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 PT |
55 | static int |
56 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, 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 PT |
387 | |
388 | stateid = lo->plh_stateid; | |
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 */ | |
394 | pnfs_send_layoutreturn(lo, stateid, iomode, false); | |
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); | |
706 | pnfs_put_layout_hdr(lo); | |
707 | iput(inode); | |
708 | } | |
709 | return ret; | |
710 | } | |
711 | ||
712 | int | |
713 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
714 | struct nfs_fsid *fsid, | |
715 | bool is_recall) | |
716 | { | |
717 | struct nfs_server *server; | |
718 | LIST_HEAD(layout_list); | |
c47abcf8 | 719 | |
974cec8c | 720 | spin_lock(&clp->cl_lock); |
6382a441 | 721 | rcu_read_lock(); |
fd9a8d71 | 722 | restart: |
6382a441 | 723 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
724 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
725 | continue; | |
726 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
727 | server, | |
728 | &layout_list) != 0) | |
729 | goto restart; | |
6382a441 WAA |
730 | } |
731 | rcu_read_unlock(); | |
974cec8c AA |
732 | spin_unlock(&clp->cl_lock); |
733 | ||
fd9a8d71 TM |
734 | if (list_empty(&layout_list)) |
735 | return 0; | |
736 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
737 | } | |
738 | ||
739 | int | |
740 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
741 | bool is_recall) | |
742 | { | |
743 | struct nfs_server *server; | |
744 | LIST_HEAD(layout_list); | |
745 | ||
746 | spin_lock(&clp->cl_lock); | |
747 | rcu_read_lock(); | |
748 | restart: | |
749 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
750 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
751 | server, | |
752 | &layout_list) != 0) | |
753 | goto restart; | |
974cec8c | 754 | } |
fd9a8d71 TM |
755 | rcu_read_unlock(); |
756 | spin_unlock(&clp->cl_lock); | |
757 | ||
758 | if (list_empty(&layout_list)) | |
759 | return 0; | |
760 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
761 | } | |
762 | ||
763 | /* | |
764 | * Called by the state manger to remove all layouts established under an | |
765 | * expired lease. | |
766 | */ | |
767 | void | |
768 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
769 | { | |
770 | nfs4_deviceid_mark_client_invalid(clp); | |
771 | nfs4_deviceid_purge_client(clp); | |
772 | ||
773 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
774 | } |
775 | ||
5a65503f TM |
776 | /* |
777 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
778 | * taking into account wraparound issues. | |
779 | */ | |
780 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
781 | { | |
2c64c57d | 782 | return (s32)(s1 - s2) > 0; |
5a65503f TM |
783 | } |
784 | ||
fd6002e9 | 785 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
786 | void |
787 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
788 | bool update_barrier) | |
b1f69b75 | 789 | { |
22aaf714 TM |
790 | u32 oldseq, newseq, new_barrier; |
791 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 792 | |
2d2f24ad TM |
793 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
794 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 795 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 796 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 797 | if (update_barrier) { |
22aaf714 | 798 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
799 | } else { |
800 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 801 | * "close" to the current seqids. |
43f1b3da | 802 | */ |
22aaf714 | 803 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 804 | } |
22aaf714 TM |
805 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
806 | lo->plh_barrier = new_barrier; | |
43f1b3da | 807 | } |
b1f69b75 AA |
808 | } |
809 | ||
cf7d63f1 | 810 | static bool |
19c54aba TM |
811 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
812 | const nfs4_stateid *stateid) | |
43f1b3da | 813 | { |
19c54aba | 814 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 815 | |
19c54aba TM |
816 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
817 | } | |
818 | ||
819 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
820 | static bool | |
e1c06f80 | 821 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo) |
19c54aba | 822 | { |
f7e8917a | 823 | return lo->plh_block_lgets || |
e1c06f80 | 824 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
cf7d63f1 FI |
825 | } |
826 | ||
fd6002e9 FI |
827 | int |
828 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
ce6ab4f2 | 829 | struct pnfs_layout_range *range, |
fd6002e9 | 830 | struct nfs4_state *open_state) |
b1f69b75 | 831 | { |
fd6002e9 | 832 | int status = 0; |
974cec8c | 833 | |
b1f69b75 | 834 | dprintk("--> %s\n", __func__); |
fd6002e9 | 835 | spin_lock(&lo->plh_inode->i_lock); |
e1c06f80 | 836 | if (pnfs_layoutgets_blocked(lo)) { |
cf7d63f1 | 837 | status = -EAGAIN; |
5d422301 TM |
838 | } else if (!nfs4_valid_open_stateid(open_state)) { |
839 | status = -EBADF; | |
47abadef CH |
840 | } else if (list_empty(&lo->plh_segs) || |
841 | test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { | |
fd6002e9 FI |
842 | int seq; |
843 | ||
844 | do { | |
845 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 846 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
847 | } while (read_seqretry(&open_state->seqlock, seq)); |
848 | } else | |
f597c537 | 849 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 850 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 851 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 852 | return status; |
b1f69b75 AA |
853 | } |
854 | ||
855 | /* | |
856 | * Get layout from server. | |
857 | * for now, assume that whole file layouts are requested. | |
858 | * arg->offset: 0 | |
859 | * arg->length: all ones | |
860 | */ | |
e5e94017 BH |
861 | static struct pnfs_layout_segment * |
862 | send_layoutget(struct pnfs_layout_hdr *lo, | |
863 | struct nfs_open_context *ctx, | |
fb3296eb | 864 | struct pnfs_layout_range *range, |
a75b9df9 | 865 | gfp_t gfp_flags) |
e5e94017 | 866 | { |
b7edfaa1 | 867 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
868 | struct nfs_server *server = NFS_SERVER(ino); |
869 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 870 | struct pnfs_layout_segment *lseg; |
2d89a1d3 | 871 | loff_t i_size; |
b1f69b75 AA |
872 | |
873 | dprintk("--> %s\n", __func__); | |
e5e94017 | 874 | |
4f2e9dce JL |
875 | /* |
876 | * Synchronously retrieve layout information from server and | |
877 | * store in lseg. If we race with a concurrent seqid morphing | |
878 | * op, then re-send the LAYOUTGET. | |
879 | */ | |
880 | do { | |
881 | lgp = kzalloc(sizeof(*lgp), gfp_flags); | |
882 | if (lgp == NULL) | |
883 | return NULL; | |
884 | ||
885 | i_size = i_size_read(ino); | |
886 | ||
887 | lgp->args.minlength = PAGE_CACHE_SIZE; | |
888 | if (lgp->args.minlength > range->length) | |
889 | lgp->args.minlength = range->length; | |
890 | if (range->iomode == IOMODE_READ) { | |
891 | if (range->offset >= i_size) | |
892 | lgp->args.minlength = 0; | |
893 | else if (i_size - range->offset < lgp->args.minlength) | |
894 | lgp->args.minlength = i_size - range->offset; | |
895 | } | |
896 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; | |
897 | lgp->args.range = *range; | |
898 | lgp->args.type = server->pnfs_curr_ld->id; | |
899 | lgp->args.inode = ino; | |
900 | lgp->args.ctx = get_nfs_open_context(ctx); | |
901 | lgp->gfp_flags = gfp_flags; | |
902 | lgp->cred = lo->plh_lc_cred; | |
35124a09 | 903 | |
4f2e9dce JL |
904 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
905 | } while (lseg == ERR_PTR(-EAGAIN)); | |
2d89a1d3 | 906 | |
a0b0a6e3 TM |
907 | if (IS_ERR(lseg)) { |
908 | switch (PTR_ERR(lseg)) { | |
909 | case -ENOMEM: | |
910 | case -ERESTARTSYS: | |
911 | break; | |
912 | default: | |
913 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 914 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
915 | } |
916 | return NULL; | |
d67ae825 TH |
917 | } else |
918 | pnfs_layout_clear_fail_bit(lo, | |
919 | pnfs_iomode_to_fail_bit(range->iomode)); | |
35124a09 | 920 | |
974cec8c AA |
921 | return lseg; |
922 | } | |
923 | ||
24956804 TM |
924 | static void pnfs_clear_layoutcommit(struct inode *inode, |
925 | struct list_head *head) | |
926 | { | |
927 | struct nfs_inode *nfsi = NFS_I(inode); | |
928 | struct pnfs_layout_segment *lseg, *tmp; | |
929 | ||
930 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
931 | return; | |
932 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
933 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
934 | continue; | |
935 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
936 | } | |
937 | } | |
938 | ||
d67ae825 TH |
939 | void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo) |
940 | { | |
941 | clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
942 | smp_mb__after_atomic(); | |
943 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN); | |
7f27392c | 944 | rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq); |
d67ae825 TH |
945 | } |
946 | ||
f40eb5d0 PT |
947 | static int |
948 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid, | |
6c16605d | 949 | enum pnfs_iomode iomode, bool sync) |
f40eb5d0 PT |
950 | { |
951 | struct inode *ino = lo->plh_inode; | |
952 | struct nfs4_layoutreturn *lrp; | |
953 | int status = 0; | |
954 | ||
e4af440a | 955 | lrp = kzalloc(sizeof(*lrp), GFP_NOFS); |
f40eb5d0 PT |
956 | if (unlikely(lrp == NULL)) { |
957 | status = -ENOMEM; | |
958 | spin_lock(&ino->i_lock); | |
d67ae825 | 959 | pnfs_clear_layoutreturn_waitbit(lo); |
f40eb5d0 PT |
960 | spin_unlock(&ino->i_lock); |
961 | pnfs_put_layout_hdr(lo); | |
962 | goto out; | |
963 | } | |
964 | ||
965 | lrp->args.stateid = stateid; | |
966 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
967 | lrp->args.inode = ino; | |
15eb67c1 PT |
968 | lrp->args.range.iomode = iomode; |
969 | lrp->args.range.offset = 0; | |
970 | lrp->args.range.length = NFS4_MAX_UINT64; | |
f40eb5d0 PT |
971 | lrp->args.layout = lo; |
972 | lrp->clp = NFS_SERVER(ino)->nfs_client; | |
973 | lrp->cred = lo->plh_lc_cred; | |
974 | ||
6c16605d | 975 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
976 | out: |
977 | dprintk("<-- %s status: %d\n", __func__, status); | |
978 | return status; | |
979 | } | |
980 | ||
293b3b06 AA |
981 | /* |
982 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
983 | * when the layout segment list is empty. | |
984 | * | |
985 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
986 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
987 | * deviceid is marked invalid. | |
988 | */ | |
cbe82603 BH |
989 | int |
990 | _pnfs_return_layout(struct inode *ino) | |
991 | { | |
992 | struct pnfs_layout_hdr *lo = NULL; | |
993 | struct nfs_inode *nfsi = NFS_I(ino); | |
994 | LIST_HEAD(tmp_list); | |
cbe82603 | 995 | nfs4_stateid stateid; |
293b3b06 | 996 | int status = 0, empty; |
7f27392c | 997 | bool send; |
cbe82603 | 998 | |
366d5052 | 999 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
1000 | |
1001 | spin_lock(&ino->i_lock); | |
1002 | lo = nfsi->layout; | |
e5929f3c | 1003 | if (!lo) { |
cbe82603 | 1004 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
1005 | dprintk("NFS: %s no layout to return\n", __func__); |
1006 | goto out; | |
cbe82603 BH |
1007 | } |
1008 | stateid = nfsi->layout->plh_stateid; | |
1009 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 1010 | pnfs_get_layout_hdr(lo); |
293b3b06 | 1011 | empty = list_empty(&lo->plh_segs); |
24956804 | 1012 | pnfs_clear_layoutcommit(ino, &tmp_list); |
49a85061 | 1013 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
c88953d8 CH |
1014 | |
1015 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
1016 | struct pnfs_layout_range range = { | |
1017 | .iomode = IOMODE_ANY, | |
1018 | .offset = 0, | |
1019 | .length = NFS4_MAX_UINT64, | |
1020 | }; | |
1021 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
1022 | } | |
1023 | ||
293b3b06 AA |
1024 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
1025 | if (empty) { | |
1026 | spin_unlock(&ino->i_lock); | |
293b3b06 | 1027 | dprintk("NFS: %s no layout segments to return\n", __func__); |
7f27392c | 1028 | goto out_put_layout_hdr; |
293b3b06 | 1029 | } |
47abadef CH |
1030 | |
1031 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
7f27392c | 1032 | send = pnfs_prepare_layoutreturn(lo); |
cbe82603 BH |
1033 | spin_unlock(&ino->i_lock); |
1034 | pnfs_free_lseg_list(&tmp_list); | |
7f27392c TM |
1035 | if (send) |
1036 | status = pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true); | |
1037 | out_put_layout_hdr: | |
1038 | pnfs_put_layout_hdr(lo); | |
cbe82603 BH |
1039 | out: |
1040 | dprintk("<-- %s status: %d\n", __func__, status); | |
1041 | return status; | |
1042 | } | |
0a57cdac | 1043 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 1044 | |
24028672 TM |
1045 | int |
1046 | pnfs_commit_and_return_layout(struct inode *inode) | |
1047 | { | |
1048 | struct pnfs_layout_hdr *lo; | |
1049 | int ret; | |
1050 | ||
1051 | spin_lock(&inode->i_lock); | |
1052 | lo = NFS_I(inode)->layout; | |
1053 | if (lo == NULL) { | |
1054 | spin_unlock(&inode->i_lock); | |
1055 | return 0; | |
1056 | } | |
1057 | pnfs_get_layout_hdr(lo); | |
1058 | /* Block new layoutgets and read/write to ds */ | |
1059 | lo->plh_block_lgets++; | |
1060 | spin_unlock(&inode->i_lock); | |
1061 | filemap_fdatawait(inode->i_mapping); | |
1062 | ret = pnfs_layoutcommit_inode(inode, true); | |
1063 | if (ret == 0) | |
1064 | ret = _pnfs_return_layout(inode); | |
1065 | spin_lock(&inode->i_lock); | |
1066 | lo->plh_block_lgets--; | |
1067 | spin_unlock(&inode->i_lock); | |
1068 | pnfs_put_layout_hdr(lo); | |
1069 | return ret; | |
1070 | } | |
1071 | ||
f7e8917a FI |
1072 | bool pnfs_roc(struct inode *ino) |
1073 | { | |
40dd4b7a TM |
1074 | struct nfs_inode *nfsi = NFS_I(ino); |
1075 | struct nfs_open_context *ctx; | |
1076 | struct nfs4_state *state; | |
f7e8917a FI |
1077 | struct pnfs_layout_hdr *lo; |
1078 | struct pnfs_layout_segment *lseg, *tmp; | |
193e3aa2 | 1079 | nfs4_stateid stateid; |
f7e8917a | 1080 | LIST_HEAD(tmp_list); |
e755d638 | 1081 | bool found = false, layoutreturn = false, roc = false; |
f7e8917a FI |
1082 | |
1083 | spin_lock(&ino->i_lock); | |
40dd4b7a | 1084 | lo = nfsi->layout; |
3976143b | 1085 | if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) |
40dd4b7a TM |
1086 | goto out_noroc; |
1087 | ||
e755d638 | 1088 | /* no roc if we hold a delegation */ |
40dd4b7a TM |
1089 | if (nfs4_check_delegation(ino, FMODE_READ)) |
1090 | goto out_noroc; | |
1091 | ||
1092 | list_for_each_entry(ctx, &nfsi->open_files, list) { | |
1093 | state = ctx->state; | |
1094 | /* Don't return layout if there is open file state */ | |
1095 | if (state != NULL && state->state != 0) | |
1096 | goto out_noroc; | |
1097 | } | |
1098 | ||
e755d638 PT |
1099 | stateid = lo->plh_stateid; |
1100 | /* always send layoutreturn if being marked so */ | |
1101 | if (test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, | |
1102 | &lo->plh_flags)) | |
1103 | layoutreturn = pnfs_prepare_layoutreturn(lo); | |
1104 | ||
c829013d | 1105 | pnfs_clear_retry_layoutget(lo); |
f7e8917a | 1106 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) |
e755d638 PT |
1107 | /* If we are sending layoutreturn, invalidate all valid lsegs */ |
1108 | if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
f7e8917a FI |
1109 | mark_lseg_invalid(lseg, &tmp_list); |
1110 | found = true; | |
1111 | } | |
500d701f | 1112 | /* ROC in two conditions: |
e755d638 PT |
1113 | * 1. there are ROC lsegs |
1114 | * 2. we don't send layoutreturn | |
e755d638 | 1115 | */ |
500d701f PT |
1116 | if (found && !layoutreturn) { |
1117 | /* lo ref dropped in pnfs_roc_release() */ | |
1118 | pnfs_get_layout_hdr(lo); | |
e755d638 | 1119 | roc = true; |
500d701f | 1120 | } |
f7e8917a | 1121 | |
40dd4b7a | 1122 | out_noroc: |
f7e8917a | 1123 | spin_unlock(&ino->i_lock); |
e755d638 PT |
1124 | pnfs_free_lseg_list(&tmp_list); |
1125 | pnfs_layoutcommit_inode(ino, true); | |
1126 | if (layoutreturn) | |
27b6f539 | 1127 | pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true); |
e755d638 | 1128 | return roc; |
f7e8917a FI |
1129 | } |
1130 | ||
1131 | void pnfs_roc_release(struct inode *ino) | |
1132 | { | |
1133 | struct pnfs_layout_hdr *lo; | |
1134 | ||
1135 | spin_lock(&ino->i_lock); | |
1136 | lo = NFS_I(ino)->layout; | |
5c4a79fb | 1137 | pnfs_clear_layoutreturn_waitbit(lo); |
6622c3ea TM |
1138 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
1139 | pnfs_detach_layout_hdr(lo); | |
1140 | spin_unlock(&ino->i_lock); | |
1141 | pnfs_free_layout_hdr(lo); | |
1142 | } else | |
1143 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
1144 | } |
1145 | ||
1146 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
1147 | { | |
1148 | struct pnfs_layout_hdr *lo; | |
1149 | ||
1150 | spin_lock(&ino->i_lock); | |
1151 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 1152 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
1153 | lo->plh_barrier = barrier; |
1154 | spin_unlock(&ino->i_lock); | |
6a463beb | 1155 | trace_nfs4_layoutreturn_on_close(ino, 0); |
f7e8917a FI |
1156 | } |
1157 | ||
4ff376fe | 1158 | void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier) |
f7e8917a FI |
1159 | { |
1160 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 1161 | struct pnfs_layout_hdr *lo; |
7fdab069 | 1162 | u32 current_seqid; |
f7e8917a FI |
1163 | |
1164 | spin_lock(&ino->i_lock); | |
7fdab069 TM |
1165 | lo = nfsi->layout; |
1166 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1167 | |
7fdab069 TM |
1168 | /* Since close does not return a layout stateid for use as |
1169 | * a barrier, we choose the worst-case barrier. | |
1170 | */ | |
1171 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
f7e8917a | 1172 | spin_unlock(&ino->i_lock); |
f7e8917a FI |
1173 | } |
1174 | ||
500d701f PT |
1175 | bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task) |
1176 | { | |
1177 | struct nfs_inode *nfsi = NFS_I(ino); | |
1178 | struct pnfs_layout_hdr *lo; | |
1179 | bool sleep = false; | |
1180 | ||
1181 | /* we might not have grabbed lo reference. so need to check under | |
1182 | * i_lock */ | |
1183 | spin_lock(&ino->i_lock); | |
1184 | lo = nfsi->layout; | |
1185 | if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) | |
1186 | sleep = true; | |
1187 | spin_unlock(&ino->i_lock); | |
1188 | ||
1189 | if (sleep) | |
1190 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
1191 | ||
1192 | return sleep; | |
1193 | } | |
1194 | ||
b1f69b75 AA |
1195 | /* |
1196 | * Compare two layout segments for sorting into layout cache. | |
1197 | * We want to preferentially return RW over RO layouts, so ensure those | |
1198 | * are seen first. | |
1199 | */ | |
1200 | static s64 | |
7dc0ac70 | 1201 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1202 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1203 | { |
fb3296eb BH |
1204 | s64 d; |
1205 | ||
1206 | /* high offset > low offset */ | |
1207 | d = l1->offset - l2->offset; | |
1208 | if (d) | |
1209 | return d; | |
1210 | ||
1211 | /* short length > long length */ | |
1212 | d = l2->length - l1->length; | |
1213 | if (d) | |
1214 | return d; | |
1215 | ||
b1f69b75 | 1216 | /* read > read/write */ |
fb3296eb | 1217 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1218 | } |
1219 | ||
03772d2f TM |
1220 | static bool |
1221 | pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1, | |
1222 | const struct pnfs_layout_range *l2) | |
1223 | { | |
1224 | return pnfs_lseg_range_cmp(l1, l2) > 0; | |
1225 | } | |
1226 | ||
1227 | static bool | |
1228 | pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg, | |
1229 | struct pnfs_layout_segment *old) | |
1230 | { | |
1231 | return false; | |
1232 | } | |
1233 | ||
1234 | void | |
1235 | pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1236 | struct pnfs_layout_segment *lseg, | |
1237 | bool (*is_after)(const struct pnfs_layout_range *, | |
1238 | const struct pnfs_layout_range *), | |
1239 | bool (*do_merge)(struct pnfs_layout_segment *, | |
1240 | struct pnfs_layout_segment *), | |
1241 | struct list_head *free_me) | |
974cec8c | 1242 | { |
03772d2f | 1243 | struct pnfs_layout_segment *lp, *tmp; |
b1f69b75 | 1244 | |
974cec8c AA |
1245 | dprintk("%s:Begin\n", __func__); |
1246 | ||
03772d2f TM |
1247 | list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) { |
1248 | if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0) | |
1249 | continue; | |
1250 | if (do_merge(lseg, lp)) { | |
1251 | mark_lseg_invalid(lp, free_me); | |
1252 | continue; | |
1253 | } | |
1254 | if (is_after(&lseg->pls_range, &lp->pls_range)) | |
b1f69b75 | 1255 | continue; |
566052c5 | 1256 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1257 | dprintk("%s: inserted lseg %p " |
1258 | "iomode %d offset %llu length %llu before " | |
1259 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1260 | __func__, lseg, lseg->pls_range.iomode, |
1261 | lseg->pls_range.offset, lseg->pls_range.length, | |
1262 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1263 | lp->pls_range.length); | |
fb3296eb | 1264 | goto out; |
974cec8c | 1265 | } |
fb3296eb BH |
1266 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1267 | dprintk("%s: inserted lseg %p " | |
1268 | "iomode %d offset %llu length %llu at tail\n", | |
1269 | __func__, lseg, lseg->pls_range.iomode, | |
1270 | lseg->pls_range.offset, lseg->pls_range.length); | |
1271 | out: | |
70c3bd2b | 1272 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1273 | |
1274 | dprintk("%s:Return\n", __func__); | |
e5e94017 | 1275 | } |
03772d2f TM |
1276 | EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg); |
1277 | ||
1278 | static void | |
1279 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1280 | struct pnfs_layout_segment *lseg, | |
1281 | struct list_head *free_me) | |
1282 | { | |
1283 | struct inode *inode = lo->plh_inode; | |
1284 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
1285 | ||
1286 | if (ld->add_lseg != NULL) | |
1287 | ld->add_lseg(lo, lseg, free_me); | |
1288 | else | |
1289 | pnfs_generic_layout_insert_lseg(lo, lseg, | |
1290 | pnfs_lseg_range_is_after, | |
1291 | pnfs_lseg_no_merge, | |
1292 | free_me); | |
1293 | } | |
e5e94017 BH |
1294 | |
1295 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1296 | alloc_init_layout_hdr(struct inode *ino, |
1297 | struct nfs_open_context *ctx, | |
1298 | gfp_t gfp_flags) | |
e5e94017 BH |
1299 | { |
1300 | struct pnfs_layout_hdr *lo; | |
1301 | ||
636fb9c8 | 1302 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1303 | if (!lo) |
1304 | return NULL; | |
cc6e5340 | 1305 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1306 | INIT_LIST_HEAD(&lo->plh_layouts); |
1307 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1308 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1309 | lo->plh_inode = ino; |
5cc2216d | 1310 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
e5e94017 BH |
1311 | return lo; |
1312 | } | |
1313 | ||
1314 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1315 | pnfs_find_alloc_layout(struct inode *ino, |
1316 | struct nfs_open_context *ctx, | |
1317 | gfp_t gfp_flags) | |
e5e94017 BH |
1318 | { |
1319 | struct nfs_inode *nfsi = NFS_I(ino); | |
1320 | struct pnfs_layout_hdr *new = NULL; | |
1321 | ||
1322 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1323 | ||
251ec410 TM |
1324 | if (nfsi->layout != NULL) |
1325 | goto out_existing; | |
e5e94017 | 1326 | spin_unlock(&ino->i_lock); |
9fa40758 | 1327 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1328 | spin_lock(&ino->i_lock); |
1329 | ||
251ec410 | 1330 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1331 | nfsi->layout = new; |
251ec410 | 1332 | return new; |
7175fe90 YN |
1333 | } else if (new != NULL) |
1334 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1335 | out_existing: |
1336 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1337 | return nfsi->layout; |
1338 | } | |
1339 | ||
b1f69b75 AA |
1340 | /* |
1341 | * iomode matching rules: | |
1342 | * iomode lseg match | |
1343 | * ----- ----- ----- | |
1344 | * ANY READ true | |
1345 | * ANY RW true | |
1346 | * RW READ false | |
1347 | * RW RW true | |
1348 | * READ READ true | |
1349 | * READ RW true | |
1350 | */ | |
3cb2df17 | 1351 | static bool |
7dc0ac70 | 1352 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
3cb2df17 | 1353 | const struct pnfs_layout_range *range) |
b1f69b75 | 1354 | { |
fb3296eb BH |
1355 | struct pnfs_layout_range range1; |
1356 | ||
1357 | if ((range->iomode == IOMODE_RW && | |
1358 | ls_range->iomode != IOMODE_RW) || | |
7dc0ac70 | 1359 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1360 | return 0; |
1361 | ||
1362 | /* range1 covers only the first byte in the range */ | |
1363 | range1 = *range; | |
1364 | range1.length = 1; | |
7dc0ac70 | 1365 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1366 | } |
1367 | ||
1368 | /* | |
1369 | * lookup range in layout | |
1370 | */ | |
e5e94017 | 1371 | static struct pnfs_layout_segment * |
fb3296eb BH |
1372 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
1373 | struct pnfs_layout_range *range) | |
e5e94017 | 1374 | { |
b1f69b75 AA |
1375 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1376 | ||
1377 | dprintk("%s:Begin\n", __func__); | |
1378 | ||
b7edfaa1 | 1379 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1380 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1381 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
7dc0ac70 | 1382 | pnfs_lseg_range_match(&lseg->pls_range, range)) { |
9369a431 | 1383 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1384 | break; |
1385 | } | |
b1f69b75 AA |
1386 | } |
1387 | ||
1388 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1389 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1390 | return ret; |
e5e94017 BH |
1391 | } |
1392 | ||
d23d61c8 AA |
1393 | /* |
1394 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1395 | * to the MDS or over pNFS | |
1396 | * | |
1397 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1398 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1399 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1400 | * WRITE request. | |
1401 | * | |
1402 | * A return of true means use MDS I/O. | |
1403 | * | |
1404 | * From rfc 5661: | |
1405 | * If a file's size is smaller than the file size threshold, data accesses | |
1406 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1407 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1408 | * server. If both file size and I/O size are provided, the client SHOULD | |
1409 | * reach or exceed both thresholds before sending its read or write | |
1410 | * requests to the data server. | |
1411 | */ | |
1412 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1413 | struct inode *ino, int iomode) | |
1414 | { | |
1415 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1416 | struct nfs_inode *nfsi = NFS_I(ino); | |
1417 | loff_t fsize = i_size_read(ino); | |
1418 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1419 | ||
1420 | if (t == NULL) | |
1421 | return ret; | |
1422 | ||
1423 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1424 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1425 | ||
1426 | switch (iomode) { | |
1427 | case IOMODE_READ: | |
1428 | if (t->bm & THRESHOLD_RD) { | |
1429 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1430 | size_set = true; | |
1431 | if (fsize < t->rd_sz) | |
1432 | size = true; | |
1433 | } | |
1434 | if (t->bm & THRESHOLD_RD_IO) { | |
1435 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1436 | nfsi->read_io); | |
1437 | io_set = true; | |
1438 | if (nfsi->read_io < t->rd_io_sz) | |
1439 | io = true; | |
1440 | } | |
1441 | break; | |
1442 | case IOMODE_RW: | |
1443 | if (t->bm & THRESHOLD_WR) { | |
1444 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1445 | size_set = true; | |
1446 | if (fsize < t->wr_sz) | |
1447 | size = true; | |
1448 | } | |
1449 | if (t->bm & THRESHOLD_WR_IO) { | |
1450 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1451 | nfsi->write_io); | |
1452 | io_set = true; | |
1453 | if (nfsi->write_io < t->wr_io_sz) | |
1454 | io = true; | |
1455 | } | |
1456 | break; | |
1457 | } | |
1458 | if (size_set && io_set) { | |
1459 | if (size && io) | |
1460 | ret = true; | |
1461 | } else if (size || io) | |
1462 | ret = true; | |
1463 | ||
1464 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1465 | return ret; | |
1466 | } | |
1467 | ||
aa8a45ee | 1468 | /* stop waiting if someone clears NFS_LAYOUT_RETRY_LAYOUTGET bit. */ |
dfd01f02 | 1469 | static int pnfs_layoutget_retry_bit_wait(struct wait_bit_key *key, int mode) |
aa8a45ee PT |
1470 | { |
1471 | if (!test_bit(NFS_LAYOUT_RETRY_LAYOUTGET, key->flags)) | |
1472 | return 1; | |
dfd01f02 | 1473 | return nfs_wait_bit_killable(key, mode); |
aa8a45ee PT |
1474 | } |
1475 | ||
1476 | static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo) | |
1477 | { | |
8f70f53a TM |
1478 | if (!pnfs_should_retry_layoutget(lo)) |
1479 | return false; | |
aa8a45ee PT |
1480 | /* |
1481 | * send layoutcommit as it can hold up layoutreturn due to lseg | |
1482 | * reference | |
1483 | */ | |
1484 | pnfs_layoutcommit_inode(lo->plh_inode, false); | |
1485 | return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
1486 | pnfs_layoutget_retry_bit_wait, | |
1487 | TASK_UNINTERRUPTIBLE); | |
1488 | } | |
1489 | ||
d67ae825 TH |
1490 | static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo) |
1491 | { | |
1492 | unsigned long *bitlock = &lo->plh_flags; | |
1493 | ||
1494 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1495 | smp_mb__after_atomic(); | |
1496 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1497 | } | |
1498 | ||
e5e94017 BH |
1499 | /* |
1500 | * Layout segment is retreived from the server if not cached. | |
1501 | * The appropriate layout segment is referenced and returned to the caller. | |
1502 | */ | |
7c24d948 | 1503 | struct pnfs_layout_segment * |
e5e94017 BH |
1504 | pnfs_update_layout(struct inode *ino, |
1505 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1506 | loff_t pos, |
1507 | u64 count, | |
a75b9df9 TM |
1508 | enum pnfs_iomode iomode, |
1509 | gfp_t gfp_flags) | |
e5e94017 | 1510 | { |
fb3296eb BH |
1511 | struct pnfs_layout_range arg = { |
1512 | .iomode = iomode, | |
1513 | .offset = pos, | |
1514 | .length = count, | |
1515 | }; | |
707ed5fd | 1516 | unsigned pg_offset; |
6382a441 WAA |
1517 | struct nfs_server *server = NFS_SERVER(ino); |
1518 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1519 | struct pnfs_layout_hdr *lo; |
1520 | struct pnfs_layout_segment *lseg = NULL; | |
30005121 | 1521 | bool first; |
e5e94017 | 1522 | |
9a4bf31d | 1523 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) { |
f4848303 | 1524 | trace_pnfs_update_layout(ino, pos, count, iomode, NULL, |
9a4bf31d | 1525 | PNFS_UPDATE_LAYOUT_NO_PNFS); |
f86bbcf8 | 1526 | goto out; |
9a4bf31d | 1527 | } |
d23d61c8 | 1528 | |
9a4bf31d | 1529 | if (iomode == IOMODE_READ && i_size_read(ino) == 0) { |
f4848303 | 1530 | trace_pnfs_update_layout(ino, pos, count, iomode, NULL, |
9a4bf31d | 1531 | PNFS_UPDATE_LAYOUT_RD_ZEROLEN); |
4ae93560 | 1532 | goto out; |
9a4bf31d | 1533 | } |
4ae93560 | 1534 | |
9a4bf31d | 1535 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) { |
f4848303 | 1536 | trace_pnfs_update_layout(ino, pos, count, iomode, NULL, |
9a4bf31d | 1537 | PNFS_UPDATE_LAYOUT_MDSTHRESH); |
f86bbcf8 | 1538 | goto out; |
9a4bf31d | 1539 | } |
d23d61c8 | 1540 | |
9bf87482 PT |
1541 | lookup_again: |
1542 | first = false; | |
e5e94017 | 1543 | spin_lock(&ino->i_lock); |
9fa40758 | 1544 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1545 | if (lo == NULL) { |
1546 | spin_unlock(&ino->i_lock); | |
f4848303 | 1547 | trace_pnfs_update_layout(ino, pos, count, iomode, NULL, |
9a4bf31d | 1548 | PNFS_UPDATE_LAYOUT_NOMEM); |
830ffb56 TM |
1549 | goto out; |
1550 | } | |
e5e94017 | 1551 | |
43f1b3da | 1552 | /* Do we even need to bother with this? */ |
a59c30ac | 1553 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
f4848303 | 1554 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1555 | PNFS_UPDATE_LAYOUT_BULK_RECALL); |
43f1b3da | 1556 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1557 | goto out_unlock; |
1558 | } | |
1559 | ||
1560 | /* if LAYOUTGET already failed once we don't try again */ | |
aa8a45ee | 1561 | if (pnfs_layout_io_test_failed(lo, iomode) && |
9a4bf31d | 1562 | !pnfs_should_retry_layoutget(lo)) { |
f4848303 | 1563 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1564 | PNFS_UPDATE_LAYOUT_IO_TEST_FAIL); |
e5e94017 | 1565 | goto out_unlock; |
9a4bf31d | 1566 | } |
e5e94017 | 1567 | |
9bf87482 PT |
1568 | first = list_empty(&lo->plh_segs); |
1569 | if (first) { | |
1570 | /* The first layoutget for the file. Need to serialize per | |
1571 | * RFC 5661 Errata 3208. | |
1572 | */ | |
1573 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1574 | &lo->plh_flags)) { | |
1575 | spin_unlock(&ino->i_lock); | |
1576 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1577 | TASK_UNINTERRUPTIBLE); | |
1578 | pnfs_put_layout_hdr(lo); | |
1579 | goto lookup_again; | |
1580 | } | |
1581 | } else { | |
1582 | /* Check to see if the layout for the given range | |
1583 | * already exists | |
1584 | */ | |
1585 | lseg = pnfs_find_lseg(lo, &arg); | |
9a4bf31d | 1586 | if (lseg) { |
f4848303 | 1587 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1588 | PNFS_UPDATE_LAYOUT_FOUND_CACHED); |
9bf87482 | 1589 | goto out_unlock; |
9a4bf31d | 1590 | } |
9bf87482 | 1591 | } |
568e8c49 | 1592 | |
aa8a45ee PT |
1593 | /* |
1594 | * Because we free lsegs before sending LAYOUTRETURN, we need to wait | |
1595 | * for LAYOUTRETURN even if first is true. | |
1596 | */ | |
8f70f53a | 1597 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
aa8a45ee PT |
1598 | spin_unlock(&ino->i_lock); |
1599 | dprintk("%s wait for layoutreturn\n", __func__); | |
1600 | if (pnfs_prepare_to_retry_layoutget(lo)) { | |
d67ae825 TH |
1601 | if (first) |
1602 | pnfs_clear_first_layoutget(lo); | |
aa8a45ee PT |
1603 | pnfs_put_layout_hdr(lo); |
1604 | dprintk("%s retrying\n", __func__); | |
1605 | goto lookup_again; | |
1606 | } | |
f4848303 | 1607 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1608 | PNFS_UPDATE_LAYOUT_RETURN); |
aa8a45ee PT |
1609 | goto out_put_layout_hdr; |
1610 | } | |
1611 | ||
9a4bf31d | 1612 | if (pnfs_layoutgets_blocked(lo)) { |
f4848303 | 1613 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1614 | PNFS_UPDATE_LAYOUT_BLOCKED); |
cf7d63f1 | 1615 | goto out_unlock; |
9a4bf31d | 1616 | } |
cf7d63f1 | 1617 | atomic_inc(&lo->plh_outstanding); |
f49f9baa | 1618 | spin_unlock(&ino->i_lock); |
30005121 | 1619 | |
abb9a007 | 1620 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1621 | /* The lo must be on the clp list if there is any |
1622 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1623 | */ | |
1624 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1625 | if (list_empty(&lo->plh_layouts)) |
1626 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1627 | spin_unlock(&clp->cl_lock); |
1628 | } | |
e5e94017 | 1629 | |
707ed5fd BH |
1630 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1631 | if (pg_offset) { | |
1632 | arg.offset -= pg_offset; | |
1633 | arg.length += pg_offset; | |
1634 | } | |
7c24d948 AA |
1635 | if (arg.length != NFS4_MAX_UINT64) |
1636 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1637 | |
fb3296eb | 1638 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
c829013d | 1639 | pnfs_clear_retry_layoutget(lo); |
cf7d63f1 | 1640 | atomic_dec(&lo->plh_outstanding); |
f4848303 | 1641 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
9a4bf31d | 1642 | PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET); |
830ffb56 | 1643 | out_put_layout_hdr: |
d67ae825 TH |
1644 | if (first) |
1645 | pnfs_clear_first_layoutget(lo); | |
70c3bd2b | 1646 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1647 | out: |
f86bbcf8 TM |
1648 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1649 | "(%s, offset: %llu, length: %llu)\n", | |
1650 | __func__, ino->i_sb->s_id, | |
1651 | (unsigned long long)NFS_FILEID(ino), | |
1652 | lseg == NULL ? "not found" : "found", | |
1653 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1654 | (unsigned long long)pos, | |
1655 | (unsigned long long)count); | |
e5e94017 BH |
1656 | return lseg; |
1657 | out_unlock: | |
1658 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1659 | goto out_put_layout_hdr; |
e5e94017 | 1660 | } |
7c24d948 | 1661 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1662 | |
540d9864 TM |
1663 | static bool |
1664 | pnfs_sanity_check_layout_range(struct pnfs_layout_range *range) | |
1665 | { | |
1666 | switch (range->iomode) { | |
1667 | case IOMODE_READ: | |
1668 | case IOMODE_RW: | |
1669 | break; | |
1670 | default: | |
1671 | return false; | |
1672 | } | |
1673 | if (range->offset == NFS4_MAX_UINT64) | |
1674 | return false; | |
1675 | if (range->length == 0) | |
1676 | return false; | |
1677 | if (range->length != NFS4_MAX_UINT64 && | |
1678 | range->length > NFS4_MAX_UINT64 - range->offset) | |
1679 | return false; | |
1680 | return true; | |
1681 | } | |
1682 | ||
a0b0a6e3 | 1683 | struct pnfs_layout_segment * |
b1f69b75 AA |
1684 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1685 | { | |
1686 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1687 | struct nfs4_layoutget_res *res = &lgp->res; | |
1688 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1689 | struct inode *ino = lo->plh_inode; |
78096cca | 1690 | LIST_HEAD(free_me); |
540d9864 TM |
1691 | int status = -EINVAL; |
1692 | ||
1693 | if (!pnfs_sanity_check_layout_range(&res->range)) | |
1694 | goto out; | |
b1f69b75 AA |
1695 | |
1696 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1697 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1698 | if (!lseg || IS_ERR(lseg)) { |
1699 | if (!lseg) | |
1700 | status = -ENOMEM; | |
1701 | else | |
1702 | status = PTR_ERR(lseg); | |
1703 | dprintk("%s: Could not allocate layout: error %d\n", | |
1704 | __func__, status); | |
1705 | goto out; | |
1706 | } | |
1707 | ||
1013df61 CH |
1708 | init_lseg(lo, lseg); |
1709 | lseg->pls_range = res->range; | |
1710 | ||
b1f69b75 | 1711 | spin_lock(&ino->i_lock); |
e1c06f80 | 1712 | if (pnfs_layoutgets_blocked(lo)) { |
43f1b3da FI |
1713 | dprintk("%s forget reply due to state\n", __func__); |
1714 | goto out_forget_reply; | |
1715 | } | |
038d6493 | 1716 | |
362f7474 CH |
1717 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1718 | /* existing state ID, make sure the sequence number matches. */ | |
1719 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1720 | dprintk("%s forget reply due to sequence\n", __func__); | |
4f2e9dce | 1721 | status = -EAGAIN; |
362f7474 CH |
1722 | goto out_forget_reply; |
1723 | } | |
1724 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1725 | } else { | |
1726 | /* | |
1727 | * We got an entirely new state ID. Mark all segments for the | |
1728 | * inode invalid, and don't bother validating the stateid | |
1729 | * sequence number. | |
1730 | */ | |
1731 | pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL); | |
1732 | ||
1733 | nfs4_stateid_copy(&lo->plh_stateid, &res->stateid); | |
1734 | lo->plh_barrier = be32_to_cpu(res->stateid.seqid); | |
1735 | } | |
038d6493 | 1736 | |
47abadef CH |
1737 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); |
1738 | ||
9369a431 | 1739 | pnfs_get_lseg(lseg); |
03772d2f | 1740 | pnfs_layout_insert_lseg(lo, lseg, &free_me); |
b1f69b75 | 1741 | |
3976143b | 1742 | if (res->return_on_close) |
f7e8917a | 1743 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); |
f7e8917a | 1744 | |
b1f69b75 | 1745 | spin_unlock(&ino->i_lock); |
78096cca | 1746 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1747 | return lseg; |
b1f69b75 | 1748 | out: |
a0b0a6e3 | 1749 | return ERR_PTR(status); |
43f1b3da FI |
1750 | |
1751 | out_forget_reply: | |
1752 | spin_unlock(&ino->i_lock); | |
1753 | lseg->pls_layout = lo; | |
1754 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1755 | goto out; | |
b1f69b75 AA |
1756 | } |
1757 | ||
016256df PT |
1758 | static void |
1759 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, | |
1760 | struct list_head *tmp_list, | |
1761 | struct pnfs_layout_range *return_range) | |
1762 | { | |
1763 | struct pnfs_layout_segment *lseg, *next; | |
1764 | ||
1765 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
1766 | ||
1767 | if (list_empty(&lo->plh_segs)) | |
1768 | return; | |
1769 | ||
1770 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
1771 | if (should_free_lseg(&lseg->pls_range, return_range)) { | |
1772 | dprintk("%s: marking lseg %p iomode %d " | |
1773 | "offset %llu length %llu\n", __func__, | |
1774 | lseg, lseg->pls_range.iomode, | |
1775 | lseg->pls_range.offset, | |
1776 | lseg->pls_range.length); | |
1777 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
1778 | mark_lseg_invalid(lseg, tmp_list); | |
0bdb8fa6 TM |
1779 | set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, |
1780 | &lo->plh_flags); | |
016256df PT |
1781 | } |
1782 | } | |
1783 | ||
1784 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
1785 | struct pnfs_layout_segment *lseg) | |
1786 | { | |
1787 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
1788 | int iomode = pnfs_iomode_to_fail_bit(lseg->pls_range.iomode); | |
1789 | struct pnfs_layout_range range = { | |
1790 | .iomode = lseg->pls_range.iomode, | |
1791 | .offset = 0, | |
1792 | .length = NFS4_MAX_UINT64, | |
1793 | }; | |
1794 | LIST_HEAD(free_me); | |
1795 | ||
1796 | spin_lock(&inode->i_lock); | |
1797 | /* set failure bit so that pnfs path will be retried later */ | |
1798 | pnfs_layout_set_fail_bit(lo, iomode); | |
016256df PT |
1799 | if (lo->plh_return_iomode == 0) |
1800 | lo->plh_return_iomode = range.iomode; | |
1801 | else if (lo->plh_return_iomode != range.iomode) | |
1802 | lo->plh_return_iomode = IOMODE_ANY; | |
1803 | /* | |
1804 | * mark all matching lsegs so that we are sure to have no live | |
1805 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
1806 | * for how it works. | |
1807 | */ | |
1808 | pnfs_mark_matching_lsegs_return(lo, &free_me, &range); | |
1809 | spin_unlock(&inode->i_lock); | |
1810 | pnfs_free_lseg_list(&free_me); | |
1811 | } | |
1812 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
1813 | ||
d8007d4d TM |
1814 | void |
1815 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1816 | { | |
1fd937bd PT |
1817 | u64 rd_size = req->wb_bytes; |
1818 | ||
cb5d04bc PT |
1819 | if (pgio->pg_lseg == NULL) { |
1820 | if (pgio->pg_dreq == NULL) | |
1821 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1822 | else | |
1823 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1824 | ||
1825 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1826 | req->wb_context, | |
1827 | req_offset(req), | |
1828 | rd_size, | |
1829 | IOMODE_READ, | |
1830 | GFP_KERNEL); | |
1831 | } | |
e885de1a TM |
1832 | /* If no lseg, fall back to read through mds */ |
1833 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1834 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1835 | |
d8007d4d TM |
1836 | } |
1837 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1838 | ||
1839 | void | |
6296556f PT |
1840 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1841 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1842 | { |
cb5d04bc PT |
1843 | if (pgio->pg_lseg == NULL) |
1844 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1845 | req->wb_context, | |
1846 | req_offset(req), | |
1847 | wb_size, | |
1848 | IOMODE_RW, | |
1849 | GFP_NOFS); | |
e885de1a TM |
1850 | /* If no lseg, fall back to write through mds */ |
1851 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1852 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1853 | } |
1854 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1855 | ||
180bb5ec WAA |
1856 | void |
1857 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
1858 | { | |
1859 | if (desc->pg_lseg) { | |
1860 | pnfs_put_lseg(desc->pg_lseg); | |
1861 | desc->pg_lseg = NULL; | |
1862 | } | |
1863 | } | |
1864 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
1865 | ||
b4fdac1a WAA |
1866 | /* |
1867 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1868 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1869 | */ | |
1870 | size_t | |
a7d42ddb WAA |
1871 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
1872 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 1873 | { |
0f9c429e | 1874 | unsigned int size; |
c5e20cb7 | 1875 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1876 | |
1877 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1878 | if (!size) |
1879 | return 0; | |
94ad1c80 | 1880 | |
19982ba8 | 1881 | /* |
c5e20cb7 WAA |
1882 | * 'size' contains the number of bytes left in the current page (up |
1883 | * to the original size asked for in @req->wb_bytes). | |
1884 | * | |
1885 | * Calculate how many bytes are left in the layout segment | |
1886 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1887 | * |
1888 | * Please also note that 'end_offset' is actually the offset of the | |
1889 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1890 | * | |
1891 | */ | |
19b54848 | 1892 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1893 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1894 | pgio->pg_lseg->pls_range.length); | |
1895 | req_start = req_offset(req); | |
7c13789e | 1896 | WARN_ON_ONCE(req_start >= seg_end); |
c5e20cb7 | 1897 | /* start of request is past the last byte of this segment */ |
7c13789e WAA |
1898 | if (req_start >= seg_end) { |
1899 | /* reference the new lseg */ | |
1900 | if (pgio->pg_ops->pg_cleanup) | |
1901 | pgio->pg_ops->pg_cleanup(pgio); | |
1902 | if (pgio->pg_ops->pg_init) | |
1903 | pgio->pg_ops->pg_init(pgio, req); | |
19b54848 | 1904 | return 0; |
7c13789e | 1905 | } |
c5e20cb7 WAA |
1906 | |
1907 | /* adjust 'size' iff there are fewer bytes left in the | |
1908 | * segment than what nfs_generic_pg_test returned */ | |
1909 | seg_left = seg_end - req_start; | |
1910 | if (seg_left < size) | |
1911 | size = (unsigned int)seg_left; | |
19b54848 | 1912 | } |
0f9c429e | 1913 | |
19b54848 | 1914 | return size; |
94ad1c80 | 1915 | } |
89a58e32 | 1916 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1917 | |
53113ad3 | 1918 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
1919 | { |
1920 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
1921 | |
1922 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
1923 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
1924 | hdr->completion_ops); | |
c7070113 | 1925 | set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags); |
53113ad3 | 1926 | return nfs_pageio_resend(&pgio, hdr); |
e2fecb21 | 1927 | } |
e7dd79af | 1928 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1929 | |
d45f60c6 | 1930 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1931 | { |
cd841605 FI |
1932 | |
1933 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1934 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1935 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1936 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1937 | } |
6c75dc0d | 1938 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1939 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
1940 | } |
1941 | ||
d20581aa BH |
1942 | /* |
1943 | * Called by non rpc-based layout drivers | |
1944 | */ | |
d45f60c6 | 1945 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 1946 | { |
f8417b48 | 1947 | if (likely(!hdr->pnfs_error)) { |
67af7611 TM |
1948 | pnfs_set_layoutcommit(hdr->inode, hdr->lseg, |
1949 | hdr->mds_offset + hdr->res.count); | |
d45f60c6 | 1950 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
1951 | } |
1952 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); | |
1953 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
1954 | pnfs_ld_handle_write_error(hdr); |
1955 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 1956 | } |
d20581aa | 1957 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1958 | |
dce81290 TM |
1959 | static void |
1960 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1961 | struct nfs_pgio_header *hdr) |
dce81290 | 1962 | { |
48d635f1 | 1963 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1964 | |
6c75dc0d | 1965 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 1966 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 1967 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 1968 | mirror->pg_recoalesce = 1; |
6c75dc0d | 1969 | } |
d45f60c6 | 1970 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 1971 | hdr->release(hdr); |
dce81290 TM |
1972 | } |
1973 | ||
1974 | static enum pnfs_try_status | |
d45f60c6 | 1975 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
1976 | const struct rpc_call_ops *call_ops, |
1977 | struct pnfs_layout_segment *lseg, | |
1978 | int how) | |
0382b744 | 1979 | { |
cd841605 | 1980 | struct inode *inode = hdr->inode; |
0382b744 AA |
1981 | enum pnfs_try_status trypnfs; |
1982 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1983 | ||
cd841605 | 1984 | hdr->mds_ops = call_ops; |
0382b744 AA |
1985 | |
1986 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
1987 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
1988 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 1989 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1990 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1991 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1992 | return trypnfs; | |
1993 | } | |
1994 | ||
dce81290 | 1995 | static void |
7f714720 WAA |
1996 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
1997 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 1998 | { |
dce81290 TM |
1999 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2000 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2001 | enum pnfs_try_status trypnfs; |
dce81290 | 2002 | |
d45f60c6 | 2003 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 2004 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 2005 | pnfs_write_through_mds(desc, hdr); |
dce81290 TM |
2006 | } |
2007 | ||
6c75dc0d FI |
2008 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
2009 | { | |
9369a431 | 2010 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2011 | nfs_pgio_header_free(hdr); |
6c75dc0d FI |
2012 | } |
2013 | ||
dce81290 TM |
2014 | int |
2015 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
2016 | { | |
48d635f1 | 2017 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2018 | |
6c75dc0d | 2019 | struct nfs_pgio_header *hdr; |
dce81290 TM |
2020 | int ret; |
2021 | ||
1e7f3a48 WAA |
2022 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2023 | if (!hdr) { | |
a7d42ddb | 2024 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); |
6c75dc0d | 2025 | return -ENOMEM; |
dce81290 | 2026 | } |
6c75dc0d | 2027 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 2028 | |
9369a431 | 2029 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2030 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2031 | if (!ret) |
7f714720 | 2032 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 2033 | |
6c75dc0d | 2034 | return ret; |
dce81290 TM |
2035 | } |
2036 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
2037 | ||
53113ad3 | 2038 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
2039 | { |
2040 | struct nfs_pageio_descriptor pgio; | |
2041 | ||
1acbbb4e | 2042 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
2043 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
2044 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 2045 | } |
e7dd79af | 2046 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 2047 | |
d45f60c6 | 2048 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2049 | { |
cd841605 FI |
2050 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
2051 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2052 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2053 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2054 | } |
4db6e0b7 | 2055 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2056 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
2057 | } |
2058 | ||
d20581aa BH |
2059 | /* |
2060 | * Called by non rpc-based layout drivers | |
2061 | */ | |
d45f60c6 | 2062 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 2063 | { |
cd841605 | 2064 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
2065 | __nfs4_read_done_cb(hdr); |
2066 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
f8417b48 KM |
2067 | } |
2068 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); | |
2069 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2070 | pnfs_ld_handle_read_error(hdr); |
2071 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
2072 | } |
2073 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
2074 | ||
493292dd TM |
2075 | static void |
2076 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2077 | struct nfs_pgio_header *hdr) |
493292dd | 2078 | { |
48d635f1 | 2079 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2080 | |
4db6e0b7 | 2081 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2082 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 2083 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 2084 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 2085 | } |
d45f60c6 | 2086 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2087 | hdr->release(hdr); |
493292dd TM |
2088 | } |
2089 | ||
64419a9b AA |
2090 | /* |
2091 | * Call the appropriate parallel I/O subsystem read function. | |
2092 | */ | |
493292dd | 2093 | static enum pnfs_try_status |
d45f60c6 | 2094 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
2095 | const struct rpc_call_ops *call_ops, |
2096 | struct pnfs_layout_segment *lseg) | |
64419a9b | 2097 | { |
cd841605 | 2098 | struct inode *inode = hdr->inode; |
64419a9b AA |
2099 | struct nfs_server *nfss = NFS_SERVER(inode); |
2100 | enum pnfs_try_status trypnfs; | |
2101 | ||
cd841605 | 2102 | hdr->mds_ops = call_ops; |
64419a9b AA |
2103 | |
2104 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 2105 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 2106 | |
d45f60c6 | 2107 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 2108 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 2109 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
2110 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2111 | return trypnfs; | |
2112 | } | |
863a3c6c | 2113 | |
ceb11e13 PT |
2114 | /* Resend all requests through pnfs. */ |
2115 | int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) | |
2116 | { | |
2117 | struct nfs_pageio_descriptor pgio; | |
2118 | ||
2119 | nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops); | |
2120 | return nfs_pageio_resend(&pgio, hdr); | |
2121 | } | |
2122 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
2123 | ||
493292dd | 2124 | static void |
7f714720 | 2125 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 2126 | { |
493292dd TM |
2127 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2128 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2129 | enum pnfs_try_status trypnfs; |
ceb11e13 | 2130 | int err = 0; |
493292dd | 2131 | |
d45f60c6 | 2132 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
ceb11e13 PT |
2133 | if (trypnfs == PNFS_TRY_AGAIN) |
2134 | err = pnfs_read_resend_pnfs(hdr); | |
2135 | if (trypnfs == PNFS_NOT_ATTEMPTED || err) | |
d45f60c6 | 2136 | pnfs_read_through_mds(desc, hdr); |
493292dd TM |
2137 | } |
2138 | ||
4db6e0b7 FI |
2139 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
2140 | { | |
9369a431 | 2141 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2142 | nfs_pgio_header_free(hdr); |
4db6e0b7 FI |
2143 | } |
2144 | ||
493292dd TM |
2145 | int |
2146 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
2147 | { | |
48d635f1 | 2148 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2149 | |
4db6e0b7 | 2150 | struct nfs_pgio_header *hdr; |
493292dd TM |
2151 | int ret; |
2152 | ||
1e7f3a48 WAA |
2153 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2154 | if (!hdr) { | |
a7d42ddb | 2155 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); |
180bb5ec | 2156 | return -ENOMEM; |
493292dd | 2157 | } |
4db6e0b7 | 2158 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 2159 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2160 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2161 | if (!ret) |
7f714720 | 2162 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 2163 | return ret; |
493292dd TM |
2164 | } |
2165 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
2166 | ||
71244d9b TM |
2167 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
2168 | { | |
2169 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
2170 | ||
2171 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 2172 | smp_mb__after_atomic(); |
71244d9b TM |
2173 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
2174 | } | |
2175 | ||
863a3c6c | 2176 | /* |
a9bae566 | 2177 | * There can be multiple RW segments. |
863a3c6c | 2178 | */ |
a9bae566 | 2179 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 2180 | { |
a9bae566 | 2181 | struct pnfs_layout_segment *lseg; |
863a3c6c | 2182 | |
a9bae566 PT |
2183 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
2184 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 2185 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
2186 | list_add(&lseg->pls_lc_list, listp); |
2187 | } | |
863a3c6c AA |
2188 | } |
2189 | ||
a073dbff TM |
2190 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
2191 | { | |
2192 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
2193 | |
2194 | /* Matched by references in pnfs_set_layoutcommit */ | |
2195 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
2196 | list_del_init(&lseg->pls_lc_list); | |
2197 | pnfs_put_lseg(lseg); | |
2198 | } | |
2199 | ||
71244d9b | 2200 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
2201 | } |
2202 | ||
1b0ae068 PT |
2203 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
2204 | { | |
b9e028fd | 2205 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
2206 | } |
2207 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
2208 | ||
863a3c6c | 2209 | void |
67af7611 TM |
2210 | pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, |
2211 | loff_t end_pos) | |
863a3c6c | 2212 | { |
cd841605 | 2213 | struct nfs_inode *nfsi = NFS_I(inode); |
79a48a1f | 2214 | bool mark_as_dirty = false; |
863a3c6c | 2215 | |
cd841605 | 2216 | spin_lock(&inode->i_lock); |
863a3c6c | 2217 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
29559b11 | 2218 | nfsi->layout->plh_lwb = end_pos; |
79a48a1f | 2219 | mark_as_dirty = true; |
863a3c6c | 2220 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 2221 | __func__, inode->i_ino); |
29559b11 TM |
2222 | } else if (end_pos > nfsi->layout->plh_lwb) |
2223 | nfsi->layout->plh_lwb = end_pos; | |
67af7611 | 2224 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) { |
a9bae566 | 2225 | /* references matched in nfs4_layoutcommit_release */ |
67af7611 | 2226 | pnfs_get_lseg(lseg); |
a9bae566 | 2227 | } |
cd841605 | 2228 | spin_unlock(&inode->i_lock); |
acff5880 | 2229 | dprintk("%s: lseg %p end_pos %llu\n", |
67af7611 | 2230 | __func__, lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
2231 | |
2232 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2233 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2234 | if (mark_as_dirty) | |
cd841605 | 2235 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
2236 | } |
2237 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
2238 | ||
db29c089 AA |
2239 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2240 | { | |
2241 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2242 | ||
2243 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2244 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2245 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2246 | } |
2247 | ||
de4b15c7 AA |
2248 | /* |
2249 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2250 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2251 | * data to disk to allow the server to recover the data if it crashes. | |
2252 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2253 | * is off, and a COMMIT is sent to a data server, or | |
2254 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2255 | */ | |
863a3c6c | 2256 | int |
ef311537 | 2257 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2258 | { |
5f919c9f | 2259 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2260 | struct nfs4_layoutcommit_data *data; |
2261 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2262 | loff_t end_pos; |
71244d9b | 2263 | int status; |
863a3c6c | 2264 | |
71244d9b | 2265 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2266 | return 0; |
2267 | ||
71244d9b | 2268 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2269 | |
71244d9b | 2270 | status = -EAGAIN; |
92407e75 | 2271 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2272 | if (!sync) |
2273 | goto out; | |
74316201 | 2274 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2275 | NFS_INO_LAYOUTCOMMITTING, |
2276 | nfs_wait_bit_killable, | |
2277 | TASK_KILLABLE); | |
92407e75 | 2278 | if (status) |
71244d9b | 2279 | goto out; |
92407e75 PT |
2280 | } |
2281 | ||
71244d9b TM |
2282 | status = -ENOMEM; |
2283 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2284 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2285 | if (!data) | |
2286 | goto clear_layoutcommitting; | |
2287 | ||
2288 | status = 0; | |
de4b15c7 | 2289 | spin_lock(&inode->i_lock); |
71244d9b TM |
2290 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2291 | goto out_unlock; | |
a9bae566 | 2292 | |
71244d9b | 2293 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2294 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2295 | |
acff5880 | 2296 | end_pos = nfsi->layout->plh_lwb; |
863a3c6c | 2297 | |
f597c537 | 2298 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2299 | spin_unlock(&inode->i_lock); |
2300 | ||
2301 | data->args.inode = inode; | |
9fa40758 | 2302 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2303 | nfs_fattr_init(&data->fattr); |
2304 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2305 | data->res.fattr = &data->fattr; | |
2306 | data->args.lastbytewritten = end_pos - 1; | |
2307 | data->res.server = NFS_SERVER(inode); | |
2308 | ||
5f919c9f CH |
2309 | if (ld->prepare_layoutcommit) { |
2310 | status = ld->prepare_layoutcommit(&data->args); | |
2311 | if (status) { | |
3471648a | 2312 | put_rpccred(data->cred); |
5f919c9f | 2313 | spin_lock(&inode->i_lock); |
29559b11 TM |
2314 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); |
2315 | if (end_pos > nfsi->layout->plh_lwb) | |
5f919c9f | 2316 | nfsi->layout->plh_lwb = end_pos; |
3471648a | 2317 | goto out_unlock; |
5f919c9f CH |
2318 | } |
2319 | } | |
2320 | ||
2321 | ||
863a3c6c AA |
2322 | status = nfs4_proc_layoutcommit(data, sync); |
2323 | out: | |
92407e75 PT |
2324 | if (status) |
2325 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2326 | dprintk("<-- %s status %d\n", __func__, status); |
2327 | return status; | |
71244d9b TM |
2328 | out_unlock: |
2329 | spin_unlock(&inode->i_lock); | |
92407e75 | 2330 | kfree(data); |
71244d9b TM |
2331 | clear_layoutcommitting: |
2332 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2333 | goto out; |
863a3c6c | 2334 | } |
72cff449 | 2335 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a | 2336 | |
5bb89b47 TM |
2337 | int |
2338 | pnfs_generic_sync(struct inode *inode, bool datasync) | |
2339 | { | |
2340 | return pnfs_layoutcommit_inode(inode, true); | |
2341 | } | |
2342 | EXPORT_SYMBOL_GPL(pnfs_generic_sync); | |
2343 | ||
82be417a AA |
2344 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) |
2345 | { | |
2346 | struct nfs4_threshold *thp; | |
2347 | ||
2348 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2349 | if (!thp) { | |
2350 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2351 | return NULL; | |
2352 | } | |
2353 | return thp; | |
2354 | } | |
8733408d | 2355 | |
865a7ecb | 2356 | #if IS_ENABLED(CONFIG_NFS_V4_2) |
8733408d | 2357 | int |
c8ad8894 | 2358 | pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags) |
8733408d PT |
2359 | { |
2360 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
2361 | struct nfs_server *server = NFS_SERVER(inode); | |
1bfe3b25 | 2362 | struct nfs_inode *nfsi = NFS_I(inode); |
8733408d PT |
2363 | struct nfs42_layoutstat_data *data; |
2364 | struct pnfs_layout_hdr *hdr; | |
2365 | int status = 0; | |
2366 | ||
2367 | if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats) | |
2368 | goto out; | |
2369 | ||
6c5a0d89 TM |
2370 | if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS)) |
2371 | goto out; | |
2372 | ||
1bfe3b25 PT |
2373 | if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags)) |
2374 | goto out; | |
2375 | ||
8733408d PT |
2376 | spin_lock(&inode->i_lock); |
2377 | if (!NFS_I(inode)->layout) { | |
2378 | spin_unlock(&inode->i_lock); | |
2379 | goto out; | |
2380 | } | |
2381 | hdr = NFS_I(inode)->layout; | |
2382 | pnfs_get_layout_hdr(hdr); | |
2383 | spin_unlock(&inode->i_lock); | |
2384 | ||
c8ad8894 | 2385 | data = kzalloc(sizeof(*data), gfp_flags); |
8733408d PT |
2386 | if (!data) { |
2387 | status = -ENOMEM; | |
2388 | goto out_put; | |
2389 | } | |
2390 | ||
2391 | data->args.fh = NFS_FH(inode); | |
2392 | data->args.inode = inode; | |
2393 | nfs4_stateid_copy(&data->args.stateid, &hdr->plh_stateid); | |
2394 | status = ld->prepare_layoutstats(&data->args); | |
2395 | if (status) | |
2396 | goto out_free; | |
2397 | ||
2398 | status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data); | |
2399 | ||
2400 | out: | |
2401 | dprintk("%s returns %d\n", __func__, status); | |
2402 | return status; | |
2403 | ||
2404 | out_free: | |
2405 | kfree(data); | |
2406 | out_put: | |
2407 | pnfs_put_layout_hdr(hdr); | |
1bfe3b25 PT |
2408 | smp_mb__before_atomic(); |
2409 | clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags); | |
2410 | smp_mb__after_atomic(); | |
8733408d PT |
2411 | goto out; |
2412 | } | |
2413 | EXPORT_SYMBOL_GPL(pnfs_report_layoutstat); | |
865a7ecb | 2414 | #endif |
bbf58bf3 TM |
2415 | |
2416 | unsigned int layoutstats_timer; | |
2417 | module_param(layoutstats_timer, uint, 0644); | |
2418 | EXPORT_SYMBOL_GPL(layoutstats_timer); |