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 | { | |
9c626381 TM |
210 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
211 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
212 | ||
213 | if (!list_empty(&lo->plh_layouts)) { | |
214 | struct nfs_client *clp = server->nfs_client; | |
215 | ||
216 | spin_lock(&clp->cl_lock); | |
217 | list_del_init(&lo->plh_layouts); | |
218 | spin_unlock(&clp->cl_lock); | |
219 | } | |
9fa40758 | 220 | put_rpccred(lo->plh_lc_cred); |
636fb9c8 BH |
221 | return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo); |
222 | } | |
223 | ||
e5e94017 | 224 | static void |
6622c3ea | 225 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 226 | { |
bb346f63 | 227 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 228 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
229 | nfsi->layout = NULL; |
230 | /* Reset MDS Threshold I/O counters */ | |
231 | nfsi->write_io = 0; | |
232 | nfsi->read_io = 0; | |
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)) { | |
6622c3ea | 241 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 242 | spin_unlock(&inode->i_lock); |
6622c3ea | 243 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 244 | } |
974cec8c AA |
245 | } |
246 | ||
b9e028fd TM |
247 | static int |
248 | pnfs_iomode_to_fail_bit(u32 iomode) | |
249 | { | |
250 | return iomode == IOMODE_RW ? | |
251 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
252 | } | |
253 | ||
254 | static void | |
3e621214 | 255 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 256 | { |
25c75333 | 257 | lo->plh_retry_timestamp = jiffies; |
3e621214 TM |
258 | if (test_and_set_bit(fail_bit, &lo->plh_flags)) |
259 | atomic_inc(&lo->plh_refcount); | |
260 | } | |
261 | ||
262 | static void | |
263 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
264 | { | |
265 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
266 | atomic_dec(&lo->plh_refcount); | |
267 | } | |
268 | ||
269 | static void | |
270 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
271 | { | |
272 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
273 | struct pnfs_layout_range range = { |
274 | .iomode = iomode, | |
275 | .offset = 0, | |
276 | .length = NFS4_MAX_UINT64, | |
277 | }; | |
278 | LIST_HEAD(head); | |
3e621214 TM |
279 | |
280 | spin_lock(&inode->i_lock); | |
281 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
115ce575 | 282 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range); |
3e621214 | 283 | spin_unlock(&inode->i_lock); |
115ce575 | 284 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
285 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
286 | iomode == IOMODE_RW ? "RW" : "READ"); | |
287 | } | |
288 | ||
289 | static bool | |
290 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
291 | { | |
25c75333 | 292 | unsigned long start, end; |
3e621214 TM |
293 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
294 | ||
295 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
296 | return false; |
297 | end = jiffies; | |
298 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
299 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
300 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 301 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
302 | return false; |
303 | } | |
304 | return true; | |
b9e028fd TM |
305 | } |
306 | ||
974cec8c AA |
307 | static void |
308 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
309 | { | |
566052c5 | 310 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 311 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c FI |
312 | atomic_set(&lseg->pls_refcount, 1); |
313 | smp_mb(); | |
314 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 315 | lseg->pls_layout = lo; |
974cec8c AA |
316 | } |
317 | ||
905ca191 | 318 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 319 | { |
b7edfaa1 | 320 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 321 | |
b1f69b75 | 322 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
57036a37 | 323 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
70c3bd2b | 324 | pnfs_put_layout_hdr(NFS_I(ino)->layout); |
974cec8c AA |
325 | } |
326 | ||
d684d2ae | 327 | static void |
57036a37 TM |
328 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
329 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 330 | { |
57036a37 | 331 | struct inode *inode = lo->plh_inode; |
d684d2ae | 332 | |
d20581aa | 333 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 334 | list_del_init(&lseg->pls_list); |
01d39ce8 | 335 | if (list_empty(&lo->plh_segs)) |
57036a37 | 336 | set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags); |
d684d2ae FI |
337 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); |
338 | } | |
339 | ||
bae724ef | 340 | void |
9369a431 | 341 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 342 | { |
57036a37 | 343 | struct pnfs_layout_hdr *lo; |
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)); | |
57036a37 TM |
352 | lo = lseg->pls_layout; |
353 | inode = lo->plh_inode; | |
d684d2ae | 354 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
57036a37 | 355 | pnfs_layout_remove_lseg(lo, lseg); |
d684d2ae | 356 | spin_unlock(&inode->i_lock); |
905ca191 | 357 | pnfs_free_lseg(lseg); |
4541d16c | 358 | } |
4541d16c | 359 | } |
9369a431 | 360 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 361 | |
fb3296eb BH |
362 | static inline u64 |
363 | end_offset(u64 start, u64 len) | |
364 | { | |
365 | u64 end; | |
366 | ||
367 | end = start + len; | |
368 | return end >= start ? end : NFS4_MAX_UINT64; | |
369 | } | |
370 | ||
371 | /* last octet in a range */ | |
372 | static inline u64 | |
373 | last_byte_offset(u64 start, u64 len) | |
374 | { | |
375 | u64 end; | |
376 | ||
377 | BUG_ON(!len); | |
378 | end = start + len; | |
379 | return end > start ? end - 1 : NFS4_MAX_UINT64; | |
380 | } | |
381 | ||
382 | /* | |
383 | * is l2 fully contained in l1? | |
384 | * start1 end1 | |
385 | * [----------------------------------) | |
386 | * start2 end2 | |
387 | * [----------------) | |
388 | */ | |
389 | static inline int | |
390 | lo_seg_contained(struct pnfs_layout_range *l1, | |
391 | struct pnfs_layout_range *l2) | |
392 | { | |
393 | u64 start1 = l1->offset; | |
394 | u64 end1 = end_offset(start1, l1->length); | |
395 | u64 start2 = l2->offset; | |
396 | u64 end2 = end_offset(start2, l2->length); | |
397 | ||
398 | return (start1 <= start2) && (end1 >= end2); | |
399 | } | |
400 | ||
401 | /* | |
402 | * is l1 and l2 intersecting? | |
403 | * start1 end1 | |
404 | * [----------------------------------) | |
405 | * start2 end2 | |
406 | * [----------------) | |
407 | */ | |
408 | static inline int | |
409 | lo_seg_intersecting(struct pnfs_layout_range *l1, | |
410 | struct pnfs_layout_range *l2) | |
411 | { | |
412 | u64 start1 = l1->offset; | |
413 | u64 end1 = end_offset(start1, l1->length); | |
414 | u64 start2 = l2->offset; | |
415 | u64 end2 = end_offset(start2, l2->length); | |
416 | ||
417 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
418 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
419 | } | |
420 | ||
4541d16c | 421 | static bool |
778b5502 BH |
422 | should_free_lseg(struct pnfs_layout_range *lseg_range, |
423 | struct pnfs_layout_range *recall_range) | |
4541d16c | 424 | { |
778b5502 BH |
425 | return (recall_range->iomode == IOMODE_ANY || |
426 | lseg_range->iomode == recall_range->iomode) && | |
427 | lo_seg_intersecting(lseg_range, recall_range); | |
974cec8c AA |
428 | } |
429 | ||
4541d16c FI |
430 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
431 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
432 | struct list_head *tmp_list) | |
433 | { | |
434 | int rv = 0; | |
435 | ||
436 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
437 | /* Remove the reference keeping the lseg in the | |
438 | * list. It will now be removed when all | |
439 | * outstanding io is finished. | |
440 | */ | |
d684d2ae FI |
441 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
442 | atomic_read(&lseg->pls_refcount)); | |
443 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
57036a37 | 444 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); |
d684d2ae FI |
445 | list_add(&lseg->pls_list, tmp_list); |
446 | rv = 1; | |
447 | } | |
4541d16c FI |
448 | } |
449 | return rv; | |
450 | } | |
451 | ||
452 | /* Returns count of number of matching invalid lsegs remaining in list | |
453 | * after call. | |
454 | */ | |
43f1b3da | 455 | int |
49a85061 | 456 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 457 | struct list_head *tmp_list, |
778b5502 | 458 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
459 | { |
460 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 461 | int invalid = 0, removed = 0; |
974cec8c AA |
462 | |
463 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
464 | ||
38511722 | 465 | if (list_empty(&lo->plh_segs)) { |
01d39ce8 | 466 | set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags); |
38511722 FI |
467 | return 0; |
468 | } | |
4541d16c | 469 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
470 | if (!recall_range || |
471 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
472 | dprintk("%s: freeing lseg %p iomode %d " |
473 | "offset %llu length %llu\n", __func__, | |
474 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
475 | lseg->pls_range.length); | |
476 | invalid++; | |
477 | removed += mark_lseg_invalid(lseg, tmp_list); | |
478 | } | |
479 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
480 | return invalid - removed; | |
974cec8c AA |
481 | } |
482 | ||
f49f9baa | 483 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 484 | void |
4541d16c | 485 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 486 | { |
4541d16c | 487 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
488 | |
489 | if (list_empty(free_me)) | |
490 | return; | |
491 | ||
4541d16c | 492 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 493 | list_del(&lseg->pls_list); |
905ca191 | 494 | pnfs_free_lseg(lseg); |
974cec8c AA |
495 | } |
496 | } | |
497 | ||
e5e94017 BH |
498 | void |
499 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
500 | { | |
501 | struct pnfs_layout_hdr *lo; | |
974cec8c | 502 | LIST_HEAD(tmp_list); |
e5e94017 BH |
503 | |
504 | spin_lock(&nfsi->vfs_inode.i_lock); | |
505 | lo = nfsi->layout; | |
506 | if (lo) { | |
38511722 | 507 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 508 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
3e621214 TM |
509 | pnfs_get_layout_hdr(lo); |
510 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); | |
511 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
512 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
513 | pnfs_free_lseg_list(&tmp_list); | |
514 | pnfs_put_layout_hdr(lo); | |
515 | } else | |
516 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 517 | } |
041245c8 | 518 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c AA |
519 | |
520 | /* | |
521 | * Called by the state manger to remove all layouts established under an | |
522 | * expired lease. | |
523 | */ | |
524 | void | |
525 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
526 | { | |
6382a441 | 527 | struct nfs_server *server; |
974cec8c AA |
528 | struct pnfs_layout_hdr *lo; |
529 | LIST_HEAD(tmp_list); | |
530 | ||
c47abcf8 AA |
531 | nfs4_deviceid_mark_client_invalid(clp); |
532 | nfs4_deviceid_purge_client(clp); | |
533 | ||
974cec8c | 534 | spin_lock(&clp->cl_lock); |
6382a441 WAA |
535 | rcu_read_lock(); |
536 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
537 | if (!list_empty(&server->layouts)) | |
538 | list_splice_init(&server->layouts, &tmp_list); | |
539 | } | |
540 | rcu_read_unlock(); | |
974cec8c AA |
541 | spin_unlock(&clp->cl_lock); |
542 | ||
543 | while (!list_empty(&tmp_list)) { | |
544 | lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, | |
b7edfaa1 | 545 | plh_layouts); |
974cec8c | 546 | dprintk("%s freeing layout for inode %lu\n", __func__, |
b7edfaa1 | 547 | lo->plh_inode->i_ino); |
2887fe45 | 548 | list_del_init(&lo->plh_layouts); |
b7edfaa1 | 549 | pnfs_destroy_layout(NFS_I(lo->plh_inode)); |
974cec8c | 550 | } |
e5e94017 BH |
551 | } |
552 | ||
fd6002e9 | 553 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
554 | void |
555 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
556 | bool update_barrier) | |
b1f69b75 | 557 | { |
fd6002e9 | 558 | u32 oldseq, newseq; |
b1f69b75 | 559 | |
2d2f24ad TM |
560 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
561 | newseq = be32_to_cpu(new->seqid); | |
43f1b3da | 562 | if ((int)(newseq - oldseq) > 0) { |
f597c537 | 563 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 564 | if (update_barrier) { |
2d2f24ad | 565 | u32 new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
566 | |
567 | if ((int)(new_barrier - lo->plh_barrier)) | |
568 | lo->plh_barrier = new_barrier; | |
569 | } else { | |
570 | /* Because of wraparound, we want to keep the barrier | |
571 | * "close" to the current seqids. It needs to be | |
572 | * within 2**31 to count as "behind", so if it | |
573 | * gets too near that limit, give us a litle leeway | |
574 | * and bring it to within 2**30. | |
575 | * NOTE - and yes, this is all unsigned arithmetic. | |
576 | */ | |
577 | if (unlikely((newseq - lo->plh_barrier) > (3 << 29))) | |
578 | lo->plh_barrier = newseq - (1 << 30); | |
579 | } | |
580 | } | |
b1f69b75 AA |
581 | } |
582 | ||
cf7d63f1 FI |
583 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
584 | static bool | |
43f1b3da FI |
585 | pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid, |
586 | int lget) | |
587 | { | |
588 | if ((stateid) && | |
2d2f24ad | 589 | (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0) |
43f1b3da | 590 | return true; |
f7e8917a | 591 | return lo->plh_block_lgets || |
38511722 | 592 | test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) || |
f7e8917a | 593 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || |
43f1b3da | 594 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
595 | (atomic_read(&lo->plh_outstanding) > lget)); |
596 | } | |
597 | ||
fd6002e9 FI |
598 | int |
599 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
600 | struct nfs4_state *open_state) | |
b1f69b75 | 601 | { |
fd6002e9 | 602 | int status = 0; |
974cec8c | 603 | |
b1f69b75 | 604 | dprintk("--> %s\n", __func__); |
fd6002e9 | 605 | spin_lock(&lo->plh_inode->i_lock); |
43f1b3da | 606 | if (pnfs_layoutgets_blocked(lo, NULL, 1)) { |
cf7d63f1 FI |
607 | status = -EAGAIN; |
608 | } else if (list_empty(&lo->plh_segs)) { | |
fd6002e9 FI |
609 | int seq; |
610 | ||
611 | do { | |
612 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 613 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
614 | } while (read_seqretry(&open_state->seqlock, seq)); |
615 | } else | |
f597c537 | 616 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 617 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 618 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 619 | return status; |
b1f69b75 AA |
620 | } |
621 | ||
622 | /* | |
623 | * Get layout from server. | |
624 | * for now, assume that whole file layouts are requested. | |
625 | * arg->offset: 0 | |
626 | * arg->length: all ones | |
627 | */ | |
e5e94017 BH |
628 | static struct pnfs_layout_segment * |
629 | send_layoutget(struct pnfs_layout_hdr *lo, | |
630 | struct nfs_open_context *ctx, | |
fb3296eb | 631 | struct pnfs_layout_range *range, |
a75b9df9 | 632 | gfp_t gfp_flags) |
e5e94017 | 633 | { |
b7edfaa1 | 634 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
635 | struct nfs_server *server = NFS_SERVER(ino); |
636 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 637 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
638 | |
639 | dprintk("--> %s\n", __func__); | |
e5e94017 | 640 | |
b1f69b75 | 641 | BUG_ON(ctx == NULL); |
a75b9df9 | 642 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 643 | if (lgp == NULL) |
b1f69b75 | 644 | return NULL; |
35124a09 | 645 | |
fb3296eb BH |
646 | lgp->args.minlength = PAGE_CACHE_SIZE; |
647 | if (lgp->args.minlength > range->length) | |
648 | lgp->args.minlength = range->length; | |
b1f69b75 | 649 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 650 | lgp->args.range = *range; |
b1f69b75 AA |
651 | lgp->args.type = server->pnfs_curr_ld->id; |
652 | lgp->args.inode = ino; | |
653 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 654 | lgp->gfp_flags = gfp_flags; |
b1f69b75 AA |
655 | |
656 | /* Synchronously retrieve layout information from server and | |
657 | * store in lseg. | |
658 | */ | |
a0b0a6e3 TM |
659 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
660 | if (IS_ERR(lseg)) { | |
661 | switch (PTR_ERR(lseg)) { | |
662 | case -ENOMEM: | |
663 | case -ERESTARTSYS: | |
664 | break; | |
665 | default: | |
666 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 667 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
668 | } |
669 | return NULL; | |
974cec8c | 670 | } |
35124a09 | 671 | |
974cec8c AA |
672 | return lseg; |
673 | } | |
674 | ||
293b3b06 AA |
675 | /* |
676 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
677 | * when the layout segment list is empty. | |
678 | * | |
679 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
680 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
681 | * deviceid is marked invalid. | |
682 | */ | |
cbe82603 BH |
683 | int |
684 | _pnfs_return_layout(struct inode *ino) | |
685 | { | |
686 | struct pnfs_layout_hdr *lo = NULL; | |
687 | struct nfs_inode *nfsi = NFS_I(ino); | |
688 | LIST_HEAD(tmp_list); | |
689 | struct nfs4_layoutreturn *lrp; | |
690 | nfs4_stateid stateid; | |
293b3b06 | 691 | int status = 0, empty; |
cbe82603 | 692 | |
366d5052 | 693 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
694 | |
695 | spin_lock(&ino->i_lock); | |
696 | lo = nfsi->layout; | |
366d5052 | 697 | if (!lo || pnfs_test_layout_returned(lo)) { |
cbe82603 | 698 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
699 | dprintk("NFS: %s no layout to return\n", __func__); |
700 | goto out; | |
cbe82603 BH |
701 | } |
702 | stateid = nfsi->layout->plh_stateid; | |
703 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 704 | pnfs_get_layout_hdr(lo); |
293b3b06 | 705 | empty = list_empty(&lo->plh_segs); |
49a85061 | 706 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
293b3b06 AA |
707 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
708 | if (empty) { | |
709 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 710 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
711 | dprintk("NFS: %s no layout segments to return\n", __func__); |
712 | goto out; | |
713 | } | |
ea0ded74 | 714 | lo->plh_block_lgets++; |
366d5052 | 715 | pnfs_mark_layout_returned(lo); |
cbe82603 BH |
716 | spin_unlock(&ino->i_lock); |
717 | pnfs_free_lseg_list(&tmp_list); | |
718 | ||
719 | WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)); | |
720 | ||
721 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); | |
722 | if (unlikely(lrp == NULL)) { | |
723 | status = -ENOMEM; | |
b9e028fd TM |
724 | pnfs_layout_io_set_failed(lo, IOMODE_RW); |
725 | pnfs_layout_io_set_failed(lo, IOMODE_READ); | |
366d5052 | 726 | pnfs_clear_layout_returned(lo); |
70c3bd2b | 727 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
728 | goto out; |
729 | } | |
730 | ||
731 | lrp->args.stateid = stateid; | |
732 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
733 | lrp->args.inode = ino; | |
a56aaa02 | 734 | lrp->args.layout = lo; |
cbe82603 BH |
735 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
736 | ||
737 | status = nfs4_proc_layoutreturn(lrp); | |
738 | out: | |
739 | dprintk("<-- %s status: %d\n", __func__, status); | |
740 | return status; | |
741 | } | |
0a57cdac | 742 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 743 | |
f7e8917a FI |
744 | bool pnfs_roc(struct inode *ino) |
745 | { | |
746 | struct pnfs_layout_hdr *lo; | |
747 | struct pnfs_layout_segment *lseg, *tmp; | |
748 | LIST_HEAD(tmp_list); | |
749 | bool found = false; | |
750 | ||
751 | spin_lock(&ino->i_lock); | |
752 | lo = NFS_I(ino)->layout; | |
753 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
754 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
755 | goto out_nolayout; | |
756 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
757 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
758 | mark_lseg_invalid(lseg, &tmp_list); | |
759 | found = true; | |
760 | } | |
761 | if (!found) | |
762 | goto out_nolayout; | |
763 | lo->plh_block_lgets++; | |
70c3bd2b | 764 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
765 | spin_unlock(&ino->i_lock); |
766 | pnfs_free_lseg_list(&tmp_list); | |
767 | return true; | |
768 | ||
769 | out_nolayout: | |
770 | spin_unlock(&ino->i_lock); | |
771 | return false; | |
772 | } | |
773 | ||
774 | void pnfs_roc_release(struct inode *ino) | |
775 | { | |
776 | struct pnfs_layout_hdr *lo; | |
777 | ||
778 | spin_lock(&ino->i_lock); | |
779 | lo = NFS_I(ino)->layout; | |
780 | lo->plh_block_lgets--; | |
6622c3ea TM |
781 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
782 | pnfs_detach_layout_hdr(lo); | |
783 | spin_unlock(&ino->i_lock); | |
784 | pnfs_free_layout_hdr(lo); | |
785 | } else | |
786 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
787 | } |
788 | ||
789 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
790 | { | |
791 | struct pnfs_layout_hdr *lo; | |
792 | ||
793 | spin_lock(&ino->i_lock); | |
794 | lo = NFS_I(ino)->layout; | |
795 | if ((int)(barrier - lo->plh_barrier) > 0) | |
796 | lo->plh_barrier = barrier; | |
797 | spin_unlock(&ino->i_lock); | |
798 | } | |
799 | ||
7fdab069 | 800 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
801 | { |
802 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 803 | struct pnfs_layout_hdr *lo; |
f7e8917a | 804 | struct pnfs_layout_segment *lseg; |
7fdab069 | 805 | u32 current_seqid; |
f7e8917a FI |
806 | bool found = false; |
807 | ||
808 | spin_lock(&ino->i_lock); | |
809 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
810 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 811 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 812 | found = true; |
7fdab069 | 813 | goto out; |
f7e8917a | 814 | } |
7fdab069 TM |
815 | lo = nfsi->layout; |
816 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 817 | |
7fdab069 TM |
818 | /* Since close does not return a layout stateid for use as |
819 | * a barrier, we choose the worst-case barrier. | |
820 | */ | |
821 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
822 | out: | |
f7e8917a FI |
823 | spin_unlock(&ino->i_lock); |
824 | return found; | |
825 | } | |
826 | ||
b1f69b75 AA |
827 | /* |
828 | * Compare two layout segments for sorting into layout cache. | |
829 | * We want to preferentially return RW over RO layouts, so ensure those | |
830 | * are seen first. | |
831 | */ | |
832 | static s64 | |
fb3296eb BH |
833 | cmp_layout(struct pnfs_layout_range *l1, |
834 | struct pnfs_layout_range *l2) | |
b1f69b75 | 835 | { |
fb3296eb BH |
836 | s64 d; |
837 | ||
838 | /* high offset > low offset */ | |
839 | d = l1->offset - l2->offset; | |
840 | if (d) | |
841 | return d; | |
842 | ||
843 | /* short length > long length */ | |
844 | d = l2->length - l1->length; | |
845 | if (d) | |
846 | return d; | |
847 | ||
b1f69b75 | 848 | /* read > read/write */ |
fb3296eb | 849 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
850 | } |
851 | ||
974cec8c | 852 | static void |
57036a37 | 853 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
854 | struct pnfs_layout_segment *lseg) |
855 | { | |
b1f69b75 | 856 | struct pnfs_layout_segment *lp; |
b1f69b75 | 857 | |
974cec8c AA |
858 | dprintk("%s:Begin\n", __func__); |
859 | ||
b7edfaa1 | 860 | assert_spin_locked(&lo->plh_inode->i_lock); |
b7edfaa1 | 861 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
fb3296eb | 862 | if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 863 | continue; |
566052c5 | 864 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
865 | dprintk("%s: inserted lseg %p " |
866 | "iomode %d offset %llu length %llu before " | |
867 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
868 | __func__, lseg, lseg->pls_range.iomode, |
869 | lseg->pls_range.offset, lseg->pls_range.length, | |
870 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
871 | lp->pls_range.length); | |
fb3296eb | 872 | goto out; |
974cec8c | 873 | } |
fb3296eb BH |
874 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
875 | dprintk("%s: inserted lseg %p " | |
876 | "iomode %d offset %llu length %llu at tail\n", | |
877 | __func__, lseg, lseg->pls_range.iomode, | |
878 | lseg->pls_range.offset, lseg->pls_range.length); | |
879 | out: | |
70c3bd2b | 880 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
881 | |
882 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
883 | } |
884 | ||
885 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
886 | alloc_init_layout_hdr(struct inode *ino, |
887 | struct nfs_open_context *ctx, | |
888 | gfp_t gfp_flags) | |
e5e94017 BH |
889 | { |
890 | struct pnfs_layout_hdr *lo; | |
891 | ||
636fb9c8 | 892 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
893 | if (!lo) |
894 | return NULL; | |
cc6e5340 | 895 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
896 | INIT_LIST_HEAD(&lo->plh_layouts); |
897 | INIT_LIST_HEAD(&lo->plh_segs); | |
43f1b3da | 898 | INIT_LIST_HEAD(&lo->plh_bulk_recall); |
b7edfaa1 | 899 | lo->plh_inode = ino; |
9fa40758 | 900 | lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred); |
e5e94017 BH |
901 | return lo; |
902 | } | |
903 | ||
904 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
905 | pnfs_find_alloc_layout(struct inode *ino, |
906 | struct nfs_open_context *ctx, | |
907 | gfp_t gfp_flags) | |
e5e94017 BH |
908 | { |
909 | struct nfs_inode *nfsi = NFS_I(ino); | |
910 | struct pnfs_layout_hdr *new = NULL; | |
911 | ||
912 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
913 | ||
914 | assert_spin_locked(&ino->i_lock); | |
4541d16c FI |
915 | if (nfsi->layout) { |
916 | if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags)) | |
917 | return NULL; | |
01d39ce8 TM |
918 | pnfs_get_layout_hdr(nfsi->layout); |
919 | return nfsi->layout; | |
4541d16c | 920 | } |
e5e94017 | 921 | spin_unlock(&ino->i_lock); |
9fa40758 | 922 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
923 | spin_lock(&ino->i_lock); |
924 | ||
925 | if (likely(nfsi->layout == NULL)) /* Won the race? */ | |
926 | nfsi->layout = new; | |
927 | else | |
636fb9c8 | 928 | pnfs_free_layout_hdr(new); |
e5e94017 BH |
929 | return nfsi->layout; |
930 | } | |
931 | ||
b1f69b75 AA |
932 | /* |
933 | * iomode matching rules: | |
934 | * iomode lseg match | |
935 | * ----- ----- ----- | |
936 | * ANY READ true | |
937 | * ANY RW true | |
938 | * RW READ false | |
939 | * RW RW true | |
940 | * READ READ true | |
941 | * READ RW true | |
942 | */ | |
943 | static int | |
fb3296eb BH |
944 | is_matching_lseg(struct pnfs_layout_range *ls_range, |
945 | struct pnfs_layout_range *range) | |
b1f69b75 | 946 | { |
fb3296eb BH |
947 | struct pnfs_layout_range range1; |
948 | ||
949 | if ((range->iomode == IOMODE_RW && | |
950 | ls_range->iomode != IOMODE_RW) || | |
951 | !lo_seg_intersecting(ls_range, range)) | |
952 | return 0; | |
953 | ||
954 | /* range1 covers only the first byte in the range */ | |
955 | range1 = *range; | |
956 | range1.length = 1; | |
957 | return lo_seg_contained(ls_range, &range1); | |
b1f69b75 AA |
958 | } |
959 | ||
960 | /* | |
961 | * lookup range in layout | |
962 | */ | |
e5e94017 | 963 | static struct pnfs_layout_segment * |
fb3296eb BH |
964 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
965 | struct pnfs_layout_range *range) | |
e5e94017 | 966 | { |
b1f69b75 AA |
967 | struct pnfs_layout_segment *lseg, *ret = NULL; |
968 | ||
969 | dprintk("%s:Begin\n", __func__); | |
970 | ||
b7edfaa1 FI |
971 | assert_spin_locked(&lo->plh_inode->i_lock); |
972 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
4541d16c | 973 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
fb3296eb | 974 | is_matching_lseg(&lseg->pls_range, range)) { |
9369a431 | 975 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
976 | break; |
977 | } | |
d771e3a4 | 978 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
979 | break; |
980 | } | |
981 | ||
982 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 983 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 984 | return ret; |
e5e94017 BH |
985 | } |
986 | ||
d23d61c8 AA |
987 | /* |
988 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
989 | * to the MDS or over pNFS | |
990 | * | |
991 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
992 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
993 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
994 | * WRITE request. | |
995 | * | |
996 | * A return of true means use MDS I/O. | |
997 | * | |
998 | * From rfc 5661: | |
999 | * If a file's size is smaller than the file size threshold, data accesses | |
1000 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1001 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1002 | * server. If both file size and I/O size are provided, the client SHOULD | |
1003 | * reach or exceed both thresholds before sending its read or write | |
1004 | * requests to the data server. | |
1005 | */ | |
1006 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1007 | struct inode *ino, int iomode) | |
1008 | { | |
1009 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1010 | struct nfs_inode *nfsi = NFS_I(ino); | |
1011 | loff_t fsize = i_size_read(ino); | |
1012 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1013 | ||
1014 | if (t == NULL) | |
1015 | return ret; | |
1016 | ||
1017 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1018 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1019 | ||
1020 | switch (iomode) { | |
1021 | case IOMODE_READ: | |
1022 | if (t->bm & THRESHOLD_RD) { | |
1023 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1024 | size_set = true; | |
1025 | if (fsize < t->rd_sz) | |
1026 | size = true; | |
1027 | } | |
1028 | if (t->bm & THRESHOLD_RD_IO) { | |
1029 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1030 | nfsi->read_io); | |
1031 | io_set = true; | |
1032 | if (nfsi->read_io < t->rd_io_sz) | |
1033 | io = true; | |
1034 | } | |
1035 | break; | |
1036 | case IOMODE_RW: | |
1037 | if (t->bm & THRESHOLD_WR) { | |
1038 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1039 | size_set = true; | |
1040 | if (fsize < t->wr_sz) | |
1041 | size = true; | |
1042 | } | |
1043 | if (t->bm & THRESHOLD_WR_IO) { | |
1044 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1045 | nfsi->write_io); | |
1046 | io_set = true; | |
1047 | if (nfsi->write_io < t->wr_io_sz) | |
1048 | io = true; | |
1049 | } | |
1050 | break; | |
1051 | } | |
1052 | if (size_set && io_set) { | |
1053 | if (size && io) | |
1054 | ret = true; | |
1055 | } else if (size || io) | |
1056 | ret = true; | |
1057 | ||
1058 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1059 | return ret; | |
1060 | } | |
1061 | ||
e5e94017 BH |
1062 | /* |
1063 | * Layout segment is retreived from the server if not cached. | |
1064 | * The appropriate layout segment is referenced and returned to the caller. | |
1065 | */ | |
7c24d948 | 1066 | struct pnfs_layout_segment * |
e5e94017 BH |
1067 | pnfs_update_layout(struct inode *ino, |
1068 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1069 | loff_t pos, |
1070 | u64 count, | |
a75b9df9 TM |
1071 | enum pnfs_iomode iomode, |
1072 | gfp_t gfp_flags) | |
e5e94017 | 1073 | { |
fb3296eb BH |
1074 | struct pnfs_layout_range arg = { |
1075 | .iomode = iomode, | |
1076 | .offset = pos, | |
1077 | .length = count, | |
1078 | }; | |
707ed5fd | 1079 | unsigned pg_offset; |
6382a441 WAA |
1080 | struct nfs_server *server = NFS_SERVER(ino); |
1081 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1082 | struct pnfs_layout_hdr *lo; |
1083 | struct pnfs_layout_segment *lseg = NULL; | |
f49f9baa | 1084 | bool first = false; |
e5e94017 BH |
1085 | |
1086 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1087 | goto out; |
d23d61c8 AA |
1088 | |
1089 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1090 | goto out; |
d23d61c8 | 1091 | |
e5e94017 | 1092 | spin_lock(&ino->i_lock); |
9fa40758 | 1093 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1094 | if (lo == NULL) { |
1095 | spin_unlock(&ino->i_lock); | |
1096 | goto out; | |
1097 | } | |
e5e94017 | 1098 | |
43f1b3da | 1099 | /* Do we even need to bother with this? */ |
a59c30ac | 1100 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1101 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1102 | goto out_unlock; |
1103 | } | |
1104 | ||
1105 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1106 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1107 | goto out_unlock; |
1108 | ||
568e8c49 | 1109 | /* Check to see if the layout for the given range already exists */ |
fb3296eb | 1110 | lseg = pnfs_find_lseg(lo, &arg); |
568e8c49 AA |
1111 | if (lseg) |
1112 | goto out_unlock; | |
1113 | ||
43f1b3da | 1114 | if (pnfs_layoutgets_blocked(lo, NULL, 0)) |
cf7d63f1 FI |
1115 | goto out_unlock; |
1116 | atomic_inc(&lo->plh_outstanding); | |
1117 | ||
f49f9baa FI |
1118 | if (list_empty(&lo->plh_segs)) |
1119 | first = true; | |
366d5052 AA |
1120 | |
1121 | /* Enable LAYOUTRETURNs */ | |
1122 | pnfs_clear_layout_returned(lo); | |
1123 | ||
f49f9baa FI |
1124 | spin_unlock(&ino->i_lock); |
1125 | if (first) { | |
2130ff66 FI |
1126 | /* The lo must be on the clp list if there is any |
1127 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1128 | */ | |
1129 | spin_lock(&clp->cl_lock); | |
1130 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
6382a441 | 1131 | list_add_tail(&lo->plh_layouts, &server->layouts); |
2130ff66 FI |
1132 | spin_unlock(&clp->cl_lock); |
1133 | } | |
e5e94017 | 1134 | |
707ed5fd BH |
1135 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1136 | if (pg_offset) { | |
1137 | arg.offset -= pg_offset; | |
1138 | arg.length += pg_offset; | |
1139 | } | |
7c24d948 AA |
1140 | if (arg.length != NFS4_MAX_UINT64) |
1141 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1142 | |
fb3296eb | 1143 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
cf7d63f1 | 1144 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1145 | out_put_layout_hdr: |
70c3bd2b | 1146 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1147 | out: |
f86bbcf8 TM |
1148 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1149 | "(%s, offset: %llu, length: %llu)\n", | |
1150 | __func__, ino->i_sb->s_id, | |
1151 | (unsigned long long)NFS_FILEID(ino), | |
1152 | lseg == NULL ? "not found" : "found", | |
1153 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1154 | (unsigned long long)pos, | |
1155 | (unsigned long long)count); | |
e5e94017 BH |
1156 | return lseg; |
1157 | out_unlock: | |
1158 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1159 | goto out_put_layout_hdr; |
e5e94017 | 1160 | } |
7c24d948 | 1161 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1162 | |
a0b0a6e3 | 1163 | struct pnfs_layout_segment * |
b1f69b75 AA |
1164 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1165 | { | |
1166 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1167 | struct nfs4_layoutget_res *res = &lgp->res; | |
1168 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1169 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
1170 | int status = 0; |
1171 | ||
1172 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1173 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1174 | if (!lseg || IS_ERR(lseg)) { |
1175 | if (!lseg) | |
1176 | status = -ENOMEM; | |
1177 | else | |
1178 | status = PTR_ERR(lseg); | |
1179 | dprintk("%s: Could not allocate layout: error %d\n", | |
1180 | __func__, status); | |
1181 | goto out; | |
1182 | } | |
1183 | ||
1184 | spin_lock(&ino->i_lock); | |
a59c30ac | 1185 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1186 | dprintk("%s forget reply due to recall\n", __func__); |
1187 | goto out_forget_reply; | |
1188 | } | |
1189 | ||
1190 | if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) { | |
1191 | dprintk("%s forget reply due to state\n", __func__); | |
1192 | goto out_forget_reply; | |
1193 | } | |
b1f69b75 | 1194 | init_lseg(lo, lseg); |
566052c5 | 1195 | lseg->pls_range = res->range; |
9369a431 | 1196 | pnfs_get_lseg(lseg); |
57036a37 | 1197 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1198 | |
f7e8917a FI |
1199 | if (res->return_on_close) { |
1200 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1201 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1202 | } | |
1203 | ||
b1f69b75 | 1204 | /* Done processing layoutget. Set the layout stateid */ |
43f1b3da | 1205 | pnfs_set_layout_stateid(lo, &res->stateid, false); |
b1f69b75 | 1206 | spin_unlock(&ino->i_lock); |
a0b0a6e3 | 1207 | return lseg; |
b1f69b75 | 1208 | out: |
a0b0a6e3 | 1209 | return ERR_PTR(status); |
43f1b3da FI |
1210 | |
1211 | out_forget_reply: | |
1212 | spin_unlock(&ino->i_lock); | |
1213 | lseg->pls_layout = lo; | |
1214 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1215 | goto out; | |
b1f69b75 AA |
1216 | } |
1217 | ||
d8007d4d TM |
1218 | void |
1219 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1220 | { | |
1221 | BUG_ON(pgio->pg_lseg != NULL); | |
1222 | ||
1825a0d0 FI |
1223 | if (req->wb_offset != req->wb_pgbase) { |
1224 | nfs_pageio_reset_read_mds(pgio); | |
1225 | return; | |
1226 | } | |
d8007d4d TM |
1227 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1228 | req->wb_context, | |
1229 | req_offset(req), | |
1230 | req->wb_bytes, | |
1231 | IOMODE_READ, | |
1232 | GFP_KERNEL); | |
e885de1a TM |
1233 | /* If no lseg, fall back to read through mds */ |
1234 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1235 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1236 | |
d8007d4d TM |
1237 | } |
1238 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1239 | ||
1240 | void | |
1241 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1242 | { | |
1243 | BUG_ON(pgio->pg_lseg != NULL); | |
1244 | ||
1825a0d0 FI |
1245 | if (req->wb_offset != req->wb_pgbase) { |
1246 | nfs_pageio_reset_write_mds(pgio); | |
1247 | return; | |
1248 | } | |
d8007d4d TM |
1249 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1250 | req->wb_context, | |
1251 | req_offset(req), | |
1252 | req->wb_bytes, | |
1253 | IOMODE_RW, | |
1254 | GFP_NOFS); | |
e885de1a TM |
1255 | /* If no lseg, fall back to write through mds */ |
1256 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1257 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1258 | } |
1259 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1260 | ||
1abb5088 | 1261 | void |
061ae2ed FI |
1262 | pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode, |
1263 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 TM |
1264 | { |
1265 | struct nfs_server *server = NFS_SERVER(inode); | |
1266 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
1267 | ||
1268 | if (ld == NULL) | |
1abb5088 BS |
1269 | nfs_pageio_init_read(pgio, inode, compl_ops); |
1270 | else | |
1271 | nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0); | |
1751c363 TM |
1272 | } |
1273 | ||
57208fa7 | 1274 | void |
061ae2ed FI |
1275 | pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode, |
1276 | int ioflags, | |
1277 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 TM |
1278 | { |
1279 | struct nfs_server *server = NFS_SERVER(inode); | |
1280 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
1281 | ||
1282 | if (ld == NULL) | |
57208fa7 BS |
1283 | nfs_pageio_init_write(pgio, inode, ioflags, compl_ops); |
1284 | else | |
1285 | nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags); | |
1751c363 TM |
1286 | } |
1287 | ||
18ad0a9f | 1288 | bool |
dfed206b BH |
1289 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
1290 | struct nfs_page *req) | |
94ad1c80 | 1291 | { |
d8007d4d TM |
1292 | if (pgio->pg_lseg == NULL) |
1293 | return nfs_generic_pg_test(pgio, prev, req); | |
94ad1c80 | 1294 | |
19982ba8 TM |
1295 | /* |
1296 | * Test if a nfs_page is fully contained in the pnfs_layout_range. | |
1297 | * Note that this test makes several assumptions: | |
1298 | * - that the previous nfs_page in the struct nfs_pageio_descriptor | |
1299 | * is known to lie within the range. | |
1300 | * - that the nfs_page being tested is known to be contiguous with the | |
1301 | * previous nfs_page. | |
1302 | * - Layout ranges are page aligned, so we only have to test the | |
1303 | * start offset of the request. | |
1304 | * | |
1305 | * Please also note that 'end_offset' is actually the offset of the | |
1306 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1307 | * | |
1308 | */ | |
1309 | return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset, | |
1310 | pgio->pg_lseg->pls_range.length); | |
94ad1c80 | 1311 | } |
89a58e32 | 1312 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1313 | |
e7dd79af | 1314 | int pnfs_write_done_resend_to_mds(struct inode *inode, |
061ae2ed FI |
1315 | struct list_head *head, |
1316 | const struct nfs_pgio_completion_ops *compl_ops) | |
e2fecb21 TM |
1317 | { |
1318 | struct nfs_pageio_descriptor pgio; | |
1319 | LIST_HEAD(failed); | |
1320 | ||
1321 | /* Resend all requests through the MDS */ | |
57208fa7 | 1322 | nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops); |
e2fecb21 TM |
1323 | while (!list_empty(head)) { |
1324 | struct nfs_page *req = nfs_list_entry(head->next); | |
1325 | ||
1326 | nfs_list_remove_request(req); | |
1327 | if (!nfs_pageio_add_request(&pgio, req)) | |
1328 | nfs_list_add_request(req, &failed); | |
1329 | } | |
1330 | nfs_pageio_complete(&pgio); | |
1331 | ||
1332 | if (!list_empty(&failed)) { | |
1333 | /* For some reason our attempt to resend pages. Mark the | |
1334 | * overall send request as having failed, and let | |
1335 | * nfs_writeback_release_full deal with the error. | |
1336 | */ | |
1337 | list_move(&failed, head); | |
1338 | return -EIO; | |
1339 | } | |
1340 | return 0; | |
1341 | } | |
e7dd79af | 1342 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1343 | |
1acbbb4e FI |
1344 | static void pnfs_ld_handle_write_error(struct nfs_write_data *data) |
1345 | { | |
cd841605 FI |
1346 | struct nfs_pgio_header *hdr = data->header; |
1347 | ||
1348 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1349 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1350 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 FI |
1351 | clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags); |
1352 | pnfs_return_layout(hdr->inode); | |
1acbbb4e | 1353 | } |
6c75dc0d FI |
1354 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
1355 | data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode, | |
061ae2ed FI |
1356 | &hdr->pages, |
1357 | hdr->completion_ops); | |
1acbbb4e FI |
1358 | } |
1359 | ||
d20581aa BH |
1360 | /* |
1361 | * Called by non rpc-based layout drivers | |
1362 | */ | |
8ce160c5 | 1363 | void pnfs_ld_write_done(struct nfs_write_data *data) |
44b83799 | 1364 | { |
cd841605 FI |
1365 | struct nfs_pgio_header *hdr = data->header; |
1366 | ||
1367 | if (!hdr->pnfs_error) { | |
d20581aa | 1368 | pnfs_set_layoutcommit(data); |
cd841605 | 1369 | hdr->mds_ops->rpc_call_done(&data->task, data); |
1acbbb4e FI |
1370 | } else |
1371 | pnfs_ld_handle_write_error(data); | |
cd841605 | 1372 | hdr->mds_ops->rpc_release(data); |
44b83799 | 1373 | } |
d20581aa | 1374 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1375 | |
dce81290 TM |
1376 | static void |
1377 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
1378 | struct nfs_write_data *data) | |
1379 | { | |
cd841605 FI |
1380 | struct nfs_pgio_header *hdr = data->header; |
1381 | ||
6c75dc0d FI |
1382 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1383 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1384 | nfs_pageio_reset_write_mds(desc); | |
1385 | desc->pg_recoalesce = 1; | |
1386 | } | |
dce81290 TM |
1387 | nfs_writedata_release(data); |
1388 | } | |
1389 | ||
1390 | static enum pnfs_try_status | |
0382b744 | 1391 | pnfs_try_to_write_data(struct nfs_write_data *wdata, |
dce81290 TM |
1392 | const struct rpc_call_ops *call_ops, |
1393 | struct pnfs_layout_segment *lseg, | |
1394 | int how) | |
0382b744 | 1395 | { |
cd841605 FI |
1396 | struct nfs_pgio_header *hdr = wdata->header; |
1397 | struct inode *inode = hdr->inode; | |
0382b744 AA |
1398 | enum pnfs_try_status trypnfs; |
1399 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1400 | ||
cd841605 | 1401 | hdr->mds_ops = call_ops; |
0382b744 AA |
1402 | |
1403 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
1404 | inode->i_ino, wdata->args.count, wdata->args.offset, how); | |
0382b744 | 1405 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how); |
6c75dc0d | 1406 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1407 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1408 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1409 | return trypnfs; | |
1410 | } | |
1411 | ||
dce81290 TM |
1412 | static void |
1413 | pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how) | |
1414 | { | |
1415 | struct nfs_write_data *data; | |
1416 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; | |
1417 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
1418 | ||
1419 | desc->pg_lseg = NULL; | |
1420 | while (!list_empty(head)) { | |
1421 | enum pnfs_try_status trypnfs; | |
1422 | ||
6c75dc0d | 1423 | data = list_first_entry(head, struct nfs_write_data, list); |
dce81290 TM |
1424 | list_del_init(&data->list); |
1425 | ||
1426 | trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how); | |
1427 | if (trypnfs == PNFS_NOT_ATTEMPTED) | |
1428 | pnfs_write_through_mds(desc, data); | |
1429 | } | |
9369a431 | 1430 | pnfs_put_lseg(lseg); |
dce81290 TM |
1431 | } |
1432 | ||
6c75dc0d FI |
1433 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1434 | { | |
9369a431 | 1435 | pnfs_put_lseg(hdr->lseg); |
6c75dc0d FI |
1436 | nfs_writehdr_free(hdr); |
1437 | } | |
89d77c8f | 1438 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1439 | |
dce81290 TM |
1440 | int |
1441 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1442 | { | |
6c75dc0d FI |
1443 | struct nfs_write_header *whdr; |
1444 | struct nfs_pgio_header *hdr; | |
dce81290 TM |
1445 | int ret; |
1446 | ||
6c75dc0d FI |
1447 | whdr = nfs_writehdr_alloc(); |
1448 | if (!whdr) { | |
9b5415b5 | 1449 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
9369a431 | 1450 | pnfs_put_lseg(desc->pg_lseg); |
dce81290 | 1451 | desc->pg_lseg = NULL; |
6c75dc0d | 1452 | return -ENOMEM; |
dce81290 | 1453 | } |
6c75dc0d FI |
1454 | hdr = &whdr->header; |
1455 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); | |
9369a431 | 1456 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
6c75dc0d FI |
1457 | atomic_inc(&hdr->refcnt); |
1458 | ret = nfs_generic_flush(desc, hdr); | |
1459 | if (ret != 0) { | |
9369a431 | 1460 | pnfs_put_lseg(desc->pg_lseg); |
6c75dc0d | 1461 | desc->pg_lseg = NULL; |
6c75dc0d FI |
1462 | } else |
1463 | pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags); | |
1464 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1465 | hdr->completion_ops->completion(hdr); |
6c75dc0d | 1466 | return ret; |
dce81290 TM |
1467 | } |
1468 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1469 | ||
e7dd79af | 1470 | int pnfs_read_done_resend_to_mds(struct inode *inode, |
061ae2ed FI |
1471 | struct list_head *head, |
1472 | const struct nfs_pgio_completion_ops *compl_ops) | |
62e4a769 TM |
1473 | { |
1474 | struct nfs_pageio_descriptor pgio; | |
1acbbb4e | 1475 | LIST_HEAD(failed); |
62e4a769 | 1476 | |
1acbbb4e | 1477 | /* Resend all requests through the MDS */ |
1abb5088 | 1478 | nfs_pageio_init_read(&pgio, inode, compl_ops); |
1acbbb4e FI |
1479 | while (!list_empty(head)) { |
1480 | struct nfs_page *req = nfs_list_entry(head->next); | |
62e4a769 TM |
1481 | |
1482 | nfs_list_remove_request(req); | |
1acbbb4e FI |
1483 | if (!nfs_pageio_add_request(&pgio, req)) |
1484 | nfs_list_add_request(req, &failed); | |
62e4a769 TM |
1485 | } |
1486 | nfs_pageio_complete(&pgio); | |
1acbbb4e FI |
1487 | |
1488 | if (!list_empty(&failed)) { | |
1489 | list_move(&failed, head); | |
1490 | return -EIO; | |
1491 | } | |
1492 | return 0; | |
1493 | } | |
e7dd79af | 1494 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e FI |
1495 | |
1496 | static void pnfs_ld_handle_read_error(struct nfs_read_data *data) | |
1497 | { | |
cd841605 FI |
1498 | struct nfs_pgio_header *hdr = data->header; |
1499 | ||
1500 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); | |
1501 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1502 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 FI |
1503 | clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags); |
1504 | pnfs_return_layout(hdr->inode); | |
1acbbb4e | 1505 | } |
4db6e0b7 FI |
1506 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
1507 | data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode, | |
061ae2ed FI |
1508 | &hdr->pages, |
1509 | hdr->completion_ops); | |
62e4a769 TM |
1510 | } |
1511 | ||
d20581aa BH |
1512 | /* |
1513 | * Called by non rpc-based layout drivers | |
1514 | */ | |
9b7eecdc | 1515 | void pnfs_ld_read_done(struct nfs_read_data *data) |
d20581aa | 1516 | { |
cd841605 FI |
1517 | struct nfs_pgio_header *hdr = data->header; |
1518 | ||
1519 | if (likely(!hdr->pnfs_error)) { | |
d20581aa | 1520 | __nfs4_read_done_cb(data); |
cd841605 | 1521 | hdr->mds_ops->rpc_call_done(&data->task, data); |
62e4a769 TM |
1522 | } else |
1523 | pnfs_ld_handle_read_error(data); | |
cd841605 | 1524 | hdr->mds_ops->rpc_release(data); |
d20581aa BH |
1525 | } |
1526 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1527 | ||
493292dd TM |
1528 | static void |
1529 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
1530 | struct nfs_read_data *data) | |
1531 | { | |
cd841605 FI |
1532 | struct nfs_pgio_header *hdr = data->header; |
1533 | ||
4db6e0b7 FI |
1534 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1535 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1536 | nfs_pageio_reset_read_mds(desc); | |
1537 | desc->pg_recoalesce = 1; | |
1538 | } | |
493292dd TM |
1539 | nfs_readdata_release(data); |
1540 | } | |
1541 | ||
64419a9b AA |
1542 | /* |
1543 | * Call the appropriate parallel I/O subsystem read function. | |
1544 | */ | |
493292dd | 1545 | static enum pnfs_try_status |
64419a9b | 1546 | pnfs_try_to_read_data(struct nfs_read_data *rdata, |
493292dd TM |
1547 | const struct rpc_call_ops *call_ops, |
1548 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1549 | { |
cd841605 FI |
1550 | struct nfs_pgio_header *hdr = rdata->header; |
1551 | struct inode *inode = hdr->inode; | |
64419a9b AA |
1552 | struct nfs_server *nfss = NFS_SERVER(inode); |
1553 | enum pnfs_try_status trypnfs; | |
1554 | ||
cd841605 | 1555 | hdr->mds_ops = call_ops; |
64419a9b AA |
1556 | |
1557 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
1558 | __func__, inode->i_ino, rdata->args.count, rdata->args.offset); | |
1559 | ||
1560 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata); | |
4db6e0b7 | 1561 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1562 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1563 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1564 | return trypnfs; | |
1565 | } | |
863a3c6c | 1566 | |
493292dd TM |
1567 | static void |
1568 | pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head) | |
1569 | { | |
1570 | struct nfs_read_data *data; | |
1571 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; | |
1572 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
1573 | ||
1574 | desc->pg_lseg = NULL; | |
1575 | while (!list_empty(head)) { | |
1576 | enum pnfs_try_status trypnfs; | |
1577 | ||
4db6e0b7 | 1578 | data = list_first_entry(head, struct nfs_read_data, list); |
493292dd TM |
1579 | list_del_init(&data->list); |
1580 | ||
1581 | trypnfs = pnfs_try_to_read_data(data, call_ops, lseg); | |
1582 | if (trypnfs == PNFS_NOT_ATTEMPTED) | |
1583 | pnfs_read_through_mds(desc, data); | |
1584 | } | |
9369a431 | 1585 | pnfs_put_lseg(lseg); |
493292dd TM |
1586 | } |
1587 | ||
4db6e0b7 FI |
1588 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1589 | { | |
9369a431 | 1590 | pnfs_put_lseg(hdr->lseg); |
4db6e0b7 FI |
1591 | nfs_readhdr_free(hdr); |
1592 | } | |
89d77c8f | 1593 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1594 | |
493292dd TM |
1595 | int |
1596 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1597 | { | |
4db6e0b7 FI |
1598 | struct nfs_read_header *rhdr; |
1599 | struct nfs_pgio_header *hdr; | |
493292dd TM |
1600 | int ret; |
1601 | ||
4db6e0b7 FI |
1602 | rhdr = nfs_readhdr_alloc(); |
1603 | if (!rhdr) { | |
061ae2ed | 1604 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
4db6e0b7 | 1605 | ret = -ENOMEM; |
9369a431 | 1606 | pnfs_put_lseg(desc->pg_lseg); |
493292dd TM |
1607 | desc->pg_lseg = NULL; |
1608 | return ret; | |
1609 | } | |
4db6e0b7 FI |
1610 | hdr = &rhdr->header; |
1611 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); | |
9369a431 | 1612 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
4db6e0b7 FI |
1613 | atomic_inc(&hdr->refcnt); |
1614 | ret = nfs_generic_pagein(desc, hdr); | |
1615 | if (ret != 0) { | |
9369a431 | 1616 | pnfs_put_lseg(desc->pg_lseg); |
4db6e0b7 | 1617 | desc->pg_lseg = NULL; |
4db6e0b7 FI |
1618 | } else |
1619 | pnfs_do_multiple_reads(desc, &hdr->rpc_list); | |
1620 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1621 | hdr->completion_ops->completion(hdr); |
4db6e0b7 | 1622 | return ret; |
493292dd TM |
1623 | } |
1624 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1625 | ||
863a3c6c | 1626 | /* |
a9bae566 | 1627 | * There can be multiple RW segments. |
863a3c6c | 1628 | */ |
a9bae566 | 1629 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1630 | { |
a9bae566 | 1631 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1632 | |
a9bae566 PT |
1633 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1634 | if (lseg->pls_range.iomode == IOMODE_RW && | |
1635 | test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
1636 | list_add(&lseg->pls_lc_list, listp); | |
1637 | } | |
863a3c6c AA |
1638 | } |
1639 | ||
1b0ae068 PT |
1640 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
1641 | { | |
b9e028fd | 1642 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
1643 | } |
1644 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
1645 | ||
863a3c6c AA |
1646 | void |
1647 | pnfs_set_layoutcommit(struct nfs_write_data *wdata) | |
1648 | { | |
cd841605 FI |
1649 | struct nfs_pgio_header *hdr = wdata->header; |
1650 | struct inode *inode = hdr->inode; | |
1651 | struct nfs_inode *nfsi = NFS_I(inode); | |
4b8ee2b8 | 1652 | loff_t end_pos = wdata->mds_offset + wdata->res.count; |
79a48a1f | 1653 | bool mark_as_dirty = false; |
863a3c6c | 1654 | |
cd841605 | 1655 | spin_lock(&inode->i_lock); |
863a3c6c | 1656 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 1657 | mark_as_dirty = true; |
863a3c6c | 1658 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 1659 | __func__, inode->i_ino); |
863a3c6c | 1660 | } |
cd841605 | 1661 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 1662 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 1663 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 1664 | } |
acff5880 PT |
1665 | if (end_pos > nfsi->layout->plh_lwb) |
1666 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 1667 | spin_unlock(&inode->i_lock); |
acff5880 | 1668 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 1669 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
1670 | |
1671 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1672 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1673 | if (mark_as_dirty) | |
cd841605 | 1674 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
1675 | } |
1676 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
1677 | ||
db29c089 AA |
1678 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
1679 | { | |
1680 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
1681 | ||
1682 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
1683 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
1684 | } | |
1685 | ||
de4b15c7 AA |
1686 | /* |
1687 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
1688 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
1689 | * data to disk to allow the server to recover the data if it crashes. | |
1690 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
1691 | * is off, and a COMMIT is sent to a data server, or | |
1692 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
1693 | */ | |
863a3c6c | 1694 | int |
ef311537 | 1695 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c AA |
1696 | { |
1697 | struct nfs4_layoutcommit_data *data; | |
1698 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c AA |
1699 | loff_t end_pos; |
1700 | int status = 0; | |
1701 | ||
1702 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); | |
1703 | ||
de4b15c7 AA |
1704 | if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1705 | return 0; | |
1706 | ||
863a3c6c AA |
1707 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ |
1708 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
de4b15c7 | 1709 | if (!data) { |
de4b15c7 AA |
1710 | status = -ENOMEM; |
1711 | goto out; | |
1712 | } | |
863a3c6c | 1713 | |
92407e75 PT |
1714 | if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1715 | goto out_free; | |
1716 | ||
1717 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { | |
1718 | if (!sync) { | |
1719 | status = -EAGAIN; | |
1720 | goto out_free; | |
1721 | } | |
1722 | status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING, | |
1723 | nfs_wait_bit_killable, TASK_KILLABLE); | |
1724 | if (status) | |
1725 | goto out_free; | |
1726 | } | |
1727 | ||
a9bae566 | 1728 | INIT_LIST_HEAD(&data->lseg_list); |
de4b15c7 | 1729 | spin_lock(&inode->i_lock); |
863a3c6c | 1730 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
92407e75 | 1731 | clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags); |
863a3c6c | 1732 | spin_unlock(&inode->i_lock); |
92407e75 PT |
1733 | wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING); |
1734 | goto out_free; | |
863a3c6c | 1735 | } |
a9bae566 PT |
1736 | |
1737 | pnfs_list_write_lseg(inode, &data->lseg_list); | |
863a3c6c | 1738 | |
acff5880 | 1739 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 1740 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 1741 | |
f597c537 | 1742 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
1743 | spin_unlock(&inode->i_lock); |
1744 | ||
1745 | data->args.inode = inode; | |
9fa40758 | 1746 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
1747 | nfs_fattr_init(&data->fattr); |
1748 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
1749 | data->res.fattr = &data->fattr; | |
1750 | data->args.lastbytewritten = end_pos - 1; | |
1751 | data->res.server = NFS_SERVER(inode); | |
1752 | ||
1753 | status = nfs4_proc_layoutcommit(data, sync); | |
1754 | out: | |
92407e75 PT |
1755 | if (status) |
1756 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
1757 | dprintk("<-- %s status %d\n", __func__, status); |
1758 | return status; | |
92407e75 PT |
1759 | out_free: |
1760 | kfree(data); | |
1761 | goto out; | |
863a3c6c | 1762 | } |
82be417a AA |
1763 | |
1764 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
1765 | { | |
1766 | struct nfs4_threshold *thp; | |
1767 | ||
1768 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
1769 | if (!thp) { | |
1770 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
1771 | return NULL; | |
1772 | } | |
1773 | return thp; | |
1774 | } |