Commit | Line | Data |
---|---|---|
85e174ba RL |
1 | /* |
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> | |
493292dd | 31 | #include <linux/nfs_page.h> |
143cb494 | 32 | #include <linux/module.h> |
974cec8c | 33 | #include "internal.h" |
85e174ba | 34 | #include "pnfs.h" |
64419a9b | 35 | #include "iostat.h" |
85e174ba RL |
36 | |
37 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
25c75333 | 38 | #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) |
85e174ba | 39 | |
02c35fca FI |
40 | /* Locking: |
41 | * | |
42 | * pnfs_spinlock: | |
43 | * protects pnfs_modules_tbl. | |
44 | */ | |
45 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
46 | ||
47 | /* | |
48 | * pnfs_modules_tbl holds all pnfs modules | |
49 | */ | |
50 | static LIST_HEAD(pnfs_modules_tbl); | |
51 | ||
52 | /* Return the registered pnfs layout driver module matching given id */ | |
53 | static struct pnfs_layoutdriver_type * | |
54 | find_pnfs_driver_locked(u32 id) | |
55 | { | |
56 | struct pnfs_layoutdriver_type *local; | |
57 | ||
58 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
59 | if (local->id == id) | |
60 | goto out; | |
61 | local = NULL; | |
62 | out: | |
63 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
64 | return local; | |
65 | } | |
66 | ||
85e174ba RL |
67 | static struct pnfs_layoutdriver_type * |
68 | find_pnfs_driver(u32 id) | |
69 | { | |
02c35fca FI |
70 | struct pnfs_layoutdriver_type *local; |
71 | ||
72 | spin_lock(&pnfs_spinlock); | |
73 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
74 | if (local != NULL && !try_module_get(local->owner)) { |
75 | dprintk("%s: Could not grab reference on module\n", __func__); | |
76 | local = NULL; | |
77 | } | |
02c35fca FI |
78 | spin_unlock(&pnfs_spinlock); |
79 | return local; | |
85e174ba RL |
80 | } |
81 | ||
82 | void | |
83 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
84 | { | |
738fd0f3 BH |
85 | if (nfss->pnfs_curr_ld) { |
86 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
87 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
88 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
89 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
90 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 91 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 92 | } |
85e174ba RL |
93 | nfss->pnfs_curr_ld = NULL; |
94 | } | |
95 | ||
96 | /* | |
97 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
98 | * Currently only one pNFS layout driver per filesystem is supported. | |
99 | * | |
100 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
101 | */ | |
102 | void | |
738fd0f3 BH |
103 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
104 | u32 id) | |
85e174ba RL |
105 | { |
106 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
107 | ||
108 | if (id == 0) | |
109 | goto out_no_driver; | |
110 | if (!(server->nfs_client->cl_exchange_flags & | |
111 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
a030889a WAA |
112 | printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n", |
113 | __func__, id, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
114 | goto out_no_driver; |
115 | } | |
116 | ld_type = find_pnfs_driver(id); | |
117 | if (!ld_type) { | |
118 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
119 | ld_type = find_pnfs_driver(id); | |
120 | if (!ld_type) { | |
121 | dprintk("%s: No pNFS module found for %u.\n", | |
122 | __func__, id); | |
123 | goto out_no_driver; | |
124 | } | |
125 | } | |
126 | server->pnfs_curr_ld = ld_type; | |
738fd0f3 BH |
127 | if (ld_type->set_layoutdriver |
128 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
129 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
130 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
131 | module_put(ld_type->owner); |
132 | goto out_no_driver; | |
133 | } | |
2a4c8994 TM |
134 | /* Bump the MDS count */ |
135 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 136 | |
85e174ba RL |
137 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
138 | return; | |
139 | ||
140 | out_no_driver: | |
141 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
142 | server->pnfs_curr_ld = NULL; | |
143 | } | |
02c35fca FI |
144 | |
145 | int | |
146 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
147 | { | |
148 | int status = -EINVAL; | |
149 | struct pnfs_layoutdriver_type *tmp; | |
150 | ||
151 | if (ld_type->id == 0) { | |
a030889a | 152 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
153 | return status; |
154 | } | |
b1f69b75 | 155 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 156 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
157 | "alloc_lseg and free_lseg.\n", __func__); |
158 | return status; | |
159 | } | |
02c35fca FI |
160 | |
161 | spin_lock(&pnfs_spinlock); | |
162 | tmp = find_pnfs_driver_locked(ld_type->id); | |
163 | if (!tmp) { | |
164 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
165 | status = 0; | |
166 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
167 | ld_type->name); | |
168 | } else { | |
a030889a | 169 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
170 | __func__, ld_type->id); |
171 | } | |
172 | spin_unlock(&pnfs_spinlock); | |
173 | ||
174 | return status; | |
175 | } | |
176 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
177 | ||
178 | void | |
179 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
180 | { | |
181 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
182 | spin_lock(&pnfs_spinlock); | |
183 | list_del(&ld_type->pnfs_tblid); | |
184 | spin_unlock(&pnfs_spinlock); | |
185 | } | |
186 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 187 | |
b1f69b75 AA |
188 | /* |
189 | * pNFS client layout cache | |
190 | */ | |
191 | ||
cc6e5340 | 192 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 193 | void |
70c3bd2b | 194 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 195 | { |
cc6e5340 | 196 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
197 | } |
198 | ||
636fb9c8 BH |
199 | static struct pnfs_layout_hdr * |
200 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
201 | { | |
202 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
203 | return ld->alloc_layout_hdr ? ld->alloc_layout_hdr(ino, gfp_flags) : | |
204 | kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags); | |
205 | } | |
206 | ||
207 | static void | |
208 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
209 | { | |
210 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(lo->plh_inode)->pnfs_curr_ld; | |
9fa40758 | 211 | put_rpccred(lo->plh_lc_cred); |
636fb9c8 BH |
212 | return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo); |
213 | } | |
214 | ||
e5e94017 | 215 | static void |
cc6e5340 | 216 | destroy_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 217 | { |
bb346f63 | 218 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 FI |
219 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
220 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
bb346f63 TM |
221 | nfsi->layout = NULL; |
222 | /* Reset MDS Threshold I/O counters */ | |
223 | nfsi->write_io = 0; | |
224 | nfsi->read_io = 0; | |
636fb9c8 | 225 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 226 | } |
e5e94017 | 227 | |
cc6e5340 | 228 | static void |
70c3bd2b | 229 | pnfs_put_layout_hdr_locked(struct pnfs_layout_hdr *lo) |
cc6e5340 FI |
230 | { |
231 | if (atomic_dec_and_test(&lo->plh_refcount)) | |
232 | destroy_layout_hdr(lo); | |
e5e94017 BH |
233 | } |
234 | ||
b1f69b75 | 235 | void |
70c3bd2b | 236 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 237 | { |
cc6e5340 FI |
238 | struct inode *inode = lo->plh_inode; |
239 | ||
240 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
241 | destroy_layout_hdr(lo); | |
242 | spin_unlock(&inode->i_lock); | |
243 | } | |
974cec8c AA |
244 | } |
245 | ||
b9e028fd TM |
246 | static int |
247 | pnfs_iomode_to_fail_bit(u32 iomode) | |
248 | { | |
249 | return iomode == IOMODE_RW ? | |
250 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
251 | } | |
252 | ||
253 | static void | |
3e621214 | 254 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 255 | { |
25c75333 | 256 | lo->plh_retry_timestamp = jiffies; |
3e621214 TM |
257 | if (test_and_set_bit(fail_bit, &lo->plh_flags)) |
258 | atomic_inc(&lo->plh_refcount); | |
259 | } | |
260 | ||
261 | static void | |
262 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
263 | { | |
264 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
265 | atomic_dec(&lo->plh_refcount); | |
266 | } | |
267 | ||
268 | static void | |
269 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
270 | { | |
271 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
272 | struct pnfs_layout_range range = { |
273 | .iomode = iomode, | |
274 | .offset = 0, | |
275 | .length = NFS4_MAX_UINT64, | |
276 | }; | |
277 | LIST_HEAD(head); | |
3e621214 TM |
278 | |
279 | spin_lock(&inode->i_lock); | |
280 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
115ce575 | 281 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range); |
3e621214 | 282 | spin_unlock(&inode->i_lock); |
115ce575 | 283 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
284 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
285 | iomode == IOMODE_RW ? "RW" : "READ"); | |
286 | } | |
287 | ||
288 | static bool | |
289 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
290 | { | |
25c75333 | 291 | unsigned long start, end; |
3e621214 TM |
292 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
293 | ||
294 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
295 | return false; |
296 | end = jiffies; | |
297 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
298 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
299 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 300 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
301 | return false; |
302 | } | |
303 | return true; | |
b9e028fd TM |
304 | } |
305 | ||
974cec8c AA |
306 | static void |
307 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
308 | { | |
566052c5 | 309 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 310 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c FI |
311 | atomic_set(&lseg->pls_refcount, 1); |
312 | smp_mb(); | |
313 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 314 | lseg->pls_layout = lo; |
974cec8c AA |
315 | } |
316 | ||
4541d16c | 317 | static void free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 318 | { |
b7edfaa1 | 319 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 320 | |
b1f69b75 | 321 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
57036a37 | 322 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
70c3bd2b | 323 | pnfs_put_layout_hdr(NFS_I(ino)->layout); |
974cec8c AA |
324 | } |
325 | ||
d684d2ae | 326 | static void |
57036a37 TM |
327 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
328 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 329 | { |
57036a37 | 330 | struct inode *inode = lo->plh_inode; |
d684d2ae | 331 | |
d20581aa | 332 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 333 | list_del_init(&lseg->pls_list); |
57036a37 TM |
334 | if (list_empty(&lo->plh_segs)) { |
335 | set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags); | |
d684d2ae | 336 | /* Matched by initial refcount set in alloc_init_layout_hdr */ |
57036a37 | 337 | pnfs_put_layout_hdr_locked(lo); |
d684d2ae FI |
338 | } |
339 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); | |
340 | } | |
341 | ||
bae724ef | 342 | void |
9369a431 | 343 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 344 | { |
57036a37 | 345 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
346 | struct inode *inode; |
347 | ||
348 | if (!lseg) | |
349 | return; | |
350 | ||
4541d16c FI |
351 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
352 | atomic_read(&lseg->pls_refcount), | |
353 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
57036a37 TM |
354 | lo = lseg->pls_layout; |
355 | inode = lo->plh_inode; | |
d684d2ae FI |
356 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
357 | LIST_HEAD(free_me); | |
4541d16c | 358 | |
57036a37 | 359 | pnfs_layout_remove_lseg(lo, lseg); |
d684d2ae | 360 | spin_unlock(&inode->i_lock); |
57036a37 | 361 | list_add(&lseg->pls_list, &free_me); |
d684d2ae | 362 | pnfs_free_lseg_list(&free_me); |
4541d16c | 363 | } |
4541d16c | 364 | } |
9369a431 | 365 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 366 | |
fb3296eb BH |
367 | static inline u64 |
368 | end_offset(u64 start, u64 len) | |
369 | { | |
370 | u64 end; | |
371 | ||
372 | end = start + len; | |
373 | return end >= start ? end : NFS4_MAX_UINT64; | |
374 | } | |
375 | ||
376 | /* last octet in a range */ | |
377 | static inline u64 | |
378 | last_byte_offset(u64 start, u64 len) | |
379 | { | |
380 | u64 end; | |
381 | ||
382 | BUG_ON(!len); | |
383 | end = start + len; | |
384 | return end > start ? end - 1 : NFS4_MAX_UINT64; | |
385 | } | |
386 | ||
387 | /* | |
388 | * is l2 fully contained in l1? | |
389 | * start1 end1 | |
390 | * [----------------------------------) | |
391 | * start2 end2 | |
392 | * [----------------) | |
393 | */ | |
394 | static inline int | |
395 | lo_seg_contained(struct pnfs_layout_range *l1, | |
396 | struct pnfs_layout_range *l2) | |
397 | { | |
398 | u64 start1 = l1->offset; | |
399 | u64 end1 = end_offset(start1, l1->length); | |
400 | u64 start2 = l2->offset; | |
401 | u64 end2 = end_offset(start2, l2->length); | |
402 | ||
403 | return (start1 <= start2) && (end1 >= end2); | |
404 | } | |
405 | ||
406 | /* | |
407 | * is l1 and l2 intersecting? | |
408 | * start1 end1 | |
409 | * [----------------------------------) | |
410 | * start2 end2 | |
411 | * [----------------) | |
412 | */ | |
413 | static inline int | |
414 | lo_seg_intersecting(struct pnfs_layout_range *l1, | |
415 | struct pnfs_layout_range *l2) | |
416 | { | |
417 | u64 start1 = l1->offset; | |
418 | u64 end1 = end_offset(start1, l1->length); | |
419 | u64 start2 = l2->offset; | |
420 | u64 end2 = end_offset(start2, l2->length); | |
421 | ||
422 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
423 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
424 | } | |
425 | ||
4541d16c | 426 | static bool |
778b5502 BH |
427 | should_free_lseg(struct pnfs_layout_range *lseg_range, |
428 | struct pnfs_layout_range *recall_range) | |
4541d16c | 429 | { |
778b5502 BH |
430 | return (recall_range->iomode == IOMODE_ANY || |
431 | lseg_range->iomode == recall_range->iomode) && | |
432 | lo_seg_intersecting(lseg_range, recall_range); | |
974cec8c AA |
433 | } |
434 | ||
4541d16c FI |
435 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
436 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
437 | struct list_head *tmp_list) | |
438 | { | |
439 | int rv = 0; | |
440 | ||
441 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
442 | /* Remove the reference keeping the lseg in the | |
443 | * list. It will now be removed when all | |
444 | * outstanding io is finished. | |
445 | */ | |
d684d2ae FI |
446 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
447 | atomic_read(&lseg->pls_refcount)); | |
448 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
57036a37 | 449 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); |
d684d2ae FI |
450 | list_add(&lseg->pls_list, tmp_list); |
451 | rv = 1; | |
452 | } | |
4541d16c FI |
453 | } |
454 | return rv; | |
455 | } | |
456 | ||
457 | /* Returns count of number of matching invalid lsegs remaining in list | |
458 | * after call. | |
459 | */ | |
43f1b3da | 460 | int |
49a85061 | 461 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 462 | struct list_head *tmp_list, |
778b5502 | 463 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
464 | { |
465 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 466 | int invalid = 0, removed = 0; |
974cec8c AA |
467 | |
468 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
469 | ||
38511722 FI |
470 | if (list_empty(&lo->plh_segs)) { |
471 | if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) | |
70c3bd2b | 472 | pnfs_put_layout_hdr_locked(lo); |
38511722 FI |
473 | return 0; |
474 | } | |
4541d16c | 475 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
476 | if (!recall_range || |
477 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
478 | dprintk("%s: freeing lseg %p iomode %d " |
479 | "offset %llu length %llu\n", __func__, | |
480 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
481 | lseg->pls_range.length); | |
482 | invalid++; | |
483 | removed += mark_lseg_invalid(lseg, tmp_list); | |
484 | } | |
485 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
486 | return invalid - removed; | |
974cec8c AA |
487 | } |
488 | ||
f49f9baa | 489 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 490 | void |
4541d16c | 491 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 492 | { |
4541d16c | 493 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
494 | struct pnfs_layout_hdr *lo; |
495 | ||
496 | if (list_empty(free_me)) | |
497 | return; | |
498 | ||
499 | lo = list_first_entry(free_me, struct pnfs_layout_segment, | |
500 | pls_list)->pls_layout; | |
974cec8c | 501 | |
f49f9baa FI |
502 | if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) { |
503 | struct nfs_client *clp; | |
504 | ||
505 | clp = NFS_SERVER(lo->plh_inode)->nfs_client; | |
506 | spin_lock(&clp->cl_lock); | |
507 | list_del_init(&lo->plh_layouts); | |
508 | spin_unlock(&clp->cl_lock); | |
509 | } | |
4541d16c | 510 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 511 | list_del(&lseg->pls_list); |
4541d16c | 512 | free_lseg(lseg); |
974cec8c AA |
513 | } |
514 | } | |
515 | ||
e5e94017 BH |
516 | void |
517 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
518 | { | |
519 | struct pnfs_layout_hdr *lo; | |
974cec8c | 520 | LIST_HEAD(tmp_list); |
e5e94017 BH |
521 | |
522 | spin_lock(&nfsi->vfs_inode.i_lock); | |
523 | lo = nfsi->layout; | |
524 | if (lo) { | |
38511722 | 525 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 526 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
3e621214 TM |
527 | pnfs_get_layout_hdr(lo); |
528 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); | |
529 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
530 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
531 | pnfs_free_lseg_list(&tmp_list); | |
532 | pnfs_put_layout_hdr(lo); | |
533 | } else | |
534 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 535 | } |
041245c8 | 536 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c AA |
537 | |
538 | /* | |
539 | * Called by the state manger to remove all layouts established under an | |
540 | * expired lease. | |
541 | */ | |
542 | void | |
543 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
544 | { | |
6382a441 | 545 | struct nfs_server *server; |
974cec8c AA |
546 | struct pnfs_layout_hdr *lo; |
547 | LIST_HEAD(tmp_list); | |
548 | ||
c47abcf8 AA |
549 | nfs4_deviceid_mark_client_invalid(clp); |
550 | nfs4_deviceid_purge_client(clp); | |
551 | ||
974cec8c | 552 | spin_lock(&clp->cl_lock); |
6382a441 WAA |
553 | rcu_read_lock(); |
554 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
555 | if (!list_empty(&server->layouts)) | |
556 | list_splice_init(&server->layouts, &tmp_list); | |
557 | } | |
558 | rcu_read_unlock(); | |
974cec8c AA |
559 | spin_unlock(&clp->cl_lock); |
560 | ||
561 | while (!list_empty(&tmp_list)) { | |
562 | lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, | |
b7edfaa1 | 563 | plh_layouts); |
974cec8c | 564 | dprintk("%s freeing layout for inode %lu\n", __func__, |
b7edfaa1 | 565 | lo->plh_inode->i_ino); |
2887fe45 | 566 | list_del_init(&lo->plh_layouts); |
b7edfaa1 | 567 | pnfs_destroy_layout(NFS_I(lo->plh_inode)); |
974cec8c | 568 | } |
e5e94017 BH |
569 | } |
570 | ||
fd6002e9 | 571 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
572 | void |
573 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
574 | bool update_barrier) | |
b1f69b75 | 575 | { |
fd6002e9 | 576 | u32 oldseq, newseq; |
b1f69b75 | 577 | |
2d2f24ad TM |
578 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
579 | newseq = be32_to_cpu(new->seqid); | |
43f1b3da | 580 | if ((int)(newseq - oldseq) > 0) { |
f597c537 | 581 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 582 | if (update_barrier) { |
2d2f24ad | 583 | u32 new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
584 | |
585 | if ((int)(new_barrier - lo->plh_barrier)) | |
586 | lo->plh_barrier = new_barrier; | |
587 | } else { | |
588 | /* Because of wraparound, we want to keep the barrier | |
589 | * "close" to the current seqids. It needs to be | |
590 | * within 2**31 to count as "behind", so if it | |
591 | * gets too near that limit, give us a litle leeway | |
592 | * and bring it to within 2**30. | |
593 | * NOTE - and yes, this is all unsigned arithmetic. | |
594 | */ | |
595 | if (unlikely((newseq - lo->plh_barrier) > (3 << 29))) | |
596 | lo->plh_barrier = newseq - (1 << 30); | |
597 | } | |
598 | } | |
b1f69b75 AA |
599 | } |
600 | ||
cf7d63f1 FI |
601 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
602 | static bool | |
43f1b3da FI |
603 | pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid, |
604 | int lget) | |
605 | { | |
606 | if ((stateid) && | |
2d2f24ad | 607 | (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0) |
43f1b3da | 608 | return true; |
f7e8917a | 609 | return lo->plh_block_lgets || |
38511722 | 610 | test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) || |
f7e8917a | 611 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || |
43f1b3da | 612 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
613 | (atomic_read(&lo->plh_outstanding) > lget)); |
614 | } | |
615 | ||
fd6002e9 FI |
616 | int |
617 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
618 | struct nfs4_state *open_state) | |
b1f69b75 | 619 | { |
fd6002e9 | 620 | int status = 0; |
974cec8c | 621 | |
b1f69b75 | 622 | dprintk("--> %s\n", __func__); |
fd6002e9 | 623 | spin_lock(&lo->plh_inode->i_lock); |
43f1b3da | 624 | if (pnfs_layoutgets_blocked(lo, NULL, 1)) { |
cf7d63f1 FI |
625 | status = -EAGAIN; |
626 | } else if (list_empty(&lo->plh_segs)) { | |
fd6002e9 FI |
627 | int seq; |
628 | ||
629 | do { | |
630 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 631 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
632 | } while (read_seqretry(&open_state->seqlock, seq)); |
633 | } else | |
f597c537 | 634 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 635 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 636 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 637 | return status; |
b1f69b75 AA |
638 | } |
639 | ||
640 | /* | |
641 | * Get layout from server. | |
642 | * for now, assume that whole file layouts are requested. | |
643 | * arg->offset: 0 | |
644 | * arg->length: all ones | |
645 | */ | |
e5e94017 BH |
646 | static struct pnfs_layout_segment * |
647 | send_layoutget(struct pnfs_layout_hdr *lo, | |
648 | struct nfs_open_context *ctx, | |
fb3296eb | 649 | struct pnfs_layout_range *range, |
a75b9df9 | 650 | gfp_t gfp_flags) |
e5e94017 | 651 | { |
b7edfaa1 | 652 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
653 | struct nfs_server *server = NFS_SERVER(ino); |
654 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 655 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
656 | |
657 | dprintk("--> %s\n", __func__); | |
e5e94017 | 658 | |
b1f69b75 | 659 | BUG_ON(ctx == NULL); |
a75b9df9 | 660 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 661 | if (lgp == NULL) |
b1f69b75 | 662 | return NULL; |
35124a09 | 663 | |
fb3296eb BH |
664 | lgp->args.minlength = PAGE_CACHE_SIZE; |
665 | if (lgp->args.minlength > range->length) | |
666 | lgp->args.minlength = range->length; | |
b1f69b75 | 667 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 668 | lgp->args.range = *range; |
b1f69b75 AA |
669 | lgp->args.type = server->pnfs_curr_ld->id; |
670 | lgp->args.inode = ino; | |
671 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 672 | lgp->gfp_flags = gfp_flags; |
b1f69b75 AA |
673 | |
674 | /* Synchronously retrieve layout information from server and | |
675 | * store in lseg. | |
676 | */ | |
a0b0a6e3 TM |
677 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
678 | if (IS_ERR(lseg)) { | |
679 | switch (PTR_ERR(lseg)) { | |
680 | case -ENOMEM: | |
681 | case -ERESTARTSYS: | |
682 | break; | |
683 | default: | |
684 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 685 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
686 | } |
687 | return NULL; | |
974cec8c | 688 | } |
35124a09 | 689 | |
974cec8c AA |
690 | return lseg; |
691 | } | |
692 | ||
293b3b06 AA |
693 | /* |
694 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
695 | * when the layout segment list is empty. | |
696 | * | |
697 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
698 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
699 | * deviceid is marked invalid. | |
700 | */ | |
cbe82603 BH |
701 | int |
702 | _pnfs_return_layout(struct inode *ino) | |
703 | { | |
704 | struct pnfs_layout_hdr *lo = NULL; | |
705 | struct nfs_inode *nfsi = NFS_I(ino); | |
706 | LIST_HEAD(tmp_list); | |
707 | struct nfs4_layoutreturn *lrp; | |
708 | nfs4_stateid stateid; | |
293b3b06 | 709 | int status = 0, empty; |
cbe82603 | 710 | |
366d5052 | 711 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
712 | |
713 | spin_lock(&ino->i_lock); | |
714 | lo = nfsi->layout; | |
366d5052 | 715 | if (!lo || pnfs_test_layout_returned(lo)) { |
cbe82603 | 716 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
717 | dprintk("NFS: %s no layout to return\n", __func__); |
718 | goto out; | |
cbe82603 BH |
719 | } |
720 | stateid = nfsi->layout->plh_stateid; | |
721 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 722 | pnfs_get_layout_hdr(lo); |
293b3b06 | 723 | empty = list_empty(&lo->plh_segs); |
49a85061 | 724 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
293b3b06 AA |
725 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
726 | if (empty) { | |
727 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 728 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
729 | dprintk("NFS: %s no layout segments to return\n", __func__); |
730 | goto out; | |
731 | } | |
ea0ded74 | 732 | lo->plh_block_lgets++; |
366d5052 | 733 | pnfs_mark_layout_returned(lo); |
cbe82603 BH |
734 | spin_unlock(&ino->i_lock); |
735 | pnfs_free_lseg_list(&tmp_list); | |
736 | ||
737 | WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)); | |
738 | ||
739 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); | |
740 | if (unlikely(lrp == NULL)) { | |
741 | status = -ENOMEM; | |
b9e028fd TM |
742 | pnfs_layout_io_set_failed(lo, IOMODE_RW); |
743 | pnfs_layout_io_set_failed(lo, IOMODE_READ); | |
366d5052 | 744 | pnfs_clear_layout_returned(lo); |
70c3bd2b | 745 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
746 | goto out; |
747 | } | |
748 | ||
749 | lrp->args.stateid = stateid; | |
750 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
751 | lrp->args.inode = ino; | |
a56aaa02 | 752 | lrp->args.layout = lo; |
cbe82603 BH |
753 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
754 | ||
755 | status = nfs4_proc_layoutreturn(lrp); | |
756 | out: | |
757 | dprintk("<-- %s status: %d\n", __func__, status); | |
758 | return status; | |
759 | } | |
0a57cdac | 760 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 761 | |
f7e8917a FI |
762 | bool pnfs_roc(struct inode *ino) |
763 | { | |
764 | struct pnfs_layout_hdr *lo; | |
765 | struct pnfs_layout_segment *lseg, *tmp; | |
766 | LIST_HEAD(tmp_list); | |
767 | bool found = false; | |
768 | ||
769 | spin_lock(&ino->i_lock); | |
770 | lo = NFS_I(ino)->layout; | |
771 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
772 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
773 | goto out_nolayout; | |
774 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
775 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
776 | mark_lseg_invalid(lseg, &tmp_list); | |
777 | found = true; | |
778 | } | |
779 | if (!found) | |
780 | goto out_nolayout; | |
781 | lo->plh_block_lgets++; | |
70c3bd2b | 782 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
783 | spin_unlock(&ino->i_lock); |
784 | pnfs_free_lseg_list(&tmp_list); | |
785 | return true; | |
786 | ||
787 | out_nolayout: | |
788 | spin_unlock(&ino->i_lock); | |
789 | return false; | |
790 | } | |
791 | ||
792 | void pnfs_roc_release(struct inode *ino) | |
793 | { | |
794 | struct pnfs_layout_hdr *lo; | |
795 | ||
796 | spin_lock(&ino->i_lock); | |
797 | lo = NFS_I(ino)->layout; | |
798 | lo->plh_block_lgets--; | |
70c3bd2b | 799 | pnfs_put_layout_hdr_locked(lo); |
f7e8917a FI |
800 | spin_unlock(&ino->i_lock); |
801 | } | |
802 | ||
803 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
804 | { | |
805 | struct pnfs_layout_hdr *lo; | |
806 | ||
807 | spin_lock(&ino->i_lock); | |
808 | lo = NFS_I(ino)->layout; | |
809 | if ((int)(barrier - lo->plh_barrier) > 0) | |
810 | lo->plh_barrier = barrier; | |
811 | spin_unlock(&ino->i_lock); | |
812 | } | |
813 | ||
7fdab069 | 814 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
815 | { |
816 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 817 | struct pnfs_layout_hdr *lo; |
f7e8917a | 818 | struct pnfs_layout_segment *lseg; |
7fdab069 | 819 | u32 current_seqid; |
f7e8917a FI |
820 | bool found = false; |
821 | ||
822 | spin_lock(&ino->i_lock); | |
823 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
824 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 825 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 826 | found = true; |
7fdab069 | 827 | goto out; |
f7e8917a | 828 | } |
7fdab069 TM |
829 | lo = nfsi->layout; |
830 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 831 | |
7fdab069 TM |
832 | /* Since close does not return a layout stateid for use as |
833 | * a barrier, we choose the worst-case barrier. | |
834 | */ | |
835 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
836 | out: | |
f7e8917a FI |
837 | spin_unlock(&ino->i_lock); |
838 | return found; | |
839 | } | |
840 | ||
b1f69b75 AA |
841 | /* |
842 | * Compare two layout segments for sorting into layout cache. | |
843 | * We want to preferentially return RW over RO layouts, so ensure those | |
844 | * are seen first. | |
845 | */ | |
846 | static s64 | |
fb3296eb BH |
847 | cmp_layout(struct pnfs_layout_range *l1, |
848 | struct pnfs_layout_range *l2) | |
b1f69b75 | 849 | { |
fb3296eb BH |
850 | s64 d; |
851 | ||
852 | /* high offset > low offset */ | |
853 | d = l1->offset - l2->offset; | |
854 | if (d) | |
855 | return d; | |
856 | ||
857 | /* short length > long length */ | |
858 | d = l2->length - l1->length; | |
859 | if (d) | |
860 | return d; | |
861 | ||
b1f69b75 | 862 | /* read > read/write */ |
fb3296eb | 863 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
864 | } |
865 | ||
974cec8c | 866 | static void |
57036a37 | 867 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
868 | struct pnfs_layout_segment *lseg) |
869 | { | |
b1f69b75 | 870 | struct pnfs_layout_segment *lp; |
b1f69b75 | 871 | |
974cec8c AA |
872 | dprintk("%s:Begin\n", __func__); |
873 | ||
b7edfaa1 | 874 | assert_spin_locked(&lo->plh_inode->i_lock); |
b7edfaa1 | 875 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
fb3296eb | 876 | if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 877 | continue; |
566052c5 | 878 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
879 | dprintk("%s: inserted lseg %p " |
880 | "iomode %d offset %llu length %llu before " | |
881 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
882 | __func__, lseg, lseg->pls_range.iomode, |
883 | lseg->pls_range.offset, lseg->pls_range.length, | |
884 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
885 | lp->pls_range.length); | |
fb3296eb | 886 | goto out; |
974cec8c | 887 | } |
fb3296eb BH |
888 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
889 | dprintk("%s: inserted lseg %p " | |
890 | "iomode %d offset %llu length %llu at tail\n", | |
891 | __func__, lseg, lseg->pls_range.iomode, | |
892 | lseg->pls_range.offset, lseg->pls_range.length); | |
893 | out: | |
70c3bd2b | 894 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
895 | |
896 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
897 | } |
898 | ||
899 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
900 | alloc_init_layout_hdr(struct inode *ino, |
901 | struct nfs_open_context *ctx, | |
902 | gfp_t gfp_flags) | |
e5e94017 BH |
903 | { |
904 | struct pnfs_layout_hdr *lo; | |
905 | ||
636fb9c8 | 906 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
907 | if (!lo) |
908 | return NULL; | |
cc6e5340 | 909 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
910 | INIT_LIST_HEAD(&lo->plh_layouts); |
911 | INIT_LIST_HEAD(&lo->plh_segs); | |
43f1b3da | 912 | INIT_LIST_HEAD(&lo->plh_bulk_recall); |
b7edfaa1 | 913 | lo->plh_inode = ino; |
9fa40758 | 914 | lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred); |
e5e94017 BH |
915 | return lo; |
916 | } | |
917 | ||
918 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
919 | pnfs_find_alloc_layout(struct inode *ino, |
920 | struct nfs_open_context *ctx, | |
921 | gfp_t gfp_flags) | |
e5e94017 BH |
922 | { |
923 | struct nfs_inode *nfsi = NFS_I(ino); | |
924 | struct pnfs_layout_hdr *new = NULL; | |
925 | ||
926 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
927 | ||
928 | assert_spin_locked(&ino->i_lock); | |
4541d16c FI |
929 | if (nfsi->layout) { |
930 | if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags)) | |
931 | return NULL; | |
932 | else | |
933 | return nfsi->layout; | |
934 | } | |
e5e94017 | 935 | spin_unlock(&ino->i_lock); |
9fa40758 | 936 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
937 | spin_lock(&ino->i_lock); |
938 | ||
939 | if (likely(nfsi->layout == NULL)) /* Won the race? */ | |
940 | nfsi->layout = new; | |
941 | else | |
636fb9c8 | 942 | pnfs_free_layout_hdr(new); |
e5e94017 BH |
943 | return nfsi->layout; |
944 | } | |
945 | ||
b1f69b75 AA |
946 | /* |
947 | * iomode matching rules: | |
948 | * iomode lseg match | |
949 | * ----- ----- ----- | |
950 | * ANY READ true | |
951 | * ANY RW true | |
952 | * RW READ false | |
953 | * RW RW true | |
954 | * READ READ true | |
955 | * READ RW true | |
956 | */ | |
957 | static int | |
fb3296eb BH |
958 | is_matching_lseg(struct pnfs_layout_range *ls_range, |
959 | struct pnfs_layout_range *range) | |
b1f69b75 | 960 | { |
fb3296eb BH |
961 | struct pnfs_layout_range range1; |
962 | ||
963 | if ((range->iomode == IOMODE_RW && | |
964 | ls_range->iomode != IOMODE_RW) || | |
965 | !lo_seg_intersecting(ls_range, range)) | |
966 | return 0; | |
967 | ||
968 | /* range1 covers only the first byte in the range */ | |
969 | range1 = *range; | |
970 | range1.length = 1; | |
971 | return lo_seg_contained(ls_range, &range1); | |
b1f69b75 AA |
972 | } |
973 | ||
974 | /* | |
975 | * lookup range in layout | |
976 | */ | |
e5e94017 | 977 | static struct pnfs_layout_segment * |
fb3296eb BH |
978 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
979 | struct pnfs_layout_range *range) | |
e5e94017 | 980 | { |
b1f69b75 AA |
981 | struct pnfs_layout_segment *lseg, *ret = NULL; |
982 | ||
983 | dprintk("%s:Begin\n", __func__); | |
984 | ||
b7edfaa1 FI |
985 | assert_spin_locked(&lo->plh_inode->i_lock); |
986 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
4541d16c | 987 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
fb3296eb | 988 | is_matching_lseg(&lseg->pls_range, range)) { |
9369a431 | 989 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
990 | break; |
991 | } | |
d771e3a4 | 992 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
993 | break; |
994 | } | |
995 | ||
996 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 997 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 998 | return ret; |
e5e94017 BH |
999 | } |
1000 | ||
d23d61c8 AA |
1001 | /* |
1002 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1003 | * to the MDS or over pNFS | |
1004 | * | |
1005 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1006 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1007 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1008 | * WRITE request. | |
1009 | * | |
1010 | * A return of true means use MDS I/O. | |
1011 | * | |
1012 | * From rfc 5661: | |
1013 | * If a file's size is smaller than the file size threshold, data accesses | |
1014 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1015 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1016 | * server. If both file size and I/O size are provided, the client SHOULD | |
1017 | * reach or exceed both thresholds before sending its read or write | |
1018 | * requests to the data server. | |
1019 | */ | |
1020 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1021 | struct inode *ino, int iomode) | |
1022 | { | |
1023 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1024 | struct nfs_inode *nfsi = NFS_I(ino); | |
1025 | loff_t fsize = i_size_read(ino); | |
1026 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1027 | ||
1028 | if (t == NULL) | |
1029 | return ret; | |
1030 | ||
1031 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1032 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1033 | ||
1034 | switch (iomode) { | |
1035 | case IOMODE_READ: | |
1036 | if (t->bm & THRESHOLD_RD) { | |
1037 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1038 | size_set = true; | |
1039 | if (fsize < t->rd_sz) | |
1040 | size = true; | |
1041 | } | |
1042 | if (t->bm & THRESHOLD_RD_IO) { | |
1043 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1044 | nfsi->read_io); | |
1045 | io_set = true; | |
1046 | if (nfsi->read_io < t->rd_io_sz) | |
1047 | io = true; | |
1048 | } | |
1049 | break; | |
1050 | case IOMODE_RW: | |
1051 | if (t->bm & THRESHOLD_WR) { | |
1052 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1053 | size_set = true; | |
1054 | if (fsize < t->wr_sz) | |
1055 | size = true; | |
1056 | } | |
1057 | if (t->bm & THRESHOLD_WR_IO) { | |
1058 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1059 | nfsi->write_io); | |
1060 | io_set = true; | |
1061 | if (nfsi->write_io < t->wr_io_sz) | |
1062 | io = true; | |
1063 | } | |
1064 | break; | |
1065 | } | |
1066 | if (size_set && io_set) { | |
1067 | if (size && io) | |
1068 | ret = true; | |
1069 | } else if (size || io) | |
1070 | ret = true; | |
1071 | ||
1072 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1073 | return ret; | |
1074 | } | |
1075 | ||
e5e94017 BH |
1076 | /* |
1077 | * Layout segment is retreived from the server if not cached. | |
1078 | * The appropriate layout segment is referenced and returned to the caller. | |
1079 | */ | |
7c24d948 | 1080 | struct pnfs_layout_segment * |
e5e94017 BH |
1081 | pnfs_update_layout(struct inode *ino, |
1082 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1083 | loff_t pos, |
1084 | u64 count, | |
a75b9df9 TM |
1085 | enum pnfs_iomode iomode, |
1086 | gfp_t gfp_flags) | |
e5e94017 | 1087 | { |
fb3296eb BH |
1088 | struct pnfs_layout_range arg = { |
1089 | .iomode = iomode, | |
1090 | .offset = pos, | |
1091 | .length = count, | |
1092 | }; | |
707ed5fd | 1093 | unsigned pg_offset; |
6382a441 WAA |
1094 | struct nfs_server *server = NFS_SERVER(ino); |
1095 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1096 | struct pnfs_layout_hdr *lo; |
1097 | struct pnfs_layout_segment *lseg = NULL; | |
f49f9baa | 1098 | bool first = false; |
e5e94017 BH |
1099 | |
1100 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1101 | goto out; |
d23d61c8 AA |
1102 | |
1103 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1104 | goto out; |
d23d61c8 | 1105 | |
e5e94017 | 1106 | spin_lock(&ino->i_lock); |
9fa40758 | 1107 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1108 | if (lo == NULL) { |
1109 | spin_unlock(&ino->i_lock); | |
1110 | goto out; | |
1111 | } | |
e5e94017 | 1112 | |
43f1b3da | 1113 | /* Do we even need to bother with this? */ |
a59c30ac | 1114 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1115 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1116 | goto out_unlock; |
1117 | } | |
1118 | ||
1119 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1120 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1121 | goto out_unlock; |
1122 | ||
568e8c49 | 1123 | /* Check to see if the layout for the given range already exists */ |
fb3296eb | 1124 | lseg = pnfs_find_lseg(lo, &arg); |
568e8c49 AA |
1125 | if (lseg) |
1126 | goto out_unlock; | |
1127 | ||
43f1b3da | 1128 | if (pnfs_layoutgets_blocked(lo, NULL, 0)) |
cf7d63f1 FI |
1129 | goto out_unlock; |
1130 | atomic_inc(&lo->plh_outstanding); | |
1131 | ||
70c3bd2b | 1132 | pnfs_get_layout_hdr(lo); |
f49f9baa FI |
1133 | if (list_empty(&lo->plh_segs)) |
1134 | first = true; | |
366d5052 AA |
1135 | |
1136 | /* Enable LAYOUTRETURNs */ | |
1137 | pnfs_clear_layout_returned(lo); | |
1138 | ||
f49f9baa FI |
1139 | spin_unlock(&ino->i_lock); |
1140 | if (first) { | |
2130ff66 FI |
1141 | /* The lo must be on the clp list if there is any |
1142 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1143 | */ | |
1144 | spin_lock(&clp->cl_lock); | |
1145 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
6382a441 | 1146 | list_add_tail(&lo->plh_layouts, &server->layouts); |
2130ff66 FI |
1147 | spin_unlock(&clp->cl_lock); |
1148 | } | |
e5e94017 | 1149 | |
707ed5fd BH |
1150 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1151 | if (pg_offset) { | |
1152 | arg.offset -= pg_offset; | |
1153 | arg.length += pg_offset; | |
1154 | } | |
7c24d948 AA |
1155 | if (arg.length != NFS4_MAX_UINT64) |
1156 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1157 | |
fb3296eb | 1158 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
f49f9baa FI |
1159 | if (!lseg && first) { |
1160 | spin_lock(&clp->cl_lock); | |
1161 | list_del_init(&lo->plh_layouts); | |
1162 | spin_unlock(&clp->cl_lock); | |
2130ff66 | 1163 | } |
cf7d63f1 | 1164 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1165 | out_put_layout_hdr: |
70c3bd2b | 1166 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1167 | out: |
f86bbcf8 TM |
1168 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1169 | "(%s, offset: %llu, length: %llu)\n", | |
1170 | __func__, ino->i_sb->s_id, | |
1171 | (unsigned long long)NFS_FILEID(ino), | |
1172 | lseg == NULL ? "not found" : "found", | |
1173 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1174 | (unsigned long long)pos, | |
1175 | (unsigned long long)count); | |
e5e94017 BH |
1176 | return lseg; |
1177 | out_unlock: | |
1178 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1179 | goto out_put_layout_hdr; |
e5e94017 | 1180 | } |
7c24d948 | 1181 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1182 | |
a0b0a6e3 | 1183 | struct pnfs_layout_segment * |
b1f69b75 AA |
1184 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1185 | { | |
1186 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1187 | struct nfs4_layoutget_res *res = &lgp->res; | |
1188 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1189 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
1190 | int status = 0; |
1191 | ||
1192 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1193 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1194 | if (!lseg || IS_ERR(lseg)) { |
1195 | if (!lseg) | |
1196 | status = -ENOMEM; | |
1197 | else | |
1198 | status = PTR_ERR(lseg); | |
1199 | dprintk("%s: Could not allocate layout: error %d\n", | |
1200 | __func__, status); | |
1201 | goto out; | |
1202 | } | |
1203 | ||
1204 | spin_lock(&ino->i_lock); | |
a59c30ac | 1205 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1206 | dprintk("%s forget reply due to recall\n", __func__); |
1207 | goto out_forget_reply; | |
1208 | } | |
1209 | ||
1210 | if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) { | |
1211 | dprintk("%s forget reply due to state\n", __func__); | |
1212 | goto out_forget_reply; | |
1213 | } | |
b1f69b75 | 1214 | init_lseg(lo, lseg); |
566052c5 | 1215 | lseg->pls_range = res->range; |
9369a431 | 1216 | pnfs_get_lseg(lseg); |
57036a37 | 1217 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1218 | |
f7e8917a FI |
1219 | if (res->return_on_close) { |
1220 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1221 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1222 | } | |
1223 | ||
b1f69b75 | 1224 | /* Done processing layoutget. Set the layout stateid */ |
43f1b3da | 1225 | pnfs_set_layout_stateid(lo, &res->stateid, false); |
b1f69b75 | 1226 | spin_unlock(&ino->i_lock); |
a0b0a6e3 | 1227 | return lseg; |
b1f69b75 | 1228 | out: |
a0b0a6e3 | 1229 | return ERR_PTR(status); |
43f1b3da FI |
1230 | |
1231 | out_forget_reply: | |
1232 | spin_unlock(&ino->i_lock); | |
1233 | lseg->pls_layout = lo; | |
1234 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1235 | goto out; | |
b1f69b75 AA |
1236 | } |
1237 | ||
d8007d4d TM |
1238 | void |
1239 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1240 | { | |
1241 | BUG_ON(pgio->pg_lseg != NULL); | |
1242 | ||
1825a0d0 FI |
1243 | if (req->wb_offset != req->wb_pgbase) { |
1244 | nfs_pageio_reset_read_mds(pgio); | |
1245 | return; | |
1246 | } | |
d8007d4d TM |
1247 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1248 | req->wb_context, | |
1249 | req_offset(req), | |
1250 | req->wb_bytes, | |
1251 | IOMODE_READ, | |
1252 | GFP_KERNEL); | |
e885de1a TM |
1253 | /* If no lseg, fall back to read through mds */ |
1254 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1255 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1256 | |
d8007d4d TM |
1257 | } |
1258 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1259 | ||
1260 | void | |
1261 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1262 | { | |
1263 | BUG_ON(pgio->pg_lseg != NULL); | |
1264 | ||
1825a0d0 FI |
1265 | if (req->wb_offset != req->wb_pgbase) { |
1266 | nfs_pageio_reset_write_mds(pgio); | |
1267 | return; | |
1268 | } | |
d8007d4d TM |
1269 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1270 | req->wb_context, | |
1271 | req_offset(req), | |
1272 | req->wb_bytes, | |
1273 | IOMODE_RW, | |
1274 | GFP_NOFS); | |
e885de1a TM |
1275 | /* If no lseg, fall back to write through mds */ |
1276 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1277 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1278 | } |
1279 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1280 | ||
1abb5088 | 1281 | void |
061ae2ed FI |
1282 | pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode, |
1283 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 TM |
1284 | { |
1285 | struct nfs_server *server = NFS_SERVER(inode); | |
1286 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
1287 | ||
1288 | if (ld == NULL) | |
1abb5088 BS |
1289 | nfs_pageio_init_read(pgio, inode, compl_ops); |
1290 | else | |
1291 | nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0); | |
1751c363 TM |
1292 | } |
1293 | ||
57208fa7 | 1294 | void |
061ae2ed FI |
1295 | pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode, |
1296 | int ioflags, | |
1297 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 TM |
1298 | { |
1299 | struct nfs_server *server = NFS_SERVER(inode); | |
1300 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
1301 | ||
1302 | if (ld == NULL) | |
57208fa7 BS |
1303 | nfs_pageio_init_write(pgio, inode, ioflags, compl_ops); |
1304 | else | |
1305 | nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags); | |
1751c363 TM |
1306 | } |
1307 | ||
18ad0a9f | 1308 | bool |
dfed206b BH |
1309 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
1310 | struct nfs_page *req) | |
94ad1c80 | 1311 | { |
d8007d4d TM |
1312 | if (pgio->pg_lseg == NULL) |
1313 | return nfs_generic_pg_test(pgio, prev, req); | |
94ad1c80 | 1314 | |
19982ba8 TM |
1315 | /* |
1316 | * Test if a nfs_page is fully contained in the pnfs_layout_range. | |
1317 | * Note that this test makes several assumptions: | |
1318 | * - that the previous nfs_page in the struct nfs_pageio_descriptor | |
1319 | * is known to lie within the range. | |
1320 | * - that the nfs_page being tested is known to be contiguous with the | |
1321 | * previous nfs_page. | |
1322 | * - Layout ranges are page aligned, so we only have to test the | |
1323 | * start offset of the request. | |
1324 | * | |
1325 | * Please also note that 'end_offset' is actually the offset of the | |
1326 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1327 | * | |
1328 | */ | |
1329 | return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset, | |
1330 | pgio->pg_lseg->pls_range.length); | |
94ad1c80 | 1331 | } |
89a58e32 | 1332 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1333 | |
e7dd79af | 1334 | int pnfs_write_done_resend_to_mds(struct inode *inode, |
061ae2ed FI |
1335 | struct list_head *head, |
1336 | const struct nfs_pgio_completion_ops *compl_ops) | |
e2fecb21 TM |
1337 | { |
1338 | struct nfs_pageio_descriptor pgio; | |
1339 | LIST_HEAD(failed); | |
1340 | ||
1341 | /* Resend all requests through the MDS */ | |
57208fa7 | 1342 | nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops); |
e2fecb21 TM |
1343 | while (!list_empty(head)) { |
1344 | struct nfs_page *req = nfs_list_entry(head->next); | |
1345 | ||
1346 | nfs_list_remove_request(req); | |
1347 | if (!nfs_pageio_add_request(&pgio, req)) | |
1348 | nfs_list_add_request(req, &failed); | |
1349 | } | |
1350 | nfs_pageio_complete(&pgio); | |
1351 | ||
1352 | if (!list_empty(&failed)) { | |
1353 | /* For some reason our attempt to resend pages. Mark the | |
1354 | * overall send request as having failed, and let | |
1355 | * nfs_writeback_release_full deal with the error. | |
1356 | */ | |
1357 | list_move(&failed, head); | |
1358 | return -EIO; | |
1359 | } | |
1360 | return 0; | |
1361 | } | |
e7dd79af | 1362 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1363 | |
1acbbb4e FI |
1364 | static void pnfs_ld_handle_write_error(struct nfs_write_data *data) |
1365 | { | |
cd841605 FI |
1366 | struct nfs_pgio_header *hdr = data->header; |
1367 | ||
1368 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1369 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1370 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 FI |
1371 | clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags); |
1372 | pnfs_return_layout(hdr->inode); | |
1acbbb4e | 1373 | } |
6c75dc0d FI |
1374 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
1375 | data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode, | |
061ae2ed FI |
1376 | &hdr->pages, |
1377 | hdr->completion_ops); | |
1acbbb4e FI |
1378 | } |
1379 | ||
d20581aa BH |
1380 | /* |
1381 | * Called by non rpc-based layout drivers | |
1382 | */ | |
8ce160c5 | 1383 | void pnfs_ld_write_done(struct nfs_write_data *data) |
44b83799 | 1384 | { |
cd841605 FI |
1385 | struct nfs_pgio_header *hdr = data->header; |
1386 | ||
1387 | if (!hdr->pnfs_error) { | |
d20581aa | 1388 | pnfs_set_layoutcommit(data); |
cd841605 | 1389 | hdr->mds_ops->rpc_call_done(&data->task, data); |
1acbbb4e FI |
1390 | } else |
1391 | pnfs_ld_handle_write_error(data); | |
cd841605 | 1392 | hdr->mds_ops->rpc_release(data); |
44b83799 | 1393 | } |
d20581aa | 1394 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1395 | |
dce81290 TM |
1396 | static void |
1397 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
1398 | struct nfs_write_data *data) | |
1399 | { | |
cd841605 FI |
1400 | struct nfs_pgio_header *hdr = data->header; |
1401 | ||
6c75dc0d FI |
1402 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1403 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1404 | nfs_pageio_reset_write_mds(desc); | |
1405 | desc->pg_recoalesce = 1; | |
1406 | } | |
dce81290 TM |
1407 | nfs_writedata_release(data); |
1408 | } | |
1409 | ||
1410 | static enum pnfs_try_status | |
0382b744 | 1411 | pnfs_try_to_write_data(struct nfs_write_data *wdata, |
dce81290 TM |
1412 | const struct rpc_call_ops *call_ops, |
1413 | struct pnfs_layout_segment *lseg, | |
1414 | int how) | |
0382b744 | 1415 | { |
cd841605 FI |
1416 | struct nfs_pgio_header *hdr = wdata->header; |
1417 | struct inode *inode = hdr->inode; | |
0382b744 AA |
1418 | enum pnfs_try_status trypnfs; |
1419 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1420 | ||
cd841605 | 1421 | hdr->mds_ops = call_ops; |
0382b744 AA |
1422 | |
1423 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
1424 | inode->i_ino, wdata->args.count, wdata->args.offset, how); | |
0382b744 | 1425 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how); |
6c75dc0d | 1426 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1427 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1428 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1429 | return trypnfs; | |
1430 | } | |
1431 | ||
dce81290 TM |
1432 | static void |
1433 | pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how) | |
1434 | { | |
1435 | struct nfs_write_data *data; | |
1436 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; | |
1437 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
1438 | ||
1439 | desc->pg_lseg = NULL; | |
1440 | while (!list_empty(head)) { | |
1441 | enum pnfs_try_status trypnfs; | |
1442 | ||
6c75dc0d | 1443 | data = list_first_entry(head, struct nfs_write_data, list); |
dce81290 TM |
1444 | list_del_init(&data->list); |
1445 | ||
1446 | trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how); | |
1447 | if (trypnfs == PNFS_NOT_ATTEMPTED) | |
1448 | pnfs_write_through_mds(desc, data); | |
1449 | } | |
9369a431 | 1450 | pnfs_put_lseg(lseg); |
dce81290 TM |
1451 | } |
1452 | ||
6c75dc0d FI |
1453 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1454 | { | |
9369a431 | 1455 | pnfs_put_lseg(hdr->lseg); |
6c75dc0d FI |
1456 | nfs_writehdr_free(hdr); |
1457 | } | |
89d77c8f | 1458 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1459 | |
dce81290 TM |
1460 | int |
1461 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1462 | { | |
6c75dc0d FI |
1463 | struct nfs_write_header *whdr; |
1464 | struct nfs_pgio_header *hdr; | |
dce81290 TM |
1465 | int ret; |
1466 | ||
6c75dc0d FI |
1467 | whdr = nfs_writehdr_alloc(); |
1468 | if (!whdr) { | |
9b5415b5 | 1469 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
9369a431 | 1470 | pnfs_put_lseg(desc->pg_lseg); |
dce81290 | 1471 | desc->pg_lseg = NULL; |
6c75dc0d | 1472 | return -ENOMEM; |
dce81290 | 1473 | } |
6c75dc0d FI |
1474 | hdr = &whdr->header; |
1475 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); | |
9369a431 | 1476 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
6c75dc0d FI |
1477 | atomic_inc(&hdr->refcnt); |
1478 | ret = nfs_generic_flush(desc, hdr); | |
1479 | if (ret != 0) { | |
9369a431 | 1480 | pnfs_put_lseg(desc->pg_lseg); |
6c75dc0d | 1481 | desc->pg_lseg = NULL; |
6c75dc0d FI |
1482 | } else |
1483 | pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags); | |
1484 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1485 | hdr->completion_ops->completion(hdr); |
6c75dc0d | 1486 | return ret; |
dce81290 TM |
1487 | } |
1488 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1489 | ||
e7dd79af | 1490 | int pnfs_read_done_resend_to_mds(struct inode *inode, |
061ae2ed FI |
1491 | struct list_head *head, |
1492 | const struct nfs_pgio_completion_ops *compl_ops) | |
62e4a769 TM |
1493 | { |
1494 | struct nfs_pageio_descriptor pgio; | |
1acbbb4e | 1495 | LIST_HEAD(failed); |
62e4a769 | 1496 | |
1acbbb4e | 1497 | /* Resend all requests through the MDS */ |
1abb5088 | 1498 | nfs_pageio_init_read(&pgio, inode, compl_ops); |
1acbbb4e FI |
1499 | while (!list_empty(head)) { |
1500 | struct nfs_page *req = nfs_list_entry(head->next); | |
62e4a769 TM |
1501 | |
1502 | nfs_list_remove_request(req); | |
1acbbb4e FI |
1503 | if (!nfs_pageio_add_request(&pgio, req)) |
1504 | nfs_list_add_request(req, &failed); | |
62e4a769 TM |
1505 | } |
1506 | nfs_pageio_complete(&pgio); | |
1acbbb4e FI |
1507 | |
1508 | if (!list_empty(&failed)) { | |
1509 | list_move(&failed, head); | |
1510 | return -EIO; | |
1511 | } | |
1512 | return 0; | |
1513 | } | |
e7dd79af | 1514 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e FI |
1515 | |
1516 | static void pnfs_ld_handle_read_error(struct nfs_read_data *data) | |
1517 | { | |
cd841605 FI |
1518 | struct nfs_pgio_header *hdr = data->header; |
1519 | ||
1520 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); | |
1521 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1522 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 FI |
1523 | clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags); |
1524 | pnfs_return_layout(hdr->inode); | |
1acbbb4e | 1525 | } |
4db6e0b7 FI |
1526 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
1527 | data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode, | |
061ae2ed FI |
1528 | &hdr->pages, |
1529 | hdr->completion_ops); | |
62e4a769 TM |
1530 | } |
1531 | ||
d20581aa BH |
1532 | /* |
1533 | * Called by non rpc-based layout drivers | |
1534 | */ | |
9b7eecdc | 1535 | void pnfs_ld_read_done(struct nfs_read_data *data) |
d20581aa | 1536 | { |
cd841605 FI |
1537 | struct nfs_pgio_header *hdr = data->header; |
1538 | ||
1539 | if (likely(!hdr->pnfs_error)) { | |
d20581aa | 1540 | __nfs4_read_done_cb(data); |
cd841605 | 1541 | hdr->mds_ops->rpc_call_done(&data->task, data); |
62e4a769 TM |
1542 | } else |
1543 | pnfs_ld_handle_read_error(data); | |
cd841605 | 1544 | hdr->mds_ops->rpc_release(data); |
d20581aa BH |
1545 | } |
1546 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1547 | ||
493292dd TM |
1548 | static void |
1549 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
1550 | struct nfs_read_data *data) | |
1551 | { | |
cd841605 FI |
1552 | struct nfs_pgio_header *hdr = data->header; |
1553 | ||
4db6e0b7 FI |
1554 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1555 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1556 | nfs_pageio_reset_read_mds(desc); | |
1557 | desc->pg_recoalesce = 1; | |
1558 | } | |
493292dd TM |
1559 | nfs_readdata_release(data); |
1560 | } | |
1561 | ||
64419a9b AA |
1562 | /* |
1563 | * Call the appropriate parallel I/O subsystem read function. | |
1564 | */ | |
493292dd | 1565 | static enum pnfs_try_status |
64419a9b | 1566 | pnfs_try_to_read_data(struct nfs_read_data *rdata, |
493292dd TM |
1567 | const struct rpc_call_ops *call_ops, |
1568 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1569 | { |
cd841605 FI |
1570 | struct nfs_pgio_header *hdr = rdata->header; |
1571 | struct inode *inode = hdr->inode; | |
64419a9b AA |
1572 | struct nfs_server *nfss = NFS_SERVER(inode); |
1573 | enum pnfs_try_status trypnfs; | |
1574 | ||
cd841605 | 1575 | hdr->mds_ops = call_ops; |
64419a9b AA |
1576 | |
1577 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
1578 | __func__, inode->i_ino, rdata->args.count, rdata->args.offset); | |
1579 | ||
1580 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata); | |
4db6e0b7 | 1581 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1582 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1583 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1584 | return trypnfs; | |
1585 | } | |
863a3c6c | 1586 | |
493292dd TM |
1587 | static void |
1588 | pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head) | |
1589 | { | |
1590 | struct nfs_read_data *data; | |
1591 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; | |
1592 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
1593 | ||
1594 | desc->pg_lseg = NULL; | |
1595 | while (!list_empty(head)) { | |
1596 | enum pnfs_try_status trypnfs; | |
1597 | ||
4db6e0b7 | 1598 | data = list_first_entry(head, struct nfs_read_data, list); |
493292dd TM |
1599 | list_del_init(&data->list); |
1600 | ||
1601 | trypnfs = pnfs_try_to_read_data(data, call_ops, lseg); | |
1602 | if (trypnfs == PNFS_NOT_ATTEMPTED) | |
1603 | pnfs_read_through_mds(desc, data); | |
1604 | } | |
9369a431 | 1605 | pnfs_put_lseg(lseg); |
493292dd TM |
1606 | } |
1607 | ||
4db6e0b7 FI |
1608 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1609 | { | |
9369a431 | 1610 | pnfs_put_lseg(hdr->lseg); |
4db6e0b7 FI |
1611 | nfs_readhdr_free(hdr); |
1612 | } | |
89d77c8f | 1613 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1614 | |
493292dd TM |
1615 | int |
1616 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1617 | { | |
4db6e0b7 FI |
1618 | struct nfs_read_header *rhdr; |
1619 | struct nfs_pgio_header *hdr; | |
493292dd TM |
1620 | int ret; |
1621 | ||
4db6e0b7 FI |
1622 | rhdr = nfs_readhdr_alloc(); |
1623 | if (!rhdr) { | |
061ae2ed | 1624 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
4db6e0b7 | 1625 | ret = -ENOMEM; |
9369a431 | 1626 | pnfs_put_lseg(desc->pg_lseg); |
493292dd TM |
1627 | desc->pg_lseg = NULL; |
1628 | return ret; | |
1629 | } | |
4db6e0b7 FI |
1630 | hdr = &rhdr->header; |
1631 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); | |
9369a431 | 1632 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
4db6e0b7 FI |
1633 | atomic_inc(&hdr->refcnt); |
1634 | ret = nfs_generic_pagein(desc, hdr); | |
1635 | if (ret != 0) { | |
9369a431 | 1636 | pnfs_put_lseg(desc->pg_lseg); |
4db6e0b7 | 1637 | desc->pg_lseg = NULL; |
4db6e0b7 FI |
1638 | } else |
1639 | pnfs_do_multiple_reads(desc, &hdr->rpc_list); | |
1640 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1641 | hdr->completion_ops->completion(hdr); |
4db6e0b7 | 1642 | return ret; |
493292dd TM |
1643 | } |
1644 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1645 | ||
863a3c6c | 1646 | /* |
a9bae566 | 1647 | * There can be multiple RW segments. |
863a3c6c | 1648 | */ |
a9bae566 | 1649 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1650 | { |
a9bae566 | 1651 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1652 | |
a9bae566 PT |
1653 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1654 | if (lseg->pls_range.iomode == IOMODE_RW && | |
1655 | test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
1656 | list_add(&lseg->pls_lc_list, listp); | |
1657 | } | |
863a3c6c AA |
1658 | } |
1659 | ||
1b0ae068 PT |
1660 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
1661 | { | |
b9e028fd | 1662 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
1663 | } |
1664 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
1665 | ||
863a3c6c AA |
1666 | void |
1667 | pnfs_set_layoutcommit(struct nfs_write_data *wdata) | |
1668 | { | |
cd841605 FI |
1669 | struct nfs_pgio_header *hdr = wdata->header; |
1670 | struct inode *inode = hdr->inode; | |
1671 | struct nfs_inode *nfsi = NFS_I(inode); | |
4b8ee2b8 | 1672 | loff_t end_pos = wdata->mds_offset + wdata->res.count; |
79a48a1f | 1673 | bool mark_as_dirty = false; |
863a3c6c | 1674 | |
cd841605 | 1675 | spin_lock(&inode->i_lock); |
863a3c6c | 1676 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 1677 | mark_as_dirty = true; |
863a3c6c | 1678 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 1679 | __func__, inode->i_ino); |
863a3c6c | 1680 | } |
cd841605 | 1681 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 1682 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 1683 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 1684 | } |
acff5880 PT |
1685 | if (end_pos > nfsi->layout->plh_lwb) |
1686 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 1687 | spin_unlock(&inode->i_lock); |
acff5880 | 1688 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 1689 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
1690 | |
1691 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1692 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1693 | if (mark_as_dirty) | |
cd841605 | 1694 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
1695 | } |
1696 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
1697 | ||
db29c089 AA |
1698 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
1699 | { | |
1700 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
1701 | ||
1702 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
1703 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
1704 | } | |
1705 | ||
de4b15c7 AA |
1706 | /* |
1707 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
1708 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
1709 | * data to disk to allow the server to recover the data if it crashes. | |
1710 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
1711 | * is off, and a COMMIT is sent to a data server, or | |
1712 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
1713 | */ | |
863a3c6c | 1714 | int |
ef311537 | 1715 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c AA |
1716 | { |
1717 | struct nfs4_layoutcommit_data *data; | |
1718 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c AA |
1719 | loff_t end_pos; |
1720 | int status = 0; | |
1721 | ||
1722 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); | |
1723 | ||
de4b15c7 AA |
1724 | if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1725 | return 0; | |
1726 | ||
863a3c6c AA |
1727 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ |
1728 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
de4b15c7 | 1729 | if (!data) { |
de4b15c7 AA |
1730 | status = -ENOMEM; |
1731 | goto out; | |
1732 | } | |
863a3c6c | 1733 | |
92407e75 PT |
1734 | if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1735 | goto out_free; | |
1736 | ||
1737 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { | |
1738 | if (!sync) { | |
1739 | status = -EAGAIN; | |
1740 | goto out_free; | |
1741 | } | |
1742 | status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING, | |
1743 | nfs_wait_bit_killable, TASK_KILLABLE); | |
1744 | if (status) | |
1745 | goto out_free; | |
1746 | } | |
1747 | ||
a9bae566 | 1748 | INIT_LIST_HEAD(&data->lseg_list); |
de4b15c7 | 1749 | spin_lock(&inode->i_lock); |
863a3c6c | 1750 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
92407e75 | 1751 | clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags); |
863a3c6c | 1752 | spin_unlock(&inode->i_lock); |
92407e75 PT |
1753 | wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING); |
1754 | goto out_free; | |
863a3c6c | 1755 | } |
a9bae566 PT |
1756 | |
1757 | pnfs_list_write_lseg(inode, &data->lseg_list); | |
863a3c6c | 1758 | |
acff5880 | 1759 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 1760 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 1761 | |
f597c537 | 1762 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
1763 | spin_unlock(&inode->i_lock); |
1764 | ||
1765 | data->args.inode = inode; | |
9fa40758 | 1766 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
1767 | nfs_fattr_init(&data->fattr); |
1768 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
1769 | data->res.fattr = &data->fattr; | |
1770 | data->args.lastbytewritten = end_pos - 1; | |
1771 | data->res.server = NFS_SERVER(inode); | |
1772 | ||
1773 | status = nfs4_proc_layoutcommit(data, sync); | |
1774 | out: | |
92407e75 PT |
1775 | if (status) |
1776 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
1777 | dprintk("<-- %s status %d\n", __func__, status); |
1778 | return status; | |
92407e75 PT |
1779 | out_free: |
1780 | kfree(data); | |
1781 | goto out; | |
863a3c6c | 1782 | } |
82be417a AA |
1783 | |
1784 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
1785 | { | |
1786 | struct nfs4_threshold *thp; | |
1787 | ||
1788 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
1789 | if (!thp) { | |
1790 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
1791 | return NULL; | |
1792 | } | |
1793 | return thp; | |
1794 | } |