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 | ||
78096cca TM |
685 | static void |
686 | pnfs_verify_layout_stateid(struct pnfs_layout_hdr *lo, | |
687 | const nfs4_stateid *new, | |
688 | struct list_head *free_me_list) | |
689 | { | |
690 | if (nfs4_stateid_match_other(&lo->plh_stateid, new)) | |
691 | return; | |
692 | /* Layout is new! Kill existing layout segments */ | |
693 | pnfs_mark_matching_lsegs_invalid(lo, free_me_list, NULL); | |
694 | } | |
695 | ||
fd6002e9 | 696 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
697 | void |
698 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
699 | bool update_barrier) | |
b1f69b75 | 700 | { |
22aaf714 TM |
701 | u32 oldseq, newseq, new_barrier; |
702 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 703 | |
2d2f24ad TM |
704 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
705 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 706 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 707 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 708 | if (update_barrier) { |
22aaf714 | 709 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
710 | } else { |
711 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 712 | * "close" to the current seqids. |
43f1b3da | 713 | */ |
22aaf714 | 714 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 715 | } |
22aaf714 TM |
716 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
717 | lo->plh_barrier = new_barrier; | |
43f1b3da | 718 | } |
b1f69b75 AA |
719 | } |
720 | ||
cf7d63f1 | 721 | static bool |
19c54aba TM |
722 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
723 | const nfs4_stateid *stateid) | |
43f1b3da | 724 | { |
19c54aba | 725 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 726 | |
19c54aba TM |
727 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
728 | } | |
729 | ||
730 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
731 | static bool | |
732 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget) | |
733 | { | |
f7e8917a FI |
734 | return lo->plh_block_lgets || |
735 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || | |
43f1b3da | 736 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
737 | (atomic_read(&lo->plh_outstanding) > lget)); |
738 | } | |
739 | ||
fd6002e9 FI |
740 | int |
741 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
742 | struct nfs4_state *open_state) | |
b1f69b75 | 743 | { |
fd6002e9 | 744 | int status = 0; |
974cec8c | 745 | |
b1f69b75 | 746 | dprintk("--> %s\n", __func__); |
fd6002e9 | 747 | spin_lock(&lo->plh_inode->i_lock); |
19c54aba | 748 | if (pnfs_layoutgets_blocked(lo, 1)) { |
cf7d63f1 | 749 | status = -EAGAIN; |
5d422301 TM |
750 | } else if (!nfs4_valid_open_stateid(open_state)) { |
751 | status = -EBADF; | |
cf7d63f1 | 752 | } else if (list_empty(&lo->plh_segs)) { |
fd6002e9 FI |
753 | int seq; |
754 | ||
755 | do { | |
756 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 757 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
758 | } while (read_seqretry(&open_state->seqlock, seq)); |
759 | } else | |
f597c537 | 760 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 761 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 762 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 763 | return status; |
b1f69b75 AA |
764 | } |
765 | ||
766 | /* | |
767 | * Get layout from server. | |
768 | * for now, assume that whole file layouts are requested. | |
769 | * arg->offset: 0 | |
770 | * arg->length: all ones | |
771 | */ | |
e5e94017 BH |
772 | static struct pnfs_layout_segment * |
773 | send_layoutget(struct pnfs_layout_hdr *lo, | |
774 | struct nfs_open_context *ctx, | |
fb3296eb | 775 | struct pnfs_layout_range *range, |
a75b9df9 | 776 | gfp_t gfp_flags) |
e5e94017 | 777 | { |
b7edfaa1 | 778 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
779 | struct nfs_server *server = NFS_SERVER(ino); |
780 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 781 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
782 | |
783 | dprintk("--> %s\n", __func__); | |
e5e94017 | 784 | |
a75b9df9 | 785 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 786 | if (lgp == NULL) |
b1f69b75 | 787 | return NULL; |
35124a09 | 788 | |
fb3296eb BH |
789 | lgp->args.minlength = PAGE_CACHE_SIZE; |
790 | if (lgp->args.minlength > range->length) | |
791 | lgp->args.minlength = range->length; | |
b1f69b75 | 792 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 793 | lgp->args.range = *range; |
b1f69b75 AA |
794 | lgp->args.type = server->pnfs_curr_ld->id; |
795 | lgp->args.inode = ino; | |
796 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 797 | lgp->gfp_flags = gfp_flags; |
6ab59344 | 798 | lgp->cred = lo->plh_lc_cred; |
b1f69b75 AA |
799 | |
800 | /* Synchronously retrieve layout information from server and | |
801 | * store in lseg. | |
802 | */ | |
a0b0a6e3 TM |
803 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
804 | if (IS_ERR(lseg)) { | |
805 | switch (PTR_ERR(lseg)) { | |
806 | case -ENOMEM: | |
807 | case -ERESTARTSYS: | |
808 | break; | |
809 | default: | |
810 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 811 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
812 | } |
813 | return NULL; | |
974cec8c | 814 | } |
35124a09 | 815 | |
974cec8c AA |
816 | return lseg; |
817 | } | |
818 | ||
24956804 TM |
819 | static void pnfs_clear_layoutcommit(struct inode *inode, |
820 | struct list_head *head) | |
821 | { | |
822 | struct nfs_inode *nfsi = NFS_I(inode); | |
823 | struct pnfs_layout_segment *lseg, *tmp; | |
824 | ||
825 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
826 | return; | |
827 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
828 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
829 | continue; | |
830 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
831 | } | |
832 | } | |
833 | ||
293b3b06 AA |
834 | /* |
835 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
836 | * when the layout segment list is empty. | |
837 | * | |
838 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
839 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
840 | * deviceid is marked invalid. | |
841 | */ | |
cbe82603 BH |
842 | int |
843 | _pnfs_return_layout(struct inode *ino) | |
844 | { | |
845 | struct pnfs_layout_hdr *lo = NULL; | |
846 | struct nfs_inode *nfsi = NFS_I(ino); | |
847 | LIST_HEAD(tmp_list); | |
848 | struct nfs4_layoutreturn *lrp; | |
849 | nfs4_stateid stateid; | |
293b3b06 | 850 | int status = 0, empty; |
cbe82603 | 851 | |
366d5052 | 852 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
853 | |
854 | spin_lock(&ino->i_lock); | |
855 | lo = nfsi->layout; | |
e5929f3c | 856 | if (!lo) { |
cbe82603 | 857 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
858 | dprintk("NFS: %s no layout to return\n", __func__); |
859 | goto out; | |
cbe82603 BH |
860 | } |
861 | stateid = nfsi->layout->plh_stateid; | |
862 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 863 | pnfs_get_layout_hdr(lo); |
293b3b06 | 864 | empty = list_empty(&lo->plh_segs); |
24956804 | 865 | pnfs_clear_layoutcommit(ino, &tmp_list); |
49a85061 | 866 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
293b3b06 AA |
867 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
868 | if (empty) { | |
869 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 870 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
871 | dprintk("NFS: %s no layout segments to return\n", __func__); |
872 | goto out; | |
873 | } | |
ea0ded74 | 874 | lo->plh_block_lgets++; |
cbe82603 BH |
875 | spin_unlock(&ino->i_lock); |
876 | pnfs_free_lseg_list(&tmp_list); | |
877 | ||
cbe82603 BH |
878 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); |
879 | if (unlikely(lrp == NULL)) { | |
880 | status = -ENOMEM; | |
65857d57 TM |
881 | spin_lock(&ino->i_lock); |
882 | lo->plh_block_lgets--; | |
883 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 884 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
885 | goto out; |
886 | } | |
887 | ||
888 | lrp->args.stateid = stateid; | |
889 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
890 | lrp->args.inode = ino; | |
a56aaa02 | 891 | lrp->args.layout = lo; |
cbe82603 | 892 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
9556000d | 893 | lrp->cred = lo->plh_lc_cred; |
cbe82603 BH |
894 | |
895 | status = nfs4_proc_layoutreturn(lrp); | |
896 | out: | |
897 | dprintk("<-- %s status: %d\n", __func__, status); | |
898 | return status; | |
899 | } | |
0a57cdac | 900 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 901 | |
24028672 TM |
902 | int |
903 | pnfs_commit_and_return_layout(struct inode *inode) | |
904 | { | |
905 | struct pnfs_layout_hdr *lo; | |
906 | int ret; | |
907 | ||
908 | spin_lock(&inode->i_lock); | |
909 | lo = NFS_I(inode)->layout; | |
910 | if (lo == NULL) { | |
911 | spin_unlock(&inode->i_lock); | |
912 | return 0; | |
913 | } | |
914 | pnfs_get_layout_hdr(lo); | |
915 | /* Block new layoutgets and read/write to ds */ | |
916 | lo->plh_block_lgets++; | |
917 | spin_unlock(&inode->i_lock); | |
918 | filemap_fdatawait(inode->i_mapping); | |
919 | ret = pnfs_layoutcommit_inode(inode, true); | |
920 | if (ret == 0) | |
921 | ret = _pnfs_return_layout(inode); | |
922 | spin_lock(&inode->i_lock); | |
923 | lo->plh_block_lgets--; | |
924 | spin_unlock(&inode->i_lock); | |
925 | pnfs_put_layout_hdr(lo); | |
926 | return ret; | |
927 | } | |
928 | ||
f7e8917a FI |
929 | bool pnfs_roc(struct inode *ino) |
930 | { | |
931 | struct pnfs_layout_hdr *lo; | |
932 | struct pnfs_layout_segment *lseg, *tmp; | |
933 | LIST_HEAD(tmp_list); | |
934 | bool found = false; | |
935 | ||
936 | spin_lock(&ino->i_lock); | |
937 | lo = NFS_I(ino)->layout; | |
938 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
939 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
940 | goto out_nolayout; | |
941 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
942 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
943 | mark_lseg_invalid(lseg, &tmp_list); | |
944 | found = true; | |
945 | } | |
946 | if (!found) | |
947 | goto out_nolayout; | |
948 | lo->plh_block_lgets++; | |
70c3bd2b | 949 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
950 | spin_unlock(&ino->i_lock); |
951 | pnfs_free_lseg_list(&tmp_list); | |
952 | return true; | |
953 | ||
954 | out_nolayout: | |
955 | spin_unlock(&ino->i_lock); | |
956 | return false; | |
957 | } | |
958 | ||
959 | void pnfs_roc_release(struct inode *ino) | |
960 | { | |
961 | struct pnfs_layout_hdr *lo; | |
962 | ||
963 | spin_lock(&ino->i_lock); | |
964 | lo = NFS_I(ino)->layout; | |
965 | lo->plh_block_lgets--; | |
6622c3ea TM |
966 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
967 | pnfs_detach_layout_hdr(lo); | |
968 | spin_unlock(&ino->i_lock); | |
969 | pnfs_free_layout_hdr(lo); | |
970 | } else | |
971 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
972 | } |
973 | ||
974 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
975 | { | |
976 | struct pnfs_layout_hdr *lo; | |
977 | ||
978 | spin_lock(&ino->i_lock); | |
979 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 980 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
981 | lo->plh_barrier = barrier; |
982 | spin_unlock(&ino->i_lock); | |
983 | } | |
984 | ||
7fdab069 | 985 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
986 | { |
987 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 988 | struct pnfs_layout_hdr *lo; |
f7e8917a | 989 | struct pnfs_layout_segment *lseg; |
7fdab069 | 990 | u32 current_seqid; |
f7e8917a FI |
991 | bool found = false; |
992 | ||
993 | spin_lock(&ino->i_lock); | |
994 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
995 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 996 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 997 | found = true; |
7fdab069 | 998 | goto out; |
f7e8917a | 999 | } |
7fdab069 TM |
1000 | lo = nfsi->layout; |
1001 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1002 | |
7fdab069 TM |
1003 | /* Since close does not return a layout stateid for use as |
1004 | * a barrier, we choose the worst-case barrier. | |
1005 | */ | |
1006 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
1007 | out: | |
f7e8917a FI |
1008 | spin_unlock(&ino->i_lock); |
1009 | return found; | |
1010 | } | |
1011 | ||
b1f69b75 AA |
1012 | /* |
1013 | * Compare two layout segments for sorting into layout cache. | |
1014 | * We want to preferentially return RW over RO layouts, so ensure those | |
1015 | * are seen first. | |
1016 | */ | |
1017 | static s64 | |
7dc0ac70 | 1018 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1019 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1020 | { |
fb3296eb BH |
1021 | s64 d; |
1022 | ||
1023 | /* high offset > low offset */ | |
1024 | d = l1->offset - l2->offset; | |
1025 | if (d) | |
1026 | return d; | |
1027 | ||
1028 | /* short length > long length */ | |
1029 | d = l2->length - l1->length; | |
1030 | if (d) | |
1031 | return d; | |
1032 | ||
b1f69b75 | 1033 | /* read > read/write */ |
fb3296eb | 1034 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1035 | } |
1036 | ||
974cec8c | 1037 | static void |
57036a37 | 1038 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
1039 | struct pnfs_layout_segment *lseg) |
1040 | { | |
b1f69b75 | 1041 | struct pnfs_layout_segment *lp; |
b1f69b75 | 1042 | |
974cec8c AA |
1043 | dprintk("%s:Begin\n", __func__); |
1044 | ||
b7edfaa1 | 1045 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
7dc0ac70 | 1046 | if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 1047 | continue; |
566052c5 | 1048 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1049 | dprintk("%s: inserted lseg %p " |
1050 | "iomode %d offset %llu length %llu before " | |
1051 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1052 | __func__, lseg, lseg->pls_range.iomode, |
1053 | lseg->pls_range.offset, lseg->pls_range.length, | |
1054 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1055 | lp->pls_range.length); | |
fb3296eb | 1056 | goto out; |
974cec8c | 1057 | } |
fb3296eb BH |
1058 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1059 | dprintk("%s: inserted lseg %p " | |
1060 | "iomode %d offset %llu length %llu at tail\n", | |
1061 | __func__, lseg, lseg->pls_range.iomode, | |
1062 | lseg->pls_range.offset, lseg->pls_range.length); | |
1063 | out: | |
70c3bd2b | 1064 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1065 | |
1066 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
1067 | } |
1068 | ||
1069 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1070 | alloc_init_layout_hdr(struct inode *ino, |
1071 | struct nfs_open_context *ctx, | |
1072 | gfp_t gfp_flags) | |
e5e94017 BH |
1073 | { |
1074 | struct pnfs_layout_hdr *lo; | |
1075 | ||
636fb9c8 | 1076 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1077 | if (!lo) |
1078 | return NULL; | |
cc6e5340 | 1079 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1080 | INIT_LIST_HEAD(&lo->plh_layouts); |
1081 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1082 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1083 | lo->plh_inode = ino; |
5cc2216d | 1084 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
e5e94017 BH |
1085 | return lo; |
1086 | } | |
1087 | ||
1088 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1089 | pnfs_find_alloc_layout(struct inode *ino, |
1090 | struct nfs_open_context *ctx, | |
1091 | gfp_t gfp_flags) | |
e5e94017 BH |
1092 | { |
1093 | struct nfs_inode *nfsi = NFS_I(ino); | |
1094 | struct pnfs_layout_hdr *new = NULL; | |
1095 | ||
1096 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1097 | ||
251ec410 TM |
1098 | if (nfsi->layout != NULL) |
1099 | goto out_existing; | |
e5e94017 | 1100 | spin_unlock(&ino->i_lock); |
9fa40758 | 1101 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1102 | spin_lock(&ino->i_lock); |
1103 | ||
251ec410 | 1104 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1105 | nfsi->layout = new; |
251ec410 | 1106 | return new; |
7175fe90 YN |
1107 | } else if (new != NULL) |
1108 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1109 | out_existing: |
1110 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1111 | return nfsi->layout; |
1112 | } | |
1113 | ||
b1f69b75 AA |
1114 | /* |
1115 | * iomode matching rules: | |
1116 | * iomode lseg match | |
1117 | * ----- ----- ----- | |
1118 | * ANY READ true | |
1119 | * ANY RW true | |
1120 | * RW READ false | |
1121 | * RW RW true | |
1122 | * READ READ true | |
1123 | * READ RW true | |
1124 | */ | |
3cb2df17 | 1125 | static bool |
7dc0ac70 | 1126 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
3cb2df17 | 1127 | const struct pnfs_layout_range *range) |
b1f69b75 | 1128 | { |
fb3296eb BH |
1129 | struct pnfs_layout_range range1; |
1130 | ||
1131 | if ((range->iomode == IOMODE_RW && | |
1132 | ls_range->iomode != IOMODE_RW) || | |
7dc0ac70 | 1133 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1134 | return 0; |
1135 | ||
1136 | /* range1 covers only the first byte in the range */ | |
1137 | range1 = *range; | |
1138 | range1.length = 1; | |
7dc0ac70 | 1139 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1140 | } |
1141 | ||
1142 | /* | |
1143 | * lookup range in layout | |
1144 | */ | |
e5e94017 | 1145 | static struct pnfs_layout_segment * |
fb3296eb BH |
1146 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
1147 | struct pnfs_layout_range *range) | |
e5e94017 | 1148 | { |
b1f69b75 AA |
1149 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1150 | ||
1151 | dprintk("%s:Begin\n", __func__); | |
1152 | ||
b7edfaa1 | 1153 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1154 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
7dc0ac70 | 1155 | pnfs_lseg_range_match(&lseg->pls_range, range)) { |
9369a431 | 1156 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1157 | break; |
1158 | } | |
d771e3a4 | 1159 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
1160 | break; |
1161 | } | |
1162 | ||
1163 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1164 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1165 | return ret; |
e5e94017 BH |
1166 | } |
1167 | ||
d23d61c8 AA |
1168 | /* |
1169 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1170 | * to the MDS or over pNFS | |
1171 | * | |
1172 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1173 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1174 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1175 | * WRITE request. | |
1176 | * | |
1177 | * A return of true means use MDS I/O. | |
1178 | * | |
1179 | * From rfc 5661: | |
1180 | * If a file's size is smaller than the file size threshold, data accesses | |
1181 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1182 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1183 | * server. If both file size and I/O size are provided, the client SHOULD | |
1184 | * reach or exceed both thresholds before sending its read or write | |
1185 | * requests to the data server. | |
1186 | */ | |
1187 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1188 | struct inode *ino, int iomode) | |
1189 | { | |
1190 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1191 | struct nfs_inode *nfsi = NFS_I(ino); | |
1192 | loff_t fsize = i_size_read(ino); | |
1193 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1194 | ||
1195 | if (t == NULL) | |
1196 | return ret; | |
1197 | ||
1198 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1199 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1200 | ||
1201 | switch (iomode) { | |
1202 | case IOMODE_READ: | |
1203 | if (t->bm & THRESHOLD_RD) { | |
1204 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1205 | size_set = true; | |
1206 | if (fsize < t->rd_sz) | |
1207 | size = true; | |
1208 | } | |
1209 | if (t->bm & THRESHOLD_RD_IO) { | |
1210 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1211 | nfsi->read_io); | |
1212 | io_set = true; | |
1213 | if (nfsi->read_io < t->rd_io_sz) | |
1214 | io = true; | |
1215 | } | |
1216 | break; | |
1217 | case IOMODE_RW: | |
1218 | if (t->bm & THRESHOLD_WR) { | |
1219 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1220 | size_set = true; | |
1221 | if (fsize < t->wr_sz) | |
1222 | size = true; | |
1223 | } | |
1224 | if (t->bm & THRESHOLD_WR_IO) { | |
1225 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1226 | nfsi->write_io); | |
1227 | io_set = true; | |
1228 | if (nfsi->write_io < t->wr_io_sz) | |
1229 | io = true; | |
1230 | } | |
1231 | break; | |
1232 | } | |
1233 | if (size_set && io_set) { | |
1234 | if (size && io) | |
1235 | ret = true; | |
1236 | } else if (size || io) | |
1237 | ret = true; | |
1238 | ||
1239 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1240 | return ret; | |
1241 | } | |
1242 | ||
e5e94017 BH |
1243 | /* |
1244 | * Layout segment is retreived from the server if not cached. | |
1245 | * The appropriate layout segment is referenced and returned to the caller. | |
1246 | */ | |
7c24d948 | 1247 | struct pnfs_layout_segment * |
e5e94017 BH |
1248 | pnfs_update_layout(struct inode *ino, |
1249 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1250 | loff_t pos, |
1251 | u64 count, | |
a75b9df9 TM |
1252 | enum pnfs_iomode iomode, |
1253 | gfp_t gfp_flags) | |
e5e94017 | 1254 | { |
fb3296eb BH |
1255 | struct pnfs_layout_range arg = { |
1256 | .iomode = iomode, | |
1257 | .offset = pos, | |
1258 | .length = count, | |
1259 | }; | |
707ed5fd | 1260 | unsigned pg_offset; |
6382a441 WAA |
1261 | struct nfs_server *server = NFS_SERVER(ino); |
1262 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1263 | struct pnfs_layout_hdr *lo; |
1264 | struct pnfs_layout_segment *lseg = NULL; | |
30005121 | 1265 | bool first; |
e5e94017 BH |
1266 | |
1267 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1268 | goto out; |
d23d61c8 AA |
1269 | |
1270 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1271 | goto out; |
d23d61c8 | 1272 | |
e5e94017 | 1273 | spin_lock(&ino->i_lock); |
9fa40758 | 1274 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1275 | if (lo == NULL) { |
1276 | spin_unlock(&ino->i_lock); | |
1277 | goto out; | |
1278 | } | |
e5e94017 | 1279 | |
43f1b3da | 1280 | /* Do we even need to bother with this? */ |
a59c30ac | 1281 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1282 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1283 | goto out_unlock; |
1284 | } | |
1285 | ||
1286 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1287 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1288 | goto out_unlock; |
1289 | ||
568e8c49 | 1290 | /* Check to see if the layout for the given range already exists */ |
fb3296eb | 1291 | lseg = pnfs_find_lseg(lo, &arg); |
568e8c49 AA |
1292 | if (lseg) |
1293 | goto out_unlock; | |
1294 | ||
19c54aba | 1295 | if (pnfs_layoutgets_blocked(lo, 0)) |
cf7d63f1 FI |
1296 | goto out_unlock; |
1297 | atomic_inc(&lo->plh_outstanding); | |
1298 | ||
30005121 | 1299 | first = list_empty(&lo->plh_layouts) ? true : false; |
f49f9baa | 1300 | spin_unlock(&ino->i_lock); |
30005121 | 1301 | |
f49f9baa | 1302 | if (first) { |
2130ff66 FI |
1303 | /* The lo must be on the clp list if there is any |
1304 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1305 | */ | |
1306 | spin_lock(&clp->cl_lock); | |
6382a441 | 1307 | list_add_tail(&lo->plh_layouts, &server->layouts); |
2130ff66 FI |
1308 | spin_unlock(&clp->cl_lock); |
1309 | } | |
e5e94017 | 1310 | |
707ed5fd BH |
1311 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1312 | if (pg_offset) { | |
1313 | arg.offset -= pg_offset; | |
1314 | arg.length += pg_offset; | |
1315 | } | |
7c24d948 AA |
1316 | if (arg.length != NFS4_MAX_UINT64) |
1317 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1318 | |
fb3296eb | 1319 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
cf7d63f1 | 1320 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1321 | out_put_layout_hdr: |
70c3bd2b | 1322 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1323 | out: |
f86bbcf8 TM |
1324 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1325 | "(%s, offset: %llu, length: %llu)\n", | |
1326 | __func__, ino->i_sb->s_id, | |
1327 | (unsigned long long)NFS_FILEID(ino), | |
1328 | lseg == NULL ? "not found" : "found", | |
1329 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1330 | (unsigned long long)pos, | |
1331 | (unsigned long long)count); | |
e5e94017 BH |
1332 | return lseg; |
1333 | out_unlock: | |
1334 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1335 | goto out_put_layout_hdr; |
e5e94017 | 1336 | } |
7c24d948 | 1337 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1338 | |
a0b0a6e3 | 1339 | struct pnfs_layout_segment * |
b1f69b75 AA |
1340 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1341 | { | |
1342 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1343 | struct nfs4_layoutget_res *res = &lgp->res; | |
1344 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1345 | struct inode *ino = lo->plh_inode; |
78096cca | 1346 | LIST_HEAD(free_me); |
b1f69b75 AA |
1347 | int status = 0; |
1348 | ||
1349 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1350 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1351 | if (!lseg || IS_ERR(lseg)) { |
1352 | if (!lseg) | |
1353 | status = -ENOMEM; | |
1354 | else | |
1355 | status = PTR_ERR(lseg); | |
1356 | dprintk("%s: Could not allocate layout: error %d\n", | |
1357 | __func__, status); | |
1358 | goto out; | |
1359 | } | |
1360 | ||
1361 | spin_lock(&ino->i_lock); | |
a59c30ac | 1362 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1363 | dprintk("%s forget reply due to recall\n", __func__); |
1364 | goto out_forget_reply; | |
1365 | } | |
1366 | ||
19c54aba TM |
1367 | if (pnfs_layoutgets_blocked(lo, 1) || |
1368 | pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
43f1b3da FI |
1369 | dprintk("%s forget reply due to state\n", __func__); |
1370 | goto out_forget_reply; | |
1371 | } | |
038d6493 | 1372 | |
78096cca TM |
1373 | /* Check that the new stateid matches the old stateid */ |
1374 | pnfs_verify_layout_stateid(lo, &res->stateid, &free_me); | |
038d6493 TM |
1375 | /* Done processing layoutget. Set the layout stateid */ |
1376 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1377 | ||
b1f69b75 | 1378 | init_lseg(lo, lseg); |
566052c5 | 1379 | lseg->pls_range = res->range; |
9369a431 | 1380 | pnfs_get_lseg(lseg); |
57036a37 | 1381 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1382 | |
f7e8917a FI |
1383 | if (res->return_on_close) { |
1384 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1385 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1386 | } | |
1387 | ||
b1f69b75 | 1388 | spin_unlock(&ino->i_lock); |
78096cca | 1389 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1390 | return lseg; |
b1f69b75 | 1391 | out: |
a0b0a6e3 | 1392 | return ERR_PTR(status); |
43f1b3da FI |
1393 | |
1394 | out_forget_reply: | |
1395 | spin_unlock(&ino->i_lock); | |
1396 | lseg->pls_layout = lo; | |
1397 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1398 | goto out; | |
b1f69b75 AA |
1399 | } |
1400 | ||
d8007d4d TM |
1401 | void |
1402 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1403 | { | |
1fd937bd PT |
1404 | u64 rd_size = req->wb_bytes; |
1405 | ||
bc5a89b3 | 1406 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d | 1407 | |
1fd937bd PT |
1408 | if (pgio->pg_dreq == NULL) |
1409 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1410 | else | |
1411 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1412 | ||
d8007d4d TM |
1413 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1414 | req->wb_context, | |
1415 | req_offset(req), | |
1fd937bd | 1416 | rd_size, |
d8007d4d TM |
1417 | IOMODE_READ, |
1418 | GFP_KERNEL); | |
e885de1a TM |
1419 | /* If no lseg, fall back to read through mds */ |
1420 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1421 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1422 | |
d8007d4d TM |
1423 | } |
1424 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1425 | ||
1426 | void | |
6296556f PT |
1427 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1428 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1429 | { |
bc5a89b3 | 1430 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d TM |
1431 | |
1432 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1433 | req->wb_context, | |
1434 | req_offset(req), | |
6296556f | 1435 | wb_size, |
d8007d4d TM |
1436 | IOMODE_RW, |
1437 | GFP_NOFS); | |
e885de1a TM |
1438 | /* If no lseg, fall back to write through mds */ |
1439 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1440 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1441 | } |
1442 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1443 | ||
b4fdac1a WAA |
1444 | /* |
1445 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1446 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1447 | */ | |
1448 | size_t | |
dfed206b BH |
1449 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
1450 | struct nfs_page *req) | |
94ad1c80 | 1451 | { |
0f9c429e | 1452 | unsigned int size; |
c5e20cb7 | 1453 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1454 | |
1455 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1456 | if (!size) |
1457 | return 0; | |
94ad1c80 | 1458 | |
19982ba8 | 1459 | /* |
c5e20cb7 WAA |
1460 | * 'size' contains the number of bytes left in the current page (up |
1461 | * to the original size asked for in @req->wb_bytes). | |
1462 | * | |
1463 | * Calculate how many bytes are left in the layout segment | |
1464 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1465 | * |
1466 | * Please also note that 'end_offset' is actually the offset of the | |
1467 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1468 | * | |
1469 | */ | |
19b54848 | 1470 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1471 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1472 | pgio->pg_lseg->pls_range.length); | |
1473 | req_start = req_offset(req); | |
1474 | WARN_ON_ONCE(req_start > seg_end); | |
1475 | /* start of request is past the last byte of this segment */ | |
1476 | if (req_start >= seg_end) | |
19b54848 | 1477 | return 0; |
c5e20cb7 WAA |
1478 | |
1479 | /* adjust 'size' iff there are fewer bytes left in the | |
1480 | * segment than what nfs_generic_pg_test returned */ | |
1481 | seg_left = seg_end - req_start; | |
1482 | if (seg_left < size) | |
1483 | size = (unsigned int)seg_left; | |
19b54848 | 1484 | } |
0f9c429e | 1485 | |
19b54848 | 1486 | return size; |
94ad1c80 | 1487 | } |
89a58e32 | 1488 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1489 | |
53113ad3 | 1490 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
1491 | { |
1492 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
1493 | |
1494 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
1495 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
1496 | hdr->completion_ops); | |
1497 | return nfs_pageio_resend(&pgio, hdr); | |
e2fecb21 | 1498 | } |
e7dd79af | 1499 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1500 | |
d45f60c6 | 1501 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1502 | { |
cd841605 FI |
1503 | |
1504 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1505 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1506 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1507 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1508 | } |
6c75dc0d | 1509 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1510 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
1511 | } |
1512 | ||
d20581aa BH |
1513 | /* |
1514 | * Called by non rpc-based layout drivers | |
1515 | */ | |
d45f60c6 | 1516 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 1517 | { |
d45f60c6 | 1518 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); |
cd841605 | 1519 | if (!hdr->pnfs_error) { |
d45f60c6 WAA |
1520 | pnfs_set_layoutcommit(hdr); |
1521 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
1acbbb4e | 1522 | } else |
d45f60c6 WAA |
1523 | pnfs_ld_handle_write_error(hdr); |
1524 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 1525 | } |
d20581aa | 1526 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1527 | |
dce81290 TM |
1528 | static void |
1529 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1530 | struct nfs_pgio_header *hdr) |
dce81290 | 1531 | { |
6c75dc0d FI |
1532 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1533 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1534 | nfs_pageio_reset_write_mds(desc); | |
1535 | desc->pg_recoalesce = 1; | |
1536 | } | |
d45f60c6 | 1537 | nfs_pgio_data_destroy(hdr); |
dce81290 TM |
1538 | } |
1539 | ||
1540 | static enum pnfs_try_status | |
d45f60c6 | 1541 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
1542 | const struct rpc_call_ops *call_ops, |
1543 | struct pnfs_layout_segment *lseg, | |
1544 | int how) | |
0382b744 | 1545 | { |
cd841605 | 1546 | struct inode *inode = hdr->inode; |
0382b744 AA |
1547 | enum pnfs_try_status trypnfs; |
1548 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1549 | ||
cd841605 | 1550 | hdr->mds_ops = call_ops; |
0382b744 AA |
1551 | |
1552 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
1553 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
1554 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 1555 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1556 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1557 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1558 | return trypnfs; | |
1559 | } | |
1560 | ||
dce81290 | 1561 | static void |
7f714720 WAA |
1562 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
1563 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 1564 | { |
dce81290 TM |
1565 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1566 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1567 | enum pnfs_try_status trypnfs; |
dce81290 TM |
1568 | |
1569 | desc->pg_lseg = NULL; | |
d45f60c6 | 1570 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 1571 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1572 | pnfs_write_through_mds(desc, hdr); |
9369a431 | 1573 | pnfs_put_lseg(lseg); |
dce81290 TM |
1574 | } |
1575 | ||
6c75dc0d FI |
1576 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1577 | { | |
9369a431 | 1578 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1579 | nfs_pgio_header_free(hdr); |
6c75dc0d | 1580 | } |
89d77c8f | 1581 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1582 | |
dce81290 TM |
1583 | int |
1584 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1585 | { | |
6c75dc0d | 1586 | struct nfs_pgio_header *hdr; |
dce81290 TM |
1587 | int ret; |
1588 | ||
1e7f3a48 WAA |
1589 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1590 | if (!hdr) { | |
9b5415b5 | 1591 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
9369a431 | 1592 | pnfs_put_lseg(desc->pg_lseg); |
dce81290 | 1593 | desc->pg_lseg = NULL; |
6c75dc0d | 1594 | return -ENOMEM; |
dce81290 | 1595 | } |
6c75dc0d | 1596 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
9369a431 | 1597 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1598 | ret = nfs_generic_pgio(desc, hdr); |
6c75dc0d | 1599 | if (ret != 0) { |
9369a431 | 1600 | pnfs_put_lseg(desc->pg_lseg); |
6c75dc0d | 1601 | desc->pg_lseg = NULL; |
6c75dc0d | 1602 | } else |
7f714720 | 1603 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
6c75dc0d | 1604 | return ret; |
dce81290 TM |
1605 | } |
1606 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1607 | ||
53113ad3 | 1608 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
1609 | { |
1610 | struct nfs_pageio_descriptor pgio; | |
1611 | ||
1acbbb4e | 1612 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
1613 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
1614 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 1615 | } |
e7dd79af | 1616 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 1617 | |
d45f60c6 | 1618 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1619 | { |
cd841605 FI |
1620 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
1621 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1622 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1623 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1624 | } |
4db6e0b7 | 1625 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1626 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
1627 | } |
1628 | ||
d20581aa BH |
1629 | /* |
1630 | * Called by non rpc-based layout drivers | |
1631 | */ | |
d45f60c6 | 1632 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 1633 | { |
d45f60c6 | 1634 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
cd841605 | 1635 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
1636 | __nfs4_read_done_cb(hdr); |
1637 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
62e4a769 | 1638 | } else |
d45f60c6 WAA |
1639 | pnfs_ld_handle_read_error(hdr); |
1640 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
1641 | } |
1642 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1643 | ||
493292dd TM |
1644 | static void |
1645 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1646 | struct nfs_pgio_header *hdr) |
493292dd | 1647 | { |
4db6e0b7 FI |
1648 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1649 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1650 | nfs_pageio_reset_read_mds(desc); | |
1651 | desc->pg_recoalesce = 1; | |
1652 | } | |
d45f60c6 | 1653 | nfs_pgio_data_destroy(hdr); |
493292dd TM |
1654 | } |
1655 | ||
64419a9b AA |
1656 | /* |
1657 | * Call the appropriate parallel I/O subsystem read function. | |
1658 | */ | |
493292dd | 1659 | static enum pnfs_try_status |
d45f60c6 | 1660 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
1661 | const struct rpc_call_ops *call_ops, |
1662 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1663 | { |
cd841605 | 1664 | struct inode *inode = hdr->inode; |
64419a9b AA |
1665 | struct nfs_server *nfss = NFS_SERVER(inode); |
1666 | enum pnfs_try_status trypnfs; | |
1667 | ||
cd841605 | 1668 | hdr->mds_ops = call_ops; |
64419a9b AA |
1669 | |
1670 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 1671 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 1672 | |
d45f60c6 | 1673 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 1674 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1675 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1676 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1677 | return trypnfs; | |
1678 | } | |
863a3c6c | 1679 | |
493292dd | 1680 | static void |
7f714720 | 1681 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 1682 | { |
493292dd TM |
1683 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1684 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1685 | enum pnfs_try_status trypnfs; |
493292dd TM |
1686 | |
1687 | desc->pg_lseg = NULL; | |
d45f60c6 | 1688 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
7f714720 | 1689 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1690 | pnfs_read_through_mds(desc, hdr); |
9369a431 | 1691 | pnfs_put_lseg(lseg); |
493292dd TM |
1692 | } |
1693 | ||
4db6e0b7 FI |
1694 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1695 | { | |
9369a431 | 1696 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1697 | nfs_pgio_header_free(hdr); |
4db6e0b7 | 1698 | } |
89d77c8f | 1699 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1700 | |
493292dd TM |
1701 | int |
1702 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1703 | { | |
4db6e0b7 | 1704 | struct nfs_pgio_header *hdr; |
493292dd TM |
1705 | int ret; |
1706 | ||
1e7f3a48 WAA |
1707 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1708 | if (!hdr) { | |
061ae2ed | 1709 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
4db6e0b7 | 1710 | ret = -ENOMEM; |
9369a431 | 1711 | pnfs_put_lseg(desc->pg_lseg); |
493292dd TM |
1712 | desc->pg_lseg = NULL; |
1713 | return ret; | |
1714 | } | |
4db6e0b7 | 1715 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 1716 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1717 | ret = nfs_generic_pgio(desc, hdr); |
4db6e0b7 | 1718 | if (ret != 0) { |
9369a431 | 1719 | pnfs_put_lseg(desc->pg_lseg); |
4db6e0b7 | 1720 | desc->pg_lseg = NULL; |
4db6e0b7 | 1721 | } else |
7f714720 | 1722 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 1723 | return ret; |
493292dd TM |
1724 | } |
1725 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1726 | ||
71244d9b TM |
1727 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
1728 | { | |
1729 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
1730 | ||
1731 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 1732 | smp_mb__after_atomic(); |
71244d9b TM |
1733 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
1734 | } | |
1735 | ||
863a3c6c | 1736 | /* |
a9bae566 | 1737 | * There can be multiple RW segments. |
863a3c6c | 1738 | */ |
a9bae566 | 1739 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1740 | { |
a9bae566 | 1741 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1742 | |
a9bae566 PT |
1743 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1744 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 1745 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
1746 | list_add(&lseg->pls_lc_list, listp); |
1747 | } | |
863a3c6c AA |
1748 | } |
1749 | ||
a073dbff TM |
1750 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
1751 | { | |
1752 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
1753 | |
1754 | /* Matched by references in pnfs_set_layoutcommit */ | |
1755 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
1756 | list_del_init(&lseg->pls_lc_list); | |
1757 | pnfs_put_lseg(lseg); | |
1758 | } | |
1759 | ||
71244d9b | 1760 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
1761 | } |
1762 | ||
1b0ae068 PT |
1763 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
1764 | { | |
b9e028fd | 1765 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
1766 | } |
1767 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
1768 | ||
863a3c6c | 1769 | void |
d45f60c6 | 1770 | pnfs_set_layoutcommit(struct nfs_pgio_header *hdr) |
863a3c6c | 1771 | { |
cd841605 FI |
1772 | struct inode *inode = hdr->inode; |
1773 | struct nfs_inode *nfsi = NFS_I(inode); | |
d45f60c6 | 1774 | loff_t end_pos = hdr->mds_offset + hdr->res.count; |
79a48a1f | 1775 | bool mark_as_dirty = false; |
863a3c6c | 1776 | |
cd841605 | 1777 | spin_lock(&inode->i_lock); |
863a3c6c | 1778 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 1779 | mark_as_dirty = true; |
863a3c6c | 1780 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 1781 | __func__, inode->i_ino); |
863a3c6c | 1782 | } |
cd841605 | 1783 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 1784 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 1785 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 1786 | } |
acff5880 PT |
1787 | if (end_pos > nfsi->layout->plh_lwb) |
1788 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 1789 | spin_unlock(&inode->i_lock); |
acff5880 | 1790 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 1791 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
1792 | |
1793 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1794 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1795 | if (mark_as_dirty) | |
cd841605 | 1796 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
1797 | } |
1798 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
1799 | ||
378520b8 PT |
1800 | void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data) |
1801 | { | |
1802 | struct inode *inode = data->inode; | |
1803 | struct nfs_inode *nfsi = NFS_I(inode); | |
1804 | bool mark_as_dirty = false; | |
1805 | ||
1806 | spin_lock(&inode->i_lock); | |
1807 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { | |
1808 | mark_as_dirty = true; | |
1809 | dprintk("%s: Set layoutcommit for inode %lu ", | |
1810 | __func__, inode->i_ino); | |
1811 | } | |
1812 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) { | |
1813 | /* references matched in nfs4_layoutcommit_release */ | |
1814 | pnfs_get_lseg(data->lseg); | |
1815 | } | |
1816 | if (data->lwb > nfsi->layout->plh_lwb) | |
1817 | nfsi->layout->plh_lwb = data->lwb; | |
1818 | spin_unlock(&inode->i_lock); | |
1819 | dprintk("%s: lseg %p end_pos %llu\n", | |
1820 | __func__, data->lseg, nfsi->layout->plh_lwb); | |
1821 | ||
1822 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1823 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1824 | if (mark_as_dirty) | |
1825 | mark_inode_dirty_sync(inode); | |
1826 | } | |
1827 | EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit); | |
1828 | ||
db29c089 AA |
1829 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
1830 | { | |
1831 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
1832 | ||
1833 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
1834 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 1835 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
1836 | } |
1837 | ||
de4b15c7 AA |
1838 | /* |
1839 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
1840 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
1841 | * data to disk to allow the server to recover the data if it crashes. | |
1842 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
1843 | * is off, and a COMMIT is sent to a data server, or | |
1844 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
1845 | */ | |
863a3c6c | 1846 | int |
ef311537 | 1847 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c AA |
1848 | { |
1849 | struct nfs4_layoutcommit_data *data; | |
1850 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 1851 | loff_t end_pos; |
71244d9b | 1852 | int status; |
863a3c6c | 1853 | |
71244d9b | 1854 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
1855 | return 0; |
1856 | ||
71244d9b | 1857 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 1858 | |
71244d9b | 1859 | status = -EAGAIN; |
92407e75 | 1860 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
1861 | if (!sync) |
1862 | goto out; | |
74316201 | 1863 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
1864 | NFS_INO_LAYOUTCOMMITTING, |
1865 | nfs_wait_bit_killable, | |
1866 | TASK_KILLABLE); | |
92407e75 | 1867 | if (status) |
71244d9b | 1868 | goto out; |
92407e75 PT |
1869 | } |
1870 | ||
71244d9b TM |
1871 | status = -ENOMEM; |
1872 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
1873 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
1874 | if (!data) | |
1875 | goto clear_layoutcommitting; | |
1876 | ||
1877 | status = 0; | |
de4b15c7 | 1878 | spin_lock(&inode->i_lock); |
71244d9b TM |
1879 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1880 | goto out_unlock; | |
a9bae566 | 1881 | |
71244d9b | 1882 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 1883 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 1884 | |
acff5880 | 1885 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 1886 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 1887 | |
f597c537 | 1888 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
1889 | spin_unlock(&inode->i_lock); |
1890 | ||
1891 | data->args.inode = inode; | |
9fa40758 | 1892 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
1893 | nfs_fattr_init(&data->fattr); |
1894 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
1895 | data->res.fattr = &data->fattr; | |
1896 | data->args.lastbytewritten = end_pos - 1; | |
1897 | data->res.server = NFS_SERVER(inode); | |
1898 | ||
1899 | status = nfs4_proc_layoutcommit(data, sync); | |
1900 | out: | |
92407e75 PT |
1901 | if (status) |
1902 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
1903 | dprintk("<-- %s status %d\n", __func__, status); |
1904 | return status; | |
71244d9b TM |
1905 | out_unlock: |
1906 | spin_unlock(&inode->i_lock); | |
92407e75 | 1907 | kfree(data); |
71244d9b TM |
1908 | clear_layoutcommitting: |
1909 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 1910 | goto out; |
863a3c6c | 1911 | } |
82be417a AA |
1912 | |
1913 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
1914 | { | |
1915 | struct nfs4_threshold *thp; | |
1916 | ||
1917 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
1918 | if (!thp) { | |
1919 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
1920 | return NULL; | |
1921 | } | |
1922 | return thp; | |
1923 | } |