blob: d737557747b91b82947ec2d1e9552a35d9fb9419 [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
Trond Myklebust25c75332012-09-18 17:01:12 -040038#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040039
Fred Isaman02c35fc2010-10-20 00:17:59 -040040/* Locking:
41 *
42 * pnfs_spinlock:
43 * protects pnfs_modules_tbl.
44 */
45static DEFINE_SPINLOCK(pnfs_spinlock);
46
47/*
48 * pnfs_modules_tbl holds all pnfs modules
49 */
50static LIST_HEAD(pnfs_modules_tbl);
51
52/* Return the registered pnfs layout driver module matching given id */
53static struct pnfs_layoutdriver_type *
54find_pnfs_driver_locked(u32 id)
55{
56 struct pnfs_layoutdriver_type *local;
57
58 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
59 if (local->id == id)
60 goto out;
61 local = NULL;
62out:
63 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
64 return local;
65}
66
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040067static struct pnfs_layoutdriver_type *
68find_pnfs_driver(u32 id)
69{
Fred Isaman02c35fc2010-10-20 00:17:59 -040070 struct pnfs_layoutdriver_type *local;
71
72 spin_lock(&pnfs_spinlock);
73 local = find_pnfs_driver_locked(id);
Trond Myklebust0a9c63f2012-06-15 13:02:58 -040074 if (local != NULL && !try_module_get(local->owner)) {
75 dprintk("%s: Could not grab reference on module\n", __func__);
76 local = NULL;
77 }
Fred Isaman02c35fc2010-10-20 00:17:59 -040078 spin_unlock(&pnfs_spinlock);
79 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040080}
81
82void
83unset_pnfs_layoutdriver(struct nfs_server *nfss)
84{
Benny Halevy738fd0f32011-07-30 20:52:36 -040085 if (nfss->pnfs_curr_ld) {
86 if (nfss->pnfs_curr_ld->clear_layoutdriver)
87 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
Trond Myklebust2a4c8992012-06-14 13:08:38 -040088 /* Decrement the MDS count. Purge the deviceid cache if zero */
89 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
90 nfs4_deviceid_purge_client(nfss->nfs_client);
Fred Isaman02c35fc2010-10-20 00:17:59 -040091 module_put(nfss->pnfs_curr_ld->owner);
Benny Halevy738fd0f32011-07-30 20:52:36 -040092 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040093 nfss->pnfs_curr_ld = NULL;
94}
95
96/*
97 * Try to set the server's pnfs module to the pnfs layout type specified by id.
98 * Currently only one pNFS layout driver per filesystem is supported.
99 *
100 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
101 */
102void
Benny Halevy738fd0f32011-07-30 20:52:36 -0400103set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
104 u32 id)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400105{
106 struct pnfs_layoutdriver_type *ld_type = NULL;
107
108 if (id == 0)
109 goto out_no_driver;
110 if (!(server->nfs_client->cl_exchange_flags &
111 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500112 printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
113 __func__, id, server->nfs_client->cl_exchange_flags);
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400114 goto out_no_driver;
115 }
116 ld_type = find_pnfs_driver(id);
117 if (!ld_type) {
118 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
119 ld_type = find_pnfs_driver(id);
120 if (!ld_type) {
121 dprintk("%s: No pNFS module found for %u.\n",
122 __func__, id);
123 goto out_no_driver;
124 }
125 }
126 server->pnfs_curr_ld = ld_type;
Benny Halevy738fd0f32011-07-30 20:52:36 -0400127 if (ld_type->set_layoutdriver
128 && ld_type->set_layoutdriver(server, mntfh)) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500129 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
130 "driver %u.\n", __func__, id);
Benny Halevy738fd0f32011-07-30 20:52:36 -0400131 module_put(ld_type->owner);
132 goto out_no_driver;
133 }
Trond Myklebust2a4c8992012-06-14 13:08:38 -0400134 /* Bump the MDS count */
135 atomic_inc(&server->nfs_client->cl_mds_count);
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000136
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400137 dprintk("%s: pNFS module for %u set\n", __func__, id);
138 return;
139
140out_no_driver:
141 dprintk("%s: Using NFSv4 I/O\n", __func__);
142 server->pnfs_curr_ld = NULL;
143}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400144
145int
146pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
147{
148 int status = -EINVAL;
149 struct pnfs_layoutdriver_type *tmp;
150
151 if (ld_type->id == 0) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500152 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
Fred Isaman02c35fc2010-10-20 00:17:59 -0400153 return status;
154 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400155 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500156 printk(KERN_ERR "NFS: %s Layout driver must provide "
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400157 "alloc_lseg and free_lseg.\n", __func__);
158 return status;
159 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400160
161 spin_lock(&pnfs_spinlock);
162 tmp = find_pnfs_driver_locked(ld_type->id);
163 if (!tmp) {
164 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
165 status = 0;
166 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
167 ld_type->name);
168 } else {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500169 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
Fred Isaman02c35fc2010-10-20 00:17:59 -0400170 __func__, ld_type->id);
171 }
172 spin_unlock(&pnfs_spinlock);
173
174 return status;
175}
176EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
177
178void
179pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
180{
181 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
182 spin_lock(&pnfs_spinlock);
183 list_del(&ld_type->pnfs_tblid);
184 spin_unlock(&pnfs_spinlock);
185}
186EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400187
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400188/*
189 * pNFS client layout cache
190 */
191
Fred Isamancc6e5342011-01-06 11:36:28 +0000192/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000193void
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400194pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400195{
Fred Isamancc6e5342011-01-06 11:36:28 +0000196 atomic_inc(&lo->plh_refcount);
197}
198
Benny Halevy636fb9c2011-05-22 19:51:33 +0300199static struct pnfs_layout_hdr *
200pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
201{
202 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
Trond Myklebust57934272012-09-20 20:37:23 -0400203 return ld->alloc_layout_hdr(ino, gfp_flags);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300204}
205
206static void
207pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
208{
Trond Myklebust9c626382012-09-20 17:31:43 -0400209 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
210 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
211
212 if (!list_empty(&lo->plh_layouts)) {
213 struct nfs_client *clp = server->nfs_client;
214
215 spin_lock(&clp->cl_lock);
216 list_del_init(&lo->plh_layouts);
217 spin_unlock(&clp->cl_lock);
218 }
Peng Tao9fa40752011-07-30 20:52:32 -0400219 put_rpccred(lo->plh_lc_cred);
Trond Myklebust57934272012-09-20 20:37:23 -0400220 return ld->free_layout_hdr(lo);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300221}
222
Fred Isamancc6e5342011-01-06 11:36:28 +0000223static void
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400224pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
Fred Isamancc6e5342011-01-06 11:36:28 +0000225{
Trond Myklebustbb346f62012-09-20 15:52:13 -0400226 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
Fred Isamancc6e5342011-01-06 11:36:28 +0000227 dprintk("%s: freeing layout cache %p\n", __func__, lo);
Trond Myklebustbb346f62012-09-20 15:52:13 -0400228 nfsi->layout = NULL;
229 /* Reset MDS Threshold I/O counters */
230 nfsi->write_io = 0;
231 nfsi->read_io = 0;
Benny Halevye5e94012010-10-20 00:18:01 -0400232}
233
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400234void
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400235pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400236{
Fred Isamancc6e5342011-01-06 11:36:28 +0000237 struct inode *inode = lo->plh_inode;
238
239 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400240 pnfs_detach_layout_hdr(lo);
Fred Isamancc6e5342011-01-06 11:36:28 +0000241 spin_unlock(&inode->i_lock);
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400242 pnfs_free_layout_hdr(lo);
Fred Isamancc6e5342011-01-06 11:36:28 +0000243 }
Andy Adamson974cec82010-10-20 00:18:02 -0400244}
245
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400246static int
247pnfs_iomode_to_fail_bit(u32 iomode)
248{
249 return iomode == IOMODE_RW ?
250 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
251}
252
253static void
Trond Myklebust3e621212012-09-24 13:07:16 -0400254pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400255{
Trond Myklebust25c75332012-09-18 17:01:12 -0400256 lo->plh_retry_timestamp = jiffies;
Trond Myklebust3e621212012-09-24 13:07:16 -0400257 if (test_and_set_bit(fail_bit, &lo->plh_flags))
258 atomic_inc(&lo->plh_refcount);
259}
260
261static void
262pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
263{
264 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
265 atomic_dec(&lo->plh_refcount);
266}
267
268static void
269pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
270{
271 struct inode *inode = lo->plh_inode;
Trond Myklebust115ce572012-09-20 21:19:43 -0400272 struct pnfs_layout_range range = {
273 .iomode = iomode,
274 .offset = 0,
275 .length = NFS4_MAX_UINT64,
276 };
277 LIST_HEAD(head);
Trond Myklebust3e621212012-09-24 13:07:16 -0400278
279 spin_lock(&inode->i_lock);
280 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
Trond Myklebust115ce572012-09-20 21:19:43 -0400281 pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
Trond Myklebust3e621212012-09-24 13:07:16 -0400282 spin_unlock(&inode->i_lock);
Trond Myklebust115ce572012-09-20 21:19:43 -0400283 pnfs_free_lseg_list(&head);
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400284 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
285 iomode == IOMODE_RW ? "RW" : "READ");
286}
287
288static bool
289pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
290{
Trond Myklebust25c75332012-09-18 17:01:12 -0400291 unsigned long start, end;
Trond Myklebust3e621212012-09-24 13:07:16 -0400292 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
293
294 if (test_bit(fail_bit, &lo->plh_flags) == 0)
Trond Myklebust25c75332012-09-18 17:01:12 -0400295 return false;
296 end = jiffies;
297 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
298 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
299 /* It is time to retry the failed layoutgets */
Trond Myklebust3e621212012-09-24 13:07:16 -0400300 pnfs_layout_clear_fail_bit(lo, fail_bit);
Trond Myklebust25c75332012-09-18 17:01:12 -0400301 return false;
302 }
303 return true;
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400304}
305
Andy Adamson974cec82010-10-20 00:18:02 -0400306static void
307init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
308{
Fred Isaman566052c2011-01-06 11:36:20 +0000309 INIT_LIST_HEAD(&lseg->pls_list);
Peng Taoa9bae562011-07-30 20:52:33 -0400310 INIT_LIST_HEAD(&lseg->pls_lc_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000311 atomic_set(&lseg->pls_refcount, 1);
312 smp_mb();
313 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000314 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400315}
316
Trond Myklebust905ca192012-09-20 20:46:49 -0400317static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400318{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000319 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400320
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400321 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400322}
323
Fred Isamand684d2a2011-03-01 01:34:13 +0000324static void
Trond Myklebust57036a32012-09-20 16:33:30 -0400325pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
326 struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400327{
Trond Myklebust57036a32012-09-20 16:33:30 -0400328 struct inode *inode = lo->plh_inode;
Fred Isamand684d2a2011-03-01 01:34:13 +0000329
Benny Halevyd20581a2011-05-22 19:52:03 +0300330 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000331 list_del_init(&lseg->pls_list);
Trond Myklebust8f0d27d2012-09-20 20:57:11 -0400332 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
333 atomic_dec(&lo->plh_refcount);
Trond Myklebust173f77e2012-09-21 15:49:42 -0400334 if (list_empty(&lo->plh_segs))
335 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
Fred Isamand684d2a2011-03-01 01:34:13 +0000336 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
337}
338
Fred Isamanbae724e2011-03-01 01:34:15 +0000339void
Trond Myklebust9369a432012-09-18 20:57:08 -0400340pnfs_put_lseg(struct pnfs_layout_segment *lseg)
Fred Isamand684d2a2011-03-01 01:34:13 +0000341{
Trond Myklebust57036a32012-09-20 16:33:30 -0400342 struct pnfs_layout_hdr *lo;
Fred Isamand684d2a2011-03-01 01:34:13 +0000343 struct inode *inode;
344
345 if (!lseg)
346 return;
347
Fred Isaman4541d162011-01-06 11:36:23 +0000348 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
349 atomic_read(&lseg->pls_refcount),
350 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Trond Myklebust57036a32012-09-20 16:33:30 -0400351 lo = lseg->pls_layout;
352 inode = lo->plh_inode;
Fred Isamand684d2a2011-03-01 01:34:13 +0000353 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
Trond Myklebust8f0d27d2012-09-20 20:57:11 -0400354 pnfs_get_layout_hdr(lo);
Trond Myklebust57036a32012-09-20 16:33:30 -0400355 pnfs_layout_remove_lseg(lo, lseg);
Fred Isamand684d2a2011-03-01 01:34:13 +0000356 spin_unlock(&inode->i_lock);
Trond Myklebust905ca192012-09-20 20:46:49 -0400357 pnfs_free_lseg(lseg);
Trond Myklebust8f0d27d2012-09-20 20:57:11 -0400358 pnfs_put_layout_hdr(lo);
Fred Isaman4541d162011-01-06 11:36:23 +0000359 }
Andy Adamson974cec82010-10-20 00:18:02 -0400360}
Trond Myklebust9369a432012-09-18 20:57:08 -0400361EXPORT_SYMBOL_GPL(pnfs_put_lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400362
Benny Halevyfb3296e2011-05-22 19:47:26 +0300363static inline u64
364end_offset(u64 start, u64 len)
Fred Isaman4541d162011-01-06 11:36:23 +0000365{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300366 u64 end;
367
368 end = start + len;
369 return end >= start ? end : NFS4_MAX_UINT64;
370}
371
372/* last octet in a range */
373static inline u64
374last_byte_offset(u64 start, u64 len)
375{
376 u64 end;
377
378 BUG_ON(!len);
379 end = start + len;
380 return end > start ? end - 1 : NFS4_MAX_UINT64;
381}
382
383/*
384 * is l2 fully contained in l1?
385 * start1 end1
386 * [----------------------------------)
387 * start2 end2
388 * [----------------)
389 */
390static inline int
391lo_seg_contained(struct pnfs_layout_range *l1,
392 struct pnfs_layout_range *l2)
393{
394 u64 start1 = l1->offset;
395 u64 end1 = end_offset(start1, l1->length);
396 u64 start2 = l2->offset;
397 u64 end2 = end_offset(start2, l2->length);
398
399 return (start1 <= start2) && (end1 >= end2);
400}
401
402/*
403 * is l1 and l2 intersecting?
404 * start1 end1
405 * [----------------------------------)
406 * start2 end2
407 * [----------------)
408 */
409static inline int
410lo_seg_intersecting(struct pnfs_layout_range *l1,
411 struct pnfs_layout_range *l2)
412{
413 u64 start1 = l1->offset;
414 u64 end1 = end_offset(start1, l1->length);
415 u64 start2 = l2->offset;
416 u64 end2 = end_offset(start2, l2->length);
417
418 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
419 (end2 == NFS4_MAX_UINT64 || end2 > start1);
420}
421
Fred Isaman4541d162011-01-06 11:36:23 +0000422static bool
Benny Halevy778b5502011-05-22 19:48:02 +0300423should_free_lseg(struct pnfs_layout_range *lseg_range,
424 struct pnfs_layout_range *recall_range)
Fred Isaman4541d162011-01-06 11:36:23 +0000425{
Benny Halevy778b5502011-05-22 19:48:02 +0300426 return (recall_range->iomode == IOMODE_ANY ||
427 lseg_range->iomode == recall_range->iomode) &&
428 lo_seg_intersecting(lseg_range, recall_range);
Fred Isaman4541d162011-01-06 11:36:23 +0000429}
430
431/* Returns 1 if lseg is removed from list, 0 otherwise */
432static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
433 struct list_head *tmp_list)
434{
435 int rv = 0;
436
437 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
438 /* Remove the reference keeping the lseg in the
439 * list. It will now be removed when all
440 * outstanding io is finished.
441 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000442 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
443 atomic_read(&lseg->pls_refcount));
444 if (atomic_dec_and_test(&lseg->pls_refcount)) {
Trond Myklebust57036a32012-09-20 16:33:30 -0400445 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
Fred Isamand684d2a2011-03-01 01:34:13 +0000446 list_add(&lseg->pls_list, tmp_list);
447 rv = 1;
448 }
Fred Isaman4541d162011-01-06 11:36:23 +0000449 }
450 return rv;
451}
452
453/* Returns count of number of matching invalid lsegs remaining in list
454 * after call.
455 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000456int
Trond Myklebust49a85062012-09-18 20:43:31 -0400457pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
Fred Isaman4541d162011-01-06 11:36:23 +0000458 struct list_head *tmp_list,
Benny Halevy778b5502011-05-22 19:48:02 +0300459 struct pnfs_layout_range *recall_range)
Andy Adamson974cec82010-10-20 00:18:02 -0400460{
461 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000462 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400463
464 dprintk("%s:Begin lo %p\n", __func__, lo);
465
Trond Myklebust8006bfb2012-09-21 14:48:04 -0400466 if (list_empty(&lo->plh_segs))
Fred Isaman38511722011-02-03 18:28:50 +0000467 return 0;
Fred Isaman4541d162011-01-06 11:36:23 +0000468 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
Benny Halevy778b5502011-05-22 19:48:02 +0300469 if (!recall_range ||
470 should_free_lseg(&lseg->pls_range, recall_range)) {
Fred Isaman4541d162011-01-06 11:36:23 +0000471 dprintk("%s: freeing lseg %p iomode %d "
472 "offset %llu length %llu\n", __func__,
473 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
474 lseg->pls_range.length);
475 invalid++;
476 removed += mark_lseg_invalid(lseg, tmp_list);
477 }
478 dprintk("%s:Return %i\n", __func__, invalid - removed);
479 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400480}
481
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000482/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000483void
Fred Isaman4541d162011-01-06 11:36:23 +0000484pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400485{
Fred Isaman4541d162011-01-06 11:36:23 +0000486 struct pnfs_layout_segment *lseg, *tmp;
Andy Adamson974cec82010-10-20 00:18:02 -0400487
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000488 if (list_empty(free_me))
489 return;
490
Fred Isaman4541d162011-01-06 11:36:23 +0000491 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000492 list_del(&lseg->pls_list);
Trond Myklebust905ca192012-09-20 20:46:49 -0400493 pnfs_free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400494 }
495}
496
Benny Halevye5e94012010-10-20 00:18:01 -0400497void
498pnfs_destroy_layout(struct nfs_inode *nfsi)
499{
500 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400501 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400502
503 spin_lock(&nfsi->vfs_inode.i_lock);
504 lo = nfsi->layout;
505 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000506 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Trond Myklebust49a85062012-09-18 20:43:31 -0400507 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
Trond Myklebust3e621212012-09-24 13:07:16 -0400508 pnfs_get_layout_hdr(lo);
509 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
510 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
511 spin_unlock(&nfsi->vfs_inode.i_lock);
512 pnfs_free_lseg_list(&tmp_list);
513 pnfs_put_layout_hdr(lo);
514 } else
515 spin_unlock(&nfsi->vfs_inode.i_lock);
Benny Halevye5e94012010-10-20 00:18:01 -0400516}
Andy Adamson041245c2012-04-27 17:53:53 -0400517EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
Benny Halevye5e94012010-10-20 00:18:01 -0400518
Andy Adamson974cec82010-10-20 00:18:02 -0400519/*
520 * Called by the state manger to remove all layouts established under an
521 * expired lease.
522 */
523void
524pnfs_destroy_all_layouts(struct nfs_client *clp)
525{
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400526 struct nfs_server *server;
Andy Adamson974cec82010-10-20 00:18:02 -0400527 struct pnfs_layout_hdr *lo;
528 LIST_HEAD(tmp_list);
529
Andy Adamsonc47abcf2011-06-15 17:52:40 -0400530 nfs4_deviceid_mark_client_invalid(clp);
531 nfs4_deviceid_purge_client(clp);
532
Andy Adamson974cec82010-10-20 00:18:02 -0400533 spin_lock(&clp->cl_lock);
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400534 rcu_read_lock();
535 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
536 if (!list_empty(&server->layouts))
537 list_splice_init(&server->layouts, &tmp_list);
538 }
539 rcu_read_unlock();
Andy Adamson974cec82010-10-20 00:18:02 -0400540 spin_unlock(&clp->cl_lock);
541
542 while (!list_empty(&tmp_list)) {
543 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000544 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400545 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000546 lo->plh_inode->i_ino);
Andy Adamson2887fe42011-05-11 01:19:58 -0400547 list_del_init(&lo->plh_layouts);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000548 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400549 }
550}
551
Fred Isamanfd6002e2011-01-06 11:36:22 +0000552/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000553void
554pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
555 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400556{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000557 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400558
Trond Myklebust2d2f24a2012-03-04 18:13:57 -0500559 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
560 newseq = be32_to_cpu(new->seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000561 if ((int)(newseq - oldseq) > 0) {
Trond Myklebustf597c532012-03-04 18:13:56 -0500562 nfs4_stateid_copy(&lo->plh_stateid, new);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000563 if (update_barrier) {
Trond Myklebust2d2f24a2012-03-04 18:13:57 -0500564 u32 new_barrier = be32_to_cpu(new->seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000565
566 if ((int)(new_barrier - lo->plh_barrier))
567 lo->plh_barrier = new_barrier;
568 } else {
569 /* Because of wraparound, we want to keep the barrier
570 * "close" to the current seqids. It needs to be
571 * within 2**31 to count as "behind", so if it
572 * gets too near that limit, give us a litle leeway
573 * and bring it to within 2**30.
574 * NOTE - and yes, this is all unsigned arithmetic.
575 */
576 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
577 lo->plh_barrier = newseq - (1 << 30);
578 }
579 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400580}
581
Fred Isamancf7d63f2011-01-06 11:36:25 +0000582/* lget is set to 1 if called from inside send_layoutget call chain */
583static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000584pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
585 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000586{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000587 if ((stateid) &&
Trond Myklebust2d2f24a2012-03-04 18:13:57 -0500588 (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0)
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000589 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000590 return lo->plh_block_lgets ||
591 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000592 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000593 (atomic_read(&lo->plh_outstanding) > lget));
594}
595
Fred Isamanfd6002e2011-01-06 11:36:22 +0000596int
597pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
598 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400599{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000600 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400601
602 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000603 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000604 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000605 status = -EAGAIN;
606 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000607 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400608
Fred Isamanfd6002e2011-01-06 11:36:22 +0000609 do {
610 seq = read_seqbegin(&open_state->seqlock);
Trond Myklebustf597c532012-03-04 18:13:56 -0500611 nfs4_stateid_copy(dst, &open_state->stateid);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000612 } while (read_seqretry(&open_state->seqlock, seq));
613 } else
Trond Myklebustf597c532012-03-04 18:13:56 -0500614 nfs4_stateid_copy(dst, &lo->plh_stateid);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000615 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400616 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000617 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400618}
619
620/*
621* Get layout from server.
622* for now, assume that whole file layouts are requested.
623* arg->offset: 0
624* arg->length: all ones
625*/
Benny Halevye5e94012010-10-20 00:18:01 -0400626static struct pnfs_layout_segment *
627send_layoutget(struct pnfs_layout_hdr *lo,
628 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +0300629 struct pnfs_layout_range *range,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400630 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400631{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000632 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400633 struct nfs_server *server = NFS_SERVER(ino);
634 struct nfs4_layoutget *lgp;
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -0400635 struct pnfs_layout_segment *lseg;
Benny Halevye5e94012010-10-20 00:18:01 -0400636
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400637 dprintk("--> %s\n", __func__);
638
639 BUG_ON(ctx == NULL);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400640 lgp = kzalloc(sizeof(*lgp), gfp_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000641 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400642 return NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400643
Benny Halevyfb3296e2011-05-22 19:47:26 +0300644 lgp->args.minlength = PAGE_CACHE_SIZE;
645 if (lgp->args.minlength > range->length)
646 lgp->args.minlength = range->length;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400647 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
Benny Halevyfb3296e2011-05-22 19:47:26 +0300648 lgp->args.range = *range;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400649 lgp->args.type = server->pnfs_curr_ld->id;
650 lgp->args.inode = ino;
651 lgp->args.ctx = get_nfs_open_context(ctx);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400652 lgp->gfp_flags = gfp_flags;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400653
654 /* Synchronously retrieve layout information from server and
655 * store in lseg.
656 */
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -0400657 lseg = nfs4_proc_layoutget(lgp, gfp_flags);
658 if (IS_ERR(lseg)) {
659 switch (PTR_ERR(lseg)) {
660 case -ENOMEM:
661 case -ERESTARTSYS:
662 break;
663 default:
664 /* remember that LAYOUTGET failed and suspend trying */
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400665 pnfs_layout_io_set_failed(lo, range->iomode);
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -0400666 }
667 return NULL;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400668 }
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400669
Andy Adamson974cec82010-10-20 00:18:02 -0400670 return lseg;
671}
672
Andy Adamson293b3b02012-06-20 15:03:34 -0400673/*
674 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
675 * when the layout segment list is empty.
676 *
677 * Note that a pnfs_layout_hdr can exist with an empty layout segment
678 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
679 * deviceid is marked invalid.
680 */
Benny Halevycbe82602011-05-22 19:52:37 +0300681int
682_pnfs_return_layout(struct inode *ino)
683{
684 struct pnfs_layout_hdr *lo = NULL;
685 struct nfs_inode *nfsi = NFS_I(ino);
686 LIST_HEAD(tmp_list);
687 struct nfs4_layoutreturn *lrp;
688 nfs4_stateid stateid;
Andy Adamson293b3b02012-06-20 15:03:34 -0400689 int status = 0, empty;
Benny Halevycbe82602011-05-22 19:52:37 +0300690
Andy Adamson366d5052012-06-20 15:03:33 -0400691 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
Benny Halevycbe82602011-05-22 19:52:37 +0300692
693 spin_lock(&ino->i_lock);
694 lo = nfsi->layout;
Trond Myklebuste5929f32012-09-21 16:37:02 -0400695 if (!lo) {
Benny Halevycbe82602011-05-22 19:52:37 +0300696 spin_unlock(&ino->i_lock);
Andy Adamson293b3b02012-06-20 15:03:34 -0400697 dprintk("NFS: %s no layout to return\n", __func__);
698 goto out;
Benny Halevycbe82602011-05-22 19:52:37 +0300699 }
700 stateid = nfsi->layout->plh_stateid;
701 /* Reference matched in nfs4_layoutreturn_release */
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400702 pnfs_get_layout_hdr(lo);
Andy Adamson293b3b02012-06-20 15:03:34 -0400703 empty = list_empty(&lo->plh_segs);
Trond Myklebust49a85062012-09-18 20:43:31 -0400704 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
Andy Adamson293b3b02012-06-20 15:03:34 -0400705 /* Don't send a LAYOUTRETURN if list was initially empty */
706 if (empty) {
707 spin_unlock(&ino->i_lock);
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400708 pnfs_put_layout_hdr(lo);
Andy Adamson293b3b02012-06-20 15:03:34 -0400709 dprintk("NFS: %s no layout segments to return\n", __func__);
710 goto out;
711 }
Fred Isamanea0ded72011-06-15 12:31:02 -0400712 lo->plh_block_lgets++;
Benny Halevycbe82602011-05-22 19:52:37 +0300713 spin_unlock(&ino->i_lock);
714 pnfs_free_lseg_list(&tmp_list);
715
716 WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags));
717
718 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
719 if (unlikely(lrp == NULL)) {
720 status = -ENOMEM;
Trond Myklebust65857d52012-09-24 13:49:27 -0400721 spin_lock(&ino->i_lock);
722 lo->plh_block_lgets--;
723 spin_unlock(&ino->i_lock);
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400724 pnfs_put_layout_hdr(lo);
Benny Halevycbe82602011-05-22 19:52:37 +0300725 goto out;
726 }
727
728 lrp->args.stateid = stateid;
729 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
730 lrp->args.inode = ino;
Trond Myklebusta56aaa02011-06-15 11:59:10 -0400731 lrp->args.layout = lo;
Benny Halevycbe82602011-05-22 19:52:37 +0300732 lrp->clp = NFS_SERVER(ino)->nfs_client;
733
734 status = nfs4_proc_layoutreturn(lrp);
735out:
736 dprintk("<-- %s status: %d\n", __func__, status);
737 return status;
738}
Andy Adamson0a57cda2012-04-27 17:53:50 -0400739EXPORT_SYMBOL_GPL(_pnfs_return_layout);
Benny Halevycbe82602011-05-22 19:52:37 +0300740
Fred Isamanf7e89172011-01-06 11:36:32 +0000741bool pnfs_roc(struct inode *ino)
742{
743 struct pnfs_layout_hdr *lo;
744 struct pnfs_layout_segment *lseg, *tmp;
745 LIST_HEAD(tmp_list);
746 bool found = false;
747
748 spin_lock(&ino->i_lock);
749 lo = NFS_I(ino)->layout;
750 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
751 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
752 goto out_nolayout;
753 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
754 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
755 mark_lseg_invalid(lseg, &tmp_list);
756 found = true;
757 }
758 if (!found)
759 goto out_nolayout;
760 lo->plh_block_lgets++;
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400761 pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
Fred Isamanf7e89172011-01-06 11:36:32 +0000762 spin_unlock(&ino->i_lock);
763 pnfs_free_lseg_list(&tmp_list);
764 return true;
765
766out_nolayout:
767 spin_unlock(&ino->i_lock);
768 return false;
769}
770
771void pnfs_roc_release(struct inode *ino)
772{
773 struct pnfs_layout_hdr *lo;
774
775 spin_lock(&ino->i_lock);
776 lo = NFS_I(ino)->layout;
777 lo->plh_block_lgets--;
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400778 if (atomic_dec_and_test(&lo->plh_refcount)) {
779 pnfs_detach_layout_hdr(lo);
780 spin_unlock(&ino->i_lock);
781 pnfs_free_layout_hdr(lo);
782 } else
783 spin_unlock(&ino->i_lock);
Fred Isamanf7e89172011-01-06 11:36:32 +0000784}
785
786void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
787{
788 struct pnfs_layout_hdr *lo;
789
790 spin_lock(&ino->i_lock);
791 lo = NFS_I(ino)->layout;
792 if ((int)(barrier - lo->plh_barrier) > 0)
793 lo->plh_barrier = barrier;
794 spin_unlock(&ino->i_lock);
795}
796
Trond Myklebust7fdab062012-09-20 20:15:57 -0400797bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
Fred Isamanf7e89172011-01-06 11:36:32 +0000798{
799 struct nfs_inode *nfsi = NFS_I(ino);
Trond Myklebust7fdab062012-09-20 20:15:57 -0400800 struct pnfs_layout_hdr *lo;
Fred Isamanf7e89172011-01-06 11:36:32 +0000801 struct pnfs_layout_segment *lseg;
Trond Myklebust7fdab062012-09-20 20:15:57 -0400802 u32 current_seqid;
Fred Isamanf7e89172011-01-06 11:36:32 +0000803 bool found = false;
804
805 spin_lock(&ino->i_lock);
806 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
807 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
Trond Myklebust7fdab062012-09-20 20:15:57 -0400808 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
Fred Isamanf7e89172011-01-06 11:36:32 +0000809 found = true;
Trond Myklebust7fdab062012-09-20 20:15:57 -0400810 goto out;
Fred Isamanf7e89172011-01-06 11:36:32 +0000811 }
Trond Myklebust7fdab062012-09-20 20:15:57 -0400812 lo = nfsi->layout;
813 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
Fred Isamanf7e89172011-01-06 11:36:32 +0000814
Trond Myklebust7fdab062012-09-20 20:15:57 -0400815 /* Since close does not return a layout stateid for use as
816 * a barrier, we choose the worst-case barrier.
817 */
818 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
819out:
Fred Isamanf7e89172011-01-06 11:36:32 +0000820 spin_unlock(&ino->i_lock);
821 return found;
822}
823
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400824/*
825 * Compare two layout segments for sorting into layout cache.
826 * We want to preferentially return RW over RO layouts, so ensure those
827 * are seen first.
828 */
829static s64
Benny Halevyfb3296e2011-05-22 19:47:26 +0300830cmp_layout(struct pnfs_layout_range *l1,
831 struct pnfs_layout_range *l2)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400832{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300833 s64 d;
834
835 /* high offset > low offset */
836 d = l1->offset - l2->offset;
837 if (d)
838 return d;
839
840 /* short length > long length */
841 d = l2->length - l1->length;
842 if (d)
843 return d;
844
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400845 /* read > read/write */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300846 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400847}
848
Andy Adamson974cec82010-10-20 00:18:02 -0400849static void
Trond Myklebust57036a32012-09-20 16:33:30 -0400850pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
Andy Adamson974cec82010-10-20 00:18:02 -0400851 struct pnfs_layout_segment *lseg)
852{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400853 struct pnfs_layout_segment *lp;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400854
Andy Adamson974cec82010-10-20 00:18:02 -0400855 dprintk("%s:Begin\n", __func__);
856
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000857 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Benny Halevyfb3296e2011-05-22 19:47:26 +0300858 if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400859 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000860 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400861 dprintk("%s: inserted lseg %p "
862 "iomode %d offset %llu length %llu before "
863 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000864 __func__, lseg, lseg->pls_range.iomode,
865 lseg->pls_range.offset, lseg->pls_range.length,
866 lp, lp->pls_range.iomode, lp->pls_range.offset,
867 lp->pls_range.length);
Benny Halevyfb3296e2011-05-22 19:47:26 +0300868 goto out;
Andy Adamson974cec82010-10-20 00:18:02 -0400869 }
Benny Halevyfb3296e2011-05-22 19:47:26 +0300870 list_add_tail(&lseg->pls_list, &lo->plh_segs);
871 dprintk("%s: inserted lseg %p "
872 "iomode %d offset %llu length %llu at tail\n",
873 __func__, lseg, lseg->pls_range.iomode,
874 lseg->pls_range.offset, lseg->pls_range.length);
875out:
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400876 pnfs_get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400877
878 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400879}
880
881static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -0400882alloc_init_layout_hdr(struct inode *ino,
883 struct nfs_open_context *ctx,
884 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400885{
886 struct pnfs_layout_hdr *lo;
887
Benny Halevy636fb9c2011-05-22 19:51:33 +0300888 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400889 if (!lo)
890 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000891 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000892 INIT_LIST_HEAD(&lo->plh_layouts);
893 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000894 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000895 lo->plh_inode = ino;
Peng Tao9fa40752011-07-30 20:52:32 -0400896 lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
Benny Halevye5e94012010-10-20 00:18:01 -0400897 return lo;
898}
899
900static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -0400901pnfs_find_alloc_layout(struct inode *ino,
902 struct nfs_open_context *ctx,
903 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400904{
905 struct nfs_inode *nfsi = NFS_I(ino);
906 struct pnfs_layout_hdr *new = NULL;
907
908 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
909
Fred Isaman4541d162011-01-06 11:36:23 +0000910 if (nfsi->layout) {
Trond Myklebust01d39ce2012-09-20 17:02:32 -0400911 pnfs_get_layout_hdr(nfsi->layout);
912 return nfsi->layout;
Fred Isaman4541d162011-01-06 11:36:23 +0000913 }
Benny Halevye5e94012010-10-20 00:18:01 -0400914 spin_unlock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -0400915 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400916 spin_lock(&ino->i_lock);
917
918 if (likely(nfsi->layout == NULL)) /* Won the race? */
919 nfsi->layout = new;
920 else
Benny Halevy636fb9c2011-05-22 19:51:33 +0300921 pnfs_free_layout_hdr(new);
Benny Halevye5e94012010-10-20 00:18:01 -0400922 return nfsi->layout;
923}
924
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400925/*
926 * iomode matching rules:
927 * iomode lseg match
928 * ----- ----- -----
929 * ANY READ true
930 * ANY RW true
931 * RW READ false
932 * RW RW true
933 * READ READ true
934 * READ RW true
935 */
936static int
Benny Halevyfb3296e2011-05-22 19:47:26 +0300937is_matching_lseg(struct pnfs_layout_range *ls_range,
938 struct pnfs_layout_range *range)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400939{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300940 struct pnfs_layout_range range1;
941
942 if ((range->iomode == IOMODE_RW &&
943 ls_range->iomode != IOMODE_RW) ||
944 !lo_seg_intersecting(ls_range, range))
945 return 0;
946
947 /* range1 covers only the first byte in the range */
948 range1 = *range;
949 range1.length = 1;
950 return lo_seg_contained(ls_range, &range1);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400951}
952
953/*
954 * lookup range in layout
955 */
Benny Halevye5e94012010-10-20 00:18:01 -0400956static struct pnfs_layout_segment *
Benny Halevyfb3296e2011-05-22 19:47:26 +0300957pnfs_find_lseg(struct pnfs_layout_hdr *lo,
958 struct pnfs_layout_range *range)
Benny Halevye5e94012010-10-20 00:18:01 -0400959{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400960 struct pnfs_layout_segment *lseg, *ret = NULL;
961
962 dprintk("%s:Begin\n", __func__);
963
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000964 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000965 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
Benny Halevyfb3296e2011-05-22 19:47:26 +0300966 is_matching_lseg(&lseg->pls_range, range)) {
Trond Myklebust9369a432012-09-18 20:57:08 -0400967 ret = pnfs_get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400968 break;
969 }
Benny Halevyd771e3a2011-06-14 16:30:16 -0400970 if (lseg->pls_range.offset > range->offset)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400971 break;
972 }
973
974 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000975 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400976 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400977}
978
979/*
Andy Adamsond23d61c2012-05-23 05:02:37 -0400980 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
981 * to the MDS or over pNFS
982 *
983 * The nfs_inode read_io and write_io fields are cumulative counters reset
984 * when there are no layout segments. Note that in pnfs_update_layout iomode
985 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
986 * WRITE request.
987 *
988 * A return of true means use MDS I/O.
989 *
990 * From rfc 5661:
991 * If a file's size is smaller than the file size threshold, data accesses
992 * SHOULD be sent to the metadata server. If an I/O request has a length that
993 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
994 * server. If both file size and I/O size are provided, the client SHOULD
995 * reach or exceed both thresholds before sending its read or write
996 * requests to the data server.
997 */
998static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
999 struct inode *ino, int iomode)
1000{
1001 struct nfs4_threshold *t = ctx->mdsthreshold;
1002 struct nfs_inode *nfsi = NFS_I(ino);
1003 loff_t fsize = i_size_read(ino);
1004 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1005
1006 if (t == NULL)
1007 return ret;
1008
1009 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1010 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1011
1012 switch (iomode) {
1013 case IOMODE_READ:
1014 if (t->bm & THRESHOLD_RD) {
1015 dprintk("%s fsize %llu\n", __func__, fsize);
1016 size_set = true;
1017 if (fsize < t->rd_sz)
1018 size = true;
1019 }
1020 if (t->bm & THRESHOLD_RD_IO) {
1021 dprintk("%s nfsi->read_io %llu\n", __func__,
1022 nfsi->read_io);
1023 io_set = true;
1024 if (nfsi->read_io < t->rd_io_sz)
1025 io = true;
1026 }
1027 break;
1028 case IOMODE_RW:
1029 if (t->bm & THRESHOLD_WR) {
1030 dprintk("%s fsize %llu\n", __func__, fsize);
1031 size_set = true;
1032 if (fsize < t->wr_sz)
1033 size = true;
1034 }
1035 if (t->bm & THRESHOLD_WR_IO) {
1036 dprintk("%s nfsi->write_io %llu\n", __func__,
1037 nfsi->write_io);
1038 io_set = true;
1039 if (nfsi->write_io < t->wr_io_sz)
1040 io = true;
1041 }
1042 break;
1043 }
1044 if (size_set && io_set) {
1045 if (size && io)
1046 ret = true;
1047 } else if (size || io)
1048 ret = true;
1049
1050 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1051 return ret;
1052}
1053
1054/*
Benny Halevye5e94012010-10-20 00:18:01 -04001055 * Layout segment is retreived from the server if not cached.
1056 * The appropriate layout segment is referenced and returned to the caller.
1057 */
Andy Adamson7c24d942011-06-13 18:22:38 -04001058struct pnfs_layout_segment *
Benny Halevye5e94012010-10-20 00:18:01 -04001059pnfs_update_layout(struct inode *ino,
1060 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +03001061 loff_t pos,
1062 u64 count,
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001063 enum pnfs_iomode iomode,
1064 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -04001065{
Benny Halevyfb3296e2011-05-22 19:47:26 +03001066 struct pnfs_layout_range arg = {
1067 .iomode = iomode,
1068 .offset = pos,
1069 .length = count,
1070 };
Benny Halevy707ed5f2011-05-22 19:47:46 +03001071 unsigned pg_offset;
Weston Andros Adamson6382a442011-06-01 16:44:44 -04001072 struct nfs_server *server = NFS_SERVER(ino);
1073 struct nfs_client *clp = server->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -04001074 struct pnfs_layout_hdr *lo;
1075 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +00001076 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -04001077
1078 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04001079 goto out;
Andy Adamsond23d61c2012-05-23 05:02:37 -04001080
1081 if (pnfs_within_mdsthreshold(ctx, ino, iomode))
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04001082 goto out;
Andy Adamsond23d61c2012-05-23 05:02:37 -04001083
Benny Halevye5e94012010-10-20 00:18:01 -04001084 spin_lock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -04001085 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
Trond Myklebust830ffb52012-09-20 21:25:19 -04001086 if (lo == NULL) {
1087 spin_unlock(&ino->i_lock);
1088 goto out;
1089 }
Benny Halevye5e94012010-10-20 00:18:01 -04001090
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001091 /* Do we even need to bother with this? */
Trond Myklebusta59c30a2012-03-01 11:17:47 -05001092 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001093 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -04001094 goto out_unlock;
1095 }
1096
1097 /* if LAYOUTGET already failed once we don't try again */
Trond Myklebustb9e028f2012-09-18 16:41:18 -04001098 if (pnfs_layout_io_test_failed(lo, iomode))
Benny Halevye5e94012010-10-20 00:18:01 -04001099 goto out_unlock;
1100
Andy Adamson568e8c42011-03-01 01:34:22 +00001101 /* Check to see if the layout for the given range already exists */
Benny Halevyfb3296e2011-05-22 19:47:26 +03001102 lseg = pnfs_find_lseg(lo, &arg);
Andy Adamson568e8c42011-03-01 01:34:22 +00001103 if (lseg)
1104 goto out_unlock;
1105
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001106 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +00001107 goto out_unlock;
1108 atomic_inc(&lo->plh_outstanding);
1109
Fred Isamanf49f9ba2011-02-03 18:28:52 +00001110 if (list_empty(&lo->plh_segs))
1111 first = true;
Andy Adamson366d5052012-06-20 15:03:33 -04001112
Fred Isamanf49f9ba2011-02-03 18:28:52 +00001113 spin_unlock(&ino->i_lock);
1114 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +00001115 /* The lo must be on the clp list if there is any
1116 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1117 */
1118 spin_lock(&clp->cl_lock);
1119 BUG_ON(!list_empty(&lo->plh_layouts));
Weston Andros Adamson6382a442011-06-01 16:44:44 -04001120 list_add_tail(&lo->plh_layouts, &server->layouts);
Fred Isaman2130ff62011-01-06 11:36:26 +00001121 spin_unlock(&clp->cl_lock);
1122 }
Benny Halevye5e94012010-10-20 00:18:01 -04001123
Benny Halevy707ed5f2011-05-22 19:47:46 +03001124 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1125 if (pg_offset) {
1126 arg.offset -= pg_offset;
1127 arg.length += pg_offset;
1128 }
Andy Adamson7c24d942011-06-13 18:22:38 -04001129 if (arg.length != NFS4_MAX_UINT64)
1130 arg.length = PAGE_CACHE_ALIGN(arg.length);
Benny Halevy707ed5f2011-05-22 19:47:46 +03001131
Benny Halevyfb3296e2011-05-22 19:47:26 +03001132 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +00001133 atomic_dec(&lo->plh_outstanding);
Trond Myklebust830ffb52012-09-20 21:25:19 -04001134out_put_layout_hdr:
Trond Myklebust70c3bd22012-09-18 20:51:13 -04001135 pnfs_put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -04001136out:
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04001137 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1138 "(%s, offset: %llu, length: %llu)\n",
1139 __func__, ino->i_sb->s_id,
1140 (unsigned long long)NFS_FILEID(ino),
1141 lseg == NULL ? "not found" : "found",
1142 iomode==IOMODE_RW ? "read/write" : "read-only",
1143 (unsigned long long)pos,
1144 (unsigned long long)count);
Benny Halevye5e94012010-10-20 00:18:01 -04001145 return lseg;
1146out_unlock:
1147 spin_unlock(&ino->i_lock);
Trond Myklebust830ffb52012-09-20 21:25:19 -04001148 goto out_put_layout_hdr;
Benny Halevye5e94012010-10-20 00:18:01 -04001149}
Andy Adamson7c24d942011-06-13 18:22:38 -04001150EXPORT_SYMBOL_GPL(pnfs_update_layout);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001151
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -04001152struct pnfs_layout_segment *
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001153pnfs_layout_process(struct nfs4_layoutget *lgp)
1154{
1155 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1156 struct nfs4_layoutget_res *res = &lgp->res;
1157 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001158 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001159 int status = 0;
1160
1161 /* Inject layout blob into I/O device driver */
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001162 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001163 if (!lseg || IS_ERR(lseg)) {
1164 if (!lseg)
1165 status = -ENOMEM;
1166 else
1167 status = PTR_ERR(lseg);
1168 dprintk("%s: Could not allocate layout: error %d\n",
1169 __func__, status);
1170 goto out;
1171 }
1172
1173 spin_lock(&ino->i_lock);
Trond Myklebusta59c30a2012-03-01 11:17:47 -05001174 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001175 dprintk("%s forget reply due to recall\n", __func__);
1176 goto out_forget_reply;
1177 }
1178
1179 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
1180 dprintk("%s forget reply due to state\n", __func__);
1181 goto out_forget_reply;
1182 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001183 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +00001184 lseg->pls_range = res->range;
Trond Myklebust9369a432012-09-18 20:57:08 -04001185 pnfs_get_lseg(lseg);
Trond Myklebust57036a32012-09-20 16:33:30 -04001186 pnfs_layout_insert_lseg(lo, lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001187
Fred Isamanf7e89172011-01-06 11:36:32 +00001188 if (res->return_on_close) {
1189 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1190 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
1191 }
1192
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001193 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001194 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001195 spin_unlock(&ino->i_lock);
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -04001196 return lseg;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001197out:
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -04001198 return ERR_PTR(status);
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001199
1200out_forget_reply:
1201 spin_unlock(&ino->i_lock);
1202 lseg->pls_layout = lo;
1203 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1204 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001205}
1206
Trond Myklebustd8007d42011-06-10 13:30:23 -04001207void
1208pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1209{
1210 BUG_ON(pgio->pg_lseg != NULL);
1211
Fred Isaman1825a0d2012-04-20 19:55:31 -04001212 if (req->wb_offset != req->wb_pgbase) {
1213 nfs_pageio_reset_read_mds(pgio);
1214 return;
1215 }
Trond Myklebustd8007d42011-06-10 13:30:23 -04001216 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1217 req->wb_context,
1218 req_offset(req),
1219 req->wb_bytes,
1220 IOMODE_READ,
1221 GFP_KERNEL);
Trond Myklebuste885de12011-06-10 13:30:23 -04001222 /* If no lseg, fall back to read through mds */
1223 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001224 nfs_pageio_reset_read_mds(pgio);
Trond Myklebuste885de12011-06-10 13:30:23 -04001225
Trond Myklebustd8007d42011-06-10 13:30:23 -04001226}
1227EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1228
1229void
1230pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1231{
1232 BUG_ON(pgio->pg_lseg != NULL);
1233
Fred Isaman1825a0d2012-04-20 19:55:31 -04001234 if (req->wb_offset != req->wb_pgbase) {
1235 nfs_pageio_reset_write_mds(pgio);
1236 return;
1237 }
Trond Myklebustd8007d42011-06-10 13:30:23 -04001238 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1239 req->wb_context,
1240 req_offset(req),
1241 req->wb_bytes,
1242 IOMODE_RW,
1243 GFP_NOFS);
Trond Myklebuste885de12011-06-10 13:30:23 -04001244 /* If no lseg, fall back to write through mds */
1245 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001246 nfs_pageio_reset_write_mds(pgio);
Trond Myklebustd8007d42011-06-10 13:30:23 -04001247}
1248EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1249
Bryan Schumaker1abb50882012-06-20 15:53:47 -04001250void
Fred Isaman061ae2e2012-04-20 14:47:48 -04001251pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode,
1252 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001253{
1254 struct nfs_server *server = NFS_SERVER(inode);
1255 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1256
1257 if (ld == NULL)
Bryan Schumaker1abb50882012-06-20 15:53:47 -04001258 nfs_pageio_init_read(pgio, inode, compl_ops);
1259 else
1260 nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0);
Trond Myklebust1751c362011-06-10 13:30:23 -04001261}
1262
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001263void
Fred Isaman061ae2e2012-04-20 14:47:48 -04001264pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode,
1265 int ioflags,
1266 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001267{
1268 struct nfs_server *server = NFS_SERVER(inode);
1269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1270
1271 if (ld == NULL)
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001272 nfs_pageio_init_write(pgio, inode, ioflags, compl_ops);
1273 else
1274 nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001275}
1276
1277bool
Benny Halevydfed2062011-05-25 20:25:22 +03001278pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1279 struct nfs_page *req)
Fred Isamanbae724e2011-03-01 01:34:15 +00001280{
Trond Myklebustd8007d42011-06-10 13:30:23 -04001281 if (pgio->pg_lseg == NULL)
1282 return nfs_generic_pg_test(pgio, prev, req);
Benny Halevy89a58e32011-05-25 20:54:40 +03001283
Trond Myklebust19982ba2011-06-10 13:30:23 -04001284 /*
1285 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1286 * Note that this test makes several assumptions:
1287 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1288 * is known to lie within the range.
1289 * - that the nfs_page being tested is known to be contiguous with the
1290 * previous nfs_page.
1291 * - Layout ranges are page aligned, so we only have to test the
1292 * start offset of the request.
1293 *
1294 * Please also note that 'end_offset' is actually the offset of the
1295 * first byte that lies outside the pnfs_layout_range. FIXME?
1296 *
1297 */
1298 return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
1299 pgio->pg_lseg->pls_range.length);
Fred Isamanbae724e2011-03-01 01:34:15 +00001300}
Benny Halevy89a58e32011-05-25 20:54:40 +03001301EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
Fred Isamanbae724e2011-03-01 01:34:15 +00001302
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001303int pnfs_write_done_resend_to_mds(struct inode *inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001304 struct list_head *head,
1305 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001306{
1307 struct nfs_pageio_descriptor pgio;
1308 LIST_HEAD(failed);
1309
1310 /* Resend all requests through the MDS */
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001311 nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops);
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001312 while (!list_empty(head)) {
1313 struct nfs_page *req = nfs_list_entry(head->next);
1314
1315 nfs_list_remove_request(req);
1316 if (!nfs_pageio_add_request(&pgio, req))
1317 nfs_list_add_request(req, &failed);
1318 }
1319 nfs_pageio_complete(&pgio);
1320
1321 if (!list_empty(&failed)) {
1322 /* For some reason our attempt to resend pages. Mark the
1323 * overall send request as having failed, and let
1324 * nfs_writeback_release_full deal with the error.
1325 */
1326 list_move(&failed, head);
1327 return -EIO;
1328 }
1329 return 0;
1330}
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001331EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001332
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001333static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
1334{
Fred Isamancd841602012-04-20 14:47:44 -04001335 struct nfs_pgio_header *hdr = data->header;
1336
1337 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1338 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001339 PNFS_LAYOUTRET_ON_ERROR) {
Fred Isamancd841602012-04-20 14:47:44 -04001340 clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
1341 pnfs_return_layout(hdr->inode);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001342 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001343 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1344 data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001345 &hdr->pages,
1346 hdr->completion_ops);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001347}
1348
Benny Halevyd20581a2011-05-22 19:52:03 +03001349/*
1350 * Called by non rpc-based layout drivers
1351 */
Peng Tao8ce160c2011-09-22 21:50:14 -04001352void pnfs_ld_write_done(struct nfs_write_data *data)
Fred Isaman94ad1c82011-03-01 01:34:14 +00001353{
Fred Isamancd841602012-04-20 14:47:44 -04001354 struct nfs_pgio_header *hdr = data->header;
1355
1356 if (!hdr->pnfs_error) {
Benny Halevyd20581a2011-05-22 19:52:03 +03001357 pnfs_set_layoutcommit(data);
Fred Isamancd841602012-04-20 14:47:44 -04001358 hdr->mds_ops->rpc_call_done(&data->task, data);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001359 } else
1360 pnfs_ld_handle_write_error(data);
Fred Isamancd841602012-04-20 14:47:44 -04001361 hdr->mds_ops->rpc_release(data);
Fred Isaman44b83792011-03-03 15:13:44 +00001362}
Benny Halevyd20581a2011-05-22 19:52:03 +03001363EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
Fred Isaman44b83792011-03-03 15:13:44 +00001364
Trond Myklebustdce81292011-07-13 15:59:19 -04001365static void
1366pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1367 struct nfs_write_data *data)
1368{
Fred Isamancd841602012-04-20 14:47:44 -04001369 struct nfs_pgio_header *hdr = data->header;
1370
Fred Isaman6c75dc02012-04-20 14:47:47 -04001371 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1372 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1373 nfs_pageio_reset_write_mds(desc);
1374 desc->pg_recoalesce = 1;
1375 }
Trond Myklebustdce81292011-07-13 15:59:19 -04001376 nfs_writedata_release(data);
1377}
1378
1379static enum pnfs_try_status
Andy Adamson0382b742011-03-03 15:13:45 +00001380pnfs_try_to_write_data(struct nfs_write_data *wdata,
Trond Myklebustdce81292011-07-13 15:59:19 -04001381 const struct rpc_call_ops *call_ops,
1382 struct pnfs_layout_segment *lseg,
1383 int how)
Andy Adamson0382b742011-03-03 15:13:45 +00001384{
Fred Isamancd841602012-04-20 14:47:44 -04001385 struct nfs_pgio_header *hdr = wdata->header;
1386 struct inode *inode = hdr->inode;
Andy Adamson0382b742011-03-03 15:13:45 +00001387 enum pnfs_try_status trypnfs;
1388 struct nfs_server *nfss = NFS_SERVER(inode);
1389
Fred Isamancd841602012-04-20 14:47:44 -04001390 hdr->mds_ops = call_ops;
Andy Adamson0382b742011-03-03 15:13:45 +00001391
1392 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1393 inode->i_ino, wdata->args.count, wdata->args.offset, how);
Andy Adamson0382b742011-03-03 15:13:45 +00001394 trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001395 if (trypnfs != PNFS_NOT_ATTEMPTED)
Andy Adamson0382b742011-03-03 15:13:45 +00001396 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
Andy Adamson0382b742011-03-03 15:13:45 +00001397 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1398 return trypnfs;
1399}
1400
Trond Myklebustdce81292011-07-13 15:59:19 -04001401static void
1402pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
1403{
1404 struct nfs_write_data *data;
1405 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1406 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1407
1408 desc->pg_lseg = NULL;
1409 while (!list_empty(head)) {
1410 enum pnfs_try_status trypnfs;
1411
Fred Isaman6c75dc02012-04-20 14:47:47 -04001412 data = list_first_entry(head, struct nfs_write_data, list);
Trond Myklebustdce81292011-07-13 15:59:19 -04001413 list_del_init(&data->list);
1414
1415 trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
1416 if (trypnfs == PNFS_NOT_ATTEMPTED)
1417 pnfs_write_through_mds(desc, data);
1418 }
Trond Myklebust9369a432012-09-18 20:57:08 -04001419 pnfs_put_lseg(lseg);
Trond Myklebustdce81292011-07-13 15:59:19 -04001420}
1421
Fred Isaman6c75dc02012-04-20 14:47:47 -04001422static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
1423{
Trond Myklebust9369a432012-09-18 20:57:08 -04001424 pnfs_put_lseg(hdr->lseg);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001425 nfs_writehdr_free(hdr);
1426}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001427EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001428
Trond Myklebustdce81292011-07-13 15:59:19 -04001429int
1430pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1431{
Fred Isaman6c75dc02012-04-20 14:47:47 -04001432 struct nfs_write_header *whdr;
1433 struct nfs_pgio_header *hdr;
Trond Myklebustdce81292011-07-13 15:59:19 -04001434 int ret;
1435
Fred Isaman6c75dc02012-04-20 14:47:47 -04001436 whdr = nfs_writehdr_alloc();
1437 if (!whdr) {
Trond Myklebust9b5415b2012-04-27 14:31:47 -04001438 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
Trond Myklebust9369a432012-09-18 20:57:08 -04001439 pnfs_put_lseg(desc->pg_lseg);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001440 desc->pg_lseg = NULL;
1441 return -ENOMEM;
1442 }
1443 hdr = &whdr->header;
1444 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
Trond Myklebust9369a432012-09-18 20:57:08 -04001445 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001446 atomic_inc(&hdr->refcnt);
1447 ret = nfs_generic_flush(desc, hdr);
Trond Myklebustdce81292011-07-13 15:59:19 -04001448 if (ret != 0) {
Trond Myklebust9369a432012-09-18 20:57:08 -04001449 pnfs_put_lseg(desc->pg_lseg);
Trond Myklebustdce81292011-07-13 15:59:19 -04001450 desc->pg_lseg = NULL;
Fred Isaman6c75dc02012-04-20 14:47:47 -04001451 } else
1452 pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
1453 if (atomic_dec_and_test(&hdr->refcnt))
Fred Isaman061ae2e2012-04-20 14:47:48 -04001454 hdr->completion_ops->completion(hdr);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001455 return ret;
Trond Myklebustdce81292011-07-13 15:59:19 -04001456}
1457EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
1458
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001459int pnfs_read_done_resend_to_mds(struct inode *inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001460 struct list_head *head,
1461 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust62e4a762011-11-10 14:30:37 -05001462{
1463 struct nfs_pageio_descriptor pgio;
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001464 LIST_HEAD(failed);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001465
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001466 /* Resend all requests through the MDS */
Bryan Schumaker1abb50882012-06-20 15:53:47 -04001467 nfs_pageio_init_read(&pgio, inode, compl_ops);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001468 while (!list_empty(head)) {
1469 struct nfs_page *req = nfs_list_entry(head->next);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001470
1471 nfs_list_remove_request(req);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001472 if (!nfs_pageio_add_request(&pgio, req))
1473 nfs_list_add_request(req, &failed);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001474 }
1475 nfs_pageio_complete(&pgio);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001476
1477 if (!list_empty(&failed)) {
1478 list_move(&failed, head);
1479 return -EIO;
1480 }
1481 return 0;
1482}
Andy Adamsone7dd79af2012-04-27 17:53:46 -04001483EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001484
1485static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
1486{
Fred Isamancd841602012-04-20 14:47:44 -04001487 struct nfs_pgio_header *hdr = data->header;
1488
1489 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
1490 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001491 PNFS_LAYOUTRET_ON_ERROR) {
Fred Isamancd841602012-04-20 14:47:44 -04001492 clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
1493 pnfs_return_layout(hdr->inode);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04001494 }
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001495 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1496 data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001497 &hdr->pages,
1498 hdr->completion_ops);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001499}
1500
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001501/*
Benny Halevyd20581a2011-05-22 19:52:03 +03001502 * Called by non rpc-based layout drivers
1503 */
Peng Tao9b7eecd2011-09-22 21:50:15 -04001504void pnfs_ld_read_done(struct nfs_read_data *data)
Benny Halevyd20581a2011-05-22 19:52:03 +03001505{
Fred Isamancd841602012-04-20 14:47:44 -04001506 struct nfs_pgio_header *hdr = data->header;
1507
1508 if (likely(!hdr->pnfs_error)) {
Benny Halevyd20581a2011-05-22 19:52:03 +03001509 __nfs4_read_done_cb(data);
Fred Isamancd841602012-04-20 14:47:44 -04001510 hdr->mds_ops->rpc_call_done(&data->task, data);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001511 } else
1512 pnfs_ld_handle_read_error(data);
Fred Isamancd841602012-04-20 14:47:44 -04001513 hdr->mds_ops->rpc_release(data);
Benny Halevyd20581a2011-05-22 19:52:03 +03001514}
1515EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
1516
Trond Myklebust493292d2011-07-13 15:58:28 -04001517static void
1518pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1519 struct nfs_read_data *data)
1520{
Fred Isamancd841602012-04-20 14:47:44 -04001521 struct nfs_pgio_header *hdr = data->header;
1522
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001523 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1524 list_splice_tail_init(&hdr->pages, &desc->pg_list);
1525 nfs_pageio_reset_read_mds(desc);
1526 desc->pg_recoalesce = 1;
1527 }
Trond Myklebust493292d2011-07-13 15:58:28 -04001528 nfs_readdata_release(data);
1529}
1530
Benny Halevyd20581a2011-05-22 19:52:03 +03001531/*
Andy Adamson64419a92011-03-01 01:34:16 +00001532 * Call the appropriate parallel I/O subsystem read function.
1533 */
Trond Myklebust493292d2011-07-13 15:58:28 -04001534static enum pnfs_try_status
Andy Adamson64419a92011-03-01 01:34:16 +00001535pnfs_try_to_read_data(struct nfs_read_data *rdata,
Trond Myklebust493292d2011-07-13 15:58:28 -04001536 const struct rpc_call_ops *call_ops,
1537 struct pnfs_layout_segment *lseg)
Andy Adamson64419a92011-03-01 01:34:16 +00001538{
Fred Isamancd841602012-04-20 14:47:44 -04001539 struct nfs_pgio_header *hdr = rdata->header;
1540 struct inode *inode = hdr->inode;
Andy Adamson64419a92011-03-01 01:34:16 +00001541 struct nfs_server *nfss = NFS_SERVER(inode);
1542 enum pnfs_try_status trypnfs;
1543
Fred Isamancd841602012-04-20 14:47:44 -04001544 hdr->mds_ops = call_ops;
Andy Adamson64419a92011-03-01 01:34:16 +00001545
1546 dprintk("%s: Reading ino:%lu %u@%llu\n",
1547 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
1548
1549 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001550 if (trypnfs != PNFS_NOT_ATTEMPTED)
Andy Adamson64419a92011-03-01 01:34:16 +00001551 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
Andy Adamson64419a92011-03-01 01:34:16 +00001552 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1553 return trypnfs;
1554}
Andy Adamson863a3c62011-03-23 13:27:54 +00001555
Trond Myklebust493292d2011-07-13 15:58:28 -04001556static void
1557pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
1558{
1559 struct nfs_read_data *data;
1560 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1561 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1562
1563 desc->pg_lseg = NULL;
1564 while (!list_empty(head)) {
1565 enum pnfs_try_status trypnfs;
1566
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001567 data = list_first_entry(head, struct nfs_read_data, list);
Trond Myklebust493292d2011-07-13 15:58:28 -04001568 list_del_init(&data->list);
1569
1570 trypnfs = pnfs_try_to_read_data(data, call_ops, lseg);
1571 if (trypnfs == PNFS_NOT_ATTEMPTED)
1572 pnfs_read_through_mds(desc, data);
1573 }
Trond Myklebust9369a432012-09-18 20:57:08 -04001574 pnfs_put_lseg(lseg);
Trond Myklebust493292d2011-07-13 15:58:28 -04001575}
1576
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001577static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
1578{
Trond Myklebust9369a432012-09-18 20:57:08 -04001579 pnfs_put_lseg(hdr->lseg);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001580 nfs_readhdr_free(hdr);
1581}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001582EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001583
Trond Myklebust493292d2011-07-13 15:58:28 -04001584int
1585pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
1586{
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001587 struct nfs_read_header *rhdr;
1588 struct nfs_pgio_header *hdr;
Trond Myklebust493292d2011-07-13 15:58:28 -04001589 int ret;
1590
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001591 rhdr = nfs_readhdr_alloc();
1592 if (!rhdr) {
Fred Isaman061ae2e2012-04-20 14:47:48 -04001593 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001594 ret = -ENOMEM;
Trond Myklebust9369a432012-09-18 20:57:08 -04001595 pnfs_put_lseg(desc->pg_lseg);
Trond Myklebust493292d2011-07-13 15:58:28 -04001596 desc->pg_lseg = NULL;
1597 return ret;
1598 }
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001599 hdr = &rhdr->header;
1600 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
Trond Myklebust9369a432012-09-18 20:57:08 -04001601 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001602 atomic_inc(&hdr->refcnt);
1603 ret = nfs_generic_pagein(desc, hdr);
1604 if (ret != 0) {
Trond Myklebust9369a432012-09-18 20:57:08 -04001605 pnfs_put_lseg(desc->pg_lseg);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001606 desc->pg_lseg = NULL;
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001607 } else
1608 pnfs_do_multiple_reads(desc, &hdr->rpc_list);
1609 if (atomic_dec_and_test(&hdr->refcnt))
Fred Isaman061ae2e2012-04-20 14:47:48 -04001610 hdr->completion_ops->completion(hdr);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04001611 return ret;
Trond Myklebust493292d2011-07-13 15:58:28 -04001612}
1613EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
1614
Andy Adamson863a3c62011-03-23 13:27:54 +00001615/*
Peng Taoa9bae562011-07-30 20:52:33 -04001616 * There can be multiple RW segments.
Andy Adamson863a3c62011-03-23 13:27:54 +00001617 */
Peng Taoa9bae562011-07-30 20:52:33 -04001618static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
Andy Adamson863a3c62011-03-23 13:27:54 +00001619{
Peng Taoa9bae562011-07-30 20:52:33 -04001620 struct pnfs_layout_segment *lseg;
Andy Adamson863a3c62011-03-23 13:27:54 +00001621
Peng Taoa9bae562011-07-30 20:52:33 -04001622 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
1623 if (lseg->pls_range.iomode == IOMODE_RW &&
1624 test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1625 list_add(&lseg->pls_lc_list, listp);
1626 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001627}
1628
Peng Tao1b0ae062011-09-22 21:50:12 -04001629void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
1630{
Trond Myklebustb9e028f2012-09-18 16:41:18 -04001631 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
Peng Tao1b0ae062011-09-22 21:50:12 -04001632}
1633EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
1634
Andy Adamson863a3c62011-03-23 13:27:54 +00001635void
1636pnfs_set_layoutcommit(struct nfs_write_data *wdata)
1637{
Fred Isamancd841602012-04-20 14:47:44 -04001638 struct nfs_pgio_header *hdr = wdata->header;
1639 struct inode *inode = hdr->inode;
1640 struct nfs_inode *nfsi = NFS_I(inode);
Vitaliy Gusev4b8ee2b2011-05-20 01:34:46 +04001641 loff_t end_pos = wdata->mds_offset + wdata->res.count;
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001642 bool mark_as_dirty = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001643
Fred Isamancd841602012-04-20 14:47:44 -04001644 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001645 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001646 mark_as_dirty = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001647 dprintk("%s: Set layoutcommit for inode %lu ",
Fred Isamancd841602012-04-20 14:47:44 -04001648 __func__, inode->i_ino);
Andy Adamson863a3c62011-03-23 13:27:54 +00001649 }
Fred Isamancd841602012-04-20 14:47:44 -04001650 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
Peng Taoa9bae562011-07-30 20:52:33 -04001651 /* references matched in nfs4_layoutcommit_release */
Trond Myklebust9369a432012-09-18 20:57:08 -04001652 pnfs_get_lseg(hdr->lseg);
Peng Taoa9bae562011-07-30 20:52:33 -04001653 }
Peng Taoacff58802011-07-30 20:52:31 -04001654 if (end_pos > nfsi->layout->plh_lwb)
1655 nfsi->layout->plh_lwb = end_pos;
Fred Isamancd841602012-04-20 14:47:44 -04001656 spin_unlock(&inode->i_lock);
Peng Taoacff58802011-07-30 20:52:31 -04001657 dprintk("%s: lseg %p end_pos %llu\n",
Fred Isamancd841602012-04-20 14:47:44 -04001658 __func__, hdr->lseg, nfsi->layout->plh_lwb);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001659
1660 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1661 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1662 if (mark_as_dirty)
Fred Isamancd841602012-04-20 14:47:44 -04001663 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001664}
1665EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1666
Andy Adamsondb29c082011-07-30 20:52:38 -04001667void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
1668{
1669 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
1670
1671 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
1672 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
1673}
1674
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001675/*
1676 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1677 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1678 * data to disk to allow the server to recover the data if it crashes.
1679 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1680 * is off, and a COMMIT is sent to a data server, or
1681 * if WRITEs to a data server return NFS_DATA_SYNC.
1682 */
Andy Adamson863a3c62011-03-23 13:27:54 +00001683int
Andy Adamsonef311532011-03-12 02:58:10 -05001684pnfs_layoutcommit_inode(struct inode *inode, bool sync)
Andy Adamson863a3c62011-03-23 13:27:54 +00001685{
1686 struct nfs4_layoutcommit_data *data;
1687 struct nfs_inode *nfsi = NFS_I(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001688 loff_t end_pos;
1689 int status = 0;
1690
1691 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1692
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001693 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1694 return 0;
1695
Andy Adamson863a3c62011-03-23 13:27:54 +00001696 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1697 data = kzalloc(sizeof(*data), GFP_NOFS);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001698 if (!data) {
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001699 status = -ENOMEM;
1700 goto out;
1701 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001702
Peng Tao92407e72011-10-23 20:21:17 -07001703 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1704 goto out_free;
1705
1706 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
1707 if (!sync) {
1708 status = -EAGAIN;
1709 goto out_free;
1710 }
1711 status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING,
1712 nfs_wait_bit_killable, TASK_KILLABLE);
1713 if (status)
1714 goto out_free;
1715 }
1716
Peng Taoa9bae562011-07-30 20:52:33 -04001717 INIT_LIST_HEAD(&data->lseg_list);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001718 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001719 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Peng Tao92407e72011-10-23 20:21:17 -07001720 clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
Andy Adamson863a3c62011-03-23 13:27:54 +00001721 spin_unlock(&inode->i_lock);
Peng Tao92407e72011-10-23 20:21:17 -07001722 wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
1723 goto out_free;
Andy Adamson863a3c62011-03-23 13:27:54 +00001724 }
Peng Taoa9bae562011-07-30 20:52:33 -04001725
1726 pnfs_list_write_lseg(inode, &data->lseg_list);
Andy Adamson863a3c62011-03-23 13:27:54 +00001727
Peng Taoacff58802011-07-30 20:52:31 -04001728 end_pos = nfsi->layout->plh_lwb;
Peng Taoacff58802011-07-30 20:52:31 -04001729 nfsi->layout->plh_lwb = 0;
Andy Adamson863a3c62011-03-23 13:27:54 +00001730
Trond Myklebustf597c532012-03-04 18:13:56 -05001731 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
Andy Adamson863a3c62011-03-23 13:27:54 +00001732 spin_unlock(&inode->i_lock);
1733
1734 data->args.inode = inode;
Peng Tao9fa40752011-07-30 20:52:32 -04001735 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
Andy Adamson863a3c62011-03-23 13:27:54 +00001736 nfs_fattr_init(&data->fattr);
1737 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1738 data->res.fattr = &data->fattr;
1739 data->args.lastbytewritten = end_pos - 1;
1740 data->res.server = NFS_SERVER(inode);
1741
1742 status = nfs4_proc_layoutcommit(data, sync);
1743out:
Peng Tao92407e72011-10-23 20:21:17 -07001744 if (status)
1745 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001746 dprintk("<-- %s status %d\n", __func__, status);
1747 return status;
Peng Tao92407e72011-10-23 20:21:17 -07001748out_free:
1749 kfree(data);
1750 goto out;
Andy Adamson863a3c62011-03-23 13:27:54 +00001751}
Andy Adamson82be4172012-05-23 05:02:35 -04001752
1753struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
1754{
1755 struct nfs4_threshold *thp;
1756
1757 thp = kzalloc(sizeof(*thp), GFP_NOFS);
1758 if (!thp) {
1759 dprintk("%s mdsthreshold allocation failed\n", __func__);
1760 return NULL;
1761 }
1762 return thp;
1763}