]>
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" |
cc668ab3 | 36 | #include "nfs4trace.h" |
85e174ba RL |
37 | |
38 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
25c75333 | 39 | #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) |
85e174ba | 40 | |
02c35fca FI |
41 | /* Locking: |
42 | * | |
43 | * pnfs_spinlock: | |
44 | * protects pnfs_modules_tbl. | |
45 | */ | |
46 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
47 | ||
48 | /* | |
49 | * pnfs_modules_tbl holds all pnfs modules | |
50 | */ | |
51 | static LIST_HEAD(pnfs_modules_tbl); | |
52 | ||
53 | /* Return the registered pnfs layout driver module matching given id */ | |
54 | static struct pnfs_layoutdriver_type * | |
55 | find_pnfs_driver_locked(u32 id) | |
56 | { | |
57 | struct pnfs_layoutdriver_type *local; | |
58 | ||
59 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
60 | if (local->id == id) | |
61 | goto out; | |
62 | local = NULL; | |
63 | out: | |
64 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
65 | return local; | |
66 | } | |
67 | ||
85e174ba RL |
68 | static struct pnfs_layoutdriver_type * |
69 | find_pnfs_driver(u32 id) | |
70 | { | |
02c35fca FI |
71 | struct pnfs_layoutdriver_type *local; |
72 | ||
73 | spin_lock(&pnfs_spinlock); | |
74 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
75 | if (local != NULL && !try_module_get(local->owner)) { |
76 | dprintk("%s: Could not grab reference on module\n", __func__); | |
77 | local = NULL; | |
78 | } | |
02c35fca FI |
79 | spin_unlock(&pnfs_spinlock); |
80 | return local; | |
85e174ba RL |
81 | } |
82 | ||
83 | void | |
84 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
85 | { | |
738fd0f3 BH |
86 | if (nfss->pnfs_curr_ld) { |
87 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
88 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
89 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
90 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
91 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 92 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 93 | } |
85e174ba RL |
94 | nfss->pnfs_curr_ld = NULL; |
95 | } | |
96 | ||
97 | /* | |
98 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
99 | * Currently only one pNFS layout driver per filesystem is supported. | |
100 | * | |
101 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
102 | */ | |
103 | void | |
738fd0f3 BH |
104 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
105 | u32 id) | |
85e174ba RL |
106 | { |
107 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
108 | ||
109 | if (id == 0) | |
110 | goto out_no_driver; | |
111 | if (!(server->nfs_client->cl_exchange_flags & | |
112 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
a030889a WAA |
113 | printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n", |
114 | __func__, id, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
115 | goto out_no_driver; |
116 | } | |
117 | ld_type = find_pnfs_driver(id); | |
118 | if (!ld_type) { | |
119 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
120 | ld_type = find_pnfs_driver(id); | |
121 | if (!ld_type) { | |
122 | dprintk("%s: No pNFS module found for %u.\n", | |
123 | __func__, id); | |
124 | goto out_no_driver; | |
125 | } | |
126 | } | |
127 | server->pnfs_curr_ld = ld_type; | |
738fd0f3 BH |
128 | if (ld_type->set_layoutdriver |
129 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
130 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
131 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
132 | module_put(ld_type->owner); |
133 | goto out_no_driver; | |
134 | } | |
2a4c8994 TM |
135 | /* Bump the MDS count */ |
136 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 137 | |
85e174ba RL |
138 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
139 | return; | |
140 | ||
141 | out_no_driver: | |
142 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
143 | server->pnfs_curr_ld = NULL; | |
144 | } | |
02c35fca FI |
145 | |
146 | int | |
147 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
148 | { | |
149 | int status = -EINVAL; | |
150 | struct pnfs_layoutdriver_type *tmp; | |
151 | ||
152 | if (ld_type->id == 0) { | |
a030889a | 153 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
154 | return status; |
155 | } | |
b1f69b75 | 156 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 157 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
158 | "alloc_lseg and free_lseg.\n", __func__); |
159 | return status; | |
160 | } | |
02c35fca FI |
161 | |
162 | spin_lock(&pnfs_spinlock); | |
163 | tmp = find_pnfs_driver_locked(ld_type->id); | |
164 | if (!tmp) { | |
165 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
166 | status = 0; | |
167 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
168 | ld_type->name); | |
169 | } else { | |
a030889a | 170 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
171 | __func__, ld_type->id); |
172 | } | |
173 | spin_unlock(&pnfs_spinlock); | |
174 | ||
175 | return status; | |
176 | } | |
177 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
178 | ||
179 | void | |
180 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
181 | { | |
182 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
183 | spin_lock(&pnfs_spinlock); | |
184 | list_del(&ld_type->pnfs_tblid); | |
185 | spin_unlock(&pnfs_spinlock); | |
186 | } | |
187 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 188 | |
b1f69b75 AA |
189 | /* |
190 | * pNFS client layout cache | |
191 | */ | |
192 | ||
cc6e5340 | 193 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 194 | void |
70c3bd2b | 195 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 196 | { |
cc6e5340 | 197 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
198 | } |
199 | ||
636fb9c8 BH |
200 | static struct pnfs_layout_hdr * |
201 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
202 | { | |
203 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
57934278 | 204 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
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); |
57934278 | 221 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
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; |
39e88fcf | 258 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
3e621214 TM |
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); |
974cec8c AA |
323 | } |
324 | ||
d684d2ae | 325 | static void |
57036a37 TM |
326 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
327 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 328 | { |
57036a37 | 329 | struct inode *inode = lo->plh_inode; |
d684d2ae | 330 | |
d20581aa | 331 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 332 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
333 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
334 | atomic_dec(&lo->plh_refcount); | |
173f77e9 TM |
335 | if (list_empty(&lo->plh_segs)) |
336 | clear_bit(NFS_LAYOUT_BULK_RECALL, &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)) { |
8f0d27dc | 355 | pnfs_get_layout_hdr(lo); |
57036a37 | 356 | pnfs_layout_remove_lseg(lo, lseg); |
d684d2ae | 357 | spin_unlock(&inode->i_lock); |
905ca191 | 358 | pnfs_free_lseg(lseg); |
8f0d27dc | 359 | pnfs_put_layout_hdr(lo); |
4541d16c | 360 | } |
4541d16c | 361 | } |
9369a431 | 362 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 363 | |
e6cf82d1 WAA |
364 | static void pnfs_put_lseg_async_work(struct work_struct *work) |
365 | { | |
366 | struct pnfs_layout_segment *lseg; | |
367 | ||
368 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
369 | ||
370 | pnfs_put_lseg(lseg); | |
371 | } | |
372 | ||
373 | void | |
374 | pnfs_put_lseg_async(struct pnfs_layout_segment *lseg) | |
375 | { | |
376 | INIT_WORK(&lseg->pls_work, pnfs_put_lseg_async_work); | |
377 | schedule_work(&lseg->pls_work); | |
378 | } | |
379 | EXPORT_SYMBOL_GPL(pnfs_put_lseg_async); | |
380 | ||
3cb2df17 | 381 | static u64 |
fb3296eb BH |
382 | end_offset(u64 start, u64 len) |
383 | { | |
384 | u64 end; | |
385 | ||
386 | end = start + len; | |
387 | return end >= start ? end : NFS4_MAX_UINT64; | |
388 | } | |
389 | ||
fb3296eb BH |
390 | /* |
391 | * is l2 fully contained in l1? | |
392 | * start1 end1 | |
393 | * [----------------------------------) | |
394 | * start2 end2 | |
395 | * [----------------) | |
396 | */ | |
3cb2df17 | 397 | static bool |
7dc0ac70 | 398 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 399 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
400 | { |
401 | u64 start1 = l1->offset; | |
402 | u64 end1 = end_offset(start1, l1->length); | |
403 | u64 start2 = l2->offset; | |
404 | u64 end2 = end_offset(start2, l2->length); | |
405 | ||
406 | return (start1 <= start2) && (end1 >= end2); | |
407 | } | |
408 | ||
409 | /* | |
410 | * is l1 and l2 intersecting? | |
411 | * start1 end1 | |
412 | * [----------------------------------) | |
413 | * start2 end2 | |
414 | * [----------------) | |
415 | */ | |
3cb2df17 | 416 | static bool |
7dc0ac70 | 417 | pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1, |
3cb2df17 | 418 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
419 | { |
420 | u64 start1 = l1->offset; | |
421 | u64 end1 = end_offset(start1, l1->length); | |
422 | u64 start2 = l2->offset; | |
423 | u64 end2 = end_offset(start2, l2->length); | |
424 | ||
425 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
426 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
427 | } | |
428 | ||
4541d16c | 429 | static bool |
3cb2df17 TM |
430 | should_free_lseg(const struct pnfs_layout_range *lseg_range, |
431 | const struct pnfs_layout_range *recall_range) | |
4541d16c | 432 | { |
778b5502 BH |
433 | return (recall_range->iomode == IOMODE_ANY || |
434 | lseg_range->iomode == recall_range->iomode) && | |
7dc0ac70 | 435 | pnfs_lseg_range_intersecting(lseg_range, recall_range); |
974cec8c AA |
436 | } |
437 | ||
24956804 TM |
438 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
439 | struct list_head *tmp_list) | |
440 | { | |
441 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
442 | return false; | |
443 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
444 | list_add(&lseg->pls_list, tmp_list); | |
445 | return true; | |
446 | } | |
447 | ||
4541d16c FI |
448 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
449 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
450 | struct list_head *tmp_list) | |
451 | { | |
452 | int rv = 0; | |
453 | ||
454 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
455 | /* Remove the reference keeping the lseg in the | |
456 | * list. It will now be removed when all | |
457 | * outstanding io is finished. | |
458 | */ | |
d684d2ae FI |
459 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
460 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 461 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 462 | rv = 1; |
4541d16c FI |
463 | } |
464 | return rv; | |
465 | } | |
466 | ||
467 | /* Returns count of number of matching invalid lsegs remaining in list | |
468 | * after call. | |
469 | */ | |
43f1b3da | 470 | int |
49a85061 | 471 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 472 | struct list_head *tmp_list, |
778b5502 | 473 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
474 | { |
475 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 476 | int invalid = 0, removed = 0; |
974cec8c AA |
477 | |
478 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
479 | ||
8006bfba | 480 | if (list_empty(&lo->plh_segs)) |
38511722 | 481 | return 0; |
4541d16c | 482 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
483 | if (!recall_range || |
484 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
485 | dprintk("%s: freeing lseg %p iomode %d " |
486 | "offset %llu length %llu\n", __func__, | |
487 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
488 | lseg->pls_range.length); | |
489 | invalid++; | |
490 | removed += mark_lseg_invalid(lseg, tmp_list); | |
491 | } | |
492 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
493 | return invalid - removed; | |
974cec8c AA |
494 | } |
495 | ||
f49f9baa | 496 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 497 | void |
4541d16c | 498 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 499 | { |
4541d16c | 500 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
501 | |
502 | if (list_empty(free_me)) | |
503 | return; | |
504 | ||
4541d16c | 505 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 506 | list_del(&lseg->pls_list); |
905ca191 | 507 | pnfs_free_lseg(lseg); |
974cec8c AA |
508 | } |
509 | } | |
510 | ||
e5e94017 BH |
511 | void |
512 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
513 | { | |
514 | struct pnfs_layout_hdr *lo; | |
974cec8c | 515 | LIST_HEAD(tmp_list); |
e5e94017 BH |
516 | |
517 | spin_lock(&nfsi->vfs_inode.i_lock); | |
518 | lo = nfsi->layout; | |
519 | if (lo) { | |
38511722 | 520 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 521 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
3e621214 TM |
522 | pnfs_get_layout_hdr(lo); |
523 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); | |
524 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
525 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
526 | pnfs_free_lseg_list(&tmp_list); | |
527 | pnfs_put_layout_hdr(lo); | |
528 | } else | |
529 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 530 | } |
041245c8 | 531 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 532 | |
fd9a8d71 TM |
533 | static bool |
534 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
535 | struct list_head *layout_list) | |
974cec8c AA |
536 | { |
537 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 538 | bool ret = false; |
974cec8c | 539 | |
fd9a8d71 TM |
540 | spin_lock(&inode->i_lock); |
541 | lo = NFS_I(inode)->layout; | |
542 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
543 | pnfs_get_layout_hdr(lo); | |
544 | list_add(&lo->plh_bulk_destroy, layout_list); | |
545 | ret = true; | |
546 | } | |
547 | spin_unlock(&inode->i_lock); | |
548 | return ret; | |
549 | } | |
550 | ||
551 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
552 | static int | |
553 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
554 | struct nfs_server *server, | |
555 | struct list_head *layout_list) | |
556 | { | |
557 | struct pnfs_layout_hdr *lo, *next; | |
558 | struct inode *inode; | |
559 | ||
560 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
561 | inode = igrab(lo->plh_inode); | |
562 | if (inode == NULL) | |
563 | continue; | |
564 | list_del_init(&lo->plh_layouts); | |
565 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
566 | continue; | |
567 | rcu_read_unlock(); | |
568 | spin_unlock(&clp->cl_lock); | |
569 | iput(inode); | |
570 | spin_lock(&clp->cl_lock); | |
571 | rcu_read_lock(); | |
572 | return -EAGAIN; | |
573 | } | |
574 | return 0; | |
575 | } | |
576 | ||
577 | static int | |
578 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
579 | bool is_bulk_recall) | |
580 | { | |
581 | struct pnfs_layout_hdr *lo; | |
582 | struct inode *inode; | |
583 | struct pnfs_layout_range range = { | |
584 | .iomode = IOMODE_ANY, | |
585 | .offset = 0, | |
586 | .length = NFS4_MAX_UINT64, | |
587 | }; | |
588 | LIST_HEAD(lseg_list); | |
589 | int ret = 0; | |
590 | ||
591 | while (!list_empty(layout_list)) { | |
592 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
593 | plh_bulk_destroy); | |
594 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
595 | lo->plh_inode->i_ino); | |
596 | inode = lo->plh_inode; | |
597 | spin_lock(&inode->i_lock); | |
598 | list_del_init(&lo->plh_bulk_destroy); | |
599 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ | |
600 | if (is_bulk_recall) | |
601 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
602 | if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range)) | |
603 | ret = -EAGAIN; | |
604 | spin_unlock(&inode->i_lock); | |
605 | pnfs_free_lseg_list(&lseg_list); | |
606 | pnfs_put_layout_hdr(lo); | |
607 | iput(inode); | |
608 | } | |
609 | return ret; | |
610 | } | |
611 | ||
612 | int | |
613 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
614 | struct nfs_fsid *fsid, | |
615 | bool is_recall) | |
616 | { | |
617 | struct nfs_server *server; | |
618 | LIST_HEAD(layout_list); | |
c47abcf8 | 619 | |
974cec8c | 620 | spin_lock(&clp->cl_lock); |
6382a441 | 621 | rcu_read_lock(); |
fd9a8d71 | 622 | restart: |
6382a441 | 623 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
624 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
625 | continue; | |
626 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
627 | server, | |
628 | &layout_list) != 0) | |
629 | goto restart; | |
6382a441 WAA |
630 | } |
631 | rcu_read_unlock(); | |
974cec8c AA |
632 | spin_unlock(&clp->cl_lock); |
633 | ||
fd9a8d71 TM |
634 | if (list_empty(&layout_list)) |
635 | return 0; | |
636 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
637 | } | |
638 | ||
639 | int | |
640 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
641 | bool is_recall) | |
642 | { | |
643 | struct nfs_server *server; | |
644 | LIST_HEAD(layout_list); | |
645 | ||
646 | spin_lock(&clp->cl_lock); | |
647 | rcu_read_lock(); | |
648 | restart: | |
649 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
650 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
651 | server, | |
652 | &layout_list) != 0) | |
653 | goto restart; | |
974cec8c | 654 | } |
fd9a8d71 TM |
655 | rcu_read_unlock(); |
656 | spin_unlock(&clp->cl_lock); | |
657 | ||
658 | if (list_empty(&layout_list)) | |
659 | return 0; | |
660 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
661 | } | |
662 | ||
663 | /* | |
664 | * Called by the state manger to remove all layouts established under an | |
665 | * expired lease. | |
666 | */ | |
667 | void | |
668 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
669 | { | |
670 | nfs4_deviceid_mark_client_invalid(clp); | |
671 | nfs4_deviceid_purge_client(clp); | |
672 | ||
673 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
674 | } |
675 | ||
5a65503f TM |
676 | /* |
677 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
678 | * taking into account wraparound issues. | |
679 | */ | |
680 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
681 | { | |
2c64c57d | 682 | return (s32)(s1 - s2) > 0; |
5a65503f TM |
683 | } |
684 | ||
fd6002e9 | 685 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
686 | void |
687 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
688 | bool update_barrier) | |
b1f69b75 | 689 | { |
22aaf714 TM |
690 | u32 oldseq, newseq, new_barrier; |
691 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 692 | |
2d2f24ad TM |
693 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
694 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 695 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 696 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 697 | if (update_barrier) { |
22aaf714 | 698 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
699 | } else { |
700 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 701 | * "close" to the current seqids. |
43f1b3da | 702 | */ |
22aaf714 | 703 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 704 | } |
22aaf714 TM |
705 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
706 | lo->plh_barrier = new_barrier; | |
43f1b3da | 707 | } |
b1f69b75 AA |
708 | } |
709 | ||
cf7d63f1 | 710 | static bool |
19c54aba TM |
711 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
712 | const nfs4_stateid *stateid) | |
43f1b3da | 713 | { |
19c54aba | 714 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 715 | |
19c54aba TM |
716 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
717 | } | |
718 | ||
719 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
720 | static bool | |
721 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget) | |
722 | { | |
f7e8917a FI |
723 | return lo->plh_block_lgets || |
724 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || | |
43f1b3da | 725 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
726 | (atomic_read(&lo->plh_outstanding) > lget)); |
727 | } | |
728 | ||
fd6002e9 FI |
729 | int |
730 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
731 | struct nfs4_state *open_state) | |
b1f69b75 | 732 | { |
fd6002e9 | 733 | int status = 0; |
974cec8c | 734 | |
b1f69b75 | 735 | dprintk("--> %s\n", __func__); |
fd6002e9 | 736 | spin_lock(&lo->plh_inode->i_lock); |
19c54aba | 737 | if (pnfs_layoutgets_blocked(lo, 1)) { |
cf7d63f1 | 738 | status = -EAGAIN; |
5d422301 TM |
739 | } else if (!nfs4_valid_open_stateid(open_state)) { |
740 | status = -EBADF; | |
47abadef CH |
741 | } else if (list_empty(&lo->plh_segs) || |
742 | test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { | |
fd6002e9 FI |
743 | int seq; |
744 | ||
745 | do { | |
746 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 747 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
748 | } while (read_seqretry(&open_state->seqlock, seq)); |
749 | } else | |
f597c537 | 750 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 751 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 752 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 753 | return status; |
b1f69b75 AA |
754 | } |
755 | ||
756 | /* | |
757 | * Get layout from server. | |
758 | * for now, assume that whole file layouts are requested. | |
759 | * arg->offset: 0 | |
760 | * arg->length: all ones | |
761 | */ | |
e5e94017 BH |
762 | static struct pnfs_layout_segment * |
763 | send_layoutget(struct pnfs_layout_hdr *lo, | |
764 | struct nfs_open_context *ctx, | |
fb3296eb | 765 | struct pnfs_layout_range *range, |
a75b9df9 | 766 | gfp_t gfp_flags) |
e5e94017 | 767 | { |
b7edfaa1 | 768 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
769 | struct nfs_server *server = NFS_SERVER(ino); |
770 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 771 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
772 | |
773 | dprintk("--> %s\n", __func__); | |
e5e94017 | 774 | |
a75b9df9 | 775 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 776 | if (lgp == NULL) |
b1f69b75 | 777 | return NULL; |
35124a09 | 778 | |
fb3296eb BH |
779 | lgp->args.minlength = PAGE_CACHE_SIZE; |
780 | if (lgp->args.minlength > range->length) | |
781 | lgp->args.minlength = range->length; | |
b1f69b75 | 782 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 783 | lgp->args.range = *range; |
b1f69b75 AA |
784 | lgp->args.type = server->pnfs_curr_ld->id; |
785 | lgp->args.inode = ino; | |
786 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 787 | lgp->gfp_flags = gfp_flags; |
6ab59344 | 788 | lgp->cred = lo->plh_lc_cred; |
b1f69b75 AA |
789 | |
790 | /* Synchronously retrieve layout information from server and | |
791 | * store in lseg. | |
792 | */ | |
a0b0a6e3 TM |
793 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
794 | if (IS_ERR(lseg)) { | |
795 | switch (PTR_ERR(lseg)) { | |
796 | case -ENOMEM: | |
797 | case -ERESTARTSYS: | |
798 | break; | |
799 | default: | |
800 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 801 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
802 | } |
803 | return NULL; | |
974cec8c | 804 | } |
35124a09 | 805 | |
974cec8c AA |
806 | return lseg; |
807 | } | |
808 | ||
24956804 TM |
809 | static void pnfs_clear_layoutcommit(struct inode *inode, |
810 | struct list_head *head) | |
811 | { | |
812 | struct nfs_inode *nfsi = NFS_I(inode); | |
813 | struct pnfs_layout_segment *lseg, *tmp; | |
814 | ||
815 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
816 | return; | |
817 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
818 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
819 | continue; | |
820 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
821 | } | |
822 | } | |
823 | ||
293b3b06 AA |
824 | /* |
825 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
826 | * when the layout segment list is empty. | |
827 | * | |
828 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
829 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
830 | * deviceid is marked invalid. | |
831 | */ | |
cbe82603 BH |
832 | int |
833 | _pnfs_return_layout(struct inode *ino) | |
834 | { | |
835 | struct pnfs_layout_hdr *lo = NULL; | |
836 | struct nfs_inode *nfsi = NFS_I(ino); | |
837 | LIST_HEAD(tmp_list); | |
838 | struct nfs4_layoutreturn *lrp; | |
839 | nfs4_stateid stateid; | |
293b3b06 | 840 | int status = 0, empty; |
cbe82603 | 841 | |
366d5052 | 842 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
843 | |
844 | spin_lock(&ino->i_lock); | |
845 | lo = nfsi->layout; | |
e5929f3c | 846 | if (!lo) { |
cbe82603 | 847 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
848 | dprintk("NFS: %s no layout to return\n", __func__); |
849 | goto out; | |
cbe82603 BH |
850 | } |
851 | stateid = nfsi->layout->plh_stateid; | |
852 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 853 | pnfs_get_layout_hdr(lo); |
293b3b06 | 854 | empty = list_empty(&lo->plh_segs); |
24956804 | 855 | pnfs_clear_layoutcommit(ino, &tmp_list); |
49a85061 | 856 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
293b3b06 AA |
857 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
858 | if (empty) { | |
859 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 860 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
861 | dprintk("NFS: %s no layout segments to return\n", __func__); |
862 | goto out; | |
863 | } | |
47abadef CH |
864 | |
865 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
ea0ded74 | 866 | lo->plh_block_lgets++; |
cbe82603 BH |
867 | spin_unlock(&ino->i_lock); |
868 | pnfs_free_lseg_list(&tmp_list); | |
869 | ||
cbe82603 BH |
870 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); |
871 | if (unlikely(lrp == NULL)) { | |
872 | status = -ENOMEM; | |
65857d57 TM |
873 | spin_lock(&ino->i_lock); |
874 | lo->plh_block_lgets--; | |
875 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 876 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
877 | goto out; |
878 | } | |
879 | ||
880 | lrp->args.stateid = stateid; | |
881 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
882 | lrp->args.inode = ino; | |
a56aaa02 | 883 | lrp->args.layout = lo; |
cbe82603 | 884 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
9556000d | 885 | lrp->cred = lo->plh_lc_cred; |
cbe82603 BH |
886 | |
887 | status = nfs4_proc_layoutreturn(lrp); | |
888 | out: | |
889 | dprintk("<-- %s status: %d\n", __func__, status); | |
890 | return status; | |
891 | } | |
0a57cdac | 892 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 893 | |
24028672 TM |
894 | int |
895 | pnfs_commit_and_return_layout(struct inode *inode) | |
896 | { | |
897 | struct pnfs_layout_hdr *lo; | |
898 | int ret; | |
899 | ||
900 | spin_lock(&inode->i_lock); | |
901 | lo = NFS_I(inode)->layout; | |
902 | if (lo == NULL) { | |
903 | spin_unlock(&inode->i_lock); | |
904 | return 0; | |
905 | } | |
906 | pnfs_get_layout_hdr(lo); | |
907 | /* Block new layoutgets and read/write to ds */ | |
908 | lo->plh_block_lgets++; | |
909 | spin_unlock(&inode->i_lock); | |
910 | filemap_fdatawait(inode->i_mapping); | |
911 | ret = pnfs_layoutcommit_inode(inode, true); | |
912 | if (ret == 0) | |
913 | ret = _pnfs_return_layout(inode); | |
914 | spin_lock(&inode->i_lock); | |
915 | lo->plh_block_lgets--; | |
916 | spin_unlock(&inode->i_lock); | |
917 | pnfs_put_layout_hdr(lo); | |
918 | return ret; | |
919 | } | |
920 | ||
f7e8917a FI |
921 | bool pnfs_roc(struct inode *ino) |
922 | { | |
923 | struct pnfs_layout_hdr *lo; | |
924 | struct pnfs_layout_segment *lseg, *tmp; | |
925 | LIST_HEAD(tmp_list); | |
926 | bool found = false; | |
927 | ||
928 | spin_lock(&ino->i_lock); | |
929 | lo = NFS_I(ino)->layout; | |
930 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
931 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
932 | goto out_nolayout; | |
933 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
934 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
935 | mark_lseg_invalid(lseg, &tmp_list); | |
936 | found = true; | |
937 | } | |
938 | if (!found) | |
939 | goto out_nolayout; | |
940 | lo->plh_block_lgets++; | |
70c3bd2b | 941 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
942 | spin_unlock(&ino->i_lock); |
943 | pnfs_free_lseg_list(&tmp_list); | |
944 | return true; | |
945 | ||
946 | out_nolayout: | |
947 | spin_unlock(&ino->i_lock); | |
948 | return false; | |
949 | } | |
950 | ||
951 | void pnfs_roc_release(struct inode *ino) | |
952 | { | |
953 | struct pnfs_layout_hdr *lo; | |
954 | ||
955 | spin_lock(&ino->i_lock); | |
956 | lo = NFS_I(ino)->layout; | |
957 | lo->plh_block_lgets--; | |
6622c3ea TM |
958 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
959 | pnfs_detach_layout_hdr(lo); | |
960 | spin_unlock(&ino->i_lock); | |
961 | pnfs_free_layout_hdr(lo); | |
962 | } else | |
963 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
964 | } |
965 | ||
966 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
967 | { | |
968 | struct pnfs_layout_hdr *lo; | |
969 | ||
970 | spin_lock(&ino->i_lock); | |
971 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 972 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
973 | lo->plh_barrier = barrier; |
974 | spin_unlock(&ino->i_lock); | |
975 | } | |
976 | ||
7fdab069 | 977 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
978 | { |
979 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 980 | struct pnfs_layout_hdr *lo; |
f7e8917a | 981 | struct pnfs_layout_segment *lseg; |
7fdab069 | 982 | u32 current_seqid; |
f7e8917a FI |
983 | bool found = false; |
984 | ||
985 | spin_lock(&ino->i_lock); | |
986 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
987 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 988 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 989 | found = true; |
7fdab069 | 990 | goto out; |
f7e8917a | 991 | } |
7fdab069 TM |
992 | lo = nfsi->layout; |
993 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 994 | |
7fdab069 TM |
995 | /* Since close does not return a layout stateid for use as |
996 | * a barrier, we choose the worst-case barrier. | |
997 | */ | |
998 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
999 | out: | |
f7e8917a FI |
1000 | spin_unlock(&ino->i_lock); |
1001 | return found; | |
1002 | } | |
1003 | ||
b1f69b75 AA |
1004 | /* |
1005 | * Compare two layout segments for sorting into layout cache. | |
1006 | * We want to preferentially return RW over RO layouts, so ensure those | |
1007 | * are seen first. | |
1008 | */ | |
1009 | static s64 | |
7dc0ac70 | 1010 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1011 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1012 | { |
fb3296eb BH |
1013 | s64 d; |
1014 | ||
1015 | /* high offset > low offset */ | |
1016 | d = l1->offset - l2->offset; | |
1017 | if (d) | |
1018 | return d; | |
1019 | ||
1020 | /* short length > long length */ | |
1021 | d = l2->length - l1->length; | |
1022 | if (d) | |
1023 | return d; | |
1024 | ||
b1f69b75 | 1025 | /* read > read/write */ |
fb3296eb | 1026 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1027 | } |
1028 | ||
974cec8c | 1029 | static void |
57036a37 | 1030 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
1031 | struct pnfs_layout_segment *lseg) |
1032 | { | |
b1f69b75 | 1033 | struct pnfs_layout_segment *lp; |
b1f69b75 | 1034 | |
974cec8c AA |
1035 | dprintk("%s:Begin\n", __func__); |
1036 | ||
b7edfaa1 | 1037 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
7dc0ac70 | 1038 | if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 1039 | continue; |
566052c5 | 1040 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1041 | dprintk("%s: inserted lseg %p " |
1042 | "iomode %d offset %llu length %llu before " | |
1043 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1044 | __func__, lseg, lseg->pls_range.iomode, |
1045 | lseg->pls_range.offset, lseg->pls_range.length, | |
1046 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1047 | lp->pls_range.length); | |
fb3296eb | 1048 | goto out; |
974cec8c | 1049 | } |
fb3296eb BH |
1050 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1051 | dprintk("%s: inserted lseg %p " | |
1052 | "iomode %d offset %llu length %llu at tail\n", | |
1053 | __func__, lseg, lseg->pls_range.iomode, | |
1054 | lseg->pls_range.offset, lseg->pls_range.length); | |
1055 | out: | |
70c3bd2b | 1056 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1057 | |
1058 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
1059 | } |
1060 | ||
1061 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1062 | alloc_init_layout_hdr(struct inode *ino, |
1063 | struct nfs_open_context *ctx, | |
1064 | gfp_t gfp_flags) | |
e5e94017 BH |
1065 | { |
1066 | struct pnfs_layout_hdr *lo; | |
1067 | ||
636fb9c8 | 1068 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1069 | if (!lo) |
1070 | return NULL; | |
cc6e5340 | 1071 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1072 | INIT_LIST_HEAD(&lo->plh_layouts); |
1073 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1074 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1075 | lo->plh_inode = ino; |
5cc2216d | 1076 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
e5e94017 BH |
1077 | return lo; |
1078 | } | |
1079 | ||
1080 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1081 | pnfs_find_alloc_layout(struct inode *ino, |
1082 | struct nfs_open_context *ctx, | |
1083 | gfp_t gfp_flags) | |
e5e94017 BH |
1084 | { |
1085 | struct nfs_inode *nfsi = NFS_I(ino); | |
1086 | struct pnfs_layout_hdr *new = NULL; | |
1087 | ||
1088 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1089 | ||
251ec410 TM |
1090 | if (nfsi->layout != NULL) |
1091 | goto out_existing; | |
e5e94017 | 1092 | spin_unlock(&ino->i_lock); |
9fa40758 | 1093 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1094 | spin_lock(&ino->i_lock); |
1095 | ||
251ec410 | 1096 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1097 | nfsi->layout = new; |
251ec410 | 1098 | return new; |
7175fe90 YN |
1099 | } else if (new != NULL) |
1100 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1101 | out_existing: |
1102 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1103 | return nfsi->layout; |
1104 | } | |
1105 | ||
b1f69b75 AA |
1106 | /* |
1107 | * iomode matching rules: | |
1108 | * iomode lseg match | |
1109 | * ----- ----- ----- | |
1110 | * ANY READ true | |
1111 | * ANY RW true | |
1112 | * RW READ false | |
1113 | * RW RW true | |
1114 | * READ READ true | |
1115 | * READ RW true | |
1116 | */ | |
3cb2df17 | 1117 | static bool |
7dc0ac70 | 1118 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
3cb2df17 | 1119 | const struct pnfs_layout_range *range) |
b1f69b75 | 1120 | { |
fb3296eb BH |
1121 | struct pnfs_layout_range range1; |
1122 | ||
1123 | if ((range->iomode == IOMODE_RW && | |
1124 | ls_range->iomode != IOMODE_RW) || | |
7dc0ac70 | 1125 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1126 | return 0; |
1127 | ||
1128 | /* range1 covers only the first byte in the range */ | |
1129 | range1 = *range; | |
1130 | range1.length = 1; | |
7dc0ac70 | 1131 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1132 | } |
1133 | ||
1134 | /* | |
1135 | * lookup range in layout | |
1136 | */ | |
e5e94017 | 1137 | static struct pnfs_layout_segment * |
fb3296eb BH |
1138 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
1139 | struct pnfs_layout_range *range) | |
e5e94017 | 1140 | { |
b1f69b75 AA |
1141 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1142 | ||
1143 | dprintk("%s:Begin\n", __func__); | |
1144 | ||
b7edfaa1 | 1145 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1146 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
7dc0ac70 | 1147 | pnfs_lseg_range_match(&lseg->pls_range, range)) { |
9369a431 | 1148 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1149 | break; |
1150 | } | |
d771e3a4 | 1151 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
1152 | break; |
1153 | } | |
1154 | ||
1155 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1156 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1157 | return ret; |
e5e94017 BH |
1158 | } |
1159 | ||
d23d61c8 AA |
1160 | /* |
1161 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1162 | * to the MDS or over pNFS | |
1163 | * | |
1164 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1165 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1166 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1167 | * WRITE request. | |
1168 | * | |
1169 | * A return of true means use MDS I/O. | |
1170 | * | |
1171 | * From rfc 5661: | |
1172 | * If a file's size is smaller than the file size threshold, data accesses | |
1173 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1174 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1175 | * server. If both file size and I/O size are provided, the client SHOULD | |
1176 | * reach or exceed both thresholds before sending its read or write | |
1177 | * requests to the data server. | |
1178 | */ | |
1179 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1180 | struct inode *ino, int iomode) | |
1181 | { | |
1182 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1183 | struct nfs_inode *nfsi = NFS_I(ino); | |
1184 | loff_t fsize = i_size_read(ino); | |
1185 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1186 | ||
1187 | if (t == NULL) | |
1188 | return ret; | |
1189 | ||
1190 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1191 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1192 | ||
1193 | switch (iomode) { | |
1194 | case IOMODE_READ: | |
1195 | if (t->bm & THRESHOLD_RD) { | |
1196 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1197 | size_set = true; | |
1198 | if (fsize < t->rd_sz) | |
1199 | size = true; | |
1200 | } | |
1201 | if (t->bm & THRESHOLD_RD_IO) { | |
1202 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1203 | nfsi->read_io); | |
1204 | io_set = true; | |
1205 | if (nfsi->read_io < t->rd_io_sz) | |
1206 | io = true; | |
1207 | } | |
1208 | break; | |
1209 | case IOMODE_RW: | |
1210 | if (t->bm & THRESHOLD_WR) { | |
1211 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1212 | size_set = true; | |
1213 | if (fsize < t->wr_sz) | |
1214 | size = true; | |
1215 | } | |
1216 | if (t->bm & THRESHOLD_WR_IO) { | |
1217 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1218 | nfsi->write_io); | |
1219 | io_set = true; | |
1220 | if (nfsi->write_io < t->wr_io_sz) | |
1221 | io = true; | |
1222 | } | |
1223 | break; | |
1224 | } | |
1225 | if (size_set && io_set) { | |
1226 | if (size && io) | |
1227 | ret = true; | |
1228 | } else if (size || io) | |
1229 | ret = true; | |
1230 | ||
1231 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1232 | return ret; | |
1233 | } | |
1234 | ||
e5e94017 BH |
1235 | /* |
1236 | * Layout segment is retreived from the server if not cached. | |
1237 | * The appropriate layout segment is referenced and returned to the caller. | |
1238 | */ | |
7c24d948 | 1239 | struct pnfs_layout_segment * |
e5e94017 BH |
1240 | pnfs_update_layout(struct inode *ino, |
1241 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1242 | loff_t pos, |
1243 | u64 count, | |
a75b9df9 TM |
1244 | enum pnfs_iomode iomode, |
1245 | gfp_t gfp_flags) | |
e5e94017 | 1246 | { |
fb3296eb BH |
1247 | struct pnfs_layout_range arg = { |
1248 | .iomode = iomode, | |
1249 | .offset = pos, | |
1250 | .length = count, | |
1251 | }; | |
707ed5fd | 1252 | unsigned pg_offset; |
6382a441 WAA |
1253 | struct nfs_server *server = NFS_SERVER(ino); |
1254 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1255 | struct pnfs_layout_hdr *lo; |
1256 | struct pnfs_layout_segment *lseg = NULL; | |
30005121 | 1257 | bool first; |
e5e94017 BH |
1258 | |
1259 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1260 | goto out; |
d23d61c8 AA |
1261 | |
1262 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1263 | goto out; |
d23d61c8 | 1264 | |
e5e94017 | 1265 | spin_lock(&ino->i_lock); |
9fa40758 | 1266 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1267 | if (lo == NULL) { |
1268 | spin_unlock(&ino->i_lock); | |
1269 | goto out; | |
1270 | } | |
e5e94017 | 1271 | |
43f1b3da | 1272 | /* Do we even need to bother with this? */ |
a59c30ac | 1273 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1274 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1275 | goto out_unlock; |
1276 | } | |
1277 | ||
1278 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1279 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1280 | goto out_unlock; |
1281 | ||
568e8c49 | 1282 | /* Check to see if the layout for the given range already exists */ |
fb3296eb | 1283 | lseg = pnfs_find_lseg(lo, &arg); |
568e8c49 AA |
1284 | if (lseg) |
1285 | goto out_unlock; | |
1286 | ||
19c54aba | 1287 | if (pnfs_layoutgets_blocked(lo, 0)) |
cf7d63f1 FI |
1288 | goto out_unlock; |
1289 | atomic_inc(&lo->plh_outstanding); | |
1290 | ||
30005121 | 1291 | first = list_empty(&lo->plh_layouts) ? true : false; |
f49f9baa | 1292 | spin_unlock(&ino->i_lock); |
30005121 | 1293 | |
f49f9baa | 1294 | if (first) { |
2130ff66 FI |
1295 | /* The lo must be on the clp list if there is any |
1296 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1297 | */ | |
1298 | spin_lock(&clp->cl_lock); | |
6382a441 | 1299 | list_add_tail(&lo->plh_layouts, &server->layouts); |
2130ff66 FI |
1300 | spin_unlock(&clp->cl_lock); |
1301 | } | |
e5e94017 | 1302 | |
707ed5fd BH |
1303 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1304 | if (pg_offset) { | |
1305 | arg.offset -= pg_offset; | |
1306 | arg.length += pg_offset; | |
1307 | } | |
7c24d948 AA |
1308 | if (arg.length != NFS4_MAX_UINT64) |
1309 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1310 | |
fb3296eb | 1311 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
cf7d63f1 | 1312 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1313 | out_put_layout_hdr: |
70c3bd2b | 1314 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1315 | out: |
f86bbcf8 TM |
1316 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1317 | "(%s, offset: %llu, length: %llu)\n", | |
1318 | __func__, ino->i_sb->s_id, | |
1319 | (unsigned long long)NFS_FILEID(ino), | |
1320 | lseg == NULL ? "not found" : "found", | |
1321 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1322 | (unsigned long long)pos, | |
1323 | (unsigned long long)count); | |
e5e94017 BH |
1324 | return lseg; |
1325 | out_unlock: | |
1326 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1327 | goto out_put_layout_hdr; |
e5e94017 | 1328 | } |
7c24d948 | 1329 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1330 | |
a0b0a6e3 | 1331 | struct pnfs_layout_segment * |
b1f69b75 AA |
1332 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1333 | { | |
1334 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1335 | struct nfs4_layoutget_res *res = &lgp->res; | |
1336 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1337 | struct inode *ino = lo->plh_inode; |
78096cca | 1338 | LIST_HEAD(free_me); |
b1f69b75 AA |
1339 | int status = 0; |
1340 | ||
1341 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1342 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1343 | if (!lseg || IS_ERR(lseg)) { |
1344 | if (!lseg) | |
1345 | status = -ENOMEM; | |
1346 | else | |
1347 | status = PTR_ERR(lseg); | |
1348 | dprintk("%s: Could not allocate layout: error %d\n", | |
1349 | __func__, status); | |
1350 | goto out; | |
1351 | } | |
1352 | ||
1013df61 CH |
1353 | init_lseg(lo, lseg); |
1354 | lseg->pls_range = res->range; | |
1355 | ||
b1f69b75 | 1356 | spin_lock(&ino->i_lock); |
a59c30ac | 1357 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1358 | dprintk("%s forget reply due to recall\n", __func__); |
1359 | goto out_forget_reply; | |
1360 | } | |
1361 | ||
362f7474 | 1362 | if (pnfs_layoutgets_blocked(lo, 1)) { |
43f1b3da FI |
1363 | dprintk("%s forget reply due to state\n", __func__); |
1364 | goto out_forget_reply; | |
1365 | } | |
038d6493 | 1366 | |
362f7474 CH |
1367 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1368 | /* existing state ID, make sure the sequence number matches. */ | |
1369 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1370 | dprintk("%s forget reply due to sequence\n", __func__); | |
1371 | goto out_forget_reply; | |
1372 | } | |
1373 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1374 | } else { | |
1375 | /* | |
1376 | * We got an entirely new state ID. Mark all segments for the | |
1377 | * inode invalid, and don't bother validating the stateid | |
1378 | * sequence number. | |
1379 | */ | |
1380 | pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL); | |
1381 | ||
1382 | nfs4_stateid_copy(&lo->plh_stateid, &res->stateid); | |
1383 | lo->plh_barrier = be32_to_cpu(res->stateid.seqid); | |
1384 | } | |
038d6493 | 1385 | |
47abadef CH |
1386 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); |
1387 | ||
9369a431 | 1388 | pnfs_get_lseg(lseg); |
57036a37 | 1389 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1390 | |
f7e8917a FI |
1391 | if (res->return_on_close) { |
1392 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1393 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1394 | } | |
1395 | ||
b1f69b75 | 1396 | spin_unlock(&ino->i_lock); |
78096cca | 1397 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1398 | return lseg; |
b1f69b75 | 1399 | out: |
a0b0a6e3 | 1400 | return ERR_PTR(status); |
43f1b3da FI |
1401 | |
1402 | out_forget_reply: | |
1403 | spin_unlock(&ino->i_lock); | |
1404 | lseg->pls_layout = lo; | |
1405 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1406 | goto out; | |
b1f69b75 AA |
1407 | } |
1408 | ||
d8007d4d TM |
1409 | void |
1410 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1411 | { | |
1fd937bd PT |
1412 | u64 rd_size = req->wb_bytes; |
1413 | ||
bc5a89b3 | 1414 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d | 1415 | |
1fd937bd PT |
1416 | if (pgio->pg_dreq == NULL) |
1417 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1418 | else | |
1419 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1420 | ||
d8007d4d TM |
1421 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1422 | req->wb_context, | |
1423 | req_offset(req), | |
1fd937bd | 1424 | rd_size, |
d8007d4d TM |
1425 | IOMODE_READ, |
1426 | GFP_KERNEL); | |
e885de1a TM |
1427 | /* If no lseg, fall back to read through mds */ |
1428 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1429 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1430 | |
d8007d4d TM |
1431 | } |
1432 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1433 | ||
1434 | void | |
6296556f PT |
1435 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1436 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1437 | { |
bc5a89b3 | 1438 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d TM |
1439 | |
1440 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1441 | req->wb_context, | |
1442 | req_offset(req), | |
6296556f | 1443 | wb_size, |
d8007d4d TM |
1444 | IOMODE_RW, |
1445 | GFP_NOFS); | |
e885de1a TM |
1446 | /* If no lseg, fall back to write through mds */ |
1447 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1448 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1449 | } |
1450 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1451 | ||
b4fdac1a WAA |
1452 | /* |
1453 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1454 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1455 | */ | |
1456 | size_t | |
dfed206b BH |
1457 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
1458 | struct nfs_page *req) | |
94ad1c80 | 1459 | { |
0f9c429e | 1460 | unsigned int size; |
c5e20cb7 | 1461 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1462 | |
1463 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1464 | if (!size) |
1465 | return 0; | |
94ad1c80 | 1466 | |
19982ba8 | 1467 | /* |
c5e20cb7 WAA |
1468 | * 'size' contains the number of bytes left in the current page (up |
1469 | * to the original size asked for in @req->wb_bytes). | |
1470 | * | |
1471 | * Calculate how many bytes are left in the layout segment | |
1472 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1473 | * |
1474 | * Please also note that 'end_offset' is actually the offset of the | |
1475 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1476 | * | |
1477 | */ | |
19b54848 | 1478 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1479 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1480 | pgio->pg_lseg->pls_range.length); | |
1481 | req_start = req_offset(req); | |
1482 | WARN_ON_ONCE(req_start > seg_end); | |
1483 | /* start of request is past the last byte of this segment */ | |
1484 | if (req_start >= seg_end) | |
19b54848 | 1485 | return 0; |
c5e20cb7 WAA |
1486 | |
1487 | /* adjust 'size' iff there are fewer bytes left in the | |
1488 | * segment than what nfs_generic_pg_test returned */ | |
1489 | seg_left = seg_end - req_start; | |
1490 | if (seg_left < size) | |
1491 | size = (unsigned int)seg_left; | |
19b54848 | 1492 | } |
0f9c429e | 1493 | |
19b54848 | 1494 | return size; |
94ad1c80 | 1495 | } |
89a58e32 | 1496 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1497 | |
53113ad3 | 1498 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
1499 | { |
1500 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
1501 | |
1502 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
1503 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
1504 | hdr->completion_ops); | |
1505 | return nfs_pageio_resend(&pgio, hdr); | |
e2fecb21 | 1506 | } |
e7dd79af | 1507 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1508 | |
d45f60c6 | 1509 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1510 | { |
cd841605 FI |
1511 | |
1512 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1513 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1514 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1515 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1516 | } |
6c75dc0d | 1517 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1518 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
1519 | } |
1520 | ||
d20581aa BH |
1521 | /* |
1522 | * Called by non rpc-based layout drivers | |
1523 | */ | |
d45f60c6 | 1524 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 1525 | { |
d45f60c6 | 1526 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); |
cd841605 | 1527 | if (!hdr->pnfs_error) { |
d45f60c6 WAA |
1528 | pnfs_set_layoutcommit(hdr); |
1529 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
1acbbb4e | 1530 | } else |
d45f60c6 WAA |
1531 | pnfs_ld_handle_write_error(hdr); |
1532 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 1533 | } |
d20581aa | 1534 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1535 | |
dce81290 TM |
1536 | static void |
1537 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1538 | struct nfs_pgio_header *hdr) |
dce81290 | 1539 | { |
6c75dc0d FI |
1540 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1541 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1542 | nfs_pageio_reset_write_mds(desc); | |
1543 | desc->pg_recoalesce = 1; | |
1544 | } | |
d45f60c6 | 1545 | nfs_pgio_data_destroy(hdr); |
dce81290 TM |
1546 | } |
1547 | ||
1548 | static enum pnfs_try_status | |
d45f60c6 | 1549 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
1550 | const struct rpc_call_ops *call_ops, |
1551 | struct pnfs_layout_segment *lseg, | |
1552 | int how) | |
0382b744 | 1553 | { |
cd841605 | 1554 | struct inode *inode = hdr->inode; |
0382b744 AA |
1555 | enum pnfs_try_status trypnfs; |
1556 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1557 | ||
cd841605 | 1558 | hdr->mds_ops = call_ops; |
0382b744 AA |
1559 | |
1560 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
1561 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
1562 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 1563 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1564 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1565 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1566 | return trypnfs; | |
1567 | } | |
1568 | ||
dce81290 | 1569 | static void |
7f714720 WAA |
1570 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
1571 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 1572 | { |
dce81290 TM |
1573 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1574 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1575 | enum pnfs_try_status trypnfs; |
dce81290 TM |
1576 | |
1577 | desc->pg_lseg = NULL; | |
d45f60c6 | 1578 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 1579 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1580 | pnfs_write_through_mds(desc, hdr); |
9369a431 | 1581 | pnfs_put_lseg(lseg); |
dce81290 TM |
1582 | } |
1583 | ||
6c75dc0d FI |
1584 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1585 | { | |
9369a431 | 1586 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1587 | nfs_pgio_header_free(hdr); |
6c75dc0d | 1588 | } |
89d77c8f | 1589 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1590 | |
dce81290 TM |
1591 | int |
1592 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1593 | { | |
6c75dc0d | 1594 | struct nfs_pgio_header *hdr; |
dce81290 TM |
1595 | int ret; |
1596 | ||
1e7f3a48 WAA |
1597 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1598 | if (!hdr) { | |
9b5415b5 | 1599 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
9369a431 | 1600 | pnfs_put_lseg(desc->pg_lseg); |
dce81290 | 1601 | desc->pg_lseg = NULL; |
6c75dc0d | 1602 | return -ENOMEM; |
dce81290 | 1603 | } |
6c75dc0d | 1604 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
9369a431 | 1605 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1606 | ret = nfs_generic_pgio(desc, hdr); |
6c75dc0d | 1607 | if (ret != 0) { |
9369a431 | 1608 | pnfs_put_lseg(desc->pg_lseg); |
6c75dc0d | 1609 | desc->pg_lseg = NULL; |
6c75dc0d | 1610 | } else |
7f714720 | 1611 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
6c75dc0d | 1612 | return ret; |
dce81290 TM |
1613 | } |
1614 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1615 | ||
53113ad3 | 1616 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
1617 | { |
1618 | struct nfs_pageio_descriptor pgio; | |
1619 | ||
1acbbb4e | 1620 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
1621 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
1622 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 1623 | } |
e7dd79af | 1624 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 1625 | |
d45f60c6 | 1626 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1627 | { |
cd841605 FI |
1628 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
1629 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1630 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1631 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1632 | } |
4db6e0b7 | 1633 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1634 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
1635 | } |
1636 | ||
d20581aa BH |
1637 | /* |
1638 | * Called by non rpc-based layout drivers | |
1639 | */ | |
d45f60c6 | 1640 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 1641 | { |
d45f60c6 | 1642 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
cd841605 | 1643 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
1644 | __nfs4_read_done_cb(hdr); |
1645 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
62e4a769 | 1646 | } else |
d45f60c6 WAA |
1647 | pnfs_ld_handle_read_error(hdr); |
1648 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
1649 | } |
1650 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1651 | ||
493292dd TM |
1652 | static void |
1653 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1654 | struct nfs_pgio_header *hdr) |
493292dd | 1655 | { |
4db6e0b7 FI |
1656 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1657 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1658 | nfs_pageio_reset_read_mds(desc); | |
1659 | desc->pg_recoalesce = 1; | |
1660 | } | |
d45f60c6 | 1661 | nfs_pgio_data_destroy(hdr); |
493292dd TM |
1662 | } |
1663 | ||
64419a9b AA |
1664 | /* |
1665 | * Call the appropriate parallel I/O subsystem read function. | |
1666 | */ | |
493292dd | 1667 | static enum pnfs_try_status |
d45f60c6 | 1668 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
1669 | const struct rpc_call_ops *call_ops, |
1670 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1671 | { |
cd841605 | 1672 | struct inode *inode = hdr->inode; |
64419a9b AA |
1673 | struct nfs_server *nfss = NFS_SERVER(inode); |
1674 | enum pnfs_try_status trypnfs; | |
1675 | ||
cd841605 | 1676 | hdr->mds_ops = call_ops; |
64419a9b AA |
1677 | |
1678 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 1679 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 1680 | |
d45f60c6 | 1681 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 1682 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1683 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1684 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1685 | return trypnfs; | |
1686 | } | |
863a3c6c | 1687 | |
493292dd | 1688 | static void |
7f714720 | 1689 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 1690 | { |
493292dd TM |
1691 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1692 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1693 | enum pnfs_try_status trypnfs; |
493292dd TM |
1694 | |
1695 | desc->pg_lseg = NULL; | |
d45f60c6 | 1696 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
7f714720 | 1697 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1698 | pnfs_read_through_mds(desc, hdr); |
9369a431 | 1699 | pnfs_put_lseg(lseg); |
493292dd TM |
1700 | } |
1701 | ||
4db6e0b7 FI |
1702 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1703 | { | |
9369a431 | 1704 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1705 | nfs_pgio_header_free(hdr); |
4db6e0b7 | 1706 | } |
89d77c8f | 1707 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1708 | |
493292dd TM |
1709 | int |
1710 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1711 | { | |
4db6e0b7 | 1712 | struct nfs_pgio_header *hdr; |
493292dd TM |
1713 | int ret; |
1714 | ||
1e7f3a48 WAA |
1715 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1716 | if (!hdr) { | |
061ae2ed | 1717 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
4db6e0b7 | 1718 | ret = -ENOMEM; |
9369a431 | 1719 | pnfs_put_lseg(desc->pg_lseg); |
493292dd TM |
1720 | desc->pg_lseg = NULL; |
1721 | return ret; | |
1722 | } | |
4db6e0b7 | 1723 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 1724 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1725 | ret = nfs_generic_pgio(desc, hdr); |
4db6e0b7 | 1726 | if (ret != 0) { |
9369a431 | 1727 | pnfs_put_lseg(desc->pg_lseg); |
4db6e0b7 | 1728 | desc->pg_lseg = NULL; |
4db6e0b7 | 1729 | } else |
7f714720 | 1730 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 1731 | return ret; |
493292dd TM |
1732 | } |
1733 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1734 | ||
71244d9b TM |
1735 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
1736 | { | |
1737 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
1738 | ||
1739 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 1740 | smp_mb__after_atomic(); |
71244d9b TM |
1741 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
1742 | } | |
1743 | ||
863a3c6c | 1744 | /* |
a9bae566 | 1745 | * There can be multiple RW segments. |
863a3c6c | 1746 | */ |
a9bae566 | 1747 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1748 | { |
a9bae566 | 1749 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1750 | |
a9bae566 PT |
1751 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1752 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 1753 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
1754 | list_add(&lseg->pls_lc_list, listp); |
1755 | } | |
863a3c6c AA |
1756 | } |
1757 | ||
a073dbff TM |
1758 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
1759 | { | |
1760 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
1761 | |
1762 | /* Matched by references in pnfs_set_layoutcommit */ | |
1763 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
1764 | list_del_init(&lseg->pls_lc_list); | |
1765 | pnfs_put_lseg(lseg); | |
1766 | } | |
1767 | ||
71244d9b | 1768 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
1769 | } |
1770 | ||
1b0ae068 PT |
1771 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
1772 | { | |
b9e028fd | 1773 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
1774 | } |
1775 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
1776 | ||
863a3c6c | 1777 | void |
d45f60c6 | 1778 | pnfs_set_layoutcommit(struct nfs_pgio_header *hdr) |
863a3c6c | 1779 | { |
cd841605 FI |
1780 | struct inode *inode = hdr->inode; |
1781 | struct nfs_inode *nfsi = NFS_I(inode); | |
d45f60c6 | 1782 | loff_t end_pos = hdr->mds_offset + hdr->res.count; |
79a48a1f | 1783 | bool mark_as_dirty = false; |
863a3c6c | 1784 | |
cd841605 | 1785 | spin_lock(&inode->i_lock); |
863a3c6c | 1786 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 1787 | mark_as_dirty = true; |
863a3c6c | 1788 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 1789 | __func__, inode->i_ino); |
863a3c6c | 1790 | } |
cd841605 | 1791 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 1792 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 1793 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 1794 | } |
acff5880 PT |
1795 | if (end_pos > nfsi->layout->plh_lwb) |
1796 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 1797 | spin_unlock(&inode->i_lock); |
acff5880 | 1798 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 1799 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
1800 | |
1801 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1802 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1803 | if (mark_as_dirty) | |
cd841605 | 1804 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
1805 | } |
1806 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
1807 | ||
378520b8 PT |
1808 | void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data) |
1809 | { | |
1810 | struct inode *inode = data->inode; | |
1811 | struct nfs_inode *nfsi = NFS_I(inode); | |
1812 | bool mark_as_dirty = false; | |
1813 | ||
1814 | spin_lock(&inode->i_lock); | |
1815 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { | |
1816 | mark_as_dirty = true; | |
1817 | dprintk("%s: Set layoutcommit for inode %lu ", | |
1818 | __func__, inode->i_ino); | |
1819 | } | |
1820 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) { | |
1821 | /* references matched in nfs4_layoutcommit_release */ | |
1822 | pnfs_get_lseg(data->lseg); | |
1823 | } | |
1824 | if (data->lwb > nfsi->layout->plh_lwb) | |
1825 | nfsi->layout->plh_lwb = data->lwb; | |
1826 | spin_unlock(&inode->i_lock); | |
1827 | dprintk("%s: lseg %p end_pos %llu\n", | |
1828 | __func__, data->lseg, nfsi->layout->plh_lwb); | |
1829 | ||
1830 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1831 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1832 | if (mark_as_dirty) | |
1833 | mark_inode_dirty_sync(inode); | |
1834 | } | |
1835 | EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit); | |
1836 | ||
db29c089 AA |
1837 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
1838 | { | |
1839 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
1840 | ||
1841 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
1842 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 1843 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
1844 | } |
1845 | ||
de4b15c7 AA |
1846 | /* |
1847 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
1848 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
1849 | * data to disk to allow the server to recover the data if it crashes. | |
1850 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
1851 | * is off, and a COMMIT is sent to a data server, or | |
1852 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
1853 | */ | |
863a3c6c | 1854 | int |
ef311537 | 1855 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 1856 | { |
5f919c9f | 1857 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
1858 | struct nfs4_layoutcommit_data *data; |
1859 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 1860 | loff_t end_pos; |
71244d9b | 1861 | int status; |
863a3c6c | 1862 | |
71244d9b | 1863 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
1864 | return 0; |
1865 | ||
71244d9b | 1866 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 1867 | |
71244d9b | 1868 | status = -EAGAIN; |
92407e75 | 1869 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
1870 | if (!sync) |
1871 | goto out; | |
74316201 | 1872 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
1873 | NFS_INO_LAYOUTCOMMITTING, |
1874 | nfs_wait_bit_killable, | |
1875 | TASK_KILLABLE); | |
92407e75 | 1876 | if (status) |
71244d9b | 1877 | goto out; |
92407e75 PT |
1878 | } |
1879 | ||
71244d9b TM |
1880 | status = -ENOMEM; |
1881 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
1882 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
1883 | if (!data) | |
1884 | goto clear_layoutcommitting; | |
1885 | ||
1886 | status = 0; | |
de4b15c7 | 1887 | spin_lock(&inode->i_lock); |
71244d9b TM |
1888 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1889 | goto out_unlock; | |
a9bae566 | 1890 | |
71244d9b | 1891 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 1892 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 1893 | |
acff5880 | 1894 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 1895 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 1896 | |
f597c537 | 1897 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
1898 | spin_unlock(&inode->i_lock); |
1899 | ||
1900 | data->args.inode = inode; | |
9fa40758 | 1901 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
1902 | nfs_fattr_init(&data->fattr); |
1903 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
1904 | data->res.fattr = &data->fattr; | |
1905 | data->args.lastbytewritten = end_pos - 1; | |
1906 | data->res.server = NFS_SERVER(inode); | |
1907 | ||
5f919c9f CH |
1908 | if (ld->prepare_layoutcommit) { |
1909 | status = ld->prepare_layoutcommit(&data->args); | |
1910 | if (status) { | |
1911 | spin_lock(&inode->i_lock); | |
1912 | if (end_pos < nfsi->layout->plh_lwb) | |
1913 | nfsi->layout->plh_lwb = end_pos; | |
1914 | spin_unlock(&inode->i_lock); | |
1915 | put_rpccred(data->cred); | |
1916 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); | |
1917 | goto clear_layoutcommitting; | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | ||
863a3c6c AA |
1922 | status = nfs4_proc_layoutcommit(data, sync); |
1923 | out: | |
92407e75 PT |
1924 | if (status) |
1925 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
1926 | dprintk("<-- %s status %d\n", __func__, status); |
1927 | return status; | |
71244d9b TM |
1928 | out_unlock: |
1929 | spin_unlock(&inode->i_lock); | |
92407e75 | 1930 | kfree(data); |
71244d9b TM |
1931 | clear_layoutcommitting: |
1932 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 1933 | goto out; |
863a3c6c | 1934 | } |
82be417a AA |
1935 | |
1936 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
1937 | { | |
1938 | struct nfs4_threshold *thp; | |
1939 | ||
1940 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
1941 | if (!thp) { | |
1942 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
1943 | return NULL; | |
1944 | } | |
1945 | return thp; | |
1946 | } |