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 | ||
aa1e0e3a PT |
53 | static int |
54 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid, | |
6c16605d | 55 | enum pnfs_iomode iomode, bool sync); |
aa1e0e3a | 56 | |
02c35fca FI |
57 | /* Return the registered pnfs layout driver module matching given id */ |
58 | static struct pnfs_layoutdriver_type * | |
59 | find_pnfs_driver_locked(u32 id) | |
60 | { | |
61 | struct pnfs_layoutdriver_type *local; | |
62 | ||
63 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
64 | if (local->id == id) | |
65 | goto out; | |
66 | local = NULL; | |
67 | out: | |
68 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
69 | return local; | |
70 | } | |
71 | ||
85e174ba RL |
72 | static struct pnfs_layoutdriver_type * |
73 | find_pnfs_driver(u32 id) | |
74 | { | |
02c35fca FI |
75 | struct pnfs_layoutdriver_type *local; |
76 | ||
77 | spin_lock(&pnfs_spinlock); | |
78 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
79 | if (local != NULL && !try_module_get(local->owner)) { |
80 | dprintk("%s: Could not grab reference on module\n", __func__); | |
81 | local = NULL; | |
82 | } | |
02c35fca FI |
83 | spin_unlock(&pnfs_spinlock); |
84 | return local; | |
85e174ba RL |
85 | } |
86 | ||
87 | void | |
88 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
89 | { | |
738fd0f3 BH |
90 | if (nfss->pnfs_curr_ld) { |
91 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
92 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
93 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
94 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
95 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 96 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 97 | } |
85e174ba RL |
98 | nfss->pnfs_curr_ld = NULL; |
99 | } | |
100 | ||
101 | /* | |
102 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
103 | * Currently only one pNFS layout driver per filesystem is supported. | |
104 | * | |
105 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
106 | */ | |
107 | void | |
738fd0f3 BH |
108 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
109 | u32 id) | |
85e174ba RL |
110 | { |
111 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
112 | ||
113 | if (id == 0) | |
114 | goto out_no_driver; | |
115 | if (!(server->nfs_client->cl_exchange_flags & | |
116 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
a030889a WAA |
117 | printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n", |
118 | __func__, id, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
119 | goto out_no_driver; |
120 | } | |
121 | ld_type = find_pnfs_driver(id); | |
122 | if (!ld_type) { | |
123 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
124 | ld_type = find_pnfs_driver(id); | |
125 | if (!ld_type) { | |
126 | dprintk("%s: No pNFS module found for %u.\n", | |
127 | __func__, id); | |
128 | goto out_no_driver; | |
129 | } | |
130 | } | |
131 | server->pnfs_curr_ld = ld_type; | |
738fd0f3 BH |
132 | if (ld_type->set_layoutdriver |
133 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
134 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
135 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
136 | module_put(ld_type->owner); |
137 | goto out_no_driver; | |
138 | } | |
2a4c8994 TM |
139 | /* Bump the MDS count */ |
140 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 141 | |
85e174ba RL |
142 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
143 | return; | |
144 | ||
145 | out_no_driver: | |
146 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
147 | server->pnfs_curr_ld = NULL; | |
148 | } | |
02c35fca FI |
149 | |
150 | int | |
151 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
152 | { | |
153 | int status = -EINVAL; | |
154 | struct pnfs_layoutdriver_type *tmp; | |
155 | ||
156 | if (ld_type->id == 0) { | |
a030889a | 157 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
158 | return status; |
159 | } | |
b1f69b75 | 160 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 161 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
162 | "alloc_lseg and free_lseg.\n", __func__); |
163 | return status; | |
164 | } | |
02c35fca FI |
165 | |
166 | spin_lock(&pnfs_spinlock); | |
167 | tmp = find_pnfs_driver_locked(ld_type->id); | |
168 | if (!tmp) { | |
169 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
170 | status = 0; | |
171 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
172 | ld_type->name); | |
173 | } else { | |
a030889a | 174 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
175 | __func__, ld_type->id); |
176 | } | |
177 | spin_unlock(&pnfs_spinlock); | |
178 | ||
179 | return status; | |
180 | } | |
181 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
182 | ||
183 | void | |
184 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
185 | { | |
186 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
187 | spin_lock(&pnfs_spinlock); | |
188 | list_del(&ld_type->pnfs_tblid); | |
189 | spin_unlock(&pnfs_spinlock); | |
190 | } | |
191 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 192 | |
b1f69b75 AA |
193 | /* |
194 | * pNFS client layout cache | |
195 | */ | |
196 | ||
cc6e5340 | 197 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 198 | void |
70c3bd2b | 199 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 200 | { |
cc6e5340 | 201 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
202 | } |
203 | ||
636fb9c8 BH |
204 | static struct pnfs_layout_hdr * |
205 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
206 | { | |
207 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
57934278 | 208 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
209 | } |
210 | ||
211 | static void | |
212 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
213 | { | |
9c626381 TM |
214 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
215 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
216 | ||
217 | if (!list_empty(&lo->plh_layouts)) { | |
218 | struct nfs_client *clp = server->nfs_client; | |
219 | ||
220 | spin_lock(&clp->cl_lock); | |
221 | list_del_init(&lo->plh_layouts); | |
222 | spin_unlock(&clp->cl_lock); | |
223 | } | |
9fa40758 | 224 | put_rpccred(lo->plh_lc_cred); |
57934278 | 225 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
226 | } |
227 | ||
e5e94017 | 228 | static void |
6622c3ea | 229 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 230 | { |
bb346f63 | 231 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 232 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
233 | nfsi->layout = NULL; |
234 | /* Reset MDS Threshold I/O counters */ | |
235 | nfsi->write_io = 0; | |
236 | nfsi->read_io = 0; | |
e5e94017 BH |
237 | } |
238 | ||
b1f69b75 | 239 | void |
70c3bd2b | 240 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 241 | { |
cc6e5340 FI |
242 | struct inode *inode = lo->plh_inode; |
243 | ||
244 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
566f8737 PT |
245 | if (!list_empty(&lo->plh_segs)) |
246 | WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n"); | |
6622c3ea | 247 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 248 | spin_unlock(&inode->i_lock); |
6622c3ea | 249 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 250 | } |
974cec8c AA |
251 | } |
252 | ||
b9e028fd TM |
253 | static int |
254 | pnfs_iomode_to_fail_bit(u32 iomode) | |
255 | { | |
256 | return iomode == IOMODE_RW ? | |
257 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
258 | } | |
259 | ||
260 | static void | |
3e621214 | 261 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 262 | { |
25c75333 | 263 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 264 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
3e621214 TM |
265 | atomic_inc(&lo->plh_refcount); |
266 | } | |
267 | ||
268 | static void | |
269 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
270 | { | |
271 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
272 | atomic_dec(&lo->plh_refcount); | |
273 | } | |
274 | ||
275 | static void | |
276 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
277 | { | |
278 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
279 | struct pnfs_layout_range range = { |
280 | .iomode = iomode, | |
281 | .offset = 0, | |
282 | .length = NFS4_MAX_UINT64, | |
283 | }; | |
284 | LIST_HEAD(head); | |
3e621214 TM |
285 | |
286 | spin_lock(&inode->i_lock); | |
287 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
115ce575 | 288 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range); |
3e621214 | 289 | spin_unlock(&inode->i_lock); |
115ce575 | 290 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
291 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
292 | iomode == IOMODE_RW ? "RW" : "READ"); | |
293 | } | |
294 | ||
295 | static bool | |
296 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
297 | { | |
25c75333 | 298 | unsigned long start, end; |
3e621214 TM |
299 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
300 | ||
301 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
302 | return false; |
303 | end = jiffies; | |
304 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
305 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
306 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 307 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
308 | return false; |
309 | } | |
310 | return true; | |
b9e028fd TM |
311 | } |
312 | ||
974cec8c AA |
313 | static void |
314 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
315 | { | |
566052c5 | 316 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 317 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c FI |
318 | atomic_set(&lseg->pls_refcount, 1); |
319 | smp_mb(); | |
320 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 321 | lseg->pls_layout = lo; |
974cec8c AA |
322 | } |
323 | ||
905ca191 | 324 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 325 | { |
b7edfaa1 | 326 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 327 | |
b1f69b75 | 328 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
974cec8c AA |
329 | } |
330 | ||
d684d2ae | 331 | static void |
57036a37 TM |
332 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
333 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 334 | { |
57036a37 | 335 | struct inode *inode = lo->plh_inode; |
d684d2ae | 336 | |
d20581aa | 337 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 338 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
339 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
340 | atomic_dec(&lo->plh_refcount); | |
173f77e9 TM |
341 | if (list_empty(&lo->plh_segs)) |
342 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
d684d2ae FI |
343 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); |
344 | } | |
345 | ||
aa1e0e3a PT |
346 | /* Return true if layoutreturn is needed */ |
347 | static bool | |
348 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo, | |
27b6f539 | 349 | struct pnfs_layout_segment *lseg) |
aa1e0e3a PT |
350 | { |
351 | struct pnfs_layout_segment *s; | |
352 | ||
353 | if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
354 | return false; | |
355 | ||
356 | list_for_each_entry(s, &lo->plh_segs, pls_list) | |
27b6f539 | 357 | if (s != lseg && test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags)) |
aa1e0e3a PT |
358 | return false; |
359 | ||
aa1e0e3a PT |
360 | return true; |
361 | } | |
362 | ||
27b6f539 PT |
363 | static void pnfs_layoutreturn_free_lseg(struct work_struct *work) |
364 | { | |
365 | struct pnfs_layout_segment *lseg; | |
366 | struct pnfs_layout_hdr *lo; | |
367 | struct inode *inode; | |
368 | ||
369 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
370 | WARN_ON(atomic_read(&lseg->pls_refcount)); | |
371 | lo = lseg->pls_layout; | |
372 | inode = lo->plh_inode; | |
373 | ||
374 | spin_lock(&inode->i_lock); | |
375 | if (pnfs_layout_need_return(lo, lseg)) { | |
376 | nfs4_stateid stateid; | |
377 | enum pnfs_iomode iomode; | |
378 | ||
379 | stateid = lo->plh_stateid; | |
380 | iomode = lo->plh_return_iomode; | |
381 | /* decreased in pnfs_send_layoutreturn() */ | |
382 | lo->plh_block_lgets++; | |
383 | lo->plh_return_iomode = 0; | |
384 | spin_unlock(&inode->i_lock); | |
385 | ||
386 | pnfs_send_layoutreturn(lo, stateid, iomode, true); | |
387 | spin_lock(&inode->i_lock); | |
388 | } else | |
389 | /* match pnfs_get_layout_hdr #2 in pnfs_put_lseg */ | |
390 | pnfs_put_layout_hdr(lo); | |
391 | pnfs_layout_remove_lseg(lo, lseg); | |
392 | spin_unlock(&inode->i_lock); | |
393 | pnfs_free_lseg(lseg); | |
394 | /* match pnfs_get_layout_hdr #1 in pnfs_put_lseg */ | |
395 | pnfs_put_layout_hdr(lo); | |
396 | } | |
397 | ||
398 | static void | |
399 | pnfs_layoutreturn_free_lseg_async(struct pnfs_layout_segment *lseg) | |
400 | { | |
401 | INIT_WORK(&lseg->pls_work, pnfs_layoutreturn_free_lseg); | |
402 | queue_work(nfsiod_workqueue, &lseg->pls_work); | |
403 | } | |
404 | ||
bae724ef | 405 | void |
9369a431 | 406 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 407 | { |
57036a37 | 408 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
409 | struct inode *inode; |
410 | ||
411 | if (!lseg) | |
412 | return; | |
413 | ||
4541d16c FI |
414 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
415 | atomic_read(&lseg->pls_refcount), | |
416 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
57036a37 TM |
417 | lo = lseg->pls_layout; |
418 | inode = lo->plh_inode; | |
d684d2ae | 419 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
8f0d27dc | 420 | pnfs_get_layout_hdr(lo); |
27b6f539 PT |
421 | if (pnfs_layout_need_return(lo, lseg)) { |
422 | spin_unlock(&inode->i_lock); | |
423 | /* hdr reference dropped in nfs4_layoutreturn_release */ | |
424 | pnfs_get_layout_hdr(lo); | |
425 | pnfs_layoutreturn_free_lseg_async(lseg); | |
426 | } else { | |
427 | pnfs_layout_remove_lseg(lo, lseg); | |
428 | spin_unlock(&inode->i_lock); | |
429 | pnfs_free_lseg(lseg); | |
aa1e0e3a | 430 | pnfs_put_layout_hdr(lo); |
27b6f539 | 431 | } |
4541d16c | 432 | } |
4541d16c | 433 | } |
9369a431 | 434 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 435 | |
6543f803 | 436 | static void pnfs_free_lseg_async_work(struct work_struct *work) |
e6cf82d1 WAA |
437 | { |
438 | struct pnfs_layout_segment *lseg; | |
6543f803 | 439 | struct pnfs_layout_hdr *lo; |
e6cf82d1 WAA |
440 | |
441 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
6543f803 | 442 | lo = lseg->pls_layout; |
e6cf82d1 | 443 | |
6543f803 TM |
444 | pnfs_free_lseg(lseg); |
445 | pnfs_put_layout_hdr(lo); | |
e6cf82d1 WAA |
446 | } |
447 | ||
6543f803 | 448 | static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg) |
e6cf82d1 | 449 | { |
6543f803 | 450 | INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work); |
e6cf82d1 WAA |
451 | schedule_work(&lseg->pls_work); |
452 | } | |
6543f803 TM |
453 | |
454 | void | |
455 | pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg) | |
456 | { | |
457 | if (!lseg) | |
458 | return; | |
459 | ||
460 | assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock); | |
461 | ||
462 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, | |
463 | atomic_read(&lseg->pls_refcount), | |
464 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
465 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
466 | struct pnfs_layout_hdr *lo = lseg->pls_layout; | |
467 | pnfs_get_layout_hdr(lo); | |
468 | pnfs_layout_remove_lseg(lo, lseg); | |
469 | pnfs_free_lseg_async(lseg); | |
470 | } | |
471 | } | |
472 | EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked); | |
e6cf82d1 | 473 | |
3cb2df17 | 474 | static u64 |
fb3296eb BH |
475 | end_offset(u64 start, u64 len) |
476 | { | |
477 | u64 end; | |
478 | ||
479 | end = start + len; | |
480 | return end >= start ? end : NFS4_MAX_UINT64; | |
481 | } | |
482 | ||
fb3296eb BH |
483 | /* |
484 | * is l2 fully contained in l1? | |
485 | * start1 end1 | |
486 | * [----------------------------------) | |
487 | * start2 end2 | |
488 | * [----------------) | |
489 | */ | |
3cb2df17 | 490 | static bool |
7dc0ac70 | 491 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 492 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
493 | { |
494 | u64 start1 = l1->offset; | |
495 | u64 end1 = end_offset(start1, l1->length); | |
496 | u64 start2 = l2->offset; | |
497 | u64 end2 = end_offset(start2, l2->length); | |
498 | ||
499 | return (start1 <= start2) && (end1 >= end2); | |
500 | } | |
501 | ||
502 | /* | |
503 | * is l1 and l2 intersecting? | |
504 | * start1 end1 | |
505 | * [----------------------------------) | |
506 | * start2 end2 | |
507 | * [----------------) | |
508 | */ | |
3cb2df17 | 509 | static bool |
7dc0ac70 | 510 | pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1, |
3cb2df17 | 511 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
512 | { |
513 | u64 start1 = l1->offset; | |
514 | u64 end1 = end_offset(start1, l1->length); | |
515 | u64 start2 = l2->offset; | |
516 | u64 end2 = end_offset(start2, l2->length); | |
517 | ||
518 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
519 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
520 | } | |
521 | ||
4541d16c | 522 | static bool |
3cb2df17 TM |
523 | should_free_lseg(const struct pnfs_layout_range *lseg_range, |
524 | const struct pnfs_layout_range *recall_range) | |
4541d16c | 525 | { |
778b5502 BH |
526 | return (recall_range->iomode == IOMODE_ANY || |
527 | lseg_range->iomode == recall_range->iomode) && | |
7dc0ac70 | 528 | pnfs_lseg_range_intersecting(lseg_range, recall_range); |
974cec8c AA |
529 | } |
530 | ||
24956804 TM |
531 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
532 | struct list_head *tmp_list) | |
533 | { | |
534 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
535 | return false; | |
536 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
537 | list_add(&lseg->pls_list, tmp_list); | |
538 | return true; | |
539 | } | |
540 | ||
4541d16c FI |
541 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
542 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
543 | struct list_head *tmp_list) | |
544 | { | |
545 | int rv = 0; | |
546 | ||
547 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
548 | /* Remove the reference keeping the lseg in the | |
549 | * list. It will now be removed when all | |
550 | * outstanding io is finished. | |
551 | */ | |
d684d2ae FI |
552 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
553 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 554 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 555 | rv = 1; |
4541d16c FI |
556 | } |
557 | return rv; | |
558 | } | |
559 | ||
560 | /* Returns count of number of matching invalid lsegs remaining in list | |
561 | * after call. | |
562 | */ | |
43f1b3da | 563 | int |
49a85061 | 564 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 565 | struct list_head *tmp_list, |
778b5502 | 566 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
567 | { |
568 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 569 | int invalid = 0, removed = 0; |
974cec8c AA |
570 | |
571 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
572 | ||
8006bfba | 573 | if (list_empty(&lo->plh_segs)) |
38511722 | 574 | return 0; |
4541d16c | 575 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
576 | if (!recall_range || |
577 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
578 | dprintk("%s: freeing lseg %p iomode %d " |
579 | "offset %llu length %llu\n", __func__, | |
580 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
581 | lseg->pls_range.length); | |
582 | invalid++; | |
583 | removed += mark_lseg_invalid(lseg, tmp_list); | |
584 | } | |
585 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
586 | return invalid - removed; | |
974cec8c AA |
587 | } |
588 | ||
f49f9baa | 589 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 590 | void |
4541d16c | 591 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 592 | { |
4541d16c | 593 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
594 | |
595 | if (list_empty(free_me)) | |
596 | return; | |
597 | ||
4541d16c | 598 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 599 | list_del(&lseg->pls_list); |
905ca191 | 600 | pnfs_free_lseg(lseg); |
974cec8c AA |
601 | } |
602 | } | |
603 | ||
e5e94017 BH |
604 | void |
605 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
606 | { | |
607 | struct pnfs_layout_hdr *lo; | |
974cec8c | 608 | LIST_HEAD(tmp_list); |
e5e94017 BH |
609 | |
610 | spin_lock(&nfsi->vfs_inode.i_lock); | |
611 | lo = nfsi->layout; | |
612 | if (lo) { | |
38511722 | 613 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 614 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
3e621214 TM |
615 | pnfs_get_layout_hdr(lo); |
616 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); | |
617 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
618 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
619 | pnfs_free_lseg_list(&tmp_list); | |
620 | pnfs_put_layout_hdr(lo); | |
621 | } else | |
622 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 623 | } |
041245c8 | 624 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 625 | |
fd9a8d71 TM |
626 | static bool |
627 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
628 | struct list_head *layout_list) | |
974cec8c AA |
629 | { |
630 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 631 | bool ret = false; |
974cec8c | 632 | |
fd9a8d71 TM |
633 | spin_lock(&inode->i_lock); |
634 | lo = NFS_I(inode)->layout; | |
635 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
636 | pnfs_get_layout_hdr(lo); | |
637 | list_add(&lo->plh_bulk_destroy, layout_list); | |
638 | ret = true; | |
639 | } | |
640 | spin_unlock(&inode->i_lock); | |
641 | return ret; | |
642 | } | |
643 | ||
644 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
645 | static int | |
646 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
647 | struct nfs_server *server, | |
648 | struct list_head *layout_list) | |
649 | { | |
650 | struct pnfs_layout_hdr *lo, *next; | |
651 | struct inode *inode; | |
652 | ||
653 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
654 | inode = igrab(lo->plh_inode); | |
655 | if (inode == NULL) | |
656 | continue; | |
657 | list_del_init(&lo->plh_layouts); | |
658 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
659 | continue; | |
660 | rcu_read_unlock(); | |
661 | spin_unlock(&clp->cl_lock); | |
662 | iput(inode); | |
663 | spin_lock(&clp->cl_lock); | |
664 | rcu_read_lock(); | |
665 | return -EAGAIN; | |
666 | } | |
667 | return 0; | |
668 | } | |
669 | ||
670 | static int | |
671 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
672 | bool is_bulk_recall) | |
673 | { | |
674 | struct pnfs_layout_hdr *lo; | |
675 | struct inode *inode; | |
676 | struct pnfs_layout_range range = { | |
677 | .iomode = IOMODE_ANY, | |
678 | .offset = 0, | |
679 | .length = NFS4_MAX_UINT64, | |
680 | }; | |
681 | LIST_HEAD(lseg_list); | |
682 | int ret = 0; | |
683 | ||
684 | while (!list_empty(layout_list)) { | |
685 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
686 | plh_bulk_destroy); | |
687 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
688 | lo->plh_inode->i_ino); | |
689 | inode = lo->plh_inode; | |
7c5d1875 CH |
690 | |
691 | pnfs_layoutcommit_inode(inode, false); | |
692 | ||
fd9a8d71 TM |
693 | spin_lock(&inode->i_lock); |
694 | list_del_init(&lo->plh_bulk_destroy); | |
695 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ | |
696 | if (is_bulk_recall) | |
697 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
698 | if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range)) | |
699 | ret = -EAGAIN; | |
700 | spin_unlock(&inode->i_lock); | |
701 | pnfs_free_lseg_list(&lseg_list); | |
702 | pnfs_put_layout_hdr(lo); | |
703 | iput(inode); | |
704 | } | |
705 | return ret; | |
706 | } | |
707 | ||
708 | int | |
709 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
710 | struct nfs_fsid *fsid, | |
711 | bool is_recall) | |
712 | { | |
713 | struct nfs_server *server; | |
714 | LIST_HEAD(layout_list); | |
c47abcf8 | 715 | |
974cec8c | 716 | spin_lock(&clp->cl_lock); |
6382a441 | 717 | rcu_read_lock(); |
fd9a8d71 | 718 | restart: |
6382a441 | 719 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
720 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
721 | continue; | |
722 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
723 | server, | |
724 | &layout_list) != 0) | |
725 | goto restart; | |
6382a441 WAA |
726 | } |
727 | rcu_read_unlock(); | |
974cec8c AA |
728 | spin_unlock(&clp->cl_lock); |
729 | ||
fd9a8d71 TM |
730 | if (list_empty(&layout_list)) |
731 | return 0; | |
732 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
733 | } | |
734 | ||
735 | int | |
736 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
737 | bool is_recall) | |
738 | { | |
739 | struct nfs_server *server; | |
740 | LIST_HEAD(layout_list); | |
741 | ||
742 | spin_lock(&clp->cl_lock); | |
743 | rcu_read_lock(); | |
744 | restart: | |
745 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
746 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
747 | server, | |
748 | &layout_list) != 0) | |
749 | goto restart; | |
974cec8c | 750 | } |
fd9a8d71 TM |
751 | rcu_read_unlock(); |
752 | spin_unlock(&clp->cl_lock); | |
753 | ||
754 | if (list_empty(&layout_list)) | |
755 | return 0; | |
756 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
757 | } | |
758 | ||
759 | /* | |
760 | * Called by the state manger to remove all layouts established under an | |
761 | * expired lease. | |
762 | */ | |
763 | void | |
764 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
765 | { | |
766 | nfs4_deviceid_mark_client_invalid(clp); | |
767 | nfs4_deviceid_purge_client(clp); | |
768 | ||
769 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
770 | } |
771 | ||
5a65503f TM |
772 | /* |
773 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
774 | * taking into account wraparound issues. | |
775 | */ | |
776 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
777 | { | |
2c64c57d | 778 | return (s32)(s1 - s2) > 0; |
5a65503f TM |
779 | } |
780 | ||
fd6002e9 | 781 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
782 | void |
783 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
784 | bool update_barrier) | |
b1f69b75 | 785 | { |
22aaf714 TM |
786 | u32 oldseq, newseq, new_barrier; |
787 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 788 | |
2d2f24ad TM |
789 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
790 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 791 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 792 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 793 | if (update_barrier) { |
22aaf714 | 794 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
795 | } else { |
796 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 797 | * "close" to the current seqids. |
43f1b3da | 798 | */ |
22aaf714 | 799 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 800 | } |
22aaf714 TM |
801 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
802 | lo->plh_barrier = new_barrier; | |
43f1b3da | 803 | } |
b1f69b75 AA |
804 | } |
805 | ||
cf7d63f1 | 806 | static bool |
19c54aba TM |
807 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
808 | const nfs4_stateid *stateid) | |
43f1b3da | 809 | { |
19c54aba | 810 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 811 | |
19c54aba TM |
812 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
813 | } | |
814 | ||
ce6ab4f2 PT |
815 | static bool |
816 | pnfs_layout_returning(const struct pnfs_layout_hdr *lo, | |
817 | struct pnfs_layout_range *range) | |
818 | { | |
819 | return test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) && | |
820 | (lo->plh_return_iomode == IOMODE_ANY || | |
821 | lo->plh_return_iomode == range->iomode); | |
822 | } | |
823 | ||
19c54aba TM |
824 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
825 | static bool | |
ce6ab4f2 PT |
826 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, |
827 | struct pnfs_layout_range *range, int lget) | |
19c54aba | 828 | { |
f7e8917a FI |
829 | return lo->plh_block_lgets || |
830 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || | |
43f1b3da | 831 | (list_empty(&lo->plh_segs) && |
ce6ab4f2 PT |
832 | (atomic_read(&lo->plh_outstanding) > lget)) || |
833 | pnfs_layout_returning(lo, range); | |
cf7d63f1 FI |
834 | } |
835 | ||
fd6002e9 FI |
836 | int |
837 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
ce6ab4f2 | 838 | struct pnfs_layout_range *range, |
fd6002e9 | 839 | struct nfs4_state *open_state) |
b1f69b75 | 840 | { |
fd6002e9 | 841 | int status = 0; |
974cec8c | 842 | |
b1f69b75 | 843 | dprintk("--> %s\n", __func__); |
fd6002e9 | 844 | spin_lock(&lo->plh_inode->i_lock); |
ce6ab4f2 | 845 | if (pnfs_layoutgets_blocked(lo, range, 1)) { |
cf7d63f1 | 846 | status = -EAGAIN; |
5d422301 TM |
847 | } else if (!nfs4_valid_open_stateid(open_state)) { |
848 | status = -EBADF; | |
47abadef CH |
849 | } else if (list_empty(&lo->plh_segs) || |
850 | test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { | |
fd6002e9 FI |
851 | int seq; |
852 | ||
853 | do { | |
854 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 855 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
856 | } while (read_seqretry(&open_state->seqlock, seq)); |
857 | } else | |
f597c537 | 858 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 859 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 860 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 861 | return status; |
b1f69b75 AA |
862 | } |
863 | ||
864 | /* | |
865 | * Get layout from server. | |
866 | * for now, assume that whole file layouts are requested. | |
867 | * arg->offset: 0 | |
868 | * arg->length: all ones | |
869 | */ | |
e5e94017 BH |
870 | static struct pnfs_layout_segment * |
871 | send_layoutget(struct pnfs_layout_hdr *lo, | |
872 | struct nfs_open_context *ctx, | |
fb3296eb | 873 | struct pnfs_layout_range *range, |
a75b9df9 | 874 | gfp_t gfp_flags) |
e5e94017 | 875 | { |
b7edfaa1 | 876 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
877 | struct nfs_server *server = NFS_SERVER(ino); |
878 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 879 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
880 | |
881 | dprintk("--> %s\n", __func__); | |
e5e94017 | 882 | |
a75b9df9 | 883 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 884 | if (lgp == NULL) |
b1f69b75 | 885 | return NULL; |
35124a09 | 886 | |
fb3296eb BH |
887 | lgp->args.minlength = PAGE_CACHE_SIZE; |
888 | if (lgp->args.minlength > range->length) | |
889 | lgp->args.minlength = range->length; | |
b1f69b75 | 890 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 891 | lgp->args.range = *range; |
b1f69b75 AA |
892 | lgp->args.type = server->pnfs_curr_ld->id; |
893 | lgp->args.inode = ino; | |
894 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 895 | lgp->gfp_flags = gfp_flags; |
6ab59344 | 896 | lgp->cred = lo->plh_lc_cred; |
b1f69b75 AA |
897 | |
898 | /* Synchronously retrieve layout information from server and | |
899 | * store in lseg. | |
900 | */ | |
a0b0a6e3 TM |
901 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
902 | if (IS_ERR(lseg)) { | |
903 | switch (PTR_ERR(lseg)) { | |
904 | case -ENOMEM: | |
905 | case -ERESTARTSYS: | |
906 | break; | |
907 | default: | |
908 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 909 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
910 | } |
911 | return NULL; | |
974cec8c | 912 | } |
35124a09 | 913 | |
974cec8c AA |
914 | return lseg; |
915 | } | |
916 | ||
24956804 TM |
917 | static void pnfs_clear_layoutcommit(struct inode *inode, |
918 | struct list_head *head) | |
919 | { | |
920 | struct nfs_inode *nfsi = NFS_I(inode); | |
921 | struct pnfs_layout_segment *lseg, *tmp; | |
922 | ||
923 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
924 | return; | |
925 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
926 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
927 | continue; | |
928 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
929 | } | |
930 | } | |
931 | ||
f40eb5d0 PT |
932 | static int |
933 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid, | |
6c16605d | 934 | enum pnfs_iomode iomode, bool sync) |
f40eb5d0 PT |
935 | { |
936 | struct inode *ino = lo->plh_inode; | |
937 | struct nfs4_layoutreturn *lrp; | |
938 | int status = 0; | |
939 | ||
940 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); | |
941 | if (unlikely(lrp == NULL)) { | |
942 | status = -ENOMEM; | |
943 | spin_lock(&ino->i_lock); | |
944 | lo->plh_block_lgets--; | |
193e3aa2 | 945 | rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq); |
f40eb5d0 PT |
946 | spin_unlock(&ino->i_lock); |
947 | pnfs_put_layout_hdr(lo); | |
948 | goto out; | |
949 | } | |
950 | ||
951 | lrp->args.stateid = stateid; | |
952 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
953 | lrp->args.inode = ino; | |
15eb67c1 PT |
954 | lrp->args.range.iomode = iomode; |
955 | lrp->args.range.offset = 0; | |
956 | lrp->args.range.length = NFS4_MAX_UINT64; | |
f40eb5d0 PT |
957 | lrp->args.layout = lo; |
958 | lrp->clp = NFS_SERVER(ino)->nfs_client; | |
959 | lrp->cred = lo->plh_lc_cred; | |
960 | ||
6c16605d | 961 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
962 | out: |
963 | dprintk("<-- %s status: %d\n", __func__, status); | |
964 | return status; | |
965 | } | |
966 | ||
293b3b06 AA |
967 | /* |
968 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
969 | * when the layout segment list is empty. | |
970 | * | |
971 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
972 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
973 | * deviceid is marked invalid. | |
974 | */ | |
cbe82603 BH |
975 | int |
976 | _pnfs_return_layout(struct inode *ino) | |
977 | { | |
978 | struct pnfs_layout_hdr *lo = NULL; | |
979 | struct nfs_inode *nfsi = NFS_I(ino); | |
980 | LIST_HEAD(tmp_list); | |
cbe82603 | 981 | nfs4_stateid stateid; |
293b3b06 | 982 | int status = 0, empty; |
cbe82603 | 983 | |
366d5052 | 984 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
985 | |
986 | spin_lock(&ino->i_lock); | |
987 | lo = nfsi->layout; | |
e5929f3c | 988 | if (!lo) { |
cbe82603 | 989 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
990 | dprintk("NFS: %s no layout to return\n", __func__); |
991 | goto out; | |
cbe82603 BH |
992 | } |
993 | stateid = nfsi->layout->plh_stateid; | |
994 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 995 | pnfs_get_layout_hdr(lo); |
293b3b06 | 996 | empty = list_empty(&lo->plh_segs); |
24956804 | 997 | pnfs_clear_layoutcommit(ino, &tmp_list); |
49a85061 | 998 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
c88953d8 CH |
999 | |
1000 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
1001 | struct pnfs_layout_range range = { | |
1002 | .iomode = IOMODE_ANY, | |
1003 | .offset = 0, | |
1004 | .length = NFS4_MAX_UINT64, | |
1005 | }; | |
1006 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
1007 | } | |
1008 | ||
293b3b06 AA |
1009 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
1010 | if (empty) { | |
1011 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 1012 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
1013 | dprintk("NFS: %s no layout segments to return\n", __func__); |
1014 | goto out; | |
1015 | } | |
47abadef CH |
1016 | |
1017 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
ea0ded74 | 1018 | lo->plh_block_lgets++; |
cbe82603 BH |
1019 | spin_unlock(&ino->i_lock); |
1020 | pnfs_free_lseg_list(&tmp_list); | |
1021 | ||
6c16605d | 1022 | status = pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true); |
cbe82603 BH |
1023 | out: |
1024 | dprintk("<-- %s status: %d\n", __func__, status); | |
1025 | return status; | |
1026 | } | |
0a57cdac | 1027 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 1028 | |
24028672 TM |
1029 | int |
1030 | pnfs_commit_and_return_layout(struct inode *inode) | |
1031 | { | |
1032 | struct pnfs_layout_hdr *lo; | |
1033 | int ret; | |
1034 | ||
1035 | spin_lock(&inode->i_lock); | |
1036 | lo = NFS_I(inode)->layout; | |
1037 | if (lo == NULL) { | |
1038 | spin_unlock(&inode->i_lock); | |
1039 | return 0; | |
1040 | } | |
1041 | pnfs_get_layout_hdr(lo); | |
1042 | /* Block new layoutgets and read/write to ds */ | |
1043 | lo->plh_block_lgets++; | |
1044 | spin_unlock(&inode->i_lock); | |
1045 | filemap_fdatawait(inode->i_mapping); | |
1046 | ret = pnfs_layoutcommit_inode(inode, true); | |
1047 | if (ret == 0) | |
1048 | ret = _pnfs_return_layout(inode); | |
1049 | spin_lock(&inode->i_lock); | |
1050 | lo->plh_block_lgets--; | |
1051 | spin_unlock(&inode->i_lock); | |
1052 | pnfs_put_layout_hdr(lo); | |
1053 | return ret; | |
1054 | } | |
1055 | ||
f7e8917a FI |
1056 | bool pnfs_roc(struct inode *ino) |
1057 | { | |
1058 | struct pnfs_layout_hdr *lo; | |
1059 | struct pnfs_layout_segment *lseg, *tmp; | |
193e3aa2 | 1060 | nfs4_stateid stateid; |
f7e8917a | 1061 | LIST_HEAD(tmp_list); |
193e3aa2 | 1062 | bool found = false, layoutreturn = false; |
f7e8917a FI |
1063 | |
1064 | spin_lock(&ino->i_lock); | |
1065 | lo = NFS_I(ino)->layout; | |
1066 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
1067 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
1068 | goto out_nolayout; | |
1069 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
1070 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
1071 | mark_lseg_invalid(lseg, &tmp_list); | |
1072 | found = true; | |
1073 | } | |
1074 | if (!found) | |
1075 | goto out_nolayout; | |
1076 | lo->plh_block_lgets++; | |
70c3bd2b | 1077 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
1078 | spin_unlock(&ino->i_lock); |
1079 | pnfs_free_lseg_list(&tmp_list); | |
1080 | return true; | |
1081 | ||
1082 | out_nolayout: | |
193e3aa2 PT |
1083 | if (lo) { |
1084 | stateid = lo->plh_stateid; | |
1085 | layoutreturn = | |
1086 | test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, | |
1087 | &lo->plh_flags); | |
1088 | if (layoutreturn) { | |
1089 | lo->plh_block_lgets++; | |
1090 | pnfs_get_layout_hdr(lo); | |
1091 | } | |
1092 | } | |
f7e8917a | 1093 | spin_unlock(&ino->i_lock); |
193e3aa2 | 1094 | if (layoutreturn) |
27b6f539 | 1095 | pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true); |
f7e8917a FI |
1096 | return false; |
1097 | } | |
1098 | ||
1099 | void pnfs_roc_release(struct inode *ino) | |
1100 | { | |
1101 | struct pnfs_layout_hdr *lo; | |
1102 | ||
1103 | spin_lock(&ino->i_lock); | |
1104 | lo = NFS_I(ino)->layout; | |
1105 | lo->plh_block_lgets--; | |
6622c3ea TM |
1106 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
1107 | pnfs_detach_layout_hdr(lo); | |
1108 | spin_unlock(&ino->i_lock); | |
1109 | pnfs_free_layout_hdr(lo); | |
1110 | } else | |
1111 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
1112 | } |
1113 | ||
1114 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
1115 | { | |
1116 | struct pnfs_layout_hdr *lo; | |
1117 | ||
1118 | spin_lock(&ino->i_lock); | |
1119 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 1120 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
1121 | lo->plh_barrier = barrier; |
1122 | spin_unlock(&ino->i_lock); | |
1123 | } | |
1124 | ||
7fdab069 | 1125 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
1126 | { |
1127 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 1128 | struct pnfs_layout_hdr *lo; |
f7e8917a | 1129 | struct pnfs_layout_segment *lseg; |
193e3aa2 | 1130 | nfs4_stateid stateid; |
7fdab069 | 1131 | u32 current_seqid; |
193e3aa2 | 1132 | bool found = false, layoutreturn = false; |
f7e8917a FI |
1133 | |
1134 | spin_lock(&ino->i_lock); | |
1135 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
1136 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 1137 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 1138 | found = true; |
7fdab069 | 1139 | goto out; |
f7e8917a | 1140 | } |
7fdab069 TM |
1141 | lo = nfsi->layout; |
1142 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1143 | |
7fdab069 TM |
1144 | /* Since close does not return a layout stateid for use as |
1145 | * a barrier, we choose the worst-case barrier. | |
1146 | */ | |
1147 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
1148 | out: | |
193e3aa2 PT |
1149 | if (!found) { |
1150 | stateid = lo->plh_stateid; | |
1151 | layoutreturn = | |
1152 | test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, | |
1153 | &lo->plh_flags); | |
1154 | if (layoutreturn) { | |
1155 | lo->plh_block_lgets++; | |
1156 | pnfs_get_layout_hdr(lo); | |
1157 | } | |
1158 | } | |
f7e8917a | 1159 | spin_unlock(&ino->i_lock); |
193e3aa2 PT |
1160 | if (layoutreturn) { |
1161 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
27b6f539 | 1162 | pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, false); |
193e3aa2 | 1163 | } |
f7e8917a FI |
1164 | return found; |
1165 | } | |
1166 | ||
b1f69b75 AA |
1167 | /* |
1168 | * Compare two layout segments for sorting into layout cache. | |
1169 | * We want to preferentially return RW over RO layouts, so ensure those | |
1170 | * are seen first. | |
1171 | */ | |
1172 | static s64 | |
7dc0ac70 | 1173 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1174 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1175 | { |
fb3296eb BH |
1176 | s64 d; |
1177 | ||
1178 | /* high offset > low offset */ | |
1179 | d = l1->offset - l2->offset; | |
1180 | if (d) | |
1181 | return d; | |
1182 | ||
1183 | /* short length > long length */ | |
1184 | d = l2->length - l1->length; | |
1185 | if (d) | |
1186 | return d; | |
1187 | ||
b1f69b75 | 1188 | /* read > read/write */ |
fb3296eb | 1189 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1190 | } |
1191 | ||
974cec8c | 1192 | static void |
57036a37 | 1193 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
1194 | struct pnfs_layout_segment *lseg) |
1195 | { | |
b1f69b75 | 1196 | struct pnfs_layout_segment *lp; |
b1f69b75 | 1197 | |
974cec8c AA |
1198 | dprintk("%s:Begin\n", __func__); |
1199 | ||
b7edfaa1 | 1200 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
7dc0ac70 | 1201 | if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 1202 | continue; |
566052c5 | 1203 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1204 | dprintk("%s: inserted lseg %p " |
1205 | "iomode %d offset %llu length %llu before " | |
1206 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1207 | __func__, lseg, lseg->pls_range.iomode, |
1208 | lseg->pls_range.offset, lseg->pls_range.length, | |
1209 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1210 | lp->pls_range.length); | |
fb3296eb | 1211 | goto out; |
974cec8c | 1212 | } |
fb3296eb BH |
1213 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1214 | dprintk("%s: inserted lseg %p " | |
1215 | "iomode %d offset %llu length %llu at tail\n", | |
1216 | __func__, lseg, lseg->pls_range.iomode, | |
1217 | lseg->pls_range.offset, lseg->pls_range.length); | |
1218 | out: | |
70c3bd2b | 1219 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1220 | |
1221 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
1222 | } |
1223 | ||
1224 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1225 | alloc_init_layout_hdr(struct inode *ino, |
1226 | struct nfs_open_context *ctx, | |
1227 | gfp_t gfp_flags) | |
e5e94017 BH |
1228 | { |
1229 | struct pnfs_layout_hdr *lo; | |
1230 | ||
636fb9c8 | 1231 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1232 | if (!lo) |
1233 | return NULL; | |
cc6e5340 | 1234 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1235 | INIT_LIST_HEAD(&lo->plh_layouts); |
1236 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1237 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1238 | lo->plh_inode = ino; |
5cc2216d | 1239 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
e5e94017 BH |
1240 | return lo; |
1241 | } | |
1242 | ||
1243 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1244 | pnfs_find_alloc_layout(struct inode *ino, |
1245 | struct nfs_open_context *ctx, | |
1246 | gfp_t gfp_flags) | |
e5e94017 BH |
1247 | { |
1248 | struct nfs_inode *nfsi = NFS_I(ino); | |
1249 | struct pnfs_layout_hdr *new = NULL; | |
1250 | ||
1251 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1252 | ||
251ec410 TM |
1253 | if (nfsi->layout != NULL) |
1254 | goto out_existing; | |
e5e94017 | 1255 | spin_unlock(&ino->i_lock); |
9fa40758 | 1256 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1257 | spin_lock(&ino->i_lock); |
1258 | ||
251ec410 | 1259 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1260 | nfsi->layout = new; |
251ec410 | 1261 | return new; |
7175fe90 YN |
1262 | } else if (new != NULL) |
1263 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1264 | out_existing: |
1265 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1266 | return nfsi->layout; |
1267 | } | |
1268 | ||
b1f69b75 AA |
1269 | /* |
1270 | * iomode matching rules: | |
1271 | * iomode lseg match | |
1272 | * ----- ----- ----- | |
1273 | * ANY READ true | |
1274 | * ANY RW true | |
1275 | * RW READ false | |
1276 | * RW RW true | |
1277 | * READ READ true | |
1278 | * READ RW true | |
1279 | */ | |
3cb2df17 | 1280 | static bool |
7dc0ac70 | 1281 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
3cb2df17 | 1282 | const struct pnfs_layout_range *range) |
b1f69b75 | 1283 | { |
fb3296eb BH |
1284 | struct pnfs_layout_range range1; |
1285 | ||
1286 | if ((range->iomode == IOMODE_RW && | |
1287 | ls_range->iomode != IOMODE_RW) || | |
7dc0ac70 | 1288 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1289 | return 0; |
1290 | ||
1291 | /* range1 covers only the first byte in the range */ | |
1292 | range1 = *range; | |
1293 | range1.length = 1; | |
7dc0ac70 | 1294 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1295 | } |
1296 | ||
1297 | /* | |
1298 | * lookup range in layout | |
1299 | */ | |
e5e94017 | 1300 | static struct pnfs_layout_segment * |
fb3296eb BH |
1301 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
1302 | struct pnfs_layout_range *range) | |
e5e94017 | 1303 | { |
b1f69b75 AA |
1304 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1305 | ||
1306 | dprintk("%s:Begin\n", __func__); | |
1307 | ||
b7edfaa1 | 1308 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1309 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1310 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
7dc0ac70 | 1311 | pnfs_lseg_range_match(&lseg->pls_range, range)) { |
9369a431 | 1312 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1313 | break; |
1314 | } | |
d771e3a4 | 1315 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
1316 | break; |
1317 | } | |
1318 | ||
1319 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1320 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1321 | return ret; |
e5e94017 BH |
1322 | } |
1323 | ||
d23d61c8 AA |
1324 | /* |
1325 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1326 | * to the MDS or over pNFS | |
1327 | * | |
1328 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1329 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1330 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1331 | * WRITE request. | |
1332 | * | |
1333 | * A return of true means use MDS I/O. | |
1334 | * | |
1335 | * From rfc 5661: | |
1336 | * If a file's size is smaller than the file size threshold, data accesses | |
1337 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1338 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1339 | * server. If both file size and I/O size are provided, the client SHOULD | |
1340 | * reach or exceed both thresholds before sending its read or write | |
1341 | * requests to the data server. | |
1342 | */ | |
1343 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1344 | struct inode *ino, int iomode) | |
1345 | { | |
1346 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1347 | struct nfs_inode *nfsi = NFS_I(ino); | |
1348 | loff_t fsize = i_size_read(ino); | |
1349 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1350 | ||
1351 | if (t == NULL) | |
1352 | return ret; | |
1353 | ||
1354 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1355 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1356 | ||
1357 | switch (iomode) { | |
1358 | case IOMODE_READ: | |
1359 | if (t->bm & THRESHOLD_RD) { | |
1360 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1361 | size_set = true; | |
1362 | if (fsize < t->rd_sz) | |
1363 | size = true; | |
1364 | } | |
1365 | if (t->bm & THRESHOLD_RD_IO) { | |
1366 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1367 | nfsi->read_io); | |
1368 | io_set = true; | |
1369 | if (nfsi->read_io < t->rd_io_sz) | |
1370 | io = true; | |
1371 | } | |
1372 | break; | |
1373 | case IOMODE_RW: | |
1374 | if (t->bm & THRESHOLD_WR) { | |
1375 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1376 | size_set = true; | |
1377 | if (fsize < t->wr_sz) | |
1378 | size = true; | |
1379 | } | |
1380 | if (t->bm & THRESHOLD_WR_IO) { | |
1381 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1382 | nfsi->write_io); | |
1383 | io_set = true; | |
1384 | if (nfsi->write_io < t->wr_io_sz) | |
1385 | io = true; | |
1386 | } | |
1387 | break; | |
1388 | } | |
1389 | if (size_set && io_set) { | |
1390 | if (size && io) | |
1391 | ret = true; | |
1392 | } else if (size || io) | |
1393 | ret = true; | |
1394 | ||
1395 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1396 | return ret; | |
1397 | } | |
1398 | ||
e5e94017 BH |
1399 | /* |
1400 | * Layout segment is retreived from the server if not cached. | |
1401 | * The appropriate layout segment is referenced and returned to the caller. | |
1402 | */ | |
7c24d948 | 1403 | struct pnfs_layout_segment * |
e5e94017 BH |
1404 | pnfs_update_layout(struct inode *ino, |
1405 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1406 | loff_t pos, |
1407 | u64 count, | |
a75b9df9 TM |
1408 | enum pnfs_iomode iomode, |
1409 | gfp_t gfp_flags) | |
e5e94017 | 1410 | { |
fb3296eb BH |
1411 | struct pnfs_layout_range arg = { |
1412 | .iomode = iomode, | |
1413 | .offset = pos, | |
1414 | .length = count, | |
1415 | }; | |
707ed5fd | 1416 | unsigned pg_offset; |
6382a441 WAA |
1417 | struct nfs_server *server = NFS_SERVER(ino); |
1418 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1419 | struct pnfs_layout_hdr *lo; |
1420 | struct pnfs_layout_segment *lseg = NULL; | |
9bf87482 | 1421 | bool first; |
e5e94017 BH |
1422 | |
1423 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1424 | goto out; |
d23d61c8 AA |
1425 | |
1426 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1427 | goto out; |
d23d61c8 | 1428 | |
9bf87482 PT |
1429 | lookup_again: |
1430 | first = false; | |
e5e94017 | 1431 | spin_lock(&ino->i_lock); |
9fa40758 | 1432 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1433 | if (lo == NULL) { |
1434 | spin_unlock(&ino->i_lock); | |
1435 | goto out; | |
1436 | } | |
e5e94017 | 1437 | |
43f1b3da | 1438 | /* Do we even need to bother with this? */ |
a59c30ac | 1439 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1440 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1441 | goto out_unlock; |
1442 | } | |
1443 | ||
1444 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1445 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1446 | goto out_unlock; |
1447 | ||
9bf87482 PT |
1448 | first = list_empty(&lo->plh_segs); |
1449 | if (first) { | |
1450 | /* The first layoutget for the file. Need to serialize per | |
1451 | * RFC 5661 Errata 3208. | |
1452 | */ | |
1453 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1454 | &lo->plh_flags)) { | |
1455 | spin_unlock(&ino->i_lock); | |
1456 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1457 | TASK_UNINTERRUPTIBLE); | |
1458 | pnfs_put_layout_hdr(lo); | |
1459 | goto lookup_again; | |
1460 | } | |
1461 | } else { | |
1462 | /* Check to see if the layout for the given range | |
1463 | * already exists | |
1464 | */ | |
1465 | lseg = pnfs_find_lseg(lo, &arg); | |
1466 | if (lseg) | |
1467 | goto out_unlock; | |
1468 | } | |
568e8c49 | 1469 | |
ce6ab4f2 | 1470 | if (pnfs_layoutgets_blocked(lo, &arg, 0)) |
cf7d63f1 FI |
1471 | goto out_unlock; |
1472 | atomic_inc(&lo->plh_outstanding); | |
f49f9baa | 1473 | spin_unlock(&ino->i_lock); |
30005121 | 1474 | |
abb9a007 | 1475 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1476 | /* The lo must be on the clp list if there is any |
1477 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1478 | */ | |
1479 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1480 | if (list_empty(&lo->plh_layouts)) |
1481 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1482 | spin_unlock(&clp->cl_lock); |
1483 | } | |
e5e94017 | 1484 | |
707ed5fd BH |
1485 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1486 | if (pg_offset) { | |
1487 | arg.offset -= pg_offset; | |
1488 | arg.length += pg_offset; | |
1489 | } | |
7c24d948 AA |
1490 | if (arg.length != NFS4_MAX_UINT64) |
1491 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1492 | |
fb3296eb | 1493 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
cf7d63f1 | 1494 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1495 | out_put_layout_hdr: |
9bf87482 PT |
1496 | if (first) { |
1497 | unsigned long *bitlock = &lo->plh_flags; | |
1498 | ||
1499 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1500 | smp_mb__after_atomic(); | |
1501 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1502 | } | |
70c3bd2b | 1503 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1504 | out: |
f86bbcf8 TM |
1505 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1506 | "(%s, offset: %llu, length: %llu)\n", | |
1507 | __func__, ino->i_sb->s_id, | |
1508 | (unsigned long long)NFS_FILEID(ino), | |
1509 | lseg == NULL ? "not found" : "found", | |
1510 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1511 | (unsigned long long)pos, | |
1512 | (unsigned long long)count); | |
e5e94017 BH |
1513 | return lseg; |
1514 | out_unlock: | |
1515 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1516 | goto out_put_layout_hdr; |
e5e94017 | 1517 | } |
7c24d948 | 1518 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1519 | |
a0b0a6e3 | 1520 | struct pnfs_layout_segment * |
b1f69b75 AA |
1521 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1522 | { | |
1523 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1524 | struct nfs4_layoutget_res *res = &lgp->res; | |
1525 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1526 | struct inode *ino = lo->plh_inode; |
78096cca | 1527 | LIST_HEAD(free_me); |
b1f69b75 AA |
1528 | int status = 0; |
1529 | ||
1530 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1531 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1532 | if (!lseg || IS_ERR(lseg)) { |
1533 | if (!lseg) | |
1534 | status = -ENOMEM; | |
1535 | else | |
1536 | status = PTR_ERR(lseg); | |
1537 | dprintk("%s: Could not allocate layout: error %d\n", | |
1538 | __func__, status); | |
1539 | goto out; | |
1540 | } | |
1541 | ||
1013df61 CH |
1542 | init_lseg(lo, lseg); |
1543 | lseg->pls_range = res->range; | |
1544 | ||
b1f69b75 | 1545 | spin_lock(&ino->i_lock); |
a59c30ac | 1546 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1547 | dprintk("%s forget reply due to recall\n", __func__); |
1548 | goto out_forget_reply; | |
1549 | } | |
1550 | ||
ce6ab4f2 | 1551 | if (pnfs_layoutgets_blocked(lo, &lgp->args.range, 1)) { |
43f1b3da FI |
1552 | dprintk("%s forget reply due to state\n", __func__); |
1553 | goto out_forget_reply; | |
1554 | } | |
038d6493 | 1555 | |
362f7474 CH |
1556 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1557 | /* existing state ID, make sure the sequence number matches. */ | |
1558 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1559 | dprintk("%s forget reply due to sequence\n", __func__); | |
1560 | goto out_forget_reply; | |
1561 | } | |
1562 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1563 | } else { | |
1564 | /* | |
1565 | * We got an entirely new state ID. Mark all segments for the | |
1566 | * inode invalid, and don't bother validating the stateid | |
1567 | * sequence number. | |
1568 | */ | |
1569 | pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL); | |
1570 | ||
1571 | nfs4_stateid_copy(&lo->plh_stateid, &res->stateid); | |
1572 | lo->plh_barrier = be32_to_cpu(res->stateid.seqid); | |
1573 | } | |
038d6493 | 1574 | |
47abadef CH |
1575 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); |
1576 | ||
9369a431 | 1577 | pnfs_get_lseg(lseg); |
57036a37 | 1578 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1579 | |
f7e8917a FI |
1580 | if (res->return_on_close) { |
1581 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1582 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1583 | } | |
1584 | ||
b1f69b75 | 1585 | spin_unlock(&ino->i_lock); |
78096cca | 1586 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1587 | return lseg; |
b1f69b75 | 1588 | out: |
a0b0a6e3 | 1589 | return ERR_PTR(status); |
43f1b3da FI |
1590 | |
1591 | out_forget_reply: | |
1592 | spin_unlock(&ino->i_lock); | |
1593 | lseg->pls_layout = lo; | |
1594 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1595 | goto out; | |
b1f69b75 AA |
1596 | } |
1597 | ||
016256df PT |
1598 | static void |
1599 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, | |
1600 | struct list_head *tmp_list, | |
1601 | struct pnfs_layout_range *return_range) | |
1602 | { | |
1603 | struct pnfs_layout_segment *lseg, *next; | |
1604 | ||
1605 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
1606 | ||
1607 | if (list_empty(&lo->plh_segs)) | |
1608 | return; | |
1609 | ||
1610 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
1611 | if (should_free_lseg(&lseg->pls_range, return_range)) { | |
1612 | dprintk("%s: marking lseg %p iomode %d " | |
1613 | "offset %llu length %llu\n", __func__, | |
1614 | lseg, lseg->pls_range.iomode, | |
1615 | lseg->pls_range.offset, | |
1616 | lseg->pls_range.length); | |
1617 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
1618 | mark_lseg_invalid(lseg, tmp_list); | |
1619 | } | |
1620 | } | |
1621 | ||
1622 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
1623 | struct pnfs_layout_segment *lseg) | |
1624 | { | |
1625 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
1626 | int iomode = pnfs_iomode_to_fail_bit(lseg->pls_range.iomode); | |
1627 | struct pnfs_layout_range range = { | |
1628 | .iomode = lseg->pls_range.iomode, | |
1629 | .offset = 0, | |
1630 | .length = NFS4_MAX_UINT64, | |
1631 | }; | |
1632 | LIST_HEAD(free_me); | |
1633 | ||
1634 | spin_lock(&inode->i_lock); | |
1635 | /* set failure bit so that pnfs path will be retried later */ | |
1636 | pnfs_layout_set_fail_bit(lo, iomode); | |
1637 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
1638 | if (lo->plh_return_iomode == 0) | |
1639 | lo->plh_return_iomode = range.iomode; | |
1640 | else if (lo->plh_return_iomode != range.iomode) | |
1641 | lo->plh_return_iomode = IOMODE_ANY; | |
1642 | /* | |
1643 | * mark all matching lsegs so that we are sure to have no live | |
1644 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
1645 | * for how it works. | |
1646 | */ | |
1647 | pnfs_mark_matching_lsegs_return(lo, &free_me, &range); | |
1648 | spin_unlock(&inode->i_lock); | |
1649 | pnfs_free_lseg_list(&free_me); | |
1650 | } | |
1651 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
1652 | ||
d8007d4d TM |
1653 | void |
1654 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1655 | { | |
1fd937bd PT |
1656 | u64 rd_size = req->wb_bytes; |
1657 | ||
bc5a89b3 | 1658 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d | 1659 | |
1fd937bd PT |
1660 | if (pgio->pg_dreq == NULL) |
1661 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1662 | else | |
1663 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1664 | ||
d8007d4d TM |
1665 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1666 | req->wb_context, | |
1667 | req_offset(req), | |
1fd937bd | 1668 | rd_size, |
d8007d4d TM |
1669 | IOMODE_READ, |
1670 | GFP_KERNEL); | |
e885de1a TM |
1671 | /* If no lseg, fall back to read through mds */ |
1672 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1673 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1674 | |
d8007d4d TM |
1675 | } |
1676 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1677 | ||
1678 | void | |
6296556f PT |
1679 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1680 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1681 | { |
bc5a89b3 | 1682 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d TM |
1683 | |
1684 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1685 | req->wb_context, | |
1686 | req_offset(req), | |
6296556f | 1687 | wb_size, |
d8007d4d TM |
1688 | IOMODE_RW, |
1689 | GFP_NOFS); | |
e885de1a TM |
1690 | /* If no lseg, fall back to write through mds */ |
1691 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1692 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1693 | } |
1694 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1695 | ||
180bb5ec WAA |
1696 | void |
1697 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
1698 | { | |
1699 | if (desc->pg_lseg) { | |
1700 | pnfs_put_lseg(desc->pg_lseg); | |
1701 | desc->pg_lseg = NULL; | |
1702 | } | |
1703 | } | |
1704 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
1705 | ||
b4fdac1a WAA |
1706 | /* |
1707 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1708 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1709 | */ | |
1710 | size_t | |
a7d42ddb WAA |
1711 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
1712 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 1713 | { |
0f9c429e | 1714 | unsigned int size; |
c5e20cb7 | 1715 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1716 | |
1717 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1718 | if (!size) |
1719 | return 0; | |
94ad1c80 | 1720 | |
19982ba8 | 1721 | /* |
c5e20cb7 WAA |
1722 | * 'size' contains the number of bytes left in the current page (up |
1723 | * to the original size asked for in @req->wb_bytes). | |
1724 | * | |
1725 | * Calculate how many bytes are left in the layout segment | |
1726 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1727 | * |
1728 | * Please also note that 'end_offset' is actually the offset of the | |
1729 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1730 | * | |
1731 | */ | |
19b54848 | 1732 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1733 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1734 | pgio->pg_lseg->pls_range.length); | |
1735 | req_start = req_offset(req); | |
1736 | WARN_ON_ONCE(req_start > seg_end); | |
1737 | /* start of request is past the last byte of this segment */ | |
1738 | if (req_start >= seg_end) | |
19b54848 | 1739 | return 0; |
c5e20cb7 WAA |
1740 | |
1741 | /* adjust 'size' iff there are fewer bytes left in the | |
1742 | * segment than what nfs_generic_pg_test returned */ | |
1743 | seg_left = seg_end - req_start; | |
1744 | if (seg_left < size) | |
1745 | size = (unsigned int)seg_left; | |
19b54848 | 1746 | } |
0f9c429e | 1747 | |
19b54848 | 1748 | return size; |
94ad1c80 | 1749 | } |
89a58e32 | 1750 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1751 | |
53113ad3 | 1752 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
1753 | { |
1754 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
1755 | |
1756 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
1757 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
1758 | hdr->completion_ops); | |
1759 | return nfs_pageio_resend(&pgio, hdr); | |
e2fecb21 | 1760 | } |
e7dd79af | 1761 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1762 | |
d45f60c6 | 1763 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1764 | { |
cd841605 FI |
1765 | |
1766 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1767 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1768 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1769 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1770 | } |
6c75dc0d | 1771 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1772 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
1773 | } |
1774 | ||
d20581aa BH |
1775 | /* |
1776 | * Called by non rpc-based layout drivers | |
1777 | */ | |
d45f60c6 | 1778 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 1779 | { |
d45f60c6 | 1780 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); |
cd841605 | 1781 | if (!hdr->pnfs_error) { |
d45f60c6 WAA |
1782 | pnfs_set_layoutcommit(hdr); |
1783 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
1acbbb4e | 1784 | } else |
d45f60c6 WAA |
1785 | pnfs_ld_handle_write_error(hdr); |
1786 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 1787 | } |
d20581aa | 1788 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1789 | |
dce81290 TM |
1790 | static void |
1791 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1792 | struct nfs_pgio_header *hdr) |
dce81290 | 1793 | { |
48d635f1 | 1794 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1795 | |
6c75dc0d | 1796 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 1797 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 1798 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 1799 | mirror->pg_recoalesce = 1; |
6c75dc0d | 1800 | } |
d45f60c6 | 1801 | nfs_pgio_data_destroy(hdr); |
dce81290 TM |
1802 | } |
1803 | ||
1804 | static enum pnfs_try_status | |
d45f60c6 | 1805 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
1806 | const struct rpc_call_ops *call_ops, |
1807 | struct pnfs_layout_segment *lseg, | |
1808 | int how) | |
0382b744 | 1809 | { |
cd841605 | 1810 | struct inode *inode = hdr->inode; |
0382b744 AA |
1811 | enum pnfs_try_status trypnfs; |
1812 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1813 | ||
cd841605 | 1814 | hdr->mds_ops = call_ops; |
0382b744 AA |
1815 | |
1816 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
1817 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
1818 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 1819 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1820 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1821 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1822 | return trypnfs; | |
1823 | } | |
1824 | ||
dce81290 | 1825 | static void |
7f714720 WAA |
1826 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
1827 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 1828 | { |
dce81290 TM |
1829 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1830 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1831 | enum pnfs_try_status trypnfs; |
dce81290 | 1832 | |
d45f60c6 | 1833 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 1834 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1835 | pnfs_write_through_mds(desc, hdr); |
dce81290 TM |
1836 | } |
1837 | ||
6c75dc0d FI |
1838 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1839 | { | |
9369a431 | 1840 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1841 | nfs_pgio_header_free(hdr); |
6c75dc0d | 1842 | } |
89d77c8f | 1843 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1844 | |
dce81290 TM |
1845 | int |
1846 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1847 | { | |
48d635f1 | 1848 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1849 | |
6c75dc0d | 1850 | struct nfs_pgio_header *hdr; |
dce81290 TM |
1851 | int ret; |
1852 | ||
1e7f3a48 WAA |
1853 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1854 | if (!hdr) { | |
a7d42ddb | 1855 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); |
6c75dc0d | 1856 | return -ENOMEM; |
dce81290 | 1857 | } |
6c75dc0d | 1858 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 1859 | |
9369a431 | 1860 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1861 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 1862 | if (!ret) |
7f714720 | 1863 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 1864 | |
6c75dc0d | 1865 | return ret; |
dce81290 TM |
1866 | } |
1867 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1868 | ||
53113ad3 | 1869 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
1870 | { |
1871 | struct nfs_pageio_descriptor pgio; | |
1872 | ||
1acbbb4e | 1873 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
1874 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
1875 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 1876 | } |
e7dd79af | 1877 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 1878 | |
d45f60c6 | 1879 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1880 | { |
cd841605 FI |
1881 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
1882 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1883 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1884 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1885 | } |
4db6e0b7 | 1886 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1887 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
1888 | } |
1889 | ||
d20581aa BH |
1890 | /* |
1891 | * Called by non rpc-based layout drivers | |
1892 | */ | |
d45f60c6 | 1893 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 1894 | { |
d45f60c6 | 1895 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
cd841605 | 1896 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
1897 | __nfs4_read_done_cb(hdr); |
1898 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
62e4a769 | 1899 | } else |
d45f60c6 WAA |
1900 | pnfs_ld_handle_read_error(hdr); |
1901 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
1902 | } |
1903 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1904 | ||
493292dd TM |
1905 | static void |
1906 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1907 | struct nfs_pgio_header *hdr) |
493292dd | 1908 | { |
48d635f1 | 1909 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1910 | |
4db6e0b7 | 1911 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 1912 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 1913 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 1914 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 1915 | } |
d45f60c6 | 1916 | nfs_pgio_data_destroy(hdr); |
493292dd TM |
1917 | } |
1918 | ||
64419a9b AA |
1919 | /* |
1920 | * Call the appropriate parallel I/O subsystem read function. | |
1921 | */ | |
493292dd | 1922 | static enum pnfs_try_status |
d45f60c6 | 1923 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
1924 | const struct rpc_call_ops *call_ops, |
1925 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1926 | { |
cd841605 | 1927 | struct inode *inode = hdr->inode; |
64419a9b AA |
1928 | struct nfs_server *nfss = NFS_SERVER(inode); |
1929 | enum pnfs_try_status trypnfs; | |
1930 | ||
cd841605 | 1931 | hdr->mds_ops = call_ops; |
64419a9b AA |
1932 | |
1933 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 1934 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 1935 | |
d45f60c6 | 1936 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 1937 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1938 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1939 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1940 | return trypnfs; | |
1941 | } | |
863a3c6c | 1942 | |
ceb11e13 PT |
1943 | /* Resend all requests through pnfs. */ |
1944 | int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) | |
1945 | { | |
1946 | struct nfs_pageio_descriptor pgio; | |
1947 | ||
1948 | nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops); | |
1949 | return nfs_pageio_resend(&pgio, hdr); | |
1950 | } | |
1951 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
1952 | ||
493292dd | 1953 | static void |
7f714720 | 1954 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 1955 | { |
493292dd TM |
1956 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1957 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1958 | enum pnfs_try_status trypnfs; |
ceb11e13 | 1959 | int err = 0; |
493292dd | 1960 | |
d45f60c6 | 1961 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
ceb11e13 PT |
1962 | if (trypnfs == PNFS_TRY_AGAIN) |
1963 | err = pnfs_read_resend_pnfs(hdr); | |
1964 | if (trypnfs == PNFS_NOT_ATTEMPTED || err) | |
d45f60c6 | 1965 | pnfs_read_through_mds(desc, hdr); |
493292dd TM |
1966 | } |
1967 | ||
4db6e0b7 FI |
1968 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1969 | { | |
9369a431 | 1970 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1971 | nfs_pgio_header_free(hdr); |
4db6e0b7 | 1972 | } |
89d77c8f | 1973 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1974 | |
493292dd TM |
1975 | int |
1976 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1977 | { | |
48d635f1 | 1978 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1979 | |
4db6e0b7 | 1980 | struct nfs_pgio_header *hdr; |
493292dd TM |
1981 | int ret; |
1982 | ||
1e7f3a48 WAA |
1983 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1984 | if (!hdr) { | |
a7d42ddb | 1985 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); |
180bb5ec | 1986 | return -ENOMEM; |
493292dd | 1987 | } |
4db6e0b7 | 1988 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 1989 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1990 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 1991 | if (!ret) |
7f714720 | 1992 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 1993 | return ret; |
493292dd TM |
1994 | } |
1995 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1996 | ||
71244d9b TM |
1997 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
1998 | { | |
1999 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
2000 | ||
2001 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 2002 | smp_mb__after_atomic(); |
71244d9b TM |
2003 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
2004 | } | |
2005 | ||
863a3c6c | 2006 | /* |
a9bae566 | 2007 | * There can be multiple RW segments. |
863a3c6c | 2008 | */ |
a9bae566 | 2009 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 2010 | { |
a9bae566 | 2011 | struct pnfs_layout_segment *lseg; |
863a3c6c | 2012 | |
a9bae566 PT |
2013 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
2014 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 2015 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
2016 | list_add(&lseg->pls_lc_list, listp); |
2017 | } | |
863a3c6c AA |
2018 | } |
2019 | ||
a073dbff TM |
2020 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
2021 | { | |
2022 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
2023 | |
2024 | /* Matched by references in pnfs_set_layoutcommit */ | |
2025 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
2026 | list_del_init(&lseg->pls_lc_list); | |
2027 | pnfs_put_lseg(lseg); | |
2028 | } | |
2029 | ||
71244d9b | 2030 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
2031 | } |
2032 | ||
1b0ae068 PT |
2033 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
2034 | { | |
b9e028fd | 2035 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
2036 | } |
2037 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
2038 | ||
863a3c6c | 2039 | void |
d45f60c6 | 2040 | pnfs_set_layoutcommit(struct nfs_pgio_header *hdr) |
863a3c6c | 2041 | { |
cd841605 FI |
2042 | struct inode *inode = hdr->inode; |
2043 | struct nfs_inode *nfsi = NFS_I(inode); | |
d45f60c6 | 2044 | loff_t end_pos = hdr->mds_offset + hdr->res.count; |
79a48a1f | 2045 | bool mark_as_dirty = false; |
863a3c6c | 2046 | |
cd841605 | 2047 | spin_lock(&inode->i_lock); |
863a3c6c | 2048 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 2049 | mark_as_dirty = true; |
863a3c6c | 2050 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 2051 | __func__, inode->i_ino); |
863a3c6c | 2052 | } |
cd841605 | 2053 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 2054 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 2055 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 2056 | } |
acff5880 PT |
2057 | if (end_pos > nfsi->layout->plh_lwb) |
2058 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 2059 | spin_unlock(&inode->i_lock); |
acff5880 | 2060 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 2061 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
2062 | |
2063 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2064 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2065 | if (mark_as_dirty) | |
cd841605 | 2066 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
2067 | } |
2068 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
2069 | ||
378520b8 PT |
2070 | void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data) |
2071 | { | |
2072 | struct inode *inode = data->inode; | |
2073 | struct nfs_inode *nfsi = NFS_I(inode); | |
2074 | bool mark_as_dirty = false; | |
2075 | ||
2076 | spin_lock(&inode->i_lock); | |
2077 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { | |
2078 | mark_as_dirty = true; | |
2079 | dprintk("%s: Set layoutcommit for inode %lu ", | |
2080 | __func__, inode->i_ino); | |
2081 | } | |
2082 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) { | |
2083 | /* references matched in nfs4_layoutcommit_release */ | |
2084 | pnfs_get_lseg(data->lseg); | |
2085 | } | |
2086 | if (data->lwb > nfsi->layout->plh_lwb) | |
2087 | nfsi->layout->plh_lwb = data->lwb; | |
2088 | spin_unlock(&inode->i_lock); | |
2089 | dprintk("%s: lseg %p end_pos %llu\n", | |
2090 | __func__, data->lseg, nfsi->layout->plh_lwb); | |
2091 | ||
2092 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2093 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2094 | if (mark_as_dirty) | |
2095 | mark_inode_dirty_sync(inode); | |
2096 | } | |
2097 | EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit); | |
2098 | ||
db29c089 AA |
2099 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2100 | { | |
2101 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2102 | ||
2103 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2104 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2105 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2106 | } |
2107 | ||
de4b15c7 AA |
2108 | /* |
2109 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2110 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2111 | * data to disk to allow the server to recover the data if it crashes. | |
2112 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2113 | * is off, and a COMMIT is sent to a data server, or | |
2114 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2115 | */ | |
863a3c6c | 2116 | int |
ef311537 | 2117 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2118 | { |
5f919c9f | 2119 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2120 | struct nfs4_layoutcommit_data *data; |
2121 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2122 | loff_t end_pos; |
71244d9b | 2123 | int status; |
863a3c6c | 2124 | |
71244d9b | 2125 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2126 | return 0; |
2127 | ||
71244d9b | 2128 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2129 | |
71244d9b | 2130 | status = -EAGAIN; |
92407e75 | 2131 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2132 | if (!sync) |
2133 | goto out; | |
74316201 | 2134 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2135 | NFS_INO_LAYOUTCOMMITTING, |
2136 | nfs_wait_bit_killable, | |
2137 | TASK_KILLABLE); | |
92407e75 | 2138 | if (status) |
71244d9b | 2139 | goto out; |
92407e75 PT |
2140 | } |
2141 | ||
71244d9b TM |
2142 | status = -ENOMEM; |
2143 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2144 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2145 | if (!data) | |
2146 | goto clear_layoutcommitting; | |
2147 | ||
2148 | status = 0; | |
de4b15c7 | 2149 | spin_lock(&inode->i_lock); |
71244d9b TM |
2150 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2151 | goto out_unlock; | |
a9bae566 | 2152 | |
71244d9b | 2153 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2154 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2155 | |
acff5880 | 2156 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 2157 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 2158 | |
f597c537 | 2159 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2160 | spin_unlock(&inode->i_lock); |
2161 | ||
2162 | data->args.inode = inode; | |
9fa40758 | 2163 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2164 | nfs_fattr_init(&data->fattr); |
2165 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2166 | data->res.fattr = &data->fattr; | |
2167 | data->args.lastbytewritten = end_pos - 1; | |
2168 | data->res.server = NFS_SERVER(inode); | |
2169 | ||
5f919c9f CH |
2170 | if (ld->prepare_layoutcommit) { |
2171 | status = ld->prepare_layoutcommit(&data->args); | |
2172 | if (status) { | |
2173 | spin_lock(&inode->i_lock); | |
2174 | if (end_pos < nfsi->layout->plh_lwb) | |
2175 | nfsi->layout->plh_lwb = end_pos; | |
2176 | spin_unlock(&inode->i_lock); | |
2177 | put_rpccred(data->cred); | |
2178 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); | |
2179 | goto clear_layoutcommitting; | |
2180 | } | |
2181 | } | |
2182 | ||
2183 | ||
863a3c6c AA |
2184 | status = nfs4_proc_layoutcommit(data, sync); |
2185 | out: | |
92407e75 PT |
2186 | if (status) |
2187 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2188 | dprintk("<-- %s status %d\n", __func__, status); |
2189 | return status; | |
71244d9b TM |
2190 | out_unlock: |
2191 | spin_unlock(&inode->i_lock); | |
92407e75 | 2192 | kfree(data); |
71244d9b TM |
2193 | clear_layoutcommitting: |
2194 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2195 | goto out; |
863a3c6c | 2196 | } |
72cff449 | 2197 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a AA |
2198 | |
2199 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
2200 | { | |
2201 | struct nfs4_threshold *thp; | |
2202 | ||
2203 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2204 | if (!thp) { | |
2205 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2206 | return NULL; | |
2207 | } | |
2208 | return thp; | |
2209 | } |