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