blob: e550e8836c3759acb041e035078387a869e1f2c8 [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>
Andy Adamson974cec82010-10-20 00:18:02 -040032#include "internal.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040033#include "pnfs.h"
Andy Adamson64419a92011-03-01 01:34:16 +000034#include "iostat.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040035
36#define NFSDBG_FACILITY NFSDBG_PNFS
37
Fred Isaman02c35fc2010-10-20 00:17:59 -040038/* Locking:
39 *
40 * pnfs_spinlock:
41 * protects pnfs_modules_tbl.
42 */
43static DEFINE_SPINLOCK(pnfs_spinlock);
44
45/*
46 * pnfs_modules_tbl holds all pnfs modules
47 */
48static LIST_HEAD(pnfs_modules_tbl);
49
50/* Return the registered pnfs layout driver module matching given id */
51static struct pnfs_layoutdriver_type *
52find_pnfs_driver_locked(u32 id)
53{
54 struct pnfs_layoutdriver_type *local;
55
56 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
57 if (local->id == id)
58 goto out;
59 local = NULL;
60out:
61 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
62 return local;
63}
64
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040065static struct pnfs_layoutdriver_type *
66find_pnfs_driver(u32 id)
67{
Fred Isaman02c35fc2010-10-20 00:17:59 -040068 struct pnfs_layoutdriver_type *local;
69
70 spin_lock(&pnfs_spinlock);
71 local = find_pnfs_driver_locked(id);
72 spin_unlock(&pnfs_spinlock);
73 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040074}
75
76void
77unset_pnfs_layoutdriver(struct nfs_server *nfss)
78{
Benny Halevy738fd0f32011-07-30 20:52:36 -040079 if (nfss->pnfs_curr_ld) {
80 if (nfss->pnfs_curr_ld->clear_layoutdriver)
81 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
Fred Isaman02c35fc2010-10-20 00:17:59 -040082 module_put(nfss->pnfs_curr_ld->owner);
Benny Halevy738fd0f32011-07-30 20:52:36 -040083 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040084 nfss->pnfs_curr_ld = NULL;
85}
86
87/*
88 * Try to set the server's pnfs module to the pnfs layout type specified by id.
89 * Currently only one pNFS layout driver per filesystem is supported.
90 *
91 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
92 */
93void
Benny Halevy738fd0f32011-07-30 20:52:36 -040094set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
95 u32 id)
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040096{
97 struct pnfs_layoutdriver_type *ld_type = NULL;
98
99 if (id == 0)
100 goto out_no_driver;
101 if (!(server->nfs_client->cl_exchange_flags &
102 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
103 printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__,
104 id, server->nfs_client->cl_exchange_flags);
105 goto out_no_driver;
106 }
107 ld_type = find_pnfs_driver(id);
108 if (!ld_type) {
109 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
110 ld_type = find_pnfs_driver(id);
111 if (!ld_type) {
112 dprintk("%s: No pNFS module found for %u.\n",
113 __func__, id);
114 goto out_no_driver;
115 }
116 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400117 if (!try_module_get(ld_type->owner)) {
118 dprintk("%s: Could not grab reference on module\n", __func__);
119 goto out_no_driver;
120 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400121 server->pnfs_curr_ld = ld_type;
Benny Halevy738fd0f32011-07-30 20:52:36 -0400122 if (ld_type->set_layoutdriver
123 && ld_type->set_layoutdriver(server, mntfh)) {
124 printk(KERN_ERR "%s: Error initializing pNFS layout driver %u.\n",
125 __func__, id);
126 module_put(ld_type->owner);
127 goto out_no_driver;
128 }
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000129
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400130 dprintk("%s: pNFS module for %u set\n", __func__, id);
131 return;
132
133out_no_driver:
134 dprintk("%s: Using NFSv4 I/O\n", __func__);
135 server->pnfs_curr_ld = NULL;
136}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400137
138int
139pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
140{
141 int status = -EINVAL;
142 struct pnfs_layoutdriver_type *tmp;
143
144 if (ld_type->id == 0) {
145 printk(KERN_ERR "%s id 0 is reserved\n", __func__);
146 return status;
147 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400148 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
149 printk(KERN_ERR "%s Layout driver must provide "
150 "alloc_lseg and free_lseg.\n", __func__);
151 return status;
152 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400153
154 spin_lock(&pnfs_spinlock);
155 tmp = find_pnfs_driver_locked(ld_type->id);
156 if (!tmp) {
157 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
158 status = 0;
159 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
160 ld_type->name);
161 } else {
162 printk(KERN_ERR "%s Module with id %d already loaded!\n",
163 __func__, ld_type->id);
164 }
165 spin_unlock(&pnfs_spinlock);
166
167 return status;
168}
169EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
170
171void
172pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
173{
174 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
175 spin_lock(&pnfs_spinlock);
176 list_del(&ld_type->pnfs_tblid);
177 spin_unlock(&pnfs_spinlock);
178}
179EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400180
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400181/*
182 * pNFS client layout cache
183 */
184
Fred Isamancc6e5342011-01-06 11:36:28 +0000185/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000186void
Fred Isamancc6e5342011-01-06 11:36:28 +0000187get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400188{
Fred Isamancc6e5342011-01-06 11:36:28 +0000189 atomic_inc(&lo->plh_refcount);
190}
191
Benny Halevy636fb9c2011-05-22 19:51:33 +0300192static struct pnfs_layout_hdr *
193pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
194{
195 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
196 return ld->alloc_layout_hdr ? ld->alloc_layout_hdr(ino, gfp_flags) :
197 kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags);
198}
199
200static void
201pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
202{
203 struct pnfs_layoutdriver_type *ld = NFS_SERVER(lo->plh_inode)->pnfs_curr_ld;
Peng Tao9fa40752011-07-30 20:52:32 -0400204 put_rpccred(lo->plh_lc_cred);
Benny Halevy636fb9c2011-05-22 19:51:33 +0300205 return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo);
206}
207
Fred Isamancc6e5342011-01-06 11:36:28 +0000208static void
209destroy_layout_hdr(struct pnfs_layout_hdr *lo)
210{
211 dprintk("%s: freeing layout cache %p\n", __func__, lo);
212 BUG_ON(!list_empty(&lo->plh_layouts));
213 NFS_I(lo->plh_inode)->layout = NULL;
Benny Halevy636fb9c2011-05-22 19:51:33 +0300214 pnfs_free_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400215}
216
217static void
218put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
219{
Fred Isamancc6e5342011-01-06 11:36:28 +0000220 if (atomic_dec_and_test(&lo->plh_refcount))
221 destroy_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400222}
223
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400224void
Fred Isamancc6e5342011-01-06 11:36:28 +0000225put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400226{
Fred Isamancc6e5342011-01-06 11:36:28 +0000227 struct inode *inode = lo->plh_inode;
228
229 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
230 destroy_layout_hdr(lo);
231 spin_unlock(&inode->i_lock);
232 }
Andy Adamson974cec82010-10-20 00:18:02 -0400233}
234
235static void
236init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
237{
Fred Isaman566052c2011-01-06 11:36:20 +0000238 INIT_LIST_HEAD(&lseg->pls_list);
Peng Taoa9bae562011-07-30 20:52:33 -0400239 INIT_LIST_HEAD(&lseg->pls_lc_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000240 atomic_set(&lseg->pls_refcount, 1);
241 smp_mb();
242 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000243 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400244}
245
Fred Isaman4541d162011-01-06 11:36:23 +0000246static void free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400247{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000248 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400249
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400250 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Fred Isaman52fabd72011-01-06 11:36:18 +0000251 /* Matched by get_layout_hdr in pnfs_insert_layout */
Fred Isamancc6e5342011-01-06 11:36:28 +0000252 put_layout_hdr(NFS_I(ino)->layout);
Andy Adamson974cec82010-10-20 00:18:02 -0400253}
254
Fred Isamand684d2a2011-03-01 01:34:13 +0000255static void
256put_lseg_common(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400257{
Fred Isamand684d2a2011-03-01 01:34:13 +0000258 struct inode *inode = lseg->pls_layout->plh_inode;
259
Benny Halevyd20581a2011-05-22 19:52:03 +0300260 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000261 list_del_init(&lseg->pls_list);
262 if (list_empty(&lseg->pls_layout->plh_segs)) {
263 set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
264 /* Matched by initial refcount set in alloc_init_layout_hdr */
265 put_layout_hdr_locked(lseg->pls_layout);
266 }
267 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
268}
269
Fred Isamanbae724e2011-03-01 01:34:15 +0000270void
Fred Isamand684d2a2011-03-01 01:34:13 +0000271put_lseg(struct pnfs_layout_segment *lseg)
272{
273 struct inode *inode;
274
275 if (!lseg)
276 return;
277
Fred Isaman4541d162011-01-06 11:36:23 +0000278 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
279 atomic_read(&lseg->pls_refcount),
280 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000281 inode = lseg->pls_layout->plh_inode;
282 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
283 LIST_HEAD(free_me);
Andy Adamson974cec82010-10-20 00:18:02 -0400284
Fred Isamand684d2a2011-03-01 01:34:13 +0000285 put_lseg_common(lseg);
286 list_add(&lseg->pls_list, &free_me);
287 spin_unlock(&inode->i_lock);
288 pnfs_free_lseg_list(&free_me);
Fred Isaman4541d162011-01-06 11:36:23 +0000289 }
Andy Adamson974cec82010-10-20 00:18:02 -0400290}
Fred Isamane0c2b382011-03-23 13:27:53 +0000291EXPORT_SYMBOL_GPL(put_lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400292
Benny Halevyfb3296e2011-05-22 19:47:26 +0300293static inline u64
294end_offset(u64 start, u64 len)
Fred Isaman4541d162011-01-06 11:36:23 +0000295{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300296 u64 end;
297
298 end = start + len;
299 return end >= start ? end : NFS4_MAX_UINT64;
300}
301
302/* last octet in a range */
303static inline u64
304last_byte_offset(u64 start, u64 len)
305{
306 u64 end;
307
308 BUG_ON(!len);
309 end = start + len;
310 return end > start ? end - 1 : NFS4_MAX_UINT64;
311}
312
313/*
314 * is l2 fully contained in l1?
315 * start1 end1
316 * [----------------------------------)
317 * start2 end2
318 * [----------------)
319 */
320static inline int
321lo_seg_contained(struct pnfs_layout_range *l1,
322 struct pnfs_layout_range *l2)
323{
324 u64 start1 = l1->offset;
325 u64 end1 = end_offset(start1, l1->length);
326 u64 start2 = l2->offset;
327 u64 end2 = end_offset(start2, l2->length);
328
329 return (start1 <= start2) && (end1 >= end2);
330}
331
332/*
333 * is l1 and l2 intersecting?
334 * start1 end1
335 * [----------------------------------)
336 * start2 end2
337 * [----------------)
338 */
339static inline int
340lo_seg_intersecting(struct pnfs_layout_range *l1,
341 struct pnfs_layout_range *l2)
342{
343 u64 start1 = l1->offset;
344 u64 end1 = end_offset(start1, l1->length);
345 u64 start2 = l2->offset;
346 u64 end2 = end_offset(start2, l2->length);
347
348 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
349 (end2 == NFS4_MAX_UINT64 || end2 > start1);
350}
351
Fred Isaman4541d162011-01-06 11:36:23 +0000352static bool
Benny Halevy778b5502011-05-22 19:48:02 +0300353should_free_lseg(struct pnfs_layout_range *lseg_range,
354 struct pnfs_layout_range *recall_range)
Fred Isaman4541d162011-01-06 11:36:23 +0000355{
Benny Halevy778b5502011-05-22 19:48:02 +0300356 return (recall_range->iomode == IOMODE_ANY ||
357 lseg_range->iomode == recall_range->iomode) &&
358 lo_seg_intersecting(lseg_range, recall_range);
Fred Isaman4541d162011-01-06 11:36:23 +0000359}
360
361/* Returns 1 if lseg is removed from list, 0 otherwise */
362static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
363 struct list_head *tmp_list)
364{
365 int rv = 0;
366
367 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
368 /* Remove the reference keeping the lseg in the
369 * list. It will now be removed when all
370 * outstanding io is finished.
371 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000372 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
373 atomic_read(&lseg->pls_refcount));
374 if (atomic_dec_and_test(&lseg->pls_refcount)) {
375 put_lseg_common(lseg);
376 list_add(&lseg->pls_list, tmp_list);
377 rv = 1;
378 }
Fred Isaman4541d162011-01-06 11:36:23 +0000379 }
380 return rv;
381}
382
383/* Returns count of number of matching invalid lsegs remaining in list
384 * after call.
385 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000386int
Fred Isaman4541d162011-01-06 11:36:23 +0000387mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
388 struct list_head *tmp_list,
Benny Halevy778b5502011-05-22 19:48:02 +0300389 struct pnfs_layout_range *recall_range)
Andy Adamson974cec82010-10-20 00:18:02 -0400390{
391 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000392 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400393
394 dprintk("%s:Begin lo %p\n", __func__, lo);
395
Fred Isaman38511722011-02-03 18:28:50 +0000396 if (list_empty(&lo->plh_segs)) {
397 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
398 put_layout_hdr_locked(lo);
399 return 0;
400 }
Fred Isaman4541d162011-01-06 11:36:23 +0000401 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
Benny Halevy778b5502011-05-22 19:48:02 +0300402 if (!recall_range ||
403 should_free_lseg(&lseg->pls_range, recall_range)) {
Fred Isaman4541d162011-01-06 11:36:23 +0000404 dprintk("%s: freeing lseg %p iomode %d "
405 "offset %llu length %llu\n", __func__,
406 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
407 lseg->pls_range.length);
408 invalid++;
409 removed += mark_lseg_invalid(lseg, tmp_list);
410 }
411 dprintk("%s:Return %i\n", __func__, invalid - removed);
412 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400413}
414
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000415/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000416void
Fred Isaman4541d162011-01-06 11:36:23 +0000417pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400418{
Fred Isaman4541d162011-01-06 11:36:23 +0000419 struct pnfs_layout_segment *lseg, *tmp;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000420 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400421
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000422 if (list_empty(free_me))
423 return;
424
425 lo = list_first_entry(free_me, struct pnfs_layout_segment,
426 pls_list)->pls_layout;
427
428 if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
429 struct nfs_client *clp;
430
431 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
432 spin_lock(&clp->cl_lock);
433 list_del_init(&lo->plh_layouts);
434 spin_unlock(&clp->cl_lock);
435 }
Fred Isaman4541d162011-01-06 11:36:23 +0000436 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000437 list_del(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000438 free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400439 }
440}
441
Benny Halevye5e94012010-10-20 00:18:01 -0400442void
443pnfs_destroy_layout(struct nfs_inode *nfsi)
444{
445 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400446 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400447
448 spin_lock(&nfsi->vfs_inode.i_lock);
449 lo = nfsi->layout;
450 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000451 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Benny Halevy778b5502011-05-22 19:48:02 +0300452 mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
Benny Halevye5e94012010-10-20 00:18:01 -0400453 }
454 spin_unlock(&nfsi->vfs_inode.i_lock);
Andy Adamson974cec82010-10-20 00:18:02 -0400455 pnfs_free_lseg_list(&tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400456}
457
Andy Adamson974cec82010-10-20 00:18:02 -0400458/*
459 * Called by the state manger to remove all layouts established under an
460 * expired lease.
461 */
462void
463pnfs_destroy_all_layouts(struct nfs_client *clp)
464{
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400465 struct nfs_server *server;
Andy Adamson974cec82010-10-20 00:18:02 -0400466 struct pnfs_layout_hdr *lo;
467 LIST_HEAD(tmp_list);
468
Andy Adamsonc47abcf2011-06-15 17:52:40 -0400469 nfs4_deviceid_mark_client_invalid(clp);
470 nfs4_deviceid_purge_client(clp);
471
Andy Adamson974cec82010-10-20 00:18:02 -0400472 spin_lock(&clp->cl_lock);
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400473 rcu_read_lock();
474 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
475 if (!list_empty(&server->layouts))
476 list_splice_init(&server->layouts, &tmp_list);
477 }
478 rcu_read_unlock();
Andy Adamson974cec82010-10-20 00:18:02 -0400479 spin_unlock(&clp->cl_lock);
480
481 while (!list_empty(&tmp_list)) {
482 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000483 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400484 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000485 lo->plh_inode->i_ino);
Andy Adamson2887fe42011-05-11 01:19:58 -0400486 list_del_init(&lo->plh_layouts);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000487 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400488 }
489}
490
Fred Isamanfd6002e2011-01-06 11:36:22 +0000491/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000492void
493pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
494 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400495{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000496 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400497
Fred Isamanfd6002e2011-01-06 11:36:22 +0000498 oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid);
499 newseq = be32_to_cpu(new->stateid.seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000500 if ((int)(newseq - oldseq) > 0) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000501 memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid));
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000502 if (update_barrier) {
503 u32 new_barrier = be32_to_cpu(new->stateid.seqid);
504
505 if ((int)(new_barrier - lo->plh_barrier))
506 lo->plh_barrier = new_barrier;
507 } else {
508 /* Because of wraparound, we want to keep the barrier
509 * "close" to the current seqids. It needs to be
510 * within 2**31 to count as "behind", so if it
511 * gets too near that limit, give us a litle leeway
512 * and bring it to within 2**30.
513 * NOTE - and yes, this is all unsigned arithmetic.
514 */
515 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
516 lo->plh_barrier = newseq - (1 << 30);
517 }
518 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400519}
520
Fred Isamancf7d63f2011-01-06 11:36:25 +0000521/* lget is set to 1 if called from inside send_layoutget call chain */
522static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000523pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
524 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000525{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000526 if ((stateid) &&
527 (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0)
528 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000529 return lo->plh_block_lgets ||
Fred Isaman38511722011-02-03 18:28:50 +0000530 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
Fred Isamanf7e89172011-01-06 11:36:32 +0000531 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000532 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000533 (atomic_read(&lo->plh_outstanding) > lget));
534}
535
Fred Isamanfd6002e2011-01-06 11:36:22 +0000536int
537pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
538 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400539{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000540 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400541
542 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000543 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000544 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000545 status = -EAGAIN;
546 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000547 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400548
Fred Isamanfd6002e2011-01-06 11:36:22 +0000549 do {
550 seq = read_seqbegin(&open_state->seqlock);
551 memcpy(dst->data, open_state->stateid.data,
552 sizeof(open_state->stateid.data));
553 } while (read_seqretry(&open_state->seqlock, seq));
554 } else
555 memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data));
556 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400557 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000558 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400559}
560
561/*
562* Get layout from server.
563* for now, assume that whole file layouts are requested.
564* arg->offset: 0
565* arg->length: all ones
566*/
Benny Halevye5e94012010-10-20 00:18:01 -0400567static struct pnfs_layout_segment *
568send_layoutget(struct pnfs_layout_hdr *lo,
569 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +0300570 struct pnfs_layout_range *range,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400571 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400572{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000573 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400574 struct nfs_server *server = NFS_SERVER(ino);
575 struct nfs4_layoutget *lgp;
576 struct pnfs_layout_segment *lseg = NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400577 struct page **pages = NULL;
578 int i;
579 u32 max_resp_sz, max_pages;
Benny Halevye5e94012010-10-20 00:18:01 -0400580
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400581 dprintk("--> %s\n", __func__);
582
583 BUG_ON(ctx == NULL);
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400584 lgp = kzalloc(sizeof(*lgp), gfp_flags);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000585 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400586 return NULL;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400587
588 /* allocate pages for xdr post processing */
589 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
590 max_pages = max_resp_sz >> PAGE_SHIFT;
591
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400592 pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400593 if (!pages)
594 goto out_err_free;
595
596 for (i = 0; i < max_pages; i++) {
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400597 pages[i] = alloc_page(gfp_flags);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400598 if (!pages[i])
599 goto out_err_free;
600 }
601
Benny Halevyfb3296e2011-05-22 19:47:26 +0300602 lgp->args.minlength = PAGE_CACHE_SIZE;
603 if (lgp->args.minlength > range->length)
604 lgp->args.minlength = range->length;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400605 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
Benny Halevyfb3296e2011-05-22 19:47:26 +0300606 lgp->args.range = *range;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400607 lgp->args.type = server->pnfs_curr_ld->id;
608 lgp->args.inode = ino;
609 lgp->args.ctx = get_nfs_open_context(ctx);
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400610 lgp->args.layout.pages = pages;
611 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400612 lgp->lsegpp = &lseg;
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400613 lgp->gfp_flags = gfp_flags;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400614
615 /* Synchronously retrieve layout information from server and
616 * store in lseg.
617 */
618 nfs4_proc_layoutget(lgp);
619 if (!lseg) {
620 /* remember that LAYOUTGET failed and suspend trying */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300621 set_bit(lo_fail_bit(range->iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400622 }
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400623
624 /* free xdr pages */
625 for (i = 0; i < max_pages; i++)
626 __free_page(pages[i]);
627 kfree(pages);
628
Andy Adamson974cec82010-10-20 00:18:02 -0400629 return lseg;
Weston Andros Adamson35124a02011-03-24 16:48:21 -0400630
631out_err_free:
632 /* free any allocated xdr pages, lgp as it's not used */
633 if (pages) {
634 for (i = 0; i < max_pages; i++) {
635 if (!pages[i])
636 break;
637 __free_page(pages[i]);
638 }
639 kfree(pages);
640 }
641 kfree(lgp);
642 return NULL;
Andy Adamson974cec82010-10-20 00:18:02 -0400643}
644
Benny Halevycbe82602011-05-22 19:52:37 +0300645/* Initiates a LAYOUTRETURN(FILE) */
646int
647_pnfs_return_layout(struct inode *ino)
648{
649 struct pnfs_layout_hdr *lo = NULL;
650 struct nfs_inode *nfsi = NFS_I(ino);
651 LIST_HEAD(tmp_list);
652 struct nfs4_layoutreturn *lrp;
653 nfs4_stateid stateid;
654 int status = 0;
655
656 dprintk("--> %s\n", __func__);
657
658 spin_lock(&ino->i_lock);
659 lo = nfsi->layout;
Fred Isamana2e1d4f2011-06-13 18:54:53 -0400660 if (!lo) {
Benny Halevycbe82602011-05-22 19:52:37 +0300661 spin_unlock(&ino->i_lock);
Fred Isamana2e1d4f2011-06-13 18:54:53 -0400662 dprintk("%s: no layout to return\n", __func__);
663 return status;
Benny Halevycbe82602011-05-22 19:52:37 +0300664 }
665 stateid = nfsi->layout->plh_stateid;
666 /* Reference matched in nfs4_layoutreturn_release */
667 get_layout_hdr(lo);
Fred Isamanea0ded72011-06-15 12:31:02 -0400668 mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
669 lo->plh_block_lgets++;
Benny Halevycbe82602011-05-22 19:52:37 +0300670 spin_unlock(&ino->i_lock);
671 pnfs_free_lseg_list(&tmp_list);
672
673 WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags));
674
675 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
676 if (unlikely(lrp == NULL)) {
677 status = -ENOMEM;
Fred Isaman9e2dfdb2011-06-15 14:32:02 -0400678 set_bit(NFS_LAYOUT_RW_FAILED, &lo->plh_flags);
679 set_bit(NFS_LAYOUT_RO_FAILED, &lo->plh_flags);
Benny Halevy1ed3a852011-06-15 11:39:57 -0400680 put_layout_hdr(lo);
Benny Halevycbe82602011-05-22 19:52:37 +0300681 goto out;
682 }
683
684 lrp->args.stateid = stateid;
685 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
686 lrp->args.inode = ino;
Trond Myklebusta56aaa02011-06-15 11:59:10 -0400687 lrp->args.layout = lo;
Benny Halevycbe82602011-05-22 19:52:37 +0300688 lrp->clp = NFS_SERVER(ino)->nfs_client;
689
690 status = nfs4_proc_layoutreturn(lrp);
691out:
692 dprintk("<-- %s status: %d\n", __func__, status);
693 return status;
694}
695
Fred Isamanf7e89172011-01-06 11:36:32 +0000696bool pnfs_roc(struct inode *ino)
697{
698 struct pnfs_layout_hdr *lo;
699 struct pnfs_layout_segment *lseg, *tmp;
700 LIST_HEAD(tmp_list);
701 bool found = false;
702
703 spin_lock(&ino->i_lock);
704 lo = NFS_I(ino)->layout;
705 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
706 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
707 goto out_nolayout;
708 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
709 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
710 mark_lseg_invalid(lseg, &tmp_list);
711 found = true;
712 }
713 if (!found)
714 goto out_nolayout;
715 lo->plh_block_lgets++;
716 get_layout_hdr(lo); /* matched in pnfs_roc_release */
717 spin_unlock(&ino->i_lock);
718 pnfs_free_lseg_list(&tmp_list);
719 return true;
720
721out_nolayout:
722 spin_unlock(&ino->i_lock);
723 return false;
724}
725
726void pnfs_roc_release(struct inode *ino)
727{
728 struct pnfs_layout_hdr *lo;
729
730 spin_lock(&ino->i_lock);
731 lo = NFS_I(ino)->layout;
732 lo->plh_block_lgets--;
733 put_layout_hdr_locked(lo);
734 spin_unlock(&ino->i_lock);
735}
736
737void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
738{
739 struct pnfs_layout_hdr *lo;
740
741 spin_lock(&ino->i_lock);
742 lo = NFS_I(ino)->layout;
743 if ((int)(barrier - lo->plh_barrier) > 0)
744 lo->plh_barrier = barrier;
745 spin_unlock(&ino->i_lock);
746}
747
748bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
749{
750 struct nfs_inode *nfsi = NFS_I(ino);
751 struct pnfs_layout_segment *lseg;
752 bool found = false;
753
754 spin_lock(&ino->i_lock);
755 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
756 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
757 found = true;
758 break;
759 }
760 if (!found) {
761 struct pnfs_layout_hdr *lo = nfsi->layout;
762 u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid);
763
764 /* Since close does not return a layout stateid for use as
765 * a barrier, we choose the worst-case barrier.
766 */
767 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
768 }
769 spin_unlock(&ino->i_lock);
770 return found;
771}
772
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400773/*
774 * Compare two layout segments for sorting into layout cache.
775 * We want to preferentially return RW over RO layouts, so ensure those
776 * are seen first.
777 */
778static s64
Benny Halevyfb3296e2011-05-22 19:47:26 +0300779cmp_layout(struct pnfs_layout_range *l1,
780 struct pnfs_layout_range *l2)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400781{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300782 s64 d;
783
784 /* high offset > low offset */
785 d = l1->offset - l2->offset;
786 if (d)
787 return d;
788
789 /* short length > long length */
790 d = l2->length - l1->length;
791 if (d)
792 return d;
793
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400794 /* read > read/write */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300795 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400796}
797
Andy Adamson974cec82010-10-20 00:18:02 -0400798static void
799pnfs_insert_layout(struct pnfs_layout_hdr *lo,
800 struct pnfs_layout_segment *lseg)
801{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400802 struct pnfs_layout_segment *lp;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400803
Andy Adamson974cec82010-10-20 00:18:02 -0400804 dprintk("%s:Begin\n", __func__);
805
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000806 assert_spin_locked(&lo->plh_inode->i_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000807 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Benny Halevyfb3296e2011-05-22 19:47:26 +0300808 if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400809 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000810 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400811 dprintk("%s: inserted lseg %p "
812 "iomode %d offset %llu length %llu before "
813 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000814 __func__, lseg, lseg->pls_range.iomode,
815 lseg->pls_range.offset, lseg->pls_range.length,
816 lp, lp->pls_range.iomode, lp->pls_range.offset,
817 lp->pls_range.length);
Benny Halevyfb3296e2011-05-22 19:47:26 +0300818 goto out;
Andy Adamson974cec82010-10-20 00:18:02 -0400819 }
Benny Halevyfb3296e2011-05-22 19:47:26 +0300820 list_add_tail(&lseg->pls_list, &lo->plh_segs);
821 dprintk("%s: inserted lseg %p "
822 "iomode %d offset %llu length %llu at tail\n",
823 __func__, lseg, lseg->pls_range.iomode,
824 lseg->pls_range.offset, lseg->pls_range.length);
825out:
Fred Isamancc6e5342011-01-06 11:36:28 +0000826 get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400827
828 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400829}
830
831static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -0400832alloc_init_layout_hdr(struct inode *ino,
833 struct nfs_open_context *ctx,
834 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400835{
836 struct pnfs_layout_hdr *lo;
837
Benny Halevy636fb9c2011-05-22 19:51:33 +0300838 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400839 if (!lo)
840 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000841 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000842 INIT_LIST_HEAD(&lo->plh_layouts);
843 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000844 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000845 lo->plh_inode = ino;
Peng Tao9fa40752011-07-30 20:52:32 -0400846 lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred);
Benny Halevye5e94012010-10-20 00:18:01 -0400847 return lo;
848}
849
850static struct pnfs_layout_hdr *
Peng Tao9fa40752011-07-30 20:52:32 -0400851pnfs_find_alloc_layout(struct inode *ino,
852 struct nfs_open_context *ctx,
853 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400854{
855 struct nfs_inode *nfsi = NFS_I(ino);
856 struct pnfs_layout_hdr *new = NULL;
857
858 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
859
860 assert_spin_locked(&ino->i_lock);
Fred Isaman4541d162011-01-06 11:36:23 +0000861 if (nfsi->layout) {
862 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
863 return NULL;
864 else
865 return nfsi->layout;
866 }
Benny Halevye5e94012010-10-20 00:18:01 -0400867 spin_unlock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -0400868 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400869 spin_lock(&ino->i_lock);
870
871 if (likely(nfsi->layout == NULL)) /* Won the race? */
872 nfsi->layout = new;
873 else
Benny Halevy636fb9c2011-05-22 19:51:33 +0300874 pnfs_free_layout_hdr(new);
Benny Halevye5e94012010-10-20 00:18:01 -0400875 return nfsi->layout;
876}
877
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400878/*
879 * iomode matching rules:
880 * iomode lseg match
881 * ----- ----- -----
882 * ANY READ true
883 * ANY RW true
884 * RW READ false
885 * RW RW true
886 * READ READ true
887 * READ RW true
888 */
889static int
Benny Halevyfb3296e2011-05-22 19:47:26 +0300890is_matching_lseg(struct pnfs_layout_range *ls_range,
891 struct pnfs_layout_range *range)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400892{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300893 struct pnfs_layout_range range1;
894
895 if ((range->iomode == IOMODE_RW &&
896 ls_range->iomode != IOMODE_RW) ||
897 !lo_seg_intersecting(ls_range, range))
898 return 0;
899
900 /* range1 covers only the first byte in the range */
901 range1 = *range;
902 range1.length = 1;
903 return lo_seg_contained(ls_range, &range1);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400904}
905
906/*
907 * lookup range in layout
908 */
Benny Halevye5e94012010-10-20 00:18:01 -0400909static struct pnfs_layout_segment *
Benny Halevyfb3296e2011-05-22 19:47:26 +0300910pnfs_find_lseg(struct pnfs_layout_hdr *lo,
911 struct pnfs_layout_range *range)
Benny Halevye5e94012010-10-20 00:18:01 -0400912{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400913 struct pnfs_layout_segment *lseg, *ret = NULL;
914
915 dprintk("%s:Begin\n", __func__);
916
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000917 assert_spin_locked(&lo->plh_inode->i_lock);
918 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000919 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
Benny Halevyfb3296e2011-05-22 19:47:26 +0300920 is_matching_lseg(&lseg->pls_range, range)) {
Fred Isamand684d2a2011-03-01 01:34:13 +0000921 ret = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400922 break;
923 }
Benny Halevyd771e3a2011-06-14 16:30:16 -0400924 if (lseg->pls_range.offset > range->offset)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400925 break;
926 }
927
928 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000929 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400930 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400931}
932
933/*
934 * Layout segment is retreived from the server if not cached.
935 * The appropriate layout segment is referenced and returned to the caller.
936 */
Andy Adamson7c24d942011-06-13 18:22:38 -0400937struct pnfs_layout_segment *
Benny Halevye5e94012010-10-20 00:18:01 -0400938pnfs_update_layout(struct inode *ino,
939 struct nfs_open_context *ctx,
Benny Halevyfb3296e2011-05-22 19:47:26 +0300940 loff_t pos,
941 u64 count,
Trond Myklebusta75b9df2011-05-11 18:00:51 -0400942 enum pnfs_iomode iomode,
943 gfp_t gfp_flags)
Benny Halevye5e94012010-10-20 00:18:01 -0400944{
Benny Halevyfb3296e2011-05-22 19:47:26 +0300945 struct pnfs_layout_range arg = {
946 .iomode = iomode,
947 .offset = pos,
948 .length = count,
949 };
Benny Halevy707ed5f2011-05-22 19:47:46 +0300950 unsigned pg_offset;
Benny Halevye5e94012010-10-20 00:18:01 -0400951 struct nfs_inode *nfsi = NFS_I(ino);
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400952 struct nfs_server *server = NFS_SERVER(ino);
953 struct nfs_client *clp = server->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -0400954 struct pnfs_layout_hdr *lo;
955 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000956 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -0400957
958 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
959 return NULL;
960 spin_lock(&ino->i_lock);
Peng Tao9fa40752011-07-30 20:52:32 -0400961 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
Benny Halevye5e94012010-10-20 00:18:01 -0400962 if (lo == NULL) {
963 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
964 goto out_unlock;
965 }
966
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000967 /* Do we even need to bother with this? */
968 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
969 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
970 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400971 goto out_unlock;
972 }
973
974 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000975 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400976 goto out_unlock;
977
Andy Adamson568e8c42011-03-01 01:34:22 +0000978 /* Check to see if the layout for the given range already exists */
Benny Halevyfb3296e2011-05-22 19:47:26 +0300979 lseg = pnfs_find_lseg(lo, &arg);
Andy Adamson568e8c42011-03-01 01:34:22 +0000980 if (lseg)
981 goto out_unlock;
982
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000983 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +0000984 goto out_unlock;
985 atomic_inc(&lo->plh_outstanding);
986
Fred Isamancc6e5342011-01-06 11:36:28 +0000987 get_layout_hdr(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000988 if (list_empty(&lo->plh_segs))
989 first = true;
990 spin_unlock(&ino->i_lock);
991 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +0000992 /* The lo must be on the clp list if there is any
993 * chance of a CB_LAYOUTRECALL(FILE) coming in.
994 */
995 spin_lock(&clp->cl_lock);
996 BUG_ON(!list_empty(&lo->plh_layouts));
Weston Andros Adamson6382a442011-06-01 16:44:44 -0400997 list_add_tail(&lo->plh_layouts, &server->layouts);
Fred Isaman2130ff62011-01-06 11:36:26 +0000998 spin_unlock(&clp->cl_lock);
999 }
Benny Halevye5e94012010-10-20 00:18:01 -04001000
Benny Halevy707ed5f2011-05-22 19:47:46 +03001001 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1002 if (pg_offset) {
1003 arg.offset -= pg_offset;
1004 arg.length += pg_offset;
1005 }
Andy Adamson7c24d942011-06-13 18:22:38 -04001006 if (arg.length != NFS4_MAX_UINT64)
1007 arg.length = PAGE_CACHE_ALIGN(arg.length);
Benny Halevy707ed5f2011-05-22 19:47:46 +03001008
Benny Halevyfb3296e2011-05-22 19:47:26 +03001009 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
Fred Isamanf49f9ba2011-02-03 18:28:52 +00001010 if (!lseg && first) {
1011 spin_lock(&clp->cl_lock);
1012 list_del_init(&lo->plh_layouts);
1013 spin_unlock(&clp->cl_lock);
Fred Isaman2130ff62011-01-06 11:36:26 +00001014 }
Fred Isamancf7d63f2011-01-06 11:36:25 +00001015 atomic_dec(&lo->plh_outstanding);
Fred Isamancc6e5342011-01-06 11:36:28 +00001016 put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -04001017out:
1018 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Andy Adamsonbf9c1382011-03-01 01:34:07 +00001019 nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
Benny Halevye5e94012010-10-20 00:18:01 -04001020 return lseg;
1021out_unlock:
1022 spin_unlock(&ino->i_lock);
1023 goto out;
1024}
Andy Adamson7c24d942011-06-13 18:22:38 -04001025EXPORT_SYMBOL_GPL(pnfs_update_layout);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001026
1027int
1028pnfs_layout_process(struct nfs4_layoutget *lgp)
1029{
1030 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1031 struct nfs4_layoutget_res *res = &lgp->res;
1032 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +00001033 struct inode *ino = lo->plh_inode;
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001034 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001035 int status = 0;
1036
1037 /* Inject layout blob into I/O device driver */
Trond Myklebusta75b9df2011-05-11 18:00:51 -04001038 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001039 if (!lseg || IS_ERR(lseg)) {
1040 if (!lseg)
1041 status = -ENOMEM;
1042 else
1043 status = PTR_ERR(lseg);
1044 dprintk("%s: Could not allocate layout: error %d\n",
1045 __func__, status);
1046 goto out;
1047 }
1048
1049 spin_lock(&ino->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001050 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
1051 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1052 dprintk("%s forget reply due to recall\n", __func__);
1053 goto out_forget_reply;
1054 }
1055
1056 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
1057 dprintk("%s forget reply due to state\n", __func__);
1058 goto out_forget_reply;
1059 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001060 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +00001061 lseg->pls_range = res->range;
Fred Isamand684d2a2011-03-01 01:34:13 +00001062 *lgp->lsegpp = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001063 pnfs_insert_layout(lo, lseg);
1064
Fred Isamanf7e89172011-01-06 11:36:32 +00001065 if (res->return_on_close) {
1066 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1067 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
1068 }
1069
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001070 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001071 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001072 spin_unlock(&ino->i_lock);
1073out:
1074 return status;
Fred Isaman43f1b3d2011-01-06 11:36:30 +00001075
1076out_forget_reply:
1077 spin_unlock(&ino->i_lock);
1078 lseg->pls_layout = lo;
1079 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1080 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001081}
1082
Trond Myklebustd8007d42011-06-10 13:30:23 -04001083void
1084pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1085{
1086 BUG_ON(pgio->pg_lseg != NULL);
1087
1088 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1089 req->wb_context,
1090 req_offset(req),
1091 req->wb_bytes,
1092 IOMODE_READ,
1093 GFP_KERNEL);
Trond Myklebuste885de12011-06-10 13:30:23 -04001094 /* If no lseg, fall back to read through mds */
1095 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001096 nfs_pageio_reset_read_mds(pgio);
Trond Myklebuste885de12011-06-10 13:30:23 -04001097
Trond Myklebustd8007d42011-06-10 13:30:23 -04001098}
1099EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1100
1101void
1102pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1103{
1104 BUG_ON(pgio->pg_lseg != NULL);
1105
1106 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1107 req->wb_context,
1108 req_offset(req),
1109 req->wb_bytes,
1110 IOMODE_RW,
1111 GFP_NOFS);
Trond Myklebuste885de12011-06-10 13:30:23 -04001112 /* If no lseg, fall back to write through mds */
1113 if (pgio->pg_lseg == NULL)
Trond Myklebust1f945352011-07-13 15:59:57 -04001114 nfs_pageio_reset_write_mds(pgio);
Trond Myklebustd8007d42011-06-10 13:30:23 -04001115}
1116EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1117
Benny Halevy18ad0a92011-05-25 21:03:56 +03001118bool
Trond Myklebust1751c362011-06-10 13:30:23 -04001119pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode)
1120{
1121 struct nfs_server *server = NFS_SERVER(inode);
1122 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1123
1124 if (ld == NULL)
1125 return false;
1126 nfs_pageio_init(pgio, inode, ld->pg_read_ops, server->rsize, 0);
1127 return true;
1128}
1129
1130bool
1131pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode, int ioflags)
1132{
1133 struct nfs_server *server = NFS_SERVER(inode);
1134 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
1135
1136 if (ld == NULL)
1137 return false;
1138 nfs_pageio_init(pgio, inode, ld->pg_write_ops, server->wsize, ioflags);
1139 return true;
1140}
1141
1142bool
Benny Halevydfed2062011-05-25 20:25:22 +03001143pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1144 struct nfs_page *req)
Fred Isamanbae724e2011-03-01 01:34:15 +00001145{
Trond Myklebustd8007d42011-06-10 13:30:23 -04001146 if (pgio->pg_lseg == NULL)
1147 return nfs_generic_pg_test(pgio, prev, req);
Benny Halevy89a58e32011-05-25 20:54:40 +03001148
Trond Myklebust19982ba2011-06-10 13:30:23 -04001149 /*
1150 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1151 * Note that this test makes several assumptions:
1152 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1153 * is known to lie within the range.
1154 * - that the nfs_page being tested is known to be contiguous with the
1155 * previous nfs_page.
1156 * - Layout ranges are page aligned, so we only have to test the
1157 * start offset of the request.
1158 *
1159 * Please also note that 'end_offset' is actually the offset of the
1160 * first byte that lies outside the pnfs_layout_range. FIXME?
1161 *
1162 */
1163 return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset,
1164 pgio->pg_lseg->pls_range.length);
Fred Isamanbae724e2011-03-01 01:34:15 +00001165}
Benny Halevy89a58e32011-05-25 20:54:40 +03001166EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
Fred Isamanbae724e2011-03-01 01:34:15 +00001167
Benny Halevyd20581a2011-05-22 19:52:03 +03001168/*
1169 * Called by non rpc-based layout drivers
1170 */
1171int
1172pnfs_ld_write_done(struct nfs_write_data *data)
Fred Isaman94ad1c82011-03-01 01:34:14 +00001173{
Benny Halevyd20581a2011-05-22 19:52:03 +03001174 int status;
Fred Isaman94ad1c82011-03-01 01:34:14 +00001175
Benny Halevyd20581a2011-05-22 19:52:03 +03001176 if (!data->pnfs_error) {
1177 pnfs_set_layoutcommit(data);
1178 data->mds_ops->rpc_call_done(&data->task, data);
1179 data->mds_ops->rpc_release(data);
1180 return 0;
Fred Isaman44b83792011-03-03 15:13:44 +00001181 }
Fred Isaman44b83792011-03-03 15:13:44 +00001182
Benny Halevyd20581a2011-05-22 19:52:03 +03001183 dprintk("%s: pnfs_error=%d, retry via MDS\n", __func__,
1184 data->pnfs_error);
1185 status = nfs_initiate_write(data, NFS_CLIENT(data->inode),
1186 data->mds_ops, NFS_FILE_SYNC);
1187 return status ? : -EAGAIN;
Fred Isaman44b83792011-03-03 15:13:44 +00001188}
Benny Halevyd20581a2011-05-22 19:52:03 +03001189EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
Fred Isaman44b83792011-03-03 15:13:44 +00001190
Trond Myklebustdce81292011-07-13 15:59:19 -04001191static void
1192pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1193 struct nfs_write_data *data)
1194{
1195 list_splice_tail_init(&data->pages, &desc->pg_list);
1196 if (data->req && list_empty(&data->req->wb_list))
1197 nfs_list_add_request(data->req, &desc->pg_list);
1198 nfs_pageio_reset_write_mds(desc);
1199 desc->pg_recoalesce = 1;
1200 nfs_writedata_release(data);
1201}
1202
1203static enum pnfs_try_status
Andy Adamson0382b742011-03-03 15:13:45 +00001204pnfs_try_to_write_data(struct nfs_write_data *wdata,
Trond Myklebustdce81292011-07-13 15:59:19 -04001205 const struct rpc_call_ops *call_ops,
1206 struct pnfs_layout_segment *lseg,
1207 int how)
Andy Adamson0382b742011-03-03 15:13:45 +00001208{
1209 struct inode *inode = wdata->inode;
1210 enum pnfs_try_status trypnfs;
1211 struct nfs_server *nfss = NFS_SERVER(inode);
1212
1213 wdata->mds_ops = call_ops;
Trond Myklebustdce81292011-07-13 15:59:19 -04001214 wdata->lseg = get_lseg(lseg);
Andy Adamson0382b742011-03-03 15:13:45 +00001215
1216 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1217 inode->i_ino, wdata->args.count, wdata->args.offset, how);
1218
1219 trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
1220 if (trypnfs == PNFS_NOT_ATTEMPTED) {
1221 put_lseg(wdata->lseg);
1222 wdata->lseg = NULL;
1223 } else
1224 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
1225
1226 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1227 return trypnfs;
1228}
1229
Trond Myklebustdce81292011-07-13 15:59:19 -04001230static void
1231pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how)
1232{
1233 struct nfs_write_data *data;
1234 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1235 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1236
1237 desc->pg_lseg = NULL;
1238 while (!list_empty(head)) {
1239 enum pnfs_try_status trypnfs;
1240
1241 data = list_entry(head->next, struct nfs_write_data, list);
1242 list_del_init(&data->list);
1243
1244 trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
1245 if (trypnfs == PNFS_NOT_ATTEMPTED)
1246 pnfs_write_through_mds(desc, data);
1247 }
1248 put_lseg(lseg);
1249}
1250
1251int
1252pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1253{
1254 LIST_HEAD(head);
1255 int ret;
1256
1257 ret = nfs_generic_flush(desc, &head);
1258 if (ret != 0) {
1259 put_lseg(desc->pg_lseg);
1260 desc->pg_lseg = NULL;
1261 return ret;
1262 }
1263 pnfs_do_multiple_writes(desc, &head, desc->pg_ioflags);
1264 return 0;
1265}
1266EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
1267
Andy Adamsonb1f69b72010-10-20 00:18:03 -04001268/*
Benny Halevyd20581a2011-05-22 19:52:03 +03001269 * Called by non rpc-based layout drivers
1270 */
1271int
1272pnfs_ld_read_done(struct nfs_read_data *data)
1273{
1274 int status;
1275
1276 if (!data->pnfs_error) {
1277 __nfs4_read_done_cb(data);
1278 data->mds_ops->rpc_call_done(&data->task, data);
1279 data->mds_ops->rpc_release(data);
1280 return 0;
1281 }
1282
1283 dprintk("%s: pnfs_error=%d, retry via MDS\n", __func__,
1284 data->pnfs_error);
1285 status = nfs_initiate_read(data, NFS_CLIENT(data->inode),
1286 data->mds_ops);
1287 return status ? : -EAGAIN;
1288}
1289EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
1290
Trond Myklebust493292d2011-07-13 15:58:28 -04001291static void
1292pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1293 struct nfs_read_data *data)
1294{
1295 list_splice_tail_init(&data->pages, &desc->pg_list);
1296 if (data->req && list_empty(&data->req->wb_list))
1297 nfs_list_add_request(data->req, &desc->pg_list);
1298 nfs_pageio_reset_read_mds(desc);
1299 desc->pg_recoalesce = 1;
1300 nfs_readdata_release(data);
1301}
1302
Benny Halevyd20581a2011-05-22 19:52:03 +03001303/*
Andy Adamson64419a92011-03-01 01:34:16 +00001304 * Call the appropriate parallel I/O subsystem read function.
1305 */
Trond Myklebust493292d2011-07-13 15:58:28 -04001306static enum pnfs_try_status
Andy Adamson64419a92011-03-01 01:34:16 +00001307pnfs_try_to_read_data(struct nfs_read_data *rdata,
Trond Myklebust493292d2011-07-13 15:58:28 -04001308 const struct rpc_call_ops *call_ops,
1309 struct pnfs_layout_segment *lseg)
Andy Adamson64419a92011-03-01 01:34:16 +00001310{
1311 struct inode *inode = rdata->inode;
1312 struct nfs_server *nfss = NFS_SERVER(inode);
1313 enum pnfs_try_status trypnfs;
1314
1315 rdata->mds_ops = call_ops;
Trond Myklebust493292d2011-07-13 15:58:28 -04001316 rdata->lseg = get_lseg(lseg);
Andy Adamson64419a92011-03-01 01:34:16 +00001317
1318 dprintk("%s: Reading ino:%lu %u@%llu\n",
1319 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
1320
1321 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
1322 if (trypnfs == PNFS_NOT_ATTEMPTED) {
1323 put_lseg(rdata->lseg);
1324 rdata->lseg = NULL;
1325 } else {
1326 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
1327 }
1328 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1329 return trypnfs;
1330}
Andy Adamson863a3c62011-03-23 13:27:54 +00001331
Trond Myklebust493292d2011-07-13 15:58:28 -04001332static void
1333pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head)
1334{
1335 struct nfs_read_data *data;
1336 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1337 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1338
1339 desc->pg_lseg = NULL;
1340 while (!list_empty(head)) {
1341 enum pnfs_try_status trypnfs;
1342
1343 data = list_entry(head->next, struct nfs_read_data, list);
1344 list_del_init(&data->list);
1345
1346 trypnfs = pnfs_try_to_read_data(data, call_ops, lseg);
1347 if (trypnfs == PNFS_NOT_ATTEMPTED)
1348 pnfs_read_through_mds(desc, data);
1349 }
1350 put_lseg(lseg);
1351}
1352
1353int
1354pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
1355{
1356 LIST_HEAD(head);
1357 int ret;
1358
1359 ret = nfs_generic_pagein(desc, &head);
1360 if (ret != 0) {
1361 put_lseg(desc->pg_lseg);
1362 desc->pg_lseg = NULL;
1363 return ret;
1364 }
1365 pnfs_do_multiple_reads(desc, &head);
1366 return 0;
1367}
1368EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
1369
Andy Adamson863a3c62011-03-23 13:27:54 +00001370/*
Peng Taoa9bae562011-07-30 20:52:33 -04001371 * There can be multiple RW segments.
Andy Adamson863a3c62011-03-23 13:27:54 +00001372 */
Peng Taoa9bae562011-07-30 20:52:33 -04001373static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
Andy Adamson863a3c62011-03-23 13:27:54 +00001374{
Peng Taoa9bae562011-07-30 20:52:33 -04001375 struct pnfs_layout_segment *lseg;
Andy Adamson863a3c62011-03-23 13:27:54 +00001376
Peng Taoa9bae562011-07-30 20:52:33 -04001377 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
1378 if (lseg->pls_range.iomode == IOMODE_RW &&
1379 test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1380 list_add(&lseg->pls_lc_list, listp);
1381 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001382}
1383
1384void
1385pnfs_set_layoutcommit(struct nfs_write_data *wdata)
1386{
1387 struct nfs_inode *nfsi = NFS_I(wdata->inode);
Vitaliy Gusev4b8ee2b2011-05-20 01:34:46 +04001388 loff_t end_pos = wdata->mds_offset + wdata->res.count;
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001389 bool mark_as_dirty = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001390
1391 spin_lock(&nfsi->vfs_inode.i_lock);
1392 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001393 mark_as_dirty = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001394 dprintk("%s: Set layoutcommit for inode %lu ",
1395 __func__, wdata->inode->i_ino);
1396 }
Peng Taoa9bae562011-07-30 20:52:33 -04001397 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &wdata->lseg->pls_flags)) {
1398 /* references matched in nfs4_layoutcommit_release */
1399 get_lseg(wdata->lseg);
1400 }
Peng Taoacff58802011-07-30 20:52:31 -04001401 if (end_pos > nfsi->layout->plh_lwb)
1402 nfsi->layout->plh_lwb = end_pos;
Andy Adamson863a3c62011-03-23 13:27:54 +00001403 spin_unlock(&nfsi->vfs_inode.i_lock);
Peng Taoacff58802011-07-30 20:52:31 -04001404 dprintk("%s: lseg %p end_pos %llu\n",
1405 __func__, wdata->lseg, nfsi->layout->plh_lwb);
Weston Andros Adamson79a48a12011-04-13 10:53:51 -04001406
1407 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1408 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1409 if (mark_as_dirty)
1410 mark_inode_dirty_sync(wdata->inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001411}
1412EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
1413
Andy Adamsondb29c082011-07-30 20:52:38 -04001414void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
1415{
1416 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
1417
1418 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
1419 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
1420}
1421
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001422/*
1423 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1424 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1425 * data to disk to allow the server to recover the data if it crashes.
1426 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1427 * is off, and a COMMIT is sent to a data server, or
1428 * if WRITEs to a data server return NFS_DATA_SYNC.
1429 */
Andy Adamson863a3c62011-03-23 13:27:54 +00001430int
Andy Adamsonef311532011-03-12 02:58:10 -05001431pnfs_layoutcommit_inode(struct inode *inode, bool sync)
Andy Adamson863a3c62011-03-23 13:27:54 +00001432{
1433 struct nfs4_layoutcommit_data *data;
1434 struct nfs_inode *nfsi = NFS_I(inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001435 loff_t end_pos;
1436 int status = 0;
1437
1438 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
1439
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001440 if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1441 return 0;
1442
Andy Adamson863a3c62011-03-23 13:27:54 +00001443 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1444 data = kzalloc(sizeof(*data), GFP_NOFS);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001445 if (!data) {
1446 mark_inode_dirty_sync(inode);
1447 status = -ENOMEM;
1448 goto out;
1449 }
Andy Adamson863a3c62011-03-23 13:27:54 +00001450
Peng Taoa9bae562011-07-30 20:52:33 -04001451 INIT_LIST_HEAD(&data->lseg_list);
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001452 spin_lock(&inode->i_lock);
Andy Adamson863a3c62011-03-23 13:27:54 +00001453 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
1454 spin_unlock(&inode->i_lock);
1455 kfree(data);
1456 goto out;
1457 }
Peng Taoa9bae562011-07-30 20:52:33 -04001458
1459 pnfs_list_write_lseg(inode, &data->lseg_list);
Andy Adamson863a3c62011-03-23 13:27:54 +00001460
Peng Taoacff58802011-07-30 20:52:31 -04001461 end_pos = nfsi->layout->plh_lwb;
Peng Taoacff58802011-07-30 20:52:31 -04001462 nfsi->layout->plh_lwb = 0;
Andy Adamson863a3c62011-03-23 13:27:54 +00001463
Andy Adamsonde4b15c2011-03-12 02:58:09 -05001464 memcpy(&data->args.stateid.data, nfsi->layout->plh_stateid.data,
1465 sizeof(nfsi->layout->plh_stateid.data));
Andy Adamson863a3c62011-03-23 13:27:54 +00001466 spin_unlock(&inode->i_lock);
1467
1468 data->args.inode = inode;
Peng Tao9fa40752011-07-30 20:52:32 -04001469 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
Andy Adamson863a3c62011-03-23 13:27:54 +00001470 nfs_fattr_init(&data->fattr);
1471 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
1472 data->res.fattr = &data->fattr;
1473 data->args.lastbytewritten = end_pos - 1;
1474 data->res.server = NFS_SERVER(inode);
1475
1476 status = nfs4_proc_layoutcommit(data, sync);
1477out:
1478 dprintk("<-- %s status %d\n", __func__, status);
1479 return status;
1480}