 goto out_forget;
43f1b3da 2391 }
038d6493 2392
d29b468d 2393 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2394 /* existing state ID, make sure the sequence number matches. */
2395 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
d29b468d
TM		 2396 if (!pnfs_layout_is_valid(lo) &&
2397 pnfs_is_first_layoutget(lo))
2398 lo->plh_barrier = 0;
362f7474 2399 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2400 goto out_forget;
362f7474 2401 }
59b56394 2402 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
d29b468d 2403 } else if (pnfs_layout_is_valid(lo)) {
362f7474
CH		 2404 /*
2405 * We got an entirely new state ID. Mark all segments for the
9888d837 2406 * inode invalid, and retry the layoutget
362f7474 2407 */
08bd8dbe
TM		 2408 struct pnfs_layout_range range = {
2409 .iomode = IOMODE_ANY,
2410 .length = NFS4_MAX_UINT64,
2411 };
fb700ef0 2412 pnfs_mark_matching_lsegs_return(lo, &free_me, &range, 0);
9888d837 2413 goto out_forget;
d29b468d
TM		 2414 } else {
2415 /* We have a completely new layout */
2416 if (!pnfs_is_first_layoutget(lo))
2417 goto out_forget;
2418 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
362f7474 2419 }
038d6493 2420
9369a431 2421 pnfs_get_lseg(lseg);
03772d2f 2422 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2423
b1f69b75 2424
3976143b 2425 if (res->return_on_close)
f7e8917a 2426 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2427
b1f69b75 2428 spin_unlock(&ino->i_lock);
78096cca 2429 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2430 return lseg;
43f1b3da 2431
1b3c6d07 2432out_forget:
43f1b3da
FI		 2433 spin_unlock(&ino->i_lock);
2434 lseg->pls_layout = lo;
2435 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2436 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2437}
2438
2f215968
TM		 2439/**
2440 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2441 * @lo: pointer to layout header
2442 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2443 * @return_range: describe layout segment ranges to be returned
e0b7d420 2444 * @seq: stateid seqid to match
2f215968
TM		 2445 *
2446 * This function is mainly intended for use by layoutrecall. It attempts
2447 * to free the layout segment immediately, or else to mark it for return
2448 * as soon as its reference count drops to zero.
e0b7d420
TM		 2449 *
2450 * Returns
2451 * - 0: a layoutreturn needs to be scheduled.
2452 * - EBUSY: there are layout segment that are still in use.
2453 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2454 */
10335556 2455int
016256df
PT		 2456pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2457 struct list_head *tmp_list,
6d597e17
JL		 2458 const struct pnfs_layout_range *return_range,
2459 u32 seq)
016256df
PT		 2460{
2461 struct pnfs_layout_segment *lseg, *next;
10335556 2462 int remaining = 0;
016256df
PT		 2463
2464 dprintk("%s:Begin lo %p\n", __func__, lo);
2465
fc7ff367 2466 assert_spin_locked(&lo->plh_inode->i_lock);
016256df 2467
39fd0186
TM		 2468 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2469 tmp_list = &lo->plh_return_segs;
2470
016256df 2471 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2472 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2473 dprintk("%s: marking lseg %p iomode %d "
2474 "offset %llu length %llu\n", __func__,
2475 lseg, lseg->pls_range.iomode,
2476 lseg->pls_range.offset,
2477 lseg->pls_range.length);
39fd0186
TM		 2478 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2479 tmp_list = &lo->plh_return_segs;
ff041727 2480 if (mark_lseg_invalid(lseg, tmp_list))
2f215968
TM		 2481 continue;
2482 remaining++;
016256df 2483 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2484 }
6d597e17 2485
e0b7d420 2486 if (remaining) {
6d597e17 2487 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2488 return -EBUSY;
2489 }
6d597e17 2490
e0b7d420
TM		 2491 if (!list_empty(&lo->plh_return_segs)) {
2492 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2493 return 0;
2494 }
2495
2496 return -ENOENT;
016256df
PT		 2497}
2498
b5fdf841
TM		 2499static void
2500pnfs_mark_layout_for_return(struct inode *inode,
2501 const struct pnfs_layout_range *range)
016256df 2502{
b5fdf841 2503 struct pnfs_layout_hdr *lo;
10335556 2504 bool return_now = false;
016256df
PT		 2505
2506 spin_lock(&inode->i_lock);
b5fdf841 2507 lo = NFS_I(inode)->layout;
bdebfccd
TM		 2508 if (!pnfs_layout_is_valid(lo)) {
2509 spin_unlock(&inode->i_lock);
2510 return;
2511 }
b5fdf841 2512 pnfs_set_plh_return_info(lo, range->iomode, 0);
016256df
PT		 2513 /*
2514 * mark all matching lsegs so that we are sure to have no live
2515 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2516 * for how it works.
2517 */
b5fdf841 2518 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
44ea8dfc 2519 const struct cred *cred;
10335556 2520 nfs4_stateid stateid;
e5fd1904 2521 enum pnfs_iomode iomode;
10335556 2522
44ea8dfc 2523 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
10335556
TM		 2524 spin_unlock(&inode->i_lock);
2525 if (return_now)
44ea8dfc 2526 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
10335556
TM		 2527 } else {
2528 spin_unlock(&inode->i_lock);
2529 nfs_commit_inode(inode, 0);
2530 }
016256df 2531}
b5fdf841
TM		 2532
2533void pnfs_error_mark_layout_for_return(struct inode *inode,
2534 struct pnfs_layout_segment *lseg)
2535{
2536 struct pnfs_layout_range range = {
2537 .iomode = lseg->pls_range.iomode,
2538 .offset = 0,
2539 .length = NFS4_MAX_UINT64,
2540 };
2541
2542 pnfs_mark_layout_for_return(inode, &range);
2543}
016256df
PT		 2544EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2545
b5fdf841
TM		 2546static bool
2547pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
2548{
2549 return pnfs_layout_is_valid(lo) &&
2550 !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
2551 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
2552}
2553
2554static struct pnfs_layout_segment *
2555pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
2556 const struct pnfs_layout_range *range,
2557 enum pnfs_iomode iomode)
2558{
2559 struct pnfs_layout_segment *lseg;
2560
2561 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
2562 if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
2563 continue;
2564 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2565 continue;
2566 if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
2567 continue;
2568 if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
2569 return lseg;
2570 }
2571 return NULL;
2572}
2573
2574/* Find open file states whose mode matches that of the range */
2575static bool
2576pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
2577 const struct pnfs_layout_range *range)
2578{
2579 struct list_head *head;
2580 struct nfs_open_context *ctx;
2581 fmode_t mode = 0;
2582
2583 if (!pnfs_layout_can_be_returned(lo) ||
2584 !pnfs_find_first_lseg(lo, range, range->iomode))
2585 return false;
2586
2587 head = &NFS_I(lo->plh_inode)->open_files;
2588 list_for_each_entry_rcu(ctx, head, list) {
2589 if (ctx->state)
2590 mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
2591 }
2592
2593 switch (range->iomode) {
2594 default:
2595 break;
2596 case IOMODE_READ:
2597 mode &= ~FMODE_WRITE;
2598 break;
2599 case IOMODE_RW:
2600 if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
2601 mode &= ~FMODE_READ;
2602 }
2603 return mode == 0;
2604}
2605
2606static int
2607pnfs_layout_return_unused_byserver(struct nfs_server *server, void *data)
2608{
2609 const struct pnfs_layout_range *range = data;
2610 struct pnfs_layout_hdr *lo;
2611 struct inode *inode;
2612restart:
2613 rcu_read_lock();
2614 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
2615 if (!pnfs_layout_can_be_returned(lo) ||
2616 test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2617 continue;
2618 inode = lo->plh_inode;
2619 spin_lock(&inode->i_lock);
2620 if (!pnfs_should_return_unused_layout(lo, range)) {
2621 spin_unlock(&inode->i_lock);
2622 continue;
2623 }
2624 spin_unlock(&inode->i_lock);
2625 inode = pnfs_grab_inode_layout_hdr(lo);
2626 if (!inode)
2627 continue;
2628 rcu_read_unlock();
2629 pnfs_mark_layout_for_return(inode, range);
2630 iput(inode);
2631 cond_resched();
2632 goto restart;
2633 }
2634 rcu_read_unlock();
2635 return 0;
2636}
2637
2638void
2639pnfs_layout_return_unused_byclid(struct nfs_client *clp,
2640 enum pnfs_iomode iomode)
2641{
2642 struct pnfs_layout_range range = {
2643 .iomode = iomode,
2644 .offset = 0,
2645 .length = NFS4_MAX_UINT64,
2646 };
2647
2648 nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
2649 &range);
2650}
2651
b3230e80
TM		 2652void
2653pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2654{
2655 if (pgio->pg_lseg == NULL ||
2656 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2657 return;
2658 pnfs_put_lseg(pgio->pg_lseg);
2659 pgio->pg_lseg = NULL;
2660}
2661EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2662
08cb5b0f
BC		 2663/*
2664 * Check for any intersection between the request and the pgio->pg_lseg,
2665 * and if none, put this pgio->pg_lseg away.
2666 */
e1e54ab7 2667void
08cb5b0f
BC		 2668pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2669{
2670 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2671 pnfs_put_lseg(pgio->pg_lseg);
2672 pgio->pg_lseg = NULL;
2673 }
2674}
e1e54ab7 2675EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
08cb5b0f 2676
d8007d4d
TM		 2677void
2678pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2679{
d1d97395 2680 u64 rd_size;
1fd937bd 2681
b3230e80 2682 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2683 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2684 if (pgio->pg_lseg == NULL) {
2685 if (pgio->pg_dreq == NULL)
2686 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2687 else
2688 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2689
2690 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2691 nfs_req_openctx(req),
cb5d04bc
PT		 2692 req_offset(req),
2693 rd_size,
2694 IOMODE_READ,
c7d73af2 2695 false,
cb5d04bc 2696 GFP_KERNEL);
d600ad1f
PT		 2697 if (IS_ERR(pgio->pg_lseg)) {
2698 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2699 pgio->pg_lseg = NULL;
2700 return;
2701 }
cb5d04bc 2702 }
e885de1a
TM		 2703 /* If no lseg, fall back to read through mds */
2704 if (pgio->pg_lseg == NULL)
1f945357 2705 nfs_pageio_reset_read_mds(pgio);
e885de1a 2706
d8007d4d
TM		 2707}
2708EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2709
2710void
6296556f
PT		 2711pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2712 struct nfs_page *req, u64 wb_size)
d8007d4d 2713{
b3230e80 2714 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2715 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2716 if (pgio->pg_lseg == NULL) {
cb5d04bc 2717 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2718 nfs_req_openctx(req),
cb5d04bc
PT		 2719 req_offset(req),
2720 wb_size,
2721 IOMODE_RW,
c7d73af2 2722 false,
2b17d725 2723 GFP_KERNEL);
d600ad1f
PT		 2724 if (IS_ERR(pgio->pg_lseg)) {
2725 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2726 pgio->pg_lseg = NULL;
2727 return;
2728 }
2729 }
e885de1a
TM		 2730 /* If no lseg, fall back to write through mds */
2731 if (pgio->pg_lseg == NULL)
1f945357 2732 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2733}
2734EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2735
180bb5ec
WAA		 2736void
2737pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2738{
2739 if (desc->pg_lseg) {
2740 pnfs_put_lseg(desc->pg_lseg);
2741 desc->pg_lseg = NULL;
2742 }
2743}
2744EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2745
b4fdac1a
WAA		 2746/*
2747 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2748 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2749 */
2750size_t
a7d42ddb
WAA		 2751pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2752 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2753{
0f9c429e 2754 unsigned int size;
c5e20cb7 2755 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2756
2757 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2758 if (!size)
2759 return 0;
94ad1c80 2760
19982ba8 2761 /*
c5e20cb7
WAA		 2762 * 'size' contains the number of bytes left in the current page (up
2763 * to the original size asked for in @req->wb_bytes).
2764 *
2765 * Calculate how many bytes are left in the layout segment
2766 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2767 *
2768 * Please also note that 'end_offset' is actually the offset of the
2769 * first byte that lies outside the pnfs_layout_range. FIXME?
2770 *
2771 */
19b54848 2772 if (pgio->pg_lseg) {
17822b20 2773 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2774 pgio->pg_lseg->pls_range.length);
2775 req_start = req_offset(req);
08cb5b0f 2776
c5e20cb7 2777 /* start of request is past the last byte of this segment */
08cb5b0f 2778 if (req_start >= seg_end)
19b54848 2779 return 0;
c5e20cb7
WAA		 2780
2781 /* adjust 'size' iff there are fewer bytes left in the
2782 * segment than what nfs_generic_pg_test returned */
2783 seg_left = seg_end - req_start;
2784 if (seg_left < size)
2785 size = (unsigned int)seg_left;
19b54848 2786 }
0f9c429e 2787
19b54848 2788 return size;
94ad1c80 2789}
89a58e32 2790EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2791
53113ad3 2792int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2793{
2794 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2795
2796 /* Resend all requests through the MDS */
53113ad3
WAA		 2797 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2798 hdr->completion_ops);
c7070113 2799 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2800 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2801}
e7dd79af 2802EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2803
d45f60c6 2804static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2805{
cd841605
FI		 2806
2807 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2808 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2809 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2810 pnfs_return_layout(hdr->inode);
1acbbb4e 2811 }
6c75dc0d 2812 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2813 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2814}
2815
d20581aa
BH		 2816/*
2817 * Called by non rpc-based layout drivers
2818 */
d45f60c6 2819void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2820{
f8417b48 2821 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2822 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2823 hdr->mds_offset + hdr->res.count);
d45f60c6 2824 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2825 }
2826 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2827 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2828 pnfs_ld_handle_write_error(hdr);
2829 hdr->mds_ops->rpc_release(hdr);
44b83799 2830}
d20581aa 2831EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2832
dce81290
TM		 2833static void
2834pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2835 struct nfs_pgio_header *hdr)
dce81290 2836{
48d635f1 2837 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2838
6c75dc0d 2839 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2840 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2841 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2842 mirror->pg_recoalesce = 1;
6c75dc0d 2843 }
ba4a76f7 2844 hdr->completion_ops->completion(hdr);
dce81290
TM		 2845}
2846
2847static enum pnfs_try_status
d45f60c6 2848pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2849 const struct rpc_call_ops *call_ops,
2850 struct pnfs_layout_segment *lseg,
2851 int how)
0382b744 2852{
cd841605 2853 struct inode *inode = hdr->inode;
0382b744
AA		 2854 enum pnfs_try_status trypnfs;
2855 struct nfs_server *nfss = NFS_SERVER(inode);
2856
cd841605 2857 hdr->mds_ops = call_ops;
0382b744
AA		 2858
2859 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2860 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2861 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2862 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2863 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2864 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2865 return trypnfs;
2866}
2867
dce81290 2868static void
7f714720
WAA		 2869pnfs_do_write(struct nfs_pageio_descriptor *desc,
2870 struct nfs_pgio_header *hdr, int how)
dce81290 2871{
dce81290
TM		 2872 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2873 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2874 enum pnfs_try_status trypnfs;
dce81290 2875
d45f60c6 2876 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2877 switch (trypnfs) {
2878 case PNFS_NOT_ATTEMPTED:
d45f60c6 2879 pnfs_write_through_mds(desc, hdr);
ffb81717 2880 break;
37f8aa16
TM		 2881 case PNFS_ATTEMPTED:
2882 break;
2883 case PNFS_TRY_AGAIN:
2884 /* cleanup hdr and prepare to redo pnfs */
2885 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2886 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2887 list_splice_init(&hdr->pages, &mirror->pg_list);
2888 mirror->pg_recoalesce = 1;
2889 }
2890 hdr->mds_ops->rpc_release(hdr);
2891 }
dce81290
TM		 2892}
2893
6c75dc0d
FI		 2894static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2895{
9369a431 2896 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2897 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2898}
2899
dce81290
TM		 2900int
2901pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2902{
6c75dc0d 2903 struct nfs_pgio_header *hdr;
dce81290
TM		 2904 int ret;
2905
1e7f3a48
WAA		 2906 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2907 if (!hdr) {
2bff2288
PT		 2908 desc->pg_error = -ENOMEM;
2909 return desc->pg_error;
dce81290 2910 }
6c75dc0d 2911 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2912
9369a431 2913 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2914 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2915 if (!ret)
7f714720 2916 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2917
6c75dc0d 2918 return ret;
dce81290
TM		 2919}
2920EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2921
53113ad3 2922int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2923{
2924 struct nfs_pageio_descriptor pgio;
2925
1acbbb4e 2926 /* Resend all requests through the MDS */
53113ad3
WAA		 2927 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2928 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2929}
e7dd79af 2930EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2931
d45f60c6 2932static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2933{
cd841605
FI		 2934 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2935 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2936 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2937 pnfs_return_layout(hdr->inode);
1acbbb4e 2938 }
4db6e0b7 2939 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2940 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2941}
2942
d20581aa
BH		 2943/*
2944 * Called by non rpc-based layout drivers
2945 */
d45f60c6 2946void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2947{
bfc505de 2948 if (likely(!hdr->pnfs_error))
d45f60c6 2949 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2950 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2951 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2952 pnfs_ld_handle_read_error(hdr);
2953 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2954}
2955EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2956
493292dd
TM		 2957static void
2958pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2959 struct nfs_pgio_header *hdr)
493292dd 2960{
48d635f1 2961 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2962
4db6e0b7 2963 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2964 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2965 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2966 mirror->pg_recoalesce = 1;
4db6e0b7 2967 }
ba4a76f7 2968 hdr->completion_ops->completion(hdr);
493292dd
TM		 2969}
2970
64419a9b
AA		 2971/*
2972 * Call the appropriate parallel I/O subsystem read function.
2973 */
493292dd 2974static enum pnfs_try_status
d45f60c6 2975pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2976 const struct rpc_call_ops *call_ops,
2977 struct pnfs_layout_segment *lseg)
64419a9b 2978{
cd841605 2979 struct inode *inode = hdr->inode;
64419a9b
AA		 2980 struct nfs_server *nfss = NFS_SERVER(inode);
2981 enum pnfs_try_status trypnfs;
2982
cd841605 2983 hdr->mds_ops = call_ops;
64419a9b
AA		 2984
2985 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2986 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2987
d45f60c6 2988 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2989 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2990 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2991 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2992 return trypnfs;
2993}
863a3c6c 2994
ceb11e13 2995/* Resend all requests through pnfs. */
563c53e7
TM		 2996void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr,
2997 unsigned int mirror_idx)
ceb11e13
PT		 2998{
2999 struct nfs_pageio_descriptor pgio;
3000
1b1bc66b 3001 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 3002 /* Prevent deadlocks with layoutreturn! */
3003 pnfs_put_lseg(hdr->lseg);
3004 hdr->lseg = NULL;
3005
1b1bc66b
WAA		 3006 nfs_pageio_init_read(&pgio, hdr->inode, false,
3007 hdr->completion_ops);
563c53e7 3008 pgio.pg_mirror_idx = mirror_idx;
1b1bc66b
WAA		 3009 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
3010 }
ceb11e13
PT		 3011}
3012EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
3013
493292dd 3014static void
7f714720 3015pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 3016{
493292dd
TM		 3017 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
3018 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 3019 enum pnfs_try_status trypnfs;
493292dd 3020
d45f60c6 3021 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 3022 switch (trypnfs) {
3023 case PNFS_NOT_ATTEMPTED:
d45f60c6 3024 pnfs_read_through_mds(desc, hdr);
ffb81717 3025 break;
6aeafd05
TM		 3026 case PNFS_ATTEMPTED:
3027 break;
3028 case PNFS_TRY_AGAIN:
3029 /* cleanup hdr and prepare to redo pnfs */
3030 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3031 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3032 list_splice_init(&hdr->pages, &mirror->pg_list);
3033 mirror->pg_recoalesce = 1;
3034 }
3035 hdr->mds_ops->rpc_release(hdr);
3036 }
493292dd
TM		 3037}
3038
4db6e0b7
FI		 3039static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
3040{
9369a431 3041 pnfs_put_lseg(hdr->lseg);
1e7f3a48 3042 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 3043}
3044
493292dd
TM		 3045int
3046pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
3047{
4db6e0b7 3048 struct nfs_pgio_header *hdr;
493292dd
TM		 3049 int ret;
3050
1e7f3a48
WAA		 3051 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
3052 if (!hdr) {
2bff2288
PT		 3053 desc->pg_error = -ENOMEM;
3054 return desc->pg_error;
493292dd 3055 }
4db6e0b7 3056 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 3057 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 3058 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 3059 if (!ret)
7f714720 3060 pnfs_do_read(desc, hdr);
4db6e0b7 3061 return ret;
493292dd
TM		 3062}
3063EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
3064
71244d9b
TM		 3065static void pnfs_clear_layoutcommitting(struct inode *inode)
3066{
3067 unsigned long *bitlock = &NFS_I(inode)->flags;
3068
3069 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 3070 smp_mb__after_atomic();
71244d9b
TM		 3071 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
3072}
3073
863a3c6c 3074/*
a9bae566 3075 * There can be multiple RW segments.
863a3c6c 3076 */
a9bae566 3077static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 3078{
a9bae566 3079 struct pnfs_layout_segment *lseg;
863a3c6c 3080
a9bae566
PT		 3081 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
3082 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 3083 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 3084 list_add(&lseg->pls_lc_list, listp);
3085 }
863a3c6c
AA		 3086}
3087
a073dbff
TM		 3088static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
3089{
3090 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 3091
3092 /* Matched by references in pnfs_set_layoutcommit */
3093 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
3094 list_del_init(&lseg->pls_lc_list);
3095 pnfs_put_lseg(lseg);
3096 }
3097
71244d9b 3098 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 3099}
3100
1b0ae068
PT		 3101void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
3102{
b9e028fd 3103 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 3104}
3105EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
3106
863a3c6c 3107void
67af7611
TM		 3108pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
3109 loff_t end_pos)
863a3c6c 3110{
cd841605 3111 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 3112 bool mark_as_dirty = false;
863a3c6c 3113
cd841605 3114 spin_lock(&inode->i_lock);
863a3c6c 3115 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 3116 nfsi->layout->plh_lwb = end_pos;
79a48a1f 3117 mark_as_dirty = true;
863a3c6c 3118 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 3119 __func__, inode->i_ino);
29559b11
TM		 3120 } else if (end_pos > nfsi->layout->plh_lwb)
3121 nfsi->layout->plh_lwb = end_pos;
67af7611 3122 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 3123 /* references matched in nfs4_layoutcommit_release */
67af7611 3124 pnfs_get_lseg(lseg);
a9bae566 3125 }
cd841605 3126 spin_unlock(&inode->i_lock);
acff5880 3127 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 3128 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 3129
3130 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3131 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3132 if (mark_as_dirty)
cd841605 3133 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3134}
3135EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
3136
db29c089
AA		 3137void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
3138{
3139 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
3140
3141 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
3142 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 3143 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 3144}
3145
de4b15c7
AA		 3146/*
3147 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3148 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3149 * data to disk to allow the server to recover the data if it crashes.
3150 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3151 * is off, and a COMMIT is sent to a data server, or
3152 * if WRITEs to a data server return NFS_DATA_SYNC.
3153 */
863a3c6c 3154int
ef311537 3155pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 3156{
5f919c9f 3157 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 3158 struct nfs4_layoutcommit_data *data;
3159 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 3160 loff_t end_pos;
71244d9b 3161 int status;
863a3c6c 3162
71244d9b 3163 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 3164 return 0;
3165
71244d9b 3166 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 3167
71244d9b 3168 status = -EAGAIN;
92407e75 3169 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 3170 if (!sync)
3171 goto out;
74316201 3172 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 3173 NFS_INO_LAYOUTCOMMITTING,
3174 nfs_wait_bit_killable,
3175 TASK_KILLABLE);
92407e75 3176 if (status)
71244d9b 3177 goto out;
92407e75
PT		 3178 }
3179
71244d9b
TM		 3180 status = -ENOMEM;
3181 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
3182 data = kzalloc(sizeof(*data), GFP_NOFS);
3183 if (!data)
3184 goto clear_layoutcommitting;
3185
3186 status = 0;
de4b15c7 3187 spin_lock(&inode->i_lock);
71244d9b
TM		 3188 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
3189 goto out_unlock;
a9bae566 3190
71244d9b 3191 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 3192 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 3193
acff5880 3194 end_pos = nfsi->layout->plh_lwb;
863a3c6c 3195
f597c537 3196 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
97a728f5 3197 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 3198 spin_unlock(&inode->i_lock);
3199
3200 data->args.inode = inode;
863a3c6c
AA		 3201 nfs_fattr_init(&data->fattr);
3202 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3203 data->res.fattr = &data->fattr;
2e18d4d8
TM		 3204 if (end_pos != 0)
3205 data->args.lastbytewritten = end_pos - 1;
3206 else
3207 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 3208 data->res.server = NFS_SERVER(inode);
3209
5f919c9f
CH		 3210 if (ld->prepare_layoutcommit) {
3211 status = ld->prepare_layoutcommit(&data->args);
3212 if (status) {
a52458b4 3213 put_cred(data->cred);
5f919c9f 3214 spin_lock(&inode->i_lock);
29559b11
TM		 3215 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3216 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 3217 nfsi->layout->plh_lwb = end_pos;
3471648a 3218 goto out_unlock;
5f919c9f
CH		 3219 }
3220 }
3221
3222
863a3c6c
AA		 3223 status = nfs4_proc_layoutcommit(data, sync);
3224out:
92407e75
PT		 3225 if (status)
3226 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3227 dprintk("<-- %s status %d\n", __func__, status);
3228 return status;
71244d9b
TM		 3229out_unlock:
3230 spin_unlock(&inode->i_lock);
92407e75 3231 kfree(data);
71244d9b
TM		 3232clear_layoutcommitting:
3233 pnfs_clear_layoutcommitting(inode);
92407e75 3234 goto out;
863a3c6c 3235}
72cff449 3236EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 3237
5bb89b47
TM		 3238int
3239pnfs_generic_sync(struct inode *inode, bool datasync)
3240{
3241 return pnfs_layoutcommit_inode(inode, true);
3242}
3243EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3244
82be417a
AA		 3245struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3246{
3247 struct nfs4_threshold *thp;
3248
3249 thp = kzalloc(sizeof(*thp), GFP_NOFS);
3250 if (!thp) {
3251 dprintk("%s mdsthreshold allocation failed\n", __func__);
3252 return NULL;
3253 }
3254 return thp;
3255}
8733408d 3256
865a7ecb 3257#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 3258int
c8ad8894 3259pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 3260{
3261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3262 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 3263 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 3264 struct nfs42_layoutstat_data *data;
3265 struct pnfs_layout_hdr *hdr;
3266 int status = 0;
3267
3268 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3269 goto out;
3270
6c5a0d89
TM		 3271 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3272 goto out;
3273
1bfe3b25
PT		 3274 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3275 goto out;
3276
8733408d
PT		 3277 spin_lock(&inode->i_lock);
3278 if (!NFS_I(inode)->layout) {
3279 spin_unlock(&inode->i_lock);
f538d0ba 3280 goto out_clear_layoutstats;
8733408d
PT		 3281 }
3282 hdr = NFS_I(inode)->layout;
3283 pnfs_get_layout_hdr(hdr);
3284 spin_unlock(&inode->i_lock);
3285
c8ad8894 3286 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3287 if (!data) {
3288 status = -ENOMEM;
3289 goto out_put;
3290 }
3291
3292 data->args.fh = NFS_FH(inode);
3293 data->args.inode = inode;
8733408d
PT		 3294 status = ld->prepare_layoutstats(&data->args);
3295 if (status)
3296 goto out_free;
3297
3298 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3299
3300out:
3301 dprintk("%s returns %d\n", __func__, status);
3302 return status;
3303
3304out_free:
3305 kfree(data);
3306out_put:
3307 pnfs_put_layout_hdr(hdr);
f538d0ba 3308out_clear_layoutstats:
1bfe3b25
PT		 3309 smp_mb__before_atomic();
3310 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3311 smp_mb__after_atomic();
8733408d
PT		 3312 goto out;
3313}
3314EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3315#endif
bbf58bf3
TM		 3316
3317unsigned int layoutstats_timer;
3318module_param(layoutstats_timer, uint, 0644);
3319EXPORT_SYMBOL_GPL(layoutstats_timer);
Linux 6.x block layer and io_uring tree(s)