]> Git Repo - linux.git/blame - fs/nfs/pnfs.c
projects / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branches 'acpica' and 'acpi-property'
[linux.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <[email protected]>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
95void
96unset_pnfs_layoutdriver(struct nfs_server *nfss)
97{
738fd0f3
BH		 98 if (nfss->pnfs_curr_ld) {
99 if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 104 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 105 }
85e174ba
RL		 106 nfss->pnfs_curr_ld = NULL;
107}
108
ca440c38
JL		 109/*
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
112 *
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
115 */
116static const u32 ld_prefs[] = {
117 LAYOUT_SCSI,
118 LAYOUT_BLOCK_VOLUME,
119 LAYOUT_OSD2_OBJECTS,
120 LAYOUT_FLEX_FILES,
121 LAYOUT_NFSV4_1_FILES,
122 0
123};
124
125static int
126ld_cmp(const void *e1, const void *e2)
127{
128 u32 ld1 = *((u32 *)e1);
129 u32 ld2 = *((u32 *)e2);
130 int i;
131
132 for (i = 0; ld_prefs[i] != 0; i++) {
133 if (ld1 == ld_prefs[i])
134 return -1;
135
136 if (ld2 == ld_prefs[i])
137 return 1;
138 }
139 return 0;
140}
141
85e174ba
RL		 142/*
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
145 *
3132e49e 146 * @ids array of layout types supported by MDS.
85e174ba
RL		 147 */
148void
738fd0f3 149set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 150 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 151{
152 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 153 u32 id;
ca440c38 154 int i;
85e174ba 155
19274716
AS		 156 if (fsinfo->nlayouttypes == 0)
157 goto out_no_driver;
85e174ba
RL		 158 if (!(server->nfs_client->cl_exchange_flags &
159 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 160 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 162 goto out_no_driver;
163 }
3132e49e 164
ca440c38
JL		 165 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 167
ca440c38
JL		 168 for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 id = fsinfo->layouttype[i];
85e174ba 170 ld_type = find_pnfs_driver(id);
ca440c38
JL		 171 if (!ld_type) {
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 id);
174 ld_type = find_pnfs_driver(id);
175 }
176 if (ld_type)
177 break;
85e174ba 178 }
3132e49e
JL		 179
180 if (!ld_type) {
ca440c38 181 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 182 goto out_no_driver;
183 }
184
85e174ba 185 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 186 if (ld_type->set_layoutdriver
187 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 188 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__, id);
738fd0f3
BH		 190 module_put(ld_type->owner);
191 goto out_no_driver;
192 }
2a4c8994
TM		 193 /* Bump the MDS count */
194 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 195
85e174ba
RL		 196 dprintk("%s: pNFS module for %u set\n", __func__, id);
197 return;
198
199out_no_driver:
200 dprintk("%s: Using NFSv4 I/O\n", __func__);
201 server->pnfs_curr_ld = NULL;
202}
02c35fca
FI		 203
204int
205pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206{
207 int status = -EINVAL;
208 struct pnfs_layoutdriver_type *tmp;
209
210 if (ld_type->id == 0) {
a030889a 211 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 212 return status;
213 }
b1f69b75 214 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 215 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 216 "alloc_lseg and free_lseg.\n", __func__);
217 return status;
218 }
02c35fca
FI		 219
220 spin_lock(&pnfs_spinlock);
221 tmp = find_pnfs_driver_locked(ld_type->id);
222 if (!tmp) {
223 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 status = 0;
225 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 ld_type->name);
227 } else {
a030889a 228 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 229 __func__, ld_type->id);
230 }
231 spin_unlock(&pnfs_spinlock);
232
233 return status;
234}
235EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236
237void
238pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239{
240 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 spin_lock(&pnfs_spinlock);
242 list_del(&ld_type->pnfs_tblid);
243 spin_unlock(&pnfs_spinlock);
244}
245EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 246
b1f69b75
AA		 247/*
248 * pNFS client layout cache
249 */
250
cc6e5340 251/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 252void
70c3bd2b 253pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 254{
2b28a7be 255 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 256}
257
636fb9c8
BH		 258static struct pnfs_layout_hdr *
259pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260{
261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 262 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 263}
264
265static void
266pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267{
9c626381
TM		 268 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270
271 if (!list_empty(&lo->plh_layouts)) {
272 struct nfs_client *clp = server->nfs_client;
273
274 spin_lock(&clp->cl_lock);
275 list_del_init(&lo->plh_layouts);
276 spin_unlock(&clp->cl_lock);
277 }
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;
93b7f7ad 1214 bool send, valid_layout;
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 }
93b7f7ad 1235 valid_layout = pnfs_layout_is_valid(lo);
24956804 1236 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1237 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1238
1239 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1240 struct pnfs_layout_range range = {
1241 .iomode = IOMODE_ANY,
1242 .offset = 0,
1243 .length = NFS4_MAX_UINT64,
1244 };
1245 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1246 }
1247
293b3b06 1248 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1249 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1250 !valid_layout) {
293b3b06 1251 spin_unlock(&ino->i_lock);
293b3b06 1252 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1253 goto out_put_layout_hdr;
293b3b06 1254 }
47abadef 1255
e5fd1904 1256 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603 1257 spin_unlock(&ino->i_lock);
7f27392c 1258 if (send)
ed429d6b 1259 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c 1260out_put_layout_hdr:
ee6625a9 1261 pnfs_free_lseg_list(&tmp_list);
7f27392c 1262 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1263out:
1264 dprintk("<-- %s status: %d\n", __func__, status);
1265 return status;
1266}
1267
24028672
TM		 1268int
1269pnfs_commit_and_return_layout(struct inode *inode)
1270{
1271 struct pnfs_layout_hdr *lo;
1272 int ret;
1273
1274 spin_lock(&inode->i_lock);
1275 lo = NFS_I(inode)->layout;
1276 if (lo == NULL) {
1277 spin_unlock(&inode->i_lock);
1278 return 0;
1279 }
1280 pnfs_get_layout_hdr(lo);
1281 /* Block new layoutgets and read/write to ds */
1282 lo->plh_block_lgets++;
1283 spin_unlock(&inode->i_lock);
1284 filemap_fdatawait(inode->i_mapping);
1285 ret = pnfs_layoutcommit_inode(inode, true);
1286 if (ret == 0)
1287 ret = _pnfs_return_layout(inode);
1288 spin_lock(&inode->i_lock);
1289 lo->plh_block_lgets--;
1290 spin_unlock(&inode->i_lock);
1291 pnfs_put_layout_hdr(lo);
1292 return ret;
1293}
1294
1c5bd76d
TM		 1295bool pnfs_roc(struct inode *ino,
1296 struct nfs4_layoutreturn_args *args,
1297 struct nfs4_layoutreturn_res *res,
1298 const struct rpc_cred *cred)
f7e8917a 1299{
40dd4b7a
TM		 1300 struct nfs_inode *nfsi = NFS_I(ino);
1301 struct nfs_open_context *ctx;
1302 struct nfs4_state *state;
f7e8917a 1303 struct pnfs_layout_hdr *lo;
1c5bd76d 1304 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1305 nfs4_stateid stateid;
1c5bd76d
TM		 1306 enum pnfs_iomode iomode = 0;
1307 bool layoutreturn = false, roc = false;
e71708d4 1308 bool skip_read = false;
f7e8917a 1309
1c5bd76d
TM		 1310 if (!nfs_have_layout(ino))
1311 return false;
29ade5db 1312retry:
f7e8917a 1313 spin_lock(&ino->i_lock);
40dd4b7a 1314 lo = nfsi->layout;
0cdc329e 1315 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1316 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1317 lo = NULL;
40dd4b7a 1318 goto out_noroc;
9c6376eb
TM		 1319 }
1320 pnfs_get_layout_hdr(lo);
29ade5db 1321 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db
TM		 1322 spin_unlock(&ino->i_lock);
1323 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1324 TASK_UNINTERRUPTIBLE);
1325 pnfs_put_layout_hdr(lo);
1326 goto retry;
1327 }
40dd4b7a 1328
e755d638 1329 /* no roc if we hold a delegation */
e71708d4
TM		 1330 if (nfs4_check_delegation(ino, FMODE_READ)) {
1331 if (nfs4_check_delegation(ino, FMODE_WRITE))
1332 goto out_noroc;
1333 skip_read = true;
1334 }
40dd4b7a
TM		 1335
1336 list_for_each_entry(ctx, &nfsi->open_files, list) {
1337 state = ctx->state;
e71708d4
TM		 1338 if (state == NULL)
1339 continue;
40dd4b7a 1340 /* Don't return layout if there is open file state */
e71708d4 1341 if (state->state & FMODE_WRITE)
40dd4b7a 1342 goto out_noroc;
e71708d4
TM		 1343 if (state->state & FMODE_READ)
1344 skip_read = true;
40dd4b7a
TM		 1345 }
1346
e755d638 1347
1c5bd76d 1348 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1349 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1350 continue;
e755d638 1351 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1352 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1353 continue;
1354 /*
1355 * Note: mark lseg for return so pnfs_layout_remove_lseg
1356 * doesn't invalidate the layout for us.
1357 */
1358 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1359 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1360 continue;
1361 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1362 }
1363
1c5bd76d
TM		 1364 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1365 goto out_noroc;
69820d22 1366
500d701f 1367 /* ROC in two conditions:
e755d638
PT		 1368 * 1. there are ROC lsegs
1369 * 2. we don't send layoutreturn
e755d638 1370 */
1c5bd76d
TM		 1371 /* lo ref dropped in pnfs_roc_release() */
1372 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1373 /* If the creds don't match, we can't compound the layoutreturn */
1374 if (!layoutreturn || cred != lo->plh_lc_cred)
1375 goto out_noroc;
1376
1377 roc = layoutreturn;
1378 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1379 res->lrs_present = 0;
1380 layoutreturn = false;
f7e8917a 1381
40dd4b7a 1382out_noroc:
f7e8917a 1383 spin_unlock(&ino->i_lock);
e755d638 1384 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1385 if (roc) {
1386 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1387 if (ld->prepare_layoutreturn)
1388 ld->prepare_layoutreturn(args);
9c6376eb 1389 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1390 return true;
1391 }
e755d638 1392 if (layoutreturn)
1c5bd76d 1393 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
9c6376eb 1394 pnfs_put_layout_hdr(lo);
287bd3e9 1395 return false;
f7e8917a
FI		 1396}
1397
1c5bd76d
TM		 1398void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1399 struct nfs4_layoutreturn_res *res,
1400 int ret)
f7e8917a 1401{
1c5bd76d
TM		 1402 struct pnfs_layout_hdr *lo = args->layout;
1403 const nfs4_stateid *arg_stateid = NULL;
1404 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1405 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1406
1c5bd76d
TM		 1407 if (ret == 0) {
1408 arg_stateid = &args->stateid;
1409 if (res->lrs_present)
1410 res_stateid = &res->stateid;
1411 }
1412 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1413 res_stateid);
287bd3e9
TM		 1414 if (ld_private && ld_private->ops && ld_private->ops->free)
1415 ld_private->ops->free(ld_private);
1c5bd76d
TM		 1416 pnfs_put_layout_hdr(lo);
1417 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1418}
1419
500d701f
PT		 1420bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1421{
1422 struct nfs_inode *nfsi = NFS_I(ino);
1423 struct pnfs_layout_hdr *lo;
1424 bool sleep = false;
1425
1426 /* we might not have grabbed lo reference. so need to check under
1427 * i_lock */
1428 spin_lock(&ino->i_lock);
1429 lo = nfsi->layout;
ee284e35
TM		 1430 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1431 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1432 sleep = true;
ee284e35 1433 }
500d701f 1434 spin_unlock(&ino->i_lock);
500d701f
PT		 1435 return sleep;
1436}
1437
b1f69b75
AA		 1438/*
1439 * Compare two layout segments for sorting into layout cache.
1440 * We want to preferentially return RW over RO layouts, so ensure those
1441 * are seen first.
1442 */
1443static s64
7dc0ac70 1444pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1445 const struct pnfs_layout_range *l2)
b1f69b75 1446{
fb3296eb
BH		 1447 s64 d;
1448
1449 /* high offset > low offset */
1450 d = l1->offset - l2->offset;
1451 if (d)
1452 return d;
1453
1454 /* short length > long length */
1455 d = l2->length - l1->length;
1456 if (d)
1457 return d;
1458
b1f69b75 1459 /* read > read/write */
fb3296eb 1460 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1461}
1462
03772d2f
TM		 1463static bool
1464pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1465 const struct pnfs_layout_range *l2)
1466{
1467 return pnfs_lseg_range_cmp(l1, l2) > 0;
1468}
1469
1470static bool
1471pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1472 struct pnfs_layout_segment *old)
1473{
1474 return false;
1475}
1476
1477void
1478pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1479 struct pnfs_layout_segment *lseg,
1480 bool (*is_after)(const struct pnfs_layout_range *,
1481 const struct pnfs_layout_range *),
1482 bool (*do_merge)(struct pnfs_layout_segment *,
1483 struct pnfs_layout_segment *),
1484 struct list_head *free_me)
974cec8c 1485{
03772d2f 1486 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1487
974cec8c
AA		 1488 dprintk("%s:Begin\n", __func__);
1489
03772d2f
TM		 1490 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1491 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1492 continue;
1493 if (do_merge(lseg, lp)) {
1494 mark_lseg_invalid(lp, free_me);
1495 continue;
1496 }
1497 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1498 continue;
566052c5 1499 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1500 dprintk("%s: inserted lseg %p "
1501 "iomode %d offset %llu length %llu before "
1502 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1503 __func__, lseg, lseg->pls_range.iomode,
1504 lseg->pls_range.offset, lseg->pls_range.length,
1505 lp, lp->pls_range.iomode, lp->pls_range.offset,
1506 lp->pls_range.length);
fb3296eb 1507 goto out;
974cec8c 1508 }
fb3296eb
BH		 1509 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1510 dprintk("%s: inserted lseg %p "
1511 "iomode %d offset %llu length %llu at tail\n",
1512 __func__, lseg, lseg->pls_range.iomode,
1513 lseg->pls_range.offset, lseg->pls_range.length);
1514out:
70c3bd2b 1515 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1516
1517 dprintk("%s:Return\n", __func__);
e5e94017 1518}
03772d2f
TM		 1519EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1520
1521static void
1522pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1523 struct pnfs_layout_segment *lseg,
1524 struct list_head *free_me)
1525{
1526 struct inode *inode = lo->plh_inode;
1527 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1528
1529 if (ld->add_lseg != NULL)
1530 ld->add_lseg(lo, lseg, free_me);
1531 else
1532 pnfs_generic_layout_insert_lseg(lo, lseg,
1533 pnfs_lseg_range_is_after,
1534 pnfs_lseg_no_merge,
1535 free_me);
1536}
e5e94017
BH		 1537
1538static struct pnfs_layout_hdr *
9fa40758
PT		 1539alloc_init_layout_hdr(struct inode *ino,
1540 struct nfs_open_context *ctx,
1541 gfp_t gfp_flags)
e5e94017
BH		 1542{
1543 struct pnfs_layout_hdr *lo;
1544
636fb9c8 1545 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1546 if (!lo)
1547 return NULL;
2b28a7be 1548 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1549 INIT_LIST_HEAD(&lo->plh_layouts);
1550 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1551 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1552 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1553 lo->plh_inode = ino;
5cc2216d 1554 lo->plh_lc_cred = get_rpccred(ctx->cred);
67a3b721 1555 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1556 return lo;
1557}
1558
1559static struct pnfs_layout_hdr *
9fa40758
PT		 1560pnfs_find_alloc_layout(struct inode *ino,
1561 struct nfs_open_context *ctx,
1562 gfp_t gfp_flags)
e5241e43
TM		 1563 __releases(&ino->i_lock)
1564 __acquires(&ino->i_lock)
e5e94017
BH		 1565{
1566 struct nfs_inode *nfsi = NFS_I(ino);
1567 struct pnfs_layout_hdr *new = NULL;
1568
1569 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1570
251ec410
TM		 1571 if (nfsi->layout != NULL)
1572 goto out_existing;
e5e94017 1573 spin_unlock(&ino->i_lock);
9fa40758 1574 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1575 spin_lock(&ino->i_lock);
1576
251ec410 1577 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1578 nfsi->layout = new;
251ec410 1579 return new;
7175fe90
YN		 1580 } else if (new != NULL)
1581 pnfs_free_layout_hdr(new);
251ec410
TM		 1582out_existing:
1583 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1584 return nfsi->layout;
1585}
1586
b1f69b75
AA		 1587/*
1588 * iomode matching rules:
c7d73af2
TH		 1589 * iomode lseg strict match
1590 * iomode
1591 * ----- ----- ------ -----
1592 * ANY READ N/A true
1593 * ANY RW N/A true
1594 * RW READ N/A false
1595 * RW RW N/A true
1596 * READ READ N/A true
1597 * READ RW true false
1598 * READ RW false true
b1f69b75 1599 */
3cb2df17 1600static bool
7dc0ac70 1601pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1602 const struct pnfs_layout_range *range,
1603 bool strict_iomode)
b1f69b75 1604{
fb3296eb
BH		 1605 struct pnfs_layout_range range1;
1606
1607 if ((range->iomode == IOMODE_RW &&
1608 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1609 (range->iomode != ls_range->iomode &&
6089dd0d 1610 strict_iomode) ||
7dc0ac70 1611 !pnfs_lseg_range_intersecting(ls_range, range))
fb3296eb
BH		 1612 return 0;
1613
1614 /* range1 covers only the first byte in the range */
1615 range1 = *range;
1616 range1.length = 1;
7dc0ac70 1617 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1618}
1619
1620/*
1621 * lookup range in layout
1622 */
e5e94017 1623static struct pnfs_layout_segment *
fb3296eb 1624pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1625 struct pnfs_layout_range *range,
1626 bool strict_iomode)
e5e94017 1627{
b1f69b75
AA		 1628 struct pnfs_layout_segment *lseg, *ret = NULL;
1629
1630 dprintk("%s:Begin\n", __func__);
1631
b7edfaa1 1632 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1633 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
ce6ab4f2 1634 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
c7d73af2
TH		 1635 pnfs_lseg_range_match(&lseg->pls_range, range,
1636 strict_iomode)) {
9369a431 1637 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1638 break;
1639 }
b1f69b75
AA		 1640 }
1641
1642 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1643 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1644 return ret;
e5e94017
BH		 1645}
1646
d23d61c8
AA		 1647/*
1648 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1649 * to the MDS or over pNFS
1650 *
1651 * The nfs_inode read_io and write_io fields are cumulative counters reset
1652 * when there are no layout segments. Note that in pnfs_update_layout iomode
1653 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1654 * WRITE request.
1655 *
1656 * A return of true means use MDS I/O.
1657 *
1658 * From rfc 5661:
1659 * If a file's size is smaller than the file size threshold, data accesses
1660 * SHOULD be sent to the metadata server. If an I/O request has a length that
1661 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1662 * server. If both file size and I/O size are provided, the client SHOULD
1663 * reach or exceed both thresholds before sending its read or write
1664 * requests to the data server.
1665 */
1666static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1667 struct inode *ino, int iomode)
1668{
1669 struct nfs4_threshold *t = ctx->mdsthreshold;
1670 struct nfs_inode *nfsi = NFS_I(ino);
1671 loff_t fsize = i_size_read(ino);
1672 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1673
1674 if (t == NULL)
1675 return ret;
1676
1677 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1678 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1679
1680 switch (iomode) {
1681 case IOMODE_READ:
1682 if (t->bm & THRESHOLD_RD) {
1683 dprintk("%s fsize %llu\n", __func__, fsize);
1684 size_set = true;
1685 if (fsize < t->rd_sz)
1686 size = true;
1687 }
1688 if (t->bm & THRESHOLD_RD_IO) {
1689 dprintk("%s nfsi->read_io %llu\n", __func__,
1690 nfsi->read_io);
1691 io_set = true;
1692 if (nfsi->read_io < t->rd_io_sz)
1693 io = true;
1694 }
1695 break;
1696 case IOMODE_RW:
1697 if (t->bm & THRESHOLD_WR) {
1698 dprintk("%s fsize %llu\n", __func__, fsize);
1699 size_set = true;
1700 if (fsize < t->wr_sz)
1701 size = true;
1702 }
1703 if (t->bm & THRESHOLD_WR_IO) {
1704 dprintk("%s nfsi->write_io %llu\n", __func__,
1705 nfsi->write_io);
1706 io_set = true;
1707 if (nfsi->write_io < t->wr_io_sz)
1708 io = true;
1709 }
1710 break;
1711 }
1712 if (size_set && io_set) {
1713 if (size && io)
1714 ret = true;
1715 } else if (size || io)
1716 ret = true;
1717
1718 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1719 return ret;
1720}
1721
aa8a45ee
PT		 1722static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1723{
1724 /*
1725 * send layoutcommit as it can hold up layoutreturn due to lseg
1726 * reference
1727 */
1728 pnfs_layoutcommit_inode(lo->plh_inode, false);
1729 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1730 nfs_wait_bit_killable,
aa8a45ee
PT		 1731 TASK_UNINTERRUPTIBLE);
1732}
1733
d67ae825
TH		 1734static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1735{
1736 unsigned long *bitlock = &lo->plh_flags;
1737
1738 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1739 smp_mb__after_atomic();
1740 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1741}
1742
78746a38
FI		 1743static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1744 struct nfs_server *server)
1745{
1746 if (list_empty(&lo->plh_layouts)) {
1747 struct nfs_client *clp = server->nfs_client;
1748
1749 /* The lo must be on the clp list if there is any
1750 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1751 */
1752 spin_lock(&clp->cl_lock);
1753 if (list_empty(&lo->plh_layouts))
1754 list_add_tail(&lo->plh_layouts, &server->layouts);
1755 spin_unlock(&clp->cl_lock);
1756 }
1757}
1758
e5e94017
BH		 1759/*
1760 * Layout segment is retreived from the server if not cached.
1761 * The appropriate layout segment is referenced and returned to the caller.
1762 */
7c24d948 1763struct pnfs_layout_segment *
e5e94017
BH		 1764pnfs_update_layout(struct inode *ino,
1765 struct nfs_open_context *ctx,
fb3296eb
BH		 1766 loff_t pos,
1767 u64 count,
a75b9df9 1768 enum pnfs_iomode iomode,
c7d73af2 1769 bool strict_iomode,
a75b9df9 1770 gfp_t gfp_flags)
e5e94017 1771{
fb3296eb
BH		 1772 struct pnfs_layout_range arg = {
1773 .iomode = iomode,
1774 .offset = pos,
1775 .length = count,
1776 };
70d2f7b1 1777 unsigned pg_offset;
6382a441
WAA		 1778 struct nfs_server *server = NFS_SERVER(ino);
1779 struct nfs_client *clp = server->nfs_client;
183d9e7b 1780 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1781 struct pnfs_layout_segment *lseg = NULL;
587f03de 1782 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1783 nfs4_stateid stateid;
1784 long timeout = 0;
66b53f32 1785 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1786 bool first;
e5e94017 1787
9a4bf31d 1788 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1789 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1790 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1791 goto out;
9a4bf31d 1792 }
d23d61c8 1793
9a4bf31d 1794 if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
183d9e7b 1795 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1796 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
4ae93560 1797 goto out;
9a4bf31d 1798 }
4ae93560 1799
9a4bf31d 1800 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1801 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1802 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1803 goto out;
9a4bf31d 1804 }
d23d61c8 1805
9bf87482 1806lookup_again:
b88fa69e 1807 nfs4_client_recover_expired_lease(clp);
9bf87482 1808 first = false;
e5e94017 1809 spin_lock(&ino->i_lock);
9fa40758 1810 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1811 if (lo == NULL) {
1812 spin_unlock(&ino->i_lock);
183d9e7b 1813 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1814 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1815 goto out;
1816 }
e5e94017 1817
43f1b3da 1818 /* Do we even need to bother with this? */
a59c30ac 1819 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1820 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1821 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1822 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1823 goto out_unlock;
1824 }
1825
1826 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1827 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1828 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1829 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1830 goto out_unlock;
9a4bf31d 1831 }
e5e94017 1832
c7d73af2 1833 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1834 if (lseg) {
1835 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1836 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1837 goto out_unlock;
1838 }
1839
1840 if (!nfs4_valid_open_stateid(ctx->state)) {
1841 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1842 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1843 goto out_unlock;
1844 }
1845
1846 /*
1847 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1848 * stateid, or it has been invalidated, then we must use the open
1849 * stateid.
1850 */
67a3b721 1851 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
183d9e7b
JL		 1852
1853 /*
1854 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1855 * RFC 5661 Errata 3208.
1856 */
1857 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1858 &lo->plh_flags)) {
1859 spin_unlock(&ino->i_lock);
1860 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1861 TASK_UNINTERRUPTIBLE);
1862 pnfs_put_layout_hdr(lo);
183d9e7b 1863 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1864 goto lookup_again;
1865 }
183d9e7b
JL		 1866
1867 first = true;
70d2f7b1
TM		 1868 if (nfs4_select_rw_stateid(ctx->state,
1869 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1870 NULL, &stateid, NULL) != 0) {
1871 trace_pnfs_update_layout(ino, pos, count,
1872 iomode, lo, lseg,
1873 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1874 goto out_unlock;
1875 }
9bf87482 1876 } else {
183d9e7b 1877 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 1878 }
568e8c49 1879
aa8a45ee
PT		 1880 /*
1881 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1882 * for LAYOUTRETURN even if first is true.
1883 */
8f70f53a 1884 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 1885 spin_unlock(&ino->i_lock);
1886 dprintk("%s wait for layoutreturn\n", __func__);
1887 if (pnfs_prepare_to_retry_layoutget(lo)) {
d67ae825
TH		 1888 if (first)
1889 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 1890 pnfs_put_layout_hdr(lo);
1891 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 1892 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1893 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 1894 goto lookup_again;
1895 }
183d9e7b 1896 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1897 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 1898 goto out_put_layout_hdr;
1899 }
1900
9a4bf31d 1901 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 1902 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1903 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 1904 goto out_unlock;
9a4bf31d 1905 }
cf7d63f1 1906 atomic_inc(&lo->plh_outstanding);
f49f9baa 1907 spin_unlock(&ino->i_lock);
30005121 1908
78746a38 1909 _add_to_server_list(lo, server);
e5e94017 1910
09cbfeaf 1911 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 1912 if (pg_offset) {
1913 arg.offset -= pg_offset;
1914 arg.length += pg_offset;
1915 }
7c24d948 1916 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 1917 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 1918
5e36e2a9 1919 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 1920 if (!lgp) {
1921 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
1922 PNFS_UPDATE_LAYOUT_NOMEM);
1923 atomic_dec(&lo->plh_outstanding);
1924 goto out_put_layout_hdr;
1925 }
1926
dacb452d 1927 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 1928 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1929 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
56b38a1f 1930 atomic_dec(&lo->plh_outstanding);
83026d80 1931 if (IS_ERR(lseg)) {
183d9e7b 1932 switch(PTR_ERR(lseg)) {
e85d7ee4 1933 case -EBUSY:
183d9e7b
JL		 1934 if (time_after(jiffies, giveup))
1935 lseg = NULL;
66b53f32
TM		 1936 break;
1937 case -ERECALLCONFLICT:
1938 /* Huh? We hold no layouts, how is there a recall? */
1939 if (first) {
1940 lseg = NULL;
1941 break;
1942 }
1943 /* Destroy the existing layout and start over */
1944 if (time_after(jiffies, giveup))
1945 pnfs_destroy_layout(NFS_I(ino));
183d9e7b
JL		 1946 /* Fallthrough */
1947 case -EAGAIN:
56b38a1f 1948 break;
183d9e7b
JL		 1949 default:
1950 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
1951 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1952 lseg = NULL;
1953 }
56b38a1f
TM		 1954 goto out_put_layout_hdr;
1955 }
1956 if (lseg) {
1957 if (first)
1958 pnfs_clear_first_layoutget(lo);
1959 trace_pnfs_update_layout(ino, pos, count,
1960 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
1961 pnfs_put_layout_hdr(lo);
1962 goto lookup_again;
83026d80
JL		 1963 }
1964 } else {
1965 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
1966 }
1967
830ffb56 1968out_put_layout_hdr:
d67ae825
TH		 1969 if (first)
1970 pnfs_clear_first_layoutget(lo);
70c3bd2b 1971 pnfs_put_layout_hdr(lo);
e5e94017 1972out:
f86bbcf8
TM		 1973 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1974 "(%s, offset: %llu, length: %llu)\n",
1975 __func__, ino->i_sb->s_id,
1976 (unsigned long long)NFS_FILEID(ino),
d600ad1f 1977 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 1978 iomode==IOMODE_RW ? "read/write" : "read-only",
1979 (unsigned long long)pos,
1980 (unsigned long long)count);
e5e94017
BH		 1981 return lseg;
1982out_unlock:
1983 spin_unlock(&ino->i_lock);
830ffb56 1984 goto out_put_layout_hdr;
e5e94017 1985}
7c24d948 1986EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 1987
540d9864
TM		 1988static bool
1989pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
1990{
1991 switch (range->iomode) {
1992 case IOMODE_READ:
1993 case IOMODE_RW:
1994 break;
1995 default:
1996 return false;
1997 }
1998 if (range->offset == NFS4_MAX_UINT64)
1999 return false;
2000 if (range->length == 0)
2001 return false;
2002 if (range->length != NFS4_MAX_UINT64 &&
2003 range->length > NFS4_MAX_UINT64 - range->offset)
2004 return false;
2005 return true;
2006}
2007
78746a38
FI		 2008static struct pnfs_layout_hdr *
2009_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2010{
2011 struct pnfs_layout_hdr *lo;
2012
2013 spin_lock(&ino->i_lock);
2014 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2015 if (!lo)
2016 goto out_unlock;
2017 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2018 goto out_unlock;
2019 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2020 goto out_unlock;
2021 if (pnfs_layoutgets_blocked(lo))
2022 goto out_unlock;
2023 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2024 goto out_unlock;
2025 atomic_inc(&lo->plh_outstanding);
2026 spin_unlock(&ino->i_lock);
2027 _add_to_server_list(lo, NFS_SERVER(ino));
2028 return lo;
2029
2030out_unlock:
2031 spin_unlock(&ino->i_lock);
2032 pnfs_put_layout_hdr(lo);
2033 return NULL;
2034}
2035
2409a976
FI		 2036extern const nfs4_stateid current_stateid;
2037
2038static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2039 struct nfs_open_context *ctx)
2040{
78746a38
FI		 2041 struct inode *ino = data->dentry->d_inode;
2042 struct pnfs_layout_range rng = {
2043 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2044 IOMODE_RW: IOMODE_READ,
2045 .offset = 0,
2046 .length = NFS4_MAX_UINT64,
2047 };
2048 struct nfs4_layoutget *lgp;
2049 struct pnfs_layout_hdr *lo;
2050
64294b08
TM		 2051 /* Heuristic: don't send layoutget if we have cached data */
2052 if (rng.iomode == IOMODE_READ &&
2053 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2054 return;
2055
78746a38
FI		 2056 lo = _pnfs_grab_empty_layout(ino, ctx);
2057 if (!lo)
2058 return;
2059 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2060 &rng, GFP_KERNEL);
2061 if (!lgp) {
2062 pnfs_clear_first_layoutget(lo);
2063 pnfs_put_layout_hdr(lo);
2064 return;
2065 }
2066 data->lgp = lgp;
2067 data->o_arg.lg_args = &lgp->args;
2068 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2069}
2070
2071static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2072 struct nfs_open_context *ctx)
2073{
2074 struct pnfs_layout_range rng = {
2075 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2076 IOMODE_RW: IOMODE_READ,
2077 .offset = 0,
2078 .length = NFS4_MAX_UINT64,
2079 };
2080 struct nfs4_layoutget *lgp;
2081
2082 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2083 &rng, GFP_KERNEL);
2084 if (!lgp)
2085 return;
2086 data->lgp = lgp;
2087 data->o_arg.lg_args = &lgp->args;
2088 data->o_res.lg_res = &lgp->res;
2089}
2090
2091void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2092 struct nfs_open_context *ctx)
2093{
2094 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2095
2096 if (!(pnfs_enabled_sb(server) &&
2097 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2098 return;
2099 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2100 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2101 return;
2409a976
FI		 2102 if (data->state)
2103 _lgopen_prepare_attached(data, ctx);
2104 else
2105 _lgopen_prepare_floating(data, ctx);
2106}
2107
2108void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2109 struct nfs_open_context *ctx)
2110{
2111 struct pnfs_layout_hdr *lo;
2112 struct pnfs_layout_segment *lseg;
c49b5209 2113 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2114 u32 iomode;
2115
6e01260c 2116 if (!lgp)
2409a976 2117 return;
6e01260c
FI		 2118 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2119 if (lgp->res.status) {
2120 switch (lgp->res.status) {
6e01260c 2121 default:
8dc96566
TM		 2122 break;
2123 /*
2124 * Halt lgopen attempts if the server doesn't recognise
2125 * the "current stateid" value, the layout type, or the
2126 * layoutget operation as being valid.
2127 * Also if it complains about too many ops in the compound
2128 * or of the request/reply being too big.
2129 */
2130 case -NFS4ERR_BAD_STATEID:
2131 case -NFS4ERR_NOTSUPP:
2132 case -NFS4ERR_REP_TOO_BIG:
2133 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2134 case -NFS4ERR_REQ_TOO_BIG:
2135 case -NFS4ERR_TOO_MANY_OPS:
2136 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2137 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2138 }
2139 return;
2140 }
2409a976 2141 if (!lgp->args.inode) {
78746a38
FI		 2142 lo = _pnfs_grab_empty_layout(ino, ctx);
2143 if (!lo)
2144 return;
2409a976
FI		 2145 lgp->args.inode = ino;
2146 } else
2147 lo = NFS_I(lgp->args.inode)->layout;
2409a976 2148
c49b5209
FI		 2149 if (read_seqcount_retry(&srv->nfs_client->cl_callback_count,
2150 lgp->callback_count))
30ae2412 2151 return;
2409a976 2152 lseg = pnfs_layout_process(lgp);
32f1c28f 2153 if (!IS_ERR(lseg)) {
2409a976
FI		 2154 iomode = lgp->args.range.iomode;
2155 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2156 pnfs_put_lseg(lseg);
2157 }
30ae2412
FI		 2158}
2159
2160void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2161{
2162 if (lgp != NULL) {
2163 struct inode *inode = lgp->args.inode;
2164 if (inode) {
2165 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2166 atomic_dec(&lo->plh_outstanding);
2167 pnfs_clear_first_layoutget(lo);
2168 }
2169 pnfs_layoutget_free(lgp);
2170 }
2409a976
FI		 2171}
2172
a0b0a6e3 2173struct pnfs_layout_segment *
b1f69b75
AA		 2174pnfs_layout_process(struct nfs4_layoutget *lgp)
2175{
2176 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2177 struct nfs4_layoutget_res *res = &lgp->res;
2178 struct pnfs_layout_segment *lseg;
b7edfaa1 2179 struct inode *ino = lo->plh_inode;
78096cca 2180 LIST_HEAD(free_me);
540d9864
TM		 2181
2182 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2183 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2184
2185 /* Inject layout blob into I/O device driver */
a75b9df9 2186 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2187 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2188 if (!lseg)
1b3c6d07
JL		 2189 lseg = ERR_PTR(-ENOMEM);
2190
2191 dprintk("%s: Could not allocate layout: error %ld\n",
2192 __func__, PTR_ERR(lseg));
2193 return lseg;
b1f69b75
AA		 2194 }
2195
119cef97 2196 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2197
b1f69b75 2198 spin_lock(&ino->i_lock);
e1c06f80 2199 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2200 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2201 goto out_forget;
43f1b3da 2202 }
038d6493 2203
9888d837
TM		 2204 if (!pnfs_layout_is_valid(lo)) {
2205 /* We have a completely new layout */
2206 pnfs_set_layout_stateid(lo, &res->stateid, true);
2207 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2208 /* existing state ID, make sure the sequence number matches. */
2209 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2210 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2211 goto out_forget;
362f7474
CH		 2212 }
2213 pnfs_set_layout_stateid(lo, &res->stateid, false);
2214 } else {
2215 /*
2216 * We got an entirely new state ID. Mark all segments for the
9888d837 2217 * inode invalid, and retry the layoutget
362f7474 2218 */
d9b61708 2219 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 2220 goto out_forget;
362f7474 2221 }
038d6493 2222
9369a431 2223 pnfs_get_lseg(lseg);
03772d2f 2224 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2225
b1f69b75 2226
3976143b 2227 if (res->return_on_close)
f7e8917a 2228 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2229
b1f69b75 2230 spin_unlock(&ino->i_lock);
78096cca 2231 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2232 return lseg;
43f1b3da 2233
1b3c6d07 2234out_forget:
43f1b3da
FI		 2235 spin_unlock(&ino->i_lock);
2236 lseg->pls_layout = lo;
2237 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2238 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2239}
2240
2f215968
TM		 2241/**
2242 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2243 * @lo: pointer to layout header
2244 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2245 * @return_range: describe layout segment ranges to be returned
2246 *
2247 * This function is mainly intended for use by layoutrecall. It attempts
2248 * to free the layout segment immediately, or else to mark it for return
2249 * as soon as its reference count drops to zero.
2250 */
10335556 2251int
016256df
PT		 2252pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2253 struct list_head *tmp_list,
6d597e17
JL		 2254 const struct pnfs_layout_range *return_range,
2255 u32 seq)
016256df
PT		 2256{
2257 struct pnfs_layout_segment *lseg, *next;
10335556 2258 int remaining = 0;
016256df
PT		 2259
2260 dprintk("%s:Begin lo %p\n", __func__, lo);
2261
2262 if (list_empty(&lo->plh_segs))
10335556 2263 return 0;
016256df 2264
fc7ff367 2265 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2266
2267 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2268 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2269 dprintk("%s: marking lseg %p iomode %d "
2270 "offset %llu length %llu\n", __func__,
2271 lseg, lseg->pls_range.iomode,
2272 lseg->pls_range.offset,
2273 lseg->pls_range.length);
2f215968
TM		 2274 if (mark_lseg_invalid(lseg, tmp_list))
2275 continue;
2276 remaining++;
016256df 2277 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2278 }
6d597e17
JL		 2279
2280 if (remaining)
2281 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2282
10335556 2283 return remaining;
016256df
PT		 2284}
2285
2286void pnfs_error_mark_layout_for_return(struct inode *inode,
2287 struct pnfs_layout_segment *lseg)
2288{
2289 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2290 struct pnfs_layout_range range = {
2291 .iomode = lseg->pls_range.iomode,
2292 .offset = 0,
2293 .length = NFS4_MAX_UINT64,
2294 };
10335556 2295 bool return_now = false;
016256df
PT		 2296
2297 spin_lock(&inode->i_lock);
bdebfccd
TM		 2298 if (!pnfs_layout_is_valid(lo)) {
2299 spin_unlock(&inode->i_lock);
2300 return;
2301 }
2d6cf5ab 2302 pnfs_set_plh_return_info(lo, range.iomode, 0);
016256df
PT		 2303 /*
2304 * mark all matching lsegs so that we are sure to have no live
2305 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2306 * for how it works.
2307 */
68f74479 2308 if (!pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0)) {
10335556 2309 nfs4_stateid stateid;
e5fd1904 2310 enum pnfs_iomode iomode;
10335556 2311
e5fd1904 2312 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2313 spin_unlock(&inode->i_lock);
2314 if (return_now)
2315 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2316 } else {
2317 spin_unlock(&inode->i_lock);
2318 nfs_commit_inode(inode, 0);
2319 }
016256df
PT		 2320}
2321EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2322
b3230e80
TM		 2323void
2324pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2325{
2326 if (pgio->pg_lseg == NULL ||
2327 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2328 return;
2329 pnfs_put_lseg(pgio->pg_lseg);
2330 pgio->pg_lseg = NULL;
2331}
2332EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2333
08cb5b0f
BC		 2334/*
2335 * Check for any intersection between the request and the pgio->pg_lseg,
2336 * and if none, put this pgio->pg_lseg away.
2337 */
2338static void
2339pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2340{
2341 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2342 pnfs_put_lseg(pgio->pg_lseg);
2343 pgio->pg_lseg = NULL;
2344 }
2345}
2346
d8007d4d
TM		 2347void
2348pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2349{
1fd937bd
PT		 2350 u64 rd_size = req->wb_bytes;
2351
b3230e80 2352 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2353 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2354 if (pgio->pg_lseg == NULL) {
2355 if (pgio->pg_dreq == NULL)
2356 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2357 else
2358 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2359
2360 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2361 req->wb_context,
2362 req_offset(req),
2363 rd_size,
2364 IOMODE_READ,
c7d73af2 2365 false,
cb5d04bc 2366 GFP_KERNEL);
d600ad1f
PT		 2367 if (IS_ERR(pgio->pg_lseg)) {
2368 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2369 pgio->pg_lseg = NULL;
2370 return;
2371 }
cb5d04bc 2372 }
e885de1a
TM		 2373 /* If no lseg, fall back to read through mds */
2374 if (pgio->pg_lseg == NULL)
1f945357 2375 nfs_pageio_reset_read_mds(pgio);
e885de1a 2376
d8007d4d
TM		 2377}
2378EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2379
2380void
6296556f
PT		 2381pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2382 struct nfs_page *req, u64 wb_size)
d8007d4d 2383{
b3230e80 2384 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2385 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2386 if (pgio->pg_lseg == NULL) {
cb5d04bc
PT		 2387 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2388 req->wb_context,
2389 req_offset(req),
2390 wb_size,
2391 IOMODE_RW,
c7d73af2 2392 false,
cb5d04bc 2393 GFP_NOFS);
d600ad1f
PT		 2394 if (IS_ERR(pgio->pg_lseg)) {
2395 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2396 pgio->pg_lseg = NULL;
2397 return;
2398 }
2399 }
e885de1a
TM		 2400 /* If no lseg, fall back to write through mds */
2401 if (pgio->pg_lseg == NULL)
1f945357 2402 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2403}
2404EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2405
180bb5ec
WAA		 2406void
2407pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2408{
2409 if (desc->pg_lseg) {
2410 pnfs_put_lseg(desc->pg_lseg);
2411 desc->pg_lseg = NULL;
2412 }
2413}
2414EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2415
b4fdac1a
WAA		 2416/*
2417 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2418 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2419 */
2420size_t
a7d42ddb
WAA		 2421pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2422 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2423{
0f9c429e 2424 unsigned int size;
c5e20cb7 2425 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2426
2427 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2428 if (!size)
2429 return 0;
94ad1c80 2430
19982ba8 2431 /*
c5e20cb7
WAA		 2432 * 'size' contains the number of bytes left in the current page (up
2433 * to the original size asked for in @req->wb_bytes).
2434 *
2435 * Calculate how many bytes are left in the layout segment
2436 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2437 *
2438 * Please also note that 'end_offset' is actually the offset of the
2439 * first byte that lies outside the pnfs_layout_range. FIXME?
2440 *
2441 */
19b54848 2442 if (pgio->pg_lseg) {
17822b20 2443 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2444 pgio->pg_lseg->pls_range.length);
2445 req_start = req_offset(req);
08cb5b0f 2446
c5e20cb7 2447 /* start of request is past the last byte of this segment */
08cb5b0f 2448 if (req_start >= seg_end)
19b54848 2449 return 0;
c5e20cb7
WAA		 2450
2451 /* adjust 'size' iff there are fewer bytes left in the
2452 * segment than what nfs_generic_pg_test returned */
2453 seg_left = seg_end - req_start;
2454 if (seg_left < size)
2455 size = (unsigned int)seg_left;
19b54848 2456 }
0f9c429e 2457
19b54848 2458 return size;
94ad1c80 2459}
89a58e32 2460EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2461
53113ad3 2462int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2463{
2464 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2465
2466 /* Resend all requests through the MDS */
53113ad3
WAA		 2467 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2468 hdr->completion_ops);
c7070113 2469 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2470 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2471}
e7dd79af 2472EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2473
d45f60c6 2474static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2475{
cd841605
FI		 2476
2477 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2478 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2479 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2480 pnfs_return_layout(hdr->inode);
1acbbb4e 2481 }
6c75dc0d 2482 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2483 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2484}
2485
d20581aa
BH		 2486/*
2487 * Called by non rpc-based layout drivers
2488 */
d45f60c6 2489void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2490{
f8417b48 2491 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2492 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2493 hdr->mds_offset + hdr->res.count);
d45f60c6 2494 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2495 }
2496 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2497 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2498 pnfs_ld_handle_write_error(hdr);
2499 hdr->mds_ops->rpc_release(hdr);
44b83799 2500}
d20581aa 2501EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2502
dce81290
TM		 2503static void
2504pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2505 struct nfs_pgio_header *hdr)
dce81290 2506{
48d635f1 2507 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2508
6c75dc0d 2509 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2510 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2511 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2512 mirror->pg_recoalesce = 1;
6c75dc0d 2513 }
ba4a76f7 2514 hdr->completion_ops->completion(hdr);
dce81290
TM		 2515}
2516
2517static enum pnfs_try_status
d45f60c6 2518pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2519 const struct rpc_call_ops *call_ops,
2520 struct pnfs_layout_segment *lseg,
2521 int how)
0382b744 2522{
cd841605 2523 struct inode *inode = hdr->inode;
0382b744
AA		 2524 enum pnfs_try_status trypnfs;
2525 struct nfs_server *nfss = NFS_SERVER(inode);
2526
cd841605 2527 hdr->mds_ops = call_ops;
0382b744
AA		 2528
2529 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2530 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2531 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2532 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2533 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2534 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2535 return trypnfs;
2536}
2537
dce81290 2538static void
7f714720
WAA		 2539pnfs_do_write(struct nfs_pageio_descriptor *desc,
2540 struct nfs_pgio_header *hdr, int how)
dce81290 2541{
dce81290
TM		 2542 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2543 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2544 enum pnfs_try_status trypnfs;
dce81290 2545
d45f60c6 2546 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2547 switch (trypnfs) {
2548 case PNFS_NOT_ATTEMPTED:
d45f60c6 2549 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2550 case PNFS_ATTEMPTED:
2551 break;
2552 case PNFS_TRY_AGAIN:
2553 /* cleanup hdr and prepare to redo pnfs */
2554 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2555 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2556 list_splice_init(&hdr->pages, &mirror->pg_list);
2557 mirror->pg_recoalesce = 1;
2558 }
2559 hdr->mds_ops->rpc_release(hdr);
2560 }
dce81290
TM		 2561}
2562
6c75dc0d
FI		 2563static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2564{
9369a431 2565 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2566 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2567}
2568
dce81290
TM		 2569int
2570pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2571{
6c75dc0d 2572 struct nfs_pgio_header *hdr;
dce81290
TM		 2573 int ret;
2574
1e7f3a48
WAA		 2575 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2576 if (!hdr) {
2bff2288
PT		 2577 desc->pg_error = -ENOMEM;
2578 return desc->pg_error;
dce81290 2579 }
6c75dc0d 2580 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2581
9369a431 2582 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2583 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2584 if (!ret)
7f714720 2585 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2586
6c75dc0d 2587 return ret;
dce81290
TM		 2588}
2589EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2590
53113ad3 2591int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2592{
2593 struct nfs_pageio_descriptor pgio;
2594
1acbbb4e 2595 /* Resend all requests through the MDS */
53113ad3
WAA		 2596 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2597 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2598}
e7dd79af 2599EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2600
d45f60c6 2601static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2602{
cd841605
FI		 2603 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2604 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2605 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2606 pnfs_return_layout(hdr->inode);
1acbbb4e 2607 }
4db6e0b7 2608 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2609 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2610}
2611
d20581aa
BH		 2612/*
2613 * Called by non rpc-based layout drivers
2614 */
d45f60c6 2615void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2616{
bfc505de 2617 if (likely(!hdr->pnfs_error))
d45f60c6 2618 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2619 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2620 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2621 pnfs_ld_handle_read_error(hdr);
2622 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2623}
2624EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2625
493292dd
TM		 2626static void
2627pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2628 struct nfs_pgio_header *hdr)
493292dd 2629{
48d635f1 2630 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2631
4db6e0b7 2632 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2633 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2634 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2635 mirror->pg_recoalesce = 1;
4db6e0b7 2636 }
ba4a76f7 2637 hdr->completion_ops->completion(hdr);
493292dd
TM		 2638}
2639
64419a9b
AA		 2640/*
2641 * Call the appropriate parallel I/O subsystem read function.
2642 */
493292dd 2643static enum pnfs_try_status
d45f60c6 2644pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2645 const struct rpc_call_ops *call_ops,
2646 struct pnfs_layout_segment *lseg)
64419a9b 2647{
cd841605 2648 struct inode *inode = hdr->inode;
64419a9b
AA		 2649 struct nfs_server *nfss = NFS_SERVER(inode);
2650 enum pnfs_try_status trypnfs;
2651
cd841605 2652 hdr->mds_ops = call_ops;
64419a9b
AA		 2653
2654 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2655 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2656
d45f60c6 2657 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2658 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2659 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2660 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2661 return trypnfs;
2662}
863a3c6c 2663
ceb11e13 2664/* Resend all requests through pnfs. */
1b1bc66b 2665void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2666{
2667 struct nfs_pageio_descriptor pgio;
2668
1b1bc66b 2669 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2670 /* Prevent deadlocks with layoutreturn! */
2671 pnfs_put_lseg(hdr->lseg);
2672 hdr->lseg = NULL;
2673
1b1bc66b
WAA		 2674 nfs_pageio_init_read(&pgio, hdr->inode, false,
2675 hdr->completion_ops);
2676 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2677 }
ceb11e13
PT		 2678}
2679EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2680
493292dd 2681static void
7f714720 2682pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2683{
493292dd
TM		 2684 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2685 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2686 enum pnfs_try_status trypnfs;
493292dd 2687
d45f60c6 2688 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2689 switch (trypnfs) {
2690 case PNFS_NOT_ATTEMPTED:
d45f60c6 2691 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2692 case PNFS_ATTEMPTED:
2693 break;
2694 case PNFS_TRY_AGAIN:
2695 /* cleanup hdr and prepare to redo pnfs */
2696 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2697 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2698 list_splice_init(&hdr->pages, &mirror->pg_list);
2699 mirror->pg_recoalesce = 1;
2700 }
2701 hdr->mds_ops->rpc_release(hdr);
2702 }
493292dd
TM		 2703}
2704
4db6e0b7
FI		 2705static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2706{
9369a431 2707 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2708 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2709}
2710
493292dd
TM		 2711int
2712pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2713{
4db6e0b7 2714 struct nfs_pgio_header *hdr;
493292dd
TM		 2715 int ret;
2716
1e7f3a48
WAA		 2717 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2718 if (!hdr) {
2bff2288
PT		 2719 desc->pg_error = -ENOMEM;
2720 return desc->pg_error;
493292dd 2721 }
4db6e0b7 2722 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2723 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2724 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2725 if (!ret)
7f714720 2726 pnfs_do_read(desc, hdr);
4db6e0b7 2727 return ret;
493292dd
TM		 2728}
2729EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2730
71244d9b
TM		 2731static void pnfs_clear_layoutcommitting(struct inode *inode)
2732{
2733 unsigned long *bitlock = &NFS_I(inode)->flags;
2734
2735 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2736 smp_mb__after_atomic();
71244d9b
TM		 2737 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2738}
2739
863a3c6c 2740/*
a9bae566 2741 * There can be multiple RW segments.
863a3c6c 2742 */
a9bae566 2743static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2744{
a9bae566 2745 struct pnfs_layout_segment *lseg;
863a3c6c 2746
a9bae566
PT		 2747 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2748 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2749 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2750 list_add(&lseg->pls_lc_list, listp);
2751 }
863a3c6c
AA		 2752}
2753
a073dbff
TM		 2754static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2755{
2756 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2757
2758 /* Matched by references in pnfs_set_layoutcommit */
2759 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2760 list_del_init(&lseg->pls_lc_list);
2761 pnfs_put_lseg(lseg);
2762 }
2763
71244d9b 2764 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2765}
2766
1b0ae068
PT		 2767void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2768{
b9e028fd 2769 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2770}
2771EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2772
863a3c6c 2773void
67af7611
TM		 2774pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2775 loff_t end_pos)
863a3c6c 2776{
cd841605 2777 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2778 bool mark_as_dirty = false;
863a3c6c 2779
cd841605 2780 spin_lock(&inode->i_lock);
863a3c6c 2781 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2782 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2783 mark_as_dirty = true;
863a3c6c 2784 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2785 __func__, inode->i_ino);
29559b11
TM		 2786 } else if (end_pos > nfsi->layout->plh_lwb)
2787 nfsi->layout->plh_lwb = end_pos;
67af7611 2788 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2789 /* references matched in nfs4_layoutcommit_release */
67af7611 2790 pnfs_get_lseg(lseg);
a9bae566 2791 }
cd841605 2792 spin_unlock(&inode->i_lock);
acff5880 2793 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2794 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2795
2796 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2797 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2798 if (mark_as_dirty)
cd841605 2799 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2800}
2801EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2802
db29c089
AA		 2803void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2804{
2805 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2806
2807 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2808 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2809 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2810}
2811
de4b15c7
AA		 2812/*
2813 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2814 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2815 * data to disk to allow the server to recover the data if it crashes.
2816 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2817 * is off, and a COMMIT is sent to a data server, or
2818 * if WRITEs to a data server return NFS_DATA_SYNC.
2819 */
863a3c6c 2820int
ef311537 2821pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2822{
5f919c9f 2823 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2824 struct nfs4_layoutcommit_data *data;
2825 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2826 loff_t end_pos;
71244d9b 2827 int status;
863a3c6c 2828
71244d9b 2829 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2830 return 0;
2831
71244d9b 2832 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2833
71244d9b 2834 status = -EAGAIN;
92407e75 2835 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2836 if (!sync)
2837 goto out;
74316201 2838 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2839 NFS_INO_LAYOUTCOMMITTING,
2840 nfs_wait_bit_killable,
2841 TASK_KILLABLE);
92407e75 2842 if (status)
71244d9b 2843 goto out;
92407e75
PT		 2844 }
2845
71244d9b
TM		 2846 status = -ENOMEM;
2847 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2848 data = kzalloc(sizeof(*data), GFP_NOFS);
2849 if (!data)
2850 goto clear_layoutcommitting;
2851
2852 status = 0;
de4b15c7 2853 spin_lock(&inode->i_lock);
71244d9b
TM		 2854 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2855 goto out_unlock;
a9bae566 2856
71244d9b 2857 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 2858 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 2859
acff5880 2860 end_pos = nfsi->layout->plh_lwb;
863a3c6c 2861
f597c537 2862 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 2863 spin_unlock(&inode->i_lock);
2864
2865 data->args.inode = inode;
9fa40758 2866 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 2867 nfs_fattr_init(&data->fattr);
2868 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2869 data->res.fattr = &data->fattr;
2e18d4d8
TM		 2870 if (end_pos != 0)
2871 data->args.lastbytewritten = end_pos - 1;
2872 else
2873 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 2874 data->res.server = NFS_SERVER(inode);
2875
5f919c9f
CH		 2876 if (ld->prepare_layoutcommit) {
2877 status = ld->prepare_layoutcommit(&data->args);
2878 if (status) {
3471648a 2879 put_rpccred(data->cred);
5f919c9f 2880 spin_lock(&inode->i_lock);
29559b11
TM		 2881 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2882 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 2883 nfsi->layout->plh_lwb = end_pos;
3471648a 2884 goto out_unlock;
5f919c9f
CH		 2885 }
2886 }
2887
2888
863a3c6c
AA		 2889 status = nfs4_proc_layoutcommit(data, sync);
2890out:
92407e75
PT		 2891 if (status)
2892 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2893 dprintk("<-- %s status %d\n", __func__, status);
2894 return status;
71244d9b
TM		 2895out_unlock:
2896 spin_unlock(&inode->i_lock);
92407e75 2897 kfree(data);
71244d9b
TM		 2898clear_layoutcommitting:
2899 pnfs_clear_layoutcommitting(inode);
92407e75 2900 goto out;
863a3c6c 2901}
72cff449 2902EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 2903
5bb89b47
TM		 2904int
2905pnfs_generic_sync(struct inode *inode, bool datasync)
2906{
2907 return pnfs_layoutcommit_inode(inode, true);
2908}
2909EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2910
82be417a
AA		 2911struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2912{
2913 struct nfs4_threshold *thp;
2914
2915 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2916 if (!thp) {
2917 dprintk("%s mdsthreshold allocation failed\n", __func__);
2918 return NULL;
2919 }
2920 return thp;
2921}
8733408d 2922
865a7ecb 2923#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 2924int
c8ad8894 2925pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 2926{
2927 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2928 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 2929 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 2930 struct nfs42_layoutstat_data *data;
2931 struct pnfs_layout_hdr *hdr;
2932 int status = 0;
2933
2934 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2935 goto out;
2936
6c5a0d89
TM		 2937 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2938 goto out;
2939
1bfe3b25
PT		 2940 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2941 goto out;
2942
8733408d
PT		 2943 spin_lock(&inode->i_lock);
2944 if (!NFS_I(inode)->layout) {
2945 spin_unlock(&inode->i_lock);
f538d0ba 2946 goto out_clear_layoutstats;
8733408d
PT		 2947 }
2948 hdr = NFS_I(inode)->layout;
2949 pnfs_get_layout_hdr(hdr);
2950 spin_unlock(&inode->i_lock);
2951
c8ad8894 2952 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 2953 if (!data) {
2954 status = -ENOMEM;
2955 goto out_put;
2956 }
2957
2958 data->args.fh = NFS_FH(inode);
2959 data->args.inode = inode;
8733408d
PT		 2960 status = ld->prepare_layoutstats(&data->args);
2961 if (status)
2962 goto out_free;
2963
2964 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2965
2966out:
2967 dprintk("%s returns %d\n", __func__, status);
2968 return status;
2969
2970out_free:
2971 kfree(data);
2972out_put:
2973 pnfs_put_layout_hdr(hdr);
f538d0ba 2974out_clear_layoutstats:
1bfe3b25
PT		 2975 smp_mb__before_atomic();
2976 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2977 smp_mb__after_atomic();
8733408d
PT		 2978 goto out;
2979}
2980EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 2981#endif
bbf58bf3
TM		 2982
2983unsigned int layoutstats_timer;
2984module_param(layoutstats_timer, uint, 0644);
2985EXPORT_SYMBOL_GPL(layoutstats_timer);
Empty description
This page took 0.929519 seconds and 4 git commands to generate.