blob: 5d09a36b2cd8a0ea2da4bd077d0e34adce3c9e15 [file] [log] [blame]
Ricardo Labiaga85e174b2010-10-20 00:17:58 -04001/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
Trond Myklebust493292d2011-07-13 15:58:28 -040031#include <linux/nfs_page.h>
Paul Gortmaker143cb492011-07-01 14:23:34 -040032#include <linux/module.h>
Andy Adamson974cec82010-10-20 00:18:02 -040033#include "internal.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040034#include "pnfs.h"
Andy Adamson64419a92011-03-01 01:34:16 +000035#include "iostat.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040036
37#define NFSDBG_FACILITY NFSDBG_PNFS
38
Fred Isaman02c35fc2010-10-20 00:17:59 -040039/* Locking:
40 *
41 * pnfs_spinlock:
42 * protects pnfs_modules_tbl.
43 */
44static DEFINE_SPINLOCK(pnfs_spinlock);
45
46/*
47 * pnfs_modules_tbl holds all pnfs modules
48 */
49static LIST_HEAD(pnfs_modules_tbl);
50
51/* Return the registered pnfs layout driver module matching given id */
52static struct pnfs_layoutdriver_type *
53find_pnfs_driver_locked(u32 id)
54{
55 struct pnfs_layoutdriver_type *local;
56
57 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
58 if (local->id == id)
59 goto out;
60 local = NULL;
61out:
62 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
63 return local;
64}
65
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040066static struct pnfs_layoutdriver_type *
67find_pnfs_driver(u32 id)
68{
Fred Isaman02c35fc2010-10-20 00:17:59 -040069 struct pnfs_layoutdriver_type *local;
70
71 spin_lock(&pnfs_spinlock);
72 local = find_pnfs_driver_locked(id);
73 spin_unlock(&pnfs_spinlock);
74 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040075}
76
77void
78unset_pnfs_layoutdriver(struct nfs_server *nfss)
79{
Benny Halevy738fd0f32011-07-30 20:52:36 -040080 if (nfss->pnfs_curr_ld) {
81 if (nfss->pnfs_curr_ld->clear_layoutdriver)
82 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
Fred Isaman02c35fc2010-10-20 00:17:59 -040083 module_put(nfss->pnfs_curr_ld->owner);
Benny Halevy738fd0f32011-07-30 20:52:36 -040084 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040085 nfss->pnfs_curr_ld = NULL;
86}
87
88/*
89 * Try to set the server's pnfs module to the pnfs layout type specified by id.
90 * Currently only one pNFS layout driver per filesystem is supported.
91 *
92 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
93 */
94void
Benny Halevy738fd0f32011-07-30 20:52:36 -040095set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
96 u32 id)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040097{
98 struct pnfs_layoutdriver_type *ld_type = NULL;
99
100 if (id == 0)
101 goto out_no_driver;
102 if (!(server->nfs_client->cl_exchange_flags &
103 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500104 printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
105 __func__, id, server->nfs_client->cl_exchange_flags);
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400106 goto out_no_driver;
107 }
108 ld_type = find_pnfs_driver(id);
109 if (!ld_type) {
110 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
111 ld_type = find_pnfs_driver(id);
112 if (!ld_type) {
113 dprintk("%s: No pNFS module found for %u.\n",
114 __func__, id);
115 goto out_no_driver;
116 }
117 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400118 if (!try_module_get(ld_type->owner)) {
119 dprintk("%s: Could not grab reference on module\n", __func__);
120 goto out_no_driver;
121 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400122 server->pnfs_curr_ld = ld_type;
Benny Halevy738fd0f32011-07-30 20:52:36 -0400123 if (ld_type->set_layoutdriver
124 && ld_type->set_layoutdriver(server, mntfh)) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500125 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
126 "driver %u.\n", __func__, id);
Benny Halevy738fd0f32011-07-30 20:52:36 -0400127 module_put(ld_type->owner);
128 goto out_no_driver;
129 }
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000130
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400131 dprintk("%s: pNFS module for %u set\n", __func__, id);
132 return;
133
134out_no_driver:
135 dprintk("%s: Using NFSv4 I/O\n", __func__);
136 server->pnfs_curr_ld = NULL;
137}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400138
139int
140pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
141{
142 int status = -EINVAL;
143 struct pnfs_layoutdriver_type *tmp;
144
145 if (ld_type->id == 0) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500146 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
Fred Isaman02c35fc2010-10-20 00:17:59 -0400147 return status;
148 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400149 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500150 printk(KERN_ERR "NFS: %s Layout driver must provide "
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400151 "alloc_lseg and free_lseg.\n", __func__);
152 return status;
153 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400154
155 spin_lock(&pnfs_spinlock);
156 tmp = find_pnfs_driver_locked(ld_type->id);
157 if (!tmp) {
158 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
159 status = 0;
160 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
161 ld_type->name);
162 } else {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500163 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
Fred Isaman02c35fc2010-10-20 00:17:59 -0400164 __func__, ld_type->id);
165 }
166 spin_unlock(&pnfs_spinlock);
167
168 return status;
169}
170EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
171
172void
173pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
174{
175 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
176 spin_lock(&pnfs_spinlock);
177 list_del(&ld_type->pnfs_tblid);
178 spin_unlock(&pnfs_spinlock);
179}
180EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400181
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400182/*
183 * pNFS client layout cache
184 */
185
Fred Isamancc6e5342011-01-06 11:36:28 +0000186/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000187void
Fred Isamancc6e5342011-01-06 11:36:28 +0000188get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400189{
Fred Isamancc6e5342011-01-06 11:36:28 +0000190 atomic_inc(&lo->plh_refcount);
191}
192
Benny Halevy636fb9c2011-05-22 19:51:33 +0300193static struct pnfs_layout_hdr *
194pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
195{
196 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
197 return ld->alloc_layout_hdr ? ld->alloc_layout_hdr(ino, gfp_flags) :
198 kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags);
199}
200
201static void
202pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
203{
204 struct pnfs_layoutdriver_type *ld = NFS_SERVER(lo->plh_inode)->pnfs_curr_ld;
Peng Tao9fa40752011-07-30 20:52:32 -0400205 put_rpccred(lo->plh_lc_cred);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300206 return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo);
207}
208
Fred Isamancc6e5342011-01-06 11:36:28 +0000209static void
210destroy_layout_hdr(struct pnfs_layout_hdr *lo)
211{
212 dprintk("%s: freeing layout cache %p\n", __func__, lo);
213 BUG_ON(!list_empty(&lo->plh_layouts));
214 NFS_I(lo->plh_inode)->layout = NULL;
Benny Halevy636fb9c2011-05-22 19:51:33 +0300215 pnfs_free_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400216}
217
218static void
219put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
220{
Fred Isamancc6e5342011-01-06 11:36:28 +0000221 if (atomic_dec_and_test(&lo->plh_refcount))
222 destroy_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400223}
224
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400225void
Fred Isamancc6e5342011-01-06 11:36:28 +0000226put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400227{
Fred Isamancc6e5342011-01-06 11:36:28 +0000228 struct inode *inode = lo->plh_inode;
229
230 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
231 destroy_layout_hdr(lo);
232 spin_unlock(&inode->i_lock);
233 }
Andy Adamson974cec82010-10-20 00:18:02 -0400234}
235
236static void
237init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
238{
Fred Isaman566052c2011-01-06 11:36:20 +0000239 INIT_LIST_HEAD(&lseg->pls_list);
Peng Taoa9bae562011-07-30 20:52:33 -0400240 INIT_LIST_HEAD(&lseg->pls_lc_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000241 atomic_set(&lseg->pls_refcount, 1);
242 smp_mb();
243 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000244 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400245}
246
Fred Isaman4541d162011-01-06 11:36:23 +0000247static void free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400248{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000249 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400250
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400251 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Fred Isaman52fabd72011-01-06 11:36:18 +0000252 /* Matched by get_layout_hdr in pnfs_insert_layout */
Fred Isamancc6e5342011-01-06 11:36:28 +0000253 put_layout_hdr(NFS_I(ino)->layout);
Andy Adamson974cec82010-10-20 00:18:02 -0400254}
255
Fred Isamand684d2a2011-03-01 01:34:13 +0000256static void
257put_lseg_common(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400258{
Fred Isamand684d2a2011-03-01 01:34:13 +0000259 struct inode *inode = lseg->pls_layout->plh_inode;
260
Benny Halevyd20581a2011-05-22 19:52:03 +0300261 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000262 list_del_init(&lseg->pls_list);
263 if (list_empty(&lseg->pls_layout->plh_segs)) {
264 set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
265 /* Matched by initial refcount set in alloc_init_layout_hdr */
266 put_layout_hdr_locked(lseg->pls_layout);
267 }
268 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
269}
270
Fred Isamanbae724e2011-03-01 01:34:15 +0000271void
Fred Isamand684d2a2011-03-01 01:34:13 +0000272put_lseg(struct pnfs_layout_segment *lseg)
273{
274 struct inode *inode;
275
276 if (!lseg)
277 return;
278
Fred Isaman4541d162011-01-06 11:36:23 +0000279 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
280 atomic_read(&lseg->pls_refcount),
281 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000282 inode = lseg->pls_layout->plh_inode;
283 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
284 LIST_HEAD(free_me);
Andy Adamson974cec82010-10-20 00:18:02 -0400285
Fred Isamand684d2a2011-03-01 01:34:13 +0000286 put_lseg_common(lseg);
287 list_add(&lseg->pls_list, &free_me);
288 spin_unlock(&inode->i_lock);
289 pnfs_free_lseg_list(&free_me);
Fred Isaman4541d162011-01-06 11:36:23 +0000290 }
Andy Adamson974cec82010-10-20 00:18:02 -0400291}
Fred Isamane0c2b382011-03-23 13:27:53 +0000292EXPORT_SYMBOL_GPL(put_lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400293
Benny Halevyfb3296e2011-05-22 19:47:26 +0300294static inline u64
295end_offset(u64 start, u64 len)
Fred Isaman4541d162011-01-06 11:36:23 +0000296{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300297 u64 end;
298
299 end = start + len;
300 return end >= start ? end : NFS4_MAX_UINT64;
301}
302
303/* last octet in a range */
304static inline u64
305last_byte_offset(u64 start, u64 len)
306{
307 u64 end;
308
309 BUG_ON(!len);
310 end = start + len;
311 return end > start ? end - 1 : NFS4_MAX_UINT64;
312}
313
314/*
315 * is l2 fully contained in l1?
316 * start1 end1
317 * [----------------------------------)
318 * start2 end2
319 * [----------------)
320 */
321static inline int
322lo_seg_contained(struct pnfs_layout_range *l1,
323 struct pnfs_layout_range *l2)
324{
325 u64 start1 = l1->offset;
326 u64 end1 = end_offset(start1, l1->length);
327 u64 start2 = l2->offset;
328 u64 end2 = end_offset(start2, l2->length);
329
330 return (start1 <= start2) && (end1 >= end2);
331}
332
333/*
334 * is l1 and l2 intersecting?
335 * start1 end1
336 * [----------------------------------)
337 * start2 end2
338 * [----------------)
339 */
340static inline int
341lo_seg_intersecting(struct pnfs_layout_range *l1,
342 struct pnfs_layout_range *l2)
343{
344 u64 start1 = l1->offset;
345 u64 end1 = end_offset(start1, l1->length);
346 u64 start2 = l2->offset;
347 u64 end2 = end_offset(start2, l2->length);
348
349 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
350 (end2 == NFS4_MAX_UINT64 || end2 > start1);
351}
352
Fred Isaman4541d162011-01-06 11:36:23 +0000353static bool
Benny Halevy778b5502011-05-22 19:48:02 +0300354should_free_lseg(struct pnfs_layout_range *lseg_range,
355 struct pnfs_layout_range *recall_range)
Fred Isaman4541d162011-01-06 11:36:23 +0000356{
Benny Halevy778b5502011-05-22 19:48:02 +0300357 return (recall_range->iomode == IOMODE_ANY ||
358 lseg_range->iomode == recall_range->iomode) &&
359 lo_seg_intersecting(lseg_range, recall_range);
Fred Isaman4541d162011-01-06 11:36:23 +0000360}
361
362/* Returns 1 if lseg is removed from list, 0 otherwise */
363static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
364 struct list_head *tmp_list)
365{
366 int rv = 0;
367
368 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
369 /* Remove the reference keeping the lseg in the
370 * list. It will now be removed when all
371 * outstanding io is finished.
372 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000373 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
374 atomic_read(&lseg->pls_refcount));
375 if (atomic_dec_and_test(&lseg->pls_refcount)) {
376 put_lseg_common(lseg);
377 list_add(&lseg->pls_list, tmp_list);
378 rv = 1;
379 }
Fred Isaman4541d162011-01-06 11:36:23 +0000380 }
381 return rv;
382}
383
384/* Returns count of number of matching invalid lsegs remaining in list
385 * after call.
386 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000387int
Fred Isaman4541d162011-01-06 11:36:23 +0000388mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
389 struct list_head *tmp_list,
Benny Halevy778b5502011-05-22 19:48:02 +0300390 struct pnfs_layout_range *recall_range)
Andy Adamson974cec82010-10-20 00:18:02 -0400391{
392 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000393 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400394
395 dprintk("%s:Begin lo %p\n", __func__, lo);
396
Fred Isaman38511722011-02-03 18:28:50 +0000397 if (list_empty(&lo->plh_segs)) {
398 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
399 put_layout_hdr_locked(lo);
400 return 0;
401 }
Fred Isaman4541d162011-01-06 11:36:23 +0000402 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
Benny Halevy778b5502011-05-22 19:48:02 +0300403 if (!recall_range ||
404 should_free_lseg(&lseg->pls_range, recall_range)) {
Fred Isaman4541d162011-01-06 11:36:23 +0000405 dprintk("%s: freeing lseg %p iomode %d "
406 "offset %llu length %llu\n", __func__,
407 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
408 lseg->pls_range.length);
409 invalid++;
410 removed += mark_lseg_invalid(lseg, tmp_list);
411 }
412 dprintk("%s:Return %i\n", __func__, invalid - removed);
413 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400414}
415
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000416/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000417void
Fred Isaman4541d162011-01-06 11:36:23 +0000418pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400419{
Fred Isaman4541d162011-01-06 11:36:23 +0000420 struct pnfs_layout_segment *lseg, *tmp;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000421 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400422
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000423 if (list_empty(free_me))
424 return;
425
426 lo = list_first_entry(free_me, struct pnfs_layout_segment,
427 pls_list)->pls_layout;
428
429 if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
430 struct nfs_client *clp;
431
432 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
433 spin_lock(&clp->cl_lock);
434 list_del_init(&lo->plh_layouts);
435 spin_unlock(&clp->cl_lock);
436 }
Fred Isaman4541d162011-01-06 11:36:23 +0000437 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000438 list_del(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000439 free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400440 }
441}
442
Benny Halevye5e94012010-10-20 00:18:01 -0400443void
444pnfs_destroy_layout(struct nfs_inode *nfsi)
445{
446 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400447 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400448
449 spin_lock(&nfsi->vfs_inode.i_lock);
450 lo = nfsi->layout;
451 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000452 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Benny Halevy778b5502011-05-22 19:48:02 +0300453 mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
Benny Halevye5e94012010-10-20 00:18:01 -0400454 }
455 spin_unlock(&nfsi->vfs_inode.i_lock);
Andy Adamson974cec82010-10-20 00:18:02 -0400456 pnfs_free_lseg_list(&tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400457}
Andy Adamson041245c2012-04-27 17:53:53 -0400458EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
Benny Halevye5e94012010-10-20 00:18:01 -0400459
Andy Adamson974cec82010-10-20 00:18:02 -0400460/*
461 * Called by the state manger to remove all layouts established under an
462 * expired lease.
463 */
464void
465pnfs_destroy_all_layouts(struct nfs_client *clp)
466{
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400467 struct nfs_server *server;
Andy Adamson974cec82010-10-20 00:18:02 -0400468 struct pnfs_layout_hdr *lo;
469 LIST_HEAD(tmp_list);
470
Andy Adamsonc47abcf2011-06-15 17:52:40 -0400471 nfs4_deviceid_mark_client_invalid(clp);
472 nfs4_deviceid_purge_client(clp);
473
Andy Adamson974cec82010-10-20 00:18:02 -0400474 spin_lock(&clp->cl_lock);
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400475 rcu_read_lock();
476 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
477 if (!list_empty(&server->layouts))
478 list_splice_init(&server->layouts, &tmp_list);
479 }
480 rcu_read_unlock();
Andy Adamson974cec82010-10-20 00:18:02 -0400481 spin_unlock(&clp->cl_lock);
482
483 while (!list_empty(&tmp_list)) {
484 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000485 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400486 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000487 lo->plh_inode->i_ino);
Andy Adamson2887fe42011-05-11 01:19:58 -0400488 list_del_init(&lo->plh_layouts);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000489 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400490 }
491}
492
Fred Isamanfd6002e2011-01-06 11:36:22 +0000493/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000494void
495pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
496 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400497{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000498 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400499
Trond Myklebust2d2f24a2012-03-04 18:13:57 -0500500 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
501 newseq = be32_to_cpu(new->seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000502 if ((int)(newseq - oldseq) > 0) {
Trond Myklebustf597c532012-03-04 18:13:56 -0500503 nfs4_stateid_copy(&lo->plh_stateid, new);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000504 if (update_barrier) {
Trond Myklebust2d2f24a2012-03-04 18:13:57 -0500505 u32 new_barrier = be32_to_cpu(new->seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000506
507 if ((int)(new_barrier - lo->plh_barrier))
508 lo->plh_barrier = new_barrier;
509 } else {
510 /* Because of wraparound, we want to keep the barrier
511 * "close" to the current seqids. It needs to be
512 * within 2**31 to count as "behind", so if it
513 * gets too near that limit, give us a litle leeway
514 * and bring it to within 2**30.
515 * NOTE - and yes, this is all unsigned arithmetic.
516 */
517 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
518 lo->plh_barrier = newseq - (1 << 30);
519 }
520 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400521}
522
Fred Isamancf7d63f2011-01-06 11:36:25 +0000523/* lget is set to 1 if called from inside send_layoutget call chain */
524static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000525pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
526 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000527{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000528 if ((stateid) &&
Trond Myklebust2d2f24a2012-03-04 18:13:57 -0500529 (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0)
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000530 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000531 return lo->plh_block_lgets ||
Fred Isaman38511722011-02-03 18:28:50 +0000532 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
Fred Isamanf7e89172011-01-06 11:36:32 +0000533 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000534 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000535 (atomic_read(&lo->plh_outstanding) > lget));
536}
537
Fred Isamanfd6002e2011-01-06 11:36:22 +0000538int
539pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
540 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400541{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000542 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400543
544 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000545 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000546 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000547 status = -EAGAIN;
548 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000549 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400550
Fred Isamanfd6002e2011-01-06 11:36:22 +0000551 do {
552 seq = read_seqbegin(&open_state->seqlock);
Trond Myklebustf597c532012-03-04 18:13:56 -0500553 nfs4_stateid_copy(dst, &open_state->stateid);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000554 } while (read_seqretry(&open_state->seqlock, seq));
555 } else
Trond Myklebustf597c532012-03-04 18:13:56 -0500556 nfs4_stateid_copy(dst, &lo->plh_stateid);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000557 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400558 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000559 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400560}
561
562/*
563* Get layout from server.
564* for now, assume that whole file layouts are requested.
565* arg->offset: 0
566* arg->length: all ones
567*/
Benny Halevye5e94012010-10-20 00:18:01 -0400568static struct pnfs_layout_segment *
569send_layoutget(struct pnfs_layout_hdr *lo,
570 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +0300571 struct pnfs_layout_range *range,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400572 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400573{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000574 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400575 struct nfs_server *server = NFS_SERVER(ino);
576 struct nfs4_layoutget *lgp;
577 struct pnfs_layout_segment *lseg = NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400578 struct page **pages = NULL;
579 int i;
580 u32 max_resp_sz, max_pages;
Benny Halevye5e94012010-10-20 00:18:01 -0400581
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400582 dprintk("--> %s\n", __func__);
583
584 BUG_ON(ctx == NULL);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400585 lgp = kzalloc(sizeof(*lgp), gfp_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000586 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400587 return NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400588
589 /* allocate pages for xdr post processing */
590 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
Andy Adamsone5265a02012-04-14 03:56:35 -0400591 max_pages = nfs_page_array_len(0, max_resp_sz);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400592
Trond Myklebust7d9dea92012-01-20 18:57:02 -0500593 pages = kcalloc(max_pages, sizeof(struct page *), gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400594 if (!pages)
595 goto out_err_free;
596
597 for (i = 0; i < max_pages; i++) {
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400598 pages[i] = alloc_page(gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400599 if (!pages[i])
600 goto out_err_free;
601 }
602
Benny Halevyfb3296e2011-05-22 19:47:26 +0300603 lgp->args.minlength = PAGE_CACHE_SIZE;
604 if (lgp->args.minlength > range->length)
605 lgp->args.minlength = range->length;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400606 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
Benny Halevyfb3296e2011-05-22 19:47:26 +0300607 lgp->args.range = *range;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400608 lgp->args.type = server->pnfs_curr_ld->id;
609 lgp->args.inode = ino;
610 lgp->args.ctx = get_nfs_open_context(ctx);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400611 lgp->args.layout.pages = pages;
612 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400613 lgp->lsegpp = &lseg;
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400614 lgp->gfp_flags = gfp_flags;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400615
616 /* Synchronously retrieve layout information from server and
617 * store in lseg.
618 */
619 nfs4_proc_layoutget(lgp);
620 if (!lseg) {
621 /* remember that LAYOUTGET failed and suspend trying */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300622 set_bit(lo_fail_bit(range->iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400623 }
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400624
625 /* free xdr pages */
626 for (i = 0; i < max_pages; i++)
627 __free_page(pages[i]);
628 kfree(pages);
629
Andy Adamson974cec82010-10-20 00:18:02 -0400630 return lseg;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400631
632out_err_free:
633 /* free any allocated xdr pages, lgp as it's not used */
634 if (pages) {
635 for (i = 0; i < max_pages; i++) {
636 if (!pages[i])
637 break;
638 __free_page(pages[i]);
639 }
640 kfree(pages);
641 }
642 kfree(lgp);
643 return NULL;
Andy Adamson974cec82010-10-20 00:18:02 -0400644}
645
Benny Halevycbe82602011-05-22 19:52:37 +0300646/* Initiates a LAYOUTRETURN(FILE) */
647int
648_pnfs_return_layout(struct inode *ino)
649{
650 struct pnfs_layout_hdr *lo = NULL;
651 struct nfs_inode *nfsi = NFS_I(ino);
652 LIST_HEAD(tmp_list);
653 struct nfs4_layoutreturn *lrp;
654 nfs4_stateid stateid;
655 int status = 0;
656
657 dprintk("--> %s\n", __func__);
658
659 spin_lock(&ino->i_lock);
660 lo = nfsi->layout;
Fred Isamana2e1d4f2011-06-13 18:54:53 -0400661 if (!lo) {
Benny Halevycbe82602011-05-22 19:52:37 +0300662 spin_unlock(&ino->i_lock);
Fred Isamana2e1d4f2011-06-13 18:54:53 -0400663 dprintk("%s: no layout to return\n", __func__);
664 return status;
Benny Halevycbe82602011-05-22 19:52:37 +0300665 }
666 stateid = nfsi->layout->plh_stateid;
667 /* Reference matched in nfs4_layoutreturn_release */
668 get_layout_hdr(lo);
Fred Isamanea0ded72011-06-15 12:31:02 -0400669 mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
670 lo->plh_block_lgets++;
Benny Halevycbe82602011-05-22 19:52:37 +0300671 spin_unlock(&ino->i_lock);
672 pnfs_free_lseg_list(&tmp_list);
673
674 WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags));
675
676 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
677 if (unlikely(lrp == NULL)) {
678 status = -ENOMEM;
Fred Isaman9e2dfdb2011-06-15 14:32:02 -0400679 set_bit(NFS_LAYOUT_RW_FAILED, &lo->plh_flags);
680 set_bit(NFS_LAYOUT_RO_FAILED, &lo->plh_flags);
Benny Halevy1ed3a852011-06-15 11:39:57 -0400681 put_layout_hdr(lo);
Benny Halevycbe82602011-05-22 19:52:37 +0300682 goto out;
683 }
684
685 lrp->args.stateid = stateid;
686 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
687 lrp->args.inode = ino;
Trond Myklebusta56aaa02011-06-15 11:59:10 -0400688 lrp->args.layout = lo;
Benny Halevycbe82602011-05-22 19:52:37 +0300689 lrp->clp = NFS_SERVER(ino)->nfs_client;
690
691 status = nfs4_proc_layoutreturn(lrp);
692out:
693 dprintk("<-- %s status: %d\n", __func__, status);
694 return status;
695}
Andy Adamson0a57cda2012-04-27 17:53:50 -0400696EXPORT_SYMBOL_GPL(_pnfs_return_layout);
Benny Halevycbe82602011-05-22 19:52:37 +0300697
Fred Isamanf7e89172011-01-06 11:36:32 +0000698bool pnfs_roc(struct inode *ino)
699{
700 struct pnfs_layout_hdr *lo;
701 struct pnfs_layout_segment *lseg, *tmp;
702 LIST_HEAD(tmp_list);
703 bool found = false;
704
705 spin_lock(&ino->i_lock);
706 lo = NFS_I(ino)->layout;
707 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
708 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
709 goto out_nolayout;
710 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
711 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
712 mark_lseg_invalid(lseg, &tmp_list);
713 found = true;
714 }
715 if (!found)
716 goto out_nolayout;
717 lo->plh_block_lgets++;
718 get_layout_hdr(lo); /* matched in pnfs_roc_release */
719 spin_unlock(&ino->i_lock);
720 pnfs_free_lseg_list(&tmp_list);
721 return true;
722
723out_nolayout:
724 spin_unlock(&ino->i_lock);
725 return false;
726}
727
728void pnfs_roc_release(struct inode *ino)
729{
730 struct pnfs_layout_hdr *lo;
731
732 spin_lock(&ino->i_lock);
733 lo = NFS_I(ino)->layout;
734 lo->plh_block_lgets--;
735 put_layout_hdr_locked(lo);
736 spin_unlock(&ino->i_lock);
737}
738
739void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
740{
741 struct pnfs_layout_hdr *lo;
742
743 spin_lock(&ino->i_lock);
744 lo = NFS_I(ino)->layout;
745 if ((int)(barrier - lo->plh_barrier) > 0)
746 lo->plh_barrier = barrier;
747 spin_unlock(&ino->i_lock);
748}
749
750bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
751{
752 struct nfs_inode *nfsi = NFS_I(ino);
753 struct pnfs_layout_segment *lseg;
754 bool found = false;
755
756 spin_lock(&ino->i_lock);
757 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
758 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
759 found = true;
760 break;
761 }
762 if (!found) {
763 struct pnfs_layout_hdr *lo = nfsi->layout;
Trond Myklebust2d2f24a2012-03-04 18:13:57 -0500764 u32 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
Fred Isamanf7e89172011-01-06 11:36:32 +0000765
766 /* Since close does not return a layout stateid for use as
767 * a barrier, we choose the worst-case barrier.
768 */
769 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
770 }
771 spin_unlock(&ino->i_lock);
772 return found;
773}
774
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400775/*
776 * Compare two layout segments for sorting into layout cache.
777 * We want to preferentially return RW over RO layouts, so ensure those
778 * are seen first.
779 */
780static s64
Benny Halevyfb3296e2011-05-22 19:47:26 +0300781cmp_layout(struct pnfs_layout_range *l1,
782 struct pnfs_layout_range *l2)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400783{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300784 s64 d;
785
786 /* high offset > low offset */
787 d = l1->offset - l2->offset;
788 if (d)
789 return d;
790
791 /* short length > long length */
792 d = l2->length - l1->length;
793 if (d)
794 return d;
795
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400796 /* read > read/write */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300797 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400798}
799
Andy Adamson974cec82010-10-20 00:18:02 -0400800static void
801pnfs_insert_layout(struct pnfs_layout_hdr *lo,
802 struct pnfs_layout_segment *lseg)
803{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400804 struct pnfs_layout_segment *lp;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400805
Andy Adamson974cec82010-10-20 00:18:02 -0400806 dprintk("%s:Begin\n", __func__);
807
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000808 assert_spin_locked(&lo->plh_inode->i_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000809 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Benny Halevyfb3296e2011-05-22 19:47:26 +0300810 if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400811 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000812 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400813 dprintk("%s: inserted lseg %p "
814 "iomode %d offset %llu length %llu before "
815 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000816 __func__, lseg, lseg->pls_range.iomode,
817 lseg->pls_range.offset, lseg->pls_range.length,
818 lp, lp->pls_range.iomode, lp->pls_range.offset,
819 lp->pls_range.length);
Benny Halevyfb3296e2011-05-22 19:47:26 +0300820 goto out;
Andy Adamson974cec82010-10-20 00:18:02 -0400821 }
Benny Halevyfb3296e2011-05-22 19:47:26 +0300822 list_add_tail(&lseg->pls_list, &lo->plh_segs);
823 dprintk("%s: inserted lseg %p "
824 "iomode %d offset %llu length %llu at tail\n",
825 __func__, lseg, lseg->pls_range.iomode,
826 lseg->pls_range.offset, lseg->pls_range.length);
827out:
Fred Isamancc6e5342011-01-06 11:36:28 +0000828 get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400829
830 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400831}
832
833static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -0400834alloc_init_layout_hdr(struct inode *ino,
835 struct nfs_open_context *ctx,
836 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400837{
838 struct pnfs_layout_hdr *lo;
839
Benny Halevy636fb9c2011-05-22 19:51:33 +0300840 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400841 if (!lo)
842 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000843 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000844 INIT_LIST_HEAD(&lo->plh_layouts);
845 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000846 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000847 lo->plh_inode = ino;
Peng Tao9fa40752011-07-30 20:52:32 -0400848 lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
Benny Halevye5e94012010-10-20 00:18:01 -0400849 return lo;
850}
851
852static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -0400853pnfs_find_alloc_layout(struct inode *ino,
854 struct nfs_open_context *ctx,
855 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400856{
857 struct nfs_inode *nfsi = NFS_I(ino);
858 struct pnfs_layout_hdr *new = NULL;
859
860 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
861
862 assert_spin_locked(&ino->i_lock);
Fred Isaman4541d162011-01-06 11:36:23 +0000863 if (nfsi->layout) {
864 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
865 return NULL;
866 else
867 return nfsi->layout;
868 }
Benny Halevye5e94012010-10-20 00:18:01 -0400869 spin_unlock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -0400870 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400871 spin_lock(&ino->i_lock);
872
873 if (likely(nfsi->layout == NULL)) /* Won the race? */
874 nfsi->layout = new;
875 else
Benny Halevy636fb9c2011-05-22 19:51:33 +0300876 pnfs_free_layout_hdr(new);
Benny Halevye5e94012010-10-20 00:18:01 -0400877 return nfsi->layout;
878}
879
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400880/*
881 * iomode matching rules:
882 * iomode lseg match
883 * ----- ----- -----
884 * ANY READ true
885 * ANY RW true
886 * RW READ false
887 * RW RW true
888 * READ READ true
889 * READ RW true
890 */
891static int
Benny Halevyfb3296e2011-05-22 19:47:26 +0300892is_matching_lseg(struct pnfs_layout_range *ls_range,
893 struct pnfs_layout_range *range)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400894{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300895 struct pnfs_layout_range range1;
896
897 if ((range->iomode == IOMODE_RW &&
898 ls_range->iomode != IOMODE_RW) ||
899 !lo_seg_intersecting(ls_range, range))
900 return 0;
901
902 /* range1 covers only the first byte in the range */
903 range1 = *range;
904 range1.length = 1;
905 return lo_seg_contained(ls_range, &range1);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400906}
907
908/*
909 * lookup range in layout
910 */
Benny Halevye5e94012010-10-20 00:18:01 -0400911static struct pnfs_layout_segment *
Benny Halevyfb3296e2011-05-22 19:47:26 +0300912pnfs_find_lseg(struct pnfs_layout_hdr *lo,
913 struct pnfs_layout_range *range)
Benny Halevye5e94012010-10-20 00:18:01 -0400914{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400915 struct pnfs_layout_segment *lseg, *ret = NULL;
916
917 dprintk("%s:Begin\n", __func__);
918
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000919 assert_spin_locked(&lo->plh_inode->i_lock);
920 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000921 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
Benny Halevyfb3296e2011-05-22 19:47:26 +0300922 is_matching_lseg(&lseg->pls_range, range)) {
Fred Isamand684d2a2011-03-01 01:34:13 +0000923 ret = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400924 break;
925 }
Benny Halevyd771e3a2011-06-14 16:30:16 -0400926 if (lseg->pls_range.offset > range->offset)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400927 break;
928 }
929
930 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000931 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400932 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400933}
934
935/*
936 * Layout segment is retreived from the server if not cached.
937 * The appropriate layout segment is referenced and returned to the caller.
938 */
Andy Adamson7c24d942011-06-13 18:22:38 -0400939struct pnfs_layout_segment *
Benny Halevye5e94012010-10-20 00:18:01 -0400940pnfs_update_layout(struct inode *ino,
941 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +0300942 loff_t pos,
943 u64 count,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400944 enum pnfs_iomode iomode,
945 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400946{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300947 struct pnfs_layout_range arg = {
948 .iomode = iomode,
949 .offset = pos,
950 .length = count,
951 };
Benny Halevy707ed5f2011-05-22 19:47:46 +0300952 unsigned pg_offset;
Benny Halevye5e94012010-10-20 00:18:01 -0400953 struct nfs_inode *nfsi = NFS_I(ino);
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400954 struct nfs_server *server = NFS_SERVER(ino);
955 struct nfs_client *clp = server->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -0400956 struct pnfs_layout_hdr *lo;
957 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000958 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -0400959
960 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
961 return NULL;
962 spin_lock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -0400963 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400964 if (lo == NULL) {
965 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
966 goto out_unlock;
967 }
968
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000969 /* Do we even need to bother with this? */
Trond Myklebusta59c30a2012-03-01 11:17:47 -0500970 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000971 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400972 goto out_unlock;
973 }
974
975 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000976 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400977 goto out_unlock;
978
Andy Adamson568e8c42011-03-01 01:34:22 +0000979 /* Check to see if the layout for the given range already exists */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300980 lseg = pnfs_find_lseg(lo, &arg);
Andy Adamson568e8c42011-03-01 01:34:22 +0000981 if (lseg)
982 goto out_unlock;
983
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000984 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +0000985 goto out_unlock;
986 atomic_inc(&lo->plh_outstanding);
987
Fred Isamancc6e5342011-01-06 11:36:28 +0000988 get_layout_hdr(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000989 if (list_empty(&lo->plh_segs))
990 first = true;
991 spin_unlock(&ino->i_lock);
992 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +0000993 /* The lo must be on the clp list if there is any
994 * chance of a CB_LAYOUTRECALL(FILE) coming in.
995 */
996 spin_lock(&clp->cl_lock);
997 BUG_ON(!list_empty(&lo->plh_layouts));
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400998 list_add_tail(&lo->plh_layouts, &server->layouts);
Fred Isaman2130ff62011-01-06 11:36:26 +0000999 spin_unlock(&clp->cl_lock);
1000 }
Benny Halevye5e94012010-10-20 00:18:01 -04001001
Benny Halevy707ed5f2011-05-22 19:47:46 +03001002 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1003 if (pg_offset) {
1004 arg.offset -= pg_offset;
1005 arg.length += pg_offset;
1006 }
Andy Adamson7c24d942011-06-13 18:22:38 -04001007 if (arg.length != NFS4_MAX_UINT64)
1008 arg.length = PAGE_CACHE_ALIGN(arg.length);
Benny Halevy707ed5f2011-05-22 19:47:46 +03001009
Benny Halevyfb3296e2011-05-22 19:47:26 +03001010 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
Fred Isamanf49f9ba2011-02-03 18:28:52 +00001011 if (!lseg && first) {
1012 spin_lock(&clp->cl_lock);
1013 list_del_init(&lo->plh_layouts);
1014 spin_unlock(&clp->cl_lock);
Fred Isaman2130ff62011-01-06 11:36:26 +00001015 }
Fred Isamancf7d63f2011-01-06 11:36:25 +00001016 atomic_dec(&lo->plh_outstanding);
Fred Isamancc6e5342011-01-06 11:36:28 +00001017 put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -04001018out:
1019 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Andy Adamsonbf9c1382011-03-01 01:34:07 +00001020 nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
Benny Halevye5e94012010-10-20 00:18:01 -04001021 return lseg;
1022out_unlock:
1023 spin_unlock(&ino->i_lock);
1024 goto out;
1025}
Andy Adamson7c24d942011-06-13 18:22:38 -04001026EXPORT_SYMBOL_GPL(pnfs_update_layout);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001027
1028int
1029pnfs_layout_process(struct nfs4_layoutget *lgp)
1030{
1031 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1032 struct nfs4_layoutget_res *res = &lgp->res;
1033 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001034 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001035 int status = 0;
1036
1037 /* Inject layout blob into I/O device driver */
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001038 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001039 if (!lseg || IS_ERR(lseg)) {
1040 if (!lseg)
1041 status = -ENOMEM;
1042 else
1043 status = PTR_ERR(lseg);
1044 dprintk("%s: Could not allocate layout: error %d\n",
1045 __func__, status);
1046 goto out;
1047 }
1048
1049 spin_lock(&ino->i_lock);
Trond Myklebusta59c30a2012-03-01 11:17:47 -05001050 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001051 dprintk("%s forget reply due to recall\n", __func__);
1052 goto out_forget_reply;
1053 }
1054
1055 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
1056 dprintk("%s forget reply due to state\n", __func__);
1057 goto out_forget_reply;
1058 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001059 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +00001060 lseg->pls_range = res->range;
Fred Isamand684d2a2011-03-01 01:34:13 +00001061 *lgp->lsegpp = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001062 pnfs_insert_layout(lo, lseg);
1063
Fred Isamanf7e89172011-01-06 11:36:32 +00001064 if (res->return_on_close) {
1065 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1066 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
1067 }
1068
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001069 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001070 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001071 spin_unlock(&ino->i_lock);
1072out:
1073 return status;
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001074
1075out_forget_reply:
1076 spin_unlock(&ino->i_lock);
1077 lseg->pls_layout = lo;
1078 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1079 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001080}
1081
Trond Myklebustd8007d42011-06-10 13:30:23 -04001082void
1083pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1084{
1085 BUG_ON(pgio->pg_lseg != NULL);
1086
Fred Isaman1825a0d2012-04-20 19:55:31 -04001087 if (req->wb_offset != req->wb_pgbase) {
1088 nfs_pageio_reset_read_mds(pgio);
1089 return;
1090 }
Trond Myklebustd8007d42011-06-10 13:30:23 -04001091 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1092 req->wb_context,
1093 req_offset(req),
1094 req->wb_bytes,
1095 IOMODE_READ,
1096 GFP_KERNEL);
Trond Myklebuste885de12011-06-10 13:30:23 -04001097 /* If no lseg, fall back to read through mds */
1098 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001099 nfs_pageio_reset_read_mds(pgio);
Trond Myklebuste885de12011-06-10 13:30:23 -04001100
Trond Myklebustd8007d42011-06-10 13:30:23 -04001101}
1102EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1103
1104void
1105pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1106{
1107 BUG_ON(pgio->pg_lseg != NULL);
1108
Fred Isaman1825a0d2012-04-20 19:55:31 -04001109 if (req->wb_offset != req->wb_pgbase) {
1110 nfs_pageio_reset_write_mds(pgio);
1111 return;
1112 }
Trond Myklebustd8007d42011-06-10 13:30:23 -04001113 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1114 req->wb_context,
1115 req_offset(req),
1116 req->wb_bytes,
1117 IOMODE_RW,
1118 GFP_NOFS);
Trond Myklebuste885de12011-06-10 13:30:23 -04001119 /* If no lseg, fall back to write through mds */
1120 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001121 nfs_pageio_reset_write_mds(pgio);
Trond Myklebustd8007d42011-06-10 13:30:23 -04001122}
1123EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1124
Benny Halevy18ad0a92011-05-25 21:03:56 +03001125bool
Fred Isaman061ae2e2012-04-20 14:47:48 -04001126pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode,
1127 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001128{
1129 struct nfs_server *server = NFS_SERVER(inode);
1130 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1131
1132 if (ld == NULL)
1133 return false;
Fred Isaman061ae2e2012-04-20 14:47:48 -04001134 nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops,
1135 server->rsize, 0);
Trond Myklebust1751c362011-06-10 13:30:23 -04001136 return true;
1137}
1138
1139bool
Fred Isaman061ae2e2012-04-20 14:47:48 -04001140pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode,
1141 int ioflags,
1142 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001143{
1144 struct nfs_server *server = NFS_SERVER(inode);
1145 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1146
1147 if (ld == NULL)
1148 return false;
Fred Isaman061ae2e2012-04-20 14:47:48 -04001149 nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops,
1150 server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001151 return true;
1152}
1153
1154bool
Benny Halevydfed2062011-05-25 20:25:22 +03001155pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1156 struct nfs_page *req)
Fred Isamanbae724e2011-03-01 01:34:15 +00001157{
Trond Myklebustd8007d42011-06-10 13:30:23 -04001158 if (pgio->pg_lseg == NULL)
1159 return nfs_generic_pg_test(pgio, prev, req);
Benny Halevy89a58e32011-05-25 20:54:40 +03001160
Trond Myklebust19982ba2011-06-10 13:30:23 -04001161 /*
1162 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1163 * Note that this test makes several assumptions:
1164 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1165 * is known to lie within the range.
1166 * - that the nfs_page being tested is known to be contiguous with the
1167 * previous nfs_page.
1168 * - Layout ranges are page aligned, so we only have to test the
1169 * start offset of the request.
1170 *
1171 * Please also note that 'end_offset' is actually the offset of the
1172 * first byte that lies outside the pnfs_layout_range. FIXME?
1173 *
1174 */
1175 return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
1176 pgio->pg_lseg->pls_range.length);
Fred Isamanbae724e2011-03-01 01:34:15 +00001177}
Benny Halevy89a58e32011-05-25 20:54:40 +03001178EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
Fred Isamanbae724e2011-03-01 01:34:15 +00001179
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001180int pnfs_write_done_resend_to_mds(struct inode *inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001181 struct list_head *head,
1182 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001183{
1184 struct nfs_pageio_descriptor pgio;
1185 LIST_HEAD(failed);
1186
1187 /* Resend all requests through the MDS */
Fred Isaman061ae2e2012-04-20 14:47:48 -04001188 nfs_pageio_init_write_mds(&pgio, inode, FLUSH_STABLE, compl_ops);
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001189 while (!list_empty(head)) {
1190 struct nfs_page *req = nfs_list_entry(head->next);
1191
1192 nfs_list_remove_request(req);
1193 if (!nfs_pageio_add_request(&pgio, req))
1194 nfs_list_add_request(req, &failed);
1195 }
1196 nfs_pageio_complete(&pgio);
1197
1198 if (!list_empty(&failed)) {
1199 /* For some reason our attempt to resend pages. Mark the
1200 * overall send request as having failed, and let
1201 * nfs_writeback_release_full deal with the error.
1202 */
1203 list_move(&failed, head);
1204 return -EIO;
1205 }
1206 return 0;
1207}
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001208EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001209
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001210static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
1211{
Fred Isamancd841602012-04-20 14:47:44 -04001212 struct nfs_pgio_header *hdr = data->header;
1213
1214 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1215 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001216 PNFS_LAYOUTRET_ON_ERROR) {
Fred Isamancd841602012-04-20 14:47:44 -04001217 clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
1218 pnfs_return_layout(hdr->inode);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001219 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001220 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1221 data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001222 &hdr->pages,
1223 hdr->completion_ops);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001224}
1225
Benny Halevyd20581a2011-05-22 19:52:03 +03001226/*
1227 * Called by non rpc-based layout drivers
1228 */
Peng Tao8ce160c2011-09-22 21:50:14 -04001229void pnfs_ld_write_done(struct nfs_write_data *data)
Fred Isaman94ad1c82011-03-01 01:34:14 +00001230{
Fred Isamancd841602012-04-20 14:47:44 -04001231 struct nfs_pgio_header *hdr = data->header;
1232
1233 if (!hdr->pnfs_error) {
Benny Halevyd20581a2011-05-22 19:52:03 +03001234 pnfs_set_layoutcommit(data);
Fred Isamancd841602012-04-20 14:47:44 -04001235 hdr->mds_ops->rpc_call_done(&data->task, data);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001236 } else
1237 pnfs_ld_handle_write_error(data);
Fred Isamancd841602012-04-20 14:47:44 -04001238 hdr->mds_ops->rpc_release(data);
Fred Isaman44b83792011-03-03 15:13:44 +00001239}
Benny Halevyd20581a2011-05-22 19:52:03 +03001240EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
Fred Isaman44b83792011-03-03 15:13:44 +00001241
Trond Myklebustdce81292011-07-13 15:59:19 -04001242static void
1243pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1244 struct nfs_write_data *data)
1245{
Fred Isamancd841602012-04-20 14:47:44 -04001246 struct nfs_pgio_header *hdr = data->header;
1247
Fred Isaman6c75dc02012-04-20 14:47:47 -04001248 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1249 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1250 nfs_pageio_reset_write_mds(desc);
1251 desc->pg_recoalesce = 1;
1252 }
Trond Myklebustdce81292011-07-13 15:59:19 -04001253 nfs_writedata_release(data);
1254}
1255
1256static enum pnfs_try_status
Andy Adamson0382b742011-03-03 15:13:45 +00001257pnfs_try_to_write_data(struct nfs_write_data *wdata,
Trond Myklebustdce81292011-07-13 15:59:19 -04001258 const struct rpc_call_ops *call_ops,
1259 struct pnfs_layout_segment *lseg,
1260 int how)
Andy Adamson0382b742011-03-03 15:13:45 +00001261{
Fred Isamancd841602012-04-20 14:47:44 -04001262 struct nfs_pgio_header *hdr = wdata->header;
1263 struct inode *inode = hdr->inode;
Andy Adamson0382b742011-03-03 15:13:45 +00001264 enum pnfs_try_status trypnfs;
1265 struct nfs_server *nfss = NFS_SERVER(inode);
1266
Fred Isamancd841602012-04-20 14:47:44 -04001267 hdr->mds_ops = call_ops;
Andy Adamson0382b742011-03-03 15:13:45 +00001268
1269 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1270 inode->i_ino, wdata->args.count, wdata->args.offset, how);
Andy Adamson0382b742011-03-03 15:13:45 +00001271 trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001272 if (trypnfs != PNFS_NOT_ATTEMPTED)
Andy Adamson0382b742011-03-03 15:13:45 +00001273 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
Andy Adamson0382b742011-03-03 15:13:45 +00001274 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1275 return trypnfs;
1276}
1277
Trond Myklebustdce81292011-07-13 15:59:19 -04001278static void
1279pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
1280{
1281 struct nfs_write_data *data;
1282 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1283 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1284
1285 desc->pg_lseg = NULL;
1286 while (!list_empty(head)) {
1287 enum pnfs_try_status trypnfs;
1288
Fred Isaman6c75dc02012-04-20 14:47:47 -04001289 data = list_first_entry(head, struct nfs_write_data, list);
Trond Myklebustdce81292011-07-13 15:59:19 -04001290 list_del_init(&data->list);
1291
1292 trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
1293 if (trypnfs == PNFS_NOT_ATTEMPTED)
1294 pnfs_write_through_mds(desc, data);
1295 }
1296 put_lseg(lseg);
1297}
1298
Fred Isaman6c75dc02012-04-20 14:47:47 -04001299static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
1300{
1301 put_lseg(hdr->lseg);
1302 nfs_writehdr_free(hdr);
1303}
1304
Trond Myklebustdce81292011-07-13 15:59:19 -04001305int
1306pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1307{
Fred Isaman6c75dc02012-04-20 14:47:47 -04001308 struct nfs_write_header *whdr;
1309 struct nfs_pgio_header *hdr;
Trond Myklebustdce81292011-07-13 15:59:19 -04001310 int ret;
1311
Fred Isaman6c75dc02012-04-20 14:47:47 -04001312 whdr = nfs_writehdr_alloc();
1313 if (!whdr) {
Trond Myklebust9b5415b2012-04-27 14:31:47 -04001314 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001315 put_lseg(desc->pg_lseg);
1316 desc->pg_lseg = NULL;
1317 return -ENOMEM;
1318 }
1319 hdr = &whdr->header;
1320 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1321 hdr->lseg = get_lseg(desc->pg_lseg);
1322 atomic_inc(&hdr->refcnt);
1323 ret = nfs_generic_flush(desc, hdr);
Trond Myklebustdce81292011-07-13 15:59:19 -04001324 if (ret != 0) {
1325 put_lseg(desc->pg_lseg);
1326 desc->pg_lseg = NULL;
Fred Isaman6c75dc02012-04-20 14:47:47 -04001327 } else
1328 pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
1329 if (atomic_dec_and_test(&hdr->refcnt))
Fred Isaman061ae2e2012-04-20 14:47:48 -04001330 hdr->completion_ops->completion(hdr);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001331 return ret;
Trond Myklebustdce81292011-07-13 15:59:19 -04001332}
1333EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
1334
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001335int pnfs_read_done_resend_to_mds(struct inode *inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001336 struct list_head *head,
1337 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust62e4a762011-11-10 14:30:37 -05001338{
1339 struct nfs_pageio_descriptor pgio;
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001340 LIST_HEAD(failed);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001341
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001342 /* Resend all requests through the MDS */
Fred Isaman061ae2e2012-04-20 14:47:48 -04001343 nfs_pageio_init_read_mds(&pgio, inode, compl_ops);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001344 while (!list_empty(head)) {
1345 struct nfs_page *req = nfs_list_entry(head->next);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001346
1347 nfs_list_remove_request(req);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001348 if (!nfs_pageio_add_request(&pgio, req))
1349 nfs_list_add_request(req, &failed);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001350 }
1351 nfs_pageio_complete(&pgio);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001352
1353 if (!list_empty(&failed)) {
1354 list_move(&failed, head);
1355 return -EIO;
1356 }
1357 return 0;
1358}
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001359EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001360
1361static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
1362{
Fred Isamancd841602012-04-20 14:47:44 -04001363 struct nfs_pgio_header *hdr = data->header;
1364
1365 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
1366 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001367 PNFS_LAYOUTRET_ON_ERROR) {
Fred Isamancd841602012-04-20 14:47:44 -04001368 clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
1369 pnfs_return_layout(hdr->inode);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001370 }
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001371 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1372 data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001373 &hdr->pages,
1374 hdr->completion_ops);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001375}
1376
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001377/*
Benny Halevyd20581a2011-05-22 19:52:03 +03001378 * Called by non rpc-based layout drivers
1379 */
Peng Tao9b7eecd2011-09-22 21:50:15 -04001380void pnfs_ld_read_done(struct nfs_read_data *data)
Benny Halevyd20581a2011-05-22 19:52:03 +03001381{
Fred Isamancd841602012-04-20 14:47:44 -04001382 struct nfs_pgio_header *hdr = data->header;
1383
1384 if (likely(!hdr->pnfs_error)) {
Benny Halevyd20581a2011-05-22 19:52:03 +03001385 __nfs4_read_done_cb(data);
Fred Isamancd841602012-04-20 14:47:44 -04001386 hdr->mds_ops->rpc_call_done(&data->task, data);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001387 } else
1388 pnfs_ld_handle_read_error(data);
Fred Isamancd841602012-04-20 14:47:44 -04001389 hdr->mds_ops->rpc_release(data);
Benny Halevyd20581a2011-05-22 19:52:03 +03001390}
1391EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
1392
Trond Myklebust493292d2011-07-13 15:58:28 -04001393static void
1394pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1395 struct nfs_read_data *data)
1396{
Fred Isamancd841602012-04-20 14:47:44 -04001397 struct nfs_pgio_header *hdr = data->header;
1398
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001399 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1400 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1401 nfs_pageio_reset_read_mds(desc);
1402 desc->pg_recoalesce = 1;
1403 }
Trond Myklebust493292d2011-07-13 15:58:28 -04001404 nfs_readdata_release(data);
1405}
1406
Benny Halevyd20581a2011-05-22 19:52:03 +03001407/*
Andy Adamson64419a92011-03-01 01:34:16 +00001408 * Call the appropriate parallel I/O subsystem read function.
1409 */
Trond Myklebust493292d2011-07-13 15:58:28 -04001410static enum pnfs_try_status
Andy Adamson64419a92011-03-01 01:34:16 +00001411pnfs_try_to_read_data(struct nfs_read_data *rdata,
Trond Myklebust493292d2011-07-13 15:58:28 -04001412 const struct rpc_call_ops *call_ops,
1413 struct pnfs_layout_segment *lseg)
Andy Adamson64419a92011-03-01 01:34:16 +00001414{
Fred Isamancd841602012-04-20 14:47:44 -04001415 struct nfs_pgio_header *hdr = rdata->header;
1416 struct inode *inode = hdr->inode;
Andy Adamson64419a92011-03-01 01:34:16 +00001417 struct nfs_server *nfss = NFS_SERVER(inode);
1418 enum pnfs_try_status trypnfs;
1419
Fred Isamancd841602012-04-20 14:47:44 -04001420 hdr->mds_ops = call_ops;
Andy Adamson64419a92011-03-01 01:34:16 +00001421
1422 dprintk("%s: Reading ino:%lu %u@%llu\n",
1423 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
1424
1425 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001426 if (trypnfs != PNFS_NOT_ATTEMPTED)
Andy Adamson64419a92011-03-01 01:34:16 +00001427 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
Andy Adamson64419a92011-03-01 01:34:16 +00001428 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1429 return trypnfs;
1430}
Andy Adamson863a3c62011-03-23 13:27:54 +00001431
Trond Myklebust493292d2011-07-13 15:58:28 -04001432static void
1433pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
1434{
1435 struct nfs_read_data *data;
1436 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1437 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1438
1439 desc->pg_lseg = NULL;
1440 while (!list_empty(head)) {
1441 enum pnfs_try_status trypnfs;
1442
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001443 data = list_first_entry(head, struct nfs_read_data, list);
Trond Myklebust493292d2011-07-13 15:58:28 -04001444 list_del_init(&data->list);
1445
1446 trypnfs = pnfs_try_to_read_data(data, call_ops, lseg);
1447 if (trypnfs == PNFS_NOT_ATTEMPTED)
1448 pnfs_read_through_mds(desc, data);
1449 }
1450 put_lseg(lseg);
1451}
1452
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001453static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
1454{
1455 put_lseg(hdr->lseg);
1456 nfs_readhdr_free(hdr);
1457}
1458
Trond Myklebust493292d2011-07-13 15:58:28 -04001459int
1460pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
1461{
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001462 struct nfs_read_header *rhdr;
1463 struct nfs_pgio_header *hdr;
Trond Myklebust493292d2011-07-13 15:58:28 -04001464 int ret;
1465
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001466 rhdr = nfs_readhdr_alloc();
1467 if (!rhdr) {
Fred Isaman061ae2e2012-04-20 14:47:48 -04001468 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001469 ret = -ENOMEM;
Trond Myklebust493292d2011-07-13 15:58:28 -04001470 put_lseg(desc->pg_lseg);
1471 desc->pg_lseg = NULL;
1472 return ret;
1473 }
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001474 hdr = &rhdr->header;
1475 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1476 hdr->lseg = get_lseg(desc->pg_lseg);
1477 atomic_inc(&hdr->refcnt);
1478 ret = nfs_generic_pagein(desc, hdr);
1479 if (ret != 0) {
1480 put_lseg(desc->pg_lseg);
1481 desc->pg_lseg = NULL;
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001482 } else
1483 pnfs_do_multiple_reads(desc, &hdr->rpc_list);
1484 if (atomic_dec_and_test(&hdr->refcnt))
Fred Isaman061ae2e2012-04-20 14:47:48 -04001485 hdr->completion_ops->completion(hdr);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001486 return ret;
Trond Myklebust493292d2011-07-13 15:58:28 -04001487}
1488EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
1489
Andy Adamson863a3c62011-03-23 13:27:54 +00001490/*
Peng Taoa9bae562011-07-30 20:52:33 -04001491 * There can be multiple RW segments.
Andy Adamson863a3c62011-03-23 13:27:54 +00001492 */
Peng Taoa9bae562011-07-30 20:52:33 -04001493static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
Andy Adamson863a3c62011-03-23 13:27:54 +00001494{
Peng Taoa9bae562011-07-30 20:52:33 -04001495 struct pnfs_layout_segment *lseg;
Andy Adamson863a3c62011-03-23 13:27:54 +00001496
Peng Taoa9bae562011-07-30 20:52:33 -04001497 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
1498 if (lseg->pls_range.iomode == IOMODE_RW &&
1499 test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1500 list_add(&lseg->pls_lc_list, listp);
1501 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001502}
1503
Peng Tao1b0ae062011-09-22 21:50:12 -04001504void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
1505{
1506 if (lseg->pls_range.iomode == IOMODE_RW) {
1507 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__);
1508 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
1509 } else {
1510 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__);
1511 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
1512 }
1513}
1514EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
1515
Andy Adamson863a3c62011-03-23 13:27:54 +00001516void
1517pnfs_set_layoutcommit(struct nfs_write_data *wdata)
1518{
Fred Isamancd841602012-04-20 14:47:44 -04001519 struct nfs_pgio_header *hdr = wdata->header;
1520 struct inode *inode = hdr->inode;
1521 struct nfs_inode *nfsi = NFS_I(inode);
Vitaliy Gusev4b8ee2b2011-05-20 01:34:46 +04001522 loff_t end_pos = wdata->mds_offset + wdata->res.count;
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001523 bool mark_as_dirty = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001524
Fred Isamancd841602012-04-20 14:47:44 -04001525 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001526 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001527 mark_as_dirty = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001528 dprintk("%s: Set layoutcommit for inode %lu ",
Fred Isamancd841602012-04-20 14:47:44 -04001529 __func__, inode->i_ino);
Andy Adamson863a3c62011-03-23 13:27:54 +00001530 }
Fred Isamancd841602012-04-20 14:47:44 -04001531 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
Peng Taoa9bae562011-07-30 20:52:33 -04001532 /* references matched in nfs4_layoutcommit_release */
Fred Isamancd841602012-04-20 14:47:44 -04001533 get_lseg(hdr->lseg);
Peng Taoa9bae562011-07-30 20:52:33 -04001534 }
Peng Taoacff58802011-07-30 20:52:31 -04001535 if (end_pos > nfsi->layout->plh_lwb)
1536 nfsi->layout->plh_lwb = end_pos;
Fred Isamancd841602012-04-20 14:47:44 -04001537 spin_unlock(&inode->i_lock);
Peng Taoacff58802011-07-30 20:52:31 -04001538 dprintk("%s: lseg %p end_pos %llu\n",
Fred Isamancd841602012-04-20 14:47:44 -04001539 __func__, hdr->lseg, nfsi->layout->plh_lwb);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001540
1541 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1542 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1543 if (mark_as_dirty)
Fred Isamancd841602012-04-20 14:47:44 -04001544 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001545}
1546EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1547
Andy Adamsondb29c082011-07-30 20:52:38 -04001548void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
1549{
1550 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
1551
1552 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
1553 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
1554}
1555
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001556/*
1557 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1558 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1559 * data to disk to allow the server to recover the data if it crashes.
1560 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1561 * is off, and a COMMIT is sent to a data server, or
1562 * if WRITEs to a data server return NFS_DATA_SYNC.
1563 */
Andy Adamson863a3c62011-03-23 13:27:54 +00001564int
Andy Adamsonef311532011-03-12 02:58:10 -05001565pnfs_layoutcommit_inode(struct inode *inode, bool sync)
Andy Adamson863a3c62011-03-23 13:27:54 +00001566{
1567 struct nfs4_layoutcommit_data *data;
1568 struct nfs_inode *nfsi = NFS_I(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001569 loff_t end_pos;
1570 int status = 0;
1571
1572 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1573
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001574 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1575 return 0;
1576
Andy Adamson863a3c62011-03-23 13:27:54 +00001577 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1578 data = kzalloc(sizeof(*data), GFP_NOFS);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001579 if (!data) {
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001580 status = -ENOMEM;
1581 goto out;
1582 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001583
Peng Tao92407e72011-10-23 20:21:17 -07001584 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1585 goto out_free;
1586
1587 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
1588 if (!sync) {
1589 status = -EAGAIN;
1590 goto out_free;
1591 }
1592 status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING,
1593 nfs_wait_bit_killable, TASK_KILLABLE);
1594 if (status)
1595 goto out_free;
1596 }
1597
Peng Taoa9bae562011-07-30 20:52:33 -04001598 INIT_LIST_HEAD(&data->lseg_list);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001599 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001600 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Peng Tao92407e72011-10-23 20:21:17 -07001601 clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
Andy Adamson863a3c62011-03-23 13:27:54 +00001602 spin_unlock(&inode->i_lock);
Peng Tao92407e72011-10-23 20:21:17 -07001603 wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
1604 goto out_free;
Andy Adamson863a3c62011-03-23 13:27:54 +00001605 }
Peng Taoa9bae562011-07-30 20:52:33 -04001606
1607 pnfs_list_write_lseg(inode, &data->lseg_list);
Andy Adamson863a3c62011-03-23 13:27:54 +00001608
Peng Taoacff58802011-07-30 20:52:31 -04001609 end_pos = nfsi->layout->plh_lwb;
Peng Taoacff58802011-07-30 20:52:31 -04001610 nfsi->layout->plh_lwb = 0;
Andy Adamson863a3c62011-03-23 13:27:54 +00001611
Trond Myklebustf597c532012-03-04 18:13:56 -05001612 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
Andy Adamson863a3c62011-03-23 13:27:54 +00001613 spin_unlock(&inode->i_lock);
1614
1615 data->args.inode = inode;
Peng Tao9fa40752011-07-30 20:52:32 -04001616 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
Andy Adamson863a3c62011-03-23 13:27:54 +00001617 nfs_fattr_init(&data->fattr);
1618 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1619 data->res.fattr = &data->fattr;
1620 data->args.lastbytewritten = end_pos - 1;
1621 data->res.server = NFS_SERVER(inode);
1622
1623 status = nfs4_proc_layoutcommit(data, sync);
1624out:
Peng Tao92407e72011-10-23 20:21:17 -07001625 if (status)
1626 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001627 dprintk("<-- %s status %d\n", __func__, status);
1628 return status;
Peng Tao92407e72011-10-23 20:21:17 -07001629out_free:
1630 kfree(data);
1631 goto out;
Andy Adamson863a3c62011-03-23 13:27:54 +00001632}