blob: 1f4c153441a10b019c883045aebc1b3dab7bc3ca [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>
Andy Adamson974cec82010-10-20 00:18:02 -040031#include "internal.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040032#include "pnfs.h"
Andy Adamson64419a92011-03-01 01:34:16 +000033#include "iostat.h"
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040034
35#define NFSDBG_FACILITY NFSDBG_PNFS
36
Fred Isaman02c35fc2010-10-20 00:17:59 -040037/* Locking:
38 *
39 * pnfs_spinlock:
40 * protects pnfs_modules_tbl.
41 */
42static DEFINE_SPINLOCK(pnfs_spinlock);
43
44/*
45 * pnfs_modules_tbl holds all pnfs modules
46 */
47static LIST_HEAD(pnfs_modules_tbl);
48
49/* Return the registered pnfs layout driver module matching given id */
50static struct pnfs_layoutdriver_type *
51find_pnfs_driver_locked(u32 id)
52{
53 struct pnfs_layoutdriver_type *local;
54
55 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
56 if (local->id == id)
57 goto out;
58 local = NULL;
59out:
60 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
61 return local;
62}
63
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040064static struct pnfs_layoutdriver_type *
65find_pnfs_driver(u32 id)
66{
Fred Isaman02c35fc2010-10-20 00:17:59 -040067 struct pnfs_layoutdriver_type *local;
68
69 spin_lock(&pnfs_spinlock);
70 local = find_pnfs_driver_locked(id);
71 spin_unlock(&pnfs_spinlock);
72 return local;
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040073}
74
75void
76unset_pnfs_layoutdriver(struct nfs_server *nfss)
77{
Christoph Hellwigea8eecd2011-03-01 01:34:21 +000078 if (nfss->pnfs_curr_ld)
Fred Isaman02c35fc2010-10-20 00:17:59 -040079 module_put(nfss->pnfs_curr_ld->owner);
Ricardo Labiaga85e174b2010-10-20 00:17:58 -040080 nfss->pnfs_curr_ld = NULL;
81}
82
83/*
84 * Try to set the server's pnfs module to the pnfs layout type specified by id.
85 * Currently only one pNFS layout driver per filesystem is supported.
86 *
87 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
88 */
89void
90set_pnfs_layoutdriver(struct nfs_server *server, u32 id)
91{
92 struct pnfs_layoutdriver_type *ld_type = NULL;
93
94 if (id == 0)
95 goto out_no_driver;
96 if (!(server->nfs_client->cl_exchange_flags &
97 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
98 printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__,
99 id, server->nfs_client->cl_exchange_flags);
100 goto out_no_driver;
101 }
102 ld_type = find_pnfs_driver(id);
103 if (!ld_type) {
104 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
105 ld_type = find_pnfs_driver(id);
106 if (!ld_type) {
107 dprintk("%s: No pNFS module found for %u.\n",
108 __func__, id);
109 goto out_no_driver;
110 }
111 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400112 if (!try_module_get(ld_type->owner)) {
113 dprintk("%s: Could not grab reference on module\n", __func__);
114 goto out_no_driver;
115 }
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400116 server->pnfs_curr_ld = ld_type;
Christoph Hellwigea8eecd2011-03-01 01:34:21 +0000117
Ricardo Labiaga85e174b2010-10-20 00:17:58 -0400118 dprintk("%s: pNFS module for %u set\n", __func__, id);
119 return;
120
121out_no_driver:
122 dprintk("%s: Using NFSv4 I/O\n", __func__);
123 server->pnfs_curr_ld = NULL;
124}
Fred Isaman02c35fc2010-10-20 00:17:59 -0400125
126int
127pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
128{
129 int status = -EINVAL;
130 struct pnfs_layoutdriver_type *tmp;
131
132 if (ld_type->id == 0) {
133 printk(KERN_ERR "%s id 0 is reserved\n", __func__);
134 return status;
135 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400136 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
137 printk(KERN_ERR "%s Layout driver must provide "
138 "alloc_lseg and free_lseg.\n", __func__);
139 return status;
140 }
Fred Isaman02c35fc2010-10-20 00:17:59 -0400141
142 spin_lock(&pnfs_spinlock);
143 tmp = find_pnfs_driver_locked(ld_type->id);
144 if (!tmp) {
145 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
146 status = 0;
147 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
148 ld_type->name);
149 } else {
150 printk(KERN_ERR "%s Module with id %d already loaded!\n",
151 __func__, ld_type->id);
152 }
153 spin_unlock(&pnfs_spinlock);
154
155 return status;
156}
157EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
158
159void
160pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
161{
162 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
163 spin_lock(&pnfs_spinlock);
164 list_del(&ld_type->pnfs_tblid);
165 spin_unlock(&pnfs_spinlock);
166}
167EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
Benny Halevye5e94012010-10-20 00:18:01 -0400168
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400169/*
170 * pNFS client layout cache
171 */
172
Fred Isamancc6e5342011-01-06 11:36:28 +0000173/* Need to hold i_lock if caller does not already hold reference */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000174void
Fred Isamancc6e5342011-01-06 11:36:28 +0000175get_layout_hdr(struct pnfs_layout_hdr *lo)
Benny Halevye5e94012010-10-20 00:18:01 -0400176{
Fred Isamancc6e5342011-01-06 11:36:28 +0000177 atomic_inc(&lo->plh_refcount);
178}
179
180static void
181destroy_layout_hdr(struct pnfs_layout_hdr *lo)
182{
183 dprintk("%s: freeing layout cache %p\n", __func__, lo);
184 BUG_ON(!list_empty(&lo->plh_layouts));
185 NFS_I(lo->plh_inode)->layout = NULL;
186 kfree(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400187}
188
189static void
190put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
191{
Fred Isamancc6e5342011-01-06 11:36:28 +0000192 if (atomic_dec_and_test(&lo->plh_refcount))
193 destroy_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400194}
195
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400196void
Fred Isamancc6e5342011-01-06 11:36:28 +0000197put_layout_hdr(struct pnfs_layout_hdr *lo)
Andy Adamson974cec82010-10-20 00:18:02 -0400198{
Fred Isamancc6e5342011-01-06 11:36:28 +0000199 struct inode *inode = lo->plh_inode;
200
201 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
202 destroy_layout_hdr(lo);
203 spin_unlock(&inode->i_lock);
204 }
Andy Adamson974cec82010-10-20 00:18:02 -0400205}
206
207static void
208init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
209{
Fred Isaman566052c2011-01-06 11:36:20 +0000210 INIT_LIST_HEAD(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000211 atomic_set(&lseg->pls_refcount, 1);
212 smp_mb();
213 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
Fred Isaman566052c2011-01-06 11:36:20 +0000214 lseg->pls_layout = lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400215}
216
Fred Isaman4541d162011-01-06 11:36:23 +0000217static void free_lseg(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400218{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000219 struct inode *ino = lseg->pls_layout->plh_inode;
Andy Adamson974cec82010-10-20 00:18:02 -0400220
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400221 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
Fred Isaman52fabd72011-01-06 11:36:18 +0000222 /* Matched by get_layout_hdr in pnfs_insert_layout */
Fred Isamancc6e5342011-01-06 11:36:28 +0000223 put_layout_hdr(NFS_I(ino)->layout);
Andy Adamson974cec82010-10-20 00:18:02 -0400224}
225
Fred Isamand684d2a2011-03-01 01:34:13 +0000226static void
227put_lseg_common(struct pnfs_layout_segment *lseg)
Andy Adamson974cec82010-10-20 00:18:02 -0400228{
Fred Isamand684d2a2011-03-01 01:34:13 +0000229 struct inode *inode = lseg->pls_layout->plh_inode;
230
231 BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
232 list_del_init(&lseg->pls_list);
233 if (list_empty(&lseg->pls_layout->plh_segs)) {
234 set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags);
235 /* Matched by initial refcount set in alloc_init_layout_hdr */
236 put_layout_hdr_locked(lseg->pls_layout);
237 }
238 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
239}
240
Fred Isamanbae724e2011-03-01 01:34:15 +0000241void
Fred Isamand684d2a2011-03-01 01:34:13 +0000242put_lseg(struct pnfs_layout_segment *lseg)
243{
244 struct inode *inode;
245
246 if (!lseg)
247 return;
248
Fred Isaman4541d162011-01-06 11:36:23 +0000249 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
250 atomic_read(&lseg->pls_refcount),
251 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
Fred Isamand684d2a2011-03-01 01:34:13 +0000252 inode = lseg->pls_layout->plh_inode;
253 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
254 LIST_HEAD(free_me);
Andy Adamson974cec82010-10-20 00:18:02 -0400255
Fred Isamand684d2a2011-03-01 01:34:13 +0000256 put_lseg_common(lseg);
257 list_add(&lseg->pls_list, &free_me);
258 spin_unlock(&inode->i_lock);
259 pnfs_free_lseg_list(&free_me);
Fred Isaman4541d162011-01-06 11:36:23 +0000260 }
Andy Adamson974cec82010-10-20 00:18:02 -0400261}
262
Fred Isaman4541d162011-01-06 11:36:23 +0000263static bool
264should_free_lseg(u32 lseg_iomode, u32 recall_iomode)
265{
266 return (recall_iomode == IOMODE_ANY ||
267 lseg_iomode == recall_iomode);
268}
269
270/* Returns 1 if lseg is removed from list, 0 otherwise */
271static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
272 struct list_head *tmp_list)
273{
274 int rv = 0;
275
276 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
277 /* Remove the reference keeping the lseg in the
278 * list. It will now be removed when all
279 * outstanding io is finished.
280 */
Fred Isamand684d2a2011-03-01 01:34:13 +0000281 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
282 atomic_read(&lseg->pls_refcount));
283 if (atomic_dec_and_test(&lseg->pls_refcount)) {
284 put_lseg_common(lseg);
285 list_add(&lseg->pls_list, tmp_list);
286 rv = 1;
287 }
Fred Isaman4541d162011-01-06 11:36:23 +0000288 }
289 return rv;
290}
291
292/* Returns count of number of matching invalid lsegs remaining in list
293 * after call.
294 */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000295int
Fred Isaman4541d162011-01-06 11:36:23 +0000296mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
297 struct list_head *tmp_list,
298 u32 iomode)
Andy Adamson974cec82010-10-20 00:18:02 -0400299{
300 struct pnfs_layout_segment *lseg, *next;
Fred Isaman4541d162011-01-06 11:36:23 +0000301 int invalid = 0, removed = 0;
Andy Adamson974cec82010-10-20 00:18:02 -0400302
303 dprintk("%s:Begin lo %p\n", __func__, lo);
304
Fred Isaman38511722011-02-03 18:28:50 +0000305 if (list_empty(&lo->plh_segs)) {
306 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags))
307 put_layout_hdr_locked(lo);
308 return 0;
309 }
Fred Isaman4541d162011-01-06 11:36:23 +0000310 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
311 if (should_free_lseg(lseg->pls_range.iomode, iomode)) {
312 dprintk("%s: freeing lseg %p iomode %d "
313 "offset %llu length %llu\n", __func__,
314 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
315 lseg->pls_range.length);
316 invalid++;
317 removed += mark_lseg_invalid(lseg, tmp_list);
318 }
319 dprintk("%s:Return %i\n", __func__, invalid - removed);
320 return invalid - removed;
Andy Adamson974cec82010-10-20 00:18:02 -0400321}
322
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000323/* note free_me must contain lsegs from a single layout_hdr */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000324void
Fred Isaman4541d162011-01-06 11:36:23 +0000325pnfs_free_lseg_list(struct list_head *free_me)
Andy Adamson974cec82010-10-20 00:18:02 -0400326{
Fred Isaman4541d162011-01-06 11:36:23 +0000327 struct pnfs_layout_segment *lseg, *tmp;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000328 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400329
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000330 if (list_empty(free_me))
331 return;
332
333 lo = list_first_entry(free_me, struct pnfs_layout_segment,
334 pls_list)->pls_layout;
335
336 if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) {
337 struct nfs_client *clp;
338
339 clp = NFS_SERVER(lo->plh_inode)->nfs_client;
340 spin_lock(&clp->cl_lock);
341 list_del_init(&lo->plh_layouts);
342 spin_unlock(&clp->cl_lock);
343 }
Fred Isaman4541d162011-01-06 11:36:23 +0000344 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000345 list_del(&lseg->pls_list);
Fred Isaman4541d162011-01-06 11:36:23 +0000346 free_lseg(lseg);
Andy Adamson974cec82010-10-20 00:18:02 -0400347 }
348}
349
Benny Halevye5e94012010-10-20 00:18:01 -0400350void
351pnfs_destroy_layout(struct nfs_inode *nfsi)
352{
353 struct pnfs_layout_hdr *lo;
Andy Adamson974cec82010-10-20 00:18:02 -0400354 LIST_HEAD(tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400355
356 spin_lock(&nfsi->vfs_inode.i_lock);
357 lo = nfsi->layout;
358 if (lo) {
Fred Isaman38511722011-02-03 18:28:50 +0000359 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
Fred Isaman4541d162011-01-06 11:36:23 +0000360 mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY);
Benny Halevye5e94012010-10-20 00:18:01 -0400361 }
362 spin_unlock(&nfsi->vfs_inode.i_lock);
Andy Adamson974cec82010-10-20 00:18:02 -0400363 pnfs_free_lseg_list(&tmp_list);
Benny Halevye5e94012010-10-20 00:18:01 -0400364}
365
Andy Adamson974cec82010-10-20 00:18:02 -0400366/*
367 * Called by the state manger to remove all layouts established under an
368 * expired lease.
369 */
370void
371pnfs_destroy_all_layouts(struct nfs_client *clp)
372{
373 struct pnfs_layout_hdr *lo;
374 LIST_HEAD(tmp_list);
375
376 spin_lock(&clp->cl_lock);
377 list_splice_init(&clp->cl_layouts, &tmp_list);
378 spin_unlock(&clp->cl_lock);
379
380 while (!list_empty(&tmp_list)) {
381 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000382 plh_layouts);
Andy Adamson974cec82010-10-20 00:18:02 -0400383 dprintk("%s freeing layout for inode %lu\n", __func__,
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000384 lo->plh_inode->i_ino);
385 pnfs_destroy_layout(NFS_I(lo->plh_inode));
Andy Adamson974cec82010-10-20 00:18:02 -0400386 }
387}
388
Fred Isamanfd6002e2011-01-06 11:36:22 +0000389/* update lo->plh_stateid with new if is more recent */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000390void
391pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
392 bool update_barrier)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400393{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000394 u32 oldseq, newseq;
Andy Adamson974cec82010-10-20 00:18:02 -0400395
Fred Isamanfd6002e2011-01-06 11:36:22 +0000396 oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid);
397 newseq = be32_to_cpu(new->stateid.seqid);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000398 if ((int)(newseq - oldseq) > 0) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000399 memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid));
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000400 if (update_barrier) {
401 u32 new_barrier = be32_to_cpu(new->stateid.seqid);
402
403 if ((int)(new_barrier - lo->plh_barrier))
404 lo->plh_barrier = new_barrier;
405 } else {
406 /* Because of wraparound, we want to keep the barrier
407 * "close" to the current seqids. It needs to be
408 * within 2**31 to count as "behind", so if it
409 * gets too near that limit, give us a litle leeway
410 * and bring it to within 2**30.
411 * NOTE - and yes, this is all unsigned arithmetic.
412 */
413 if (unlikely((newseq - lo->plh_barrier) > (3 << 29)))
414 lo->plh_barrier = newseq - (1 << 30);
415 }
416 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400417}
418
Fred Isamancf7d63f2011-01-06 11:36:25 +0000419/* lget is set to 1 if called from inside send_layoutget call chain */
420static bool
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000421pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid,
422 int lget)
Fred Isamancf7d63f2011-01-06 11:36:25 +0000423{
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000424 if ((stateid) &&
425 (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0)
426 return true;
Fred Isamanf7e89172011-01-06 11:36:32 +0000427 return lo->plh_block_lgets ||
Fred Isaman38511722011-02-03 18:28:50 +0000428 test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) ||
Fred Isamanf7e89172011-01-06 11:36:32 +0000429 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000430 (list_empty(&lo->plh_segs) &&
Fred Isamancf7d63f2011-01-06 11:36:25 +0000431 (atomic_read(&lo->plh_outstanding) > lget));
432}
433
Fred Isamanfd6002e2011-01-06 11:36:22 +0000434int
435pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
436 struct nfs4_state *open_state)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400437{
Fred Isamanfd6002e2011-01-06 11:36:22 +0000438 int status = 0;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400439
440 dprintk("--> %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000441 spin_lock(&lo->plh_inode->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000442 if (pnfs_layoutgets_blocked(lo, NULL, 1)) {
Fred Isamancf7d63f2011-01-06 11:36:25 +0000443 status = -EAGAIN;
444 } else if (list_empty(&lo->plh_segs)) {
Fred Isamanfd6002e2011-01-06 11:36:22 +0000445 int seq;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400446
Fred Isamanfd6002e2011-01-06 11:36:22 +0000447 do {
448 seq = read_seqbegin(&open_state->seqlock);
449 memcpy(dst->data, open_state->stateid.data,
450 sizeof(open_state->stateid.data));
451 } while (read_seqretry(&open_state->seqlock, seq));
452 } else
453 memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data));
454 spin_unlock(&lo->plh_inode->i_lock);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400455 dprintk("<-- %s\n", __func__);
Fred Isamanfd6002e2011-01-06 11:36:22 +0000456 return status;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400457}
458
459/*
460* Get layout from server.
461* for now, assume that whole file layouts are requested.
462* arg->offset: 0
463* arg->length: all ones
464*/
Benny Halevye5e94012010-10-20 00:18:01 -0400465static struct pnfs_layout_segment *
466send_layoutget(struct pnfs_layout_hdr *lo,
467 struct nfs_open_context *ctx,
468 u32 iomode)
469{
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000470 struct inode *ino = lo->plh_inode;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400471 struct nfs_server *server = NFS_SERVER(ino);
472 struct nfs4_layoutget *lgp;
473 struct pnfs_layout_segment *lseg = NULL;
Benny Halevye5e94012010-10-20 00:18:01 -0400474
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400475 dprintk("--> %s\n", __func__);
476
477 BUG_ON(ctx == NULL);
478 lgp = kzalloc(sizeof(*lgp), GFP_KERNEL);
Fred Isamancf7d63f2011-01-06 11:36:25 +0000479 if (lgp == NULL)
Andy Adamson974cec82010-10-20 00:18:02 -0400480 return NULL;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400481 lgp->args.minlength = NFS4_MAX_UINT64;
482 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
483 lgp->args.range.iomode = iomode;
484 lgp->args.range.offset = 0;
485 lgp->args.range.length = NFS4_MAX_UINT64;
486 lgp->args.type = server->pnfs_curr_ld->id;
487 lgp->args.inode = ino;
488 lgp->args.ctx = get_nfs_open_context(ctx);
489 lgp->lsegpp = &lseg;
490
491 /* Synchronously retrieve layout information from server and
492 * store in lseg.
493 */
494 nfs4_proc_layoutget(lgp);
495 if (!lseg) {
496 /* remember that LAYOUTGET failed and suspend trying */
Fred Isaman566052c2011-01-06 11:36:20 +0000497 set_bit(lo_fail_bit(iomode), &lo->plh_flags);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400498 }
Andy Adamson974cec82010-10-20 00:18:02 -0400499 return lseg;
500}
501
Fred Isamanf7e89172011-01-06 11:36:32 +0000502bool pnfs_roc(struct inode *ino)
503{
504 struct pnfs_layout_hdr *lo;
505 struct pnfs_layout_segment *lseg, *tmp;
506 LIST_HEAD(tmp_list);
507 bool found = false;
508
509 spin_lock(&ino->i_lock);
510 lo = NFS_I(ino)->layout;
511 if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
512 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
513 goto out_nolayout;
514 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
515 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
516 mark_lseg_invalid(lseg, &tmp_list);
517 found = true;
518 }
519 if (!found)
520 goto out_nolayout;
521 lo->plh_block_lgets++;
522 get_layout_hdr(lo); /* matched in pnfs_roc_release */
523 spin_unlock(&ino->i_lock);
524 pnfs_free_lseg_list(&tmp_list);
525 return true;
526
527out_nolayout:
528 spin_unlock(&ino->i_lock);
529 return false;
530}
531
532void pnfs_roc_release(struct inode *ino)
533{
534 struct pnfs_layout_hdr *lo;
535
536 spin_lock(&ino->i_lock);
537 lo = NFS_I(ino)->layout;
538 lo->plh_block_lgets--;
539 put_layout_hdr_locked(lo);
540 spin_unlock(&ino->i_lock);
541}
542
543void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
544{
545 struct pnfs_layout_hdr *lo;
546
547 spin_lock(&ino->i_lock);
548 lo = NFS_I(ino)->layout;
549 if ((int)(barrier - lo->plh_barrier) > 0)
550 lo->plh_barrier = barrier;
551 spin_unlock(&ino->i_lock);
552}
553
554bool pnfs_roc_drain(struct inode *ino, u32 *barrier)
555{
556 struct nfs_inode *nfsi = NFS_I(ino);
557 struct pnfs_layout_segment *lseg;
558 bool found = false;
559
560 spin_lock(&ino->i_lock);
561 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
562 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
563 found = true;
564 break;
565 }
566 if (!found) {
567 struct pnfs_layout_hdr *lo = nfsi->layout;
568 u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid);
569
570 /* Since close does not return a layout stateid for use as
571 * a barrier, we choose the worst-case barrier.
572 */
573 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
574 }
575 spin_unlock(&ino->i_lock);
576 return found;
577}
578
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400579/*
580 * Compare two layout segments for sorting into layout cache.
581 * We want to preferentially return RW over RO layouts, so ensure those
582 * are seen first.
583 */
584static s64
585cmp_layout(u32 iomode1, u32 iomode2)
586{
587 /* read > read/write */
588 return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ);
589}
590
Andy Adamson974cec82010-10-20 00:18:02 -0400591static void
592pnfs_insert_layout(struct pnfs_layout_hdr *lo,
593 struct pnfs_layout_segment *lseg)
594{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400595 struct pnfs_layout_segment *lp;
596 int found = 0;
597
Andy Adamson974cec82010-10-20 00:18:02 -0400598 dprintk("%s:Begin\n", __func__);
599
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000600 assert_spin_locked(&lo->plh_inode->i_lock);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000601 list_for_each_entry(lp, &lo->plh_segs, pls_list) {
Fred Isaman566052c2011-01-06 11:36:20 +0000602 if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400603 continue;
Fred Isaman566052c2011-01-06 11:36:20 +0000604 list_add_tail(&lseg->pls_list, &lp->pls_list);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400605 dprintk("%s: inserted lseg %p "
606 "iomode %d offset %llu length %llu before "
607 "lp %p iomode %d offset %llu length %llu\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000608 __func__, lseg, lseg->pls_range.iomode,
609 lseg->pls_range.offset, lseg->pls_range.length,
610 lp, lp->pls_range.iomode, lp->pls_range.offset,
611 lp->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400612 found = 1;
613 break;
Andy Adamson974cec82010-10-20 00:18:02 -0400614 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400615 if (!found) {
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000616 list_add_tail(&lseg->pls_list, &lo->plh_segs);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400617 dprintk("%s: inserted lseg %p "
618 "iomode %d offset %llu length %llu at tail\n",
Fred Isaman566052c2011-01-06 11:36:20 +0000619 __func__, lseg, lseg->pls_range.iomode,
620 lseg->pls_range.offset, lseg->pls_range.length);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400621 }
Fred Isamancc6e5342011-01-06 11:36:28 +0000622 get_layout_hdr(lo);
Andy Adamson974cec82010-10-20 00:18:02 -0400623
624 dprintk("%s:Return\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400625}
626
627static struct pnfs_layout_hdr *
628alloc_init_layout_hdr(struct inode *ino)
629{
630 struct pnfs_layout_hdr *lo;
631
632 lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL);
633 if (!lo)
634 return NULL;
Fred Isamancc6e5342011-01-06 11:36:28 +0000635 atomic_set(&lo->plh_refcount, 1);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000636 INIT_LIST_HEAD(&lo->plh_layouts);
637 INIT_LIST_HEAD(&lo->plh_segs);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000638 INIT_LIST_HEAD(&lo->plh_bulk_recall);
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000639 lo->plh_inode = ino;
Benny Halevye5e94012010-10-20 00:18:01 -0400640 return lo;
641}
642
643static struct pnfs_layout_hdr *
644pnfs_find_alloc_layout(struct inode *ino)
645{
646 struct nfs_inode *nfsi = NFS_I(ino);
647 struct pnfs_layout_hdr *new = NULL;
648
649 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
650
651 assert_spin_locked(&ino->i_lock);
Fred Isaman4541d162011-01-06 11:36:23 +0000652 if (nfsi->layout) {
653 if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags))
654 return NULL;
655 else
656 return nfsi->layout;
657 }
Benny Halevye5e94012010-10-20 00:18:01 -0400658 spin_unlock(&ino->i_lock);
659 new = alloc_init_layout_hdr(ino);
660 spin_lock(&ino->i_lock);
661
662 if (likely(nfsi->layout == NULL)) /* Won the race? */
663 nfsi->layout = new;
664 else
665 kfree(new);
666 return nfsi->layout;
667}
668
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400669/*
670 * iomode matching rules:
671 * iomode lseg match
672 * ----- ----- -----
673 * ANY READ true
674 * ANY RW true
675 * RW READ false
676 * RW RW true
677 * READ READ true
678 * READ RW true
679 */
680static int
681is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode)
682{
Fred Isaman566052c2011-01-06 11:36:20 +0000683 return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400684}
685
686/*
687 * lookup range in layout
688 */
Benny Halevye5e94012010-10-20 00:18:01 -0400689static struct pnfs_layout_segment *
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000690pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode)
Benny Halevye5e94012010-10-20 00:18:01 -0400691{
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400692 struct pnfs_layout_segment *lseg, *ret = NULL;
693
694 dprintk("%s:Begin\n", __func__);
695
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000696 assert_spin_locked(&lo->plh_inode->i_lock);
697 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
Fred Isaman4541d162011-01-06 11:36:23 +0000698 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
699 is_matching_lseg(lseg, iomode)) {
Fred Isamand684d2a2011-03-01 01:34:13 +0000700 ret = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400701 break;
702 }
Fred Isaman566052c2011-01-06 11:36:20 +0000703 if (cmp_layout(iomode, lseg->pls_range.iomode) > 0)
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400704 break;
705 }
706
707 dprintk("%s:Return lseg %p ref %d\n",
Fred Isaman4541d162011-01-06 11:36:23 +0000708 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400709 return ret;
Benny Halevye5e94012010-10-20 00:18:01 -0400710}
711
712/*
713 * Layout segment is retreived from the server if not cached.
714 * The appropriate layout segment is referenced and returned to the caller.
715 */
716struct pnfs_layout_segment *
717pnfs_update_layout(struct inode *ino,
718 struct nfs_open_context *ctx,
719 enum pnfs_iomode iomode)
720{
721 struct nfs_inode *nfsi = NFS_I(ino);
Fred Isaman2130ff62011-01-06 11:36:26 +0000722 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Benny Halevye5e94012010-10-20 00:18:01 -0400723 struct pnfs_layout_hdr *lo;
724 struct pnfs_layout_segment *lseg = NULL;
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000725 bool first = false;
Benny Halevye5e94012010-10-20 00:18:01 -0400726
727 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
728 return NULL;
729 spin_lock(&ino->i_lock);
730 lo = pnfs_find_alloc_layout(ino);
731 if (lo == NULL) {
732 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
733 goto out_unlock;
734 }
735
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000736 /* Do we even need to bother with this? */
737 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
738 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
739 dprintk("%s matches recall, use MDS\n", __func__);
Benny Halevye5e94012010-10-20 00:18:01 -0400740 goto out_unlock;
741 }
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000742 /* Check to see if the layout for the given range already exists */
743 lseg = pnfs_find_lseg(lo, iomode);
744 if (lseg)
745 goto out_unlock;
Benny Halevye5e94012010-10-20 00:18:01 -0400746
747 /* if LAYOUTGET already failed once we don't try again */
Fred Isaman566052c2011-01-06 11:36:20 +0000748 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags))
Benny Halevye5e94012010-10-20 00:18:01 -0400749 goto out_unlock;
750
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000751 if (pnfs_layoutgets_blocked(lo, NULL, 0))
Fred Isamancf7d63f2011-01-06 11:36:25 +0000752 goto out_unlock;
753 atomic_inc(&lo->plh_outstanding);
754
Fred Isamancc6e5342011-01-06 11:36:28 +0000755 get_layout_hdr(lo);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000756 if (list_empty(&lo->plh_segs))
757 first = true;
758 spin_unlock(&ino->i_lock);
759 if (first) {
Fred Isaman2130ff62011-01-06 11:36:26 +0000760 /* The lo must be on the clp list if there is any
761 * chance of a CB_LAYOUTRECALL(FILE) coming in.
762 */
763 spin_lock(&clp->cl_lock);
764 BUG_ON(!list_empty(&lo->plh_layouts));
765 list_add_tail(&lo->plh_layouts, &clp->cl_layouts);
766 spin_unlock(&clp->cl_lock);
767 }
Benny Halevye5e94012010-10-20 00:18:01 -0400768
769 lseg = send_layoutget(lo, ctx, iomode);
Fred Isamanf49f9ba2011-02-03 18:28:52 +0000770 if (!lseg && first) {
771 spin_lock(&clp->cl_lock);
772 list_del_init(&lo->plh_layouts);
773 spin_unlock(&clp->cl_lock);
Fred Isaman2130ff62011-01-06 11:36:26 +0000774 }
Fred Isamancf7d63f2011-01-06 11:36:25 +0000775 atomic_dec(&lo->plh_outstanding);
Fred Isamancc6e5342011-01-06 11:36:28 +0000776 put_layout_hdr(lo);
Benny Halevye5e94012010-10-20 00:18:01 -0400777out:
778 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
Andy Adamsonbf9c1382011-03-01 01:34:07 +0000779 nfsi->layout ? nfsi->layout->plh_flags : -1, lseg);
Benny Halevye5e94012010-10-20 00:18:01 -0400780 return lseg;
781out_unlock:
782 spin_unlock(&ino->i_lock);
783 goto out;
784}
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400785
786int
787pnfs_layout_process(struct nfs4_layoutget *lgp)
788{
789 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
790 struct nfs4_layoutget_res *res = &lgp->res;
791 struct pnfs_layout_segment *lseg;
Fred Isamanb7edfaa2011-01-06 11:36:21 +0000792 struct inode *ino = lo->plh_inode;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000793 struct nfs_client *clp = NFS_SERVER(ino)->nfs_client;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400794 int status = 0;
795
Fred Isamanfc1794c2011-01-06 11:36:27 +0000796 /* Verify we got what we asked for.
797 * Note that because the xdr parsing only accepts a single
798 * element array, this can fail even if the server is behaving
799 * correctly.
800 */
801 if (lgp->args.range.iomode > res->range.iomode ||
802 res->range.offset != 0 ||
803 res->range.length != NFS4_MAX_UINT64) {
804 status = -EINVAL;
805 goto out;
806 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400807 /* Inject layout blob into I/O device driver */
808 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res);
809 if (!lseg || IS_ERR(lseg)) {
810 if (!lseg)
811 status = -ENOMEM;
812 else
813 status = PTR_ERR(lseg);
814 dprintk("%s: Could not allocate layout: error %d\n",
815 __func__, status);
816 goto out;
817 }
818
819 spin_lock(&ino->i_lock);
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000820 if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) ||
821 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
822 dprintk("%s forget reply due to recall\n", __func__);
823 goto out_forget_reply;
824 }
825
826 if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) {
827 dprintk("%s forget reply due to state\n", __func__);
828 goto out_forget_reply;
829 }
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400830 init_lseg(lo, lseg);
Fred Isaman566052c2011-01-06 11:36:20 +0000831 lseg->pls_range = res->range;
Fred Isamand684d2a2011-03-01 01:34:13 +0000832 *lgp->lsegpp = get_lseg(lseg);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400833 pnfs_insert_layout(lo, lseg);
834
Fred Isamanf7e89172011-01-06 11:36:32 +0000835 if (res->return_on_close) {
836 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
837 set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
838 }
839
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400840 /* Done processing layoutget. Set the layout stateid */
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000841 pnfs_set_layout_stateid(lo, &res->stateid, false);
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400842 spin_unlock(&ino->i_lock);
843out:
844 return status;
Fred Isaman43f1b3d2011-01-06 11:36:30 +0000845
846out_forget_reply:
847 spin_unlock(&ino->i_lock);
848 lseg->pls_layout = lo;
849 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
850 goto out;
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400851}
852
Fred Isamanbae724e2011-03-01 01:34:15 +0000853static int pnfs_read_pg_test(struct nfs_pageio_descriptor *pgio,
854 struct nfs_page *prev,
855 struct nfs_page *req)
856{
857 if (pgio->pg_count == prev->wb_bytes) {
858 /* This is first coelesce call for a series of nfs_pages */
859 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
860 prev->wb_context,
861 IOMODE_READ);
862 }
863 return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req);
864}
865
866void
867pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode)
Fred Isaman94ad1c82011-03-01 01:34:14 +0000868{
869 struct pnfs_layoutdriver_type *ld;
870
871 ld = NFS_SERVER(inode)->pnfs_curr_ld;
Fred Isamanbae724e2011-03-01 01:34:15 +0000872 pgio->pg_test = (ld && ld->pg_test) ? pnfs_read_pg_test : NULL;
Fred Isaman94ad1c82011-03-01 01:34:14 +0000873}
874
Andy Adamsonb1f69b72010-10-20 00:18:03 -0400875/*
Andy Adamson64419a92011-03-01 01:34:16 +0000876 * Call the appropriate parallel I/O subsystem read function.
877 */
878enum pnfs_try_status
879pnfs_try_to_read_data(struct nfs_read_data *rdata,
880 const struct rpc_call_ops *call_ops)
881{
882 struct inode *inode = rdata->inode;
883 struct nfs_server *nfss = NFS_SERVER(inode);
884 enum pnfs_try_status trypnfs;
885
886 rdata->mds_ops = call_ops;
887
888 dprintk("%s: Reading ino:%lu %u@%llu\n",
889 __func__, inode->i_ino, rdata->args.count, rdata->args.offset);
890
891 trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata);
892 if (trypnfs == PNFS_NOT_ATTEMPTED) {
893 put_lseg(rdata->lseg);
894 rdata->lseg = NULL;
895 } else {
896 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
897 }
898 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
899 return trypnfs;
900}