blob: 17149a4900653af5326dfd6e0490e6e439b1426f [file] [log] [blame]
Ricardo Labiaga85e174b2010-10-20 00:17:58 -04001/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
Trond Myklebust493292d2011-07-13 15:58:28 -040031#include <linux/nfs_page.h>
Paul Gortmaker143cb492011-07-01 14:23:34 -040032#include <linux/module.h>
Andy Adamson974cec82010-10-20 00:18:02 -040033#include "internal.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040034#include "pnfs.h"
Andy Adamson64419a92011-03-01 01:34:16 +000035#include "iostat.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040036
37#define NFSDBG_FACILITY NFSDBG_PNFS
38
Fred Isaman02c35fc2010-10-20 00:17:59 -040039/* Locking:
40 *
41 * pnfs_spinlock:
42 * protects pnfs_modules_tbl.
43 */
44static DEFINE_SPINLOCK(pnfs_spinlock);
45
46/*
47 * pnfs_modules_tbl holds all pnfs modules
48 */
49static LIST_HEAD(pnfs_modules_tbl);
50
51/* Return the registered pnfs layout driver module matching given id */
52static struct pnfs_layoutdriver_type *
53find_pnfs_driver_locked(u32 id)
54{
55 struct pnfs_layoutdriver_type *local;
56
57 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
58 if (local->id == id)
59 goto out;
60 local = NULL;
61out:
62 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
63 return local;
64}
65
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040066static struct pnfs_layoutdriver_type *
67find_pnfs_driver(u32 id)
68{
Fred Isaman02c35fc2010-10-20 00:17:59 -040069 struct pnfs_layoutdriver_type *local;
70
71 spin_lock(&pnfs_spinlock);
72 local = find_pnfs_driver_locked(id);
73 spin_unlock(&pnfs_spinlock);
74 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040075}
76
77void
78unset_pnfs_layoutdriver(struct nfs_server *nfss)
79{
Benny Halevy738fd0f32011-07-30 20:52:36 -040080 if (nfss->pnfs_curr_ld) {
81 if (nfss->pnfs_curr_ld->clear_layoutdriver)
82 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
Fred Isaman02c35fc2010-10-20 00:17:59 -040083 module_put(nfss->pnfs_curr_ld->owner);
Benny Halevy738fd0f32011-07-30 20:52:36 -040084 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040085 nfss->pnfs_curr_ld = NULL;
86}
87
88/*
89 * Try to set the server's pnfs module to the pnfs layout type specified by id.
90 * Currently only one pNFS layout driver per filesystem is supported.
91 *
92 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
93 */
94void
Benny Halevy738fd0f32011-07-30 20:52:36 -040095set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
96 u32 id)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040097{
98 struct pnfs_layoutdriver_type *ld_type = NULL;
99
100 if (id == 0)
101 goto out_no_driver;
102 if (!(server->nfs_client->cl_exchange_flags &
103 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
104 printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__,
105 id, server->nfs_client->cl_exchange_flags);
106 goto out_no_driver;
107 }
108 ld_type = find_pnfs_driver(id);
109 if (!ld_type) {
110 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
111 ld_type = find_pnfs_driver(id);
112 if (!ld_type) {
113 dprintk("%s: No pNFS module found for %u.\n",
114 __func__, id);
115 goto out_no_driver;
116 }
117 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400118 if (!try_module_get(ld_type->owner)) {
119 dprintk("%s: Could not grab reference on module\n", __func__);
120 goto out_no_driver;
121 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400122 server->pnfs_curr_ld = ld_type;
Benny Halevy738fd0f32011-07-30 20:52:36 -0400123 if (ld_type->set_layoutdriver
124 && ld_type->set_layoutdriver(server, mntfh)) {
125 printk(KERN_ERR "%s: Error initializing pNFS layout driver %u.\n",
126 __func__, id);
127 module_put(ld_type->owner);
128 goto out_no_driver;
129 }
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000130
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400131 dprintk("%s: pNFS module for %u set\n", __func__, id);
132 return;
133
134out_no_driver:
135 dprintk("%s: Using NFSv4 I/O\n", __func__);
136 server->pnfs_curr_ld = NULL;
137}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400138
139int
140pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
141{
142 int status = -EINVAL;
143 struct pnfs_layoutdriver_type *tmp;
144
145 if (ld_type->id == 0) {
146 printk(KERN_ERR "%s id 0 is reserved\n", __func__);
147 return status;
148 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400149 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
150 printk(KERN_ERR "%s Layout driver must provide "
151 "alloc_lseg and free_lseg.\n", __func__);
152 return status;
153 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400154
155 spin_lock(&pnfs_spinlock);
156 tmp = find_pnfs_driver_locked(ld_type->id);
157 if (!tmp) {
158 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
159 status = 0;
160 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
161 ld_type->name);
162 } else {
163 printk(KERN_ERR "%s Module with id %d already loaded!\n",
164 __func__, ld_type->id);
165 }
166 spin_unlock(&pnfs_spinlock);
167
168 return status;
169}
170EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
171
172void
173pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
174{
175 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
176 spin_lock(&pnfs_spinlock);
177 list_del(&ld_type->pnfs_tblid);
178 spin_unlock(&pnfs_spinlock);
179}
180EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400181
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400182/*
183 * pNFS client layout cache
184 */
185
Fred Isamancc6e5342011-01-06 11:36:28 +0000186/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000187void
Fred Isamancc6e5342011-01-06 11:36:28 +0000188get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400189{
Fred Isamancc6e5342011-01-06 11:36:28 +0000190 atomic_inc(&lo->plh_refcount);
191}
192
Benny Halevy636fb9c2011-05-22 19:51:33 +0300193static struct pnfs_layout_hdr *
194pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
195{
196 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
197 return ld->alloc_layout_hdr ? ld->alloc_layout_hdr(ino, gfp_flags) :
198 kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags);
199}
200
201static void
202pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
203{
204 struct pnfs_layoutdriver_type *ld = NFS_SERVER(lo->plh_inode)->pnfs_curr_ld;
Peng Tao9fa40752011-07-30 20:52:32 -0400205 put_rpccred(lo->plh_lc_cred);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300206 return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo);
207}
208
Fred Isamancc6e5342011-01-06 11:36:28 +0000209static void
210destroy_layout_hdr(struct pnfs_layout_hdr *lo)
211{
212 dprintk("%s: freeing layout cache %p\n", __func__, lo);
213 BUG_ON(!list_empty(&lo->plh_layouts));
214 NFS_I(lo->plh_inode)->layout = NULL;
Benny Halevy636fb9c2011-05-22 19:51:33 +0300215 pnfs_free_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400216}
217
218static void
219put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
220{
Fred Isamancc6e5342011-01-06 11:36:28 +0000221 if (atomic_dec_and_test(&lo->plh_refcount))
222 destroy_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400223}
224
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400225void
Fred Isamancc6e5342011-01-06 11:36:28 +0000226put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400227{
Fred Isamancc6e5342011-01-06 11:36:28 +0000228 struct inode *inode = lo->plh_inode;
229
230 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
231 destroy_layout_hdr(lo);
232 spin_unlock(&inode->i_lock);
233 }
Andy Adamson974cec82010-10-20 00:18:02 -0400234}
235
236static void
237init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
238{
Fred Isaman566052c2011-01-06 11:36:20 +0000239 INIT_LIST_HEAD(&lseg->pls_list);
Peng Taoa9bae562011-07-30 20:52:33 -0400240 INIT_LIST_HEAD(&lseg->pls_lc_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000241 atomic_set(&lseg->pls_refcount, 1);
242 smp_mb();
243 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000244 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400245}
246
Fred Isaman4541d162011-01-06 11:36:23 +0000247static void free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400248{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000249 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400250
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400251 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Fred Isaman52fabd72011-01-06 11:36:18 +0000252 /* Matched by get_layout_hdr in pnfs_insert_layout */
Fred Isamancc6e5342011-01-06 11:36:28 +0000253 put_layout_hdr(NFS_I(ino)->layout);
Andy Adamson974cec82010-10-20 00:18:02 -0400254}
255
Fred Isamand684d2a2011-03-01 01:34:13 +0000256static void
257put_lseg_common(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400258{
Fred Isamand684d2a2011-03-01 01:34:13 +0000259 struct inode *inode = lseg->pls_layout->plh_inode;
260
Benny Halevyd20581a2011-05-22 19:52:03 +0300261 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000262 list_del_init(&lseg->pls_list);
263 if (list_empty(&lseg->pls_layout->plh_segs)) {
264 set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
265 /* Matched by initial refcount set in alloc_init_layout_hdr */
266 put_layout_hdr_locked(lseg->pls_layout);
267 }
268 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
269}
270
Fred Isamanbae724e2011-03-01 01:34:15 +0000271void
Fred Isamand684d2a2011-03-01 01:34:13 +0000272put_lseg(struct pnfs_layout_segment *lseg)
273{
274 struct inode *inode;
275
276 if (!lseg)
277 return;
278
Fred Isaman4541d162011-01-06 11:36:23 +0000279 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
280 atomic_read(&lseg->pls_refcount),
281 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000282 inode = lseg->pls_layout->plh_inode;
283 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
284 LIST_HEAD(free_me);
Andy Adamson974cec82010-10-20 00:18:02 -0400285
Fred Isamand684d2a2011-03-01 01:34:13 +0000286 put_lseg_common(lseg);
287 list_add(&lseg->pls_list, &free_me);
288 spin_unlock(&inode->i_lock);
289 pnfs_free_lseg_list(&free_me);
Fred Isaman4541d162011-01-06 11:36:23 +0000290 }
Andy Adamson974cec82010-10-20 00:18:02 -0400291}
Fred Isamane0c2b382011-03-23 13:27:53 +0000292EXPORT_SYMBOL_GPL(put_lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400293
Benny Halevyfb3296e2011-05-22 19:47:26 +0300294static inline u64
295end_offset(u64 start, u64 len)
Fred Isaman4541d162011-01-06 11:36:23 +0000296{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300297 u64 end;
298
299 end = start + len;
300 return end >= start ? end : NFS4_MAX_UINT64;
301}
302
303/* last octet in a range */
304static inline u64
305last_byte_offset(u64 start, u64 len)
306{
307 u64 end;
308
309 BUG_ON(!len);
310 end = start + len;
311 return end > start ? end - 1 : NFS4_MAX_UINT64;
312}
313
314/*
315 * is l2 fully contained in l1?
316 * start1 end1
317 * [----------------------------------)
318 * start2 end2
319 * [----------------)
320 */
321static inline int
322lo_seg_contained(struct pnfs_layout_range *l1,
323 struct pnfs_layout_range *l2)
324{
325 u64 start1 = l1->offset;
326 u64 end1 = end_offset(start1, l1->length);
327 u64 start2 = l2->offset;
328 u64 end2 = end_offset(start2, l2->length);
329
330 return (start1 <= start2) && (end1 >= end2);
331}
332
333/*
334 * is l1 and l2 intersecting?
335 * start1 end1
336 * [----------------------------------)
337 * start2 end2
338 * [----------------)
339 */
340static inline int
341lo_seg_intersecting(struct pnfs_layout_range *l1,
342 struct pnfs_layout_range *l2)
343{
344 u64 start1 = l1->offset;
345 u64 end1 = end_offset(start1, l1->length);
346 u64 start2 = l2->offset;
347 u64 end2 = end_offset(start2, l2->length);
348
349 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
350 (end2 == NFS4_MAX_UINT64 || end2 > start1);
351}
352
Fred Isaman4541d162011-01-06 11:36:23 +0000353static bool
Benny Halevy778b5502011-05-22 19:48:02 +0300354should_free_lseg(struct pnfs_layout_range *lseg_range,
355 struct pnfs_layout_range *recall_range)
Fred Isaman4541d162011-01-06 11:36:23 +0000356{
Benny Halevy778b5502011-05-22 19:48:02 +0300357 return (recall_range->iomode == IOMODE_ANY ||
358 lseg_range->iomode == recall_range->iomode) &&
359 lo_seg_intersecting(lseg_range, recall_range);
Fred Isaman4541d162011-01-06 11:36:23 +0000360}
361
362/* Returns 1 if lseg is removed from list, 0 otherwise */
363static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
364 struct list_head *tmp_list)
365{
366 int rv = 0;
367
368 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
369 /* Remove the reference keeping the lseg in the
370 * list. It will now be removed when all
371 * outstanding io is finished.
372 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000373 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
374 atomic_read(&lseg->pls_refcount));
375 if (atomic_dec_and_test(&lseg->pls_refcount)) {
376 put_lseg_common(lseg);
377 list_add(&lseg->pls_list, tmp_list);
378 rv = 1;
379 }
Fred Isaman4541d162011-01-06 11:36:23 +0000380 }
381 return rv;
382}
383
384/* Returns count of number of matching invalid lsegs remaining in list
385 * after call.
386 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000387int
Fred Isaman4541d162011-01-06 11:36:23 +0000388mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
389 struct list_head *tmp_list,
Benny Halevy778b5502011-05-22 19:48:02 +0300390 struct pnfs_layout_range *recall_range)
Andy Adamson974cec82010-10-20 00:18:02 -0400391{
392 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000393 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400394
395 dprintk("%s:Begin lo %p\n", __func__, lo);
396
Fred Isaman38511722011-02-03 18:28:50 +0000397 if (list_empty(&lo->plh_segs)) {
398 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
399 put_layout_hdr_locked(lo);
400 return 0;
401 }
Fred Isaman4541d162011-01-06 11:36:23 +0000402 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
Benny Halevy778b5502011-05-22 19:48:02 +0300403 if (!recall_range ||
404 should_free_lseg(&lseg->pls_range, recall_range)) {
Fred Isaman4541d162011-01-06 11:36:23 +0000405 dprintk("%s: freeing lseg %p iomode %d "
406 "offset %llu length %llu\n", __func__,
407 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
408 lseg->pls_range.length);
409 invalid++;
410 removed += mark_lseg_invalid(lseg, tmp_list);
411 }
412 dprintk("%s:Return %i\n", __func__, invalid - removed);
413 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400414}
415
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000416/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000417void
Fred Isaman4541d162011-01-06 11:36:23 +0000418pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400419{
Fred Isaman4541d162011-01-06 11:36:23 +0000420 struct pnfs_layout_segment *lseg, *tmp;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000421 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400422
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000423 if (list_empty(free_me))
424 return;
425
426 lo = list_first_entry(free_me, struct pnfs_layout_segment,
427 pls_list)->pls_layout;
428
429 if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
430 struct nfs_client *clp;
431
432 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
433 spin_lock(&clp->cl_lock);
434 list_del_init(&lo->plh_layouts);
435 spin_unlock(&clp->cl_lock);
436 }
Fred Isaman4541d162011-01-06 11:36:23 +0000437 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000438 list_del(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000439 free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400440 }
441}
442
Benny Halevye5e94012010-10-20 00:18:01 -0400443void
444pnfs_destroy_layout(struct nfs_inode *nfsi)
445{
446 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400447 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400448
449 spin_lock(&nfsi->vfs_inode.i_lock);
450 lo = nfsi->layout;
451 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000452 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Benny Halevy778b5502011-05-22 19:48:02 +0300453 mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
Benny Halevye5e94012010-10-20 00:18:01 -0400454 }
455 spin_unlock(&nfsi->vfs_inode.i_lock);
Andy Adamson974cec82010-10-20 00:18:02 -0400456 pnfs_free_lseg_list(&tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400457}
458
Andy Adamson974cec82010-10-20 00:18:02 -0400459/*
460 * Called by the state manger to remove all layouts established under an
461 * expired lease.
462 */
463void
464pnfs_destroy_all_layouts(struct nfs_client *clp)
465{
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400466 struct nfs_server *server;
Andy Adamson974cec82010-10-20 00:18:02 -0400467 struct pnfs_layout_hdr *lo;
468 LIST_HEAD(tmp_list);
469
Andy Adamsonc47abcf2011-06-15 17:52:40 -0400470 nfs4_deviceid_mark_client_invalid(clp);
471 nfs4_deviceid_purge_client(clp);
472
Andy Adamson974cec82010-10-20 00:18:02 -0400473 spin_lock(&clp->cl_lock);
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400474 rcu_read_lock();
475 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
476 if (!list_empty(&server->layouts))
477 list_splice_init(&server->layouts, &tmp_list);
478 }
479 rcu_read_unlock();
Andy Adamson974cec82010-10-20 00:18:02 -0400480 spin_unlock(&clp->cl_lock);
481
482 while (!list_empty(&tmp_list)) {
483 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000484 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400485 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000486 lo->plh_inode->i_ino);
Andy Adamson2887fe42011-05-11 01:19:58 -0400487 list_del_init(&lo->plh_layouts);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000488 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400489 }
490}
491
Fred Isamanfd6002e2011-01-06 11:36:22 +0000492/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000493void
494pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
495 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400496{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000497 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400498
Fred Isamanfd6002e2011-01-06 11:36:22 +0000499 oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid);
500 newseq = be32_to_cpu(new->stateid.seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000501 if ((int)(newseq - oldseq) > 0) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000502 memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid));
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000503 if (update_barrier) {
504 u32 new_barrier = be32_to_cpu(new->stateid.seqid);
505
506 if ((int)(new_barrier - lo->plh_barrier))
507 lo->plh_barrier = new_barrier;
508 } else {
509 /* Because of wraparound, we want to keep the barrier
510 * "close" to the current seqids. It needs to be
511 * within 2**31 to count as "behind", so if it
512 * gets too near that limit, give us a litle leeway
513 * and bring it to within 2**30.
514 * NOTE - and yes, this is all unsigned arithmetic.
515 */
516 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
517 lo->plh_barrier = newseq - (1 << 30);
518 }
519 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400520}
521
Fred Isamancf7d63f2011-01-06 11:36:25 +0000522/* lget is set to 1 if called from inside send_layoutget call chain */
523static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000524pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
525 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000526{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000527 if ((stateid) &&
528 (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0)
529 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000530 return lo->plh_block_lgets ||
Fred Isaman38511722011-02-03 18:28:50 +0000531 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
Fred Isamanf7e89172011-01-06 11:36:32 +0000532 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000533 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000534 (atomic_read(&lo->plh_outstanding) > lget));
535}
536
Fred Isamanfd6002e2011-01-06 11:36:22 +0000537int
538pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
539 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400540{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000541 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400542
543 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000544 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000545 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000546 status = -EAGAIN;
547 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000548 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400549
Fred Isamanfd6002e2011-01-06 11:36:22 +0000550 do {
551 seq = read_seqbegin(&open_state->seqlock);
552 memcpy(dst->data, open_state->stateid.data,
553 sizeof(open_state->stateid.data));
554 } while (read_seqretry(&open_state->seqlock, seq));
555 } else
556 memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data));
557 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400558 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000559 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400560}
561
562/*
563* Get layout from server.
564* for now, assume that whole file layouts are requested.
565* arg->offset: 0
566* arg->length: all ones
567*/
Benny Halevye5e94012010-10-20 00:18:01 -0400568static struct pnfs_layout_segment *
569send_layoutget(struct pnfs_layout_hdr *lo,
570 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +0300571 struct pnfs_layout_range *range,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400572 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400573{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000574 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400575 struct nfs_server *server = NFS_SERVER(ino);
576 struct nfs4_layoutget *lgp;
577 struct pnfs_layout_segment *lseg = NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400578 struct page **pages = NULL;
579 int i;
580 u32 max_resp_sz, max_pages;
Benny Halevye5e94012010-10-20 00:18:01 -0400581
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400582 dprintk("--> %s\n", __func__);
583
584 BUG_ON(ctx == NULL);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400585 lgp = kzalloc(sizeof(*lgp), gfp_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000586 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400587 return NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400588
589 /* allocate pages for xdr post processing */
590 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
591 max_pages = max_resp_sz >> PAGE_SHIFT;
592
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400593 pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400594 if (!pages)
595 goto out_err_free;
596
597 for (i = 0; i < max_pages; i++) {
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400598 pages[i] = alloc_page(gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400599 if (!pages[i])
600 goto out_err_free;
601 }
602
Benny Halevyfb3296e2011-05-22 19:47:26 +0300603 lgp->args.minlength = PAGE_CACHE_SIZE;
604 if (lgp->args.minlength > range->length)
605 lgp->args.minlength = range->length;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400606 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
Benny Halevyfb3296e2011-05-22 19:47:26 +0300607 lgp->args.range = *range;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400608 lgp->args.type = server->pnfs_curr_ld->id;
609 lgp->args.inode = ino;
610 lgp->args.ctx = get_nfs_open_context(ctx);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400611 lgp->args.layout.pages = pages;
612 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400613 lgp->lsegpp = &lseg;
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400614 lgp->gfp_flags = gfp_flags;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400615
616 /* Synchronously retrieve layout information from server and
617 * store in lseg.
618 */
619 nfs4_proc_layoutget(lgp);
620 if (!lseg) {
621 /* remember that LAYOUTGET failed and suspend trying */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300622 set_bit(lo_fail_bit(range->iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400623 }
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400624
625 /* free xdr pages */
626 for (i = 0; i < max_pages; i++)
627 __free_page(pages[i]);
628 kfree(pages);
629
Andy Adamson974cec82010-10-20 00:18:02 -0400630 return lseg;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400631
632out_err_free:
633 /* free any allocated xdr pages, lgp as it's not used */
634 if (pages) {
635 for (i = 0; i < max_pages; i++) {
636 if (!pages[i])
637 break;
638 __free_page(pages[i]);
639 }
640 kfree(pages);
641 }
642 kfree(lgp);
643 return NULL;
Andy Adamson974cec82010-10-20 00:18:02 -0400644}
645
Benny Halevycbe82602011-05-22 19:52:37 +0300646/* Initiates a LAYOUTRETURN(FILE) */
647int
648_pnfs_return_layout(struct inode *ino)
649{
650 struct pnfs_layout_hdr *lo = NULL;
651 struct nfs_inode *nfsi = NFS_I(ino);
652 LIST_HEAD(tmp_list);
653 struct nfs4_layoutreturn *lrp;
654 nfs4_stateid stateid;
655 int status = 0;
656
657 dprintk("--> %s\n", __func__);
658
659 spin_lock(&ino->i_lock);
660 lo = nfsi->layout;
Fred Isamana2e1d4f2011-06-13 18:54:53 -0400661 if (!lo) {
Benny Halevycbe82602011-05-22 19:52:37 +0300662 spin_unlock(&ino->i_lock);
Fred Isamana2e1d4f2011-06-13 18:54:53 -0400663 dprintk("%s: no layout to return\n", __func__);
664 return status;
Benny Halevycbe82602011-05-22 19:52:37 +0300665 }
666 stateid = nfsi->layout->plh_stateid;
667 /* Reference matched in nfs4_layoutreturn_release */
668 get_layout_hdr(lo);
Fred Isamanea0ded72011-06-15 12:31:02 -0400669 mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
670 lo->plh_block_lgets++;
Benny Halevycbe82602011-05-22 19:52:37 +0300671 spin_unlock(&ino->i_lock);
672 pnfs_free_lseg_list(&tmp_list);
673
674 WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags));
675
676 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
677 if (unlikely(lrp == NULL)) {
678 status = -ENOMEM;
Fred Isaman9e2dfdb2011-06-15 14:32:02 -0400679 set_bit(NFS_LAYOUT_RW_FAILED, &lo->plh_flags);
680 set_bit(NFS_LAYOUT_RO_FAILED, &lo->plh_flags);
Benny Halevy1ed3a852011-06-15 11:39:57 -0400681 put_layout_hdr(lo);
Benny Halevycbe82602011-05-22 19:52:37 +0300682 goto out;
683 }
684
685 lrp->args.stateid = stateid;
686 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
687 lrp->args.inode = ino;
Trond Myklebusta56aaa02011-06-15 11:59:10 -0400688 lrp->args.layout = lo;
Benny Halevycbe82602011-05-22 19:52:37 +0300689 lrp->clp = NFS_SERVER(ino)->nfs_client;
690
691 status = nfs4_proc_layoutreturn(lrp);
692out:
693 dprintk("<-- %s status: %d\n", __func__, status);
694 return status;
695}
696
Fred Isamanf7e89172011-01-06 11:36:32 +0000697bool pnfs_roc(struct inode *ino)
698{
699 struct pnfs_layout_hdr *lo;
700 struct pnfs_layout_segment *lseg, *tmp;
701 LIST_HEAD(tmp_list);
702 bool found = false;
703
704 spin_lock(&ino->i_lock);
705 lo = NFS_I(ino)->layout;
706 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
707 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
708 goto out_nolayout;
709 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
710 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
711 mark_lseg_invalid(lseg, &tmp_list);
712 found = true;
713 }
714 if (!found)
715 goto out_nolayout;
716 lo->plh_block_lgets++;
717 get_layout_hdr(lo); /* matched in pnfs_roc_release */
718 spin_unlock(&ino->i_lock);
719 pnfs_free_lseg_list(&tmp_list);
720 return true;
721
722out_nolayout:
723 spin_unlock(&ino->i_lock);
724 return false;
725}
726
727void pnfs_roc_release(struct inode *ino)
728{
729 struct pnfs_layout_hdr *lo;
730
731 spin_lock(&ino->i_lock);
732 lo = NFS_I(ino)->layout;
733 lo->plh_block_lgets--;
734 put_layout_hdr_locked(lo);
735 spin_unlock(&ino->i_lock);
736}
737
738void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
739{
740 struct pnfs_layout_hdr *lo;
741
742 spin_lock(&ino->i_lock);
743 lo = NFS_I(ino)->layout;
744 if ((int)(barrier - lo->plh_barrier) > 0)
745 lo->plh_barrier = barrier;
746 spin_unlock(&ino->i_lock);
747}
748
749bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
750{
751 struct nfs_inode *nfsi = NFS_I(ino);
752 struct pnfs_layout_segment *lseg;
753 bool found = false;
754
755 spin_lock(&ino->i_lock);
756 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
757 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
758 found = true;
759 break;
760 }
761 if (!found) {
762 struct pnfs_layout_hdr *lo = nfsi->layout;
763 u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid);
764
765 /* Since close does not return a layout stateid for use as
766 * a barrier, we choose the worst-case barrier.
767 */
768 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
769 }
770 spin_unlock(&ino->i_lock);
771 return found;
772}
773
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400774/*
775 * Compare two layout segments for sorting into layout cache.
776 * We want to preferentially return RW over RO layouts, so ensure those
777 * are seen first.
778 */
779static s64
Benny Halevyfb3296e2011-05-22 19:47:26 +0300780cmp_layout(struct pnfs_layout_range *l1,
781 struct pnfs_layout_range *l2)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400782{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300783 s64 d;
784
785 /* high offset > low offset */
786 d = l1->offset - l2->offset;
787 if (d)
788 return d;
789
790 /* short length > long length */
791 d = l2->length - l1->length;
792 if (d)
793 return d;
794
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400795 /* read > read/write */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300796 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400797}
798
Andy Adamson974cec82010-10-20 00:18:02 -0400799static void
800pnfs_insert_layout(struct pnfs_layout_hdr *lo,
801 struct pnfs_layout_segment *lseg)
802{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400803 struct pnfs_layout_segment *lp;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400804
Andy Adamson974cec82010-10-20 00:18:02 -0400805 dprintk("%s:Begin\n", __func__);
806
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000807 assert_spin_locked(&lo->plh_inode->i_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000808 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Benny Halevyfb3296e2011-05-22 19:47:26 +0300809 if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400810 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000811 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400812 dprintk("%s: inserted lseg %p "
813 "iomode %d offset %llu length %llu before "
814 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000815 __func__, lseg, lseg->pls_range.iomode,
816 lseg->pls_range.offset, lseg->pls_range.length,
817 lp, lp->pls_range.iomode, lp->pls_range.offset,
818 lp->pls_range.length);
Benny Halevyfb3296e2011-05-22 19:47:26 +0300819 goto out;
Andy Adamson974cec82010-10-20 00:18:02 -0400820 }
Benny Halevyfb3296e2011-05-22 19:47:26 +0300821 list_add_tail(&lseg->pls_list, &lo->plh_segs);
822 dprintk("%s: inserted lseg %p "
823 "iomode %d offset %llu length %llu at tail\n",
824 __func__, lseg, lseg->pls_range.iomode,
825 lseg->pls_range.offset, lseg->pls_range.length);
826out:
Fred Isamancc6e5342011-01-06 11:36:28 +0000827 get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400828
829 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400830}
831
832static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -0400833alloc_init_layout_hdr(struct inode *ino,
834 struct nfs_open_context *ctx,
835 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400836{
837 struct pnfs_layout_hdr *lo;
838
Benny Halevy636fb9c2011-05-22 19:51:33 +0300839 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400840 if (!lo)
841 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000842 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000843 INIT_LIST_HEAD(&lo->plh_layouts);
844 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000845 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000846 lo->plh_inode = ino;
Peng Tao9fa40752011-07-30 20:52:32 -0400847 lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
Benny Halevye5e94012010-10-20 00:18:01 -0400848 return lo;
849}
850
851static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -0400852pnfs_find_alloc_layout(struct inode *ino,
853 struct nfs_open_context *ctx,
854 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400855{
856 struct nfs_inode *nfsi = NFS_I(ino);
857 struct pnfs_layout_hdr *new = NULL;
858
859 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
860
861 assert_spin_locked(&ino->i_lock);
Fred Isaman4541d162011-01-06 11:36:23 +0000862 if (nfsi->layout) {
863 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
864 return NULL;
865 else
866 return nfsi->layout;
867 }
Benny Halevye5e94012010-10-20 00:18:01 -0400868 spin_unlock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -0400869 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400870 spin_lock(&ino->i_lock);
871
872 if (likely(nfsi->layout == NULL)) /* Won the race? */
873 nfsi->layout = new;
874 else
Benny Halevy636fb9c2011-05-22 19:51:33 +0300875 pnfs_free_layout_hdr(new);
Benny Halevye5e94012010-10-20 00:18:01 -0400876 return nfsi->layout;
877}
878
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400879/*
880 * iomode matching rules:
881 * iomode lseg match
882 * ----- ----- -----
883 * ANY READ true
884 * ANY RW true
885 * RW READ false
886 * RW RW true
887 * READ READ true
888 * READ RW true
889 */
890static int
Benny Halevyfb3296e2011-05-22 19:47:26 +0300891is_matching_lseg(struct pnfs_layout_range *ls_range,
892 struct pnfs_layout_range *range)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400893{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300894 struct pnfs_layout_range range1;
895
896 if ((range->iomode == IOMODE_RW &&
897 ls_range->iomode != IOMODE_RW) ||
898 !lo_seg_intersecting(ls_range, range))
899 return 0;
900
901 /* range1 covers only the first byte in the range */
902 range1 = *range;
903 range1.length = 1;
904 return lo_seg_contained(ls_range, &range1);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400905}
906
907/*
908 * lookup range in layout
909 */
Benny Halevye5e94012010-10-20 00:18:01 -0400910static struct pnfs_layout_segment *
Benny Halevyfb3296e2011-05-22 19:47:26 +0300911pnfs_find_lseg(struct pnfs_layout_hdr *lo,
912 struct pnfs_layout_range *range)
Benny Halevye5e94012010-10-20 00:18:01 -0400913{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400914 struct pnfs_layout_segment *lseg, *ret = NULL;
915
916 dprintk("%s:Begin\n", __func__);
917
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000918 assert_spin_locked(&lo->plh_inode->i_lock);
919 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000920 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
Benny Halevyfb3296e2011-05-22 19:47:26 +0300921 is_matching_lseg(&lseg->pls_range, range)) {
Fred Isamand684d2a2011-03-01 01:34:13 +0000922 ret = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400923 break;
924 }
Benny Halevyd771e3a2011-06-14 16:30:16 -0400925 if (lseg->pls_range.offset > range->offset)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400926 break;
927 }
928
929 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000930 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400931 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400932}
933
934/*
935 * Layout segment is retreived from the server if not cached.
936 * The appropriate layout segment is referenced and returned to the caller.
937 */
Andy Adamson7c24d942011-06-13 18:22:38 -0400938struct pnfs_layout_segment *
Benny Halevye5e94012010-10-20 00:18:01 -0400939pnfs_update_layout(struct inode *ino,
940 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +0300941 loff_t pos,
942 u64 count,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400943 enum pnfs_iomode iomode,
944 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400945{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300946 struct pnfs_layout_range arg = {
947 .iomode = iomode,
948 .offset = pos,
949 .length = count,
950 };
Benny Halevy707ed5f2011-05-22 19:47:46 +0300951 unsigned pg_offset;
Benny Halevye5e94012010-10-20 00:18:01 -0400952 struct nfs_inode *nfsi = NFS_I(ino);
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400953 struct nfs_server *server = NFS_SERVER(ino);
954 struct nfs_client *clp = server->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -0400955 struct pnfs_layout_hdr *lo;
956 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000957 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -0400958
959 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
960 return NULL;
961 spin_lock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -0400962 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400963 if (lo == NULL) {
964 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
965 goto out_unlock;
966 }
967
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000968 /* Do we even need to bother with this? */
969 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
970 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
971 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400972 goto out_unlock;
973 }
974
975 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000976 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400977 goto out_unlock;
978
Andy Adamson568e8c42011-03-01 01:34:22 +0000979 /* Check to see if the layout for the given range already exists */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300980 lseg = pnfs_find_lseg(lo, &arg);
Andy Adamson568e8c42011-03-01 01:34:22 +0000981 if (lseg)
982 goto out_unlock;
983
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000984 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +0000985 goto out_unlock;
986 atomic_inc(&lo->plh_outstanding);
987
Fred Isamancc6e5342011-01-06 11:36:28 +0000988 get_layout_hdr(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000989 if (list_empty(&lo->plh_segs))
990 first = true;
991 spin_unlock(&ino->i_lock);
992 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +0000993 /* The lo must be on the clp list if there is any
994 * chance of a CB_LAYOUTRECALL(FILE) coming in.
995 */
996 spin_lock(&clp->cl_lock);
997 BUG_ON(!list_empty(&lo->plh_layouts));
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400998 list_add_tail(&lo->plh_layouts, &server->layouts);
Fred Isaman2130ff62011-01-06 11:36:26 +0000999 spin_unlock(&clp->cl_lock);
1000 }
Benny Halevye5e94012010-10-20 00:18:01 -04001001
Benny Halevy707ed5f2011-05-22 19:47:46 +03001002 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1003 if (pg_offset) {
1004 arg.offset -= pg_offset;
1005 arg.length += pg_offset;
1006 }
Andy Adamson7c24d942011-06-13 18:22:38 -04001007 if (arg.length != NFS4_MAX_UINT64)
1008 arg.length = PAGE_CACHE_ALIGN(arg.length);
Benny Halevy707ed5f2011-05-22 19:47:46 +03001009
Benny Halevyfb3296e2011-05-22 19:47:26 +03001010 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
Fred Isamanf49f9ba2011-02-03 18:28:52 +00001011 if (!lseg && first) {
1012 spin_lock(&clp->cl_lock);
1013 list_del_init(&lo->plh_layouts);
1014 spin_unlock(&clp->cl_lock);
Fred Isaman2130ff62011-01-06 11:36:26 +00001015 }
Fred Isamancf7d63f2011-01-06 11:36:25 +00001016 atomic_dec(&lo->plh_outstanding);
Fred Isamancc6e5342011-01-06 11:36:28 +00001017 put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -04001018out:
1019 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Andy Adamsonbf9c1382011-03-01 01:34:07 +00001020 nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
Benny Halevye5e94012010-10-20 00:18:01 -04001021 return lseg;
1022out_unlock:
1023 spin_unlock(&ino->i_lock);
1024 goto out;
1025}
Andy Adamson7c24d942011-06-13 18:22:38 -04001026EXPORT_SYMBOL_GPL(pnfs_update_layout);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001027
1028int
1029pnfs_layout_process(struct nfs4_layoutget *lgp)
1030{
1031 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1032 struct nfs4_layoutget_res *res = &lgp->res;
1033 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001034 struct inode *ino = lo->plh_inode;
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001035 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001036 int status = 0;
1037
1038 /* Inject layout blob into I/O device driver */
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001039 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001040 if (!lseg || IS_ERR(lseg)) {
1041 if (!lseg)
1042 status = -ENOMEM;
1043 else
1044 status = PTR_ERR(lseg);
1045 dprintk("%s: Could not allocate layout: error %d\n",
1046 __func__, status);
1047 goto out;
1048 }
1049
1050 spin_lock(&ino->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001051 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
1052 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1053 dprintk("%s forget reply due to recall\n", __func__);
1054 goto out_forget_reply;
1055 }
1056
1057 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
1058 dprintk("%s forget reply due to state\n", __func__);
1059 goto out_forget_reply;
1060 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001061 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +00001062 lseg->pls_range = res->range;
Fred Isamand684d2a2011-03-01 01:34:13 +00001063 *lgp->lsegpp = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001064 pnfs_insert_layout(lo, lseg);
1065
Fred Isamanf7e89172011-01-06 11:36:32 +00001066 if (res->return_on_close) {
1067 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1068 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
1069 }
1070
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001071 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001072 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001073 spin_unlock(&ino->i_lock);
1074out:
1075 return status;
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001076
1077out_forget_reply:
1078 spin_unlock(&ino->i_lock);
1079 lseg->pls_layout = lo;
1080 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1081 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001082}
1083
Trond Myklebustd8007d42011-06-10 13:30:23 -04001084void
1085pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1086{
1087 BUG_ON(pgio->pg_lseg != NULL);
1088
1089 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1090 req->wb_context,
1091 req_offset(req),
1092 req->wb_bytes,
1093 IOMODE_READ,
1094 GFP_KERNEL);
Trond Myklebuste885de12011-06-10 13:30:23 -04001095 /* If no lseg, fall back to read through mds */
1096 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001097 nfs_pageio_reset_read_mds(pgio);
Trond Myklebuste885de12011-06-10 13:30:23 -04001098
Trond Myklebustd8007d42011-06-10 13:30:23 -04001099}
1100EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1101
1102void
1103pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1104{
1105 BUG_ON(pgio->pg_lseg != NULL);
1106
1107 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1108 req->wb_context,
1109 req_offset(req),
1110 req->wb_bytes,
1111 IOMODE_RW,
1112 GFP_NOFS);
Trond Myklebuste885de12011-06-10 13:30:23 -04001113 /* If no lseg, fall back to write through mds */
1114 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001115 nfs_pageio_reset_write_mds(pgio);
Trond Myklebustd8007d42011-06-10 13:30:23 -04001116}
1117EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1118
Benny Halevy18ad0a92011-05-25 21:03:56 +03001119bool
Trond Myklebust1751c362011-06-10 13:30:23 -04001120pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode)
1121{
1122 struct nfs_server *server = NFS_SERVER(inode);
1123 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1124
1125 if (ld == NULL)
1126 return false;
1127 nfs_pageio_init(pgio, inode, ld->pg_read_ops, server->rsize, 0);
1128 return true;
1129}
1130
1131bool
1132pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode, int ioflags)
1133{
1134 struct nfs_server *server = NFS_SERVER(inode);
1135 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1136
1137 if (ld == NULL)
1138 return false;
1139 nfs_pageio_init(pgio, inode, ld->pg_write_ops, server->wsize, ioflags);
1140 return true;
1141}
1142
1143bool
Benny Halevydfed2062011-05-25 20:25:22 +03001144pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1145 struct nfs_page *req)
Fred Isamanbae724e2011-03-01 01:34:15 +00001146{
Trond Myklebustd8007d42011-06-10 13:30:23 -04001147 if (pgio->pg_lseg == NULL)
1148 return nfs_generic_pg_test(pgio, prev, req);
Benny Halevy89a58e32011-05-25 20:54:40 +03001149
Trond Myklebust19982ba2011-06-10 13:30:23 -04001150 /*
1151 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1152 * Note that this test makes several assumptions:
1153 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1154 * is known to lie within the range.
1155 * - that the nfs_page being tested is known to be contiguous with the
1156 * previous nfs_page.
1157 * - Layout ranges are page aligned, so we only have to test the
1158 * start offset of the request.
1159 *
1160 * Please also note that 'end_offset' is actually the offset of the
1161 * first byte that lies outside the pnfs_layout_range. FIXME?
1162 *
1163 */
1164 return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
1165 pgio->pg_lseg->pls_range.length);
Fred Isamanbae724e2011-03-01 01:34:15 +00001166}
Benny Halevy89a58e32011-05-25 20:54:40 +03001167EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
Fred Isamanbae724e2011-03-01 01:34:15 +00001168
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001169static int pnfs_write_done_resend_to_mds(struct inode *inode, struct list_head *head)
1170{
1171 struct nfs_pageio_descriptor pgio;
1172 LIST_HEAD(failed);
1173
1174 /* Resend all requests through the MDS */
1175 nfs_pageio_init_write_mds(&pgio, inode, FLUSH_STABLE);
1176 while (!list_empty(head)) {
1177 struct nfs_page *req = nfs_list_entry(head->next);
1178
1179 nfs_list_remove_request(req);
1180 if (!nfs_pageio_add_request(&pgio, req))
1181 nfs_list_add_request(req, &failed);
1182 }
1183 nfs_pageio_complete(&pgio);
1184
1185 if (!list_empty(&failed)) {
1186 /* For some reason our attempt to resend pages. Mark the
1187 * overall send request as having failed, and let
1188 * nfs_writeback_release_full deal with the error.
1189 */
1190 list_move(&failed, head);
1191 return -EIO;
1192 }
1193 return 0;
1194}
1195
Benny Halevyd20581a2011-05-22 19:52:03 +03001196/*
1197 * Called by non rpc-based layout drivers
1198 */
Peng Tao8ce160c2011-09-22 21:50:14 -04001199void pnfs_ld_write_done(struct nfs_write_data *data)
Fred Isaman94ad1c82011-03-01 01:34:14 +00001200{
Peng Tao8ce160c2011-09-22 21:50:14 -04001201 if (likely(!data->pnfs_error)) {
Benny Halevyd20581a2011-05-22 19:52:03 +03001202 pnfs_set_layoutcommit(data);
1203 data->mds_ops->rpc_call_done(&data->task, data);
Peng Tao8ce160c2011-09-22 21:50:14 -04001204 } else {
Peng Tao8ce160c2011-09-22 21:50:14 -04001205 dprintk("pnfs write error = %d\n", data->pnfs_error);
Boaz Harroshfe0fe832012-01-06 09:31:20 +02001206 if (NFS_SERVER(data->inode)->pnfs_curr_ld->flags &
1207 PNFS_LAYOUTRET_ON_ERROR) {
1208 /* Don't lo_commit on error, Server will needs to
1209 * preform a file recovery.
1210 */
1211 clear_bit(NFS_INO_LAYOUTCOMMIT,
1212 &NFS_I(data->inode)->flags);
1213 pnfs_return_layout(data->inode);
1214 }
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001215 data->task.tk_status = pnfs_write_done_resend_to_mds(data->inode, &data->pages);
Fred Isaman44b83792011-03-03 15:13:44 +00001216 }
Peng Tao8ce160c2011-09-22 21:50:14 -04001217 data->mds_ops->rpc_release(data);
Fred Isaman44b83792011-03-03 15:13:44 +00001218}
Benny Halevyd20581a2011-05-22 19:52:03 +03001219EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
Fred Isaman44b83792011-03-03 15:13:44 +00001220
Trond Myklebustdce81292011-07-13 15:59:19 -04001221static void
1222pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1223 struct nfs_write_data *data)
1224{
1225 list_splice_tail_init(&data->pages, &desc->pg_list);
1226 if (data->req && list_empty(&data->req->wb_list))
1227 nfs_list_add_request(data->req, &desc->pg_list);
1228 nfs_pageio_reset_write_mds(desc);
1229 desc->pg_recoalesce = 1;
1230 nfs_writedata_release(data);
1231}
1232
1233static enum pnfs_try_status
Andy Adamson0382b742011-03-03 15:13:45 +00001234pnfs_try_to_write_data(struct nfs_write_data *wdata,
Trond Myklebustdce81292011-07-13 15:59:19 -04001235 const struct rpc_call_ops *call_ops,
1236 struct pnfs_layout_segment *lseg,
1237 int how)
Andy Adamson0382b742011-03-03 15:13:45 +00001238{
1239 struct inode *inode = wdata->inode;
1240 enum pnfs_try_status trypnfs;
1241 struct nfs_server *nfss = NFS_SERVER(inode);
1242
1243 wdata->mds_ops = call_ops;
Trond Myklebustdce81292011-07-13 15:59:19 -04001244 wdata->lseg = get_lseg(lseg);
Andy Adamson0382b742011-03-03 15:13:45 +00001245
1246 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1247 inode->i_ino, wdata->args.count, wdata->args.offset, how);
1248
1249 trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
1250 if (trypnfs == PNFS_NOT_ATTEMPTED) {
1251 put_lseg(wdata->lseg);
1252 wdata->lseg = NULL;
1253 } else
1254 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
1255
1256 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1257 return trypnfs;
1258}
1259
Trond Myklebustdce81292011-07-13 15:59:19 -04001260static void
1261pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
1262{
1263 struct nfs_write_data *data;
1264 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1265 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1266
1267 desc->pg_lseg = NULL;
1268 while (!list_empty(head)) {
1269 enum pnfs_try_status trypnfs;
1270
1271 data = list_entry(head->next, struct nfs_write_data, list);
1272 list_del_init(&data->list);
1273
1274 trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
1275 if (trypnfs == PNFS_NOT_ATTEMPTED)
1276 pnfs_write_through_mds(desc, data);
1277 }
1278 put_lseg(lseg);
1279}
1280
1281int
1282pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1283{
1284 LIST_HEAD(head);
1285 int ret;
1286
1287 ret = nfs_generic_flush(desc, &head);
1288 if (ret != 0) {
1289 put_lseg(desc->pg_lseg);
1290 desc->pg_lseg = NULL;
1291 return ret;
1292 }
1293 pnfs_do_multiple_writes(desc, &head, desc->pg_ioflags);
1294 return 0;
1295}
1296EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
1297
Trond Myklebust62e4a762011-11-10 14:30:37 -05001298static void pnfs_ld_handle_read_error(struct nfs_read_data *data)
1299{
1300 struct nfs_pageio_descriptor pgio;
1301
1302 put_lseg(data->lseg);
1303 data->lseg = NULL;
1304 dprintk("pnfs write error = %d\n", data->pnfs_error);
Boaz Harroshfe0fe832012-01-06 09:31:20 +02001305 if (NFS_SERVER(data->inode)->pnfs_curr_ld->flags &
1306 PNFS_LAYOUTRET_ON_ERROR)
1307 pnfs_return_layout(data->inode);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001308
1309 nfs_pageio_init_read_mds(&pgio, data->inode);
1310
1311 while (!list_empty(&data->pages)) {
1312 struct nfs_page *req = nfs_list_entry(data->pages.next);
1313
1314 nfs_list_remove_request(req);
1315 nfs_pageio_add_request(&pgio, req);
1316 }
1317 nfs_pageio_complete(&pgio);
1318}
1319
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001320/*
Benny Halevyd20581a2011-05-22 19:52:03 +03001321 * Called by non rpc-based layout drivers
1322 */
Peng Tao9b7eecd2011-09-22 21:50:15 -04001323void pnfs_ld_read_done(struct nfs_read_data *data)
Benny Halevyd20581a2011-05-22 19:52:03 +03001324{
Peng Tao9b7eecd2011-09-22 21:50:15 -04001325 if (likely(!data->pnfs_error)) {
Benny Halevyd20581a2011-05-22 19:52:03 +03001326 __nfs4_read_done_cb(data);
1327 data->mds_ops->rpc_call_done(&data->task, data);
Trond Myklebust62e4a762011-11-10 14:30:37 -05001328 } else
1329 pnfs_ld_handle_read_error(data);
Peng Tao9b7eecd2011-09-22 21:50:15 -04001330 data->mds_ops->rpc_release(data);
Benny Halevyd20581a2011-05-22 19:52:03 +03001331}
1332EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
1333
Trond Myklebust493292d2011-07-13 15:58:28 -04001334static void
1335pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1336 struct nfs_read_data *data)
1337{
1338 list_splice_tail_init(&data->pages, &desc->pg_list);
1339 if (data->req && list_empty(&data->req->wb_list))
1340 nfs_list_add_request(data->req, &desc->pg_list);
1341 nfs_pageio_reset_read_mds(desc);
1342 desc->pg_recoalesce = 1;
1343 nfs_readdata_release(data);
1344}
1345
Benny Halevyd20581a2011-05-22 19:52:03 +03001346/*
Andy Adamson64419a92011-03-01 01:34:16 +00001347 * Call the appropriate parallel I/O subsystem read function.
1348 */
Trond Myklebust493292d2011-07-13 15:58:28 -04001349static enum pnfs_try_status
Andy Adamson64419a92011-03-01 01:34:16 +00001350pnfs_try_to_read_data(struct nfs_read_data *rdata,
Trond Myklebust493292d2011-07-13 15:58:28 -04001351 const struct rpc_call_ops *call_ops,
1352 struct pnfs_layout_segment *lseg)
Andy Adamson64419a92011-03-01 01:34:16 +00001353{
1354 struct inode *inode = rdata->inode;
1355 struct nfs_server *nfss = NFS_SERVER(inode);
1356 enum pnfs_try_status trypnfs;
1357
1358 rdata->mds_ops = call_ops;
Trond Myklebust493292d2011-07-13 15:58:28 -04001359 rdata->lseg = get_lseg(lseg);
Andy Adamson64419a92011-03-01 01:34:16 +00001360
1361 dprintk("%s: Reading ino:%lu %u@%llu\n",
1362 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
1363
1364 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
1365 if (trypnfs == PNFS_NOT_ATTEMPTED) {
1366 put_lseg(rdata->lseg);
1367 rdata->lseg = NULL;
1368 } else {
1369 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
1370 }
1371 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1372 return trypnfs;
1373}
Andy Adamson863a3c62011-03-23 13:27:54 +00001374
Trond Myklebust493292d2011-07-13 15:58:28 -04001375static void
1376pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
1377{
1378 struct nfs_read_data *data;
1379 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1380 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1381
1382 desc->pg_lseg = NULL;
1383 while (!list_empty(head)) {
1384 enum pnfs_try_status trypnfs;
1385
1386 data = list_entry(head->next, struct nfs_read_data, list);
1387 list_del_init(&data->list);
1388
1389 trypnfs = pnfs_try_to_read_data(data, call_ops, lseg);
1390 if (trypnfs == PNFS_NOT_ATTEMPTED)
1391 pnfs_read_through_mds(desc, data);
1392 }
1393 put_lseg(lseg);
1394}
1395
1396int
1397pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
1398{
1399 LIST_HEAD(head);
1400 int ret;
1401
1402 ret = nfs_generic_pagein(desc, &head);
1403 if (ret != 0) {
1404 put_lseg(desc->pg_lseg);
1405 desc->pg_lseg = NULL;
1406 return ret;
1407 }
1408 pnfs_do_multiple_reads(desc, &head);
1409 return 0;
1410}
1411EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
1412
Andy Adamson863a3c62011-03-23 13:27:54 +00001413/*
Peng Taoa9bae562011-07-30 20:52:33 -04001414 * There can be multiple RW segments.
Andy Adamson863a3c62011-03-23 13:27:54 +00001415 */
Peng Taoa9bae562011-07-30 20:52:33 -04001416static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
Andy Adamson863a3c62011-03-23 13:27:54 +00001417{
Peng Taoa9bae562011-07-30 20:52:33 -04001418 struct pnfs_layout_segment *lseg;
Andy Adamson863a3c62011-03-23 13:27:54 +00001419
Peng Taoa9bae562011-07-30 20:52:33 -04001420 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
1421 if (lseg->pls_range.iomode == IOMODE_RW &&
1422 test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1423 list_add(&lseg->pls_lc_list, listp);
1424 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001425}
1426
Peng Tao1b0ae062011-09-22 21:50:12 -04001427void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
1428{
1429 if (lseg->pls_range.iomode == IOMODE_RW) {
1430 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__);
1431 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
1432 } else {
1433 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__);
1434 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
1435 }
1436}
1437EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
1438
Andy Adamson863a3c62011-03-23 13:27:54 +00001439void
1440pnfs_set_layoutcommit(struct nfs_write_data *wdata)
1441{
1442 struct nfs_inode *nfsi = NFS_I(wdata->inode);
Vitaliy Gusev4b8ee2b2011-05-20 01:34:46 +04001443 loff_t end_pos = wdata->mds_offset + wdata->res.count;
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001444 bool mark_as_dirty = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001445
1446 spin_lock(&nfsi->vfs_inode.i_lock);
1447 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001448 mark_as_dirty = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001449 dprintk("%s: Set layoutcommit for inode %lu ",
1450 __func__, wdata->inode->i_ino);
1451 }
Peng Taoa9bae562011-07-30 20:52:33 -04001452 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &wdata->lseg->pls_flags)) {
1453 /* references matched in nfs4_layoutcommit_release */
1454 get_lseg(wdata->lseg);
1455 }
Peng Taoacff58802011-07-30 20:52:31 -04001456 if (end_pos > nfsi->layout->plh_lwb)
1457 nfsi->layout->plh_lwb = end_pos;
Andy Adamson863a3c62011-03-23 13:27:54 +00001458 spin_unlock(&nfsi->vfs_inode.i_lock);
Peng Taoacff58802011-07-30 20:52:31 -04001459 dprintk("%s: lseg %p end_pos %llu\n",
1460 __func__, wdata->lseg, nfsi->layout->plh_lwb);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001461
1462 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1463 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1464 if (mark_as_dirty)
1465 mark_inode_dirty_sync(wdata->inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001466}
1467EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1468
Andy Adamsondb29c082011-07-30 20:52:38 -04001469void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
1470{
1471 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
1472
1473 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
1474 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
1475}
1476
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001477/*
1478 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1479 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1480 * data to disk to allow the server to recover the data if it crashes.
1481 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1482 * is off, and a COMMIT is sent to a data server, or
1483 * if WRITEs to a data server return NFS_DATA_SYNC.
1484 */
Andy Adamson863a3c62011-03-23 13:27:54 +00001485int
Andy Adamsonef311532011-03-12 02:58:10 -05001486pnfs_layoutcommit_inode(struct inode *inode, bool sync)
Andy Adamson863a3c62011-03-23 13:27:54 +00001487{
1488 struct nfs4_layoutcommit_data *data;
1489 struct nfs_inode *nfsi = NFS_I(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001490 loff_t end_pos;
1491 int status = 0;
1492
1493 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1494
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001495 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1496 return 0;
1497
Andy Adamson863a3c62011-03-23 13:27:54 +00001498 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1499 data = kzalloc(sizeof(*data), GFP_NOFS);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001500 if (!data) {
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001501 status = -ENOMEM;
1502 goto out;
1503 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001504
Peng Tao92407e72011-10-23 20:21:17 -07001505 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1506 goto out_free;
1507
1508 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
1509 if (!sync) {
1510 status = -EAGAIN;
1511 goto out_free;
1512 }
1513 status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING,
1514 nfs_wait_bit_killable, TASK_KILLABLE);
1515 if (status)
1516 goto out_free;
1517 }
1518
Peng Taoa9bae562011-07-30 20:52:33 -04001519 INIT_LIST_HEAD(&data->lseg_list);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001520 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001521 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Peng Tao92407e72011-10-23 20:21:17 -07001522 clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
Andy Adamson863a3c62011-03-23 13:27:54 +00001523 spin_unlock(&inode->i_lock);
Peng Tao92407e72011-10-23 20:21:17 -07001524 wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
1525 goto out_free;
Andy Adamson863a3c62011-03-23 13:27:54 +00001526 }
Peng Taoa9bae562011-07-30 20:52:33 -04001527
1528 pnfs_list_write_lseg(inode, &data->lseg_list);
Andy Adamson863a3c62011-03-23 13:27:54 +00001529
Peng Taoacff58802011-07-30 20:52:31 -04001530 end_pos = nfsi->layout->plh_lwb;
Peng Taoacff58802011-07-30 20:52:31 -04001531 nfsi->layout->plh_lwb = 0;
Andy Adamson863a3c62011-03-23 13:27:54 +00001532
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001533 memcpy(&data->args.stateid.data, nfsi->layout->plh_stateid.data,
1534 sizeof(nfsi->layout->plh_stateid.data));
Andy Adamson863a3c62011-03-23 13:27:54 +00001535 spin_unlock(&inode->i_lock);
1536
1537 data->args.inode = inode;
Peng Tao9fa40752011-07-30 20:52:32 -04001538 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
Andy Adamson863a3c62011-03-23 13:27:54 +00001539 nfs_fattr_init(&data->fattr);
1540 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1541 data->res.fattr = &data->fattr;
1542 data->args.lastbytewritten = end_pos - 1;
1543 data->res.server = NFS_SERVER(inode);
1544
1545 status = nfs4_proc_layoutcommit(data, sync);
1546out:
Peng Tao92407e72011-10-23 20:21:17 -07001547 if (status)
1548 mark_inode_dirty_sync(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001549 dprintk("<-- %s status %d\n", __func__, status);
1550 return status;
Peng Tao92407e72011-10-23 20:21:17 -07001551out_free:
1552 kfree(data);
1553 goto out;
Andy Adamson863a3c62011-03-23 13:27:54 +00001554}