projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
NFSv4/pnfs: Don't switch off layoutget-on-open for transient errors
[linux-block.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
95void
96unset_pnfs_layoutdriver(struct nfs_server *nfss)
97{
738fd0f3
BH		 98 if (nfss->pnfs_curr_ld) {
99 if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 104 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 105 }
85e174ba
RL		 106 nfss->pnfs_curr_ld = NULL;
107}
108
ca440c38
JL		 109/*
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
112 *
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
115 */
116static const u32 ld_prefs[] = {
117 LAYOUT_SCSI,
118 LAYOUT_BLOCK_VOLUME,
119 LAYOUT_OSD2_OBJECTS,
120 LAYOUT_FLEX_FILES,
121 LAYOUT_NFSV4_1_FILES,
122 0
123};
124
125static int
126ld_cmp(const void *e1, const void *e2)
127{
128 u32 ld1 = *((u32 *)e1);
129 u32 ld2 = *((u32 *)e2);
130 int i;
131
132 for (i = 0; ld_prefs[i] != 0; i++) {
133 if (ld1 == ld_prefs[i])
134 return -1;
135
136 if (ld2 == ld_prefs[i])
137 return 1;
138 }
139 return 0;
140}
141
85e174ba
RL		 142/*
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
145 *
3132e49e 146 * @ids array of layout types supported by MDS.
85e174ba
RL		 147 */
148void
738fd0f3 149set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 150 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 151{
152 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 153 u32 id;
ca440c38 154 int i;
85e174ba 155
19274716
AS		 156 if (fsinfo->nlayouttypes == 0)
157 goto out_no_driver;
85e174ba
RL		 158 if (!(server->nfs_client->cl_exchange_flags &
159 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 160 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 162 goto out_no_driver;
163 }
3132e49e 164
ca440c38
JL		 165 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 167
ca440c38
JL		 168 for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 id = fsinfo->layouttype[i];
85e174ba 170 ld_type = find_pnfs_driver(id);
ca440c38
JL		 171 if (!ld_type) {
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 id);
174 ld_type = find_pnfs_driver(id);
175 }
176 if (ld_type)
177 break;
85e174ba 178 }
3132e49e
JL		 179
180 if (!ld_type) {
ca440c38 181 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 182 goto out_no_driver;
183 }
184
85e174ba 185 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 186 if (ld_type->set_layoutdriver
187 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 188 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__, id);
738fd0f3
BH		 190 module_put(ld_type->owner);
191 goto out_no_driver;
192 }
2a4c8994
TM		 193 /* Bump the MDS count */
194 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 195
85e174ba
RL		 196 dprintk("%s: pNFS module for %u set\n", __func__, id);
197 return;
198
199out_no_driver:
200 dprintk("%s: Using NFSv4 I/O\n", __func__);
201 server->pnfs_curr_ld = NULL;
202}
02c35fca
FI		 203
204int
205pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206{
207 int status = -EINVAL;
208 struct pnfs_layoutdriver_type *tmp;
209
210 if (ld_type->id == 0) {
a030889a 211 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 212 return status;
213 }
b1f69b75 214 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 215 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 216 "alloc_lseg and free_lseg.\n", __func__);
217 return status;
218 }
02c35fca
FI		 219
220 spin_lock(&pnfs_spinlock);
221 tmp = find_pnfs_driver_locked(ld_type->id);
222 if (!tmp) {
223 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 status = 0;
225 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 ld_type->name);
227 } else {
a030889a 228 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 229 __func__, ld_type->id);
230 }
231 spin_unlock(&pnfs_spinlock);
232
233 return status;
234}
235EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236
237void
238pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239{
240 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 spin_lock(&pnfs_spinlock);
242 list_del(&ld_type->pnfs_tblid);
243 spin_unlock(&pnfs_spinlock);
244}
245EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 246
b1f69b75
AA		 247/*
248 * pNFS client layout cache
249 */
250
cc6e5340 251/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 252void
70c3bd2b 253pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 254{
2b28a7be 255 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 256}
257
636fb9c8
BH		 258static struct pnfs_layout_hdr *
259pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260{
261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 262 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 263}
264
265static void
266pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267{
9c626381
TM		 268 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270
271 if (!list_empty(&lo->plh_layouts)) {
272 struct nfs_client *clp = server->nfs_client;
273
274 spin_lock(&clp->cl_lock);
275 list_del_init(&lo->plh_layouts);
276 spin_unlock(&clp->cl_lock);
277 }
9fa40758 278 put_rpccred(lo->plh_lc_cred);
57934278 279 return ld->free_layout_hdr(lo);
636fb9c8
BH		 280}
281
e5e94017 282static void
6622c3ea 283pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 284{
bb346f63 285 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 286 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 287 nfsi->layout = NULL;
288 /* Reset MDS Threshold I/O counters */
289 nfsi->write_io = 0;
290 nfsi->read_io = 0;
e5e94017
BH		 291}
292
b1f69b75 293void
70c3bd2b 294pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 295{
9c6376eb 296 struct inode *inode;
cc6e5340 297
9c6376eb
TM		 298 if (!lo)
299 return;
300 inode = lo->plh_inode;
13c13a6a
TM		 301 pnfs_layoutreturn_before_put_layout_hdr(lo);
302
2b28a7be 303 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 304 if (!list_empty(&lo->plh_segs))
305 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 306 pnfs_detach_layout_hdr(lo);
cc6e5340 307 spin_unlock(&inode->i_lock);
6622c3ea 308 pnfs_free_layout_hdr(lo);
cc6e5340 309 }
974cec8c
AA		 310}
311
4aab9732
TM		 312static void
313pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
314 u32 seq)
315{
316 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
317 iomode = IOMODE_ANY;
318 lo->plh_return_iomode = iomode;
319 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
320 if (seq != 0) {
321 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
322 lo->plh_return_seq = seq;
323 }
324}
325
ae5a459d
TM		 326static void
327pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
328{
5466d214 329 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 330 lo->plh_return_iomode = 0;
331 lo->plh_return_seq = 0;
332 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 333 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
334 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
335 continue;
336 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
337 }
ae5a459d
TM		 338}
339
362fb578
TM		 340static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
341{
342 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
343 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
344 smp_mb__after_atomic();
345 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
346 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
347}
348
fe1cf946
TM		 349static void
350pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
351 struct list_head *free_me)
352{
353 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
354 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
355 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
356 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
357 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
358 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
359}
360
7380020e
TM		 361/*
362 * Update the seqid of a layout stateid
363 */
364bool nfs4_refresh_layout_stateid(nfs4_stateid *dst, struct inode *inode)
365{
366 struct pnfs_layout_hdr *lo;
367 bool ret = false;
368
369 spin_lock(&inode->i_lock);
370 lo = NFS_I(inode)->layout;
371 if (lo && nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
372 dst->seqid = lo->plh_stateid.seqid;
373 ret = true;
374 }
375 spin_unlock(&inode->i_lock);
376 return ret;
377}
378
2454dfea
TM		 379/*
380 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
381 *
382 * In order to continue using the pnfs_layout_hdr, a full recovery
383 * is required.
384 * Note that caller must hold inode->i_lock.
385 */
5f46be04 386int
2454dfea
TM		 387pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
388 struct list_head *lseg_list)
389{
390 struct pnfs_layout_range range = {
391 .iomode = IOMODE_ANY,
392 .offset = 0,
393 .length = NFS4_MAX_UINT64,
394 };
fe1cf946 395 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 396
397 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 398 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
399 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 400 pnfs_clear_layoutreturn_info(lo);
68f74479 401 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 402 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
403 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
404 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 405 return !list_empty(&lo->plh_segs);
2454dfea
TM		 406}
407
b9e028fd
TM		 408static int
409pnfs_iomode_to_fail_bit(u32 iomode)
410{
411 return iomode == IOMODE_RW ?
412 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
413}
414
415static void
3e621214 416pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 417{
25c75333 418 lo->plh_retry_timestamp = jiffies;
39e88fcf 419 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 420 refcount_inc(&lo->plh_refcount);
3e621214
TM		 421}
422
423static void
424pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
425{
426 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 427 refcount_dec(&lo->plh_refcount);
3e621214
TM		 428}
429
430static void
431pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
432{
433 struct inode *inode = lo->plh_inode;
115ce575
TM		 434 struct pnfs_layout_range range = {
435 .iomode = iomode,
436 .offset = 0,
437 .length = NFS4_MAX_UINT64,
438 };
439 LIST_HEAD(head);
3e621214
TM		 440
441 spin_lock(&inode->i_lock);
442 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 443 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 444 spin_unlock(&inode->i_lock);
115ce575 445 pnfs_free_lseg_list(&head);
b9e028fd
TM		 446 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
447 iomode == IOMODE_RW ? "RW" : "READ");
448}
449
450static bool
451pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
452{
25c75333 453 unsigned long start, end;
3e621214
TM		 454 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
455
456 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 457 return false;
458 end = jiffies;
459 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
460 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
461 /* It is time to retry the failed layoutgets */
3e621214 462 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 463 return false;
464 }
465 return true;
b9e028fd
TM		 466}
467
974cec8c 468static void
119cef97
TM		 469pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
470 const struct pnfs_layout_range *range,
471 const nfs4_stateid *stateid)
974cec8c 472{
566052c5 473 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 474 INIT_LIST_HEAD(&lseg->pls_lc_list);
eba6dd69 475 refcount_set(&lseg->pls_refcount, 1);
4541d16c 476 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 477 lseg->pls_layout = lo;
119cef97
TM		 478 lseg->pls_range = *range;
479 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 480}
481
905ca191 482static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 483{
68f74479
TM		 484 if (lseg != NULL) {
485 struct inode *inode = lseg->pls_layout->plh_inode;
486 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
487 }
974cec8c
AA		 488}
489
d684d2ae 490static void
57036a37
TM		 491pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
492 struct pnfs_layout_segment *lseg)
d684d2ae 493{
d20581aa 494 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 495 list_del_init(&lseg->pls_list);
8f0d27dc 496 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 497 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 498 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
499 return;
7b650994
TM		 500 if (list_empty(&lo->plh_segs) &&
501 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
502 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 503 if (atomic_read(&lo->plh_outstanding) == 0)
504 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 505 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 506 }
d684d2ae
FI		 507}
508
68f74479
TM		 509static bool
510pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
511 struct pnfs_layout_segment *lseg)
512{
513 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
514 pnfs_layout_is_valid(lo)) {
4aab9732 515 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 516 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
517 return true;
518 }
519 return false;
520}
521
bae724ef 522void
9369a431 523pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 524{
57036a37 525 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 526 struct inode *inode;
527
528 if (!lseg)
529 return;
530
4541d16c 531 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 532 refcount_read(&lseg->pls_refcount),
4541d16c 533 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 534
57036a37
TM		 535 lo = lseg->pls_layout;
536 inode = lo->plh_inode;
4ef2e4f8 537
eba6dd69 538 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 539 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
540 spin_unlock(&inode->i_lock);
541 return;
542 }
8f0d27dc 543 pnfs_get_layout_hdr(lo);
4ef2e4f8 544 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 545 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
546 lseg = NULL;
4ef2e4f8
TM		 547 spin_unlock(&inode->i_lock);
548 pnfs_free_lseg(lseg);
549 pnfs_put_layout_hdr(lo);
4541d16c 550 }
4541d16c 551}
9369a431 552EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 553
fb3296eb
BH		 554/*
555 * is l2 fully contained in l1?
556 * start1 end1
557 * [----------------------------------)
558 * start2 end2
559 * [----------------)
560 */
3cb2df17 561static bool
7dc0ac70 562pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 563 const struct pnfs_layout_range *l2)
fb3296eb
BH		 564{
565 u64 start1 = l1->offset;
17822b20 566 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 567 u64 start2 = l2->offset;
17822b20 568 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 569
570 return (start1 <= start2) && (end1 >= end2);
571}
572
24956804
TM		 573static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
574 struct list_head *tmp_list)
575{
eba6dd69 576 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 577 return false;
578 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
579 list_add(&lseg->pls_list, tmp_list);
580 return true;
581}
582
4541d16c
FI		 583/* Returns 1 if lseg is removed from list, 0 otherwise */
584static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
585 struct list_head *tmp_list)
586{
587 int rv = 0;
588
589 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
590 /* Remove the reference keeping the lseg in the
591 * list. It will now be removed when all
592 * outstanding io is finished.
593 */
d684d2ae 594 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 595 refcount_read(&lseg->pls_refcount));
24956804 596 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 597 rv = 1;
4541d16c
FI		 598 }
599 return rv;
600}
601
6d597e17
JL		 602/*
603 * Compare 2 layout stateid sequence ids, to see which is newer,
604 * taking into account wraparound issues.
605 */
606static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
607{
608 return (s32)(s1 - s2) > 0;
609}
610
e036f464
TM		 611static bool
612pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
613 const struct pnfs_layout_range *recall_range)
614{
615 return (recall_range->iomode == IOMODE_ANY ||
616 lseg_range->iomode == recall_range->iomode) &&
617 pnfs_lseg_range_intersecting(lseg_range, recall_range);
618}
619
620static bool
621pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
622 const struct pnfs_layout_range *recall_range,
623 u32 seq)
624{
625 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
626 return false;
627 if (recall_range == NULL)
628 return true;
629 return pnfs_should_free_range(&lseg->pls_range, recall_range);
630}
631
6d597e17
JL		 632/**
633 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
634 * @lo: layout header containing the lsegs
635 * @tmp_list: list head where doomed lsegs should go
636 * @recall_range: optional recall range argument to match (may be NULL)
637 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
638 *
639 * Walk the list of lsegs in the layout header, and tear down any that should
640 * be destroyed. If "recall_range" is specified then the segment must match
641 * that range. If "seq" is non-zero, then only match segments that were handed
642 * out at or before that sequence.
643 *
644 * Returns number of matching invalid lsegs remaining in list after scanning
645 * it and purging them.
4541d16c 646 */
43f1b3da 647int
49a85061 648pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 649 struct list_head *tmp_list,
6d597e17
JL		 650 const struct pnfs_layout_range *recall_range,
651 u32 seq)
974cec8c
AA		 652{
653 struct pnfs_layout_segment *lseg, *next;
71b39854 654 int remaining = 0;
974cec8c
AA		 655
656 dprintk("%s:Begin lo %p\n", __func__, lo);
657
8006bfba 658 if (list_empty(&lo->plh_segs))
38511722 659 return 0;
4541d16c 660 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 661 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 662 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 663 "offset %llu length %llu\n", __func__,
6d597e17
JL		 664 lseg, lseg->pls_range.iomode, lseg->pls_seq,
665 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 666 if (!mark_lseg_invalid(lseg, tmp_list))
667 remaining++;
4541d16c 668 }
71b39854
TM		 669 dprintk("%s:Return %i\n", __func__, remaining);
670 return remaining;
974cec8c
AA		 671}
672
68f74479
TM		 673static void
674pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
675 struct list_head *free_me,
676 const struct pnfs_layout_range *range,
677 u32 seq)
678{
679 struct pnfs_layout_segment *lseg, *next;
680
681 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
682 if (pnfs_match_lseg_recall(lseg, range, seq))
683 list_move_tail(&lseg->pls_list, free_me);
684 }
685}
686
f49f9baa 687/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 688void
4541d16c 689pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 690{
4541d16c 691 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 692
693 if (list_empty(free_me))
694 return;
695
4541d16c 696 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 697 list_del(&lseg->pls_list);
905ca191 698 pnfs_free_lseg(lseg);
974cec8c
AA		 699 }
700}
701
e5e94017
BH		 702void
703pnfs_destroy_layout(struct nfs_inode *nfsi)
704{
705 struct pnfs_layout_hdr *lo;
974cec8c 706 LIST_HEAD(tmp_list);
e5e94017
BH		 707
708 spin_lock(&nfsi->vfs_inode.i_lock);
709 lo = nfsi->layout;
710 if (lo) {
3e621214 711 pnfs_get_layout_hdr(lo);
2454dfea 712 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 713 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
714 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
715 spin_unlock(&nfsi->vfs_inode.i_lock);
716 pnfs_free_lseg_list(&tmp_list);
1f18b82c 717 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 718 pnfs_put_layout_hdr(lo);
719 } else
720 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 721}
041245c8 722EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 723
fd9a8d71
TM		 724static bool
725pnfs_layout_add_bulk_destroy_list(struct inode *inode,
726 struct list_head *layout_list)
974cec8c
AA		 727{
728 struct pnfs_layout_hdr *lo;
fd9a8d71 729 bool ret = false;
974cec8c 730
fd9a8d71
TM		 731 spin_lock(&inode->i_lock);
732 lo = NFS_I(inode)->layout;
733 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
734 pnfs_get_layout_hdr(lo);
735 list_add(&lo->plh_bulk_destroy, layout_list);
736 ret = true;
737 }
738 spin_unlock(&inode->i_lock);
739 return ret;
740}
741
742/* Caller must hold rcu_read_lock and clp->cl_lock */
743static int
744pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
745 struct nfs_server *server,
746 struct list_head *layout_list)
747{
748 struct pnfs_layout_hdr *lo, *next;
749 struct inode *inode;
750
751 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
b85f5620
TM		 752 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
753 continue;
fd9a8d71
TM		 754 inode = igrab(lo->plh_inode);
755 if (inode == NULL)
756 continue;
757 list_del_init(&lo->plh_layouts);
758 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
759 continue;
760 rcu_read_unlock();
761 spin_unlock(&clp->cl_lock);
762 iput(inode);
763 spin_lock(&clp->cl_lock);
764 rcu_read_lock();
765 return -EAGAIN;
766 }
767 return 0;
768}
769
770static int
771pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
772 bool is_bulk_recall)
773{
774 struct pnfs_layout_hdr *lo;
775 struct inode *inode;
fd9a8d71
TM		 776 LIST_HEAD(lseg_list);
777 int ret = 0;
778
779 while (!list_empty(layout_list)) {
780 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
781 plh_bulk_destroy);
782 dprintk("%s freeing layout for inode %lu\n", __func__,
783 lo->plh_inode->i_ino);
784 inode = lo->plh_inode;
7c5d1875
CH		 785
786 pnfs_layoutcommit_inode(inode, false);
787
fd9a8d71
TM		 788 spin_lock(&inode->i_lock);
789 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 790 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
791 if (is_bulk_recall)
792 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 793 ret = -EAGAIN;
9fd4b9fc 794 }
fd9a8d71
TM		 795 spin_unlock(&inode->i_lock);
796 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 797 /* Free all lsegs that are attached to commit buckets */
798 nfs_commit_inode(inode, 0);
fd9a8d71
TM		 799 pnfs_put_layout_hdr(lo);
800 iput(inode);
801 }
802 return ret;
803}
804
805int
806pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
807 struct nfs_fsid *fsid,
808 bool is_recall)
809{
810 struct nfs_server *server;
811 LIST_HEAD(layout_list);
c47abcf8 812
974cec8c 813 spin_lock(&clp->cl_lock);
6382a441 814 rcu_read_lock();
fd9a8d71 815restart:
6382a441 816 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 817 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
818 continue;
819 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
820 server,
821 &layout_list) != 0)
822 goto restart;
6382a441
WAA		 823 }
824 rcu_read_unlock();
974cec8c
AA		 825 spin_unlock(&clp->cl_lock);
826
fd9a8d71
TM		 827 if (list_empty(&layout_list))
828 return 0;
829 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
830}
831
832int
833pnfs_destroy_layouts_byclid(struct nfs_client *clp,
834 bool is_recall)
835{
836 struct nfs_server *server;
837 LIST_HEAD(layout_list);
838
839 spin_lock(&clp->cl_lock);
840 rcu_read_lock();
841restart:
842 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
843 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
844 server,
845 &layout_list) != 0)
846 goto restart;
974cec8c 847 }
fd9a8d71
TM		 848 rcu_read_unlock();
849 spin_unlock(&clp->cl_lock);
850
851 if (list_empty(&layout_list))
852 return 0;
853 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
854}
855
856/*
857 * Called by the state manger to remove all layouts established under an
858 * expired lease.
859 */
860void
861pnfs_destroy_all_layouts(struct nfs_client *clp)
862{
863 nfs4_deviceid_mark_client_invalid(clp);
864 nfs4_deviceid_purge_client(clp);
865
866 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 867}
868
fd6002e9 869/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 870void
871pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
872 bool update_barrier)
b1f69b75 873{
ecebb80b 874 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 875
2d2f24ad
TM		 876 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
877 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 878
879 if (!pnfs_layout_is_valid(lo)) {
880 nfs4_stateid_copy(&lo->plh_stateid, new);
881 lo->plh_barrier = newseq;
882 pnfs_clear_layoutreturn_info(lo);
883 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
884 return;
885 }
886 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 887 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 888 /*
889 * Because of wraparound, we want to keep the barrier
890 * "close" to the current seqids.
891 */
892 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 893 }
ecebb80b
TM		 894 if (update_barrier)
895 new_barrier = be32_to_cpu(new->seqid);
896 else if (new_barrier == 0)
897 return;
2a59a041 898 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 899 lo->plh_barrier = new_barrier;
b1f69b75
AA		 900}
901
cf7d63f1 902static bool
19c54aba
TM		 903pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
904 const nfs4_stateid *stateid)
43f1b3da 905{
19c54aba 906 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 907
19c54aba
TM		 908 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
909}
910
911/* lget is set to 1 if called from inside send_layoutget call chain */
912static bool
e1c06f80 913pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 914{
f7e8917a 915 return lo->plh_block_lgets ||
e1c06f80 916 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 917}
918
5e36e2a9
FI		 919static struct nfs_server *
920pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
921{
922 struct nfs_server *server;
923
78746a38 924 if (inode) {
5e36e2a9 925 server = NFS_SERVER(inode);
78746a38 926 } else {
5e36e2a9
FI		 927 struct dentry *parent_dir = dget_parent(ctx->dentry);
928 server = NFS_SERVER(parent_dir->d_inode);
929 dput(parent_dir);
930 }
931 return server;
932}
933
29a8bfe5
TM		 934static void nfs4_free_pages(struct page **pages, size_t size)
935{
936 int i;
937
938 if (!pages)
939 return;
940
941 for (i = 0; i < size; i++) {
942 if (!pages[i])
943 break;
944 __free_page(pages[i]);
945 }
946 kfree(pages);
947}
948
949static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
950{
951 struct page **pages;
952 int i;
953
954 pages = kcalloc(size, sizeof(struct page *), gfp_flags);
955 if (!pages) {
956 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
957 return NULL;
958 }
959
960 for (i = 0; i < size; i++) {
961 pages[i] = alloc_page(gfp_flags);
962 if (!pages[i]) {
963 dprintk("%s: failed to allocate page\n", __func__);
964 nfs4_free_pages(pages, size);
965 return NULL;
966 }
967 }
968
969 return pages;
970}
971
587f03de 972static struct nfs4_layoutget *
5e36e2a9 973pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 974 struct nfs_open_context *ctx,
2409a976 975 const nfs4_stateid *stateid,
e144e539 976 const struct pnfs_layout_range *range,
587f03de 977 gfp_t gfp_flags)
e5e94017 978{
5e36e2a9 979 struct nfs_server *server = pnfs_find_server(ino, ctx);
dacb452d 980 size_t max_pages = max_response_pages(server);
b1f69b75 981 struct nfs4_layoutget *lgp;
b1f69b75
AA		 982
983 dprintk("--> %s\n", __func__);
e5e94017 984
83026d80
JL		 985 lgp = kzalloc(sizeof(*lgp), gfp_flags);
986 if (lgp == NULL)
587f03de 987 return NULL;
83026d80 988
dacb452d
FI		 989 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
990 if (!lgp->args.layout.pages) {
991 kfree(lgp);
992 return NULL;
993 }
994 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
995 lgp->res.layoutp = &lgp->args.layout;
996
d49e0d5b
TM		 997 /* Don't confuse uninitialised result and success */
998 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 999
1000 lgp->args.minlength = PAGE_SIZE;
1001 if (lgp->args.minlength > range->length)
1002 lgp->args.minlength = range->length;
5e36e2a9
FI		 1003 if (ino) {
1004 loff_t i_size = i_size_read(ino);
1005
1006 if (range->iomode == IOMODE_READ) {
1007 if (range->offset >= i_size)
1008 lgp->args.minlength = 0;
1009 else if (i_size - range->offset < lgp->args.minlength)
1010 lgp->args.minlength = i_size - range->offset;
1011 }
d03ab29d 1012 }
83026d80
JL		 1013 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1014 pnfs_copy_range(&lgp->args.range, range);
1015 lgp->args.type = server->pnfs_curr_ld->id;
1016 lgp->args.inode = ino;
1017 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1018 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1019 lgp->gfp_flags = gfp_flags;
2409a976 1020 lgp->cred = get_rpccred(ctx->cred);
c49b5209 1021 lgp->callback_count = raw_seqcount_begin(&server->nfs_client->cl_callback_count);
587f03de 1022 return lgp;
974cec8c
AA		 1023}
1024
29a8bfe5
TM		 1025void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1026{
1027 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1028
1029 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1030 if (lgp->args.inode)
1031 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
1032 put_rpccred(lgp->cred);
1033 put_nfs_open_context(lgp->args.ctx);
1034 kfree(lgp);
1035}
1036
24956804
TM		 1037static void pnfs_clear_layoutcommit(struct inode *inode,
1038 struct list_head *head)
1039{
1040 struct nfs_inode *nfsi = NFS_I(inode);
1041 struct pnfs_layout_segment *lseg, *tmp;
1042
1043 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1044 return;
1045 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1046 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1047 continue;
1048 pnfs_lseg_dec_and_remove_zero(lseg, head);
1049 }
1050}
1051
68f74479 1052void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1053 const nfs4_stateid *arg_stateid,
68f74479 1054 const struct pnfs_layout_range *range,
68f74479
TM		 1055 const nfs4_stateid *stateid)
1056{
1057 struct inode *inode = lo->plh_inode;
1058 LIST_HEAD(freeme);
1059
1060 spin_lock(&inode->i_lock);
2a974425
TM		 1061 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1062 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1063 goto out_unlock;
68f74479 1064 if (stateid) {
2a974425
TM		 1065 u32 seq = be32_to_cpu(arg_stateid->seqid);
1066
68f74479
TM		 1067 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1068 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1069 pnfs_set_layout_stateid(lo, stateid, true);
1070 } else
1071 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1072out_unlock:
68f74479
TM		 1073 pnfs_clear_layoutreturn_waitbit(lo);
1074 spin_unlock(&inode->i_lock);
1075 pnfs_free_lseg_list(&freeme);
1076
1077}
1078
13c13a6a 1079static bool
e5fd1904
TM		 1080pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1081 nfs4_stateid *stateid,
1082 enum pnfs_iomode *iomode)
13c13a6a 1083{
bf0291dd
TM		 1084 /* Serialise LAYOUTGET/LAYOUTRETURN */
1085 if (atomic_read(&lo->plh_outstanding) != 0)
1086 return false;
6604b203 1087 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1088 return false;
6604b203 1089 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1090 pnfs_get_layout_hdr(lo);
e5fd1904
TM		 1091 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1092 if (stateid != NULL) {
1093 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1094 if (lo->plh_return_seq != 0)
1095 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1096 }
1097 if (iomode != NULL)
1098 *iomode = lo->plh_return_iomode;
1099 pnfs_clear_layoutreturn_info(lo);
1100 return true;
1101 }
1102 if (stateid != NULL)
1103 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1104 if (iomode != NULL)
1105 *iomode = IOMODE_ANY;
13c13a6a
TM		 1106 return true;
1107}
1108
828ed9ec
TM		 1109static void
1110pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1111 struct pnfs_layout_hdr *lo,
1112 const nfs4_stateid *stateid,
1113 enum pnfs_iomode iomode)
1114{
1115 struct inode *inode = lo->plh_inode;
1116
1117 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1118 args->inode = inode;
1119 args->range.iomode = iomode;
1120 args->range.offset = 0;
1121 args->range.length = NFS4_MAX_UINT64;
1122 args->layout = lo;
1123 nfs4_stateid_copy(&args->stateid, stateid);
1124}
1125
f40eb5d0 1126static int
ed429d6b 1127pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
6c16605d 1128 enum pnfs_iomode iomode, bool sync)
f40eb5d0
PT		 1129{
1130 struct inode *ino = lo->plh_inode;
287bd3e9 1131 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0
PT		 1132 struct nfs4_layoutreturn *lrp;
1133 int status = 0;
1134
e4af440a 1135 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1136 if (unlikely(lrp == NULL)) {
1137 status = -ENOMEM;
1138 spin_lock(&ino->i_lock);
d67ae825 1139 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0
PT		 1140 spin_unlock(&ino->i_lock);
1141 pnfs_put_layout_hdr(lo);
1142 goto out;
1143 }
1144
828ed9ec 1145 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1146 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0
PT		 1147 lrp->clp = NFS_SERVER(ino)->nfs_client;
1148 lrp->cred = lo->plh_lc_cred;
287bd3e9
TM		 1149 if (ld->prepare_layoutreturn)
1150 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1151
6c16605d 1152 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1153out:
1154 dprintk("<-- %s status: %d\n", __func__, status);
1155 return status;
1156}
1157
13c13a6a
TM		 1158/* Return true if layoutreturn is needed */
1159static bool
1160pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1161{
1162 struct pnfs_layout_segment *s;
1163
2370abda 1164 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1165 return false;
1166
1167 /* Defer layoutreturn until all lsegs are done */
1168 list_for_each_entry(s, &lo->plh_segs, pls_list) {
1169 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1170 return false;
1171 }
1172
1173 return true;
1174}
1175
1176static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1177{
1178 struct inode *inode= lo->plh_inode;
1179
2370abda 1180 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1181 return;
1182 spin_lock(&inode->i_lock);
1183 if (pnfs_layout_need_return(lo)) {
1184 nfs4_stateid stateid;
1185 enum pnfs_iomode iomode;
1186 bool send;
1187
e5fd1904 1188 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
13c13a6a
TM		 1189 spin_unlock(&inode->i_lock);
1190 if (send) {
1191 /* Send an async layoutreturn so we dont deadlock */
1192 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1193 }
1194 } else
1195 spin_unlock(&inode->i_lock);
1196}
1197
293b3b06
AA		 1198/*
1199 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1200 * when the layout segment list is empty.
1201 *
1202 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1203 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1204 * deviceid is marked invalid.
1205 */
cbe82603
BH		 1206int
1207_pnfs_return_layout(struct inode *ino)
1208{
1209 struct pnfs_layout_hdr *lo = NULL;
1210 struct nfs_inode *nfsi = NFS_I(ino);
1211 LIST_HEAD(tmp_list);
cbe82603 1212 nfs4_stateid stateid;
24408f52 1213 int status = 0;
7f27392c 1214 bool send;
cbe82603 1215
366d5052 1216 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1217
1218 spin_lock(&ino->i_lock);
1219 lo = nfsi->layout;
e5929f3c 1220 if (!lo) {
cbe82603 1221 spin_unlock(&ino->i_lock);
293b3b06
AA		 1222 dprintk("NFS: %s no layout to return\n", __func__);
1223 goto out;
cbe82603 1224 }
cbe82603 1225 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1226 pnfs_get_layout_hdr(lo);
24408f52
TM		 1227 /* Is there an outstanding layoutreturn ? */
1228 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1229 spin_unlock(&ino->i_lock);
1230 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1231 TASK_UNINTERRUPTIBLE))
1232 goto out_put_layout_hdr;
1233 spin_lock(&ino->i_lock);
1234 }
24956804 1235 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1236 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1237
1238 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1239 struct pnfs_layout_range range = {
1240 .iomode = IOMODE_ANY,
1241 .offset = 0,
1242 .length = NFS4_MAX_UINT64,
1243 };
1244 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1245 }
1246
293b3b06 1247 /* Don't send a LAYOUTRETURN if list was initially empty */
24408f52 1248 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
293b3b06 1249 spin_unlock(&ino->i_lock);
293b3b06 1250 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1251 goto out_put_layout_hdr;
293b3b06 1252 }
47abadef 1253
e5fd1904 1254 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603 1255 spin_unlock(&ino->i_lock);
7f27392c 1256 if (send)
ed429d6b 1257 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c 1258out_put_layout_hdr:
ee6625a9 1259 pnfs_free_lseg_list(&tmp_list);
7f27392c 1260 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1261out:
1262 dprintk("<-- %s status: %d\n", __func__, status);
1263 return status;
1264}
1265
24028672
TM		 1266int
1267pnfs_commit_and_return_layout(struct inode *inode)
1268{
1269 struct pnfs_layout_hdr *lo;
1270 int ret;
1271
1272 spin_lock(&inode->i_lock);
1273 lo = NFS_I(inode)->layout;
1274 if (lo == NULL) {
1275 spin_unlock(&inode->i_lock);
1276 return 0;
1277 }
1278 pnfs_get_layout_hdr(lo);
1279 /* Block new layoutgets and read/write to ds */
1280 lo->plh_block_lgets++;
1281 spin_unlock(&inode->i_lock);
1282 filemap_fdatawait(inode->i_mapping);
1283 ret = pnfs_layoutcommit_inode(inode, true);
1284 if (ret == 0)
1285 ret = _pnfs_return_layout(inode);
1286 spin_lock(&inode->i_lock);
1287 lo->plh_block_lgets--;
1288 spin_unlock(&inode->i_lock);
1289 pnfs_put_layout_hdr(lo);
1290 return ret;
1291}
1292
1c5bd76d
TM		 1293bool pnfs_roc(struct inode *ino,
1294 struct nfs4_layoutreturn_args *args,
1295 struct nfs4_layoutreturn_res *res,
1296 const struct rpc_cred *cred)
f7e8917a 1297{
40dd4b7a
TM		 1298 struct nfs_inode *nfsi = NFS_I(ino);
1299 struct nfs_open_context *ctx;
1300 struct nfs4_state *state;
f7e8917a 1301 struct pnfs_layout_hdr *lo;
1c5bd76d 1302 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1303 nfs4_stateid stateid;
1c5bd76d
TM		 1304 enum pnfs_iomode iomode = 0;
1305 bool layoutreturn = false, roc = false;
e71708d4 1306 bool skip_read = false;
f7e8917a 1307
1c5bd76d
TM		 1308 if (!nfs_have_layout(ino))
1309 return false;
29ade5db 1310retry:
f7e8917a 1311 spin_lock(&ino->i_lock);
40dd4b7a 1312 lo = nfsi->layout;
0cdc329e 1313 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1314 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1315 lo = NULL;
40dd4b7a 1316 goto out_noroc;
9c6376eb
TM		 1317 }
1318 pnfs_get_layout_hdr(lo);
29ade5db 1319 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db
TM		 1320 spin_unlock(&ino->i_lock);
1321 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1322 TASK_UNINTERRUPTIBLE);
1323 pnfs_put_layout_hdr(lo);
1324 goto retry;
1325 }
40dd4b7a 1326
e755d638 1327 /* no roc if we hold a delegation */
e71708d4
TM		 1328 if (nfs4_check_delegation(ino, FMODE_READ)) {
1329 if (nfs4_check_delegation(ino, FMODE_WRITE))
1330 goto out_noroc;
1331 skip_read = true;
1332 }
40dd4b7a
TM		 1333
1334 list_for_each_entry(ctx, &nfsi->open_files, list) {
1335 state = ctx->state;
e71708d4
TM		 1336 if (state == NULL)
1337 continue;
40dd4b7a 1338 /* Don't return layout if there is open file state */
e71708d4 1339 if (state->state & FMODE_WRITE)
40dd4b7a 1340 goto out_noroc;
e71708d4
TM		 1341 if (state->state & FMODE_READ)
1342 skip_read = true;
40dd4b7a
TM		 1343 }
1344
e755d638 1345
1c5bd76d 1346 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1347 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1348 continue;
e755d638 1349 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1350 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1351 continue;
1352 /*
1353 * Note: mark lseg for return so pnfs_layout_remove_lseg
1354 * doesn't invalidate the layout for us.
1355 */
1356 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1357 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1358 continue;
1359 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1360 }
1361
1c5bd76d
TM		 1362 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1363 goto out_noroc;
69820d22 1364
500d701f 1365 /* ROC in two conditions:
e755d638
PT		 1366 * 1. there are ROC lsegs
1367 * 2. we don't send layoutreturn
e755d638 1368 */
1c5bd76d
TM		 1369 /* lo ref dropped in pnfs_roc_release() */
1370 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1371 /* If the creds don't match, we can't compound the layoutreturn */
1372 if (!layoutreturn || cred != lo->plh_lc_cred)
1373 goto out_noroc;
1374
1375 roc = layoutreturn;
1376 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1377 res->lrs_present = 0;
1378 layoutreturn = false;
f7e8917a 1379
40dd4b7a 1380out_noroc:
f7e8917a 1381 spin_unlock(&ino->i_lock);
e755d638 1382 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1383 if (roc) {
1384 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1385 if (ld->prepare_layoutreturn)
1386 ld->prepare_layoutreturn(args);
9c6376eb 1387 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1388 return true;
1389 }
e755d638 1390 if (layoutreturn)
1c5bd76d 1391 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
9c6376eb 1392 pnfs_put_layout_hdr(lo);
287bd3e9 1393 return false;
f7e8917a
FI		 1394}
1395
1c5bd76d
TM		 1396void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1397 struct nfs4_layoutreturn_res *res,
1398 int ret)
f7e8917a 1399{
1c5bd76d
TM		 1400 struct pnfs_layout_hdr *lo = args->layout;
1401 const nfs4_stateid *arg_stateid = NULL;
1402 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1403 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1404
1c5bd76d
TM		 1405 if (ret == 0) {
1406 arg_stateid = &args->stateid;
1407 if (res->lrs_present)
1408 res_stateid = &res->stateid;
1409 }
1410 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1411 res_stateid);
287bd3e9
TM		 1412 if (ld_private && ld_private->ops && ld_private->ops->free)
1413 ld_private->ops->free(ld_private);
1c5bd76d
TM		 1414 pnfs_put_layout_hdr(lo);
1415 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1416}
1417
500d701f
PT		 1418bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1419{
1420 struct nfs_inode *nfsi = NFS_I(ino);
1421 struct pnfs_layout_hdr *lo;
1422 bool sleep = false;
1423
1424 /* we might not have grabbed lo reference. so need to check under
1425 * i_lock */
1426 spin_lock(&ino->i_lock);
1427 lo = nfsi->layout;
ee284e35
TM		 1428 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1429 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1430 sleep = true;
ee284e35 1431 }
500d701f 1432 spin_unlock(&ino->i_lock);
500d701f
PT		 1433 return sleep;
1434}
1435
b1f69b75
AA		 1436/*
1437 * Compare two layout segments for sorting into layout cache.
1438 * We want to preferentially return RW over RO layouts, so ensure those
1439 * are seen first.
1440 */
1441static s64
7dc0ac70 1442pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1443 const struct pnfs_layout_range *l2)
b1f69b75 1444{
fb3296eb
BH		 1445 s64 d;
1446
1447 /* high offset > low offset */
1448 d = l1->offset - l2->offset;
1449 if (d)
1450 return d;
1451
1452 /* short length > long length */
1453 d = l2->length - l1->length;
1454 if (d)
1455 return d;
1456
b1f69b75 1457 /* read > read/write */
fb3296eb 1458 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1459}
1460
03772d2f
TM		 1461static bool
1462pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1463 const struct pnfs_layout_range *l2)
1464{
1465 return pnfs_lseg_range_cmp(l1, l2) > 0;
1466}
1467
1468static bool
1469pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1470 struct pnfs_layout_segment *old)
1471{
1472 return false;
1473}
1474
1475void
1476pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1477 struct pnfs_layout_segment *lseg,
1478 bool (*is_after)(const struct pnfs_layout_range *,
1479 const struct pnfs_layout_range *),
1480 bool (*do_merge)(struct pnfs_layout_segment *,
1481 struct pnfs_layout_segment *),
1482 struct list_head *free_me)
974cec8c 1483{
03772d2f 1484 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1485
974cec8c
AA		 1486 dprintk("%s:Begin\n", __func__);
1487
03772d2f
TM		 1488 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1489 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1490 continue;
1491 if (do_merge(lseg, lp)) {
1492 mark_lseg_invalid(lp, free_me);
1493 continue;
1494 }
1495 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1496 continue;
566052c5 1497 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1498 dprintk("%s: inserted lseg %p "
1499 "iomode %d offset %llu length %llu before "
1500 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1501 __func__, lseg, lseg->pls_range.iomode,
1502 lseg->pls_range.offset, lseg->pls_range.length,
1503 lp, lp->pls_range.iomode, lp->pls_range.offset,
1504 lp->pls_range.length);
fb3296eb 1505 goto out;
974cec8c 1506 }
fb3296eb
BH		 1507 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1508 dprintk("%s: inserted lseg %p "
1509 "iomode %d offset %llu length %llu at tail\n",
1510 __func__, lseg, lseg->pls_range.iomode,
1511 lseg->pls_range.offset, lseg->pls_range.length);
1512out:
70c3bd2b 1513 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1514
1515 dprintk("%s:Return\n", __func__);
e5e94017 1516}
03772d2f
TM		 1517EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1518
1519static void
1520pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1521 struct pnfs_layout_segment *lseg,
1522 struct list_head *free_me)
1523{
1524 struct inode *inode = lo->plh_inode;
1525 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1526
1527 if (ld->add_lseg != NULL)
1528 ld->add_lseg(lo, lseg, free_me);
1529 else
1530 pnfs_generic_layout_insert_lseg(lo, lseg,
1531 pnfs_lseg_range_is_after,
1532 pnfs_lseg_no_merge,
1533 free_me);
1534}
e5e94017
BH		 1535
1536static struct pnfs_layout_hdr *
9fa40758
PT		 1537alloc_init_layout_hdr(struct inode *ino,
1538 struct nfs_open_context *ctx,
1539 gfp_t gfp_flags)
e5e94017
BH		 1540{
1541 struct pnfs_layout_hdr *lo;
1542
636fb9c8 1543 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1544 if (!lo)
1545 return NULL;
2b28a7be 1546 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1547 INIT_LIST_HEAD(&lo->plh_layouts);
1548 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1549 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1550 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1551 lo->plh_inode = ino;
5cc2216d 1552 lo->plh_lc_cred = get_rpccred(ctx->cred);
67a3b721 1553 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1554 return lo;
1555}
1556
1557static struct pnfs_layout_hdr *
9fa40758
PT		 1558pnfs_find_alloc_layout(struct inode *ino,
1559 struct nfs_open_context *ctx,
1560 gfp_t gfp_flags)
e5241e43
TM		 1561 __releases(&ino->i_lock)
1562 __acquires(&ino->i_lock)
e5e94017
BH		 1563{
1564 struct nfs_inode *nfsi = NFS_I(ino);
1565 struct pnfs_layout_hdr *new = NULL;
1566
1567 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1568
251ec410
TM		 1569 if (nfsi->layout != NULL)
1570 goto out_existing;
e5e94017 1571 spin_unlock(&ino->i_lock);
9fa40758 1572 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1573 spin_lock(&ino->i_lock);
1574
251ec410 1575 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1576 nfsi->layout = new;
251ec410 1577 return new;
7175fe90
YN		 1578 } else if (new != NULL)
1579 pnfs_free_layout_hdr(new);
251ec410
TM		 1580out_existing:
1581 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1582 return nfsi->layout;
1583}
1584
b1f69b75
AA		 1585/*
1586 * iomode matching rules:
c7d73af2
TH		 1587 * iomode lseg strict match
1588 * iomode
1589 * ----- ----- ------ -----
1590 * ANY READ N/A true
1591 * ANY RW N/A true
1592 * RW READ N/A false
1593 * RW RW N/A true
1594 * READ READ N/A true
1595 * READ RW true false
1596 * READ RW false true
b1f69b75 1597 */
3cb2df17 1598static bool
7dc0ac70 1599pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1600 const struct pnfs_layout_range *range,
1601 bool strict_iomode)
b1f69b75 1602{
fb3296eb
BH		 1603 struct pnfs_layout_range range1;
1604
1605 if ((range->iomode == IOMODE_RW &&
1606 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1607 (range->iomode != ls_range->iomode &&
6089dd0d 1608 strict_iomode) ||
7dc0ac70 1609 !pnfs_lseg_range_intersecting(ls_range, range))
fb3296eb
BH		 1610 return 0;
1611
1612 /* range1 covers only the first byte in the range */
1613 range1 = *range;
1614 range1.length = 1;
7dc0ac70 1615 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1616}
1617
1618/*
1619 * lookup range in layout
1620 */
e5e94017 1621static struct pnfs_layout_segment *
fb3296eb 1622pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1623 struct pnfs_layout_range *range,
1624 bool strict_iomode)
e5e94017 1625{
b1f69b75
AA		 1626 struct pnfs_layout_segment *lseg, *ret = NULL;
1627
1628 dprintk("%s:Begin\n", __func__);
1629
b7edfaa1 1630 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1631 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
ce6ab4f2 1632 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
c7d73af2
TH		 1633 pnfs_lseg_range_match(&lseg->pls_range, range,
1634 strict_iomode)) {
9369a431 1635 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1636 break;
1637 }
b1f69b75
AA		 1638 }
1639
1640 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1641 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1642 return ret;
e5e94017
BH		 1643}
1644
d23d61c8
AA		 1645/*
1646 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1647 * to the MDS or over pNFS
1648 *
1649 * The nfs_inode read_io and write_io fields are cumulative counters reset
1650 * when there are no layout segments. Note that in pnfs_update_layout iomode
1651 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1652 * WRITE request.
1653 *
1654 * A return of true means use MDS I/O.
1655 *
1656 * From rfc 5661:
1657 * If a file's size is smaller than the file size threshold, data accesses
1658 * SHOULD be sent to the metadata server. If an I/O request has a length that
1659 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1660 * server. If both file size and I/O size are provided, the client SHOULD
1661 * reach or exceed both thresholds before sending its read or write
1662 * requests to the data server.
1663 */
1664static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1665 struct inode *ino, int iomode)
1666{
1667 struct nfs4_threshold *t = ctx->mdsthreshold;
1668 struct nfs_inode *nfsi = NFS_I(ino);
1669 loff_t fsize = i_size_read(ino);
1670 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1671
1672 if (t == NULL)
1673 return ret;
1674
1675 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1676 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1677
1678 switch (iomode) {
1679 case IOMODE_READ:
1680 if (t->bm & THRESHOLD_RD) {
1681 dprintk("%s fsize %llu\n", __func__, fsize);
1682 size_set = true;
1683 if (fsize < t->rd_sz)
1684 size = true;
1685 }
1686 if (t->bm & THRESHOLD_RD_IO) {
1687 dprintk("%s nfsi->read_io %llu\n", __func__,
1688 nfsi->read_io);
1689 io_set = true;
1690 if (nfsi->read_io < t->rd_io_sz)
1691 io = true;
1692 }
1693 break;
1694 case IOMODE_RW:
1695 if (t->bm & THRESHOLD_WR) {
1696 dprintk("%s fsize %llu\n", __func__, fsize);
1697 size_set = true;
1698 if (fsize < t->wr_sz)
1699 size = true;
1700 }
1701 if (t->bm & THRESHOLD_WR_IO) {
1702 dprintk("%s nfsi->write_io %llu\n", __func__,
1703 nfsi->write_io);
1704 io_set = true;
1705 if (nfsi->write_io < t->wr_io_sz)
1706 io = true;
1707 }
1708 break;
1709 }
1710 if (size_set && io_set) {
1711 if (size && io)
1712 ret = true;
1713 } else if (size || io)
1714 ret = true;
1715
1716 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1717 return ret;
1718}
1719
aa8a45ee
PT		 1720static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1721{
1722 /*
1723 * send layoutcommit as it can hold up layoutreturn due to lseg
1724 * reference
1725 */
1726 pnfs_layoutcommit_inode(lo->plh_inode, false);
1727 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1728 nfs_wait_bit_killable,
aa8a45ee
PT		 1729 TASK_UNINTERRUPTIBLE);
1730}
1731
d67ae825
TH		 1732static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1733{
1734 unsigned long *bitlock = &lo->plh_flags;
1735
1736 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1737 smp_mb__after_atomic();
1738 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1739}
1740
78746a38
FI		 1741static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1742 struct nfs_server *server)
1743{
1744 if (list_empty(&lo->plh_layouts)) {
1745 struct nfs_client *clp = server->nfs_client;
1746
1747 /* The lo must be on the clp list if there is any
1748 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1749 */
1750 spin_lock(&clp->cl_lock);
1751 if (list_empty(&lo->plh_layouts))
1752 list_add_tail(&lo->plh_layouts, &server->layouts);
1753 spin_unlock(&clp->cl_lock);
1754 }
1755}
1756
e5e94017
BH		 1757/*
1758 * Layout segment is retreived from the server if not cached.
1759 * The appropriate layout segment is referenced and returned to the caller.
1760 */
7c24d948 1761struct pnfs_layout_segment *
e5e94017
BH		 1762pnfs_update_layout(struct inode *ino,
1763 struct nfs_open_context *ctx,
fb3296eb
BH		 1764 loff_t pos,
1765 u64 count,
a75b9df9 1766 enum pnfs_iomode iomode,
c7d73af2 1767 bool strict_iomode,
a75b9df9 1768 gfp_t gfp_flags)
e5e94017 1769{
fb3296eb
BH		 1770 struct pnfs_layout_range arg = {
1771 .iomode = iomode,
1772 .offset = pos,
1773 .length = count,
1774 };
70d2f7b1 1775 unsigned pg_offset;
6382a441
WAA		 1776 struct nfs_server *server = NFS_SERVER(ino);
1777 struct nfs_client *clp = server->nfs_client;
183d9e7b 1778 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1779 struct pnfs_layout_segment *lseg = NULL;
587f03de 1780 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1781 nfs4_stateid stateid;
1782 long timeout = 0;
66b53f32 1783 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1784 bool first;
e5e94017 1785
9a4bf31d 1786 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1787 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1788 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1789 goto out;
9a4bf31d 1790 }
d23d61c8 1791
9a4bf31d 1792 if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
183d9e7b 1793 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1794 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
4ae93560 1795 goto out;
9a4bf31d 1796 }
4ae93560 1797
9a4bf31d 1798 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1799 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1800 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1801 goto out;
9a4bf31d 1802 }
d23d61c8 1803
9bf87482 1804lookup_again:
b88fa69e 1805 nfs4_client_recover_expired_lease(clp);
9bf87482 1806 first = false;
e5e94017 1807 spin_lock(&ino->i_lock);
9fa40758 1808 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1809 if (lo == NULL) {
1810 spin_unlock(&ino->i_lock);
183d9e7b 1811 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1812 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1813 goto out;
1814 }
e5e94017 1815
43f1b3da 1816 /* Do we even need to bother with this? */
a59c30ac 1817 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1818 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1819 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1820 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1821 goto out_unlock;
1822 }
1823
1824 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1825 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1826 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1827 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1828 goto out_unlock;
9a4bf31d 1829 }
e5e94017 1830
c7d73af2 1831 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1832 if (lseg) {
1833 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1834 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1835 goto out_unlock;
1836 }
1837
1838 if (!nfs4_valid_open_stateid(ctx->state)) {
1839 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1840 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1841 goto out_unlock;
1842 }
1843
1844 /*
1845 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1846 * stateid, or it has been invalidated, then we must use the open
1847 * stateid.
1848 */
67a3b721 1849 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
183d9e7b
JL		 1850
1851 /*
1852 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1853 * RFC 5661 Errata 3208.
1854 */
1855 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1856 &lo->plh_flags)) {
1857 spin_unlock(&ino->i_lock);
1858 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1859 TASK_UNINTERRUPTIBLE);
1860 pnfs_put_layout_hdr(lo);
183d9e7b 1861 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1862 goto lookup_again;
1863 }
183d9e7b
JL		 1864
1865 first = true;
70d2f7b1
TM		 1866 if (nfs4_select_rw_stateid(ctx->state,
1867 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1868 NULL, &stateid, NULL) != 0) {
1869 trace_pnfs_update_layout(ino, pos, count,
1870 iomode, lo, lseg,
1871 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1872 goto out_unlock;
1873 }
9bf87482 1874 } else {
183d9e7b 1875 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 1876 }
568e8c49 1877
aa8a45ee
PT		 1878 /*
1879 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1880 * for LAYOUTRETURN even if first is true.
1881 */
8f70f53a 1882 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 1883 spin_unlock(&ino->i_lock);
1884 dprintk("%s wait for layoutreturn\n", __func__);
1885 if (pnfs_prepare_to_retry_layoutget(lo)) {
d67ae825
TH		 1886 if (first)
1887 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 1888 pnfs_put_layout_hdr(lo);
1889 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 1890 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1891 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 1892 goto lookup_again;
1893 }
183d9e7b 1894 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1895 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 1896 goto out_put_layout_hdr;
1897 }
1898
9a4bf31d 1899 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 1900 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1901 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 1902 goto out_unlock;
9a4bf31d 1903 }
cf7d63f1 1904 atomic_inc(&lo->plh_outstanding);
f49f9baa 1905 spin_unlock(&ino->i_lock);
30005121 1906
78746a38 1907 _add_to_server_list(lo, server);
e5e94017 1908
09cbfeaf 1909 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 1910 if (pg_offset) {
1911 arg.offset -= pg_offset;
1912 arg.length += pg_offset;
1913 }
7c24d948 1914 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 1915 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 1916
5e36e2a9 1917 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 1918 if (!lgp) {
1919 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
1920 PNFS_UPDATE_LAYOUT_NOMEM);
1921 atomic_dec(&lo->plh_outstanding);
1922 goto out_put_layout_hdr;
1923 }
1924
dacb452d 1925 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 1926 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1927 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
56b38a1f 1928 atomic_dec(&lo->plh_outstanding);
83026d80 1929 if (IS_ERR(lseg)) {
183d9e7b 1930 switch(PTR_ERR(lseg)) {
e85d7ee4 1931 case -EBUSY:
183d9e7b
JL		 1932 if (time_after(jiffies, giveup))
1933 lseg = NULL;
66b53f32
TM		 1934 break;
1935 case -ERECALLCONFLICT:
1936 /* Huh? We hold no layouts, how is there a recall? */
1937 if (first) {
1938 lseg = NULL;
1939 break;
1940 }
1941 /* Destroy the existing layout and start over */
1942 if (time_after(jiffies, giveup))
1943 pnfs_destroy_layout(NFS_I(ino));
183d9e7b
JL		 1944 /* Fallthrough */
1945 case -EAGAIN:
56b38a1f 1946 break;
183d9e7b
JL		 1947 default:
1948 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
1949 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1950 lseg = NULL;
1951 }
56b38a1f
TM		 1952 goto out_put_layout_hdr;
1953 }
1954 if (lseg) {
1955 if (first)
1956 pnfs_clear_first_layoutget(lo);
1957 trace_pnfs_update_layout(ino, pos, count,
1958 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
1959 pnfs_put_layout_hdr(lo);
1960 goto lookup_again;
83026d80
JL		 1961 }
1962 } else {
1963 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1964 }
1965
830ffb56 1966out_put_layout_hdr:
d67ae825
TH		 1967 if (first)
1968 pnfs_clear_first_layoutget(lo);
70c3bd2b 1969 pnfs_put_layout_hdr(lo);
e5e94017 1970out:
f86bbcf8
TM		 1971 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1972 "(%s, offset: %llu, length: %llu)\n",
1973 __func__, ino->i_sb->s_id,
1974 (unsigned long long)NFS_FILEID(ino),
d600ad1f 1975 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 1976 iomode==IOMODE_RW ? "read/write" : "read-only",
1977 (unsigned long long)pos,
1978 (unsigned long long)count);
e5e94017
BH		 1979 return lseg;
1980out_unlock:
1981 spin_unlock(&ino->i_lock);
830ffb56 1982 goto out_put_layout_hdr;
e5e94017 1983}
7c24d948 1984EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 1985
540d9864
TM		 1986static bool
1987pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
1988{
1989 switch (range->iomode) {
1990 case IOMODE_READ:
1991 case IOMODE_RW:
1992 break;
1993 default:
1994 return false;
1995 }
1996 if (range->offset == NFS4_MAX_UINT64)
1997 return false;
1998 if (range->length == 0)
1999 return false;
2000 if (range->length != NFS4_MAX_UINT64 &&
2001 range->length > NFS4_MAX_UINT64 - range->offset)
2002 return false;
2003 return true;
2004}
2005
78746a38
FI		 2006static struct pnfs_layout_hdr *
2007_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2008{
2009 struct pnfs_layout_hdr *lo;
2010
2011 spin_lock(&ino->i_lock);
2012 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2013 if (!lo)
2014 goto out_unlock;
2015 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2016 goto out_unlock;
2017 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2018 goto out_unlock;
2019 if (pnfs_layoutgets_blocked(lo))
2020 goto out_unlock;
2021 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2022 goto out_unlock;
2023 atomic_inc(&lo->plh_outstanding);
2024 spin_unlock(&ino->i_lock);
2025 _add_to_server_list(lo, NFS_SERVER(ino));
2026 return lo;
2027
2028out_unlock:
2029 spin_unlock(&ino->i_lock);
2030 pnfs_put_layout_hdr(lo);
2031 return NULL;
2032}
2033
2409a976
FI		 2034extern const nfs4_stateid current_stateid;
2035
2036static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2037 struct nfs_open_context *ctx)
2038{
78746a38
FI		 2039 struct inode *ino = data->dentry->d_inode;
2040 struct pnfs_layout_range rng = {
2041 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2042 IOMODE_RW: IOMODE_READ,
2043 .offset = 0,
2044 .length = NFS4_MAX_UINT64,
2045 };
2046 struct nfs4_layoutget *lgp;
2047 struct pnfs_layout_hdr *lo;
2048
2049 lo = _pnfs_grab_empty_layout(ino, ctx);
2050 if (!lo)
2051 return;
2052 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2053 &rng, GFP_KERNEL);
2054 if (!lgp) {
2055 pnfs_clear_first_layoutget(lo);
2056 pnfs_put_layout_hdr(lo);
2057 return;
2058 }
2059 data->lgp = lgp;
2060 data->o_arg.lg_args = &lgp->args;
2061 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2062}
2063
2064static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2065 struct nfs_open_context *ctx)
2066{
2067 struct pnfs_layout_range rng = {
2068 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2069 IOMODE_RW: IOMODE_READ,
2070 .offset = 0,
2071 .length = NFS4_MAX_UINT64,
2072 };
2073 struct nfs4_layoutget *lgp;
2074
2075 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2076 &rng, GFP_KERNEL);
2077 if (!lgp)
2078 return;
2079 data->lgp = lgp;
2080 data->o_arg.lg_args = &lgp->args;
2081 data->o_res.lg_res = &lgp->res;
2082}
2083
2084void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2085 struct nfs_open_context *ctx)
2086{
2087 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2088
2089 if (!(pnfs_enabled_sb(server) &&
2090 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2091 return;
2092 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2093 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2094 return;
2409a976
FI		 2095 if (data->state)
2096 _lgopen_prepare_attached(data, ctx);
2097 else
2098 _lgopen_prepare_floating(data, ctx);
2099}
2100
2101void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2102 struct nfs_open_context *ctx)
2103{
2104 struct pnfs_layout_hdr *lo;
2105 struct pnfs_layout_segment *lseg;
c49b5209 2106 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2107 u32 iomode;
2108
6e01260c 2109 if (!lgp)
2409a976 2110 return;
6e01260c
FI		 2111 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2112 if (lgp->res.status) {
2113 switch (lgp->res.status) {
6e01260c 2114 default:
8dc96566
TM		 2115 break;
2116 /*
2117 * Halt lgopen attempts if the server doesn't recognise
2118 * the "current stateid" value, the layout type, or the
2119 * layoutget operation as being valid.
2120 * Also if it complains about too many ops in the compound
2121 * or of the request/reply being too big.
2122 */
2123 case -NFS4ERR_BAD_STATEID:
2124 case -NFS4ERR_NOTSUPP:
2125 case -NFS4ERR_REP_TOO_BIG:
2126 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2127 case -NFS4ERR_REQ_TOO_BIG:
2128 case -NFS4ERR_TOO_MANY_OPS:
2129 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2130 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2131 }
2132 return;
2133 }
2409a976 2134 if (!lgp->args.inode) {
78746a38
FI		 2135 lo = _pnfs_grab_empty_layout(ino, ctx);
2136 if (!lo)
2137 return;
2409a976
FI		 2138 lgp->args.inode = ino;
2139 } else
2140 lo = NFS_I(lgp->args.inode)->layout;
2409a976 2141
c49b5209
FI		 2142 if (read_seqcount_retry(&srv->nfs_client->cl_callback_count,
2143 lgp->callback_count))
30ae2412 2144 return;
2409a976 2145 lseg = pnfs_layout_process(lgp);
2409a976
FI		 2146 if (IS_ERR(lseg)) {
2147 /* ignore lseg, but would like to mark not to try lgopen */
2148 /* clear some lo flags - first and fail ???? */
2149 } else {
2150 iomode = lgp->args.range.iomode;
2151 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2152 pnfs_put_lseg(lseg);
2153 }
30ae2412
FI		 2154}
2155
2156void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2157{
2158 if (lgp != NULL) {
2159 struct inode *inode = lgp->args.inode;
2160 if (inode) {
2161 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2162 atomic_dec(&lo->plh_outstanding);
2163 pnfs_clear_first_layoutget(lo);
2164 }
2165 pnfs_layoutget_free(lgp);
2166 }
2409a976
FI		 2167}
2168
a0b0a6e3 2169struct pnfs_layout_segment *
b1f69b75
AA		 2170pnfs_layout_process(struct nfs4_layoutget *lgp)
2171{
2172 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2173 struct nfs4_layoutget_res *res = &lgp->res;
2174 struct pnfs_layout_segment *lseg;
b7edfaa1 2175 struct inode *ino = lo->plh_inode;
78096cca 2176 LIST_HEAD(free_me);
540d9864
TM		 2177
2178 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2179 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2180
2181 /* Inject layout blob into I/O device driver */
a75b9df9 2182 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2183 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2184 if (!lseg)
1b3c6d07
JL		 2185 lseg = ERR_PTR(-ENOMEM);
2186
2187 dprintk("%s: Could not allocate layout: error %ld\n",
2188 __func__, PTR_ERR(lseg));
2189 return lseg;
b1f69b75
AA		 2190 }
2191
119cef97 2192 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2193
b1f69b75 2194 spin_lock(&ino->i_lock);
e1c06f80 2195 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2196 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2197 goto out_forget;
43f1b3da 2198 }
038d6493 2199
9888d837
TM		 2200 if (!pnfs_layout_is_valid(lo)) {
2201 /* We have a completely new layout */
2202 pnfs_set_layout_stateid(lo, &res->stateid, true);
2203 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2204 /* existing state ID, make sure the sequence number matches. */
2205 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2206 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2207 goto out_forget;
362f7474
CH		 2208 }
2209 pnfs_set_layout_stateid(lo, &res->stateid, false);
2210 } else {
2211 /*
2212 * We got an entirely new state ID. Mark all segments for the
9888d837 2213 * inode invalid, and retry the layoutget
362f7474 2214 */
d9b61708 2215 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 2216 goto out_forget;
362f7474 2217 }
038d6493 2218
9369a431 2219 pnfs_get_lseg(lseg);
03772d2f 2220 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2221
b1f69b75 2222
3976143b 2223 if (res->return_on_close)
f7e8917a 2224 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2225
b1f69b75 2226 spin_unlock(&ino->i_lock);
78096cca 2227 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2228 return lseg;
43f1b3da 2229
1b3c6d07 2230out_forget:
43f1b3da
FI		 2231 spin_unlock(&ino->i_lock);
2232 lseg->pls_layout = lo;
2233 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1f18b82c
TM		 2234 if (!pnfs_layout_is_valid(lo))
2235 nfs_commit_inode(ino, 0);
1b3c6d07 2236 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2237}
2238
2f215968
TM		 2239/**
2240 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2241 * @lo: pointer to layout header
2242 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2243 * @return_range: describe layout segment ranges to be returned
2244 *
2245 * This function is mainly intended for use by layoutrecall. It attempts
2246 * to free the layout segment immediately, or else to mark it for return
2247 * as soon as its reference count drops to zero.
2248 */
10335556 2249int
016256df
PT		 2250pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2251 struct list_head *tmp_list,
6d597e17
JL		 2252 const struct pnfs_layout_range *return_range,
2253 u32 seq)
016256df
PT		 2254{
2255 struct pnfs_layout_segment *lseg, *next;
10335556 2256 int remaining = 0;
016256df
PT		 2257
2258 dprintk("%s:Begin lo %p\n", __func__, lo);
2259
2260 if (list_empty(&lo->plh_segs))
10335556 2261 return 0;
016256df 2262
fc7ff367 2263 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2264
2265 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2266 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2267 dprintk("%s: marking lseg %p iomode %d "
2268 "offset %llu length %llu\n", __func__,
2269 lseg, lseg->pls_range.iomode,
2270 lseg->pls_range.offset,
2271 lseg->pls_range.length);
2f215968
TM		 2272 if (mark_lseg_invalid(lseg, tmp_list))
2273 continue;
2274 remaining++;
016256df 2275 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2276 }
6d597e17
JL		 2277
2278 if (remaining)
2279 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2280
10335556 2281 return remaining;
016256df
PT		 2282}
2283
2284void pnfs_error_mark_layout_for_return(struct inode *inode,
2285 struct pnfs_layout_segment *lseg)
2286{
2287 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2288 struct pnfs_layout_range range = {
2289 .iomode = lseg->pls_range.iomode,
2290 .offset = 0,
2291 .length = NFS4_MAX_UINT64,
2292 };
10335556 2293 bool return_now = false;
016256df
PT		 2294
2295 spin_lock(&inode->i_lock);
bdebfccd
TM		 2296 if (!pnfs_layout_is_valid(lo)) {
2297 spin_unlock(&inode->i_lock);
2298 return;
2299 }
2d6cf5ab 2300 pnfs_set_plh_return_info(lo, range.iomode, 0);
016256df
PT		 2301 /*
2302 * mark all matching lsegs so that we are sure to have no live
2303 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2304 * for how it works.
2305 */
68f74479 2306 if (!pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0)) {
10335556 2307 nfs4_stateid stateid;
e5fd1904 2308 enum pnfs_iomode iomode;
10335556 2309
e5fd1904 2310 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2311 spin_unlock(&inode->i_lock);
2312 if (return_now)
2313 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2314 } else {
2315 spin_unlock(&inode->i_lock);
2316 nfs_commit_inode(inode, 0);
2317 }
016256df
PT		 2318}
2319EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2320
b3230e80
TM		 2321void
2322pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2323{
2324 if (pgio->pg_lseg == NULL ||
2325 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2326 return;
2327 pnfs_put_lseg(pgio->pg_lseg);
2328 pgio->pg_lseg = NULL;
2329}
2330EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2331
08cb5b0f
BC		 2332/*
2333 * Check for any intersection between the request and the pgio->pg_lseg,
2334 * and if none, put this pgio->pg_lseg away.
2335 */
2336static void
2337pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2338{
2339 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2340 pnfs_put_lseg(pgio->pg_lseg);
2341 pgio->pg_lseg = NULL;
2342 }
2343}
2344
d8007d4d
TM		 2345void
2346pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2347{
1fd937bd
PT		 2348 u64 rd_size = req->wb_bytes;
2349
b3230e80 2350 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2351 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2352 if (pgio->pg_lseg == NULL) {
2353 if (pgio->pg_dreq == NULL)
2354 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2355 else
2356 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2357
2358 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2359 req->wb_context,
2360 req_offset(req),
2361 rd_size,
2362 IOMODE_READ,
c7d73af2 2363 false,
cb5d04bc 2364 GFP_KERNEL);
d600ad1f
PT		 2365 if (IS_ERR(pgio->pg_lseg)) {
2366 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2367 pgio->pg_lseg = NULL;
2368 return;
2369 }
cb5d04bc 2370 }
e885de1a
TM		 2371 /* If no lseg, fall back to read through mds */
2372 if (pgio->pg_lseg == NULL)
1f945357 2373 nfs_pageio_reset_read_mds(pgio);
e885de1a 2374
d8007d4d
TM		 2375}
2376EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2377
2378void
6296556f
PT		 2379pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2380 struct nfs_page *req, u64 wb_size)
d8007d4d 2381{
b3230e80 2382 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2383 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2384 if (pgio->pg_lseg == NULL) {
cb5d04bc
PT		 2385 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2386 req->wb_context,
2387 req_offset(req),
2388 wb_size,
2389 IOMODE_RW,
c7d73af2 2390 false,
cb5d04bc 2391 GFP_NOFS);
d600ad1f
PT		 2392 if (IS_ERR(pgio->pg_lseg)) {
2393 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2394 pgio->pg_lseg = NULL;
2395 return;
2396 }
2397 }
e885de1a
TM		 2398 /* If no lseg, fall back to write through mds */
2399 if (pgio->pg_lseg == NULL)
1f945357 2400 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2401}
2402EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2403
180bb5ec
WAA		 2404void
2405pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2406{
2407 if (desc->pg_lseg) {
2408 pnfs_put_lseg(desc->pg_lseg);
2409 desc->pg_lseg = NULL;
2410 }
2411}
2412EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2413
b4fdac1a
WAA		 2414/*
2415 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2416 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2417 */
2418size_t
a7d42ddb
WAA		 2419pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2420 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2421{
0f9c429e 2422 unsigned int size;
c5e20cb7 2423 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2424
2425 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2426 if (!size)
2427 return 0;
94ad1c80 2428
19982ba8 2429 /*
c5e20cb7
WAA		 2430 * 'size' contains the number of bytes left in the current page (up
2431 * to the original size asked for in @req->wb_bytes).
2432 *
2433 * Calculate how many bytes are left in the layout segment
2434 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2435 *
2436 * Please also note that 'end_offset' is actually the offset of the
2437 * first byte that lies outside the pnfs_layout_range. FIXME?
2438 *
2439 */
19b54848 2440 if (pgio->pg_lseg) {
17822b20 2441 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2442 pgio->pg_lseg->pls_range.length);
2443 req_start = req_offset(req);
08cb5b0f 2444
c5e20cb7 2445 /* start of request is past the last byte of this segment */
08cb5b0f 2446 if (req_start >= seg_end)
19b54848 2447 return 0;
c5e20cb7
WAA		 2448
2449 /* adjust 'size' iff there are fewer bytes left in the
2450 * segment than what nfs_generic_pg_test returned */
2451 seg_left = seg_end - req_start;
2452 if (seg_left < size)
2453 size = (unsigned int)seg_left;
19b54848 2454 }
0f9c429e 2455
19b54848 2456 return size;
94ad1c80 2457}
89a58e32 2458EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2459
53113ad3 2460int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2461{
2462 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2463
2464 /* Resend all requests through the MDS */
53113ad3
WAA		 2465 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2466 hdr->completion_ops);
c7070113 2467 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2468 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2469}
e7dd79af 2470EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2471
d45f60c6 2472static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2473{
cd841605
FI		 2474
2475 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2476 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2477 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2478 pnfs_return_layout(hdr->inode);
1acbbb4e 2479 }
6c75dc0d 2480 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2481 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2482}
2483
d20581aa
BH		 2484/*
2485 * Called by non rpc-based layout drivers
2486 */
d45f60c6 2487void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2488{
f8417b48 2489 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2490 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2491 hdr->mds_offset + hdr->res.count);
d45f60c6 2492 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2493 }
2494 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2495 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2496 pnfs_ld_handle_write_error(hdr);
2497 hdr->mds_ops->rpc_release(hdr);
44b83799 2498}
d20581aa 2499EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2500
dce81290
TM		 2501static void
2502pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2503 struct nfs_pgio_header *hdr)
dce81290 2504{
48d635f1 2505 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2506
6c75dc0d 2507 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2508 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2509 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2510 mirror->pg_recoalesce = 1;
6c75dc0d 2511 }
ba4a76f7 2512 hdr->completion_ops->completion(hdr);
dce81290
TM		 2513}
2514
2515static enum pnfs_try_status
d45f60c6 2516pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2517 const struct rpc_call_ops *call_ops,
2518 struct pnfs_layout_segment *lseg,
2519 int how)
0382b744 2520{
cd841605 2521 struct inode *inode = hdr->inode;
0382b744
AA		 2522 enum pnfs_try_status trypnfs;
2523 struct nfs_server *nfss = NFS_SERVER(inode);
2524
cd841605 2525 hdr->mds_ops = call_ops;
0382b744
AA		 2526
2527 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2528 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2529 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2530 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2531 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2532 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2533 return trypnfs;
2534}
2535
dce81290 2536static void
7f714720
WAA		 2537pnfs_do_write(struct nfs_pageio_descriptor *desc,
2538 struct nfs_pgio_header *hdr, int how)
dce81290 2539{
dce81290
TM		 2540 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2541 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2542 enum pnfs_try_status trypnfs;
dce81290 2543
d45f60c6 2544 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2545 switch (trypnfs) {
2546 case PNFS_NOT_ATTEMPTED:
d45f60c6 2547 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2548 case PNFS_ATTEMPTED:
2549 break;
2550 case PNFS_TRY_AGAIN:
2551 /* cleanup hdr and prepare to redo pnfs */
2552 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2553 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2554 list_splice_init(&hdr->pages, &mirror->pg_list);
2555 mirror->pg_recoalesce = 1;
2556 }
2557 hdr->mds_ops->rpc_release(hdr);
2558 }
dce81290
TM		 2559}
2560
6c75dc0d
FI		 2561static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2562{
9369a431 2563 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2564 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2565}
2566
dce81290
TM		 2567int
2568pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2569{
6c75dc0d 2570 struct nfs_pgio_header *hdr;
dce81290
TM		 2571 int ret;
2572
1e7f3a48
WAA		 2573 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2574 if (!hdr) {
2bff2288
PT		 2575 desc->pg_error = -ENOMEM;
2576 return desc->pg_error;
dce81290 2577 }
6c75dc0d 2578 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2579
9369a431 2580 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2581 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2582 if (!ret)
7f714720 2583 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2584
6c75dc0d 2585 return ret;
dce81290
TM		 2586}
2587EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2588
53113ad3 2589int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2590{
2591 struct nfs_pageio_descriptor pgio;
2592
1acbbb4e 2593 /* Resend all requests through the MDS */
53113ad3
WAA		 2594 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2595 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2596}
e7dd79af 2597EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2598
d45f60c6 2599static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2600{
cd841605
FI		 2601 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2602 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2603 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2604 pnfs_return_layout(hdr->inode);
1acbbb4e 2605 }
4db6e0b7 2606 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2607 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2608}
2609
d20581aa
BH		 2610/*
2611 * Called by non rpc-based layout drivers
2612 */
d45f60c6 2613void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2614{
bfc505de 2615 if (likely(!hdr->pnfs_error))
d45f60c6 2616 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2617 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2618 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2619 pnfs_ld_handle_read_error(hdr);
2620 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2621}
2622EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2623
493292dd
TM		 2624static void
2625pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2626 struct nfs_pgio_header *hdr)
493292dd 2627{
48d635f1 2628 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2629
4db6e0b7 2630 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2631 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2632 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2633 mirror->pg_recoalesce = 1;
4db6e0b7 2634 }
ba4a76f7 2635 hdr->completion_ops->completion(hdr);
493292dd
TM		 2636}
2637
64419a9b
AA		 2638/*
2639 * Call the appropriate parallel I/O subsystem read function.
2640 */
493292dd 2641static enum pnfs_try_status
d45f60c6 2642pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2643 const struct rpc_call_ops *call_ops,
2644 struct pnfs_layout_segment *lseg)
64419a9b 2645{
cd841605 2646 struct inode *inode = hdr->inode;
64419a9b
AA		 2647 struct nfs_server *nfss = NFS_SERVER(inode);
2648 enum pnfs_try_status trypnfs;
2649
cd841605 2650 hdr->mds_ops = call_ops;
64419a9b
AA		 2651
2652 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2653 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2654
d45f60c6 2655 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2656 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2657 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2658 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2659 return trypnfs;
2660}
863a3c6c 2661
ceb11e13 2662/* Resend all requests through pnfs. */
1b1bc66b 2663void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2664{
2665 struct nfs_pageio_descriptor pgio;
2666
1b1bc66b 2667 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2668 /* Prevent deadlocks with layoutreturn! */
2669 pnfs_put_lseg(hdr->lseg);
2670 hdr->lseg = NULL;
2671
1b1bc66b
WAA		 2672 nfs_pageio_init_read(&pgio, hdr->inode, false,
2673 hdr->completion_ops);
2674 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2675 }
ceb11e13
PT		 2676}
2677EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2678
493292dd 2679static void
7f714720 2680pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2681{
493292dd
TM		 2682 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2683 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2684 enum pnfs_try_status trypnfs;
493292dd 2685
d45f60c6 2686 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2687 switch (trypnfs) {
2688 case PNFS_NOT_ATTEMPTED:
d45f60c6 2689 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2690 case PNFS_ATTEMPTED:
2691 break;
2692 case PNFS_TRY_AGAIN:
2693 /* cleanup hdr and prepare to redo pnfs */
2694 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2695 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2696 list_splice_init(&hdr->pages, &mirror->pg_list);
2697 mirror->pg_recoalesce = 1;
2698 }
2699 hdr->mds_ops->rpc_release(hdr);
2700 }
493292dd
TM		 2701}
2702
4db6e0b7
FI		 2703static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2704{
9369a431 2705 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2706 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2707}
2708
493292dd
TM		 2709int
2710pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2711{
4db6e0b7 2712 struct nfs_pgio_header *hdr;
493292dd
TM		 2713 int ret;
2714
1e7f3a48
WAA		 2715 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2716 if (!hdr) {
2bff2288
PT		 2717 desc->pg_error = -ENOMEM;
2718 return desc->pg_error;
493292dd 2719 }
4db6e0b7 2720 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2721 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2722 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2723 if (!ret)
7f714720 2724 pnfs_do_read(desc, hdr);
4db6e0b7 2725 return ret;
493292dd
TM		 2726}
2727EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2728
71244d9b
TM		 2729static void pnfs_clear_layoutcommitting(struct inode *inode)
2730{
2731 unsigned long *bitlock = &NFS_I(inode)->flags;
2732
2733 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2734 smp_mb__after_atomic();
71244d9b
TM		 2735 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2736}
2737
863a3c6c 2738/*
a9bae566 2739 * There can be multiple RW segments.
863a3c6c 2740 */
a9bae566 2741static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2742{
a9bae566 2743 struct pnfs_layout_segment *lseg;
863a3c6c 2744
a9bae566
PT		 2745 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2746 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2747 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2748 list_add(&lseg->pls_lc_list, listp);
2749 }
863a3c6c
AA		 2750}
2751
a073dbff
TM		 2752static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2753{
2754 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2755
2756 /* Matched by references in pnfs_set_layoutcommit */
2757 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2758 list_del_init(&lseg->pls_lc_list);
2759 pnfs_put_lseg(lseg);
2760 }
2761
71244d9b 2762 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2763}
2764
1b0ae068
PT		 2765void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2766{
b9e028fd 2767 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2768}
2769EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2770
863a3c6c 2771void
67af7611
TM		 2772pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2773 loff_t end_pos)
863a3c6c 2774{
cd841605 2775 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2776 bool mark_as_dirty = false;
863a3c6c 2777
cd841605 2778 spin_lock(&inode->i_lock);
863a3c6c 2779 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2780 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2781 mark_as_dirty = true;
863a3c6c 2782 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2783 __func__, inode->i_ino);
29559b11
TM		 2784 } else if (end_pos > nfsi->layout->plh_lwb)
2785 nfsi->layout->plh_lwb = end_pos;
67af7611 2786 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2787 /* references matched in nfs4_layoutcommit_release */
67af7611 2788 pnfs_get_lseg(lseg);
a9bae566 2789 }
cd841605 2790 spin_unlock(&inode->i_lock);
acff5880 2791 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2792 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2793
2794 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2795 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2796 if (mark_as_dirty)
cd841605 2797 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2798}
2799EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2800
db29c089
AA		 2801void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2802{
2803 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2804
2805 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2806 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2807 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2808}
2809
de4b15c7
AA		 2810/*
2811 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2812 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2813 * data to disk to allow the server to recover the data if it crashes.
2814 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2815 * is off, and a COMMIT is sent to a data server, or
2816 * if WRITEs to a data server return NFS_DATA_SYNC.
2817 */
863a3c6c 2818int
ef311537 2819pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2820{
5f919c9f 2821 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2822 struct nfs4_layoutcommit_data *data;
2823 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2824 loff_t end_pos;
71244d9b 2825 int status;
863a3c6c 2826
71244d9b 2827 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2828 return 0;
2829
71244d9b 2830 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2831
71244d9b 2832 status = -EAGAIN;
92407e75 2833 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2834 if (!sync)
2835 goto out;
74316201 2836 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2837 NFS_INO_LAYOUTCOMMITTING,
2838 nfs_wait_bit_killable,
2839 TASK_KILLABLE);
92407e75 2840 if (status)
71244d9b 2841 goto out;
92407e75
PT		 2842 }
2843
71244d9b
TM		 2844 status = -ENOMEM;
2845 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2846 data = kzalloc(sizeof(*data), GFP_NOFS);
2847 if (!data)
2848 goto clear_layoutcommitting;
2849
2850 status = 0;
de4b15c7 2851 spin_lock(&inode->i_lock);
71244d9b
TM		 2852 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2853 goto out_unlock;
a9bae566 2854
71244d9b 2855 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 2856 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 2857
acff5880 2858 end_pos = nfsi->layout->plh_lwb;
863a3c6c 2859
f597c537 2860 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 2861 spin_unlock(&inode->i_lock);
2862
2863 data->args.inode = inode;
9fa40758 2864 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 2865 nfs_fattr_init(&data->fattr);
2866 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2867 data->res.fattr = &data->fattr;
2e18d4d8
TM		 2868 if (end_pos != 0)
2869 data->args.lastbytewritten = end_pos - 1;
2870 else
2871 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 2872 data->res.server = NFS_SERVER(inode);
2873
5f919c9f
CH		 2874 if (ld->prepare_layoutcommit) {
2875 status = ld->prepare_layoutcommit(&data->args);
2876 if (status) {
3471648a 2877 put_rpccred(data->cred);
5f919c9f 2878 spin_lock(&inode->i_lock);
29559b11
TM		 2879 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2880 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 2881 nfsi->layout->plh_lwb = end_pos;
3471648a 2882 goto out_unlock;
5f919c9f
CH		 2883 }
2884 }
2885
2886
863a3c6c
AA		 2887 status = nfs4_proc_layoutcommit(data, sync);
2888out:
92407e75
PT		 2889 if (status)
2890 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2891 dprintk("<-- %s status %d\n", __func__, status);
2892 return status;
71244d9b
TM		 2893out_unlock:
2894 spin_unlock(&inode->i_lock);
92407e75 2895 kfree(data);
71244d9b
TM		 2896clear_layoutcommitting:
2897 pnfs_clear_layoutcommitting(inode);
92407e75 2898 goto out;
863a3c6c 2899}
72cff449 2900EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 2901
5bb89b47
TM		 2902int
2903pnfs_generic_sync(struct inode *inode, bool datasync)
2904{
2905 return pnfs_layoutcommit_inode(inode, true);
2906}
2907EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2908
82be417a
AA		 2909struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2910{
2911 struct nfs4_threshold *thp;
2912
2913 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2914 if (!thp) {
2915 dprintk("%s mdsthreshold allocation failed\n", __func__);
2916 return NULL;
2917 }
2918 return thp;
2919}
8733408d 2920
865a7ecb 2921#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 2922int
c8ad8894 2923pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 2924{
2925 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2926 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 2927 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 2928 struct nfs42_layoutstat_data *data;
2929 struct pnfs_layout_hdr *hdr;
2930 int status = 0;
2931
2932 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2933 goto out;
2934
6c5a0d89
TM		 2935 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2936 goto out;
2937
1bfe3b25
PT		 2938 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2939 goto out;
2940
8733408d
PT		 2941 spin_lock(&inode->i_lock);
2942 if (!NFS_I(inode)->layout) {
2943 spin_unlock(&inode->i_lock);
f538d0ba 2944 goto out_clear_layoutstats;
8733408d
PT		 2945 }
2946 hdr = NFS_I(inode)->layout;
2947 pnfs_get_layout_hdr(hdr);
2948 spin_unlock(&inode->i_lock);
2949
c8ad8894 2950 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 2951 if (!data) {
2952 status = -ENOMEM;
2953 goto out_put;
2954 }
2955
2956 data->args.fh = NFS_FH(inode);
2957 data->args.inode = inode;
8733408d
PT		 2958 status = ld->prepare_layoutstats(&data->args);
2959 if (status)
2960 goto out_free;
2961
2962 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2963
2964out:
2965 dprintk("%s returns %d\n", __func__, status);
2966 return status;
2967
2968out_free:
2969 kfree(data);
2970out_put:
2971 pnfs_put_layout_hdr(hdr);
f538d0ba 2972out_clear_layoutstats:
1bfe3b25
PT		 2973 smp_mb__before_atomic();
2974 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2975 smp_mb__after_atomic();
8733408d
PT		 2976 goto out;
2977}
2978EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 2979#endif
bbf58bf3
TM		 2980
2981unsigned int layoutstats_timer;
2982module_param(layoutstats_timer, uint, 0644);
2983EXPORT_SYMBOL_GPL(layoutstats_timer);
Linux 6.x block layer and io_uring tree(s)