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; | |
7c5d1875 CH |
597 | |
598 | pnfs_layoutcommit_inode(inode, false); | |
599 | ||
fd9a8d71 TM |
600 | spin_lock(&inode->i_lock); |
601 | list_del_init(&lo->plh_bulk_destroy); | |
602 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ | |
603 | if (is_bulk_recall) | |
604 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
605 | if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range)) | |
606 | ret = -EAGAIN; | |
607 | spin_unlock(&inode->i_lock); | |
608 | pnfs_free_lseg_list(&lseg_list); | |
609 | pnfs_put_layout_hdr(lo); | |
610 | iput(inode); | |
611 | } | |
612 | return ret; | |
613 | } | |
614 | ||
615 | int | |
616 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
617 | struct nfs_fsid *fsid, | |
618 | bool is_recall) | |
619 | { | |
620 | struct nfs_server *server; | |
621 | LIST_HEAD(layout_list); | |
c47abcf8 | 622 | |
974cec8c | 623 | spin_lock(&clp->cl_lock); |
6382a441 | 624 | rcu_read_lock(); |
fd9a8d71 | 625 | restart: |
6382a441 | 626 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
627 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
628 | continue; | |
629 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
630 | server, | |
631 | &layout_list) != 0) | |
632 | goto restart; | |
6382a441 WAA |
633 | } |
634 | rcu_read_unlock(); | |
974cec8c AA |
635 | spin_unlock(&clp->cl_lock); |
636 | ||
fd9a8d71 TM |
637 | if (list_empty(&layout_list)) |
638 | return 0; | |
639 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
640 | } | |
641 | ||
642 | int | |
643 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
644 | bool is_recall) | |
645 | { | |
646 | struct nfs_server *server; | |
647 | LIST_HEAD(layout_list); | |
648 | ||
649 | spin_lock(&clp->cl_lock); | |
650 | rcu_read_lock(); | |
651 | restart: | |
652 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
653 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
654 | server, | |
655 | &layout_list) != 0) | |
656 | goto restart; | |
974cec8c | 657 | } |
fd9a8d71 TM |
658 | rcu_read_unlock(); |
659 | spin_unlock(&clp->cl_lock); | |
660 | ||
661 | if (list_empty(&layout_list)) | |
662 | return 0; | |
663 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
664 | } | |
665 | ||
666 | /* | |
667 | * Called by the state manger to remove all layouts established under an | |
668 | * expired lease. | |
669 | */ | |
670 | void | |
671 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
672 | { | |
673 | nfs4_deviceid_mark_client_invalid(clp); | |
674 | nfs4_deviceid_purge_client(clp); | |
675 | ||
676 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
677 | } |
678 | ||
5a65503f TM |
679 | /* |
680 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
681 | * taking into account wraparound issues. | |
682 | */ | |
683 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
684 | { | |
2c64c57d | 685 | return (s32)(s1 - s2) > 0; |
5a65503f TM |
686 | } |
687 | ||
fd6002e9 | 688 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
689 | void |
690 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
691 | bool update_barrier) | |
b1f69b75 | 692 | { |
22aaf714 TM |
693 | u32 oldseq, newseq, new_barrier; |
694 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 695 | |
2d2f24ad TM |
696 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
697 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 698 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 699 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 700 | if (update_barrier) { |
22aaf714 | 701 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
702 | } else { |
703 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 704 | * "close" to the current seqids. |
43f1b3da | 705 | */ |
22aaf714 | 706 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 707 | } |
22aaf714 TM |
708 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
709 | lo->plh_barrier = new_barrier; | |
43f1b3da | 710 | } |
b1f69b75 AA |
711 | } |
712 | ||
cf7d63f1 | 713 | static bool |
19c54aba TM |
714 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
715 | const nfs4_stateid *stateid) | |
43f1b3da | 716 | { |
19c54aba | 717 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 718 | |
19c54aba TM |
719 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
720 | } | |
721 | ||
722 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
723 | static bool | |
724 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget) | |
725 | { | |
f7e8917a FI |
726 | return lo->plh_block_lgets || |
727 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || | |
43f1b3da | 728 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
729 | (atomic_read(&lo->plh_outstanding) > lget)); |
730 | } | |
731 | ||
fd6002e9 FI |
732 | int |
733 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
734 | struct nfs4_state *open_state) | |
b1f69b75 | 735 | { |
fd6002e9 | 736 | int status = 0; |
974cec8c | 737 | |
b1f69b75 | 738 | dprintk("--> %s\n", __func__); |
fd6002e9 | 739 | spin_lock(&lo->plh_inode->i_lock); |
19c54aba | 740 | if (pnfs_layoutgets_blocked(lo, 1)) { |
cf7d63f1 | 741 | status = -EAGAIN; |
5d422301 TM |
742 | } else if (!nfs4_valid_open_stateid(open_state)) { |
743 | status = -EBADF; | |
47abadef CH |
744 | } else if (list_empty(&lo->plh_segs) || |
745 | test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { | |
fd6002e9 FI |
746 | int seq; |
747 | ||
748 | do { | |
749 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 750 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
751 | } while (read_seqretry(&open_state->seqlock, seq)); |
752 | } else | |
f597c537 | 753 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 754 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 755 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 756 | return status; |
b1f69b75 AA |
757 | } |
758 | ||
759 | /* | |
760 | * Get layout from server. | |
761 | * for now, assume that whole file layouts are requested. | |
762 | * arg->offset: 0 | |
763 | * arg->length: all ones | |
764 | */ | |
e5e94017 BH |
765 | static struct pnfs_layout_segment * |
766 | send_layoutget(struct pnfs_layout_hdr *lo, | |
767 | struct nfs_open_context *ctx, | |
fb3296eb | 768 | struct pnfs_layout_range *range, |
a75b9df9 | 769 | gfp_t gfp_flags) |
e5e94017 | 770 | { |
b7edfaa1 | 771 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
772 | struct nfs_server *server = NFS_SERVER(ino); |
773 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 774 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
775 | |
776 | dprintk("--> %s\n", __func__); | |
e5e94017 | 777 | |
a75b9df9 | 778 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 779 | if (lgp == NULL) |
b1f69b75 | 780 | return NULL; |
35124a09 | 781 | |
fb3296eb BH |
782 | lgp->args.minlength = PAGE_CACHE_SIZE; |
783 | if (lgp->args.minlength > range->length) | |
784 | lgp->args.minlength = range->length; | |
b1f69b75 | 785 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 786 | lgp->args.range = *range; |
b1f69b75 AA |
787 | lgp->args.type = server->pnfs_curr_ld->id; |
788 | lgp->args.inode = ino; | |
789 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 790 | lgp->gfp_flags = gfp_flags; |
6ab59344 | 791 | lgp->cred = lo->plh_lc_cred; |
b1f69b75 AA |
792 | |
793 | /* Synchronously retrieve layout information from server and | |
794 | * store in lseg. | |
795 | */ | |
a0b0a6e3 TM |
796 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
797 | if (IS_ERR(lseg)) { | |
798 | switch (PTR_ERR(lseg)) { | |
799 | case -ENOMEM: | |
800 | case -ERESTARTSYS: | |
801 | break; | |
802 | default: | |
803 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 804 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
805 | } |
806 | return NULL; | |
974cec8c | 807 | } |
35124a09 | 808 | |
974cec8c AA |
809 | return lseg; |
810 | } | |
811 | ||
24956804 TM |
812 | static void pnfs_clear_layoutcommit(struct inode *inode, |
813 | struct list_head *head) | |
814 | { | |
815 | struct nfs_inode *nfsi = NFS_I(inode); | |
816 | struct pnfs_layout_segment *lseg, *tmp; | |
817 | ||
818 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
819 | return; | |
820 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
821 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
822 | continue; | |
823 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
824 | } | |
825 | } | |
826 | ||
293b3b06 AA |
827 | /* |
828 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
829 | * when the layout segment list is empty. | |
830 | * | |
831 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
832 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
833 | * deviceid is marked invalid. | |
834 | */ | |
cbe82603 BH |
835 | int |
836 | _pnfs_return_layout(struct inode *ino) | |
837 | { | |
838 | struct pnfs_layout_hdr *lo = NULL; | |
839 | struct nfs_inode *nfsi = NFS_I(ino); | |
840 | LIST_HEAD(tmp_list); | |
841 | struct nfs4_layoutreturn *lrp; | |
842 | nfs4_stateid stateid; | |
293b3b06 | 843 | int status = 0, empty; |
cbe82603 | 844 | |
366d5052 | 845 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
846 | |
847 | spin_lock(&ino->i_lock); | |
848 | lo = nfsi->layout; | |
e5929f3c | 849 | if (!lo) { |
cbe82603 | 850 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
851 | dprintk("NFS: %s no layout to return\n", __func__); |
852 | goto out; | |
cbe82603 BH |
853 | } |
854 | stateid = nfsi->layout->plh_stateid; | |
855 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 856 | pnfs_get_layout_hdr(lo); |
293b3b06 | 857 | empty = list_empty(&lo->plh_segs); |
24956804 | 858 | pnfs_clear_layoutcommit(ino, &tmp_list); |
49a85061 | 859 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
293b3b06 AA |
860 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
861 | if (empty) { | |
862 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 863 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
864 | dprintk("NFS: %s no layout segments to return\n", __func__); |
865 | goto out; | |
866 | } | |
47abadef CH |
867 | |
868 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
ea0ded74 | 869 | lo->plh_block_lgets++; |
cbe82603 BH |
870 | spin_unlock(&ino->i_lock); |
871 | pnfs_free_lseg_list(&tmp_list); | |
872 | ||
cbe82603 BH |
873 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); |
874 | if (unlikely(lrp == NULL)) { | |
875 | status = -ENOMEM; | |
65857d57 TM |
876 | spin_lock(&ino->i_lock); |
877 | lo->plh_block_lgets--; | |
878 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 879 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
880 | goto out; |
881 | } | |
882 | ||
883 | lrp->args.stateid = stateid; | |
884 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
885 | lrp->args.inode = ino; | |
a56aaa02 | 886 | lrp->args.layout = lo; |
cbe82603 | 887 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
9556000d | 888 | lrp->cred = lo->plh_lc_cred; |
cbe82603 BH |
889 | |
890 | status = nfs4_proc_layoutreturn(lrp); | |
891 | out: | |
892 | dprintk("<-- %s status: %d\n", __func__, status); | |
893 | return status; | |
894 | } | |
0a57cdac | 895 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 896 | |
24028672 TM |
897 | int |
898 | pnfs_commit_and_return_layout(struct inode *inode) | |
899 | { | |
900 | struct pnfs_layout_hdr *lo; | |
901 | int ret; | |
902 | ||
903 | spin_lock(&inode->i_lock); | |
904 | lo = NFS_I(inode)->layout; | |
905 | if (lo == NULL) { | |
906 | spin_unlock(&inode->i_lock); | |
907 | return 0; | |
908 | } | |
909 | pnfs_get_layout_hdr(lo); | |
910 | /* Block new layoutgets and read/write to ds */ | |
911 | lo->plh_block_lgets++; | |
912 | spin_unlock(&inode->i_lock); | |
913 | filemap_fdatawait(inode->i_mapping); | |
914 | ret = pnfs_layoutcommit_inode(inode, true); | |
915 | if (ret == 0) | |
916 | ret = _pnfs_return_layout(inode); | |
917 | spin_lock(&inode->i_lock); | |
918 | lo->plh_block_lgets--; | |
919 | spin_unlock(&inode->i_lock); | |
920 | pnfs_put_layout_hdr(lo); | |
921 | return ret; | |
922 | } | |
923 | ||
f7e8917a FI |
924 | bool pnfs_roc(struct inode *ino) |
925 | { | |
926 | struct pnfs_layout_hdr *lo; | |
927 | struct pnfs_layout_segment *lseg, *tmp; | |
928 | LIST_HEAD(tmp_list); | |
929 | bool found = false; | |
930 | ||
931 | spin_lock(&ino->i_lock); | |
932 | lo = NFS_I(ino)->layout; | |
933 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
934 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
935 | goto out_nolayout; | |
936 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
937 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
938 | mark_lseg_invalid(lseg, &tmp_list); | |
939 | found = true; | |
940 | } | |
941 | if (!found) | |
942 | goto out_nolayout; | |
943 | lo->plh_block_lgets++; | |
70c3bd2b | 944 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
945 | spin_unlock(&ino->i_lock); |
946 | pnfs_free_lseg_list(&tmp_list); | |
947 | return true; | |
948 | ||
949 | out_nolayout: | |
950 | spin_unlock(&ino->i_lock); | |
951 | return false; | |
952 | } | |
953 | ||
954 | void pnfs_roc_release(struct inode *ino) | |
955 | { | |
956 | struct pnfs_layout_hdr *lo; | |
957 | ||
958 | spin_lock(&ino->i_lock); | |
959 | lo = NFS_I(ino)->layout; | |
960 | lo->plh_block_lgets--; | |
6622c3ea TM |
961 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
962 | pnfs_detach_layout_hdr(lo); | |
963 | spin_unlock(&ino->i_lock); | |
964 | pnfs_free_layout_hdr(lo); | |
965 | } else | |
966 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
967 | } |
968 | ||
969 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
970 | { | |
971 | struct pnfs_layout_hdr *lo; | |
972 | ||
973 | spin_lock(&ino->i_lock); | |
974 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 975 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
976 | lo->plh_barrier = barrier; |
977 | spin_unlock(&ino->i_lock); | |
978 | } | |
979 | ||
7fdab069 | 980 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
981 | { |
982 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 983 | struct pnfs_layout_hdr *lo; |
f7e8917a | 984 | struct pnfs_layout_segment *lseg; |
7fdab069 | 985 | u32 current_seqid; |
f7e8917a FI |
986 | bool found = false; |
987 | ||
988 | spin_lock(&ino->i_lock); | |
989 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
990 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 991 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 992 | found = true; |
7fdab069 | 993 | goto out; |
f7e8917a | 994 | } |
7fdab069 TM |
995 | lo = nfsi->layout; |
996 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 997 | |
7fdab069 TM |
998 | /* Since close does not return a layout stateid for use as |
999 | * a barrier, we choose the worst-case barrier. | |
1000 | */ | |
1001 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
1002 | out: | |
f7e8917a FI |
1003 | spin_unlock(&ino->i_lock); |
1004 | return found; | |
1005 | } | |
1006 | ||
b1f69b75 AA |
1007 | /* |
1008 | * Compare two layout segments for sorting into layout cache. | |
1009 | * We want to preferentially return RW over RO layouts, so ensure those | |
1010 | * are seen first. | |
1011 | */ | |
1012 | static s64 | |
7dc0ac70 | 1013 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1014 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1015 | { |
fb3296eb BH |
1016 | s64 d; |
1017 | ||
1018 | /* high offset > low offset */ | |
1019 | d = l1->offset - l2->offset; | |
1020 | if (d) | |
1021 | return d; | |
1022 | ||
1023 | /* short length > long length */ | |
1024 | d = l2->length - l1->length; | |
1025 | if (d) | |
1026 | return d; | |
1027 | ||
b1f69b75 | 1028 | /* read > read/write */ |
fb3296eb | 1029 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1030 | } |
1031 | ||
974cec8c | 1032 | static void |
57036a37 | 1033 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
1034 | struct pnfs_layout_segment *lseg) |
1035 | { | |
b1f69b75 | 1036 | struct pnfs_layout_segment *lp; |
b1f69b75 | 1037 | |
974cec8c AA |
1038 | dprintk("%s:Begin\n", __func__); |
1039 | ||
b7edfaa1 | 1040 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
7dc0ac70 | 1041 | if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 1042 | continue; |
566052c5 | 1043 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1044 | dprintk("%s: inserted lseg %p " |
1045 | "iomode %d offset %llu length %llu before " | |
1046 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1047 | __func__, lseg, lseg->pls_range.iomode, |
1048 | lseg->pls_range.offset, lseg->pls_range.length, | |
1049 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1050 | lp->pls_range.length); | |
fb3296eb | 1051 | goto out; |
974cec8c | 1052 | } |
fb3296eb BH |
1053 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1054 | dprintk("%s: inserted lseg %p " | |
1055 | "iomode %d offset %llu length %llu at tail\n", | |
1056 | __func__, lseg, lseg->pls_range.iomode, | |
1057 | lseg->pls_range.offset, lseg->pls_range.length); | |
1058 | out: | |
70c3bd2b | 1059 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1060 | |
1061 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
1062 | } |
1063 | ||
1064 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1065 | alloc_init_layout_hdr(struct inode *ino, |
1066 | struct nfs_open_context *ctx, | |
1067 | gfp_t gfp_flags) | |
e5e94017 BH |
1068 | { |
1069 | struct pnfs_layout_hdr *lo; | |
1070 | ||
636fb9c8 | 1071 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1072 | if (!lo) |
1073 | return NULL; | |
cc6e5340 | 1074 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1075 | INIT_LIST_HEAD(&lo->plh_layouts); |
1076 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1077 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1078 | lo->plh_inode = ino; |
5cc2216d | 1079 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
e5e94017 BH |
1080 | return lo; |
1081 | } | |
1082 | ||
1083 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1084 | pnfs_find_alloc_layout(struct inode *ino, |
1085 | struct nfs_open_context *ctx, | |
1086 | gfp_t gfp_flags) | |
e5e94017 BH |
1087 | { |
1088 | struct nfs_inode *nfsi = NFS_I(ino); | |
1089 | struct pnfs_layout_hdr *new = NULL; | |
1090 | ||
1091 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1092 | ||
251ec410 TM |
1093 | if (nfsi->layout != NULL) |
1094 | goto out_existing; | |
e5e94017 | 1095 | spin_unlock(&ino->i_lock); |
9fa40758 | 1096 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1097 | spin_lock(&ino->i_lock); |
1098 | ||
251ec410 | 1099 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1100 | nfsi->layout = new; |
251ec410 | 1101 | return new; |
7175fe90 YN |
1102 | } else if (new != NULL) |
1103 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1104 | out_existing: |
1105 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1106 | return nfsi->layout; |
1107 | } | |
1108 | ||
b1f69b75 AA |
1109 | /* |
1110 | * iomode matching rules: | |
1111 | * iomode lseg match | |
1112 | * ----- ----- ----- | |
1113 | * ANY READ true | |
1114 | * ANY RW true | |
1115 | * RW READ false | |
1116 | * RW RW true | |
1117 | * READ READ true | |
1118 | * READ RW true | |
1119 | */ | |
3cb2df17 | 1120 | static bool |
7dc0ac70 | 1121 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
3cb2df17 | 1122 | const struct pnfs_layout_range *range) |
b1f69b75 | 1123 | { |
fb3296eb BH |
1124 | struct pnfs_layout_range range1; |
1125 | ||
1126 | if ((range->iomode == IOMODE_RW && | |
1127 | ls_range->iomode != IOMODE_RW) || | |
7dc0ac70 | 1128 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1129 | return 0; |
1130 | ||
1131 | /* range1 covers only the first byte in the range */ | |
1132 | range1 = *range; | |
1133 | range1.length = 1; | |
7dc0ac70 | 1134 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1135 | } |
1136 | ||
1137 | /* | |
1138 | * lookup range in layout | |
1139 | */ | |
e5e94017 | 1140 | static struct pnfs_layout_segment * |
fb3296eb BH |
1141 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
1142 | struct pnfs_layout_range *range) | |
e5e94017 | 1143 | { |
b1f69b75 AA |
1144 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1145 | ||
1146 | dprintk("%s:Begin\n", __func__); | |
1147 | ||
b7edfaa1 | 1148 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1149 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
7dc0ac70 | 1150 | pnfs_lseg_range_match(&lseg->pls_range, range)) { |
9369a431 | 1151 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1152 | break; |
1153 | } | |
d771e3a4 | 1154 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
1155 | break; |
1156 | } | |
1157 | ||
1158 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1159 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1160 | return ret; |
e5e94017 BH |
1161 | } |
1162 | ||
d23d61c8 AA |
1163 | /* |
1164 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1165 | * to the MDS or over pNFS | |
1166 | * | |
1167 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1168 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1169 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1170 | * WRITE request. | |
1171 | * | |
1172 | * A return of true means use MDS I/O. | |
1173 | * | |
1174 | * From rfc 5661: | |
1175 | * If a file's size is smaller than the file size threshold, data accesses | |
1176 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1177 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1178 | * server. If both file size and I/O size are provided, the client SHOULD | |
1179 | * reach or exceed both thresholds before sending its read or write | |
1180 | * requests to the data server. | |
1181 | */ | |
1182 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1183 | struct inode *ino, int iomode) | |
1184 | { | |
1185 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1186 | struct nfs_inode *nfsi = NFS_I(ino); | |
1187 | loff_t fsize = i_size_read(ino); | |
1188 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1189 | ||
1190 | if (t == NULL) | |
1191 | return ret; | |
1192 | ||
1193 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1194 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1195 | ||
1196 | switch (iomode) { | |
1197 | case IOMODE_READ: | |
1198 | if (t->bm & THRESHOLD_RD) { | |
1199 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1200 | size_set = true; | |
1201 | if (fsize < t->rd_sz) | |
1202 | size = true; | |
1203 | } | |
1204 | if (t->bm & THRESHOLD_RD_IO) { | |
1205 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1206 | nfsi->read_io); | |
1207 | io_set = true; | |
1208 | if (nfsi->read_io < t->rd_io_sz) | |
1209 | io = true; | |
1210 | } | |
1211 | break; | |
1212 | case IOMODE_RW: | |
1213 | if (t->bm & THRESHOLD_WR) { | |
1214 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1215 | size_set = true; | |
1216 | if (fsize < t->wr_sz) | |
1217 | size = true; | |
1218 | } | |
1219 | if (t->bm & THRESHOLD_WR_IO) { | |
1220 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1221 | nfsi->write_io); | |
1222 | io_set = true; | |
1223 | if (nfsi->write_io < t->wr_io_sz) | |
1224 | io = true; | |
1225 | } | |
1226 | break; | |
1227 | } | |
1228 | if (size_set && io_set) { | |
1229 | if (size && io) | |
1230 | ret = true; | |
1231 | } else if (size || io) | |
1232 | ret = true; | |
1233 | ||
1234 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1235 | return ret; | |
1236 | } | |
1237 | ||
e5e94017 BH |
1238 | /* |
1239 | * Layout segment is retreived from the server if not cached. | |
1240 | * The appropriate layout segment is referenced and returned to the caller. | |
1241 | */ | |
7c24d948 | 1242 | struct pnfs_layout_segment * |
e5e94017 BH |
1243 | pnfs_update_layout(struct inode *ino, |
1244 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1245 | loff_t pos, |
1246 | u64 count, | |
a75b9df9 TM |
1247 | enum pnfs_iomode iomode, |
1248 | gfp_t gfp_flags) | |
e5e94017 | 1249 | { |
fb3296eb BH |
1250 | struct pnfs_layout_range arg = { |
1251 | .iomode = iomode, | |
1252 | .offset = pos, | |
1253 | .length = count, | |
1254 | }; | |
707ed5fd | 1255 | unsigned pg_offset; |
6382a441 WAA |
1256 | struct nfs_server *server = NFS_SERVER(ino); |
1257 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1258 | struct pnfs_layout_hdr *lo; |
1259 | struct pnfs_layout_segment *lseg = NULL; | |
30005121 | 1260 | bool first; |
e5e94017 BH |
1261 | |
1262 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1263 | goto out; |
d23d61c8 AA |
1264 | |
1265 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1266 | goto out; |
d23d61c8 | 1267 | |
e5e94017 | 1268 | spin_lock(&ino->i_lock); |
9fa40758 | 1269 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1270 | if (lo == NULL) { |
1271 | spin_unlock(&ino->i_lock); | |
1272 | goto out; | |
1273 | } | |
e5e94017 | 1274 | |
43f1b3da | 1275 | /* Do we even need to bother with this? */ |
a59c30ac | 1276 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1277 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1278 | goto out_unlock; |
1279 | } | |
1280 | ||
1281 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1282 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1283 | goto out_unlock; |
1284 | ||
568e8c49 | 1285 | /* Check to see if the layout for the given range already exists */ |
fb3296eb | 1286 | lseg = pnfs_find_lseg(lo, &arg); |
568e8c49 AA |
1287 | if (lseg) |
1288 | goto out_unlock; | |
1289 | ||
19c54aba | 1290 | if (pnfs_layoutgets_blocked(lo, 0)) |
cf7d63f1 FI |
1291 | goto out_unlock; |
1292 | atomic_inc(&lo->plh_outstanding); | |
1293 | ||
30005121 | 1294 | first = list_empty(&lo->plh_layouts) ? true : false; |
f49f9baa | 1295 | spin_unlock(&ino->i_lock); |
30005121 | 1296 | |
f49f9baa | 1297 | if (first) { |
2130ff66 FI |
1298 | /* The lo must be on the clp list if there is any |
1299 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1300 | */ | |
1301 | spin_lock(&clp->cl_lock); | |
6382a441 | 1302 | list_add_tail(&lo->plh_layouts, &server->layouts); |
2130ff66 FI |
1303 | spin_unlock(&clp->cl_lock); |
1304 | } | |
e5e94017 | 1305 | |
707ed5fd BH |
1306 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1307 | if (pg_offset) { | |
1308 | arg.offset -= pg_offset; | |
1309 | arg.length += pg_offset; | |
1310 | } | |
7c24d948 AA |
1311 | if (arg.length != NFS4_MAX_UINT64) |
1312 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1313 | |
fb3296eb | 1314 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
cf7d63f1 | 1315 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1316 | out_put_layout_hdr: |
70c3bd2b | 1317 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1318 | out: |
f86bbcf8 TM |
1319 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1320 | "(%s, offset: %llu, length: %llu)\n", | |
1321 | __func__, ino->i_sb->s_id, | |
1322 | (unsigned long long)NFS_FILEID(ino), | |
1323 | lseg == NULL ? "not found" : "found", | |
1324 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1325 | (unsigned long long)pos, | |
1326 | (unsigned long long)count); | |
e5e94017 BH |
1327 | return lseg; |
1328 | out_unlock: | |
1329 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1330 | goto out_put_layout_hdr; |
e5e94017 | 1331 | } |
7c24d948 | 1332 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1333 | |
a0b0a6e3 | 1334 | struct pnfs_layout_segment * |
b1f69b75 AA |
1335 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1336 | { | |
1337 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1338 | struct nfs4_layoutget_res *res = &lgp->res; | |
1339 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1340 | struct inode *ino = lo->plh_inode; |
78096cca | 1341 | LIST_HEAD(free_me); |
b1f69b75 AA |
1342 | int status = 0; |
1343 | ||
1344 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1345 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1346 | if (!lseg || IS_ERR(lseg)) { |
1347 | if (!lseg) | |
1348 | status = -ENOMEM; | |
1349 | else | |
1350 | status = PTR_ERR(lseg); | |
1351 | dprintk("%s: Could not allocate layout: error %d\n", | |
1352 | __func__, status); | |
1353 | goto out; | |
1354 | } | |
1355 | ||
1013df61 CH |
1356 | init_lseg(lo, lseg); |
1357 | lseg->pls_range = res->range; | |
1358 | ||
b1f69b75 | 1359 | spin_lock(&ino->i_lock); |
a59c30ac | 1360 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1361 | dprintk("%s forget reply due to recall\n", __func__); |
1362 | goto out_forget_reply; | |
1363 | } | |
1364 | ||
362f7474 | 1365 | if (pnfs_layoutgets_blocked(lo, 1)) { |
43f1b3da FI |
1366 | dprintk("%s forget reply due to state\n", __func__); |
1367 | goto out_forget_reply; | |
1368 | } | |
038d6493 | 1369 | |
362f7474 CH |
1370 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1371 | /* existing state ID, make sure the sequence number matches. */ | |
1372 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1373 | dprintk("%s forget reply due to sequence\n", __func__); | |
1374 | goto out_forget_reply; | |
1375 | } | |
1376 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1377 | } else { | |
1378 | /* | |
1379 | * We got an entirely new state ID. Mark all segments for the | |
1380 | * inode invalid, and don't bother validating the stateid | |
1381 | * sequence number. | |
1382 | */ | |
1383 | pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL); | |
1384 | ||
1385 | nfs4_stateid_copy(&lo->plh_stateid, &res->stateid); | |
1386 | lo->plh_barrier = be32_to_cpu(res->stateid.seqid); | |
1387 | } | |
038d6493 | 1388 | |
47abadef CH |
1389 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); |
1390 | ||
9369a431 | 1391 | pnfs_get_lseg(lseg); |
57036a37 | 1392 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1393 | |
f7e8917a FI |
1394 | if (res->return_on_close) { |
1395 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1396 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1397 | } | |
1398 | ||
b1f69b75 | 1399 | spin_unlock(&ino->i_lock); |
78096cca | 1400 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1401 | return lseg; |
b1f69b75 | 1402 | out: |
a0b0a6e3 | 1403 | return ERR_PTR(status); |
43f1b3da FI |
1404 | |
1405 | out_forget_reply: | |
1406 | spin_unlock(&ino->i_lock); | |
1407 | lseg->pls_layout = lo; | |
1408 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1409 | goto out; | |
b1f69b75 AA |
1410 | } |
1411 | ||
d8007d4d TM |
1412 | void |
1413 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1414 | { | |
1fd937bd PT |
1415 | u64 rd_size = req->wb_bytes; |
1416 | ||
bc5a89b3 | 1417 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d | 1418 | |
1fd937bd PT |
1419 | if (pgio->pg_dreq == NULL) |
1420 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1421 | else | |
1422 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1423 | ||
d8007d4d TM |
1424 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1425 | req->wb_context, | |
1426 | req_offset(req), | |
1fd937bd | 1427 | rd_size, |
d8007d4d TM |
1428 | IOMODE_READ, |
1429 | GFP_KERNEL); | |
e885de1a TM |
1430 | /* If no lseg, fall back to read through mds */ |
1431 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1432 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1433 | |
d8007d4d TM |
1434 | } |
1435 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1436 | ||
1437 | void | |
6296556f PT |
1438 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1439 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1440 | { |
bc5a89b3 | 1441 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d TM |
1442 | |
1443 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1444 | req->wb_context, | |
1445 | req_offset(req), | |
6296556f | 1446 | wb_size, |
d8007d4d TM |
1447 | IOMODE_RW, |
1448 | GFP_NOFS); | |
e885de1a TM |
1449 | /* If no lseg, fall back to write through mds */ |
1450 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1451 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1452 | } |
1453 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1454 | ||
b4fdac1a WAA |
1455 | /* |
1456 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1457 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1458 | */ | |
1459 | size_t | |
dfed206b BH |
1460 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
1461 | struct nfs_page *req) | |
94ad1c80 | 1462 | { |
0f9c429e | 1463 | unsigned int size; |
c5e20cb7 | 1464 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1465 | |
1466 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1467 | if (!size) |
1468 | return 0; | |
94ad1c80 | 1469 | |
19982ba8 | 1470 | /* |
c5e20cb7 WAA |
1471 | * 'size' contains the number of bytes left in the current page (up |
1472 | * to the original size asked for in @req->wb_bytes). | |
1473 | * | |
1474 | * Calculate how many bytes are left in the layout segment | |
1475 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1476 | * |
1477 | * Please also note that 'end_offset' is actually the offset of the | |
1478 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1479 | * | |
1480 | */ | |
19b54848 | 1481 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1482 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1483 | pgio->pg_lseg->pls_range.length); | |
1484 | req_start = req_offset(req); | |
1485 | WARN_ON_ONCE(req_start > seg_end); | |
1486 | /* start of request is past the last byte of this segment */ | |
1487 | if (req_start >= seg_end) | |
19b54848 | 1488 | return 0; |
c5e20cb7 WAA |
1489 | |
1490 | /* adjust 'size' iff there are fewer bytes left in the | |
1491 | * segment than what nfs_generic_pg_test returned */ | |
1492 | seg_left = seg_end - req_start; | |
1493 | if (seg_left < size) | |
1494 | size = (unsigned int)seg_left; | |
19b54848 | 1495 | } |
0f9c429e | 1496 | |
19b54848 | 1497 | return size; |
94ad1c80 | 1498 | } |
89a58e32 | 1499 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1500 | |
53113ad3 | 1501 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
1502 | { |
1503 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
1504 | |
1505 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
1506 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
1507 | hdr->completion_ops); | |
1508 | return nfs_pageio_resend(&pgio, hdr); | |
e2fecb21 | 1509 | } |
e7dd79af | 1510 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1511 | |
d45f60c6 | 1512 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1513 | { |
cd841605 FI |
1514 | |
1515 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1516 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1517 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1518 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1519 | } |
6c75dc0d | 1520 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1521 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
1522 | } |
1523 | ||
d20581aa BH |
1524 | /* |
1525 | * Called by non rpc-based layout drivers | |
1526 | */ | |
d45f60c6 | 1527 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 1528 | { |
d45f60c6 | 1529 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); |
cd841605 | 1530 | if (!hdr->pnfs_error) { |
d45f60c6 WAA |
1531 | pnfs_set_layoutcommit(hdr); |
1532 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
1acbbb4e | 1533 | } else |
d45f60c6 WAA |
1534 | pnfs_ld_handle_write_error(hdr); |
1535 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 1536 | } |
d20581aa | 1537 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1538 | |
dce81290 TM |
1539 | static void |
1540 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1541 | struct nfs_pgio_header *hdr) |
dce81290 | 1542 | { |
6c75dc0d FI |
1543 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1544 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1545 | nfs_pageio_reset_write_mds(desc); | |
1546 | desc->pg_recoalesce = 1; | |
1547 | } | |
d45f60c6 | 1548 | nfs_pgio_data_destroy(hdr); |
dce81290 TM |
1549 | } |
1550 | ||
1551 | static enum pnfs_try_status | |
d45f60c6 | 1552 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
1553 | const struct rpc_call_ops *call_ops, |
1554 | struct pnfs_layout_segment *lseg, | |
1555 | int how) | |
0382b744 | 1556 | { |
cd841605 | 1557 | struct inode *inode = hdr->inode; |
0382b744 AA |
1558 | enum pnfs_try_status trypnfs; |
1559 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1560 | ||
cd841605 | 1561 | hdr->mds_ops = call_ops; |
0382b744 AA |
1562 | |
1563 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
1564 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
1565 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 1566 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1567 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1568 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1569 | return trypnfs; | |
1570 | } | |
1571 | ||
dce81290 | 1572 | static void |
7f714720 WAA |
1573 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
1574 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 1575 | { |
dce81290 TM |
1576 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1577 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1578 | enum pnfs_try_status trypnfs; |
dce81290 TM |
1579 | |
1580 | desc->pg_lseg = NULL; | |
d45f60c6 | 1581 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 1582 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1583 | pnfs_write_through_mds(desc, hdr); |
9369a431 | 1584 | pnfs_put_lseg(lseg); |
dce81290 TM |
1585 | } |
1586 | ||
6c75dc0d FI |
1587 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1588 | { | |
9369a431 | 1589 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1590 | nfs_pgio_header_free(hdr); |
6c75dc0d | 1591 | } |
89d77c8f | 1592 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1593 | |
dce81290 TM |
1594 | int |
1595 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1596 | { | |
6c75dc0d | 1597 | struct nfs_pgio_header *hdr; |
dce81290 TM |
1598 | int ret; |
1599 | ||
1e7f3a48 WAA |
1600 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1601 | if (!hdr) { | |
9b5415b5 | 1602 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
9369a431 | 1603 | pnfs_put_lseg(desc->pg_lseg); |
dce81290 | 1604 | desc->pg_lseg = NULL; |
6c75dc0d | 1605 | return -ENOMEM; |
dce81290 | 1606 | } |
6c75dc0d | 1607 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
9369a431 | 1608 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1609 | ret = nfs_generic_pgio(desc, hdr); |
6c75dc0d | 1610 | if (ret != 0) { |
9369a431 | 1611 | pnfs_put_lseg(desc->pg_lseg); |
6c75dc0d | 1612 | desc->pg_lseg = NULL; |
6c75dc0d | 1613 | } else |
7f714720 | 1614 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
6c75dc0d | 1615 | return ret; |
dce81290 TM |
1616 | } |
1617 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1618 | ||
53113ad3 | 1619 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
1620 | { |
1621 | struct nfs_pageio_descriptor pgio; | |
1622 | ||
1acbbb4e | 1623 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
1624 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
1625 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 1626 | } |
e7dd79af | 1627 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 1628 | |
d45f60c6 | 1629 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1630 | { |
cd841605 FI |
1631 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
1632 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1633 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1634 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1635 | } |
4db6e0b7 | 1636 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1637 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
1638 | } |
1639 | ||
d20581aa BH |
1640 | /* |
1641 | * Called by non rpc-based layout drivers | |
1642 | */ | |
d45f60c6 | 1643 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 1644 | { |
d45f60c6 | 1645 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
cd841605 | 1646 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
1647 | __nfs4_read_done_cb(hdr); |
1648 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
62e4a769 | 1649 | } else |
d45f60c6 WAA |
1650 | pnfs_ld_handle_read_error(hdr); |
1651 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
1652 | } |
1653 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1654 | ||
493292dd TM |
1655 | static void |
1656 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1657 | struct nfs_pgio_header *hdr) |
493292dd | 1658 | { |
4db6e0b7 FI |
1659 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1660 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1661 | nfs_pageio_reset_read_mds(desc); | |
1662 | desc->pg_recoalesce = 1; | |
1663 | } | |
d45f60c6 | 1664 | nfs_pgio_data_destroy(hdr); |
493292dd TM |
1665 | } |
1666 | ||
64419a9b AA |
1667 | /* |
1668 | * Call the appropriate parallel I/O subsystem read function. | |
1669 | */ | |
493292dd | 1670 | static enum pnfs_try_status |
d45f60c6 | 1671 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
1672 | const struct rpc_call_ops *call_ops, |
1673 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1674 | { |
cd841605 | 1675 | struct inode *inode = hdr->inode; |
64419a9b AA |
1676 | struct nfs_server *nfss = NFS_SERVER(inode); |
1677 | enum pnfs_try_status trypnfs; | |
1678 | ||
cd841605 | 1679 | hdr->mds_ops = call_ops; |
64419a9b AA |
1680 | |
1681 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 1682 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 1683 | |
d45f60c6 | 1684 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 1685 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1686 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1687 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1688 | return trypnfs; | |
1689 | } | |
863a3c6c | 1690 | |
493292dd | 1691 | static void |
7f714720 | 1692 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 1693 | { |
493292dd TM |
1694 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1695 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1696 | enum pnfs_try_status trypnfs; |
493292dd TM |
1697 | |
1698 | desc->pg_lseg = NULL; | |
d45f60c6 | 1699 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
7f714720 | 1700 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1701 | pnfs_read_through_mds(desc, hdr); |
9369a431 | 1702 | pnfs_put_lseg(lseg); |
493292dd TM |
1703 | } |
1704 | ||
4db6e0b7 FI |
1705 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1706 | { | |
9369a431 | 1707 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1708 | nfs_pgio_header_free(hdr); |
4db6e0b7 | 1709 | } |
89d77c8f | 1710 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1711 | |
493292dd TM |
1712 | int |
1713 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1714 | { | |
4db6e0b7 | 1715 | struct nfs_pgio_header *hdr; |
493292dd TM |
1716 | int ret; |
1717 | ||
1e7f3a48 WAA |
1718 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1719 | if (!hdr) { | |
061ae2ed | 1720 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
4db6e0b7 | 1721 | ret = -ENOMEM; |
9369a431 | 1722 | pnfs_put_lseg(desc->pg_lseg); |
493292dd TM |
1723 | desc->pg_lseg = NULL; |
1724 | return ret; | |
1725 | } | |
4db6e0b7 | 1726 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 1727 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1728 | ret = nfs_generic_pgio(desc, hdr); |
4db6e0b7 | 1729 | if (ret != 0) { |
9369a431 | 1730 | pnfs_put_lseg(desc->pg_lseg); |
4db6e0b7 | 1731 | desc->pg_lseg = NULL; |
4db6e0b7 | 1732 | } else |
7f714720 | 1733 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 1734 | return ret; |
493292dd TM |
1735 | } |
1736 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1737 | ||
71244d9b TM |
1738 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
1739 | { | |
1740 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
1741 | ||
1742 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 1743 | smp_mb__after_atomic(); |
71244d9b TM |
1744 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
1745 | } | |
1746 | ||
863a3c6c | 1747 | /* |
a9bae566 | 1748 | * There can be multiple RW segments. |
863a3c6c | 1749 | */ |
a9bae566 | 1750 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1751 | { |
a9bae566 | 1752 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1753 | |
a9bae566 PT |
1754 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1755 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 1756 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
1757 | list_add(&lseg->pls_lc_list, listp); |
1758 | } | |
863a3c6c AA |
1759 | } |
1760 | ||
a073dbff TM |
1761 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
1762 | { | |
1763 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
1764 | |
1765 | /* Matched by references in pnfs_set_layoutcommit */ | |
1766 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
1767 | list_del_init(&lseg->pls_lc_list); | |
1768 | pnfs_put_lseg(lseg); | |
1769 | } | |
1770 | ||
71244d9b | 1771 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
1772 | } |
1773 | ||
1b0ae068 PT |
1774 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
1775 | { | |
b9e028fd | 1776 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
1777 | } |
1778 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
1779 | ||
863a3c6c | 1780 | void |
d45f60c6 | 1781 | pnfs_set_layoutcommit(struct nfs_pgio_header *hdr) |
863a3c6c | 1782 | { |
cd841605 FI |
1783 | struct inode *inode = hdr->inode; |
1784 | struct nfs_inode *nfsi = NFS_I(inode); | |
d45f60c6 | 1785 | loff_t end_pos = hdr->mds_offset + hdr->res.count; |
79a48a1f | 1786 | bool mark_as_dirty = false; |
863a3c6c | 1787 | |
cd841605 | 1788 | spin_lock(&inode->i_lock); |
863a3c6c | 1789 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 1790 | mark_as_dirty = true; |
863a3c6c | 1791 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 1792 | __func__, inode->i_ino); |
863a3c6c | 1793 | } |
cd841605 | 1794 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 1795 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 1796 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 1797 | } |
acff5880 PT |
1798 | if (end_pos > nfsi->layout->plh_lwb) |
1799 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 1800 | spin_unlock(&inode->i_lock); |
acff5880 | 1801 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 1802 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
1803 | |
1804 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1805 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1806 | if (mark_as_dirty) | |
cd841605 | 1807 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
1808 | } |
1809 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
1810 | ||
378520b8 PT |
1811 | void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data) |
1812 | { | |
1813 | struct inode *inode = data->inode; | |
1814 | struct nfs_inode *nfsi = NFS_I(inode); | |
1815 | bool mark_as_dirty = false; | |
1816 | ||
1817 | spin_lock(&inode->i_lock); | |
1818 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { | |
1819 | mark_as_dirty = true; | |
1820 | dprintk("%s: Set layoutcommit for inode %lu ", | |
1821 | __func__, inode->i_ino); | |
1822 | } | |
1823 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) { | |
1824 | /* references matched in nfs4_layoutcommit_release */ | |
1825 | pnfs_get_lseg(data->lseg); | |
1826 | } | |
1827 | if (data->lwb > nfsi->layout->plh_lwb) | |
1828 | nfsi->layout->plh_lwb = data->lwb; | |
1829 | spin_unlock(&inode->i_lock); | |
1830 | dprintk("%s: lseg %p end_pos %llu\n", | |
1831 | __func__, data->lseg, nfsi->layout->plh_lwb); | |
1832 | ||
1833 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1834 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1835 | if (mark_as_dirty) | |
1836 | mark_inode_dirty_sync(inode); | |
1837 | } | |
1838 | EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit); | |
1839 | ||
db29c089 AA |
1840 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
1841 | { | |
1842 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
1843 | ||
1844 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
1845 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 1846 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
1847 | } |
1848 | ||
de4b15c7 AA |
1849 | /* |
1850 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
1851 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
1852 | * data to disk to allow the server to recover the data if it crashes. | |
1853 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
1854 | * is off, and a COMMIT is sent to a data server, or | |
1855 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
1856 | */ | |
863a3c6c | 1857 | int |
ef311537 | 1858 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 1859 | { |
5f919c9f | 1860 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
1861 | struct nfs4_layoutcommit_data *data; |
1862 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 1863 | loff_t end_pos; |
71244d9b | 1864 | int status; |
863a3c6c | 1865 | |
71244d9b | 1866 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
1867 | return 0; |
1868 | ||
71244d9b | 1869 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 1870 | |
71244d9b | 1871 | status = -EAGAIN; |
92407e75 | 1872 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
1873 | if (!sync) |
1874 | goto out; | |
74316201 | 1875 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
1876 | NFS_INO_LAYOUTCOMMITTING, |
1877 | nfs_wait_bit_killable, | |
1878 | TASK_KILLABLE); | |
92407e75 | 1879 | if (status) |
71244d9b | 1880 | goto out; |
92407e75 PT |
1881 | } |
1882 | ||
71244d9b TM |
1883 | status = -ENOMEM; |
1884 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
1885 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
1886 | if (!data) | |
1887 | goto clear_layoutcommitting; | |
1888 | ||
1889 | status = 0; | |
de4b15c7 | 1890 | spin_lock(&inode->i_lock); |
71244d9b TM |
1891 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1892 | goto out_unlock; | |
a9bae566 | 1893 | |
71244d9b | 1894 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 1895 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 1896 | |
acff5880 | 1897 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 1898 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 1899 | |
f597c537 | 1900 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
1901 | spin_unlock(&inode->i_lock); |
1902 | ||
1903 | data->args.inode = inode; | |
9fa40758 | 1904 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
1905 | nfs_fattr_init(&data->fattr); |
1906 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
1907 | data->res.fattr = &data->fattr; | |
1908 | data->args.lastbytewritten = end_pos - 1; | |
1909 | data->res.server = NFS_SERVER(inode); | |
1910 | ||
5f919c9f CH |
1911 | if (ld->prepare_layoutcommit) { |
1912 | status = ld->prepare_layoutcommit(&data->args); | |
1913 | if (status) { | |
1914 | spin_lock(&inode->i_lock); | |
1915 | if (end_pos < nfsi->layout->plh_lwb) | |
1916 | nfsi->layout->plh_lwb = end_pos; | |
1917 | spin_unlock(&inode->i_lock); | |
1918 | put_rpccred(data->cred); | |
1919 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); | |
1920 | goto clear_layoutcommitting; | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | ||
863a3c6c AA |
1925 | status = nfs4_proc_layoutcommit(data, sync); |
1926 | out: | |
92407e75 PT |
1927 | if (status) |
1928 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
1929 | dprintk("<-- %s status %d\n", __func__, status); |
1930 | return status; | |
71244d9b TM |
1931 | out_unlock: |
1932 | spin_unlock(&inode->i_lock); | |
92407e75 | 1933 | kfree(data); |
71244d9b TM |
1934 | clear_layoutcommitting: |
1935 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 1936 | goto out; |
863a3c6c | 1937 | } |
82be417a AA |
1938 | |
1939 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
1940 | { | |
1941 | struct nfs4_threshold *thp; | |
1942 | ||
1943 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
1944 | if (!thp) { | |
1945 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
1946 | return NULL; | |
1947 | } | |
1948 | return thp; | |
1949 | } |