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; | |
57934278 | 203 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
204 | } |
205 | ||
206 | static void | |
207 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
208 | { | |
9c626381 TM |
209 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
210 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
211 | ||
212 | if (!list_empty(&lo->plh_layouts)) { | |
213 | struct nfs_client *clp = server->nfs_client; | |
214 | ||
215 | spin_lock(&clp->cl_lock); | |
216 | list_del_init(&lo->plh_layouts); | |
217 | spin_unlock(&clp->cl_lock); | |
218 | } | |
9fa40758 | 219 | put_rpccred(lo->plh_lc_cred); |
57934278 | 220 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
221 | } |
222 | ||
e5e94017 | 223 | static void |
6622c3ea | 224 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 225 | { |
bb346f63 | 226 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 227 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
228 | nfsi->layout = NULL; |
229 | /* Reset MDS Threshold I/O counters */ | |
230 | nfsi->write_io = 0; | |
231 | nfsi->read_io = 0; | |
e5e94017 BH |
232 | } |
233 | ||
b1f69b75 | 234 | void |
70c3bd2b | 235 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 236 | { |
cc6e5340 FI |
237 | struct inode *inode = lo->plh_inode; |
238 | ||
239 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
6622c3ea | 240 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 241 | spin_unlock(&inode->i_lock); |
6622c3ea | 242 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 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 | ||
905ca191 | 317 | static void pnfs_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); |
974cec8c AA |
322 | } |
323 | ||
d684d2ae | 324 | static void |
57036a37 TM |
325 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
326 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 327 | { |
57036a37 | 328 | struct inode *inode = lo->plh_inode; |
d684d2ae | 329 | |
d20581aa | 330 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 331 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
332 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
333 | atomic_dec(&lo->plh_refcount); | |
173f77e9 TM |
334 | if (list_empty(&lo->plh_segs)) |
335 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
d684d2ae FI |
336 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); |
337 | } | |
338 | ||
bae724ef | 339 | void |
9369a431 | 340 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 341 | { |
57036a37 | 342 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
343 | struct inode *inode; |
344 | ||
345 | if (!lseg) | |
346 | return; | |
347 | ||
4541d16c FI |
348 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
349 | atomic_read(&lseg->pls_refcount), | |
350 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
57036a37 TM |
351 | lo = lseg->pls_layout; |
352 | inode = lo->plh_inode; | |
d684d2ae | 353 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
8f0d27dc | 354 | pnfs_get_layout_hdr(lo); |
57036a37 | 355 | pnfs_layout_remove_lseg(lo, lseg); |
d684d2ae | 356 | spin_unlock(&inode->i_lock); |
905ca191 | 357 | pnfs_free_lseg(lseg); |
8f0d27dc | 358 | pnfs_put_layout_hdr(lo); |
4541d16c | 359 | } |
4541d16c | 360 | } |
9369a431 | 361 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 362 | |
fb3296eb BH |
363 | static inline u64 |
364 | end_offset(u64 start, u64 len) | |
365 | { | |
366 | u64 end; | |
367 | ||
368 | end = start + len; | |
369 | return end >= start ? end : NFS4_MAX_UINT64; | |
370 | } | |
371 | ||
fb3296eb BH |
372 | /* |
373 | * is l2 fully contained in l1? | |
374 | * start1 end1 | |
375 | * [----------------------------------) | |
376 | * start2 end2 | |
377 | * [----------------) | |
378 | */ | |
379 | static inline int | |
380 | lo_seg_contained(struct pnfs_layout_range *l1, | |
381 | struct pnfs_layout_range *l2) | |
382 | { | |
383 | u64 start1 = l1->offset; | |
384 | u64 end1 = end_offset(start1, l1->length); | |
385 | u64 start2 = l2->offset; | |
386 | u64 end2 = end_offset(start2, l2->length); | |
387 | ||
388 | return (start1 <= start2) && (end1 >= end2); | |
389 | } | |
390 | ||
391 | /* | |
392 | * is l1 and l2 intersecting? | |
393 | * start1 end1 | |
394 | * [----------------------------------) | |
395 | * start2 end2 | |
396 | * [----------------) | |
397 | */ | |
398 | static inline int | |
399 | lo_seg_intersecting(struct pnfs_layout_range *l1, | |
400 | struct pnfs_layout_range *l2) | |
401 | { | |
402 | u64 start1 = l1->offset; | |
403 | u64 end1 = end_offset(start1, l1->length); | |
404 | u64 start2 = l2->offset; | |
405 | u64 end2 = end_offset(start2, l2->length); | |
406 | ||
407 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
408 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
409 | } | |
410 | ||
4541d16c | 411 | static bool |
778b5502 BH |
412 | should_free_lseg(struct pnfs_layout_range *lseg_range, |
413 | struct pnfs_layout_range *recall_range) | |
4541d16c | 414 | { |
778b5502 BH |
415 | return (recall_range->iomode == IOMODE_ANY || |
416 | lseg_range->iomode == recall_range->iomode) && | |
417 | lo_seg_intersecting(lseg_range, recall_range); | |
974cec8c AA |
418 | } |
419 | ||
4541d16c FI |
420 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
421 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
422 | struct list_head *tmp_list) | |
423 | { | |
424 | int rv = 0; | |
425 | ||
426 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
427 | /* Remove the reference keeping the lseg in the | |
428 | * list. It will now be removed when all | |
429 | * outstanding io is finished. | |
430 | */ | |
d684d2ae FI |
431 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
432 | atomic_read(&lseg->pls_refcount)); | |
433 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
57036a37 | 434 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); |
d684d2ae FI |
435 | list_add(&lseg->pls_list, tmp_list); |
436 | rv = 1; | |
437 | } | |
4541d16c FI |
438 | } |
439 | return rv; | |
440 | } | |
441 | ||
442 | /* Returns count of number of matching invalid lsegs remaining in list | |
443 | * after call. | |
444 | */ | |
43f1b3da | 445 | int |
49a85061 | 446 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 447 | struct list_head *tmp_list, |
778b5502 | 448 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
449 | { |
450 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 451 | int invalid = 0, removed = 0; |
974cec8c AA |
452 | |
453 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
454 | ||
8006bfba | 455 | if (list_empty(&lo->plh_segs)) |
38511722 | 456 | return 0; |
4541d16c | 457 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
458 | if (!recall_range || |
459 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
460 | dprintk("%s: freeing lseg %p iomode %d " |
461 | "offset %llu length %llu\n", __func__, | |
462 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
463 | lseg->pls_range.length); | |
464 | invalid++; | |
465 | removed += mark_lseg_invalid(lseg, tmp_list); | |
466 | } | |
467 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
468 | return invalid - removed; | |
974cec8c AA |
469 | } |
470 | ||
f49f9baa | 471 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 472 | void |
4541d16c | 473 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 474 | { |
4541d16c | 475 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
476 | |
477 | if (list_empty(free_me)) | |
478 | return; | |
479 | ||
4541d16c | 480 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 481 | list_del(&lseg->pls_list); |
905ca191 | 482 | pnfs_free_lseg(lseg); |
974cec8c AA |
483 | } |
484 | } | |
485 | ||
e5e94017 BH |
486 | void |
487 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
488 | { | |
489 | struct pnfs_layout_hdr *lo; | |
974cec8c | 490 | LIST_HEAD(tmp_list); |
e5e94017 BH |
491 | |
492 | spin_lock(&nfsi->vfs_inode.i_lock); | |
493 | lo = nfsi->layout; | |
494 | if (lo) { | |
38511722 | 495 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 496 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
3e621214 TM |
497 | pnfs_get_layout_hdr(lo); |
498 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); | |
499 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
500 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
501 | pnfs_free_lseg_list(&tmp_list); | |
502 | pnfs_put_layout_hdr(lo); | |
503 | } else | |
504 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 505 | } |
041245c8 | 506 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c AA |
507 | |
508 | /* | |
509 | * Called by the state manger to remove all layouts established under an | |
510 | * expired lease. | |
511 | */ | |
512 | void | |
513 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
514 | { | |
6382a441 | 515 | struct nfs_server *server; |
974cec8c AA |
516 | struct pnfs_layout_hdr *lo; |
517 | LIST_HEAD(tmp_list); | |
518 | ||
c47abcf8 AA |
519 | nfs4_deviceid_mark_client_invalid(clp); |
520 | nfs4_deviceid_purge_client(clp); | |
521 | ||
974cec8c | 522 | spin_lock(&clp->cl_lock); |
6382a441 WAA |
523 | rcu_read_lock(); |
524 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
525 | if (!list_empty(&server->layouts)) | |
526 | list_splice_init(&server->layouts, &tmp_list); | |
527 | } | |
528 | rcu_read_unlock(); | |
974cec8c AA |
529 | spin_unlock(&clp->cl_lock); |
530 | ||
531 | while (!list_empty(&tmp_list)) { | |
532 | lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, | |
b7edfaa1 | 533 | plh_layouts); |
974cec8c | 534 | dprintk("%s freeing layout for inode %lu\n", __func__, |
b7edfaa1 | 535 | lo->plh_inode->i_ino); |
2887fe45 | 536 | list_del_init(&lo->plh_layouts); |
b7edfaa1 | 537 | pnfs_destroy_layout(NFS_I(lo->plh_inode)); |
974cec8c | 538 | } |
e5e94017 BH |
539 | } |
540 | ||
5a65503f TM |
541 | /* |
542 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
543 | * taking into account wraparound issues. | |
544 | */ | |
545 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
546 | { | |
547 | return (s32)s1 - (s32)s2 > 0; | |
548 | } | |
549 | ||
fd6002e9 | 550 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
551 | void |
552 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
553 | bool update_barrier) | |
b1f69b75 | 554 | { |
22aaf714 TM |
555 | u32 oldseq, newseq, new_barrier; |
556 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 557 | |
2d2f24ad TM |
558 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
559 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 560 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 561 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 562 | if (update_barrier) { |
22aaf714 | 563 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
564 | } else { |
565 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 566 | * "close" to the current seqids. |
43f1b3da | 567 | */ |
22aaf714 | 568 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 569 | } |
22aaf714 TM |
570 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
571 | lo->plh_barrier = new_barrier; | |
43f1b3da | 572 | } |
b1f69b75 AA |
573 | } |
574 | ||
cf7d63f1 | 575 | static bool |
19c54aba TM |
576 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
577 | const nfs4_stateid *stateid) | |
43f1b3da | 578 | { |
19c54aba | 579 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 580 | |
19c54aba TM |
581 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
582 | } | |
583 | ||
584 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
585 | static bool | |
586 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget) | |
587 | { | |
f7e8917a FI |
588 | return lo->plh_block_lgets || |
589 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || | |
43f1b3da | 590 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
591 | (atomic_read(&lo->plh_outstanding) > lget)); |
592 | } | |
593 | ||
fd6002e9 FI |
594 | int |
595 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
596 | struct nfs4_state *open_state) | |
b1f69b75 | 597 | { |
fd6002e9 | 598 | int status = 0; |
974cec8c | 599 | |
b1f69b75 | 600 | dprintk("--> %s\n", __func__); |
fd6002e9 | 601 | spin_lock(&lo->plh_inode->i_lock); |
19c54aba | 602 | if (pnfs_layoutgets_blocked(lo, 1)) { |
cf7d63f1 FI |
603 | status = -EAGAIN; |
604 | } else if (list_empty(&lo->plh_segs)) { | |
fd6002e9 FI |
605 | int seq; |
606 | ||
607 | do { | |
608 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 609 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
610 | } while (read_seqretry(&open_state->seqlock, seq)); |
611 | } else | |
f597c537 | 612 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 613 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 614 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 615 | return status; |
b1f69b75 AA |
616 | } |
617 | ||
618 | /* | |
619 | * Get layout from server. | |
620 | * for now, assume that whole file layouts are requested. | |
621 | * arg->offset: 0 | |
622 | * arg->length: all ones | |
623 | */ | |
e5e94017 BH |
624 | static struct pnfs_layout_segment * |
625 | send_layoutget(struct pnfs_layout_hdr *lo, | |
626 | struct nfs_open_context *ctx, | |
fb3296eb | 627 | struct pnfs_layout_range *range, |
a75b9df9 | 628 | gfp_t gfp_flags) |
e5e94017 | 629 | { |
b7edfaa1 | 630 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
631 | struct nfs_server *server = NFS_SERVER(ino); |
632 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 633 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
634 | |
635 | dprintk("--> %s\n", __func__); | |
e5e94017 | 636 | |
b1f69b75 | 637 | BUG_ON(ctx == NULL); |
a75b9df9 | 638 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 639 | if (lgp == NULL) |
b1f69b75 | 640 | return NULL; |
35124a09 | 641 | |
fb3296eb BH |
642 | lgp->args.minlength = PAGE_CACHE_SIZE; |
643 | if (lgp->args.minlength > range->length) | |
644 | lgp->args.minlength = range->length; | |
b1f69b75 | 645 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 646 | lgp->args.range = *range; |
b1f69b75 AA |
647 | lgp->args.type = server->pnfs_curr_ld->id; |
648 | lgp->args.inode = ino; | |
649 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 650 | lgp->gfp_flags = gfp_flags; |
b1f69b75 AA |
651 | |
652 | /* Synchronously retrieve layout information from server and | |
653 | * store in lseg. | |
654 | */ | |
a0b0a6e3 TM |
655 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
656 | if (IS_ERR(lseg)) { | |
657 | switch (PTR_ERR(lseg)) { | |
658 | case -ENOMEM: | |
659 | case -ERESTARTSYS: | |
660 | break; | |
661 | default: | |
662 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 663 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
664 | } |
665 | return NULL; | |
974cec8c | 666 | } |
35124a09 | 667 | |
974cec8c AA |
668 | return lseg; |
669 | } | |
670 | ||
293b3b06 AA |
671 | /* |
672 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
673 | * when the layout segment list is empty. | |
674 | * | |
675 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
676 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
677 | * deviceid is marked invalid. | |
678 | */ | |
cbe82603 BH |
679 | int |
680 | _pnfs_return_layout(struct inode *ino) | |
681 | { | |
682 | struct pnfs_layout_hdr *lo = NULL; | |
683 | struct nfs_inode *nfsi = NFS_I(ino); | |
684 | LIST_HEAD(tmp_list); | |
685 | struct nfs4_layoutreturn *lrp; | |
686 | nfs4_stateid stateid; | |
293b3b06 | 687 | int status = 0, empty; |
cbe82603 | 688 | |
366d5052 | 689 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
690 | |
691 | spin_lock(&ino->i_lock); | |
692 | lo = nfsi->layout; | |
e5929f3c | 693 | if (!lo) { |
cbe82603 | 694 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
695 | dprintk("NFS: %s no layout to return\n", __func__); |
696 | goto out; | |
cbe82603 BH |
697 | } |
698 | stateid = nfsi->layout->plh_stateid; | |
699 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 700 | pnfs_get_layout_hdr(lo); |
293b3b06 | 701 | empty = list_empty(&lo->plh_segs); |
49a85061 | 702 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
293b3b06 AA |
703 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
704 | if (empty) { | |
705 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 706 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
707 | dprintk("NFS: %s no layout segments to return\n", __func__); |
708 | goto out; | |
709 | } | |
ea0ded74 | 710 | lo->plh_block_lgets++; |
cbe82603 BH |
711 | spin_unlock(&ino->i_lock); |
712 | pnfs_free_lseg_list(&tmp_list); | |
713 | ||
714 | WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)); | |
715 | ||
716 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); | |
717 | if (unlikely(lrp == NULL)) { | |
718 | status = -ENOMEM; | |
65857d57 TM |
719 | spin_lock(&ino->i_lock); |
720 | lo->plh_block_lgets--; | |
721 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 722 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
723 | goto out; |
724 | } | |
725 | ||
726 | lrp->args.stateid = stateid; | |
727 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
728 | lrp->args.inode = ino; | |
a56aaa02 | 729 | lrp->args.layout = lo; |
cbe82603 BH |
730 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
731 | ||
732 | status = nfs4_proc_layoutreturn(lrp); | |
733 | out: | |
734 | dprintk("<-- %s status: %d\n", __func__, status); | |
735 | return status; | |
736 | } | |
0a57cdac | 737 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 738 | |
f7e8917a FI |
739 | bool pnfs_roc(struct inode *ino) |
740 | { | |
741 | struct pnfs_layout_hdr *lo; | |
742 | struct pnfs_layout_segment *lseg, *tmp; | |
743 | LIST_HEAD(tmp_list); | |
744 | bool found = false; | |
745 | ||
746 | spin_lock(&ino->i_lock); | |
747 | lo = NFS_I(ino)->layout; | |
748 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
749 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
750 | goto out_nolayout; | |
751 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
752 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
753 | mark_lseg_invalid(lseg, &tmp_list); | |
754 | found = true; | |
755 | } | |
756 | if (!found) | |
757 | goto out_nolayout; | |
758 | lo->plh_block_lgets++; | |
70c3bd2b | 759 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
760 | spin_unlock(&ino->i_lock); |
761 | pnfs_free_lseg_list(&tmp_list); | |
762 | return true; | |
763 | ||
764 | out_nolayout: | |
765 | spin_unlock(&ino->i_lock); | |
766 | return false; | |
767 | } | |
768 | ||
769 | void pnfs_roc_release(struct inode *ino) | |
770 | { | |
771 | struct pnfs_layout_hdr *lo; | |
772 | ||
773 | spin_lock(&ino->i_lock); | |
774 | lo = NFS_I(ino)->layout; | |
775 | lo->plh_block_lgets--; | |
6622c3ea TM |
776 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
777 | pnfs_detach_layout_hdr(lo); | |
778 | spin_unlock(&ino->i_lock); | |
779 | pnfs_free_layout_hdr(lo); | |
780 | } else | |
781 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
782 | } |
783 | ||
784 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
785 | { | |
786 | struct pnfs_layout_hdr *lo; | |
787 | ||
788 | spin_lock(&ino->i_lock); | |
789 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 790 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
791 | lo->plh_barrier = barrier; |
792 | spin_unlock(&ino->i_lock); | |
793 | } | |
794 | ||
7fdab069 | 795 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
796 | { |
797 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 798 | struct pnfs_layout_hdr *lo; |
f7e8917a | 799 | struct pnfs_layout_segment *lseg; |
7fdab069 | 800 | u32 current_seqid; |
f7e8917a FI |
801 | bool found = false; |
802 | ||
803 | spin_lock(&ino->i_lock); | |
804 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
805 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 806 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 807 | found = true; |
7fdab069 | 808 | goto out; |
f7e8917a | 809 | } |
7fdab069 TM |
810 | lo = nfsi->layout; |
811 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 812 | |
7fdab069 TM |
813 | /* Since close does not return a layout stateid for use as |
814 | * a barrier, we choose the worst-case barrier. | |
815 | */ | |
816 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
817 | out: | |
f7e8917a FI |
818 | spin_unlock(&ino->i_lock); |
819 | return found; | |
820 | } | |
821 | ||
b1f69b75 AA |
822 | /* |
823 | * Compare two layout segments for sorting into layout cache. | |
824 | * We want to preferentially return RW over RO layouts, so ensure those | |
825 | * are seen first. | |
826 | */ | |
827 | static s64 | |
fb3296eb BH |
828 | cmp_layout(struct pnfs_layout_range *l1, |
829 | struct pnfs_layout_range *l2) | |
b1f69b75 | 830 | { |
fb3296eb BH |
831 | s64 d; |
832 | ||
833 | /* high offset > low offset */ | |
834 | d = l1->offset - l2->offset; | |
835 | if (d) | |
836 | return d; | |
837 | ||
838 | /* short length > long length */ | |
839 | d = l2->length - l1->length; | |
840 | if (d) | |
841 | return d; | |
842 | ||
b1f69b75 | 843 | /* read > read/write */ |
fb3296eb | 844 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
845 | } |
846 | ||
974cec8c | 847 | static void |
57036a37 | 848 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
849 | struct pnfs_layout_segment *lseg) |
850 | { | |
b1f69b75 | 851 | struct pnfs_layout_segment *lp; |
b1f69b75 | 852 | |
974cec8c AA |
853 | dprintk("%s:Begin\n", __func__); |
854 | ||
b7edfaa1 | 855 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
fb3296eb | 856 | if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 857 | continue; |
566052c5 | 858 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
859 | dprintk("%s: inserted lseg %p " |
860 | "iomode %d offset %llu length %llu before " | |
861 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
862 | __func__, lseg, lseg->pls_range.iomode, |
863 | lseg->pls_range.offset, lseg->pls_range.length, | |
864 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
865 | lp->pls_range.length); | |
fb3296eb | 866 | goto out; |
974cec8c | 867 | } |
fb3296eb BH |
868 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
869 | dprintk("%s: inserted lseg %p " | |
870 | "iomode %d offset %llu length %llu at tail\n", | |
871 | __func__, lseg, lseg->pls_range.iomode, | |
872 | lseg->pls_range.offset, lseg->pls_range.length); | |
873 | out: | |
70c3bd2b | 874 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
875 | |
876 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
877 | } |
878 | ||
879 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
880 | alloc_init_layout_hdr(struct inode *ino, |
881 | struct nfs_open_context *ctx, | |
882 | gfp_t gfp_flags) | |
e5e94017 BH |
883 | { |
884 | struct pnfs_layout_hdr *lo; | |
885 | ||
636fb9c8 | 886 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
887 | if (!lo) |
888 | return NULL; | |
cc6e5340 | 889 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
890 | INIT_LIST_HEAD(&lo->plh_layouts); |
891 | INIT_LIST_HEAD(&lo->plh_segs); | |
43f1b3da | 892 | INIT_LIST_HEAD(&lo->plh_bulk_recall); |
b7edfaa1 | 893 | lo->plh_inode = ino; |
9fa40758 | 894 | lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred); |
e5e94017 BH |
895 | return lo; |
896 | } | |
897 | ||
898 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
899 | pnfs_find_alloc_layout(struct inode *ino, |
900 | struct nfs_open_context *ctx, | |
901 | gfp_t gfp_flags) | |
e5e94017 BH |
902 | { |
903 | struct nfs_inode *nfsi = NFS_I(ino); | |
904 | struct pnfs_layout_hdr *new = NULL; | |
905 | ||
906 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
907 | ||
251ec410 TM |
908 | if (nfsi->layout != NULL) |
909 | goto out_existing; | |
e5e94017 | 910 | spin_unlock(&ino->i_lock); |
9fa40758 | 911 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
912 | spin_lock(&ino->i_lock); |
913 | ||
251ec410 | 914 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 915 | nfsi->layout = new; |
251ec410 | 916 | return new; |
7175fe90 YN |
917 | } else if (new != NULL) |
918 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
919 | out_existing: |
920 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
921 | return nfsi->layout; |
922 | } | |
923 | ||
b1f69b75 AA |
924 | /* |
925 | * iomode matching rules: | |
926 | * iomode lseg match | |
927 | * ----- ----- ----- | |
928 | * ANY READ true | |
929 | * ANY RW true | |
930 | * RW READ false | |
931 | * RW RW true | |
932 | * READ READ true | |
933 | * READ RW true | |
934 | */ | |
935 | static int | |
fb3296eb BH |
936 | is_matching_lseg(struct pnfs_layout_range *ls_range, |
937 | struct pnfs_layout_range *range) | |
b1f69b75 | 938 | { |
fb3296eb BH |
939 | struct pnfs_layout_range range1; |
940 | ||
941 | if ((range->iomode == IOMODE_RW && | |
942 | ls_range->iomode != IOMODE_RW) || | |
943 | !lo_seg_intersecting(ls_range, range)) | |
944 | return 0; | |
945 | ||
946 | /* range1 covers only the first byte in the range */ | |
947 | range1 = *range; | |
948 | range1.length = 1; | |
949 | return lo_seg_contained(ls_range, &range1); | |
b1f69b75 AA |
950 | } |
951 | ||
952 | /* | |
953 | * lookup range in layout | |
954 | */ | |
e5e94017 | 955 | static struct pnfs_layout_segment * |
fb3296eb BH |
956 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
957 | struct pnfs_layout_range *range) | |
e5e94017 | 958 | { |
b1f69b75 AA |
959 | struct pnfs_layout_segment *lseg, *ret = NULL; |
960 | ||
961 | dprintk("%s:Begin\n", __func__); | |
962 | ||
b7edfaa1 | 963 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 964 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
fb3296eb | 965 | is_matching_lseg(&lseg->pls_range, range)) { |
9369a431 | 966 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
967 | break; |
968 | } | |
d771e3a4 | 969 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
970 | break; |
971 | } | |
972 | ||
973 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 974 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 975 | return ret; |
e5e94017 BH |
976 | } |
977 | ||
d23d61c8 AA |
978 | /* |
979 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
980 | * to the MDS or over pNFS | |
981 | * | |
982 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
983 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
984 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
985 | * WRITE request. | |
986 | * | |
987 | * A return of true means use MDS I/O. | |
988 | * | |
989 | * From rfc 5661: | |
990 | * If a file's size is smaller than the file size threshold, data accesses | |
991 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
992 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
993 | * server. If both file size and I/O size are provided, the client SHOULD | |
994 | * reach or exceed both thresholds before sending its read or write | |
995 | * requests to the data server. | |
996 | */ | |
997 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
998 | struct inode *ino, int iomode) | |
999 | { | |
1000 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1001 | struct nfs_inode *nfsi = NFS_I(ino); | |
1002 | loff_t fsize = i_size_read(ino); | |
1003 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1004 | ||
1005 | if (t == NULL) | |
1006 | return ret; | |
1007 | ||
1008 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1009 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1010 | ||
1011 | switch (iomode) { | |
1012 | case IOMODE_READ: | |
1013 | if (t->bm & THRESHOLD_RD) { | |
1014 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1015 | size_set = true; | |
1016 | if (fsize < t->rd_sz) | |
1017 | size = true; | |
1018 | } | |
1019 | if (t->bm & THRESHOLD_RD_IO) { | |
1020 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1021 | nfsi->read_io); | |
1022 | io_set = true; | |
1023 | if (nfsi->read_io < t->rd_io_sz) | |
1024 | io = true; | |
1025 | } | |
1026 | break; | |
1027 | case IOMODE_RW: | |
1028 | if (t->bm & THRESHOLD_WR) { | |
1029 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1030 | size_set = true; | |
1031 | if (fsize < t->wr_sz) | |
1032 | size = true; | |
1033 | } | |
1034 | if (t->bm & THRESHOLD_WR_IO) { | |
1035 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1036 | nfsi->write_io); | |
1037 | io_set = true; | |
1038 | if (nfsi->write_io < t->wr_io_sz) | |
1039 | io = true; | |
1040 | } | |
1041 | break; | |
1042 | } | |
1043 | if (size_set && io_set) { | |
1044 | if (size && io) | |
1045 | ret = true; | |
1046 | } else if (size || io) | |
1047 | ret = true; | |
1048 | ||
1049 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1050 | return ret; | |
1051 | } | |
1052 | ||
e5e94017 BH |
1053 | /* |
1054 | * Layout segment is retreived from the server if not cached. | |
1055 | * The appropriate layout segment is referenced and returned to the caller. | |
1056 | */ | |
7c24d948 | 1057 | struct pnfs_layout_segment * |
e5e94017 BH |
1058 | pnfs_update_layout(struct inode *ino, |
1059 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1060 | loff_t pos, |
1061 | u64 count, | |
a75b9df9 TM |
1062 | enum pnfs_iomode iomode, |
1063 | gfp_t gfp_flags) | |
e5e94017 | 1064 | { |
fb3296eb BH |
1065 | struct pnfs_layout_range arg = { |
1066 | .iomode = iomode, | |
1067 | .offset = pos, | |
1068 | .length = count, | |
1069 | }; | |
707ed5fd | 1070 | unsigned pg_offset; |
6382a441 WAA |
1071 | struct nfs_server *server = NFS_SERVER(ino); |
1072 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1073 | struct pnfs_layout_hdr *lo; |
1074 | struct pnfs_layout_segment *lseg = NULL; | |
f49f9baa | 1075 | bool first = false; |
e5e94017 BH |
1076 | |
1077 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1078 | goto out; |
d23d61c8 AA |
1079 | |
1080 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1081 | goto out; |
d23d61c8 | 1082 | |
e5e94017 | 1083 | spin_lock(&ino->i_lock); |
9fa40758 | 1084 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1085 | if (lo == NULL) { |
1086 | spin_unlock(&ino->i_lock); | |
1087 | goto out; | |
1088 | } | |
e5e94017 | 1089 | |
43f1b3da | 1090 | /* Do we even need to bother with this? */ |
a59c30ac | 1091 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1092 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1093 | goto out_unlock; |
1094 | } | |
1095 | ||
1096 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1097 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1098 | goto out_unlock; |
1099 | ||
568e8c49 | 1100 | /* Check to see if the layout for the given range already exists */ |
fb3296eb | 1101 | lseg = pnfs_find_lseg(lo, &arg); |
568e8c49 AA |
1102 | if (lseg) |
1103 | goto out_unlock; | |
1104 | ||
19c54aba | 1105 | if (pnfs_layoutgets_blocked(lo, 0)) |
cf7d63f1 FI |
1106 | goto out_unlock; |
1107 | atomic_inc(&lo->plh_outstanding); | |
1108 | ||
f49f9baa FI |
1109 | if (list_empty(&lo->plh_segs)) |
1110 | first = true; | |
366d5052 | 1111 | |
f49f9baa FI |
1112 | spin_unlock(&ino->i_lock); |
1113 | if (first) { | |
2130ff66 FI |
1114 | /* The lo must be on the clp list if there is any |
1115 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1116 | */ | |
1117 | spin_lock(&clp->cl_lock); | |
1118 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
6382a441 | 1119 | list_add_tail(&lo->plh_layouts, &server->layouts); |
2130ff66 FI |
1120 | spin_unlock(&clp->cl_lock); |
1121 | } | |
e5e94017 | 1122 | |
707ed5fd BH |
1123 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1124 | if (pg_offset) { | |
1125 | arg.offset -= pg_offset; | |
1126 | arg.length += pg_offset; | |
1127 | } | |
7c24d948 AA |
1128 | if (arg.length != NFS4_MAX_UINT64) |
1129 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1130 | |
fb3296eb | 1131 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
cf7d63f1 | 1132 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1133 | out_put_layout_hdr: |
70c3bd2b | 1134 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1135 | out: |
f86bbcf8 TM |
1136 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1137 | "(%s, offset: %llu, length: %llu)\n", | |
1138 | __func__, ino->i_sb->s_id, | |
1139 | (unsigned long long)NFS_FILEID(ino), | |
1140 | lseg == NULL ? "not found" : "found", | |
1141 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1142 | (unsigned long long)pos, | |
1143 | (unsigned long long)count); | |
e5e94017 BH |
1144 | return lseg; |
1145 | out_unlock: | |
1146 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1147 | goto out_put_layout_hdr; |
e5e94017 | 1148 | } |
7c24d948 | 1149 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1150 | |
a0b0a6e3 | 1151 | struct pnfs_layout_segment * |
b1f69b75 AA |
1152 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1153 | { | |
1154 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1155 | struct nfs4_layoutget_res *res = &lgp->res; | |
1156 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1157 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
1158 | int status = 0; |
1159 | ||
1160 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1161 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1162 | if (!lseg || IS_ERR(lseg)) { |
1163 | if (!lseg) | |
1164 | status = -ENOMEM; | |
1165 | else | |
1166 | status = PTR_ERR(lseg); | |
1167 | dprintk("%s: Could not allocate layout: error %d\n", | |
1168 | __func__, status); | |
1169 | goto out; | |
1170 | } | |
1171 | ||
1172 | spin_lock(&ino->i_lock); | |
a59c30ac | 1173 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1174 | dprintk("%s forget reply due to recall\n", __func__); |
1175 | goto out_forget_reply; | |
1176 | } | |
1177 | ||
19c54aba TM |
1178 | if (pnfs_layoutgets_blocked(lo, 1) || |
1179 | pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
43f1b3da FI |
1180 | dprintk("%s forget reply due to state\n", __func__); |
1181 | goto out_forget_reply; | |
1182 | } | |
038d6493 TM |
1183 | |
1184 | /* Done processing layoutget. Set the layout stateid */ | |
1185 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1186 | ||
b1f69b75 | 1187 | init_lseg(lo, lseg); |
566052c5 | 1188 | lseg->pls_range = res->range; |
9369a431 | 1189 | pnfs_get_lseg(lseg); |
57036a37 | 1190 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1191 | |
f7e8917a FI |
1192 | if (res->return_on_close) { |
1193 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1194 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1195 | } | |
1196 | ||
b1f69b75 | 1197 | spin_unlock(&ino->i_lock); |
a0b0a6e3 | 1198 | return lseg; |
b1f69b75 | 1199 | out: |
a0b0a6e3 | 1200 | return ERR_PTR(status); |
43f1b3da FI |
1201 | |
1202 | out_forget_reply: | |
1203 | spin_unlock(&ino->i_lock); | |
1204 | lseg->pls_layout = lo; | |
1205 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1206 | goto out; | |
b1f69b75 AA |
1207 | } |
1208 | ||
d8007d4d TM |
1209 | void |
1210 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1211 | { | |
1fd937bd PT |
1212 | u64 rd_size = req->wb_bytes; |
1213 | ||
d8007d4d TM |
1214 | BUG_ON(pgio->pg_lseg != NULL); |
1215 | ||
1825a0d0 FI |
1216 | if (req->wb_offset != req->wb_pgbase) { |
1217 | nfs_pageio_reset_read_mds(pgio); | |
1218 | return; | |
1219 | } | |
1fd937bd PT |
1220 | |
1221 | if (pgio->pg_dreq == NULL) | |
1222 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1223 | else | |
1224 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1225 | ||
d8007d4d TM |
1226 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1227 | req->wb_context, | |
1228 | req_offset(req), | |
1fd937bd | 1229 | rd_size, |
d8007d4d TM |
1230 | IOMODE_READ, |
1231 | GFP_KERNEL); | |
e885de1a TM |
1232 | /* If no lseg, fall back to read through mds */ |
1233 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1234 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1235 | |
d8007d4d TM |
1236 | } |
1237 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1238 | ||
1239 | void | |
6296556f PT |
1240 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1241 | struct nfs_page *req, u64 wb_size) | |
d8007d4d TM |
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 | } | |
6296556f | 1249 | |
d8007d4d TM |
1250 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1251 | req->wb_context, | |
1252 | req_offset(req), | |
6296556f | 1253 | wb_size, |
d8007d4d TM |
1254 | IOMODE_RW, |
1255 | GFP_NOFS); | |
e885de1a TM |
1256 | /* If no lseg, fall back to write through mds */ |
1257 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1258 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1259 | } |
1260 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1261 | ||
1abb5088 | 1262 | void |
061ae2ed FI |
1263 | pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode, |
1264 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 TM |
1265 | { |
1266 | struct nfs_server *server = NFS_SERVER(inode); | |
1267 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
1268 | ||
1269 | if (ld == NULL) | |
1abb5088 BS |
1270 | nfs_pageio_init_read(pgio, inode, compl_ops); |
1271 | else | |
1272 | nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0); | |
1751c363 TM |
1273 | } |
1274 | ||
57208fa7 | 1275 | void |
061ae2ed FI |
1276 | pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode, |
1277 | int ioflags, | |
1278 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 TM |
1279 | { |
1280 | struct nfs_server *server = NFS_SERVER(inode); | |
1281 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
1282 | ||
1283 | if (ld == NULL) | |
57208fa7 BS |
1284 | nfs_pageio_init_write(pgio, inode, ioflags, compl_ops); |
1285 | else | |
1286 | nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags); | |
1751c363 TM |
1287 | } |
1288 | ||
18ad0a9f | 1289 | bool |
dfed206b BH |
1290 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
1291 | struct nfs_page *req) | |
94ad1c80 | 1292 | { |
d8007d4d TM |
1293 | if (pgio->pg_lseg == NULL) |
1294 | return nfs_generic_pg_test(pgio, prev, req); | |
94ad1c80 | 1295 | |
19982ba8 TM |
1296 | /* |
1297 | * Test if a nfs_page is fully contained in the pnfs_layout_range. | |
1298 | * Note that this test makes several assumptions: | |
1299 | * - that the previous nfs_page in the struct nfs_pageio_descriptor | |
1300 | * is known to lie within the range. | |
1301 | * - that the nfs_page being tested is known to be contiguous with the | |
1302 | * previous nfs_page. | |
1303 | * - Layout ranges are page aligned, so we only have to test the | |
1304 | * start offset of the request. | |
1305 | * | |
1306 | * Please also note that 'end_offset' is actually the offset of the | |
1307 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1308 | * | |
1309 | */ | |
1310 | return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset, | |
1311 | pgio->pg_lseg->pls_range.length); | |
94ad1c80 | 1312 | } |
89a58e32 | 1313 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1314 | |
e7dd79af | 1315 | int pnfs_write_done_resend_to_mds(struct inode *inode, |
061ae2ed FI |
1316 | struct list_head *head, |
1317 | const struct nfs_pgio_completion_ops *compl_ops) | |
e2fecb21 TM |
1318 | { |
1319 | struct nfs_pageio_descriptor pgio; | |
1320 | LIST_HEAD(failed); | |
1321 | ||
1322 | /* Resend all requests through the MDS */ | |
57208fa7 | 1323 | nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops); |
e2fecb21 TM |
1324 | while (!list_empty(head)) { |
1325 | struct nfs_page *req = nfs_list_entry(head->next); | |
1326 | ||
1327 | nfs_list_remove_request(req); | |
1328 | if (!nfs_pageio_add_request(&pgio, req)) | |
1329 | nfs_list_add_request(req, &failed); | |
1330 | } | |
1331 | nfs_pageio_complete(&pgio); | |
1332 | ||
1333 | if (!list_empty(&failed)) { | |
1334 | /* For some reason our attempt to resend pages. Mark the | |
1335 | * overall send request as having failed, and let | |
1336 | * nfs_writeback_release_full deal with the error. | |
1337 | */ | |
1338 | list_move(&failed, head); | |
1339 | return -EIO; | |
1340 | } | |
1341 | return 0; | |
1342 | } | |
e7dd79af | 1343 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1344 | |
1acbbb4e FI |
1345 | static void pnfs_ld_handle_write_error(struct nfs_write_data *data) |
1346 | { | |
cd841605 FI |
1347 | struct nfs_pgio_header *hdr = data->header; |
1348 | ||
1349 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1350 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1351 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 FI |
1352 | clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags); |
1353 | pnfs_return_layout(hdr->inode); | |
1acbbb4e | 1354 | } |
6c75dc0d FI |
1355 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
1356 | data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode, | |
061ae2ed FI |
1357 | &hdr->pages, |
1358 | hdr->completion_ops); | |
1acbbb4e FI |
1359 | } |
1360 | ||
d20581aa BH |
1361 | /* |
1362 | * Called by non rpc-based layout drivers | |
1363 | */ | |
8ce160c5 | 1364 | void pnfs_ld_write_done(struct nfs_write_data *data) |
44b83799 | 1365 | { |
cd841605 FI |
1366 | struct nfs_pgio_header *hdr = data->header; |
1367 | ||
1368 | if (!hdr->pnfs_error) { | |
d20581aa | 1369 | pnfs_set_layoutcommit(data); |
cd841605 | 1370 | hdr->mds_ops->rpc_call_done(&data->task, data); |
1acbbb4e FI |
1371 | } else |
1372 | pnfs_ld_handle_write_error(data); | |
cd841605 | 1373 | hdr->mds_ops->rpc_release(data); |
44b83799 | 1374 | } |
d20581aa | 1375 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1376 | |
dce81290 TM |
1377 | static void |
1378 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
1379 | struct nfs_write_data *data) | |
1380 | { | |
cd841605 FI |
1381 | struct nfs_pgio_header *hdr = data->header; |
1382 | ||
6c75dc0d FI |
1383 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1384 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1385 | nfs_pageio_reset_write_mds(desc); | |
1386 | desc->pg_recoalesce = 1; | |
1387 | } | |
dce81290 TM |
1388 | nfs_writedata_release(data); |
1389 | } | |
1390 | ||
1391 | static enum pnfs_try_status | |
0382b744 | 1392 | pnfs_try_to_write_data(struct nfs_write_data *wdata, |
dce81290 TM |
1393 | const struct rpc_call_ops *call_ops, |
1394 | struct pnfs_layout_segment *lseg, | |
1395 | int how) | |
0382b744 | 1396 | { |
cd841605 FI |
1397 | struct nfs_pgio_header *hdr = wdata->header; |
1398 | struct inode *inode = hdr->inode; | |
0382b744 AA |
1399 | enum pnfs_try_status trypnfs; |
1400 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1401 | ||
cd841605 | 1402 | hdr->mds_ops = call_ops; |
0382b744 AA |
1403 | |
1404 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
1405 | inode->i_ino, wdata->args.count, wdata->args.offset, how); | |
0382b744 | 1406 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how); |
6c75dc0d | 1407 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1408 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1409 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1410 | return trypnfs; | |
1411 | } | |
1412 | ||
dce81290 TM |
1413 | static void |
1414 | pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how) | |
1415 | { | |
1416 | struct nfs_write_data *data; | |
1417 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; | |
1418 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
1419 | ||
1420 | desc->pg_lseg = NULL; | |
1421 | while (!list_empty(head)) { | |
1422 | enum pnfs_try_status trypnfs; | |
1423 | ||
6c75dc0d | 1424 | data = list_first_entry(head, struct nfs_write_data, list); |
dce81290 TM |
1425 | list_del_init(&data->list); |
1426 | ||
1427 | trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how); | |
1428 | if (trypnfs == PNFS_NOT_ATTEMPTED) | |
1429 | pnfs_write_through_mds(desc, data); | |
1430 | } | |
9369a431 | 1431 | pnfs_put_lseg(lseg); |
dce81290 TM |
1432 | } |
1433 | ||
6c75dc0d FI |
1434 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1435 | { | |
9369a431 | 1436 | pnfs_put_lseg(hdr->lseg); |
6c75dc0d FI |
1437 | nfs_writehdr_free(hdr); |
1438 | } | |
89d77c8f | 1439 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1440 | |
dce81290 TM |
1441 | int |
1442 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1443 | { | |
6c75dc0d FI |
1444 | struct nfs_write_header *whdr; |
1445 | struct nfs_pgio_header *hdr; | |
dce81290 TM |
1446 | int ret; |
1447 | ||
6c75dc0d FI |
1448 | whdr = nfs_writehdr_alloc(); |
1449 | if (!whdr) { | |
9b5415b5 | 1450 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
9369a431 | 1451 | pnfs_put_lseg(desc->pg_lseg); |
dce81290 | 1452 | desc->pg_lseg = NULL; |
6c75dc0d | 1453 | return -ENOMEM; |
dce81290 | 1454 | } |
6c75dc0d FI |
1455 | hdr = &whdr->header; |
1456 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); | |
9369a431 | 1457 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
6c75dc0d FI |
1458 | atomic_inc(&hdr->refcnt); |
1459 | ret = nfs_generic_flush(desc, hdr); | |
1460 | if (ret != 0) { | |
9369a431 | 1461 | pnfs_put_lseg(desc->pg_lseg); |
6c75dc0d | 1462 | desc->pg_lseg = NULL; |
6c75dc0d FI |
1463 | } else |
1464 | pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags); | |
1465 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1466 | hdr->completion_ops->completion(hdr); |
6c75dc0d | 1467 | return ret; |
dce81290 TM |
1468 | } |
1469 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1470 | ||
e7dd79af | 1471 | int pnfs_read_done_resend_to_mds(struct inode *inode, |
061ae2ed FI |
1472 | struct list_head *head, |
1473 | const struct nfs_pgio_completion_ops *compl_ops) | |
62e4a769 TM |
1474 | { |
1475 | struct nfs_pageio_descriptor pgio; | |
1acbbb4e | 1476 | LIST_HEAD(failed); |
62e4a769 | 1477 | |
1acbbb4e | 1478 | /* Resend all requests through the MDS */ |
1abb5088 | 1479 | nfs_pageio_init_read(&pgio, inode, compl_ops); |
1acbbb4e FI |
1480 | while (!list_empty(head)) { |
1481 | struct nfs_page *req = nfs_list_entry(head->next); | |
62e4a769 TM |
1482 | |
1483 | nfs_list_remove_request(req); | |
1acbbb4e FI |
1484 | if (!nfs_pageio_add_request(&pgio, req)) |
1485 | nfs_list_add_request(req, &failed); | |
62e4a769 TM |
1486 | } |
1487 | nfs_pageio_complete(&pgio); | |
1acbbb4e FI |
1488 | |
1489 | if (!list_empty(&failed)) { | |
1490 | list_move(&failed, head); | |
1491 | return -EIO; | |
1492 | } | |
1493 | return 0; | |
1494 | } | |
e7dd79af | 1495 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e FI |
1496 | |
1497 | static void pnfs_ld_handle_read_error(struct nfs_read_data *data) | |
1498 | { | |
cd841605 FI |
1499 | struct nfs_pgio_header *hdr = data->header; |
1500 | ||
1501 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); | |
1502 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1503 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 FI |
1504 | clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags); |
1505 | pnfs_return_layout(hdr->inode); | |
1acbbb4e | 1506 | } |
4db6e0b7 FI |
1507 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
1508 | data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode, | |
061ae2ed FI |
1509 | &hdr->pages, |
1510 | hdr->completion_ops); | |
62e4a769 TM |
1511 | } |
1512 | ||
d20581aa BH |
1513 | /* |
1514 | * Called by non rpc-based layout drivers | |
1515 | */ | |
9b7eecdc | 1516 | void pnfs_ld_read_done(struct nfs_read_data *data) |
d20581aa | 1517 | { |
cd841605 FI |
1518 | struct nfs_pgio_header *hdr = data->header; |
1519 | ||
1520 | if (likely(!hdr->pnfs_error)) { | |
d20581aa | 1521 | __nfs4_read_done_cb(data); |
cd841605 | 1522 | hdr->mds_ops->rpc_call_done(&data->task, data); |
62e4a769 TM |
1523 | } else |
1524 | pnfs_ld_handle_read_error(data); | |
cd841605 | 1525 | hdr->mds_ops->rpc_release(data); |
d20581aa BH |
1526 | } |
1527 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1528 | ||
493292dd TM |
1529 | static void |
1530 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
1531 | struct nfs_read_data *data) | |
1532 | { | |
cd841605 FI |
1533 | struct nfs_pgio_header *hdr = data->header; |
1534 | ||
4db6e0b7 FI |
1535 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1536 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1537 | nfs_pageio_reset_read_mds(desc); | |
1538 | desc->pg_recoalesce = 1; | |
1539 | } | |
493292dd TM |
1540 | nfs_readdata_release(data); |
1541 | } | |
1542 | ||
64419a9b AA |
1543 | /* |
1544 | * Call the appropriate parallel I/O subsystem read function. | |
1545 | */ | |
493292dd | 1546 | static enum pnfs_try_status |
64419a9b | 1547 | pnfs_try_to_read_data(struct nfs_read_data *rdata, |
493292dd TM |
1548 | const struct rpc_call_ops *call_ops, |
1549 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1550 | { |
cd841605 FI |
1551 | struct nfs_pgio_header *hdr = rdata->header; |
1552 | struct inode *inode = hdr->inode; | |
64419a9b AA |
1553 | struct nfs_server *nfss = NFS_SERVER(inode); |
1554 | enum pnfs_try_status trypnfs; | |
1555 | ||
cd841605 | 1556 | hdr->mds_ops = call_ops; |
64419a9b AA |
1557 | |
1558 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
1559 | __func__, inode->i_ino, rdata->args.count, rdata->args.offset); | |
1560 | ||
1561 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata); | |
4db6e0b7 | 1562 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1563 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1564 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1565 | return trypnfs; | |
1566 | } | |
863a3c6c | 1567 | |
493292dd TM |
1568 | static void |
1569 | pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head) | |
1570 | { | |
1571 | struct nfs_read_data *data; | |
1572 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; | |
1573 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
1574 | ||
1575 | desc->pg_lseg = NULL; | |
1576 | while (!list_empty(head)) { | |
1577 | enum pnfs_try_status trypnfs; | |
1578 | ||
4db6e0b7 | 1579 | data = list_first_entry(head, struct nfs_read_data, list); |
493292dd TM |
1580 | list_del_init(&data->list); |
1581 | ||
1582 | trypnfs = pnfs_try_to_read_data(data, call_ops, lseg); | |
1583 | if (trypnfs == PNFS_NOT_ATTEMPTED) | |
1584 | pnfs_read_through_mds(desc, data); | |
1585 | } | |
9369a431 | 1586 | pnfs_put_lseg(lseg); |
493292dd TM |
1587 | } |
1588 | ||
4db6e0b7 FI |
1589 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1590 | { | |
9369a431 | 1591 | pnfs_put_lseg(hdr->lseg); |
4db6e0b7 FI |
1592 | nfs_readhdr_free(hdr); |
1593 | } | |
89d77c8f | 1594 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1595 | |
493292dd TM |
1596 | int |
1597 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1598 | { | |
4db6e0b7 FI |
1599 | struct nfs_read_header *rhdr; |
1600 | struct nfs_pgio_header *hdr; | |
493292dd TM |
1601 | int ret; |
1602 | ||
4db6e0b7 FI |
1603 | rhdr = nfs_readhdr_alloc(); |
1604 | if (!rhdr) { | |
061ae2ed | 1605 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
4db6e0b7 | 1606 | ret = -ENOMEM; |
9369a431 | 1607 | pnfs_put_lseg(desc->pg_lseg); |
493292dd TM |
1608 | desc->pg_lseg = NULL; |
1609 | return ret; | |
1610 | } | |
4db6e0b7 FI |
1611 | hdr = &rhdr->header; |
1612 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); | |
9369a431 | 1613 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
4db6e0b7 FI |
1614 | atomic_inc(&hdr->refcnt); |
1615 | ret = nfs_generic_pagein(desc, hdr); | |
1616 | if (ret != 0) { | |
9369a431 | 1617 | pnfs_put_lseg(desc->pg_lseg); |
4db6e0b7 | 1618 | desc->pg_lseg = NULL; |
4db6e0b7 FI |
1619 | } else |
1620 | pnfs_do_multiple_reads(desc, &hdr->rpc_list); | |
1621 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1622 | hdr->completion_ops->completion(hdr); |
4db6e0b7 | 1623 | return ret; |
493292dd TM |
1624 | } |
1625 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1626 | ||
863a3c6c | 1627 | /* |
a9bae566 | 1628 | * There can be multiple RW segments. |
863a3c6c | 1629 | */ |
a9bae566 | 1630 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1631 | { |
a9bae566 | 1632 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1633 | |
a9bae566 PT |
1634 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1635 | if (lseg->pls_range.iomode == IOMODE_RW && | |
1636 | test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
1637 | list_add(&lseg->pls_lc_list, listp); | |
1638 | } | |
863a3c6c AA |
1639 | } |
1640 | ||
1b0ae068 PT |
1641 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
1642 | { | |
b9e028fd | 1643 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
1644 | } |
1645 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
1646 | ||
863a3c6c AA |
1647 | void |
1648 | pnfs_set_layoutcommit(struct nfs_write_data *wdata) | |
1649 | { | |
cd841605 FI |
1650 | struct nfs_pgio_header *hdr = wdata->header; |
1651 | struct inode *inode = hdr->inode; | |
1652 | struct nfs_inode *nfsi = NFS_I(inode); | |
4b8ee2b8 | 1653 | loff_t end_pos = wdata->mds_offset + wdata->res.count; |
79a48a1f | 1654 | bool mark_as_dirty = false; |
863a3c6c | 1655 | |
cd841605 | 1656 | spin_lock(&inode->i_lock); |
863a3c6c | 1657 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 1658 | mark_as_dirty = true; |
863a3c6c | 1659 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 1660 | __func__, inode->i_ino); |
863a3c6c | 1661 | } |
cd841605 | 1662 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 1663 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 1664 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 1665 | } |
acff5880 PT |
1666 | if (end_pos > nfsi->layout->plh_lwb) |
1667 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 1668 | spin_unlock(&inode->i_lock); |
acff5880 | 1669 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 1670 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
1671 | |
1672 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1673 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1674 | if (mark_as_dirty) | |
cd841605 | 1675 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
1676 | } |
1677 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
1678 | ||
db29c089 AA |
1679 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
1680 | { | |
1681 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
1682 | ||
1683 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
1684 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
1685 | } | |
1686 | ||
de4b15c7 AA |
1687 | /* |
1688 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
1689 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
1690 | * data to disk to allow the server to recover the data if it crashes. | |
1691 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
1692 | * is off, and a COMMIT is sent to a data server, or | |
1693 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
1694 | */ | |
863a3c6c | 1695 | int |
ef311537 | 1696 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c AA |
1697 | { |
1698 | struct nfs4_layoutcommit_data *data; | |
1699 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c AA |
1700 | loff_t end_pos; |
1701 | int status = 0; | |
1702 | ||
1703 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); | |
1704 | ||
de4b15c7 AA |
1705 | if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1706 | return 0; | |
1707 | ||
863a3c6c AA |
1708 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ |
1709 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
de4b15c7 | 1710 | if (!data) { |
de4b15c7 AA |
1711 | status = -ENOMEM; |
1712 | goto out; | |
1713 | } | |
863a3c6c | 1714 | |
92407e75 PT |
1715 | if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1716 | goto out_free; | |
1717 | ||
1718 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { | |
1719 | if (!sync) { | |
1720 | status = -EAGAIN; | |
1721 | goto out_free; | |
1722 | } | |
1723 | status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING, | |
1724 | nfs_wait_bit_killable, TASK_KILLABLE); | |
1725 | if (status) | |
1726 | goto out_free; | |
1727 | } | |
1728 | ||
a9bae566 | 1729 | INIT_LIST_HEAD(&data->lseg_list); |
de4b15c7 | 1730 | spin_lock(&inode->i_lock); |
863a3c6c | 1731 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
92407e75 | 1732 | clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags); |
863a3c6c | 1733 | spin_unlock(&inode->i_lock); |
92407e75 PT |
1734 | wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING); |
1735 | goto out_free; | |
863a3c6c | 1736 | } |
a9bae566 PT |
1737 | |
1738 | pnfs_list_write_lseg(inode, &data->lseg_list); | |
863a3c6c | 1739 | |
acff5880 | 1740 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 1741 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 1742 | |
f597c537 | 1743 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
1744 | spin_unlock(&inode->i_lock); |
1745 | ||
1746 | data->args.inode = inode; | |
9fa40758 | 1747 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
1748 | nfs_fattr_init(&data->fattr); |
1749 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
1750 | data->res.fattr = &data->fattr; | |
1751 | data->args.lastbytewritten = end_pos - 1; | |
1752 | data->res.server = NFS_SERVER(inode); | |
1753 | ||
1754 | status = nfs4_proc_layoutcommit(data, sync); | |
1755 | out: | |
92407e75 PT |
1756 | if (status) |
1757 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
1758 | dprintk("<-- %s status %d\n", __func__, status); |
1759 | return status; | |
92407e75 PT |
1760 | out_free: |
1761 | kfree(data); | |
1762 | goto out; | |
863a3c6c | 1763 | } |
82be417a AA |
1764 | |
1765 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
1766 | { | |
1767 | struct nfs4_threshold *thp; | |
1768 | ||
1769 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
1770 | if (!thp) { | |
1771 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
1772 | return NULL; | |
1773 | } | |
1774 | return thp; | |
1775 | } |