]>
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" |
40dd4b7a | 37 | #include "delegation.h" |
8733408d | 38 | #include "nfs42.h" |
85e174ba RL |
39 | |
40 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
25c75333 | 41 | #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) |
85e174ba | 42 | |
02c35fca FI |
43 | /* Locking: |
44 | * | |
45 | * pnfs_spinlock: | |
46 | * protects pnfs_modules_tbl. | |
47 | */ | |
48 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
49 | ||
50 | /* | |
51 | * pnfs_modules_tbl holds all pnfs modules | |
52 | */ | |
53 | static LIST_HEAD(pnfs_modules_tbl); | |
54 | ||
aa1e0e3a | 55 | static int |
ed429d6b | 56 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid, |
6c16605d | 57 | enum pnfs_iomode iomode, bool sync); |
aa1e0e3a | 58 | |
02c35fca FI |
59 | /* Return the registered pnfs layout driver module matching given id */ |
60 | static struct pnfs_layoutdriver_type * | |
61 | find_pnfs_driver_locked(u32 id) | |
62 | { | |
63 | struct pnfs_layoutdriver_type *local; | |
64 | ||
65 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
66 | if (local->id == id) | |
67 | goto out; | |
68 | local = NULL; | |
69 | out: | |
70 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
71 | return local; | |
72 | } | |
73 | ||
85e174ba RL |
74 | static struct pnfs_layoutdriver_type * |
75 | find_pnfs_driver(u32 id) | |
76 | { | |
02c35fca FI |
77 | struct pnfs_layoutdriver_type *local; |
78 | ||
79 | spin_lock(&pnfs_spinlock); | |
80 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
81 | if (local != NULL && !try_module_get(local->owner)) { |
82 | dprintk("%s: Could not grab reference on module\n", __func__); | |
83 | local = NULL; | |
84 | } | |
02c35fca FI |
85 | spin_unlock(&pnfs_spinlock); |
86 | return local; | |
85e174ba RL |
87 | } |
88 | ||
89 | void | |
90 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
91 | { | |
738fd0f3 BH |
92 | if (nfss->pnfs_curr_ld) { |
93 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
94 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
95 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
96 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
97 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 98 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 99 | } |
85e174ba RL |
100 | nfss->pnfs_curr_ld = NULL; |
101 | } | |
102 | ||
103 | /* | |
104 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
105 | * Currently only one pNFS layout driver per filesystem is supported. | |
106 | * | |
107 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
108 | */ | |
109 | void | |
738fd0f3 BH |
110 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
111 | u32 id) | |
85e174ba RL |
112 | { |
113 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
114 | ||
115 | if (id == 0) | |
116 | goto out_no_driver; | |
117 | if (!(server->nfs_client->cl_exchange_flags & | |
118 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
a030889a WAA |
119 | printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n", |
120 | __func__, id, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
121 | goto out_no_driver; |
122 | } | |
123 | ld_type = find_pnfs_driver(id); | |
124 | if (!ld_type) { | |
125 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
126 | ld_type = find_pnfs_driver(id); | |
127 | if (!ld_type) { | |
128 | dprintk("%s: No pNFS module found for %u.\n", | |
129 | __func__, id); | |
130 | goto out_no_driver; | |
131 | } | |
132 | } | |
133 | server->pnfs_curr_ld = ld_type; | |
738fd0f3 BH |
134 | if (ld_type->set_layoutdriver |
135 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
136 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
137 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
138 | module_put(ld_type->owner); |
139 | goto out_no_driver; | |
140 | } | |
2a4c8994 TM |
141 | /* Bump the MDS count */ |
142 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 143 | |
85e174ba RL |
144 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
145 | return; | |
146 | ||
147 | out_no_driver: | |
148 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
149 | server->pnfs_curr_ld = NULL; | |
150 | } | |
02c35fca FI |
151 | |
152 | int | |
153 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
154 | { | |
155 | int status = -EINVAL; | |
156 | struct pnfs_layoutdriver_type *tmp; | |
157 | ||
158 | if (ld_type->id == 0) { | |
a030889a | 159 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
160 | return status; |
161 | } | |
b1f69b75 | 162 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 163 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
164 | "alloc_lseg and free_lseg.\n", __func__); |
165 | return status; | |
166 | } | |
02c35fca FI |
167 | |
168 | spin_lock(&pnfs_spinlock); | |
169 | tmp = find_pnfs_driver_locked(ld_type->id); | |
170 | if (!tmp) { | |
171 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
172 | status = 0; | |
173 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
174 | ld_type->name); | |
175 | } else { | |
a030889a | 176 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
177 | __func__, ld_type->id); |
178 | } | |
179 | spin_unlock(&pnfs_spinlock); | |
180 | ||
181 | return status; | |
182 | } | |
183 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
184 | ||
185 | void | |
186 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
187 | { | |
188 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
189 | spin_lock(&pnfs_spinlock); | |
190 | list_del(&ld_type->pnfs_tblid); | |
191 | spin_unlock(&pnfs_spinlock); | |
192 | } | |
193 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 194 | |
b1f69b75 AA |
195 | /* |
196 | * pNFS client layout cache | |
197 | */ | |
198 | ||
cc6e5340 | 199 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 200 | void |
70c3bd2b | 201 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 202 | { |
cc6e5340 | 203 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
204 | } |
205 | ||
636fb9c8 BH |
206 | static struct pnfs_layout_hdr * |
207 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
208 | { | |
209 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
57934278 | 210 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
211 | } |
212 | ||
213 | static void | |
214 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
215 | { | |
9c626381 TM |
216 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
217 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
218 | ||
219 | if (!list_empty(&lo->plh_layouts)) { | |
220 | struct nfs_client *clp = server->nfs_client; | |
221 | ||
222 | spin_lock(&clp->cl_lock); | |
223 | list_del_init(&lo->plh_layouts); | |
224 | spin_unlock(&clp->cl_lock); | |
225 | } | |
9fa40758 | 226 | put_rpccred(lo->plh_lc_cred); |
57934278 | 227 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
228 | } |
229 | ||
e5e94017 | 230 | static void |
6622c3ea | 231 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 232 | { |
bb346f63 | 233 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 234 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
235 | nfsi->layout = NULL; |
236 | /* Reset MDS Threshold I/O counters */ | |
237 | nfsi->write_io = 0; | |
238 | nfsi->read_io = 0; | |
e5e94017 BH |
239 | } |
240 | ||
b1f69b75 | 241 | void |
70c3bd2b | 242 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 243 | { |
cc6e5340 FI |
244 | struct inode *inode = lo->plh_inode; |
245 | ||
246 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
566f8737 PT |
247 | if (!list_empty(&lo->plh_segs)) |
248 | WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n"); | |
6622c3ea | 249 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 250 | spin_unlock(&inode->i_lock); |
6622c3ea | 251 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 252 | } |
974cec8c AA |
253 | } |
254 | ||
b9e028fd TM |
255 | static int |
256 | pnfs_iomode_to_fail_bit(u32 iomode) | |
257 | { | |
258 | return iomode == IOMODE_RW ? | |
259 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
260 | } | |
261 | ||
262 | static void | |
3e621214 | 263 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 264 | { |
25c75333 | 265 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 266 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
3e621214 TM |
267 | atomic_inc(&lo->plh_refcount); |
268 | } | |
269 | ||
270 | static void | |
271 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
272 | { | |
273 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
274 | atomic_dec(&lo->plh_refcount); | |
275 | } | |
276 | ||
277 | static void | |
278 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
279 | { | |
280 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
281 | struct pnfs_layout_range range = { |
282 | .iomode = iomode, | |
283 | .offset = 0, | |
284 | .length = NFS4_MAX_UINT64, | |
285 | }; | |
286 | LIST_HEAD(head); | |
3e621214 TM |
287 | |
288 | spin_lock(&inode->i_lock); | |
289 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
115ce575 | 290 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range); |
3e621214 | 291 | spin_unlock(&inode->i_lock); |
115ce575 | 292 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
293 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
294 | iomode == IOMODE_RW ? "RW" : "READ"); | |
295 | } | |
296 | ||
297 | static bool | |
298 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
299 | { | |
25c75333 | 300 | unsigned long start, end; |
3e621214 TM |
301 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
302 | ||
303 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
304 | return false; |
305 | end = jiffies; | |
306 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
307 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
308 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 309 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
310 | return false; |
311 | } | |
312 | return true; | |
b9e028fd TM |
313 | } |
314 | ||
974cec8c AA |
315 | static void |
316 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
317 | { | |
566052c5 | 318 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 319 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c FI |
320 | atomic_set(&lseg->pls_refcount, 1); |
321 | smp_mb(); | |
322 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 323 | lseg->pls_layout = lo; |
974cec8c AA |
324 | } |
325 | ||
905ca191 | 326 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 327 | { |
b7edfaa1 | 328 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 329 | |
b1f69b75 | 330 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
974cec8c AA |
331 | } |
332 | ||
d684d2ae | 333 | static void |
57036a37 TM |
334 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
335 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 336 | { |
57036a37 | 337 | struct inode *inode = lo->plh_inode; |
d684d2ae | 338 | |
d20581aa | 339 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 340 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
341 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
342 | atomic_dec(&lo->plh_refcount); | |
173f77e9 TM |
343 | if (list_empty(&lo->plh_segs)) |
344 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
d684d2ae FI |
345 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); |
346 | } | |
347 | ||
aa1e0e3a PT |
348 | /* Return true if layoutreturn is needed */ |
349 | static bool | |
350 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo, | |
27b6f539 | 351 | struct pnfs_layout_segment *lseg) |
aa1e0e3a PT |
352 | { |
353 | struct pnfs_layout_segment *s; | |
354 | ||
7f27392c | 355 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) |
aa1e0e3a PT |
356 | return false; |
357 | ||
358 | list_for_each_entry(s, &lo->plh_segs, pls_list) | |
27b6f539 | 359 | if (s != lseg && test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags)) |
aa1e0e3a PT |
360 | return false; |
361 | ||
aa1e0e3a PT |
362 | return true; |
363 | } | |
364 | ||
7f27392c TM |
365 | static bool |
366 | pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo) | |
367 | { | |
368 | if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) | |
369 | return false; | |
370 | lo->plh_return_iomode = 0; | |
7f27392c TM |
371 | pnfs_get_layout_hdr(lo); |
372 | clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, &lo->plh_flags); | |
373 | return true; | |
374 | } | |
375 | ||
4ef2e4f8 TM |
376 | static void pnfs_layoutreturn_before_put_lseg(struct pnfs_layout_segment *lseg, |
377 | struct pnfs_layout_hdr *lo, struct inode *inode) | |
27b6f539 | 378 | { |
27b6f539 PT |
379 | lo = lseg->pls_layout; |
380 | inode = lo->plh_inode; | |
381 | ||
382 | spin_lock(&inode->i_lock); | |
383 | if (pnfs_layout_need_return(lo, lseg)) { | |
384 | nfs4_stateid stateid; | |
385 | enum pnfs_iomode iomode; | |
7f27392c | 386 | bool send; |
27b6f539 | 387 | |
50f563ef | 388 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
27b6f539 | 389 | iomode = lo->plh_return_iomode; |
7f27392c | 390 | send = pnfs_prepare_layoutreturn(lo); |
27b6f539 | 391 | spin_unlock(&inode->i_lock); |
7f27392c TM |
392 | if (send) { |
393 | /* Send an async layoutreturn so we dont deadlock */ | |
ed429d6b | 394 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); |
7f27392c | 395 | } |
27b6f539 | 396 | } else |
4ef2e4f8 | 397 | spin_unlock(&inode->i_lock); |
27b6f539 PT |
398 | } |
399 | ||
bae724ef | 400 | void |
9369a431 | 401 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 402 | { |
57036a37 | 403 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
404 | struct inode *inode; |
405 | ||
406 | if (!lseg) | |
407 | return; | |
408 | ||
4541d16c FI |
409 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
410 | atomic_read(&lseg->pls_refcount), | |
411 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
4ef2e4f8 TM |
412 | |
413 | /* Handle the case where refcount != 1 */ | |
414 | if (atomic_add_unless(&lseg->pls_refcount, -1, 1)) | |
415 | return; | |
416 | ||
57036a37 TM |
417 | lo = lseg->pls_layout; |
418 | inode = lo->plh_inode; | |
4ef2e4f8 TM |
419 | /* Do we need a layoutreturn? */ |
420 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
421 | pnfs_layoutreturn_before_put_lseg(lseg, lo, inode); | |
422 | ||
d684d2ae | 423 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
faa4a54f TM |
424 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { |
425 | spin_unlock(&inode->i_lock); | |
426 | return; | |
427 | } | |
8f0d27dc | 428 | pnfs_get_layout_hdr(lo); |
4ef2e4f8 TM |
429 | pnfs_layout_remove_lseg(lo, lseg); |
430 | spin_unlock(&inode->i_lock); | |
431 | pnfs_free_lseg(lseg); | |
432 | pnfs_put_layout_hdr(lo); | |
4541d16c | 433 | } |
4541d16c | 434 | } |
9369a431 | 435 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 436 | |
6543f803 | 437 | static void pnfs_free_lseg_async_work(struct work_struct *work) |
e6cf82d1 WAA |
438 | { |
439 | struct pnfs_layout_segment *lseg; | |
6543f803 | 440 | struct pnfs_layout_hdr *lo; |
e6cf82d1 WAA |
441 | |
442 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
6543f803 | 443 | lo = lseg->pls_layout; |
e6cf82d1 | 444 | |
6543f803 TM |
445 | pnfs_free_lseg(lseg); |
446 | pnfs_put_layout_hdr(lo); | |
e6cf82d1 WAA |
447 | } |
448 | ||
6543f803 | 449 | static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg) |
e6cf82d1 | 450 | { |
6543f803 | 451 | INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work); |
e6cf82d1 WAA |
452 | schedule_work(&lseg->pls_work); |
453 | } | |
6543f803 TM |
454 | |
455 | void | |
456 | pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg) | |
457 | { | |
458 | if (!lseg) | |
459 | return; | |
460 | ||
461 | assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock); | |
462 | ||
463 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, | |
464 | atomic_read(&lseg->pls_refcount), | |
465 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
466 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
467 | struct pnfs_layout_hdr *lo = lseg->pls_layout; | |
faa4a54f TM |
468 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) |
469 | return; | |
6543f803 TM |
470 | pnfs_get_layout_hdr(lo); |
471 | pnfs_layout_remove_lseg(lo, lseg); | |
472 | pnfs_free_lseg_async(lseg); | |
473 | } | |
474 | } | |
475 | EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked); | |
e6cf82d1 | 476 | |
3cb2df17 | 477 | static u64 |
fb3296eb BH |
478 | end_offset(u64 start, u64 len) |
479 | { | |
480 | u64 end; | |
481 | ||
482 | end = start + len; | |
483 | return end >= start ? end : NFS4_MAX_UINT64; | |
484 | } | |
485 | ||
fb3296eb BH |
486 | /* |
487 | * is l2 fully contained in l1? | |
488 | * start1 end1 | |
489 | * [----------------------------------) | |
490 | * start2 end2 | |
491 | * [----------------) | |
492 | */ | |
3cb2df17 | 493 | static bool |
7dc0ac70 | 494 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 495 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
496 | { |
497 | u64 start1 = l1->offset; | |
498 | u64 end1 = end_offset(start1, l1->length); | |
499 | u64 start2 = l2->offset; | |
500 | u64 end2 = end_offset(start2, l2->length); | |
501 | ||
502 | return (start1 <= start2) && (end1 >= end2); | |
503 | } | |
504 | ||
505 | /* | |
506 | * is l1 and l2 intersecting? | |
507 | * start1 end1 | |
508 | * [----------------------------------) | |
509 | * start2 end2 | |
510 | * [----------------) | |
511 | */ | |
3cb2df17 | 512 | static bool |
7dc0ac70 | 513 | pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1, |
3cb2df17 | 514 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
515 | { |
516 | u64 start1 = l1->offset; | |
517 | u64 end1 = end_offset(start1, l1->length); | |
518 | u64 start2 = l2->offset; | |
519 | u64 end2 = end_offset(start2, l2->length); | |
520 | ||
521 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
522 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
523 | } | |
524 | ||
4541d16c | 525 | static bool |
3cb2df17 TM |
526 | should_free_lseg(const struct pnfs_layout_range *lseg_range, |
527 | const struct pnfs_layout_range *recall_range) | |
4541d16c | 528 | { |
778b5502 BH |
529 | return (recall_range->iomode == IOMODE_ANY || |
530 | lseg_range->iomode == recall_range->iomode) && | |
7dc0ac70 | 531 | pnfs_lseg_range_intersecting(lseg_range, recall_range); |
974cec8c AA |
532 | } |
533 | ||
24956804 TM |
534 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
535 | struct list_head *tmp_list) | |
536 | { | |
537 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
538 | return false; | |
539 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
540 | list_add(&lseg->pls_list, tmp_list); | |
541 | return true; | |
542 | } | |
543 | ||
4541d16c FI |
544 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
545 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
546 | struct list_head *tmp_list) | |
547 | { | |
548 | int rv = 0; | |
549 | ||
550 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
551 | /* Remove the reference keeping the lseg in the | |
552 | * list. It will now be removed when all | |
553 | * outstanding io is finished. | |
554 | */ | |
d684d2ae FI |
555 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
556 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 557 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 558 | rv = 1; |
4541d16c FI |
559 | } |
560 | return rv; | |
561 | } | |
562 | ||
563 | /* Returns count of number of matching invalid lsegs remaining in list | |
564 | * after call. | |
565 | */ | |
43f1b3da | 566 | int |
49a85061 | 567 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 568 | struct list_head *tmp_list, |
778b5502 | 569 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
570 | { |
571 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 572 | int invalid = 0, removed = 0; |
974cec8c AA |
573 | |
574 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
575 | ||
8006bfba | 576 | if (list_empty(&lo->plh_segs)) |
38511722 | 577 | return 0; |
4541d16c | 578 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
579 | if (!recall_range || |
580 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
581 | dprintk("%s: freeing lseg %p iomode %d " |
582 | "offset %llu length %llu\n", __func__, | |
583 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
584 | lseg->pls_range.length); | |
585 | invalid++; | |
586 | removed += mark_lseg_invalid(lseg, tmp_list); | |
587 | } | |
588 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
589 | return invalid - removed; | |
974cec8c AA |
590 | } |
591 | ||
f49f9baa | 592 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 593 | void |
4541d16c | 594 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 595 | { |
4541d16c | 596 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
597 | |
598 | if (list_empty(free_me)) | |
599 | return; | |
600 | ||
4541d16c | 601 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 602 | list_del(&lseg->pls_list); |
905ca191 | 603 | pnfs_free_lseg(lseg); |
974cec8c AA |
604 | } |
605 | } | |
606 | ||
e5e94017 BH |
607 | void |
608 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
609 | { | |
610 | struct pnfs_layout_hdr *lo; | |
974cec8c | 611 | LIST_HEAD(tmp_list); |
e5e94017 BH |
612 | |
613 | spin_lock(&nfsi->vfs_inode.i_lock); | |
614 | lo = nfsi->layout; | |
615 | if (lo) { | |
38511722 | 616 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 617 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
3e621214 TM |
618 | pnfs_get_layout_hdr(lo); |
619 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); | |
620 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
c829013d | 621 | pnfs_clear_retry_layoutget(lo); |
3e621214 TM |
622 | spin_unlock(&nfsi->vfs_inode.i_lock); |
623 | pnfs_free_lseg_list(&tmp_list); | |
624 | pnfs_put_layout_hdr(lo); | |
625 | } else | |
626 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 627 | } |
041245c8 | 628 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 629 | |
fd9a8d71 TM |
630 | static bool |
631 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
632 | struct list_head *layout_list) | |
974cec8c AA |
633 | { |
634 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 635 | bool ret = false; |
974cec8c | 636 | |
fd9a8d71 TM |
637 | spin_lock(&inode->i_lock); |
638 | lo = NFS_I(inode)->layout; | |
639 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
640 | pnfs_get_layout_hdr(lo); | |
641 | list_add(&lo->plh_bulk_destroy, layout_list); | |
642 | ret = true; | |
643 | } | |
644 | spin_unlock(&inode->i_lock); | |
645 | return ret; | |
646 | } | |
647 | ||
648 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
649 | static int | |
650 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
651 | struct nfs_server *server, | |
652 | struct list_head *layout_list) | |
653 | { | |
654 | struct pnfs_layout_hdr *lo, *next; | |
655 | struct inode *inode; | |
656 | ||
657 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
658 | inode = igrab(lo->plh_inode); | |
659 | if (inode == NULL) | |
660 | continue; | |
661 | list_del_init(&lo->plh_layouts); | |
662 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
663 | continue; | |
664 | rcu_read_unlock(); | |
665 | spin_unlock(&clp->cl_lock); | |
666 | iput(inode); | |
667 | spin_lock(&clp->cl_lock); | |
668 | rcu_read_lock(); | |
669 | return -EAGAIN; | |
670 | } | |
671 | return 0; | |
672 | } | |
673 | ||
674 | static int | |
675 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
676 | bool is_bulk_recall) | |
677 | { | |
678 | struct pnfs_layout_hdr *lo; | |
679 | struct inode *inode; | |
680 | struct pnfs_layout_range range = { | |
681 | .iomode = IOMODE_ANY, | |
682 | .offset = 0, | |
683 | .length = NFS4_MAX_UINT64, | |
684 | }; | |
685 | LIST_HEAD(lseg_list); | |
686 | int ret = 0; | |
687 | ||
688 | while (!list_empty(layout_list)) { | |
689 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
690 | plh_bulk_destroy); | |
691 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
692 | lo->plh_inode->i_ino); | |
693 | inode = lo->plh_inode; | |
7c5d1875 CH |
694 | |
695 | pnfs_layoutcommit_inode(inode, false); | |
696 | ||
fd9a8d71 TM |
697 | spin_lock(&inode->i_lock); |
698 | list_del_init(&lo->plh_bulk_destroy); | |
699 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ | |
700 | if (is_bulk_recall) | |
701 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
702 | if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range)) | |
703 | ret = -EAGAIN; | |
704 | spin_unlock(&inode->i_lock); | |
705 | pnfs_free_lseg_list(&lseg_list); | |
b20135d0 TM |
706 | /* Free all lsegs that are attached to commit buckets */ |
707 | nfs_commit_inode(inode, 0); | |
fd9a8d71 TM |
708 | pnfs_put_layout_hdr(lo); |
709 | iput(inode); | |
710 | } | |
711 | return ret; | |
712 | } | |
713 | ||
714 | int | |
715 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
716 | struct nfs_fsid *fsid, | |
717 | bool is_recall) | |
718 | { | |
719 | struct nfs_server *server; | |
720 | LIST_HEAD(layout_list); | |
c47abcf8 | 721 | |
974cec8c | 722 | spin_lock(&clp->cl_lock); |
6382a441 | 723 | rcu_read_lock(); |
fd9a8d71 | 724 | restart: |
6382a441 | 725 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
726 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
727 | continue; | |
728 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
729 | server, | |
730 | &layout_list) != 0) | |
731 | goto restart; | |
6382a441 WAA |
732 | } |
733 | rcu_read_unlock(); | |
974cec8c AA |
734 | spin_unlock(&clp->cl_lock); |
735 | ||
fd9a8d71 TM |
736 | if (list_empty(&layout_list)) |
737 | return 0; | |
738 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
739 | } | |
740 | ||
741 | int | |
742 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
743 | bool is_recall) | |
744 | { | |
745 | struct nfs_server *server; | |
746 | LIST_HEAD(layout_list); | |
747 | ||
748 | spin_lock(&clp->cl_lock); | |
749 | rcu_read_lock(); | |
750 | restart: | |
751 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
752 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
753 | server, | |
754 | &layout_list) != 0) | |
755 | goto restart; | |
974cec8c | 756 | } |
fd9a8d71 TM |
757 | rcu_read_unlock(); |
758 | spin_unlock(&clp->cl_lock); | |
759 | ||
760 | if (list_empty(&layout_list)) | |
761 | return 0; | |
762 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
763 | } | |
764 | ||
765 | /* | |
766 | * Called by the state manger to remove all layouts established under an | |
767 | * expired lease. | |
768 | */ | |
769 | void | |
770 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
771 | { | |
772 | nfs4_deviceid_mark_client_invalid(clp); | |
773 | nfs4_deviceid_purge_client(clp); | |
774 | ||
775 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
776 | } |
777 | ||
5a65503f TM |
778 | /* |
779 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
780 | * taking into account wraparound issues. | |
781 | */ | |
782 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
783 | { | |
2c64c57d | 784 | return (s32)(s1 - s2) > 0; |
5a65503f TM |
785 | } |
786 | ||
fd6002e9 | 787 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
788 | void |
789 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
790 | bool update_barrier) | |
b1f69b75 | 791 | { |
22aaf714 TM |
792 | u32 oldseq, newseq, new_barrier; |
793 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 794 | |
2d2f24ad TM |
795 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
796 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 797 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 798 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 799 | if (update_barrier) { |
22aaf714 | 800 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
801 | } else { |
802 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 803 | * "close" to the current seqids. |
43f1b3da | 804 | */ |
22aaf714 | 805 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 806 | } |
22aaf714 TM |
807 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
808 | lo->plh_barrier = new_barrier; | |
43f1b3da | 809 | } |
b1f69b75 AA |
810 | } |
811 | ||
cf7d63f1 | 812 | static bool |
19c54aba TM |
813 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
814 | const nfs4_stateid *stateid) | |
43f1b3da | 815 | { |
19c54aba | 816 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 817 | |
19c54aba TM |
818 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
819 | } | |
820 | ||
821 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
822 | static bool | |
e1c06f80 | 823 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo) |
19c54aba | 824 | { |
f7e8917a | 825 | return lo->plh_block_lgets || |
e1c06f80 | 826 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
cf7d63f1 FI |
827 | } |
828 | ||
fd6002e9 FI |
829 | int |
830 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
ce6ab4f2 | 831 | struct pnfs_layout_range *range, |
fd6002e9 | 832 | struct nfs4_state *open_state) |
b1f69b75 | 833 | { |
fd6002e9 | 834 | int status = 0; |
974cec8c | 835 | |
b1f69b75 | 836 | dprintk("--> %s\n", __func__); |
fd6002e9 | 837 | spin_lock(&lo->plh_inode->i_lock); |
e1c06f80 | 838 | if (pnfs_layoutgets_blocked(lo)) { |
cf7d63f1 | 839 | status = -EAGAIN; |
5d422301 TM |
840 | } else if (!nfs4_valid_open_stateid(open_state)) { |
841 | status = -EBADF; | |
47abadef CH |
842 | } else if (list_empty(&lo->plh_segs) || |
843 | test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { | |
fd6002e9 FI |
844 | int seq; |
845 | ||
846 | do { | |
847 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 848 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
849 | } while (read_seqretry(&open_state->seqlock, seq)); |
850 | } else | |
f597c537 | 851 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 852 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 853 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 854 | return status; |
b1f69b75 AA |
855 | } |
856 | ||
857 | /* | |
858 | * Get layout from server. | |
859 | * for now, assume that whole file layouts are requested. | |
860 | * arg->offset: 0 | |
861 | * arg->length: all ones | |
862 | */ | |
e5e94017 BH |
863 | static struct pnfs_layout_segment * |
864 | send_layoutget(struct pnfs_layout_hdr *lo, | |
865 | struct nfs_open_context *ctx, | |
fb3296eb | 866 | struct pnfs_layout_range *range, |
a75b9df9 | 867 | gfp_t gfp_flags) |
e5e94017 | 868 | { |
b7edfaa1 | 869 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
870 | struct nfs_server *server = NFS_SERVER(ino); |
871 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 872 | struct pnfs_layout_segment *lseg; |
2d89a1d3 | 873 | loff_t i_size; |
b1f69b75 AA |
874 | |
875 | dprintk("--> %s\n", __func__); | |
e5e94017 | 876 | |
4f2e9dce JL |
877 | /* |
878 | * Synchronously retrieve layout information from server and | |
879 | * store in lseg. If we race with a concurrent seqid morphing | |
880 | * op, then re-send the LAYOUTGET. | |
881 | */ | |
882 | do { | |
883 | lgp = kzalloc(sizeof(*lgp), gfp_flags); | |
884 | if (lgp == NULL) | |
885 | return NULL; | |
886 | ||
887 | i_size = i_size_read(ino); | |
888 | ||
889 | lgp->args.minlength = PAGE_CACHE_SIZE; | |
890 | if (lgp->args.minlength > range->length) | |
891 | lgp->args.minlength = range->length; | |
892 | if (range->iomode == IOMODE_READ) { | |
893 | if (range->offset >= i_size) | |
894 | lgp->args.minlength = 0; | |
895 | else if (i_size - range->offset < lgp->args.minlength) | |
896 | lgp->args.minlength = i_size - range->offset; | |
897 | } | |
898 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; | |
899 | lgp->args.range = *range; | |
900 | lgp->args.type = server->pnfs_curr_ld->id; | |
901 | lgp->args.inode = ino; | |
902 | lgp->args.ctx = get_nfs_open_context(ctx); | |
903 | lgp->gfp_flags = gfp_flags; | |
904 | lgp->cred = lo->plh_lc_cred; | |
35124a09 | 905 | |
4f2e9dce JL |
906 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
907 | } while (lseg == ERR_PTR(-EAGAIN)); | |
2d89a1d3 | 908 | |
a0b0a6e3 TM |
909 | if (IS_ERR(lseg)) { |
910 | switch (PTR_ERR(lseg)) { | |
911 | case -ENOMEM: | |
912 | case -ERESTARTSYS: | |
913 | break; | |
914 | default: | |
915 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 916 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
917 | } |
918 | return NULL; | |
d67ae825 TH |
919 | } else |
920 | pnfs_layout_clear_fail_bit(lo, | |
921 | pnfs_iomode_to_fail_bit(range->iomode)); | |
35124a09 | 922 | |
974cec8c AA |
923 | return lseg; |
924 | } | |
925 | ||
24956804 TM |
926 | static void pnfs_clear_layoutcommit(struct inode *inode, |
927 | struct list_head *head) | |
928 | { | |
929 | struct nfs_inode *nfsi = NFS_I(inode); | |
930 | struct pnfs_layout_segment *lseg, *tmp; | |
931 | ||
932 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
933 | return; | |
934 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
935 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
936 | continue; | |
937 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
938 | } | |
939 | } | |
940 | ||
d67ae825 TH |
941 | void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo) |
942 | { | |
943 | clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
944 | smp_mb__after_atomic(); | |
945 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN); | |
7f27392c | 946 | rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq); |
d67ae825 TH |
947 | } |
948 | ||
f40eb5d0 | 949 | static int |
ed429d6b | 950 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid, |
6c16605d | 951 | enum pnfs_iomode iomode, bool sync) |
f40eb5d0 PT |
952 | { |
953 | struct inode *ino = lo->plh_inode; | |
954 | struct nfs4_layoutreturn *lrp; | |
955 | int status = 0; | |
956 | ||
e4af440a | 957 | lrp = kzalloc(sizeof(*lrp), GFP_NOFS); |
f40eb5d0 PT |
958 | if (unlikely(lrp == NULL)) { |
959 | status = -ENOMEM; | |
960 | spin_lock(&ino->i_lock); | |
d67ae825 | 961 | pnfs_clear_layoutreturn_waitbit(lo); |
f40eb5d0 PT |
962 | spin_unlock(&ino->i_lock); |
963 | pnfs_put_layout_hdr(lo); | |
964 | goto out; | |
965 | } | |
966 | ||
ed429d6b | 967 | nfs4_stateid_copy(&lrp->args.stateid, stateid); |
f40eb5d0 PT |
968 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; |
969 | lrp->args.inode = ino; | |
15eb67c1 PT |
970 | lrp->args.range.iomode = iomode; |
971 | lrp->args.range.offset = 0; | |
972 | lrp->args.range.length = NFS4_MAX_UINT64; | |
f40eb5d0 PT |
973 | lrp->args.layout = lo; |
974 | lrp->clp = NFS_SERVER(ino)->nfs_client; | |
975 | lrp->cred = lo->plh_lc_cred; | |
976 | ||
6c16605d | 977 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
978 | out: |
979 | dprintk("<-- %s status: %d\n", __func__, status); | |
980 | return status; | |
981 | } | |
982 | ||
293b3b06 AA |
983 | /* |
984 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
985 | * when the layout segment list is empty. | |
986 | * | |
987 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
988 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
989 | * deviceid is marked invalid. | |
990 | */ | |
cbe82603 BH |
991 | int |
992 | _pnfs_return_layout(struct inode *ino) | |
993 | { | |
994 | struct pnfs_layout_hdr *lo = NULL; | |
995 | struct nfs_inode *nfsi = NFS_I(ino); | |
996 | LIST_HEAD(tmp_list); | |
cbe82603 | 997 | nfs4_stateid stateid; |
293b3b06 | 998 | int status = 0, empty; |
7f27392c | 999 | bool send; |
cbe82603 | 1000 | |
366d5052 | 1001 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
1002 | |
1003 | spin_lock(&ino->i_lock); | |
1004 | lo = nfsi->layout; | |
e5929f3c | 1005 | if (!lo) { |
cbe82603 | 1006 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
1007 | dprintk("NFS: %s no layout to return\n", __func__); |
1008 | goto out; | |
cbe82603 | 1009 | } |
50f563ef | 1010 | nfs4_stateid_copy(&stateid, &nfsi->layout->plh_stateid); |
cbe82603 | 1011 | /* Reference matched in nfs4_layoutreturn_release */ |
70c3bd2b | 1012 | pnfs_get_layout_hdr(lo); |
293b3b06 | 1013 | empty = list_empty(&lo->plh_segs); |
24956804 | 1014 | pnfs_clear_layoutcommit(ino, &tmp_list); |
49a85061 | 1015 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
c88953d8 CH |
1016 | |
1017 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
1018 | struct pnfs_layout_range range = { | |
1019 | .iomode = IOMODE_ANY, | |
1020 | .offset = 0, | |
1021 | .length = NFS4_MAX_UINT64, | |
1022 | }; | |
1023 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
1024 | } | |
1025 | ||
293b3b06 AA |
1026 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
1027 | if (empty) { | |
1028 | spin_unlock(&ino->i_lock); | |
293b3b06 | 1029 | dprintk("NFS: %s no layout segments to return\n", __func__); |
7f27392c | 1030 | goto out_put_layout_hdr; |
293b3b06 | 1031 | } |
47abadef CH |
1032 | |
1033 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
7f27392c | 1034 | send = pnfs_prepare_layoutreturn(lo); |
cbe82603 BH |
1035 | spin_unlock(&ino->i_lock); |
1036 | pnfs_free_lseg_list(&tmp_list); | |
7f27392c | 1037 | if (send) |
ed429d6b | 1038 | status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
7f27392c TM |
1039 | out_put_layout_hdr: |
1040 | pnfs_put_layout_hdr(lo); | |
cbe82603 BH |
1041 | out: |
1042 | dprintk("<-- %s status: %d\n", __func__, status); | |
1043 | return status; | |
1044 | } | |
0a57cdac | 1045 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 1046 | |
24028672 TM |
1047 | int |
1048 | pnfs_commit_and_return_layout(struct inode *inode) | |
1049 | { | |
1050 | struct pnfs_layout_hdr *lo; | |
1051 | int ret; | |
1052 | ||
1053 | spin_lock(&inode->i_lock); | |
1054 | lo = NFS_I(inode)->layout; | |
1055 | if (lo == NULL) { | |
1056 | spin_unlock(&inode->i_lock); | |
1057 | return 0; | |
1058 | } | |
1059 | pnfs_get_layout_hdr(lo); | |
1060 | /* Block new layoutgets and read/write to ds */ | |
1061 | lo->plh_block_lgets++; | |
1062 | spin_unlock(&inode->i_lock); | |
1063 | filemap_fdatawait(inode->i_mapping); | |
1064 | ret = pnfs_layoutcommit_inode(inode, true); | |
1065 | if (ret == 0) | |
1066 | ret = _pnfs_return_layout(inode); | |
1067 | spin_lock(&inode->i_lock); | |
1068 | lo->plh_block_lgets--; | |
1069 | spin_unlock(&inode->i_lock); | |
1070 | pnfs_put_layout_hdr(lo); | |
1071 | return ret; | |
1072 | } | |
1073 | ||
f7e8917a FI |
1074 | bool pnfs_roc(struct inode *ino) |
1075 | { | |
40dd4b7a TM |
1076 | struct nfs_inode *nfsi = NFS_I(ino); |
1077 | struct nfs_open_context *ctx; | |
1078 | struct nfs4_state *state; | |
f7e8917a FI |
1079 | struct pnfs_layout_hdr *lo; |
1080 | struct pnfs_layout_segment *lseg, *tmp; | |
193e3aa2 | 1081 | nfs4_stateid stateid; |
f7e8917a | 1082 | LIST_HEAD(tmp_list); |
e755d638 | 1083 | bool found = false, layoutreturn = false, roc = false; |
f7e8917a FI |
1084 | |
1085 | spin_lock(&ino->i_lock); | |
40dd4b7a | 1086 | lo = nfsi->layout; |
3976143b | 1087 | if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) |
40dd4b7a TM |
1088 | goto out_noroc; |
1089 | ||
e755d638 | 1090 | /* no roc if we hold a delegation */ |
40dd4b7a TM |
1091 | if (nfs4_check_delegation(ino, FMODE_READ)) |
1092 | goto out_noroc; | |
1093 | ||
1094 | list_for_each_entry(ctx, &nfsi->open_files, list) { | |
1095 | state = ctx->state; | |
1096 | /* Don't return layout if there is open file state */ | |
1097 | if (state != NULL && state->state != 0) | |
1098 | goto out_noroc; | |
1099 | } | |
1100 | ||
50f563ef | 1101 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
e755d638 PT |
1102 | /* always send layoutreturn if being marked so */ |
1103 | if (test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, | |
1104 | &lo->plh_flags)) | |
1105 | layoutreturn = pnfs_prepare_layoutreturn(lo); | |
1106 | ||
c829013d | 1107 | pnfs_clear_retry_layoutget(lo); |
f7e8917a | 1108 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) |
e755d638 PT |
1109 | /* If we are sending layoutreturn, invalidate all valid lsegs */ |
1110 | if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
f7e8917a FI |
1111 | mark_lseg_invalid(lseg, &tmp_list); |
1112 | found = true; | |
1113 | } | |
500d701f | 1114 | /* ROC in two conditions: |
e755d638 PT |
1115 | * 1. there are ROC lsegs |
1116 | * 2. we don't send layoutreturn | |
e755d638 | 1117 | */ |
500d701f PT |
1118 | if (found && !layoutreturn) { |
1119 | /* lo ref dropped in pnfs_roc_release() */ | |
1120 | pnfs_get_layout_hdr(lo); | |
e755d638 | 1121 | roc = true; |
500d701f | 1122 | } |
f7e8917a | 1123 | |
40dd4b7a | 1124 | out_noroc: |
f7e8917a | 1125 | spin_unlock(&ino->i_lock); |
e755d638 PT |
1126 | pnfs_free_lseg_list(&tmp_list); |
1127 | pnfs_layoutcommit_inode(ino, true); | |
1128 | if (layoutreturn) | |
ed429d6b | 1129 | pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
e755d638 | 1130 | return roc; |
f7e8917a FI |
1131 | } |
1132 | ||
1133 | void pnfs_roc_release(struct inode *ino) | |
1134 | { | |
1135 | struct pnfs_layout_hdr *lo; | |
1136 | ||
1137 | spin_lock(&ino->i_lock); | |
1138 | lo = NFS_I(ino)->layout; | |
5c4a79fb | 1139 | pnfs_clear_layoutreturn_waitbit(lo); |
6622c3ea TM |
1140 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
1141 | pnfs_detach_layout_hdr(lo); | |
1142 | spin_unlock(&ino->i_lock); | |
1143 | pnfs_free_layout_hdr(lo); | |
1144 | } else | |
1145 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
1146 | } |
1147 | ||
1148 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
1149 | { | |
1150 | struct pnfs_layout_hdr *lo; | |
1151 | ||
1152 | spin_lock(&ino->i_lock); | |
1153 | lo = NFS_I(ino)->layout; | |
0654cc72 | 1154 | pnfs_mark_layout_returned_if_empty(lo); |
0f35ad6f | 1155 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
1156 | lo->plh_barrier = barrier; |
1157 | spin_unlock(&ino->i_lock); | |
6a463beb | 1158 | trace_nfs4_layoutreturn_on_close(ino, 0); |
f7e8917a FI |
1159 | } |
1160 | ||
4ff376fe | 1161 | void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier) |
f7e8917a FI |
1162 | { |
1163 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 1164 | struct pnfs_layout_hdr *lo; |
7fdab069 | 1165 | u32 current_seqid; |
f7e8917a FI |
1166 | |
1167 | spin_lock(&ino->i_lock); | |
7fdab069 TM |
1168 | lo = nfsi->layout; |
1169 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1170 | |
7fdab069 TM |
1171 | /* Since close does not return a layout stateid for use as |
1172 | * a barrier, we choose the worst-case barrier. | |
1173 | */ | |
1174 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
f7e8917a | 1175 | spin_unlock(&ino->i_lock); |
f7e8917a FI |
1176 | } |
1177 | ||
500d701f PT |
1178 | bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task) |
1179 | { | |
1180 | struct nfs_inode *nfsi = NFS_I(ino); | |
1181 | struct pnfs_layout_hdr *lo; | |
1182 | bool sleep = false; | |
1183 | ||
1184 | /* we might not have grabbed lo reference. so need to check under | |
1185 | * i_lock */ | |
1186 | spin_lock(&ino->i_lock); | |
1187 | lo = nfsi->layout; | |
1188 | if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) | |
1189 | sleep = true; | |
1190 | spin_unlock(&ino->i_lock); | |
1191 | ||
1192 | if (sleep) | |
1193 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
1194 | ||
1195 | return sleep; | |
1196 | } | |
1197 | ||
b1f69b75 AA |
1198 | /* |
1199 | * Compare two layout segments for sorting into layout cache. | |
1200 | * We want to preferentially return RW over RO layouts, so ensure those | |
1201 | * are seen first. | |
1202 | */ | |
1203 | static s64 | |
7dc0ac70 | 1204 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1205 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1206 | { |
fb3296eb BH |
1207 | s64 d; |
1208 | ||
1209 | /* high offset > low offset */ | |
1210 | d = l1->offset - l2->offset; | |
1211 | if (d) | |
1212 | return d; | |
1213 | ||
1214 | /* short length > long length */ | |
1215 | d = l2->length - l1->length; | |
1216 | if (d) | |
1217 | return d; | |
1218 | ||
b1f69b75 | 1219 | /* read > read/write */ |
fb3296eb | 1220 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1221 | } |
1222 | ||
03772d2f TM |
1223 | static bool |
1224 | pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1, | |
1225 | const struct pnfs_layout_range *l2) | |
1226 | { | |
1227 | return pnfs_lseg_range_cmp(l1, l2) > 0; | |
1228 | } | |
1229 | ||
1230 | static bool | |
1231 | pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg, | |
1232 | struct pnfs_layout_segment *old) | |
1233 | { | |
1234 | return false; | |
1235 | } | |
1236 | ||
1237 | void | |
1238 | pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1239 | struct pnfs_layout_segment *lseg, | |
1240 | bool (*is_after)(const struct pnfs_layout_range *, | |
1241 | const struct pnfs_layout_range *), | |
1242 | bool (*do_merge)(struct pnfs_layout_segment *, | |
1243 | struct pnfs_layout_segment *), | |
1244 | struct list_head *free_me) | |
974cec8c | 1245 | { |
03772d2f | 1246 | struct pnfs_layout_segment *lp, *tmp; |
b1f69b75 | 1247 | |
974cec8c AA |
1248 | dprintk("%s:Begin\n", __func__); |
1249 | ||
03772d2f TM |
1250 | list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) { |
1251 | if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0) | |
1252 | continue; | |
1253 | if (do_merge(lseg, lp)) { | |
1254 | mark_lseg_invalid(lp, free_me); | |
1255 | continue; | |
1256 | } | |
1257 | if (is_after(&lseg->pls_range, &lp->pls_range)) | |
b1f69b75 | 1258 | continue; |
566052c5 | 1259 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1260 | dprintk("%s: inserted lseg %p " |
1261 | "iomode %d offset %llu length %llu before " | |
1262 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1263 | __func__, lseg, lseg->pls_range.iomode, |
1264 | lseg->pls_range.offset, lseg->pls_range.length, | |
1265 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1266 | lp->pls_range.length); | |
fb3296eb | 1267 | goto out; |
974cec8c | 1268 | } |
fb3296eb BH |
1269 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1270 | dprintk("%s: inserted lseg %p " | |
1271 | "iomode %d offset %llu length %llu at tail\n", | |
1272 | __func__, lseg, lseg->pls_range.iomode, | |
1273 | lseg->pls_range.offset, lseg->pls_range.length); | |
1274 | out: | |
70c3bd2b | 1275 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1276 | |
1277 | dprintk("%s:Return\n", __func__); | |
e5e94017 | 1278 | } |
03772d2f TM |
1279 | EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg); |
1280 | ||
1281 | static void | |
1282 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1283 | struct pnfs_layout_segment *lseg, | |
1284 | struct list_head *free_me) | |
1285 | { | |
1286 | struct inode *inode = lo->plh_inode; | |
1287 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
1288 | ||
1289 | if (ld->add_lseg != NULL) | |
1290 | ld->add_lseg(lo, lseg, free_me); | |
1291 | else | |
1292 | pnfs_generic_layout_insert_lseg(lo, lseg, | |
1293 | pnfs_lseg_range_is_after, | |
1294 | pnfs_lseg_no_merge, | |
1295 | free_me); | |
1296 | } | |
e5e94017 BH |
1297 | |
1298 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1299 | alloc_init_layout_hdr(struct inode *ino, |
1300 | struct nfs_open_context *ctx, | |
1301 | gfp_t gfp_flags) | |
e5e94017 BH |
1302 | { |
1303 | struct pnfs_layout_hdr *lo; | |
1304 | ||
636fb9c8 | 1305 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1306 | if (!lo) |
1307 | return NULL; | |
cc6e5340 | 1308 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1309 | INIT_LIST_HEAD(&lo->plh_layouts); |
1310 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1311 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1312 | lo->plh_inode = ino; |
5cc2216d | 1313 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
e5e94017 BH |
1314 | return lo; |
1315 | } | |
1316 | ||
1317 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1318 | pnfs_find_alloc_layout(struct inode *ino, |
1319 | struct nfs_open_context *ctx, | |
1320 | gfp_t gfp_flags) | |
e5e94017 BH |
1321 | { |
1322 | struct nfs_inode *nfsi = NFS_I(ino); | |
1323 | struct pnfs_layout_hdr *new = NULL; | |
1324 | ||
1325 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1326 | ||
251ec410 TM |
1327 | if (nfsi->layout != NULL) |
1328 | goto out_existing; | |
e5e94017 | 1329 | spin_unlock(&ino->i_lock); |
9fa40758 | 1330 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1331 | spin_lock(&ino->i_lock); |
1332 | ||
251ec410 | 1333 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1334 | nfsi->layout = new; |
251ec410 | 1335 | return new; |
7175fe90 YN |
1336 | } else if (new != NULL) |
1337 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1338 | out_existing: |
1339 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1340 | return nfsi->layout; |
1341 | } | |
1342 | ||
b1f69b75 AA |
1343 | /* |
1344 | * iomode matching rules: | |
1345 | * iomode lseg match | |
1346 | * ----- ----- ----- | |
1347 | * ANY READ true | |
1348 | * ANY RW true | |
1349 | * RW READ false | |
1350 | * RW RW true | |
1351 | * READ READ true | |
1352 | * READ RW true | |
1353 | */ | |
3cb2df17 | 1354 | static bool |
7dc0ac70 | 1355 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
3cb2df17 | 1356 | const struct pnfs_layout_range *range) |
b1f69b75 | 1357 | { |
fb3296eb BH |
1358 | struct pnfs_layout_range range1; |
1359 | ||
1360 | if ((range->iomode == IOMODE_RW && | |
1361 | ls_range->iomode != IOMODE_RW) || | |
7dc0ac70 | 1362 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1363 | return 0; |
1364 | ||
1365 | /* range1 covers only the first byte in the range */ | |
1366 | range1 = *range; | |
1367 | range1.length = 1; | |
7dc0ac70 | 1368 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1369 | } |
1370 | ||
1371 | /* | |
1372 | * lookup range in layout | |
1373 | */ | |
e5e94017 | 1374 | static struct pnfs_layout_segment * |
fb3296eb BH |
1375 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
1376 | struct pnfs_layout_range *range) | |
e5e94017 | 1377 | { |
b1f69b75 AA |
1378 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1379 | ||
1380 | dprintk("%s:Begin\n", __func__); | |
1381 | ||
b7edfaa1 | 1382 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1383 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1384 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
7dc0ac70 | 1385 | pnfs_lseg_range_match(&lseg->pls_range, range)) { |
9369a431 | 1386 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1387 | break; |
1388 | } | |
b1f69b75 AA |
1389 | } |
1390 | ||
1391 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1392 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1393 | return ret; |
e5e94017 BH |
1394 | } |
1395 | ||
d23d61c8 AA |
1396 | /* |
1397 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1398 | * to the MDS or over pNFS | |
1399 | * | |
1400 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1401 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1402 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1403 | * WRITE request. | |
1404 | * | |
1405 | * A return of true means use MDS I/O. | |
1406 | * | |
1407 | * From rfc 5661: | |
1408 | * If a file's size is smaller than the file size threshold, data accesses | |
1409 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1410 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1411 | * server. If both file size and I/O size are provided, the client SHOULD | |
1412 | * reach or exceed both thresholds before sending its read or write | |
1413 | * requests to the data server. | |
1414 | */ | |
1415 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1416 | struct inode *ino, int iomode) | |
1417 | { | |
1418 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1419 | struct nfs_inode *nfsi = NFS_I(ino); | |
1420 | loff_t fsize = i_size_read(ino); | |
1421 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1422 | ||
1423 | if (t == NULL) | |
1424 | return ret; | |
1425 | ||
1426 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1427 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1428 | ||
1429 | switch (iomode) { | |
1430 | case IOMODE_READ: | |
1431 | if (t->bm & THRESHOLD_RD) { | |
1432 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1433 | size_set = true; | |
1434 | if (fsize < t->rd_sz) | |
1435 | size = true; | |
1436 | } | |
1437 | if (t->bm & THRESHOLD_RD_IO) { | |
1438 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1439 | nfsi->read_io); | |
1440 | io_set = true; | |
1441 | if (nfsi->read_io < t->rd_io_sz) | |
1442 | io = true; | |
1443 | } | |
1444 | break; | |
1445 | case IOMODE_RW: | |
1446 | if (t->bm & THRESHOLD_WR) { | |
1447 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1448 | size_set = true; | |
1449 | if (fsize < t->wr_sz) | |
1450 | size = true; | |
1451 | } | |
1452 | if (t->bm & THRESHOLD_WR_IO) { | |
1453 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1454 | nfsi->write_io); | |
1455 | io_set = true; | |
1456 | if (nfsi->write_io < t->wr_io_sz) | |
1457 | io = true; | |
1458 | } | |
1459 | break; | |
1460 | } | |
1461 | if (size_set && io_set) { | |
1462 | if (size && io) | |
1463 | ret = true; | |
1464 | } else if (size || io) | |
1465 | ret = true; | |
1466 | ||
1467 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1468 | return ret; | |
1469 | } | |
1470 | ||
aa8a45ee | 1471 | /* stop waiting if someone clears NFS_LAYOUT_RETRY_LAYOUTGET bit. */ |
dfd01f02 | 1472 | static int pnfs_layoutget_retry_bit_wait(struct wait_bit_key *key, int mode) |
aa8a45ee PT |
1473 | { |
1474 | if (!test_bit(NFS_LAYOUT_RETRY_LAYOUTGET, key->flags)) | |
1475 | return 1; | |
dfd01f02 | 1476 | return nfs_wait_bit_killable(key, mode); |
aa8a45ee PT |
1477 | } |
1478 | ||
1479 | static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo) | |
1480 | { | |
8f70f53a TM |
1481 | if (!pnfs_should_retry_layoutget(lo)) |
1482 | return false; | |
aa8a45ee PT |
1483 | /* |
1484 | * send layoutcommit as it can hold up layoutreturn due to lseg | |
1485 | * reference | |
1486 | */ | |
1487 | pnfs_layoutcommit_inode(lo->plh_inode, false); | |
1488 | return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
1489 | pnfs_layoutget_retry_bit_wait, | |
1490 | TASK_UNINTERRUPTIBLE); | |
1491 | } | |
1492 | ||
d67ae825 TH |
1493 | static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo) |
1494 | { | |
1495 | unsigned long *bitlock = &lo->plh_flags; | |
1496 | ||
1497 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1498 | smp_mb__after_atomic(); | |
1499 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1500 | } | |
1501 | ||
e5e94017 BH |
1502 | /* |
1503 | * Layout segment is retreived from the server if not cached. | |
1504 | * The appropriate layout segment is referenced and returned to the caller. | |
1505 | */ | |
7c24d948 | 1506 | struct pnfs_layout_segment * |
e5e94017 BH |
1507 | pnfs_update_layout(struct inode *ino, |
1508 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1509 | loff_t pos, |
1510 | u64 count, | |
a75b9df9 TM |
1511 | enum pnfs_iomode iomode, |
1512 | gfp_t gfp_flags) | |
e5e94017 | 1513 | { |
fb3296eb BH |
1514 | struct pnfs_layout_range arg = { |
1515 | .iomode = iomode, | |
1516 | .offset = pos, | |
1517 | .length = count, | |
1518 | }; | |
707ed5fd | 1519 | unsigned pg_offset; |
6382a441 WAA |
1520 | struct nfs_server *server = NFS_SERVER(ino); |
1521 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1522 | struct pnfs_layout_hdr *lo; |
1523 | struct pnfs_layout_segment *lseg = NULL; | |
30005121 | 1524 | bool first; |
e5e94017 | 1525 | |
9a4bf31d | 1526 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) { |
f4848303 | 1527 | trace_pnfs_update_layout(ino, pos, count, iomode, NULL, |
9a4bf31d | 1528 | PNFS_UPDATE_LAYOUT_NO_PNFS); |
f86bbcf8 | 1529 | goto out; |
9a4bf31d | 1530 | } |
d23d61c8 | 1531 | |
9a4bf31d | 1532 | if (iomode == IOMODE_READ && i_size_read(ino) == 0) { |
f4848303 | 1533 | trace_pnfs_update_layout(ino, pos, count, iomode, NULL, |
9a4bf31d | 1534 | PNFS_UPDATE_LAYOUT_RD_ZEROLEN); |
4ae93560 | 1535 | goto out; |
9a4bf31d | 1536 | } |
4ae93560 | 1537 | |
9a4bf31d | 1538 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) { |
f4848303 | 1539 | trace_pnfs_update_layout(ino, pos, count, iomode, NULL, |
9a4bf31d | 1540 | PNFS_UPDATE_LAYOUT_MDSTHRESH); |
f86bbcf8 | 1541 | goto out; |
9a4bf31d | 1542 | } |
d23d61c8 | 1543 | |
9bf87482 PT |
1544 | lookup_again: |
1545 | first = false; | |
e5e94017 | 1546 | spin_lock(&ino->i_lock); |
9fa40758 | 1547 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1548 | if (lo == NULL) { |
1549 | spin_unlock(&ino->i_lock); | |
f4848303 | 1550 | trace_pnfs_update_layout(ino, pos, count, iomode, NULL, |
9a4bf31d | 1551 | PNFS_UPDATE_LAYOUT_NOMEM); |
830ffb56 TM |
1552 | goto out; |
1553 | } | |
e5e94017 | 1554 | |
43f1b3da | 1555 | /* Do we even need to bother with this? */ |
a59c30ac | 1556 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
f4848303 | 1557 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1558 | PNFS_UPDATE_LAYOUT_BULK_RECALL); |
43f1b3da | 1559 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1560 | goto out_unlock; |
1561 | } | |
1562 | ||
1563 | /* if LAYOUTGET already failed once we don't try again */ | |
aa8a45ee | 1564 | if (pnfs_layout_io_test_failed(lo, iomode) && |
9a4bf31d | 1565 | !pnfs_should_retry_layoutget(lo)) { |
f4848303 | 1566 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1567 | PNFS_UPDATE_LAYOUT_IO_TEST_FAIL); |
e5e94017 | 1568 | goto out_unlock; |
9a4bf31d | 1569 | } |
e5e94017 | 1570 | |
9bf87482 PT |
1571 | first = list_empty(&lo->plh_segs); |
1572 | if (first) { | |
1573 | /* The first layoutget for the file. Need to serialize per | |
1574 | * RFC 5661 Errata 3208. | |
1575 | */ | |
1576 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1577 | &lo->plh_flags)) { | |
1578 | spin_unlock(&ino->i_lock); | |
1579 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1580 | TASK_UNINTERRUPTIBLE); | |
1581 | pnfs_put_layout_hdr(lo); | |
1582 | goto lookup_again; | |
1583 | } | |
1584 | } else { | |
1585 | /* Check to see if the layout for the given range | |
1586 | * already exists | |
1587 | */ | |
1588 | lseg = pnfs_find_lseg(lo, &arg); | |
9a4bf31d | 1589 | if (lseg) { |
f4848303 | 1590 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1591 | PNFS_UPDATE_LAYOUT_FOUND_CACHED); |
9bf87482 | 1592 | goto out_unlock; |
9a4bf31d | 1593 | } |
9bf87482 | 1594 | } |
568e8c49 | 1595 | |
aa8a45ee PT |
1596 | /* |
1597 | * Because we free lsegs before sending LAYOUTRETURN, we need to wait | |
1598 | * for LAYOUTRETURN even if first is true. | |
1599 | */ | |
8f70f53a | 1600 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
aa8a45ee PT |
1601 | spin_unlock(&ino->i_lock); |
1602 | dprintk("%s wait for layoutreturn\n", __func__); | |
1603 | if (pnfs_prepare_to_retry_layoutget(lo)) { | |
d67ae825 TH |
1604 | if (first) |
1605 | pnfs_clear_first_layoutget(lo); | |
aa8a45ee PT |
1606 | pnfs_put_layout_hdr(lo); |
1607 | dprintk("%s retrying\n", __func__); | |
1608 | goto lookup_again; | |
1609 | } | |
f4848303 | 1610 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1611 | PNFS_UPDATE_LAYOUT_RETURN); |
aa8a45ee PT |
1612 | goto out_put_layout_hdr; |
1613 | } | |
1614 | ||
9a4bf31d | 1615 | if (pnfs_layoutgets_blocked(lo)) { |
f4848303 | 1616 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1617 | PNFS_UPDATE_LAYOUT_BLOCKED); |
cf7d63f1 | 1618 | goto out_unlock; |
9a4bf31d | 1619 | } |
cf7d63f1 | 1620 | atomic_inc(&lo->plh_outstanding); |
f49f9baa | 1621 | spin_unlock(&ino->i_lock); |
30005121 | 1622 | |
abb9a007 | 1623 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1624 | /* The lo must be on the clp list if there is any |
1625 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1626 | */ | |
1627 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1628 | if (list_empty(&lo->plh_layouts)) |
1629 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1630 | spin_unlock(&clp->cl_lock); |
1631 | } | |
e5e94017 | 1632 | |
707ed5fd BH |
1633 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1634 | if (pg_offset) { | |
1635 | arg.offset -= pg_offset; | |
1636 | arg.length += pg_offset; | |
1637 | } | |
7c24d948 AA |
1638 | if (arg.length != NFS4_MAX_UINT64) |
1639 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1640 | |
fb3296eb | 1641 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
c829013d | 1642 | pnfs_clear_retry_layoutget(lo); |
cf7d63f1 | 1643 | atomic_dec(&lo->plh_outstanding); |
f4848303 | 1644 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1645 | PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET); |
830ffb56 | 1646 | out_put_layout_hdr: |
d67ae825 TH |
1647 | if (first) |
1648 | pnfs_clear_first_layoutget(lo); | |
70c3bd2b | 1649 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1650 | out: |
f86bbcf8 TM |
1651 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1652 | "(%s, offset: %llu, length: %llu)\n", | |
1653 | __func__, ino->i_sb->s_id, | |
1654 | (unsigned long long)NFS_FILEID(ino), | |
1655 | lseg == NULL ? "not found" : "found", | |
1656 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1657 | (unsigned long long)pos, | |
1658 | (unsigned long long)count); | |
e5e94017 BH |
1659 | return lseg; |
1660 | out_unlock: | |
1661 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1662 | goto out_put_layout_hdr; |
e5e94017 | 1663 | } |
7c24d948 | 1664 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1665 | |
540d9864 TM |
1666 | static bool |
1667 | pnfs_sanity_check_layout_range(struct pnfs_layout_range *range) | |
1668 | { | |
1669 | switch (range->iomode) { | |
1670 | case IOMODE_READ: | |
1671 | case IOMODE_RW: | |
1672 | break; | |
1673 | default: | |
1674 | return false; | |
1675 | } | |
1676 | if (range->offset == NFS4_MAX_UINT64) | |
1677 | return false; | |
1678 | if (range->length == 0) | |
1679 | return false; | |
1680 | if (range->length != NFS4_MAX_UINT64 && | |
1681 | range->length > NFS4_MAX_UINT64 - range->offset) | |
1682 | return false; | |
1683 | return true; | |
1684 | } | |
1685 | ||
a0b0a6e3 | 1686 | struct pnfs_layout_segment * |
b1f69b75 AA |
1687 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1688 | { | |
1689 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1690 | struct nfs4_layoutget_res *res = &lgp->res; | |
1691 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1692 | struct inode *ino = lo->plh_inode; |
78096cca | 1693 | LIST_HEAD(free_me); |
540d9864 TM |
1694 | int status = -EINVAL; |
1695 | ||
1696 | if (!pnfs_sanity_check_layout_range(&res->range)) | |
1697 | goto out; | |
b1f69b75 AA |
1698 | |
1699 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1700 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1701 | if (!lseg || IS_ERR(lseg)) { |
1702 | if (!lseg) | |
1703 | status = -ENOMEM; | |
1704 | else | |
1705 | status = PTR_ERR(lseg); | |
1706 | dprintk("%s: Could not allocate layout: error %d\n", | |
1707 | __func__, status); | |
1708 | goto out; | |
1709 | } | |
1710 | ||
1013df61 CH |
1711 | init_lseg(lo, lseg); |
1712 | lseg->pls_range = res->range; | |
1713 | ||
b1f69b75 | 1714 | spin_lock(&ino->i_lock); |
e1c06f80 | 1715 | if (pnfs_layoutgets_blocked(lo)) { |
43f1b3da FI |
1716 | dprintk("%s forget reply due to state\n", __func__); |
1717 | goto out_forget_reply; | |
1718 | } | |
038d6493 | 1719 | |
362f7474 CH |
1720 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1721 | /* existing state ID, make sure the sequence number matches. */ | |
1722 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1723 | dprintk("%s forget reply due to sequence\n", __func__); | |
4f2e9dce | 1724 | status = -EAGAIN; |
362f7474 CH |
1725 | goto out_forget_reply; |
1726 | } | |
1727 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1728 | } else { | |
1729 | /* | |
1730 | * We got an entirely new state ID. Mark all segments for the | |
1731 | * inode invalid, and don't bother validating the stateid | |
1732 | * sequence number. | |
1733 | */ | |
1734 | pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL); | |
1735 | ||
1736 | nfs4_stateid_copy(&lo->plh_stateid, &res->stateid); | |
1737 | lo->plh_barrier = be32_to_cpu(res->stateid.seqid); | |
1738 | } | |
038d6493 | 1739 | |
47abadef CH |
1740 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); |
1741 | ||
9369a431 | 1742 | pnfs_get_lseg(lseg); |
03772d2f | 1743 | pnfs_layout_insert_lseg(lo, lseg, &free_me); |
b1f69b75 | 1744 | |
3976143b | 1745 | if (res->return_on_close) |
f7e8917a | 1746 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); |
f7e8917a | 1747 | |
b1f69b75 | 1748 | spin_unlock(&ino->i_lock); |
78096cca | 1749 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1750 | return lseg; |
b1f69b75 | 1751 | out: |
a0b0a6e3 | 1752 | return ERR_PTR(status); |
43f1b3da FI |
1753 | |
1754 | out_forget_reply: | |
1755 | spin_unlock(&ino->i_lock); | |
1756 | lseg->pls_layout = lo; | |
1757 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1758 | goto out; | |
b1f69b75 AA |
1759 | } |
1760 | ||
5c97f5de TM |
1761 | static void |
1762 | pnfs_set_plh_return_iomode(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode) | |
1763 | { | |
1764 | if (lo->plh_return_iomode == iomode) | |
1765 | return; | |
1766 | if (lo->plh_return_iomode != 0) | |
1767 | iomode = IOMODE_ANY; | |
1768 | lo->plh_return_iomode = iomode; | |
1769 | } | |
1770 | ||
fc7ff367 | 1771 | void |
016256df PT |
1772 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, |
1773 | struct list_head *tmp_list, | |
1774 | struct pnfs_layout_range *return_range) | |
1775 | { | |
1776 | struct pnfs_layout_segment *lseg, *next; | |
1777 | ||
1778 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
1779 | ||
1780 | if (list_empty(&lo->plh_segs)) | |
1781 | return; | |
1782 | ||
fc7ff367 TM |
1783 | assert_spin_locked(&lo->plh_inode->i_lock); |
1784 | ||
016256df PT |
1785 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
1786 | if (should_free_lseg(&lseg->pls_range, return_range)) { | |
1787 | dprintk("%s: marking lseg %p iomode %d " | |
1788 | "offset %llu length %llu\n", __func__, | |
1789 | lseg, lseg->pls_range.iomode, | |
1790 | lseg->pls_range.offset, | |
1791 | lseg->pls_range.length); | |
1792 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
5c97f5de | 1793 | pnfs_set_plh_return_iomode(lo, return_range->iomode); |
016256df | 1794 | mark_lseg_invalid(lseg, tmp_list); |
0bdb8fa6 TM |
1795 | set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, |
1796 | &lo->plh_flags); | |
016256df PT |
1797 | } |
1798 | } | |
1799 | ||
1800 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
1801 | struct pnfs_layout_segment *lseg) | |
1802 | { | |
1803 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
1804 | int iomode = pnfs_iomode_to_fail_bit(lseg->pls_range.iomode); | |
1805 | struct pnfs_layout_range range = { | |
1806 | .iomode = lseg->pls_range.iomode, | |
1807 | .offset = 0, | |
1808 | .length = NFS4_MAX_UINT64, | |
1809 | }; | |
1810 | LIST_HEAD(free_me); | |
1811 | ||
1812 | spin_lock(&inode->i_lock); | |
1813 | /* set failure bit so that pnfs path will be retried later */ | |
1814 | pnfs_layout_set_fail_bit(lo, iomode); | |
5c97f5de | 1815 | pnfs_set_plh_return_iomode(lo, range.iomode); |
016256df PT |
1816 | /* |
1817 | * mark all matching lsegs so that we are sure to have no live | |
1818 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
1819 | * for how it works. | |
1820 | */ | |
1821 | pnfs_mark_matching_lsegs_return(lo, &free_me, &range); | |
1822 | spin_unlock(&inode->i_lock); | |
1823 | pnfs_free_lseg_list(&free_me); | |
b20135d0 | 1824 | nfs_commit_inode(inode, 0); |
016256df PT |
1825 | } |
1826 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
1827 | ||
d8007d4d TM |
1828 | void |
1829 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1830 | { | |
1fd937bd PT |
1831 | u64 rd_size = req->wb_bytes; |
1832 | ||
cb5d04bc PT |
1833 | if (pgio->pg_lseg == NULL) { |
1834 | if (pgio->pg_dreq == NULL) | |
1835 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1836 | else | |
1837 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1838 | ||
1839 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1840 | req->wb_context, | |
1841 | req_offset(req), | |
1842 | rd_size, | |
1843 | IOMODE_READ, | |
1844 | GFP_KERNEL); | |
1845 | } | |
e885de1a TM |
1846 | /* If no lseg, fall back to read through mds */ |
1847 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1848 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1849 | |
d8007d4d TM |
1850 | } |
1851 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1852 | ||
1853 | void | |
6296556f PT |
1854 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1855 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1856 | { |
cb5d04bc PT |
1857 | if (pgio->pg_lseg == NULL) |
1858 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1859 | req->wb_context, | |
1860 | req_offset(req), | |
1861 | wb_size, | |
1862 | IOMODE_RW, | |
1863 | GFP_NOFS); | |
e885de1a TM |
1864 | /* If no lseg, fall back to write through mds */ |
1865 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1866 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1867 | } |
1868 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1869 | ||
180bb5ec WAA |
1870 | void |
1871 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
1872 | { | |
1873 | if (desc->pg_lseg) { | |
1874 | pnfs_put_lseg(desc->pg_lseg); | |
1875 | desc->pg_lseg = NULL; | |
1876 | } | |
1877 | } | |
1878 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
1879 | ||
b4fdac1a WAA |
1880 | /* |
1881 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1882 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1883 | */ | |
1884 | size_t | |
a7d42ddb WAA |
1885 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
1886 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 1887 | { |
0f9c429e | 1888 | unsigned int size; |
c5e20cb7 | 1889 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1890 | |
1891 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1892 | if (!size) |
1893 | return 0; | |
94ad1c80 | 1894 | |
19982ba8 | 1895 | /* |
c5e20cb7 WAA |
1896 | * 'size' contains the number of bytes left in the current page (up |
1897 | * to the original size asked for in @req->wb_bytes). | |
1898 | * | |
1899 | * Calculate how many bytes are left in the layout segment | |
1900 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1901 | * |
1902 | * Please also note that 'end_offset' is actually the offset of the | |
1903 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1904 | * | |
1905 | */ | |
19b54848 | 1906 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1907 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1908 | pgio->pg_lseg->pls_range.length); | |
1909 | req_start = req_offset(req); | |
7c13789e | 1910 | WARN_ON_ONCE(req_start >= seg_end); |
c5e20cb7 | 1911 | /* start of request is past the last byte of this segment */ |
7c13789e WAA |
1912 | if (req_start >= seg_end) { |
1913 | /* reference the new lseg */ | |
1914 | if (pgio->pg_ops->pg_cleanup) | |
1915 | pgio->pg_ops->pg_cleanup(pgio); | |
1916 | if (pgio->pg_ops->pg_init) | |
1917 | pgio->pg_ops->pg_init(pgio, req); | |
19b54848 | 1918 | return 0; |
7c13789e | 1919 | } |
c5e20cb7 WAA |
1920 | |
1921 | /* adjust 'size' iff there are fewer bytes left in the | |
1922 | * segment than what nfs_generic_pg_test returned */ | |
1923 | seg_left = seg_end - req_start; | |
1924 | if (seg_left < size) | |
1925 | size = (unsigned int)seg_left; | |
19b54848 | 1926 | } |
0f9c429e | 1927 | |
19b54848 | 1928 | return size; |
94ad1c80 | 1929 | } |
89a58e32 | 1930 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1931 | |
53113ad3 | 1932 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
1933 | { |
1934 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
1935 | |
1936 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
1937 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
1938 | hdr->completion_ops); | |
c7070113 | 1939 | set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags); |
53113ad3 | 1940 | return nfs_pageio_resend(&pgio, hdr); |
e2fecb21 | 1941 | } |
e7dd79af | 1942 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1943 | |
d45f60c6 | 1944 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1945 | { |
cd841605 FI |
1946 | |
1947 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1948 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1949 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1950 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1951 | } |
6c75dc0d | 1952 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1953 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
1954 | } |
1955 | ||
d20581aa BH |
1956 | /* |
1957 | * Called by non rpc-based layout drivers | |
1958 | */ | |
d45f60c6 | 1959 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 1960 | { |
f8417b48 | 1961 | if (likely(!hdr->pnfs_error)) { |
67af7611 TM |
1962 | pnfs_set_layoutcommit(hdr->inode, hdr->lseg, |
1963 | hdr->mds_offset + hdr->res.count); | |
d45f60c6 | 1964 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
1965 | } |
1966 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); | |
1967 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
1968 | pnfs_ld_handle_write_error(hdr); |
1969 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 1970 | } |
d20581aa | 1971 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1972 | |
dce81290 TM |
1973 | static void |
1974 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1975 | struct nfs_pgio_header *hdr) |
dce81290 | 1976 | { |
48d635f1 | 1977 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1978 | |
6c75dc0d | 1979 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 1980 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 1981 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 1982 | mirror->pg_recoalesce = 1; |
6c75dc0d | 1983 | } |
d45f60c6 | 1984 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 1985 | hdr->release(hdr); |
dce81290 TM |
1986 | } |
1987 | ||
1988 | static enum pnfs_try_status | |
d45f60c6 | 1989 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
1990 | const struct rpc_call_ops *call_ops, |
1991 | struct pnfs_layout_segment *lseg, | |
1992 | int how) | |
0382b744 | 1993 | { |
cd841605 | 1994 | struct inode *inode = hdr->inode; |
0382b744 AA |
1995 | enum pnfs_try_status trypnfs; |
1996 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1997 | ||
cd841605 | 1998 | hdr->mds_ops = call_ops; |
0382b744 AA |
1999 | |
2000 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
2001 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
2002 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 2003 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 2004 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
2005 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2006 | return trypnfs; | |
2007 | } | |
2008 | ||
dce81290 | 2009 | static void |
7f714720 WAA |
2010 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
2011 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 2012 | { |
dce81290 TM |
2013 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2014 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2015 | enum pnfs_try_status trypnfs; |
dce81290 | 2016 | |
d45f60c6 | 2017 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 2018 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 2019 | pnfs_write_through_mds(desc, hdr); |
dce81290 TM |
2020 | } |
2021 | ||
6c75dc0d FI |
2022 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
2023 | { | |
9369a431 | 2024 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2025 | nfs_pgio_header_free(hdr); |
6c75dc0d FI |
2026 | } |
2027 | ||
dce81290 TM |
2028 | int |
2029 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
2030 | { | |
48d635f1 | 2031 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2032 | |
6c75dc0d | 2033 | struct nfs_pgio_header *hdr; |
dce81290 TM |
2034 | int ret; |
2035 | ||
1e7f3a48 WAA |
2036 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2037 | if (!hdr) { | |
a7d42ddb | 2038 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); |
6c75dc0d | 2039 | return -ENOMEM; |
dce81290 | 2040 | } |
6c75dc0d | 2041 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 2042 | |
9369a431 | 2043 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2044 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2045 | if (!ret) |
7f714720 | 2046 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 2047 | |
6c75dc0d | 2048 | return ret; |
dce81290 TM |
2049 | } |
2050 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
2051 | ||
53113ad3 | 2052 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
2053 | { |
2054 | struct nfs_pageio_descriptor pgio; | |
2055 | ||
1acbbb4e | 2056 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
2057 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
2058 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 2059 | } |
e7dd79af | 2060 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 2061 | |
d45f60c6 | 2062 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2063 | { |
cd841605 FI |
2064 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
2065 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2066 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2067 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2068 | } |
4db6e0b7 | 2069 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2070 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
2071 | } |
2072 | ||
d20581aa BH |
2073 | /* |
2074 | * Called by non rpc-based layout drivers | |
2075 | */ | |
d45f60c6 | 2076 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 2077 | { |
cd841605 | 2078 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
2079 | __nfs4_read_done_cb(hdr); |
2080 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
f8417b48 KM |
2081 | } |
2082 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); | |
2083 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2084 | pnfs_ld_handle_read_error(hdr); |
2085 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
2086 | } |
2087 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
2088 | ||
493292dd TM |
2089 | static void |
2090 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2091 | struct nfs_pgio_header *hdr) |
493292dd | 2092 | { |
48d635f1 | 2093 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2094 | |
4db6e0b7 | 2095 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2096 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 2097 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 2098 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 2099 | } |
d45f60c6 | 2100 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2101 | hdr->release(hdr); |
493292dd TM |
2102 | } |
2103 | ||
64419a9b AA |
2104 | /* |
2105 | * Call the appropriate parallel I/O subsystem read function. | |
2106 | */ | |
493292dd | 2107 | static enum pnfs_try_status |
d45f60c6 | 2108 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
2109 | const struct rpc_call_ops *call_ops, |
2110 | struct pnfs_layout_segment *lseg) | |
64419a9b | 2111 | { |
cd841605 | 2112 | struct inode *inode = hdr->inode; |
64419a9b AA |
2113 | struct nfs_server *nfss = NFS_SERVER(inode); |
2114 | enum pnfs_try_status trypnfs; | |
2115 | ||
cd841605 | 2116 | hdr->mds_ops = call_ops; |
64419a9b AA |
2117 | |
2118 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 2119 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 2120 | |
d45f60c6 | 2121 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 2122 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 2123 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
2124 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2125 | return trypnfs; | |
2126 | } | |
863a3c6c | 2127 | |
ceb11e13 PT |
2128 | /* Resend all requests through pnfs. */ |
2129 | int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) | |
2130 | { | |
2131 | struct nfs_pageio_descriptor pgio; | |
2132 | ||
2133 | nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops); | |
2134 | return nfs_pageio_resend(&pgio, hdr); | |
2135 | } | |
2136 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
2137 | ||
493292dd | 2138 | static void |
7f714720 | 2139 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 2140 | { |
493292dd TM |
2141 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2142 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2143 | enum pnfs_try_status trypnfs; |
ceb11e13 | 2144 | int err = 0; |
493292dd | 2145 | |
d45f60c6 | 2146 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
ceb11e13 PT |
2147 | if (trypnfs == PNFS_TRY_AGAIN) |
2148 | err = pnfs_read_resend_pnfs(hdr); | |
2149 | if (trypnfs == PNFS_NOT_ATTEMPTED || err) | |
d45f60c6 | 2150 | pnfs_read_through_mds(desc, hdr); |
493292dd TM |
2151 | } |
2152 | ||
4db6e0b7 FI |
2153 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
2154 | { | |
9369a431 | 2155 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2156 | nfs_pgio_header_free(hdr); |
4db6e0b7 FI |
2157 | } |
2158 | ||
493292dd TM |
2159 | int |
2160 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
2161 | { | |
48d635f1 | 2162 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2163 | |
4db6e0b7 | 2164 | struct nfs_pgio_header *hdr; |
493292dd TM |
2165 | int ret; |
2166 | ||
1e7f3a48 WAA |
2167 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2168 | if (!hdr) { | |
a7d42ddb | 2169 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); |
180bb5ec | 2170 | return -ENOMEM; |
493292dd | 2171 | } |
4db6e0b7 | 2172 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 2173 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2174 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2175 | if (!ret) |
7f714720 | 2176 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 2177 | return ret; |
493292dd TM |
2178 | } |
2179 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
2180 | ||
71244d9b TM |
2181 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
2182 | { | |
2183 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
2184 | ||
2185 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 2186 | smp_mb__after_atomic(); |
71244d9b TM |
2187 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
2188 | } | |
2189 | ||
863a3c6c | 2190 | /* |
a9bae566 | 2191 | * There can be multiple RW segments. |
863a3c6c | 2192 | */ |
a9bae566 | 2193 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 2194 | { |
a9bae566 | 2195 | struct pnfs_layout_segment *lseg; |
863a3c6c | 2196 | |
a9bae566 PT |
2197 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
2198 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 2199 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
2200 | list_add(&lseg->pls_lc_list, listp); |
2201 | } | |
863a3c6c AA |
2202 | } |
2203 | ||
a073dbff TM |
2204 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
2205 | { | |
2206 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
2207 | |
2208 | /* Matched by references in pnfs_set_layoutcommit */ | |
2209 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
2210 | list_del_init(&lseg->pls_lc_list); | |
2211 | pnfs_put_lseg(lseg); | |
2212 | } | |
2213 | ||
71244d9b | 2214 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
2215 | } |
2216 | ||
1b0ae068 PT |
2217 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
2218 | { | |
b9e028fd | 2219 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
2220 | } |
2221 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
2222 | ||
863a3c6c | 2223 | void |
67af7611 TM |
2224 | pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, |
2225 | loff_t end_pos) | |
863a3c6c | 2226 | { |
cd841605 | 2227 | struct nfs_inode *nfsi = NFS_I(inode); |
79a48a1f | 2228 | bool mark_as_dirty = false; |
863a3c6c | 2229 | |
cd841605 | 2230 | spin_lock(&inode->i_lock); |
863a3c6c | 2231 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
29559b11 | 2232 | nfsi->layout->plh_lwb = end_pos; |
79a48a1f | 2233 | mark_as_dirty = true; |
863a3c6c | 2234 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 2235 | __func__, inode->i_ino); |
29559b11 TM |
2236 | } else if (end_pos > nfsi->layout->plh_lwb) |
2237 | nfsi->layout->plh_lwb = end_pos; | |
67af7611 | 2238 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) { |
a9bae566 | 2239 | /* references matched in nfs4_layoutcommit_release */ |
67af7611 | 2240 | pnfs_get_lseg(lseg); |
a9bae566 | 2241 | } |
cd841605 | 2242 | spin_unlock(&inode->i_lock); |
acff5880 | 2243 | dprintk("%s: lseg %p end_pos %llu\n", |
67af7611 | 2244 | __func__, lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
2245 | |
2246 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2247 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2248 | if (mark_as_dirty) | |
cd841605 | 2249 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
2250 | } |
2251 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
2252 | ||
db29c089 AA |
2253 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2254 | { | |
2255 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2256 | ||
2257 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2258 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2259 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2260 | } |
2261 | ||
de4b15c7 AA |
2262 | /* |
2263 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2264 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2265 | * data to disk to allow the server to recover the data if it crashes. | |
2266 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2267 | * is off, and a COMMIT is sent to a data server, or | |
2268 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2269 | */ | |
863a3c6c | 2270 | int |
ef311537 | 2271 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2272 | { |
5f919c9f | 2273 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2274 | struct nfs4_layoutcommit_data *data; |
2275 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2276 | loff_t end_pos; |
71244d9b | 2277 | int status; |
863a3c6c | 2278 | |
71244d9b | 2279 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2280 | return 0; |
2281 | ||
71244d9b | 2282 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2283 | |
71244d9b | 2284 | status = -EAGAIN; |
92407e75 | 2285 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2286 | if (!sync) |
2287 | goto out; | |
74316201 | 2288 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2289 | NFS_INO_LAYOUTCOMMITTING, |
2290 | nfs_wait_bit_killable, | |
2291 | TASK_KILLABLE); | |
92407e75 | 2292 | if (status) |
71244d9b | 2293 | goto out; |
92407e75 PT |
2294 | } |
2295 | ||
71244d9b TM |
2296 | status = -ENOMEM; |
2297 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2298 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2299 | if (!data) | |
2300 | goto clear_layoutcommitting; | |
2301 | ||
2302 | status = 0; | |
de4b15c7 | 2303 | spin_lock(&inode->i_lock); |
71244d9b TM |
2304 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2305 | goto out_unlock; | |
a9bae566 | 2306 | |
71244d9b | 2307 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2308 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2309 | |
acff5880 | 2310 | end_pos = nfsi->layout->plh_lwb; |
863a3c6c | 2311 | |
f597c537 | 2312 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2313 | spin_unlock(&inode->i_lock); |
2314 | ||
2315 | data->args.inode = inode; | |
9fa40758 | 2316 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2317 | nfs_fattr_init(&data->fattr); |
2318 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2319 | data->res.fattr = &data->fattr; | |
2320 | data->args.lastbytewritten = end_pos - 1; | |
2321 | data->res.server = NFS_SERVER(inode); | |
2322 | ||
5f919c9f CH |
2323 | if (ld->prepare_layoutcommit) { |
2324 | status = ld->prepare_layoutcommit(&data->args); | |
2325 | if (status) { | |
3471648a | 2326 | put_rpccred(data->cred); |
5f919c9f | 2327 | spin_lock(&inode->i_lock); |
29559b11 TM |
2328 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); |
2329 | if (end_pos > nfsi->layout->plh_lwb) | |
5f919c9f | 2330 | nfsi->layout->plh_lwb = end_pos; |
3471648a | 2331 | goto out_unlock; |
5f919c9f CH |
2332 | } |
2333 | } | |
2334 | ||
2335 | ||
863a3c6c AA |
2336 | status = nfs4_proc_layoutcommit(data, sync); |
2337 | out: | |
92407e75 PT |
2338 | if (status) |
2339 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2340 | dprintk("<-- %s status %d\n", __func__, status); |
2341 | return status; | |
71244d9b TM |
2342 | out_unlock: |
2343 | spin_unlock(&inode->i_lock); | |
92407e75 | 2344 | kfree(data); |
71244d9b TM |
2345 | clear_layoutcommitting: |
2346 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2347 | goto out; |
863a3c6c | 2348 | } |
72cff449 | 2349 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a | 2350 | |
5bb89b47 TM |
2351 | int |
2352 | pnfs_generic_sync(struct inode *inode, bool datasync) | |
2353 | { | |
2354 | return pnfs_layoutcommit_inode(inode, true); | |
2355 | } | |
2356 | EXPORT_SYMBOL_GPL(pnfs_generic_sync); | |
2357 | ||
82be417a AA |
2358 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) |
2359 | { | |
2360 | struct nfs4_threshold *thp; | |
2361 | ||
2362 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2363 | if (!thp) { | |
2364 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2365 | return NULL; | |
2366 | } | |
2367 | return thp; | |
2368 | } | |
8733408d | 2369 | |
865a7ecb | 2370 | #if IS_ENABLED(CONFIG_NFS_V4_2) |
8733408d | 2371 | int |
c8ad8894 | 2372 | pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags) |
8733408d PT |
2373 | { |
2374 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
2375 | struct nfs_server *server = NFS_SERVER(inode); | |
1bfe3b25 | 2376 | struct nfs_inode *nfsi = NFS_I(inode); |
8733408d PT |
2377 | struct nfs42_layoutstat_data *data; |
2378 | struct pnfs_layout_hdr *hdr; | |
2379 | int status = 0; | |
2380 | ||
2381 | if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats) | |
2382 | goto out; | |
2383 | ||
6c5a0d89 TM |
2384 | if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS)) |
2385 | goto out; | |
2386 | ||
1bfe3b25 PT |
2387 | if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags)) |
2388 | goto out; | |
2389 | ||
8733408d PT |
2390 | spin_lock(&inode->i_lock); |
2391 | if (!NFS_I(inode)->layout) { | |
2392 | spin_unlock(&inode->i_lock); | |
2393 | goto out; | |
2394 | } | |
2395 | hdr = NFS_I(inode)->layout; | |
2396 | pnfs_get_layout_hdr(hdr); | |
2397 | spin_unlock(&inode->i_lock); | |
2398 | ||
c8ad8894 | 2399 | data = kzalloc(sizeof(*data), gfp_flags); |
8733408d PT |
2400 | if (!data) { |
2401 | status = -ENOMEM; | |
2402 | goto out_put; | |
2403 | } | |
2404 | ||
2405 | data->args.fh = NFS_FH(inode); | |
2406 | data->args.inode = inode; | |
2407 | nfs4_stateid_copy(&data->args.stateid, &hdr->plh_stateid); | |
2408 | status = ld->prepare_layoutstats(&data->args); | |
2409 | if (status) | |
2410 | goto out_free; | |
2411 | ||
2412 | status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data); | |
2413 | ||
2414 | out: | |
2415 | dprintk("%s returns %d\n", __func__, status); | |
2416 | return status; | |
2417 | ||
2418 | out_free: | |
2419 | kfree(data); | |
2420 | out_put: | |
2421 | pnfs_put_layout_hdr(hdr); | |
1bfe3b25 PT |
2422 | smp_mb__before_atomic(); |
2423 | clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags); | |
2424 | smp_mb__after_atomic(); | |
8733408d PT |
2425 | goto out; |
2426 | } | |
2427 | EXPORT_SYMBOL_GPL(pnfs_report_layoutstat); | |
865a7ecb | 2428 | #endif |
bbf58bf3 TM |
2429 | |
2430 | unsigned int layoutstats_timer; | |
2431 | module_param(layoutstats_timer, uint, 0644); | |
2432 | EXPORT_SYMBOL_GPL(layoutstats_timer); |