blob: 306cba0b9e69af45b7fd5e57926dccb9c886ec77 [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>
Jeff Laytonca440c32016-09-15 14:40:49 -040033#include <linux/sort.h>
Andy Adamson974cec82010-10-20 00:18:02 -040034#include "internal.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040035#include "pnfs.h"
Andy Adamson64419a92011-03-01 01:34:16 +000036#include "iostat.h"
Trond Myklebustcc668ab2013-08-14 15:31:28 -040037#include "nfs4trace.h"
Trond Myklebust40dd4b72015-01-24 13:54:37 -050038#include "delegation.h"
Peng Tao87334082015-06-23 19:51:57 +080039#include "nfs42.h"
Fred Isaman1b146fc2016-09-21 15:24:26 -040040#include "nfs4_fs.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040041
42#define NFSDBG_FACILITY NFSDBG_PNFS
Trond Myklebust25c75332012-09-18 17:01:12 -040043#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040044
Fred Isaman02c35fc2010-10-20 00:17:59 -040045/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
Trond Myklebust13c13a62016-01-26 23:12:11 -050057static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
Trond Myklebust68f74472016-10-12 19:50:54 -040058static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
Trond Myklebustfe1cf942016-11-30 12:32:55 -050062static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
Peng Taoaa1e0e3a2014-09-06 00:53:25 +080064
Fred Isaman02c35fc2010-10-20 00:17:59 -040065/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040080static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
Fred Isaman02c35fc2010-10-20 00:17:59 -040083 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
Trond Myklebust0a9c63f2012-06-15 13:02:58 -040087 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
Fred Isaman02c35fc2010-10-20 00:17:59 -040091 spin_unlock(&pnfs_spinlock);
92 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040093}
94
Trond Myklebust7c9d8452022-03-28 08:36:34 -040095const struct pnfs_layoutdriver_type *pnfs_find_layoutdriver(u32 id)
96{
97 return find_pnfs_driver(id);
98}
99
100void pnfs_put_layoutdriver(const struct pnfs_layoutdriver_type *ld)
101{
102 if (ld)
103 module_put(ld->owner);
104}
105
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400106void
107unset_pnfs_layoutdriver(struct nfs_server *nfss)
108{
Benny Halevy738fd0f32011-07-30 20:52:36 -0400109 if (nfss->pnfs_curr_ld) {
110 if (nfss->pnfs_curr_ld->clear_layoutdriver)
111 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
Trond Myklebust2a4c8992012-06-14 13:08:38 -0400112 /* Decrement the MDS count. Purge the deviceid cache if zero */
113 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
114 nfs4_deviceid_purge_client(nfss->nfs_client);
Fred Isaman02c35fc2010-10-20 00:17:59 -0400115 module_put(nfss->pnfs_curr_ld->owner);
Benny Halevy738fd0f32011-07-30 20:52:36 -0400116 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400117 nfss->pnfs_curr_ld = NULL;
118}
119
120/*
Jeff Laytonca440c32016-09-15 14:40:49 -0400121 * When the server sends a list of layout types, we choose one in the order
122 * given in the list below.
123 *
124 * FIXME: should this list be configurable in some fashion? module param?
125 * mount option? something else?
126 */
127static const u32 ld_prefs[] = {
128 LAYOUT_SCSI,
129 LAYOUT_BLOCK_VOLUME,
130 LAYOUT_OSD2_OBJECTS,
131 LAYOUT_FLEX_FILES,
132 LAYOUT_NFSV4_1_FILES,
133 0
134};
135
136static int
137ld_cmp(const void *e1, const void *e2)
138{
139 u32 ld1 = *((u32 *)e1);
140 u32 ld2 = *((u32 *)e2);
141 int i;
142
143 for (i = 0; ld_prefs[i] != 0; i++) {
144 if (ld1 == ld_prefs[i])
145 return -1;
146
147 if (ld2 == ld_prefs[i])
148 return 1;
149 }
150 return 0;
151}
152
153/*
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400154 * Try to set the server's pnfs module to the pnfs layout type specified by id.
155 * Currently only one pNFS layout driver per filesystem is supported.
156 *
Jeff Layton3132e492016-08-10 15:58:24 -0400157 * @ids array of layout types supported by MDS.
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400158 */
159void
Benny Halevy738fd0f32011-07-30 20:52:36 -0400160set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
Jeff Laytonca440c32016-09-15 14:40:49 -0400161 struct nfs_fsinfo *fsinfo)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400162{
163 struct pnfs_layoutdriver_type *ld_type = NULL;
Jeff Layton3132e492016-08-10 15:58:24 -0400164 u32 id;
Jeff Laytonca440c32016-09-15 14:40:49 -0400165 int i;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400166
Anna Schumaker19274712016-10-26 15:54:31 -0400167 if (fsinfo->nlayouttypes == 0)
168 goto out_no_driver;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400169 if (!(server->nfs_client->cl_exchange_flags &
170 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
Jeff Layton3132e492016-08-10 15:58:24 -0400171 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
172 __func__, server->nfs_client->cl_exchange_flags);
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400173 goto out_no_driver;
174 }
Jeff Layton3132e492016-08-10 15:58:24 -0400175
Jeff Laytonca440c32016-09-15 14:40:49 -0400176 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
177 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
Jeff Layton3132e492016-08-10 15:58:24 -0400178
Jeff Laytonca440c32016-09-15 14:40:49 -0400179 for (i = 0; i < fsinfo->nlayouttypes; i++) {
180 id = fsinfo->layouttype[i];
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400181 ld_type = find_pnfs_driver(id);
Jeff Laytonca440c32016-09-15 14:40:49 -0400182 if (!ld_type) {
183 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
184 id);
185 ld_type = find_pnfs_driver(id);
186 }
187 if (ld_type)
188 break;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400189 }
Jeff Layton3132e492016-08-10 15:58:24 -0400190
191 if (!ld_type) {
Jeff Laytonca440c32016-09-15 14:40:49 -0400192 dprintk("%s: No pNFS module found!\n", __func__);
Jeff Layton3132e492016-08-10 15:58:24 -0400193 goto out_no_driver;
194 }
195
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400196 server->pnfs_curr_ld = ld_type;
Benny Halevy738fd0f32011-07-30 20:52:36 -0400197 if (ld_type->set_layoutdriver
198 && ld_type->set_layoutdriver(server, mntfh)) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500199 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
200 "driver %u.\n", __func__, id);
Benny Halevy738fd0f32011-07-30 20:52:36 -0400201 module_put(ld_type->owner);
202 goto out_no_driver;
203 }
Trond Myklebust2a4c8992012-06-14 13:08:38 -0400204 /* Bump the MDS count */
205 atomic_inc(&server->nfs_client->cl_mds_count);
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000206
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400207 dprintk("%s: pNFS module for %u set\n", __func__, id);
208 return;
209
210out_no_driver:
211 dprintk("%s: Using NFSv4 I/O\n", __func__);
212 server->pnfs_curr_ld = NULL;
213}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400214
215int
216pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
217{
218 int status = -EINVAL;
219 struct pnfs_layoutdriver_type *tmp;
220
221 if (ld_type->id == 0) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500222 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
Fred Isaman02c35fc2010-10-20 00:17:59 -0400223 return status;
224 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400225 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500226 printk(KERN_ERR "NFS: %s Layout driver must provide "
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400227 "alloc_lseg and free_lseg.\n", __func__);
228 return status;
229 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400230
231 spin_lock(&pnfs_spinlock);
232 tmp = find_pnfs_driver_locked(ld_type->id);
233 if (!tmp) {
234 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
235 status = 0;
236 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
237 ld_type->name);
238 } else {
Weston Andros Adamsona0308892012-01-26 13:32:23 -0500239 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
Fred Isaman02c35fc2010-10-20 00:17:59 -0400240 __func__, ld_type->id);
241 }
242 spin_unlock(&pnfs_spinlock);
243
244 return status;
245}
246EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
247
248void
249pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
250{
251 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
252 spin_lock(&pnfs_spinlock);
253 list_del(&ld_type->pnfs_tblid);
254 spin_unlock(&pnfs_spinlock);
255}
256EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400257
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400258/*
259 * pNFS client layout cache
260 */
261
Fred Isamancc6e5342011-01-06 11:36:28 +0000262/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000263void
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400264pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400265{
Elena Reshetova2b28a7b2017-10-20 12:53:33 +0300266 refcount_inc(&lo->plh_refcount);
Fred Isamancc6e5342011-01-06 11:36:28 +0000267}
268
Benny Halevy636fb9c2011-05-22 19:51:33 +0300269static struct pnfs_layout_hdr *
270pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
271{
272 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
Trond Myklebust57934272012-09-20 20:37:23 -0400273 return ld->alloc_layout_hdr(ino, gfp_flags);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300274}
275
276static void
277pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
278{
Trond Myklebust9c626382012-09-20 17:31:43 -0400279 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
280 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
281
Trond Myklebustcf6605d2020-02-18 17:14:40 -0500282 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
Trond Myklebust9c626382012-09-20 17:31:43 -0400283 struct nfs_client *clp = server->nfs_client;
284
285 spin_lock(&clp->cl_lock);
Trond Myklebustcf6605d2020-02-18 17:14:40 -0500286 list_del_rcu(&lo->plh_layouts);
Trond Myklebust9c626382012-09-20 17:31:43 -0400287 spin_unlock(&clp->cl_lock);
288 }
NeilBrowna52458b2018-12-03 11:30:31 +1100289 put_cred(lo->plh_lc_cred);
Trond Myklebust57934272012-09-20 20:37:23 -0400290 return ld->free_layout_hdr(lo);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300291}
292
Fred Isamancc6e5342011-01-06 11:36:28 +0000293static void
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400294pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
Fred Isamancc6e5342011-01-06 11:36:28 +0000295{
Trond Myklebustbb346f62012-09-20 15:52:13 -0400296 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
Fred Isamancc6e5342011-01-06 11:36:28 +0000297 dprintk("%s: freeing layout cache %p\n", __func__, lo);
Trond Myklebustbb346f62012-09-20 15:52:13 -0400298 nfsi->layout = NULL;
299 /* Reset MDS Threshold I/O counters */
300 nfsi->write_io = 0;
301 nfsi->read_io = 0;
Benny Halevye5e94012010-10-20 00:18:01 -0400302}
303
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400304void
Trond Myklebust70c3bd22012-09-18 20:51:13 -0400305pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400306{
Trond Myklebust9c6376e2018-03-07 14:49:06 -0500307 struct inode *inode;
Trond Myklebustb6d49ec2020-11-25 12:06:14 -0500308 unsigned long i_state;
Fred Isamancc6e5342011-01-06 11:36:28 +0000309
Trond Myklebust9c6376e2018-03-07 14:49:06 -0500310 if (!lo)
311 return;
312 inode = lo->plh_inode;
Trond Myklebust13c13a62016-01-26 23:12:11 -0500313 pnfs_layoutreturn_before_put_layout_hdr(lo);
314
Elena Reshetova2b28a7b2017-10-20 12:53:33 +0300315 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
Peng Tao566f8732014-10-10 23:25:46 +0800316 if (!list_empty(&lo->plh_segs))
317 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400318 pnfs_detach_layout_hdr(lo);
Trond Myklebustb6d49ec2020-11-25 12:06:14 -0500319 i_state = inode->i_state;
Fred Isamancc6e5342011-01-06 11:36:28 +0000320 spin_unlock(&inode->i_lock);
Trond Myklebust6622c3e2012-09-20 17:23:11 -0400321 pnfs_free_layout_hdr(lo);
Trond Myklebustb6d49ec2020-11-25 12:06:14 -0500322 /* Notify pnfs_destroy_layout_final() that we're done */
323 if (i_state & (I_FREEING | I_CLEAR))
324 wake_up_var(lo);
Fred Isamancc6e5342011-01-06 11:36:28 +0000325 }
Andy Adamson974cec82010-10-20 00:18:02 -0400326}
327
Trond Myklebustb5fdf842020-02-18 15:58:31 -0500328static struct inode *
329pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
330{
331 struct inode *inode = igrab(lo->plh_inode);
332 if (inode)
333 return inode;
334 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
335 return NULL;
336}
337
Trond Myklebust1bcf34f2021-01-21 13:51:50 -0500338/*
339 * Compare 2 layout stateid sequence ids, to see which is newer,
340 * taking into account wraparound issues.
341 */
342static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
343{
344 return (s32)(s1 - s2) > 0;
345}
346
347static void pnfs_barrier_update(struct pnfs_layout_hdr *lo, u32 newseq)
348{
Trond Myklebust45baada2021-07-23 08:57:20 -0400349 if (pnfs_seqid_is_newer(newseq, lo->plh_barrier) || !lo->plh_barrier)
Trond Myklebust1bcf34f2021-01-21 13:51:50 -0500350 lo->plh_barrier = newseq;
351}
352
Trond Myklebustae5a4592016-11-14 14:34:18 -0500353static void
Trond Myklebust4aab9732016-11-30 10:47:48 -0500354pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
355 u32 seq)
356{
357 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
358 iomode = IOMODE_ANY;
359 lo->plh_return_iomode = iomode;
360 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
Trond Myklebuste20772c2021-07-26 07:58:49 -0400361 /*
362 * We must set lo->plh_return_seq to avoid livelocks with
363 * pnfs_layout_need_return()
364 */
365 if (seq == 0)
366 seq = be32_to_cpu(lo->plh_stateid.seqid);
367 if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq))
Trond Myklebust4aab9732016-11-30 10:47:48 -0500368 lo->plh_return_seq = seq;
Trond Myklebuste20772c2021-07-26 07:58:49 -0400369 pnfs_barrier_update(lo, seq);
Trond Myklebust4aab9732016-11-30 10:47:48 -0500370}
371
372static void
Trond Myklebustae5a4592016-11-14 14:34:18 -0500373pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
374{
Trond Myklebust5466d212017-05-01 17:03:44 -0400375 struct pnfs_layout_segment *lseg;
Trond Myklebustae5a4592016-11-14 14:34:18 -0500376 lo->plh_return_iomode = 0;
377 lo->plh_return_seq = 0;
378 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
Trond Myklebust5466d212017-05-01 17:03:44 -0400379 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
380 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
381 continue;
382 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
383 }
Trond Myklebustae5a4592016-11-14 14:34:18 -0500384}
385
Trond Myklebust362fb5782016-12-05 17:33:07 -0500386static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
387{
388 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
389 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
390 smp_mb__after_atomic();
391 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
392 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
393}
394
Trond Myklebustfe1cf942016-11-30 12:32:55 -0500395static void
396pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
397 struct list_head *free_me)
398{
399 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
400 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
401 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
402 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
403 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
404 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
405}
406
Trond Myklebust2454dfe2016-02-22 17:34:59 -0500407/*
Trond Myklebust30cb3ee2019-09-20 07:23:45 -0400408 * Update the seqid of a layout stateid after receiving
409 * NFS4ERR_OLD_STATEID
Trond Myklebust73800202017-11-06 15:28:07 -0500410 */
Trond Myklebust30cb3ee2019-09-20 07:23:45 -0400411bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
Trond Myklebustecf84022018-08-15 21:35:46 -0400412 struct pnfs_layout_range *dst_range,
413 struct inode *inode)
Trond Myklebust73800202017-11-06 15:28:07 -0500414{
415 struct pnfs_layout_hdr *lo;
Trond Myklebustc16467d2018-07-29 22:39:15 -0400416 struct pnfs_layout_range range = {
417 .iomode = IOMODE_ANY,
418 .offset = 0,
419 .length = NFS4_MAX_UINT64,
420 };
Trond Myklebust73800202017-11-06 15:28:07 -0500421 bool ret = false;
Trond Myklebustc16467d2018-07-29 22:39:15 -0400422 LIST_HEAD(head);
423 int err;
Trond Myklebust73800202017-11-06 15:28:07 -0500424
425 spin_lock(&inode->i_lock);
426 lo = NFS_I(inode)->layout;
Trond Myklebust30cb3ee2019-09-20 07:23:45 -0400427 if (lo && pnfs_layout_is_valid(lo) &&
428 nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
429 /* Is our call using the most recent seqid? If so, bump it */
430 if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
431 nfs4_stateid_seqid_inc(dst);
432 ret = true;
433 goto out;
434 }
435 /* Try to update the seqid to the most recent */
Trond Myklebustc16467d2018-07-29 22:39:15 -0400436 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
437 if (err != -EBUSY) {
438 dst->seqid = lo->plh_stateid.seqid;
Trond Myklebustecf84022018-08-15 21:35:46 -0400439 *dst_range = range;
Trond Myklebustc16467d2018-07-29 22:39:15 -0400440 ret = true;
441 }
Trond Myklebust73800202017-11-06 15:28:07 -0500442 }
Trond Myklebust30cb3ee2019-09-20 07:23:45 -0400443out:
Trond Myklebust73800202017-11-06 15:28:07 -0500444 spin_unlock(&inode->i_lock);
Trond Myklebustc16467d2018-07-29 22:39:15 -0400445 pnfs_free_lseg_list(&head);
Trond Myklebust73800202017-11-06 15:28:07 -0500446 return ret;
447}
448
449/*
Trond Myklebust2454dfe2016-02-22 17:34:59 -0500450 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
451 *
452 * In order to continue using the pnfs_layout_hdr, a full recovery
453 * is required.
454 * Note that caller must hold inode->i_lock.
455 */
Trond Myklebust5f46be02016-07-22 11:25:27 -0400456int
Trond Myklebust2454dfe2016-02-22 17:34:59 -0500457pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
458 struct list_head *lseg_list)
459{
460 struct pnfs_layout_range range = {
461 .iomode = IOMODE_ANY,
462 .offset = 0,
463 .length = NFS4_MAX_UINT64,
464 };
Trond Myklebustfe1cf942016-11-30 12:32:55 -0500465 struct pnfs_layout_segment *lseg, *next;
Trond Myklebust2454dfe2016-02-22 17:34:59 -0500466
467 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
Trond Myklebustfe1cf942016-11-30 12:32:55 -0500468 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
469 pnfs_clear_lseg_state(lseg, lseg_list);
Trond Myklebust5466d212017-05-01 17:03:44 -0400470 pnfs_clear_layoutreturn_info(lo);
Trond Myklebust68f74472016-10-12 19:50:54 -0400471 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
Trond Myklebust880265c2022-05-31 11:03:07 -0400472 set_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags);
Trond Myklebust362fb5782016-12-05 17:33:07 -0500473 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
474 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
475 pnfs_clear_layoutreturn_waitbit(lo);
Trond Myklebustfe1cf942016-11-30 12:32:55 -0500476 return !list_empty(&lo->plh_segs);
Trond Myklebust2454dfe2016-02-22 17:34:59 -0500477}
478
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400479static int
480pnfs_iomode_to_fail_bit(u32 iomode)
481{
482 return iomode == IOMODE_RW ?
483 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
484}
485
486static void
Trond Myklebust3e621212012-09-24 13:07:16 -0400487pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400488{
Trond Myklebust25c75332012-09-18 17:01:12 -0400489 lo->plh_retry_timestamp = jiffies;
Yanchuan Nian39e88fc2013-01-04 20:19:49 +0800490 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
Elena Reshetova2b28a7b2017-10-20 12:53:33 +0300491 refcount_inc(&lo->plh_refcount);
Trond Myklebust3e621212012-09-24 13:07:16 -0400492}
493
494static void
495pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
496{
497 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
Elena Reshetova2b28a7b2017-10-20 12:53:33 +0300498 refcount_dec(&lo->plh_refcount);
Trond Myklebust3e621212012-09-24 13:07:16 -0400499}
500
501static void
502pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
503{
504 struct inode *inode = lo->plh_inode;
Trond Myklebust115ce572012-09-20 21:19:43 -0400505 struct pnfs_layout_range range = {
506 .iomode = iomode,
507 .offset = 0,
508 .length = NFS4_MAX_UINT64,
509 };
510 LIST_HEAD(head);
Trond Myklebust3e621212012-09-24 13:07:16 -0400511
512 spin_lock(&inode->i_lock);
513 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
Olga Kornievskaia28d44112023-01-20 13:40:20 -0500514 pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
Trond Myklebust3e621212012-09-24 13:07:16 -0400515 spin_unlock(&inode->i_lock);
Trond Myklebust115ce572012-09-20 21:19:43 -0400516 pnfs_free_lseg_list(&head);
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400517 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
518 iomode == IOMODE_RW ? "RW" : "READ");
519}
520
521static bool
522pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
523{
Trond Myklebust25c75332012-09-18 17:01:12 -0400524 unsigned long start, end;
Trond Myklebust3e621212012-09-24 13:07:16 -0400525 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
526
527 if (test_bit(fail_bit, &lo->plh_flags) == 0)
Trond Myklebust25c75332012-09-18 17:01:12 -0400528 return false;
529 end = jiffies;
530 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
531 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
532 /* It is time to retry the failed layoutgets */
Trond Myklebust3e621212012-09-24 13:07:16 -0400533 pnfs_layout_clear_fail_bit(lo, fail_bit);
Trond Myklebust25c75332012-09-18 17:01:12 -0400534 return false;
535 }
536 return true;
Trond Myklebustb9e028f2012-09-18 16:41:18 -0400537}
538
Andy Adamson974cec82010-10-20 00:18:02 -0400539static void
Trond Myklebust119cef972016-07-24 15:10:12 -0400540pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
541 const struct pnfs_layout_range *range,
542 const nfs4_stateid *stateid)
Andy Adamson974cec82010-10-20 00:18:02 -0400543{
Fred Isaman566052c2011-01-06 11:36:20 +0000544 INIT_LIST_HEAD(&lseg->pls_list);
Peng Taoa9bae562011-07-30 20:52:33 -0400545 INIT_LIST_HEAD(&lseg->pls_lc_list);
Trond Myklebusta9901892020-03-20 16:04:06 -0400546 INIT_LIST_HEAD(&lseg->pls_commits);
Elena Reshetovaeba6dd62017-10-20 12:53:32 +0300547 refcount_set(&lseg->pls_refcount, 1);
Fred Isaman4541d162011-01-06 11:36:23 +0000548 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000549 lseg->pls_layout = lo;
Trond Myklebust119cef972016-07-24 15:10:12 -0400550 lseg->pls_range = *range;
551 lseg->pls_seq = be32_to_cpu(stateid->seqid);
Andy Adamson974cec82010-10-20 00:18:02 -0400552}
553
Trond Myklebust905ca192012-09-20 20:46:49 -0400554static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400555{
Trond Myklebust68f74472016-10-12 19:50:54 -0400556 if (lseg != NULL) {
557 struct inode *inode = lseg->pls_layout->plh_inode;
558 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
559 }
Andy Adamson974cec82010-10-20 00:18:02 -0400560}
561
Fred Isamand684d2a2011-03-01 01:34:13 +0000562static void
Trond Myklebust57036a32012-09-20 16:33:30 -0400563pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
564 struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400565{
Benny Halevyd20581a2011-05-22 19:52:03 +0300566 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000567 list_del_init(&lseg->pls_list);
Trond Myklebust8f0d27d2012-09-20 20:57:11 -0400568 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
Elena Reshetova2b28a7b2017-10-20 12:53:33 +0300569 refcount_dec(&lo->plh_refcount);
Trond Myklebustabb3e1c2016-11-30 11:38:10 -0500570 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
571 return;
Trond Myklebust7b650992016-11-14 13:10:48 -0500572 if (list_empty(&lo->plh_segs) &&
573 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
574 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
Trond Myklebust334a8f32016-09-04 12:46:35 -0400575 if (atomic_read(&lo->plh_outstanding) == 0)
576 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
Trond Myklebust173f77e9c2012-09-21 15:49:42 -0400577 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
Trond Myklebust2d148c72016-06-17 16:48:26 -0400578 }
Fred Isamand684d2a2011-03-01 01:34:13 +0000579}
580
Trond Myklebust68f74472016-10-12 19:50:54 -0400581static bool
582pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
583 struct pnfs_layout_segment *lseg)
584{
585 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
586 pnfs_layout_is_valid(lo)) {
Trond Myklebust4aab9732016-11-30 10:47:48 -0500587 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
Trond Myklebust68f74472016-10-12 19:50:54 -0400588 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
589 return true;
590 }
591 return false;
592}
593
Fred Isamanbae724e2011-03-01 01:34:15 +0000594void
Trond Myklebust9369a432012-09-18 20:57:08 -0400595pnfs_put_lseg(struct pnfs_layout_segment *lseg)
Fred Isamand684d2a2011-03-01 01:34:13 +0000596{
Trond Myklebust57036a32012-09-20 16:33:30 -0400597 struct pnfs_layout_hdr *lo;
Fred Isamand684d2a2011-03-01 01:34:13 +0000598 struct inode *inode;
599
600 if (!lseg)
601 return;
602
Fred Isaman4541d162011-01-06 11:36:23 +0000603 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
Elena Reshetovaeba6dd62017-10-20 12:53:32 +0300604 refcount_read(&lseg->pls_refcount),
Fred Isaman4541d162011-01-06 11:36:23 +0000605 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Trond Myklebust4ef2e4f2015-02-05 17:27:39 -0500606
Trond Myklebust57036a32012-09-20 16:33:30 -0400607 lo = lseg->pls_layout;
608 inode = lo->plh_inode;
Trond Myklebust4ef2e4f2015-02-05 17:27:39 -0500609
Elena Reshetovaeba6dd62017-10-20 12:53:32 +0300610 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
Trond Myklebust8f0d27d2012-09-20 20:57:11 -0400611 pnfs_get_layout_hdr(lo);
Trond Myklebust4ef2e4f2015-02-05 17:27:39 -0500612 pnfs_layout_remove_lseg(lo, lseg);
Trond Myklebust68f74472016-10-12 19:50:54 -0400613 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
614 lseg = NULL;
Trond Myklebust4ef2e4f2015-02-05 17:27:39 -0500615 spin_unlock(&inode->i_lock);
616 pnfs_free_lseg(lseg);
617 pnfs_put_layout_hdr(lo);
Fred Isaman4541d162011-01-06 11:36:23 +0000618 }
Andy Adamson974cec82010-10-20 00:18:02 -0400619}
Trond Myklebust9369a432012-09-18 20:57:08 -0400620EXPORT_SYMBOL_GPL(pnfs_put_lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400621
Benny Halevyfb3296e2011-05-22 19:47:26 +0300622/*
623 * is l2 fully contained in l1?
624 * start1 end1
625 * [----------------------------------)
626 * start2 end2
627 * [----------------)
628 */
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400629static bool
Trond Myklebust7dc0ac72013-06-03 11:30:24 -0400630pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
Trond Myklebust3cb2df12013-06-03 11:24:36 -0400631 const struct pnfs_layout_range *l2)
Benny Halevyfb3296e2011-05-22 19:47:26 +0300632{
633 u64 start1 = l1->offset;
Trond Myklebust17822b22016-10-25 12:24:25 -0400634 u64 end1 = pnfs_end_offset(start1, l1->length);
Benny Halevyfb3296e2011-05-22 19:47:26 +0300635 u64 start2 = l2->offset;
Trond Myklebust17822b22016-10-25 12:24:25 -0400636 u64 end2 = pnfs_end_offset(start2, l2->length);
Benny Halevyfb3296e2011-05-22 19:47:26 +0300637
638 return (start1 <= start2) && (end1 >= end2);
639}
640
Trond Myklebust24956802013-03-20 13:03:00 -0400641static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
642 struct list_head *tmp_list)
643{
Elena Reshetovaeba6dd62017-10-20 12:53:32 +0300644 if (!refcount_dec_and_test(&lseg->pls_refcount))
Trond Myklebust24956802013-03-20 13:03:00 -0400645 return false;
646 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
647 list_add(&lseg->pls_list, tmp_list);
648 return true;
649}
650
Fred Isaman4541d162011-01-06 11:36:23 +0000651/* Returns 1 if lseg is removed from list, 0 otherwise */
652static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
653 struct list_head *tmp_list)
654{
655 int rv = 0;
656
657 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
658 /* Remove the reference keeping the lseg in the
659 * list. It will now be removed when all
660 * outstanding io is finished.
661 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000662 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
Elena Reshetovaeba6dd62017-10-20 12:53:32 +0300663 refcount_read(&lseg->pls_refcount));
Trond Myklebust24956802013-03-20 13:03:00 -0400664 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
Fred Isamand684d2a2011-03-01 01:34:13 +0000665 rv = 1;
Fred Isaman4541d162011-01-06 11:36:23 +0000666 }
667 return rv;
668}
669
Trond Myklebuste036f462016-07-22 11:13:22 -0400670static bool
671pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
672 const struct pnfs_layout_range *recall_range)
673{
674 return (recall_range->iomode == IOMODE_ANY ||
675 lseg_range->iomode == recall_range->iomode) &&
676 pnfs_lseg_range_intersecting(lseg_range, recall_range);
677}
678
679static bool
680pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
681 const struct pnfs_layout_range *recall_range,
682 u32 seq)
683{
684 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
685 return false;
686 if (recall_range == NULL)
687 return true;
688 return pnfs_should_free_range(&lseg->pls_range, recall_range);
689}
690
Jeff Layton6d597e12016-05-17 12:28:42 -0400691/**
692 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
693 * @lo: layout header containing the lsegs
694 * @tmp_list: list head where doomed lsegs should go
695 * @recall_range: optional recall range argument to match (may be NULL)
696 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
697 *
698 * Walk the list of lsegs in the layout header, and tear down any that should
699 * be destroyed. If "recall_range" is specified then the segment must match
700 * that range. If "seq" is non-zero, then only match segments that were handed
701 * out at or before that sequence.
702 *
703 * Returns number of matching invalid lsegs remaining in list after scanning
704 * it and purging them.
Fred Isaman4541d162011-01-06 11:36:23 +0000705 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000706int
Trond Myklebust49a85062012-09-18 20:43:31 -0400707pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
Fred Isaman4541d162011-01-06 11:36:23 +0000708 struct list_head *tmp_list,
Jeff Layton6d597e12016-05-17 12:28:42 -0400709 const struct pnfs_layout_range *recall_range,
710 u32 seq)
Andy Adamson974cec82010-10-20 00:18:02 -0400711{
712 struct pnfs_layout_segment *lseg, *next;
Trond Myklebustb739a5b2022-10-05 15:57:38 -0400713 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
Trond Myklebust71b39852016-01-04 12:41:15 -0500714 int remaining = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400715
716 dprintk("%s:Begin lo %p\n", __func__, lo);
717
Trond Myklebust8006bfb2012-09-21 14:48:04 -0400718 if (list_empty(&lo->plh_segs))
Fred Isaman385117222011-02-03 18:28:50 +0000719 return 0;
Fred Isaman4541d162011-01-06 11:36:23 +0000720 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
Trond Myklebuste036f462016-07-22 11:13:22 -0400721 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
Benjamin Coddingtonb3dce6a2017-12-08 12:52:59 -0500722 dprintk("%s: freeing lseg %p iomode %d seq %u "
Fred Isaman4541d162011-01-06 11:36:23 +0000723 "offset %llu length %llu\n", __func__,
Jeff Layton6d597e12016-05-17 12:28:42 -0400724 lseg, lseg->pls_range.iomode, lseg->pls_seq,
725 lseg->pls_range.offset, lseg->pls_range.length);
Trond Myklebustb739a5b2022-10-05 15:57:38 -0400726 if (mark_lseg_invalid(lseg, tmp_list))
727 continue;
728 remaining++;
729 pnfs_lseg_cancel_io(server, lseg);
Fred Isaman4541d162011-01-06 11:36:23 +0000730 }
Trond Myklebust71b39852016-01-04 12:41:15 -0500731 dprintk("%s:Return %i\n", __func__, remaining);
732 return remaining;
Andy Adamson974cec82010-10-20 00:18:02 -0400733}
734
Trond Myklebust68f74472016-10-12 19:50:54 -0400735static void
736pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
737 struct list_head *free_me,
738 const struct pnfs_layout_range *range,
739 u32 seq)
740{
741 struct pnfs_layout_segment *lseg, *next;
742
743 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
744 if (pnfs_match_lseg_recall(lseg, range, seq))
745 list_move_tail(&lseg->pls_list, free_me);
746 }
747}
748
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000749/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000750void
Fred Isaman4541d162011-01-06 11:36:23 +0000751pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400752{
Fred Isaman4541d162011-01-06 11:36:23 +0000753 struct pnfs_layout_segment *lseg, *tmp;
Andy Adamson974cec82010-10-20 00:18:02 -0400754
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000755 if (list_empty(free_me))
756 return;
757
Fred Isaman4541d162011-01-06 11:36:23 +0000758 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000759 list_del(&lseg->pls_list);
Trond Myklebust905ca192012-09-20 20:46:49 -0400760 pnfs_free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400761 }
762}
763
Trond Myklebustb6d49ec2020-11-25 12:06:14 -0500764static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi)
Benny Halevye5e94012010-10-20 00:18:01 -0400765{
766 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400767 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400768
769 spin_lock(&nfsi->vfs_inode.i_lock);
770 lo = nfsi->layout;
771 if (lo) {
Trond Myklebust3e621212012-09-24 13:07:16 -0400772 pnfs_get_layout_hdr(lo);
Trond Myklebust2454dfe2016-02-22 17:34:59 -0500773 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
Trond Myklebust3e621212012-09-24 13:07:16 -0400774 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
775 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
776 spin_unlock(&nfsi->vfs_inode.i_lock);
777 pnfs_free_lseg_list(&tmp_list);
Trond Myklebust1f18b822017-04-29 10:10:17 -0400778 nfs_commit_inode(&nfsi->vfs_inode, 0);
Trond Myklebust3e621212012-09-24 13:07:16 -0400779 pnfs_put_layout_hdr(lo);
780 } else
781 spin_unlock(&nfsi->vfs_inode.i_lock);
Trond Myklebustb6d49ec2020-11-25 12:06:14 -0500782 return lo;
783}
784
785void pnfs_destroy_layout(struct nfs_inode *nfsi)
786{
787 __pnfs_destroy_layout(nfsi);
Benny Halevye5e94012010-10-20 00:18:01 -0400788}
Andy Adamson041245c2012-04-27 17:53:53 -0400789EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
Benny Halevye5e94012010-10-20 00:18:01 -0400790
Trond Myklebustb6d49ec2020-11-25 12:06:14 -0500791static bool pnfs_layout_removed(struct nfs_inode *nfsi,
792 struct pnfs_layout_hdr *lo)
793{
794 bool ret;
795
796 spin_lock(&nfsi->vfs_inode.i_lock);
797 ret = nfsi->layout != lo;
798 spin_unlock(&nfsi->vfs_inode.i_lock);
799 return ret;
800}
801
802void pnfs_destroy_layout_final(struct nfs_inode *nfsi)
803{
804 struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi);
805
806 if (lo)
807 wait_var_event(lo, pnfs_layout_removed(nfsi, lo));
808}
809
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500810static bool
811pnfs_layout_add_bulk_destroy_list(struct inode *inode,
812 struct list_head *layout_list)
813{
814 struct pnfs_layout_hdr *lo;
815 bool ret = false;
816
817 spin_lock(&inode->i_lock);
818 lo = NFS_I(inode)->layout;
819 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
820 pnfs_get_layout_hdr(lo);
821 list_add(&lo->plh_bulk_destroy, layout_list);
822 ret = true;
823 }
824 spin_unlock(&inode->i_lock);
825 return ret;
826}
827
828/* Caller must hold rcu_read_lock and clp->cl_lock */
829static int
830pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
831 struct nfs_server *server,
832 struct list_head *layout_list)
Trond Myklebust50856072019-02-22 14:20:27 -0500833 __must_hold(&clp->cl_lock)
834 __must_hold(RCU)
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500835{
836 struct pnfs_layout_hdr *lo, *next;
837 struct inode *inode;
838
839 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
Trond Myklebust50856072019-02-22 14:20:27 -0500840 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
841 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
842 !list_empty(&lo->plh_bulk_destroy))
Trond Myklebustb85f5622016-11-30 18:00:07 -0500843 continue;
Trond Myklebust50856072019-02-22 14:20:27 -0500844 /* If the sb is being destroyed, just bail */
845 if (!nfs_sb_active(server->super))
846 break;
Trond Myklebustb5fdf842020-02-18 15:58:31 -0500847 inode = pnfs_grab_inode_layout_hdr(lo);
Trond Myklebust50856072019-02-22 14:20:27 -0500848 if (inode != NULL) {
Trond Myklebustcf6605d2020-02-18 17:14:40 -0500849 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
850 list_del_rcu(&lo->plh_layouts);
Trond Myklebust50856072019-02-22 14:20:27 -0500851 if (pnfs_layout_add_bulk_destroy_list(inode,
852 layout_list))
853 continue;
854 rcu_read_unlock();
855 spin_unlock(&clp->cl_lock);
856 iput(inode);
857 } else {
858 rcu_read_unlock();
859 spin_unlock(&clp->cl_lock);
Trond Myklebust50856072019-02-22 14:20:27 -0500860 }
861 nfs_sb_deactive(server->super);
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500862 spin_lock(&clp->cl_lock);
863 rcu_read_lock();
864 return -EAGAIN;
865 }
866 return 0;
867}
868
869static int
870pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
871 bool is_bulk_recall)
872{
873 struct pnfs_layout_hdr *lo;
874 struct inode *inode;
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500875 LIST_HEAD(lseg_list);
876 int ret = 0;
877
878 while (!list_empty(layout_list)) {
879 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
880 plh_bulk_destroy);
881 dprintk("%s freeing layout for inode %lu\n", __func__,
882 lo->plh_inode->i_ino);
883 inode = lo->plh_inode;
Christoph Hellwig7c5d1872014-09-10 08:23:29 -0700884
885 pnfs_layoutcommit_inode(inode, false);
886
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500887 spin_lock(&inode->i_lock);
888 list_del_init(&lo->plh_bulk_destroy);
Trond Myklebust9fd4b9f2016-02-22 17:46:34 -0500889 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
890 if (is_bulk_recall)
891 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500892 ret = -EAGAIN;
Trond Myklebust9fd4b9f2016-02-22 17:46:34 -0500893 }
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500894 spin_unlock(&inode->i_lock);
895 pnfs_free_lseg_list(&lseg_list);
Trond Myklebustb20135d2015-12-31 09:28:06 -0500896 /* Free all lsegs that are attached to commit buckets */
897 nfs_commit_inode(inode, 0);
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500898 pnfs_put_layout_hdr(lo);
Trond Myklebust50856072019-02-22 14:20:27 -0500899 nfs_iput_and_deactive(inode);
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500900 }
901 return ret;
902}
903
904int
905pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
906 struct nfs_fsid *fsid,
907 bool is_recall)
908{
909 struct nfs_server *server;
910 LIST_HEAD(layout_list);
911
912 spin_lock(&clp->cl_lock);
913 rcu_read_lock();
914restart:
915 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
916 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
917 continue;
918 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
919 server,
920 &layout_list) != 0)
921 goto restart;
922 }
923 rcu_read_unlock();
924 spin_unlock(&clp->cl_lock);
925
926 if (list_empty(&layout_list))
927 return 0;
928 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
929}
930
931int
932pnfs_destroy_layouts_byclid(struct nfs_client *clp,
933 bool is_recall)
934{
935 struct nfs_server *server;
936 LIST_HEAD(layout_list);
937
938 spin_lock(&clp->cl_lock);
939 rcu_read_lock();
940restart:
941 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
942 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
943 server,
944 &layout_list) != 0)
945 goto restart;
946 }
947 rcu_read_unlock();
948 spin_unlock(&clp->cl_lock);
949
950 if (list_empty(&layout_list))
951 return 0;
952 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
953}
954
Andy Adamson974cec82010-10-20 00:18:02 -0400955/*
Wang Qing9f266452020-09-17 10:19:00 +0800956 * Called by the state manager to remove all layouts established under an
Andy Adamson974cec82010-10-20 00:18:02 -0400957 * expired lease.
958 */
959void
960pnfs_destroy_all_layouts(struct nfs_client *clp)
961{
Andy Adamsonc47abcf2011-06-15 17:52:40 -0400962 nfs4_deviceid_mark_client_invalid(clp);
963 nfs4_deviceid_purge_client(clp);
964
Trond Myklebustfd9a8d72013-02-12 09:48:42 -0500965 pnfs_destroy_layouts_byclid(clp, false);
Andy Adamson974cec82010-10-20 00:18:02 -0400966}
967
Trond Myklebust59b56392020-01-27 13:07:26 -0500968static void
969pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
970{
971 const struct cred *old;
972
973 if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
974 old = xchg(&lo->plh_lc_cred, get_cred(cred));
975 put_cred(old);
976 }
977}
978
Fred Isamanfd6002e2011-01-06 11:36:22 +0000979/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000980void
981pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
Trond Myklebust59b56392020-01-27 13:07:26 -0500982 const struct cred *cred, bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400983{
Trond Myklebustaa95edf2021-07-02 16:37:15 -0400984 u32 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
985 u32 newseq = be32_to_cpu(new->seqid);
Trond Myklebust2a59a042016-09-03 11:20:04 -0400986
987 if (!pnfs_layout_is_valid(lo)) {
Trond Myklebust59b56392020-01-27 13:07:26 -0500988 pnfs_set_layout_cred(lo, cred);
Trond Myklebust2a59a042016-09-03 11:20:04 -0400989 nfs4_stateid_copy(&lo->plh_stateid, new);
990 lo->plh_barrier = newseq;
991 pnfs_clear_layoutreturn_info(lo);
992 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
993 return;
994 }
Trond Myklebustaa95edf2021-07-02 16:37:15 -0400995
996 if (pnfs_seqid_is_newer(newseq, oldseq))
Trond Myklebustf597c532012-03-04 18:13:56 -0500997 nfs4_stateid_copy(&lo->plh_stateid, new);
Trond Myklebustaa95edf2021-07-02 16:37:15 -0400998
999 if (update_barrier) {
1000 pnfs_barrier_update(lo, newseq);
Trond Myklebustecebb802016-07-24 11:46:06 -04001001 return;
Trond Myklebustaa95edf2021-07-02 16:37:15 -04001002 }
1003 /*
1004 * Because of wraparound, we want to keep the barrier
1005 * "close" to the current seqids. We really only want to
1006 * get here from a layoutget call.
1007 */
1008 if (atomic_read(&lo->plh_outstanding) == 1)
1009 pnfs_barrier_update(lo, be32_to_cpu(lo->plh_stateid.seqid));
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001010}
1011
Trond Myklebust19c54ab2012-10-05 16:56:58 -07001012static bool
1013pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
1014 const nfs4_stateid *stateid)
1015{
1016 u32 seqid = be32_to_cpu(stateid->seqid);
1017
Trond Myklebustd6236a92021-07-23 08:57:21 -04001018 return lo->plh_barrier && pnfs_seqid_is_newer(lo->plh_barrier, seqid);
Trond Myklebust19c54ab2012-10-05 16:56:58 -07001019}
1020
Fred Isamancf7d63f2011-01-06 11:36:25 +00001021/* lget is set to 1 if called from inside send_layoutget call chain */
1022static bool
Trond Myklebuste1c06f82015-08-04 16:40:08 -04001023pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
Fred Isamancf7d63f2011-01-06 11:36:25 +00001024{
Fred Isamanf7e89172011-01-06 11:36:32 +00001025 return lo->plh_block_lgets ||
Trond Myklebuste1c06f82015-08-04 16:40:08 -04001026 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +00001027}
1028
Fred Isaman5e36e2a2016-10-06 12:08:51 -04001029static struct nfs_server *
1030pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
1031{
1032 struct nfs_server *server;
1033
Fred Isaman78746a32016-09-22 12:30:20 -04001034 if (inode) {
Fred Isaman5e36e2a2016-10-06 12:08:51 -04001035 server = NFS_SERVER(inode);
Fred Isaman78746a32016-09-22 12:30:20 -04001036 } else {
Fred Isaman5e36e2a2016-10-06 12:08:51 -04001037 struct dentry *parent_dir = dget_parent(ctx->dentry);
1038 server = NFS_SERVER(parent_dir->d_inode);
1039 dput(parent_dir);
1040 }
1041 return server;
1042}
1043
Trond Myklebust29a8bfe2018-05-30 17:16:20 -04001044static void nfs4_free_pages(struct page **pages, size_t size)
1045{
1046 int i;
1047
1048 if (!pages)
1049 return;
1050
1051 for (i = 0; i < size; i++) {
1052 if (!pages[i])
1053 break;
1054 __free_page(pages[i]);
1055 }
1056 kfree(pages);
1057}
1058
1059static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1060{
1061 struct page **pages;
1062 int i;
1063
Trond Myklebusta2791d32018-09-03 12:57:52 -04001064 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
Trond Myklebust29a8bfe2018-05-30 17:16:20 -04001065 if (!pages) {
1066 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1067 return NULL;
1068 }
1069
1070 for (i = 0; i < size; i++) {
1071 pages[i] = alloc_page(gfp_flags);
1072 if (!pages[i]) {
1073 dprintk("%s: failed to allocate page\n", __func__);
Trond Myklebusta2791d32018-09-03 12:57:52 -04001074 nfs4_free_pages(pages, i);
Trond Myklebust29a8bfe2018-05-30 17:16:20 -04001075 return NULL;
1076 }
1077 }
1078
1079 return pages;
1080}
1081
Fred Isaman587f03d2016-09-21 05:14:28 -04001082static struct nfs4_layoutget *
Fred Isaman5e36e2a2016-10-06 12:08:51 -04001083pnfs_alloc_init_layoutget_args(struct inode *ino,
Benny Halevye5e94012010-10-20 00:18:01 -04001084 struct nfs_open_context *ctx,
Fred Isaman2409a972016-10-06 12:11:21 -04001085 const nfs4_stateid *stateid,
Trond Myklebuste144e532016-01-04 12:52:53 -05001086 const struct pnfs_layout_range *range,
Fred Isaman587f03d2016-09-21 05:14:28 -04001087 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -04001088{
Fred Isaman5e36e2a2016-10-06 12:08:51 -04001089 struct nfs_server *server = pnfs_find_server(ino, ctx);
Trond Myklebust28ced9a2018-09-03 13:12:15 -04001090 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
Fred Isamandacb4522016-09-19 17:47:09 -04001091 size_t max_pages = max_response_pages(server);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001092 struct nfs4_layoutget *lgp;
Benny Halevye5e94012010-10-20 00:18:01 -04001093
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001094 dprintk("--> %s\n", __func__);
1095
Jeff Layton83026d802016-05-17 12:28:46 -04001096 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1097 if (lgp == NULL)
Fred Isaman587f03d2016-09-21 05:14:28 -04001098 return NULL;
Jeff Layton4f2e9dc2015-11-25 13:43:14 -05001099
Trond Myklebust28ced9a2018-09-03 13:12:15 -04001100 if (max_reply_sz) {
1101 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1102 if (npages < max_pages)
1103 max_pages = npages;
1104 }
1105
Fred Isamandacb4522016-09-19 17:47:09 -04001106 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1107 if (!lgp->args.layout.pages) {
1108 kfree(lgp);
1109 return NULL;
1110 }
1111 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1112 lgp->res.layoutp = &lgp->args.layout;
1113
Trond Myklebustd49e0d52017-02-01 20:42:44 -05001114 /* Don't confuse uninitialised result and success */
1115 lgp->res.status = -NFS4ERR_DELAY;
Jeff Layton4f2e9dc2015-11-25 13:43:14 -05001116
Jeff Layton83026d802016-05-17 12:28:46 -04001117 lgp->args.minlength = PAGE_SIZE;
1118 if (lgp->args.minlength > range->length)
1119 lgp->args.minlength = range->length;
Fred Isaman5e36e2a2016-10-06 12:08:51 -04001120 if (ino) {
1121 loff_t i_size = i_size_read(ino);
1122
1123 if (range->iomode == IOMODE_READ) {
1124 if (range->offset >= i_size)
1125 lgp->args.minlength = 0;
1126 else if (i_size - range->offset < lgp->args.minlength)
1127 lgp->args.minlength = i_size - range->offset;
1128 }
Jeff Laytond03ab292016-05-17 12:28:45 -04001129 }
Jeff Layton83026d802016-05-17 12:28:46 -04001130 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1131 pnfs_copy_range(&lgp->args.range, range);
1132 lgp->args.type = server->pnfs_curr_ld->id;
1133 lgp->args.inode = ino;
1134 lgp->args.ctx = get_nfs_open_context(ctx);
Jeff Layton183d9e72016-05-17 12:28:47 -04001135 nfs4_stateid_copy(&lgp->args.stateid, stateid);
Jeff Layton83026d802016-05-17 12:28:46 -04001136 lgp->gfp_flags = gfp_flags;
Trond Myklebust63ec2b62020-02-07 19:40:14 -05001137 lgp->cred = ctx->cred;
Fred Isaman587f03d2016-09-21 05:14:28 -04001138 return lgp;
Andy Adamson974cec82010-10-20 00:18:02 -04001139}
1140
Trond Myklebust29a8bfe2018-05-30 17:16:20 -04001141void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1142{
1143 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1144
1145 nfs4_free_pages(lgp->args.layout.pages, max_pages);
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04001146 pnfs_put_layout_hdr(lgp->lo);
Trond Myklebust29a8bfe2018-05-30 17:16:20 -04001147 put_nfs_open_context(lgp->args.ctx);
1148 kfree(lgp);
1149}
1150
Trond Myklebust24956802013-03-20 13:03:00 -04001151static void pnfs_clear_layoutcommit(struct inode *inode,
1152 struct list_head *head)
1153{
1154 struct nfs_inode *nfsi = NFS_I(inode);
1155 struct pnfs_layout_segment *lseg, *tmp;
1156
1157 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1158 return;
1159 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1160 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1161 continue;
1162 pnfs_lseg_dec_and_remove_zero(lseg, head);
1163 }
1164}
1165
Trond Myklebust68f74472016-10-12 19:50:54 -04001166void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
Trond Myklebust2a974422016-11-20 13:13:54 -05001167 const nfs4_stateid *arg_stateid,
Trond Myklebust68f74472016-10-12 19:50:54 -04001168 const struct pnfs_layout_range *range,
Trond Myklebust68f74472016-10-12 19:50:54 -04001169 const nfs4_stateid *stateid)
1170{
1171 struct inode *inode = lo->plh_inode;
1172 LIST_HEAD(freeme);
1173
1174 spin_lock(&inode->i_lock);
Trond Myklebustc18d1e12021-01-04 15:01:18 -05001175 if (!pnfs_layout_is_valid(lo) ||
Trond Myklebust2a974422016-11-20 13:13:54 -05001176 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1177 goto out_unlock;
Trond Myklebust68f74472016-10-12 19:50:54 -04001178 if (stateid) {
Trond Myklebust2a974422016-11-20 13:13:54 -05001179 u32 seq = be32_to_cpu(arg_stateid->seqid);
1180
Trond Myklebust68f74472016-10-12 19:50:54 -04001181 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1182 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
Trond Myklebust59b56392020-01-27 13:07:26 -05001183 pnfs_set_layout_stateid(lo, stateid, NULL, true);
Trond Myklebust68f74472016-10-12 19:50:54 -04001184 } else
1185 pnfs_mark_layout_stateid_invalid(lo, &freeme);
Trond Myklebust2a974422016-11-20 13:13:54 -05001186out_unlock:
Trond Myklebust68f74472016-10-12 19:50:54 -04001187 pnfs_clear_layoutreturn_waitbit(lo);
1188 spin_unlock(&inode->i_lock);
1189 pnfs_free_lseg_list(&freeme);
1190
1191}
1192
Trond Myklebust13c13a62016-01-26 23:12:11 -05001193static bool
Trond Myklebuste5fd1902016-07-21 12:44:15 -04001194pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1195 nfs4_stateid *stateid,
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001196 const struct cred **cred,
Trond Myklebuste5fd1902016-07-21 12:44:15 -04001197 enum pnfs_iomode *iomode)
Trond Myklebust13c13a62016-01-26 23:12:11 -05001198{
Trond Myklebustbf0291d2016-09-03 10:39:51 -04001199 /* Serialise LAYOUTGET/LAYOUTRETURN */
1200 if (atomic_read(&lo->plh_outstanding) != 0)
1201 return false;
Trond Myklebust6604b202016-10-17 17:54:32 -04001202 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
Trond Myklebust13c13a62016-01-26 23:12:11 -05001203 return false;
Trond Myklebust6604b202016-10-17 17:54:32 -04001204 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
Trond Myklebust13c13a62016-01-26 23:12:11 -05001205 pnfs_get_layout_hdr(lo);
Trond Myklebust1bcf34f2021-01-21 13:51:50 -05001206 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1207 *cred = get_cred(lo->plh_lc_cred);
Trond Myklebuste5fd1902016-07-21 12:44:15 -04001208 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001209 if (lo->plh_return_seq != 0)
1210 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
Trond Myklebuste5fd1902016-07-21 12:44:15 -04001211 if (iomode != NULL)
1212 *iomode = lo->plh_return_iomode;
1213 pnfs_clear_layoutreturn_info(lo);
Trond Myklebust1bcf34f2021-01-21 13:51:50 -05001214 } else if (iomode != NULL)
Trond Myklebuste5fd1902016-07-21 12:44:15 -04001215 *iomode = IOMODE_ANY;
Trond Myklebust1bcf34f2021-01-21 13:51:50 -05001216 pnfs_barrier_update(lo, be32_to_cpu(stateid->seqid));
Trond Myklebust13c13a62016-01-26 23:12:11 -05001217 return true;
1218}
1219
Trond Myklebust828ed9e2016-11-15 21:47:27 -05001220static void
1221pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1222 struct pnfs_layout_hdr *lo,
1223 const nfs4_stateid *stateid,
1224 enum pnfs_iomode iomode)
1225{
1226 struct inode *inode = lo->plh_inode;
1227
1228 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1229 args->inode = inode;
1230 args->range.iomode = iomode;
1231 args->range.offset = 0;
1232 args->range.length = NFS4_MAX_UINT64;
1233 args->layout = lo;
1234 nfs4_stateid_copy(&args->stateid, stateid);
1235}
1236
Peng Taof40eb5d2014-09-06 00:53:22 +08001237static int
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001238pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
1239 const nfs4_stateid *stateid,
1240 const struct cred **pcred,
1241 enum pnfs_iomode iomode,
1242 bool sync)
Peng Taof40eb5d2014-09-06 00:53:22 +08001243{
1244 struct inode *ino = lo->plh_inode;
Trond Myklebust287bd3e2016-12-02 16:12:12 -05001245 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
Peng Taof40eb5d2014-09-06 00:53:22 +08001246 struct nfs4_layoutreturn *lrp;
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001247 const struct cred *cred = *pcred;
Peng Taof40eb5d2014-09-06 00:53:22 +08001248 int status = 0;
1249
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001250 *pcred = NULL;
Trond Myklebust63d8a412022-03-21 15:32:09 -04001251 lrp = kzalloc(sizeof(*lrp), nfs_io_gfp_mask());
Peng Taof40eb5d2014-09-06 00:53:22 +08001252 if (unlikely(lrp == NULL)) {
1253 status = -ENOMEM;
1254 spin_lock(&ino->i_lock);
Tom Haynesd67ae822014-12-11 17:02:04 -05001255 pnfs_clear_layoutreturn_waitbit(lo);
Peng Taof40eb5d2014-09-06 00:53:22 +08001256 spin_unlock(&ino->i_lock);
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001257 put_cred(cred);
Peng Taof40eb5d2014-09-06 00:53:22 +08001258 pnfs_put_layout_hdr(lo);
1259 goto out;
1260 }
1261
Trond Myklebust828ed9e2016-11-15 21:47:27 -05001262 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
Trond Myklebust4d796d72016-09-23 11:38:08 -04001263 lrp->args.ld_private = &lrp->ld_private;
Peng Taof40eb5d2014-09-06 00:53:22 +08001264 lrp->clp = NFS_SERVER(ino)->nfs_client;
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001265 lrp->cred = cred;
Trond Myklebust287bd3e2016-12-02 16:12:12 -05001266 if (ld->prepare_layoutreturn)
1267 ld->prepare_layoutreturn(&lrp->args);
Peng Taof40eb5d2014-09-06 00:53:22 +08001268
Peng Tao6c166052014-11-17 09:30:40 +08001269 status = nfs4_proc_layoutreturn(lrp, sync);
Peng Taof40eb5d2014-09-06 00:53:22 +08001270out:
1271 dprintk("<-- %s status: %d\n", __func__, status);
1272 return status;
1273}
1274
Trond Myklebustd474f962020-08-05 09:03:56 -04001275static bool
1276pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1277 enum pnfs_iomode iomode,
1278 u32 seq)
1279{
1280 struct pnfs_layout_range recall_range = {
1281 .length = NFS4_MAX_UINT64,
1282 .iomode = iomode,
1283 };
1284 return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1285 &recall_range, seq) != -EBUSY;
1286}
1287
Trond Myklebust13c13a62016-01-26 23:12:11 -05001288/* Return true if layoutreturn is needed */
1289static bool
1290pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1291{
Trond Myklebust2370abd2016-01-27 20:32:50 -05001292 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
Trond Myklebust13c13a62016-01-26 23:12:11 -05001293 return false;
Trond Myklebustd474f962020-08-05 09:03:56 -04001294 return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1295 lo->plh_return_seq);
Trond Myklebust13c13a62016-01-26 23:12:11 -05001296}
1297
1298static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1299{
1300 struct inode *inode= lo->plh_inode;
1301
Trond Myklebust2370abd2016-01-27 20:32:50 -05001302 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
Trond Myklebust13c13a62016-01-26 23:12:11 -05001303 return;
1304 spin_lock(&inode->i_lock);
1305 if (pnfs_layout_need_return(lo)) {
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001306 const struct cred *cred;
Trond Myklebust13c13a62016-01-26 23:12:11 -05001307 nfs4_stateid stateid;
1308 enum pnfs_iomode iomode;
1309 bool send;
1310
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001311 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
Trond Myklebust13c13a62016-01-26 23:12:11 -05001312 spin_unlock(&inode->i_lock);
1313 if (send) {
1314 /* Send an async layoutreturn so we dont deadlock */
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001315 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
Trond Myklebust13c13a62016-01-26 23:12:11 -05001316 }
1317 } else
1318 spin_unlock(&inode->i_lock);
1319}
1320
Andy Adamson293b3b02012-06-20 15:03:34 -04001321/*
1322 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1323 * when the layout segment list is empty.
1324 *
1325 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1326 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1327 * deviceid is marked invalid.
1328 */
Benny Halevycbe82602011-05-22 19:52:37 +03001329int
1330_pnfs_return_layout(struct inode *ino)
1331{
1332 struct pnfs_layout_hdr *lo = NULL;
1333 struct nfs_inode *nfsi = NFS_I(ino);
Anna Schumakera421d212021-05-19 12:54:51 -04001334 struct pnfs_layout_range range = {
1335 .iomode = IOMODE_ANY,
1336 .offset = 0,
1337 .length = NFS4_MAX_UINT64,
1338 };
Benny Halevycbe82602011-05-22 19:52:37 +03001339 LIST_HEAD(tmp_list);
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001340 const struct cred *cred;
Benny Halevycbe82602011-05-22 19:52:37 +03001341 nfs4_stateid stateid;
Trond Myklebust24408f52016-11-30 10:19:09 -05001342 int status = 0;
Olga Kornievskaia93b7f7a2018-06-11 15:32:06 -04001343 bool send, valid_layout;
Benny Halevycbe82602011-05-22 19:52:37 +03001344
Andy Adamson366d5052012-06-20 15:03:33 -04001345 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
Benny Halevycbe82602011-05-22 19:52:37 +03001346
1347 spin_lock(&ino->i_lock);
1348 lo = nfsi->layout;
Trond Myklebuste5929f32012-09-21 16:37:02 -04001349 if (!lo) {
Benny Halevycbe82602011-05-22 19:52:37 +03001350 spin_unlock(&ino->i_lock);
Andy Adamson293b3b02012-06-20 15:03:34 -04001351 dprintk("NFS: %s no layout to return\n", __func__);
1352 goto out;
Benny Halevycbe82602011-05-22 19:52:37 +03001353 }
Benny Halevycbe82602011-05-22 19:52:37 +03001354 /* Reference matched in nfs4_layoutreturn_release */
Trond Myklebust70c3bd22012-09-18 20:51:13 -04001355 pnfs_get_layout_hdr(lo);
Trond Myklebust24408f52016-11-30 10:19:09 -05001356 /* Is there an outstanding layoutreturn ? */
1357 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1358 spin_unlock(&ino->i_lock);
1359 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1360 TASK_UNINTERRUPTIBLE))
1361 goto out_put_layout_hdr;
1362 spin_lock(&ino->i_lock);
1363 }
Olga Kornievskaia93b7f7a2018-06-11 15:32:06 -04001364 valid_layout = pnfs_layout_is_valid(lo);
Trond Myklebust24956802013-03-20 13:03:00 -04001365 pnfs_clear_layoutcommit(ino, &tmp_list);
Anna Schumakera421d212021-05-19 12:54:51 -04001366 pnfs_mark_matching_lsegs_return(lo, &tmp_list, &range, 0);
Christoph Hellwigc88953d2014-09-10 08:23:31 -07001367
Anna Schumakera421d212021-05-19 12:54:51 -04001368 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range)
Christoph Hellwigc88953d2014-09-10 08:23:31 -07001369 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
Christoph Hellwigc88953d2014-09-10 08:23:31 -07001370
Andy Adamson293b3b02012-06-20 15:03:34 -04001371 /* Don't send a LAYOUTRETURN if list was initially empty */
Olga Kornievskaia93b7f7a2018-06-11 15:32:06 -04001372 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1373 !valid_layout) {
Andy Adamson293b3b02012-06-20 15:03:34 -04001374 spin_unlock(&ino->i_lock);
Andy Adamson293b3b02012-06-20 15:03:34 -04001375 dprintk("NFS: %s no layout segments to return\n", __func__);
Trond Myklebust7bcc1052020-04-19 14:38:00 -04001376 goto out_wait_layoutreturn;
Andy Adamson293b3b02012-06-20 15:03:34 -04001377 }
Christoph Hellwig47abade2014-08-21 11:09:22 -05001378
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001379 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
Benny Halevycbe82602011-05-22 19:52:37 +03001380 spin_unlock(&ino->i_lock);
Trond Myklebust7f273922015-07-09 18:40:01 +02001381 if (send)
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001382 status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
Trond Myklebust7bcc1052020-04-19 14:38:00 -04001383out_wait_layoutreturn:
1384 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE);
Trond Myklebust7f273922015-07-09 18:40:01 +02001385out_put_layout_hdr:
Trond Myklebustee6625a2017-01-26 15:50:41 -05001386 pnfs_free_lseg_list(&tmp_list);
Trond Myklebust7f273922015-07-09 18:40:01 +02001387 pnfs_put_layout_hdr(lo);
Benny Halevycbe82602011-05-22 19:52:37 +03001388out:
1389 dprintk("<-- %s status: %d\n", __func__, status);
1390 return status;
1391}
1392
Trond Myklebust24028672013-03-20 13:23:33 -04001393int
1394pnfs_commit_and_return_layout(struct inode *inode)
1395{
1396 struct pnfs_layout_hdr *lo;
1397 int ret;
1398
1399 spin_lock(&inode->i_lock);
1400 lo = NFS_I(inode)->layout;
1401 if (lo == NULL) {
1402 spin_unlock(&inode->i_lock);
1403 return 0;
1404 }
1405 pnfs_get_layout_hdr(lo);
1406 /* Block new layoutgets and read/write to ds */
1407 lo->plh_block_lgets++;
1408 spin_unlock(&inode->i_lock);
1409 filemap_fdatawait(inode->i_mapping);
1410 ret = pnfs_layoutcommit_inode(inode, true);
1411 if (ret == 0)
1412 ret = _pnfs_return_layout(inode);
1413 spin_lock(&inode->i_lock);
1414 lo->plh_block_lgets--;
1415 spin_unlock(&inode->i_lock);
1416 pnfs_put_layout_hdr(lo);
1417 return ret;
1418}
1419
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001420bool pnfs_roc(struct inode *ino,
1421 struct nfs4_layoutreturn_args *args,
1422 struct nfs4_layoutreturn_res *res,
NeilBrowna52458b2018-12-03 11:30:31 +11001423 const struct cred *cred)
Fred Isamanf7e89172011-01-06 11:36:32 +00001424{
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001425 struct nfs_inode *nfsi = NFS_I(ino);
1426 struct nfs_open_context *ctx;
1427 struct nfs4_state *state;
Fred Isamanf7e89172011-01-06 11:36:32 +00001428 struct pnfs_layout_hdr *lo;
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001429 struct pnfs_layout_segment *lseg, *next;
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001430 const struct cred *lc_cred;
Peng Tao193e3aa2014-11-17 09:30:41 +08001431 nfs4_stateid stateid;
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001432 enum pnfs_iomode iomode = 0;
1433 bool layoutreturn = false, roc = false;
Trond Myklebuste71708d2016-11-21 10:56:38 -05001434 bool skip_read = false;
Fred Isamanf7e89172011-01-06 11:36:32 +00001435
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001436 if (!nfs_have_layout(ino))
1437 return false;
Trond Myklebust29ade5d2016-11-30 16:23:38 -05001438retry:
Trond Myklebust0de43972018-09-02 15:57:01 -04001439 rcu_read_lock();
Fred Isamanf7e89172011-01-06 11:36:32 +00001440 spin_lock(&ino->i_lock);
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001441 lo = nfsi->layout;
Trond Myklebust0cdc3292016-11-21 11:05:33 -05001442 if (!lo || !pnfs_layout_is_valid(lo) ||
Trond Myklebust9c6376e2018-03-07 14:49:06 -05001443 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1444 lo = NULL;
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001445 goto out_noroc;
Trond Myklebust9c6376e2018-03-07 14:49:06 -05001446 }
1447 pnfs_get_layout_hdr(lo);
Trond Myklebust29ade5d2016-11-30 16:23:38 -05001448 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
Trond Myklebust29ade5d2016-11-30 16:23:38 -05001449 spin_unlock(&ino->i_lock);
Trond Myklebust0de43972018-09-02 15:57:01 -04001450 rcu_read_unlock();
Trond Myklebust29ade5d2016-11-30 16:23:38 -05001451 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1452 TASK_UNINTERRUPTIBLE);
1453 pnfs_put_layout_hdr(lo);
1454 goto retry;
1455 }
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001456
Peng Taoe755d632015-08-19 13:49:19 +08001457 /* no roc if we hold a delegation */
Trond Myklebuste71708d2016-11-21 10:56:38 -05001458 if (nfs4_check_delegation(ino, FMODE_READ)) {
1459 if (nfs4_check_delegation(ino, FMODE_WRITE))
1460 goto out_noroc;
1461 skip_read = true;
1462 }
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001463
Trond Myklebust0de43972018-09-02 15:57:01 -04001464 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001465 state = ctx->state;
Trond Myklebuste71708d2016-11-21 10:56:38 -05001466 if (state == NULL)
1467 continue;
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001468 /* Don't return layout if there is open file state */
Trond Myklebuste71708d2016-11-21 10:56:38 -05001469 if (state->state & FMODE_WRITE)
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001470 goto out_noroc;
Trond Myklebuste71708d2016-11-21 10:56:38 -05001471 if (state->state & FMODE_READ)
1472 skip_read = true;
Trond Myklebust40dd4b72015-01-24 13:54:37 -05001473 }
1474
Peng Taoe755d632015-08-19 13:49:19 +08001475
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001476 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
Trond Myklebuste71708d2016-11-21 10:56:38 -05001477 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1478 continue;
Peng Taoe755d632015-08-19 13:49:19 +08001479 /* If we are sending layoutreturn, invalidate all valid lsegs */
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001480 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1481 continue;
1482 /*
1483 * Note: mark lseg for return so pnfs_layout_remove_lseg
1484 * doesn't invalidate the layout for us.
1485 */
1486 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1487 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1488 continue;
1489 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
Trond Myklebust69820d22016-11-15 18:29:59 -05001490 }
1491
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001492 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1493 goto out_noroc;
Trond Myklebust69820d22016-11-15 18:29:59 -05001494
Peng Tao500d7012015-09-22 11:35:22 +08001495 /* ROC in two conditions:
Peng Taoe755d632015-08-19 13:49:19 +08001496 * 1. there are ROC lsegs
1497 * 2. we don't send layoutreturn
Peng Taoe755d632015-08-19 13:49:19 +08001498 */
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001499 /* lo ref dropped in pnfs_roc_release() */
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001500 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001501 /* If the creds don't match, we can't compound the layoutreturn */
Trond Myklebust4d8948c2020-04-19 23:53:52 -04001502 if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0)
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001503 goto out_noroc;
1504
1505 roc = layoutreturn;
1506 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1507 res->lrs_present = 0;
1508 layoutreturn = false;
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001509 put_cred(lc_cred);
Trond Myklebust4d8948c2020-04-19 23:53:52 -04001510
Peng Taoe755d632015-08-19 13:49:19 +08001511out_noroc:
Fred Isamanf7e89172011-01-06 11:36:32 +00001512 spin_unlock(&ino->i_lock);
Trond Myklebust0de43972018-09-02 15:57:01 -04001513 rcu_read_unlock();
Trond Myklebust71401712015-03-25 12:36:13 -04001514 pnfs_layoutcommit_inode(ino, true);
Trond Myklebust287bd3e2016-12-02 16:12:12 -05001515 if (roc) {
1516 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1517 if (ld->prepare_layoutreturn)
1518 ld->prepare_layoutreturn(args);
Trond Myklebust9c6376e2018-03-07 14:49:06 -05001519 pnfs_put_layout_hdr(lo);
Trond Myklebust287bd3e2016-12-02 16:12:12 -05001520 return true;
1521 }
Peng Taoe755d632015-08-19 13:49:19 +08001522 if (layoutreturn)
Trond Myklebust44ea8df2020-04-02 15:37:02 -04001523 pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
Trond Myklebust9c6376e2018-03-07 14:49:06 -05001524 pnfs_put_layout_hdr(lo);
Trond Myklebust287bd3e2016-12-02 16:12:12 -05001525 return false;
Fred Isamanf7e89172011-01-06 11:36:32 +00001526}
1527
Trond Myklebust078000d2021-01-04 13:18:03 -05001528int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
1529 struct nfs4_layoutreturn_res **respp, int *ret)
Trond Myklebust287a9c52019-09-20 07:23:41 -04001530{
1531 struct nfs4_layoutreturn_args *arg = *argpp;
1532 int retval = -EAGAIN;
1533
1534 if (!arg)
1535 return 0;
1536 /* Handle Layoutreturn errors */
1537 switch (*ret) {
1538 case 0:
1539 retval = 0;
1540 break;
Trond Myklebust6109bcf2019-09-20 07:23:43 -04001541 case -NFS4ERR_NOMATCHING_LAYOUT:
1542 /* Was there an RPC level error? If not, retry */
1543 if (task->tk_rpc_status == 0)
1544 break;
1545 /* If the call was not sent, let caller handle it */
1546 if (!RPC_WAS_SENT(task))
1547 return 0;
1548 /*
1549 * Otherwise, assume the call succeeded and
1550 * that we need to release the layout
1551 */
1552 *ret = 0;
1553 (*respp)->lrs_present = 0;
1554 retval = 0;
1555 break;
Trond Myklebust078a4322019-09-20 07:23:42 -04001556 case -NFS4ERR_DELAY:
1557 /* Let the caller handle the retry */
1558 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1559 return 0;
Trond Myklebust287a9c52019-09-20 07:23:41 -04001560 case -NFS4ERR_OLD_STATEID:
Trond Myklebust30cb3ee2019-09-20 07:23:45 -04001561 if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
Trond Myklebust078000d2021-01-04 13:18:03 -05001562 &arg->range, arg->inode))
Trond Myklebust287a9c52019-09-20 07:23:41 -04001563 break;
1564 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1565 return -EAGAIN;
1566 }
1567 *argpp = NULL;
1568 *respp = NULL;
1569 return retval;
1570}
1571
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001572void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1573 struct nfs4_layoutreturn_res *res,
1574 int ret)
Fred Isamanf7e89172011-01-06 11:36:32 +00001575{
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001576 struct pnfs_layout_hdr *lo = args->layout;
Trond Myklebust67bbcee2021-01-04 13:35:46 -05001577 struct inode *inode = args->inode;
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001578 const nfs4_stateid *res_stateid = NULL;
Trond Myklebust287bd3e2016-12-02 16:12:12 -05001579 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
Fred Isamanf7e89172011-01-06 11:36:32 +00001580
Trond Myklebust9c47b182019-09-20 07:23:40 -04001581 switch (ret) {
1582 case -NFS4ERR_NOMATCHING_LAYOUT:
Trond Myklebust67bbcee2021-01-04 13:35:46 -05001583 spin_lock(&inode->i_lock);
1584 if (pnfs_layout_is_valid(lo) &&
1585 nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
1586 pnfs_set_plh_return_info(lo, args->range.iomode, 0);
Trond Myklebustc18d1e12021-01-04 15:01:18 -05001587 pnfs_clear_layoutreturn_waitbit(lo);
Trond Myklebust67bbcee2021-01-04 13:35:46 -05001588 spin_unlock(&inode->i_lock);
Trond Myklebust9c47b182019-09-20 07:23:40 -04001589 break;
1590 case 0:
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001591 if (res->lrs_present)
1592 res_stateid = &res->stateid;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001593 fallthrough;
Trond Myklebust9c47b182019-09-20 07:23:40 -04001594 default:
Trond Myklebustc18d1e12021-01-04 15:01:18 -05001595 pnfs_layoutreturn_free_lsegs(lo, &args->stateid, &args->range,
1596 res_stateid);
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001597 }
Trond Myklebusta19b4782020-08-03 18:32:28 -04001598 trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
Trond Myklebust287bd3e2016-12-02 16:12:12 -05001599 if (ld_private && ld_private->ops && ld_private->ops->free)
1600 ld_private->ops->free(ld_private);
Trond Myklebust1c5bd76d2016-11-16 01:11:25 -05001601 pnfs_put_layout_hdr(lo);
Fred Isamanf7e89172011-01-06 11:36:32 +00001602}
1603
Peng Tao500d7012015-09-22 11:35:22 +08001604bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1605{
1606 struct nfs_inode *nfsi = NFS_I(ino);
1607 struct pnfs_layout_hdr *lo;
1608 bool sleep = false;
1609
1610 /* we might not have grabbed lo reference. so need to check under
1611 * i_lock */
1612 spin_lock(&ino->i_lock);
1613 lo = nfsi->layout;
Trond Myklebustee284e32016-11-18 15:21:30 -05001614 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
Peng Tao500d7012015-09-22 11:35:22 +08001615 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
Trond Myklebustee284e32016-11-18 15:21:30 -05001616 sleep = true;
1617 }
1618 spin_unlock(&ino->i_lock);
Peng Tao500d7012015-09-22 11:35:22 +08001619 return sleep;
1620}
1621
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001622/*
1623 * Compare two layout segments for sorting into layout cache.
1624 * We want to preferentially return RW over RO layouts, so ensure those
1625 * are seen first.
1626 */
1627static s64
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001628pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
Trond Myklebust3cb2df12013-06-03 11:24:36 -04001629 const struct pnfs_layout_range *l2)
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001630{
Benny Halevyfb3296e2011-05-22 19:47:26 +03001631 s64 d;
1632
1633 /* high offset > low offset */
1634 d = l1->offset - l2->offset;
1635 if (d)
1636 return d;
1637
1638 /* short length > long length */
1639 d = l2->length - l1->length;
1640 if (d)
1641 return d;
1642
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001643 /* read > read/write */
Benny Halevyfb3296e2011-05-22 19:47:26 +03001644 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001645}
1646
Trond Myklebust03772d22015-08-25 08:54:17 -04001647static bool
1648pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1649 const struct pnfs_layout_range *l2)
Andy Adamson974cec82010-10-20 00:18:02 -04001650{
Trond Myklebust03772d22015-08-25 08:54:17 -04001651 return pnfs_lseg_range_cmp(l1, l2) > 0;
1652}
1653
1654static bool
1655pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1656 struct pnfs_layout_segment *old)
1657{
1658 return false;
1659}
1660
1661void
1662pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1663 struct pnfs_layout_segment *lseg,
1664 bool (*is_after)(const struct pnfs_layout_range *,
1665 const struct pnfs_layout_range *),
1666 bool (*do_merge)(struct pnfs_layout_segment *,
1667 struct pnfs_layout_segment *),
1668 struct list_head *free_me)
1669{
1670 struct pnfs_layout_segment *lp, *tmp;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001671
Andy Adamson974cec82010-10-20 00:18:02 -04001672 dprintk("%s:Begin\n", __func__);
1673
Trond Myklebust03772d22015-08-25 08:54:17 -04001674 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1675 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1676 continue;
1677 if (do_merge(lseg, lp)) {
1678 mark_lseg_invalid(lp, free_me);
1679 continue;
1680 }
1681 if (is_after(&lseg->pls_range, &lp->pls_range))
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001682 continue;
Fred Isaman566052c2011-01-06 11:36:20 +00001683 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001684 dprintk("%s: inserted lseg %p "
1685 "iomode %d offset %llu length %llu before "
1686 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +00001687 __func__, lseg, lseg->pls_range.iomode,
1688 lseg->pls_range.offset, lseg->pls_range.length,
1689 lp, lp->pls_range.iomode, lp->pls_range.offset,
1690 lp->pls_range.length);
Benny Halevyfb3296e2011-05-22 19:47:26 +03001691 goto out;
Andy Adamson974cec82010-10-20 00:18:02 -04001692 }
Benny Halevyfb3296e2011-05-22 19:47:26 +03001693 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1694 dprintk("%s: inserted lseg %p "
1695 "iomode %d offset %llu length %llu at tail\n",
1696 __func__, lseg, lseg->pls_range.iomode,
1697 lseg->pls_range.offset, lseg->pls_range.length);
1698out:
Trond Myklebust70c3bd22012-09-18 20:51:13 -04001699 pnfs_get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -04001700
1701 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -04001702}
Trond Myklebust03772d22015-08-25 08:54:17 -04001703EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1704
1705static void
1706pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1707 struct pnfs_layout_segment *lseg,
1708 struct list_head *free_me)
1709{
1710 struct inode *inode = lo->plh_inode;
1711 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1712
1713 if (ld->add_lseg != NULL)
1714 ld->add_lseg(lo, lseg, free_me);
1715 else
1716 pnfs_generic_layout_insert_lseg(lo, lseg,
1717 pnfs_lseg_range_is_after,
1718 pnfs_lseg_no_merge,
1719 free_me);
1720}
Benny Halevye5e94012010-10-20 00:18:01 -04001721
1722static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -04001723alloc_init_layout_hdr(struct inode *ino,
1724 struct nfs_open_context *ctx,
1725 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -04001726{
1727 struct pnfs_layout_hdr *lo;
1728
Benny Halevy636fb9c2011-05-22 19:51:33 +03001729 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -04001730 if (!lo)
1731 return NULL;
Elena Reshetova2b28a7b2017-10-20 12:53:33 +03001732 refcount_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001733 INIT_LIST_HEAD(&lo->plh_layouts);
1734 INIT_LIST_HEAD(&lo->plh_segs);
Trond Myklebust68f74472016-10-12 19:50:54 -04001735 INIT_LIST_HEAD(&lo->plh_return_segs);
Trond Myklebustfd9a8d72013-02-12 09:48:42 -05001736 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001737 lo->plh_inode = ino;
NeilBrowna52458b2018-12-03 11:30:31 +11001738 lo->plh_lc_cred = get_cred(ctx->cred);
Trond Myklebust67a3b722016-06-17 16:48:19 -04001739 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
Benny Halevye5e94012010-10-20 00:18:01 -04001740 return lo;
1741}
1742
1743static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -04001744pnfs_find_alloc_layout(struct inode *ino,
1745 struct nfs_open_context *ctx,
1746 gfp_t gfp_flags)
Trond Myklebuste5241e42016-06-17 16:48:20 -04001747 __releases(&ino->i_lock)
1748 __acquires(&ino->i_lock)
Benny Halevye5e94012010-10-20 00:18:01 -04001749{
1750 struct nfs_inode *nfsi = NFS_I(ino);
1751 struct pnfs_layout_hdr *new = NULL;
1752
1753 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1754
Trond Myklebust251ec412012-10-02 15:41:05 -07001755 if (nfsi->layout != NULL)
1756 goto out_existing;
Benny Halevye5e94012010-10-20 00:18:01 -04001757 spin_unlock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -04001758 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -04001759 spin_lock(&ino->i_lock);
1760
Trond Myklebust251ec412012-10-02 15:41:05 -07001761 if (likely(nfsi->layout == NULL)) { /* Won the race? */
Benny Halevye5e94012010-10-20 00:18:01 -04001762 nfsi->layout = new;
Trond Myklebust251ec412012-10-02 15:41:05 -07001763 return new;
Yanchuan Nian7175fe92012-10-31 16:05:48 +08001764 } else if (new != NULL)
1765 pnfs_free_layout_hdr(new);
Trond Myklebust251ec412012-10-02 15:41:05 -07001766out_existing:
1767 pnfs_get_layout_hdr(nfsi->layout);
Benny Halevye5e94012010-10-20 00:18:01 -04001768 return nfsi->layout;
1769}
1770
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001771/*
1772 * iomode matching rules:
Tom Haynesc7d73af2016-05-25 07:31:14 -07001773 * iomode lseg strict match
1774 * iomode
1775 * ----- ----- ------ -----
1776 * ANY READ N/A true
1777 * ANY RW N/A true
1778 * RW READ N/A false
1779 * RW RW N/A true
1780 * READ READ N/A true
1781 * READ RW true false
1782 * READ RW false true
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001783 */
Trond Myklebust3cb2df12013-06-03 11:24:36 -04001784static bool
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001785pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
Tom Haynesc7d73af2016-05-25 07:31:14 -07001786 const struct pnfs_layout_range *range,
1787 bool strict_iomode)
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001788{
Benny Halevyfb3296e2011-05-22 19:47:26 +03001789 struct pnfs_layout_range range1;
1790
1791 if ((range->iomode == IOMODE_RW &&
1792 ls_range->iomode != IOMODE_RW) ||
Tom Haynesc7d73af2016-05-25 07:31:14 -07001793 (range->iomode != ls_range->iomode &&
Thomas Meyer6089dd02017-10-07 16:02:21 +02001794 strict_iomode) ||
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001795 !pnfs_lseg_range_intersecting(ls_range, range))
Gustavo A. R. Silva10db5b72018-08-01 19:55:51 -05001796 return false;
Benny Halevyfb3296e2011-05-22 19:47:26 +03001797
1798 /* range1 covers only the first byte in the range */
1799 range1 = *range;
1800 range1.length = 1;
Trond Myklebust7dc0ac72013-06-03 11:30:24 -04001801 return pnfs_lseg_range_contained(ls_range, &range1);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001802}
1803
1804/*
1805 * lookup range in layout
1806 */
Benny Halevye5e94012010-10-20 00:18:01 -04001807static struct pnfs_layout_segment *
Benny Halevyfb3296e2011-05-22 19:47:26 +03001808pnfs_find_lseg(struct pnfs_layout_hdr *lo,
Tom Haynesc7d73af2016-05-25 07:31:14 -07001809 struct pnfs_layout_range *range,
1810 bool strict_iomode)
Benny Halevye5e94012010-10-20 00:18:01 -04001811{
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001812 struct pnfs_layout_segment *lseg, *ret = NULL;
1813
1814 dprintk("%s:Begin\n", __func__);
1815
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001816 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +00001817 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
Tom Haynesc7d73af2016-05-25 07:31:14 -07001818 pnfs_lseg_range_match(&lseg->pls_range, range,
1819 strict_iomode)) {
Trond Myklebust9369a432012-09-18 20:57:08 -04001820 ret = pnfs_get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001821 break;
1822 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001823 }
1824
1825 dprintk("%s:Return lseg %p ref %d\n",
Elena Reshetovaeba6dd62017-10-20 12:53:32 +03001826 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001827 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -04001828}
1829
1830/*
Andy Adamsond23d61c2012-05-23 05:02:37 -04001831 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1832 * to the MDS or over pNFS
1833 *
1834 * The nfs_inode read_io and write_io fields are cumulative counters reset
1835 * when there are no layout segments. Note that in pnfs_update_layout iomode
1836 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1837 * WRITE request.
1838 *
1839 * A return of true means use MDS I/O.
1840 *
1841 * From rfc 5661:
1842 * If a file's size is smaller than the file size threshold, data accesses
1843 * SHOULD be sent to the metadata server. If an I/O request has a length that
1844 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1845 * server. If both file size and I/O size are provided, the client SHOULD
1846 * reach or exceed both thresholds before sending its read or write
1847 * requests to the data server.
1848 */
1849static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1850 struct inode *ino, int iomode)
1851{
1852 struct nfs4_threshold *t = ctx->mdsthreshold;
1853 struct nfs_inode *nfsi = NFS_I(ino);
1854 loff_t fsize = i_size_read(ino);
1855 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1856
1857 if (t == NULL)
1858 return ret;
1859
1860 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1861 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1862
1863 switch (iomode) {
1864 case IOMODE_READ:
1865 if (t->bm & THRESHOLD_RD) {
1866 dprintk("%s fsize %llu\n", __func__, fsize);
1867 size_set = true;
1868 if (fsize < t->rd_sz)
1869 size = true;
1870 }
1871 if (t->bm & THRESHOLD_RD_IO) {
1872 dprintk("%s nfsi->read_io %llu\n", __func__,
1873 nfsi->read_io);
1874 io_set = true;
1875 if (nfsi->read_io < t->rd_io_sz)
1876 io = true;
1877 }
1878 break;
1879 case IOMODE_RW:
1880 if (t->bm & THRESHOLD_WR) {
1881 dprintk("%s fsize %llu\n", __func__, fsize);
1882 size_set = true;
1883 if (fsize < t->wr_sz)
1884 size = true;
1885 }
1886 if (t->bm & THRESHOLD_WR_IO) {
1887 dprintk("%s nfsi->write_io %llu\n", __func__,
1888 nfsi->write_io);
1889 io_set = true;
1890 if (nfsi->write_io < t->wr_io_sz)
1891 io = true;
1892 }
1893 break;
1894 }
1895 if (size_set && io_set) {
1896 if (size && io)
1897 ret = true;
1898 } else if (size || io)
1899 ret = true;
1900
1901 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1902 return ret;
1903}
1904
Trond Myklebustd03360a2018-09-05 14:07:12 -04001905static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
Peng Taoaa8a45e2014-12-01 08:22:23 +08001906{
1907 /*
1908 * send layoutcommit as it can hold up layoutreturn due to lseg
1909 * reference
1910 */
1911 pnfs_layoutcommit_inode(lo->plh_inode, false);
Trond Myklebustd03360a2018-09-05 14:07:12 -04001912 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
Trond Myklebust2e5b29f2015-12-14 16:25:11 -05001913 nfs_wait_bit_killable,
Peter Zijlstraf5d39b02022-08-22 13:18:22 +02001914 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
Peng Taoaa8a45e2014-12-01 08:22:23 +08001915}
1916
Trond Myklebust411ae722018-06-23 22:54:33 -04001917static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1918{
1919 atomic_inc(&lo->plh_outstanding);
1920}
1921
1922static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1923{
Trond Myklebust880265c2022-05-31 11:03:07 -04001924 if (atomic_dec_and_test(&lo->plh_outstanding) &&
1925 test_and_clear_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags))
1926 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN);
Trond Myklebust411ae722018-06-23 22:54:33 -04001927}
1928
Trond Myklebustd29b468d2021-01-22 10:05:51 -05001929static bool pnfs_is_first_layoutget(struct pnfs_layout_hdr *lo)
1930{
1931 return test_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags);
1932}
1933
Tom Haynesd67ae822014-12-11 17:02:04 -05001934static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1935{
1936 unsigned long *bitlock = &lo->plh_flags;
1937
1938 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1939 smp_mb__after_atomic();
1940 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1941}
1942
Fred Isaman78746a32016-09-22 12:30:20 -04001943static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1944 struct nfs_server *server)
1945{
Trond Myklebustcf6605d2020-02-18 17:14:40 -05001946 if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
Fred Isaman78746a32016-09-22 12:30:20 -04001947 struct nfs_client *clp = server->nfs_client;
1948
1949 /* The lo must be on the clp list if there is any
1950 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1951 */
1952 spin_lock(&clp->cl_lock);
Trond Myklebustcf6605d2020-02-18 17:14:40 -05001953 list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
Fred Isaman78746a32016-09-22 12:30:20 -04001954 spin_unlock(&clp->cl_lock);
1955 }
1956}
1957
Andy Adamsond23d61c2012-05-23 05:02:37 -04001958/*
Benny Halevye5e94012010-10-20 00:18:01 -04001959 * Layout segment is retreived from the server if not cached.
1960 * The appropriate layout segment is referenced and returned to the caller.
1961 */
Andy Adamson7c24d942011-06-13 18:22:38 -04001962struct pnfs_layout_segment *
Benny Halevye5e94012010-10-20 00:18:01 -04001963pnfs_update_layout(struct inode *ino,
1964 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +03001965 loff_t pos,
1966 u64 count,
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001967 enum pnfs_iomode iomode,
Tom Haynesc7d73af2016-05-25 07:31:14 -07001968 bool strict_iomode,
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001969 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -04001970{
Benny Halevyfb3296e2011-05-22 19:47:26 +03001971 struct pnfs_layout_range arg = {
1972 .iomode = iomode,
1973 .offset = pos,
1974 .length = count,
1975 };
Trond Myklebust70d2f7b2017-09-11 13:09:37 -04001976 unsigned pg_offset;
Weston Andros Adamson6382a442011-06-01 16:44:44 -04001977 struct nfs_server *server = NFS_SERVER(ino);
1978 struct nfs_client *clp = server->nfs_client;
Jeff Layton183d9e72016-05-17 12:28:47 -04001979 struct pnfs_layout_hdr *lo = NULL;
Benny Halevye5e94012010-10-20 00:18:01 -04001980 struct pnfs_layout_segment *lseg = NULL;
Fred Isaman587f03d2016-09-21 05:14:28 -04001981 struct nfs4_layoutget *lgp;
Jeff Layton183d9e72016-05-17 12:28:47 -04001982 nfs4_stateid stateid;
1983 long timeout = 0;
Trond Myklebust66b53f32016-07-14 14:28:31 -04001984 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
Weston Andros Adamson30005122013-02-28 20:30:10 -05001985 bool first;
Benny Halevye5e94012010-10-20 00:18:01 -04001986
Jeff Layton9a4bf312015-12-10 10:41:58 -05001987 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
Jeff Layton183d9e72016-05-17 12:28:47 -04001988 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
Jeff Layton9a4bf312015-12-10 10:41:58 -05001989 PNFS_UPDATE_LAYOUT_NO_PNFS);
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04001990 goto out;
Jeff Layton9a4bf312015-12-10 10:41:58 -05001991 }
Andy Adamsond23d61c2012-05-23 05:02:37 -04001992
Jeff Layton9a4bf312015-12-10 10:41:58 -05001993 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
Jeff Layton183d9e72016-05-17 12:28:47 -04001994 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
Jeff Layton9a4bf312015-12-10 10:41:58 -05001995 PNFS_UPDATE_LAYOUT_MDSTHRESH);
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04001996 goto out;
Jeff Layton9a4bf312015-12-10 10:41:58 -05001997 }
Andy Adamsond23d61c2012-05-23 05:02:37 -04001998
Peng Tao9bf87482014-08-22 17:37:41 +08001999lookup_again:
Trond Myklebustd03360a2018-09-05 14:07:12 -04002000 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
2001 if (IS_ERR(lseg))
2002 goto out;
Peng Tao9bf87482014-08-22 17:37:41 +08002003 first = false;
Benny Halevye5e94012010-10-20 00:18:01 -04002004 spin_lock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -04002005 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
Trond Myklebust830ffb52012-09-20 21:25:19 -04002006 if (lo == NULL) {
2007 spin_unlock(&ino->i_lock);
Trond Myklebust3764a172022-05-14 10:08:11 -04002008 lseg = ERR_PTR(-ENOMEM);
Jeff Layton183d9e72016-05-17 12:28:47 -04002009 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
Jeff Layton9a4bf312015-12-10 10:41:58 -05002010 PNFS_UPDATE_LAYOUT_NOMEM);
Trond Myklebust830ffb52012-09-20 21:25:19 -04002011 goto out;
2012 }
Benny Halevye5e94012010-10-20 00:18:01 -04002013
Fred Isaman43f1b3d2011-01-06 11:36:30 +00002014 /* Do we even need to bother with this? */
Trond Myklebusta59c30a2012-03-01 11:17:47 -05002015 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
Jeff Layton183d9e72016-05-17 12:28:47 -04002016 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
Jeff Layton9a4bf312015-12-10 10:41:58 -05002017 PNFS_UPDATE_LAYOUT_BULK_RECALL);
Fred Isaman43f1b3d2011-01-06 11:36:30 +00002018 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -04002019 goto out_unlock;
2020 }
2021
2022 /* if LAYOUTGET already failed once we don't try again */
Trond Myklebust2e5b29f2015-12-14 16:25:11 -05002023 if (pnfs_layout_io_test_failed(lo, iomode)) {
Jeff Layton183d9e72016-05-17 12:28:47 -04002024 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
Jeff Layton9a4bf312015-12-10 10:41:58 -05002025 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
Benny Halevye5e94012010-10-20 00:18:01 -04002026 goto out_unlock;
Jeff Layton9a4bf312015-12-10 10:41:58 -05002027 }
Benny Halevye5e94012010-10-20 00:18:01 -04002028
Trond Myklebust411ae722018-06-23 22:54:33 -04002029 /*
2030 * If the layout segment list is empty, but there are outstanding
2031 * layoutget calls, then they might be subject to a layoutrecall.
2032 */
Trond Myklebust880265c2022-05-31 11:03:07 -04002033 if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) &&
Trond Myklebust411ae722018-06-23 22:54:33 -04002034 atomic_read(&lo->plh_outstanding) != 0) {
2035 spin_unlock(&ino->i_lock);
Trond Myklebust880265c2022-05-31 11:03:07 -04002036 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN,
2037 TASK_KILLABLE));
Trond Myklebust58bbeab2019-07-18 15:33:42 -04002038 if (IS_ERR(lseg))
Trond Myklebust411ae722018-06-23 22:54:33 -04002039 goto out_put_layout_hdr;
2040 pnfs_put_layout_hdr(lo);
2041 goto lookup_again;
2042 }
2043
Trond Myklebust2c8d5fc2021-01-05 06:43:45 -05002044 /*
2045 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
2046 * for LAYOUTRETURN.
2047 */
2048 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
2049 spin_unlock(&ino->i_lock);
2050 dprintk("%s wait for layoutreturn\n", __func__);
2051 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2052 if (!IS_ERR(lseg)) {
2053 pnfs_put_layout_hdr(lo);
2054 dprintk("%s retrying\n", __func__);
2055 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2056 lseg,
2057 PNFS_UPDATE_LAYOUT_RETRY);
2058 goto lookup_again;
2059 }
2060 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2061 PNFS_UPDATE_LAYOUT_RETURN);
2062 goto out_put_layout_hdr;
2063 }
2064
Tom Haynesc7d73af2016-05-25 07:31:14 -07002065 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
Jeff Layton183d9e72016-05-17 12:28:47 -04002066 if (lseg) {
2067 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2068 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
2069 goto out_unlock;
2070 }
2071
Jeff Layton183d9e72016-05-17 12:28:47 -04002072 /*
2073 * Choose a stateid for the LAYOUTGET. If we don't have a layout
2074 * stateid, or it has been invalidated, then we must use the open
2075 * stateid.
2076 */
Trond Myklebust67a3b722016-06-17 16:48:19 -04002077 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
Trond Myklebustd9aba2b2019-07-16 15:38:28 -04002078 int status;
Jeff Layton183d9e72016-05-17 12:28:47 -04002079
2080 /*
2081 * The first layoutget for the file. Need to serialize per
Peng Tao9bf87482014-08-22 17:37:41 +08002082 * RFC 5661 Errata 3208.
2083 */
2084 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
2085 &lo->plh_flags)) {
2086 spin_unlock(&ino->i_lock);
Trond Myklebustd03360a2018-09-05 14:07:12 -04002087 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
2088 NFS_LAYOUT_FIRST_LAYOUTGET,
2089 TASK_KILLABLE));
2090 if (IS_ERR(lseg))
2091 goto out_put_layout_hdr;
Peng Tao9bf87482014-08-22 17:37:41 +08002092 pnfs_put_layout_hdr(lo);
Jeff Layton183d9e72016-05-17 12:28:47 -04002093 dprintk("%s retrying\n", __func__);
Peng Tao9bf87482014-08-22 17:37:41 +08002094 goto lookup_again;
2095 }
Jeff Layton183d9e72016-05-17 12:28:47 -04002096
Trond Myklebustfbf4bcc2020-04-13 15:55:21 -04002097 spin_unlock(&ino->i_lock);
Jeff Layton183d9e72016-05-17 12:28:47 -04002098 first = true;
Trond Myklebustd9aba2b2019-07-16 15:38:28 -04002099 status = nfs4_select_rw_stateid(ctx->state,
Trond Myklebust70d2f7b2017-09-11 13:09:37 -04002100 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
Trond Myklebustd9aba2b2019-07-16 15:38:28 -04002101 NULL, &stateid, NULL);
2102 if (status != 0) {
Trond Myklebust731c74d2019-07-22 18:06:17 +01002103 lseg = ERR_PTR(status);
Trond Myklebust70d2f7b2017-09-11 13:09:37 -04002104 trace_pnfs_update_layout(ino, pos, count,
2105 iomode, lo, lseg,
2106 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
Trond Myklebustd9aba2b2019-07-16 15:38:28 -04002107 nfs4_schedule_stateid_recovery(server, ctx->state);
2108 pnfs_clear_first_layoutget(lo);
2109 pnfs_put_layout_hdr(lo);
2110 goto lookup_again;
Trond Myklebust70d2f7b2017-09-11 13:09:37 -04002111 }
Trond Myklebustfbf4bcc2020-04-13 15:55:21 -04002112 spin_lock(&ino->i_lock);
Peng Tao9bf87482014-08-22 17:37:41 +08002113 } else {
Jeff Layton183d9e72016-05-17 12:28:47 -04002114 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
Peng Tao9bf87482014-08-22 17:37:41 +08002115 }
Andy Adamson568e8c42011-03-01 01:34:22 +00002116
Jeff Layton9a4bf312015-12-10 10:41:58 -05002117 if (pnfs_layoutgets_blocked(lo)) {
Jeff Layton183d9e72016-05-17 12:28:47 -04002118 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
Jeff Layton9a4bf312015-12-10 10:41:58 -05002119 PNFS_UPDATE_LAYOUT_BLOCKED);
Fred Isamancf7d63f2011-01-06 11:36:25 +00002120 goto out_unlock;
Jeff Layton9a4bf312015-12-10 10:41:58 -05002121 }
Trond Myklebust411ae722018-06-23 22:54:33 -04002122 nfs_layoutget_begin(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +00002123 spin_unlock(&ino->i_lock);
Weston Andros Adamson30005122013-02-28 20:30:10 -05002124
Fred Isaman78746a32016-09-22 12:30:20 -04002125 _add_to_server_list(lo, server);
Benny Halevye5e94012010-10-20 00:18:01 -04002126
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002127 pg_offset = arg.offset & ~PAGE_MASK;
Benny Halevy707ed5f2011-05-22 19:47:46 +03002128 if (pg_offset) {
2129 arg.offset -= pg_offset;
2130 arg.length += pg_offset;
2131 }
Andy Adamson7c24d942011-06-13 18:22:38 -04002132 if (arg.length != NFS4_MAX_UINT64)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002133 arg.length = PAGE_ALIGN(arg.length);
Benny Halevy707ed5f2011-05-22 19:47:46 +03002134
Fred Isaman5e36e2a2016-10-06 12:08:51 -04002135 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
Fred Isaman587f03d2016-09-21 05:14:28 -04002136 if (!lgp) {
Trond Myklebust3764a172022-05-14 10:08:11 -04002137 lseg = ERR_PTR(-ENOMEM);
Fred Isaman587f03d2016-09-21 05:14:28 -04002138 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2139 PNFS_UPDATE_LAYOUT_NOMEM);
Trond Myklebust411ae722018-06-23 22:54:33 -04002140 nfs_layoutget_end(lo);
Fred Isaman587f03d2016-09-21 05:14:28 -04002141 goto out_put_layout_hdr;
2142 }
2143
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002144 lgp->lo = lo;
2145 pnfs_get_layout_hdr(lo);
2146
Fred Isamandacb4522016-09-19 17:47:09 -04002147 lseg = nfs4_proc_layoutget(lgp, &timeout);
Jeff Layton183d9e72016-05-17 12:28:47 -04002148 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2149 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
Trond Myklebust411ae722018-06-23 22:54:33 -04002150 nfs_layoutget_end(lo);
Jeff Layton83026d802016-05-17 12:28:46 -04002151 if (IS_ERR(lseg)) {
Jeff Layton183d9e72016-05-17 12:28:47 -04002152 switch(PTR_ERR(lseg)) {
Trond Myklebuste85d7ee2016-07-14 18:46:24 -04002153 case -EBUSY:
Jeff Layton183d9e72016-05-17 12:28:47 -04002154 if (time_after(jiffies, giveup))
2155 lseg = NULL;
Trond Myklebust66b53f32016-07-14 14:28:31 -04002156 break;
2157 case -ERECALLCONFLICT:
Jeff Layton183d9e72016-05-17 12:28:47 -04002158 case -EAGAIN:
Trond Myklebust56b38a12016-07-14 18:34:12 -04002159 break;
Trond Myklebustfe44fb22022-05-31 11:03:06 -04002160 case -ENODATA:
2161 /* The server returned NFS4ERR_LAYOUTUNAVAILABLE */
2162 pnfs_layout_set_fail_bit(
2163 lo, pnfs_iomode_to_fail_bit(iomode));
2164 lseg = NULL;
2165 goto out_put_layout_hdr;
Jeff Layton183d9e72016-05-17 12:28:47 -04002166 default:
2167 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2168 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2169 lseg = NULL;
2170 }
Trond Myklebust56b38a12016-07-14 18:34:12 -04002171 goto out_put_layout_hdr;
2172 }
2173 if (lseg) {
2174 if (first)
2175 pnfs_clear_first_layoutget(lo);
2176 trace_pnfs_update_layout(ino, pos, count,
2177 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2178 pnfs_put_layout_hdr(lo);
2179 goto lookup_again;
Jeff Layton83026d802016-05-17 12:28:46 -04002180 }
2181 } else {
2182 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2183 }
2184
Trond Myklebust830ffb52012-09-20 21:25:19 -04002185out_put_layout_hdr:
Tom Haynesd67ae822014-12-11 17:02:04 -05002186 if (first)
2187 pnfs_clear_first_layoutget(lo);
Trond Myklebustd5b92162019-07-18 09:32:17 -04002188 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2189 PNFS_UPDATE_LAYOUT_EXIT);
Trond Myklebust70c3bd22012-09-18 20:51:13 -04002190 pnfs_put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -04002191out:
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04002192 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2193 "(%s, offset: %llu, length: %llu)\n",
2194 __func__, ino->i_sb->s_id,
2195 (unsigned long long)NFS_FILEID(ino),
Peng Taod600ad12015-12-04 02:57:48 +08002196 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
Trond Myklebustf86bbcf2012-09-26 11:21:40 -04002197 iomode==IOMODE_RW ? "read/write" : "read-only",
2198 (unsigned long long)pos,
2199 (unsigned long long)count);
Benny Halevye5e94012010-10-20 00:18:01 -04002200 return lseg;
2201out_unlock:
2202 spin_unlock(&ino->i_lock);
Trond Myklebust830ffb52012-09-20 21:25:19 -04002203 goto out_put_layout_hdr;
Benny Halevye5e94012010-10-20 00:18:01 -04002204}
Andy Adamson7c24d942011-06-13 18:22:38 -04002205EXPORT_SYMBOL_GPL(pnfs_update_layout);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002206
Trond Myklebust540d9862015-08-25 11:16:13 -04002207static bool
2208pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2209{
2210 switch (range->iomode) {
2211 case IOMODE_READ:
2212 case IOMODE_RW:
2213 break;
2214 default:
2215 return false;
2216 }
2217 if (range->offset == NFS4_MAX_UINT64)
2218 return false;
2219 if (range->length == 0)
2220 return false;
2221 if (range->length != NFS4_MAX_UINT64 &&
2222 range->length > NFS4_MAX_UINT64 - range->offset)
2223 return false;
2224 return true;
2225}
2226
Fred Isaman78746a32016-09-22 12:30:20 -04002227static struct pnfs_layout_hdr *
2228_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2229{
2230 struct pnfs_layout_hdr *lo;
2231
2232 spin_lock(&ino->i_lock);
Trond Myklebust63d8a412022-03-21 15:32:09 -04002233 lo = pnfs_find_alloc_layout(ino, ctx, nfs_io_gfp_mask());
Fred Isaman78746a32016-09-22 12:30:20 -04002234 if (!lo)
2235 goto out_unlock;
2236 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2237 goto out_unlock;
2238 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2239 goto out_unlock;
2240 if (pnfs_layoutgets_blocked(lo))
2241 goto out_unlock;
2242 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2243 goto out_unlock;
Trond Myklebust411ae722018-06-23 22:54:33 -04002244 nfs_layoutget_begin(lo);
Fred Isaman78746a32016-09-22 12:30:20 -04002245 spin_unlock(&ino->i_lock);
2246 _add_to_server_list(lo, NFS_SERVER(ino));
2247 return lo;
2248
2249out_unlock:
2250 spin_unlock(&ino->i_lock);
2251 pnfs_put_layout_hdr(lo);
2252 return NULL;
2253}
2254
Fred Isaman2409a972016-10-06 12:11:21 -04002255static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2256 struct nfs_open_context *ctx)
2257{
Fred Isaman78746a32016-09-22 12:30:20 -04002258 struct inode *ino = data->dentry->d_inode;
2259 struct pnfs_layout_range rng = {
2260 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2261 IOMODE_RW: IOMODE_READ,
2262 .offset = 0,
2263 .length = NFS4_MAX_UINT64,
2264 };
2265 struct nfs4_layoutget *lgp;
2266 struct pnfs_layout_hdr *lo;
2267
Trond Myklebust64294b02017-02-02 12:26:38 -05002268 /* Heuristic: don't send layoutget if we have cached data */
2269 if (rng.iomode == IOMODE_READ &&
2270 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2271 return;
2272
Fred Isaman78746a32016-09-22 12:30:20 -04002273 lo = _pnfs_grab_empty_layout(ino, ctx);
2274 if (!lo)
2275 return;
Trond Myklebust63d8a412022-03-21 15:32:09 -04002276 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid, &rng,
2277 nfs_io_gfp_mask());
Fred Isaman78746a32016-09-22 12:30:20 -04002278 if (!lgp) {
2279 pnfs_clear_first_layoutget(lo);
Trond Myklebustcb2856c2021-01-06 14:13:22 -05002280 nfs_layoutget_end(lo);
Fred Isaman78746a32016-09-22 12:30:20 -04002281 pnfs_put_layout_hdr(lo);
2282 return;
2283 }
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002284 lgp->lo = lo;
Fred Isaman78746a32016-09-22 12:30:20 -04002285 data->lgp = lgp;
2286 data->o_arg.lg_args = &lgp->args;
2287 data->o_res.lg_res = &lgp->res;
Fred Isaman2409a972016-10-06 12:11:21 -04002288}
2289
2290static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2291 struct nfs_open_context *ctx)
2292{
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002293 struct inode *ino = data->dentry->d_inode;
Fred Isaman2409a972016-10-06 12:11:21 -04002294 struct pnfs_layout_range rng = {
2295 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2296 IOMODE_RW: IOMODE_READ,
2297 .offset = 0,
2298 .length = NFS4_MAX_UINT64,
2299 };
2300 struct nfs4_layoutget *lgp;
2301
Trond Myklebust63d8a412022-03-21 15:32:09 -04002302 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid, &rng,
2303 nfs_io_gfp_mask());
Fred Isaman2409a972016-10-06 12:11:21 -04002304 if (!lgp)
2305 return;
2306 data->lgp = lgp;
2307 data->o_arg.lg_args = &lgp->args;
2308 data->o_res.lg_res = &lgp->res;
2309}
2310
2311void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2312 struct nfs_open_context *ctx)
2313{
2314 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2315
2316 if (!(pnfs_enabled_sb(server) &&
2317 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2318 return;
2319 /* Could check on max_ops, but currently hardcoded high enough */
Fred Isaman6e012602016-10-04 15:26:41 -04002320 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2321 return;
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002322 if (data->lgp)
2323 return;
Fred Isaman2409a972016-10-06 12:11:21 -04002324 if (data->state)
2325 _lgopen_prepare_attached(data, ctx);
2326 else
2327 _lgopen_prepare_floating(data, ctx);
2328}
2329
2330void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2331 struct nfs_open_context *ctx)
2332{
2333 struct pnfs_layout_hdr *lo;
2334 struct pnfs_layout_segment *lseg;
Fred Isamanc49b5202016-10-05 09:37:12 -04002335 struct nfs_server *srv = NFS_SERVER(ino);
Fred Isaman2409a972016-10-06 12:11:21 -04002336 u32 iomode;
2337
Fred Isaman6e012602016-10-04 15:26:41 -04002338 if (!lgp)
Fred Isaman2409a972016-10-06 12:11:21 -04002339 return;
Fred Isaman6e012602016-10-04 15:26:41 -04002340 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2341 if (lgp->res.status) {
2342 switch (lgp->res.status) {
Fred Isaman6e012602016-10-04 15:26:41 -04002343 default:
Trond Myklebust8dc96562017-02-01 21:02:07 -05002344 break;
2345 /*
2346 * Halt lgopen attempts if the server doesn't recognise
2347 * the "current stateid" value, the layout type, or the
2348 * layoutget operation as being valid.
2349 * Also if it complains about too many ops in the compound
2350 * or of the request/reply being too big.
2351 */
2352 case -NFS4ERR_BAD_STATEID:
2353 case -NFS4ERR_NOTSUPP:
2354 case -NFS4ERR_REP_TOO_BIG:
2355 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2356 case -NFS4ERR_REQ_TOO_BIG:
2357 case -NFS4ERR_TOO_MANY_OPS:
2358 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
Fred Isamanc49b5202016-10-05 09:37:12 -04002359 srv->caps &= ~NFS_CAP_LGOPEN;
Fred Isaman6e012602016-10-04 15:26:41 -04002360 }
2361 return;
2362 }
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002363 if (!lgp->lo) {
Fred Isaman78746a32016-09-22 12:30:20 -04002364 lo = _pnfs_grab_empty_layout(ino, ctx);
2365 if (!lo)
2366 return;
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002367 lgp->lo = lo;
Fred Isaman2409a972016-10-06 12:11:21 -04002368 } else
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002369 lo = lgp->lo;
Fred Isaman2409a972016-10-06 12:11:21 -04002370
2371 lseg = pnfs_layout_process(lgp);
Trond Myklebust32f1c282018-05-22 11:15:32 -04002372 if (!IS_ERR(lseg)) {
Fred Isaman2409a972016-10-06 12:11:21 -04002373 iomode = lgp->args.range.iomode;
2374 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2375 pnfs_put_lseg(lseg);
2376 }
Fred Isaman30ae2412016-10-18 13:39:51 -04002377}
2378
2379void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2380{
2381 if (lgp != NULL) {
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002382 if (lgp->lo) {
2383 pnfs_clear_first_layoutget(lgp->lo);
2384 nfs_layoutget_end(lgp->lo);
Fred Isaman30ae2412016-10-18 13:39:51 -04002385 }
2386 pnfs_layoutget_free(lgp);
2387 }
Fred Isaman2409a972016-10-06 12:11:21 -04002388}
2389
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -04002390struct pnfs_layout_segment *
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002391pnfs_layout_process(struct nfs4_layoutget *lgp)
2392{
Trond Myklebustb4e89bc2021-07-02 17:24:22 -04002393 struct pnfs_layout_hdr *lo = lgp->lo;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002394 struct nfs4_layoutget_res *res = &lgp->res;
2395 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +00002396 struct inode *ino = lo->plh_inode;
Trond Myklebust78096cc2014-02-12 10:02:27 -05002397 LIST_HEAD(free_me);
Trond Myklebust540d9862015-08-25 11:16:13 -04002398
2399 if (!pnfs_sanity_check_layout_range(&res->range))
Jeff Layton1b3c6d02016-05-17 12:28:48 -04002400 return ERR_PTR(-EINVAL);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002401
2402 /* Inject layout blob into I/O device driver */
Trond Myklebusta75b9df2011-05-11 18:00:51 -04002403 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
Jeff Layton1b3c6d02016-05-17 12:28:48 -04002404 if (IS_ERR_OR_NULL(lseg)) {
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002405 if (!lseg)
Jeff Layton1b3c6d02016-05-17 12:28:48 -04002406 lseg = ERR_PTR(-ENOMEM);
2407
2408 dprintk("%s: Could not allocate layout: error %ld\n",
2409 __func__, PTR_ERR(lseg));
2410 return lseg;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002411 }
2412
Trond Myklebust119cef972016-07-24 15:10:12 -04002413 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
Christoph Hellwig1013df612014-08-21 11:09:18 -05002414
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002415 spin_lock(&ino->i_lock);
Trond Myklebuste1c06f82015-08-04 16:40:08 -04002416 if (pnfs_layoutgets_blocked(lo)) {
Fred Isaman43f1b3d2011-01-06 11:36:30 +00002417 dprintk("%s forget reply due to state\n", __func__);
Jeff Layton1b3c6d02016-05-17 12:28:48 -04002418 goto out_forget;
Fred Isaman43f1b3d2011-01-06 11:36:30 +00002419 }
Trond Myklebust038d6492012-10-02 16:38:41 -07002420
Trond Myklebust880265c2022-05-31 11:03:07 -04002421 if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) &&
2422 !pnfs_is_first_layoutget(lo))
Trond Myklebust0b77f972021-07-02 19:48:41 -04002423 goto out_forget;
2424
Trond Myklebustd29b468d2021-01-22 10:05:51 -05002425 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
Christoph Hellwig362f7472014-08-21 11:09:20 -05002426 /* existing state ID, make sure the sequence number matches. */
2427 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
Trond Myklebust0b77f972021-07-02 19:48:41 -04002428 if (!pnfs_layout_is_valid(lo))
Trond Myklebustd29b468d2021-01-22 10:05:51 -05002429 lo->plh_barrier = 0;
Christoph Hellwig362f7472014-08-21 11:09:20 -05002430 dprintk("%s forget reply due to sequence\n", __func__);
Jeff Layton1b3c6d02016-05-17 12:28:48 -04002431 goto out_forget;
Christoph Hellwig362f7472014-08-21 11:09:20 -05002432 }
Trond Myklebust59b56392020-01-27 13:07:26 -05002433 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
Trond Myklebustd29b468d2021-01-22 10:05:51 -05002434 } else if (pnfs_layout_is_valid(lo)) {
Christoph Hellwig362f7472014-08-21 11:09:20 -05002435 /*
2436 * We got an entirely new state ID. Mark all segments for the
Trond Myklebust9888d832016-11-23 12:36:04 -05002437 * inode invalid, and retry the layoutget
Christoph Hellwig362f7472014-08-21 11:09:20 -05002438 */
Trond Myklebust08bd8db2021-01-21 17:11:42 -05002439 struct pnfs_layout_range range = {
2440 .iomode = IOMODE_ANY,
2441 .length = NFS4_MAX_UINT64,
2442 };
Trond Myklebustfb700ef2021-04-15 17:33:07 -04002443 pnfs_mark_matching_lsegs_return(lo, &free_me, &range, 0);
Trond Myklebust9888d832016-11-23 12:36:04 -05002444 goto out_forget;
Trond Myklebustd29b468d2021-01-22 10:05:51 -05002445 } else {
2446 /* We have a completely new layout */
Trond Myklebustd29b468d2021-01-22 10:05:51 -05002447 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
Christoph Hellwig362f7472014-08-21 11:09:20 -05002448 }
Trond Myklebust038d6492012-10-02 16:38:41 -07002449
Trond Myklebust9369a432012-09-18 20:57:08 -04002450 pnfs_get_lseg(lseg);
Trond Myklebust03772d22015-08-25 08:54:17 -04002451 pnfs_layout_insert_lseg(lo, lseg, &free_me);
Trond Myklebust8e0acf92016-07-21 11:53:29 -04002452
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002453
Peng Tao39761432015-08-21 12:49:44 +08002454 if (res->return_on_close)
Fred Isamanf7e89172011-01-06 11:36:32 +00002455 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
Fred Isamanf7e89172011-01-06 11:36:32 +00002456
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002457 spin_unlock(&ino->i_lock);
Trond Myklebust78096cc2014-02-12 10:02:27 -05002458 pnfs_free_lseg_list(&free_me);
Trond Myklebusta0b0a6e2012-09-17 17:12:15 -04002459 return lseg;
Fred Isaman43f1b3d2011-01-06 11:36:30 +00002460
Jeff Layton1b3c6d02016-05-17 12:28:48 -04002461out_forget:
Fred Isaman43f1b3d2011-01-06 11:36:30 +00002462 spin_unlock(&ino->i_lock);
2463 lseg->pls_layout = lo;
2464 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Jeff Layton1b3c6d02016-05-17 12:28:48 -04002465 return ERR_PTR(-EAGAIN);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002466}
2467
Trond Myklebust2f215962016-02-15 12:36:04 -05002468/**
2469 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2470 * @lo: pointer to layout header
2471 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2472 * @return_range: describe layout segment ranges to be returned
Trond Myklebuste0b7d422018-06-23 10:28:40 -04002473 * @seq: stateid seqid to match
Trond Myklebust2f215962016-02-15 12:36:04 -05002474 *
2475 * This function is mainly intended for use by layoutrecall. It attempts
2476 * to free the layout segment immediately, or else to mark it for return
2477 * as soon as its reference count drops to zero.
Trond Myklebuste0b7d422018-06-23 10:28:40 -04002478 *
2479 * Returns
2480 * - 0: a layoutreturn needs to be scheduled.
2481 * - EBUSY: there are layout segment that are still in use.
2482 * - ENOENT: there are no layout segments that need to be returned.
Trond Myklebust2f215962016-02-15 12:36:04 -05002483 */
Trond Myklebust10335552016-01-04 11:23:52 -05002484int
Peng Tao016256d2014-09-06 00:53:23 +08002485pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2486 struct list_head *tmp_list,
Jeff Layton6d597e12016-05-17 12:28:42 -04002487 const struct pnfs_layout_range *return_range,
2488 u32 seq)
Peng Tao016256d2014-09-06 00:53:23 +08002489{
2490 struct pnfs_layout_segment *lseg, *next;
Trond Myklebustb739a5b2022-10-05 15:57:38 -04002491 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
Trond Myklebust10335552016-01-04 11:23:52 -05002492 int remaining = 0;
Peng Tao016256d2014-09-06 00:53:23 +08002493
2494 dprintk("%s:Begin lo %p\n", __func__, lo);
2495
Trond Myklebustfc7ff362015-12-28 10:28:59 -05002496 assert_spin_locked(&lo->plh_inode->i_lock);
Peng Tao016256d2014-09-06 00:53:23 +08002497
Trond Myklebust39fd0182021-04-15 15:41:57 -04002498 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2499 tmp_list = &lo->plh_return_segs;
2500
Peng Tao016256d2014-09-06 00:53:23 +08002501 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
Trond Myklebuste036f462016-07-22 11:13:22 -04002502 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
Peng Tao016256d2014-09-06 00:53:23 +08002503 dprintk("%s: marking lseg %p iomode %d "
2504 "offset %llu length %llu\n", __func__,
2505 lseg, lseg->pls_range.iomode,
2506 lseg->pls_range.offset,
2507 lseg->pls_range.length);
Trond Myklebust39fd0182021-04-15 15:41:57 -04002508 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2509 tmp_list = &lo->plh_return_segs;
Trond Myklebustff041722020-08-04 16:30:30 -04002510 if (mark_lseg_invalid(lseg, tmp_list))
Trond Myklebust2f215962016-02-15 12:36:04 -05002511 continue;
2512 remaining++;
Peng Tao016256d2014-09-06 00:53:23 +08002513 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
Trond Myklebustb739a5b2022-10-05 15:57:38 -04002514 pnfs_lseg_cancel_io(server, lseg);
Peng Tao016256d2014-09-06 00:53:23 +08002515 }
Jeff Layton6d597e12016-05-17 12:28:42 -04002516
Trond Myklebuste0b7d422018-06-23 10:28:40 -04002517 if (remaining) {
Jeff Layton6d597e12016-05-17 12:28:42 -04002518 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
Trond Myklebuste0b7d422018-06-23 10:28:40 -04002519 return -EBUSY;
2520 }
Jeff Layton6d597e12016-05-17 12:28:42 -04002521
Trond Myklebuste0b7d422018-06-23 10:28:40 -04002522 if (!list_empty(&lo->plh_return_segs)) {
2523 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2524 return 0;
2525 }
2526
2527 return -ENOENT;
Peng Tao016256d2014-09-06 00:53:23 +08002528}
2529
Trond Myklebustb5fdf842020-02-18 15:58:31 -05002530static void
2531pnfs_mark_layout_for_return(struct inode *inode,
2532 const struct pnfs_layout_range *range)
Peng Tao016256d2014-09-06 00:53:23 +08002533{
Trond Myklebustb5fdf842020-02-18 15:58:31 -05002534 struct pnfs_layout_hdr *lo;
Trond Myklebust10335552016-01-04 11:23:52 -05002535 bool return_now = false;
Peng Tao016256d2014-09-06 00:53:23 +08002536
2537 spin_lock(&inode->i_lock);
Trond Myklebustb5fdf842020-02-18 15:58:31 -05002538 lo = NFS_I(inode)->layout;
Trond Myklebustbdebfcc2017-04-27 15:30:00 -04002539 if (!pnfs_layout_is_valid(lo)) {
2540 spin_unlock(&inode->i_lock);
2541 return;
2542 }
Trond Myklebustb5fdf842020-02-18 15:58:31 -05002543 pnfs_set_plh_return_info(lo, range->iomode, 0);
Peng Tao016256d2014-09-06 00:53:23 +08002544 /*
2545 * mark all matching lsegs so that we are sure to have no live
2546 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2547 * for how it works.
2548 */
Trond Myklebustb5fdf842020-02-18 15:58:31 -05002549 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
Trond Myklebust44ea8df2020-04-02 15:37:02 -04002550 const struct cred *cred;
Trond Myklebust10335552016-01-04 11:23:52 -05002551 nfs4_stateid stateid;
Trond Myklebuste5fd1902016-07-21 12:44:15 -04002552 enum pnfs_iomode iomode;
Trond Myklebust10335552016-01-04 11:23:52 -05002553
Trond Myklebust44ea8df2020-04-02 15:37:02 -04002554 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
Trond Myklebust10335552016-01-04 11:23:52 -05002555 spin_unlock(&inode->i_lock);
2556 if (return_now)
Trond Myklebust44ea8df2020-04-02 15:37:02 -04002557 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
Trond Myklebust10335552016-01-04 11:23:52 -05002558 } else {
2559 spin_unlock(&inode->i_lock);
2560 nfs_commit_inode(inode, 0);
2561 }
Peng Tao016256d2014-09-06 00:53:23 +08002562}
Trond Myklebustb5fdf842020-02-18 15:58:31 -05002563
2564void pnfs_error_mark_layout_for_return(struct inode *inode,
2565 struct pnfs_layout_segment *lseg)
2566{
2567 struct pnfs_layout_range range = {
2568 .iomode = lseg->pls_range.iomode,
2569 .offset = 0,
2570 .length = NFS4_MAX_UINT64,
2571 };
2572
2573 pnfs_mark_layout_for_return(inode, &range);
2574}
Peng Tao016256d2014-09-06 00:53:23 +08002575EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2576
Trond Myklebustb5fdf842020-02-18 15:58:31 -05002577static bool
2578pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
2579{
2580 return pnfs_layout_is_valid(lo) &&
2581 !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
2582 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
2583}
2584
2585static struct pnfs_layout_segment *
2586pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
2587 const struct pnfs_layout_range *range,
2588 enum pnfs_iomode iomode)
2589{
2590 struct pnfs_layout_segment *lseg;
2591
2592 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
2593 if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
2594 continue;
2595 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2596 continue;
2597 if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
2598 continue;
2599 if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
2600 return lseg;
2601 }
2602 return NULL;
2603}
2604
2605/* Find open file states whose mode matches that of the range */
2606static bool
2607pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
2608 const struct pnfs_layout_range *range)
2609{
2610 struct list_head *head;
2611 struct nfs_open_context *ctx;
2612 fmode_t mode = 0;
2613
2614 if (!pnfs_layout_can_be_returned(lo) ||
2615 !pnfs_find_first_lseg(lo, range, range->iomode))
2616 return false;
2617
2618 head = &NFS_I(lo->plh_inode)->open_files;
2619 list_for_each_entry_rcu(ctx, head, list) {
2620 if (ctx->state)
2621 mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
2622 }
2623
2624 switch (range->iomode) {
2625 default:
2626 break;
2627 case IOMODE_READ:
2628 mode &= ~FMODE_WRITE;
2629 break;
2630 case IOMODE_RW:
2631 if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
2632 mode &= ~FMODE_READ;
2633 }
2634 return mode == 0;
2635}
2636
2637static int
2638pnfs_layout_return_unused_byserver(struct nfs_server *server, void *data)
2639{
2640 const struct pnfs_layout_range *range = data;
2641 struct pnfs_layout_hdr *lo;
2642 struct inode *inode;
2643restart:
2644 rcu_read_lock();
2645 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
2646 if (!pnfs_layout_can_be_returned(lo) ||
2647 test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2648 continue;
2649 inode = lo->plh_inode;
2650 spin_lock(&inode->i_lock);
2651 if (!pnfs_should_return_unused_layout(lo, range)) {
2652 spin_unlock(&inode->i_lock);
2653 continue;
2654 }
2655 spin_unlock(&inode->i_lock);
2656 inode = pnfs_grab_inode_layout_hdr(lo);
2657 if (!inode)
2658 continue;
2659 rcu_read_unlock();
2660 pnfs_mark_layout_for_return(inode, range);
2661 iput(inode);
2662 cond_resched();
2663 goto restart;
2664 }
2665 rcu_read_unlock();
2666 return 0;
2667}
2668
2669void
2670pnfs_layout_return_unused_byclid(struct nfs_client *clp,
2671 enum pnfs_iomode iomode)
2672{
2673 struct pnfs_layout_range range = {
2674 .iomode = iomode,
2675 .offset = 0,
2676 .length = NFS4_MAX_UINT64,
2677 };
2678
2679 nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
2680 &range);
2681}
2682
Trond Myklebustd8007d42011-06-10 13:30:23 -04002683void
Trond Myklebustb3230e82017-04-25 10:56:19 -04002684pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2685{
2686 if (pgio->pg_lseg == NULL ||
2687 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2688 return;
2689 pnfs_put_lseg(pgio->pg_lseg);
2690 pgio->pg_lseg = NULL;
2691}
2692EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2693
Benjamin Coddington08cb5b02017-05-22 20:20:23 -04002694/*
2695 * Check for any intersection between the request and the pgio->pg_lseg,
2696 * and if none, put this pgio->pg_lseg away.
2697 */
Trond Myklebuste1e54ab2020-03-23 14:48:23 -04002698void
Benjamin Coddington08cb5b02017-05-22 20:20:23 -04002699pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2700{
2701 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2702 pnfs_put_lseg(pgio->pg_lseg);
2703 pgio->pg_lseg = NULL;
2704 }
2705}
Trond Myklebuste1e54ab2020-03-23 14:48:23 -04002706EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
Benjamin Coddington08cb5b02017-05-22 20:20:23 -04002707
Trond Myklebustb3230e82017-04-25 10:56:19 -04002708void
Trond Myklebustd8007d42011-06-10 13:30:23 -04002709pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2710{
Yang Lid1d97392021-05-12 17:20:04 +08002711 u64 rd_size;
Peng Tao1fd937b2012-09-25 14:55:57 +08002712
Trond Myklebustb3230e82017-04-25 10:56:19 -04002713 pnfs_generic_pg_check_layout(pgio);
Benjamin Coddington08cb5b02017-05-22 20:20:23 -04002714 pnfs_generic_pg_check_range(pgio, req);
Peng Taocb5d04b2015-01-24 22:14:52 +08002715 if (pgio->pg_lseg == NULL) {
2716 if (pgio->pg_dreq == NULL)
2717 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2718 else
2719 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
Trond Myklebustd8007d42011-06-10 13:30:23 -04002720
Trond Myklebust63d8a412022-03-21 15:32:09 -04002721 pgio->pg_lseg =
2722 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
2723 req_offset(req), rd_size,
2724 IOMODE_READ, false,
2725 nfs_io_gfp_mask());
Peng Taod600ad12015-12-04 02:57:48 +08002726 if (IS_ERR(pgio->pg_lseg)) {
2727 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2728 pgio->pg_lseg = NULL;
2729 return;
2730 }
Peng Taocb5d04b2015-01-24 22:14:52 +08002731 }
Trond Myklebuste885de12011-06-10 13:30:23 -04002732 /* If no lseg, fall back to read through mds */
2733 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04002734 nfs_pageio_reset_read_mds(pgio);
Trond Myklebuste885de12011-06-10 13:30:23 -04002735
Trond Myklebustd8007d42011-06-10 13:30:23 -04002736}
2737EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2738
2739void
Peng Tao62965562012-09-25 14:55:57 +08002740pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2741 struct nfs_page *req, u64 wb_size)
Trond Myklebustd8007d42011-06-10 13:30:23 -04002742{
Trond Myklebustb3230e82017-04-25 10:56:19 -04002743 pnfs_generic_pg_check_layout(pgio);
Benjamin Coddington08cb5b02017-05-22 20:20:23 -04002744 pnfs_generic_pg_check_range(pgio, req);
Peng Taod600ad12015-12-04 02:57:48 +08002745 if (pgio->pg_lseg == NULL) {
Trond Myklebust63d8a412022-03-21 15:32:09 -04002746 pgio->pg_lseg =
2747 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
2748 req_offset(req), wb_size, IOMODE_RW,
2749 false, nfs_io_gfp_mask());
Peng Taod600ad12015-12-04 02:57:48 +08002750 if (IS_ERR(pgio->pg_lseg)) {
2751 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2752 pgio->pg_lseg = NULL;
2753 return;
2754 }
2755 }
Trond Myklebuste885de12011-06-10 13:30:23 -04002756 /* If no lseg, fall back to write through mds */
2757 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04002758 nfs_pageio_reset_write_mds(pgio);
Trond Myklebustd8007d42011-06-10 13:30:23 -04002759}
2760EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2761
Weston Andros Adamson180bb5e2014-09-10 15:48:01 -04002762void
2763pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2764{
2765 if (desc->pg_lseg) {
2766 pnfs_put_lseg(desc->pg_lseg);
2767 desc->pg_lseg = NULL;
2768 }
2769}
2770EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2771
Weston Andros Adamsonb4fdac12014-05-15 11:56:43 -04002772/*
2773 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2774 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2775 */
2776size_t
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04002777pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2778 struct nfs_page *prev, struct nfs_page *req)
Fred Isamanbae724e2011-03-01 01:34:15 +00002779{
Weston Andros Adamson0f9c429e2014-05-15 11:56:51 -04002780 unsigned int size;
Weston Andros Adamsonc5e20cb2014-06-09 17:47:26 -04002781 u64 seg_end, req_start, seg_left;
Weston Andros Adamson0f9c429e2014-05-15 11:56:51 -04002782
2783 size = nfs_generic_pg_test(pgio, prev, req);
Weston Andros Adamson0f9c429e2014-05-15 11:56:51 -04002784 if (!size)
2785 return 0;
Benny Halevy89a58e32011-05-25 20:54:40 +03002786
Trond Myklebust19982ba2011-06-10 13:30:23 -04002787 /*
Weston Andros Adamsonc5e20cb2014-06-09 17:47:26 -04002788 * 'size' contains the number of bytes left in the current page (up
2789 * to the original size asked for in @req->wb_bytes).
2790 *
2791 * Calculate how many bytes are left in the layout segment
2792 * and if there are less bytes than 'size', return that instead.
Trond Myklebust19982ba2011-06-10 13:30:23 -04002793 *
2794 * Please also note that 'end_offset' is actually the offset of the
2795 * first byte that lies outside the pnfs_layout_range. FIXME?
2796 *
2797 */
Weston Andros Adamson19b54842014-05-15 11:56:55 -04002798 if (pgio->pg_lseg) {
Trond Myklebust17822b22016-10-25 12:24:25 -04002799 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
Weston Andros Adamsonc5e20cb2014-06-09 17:47:26 -04002800 pgio->pg_lseg->pls_range.length);
2801 req_start = req_offset(req);
Benjamin Coddington08cb5b02017-05-22 20:20:23 -04002802
Weston Andros Adamsonc5e20cb2014-06-09 17:47:26 -04002803 /* start of request is past the last byte of this segment */
Benjamin Coddington08cb5b02017-05-22 20:20:23 -04002804 if (req_start >= seg_end)
Weston Andros Adamson19b54842014-05-15 11:56:55 -04002805 return 0;
Weston Andros Adamsonc5e20cb2014-06-09 17:47:26 -04002806
2807 /* adjust 'size' iff there are fewer bytes left in the
2808 * segment than what nfs_generic_pg_test returned */
2809 seg_left = seg_end - req_start;
2810 if (seg_left < size)
2811 size = (unsigned int)seg_left;
Weston Andros Adamson19b54842014-05-15 11:56:55 -04002812 }
Weston Andros Adamson0f9c429e2014-05-15 11:56:51 -04002813
Weston Andros Adamson19b54842014-05-15 11:56:55 -04002814 return size;
Fred Isamanbae724e2011-03-01 01:34:15 +00002815}
Benny Halevy89a58e32011-05-25 20:54:40 +03002816EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
Fred Isamanbae724e2011-03-01 01:34:15 +00002817
Weston Andros Adamson53113ad2014-06-09 11:48:38 -04002818int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
Trond Myklebuste2fecb22012-01-06 08:57:46 -05002819{
2820 struct nfs_pageio_descriptor pgio;
Trond Myklebuste2fecb22012-01-06 08:57:46 -05002821
2822 /* Resend all requests through the MDS */
Weston Andros Adamson53113ad2014-06-09 11:48:38 -04002823 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2824 hdr->completion_ops);
2825 return nfs_pageio_resend(&pgio, hdr);
Trond Myklebuste2fecb22012-01-06 08:57:46 -05002826}
Andy Adamsone7dd79af2012-04-27 17:53:46 -04002827EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
Trond Myklebuste2fecb22012-01-06 08:57:46 -05002828
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002829static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002830{
Fred Isamancd841602012-04-20 14:47:44 -04002831
2832 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2833 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002834 PNFS_LAYOUTRET_ON_ERROR) {
Fred Isamancd841602012-04-20 14:47:44 -04002835 pnfs_return_layout(hdr->inode);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002836 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04002837 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
Weston Andros Adamson53113ad2014-06-09 11:48:38 -04002838 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002839}
2840
Benny Halevyd20581a2011-05-22 19:52:03 +03002841/*
2842 * Called by non rpc-based layout drivers
2843 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002844void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
Fred Isaman94ad1c82011-03-01 01:34:14 +00002845{
Kinglong Meef8417b482015-10-16 17:23:29 +08002846 if (likely(!hdr->pnfs_error)) {
Trond Myklebust67af7612015-03-25 20:40:38 -04002847 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2848 hdr->mds_offset + hdr->res.count);
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002849 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
Kinglong Meef8417b482015-10-16 17:23:29 +08002850 }
2851 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2852 if (unlikely(hdr->pnfs_error))
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002853 pnfs_ld_handle_write_error(hdr);
2854 hdr->mds_ops->rpc_release(hdr);
Fred Isaman44b83792011-03-03 15:13:44 +00002855}
Benny Halevyd20581a2011-05-22 19:52:03 +03002856EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
Fred Isaman44b83792011-03-03 15:13:44 +00002857
Trond Myklebustdce81292011-07-13 15:59:19 -04002858static void
2859pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002860 struct nfs_pgio_header *hdr)
Trond Myklebustdce81292011-07-13 15:59:19 -04002861{
Peng Tao48d635f2014-11-10 08:35:35 +08002862 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04002863
Fred Isaman6c75dc02012-04-20 14:47:47 -04002864 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04002865 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
Fred Isaman6c75dc02012-04-20 14:47:47 -04002866 nfs_pageio_reset_write_mds(desc);
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04002867 mirror->pg_recoalesce = 1;
Fred Isaman6c75dc02012-04-20 14:47:47 -04002868 }
Scott Mayhewba4a76f2017-12-15 16:12:32 -05002869 hdr->completion_ops->completion(hdr);
Trond Myklebustdce81292011-07-13 15:59:19 -04002870}
2871
2872static enum pnfs_try_status
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002873pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
Trond Myklebustdce81292011-07-13 15:59:19 -04002874 const struct rpc_call_ops *call_ops,
2875 struct pnfs_layout_segment *lseg,
2876 int how)
Andy Adamson0382b742011-03-03 15:13:45 +00002877{
Fred Isamancd841602012-04-20 14:47:44 -04002878 struct inode *inode = hdr->inode;
Andy Adamson0382b742011-03-03 15:13:45 +00002879 enum pnfs_try_status trypnfs;
2880 struct nfs_server *nfss = NFS_SERVER(inode);
2881
Fred Isamancd841602012-04-20 14:47:44 -04002882 hdr->mds_ops = call_ops;
Andy Adamson0382b742011-03-03 15:13:45 +00002883
2884 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002885 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2886 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
Fred Isaman6c75dc02012-04-20 14:47:47 -04002887 if (trypnfs != PNFS_NOT_ATTEMPTED)
Andy Adamson0382b742011-03-03 15:13:45 +00002888 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
Andy Adamson0382b742011-03-03 15:13:45 +00002889 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2890 return trypnfs;
2891}
2892
Trond Myklebustdce81292011-07-13 15:59:19 -04002893static void
Weston Andros Adamson7f714722014-05-15 11:56:53 -04002894pnfs_do_write(struct nfs_pageio_descriptor *desc,
2895 struct nfs_pgio_header *hdr, int how)
Trond Myklebustdce81292011-07-13 15:59:19 -04002896{
Trond Myklebustdce81292011-07-13 15:59:19 -04002897 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2898 struct pnfs_layout_segment *lseg = desc->pg_lseg;
Weston Andros Adamson7f714722014-05-15 11:56:53 -04002899 enum pnfs_try_status trypnfs;
Trond Myklebustdce81292011-07-13 15:59:19 -04002900
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002901 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
Trond Myklebust37f8aa12017-04-29 00:02:37 -04002902 switch (trypnfs) {
2903 case PNFS_NOT_ATTEMPTED:
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002904 pnfs_write_through_mds(desc, hdr);
Gustavo A. R. Silvaffb81712020-11-20 12:26:46 -06002905 break;
Trond Myklebust37f8aa12017-04-29 00:02:37 -04002906 case PNFS_ATTEMPTED:
2907 break;
2908 case PNFS_TRY_AGAIN:
2909 /* cleanup hdr and prepare to redo pnfs */
2910 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2911 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2912 list_splice_init(&hdr->pages, &mirror->pg_list);
2913 mirror->pg_recoalesce = 1;
2914 }
2915 hdr->mds_ops->rpc_release(hdr);
2916 }
Trond Myklebustdce81292011-07-13 15:59:19 -04002917}
2918
Fred Isaman6c75dc02012-04-20 14:47:47 -04002919static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2920{
Trond Myklebust9369a432012-09-18 20:57:08 -04002921 pnfs_put_lseg(hdr->lseg);
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -04002922 nfs_pgio_header_free(hdr);
Fred Isaman6c75dc02012-04-20 14:47:47 -04002923}
2924
Trond Myklebustdce81292011-07-13 15:59:19 -04002925int
2926pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2927{
Fred Isaman6c75dc02012-04-20 14:47:47 -04002928 struct nfs_pgio_header *hdr;
Trond Myklebustdce81292011-07-13 15:59:19 -04002929 int ret;
2930
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -04002931 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2932 if (!hdr) {
Peng Tao2bff2282015-12-05 02:03:17 +08002933 desc->pg_error = -ENOMEM;
2934 return desc->pg_error;
Fred Isaman6c75dc02012-04-20 14:47:47 -04002935 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04002936 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
Weston Andros Adamson180bb5e2014-09-10 15:48:01 -04002937
Trond Myklebust9369a432012-09-18 20:57:08 -04002938 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
Anna Schumakeref2c4882014-05-06 09:12:36 -04002939 ret = nfs_generic_pgio(desc, hdr);
Weston Andros Adamson180bb5e2014-09-10 15:48:01 -04002940 if (!ret)
Weston Andros Adamson7f714722014-05-15 11:56:53 -04002941 pnfs_do_write(desc, hdr, desc->pg_ioflags);
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04002942
Fred Isaman6c75dc02012-04-20 14:47:47 -04002943 return ret;
Trond Myklebustdce81292011-07-13 15:59:19 -04002944}
2945EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2946
Weston Andros Adamson53113ad2014-06-09 11:48:38 -04002947int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
Trond Myklebust62e4a762011-11-10 14:30:37 -05002948{
2949 struct nfs_pageio_descriptor pgio;
2950
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002951 /* Resend all requests through the MDS */
Weston Andros Adamson53113ad2014-06-09 11:48:38 -04002952 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2953 return nfs_pageio_resend(&pgio, hdr);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002954}
Andy Adamsone7dd79af2012-04-27 17:53:46 -04002955EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002956
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002957static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002958{
Fred Isamancd841602012-04-20 14:47:44 -04002959 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2960 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002961 PNFS_LAYOUTRET_ON_ERROR) {
Fred Isamancd841602012-04-20 14:47:44 -04002962 pnfs_return_layout(hdr->inode);
Fred Isaman1acbbb4e12012-04-20 14:47:37 -04002963 }
Fred Isaman4db6e0b2012-04-20 14:47:46 -04002964 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
Weston Andros Adamson53113ad2014-06-09 11:48:38 -04002965 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
Trond Myklebust62e4a762011-11-10 14:30:37 -05002966}
2967
Andy Adamsonb1f69b72010-10-20 00:18:03 -04002968/*
Benny Halevyd20581a2011-05-22 19:52:03 +03002969 * Called by non rpc-based layout drivers
2970 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002971void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
Benny Halevyd20581a2011-05-22 19:52:03 +03002972{
Trond Myklebustbfc505d2016-09-15 18:26:05 -04002973 if (likely(!hdr->pnfs_error))
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002974 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
Kinglong Meef8417b482015-10-16 17:23:29 +08002975 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2976 if (unlikely(hdr->pnfs_error))
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002977 pnfs_ld_handle_read_error(hdr);
2978 hdr->mds_ops->rpc_release(hdr);
Benny Halevyd20581a2011-05-22 19:52:03 +03002979}
2980EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2981
Trond Myklebust493292d2011-07-13 15:58:28 -04002982static void
2983pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04002984 struct nfs_pgio_header *hdr)
Trond Myklebust493292d2011-07-13 15:58:28 -04002985{
Peng Tao48d635f2014-11-10 08:35:35 +08002986 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04002987
Fred Isaman4db6e0b2012-04-20 14:47:46 -04002988 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04002989 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04002990 nfs_pageio_reset_read_mds(desc);
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04002991 mirror->pg_recoalesce = 1;
Fred Isaman4db6e0b2012-04-20 14:47:46 -04002992 }
Scott Mayhewba4a76f2017-12-15 16:12:32 -05002993 hdr->completion_ops->completion(hdr);
Trond Myklebust493292d2011-07-13 15:58:28 -04002994}
2995
Benny Halevyd20581a2011-05-22 19:52:03 +03002996/*
Andy Adamson64419a92011-03-01 01:34:16 +00002997 * Call the appropriate parallel I/O subsystem read function.
2998 */
Trond Myklebust493292d2011-07-13 15:58:28 -04002999static enum pnfs_try_status
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04003000pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
Trond Myklebust493292d2011-07-13 15:58:28 -04003001 const struct rpc_call_ops *call_ops,
3002 struct pnfs_layout_segment *lseg)
Andy Adamson64419a92011-03-01 01:34:16 +00003003{
Fred Isamancd841602012-04-20 14:47:44 -04003004 struct inode *inode = hdr->inode;
Andy Adamson64419a92011-03-01 01:34:16 +00003005 struct nfs_server *nfss = NFS_SERVER(inode);
3006 enum pnfs_try_status trypnfs;
3007
Fred Isamancd841602012-04-20 14:47:44 -04003008 hdr->mds_ops = call_ops;
Andy Adamson64419a92011-03-01 01:34:16 +00003009
3010 dprintk("%s: Reading ino:%lu %u@%llu\n",
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04003011 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
Andy Adamson64419a92011-03-01 01:34:16 +00003012
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04003013 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04003014 if (trypnfs != PNFS_NOT_ATTEMPTED)
Andy Adamson64419a92011-03-01 01:34:16 +00003015 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
Andy Adamson64419a92011-03-01 01:34:16 +00003016 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
3017 return trypnfs;
3018}
Andy Adamson863a3c62011-03-23 13:27:54 +00003019
Peng Taoceb11e12014-11-10 08:35:38 +08003020/* Resend all requests through pnfs. */
Trond Myklebust563c53e2020-08-11 13:36:32 -04003021void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr,
3022 unsigned int mirror_idx)
Peng Taoceb11e12014-11-10 08:35:38 +08003023{
3024 struct nfs_pageio_descriptor pgio;
3025
Weston Andros Adamson1b1bc662016-04-01 11:42:28 -04003026 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
Trond Myklebust54e4a0d2016-11-27 15:12:39 -05003027 /* Prevent deadlocks with layoutreturn! */
3028 pnfs_put_lseg(hdr->lseg);
3029 hdr->lseg = NULL;
3030
Weston Andros Adamson1b1bc662016-04-01 11:42:28 -04003031 nfs_pageio_init_read(&pgio, hdr->inode, false,
3032 hdr->completion_ops);
Trond Myklebust563c53e2020-08-11 13:36:32 -04003033 pgio.pg_mirror_idx = mirror_idx;
Weston Andros Adamson1b1bc662016-04-01 11:42:28 -04003034 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
3035 }
Peng Taoceb11e12014-11-10 08:35:38 +08003036}
3037EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
3038
Trond Myklebust493292d2011-07-13 15:58:28 -04003039static void
Weston Andros Adamson7f714722014-05-15 11:56:53 -04003040pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
Trond Myklebust493292d2011-07-13 15:58:28 -04003041{
Trond Myklebust493292d2011-07-13 15:58:28 -04003042 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
3043 struct pnfs_layout_segment *lseg = desc->pg_lseg;
Weston Andros Adamson7f714722014-05-15 11:56:53 -04003044 enum pnfs_try_status trypnfs;
Trond Myklebust493292d2011-07-13 15:58:28 -04003045
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04003046 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
Trond Myklebust6aeafd02017-04-25 11:26:53 -04003047 switch (trypnfs) {
3048 case PNFS_NOT_ATTEMPTED:
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04003049 pnfs_read_through_mds(desc, hdr);
Gustavo A. R. Silvaffb81712020-11-20 12:26:46 -06003050 break;
Trond Myklebust6aeafd02017-04-25 11:26:53 -04003051 case PNFS_ATTEMPTED:
3052 break;
3053 case PNFS_TRY_AGAIN:
3054 /* cleanup hdr and prepare to redo pnfs */
3055 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3056 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3057 list_splice_init(&hdr->pages, &mirror->pg_list);
3058 mirror->pg_recoalesce = 1;
3059 }
3060 hdr->mds_ops->rpc_release(hdr);
3061 }
Trond Myklebust493292d2011-07-13 15:58:28 -04003062}
3063
Fred Isaman4db6e0b2012-04-20 14:47:46 -04003064static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
3065{
Trond Myklebust9369a432012-09-18 20:57:08 -04003066 pnfs_put_lseg(hdr->lseg);
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -04003067 nfs_pgio_header_free(hdr);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04003068}
3069
Trond Myklebust493292d2011-07-13 15:58:28 -04003070int
3071pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
3072{
Fred Isaman4db6e0b2012-04-20 14:47:46 -04003073 struct nfs_pgio_header *hdr;
Trond Myklebust493292d2011-07-13 15:58:28 -04003074 int ret;
3075
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -04003076 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
3077 if (!hdr) {
Peng Tao2bff2282015-12-05 02:03:17 +08003078 desc->pg_error = -ENOMEM;
3079 return desc->pg_error;
Trond Myklebust493292d2011-07-13 15:58:28 -04003080 }
Fred Isaman4db6e0b2012-04-20 14:47:46 -04003081 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
Trond Myklebust9369a432012-09-18 20:57:08 -04003082 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
Anna Schumakeref2c4882014-05-06 09:12:36 -04003083 ret = nfs_generic_pgio(desc, hdr);
Weston Andros Adamson180bb5e2014-09-10 15:48:01 -04003084 if (!ret)
Weston Andros Adamson7f714722014-05-15 11:56:53 -04003085 pnfs_do_read(desc, hdr);
Fred Isaman4db6e0b2012-04-20 14:47:46 -04003086 return ret;
Trond Myklebust493292d2011-07-13 15:58:28 -04003087}
3088EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
3089
Trond Myklebust71244d92014-01-13 13:34:36 -05003090static void pnfs_clear_layoutcommitting(struct inode *inode)
3091{
3092 unsigned long *bitlock = &NFS_I(inode)->flags;
3093
3094 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
Peter Zijlstra4e857c52014-03-17 18:06:10 +01003095 smp_mb__after_atomic();
Trond Myklebust71244d92014-01-13 13:34:36 -05003096 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
3097}
3098
Andy Adamson863a3c62011-03-23 13:27:54 +00003099/*
Peng Taoa9bae562011-07-30 20:52:33 -04003100 * There can be multiple RW segments.
Andy Adamson863a3c62011-03-23 13:27:54 +00003101 */
Peng Taoa9bae562011-07-30 20:52:33 -04003102static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
Andy Adamson863a3c62011-03-23 13:27:54 +00003103{
Peng Taoa9bae562011-07-30 20:52:33 -04003104 struct pnfs_layout_segment *lseg;
Andy Adamson863a3c62011-03-23 13:27:54 +00003105
Peng Taoa9bae562011-07-30 20:52:33 -04003106 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
3107 if (lseg->pls_range.iomode == IOMODE_RW &&
Trond Myklebusta073dbf2013-03-20 12:34:32 -04003108 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
Peng Taoa9bae562011-07-30 20:52:33 -04003109 list_add(&lseg->pls_lc_list, listp);
3110 }
Andy Adamson863a3c62011-03-23 13:27:54 +00003111}
3112
Trond Myklebusta073dbf2013-03-20 12:34:32 -04003113static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
3114{
3115 struct pnfs_layout_segment *lseg, *tmp;
Trond Myklebusta073dbf2013-03-20 12:34:32 -04003116
3117 /* Matched by references in pnfs_set_layoutcommit */
3118 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
3119 list_del_init(&lseg->pls_lc_list);
3120 pnfs_put_lseg(lseg);
3121 }
3122
Trond Myklebust71244d92014-01-13 13:34:36 -05003123 pnfs_clear_layoutcommitting(inode);
Trond Myklebusta073dbf2013-03-20 12:34:32 -04003124}
3125
Peng Tao1b0ae062011-09-22 21:50:12 -04003126void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
3127{
Trond Myklebustb9e028f2012-09-18 16:41:18 -04003128 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
Peng Tao1b0ae062011-09-22 21:50:12 -04003129}
3130EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
3131
Andy Adamson863a3c62011-03-23 13:27:54 +00003132void
Trond Myklebust67af7612015-03-25 20:40:38 -04003133pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
3134 loff_t end_pos)
Andy Adamson863a3c62011-03-23 13:27:54 +00003135{
Fred Isamancd841602012-04-20 14:47:44 -04003136 struct nfs_inode *nfsi = NFS_I(inode);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04003137 bool mark_as_dirty = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00003138
Fred Isamancd841602012-04-20 14:47:44 -04003139 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00003140 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Trond Myklebust29559b12015-03-25 20:10:20 -04003141 nfsi->layout->plh_lwb = end_pos;
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04003142 mark_as_dirty = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00003143 dprintk("%s: Set layoutcommit for inode %lu ",
Fred Isamancd841602012-04-20 14:47:44 -04003144 __func__, inode->i_ino);
Trond Myklebust29559b12015-03-25 20:10:20 -04003145 } else if (end_pos > nfsi->layout->plh_lwb)
3146 nfsi->layout->plh_lwb = end_pos;
Trond Myklebust67af7612015-03-25 20:40:38 -04003147 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
Peng Taoa9bae562011-07-30 20:52:33 -04003148 /* references matched in nfs4_layoutcommit_release */
Trond Myklebust67af7612015-03-25 20:40:38 -04003149 pnfs_get_lseg(lseg);
Peng Taoa9bae562011-07-30 20:52:33 -04003150 }
Fred Isamancd841602012-04-20 14:47:44 -04003151 spin_unlock(&inode->i_lock);
Peng Taoacff58802011-07-30 20:52:31 -04003152 dprintk("%s: lseg %p end_pos %llu\n",
Trond Myklebust67af7612015-03-25 20:40:38 -04003153 __func__, lseg, nfsi->layout->plh_lwb);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04003154
3155 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3156 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3157 if (mark_as_dirty)
Fred Isamancd841602012-04-20 14:47:44 -04003158 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00003159}
3160EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
3161
Andy Adamsondb29c082011-07-30 20:52:38 -04003162void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
3163{
3164 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
3165
3166 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
3167 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
Trond Myklebusta073dbf2013-03-20 12:34:32 -04003168 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
Andy Adamsondb29c082011-07-30 20:52:38 -04003169}
3170
Andy Adamsonde4b15c2011-03-12 02:58:09 -05003171/*
3172 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3173 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3174 * data to disk to allow the server to recover the data if it crashes.
3175 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3176 * is off, and a COMMIT is sent to a data server, or
3177 * if WRITEs to a data server return NFS_DATA_SYNC.
3178 */
Andy Adamson863a3c62011-03-23 13:27:54 +00003179int
Andy Adamsonef311532011-03-12 02:58:10 -05003180pnfs_layoutcommit_inode(struct inode *inode, bool sync)
Andy Adamson863a3c62011-03-23 13:27:54 +00003181{
Christoph Hellwig5f919c92014-08-21 11:09:25 -05003182 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
Andy Adamson863a3c62011-03-23 13:27:54 +00003183 struct nfs4_layoutcommit_data *data;
3184 struct nfs_inode *nfsi = NFS_I(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00003185 loff_t end_pos;
Trond Myklebust71244d92014-01-13 13:34:36 -05003186 int status;
3187
3188 if (!pnfs_layoutcommit_outstanding(inode))
3189 return 0;
Andy Adamson863a3c62011-03-23 13:27:54 +00003190
3191 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
3192
Trond Myklebust71244d92014-01-13 13:34:36 -05003193 status = -EAGAIN;
3194 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
3195 if (!sync)
3196 goto out;
NeilBrown74316202014-07-07 15:16:04 +10003197 status = wait_on_bit_lock_action(&nfsi->flags,
Trond Myklebust71244d92014-01-13 13:34:36 -05003198 NFS_INO_LAYOUTCOMMITTING,
3199 nfs_wait_bit_killable,
Peter Zijlstraf5d39b02022-08-22 13:18:22 +02003200 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
Trond Myklebust71244d92014-01-13 13:34:36 -05003201 if (status)
3202 goto out;
3203 }
Andy Adamsonde4b15c2011-03-12 02:58:09 -05003204
Trond Myklebust71244d92014-01-13 13:34:36 -05003205 status = -ENOMEM;
Andy Adamson863a3c62011-03-23 13:27:54 +00003206 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
Trond Myklebust63d8a412022-03-21 15:32:09 -04003207 data = kzalloc(sizeof(*data), nfs_io_gfp_mask());
Trond Myklebust71244d92014-01-13 13:34:36 -05003208 if (!data)
3209 goto clear_layoutcommitting;
Andy Adamson863a3c62011-03-23 13:27:54 +00003210
Trond Myklebust71244d92014-01-13 13:34:36 -05003211 status = 0;
3212 spin_lock(&inode->i_lock);
3213 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
3214 goto out_unlock;
Peng Tao92407e72011-10-23 20:21:17 -07003215
Peng Taoa9bae562011-07-30 20:52:33 -04003216 INIT_LIST_HEAD(&data->lseg_list);
Peng Taoa9bae562011-07-30 20:52:33 -04003217 pnfs_list_write_lseg(inode, &data->lseg_list);
Andy Adamson863a3c62011-03-23 13:27:54 +00003218
Peng Taoacff58802011-07-30 20:52:31 -04003219 end_pos = nfsi->layout->plh_lwb;
Andy Adamson863a3c62011-03-23 13:27:54 +00003220
Trond Myklebustf597c532012-03-04 18:13:56 -05003221 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
Trond Myklebust97a728f2020-04-02 15:47:08 -04003222 data->cred = get_cred(nfsi->layout->plh_lc_cred);
Andy Adamson863a3c62011-03-23 13:27:54 +00003223 spin_unlock(&inode->i_lock);
3224
3225 data->args.inode = inode;
Andy Adamson863a3c62011-03-23 13:27:54 +00003226 nfs_fattr_init(&data->fattr);
3227 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3228 data->res.fattr = &data->fattr;
Trond Myklebust2e18d4d2016-06-26 18:54:58 -04003229 if (end_pos != 0)
3230 data->args.lastbytewritten = end_pos - 1;
3231 else
3232 data->args.lastbytewritten = U64_MAX;
Andy Adamson863a3c62011-03-23 13:27:54 +00003233 data->res.server = NFS_SERVER(inode);
3234
Christoph Hellwig5f919c92014-08-21 11:09:25 -05003235 if (ld->prepare_layoutcommit) {
3236 status = ld->prepare_layoutcommit(&data->args);
3237 if (status) {
NeilBrowna52458b2018-12-03 11:30:31 +11003238 put_cred(data->cred);
Christoph Hellwig5f919c92014-08-21 11:09:25 -05003239 spin_lock(&inode->i_lock);
Trond Myklebust29559b12015-03-25 20:10:20 -04003240 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3241 if (end_pos > nfsi->layout->plh_lwb)
Christoph Hellwig5f919c92014-08-21 11:09:25 -05003242 nfsi->layout->plh_lwb = end_pos;
Jeff Layton3471648a2015-07-10 15:58:42 -04003243 goto out_unlock;
Christoph Hellwig5f919c92014-08-21 11:09:25 -05003244 }
3245 }
3246
3247
Andy Adamson863a3c62011-03-23 13:27:54 +00003248 status = nfs4_proc_layoutcommit(data, sync);
3249out:
Peng Tao92407e72011-10-23 20:21:17 -07003250 if (status)
3251 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00003252 dprintk("<-- %s status %d\n", __func__, status);
3253 return status;
Trond Myklebust71244d92014-01-13 13:34:36 -05003254out_unlock:
3255 spin_unlock(&inode->i_lock);
Peng Tao92407e72011-10-23 20:21:17 -07003256 kfree(data);
Trond Myklebust71244d92014-01-13 13:34:36 -05003257clear_layoutcommitting:
3258 pnfs_clear_layoutcommitting(inode);
Peng Tao92407e72011-10-23 20:21:17 -07003259 goto out;
Andy Adamson863a3c62011-03-23 13:27:54 +00003260}
Peng Tao72cff442014-08-07 10:12:38 +08003261EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
Andy Adamson82be4172012-05-23 05:02:35 -04003262
Trond Myklebust5bb89b42015-03-25 14:14:42 -04003263int
3264pnfs_generic_sync(struct inode *inode, bool datasync)
3265{
3266 return pnfs_layoutcommit_inode(inode, true);
3267}
3268EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3269
Andy Adamson82be4172012-05-23 05:02:35 -04003270struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3271{
3272 struct nfs4_threshold *thp;
3273
Trond Myklebust63d8a412022-03-21 15:32:09 -04003274 thp = kzalloc(sizeof(*thp), nfs_io_gfp_mask());
Andy Adamson82be4172012-05-23 05:02:35 -04003275 if (!thp) {
3276 dprintk("%s mdsthreshold allocation failed\n", __func__);
3277 return NULL;
3278 }
3279 return thp;
3280}
Peng Tao87334082015-06-23 19:51:57 +08003281
Peng Tao865a7ec2015-06-25 18:19:32 +08003282#if IS_ENABLED(CONFIG_NFS_V4_2)
Peng Tao87334082015-06-23 19:51:57 +08003283int
Trond Myklebustc8ad8892015-08-05 17:31:58 -04003284pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
Peng Tao87334082015-06-23 19:51:57 +08003285{
3286 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3287 struct nfs_server *server = NFS_SERVER(inode);
Peng Tao1bfe3b22015-06-23 19:52:03 +08003288 struct nfs_inode *nfsi = NFS_I(inode);
Peng Tao87334082015-06-23 19:51:57 +08003289 struct nfs42_layoutstat_data *data;
3290 struct pnfs_layout_hdr *hdr;
3291 int status = 0;
3292
3293 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3294 goto out;
3295
Trond Myklebust6c5a0d82015-06-27 11:45:46 -04003296 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3297 goto out;
3298
Peng Tao1bfe3b22015-06-23 19:52:03 +08003299 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3300 goto out;
3301
Peng Tao87334082015-06-23 19:51:57 +08003302 spin_lock(&inode->i_lock);
3303 if (!NFS_I(inode)->layout) {
3304 spin_unlock(&inode->i_lock);
Trond Myklebustf538d0b2016-05-16 14:41:14 -04003305 goto out_clear_layoutstats;
Peng Tao87334082015-06-23 19:51:57 +08003306 }
3307 hdr = NFS_I(inode)->layout;
3308 pnfs_get_layout_hdr(hdr);
3309 spin_unlock(&inode->i_lock);
3310
Trond Myklebustc8ad8892015-08-05 17:31:58 -04003311 data = kzalloc(sizeof(*data), gfp_flags);
Peng Tao87334082015-06-23 19:51:57 +08003312 if (!data) {
3313 status = -ENOMEM;
3314 goto out_put;
3315 }
3316
3317 data->args.fh = NFS_FH(inode);
3318 data->args.inode = inode;
Peng Tao87334082015-06-23 19:51:57 +08003319 status = ld->prepare_layoutstats(&data->args);
3320 if (status)
3321 goto out_free;
3322
3323 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3324
3325out:
3326 dprintk("%s returns %d\n", __func__, status);
3327 return status;
3328
3329out_free:
3330 kfree(data);
3331out_put:
3332 pnfs_put_layout_hdr(hdr);
Trond Myklebustf538d0b2016-05-16 14:41:14 -04003333out_clear_layoutstats:
Peng Tao1bfe3b22015-06-23 19:52:03 +08003334 smp_mb__before_atomic();
3335 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3336 smp_mb__after_atomic();
Peng Tao87334082015-06-23 19:51:57 +08003337 goto out;
3338}
3339EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
Peng Tao865a7ec2015-06-25 18:19:32 +08003340#endif
Trond Myklebustbbf58bf2015-08-24 20:39:18 -04003341
3342unsigned int layoutstats_timer;
3343module_param(layoutstats_timer, uint, 0644);
3344EXPORT_SYMBOL_GPL(layoutstats_timer);