Commit | Line | Data |
---|---|---|
85e174ba RL |
1 | /* |
2 | * pNFS functions to call and manage layout drivers. | |
3 | * | |
4 | * Copyright (c) 2002 [year of first publication] | |
5 | * The Regents of the University of Michigan | |
6 | * All Rights Reserved | |
7 | * | |
8 | * Dean Hildebrand <dhildebz@umich.edu> | |
9 | * | |
10 | * Permission is granted to use, copy, create derivative works, and | |
11 | * redistribute this software and such derivative works for any purpose, | |
12 | * so long as the name of the University of Michigan is not used in | |
13 | * any advertising or publicity pertaining to the use or distribution | |
14 | * of this software without specific, written prior authorization. If | |
15 | * the above copyright notice or any other identification of the | |
16 | * University of Michigan is included in any copy of any portion of | |
17 | * this software, then the disclaimer below must also be included. | |
18 | * | |
19 | * This software is provided as is, without representation or warranty | |
20 | * of any kind either express or implied, including without limitation | |
21 | * the implied warranties of merchantability, fitness for a particular | |
22 | * purpose, or noninfringement. The Regents of the University of | |
23 | * Michigan shall not be liable for any damages, including special, | |
24 | * indirect, incidental, or consequential damages, with respect to any | |
25 | * claim arising out of or in connection with the use of the software, | |
26 | * even if it has been or is hereafter advised of the possibility of | |
27 | * such damages. | |
28 | */ | |
29 | ||
30 | #include <linux/nfs_fs.h> | |
493292dd | 31 | #include <linux/nfs_page.h> |
143cb494 | 32 | #include <linux/module.h> |
974cec8c | 33 | #include "internal.h" |
85e174ba | 34 | #include "pnfs.h" |
64419a9b | 35 | #include "iostat.h" |
cc668ab3 | 36 | #include "nfs4trace.h" |
85e174ba RL |
37 | |
38 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
25c75333 | 39 | #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) |
85e174ba | 40 | |
02c35fca FI |
41 | /* Locking: |
42 | * | |
43 | * pnfs_spinlock: | |
44 | * protects pnfs_modules_tbl. | |
45 | */ | |
46 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
47 | ||
48 | /* | |
49 | * pnfs_modules_tbl holds all pnfs modules | |
50 | */ | |
51 | static LIST_HEAD(pnfs_modules_tbl); | |
52 | ||
aa1e0e3a PT |
53 | static int |
54 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid, | |
6c16605d | 55 | enum pnfs_iomode iomode, bool sync); |
aa1e0e3a | 56 | |
02c35fca FI |
57 | /* Return the registered pnfs layout driver module matching given id */ |
58 | static struct pnfs_layoutdriver_type * | |
59 | find_pnfs_driver_locked(u32 id) | |
60 | { | |
61 | struct pnfs_layoutdriver_type *local; | |
62 | ||
63 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
64 | if (local->id == id) | |
65 | goto out; | |
66 | local = NULL; | |
67 | out: | |
68 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
69 | return local; | |
70 | } | |
71 | ||
85e174ba RL |
72 | static struct pnfs_layoutdriver_type * |
73 | find_pnfs_driver(u32 id) | |
74 | { | |
02c35fca FI |
75 | struct pnfs_layoutdriver_type *local; |
76 | ||
77 | spin_lock(&pnfs_spinlock); | |
78 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
79 | if (local != NULL && !try_module_get(local->owner)) { |
80 | dprintk("%s: Could not grab reference on module\n", __func__); | |
81 | local = NULL; | |
82 | } | |
02c35fca FI |
83 | spin_unlock(&pnfs_spinlock); |
84 | return local; | |
85e174ba RL |
85 | } |
86 | ||
87 | void | |
88 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
89 | { | |
738fd0f3 BH |
90 | if (nfss->pnfs_curr_ld) { |
91 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
92 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
93 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
94 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
95 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 96 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 97 | } |
85e174ba RL |
98 | nfss->pnfs_curr_ld = NULL; |
99 | } | |
100 | ||
101 | /* | |
102 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
103 | * Currently only one pNFS layout driver per filesystem is supported. | |
104 | * | |
105 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
106 | */ | |
107 | void | |
738fd0f3 BH |
108 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
109 | u32 id) | |
85e174ba RL |
110 | { |
111 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
112 | ||
113 | if (id == 0) | |
114 | goto out_no_driver; | |
115 | if (!(server->nfs_client->cl_exchange_flags & | |
116 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
a030889a WAA |
117 | printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n", |
118 | __func__, id, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
119 | goto out_no_driver; |
120 | } | |
121 | ld_type = find_pnfs_driver(id); | |
122 | if (!ld_type) { | |
123 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
124 | ld_type = find_pnfs_driver(id); | |
125 | if (!ld_type) { | |
126 | dprintk("%s: No pNFS module found for %u.\n", | |
127 | __func__, id); | |
128 | goto out_no_driver; | |
129 | } | |
130 | } | |
131 | server->pnfs_curr_ld = ld_type; | |
738fd0f3 BH |
132 | if (ld_type->set_layoutdriver |
133 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
134 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
135 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
136 | module_put(ld_type->owner); |
137 | goto out_no_driver; | |
138 | } | |
2a4c8994 TM |
139 | /* Bump the MDS count */ |
140 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 141 | |
85e174ba RL |
142 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
143 | return; | |
144 | ||
145 | out_no_driver: | |
146 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
147 | server->pnfs_curr_ld = NULL; | |
148 | } | |
02c35fca FI |
149 | |
150 | int | |
151 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
152 | { | |
153 | int status = -EINVAL; | |
154 | struct pnfs_layoutdriver_type *tmp; | |
155 | ||
156 | if (ld_type->id == 0) { | |
a030889a | 157 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
158 | return status; |
159 | } | |
b1f69b75 | 160 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 161 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
162 | "alloc_lseg and free_lseg.\n", __func__); |
163 | return status; | |
164 | } | |
02c35fca FI |
165 | |
166 | spin_lock(&pnfs_spinlock); | |
167 | tmp = find_pnfs_driver_locked(ld_type->id); | |
168 | if (!tmp) { | |
169 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
170 | status = 0; | |
171 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
172 | ld_type->name); | |
173 | } else { | |
a030889a | 174 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
175 | __func__, ld_type->id); |
176 | } | |
177 | spin_unlock(&pnfs_spinlock); | |
178 | ||
179 | return status; | |
180 | } | |
181 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
182 | ||
183 | void | |
184 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
185 | { | |
186 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
187 | spin_lock(&pnfs_spinlock); | |
188 | list_del(&ld_type->pnfs_tblid); | |
189 | spin_unlock(&pnfs_spinlock); | |
190 | } | |
191 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 192 | |
b1f69b75 AA |
193 | /* |
194 | * pNFS client layout cache | |
195 | */ | |
196 | ||
cc6e5340 | 197 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 198 | void |
70c3bd2b | 199 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 200 | { |
cc6e5340 | 201 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
202 | } |
203 | ||
636fb9c8 BH |
204 | static struct pnfs_layout_hdr * |
205 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
206 | { | |
207 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
57934278 | 208 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
209 | } |
210 | ||
211 | static void | |
212 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
213 | { | |
9c626381 TM |
214 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
215 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
216 | ||
217 | if (!list_empty(&lo->plh_layouts)) { | |
218 | struct nfs_client *clp = server->nfs_client; | |
219 | ||
220 | spin_lock(&clp->cl_lock); | |
221 | list_del_init(&lo->plh_layouts); | |
222 | spin_unlock(&clp->cl_lock); | |
223 | } | |
9fa40758 | 224 | put_rpccred(lo->plh_lc_cred); |
57934278 | 225 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
226 | } |
227 | ||
e5e94017 | 228 | static void |
6622c3ea | 229 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 230 | { |
bb346f63 | 231 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 232 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
233 | nfsi->layout = NULL; |
234 | /* Reset MDS Threshold I/O counters */ | |
235 | nfsi->write_io = 0; | |
236 | nfsi->read_io = 0; | |
e5e94017 BH |
237 | } |
238 | ||
b1f69b75 | 239 | void |
70c3bd2b | 240 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 241 | { |
cc6e5340 FI |
242 | struct inode *inode = lo->plh_inode; |
243 | ||
244 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
566f8737 PT |
245 | if (!list_empty(&lo->plh_segs)) |
246 | WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n"); | |
6622c3ea | 247 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 248 | spin_unlock(&inode->i_lock); |
6622c3ea | 249 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 250 | } |
974cec8c AA |
251 | } |
252 | ||
b9e028fd TM |
253 | static int |
254 | pnfs_iomode_to_fail_bit(u32 iomode) | |
255 | { | |
256 | return iomode == IOMODE_RW ? | |
257 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
258 | } | |
259 | ||
260 | static void | |
3e621214 | 261 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 262 | { |
25c75333 | 263 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 264 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
3e621214 TM |
265 | atomic_inc(&lo->plh_refcount); |
266 | } | |
267 | ||
268 | static void | |
269 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
270 | { | |
271 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
272 | atomic_dec(&lo->plh_refcount); | |
273 | } | |
274 | ||
275 | static void | |
276 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
277 | { | |
278 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
279 | struct pnfs_layout_range range = { |
280 | .iomode = iomode, | |
281 | .offset = 0, | |
282 | .length = NFS4_MAX_UINT64, | |
283 | }; | |
284 | LIST_HEAD(head); | |
3e621214 TM |
285 | |
286 | spin_lock(&inode->i_lock); | |
287 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
115ce575 | 288 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range); |
3e621214 | 289 | spin_unlock(&inode->i_lock); |
115ce575 | 290 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
291 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
292 | iomode == IOMODE_RW ? "RW" : "READ"); | |
293 | } | |
294 | ||
295 | static bool | |
296 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
297 | { | |
25c75333 | 298 | unsigned long start, end; |
3e621214 TM |
299 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
300 | ||
301 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
302 | return false; |
303 | end = jiffies; | |
304 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
305 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
306 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 307 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
308 | return false; |
309 | } | |
310 | return true; | |
b9e028fd TM |
311 | } |
312 | ||
974cec8c AA |
313 | static void |
314 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
315 | { | |
566052c5 | 316 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 317 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c FI |
318 | atomic_set(&lseg->pls_refcount, 1); |
319 | smp_mb(); | |
320 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 321 | lseg->pls_layout = lo; |
974cec8c AA |
322 | } |
323 | ||
905ca191 | 324 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 325 | { |
b7edfaa1 | 326 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 327 | |
b1f69b75 | 328 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
974cec8c AA |
329 | } |
330 | ||
d684d2ae | 331 | static void |
57036a37 TM |
332 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
333 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 334 | { |
57036a37 | 335 | struct inode *inode = lo->plh_inode; |
d684d2ae | 336 | |
d20581aa | 337 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 338 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
339 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
340 | atomic_dec(&lo->plh_refcount); | |
173f77e9 TM |
341 | if (list_empty(&lo->plh_segs)) |
342 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
d684d2ae FI |
343 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); |
344 | } | |
345 | ||
aa1e0e3a PT |
346 | /* Return true if layoutreturn is needed */ |
347 | static bool | |
348 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo, | |
349 | struct pnfs_layout_segment *lseg, | |
350 | nfs4_stateid *stateid, enum pnfs_iomode *iomode) | |
351 | { | |
352 | struct pnfs_layout_segment *s; | |
353 | ||
354 | if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
355 | return false; | |
356 | ||
357 | list_for_each_entry(s, &lo->plh_segs, pls_list) | |
358 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
359 | return false; | |
360 | ||
361 | *stateid = lo->plh_stateid; | |
362 | *iomode = lo->plh_return_iomode; | |
363 | /* decreased in pnfs_send_layoutreturn() */ | |
364 | lo->plh_block_lgets++; | |
365 | lo->plh_return_iomode = 0; | |
366 | return true; | |
367 | } | |
368 | ||
bae724ef | 369 | void |
9369a431 | 370 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 371 | { |
57036a37 | 372 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
373 | struct inode *inode; |
374 | ||
375 | if (!lseg) | |
376 | return; | |
377 | ||
4541d16c FI |
378 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
379 | atomic_read(&lseg->pls_refcount), | |
380 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
57036a37 TM |
381 | lo = lseg->pls_layout; |
382 | inode = lo->plh_inode; | |
d684d2ae | 383 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
aa1e0e3a PT |
384 | bool need_return; |
385 | nfs4_stateid stateid; | |
386 | enum pnfs_iomode iomode; | |
387 | ||
8f0d27dc | 388 | pnfs_get_layout_hdr(lo); |
57036a37 | 389 | pnfs_layout_remove_lseg(lo, lseg); |
aa1e0e3a PT |
390 | need_return = pnfs_layout_need_return(lo, lseg, |
391 | &stateid, &iomode); | |
d684d2ae | 392 | spin_unlock(&inode->i_lock); |
905ca191 | 393 | pnfs_free_lseg(lseg); |
aa1e0e3a | 394 | if (need_return) |
6c16605d PT |
395 | pnfs_send_layoutreturn(lo, stateid, iomode, |
396 | true); | |
aa1e0e3a PT |
397 | else |
398 | pnfs_put_layout_hdr(lo); | |
4541d16c | 399 | } |
4541d16c | 400 | } |
9369a431 | 401 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 402 | |
6543f803 | 403 | static void pnfs_free_lseg_async_work(struct work_struct *work) |
e6cf82d1 WAA |
404 | { |
405 | struct pnfs_layout_segment *lseg; | |
6543f803 | 406 | struct pnfs_layout_hdr *lo; |
e6cf82d1 WAA |
407 | |
408 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
6543f803 | 409 | lo = lseg->pls_layout; |
e6cf82d1 | 410 | |
6543f803 TM |
411 | pnfs_free_lseg(lseg); |
412 | pnfs_put_layout_hdr(lo); | |
e6cf82d1 WAA |
413 | } |
414 | ||
6543f803 | 415 | static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg) |
e6cf82d1 | 416 | { |
6543f803 | 417 | INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work); |
e6cf82d1 WAA |
418 | schedule_work(&lseg->pls_work); |
419 | } | |
6543f803 TM |
420 | |
421 | void | |
422 | pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg) | |
423 | { | |
424 | if (!lseg) | |
425 | return; | |
426 | ||
427 | assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock); | |
428 | ||
429 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, | |
430 | atomic_read(&lseg->pls_refcount), | |
431 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
432 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
433 | struct pnfs_layout_hdr *lo = lseg->pls_layout; | |
434 | pnfs_get_layout_hdr(lo); | |
435 | pnfs_layout_remove_lseg(lo, lseg); | |
436 | pnfs_free_lseg_async(lseg); | |
437 | } | |
438 | } | |
439 | EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked); | |
e6cf82d1 | 440 | |
3cb2df17 | 441 | static u64 |
fb3296eb BH |
442 | end_offset(u64 start, u64 len) |
443 | { | |
444 | u64 end; | |
445 | ||
446 | end = start + len; | |
447 | return end >= start ? end : NFS4_MAX_UINT64; | |
448 | } | |
449 | ||
fb3296eb BH |
450 | /* |
451 | * is l2 fully contained in l1? | |
452 | * start1 end1 | |
453 | * [----------------------------------) | |
454 | * start2 end2 | |
455 | * [----------------) | |
456 | */ | |
3cb2df17 | 457 | static bool |
7dc0ac70 | 458 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 459 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
460 | { |
461 | u64 start1 = l1->offset; | |
462 | u64 end1 = end_offset(start1, l1->length); | |
463 | u64 start2 = l2->offset; | |
464 | u64 end2 = end_offset(start2, l2->length); | |
465 | ||
466 | return (start1 <= start2) && (end1 >= end2); | |
467 | } | |
468 | ||
469 | /* | |
470 | * is l1 and l2 intersecting? | |
471 | * start1 end1 | |
472 | * [----------------------------------) | |
473 | * start2 end2 | |
474 | * [----------------) | |
475 | */ | |
3cb2df17 | 476 | static bool |
7dc0ac70 | 477 | pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1, |
3cb2df17 | 478 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
479 | { |
480 | u64 start1 = l1->offset; | |
481 | u64 end1 = end_offset(start1, l1->length); | |
482 | u64 start2 = l2->offset; | |
483 | u64 end2 = end_offset(start2, l2->length); | |
484 | ||
485 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
486 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
487 | } | |
488 | ||
4541d16c | 489 | static bool |
3cb2df17 TM |
490 | should_free_lseg(const struct pnfs_layout_range *lseg_range, |
491 | const struct pnfs_layout_range *recall_range) | |
4541d16c | 492 | { |
778b5502 BH |
493 | return (recall_range->iomode == IOMODE_ANY || |
494 | lseg_range->iomode == recall_range->iomode) && | |
7dc0ac70 | 495 | pnfs_lseg_range_intersecting(lseg_range, recall_range); |
974cec8c AA |
496 | } |
497 | ||
24956804 TM |
498 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
499 | struct list_head *tmp_list) | |
500 | { | |
501 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
502 | return false; | |
503 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
504 | list_add(&lseg->pls_list, tmp_list); | |
505 | return true; | |
506 | } | |
507 | ||
4541d16c FI |
508 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
509 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
510 | struct list_head *tmp_list) | |
511 | { | |
512 | int rv = 0; | |
513 | ||
514 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
515 | /* Remove the reference keeping the lseg in the | |
516 | * list. It will now be removed when all | |
517 | * outstanding io is finished. | |
518 | */ | |
d684d2ae FI |
519 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
520 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 521 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 522 | rv = 1; |
4541d16c FI |
523 | } |
524 | return rv; | |
525 | } | |
526 | ||
527 | /* Returns count of number of matching invalid lsegs remaining in list | |
528 | * after call. | |
529 | */ | |
43f1b3da | 530 | int |
49a85061 | 531 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 532 | struct list_head *tmp_list, |
778b5502 | 533 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
534 | { |
535 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 536 | int invalid = 0, removed = 0; |
974cec8c AA |
537 | |
538 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
539 | ||
8006bfba | 540 | if (list_empty(&lo->plh_segs)) |
38511722 | 541 | return 0; |
4541d16c | 542 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
543 | if (!recall_range || |
544 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
545 | dprintk("%s: freeing lseg %p iomode %d " |
546 | "offset %llu length %llu\n", __func__, | |
547 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
548 | lseg->pls_range.length); | |
549 | invalid++; | |
550 | removed += mark_lseg_invalid(lseg, tmp_list); | |
551 | } | |
552 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
553 | return invalid - removed; | |
974cec8c AA |
554 | } |
555 | ||
f49f9baa | 556 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 557 | void |
4541d16c | 558 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 559 | { |
4541d16c | 560 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
561 | |
562 | if (list_empty(free_me)) | |
563 | return; | |
564 | ||
4541d16c | 565 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 566 | list_del(&lseg->pls_list); |
905ca191 | 567 | pnfs_free_lseg(lseg); |
974cec8c AA |
568 | } |
569 | } | |
570 | ||
e5e94017 BH |
571 | void |
572 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
573 | { | |
574 | struct pnfs_layout_hdr *lo; | |
974cec8c | 575 | LIST_HEAD(tmp_list); |
e5e94017 BH |
576 | |
577 | spin_lock(&nfsi->vfs_inode.i_lock); | |
578 | lo = nfsi->layout; | |
579 | if (lo) { | |
38511722 | 580 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 581 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
3e621214 TM |
582 | pnfs_get_layout_hdr(lo); |
583 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); | |
584 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
585 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
586 | pnfs_free_lseg_list(&tmp_list); | |
587 | pnfs_put_layout_hdr(lo); | |
588 | } else | |
589 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 590 | } |
041245c8 | 591 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 592 | |
fd9a8d71 TM |
593 | static bool |
594 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
595 | struct list_head *layout_list) | |
974cec8c AA |
596 | { |
597 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 598 | bool ret = false; |
974cec8c | 599 | |
fd9a8d71 TM |
600 | spin_lock(&inode->i_lock); |
601 | lo = NFS_I(inode)->layout; | |
602 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
603 | pnfs_get_layout_hdr(lo); | |
604 | list_add(&lo->plh_bulk_destroy, layout_list); | |
605 | ret = true; | |
606 | } | |
607 | spin_unlock(&inode->i_lock); | |
608 | return ret; | |
609 | } | |
610 | ||
611 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
612 | static int | |
613 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
614 | struct nfs_server *server, | |
615 | struct list_head *layout_list) | |
616 | { | |
617 | struct pnfs_layout_hdr *lo, *next; | |
618 | struct inode *inode; | |
619 | ||
620 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
621 | inode = igrab(lo->plh_inode); | |
622 | if (inode == NULL) | |
623 | continue; | |
624 | list_del_init(&lo->plh_layouts); | |
625 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
626 | continue; | |
627 | rcu_read_unlock(); | |
628 | spin_unlock(&clp->cl_lock); | |
629 | iput(inode); | |
630 | spin_lock(&clp->cl_lock); | |
631 | rcu_read_lock(); | |
632 | return -EAGAIN; | |
633 | } | |
634 | return 0; | |
635 | } | |
636 | ||
637 | static int | |
638 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
639 | bool is_bulk_recall) | |
640 | { | |
641 | struct pnfs_layout_hdr *lo; | |
642 | struct inode *inode; | |
643 | struct pnfs_layout_range range = { | |
644 | .iomode = IOMODE_ANY, | |
645 | .offset = 0, | |
646 | .length = NFS4_MAX_UINT64, | |
647 | }; | |
648 | LIST_HEAD(lseg_list); | |
649 | int ret = 0; | |
650 | ||
651 | while (!list_empty(layout_list)) { | |
652 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
653 | plh_bulk_destroy); | |
654 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
655 | lo->plh_inode->i_ino); | |
656 | inode = lo->plh_inode; | |
7c5d1875 CH |
657 | |
658 | pnfs_layoutcommit_inode(inode, false); | |
659 | ||
fd9a8d71 TM |
660 | spin_lock(&inode->i_lock); |
661 | list_del_init(&lo->plh_bulk_destroy); | |
662 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ | |
663 | if (is_bulk_recall) | |
664 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
665 | if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range)) | |
666 | ret = -EAGAIN; | |
667 | spin_unlock(&inode->i_lock); | |
668 | pnfs_free_lseg_list(&lseg_list); | |
669 | pnfs_put_layout_hdr(lo); | |
670 | iput(inode); | |
671 | } | |
672 | return ret; | |
673 | } | |
674 | ||
675 | int | |
676 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
677 | struct nfs_fsid *fsid, | |
678 | bool is_recall) | |
679 | { | |
680 | struct nfs_server *server; | |
681 | LIST_HEAD(layout_list); | |
c47abcf8 | 682 | |
974cec8c | 683 | spin_lock(&clp->cl_lock); |
6382a441 | 684 | rcu_read_lock(); |
fd9a8d71 | 685 | restart: |
6382a441 | 686 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
687 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
688 | continue; | |
689 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
690 | server, | |
691 | &layout_list) != 0) | |
692 | goto restart; | |
6382a441 WAA |
693 | } |
694 | rcu_read_unlock(); | |
974cec8c AA |
695 | spin_unlock(&clp->cl_lock); |
696 | ||
fd9a8d71 TM |
697 | if (list_empty(&layout_list)) |
698 | return 0; | |
699 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
700 | } | |
701 | ||
702 | int | |
703 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
704 | bool is_recall) | |
705 | { | |
706 | struct nfs_server *server; | |
707 | LIST_HEAD(layout_list); | |
708 | ||
709 | spin_lock(&clp->cl_lock); | |
710 | rcu_read_lock(); | |
711 | restart: | |
712 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
713 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
714 | server, | |
715 | &layout_list) != 0) | |
716 | goto restart; | |
974cec8c | 717 | } |
fd9a8d71 TM |
718 | rcu_read_unlock(); |
719 | spin_unlock(&clp->cl_lock); | |
720 | ||
721 | if (list_empty(&layout_list)) | |
722 | return 0; | |
723 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
724 | } | |
725 | ||
726 | /* | |
727 | * Called by the state manger to remove all layouts established under an | |
728 | * expired lease. | |
729 | */ | |
730 | void | |
731 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
732 | { | |
733 | nfs4_deviceid_mark_client_invalid(clp); | |
734 | nfs4_deviceid_purge_client(clp); | |
735 | ||
736 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
737 | } |
738 | ||
5a65503f TM |
739 | /* |
740 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
741 | * taking into account wraparound issues. | |
742 | */ | |
743 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
744 | { | |
2c64c57d | 745 | return (s32)(s1 - s2) > 0; |
5a65503f TM |
746 | } |
747 | ||
fd6002e9 | 748 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
749 | void |
750 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
751 | bool update_barrier) | |
b1f69b75 | 752 | { |
22aaf714 TM |
753 | u32 oldseq, newseq, new_barrier; |
754 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 755 | |
2d2f24ad TM |
756 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
757 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 758 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 759 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 760 | if (update_barrier) { |
22aaf714 | 761 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
762 | } else { |
763 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 764 | * "close" to the current seqids. |
43f1b3da | 765 | */ |
22aaf714 | 766 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 767 | } |
22aaf714 TM |
768 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
769 | lo->plh_barrier = new_barrier; | |
43f1b3da | 770 | } |
b1f69b75 AA |
771 | } |
772 | ||
cf7d63f1 | 773 | static bool |
19c54aba TM |
774 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
775 | const nfs4_stateid *stateid) | |
43f1b3da | 776 | { |
19c54aba | 777 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 778 | |
19c54aba TM |
779 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
780 | } | |
781 | ||
ce6ab4f2 PT |
782 | static bool |
783 | pnfs_layout_returning(const struct pnfs_layout_hdr *lo, | |
784 | struct pnfs_layout_range *range) | |
785 | { | |
786 | return test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) && | |
787 | (lo->plh_return_iomode == IOMODE_ANY || | |
788 | lo->plh_return_iomode == range->iomode); | |
789 | } | |
790 | ||
19c54aba TM |
791 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
792 | static bool | |
ce6ab4f2 PT |
793 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, |
794 | struct pnfs_layout_range *range, int lget) | |
19c54aba | 795 | { |
f7e8917a FI |
796 | return lo->plh_block_lgets || |
797 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || | |
43f1b3da | 798 | (list_empty(&lo->plh_segs) && |
ce6ab4f2 PT |
799 | (atomic_read(&lo->plh_outstanding) > lget)) || |
800 | pnfs_layout_returning(lo, range); | |
cf7d63f1 FI |
801 | } |
802 | ||
fd6002e9 FI |
803 | int |
804 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
ce6ab4f2 | 805 | struct pnfs_layout_range *range, |
fd6002e9 | 806 | struct nfs4_state *open_state) |
b1f69b75 | 807 | { |
fd6002e9 | 808 | int status = 0; |
974cec8c | 809 | |
b1f69b75 | 810 | dprintk("--> %s\n", __func__); |
fd6002e9 | 811 | spin_lock(&lo->plh_inode->i_lock); |
ce6ab4f2 | 812 | if (pnfs_layoutgets_blocked(lo, range, 1)) { |
cf7d63f1 | 813 | status = -EAGAIN; |
5d422301 TM |
814 | } else if (!nfs4_valid_open_stateid(open_state)) { |
815 | status = -EBADF; | |
47abadef CH |
816 | } else if (list_empty(&lo->plh_segs) || |
817 | test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { | |
fd6002e9 FI |
818 | int seq; |
819 | ||
820 | do { | |
821 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 822 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
823 | } while (read_seqretry(&open_state->seqlock, seq)); |
824 | } else | |
f597c537 | 825 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 826 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 827 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 828 | return status; |
b1f69b75 AA |
829 | } |
830 | ||
831 | /* | |
832 | * Get layout from server. | |
833 | * for now, assume that whole file layouts are requested. | |
834 | * arg->offset: 0 | |
835 | * arg->length: all ones | |
836 | */ | |
e5e94017 BH |
837 | static struct pnfs_layout_segment * |
838 | send_layoutget(struct pnfs_layout_hdr *lo, | |
839 | struct nfs_open_context *ctx, | |
fb3296eb | 840 | struct pnfs_layout_range *range, |
a75b9df9 | 841 | gfp_t gfp_flags) |
e5e94017 | 842 | { |
b7edfaa1 | 843 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
844 | struct nfs_server *server = NFS_SERVER(ino); |
845 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 846 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
847 | |
848 | dprintk("--> %s\n", __func__); | |
e5e94017 | 849 | |
a75b9df9 | 850 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 851 | if (lgp == NULL) |
b1f69b75 | 852 | return NULL; |
35124a09 | 853 | |
fb3296eb BH |
854 | lgp->args.minlength = PAGE_CACHE_SIZE; |
855 | if (lgp->args.minlength > range->length) | |
856 | lgp->args.minlength = range->length; | |
b1f69b75 | 857 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 858 | lgp->args.range = *range; |
b1f69b75 AA |
859 | lgp->args.type = server->pnfs_curr_ld->id; |
860 | lgp->args.inode = ino; | |
861 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 862 | lgp->gfp_flags = gfp_flags; |
6ab59344 | 863 | lgp->cred = lo->plh_lc_cred; |
b1f69b75 AA |
864 | |
865 | /* Synchronously retrieve layout information from server and | |
866 | * store in lseg. | |
867 | */ | |
a0b0a6e3 TM |
868 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
869 | if (IS_ERR(lseg)) { | |
870 | switch (PTR_ERR(lseg)) { | |
871 | case -ENOMEM: | |
872 | case -ERESTARTSYS: | |
873 | break; | |
874 | default: | |
875 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 876 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
877 | } |
878 | return NULL; | |
974cec8c | 879 | } |
35124a09 | 880 | |
974cec8c AA |
881 | return lseg; |
882 | } | |
883 | ||
24956804 TM |
884 | static void pnfs_clear_layoutcommit(struct inode *inode, |
885 | struct list_head *head) | |
886 | { | |
887 | struct nfs_inode *nfsi = NFS_I(inode); | |
888 | struct pnfs_layout_segment *lseg, *tmp; | |
889 | ||
890 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
891 | return; | |
892 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
893 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
894 | continue; | |
895 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
896 | } | |
897 | } | |
898 | ||
f40eb5d0 PT |
899 | static int |
900 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid, | |
6c16605d | 901 | enum pnfs_iomode iomode, bool sync) |
f40eb5d0 PT |
902 | { |
903 | struct inode *ino = lo->plh_inode; | |
904 | struct nfs4_layoutreturn *lrp; | |
905 | int status = 0; | |
906 | ||
907 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); | |
908 | if (unlikely(lrp == NULL)) { | |
909 | status = -ENOMEM; | |
910 | spin_lock(&ino->i_lock); | |
911 | lo->plh_block_lgets--; | |
193e3aa2 | 912 | rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq); |
f40eb5d0 PT |
913 | spin_unlock(&ino->i_lock); |
914 | pnfs_put_layout_hdr(lo); | |
915 | goto out; | |
916 | } | |
917 | ||
918 | lrp->args.stateid = stateid; | |
919 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
920 | lrp->args.inode = ino; | |
15eb67c1 PT |
921 | lrp->args.range.iomode = iomode; |
922 | lrp->args.range.offset = 0; | |
923 | lrp->args.range.length = NFS4_MAX_UINT64; | |
f40eb5d0 PT |
924 | lrp->args.layout = lo; |
925 | lrp->clp = NFS_SERVER(ino)->nfs_client; | |
926 | lrp->cred = lo->plh_lc_cred; | |
927 | ||
6c16605d | 928 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
929 | out: |
930 | dprintk("<-- %s status: %d\n", __func__, status); | |
931 | return status; | |
932 | } | |
933 | ||
293b3b06 AA |
934 | /* |
935 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
936 | * when the layout segment list is empty. | |
937 | * | |
938 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
939 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
940 | * deviceid is marked invalid. | |
941 | */ | |
cbe82603 BH |
942 | int |
943 | _pnfs_return_layout(struct inode *ino) | |
944 | { | |
945 | struct pnfs_layout_hdr *lo = NULL; | |
946 | struct nfs_inode *nfsi = NFS_I(ino); | |
947 | LIST_HEAD(tmp_list); | |
cbe82603 | 948 | nfs4_stateid stateid; |
293b3b06 | 949 | int status = 0, empty; |
cbe82603 | 950 | |
366d5052 | 951 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
952 | |
953 | spin_lock(&ino->i_lock); | |
954 | lo = nfsi->layout; | |
e5929f3c | 955 | if (!lo) { |
cbe82603 | 956 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
957 | dprintk("NFS: %s no layout to return\n", __func__); |
958 | goto out; | |
cbe82603 BH |
959 | } |
960 | stateid = nfsi->layout->plh_stateid; | |
961 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 962 | pnfs_get_layout_hdr(lo); |
293b3b06 | 963 | empty = list_empty(&lo->plh_segs); |
24956804 | 964 | pnfs_clear_layoutcommit(ino, &tmp_list); |
49a85061 | 965 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
c88953d8 CH |
966 | |
967 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
968 | struct pnfs_layout_range range = { | |
969 | .iomode = IOMODE_ANY, | |
970 | .offset = 0, | |
971 | .length = NFS4_MAX_UINT64, | |
972 | }; | |
973 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
974 | } | |
975 | ||
293b3b06 AA |
976 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
977 | if (empty) { | |
978 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 979 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
980 | dprintk("NFS: %s no layout segments to return\n", __func__); |
981 | goto out; | |
982 | } | |
47abadef CH |
983 | |
984 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
ea0ded74 | 985 | lo->plh_block_lgets++; |
cbe82603 BH |
986 | spin_unlock(&ino->i_lock); |
987 | pnfs_free_lseg_list(&tmp_list); | |
988 | ||
6c16605d | 989 | status = pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true); |
cbe82603 BH |
990 | out: |
991 | dprintk("<-- %s status: %d\n", __func__, status); | |
992 | return status; | |
993 | } | |
0a57cdac | 994 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 995 | |
24028672 TM |
996 | int |
997 | pnfs_commit_and_return_layout(struct inode *inode) | |
998 | { | |
999 | struct pnfs_layout_hdr *lo; | |
1000 | int ret; | |
1001 | ||
1002 | spin_lock(&inode->i_lock); | |
1003 | lo = NFS_I(inode)->layout; | |
1004 | if (lo == NULL) { | |
1005 | spin_unlock(&inode->i_lock); | |
1006 | return 0; | |
1007 | } | |
1008 | pnfs_get_layout_hdr(lo); | |
1009 | /* Block new layoutgets and read/write to ds */ | |
1010 | lo->plh_block_lgets++; | |
1011 | spin_unlock(&inode->i_lock); | |
1012 | filemap_fdatawait(inode->i_mapping); | |
1013 | ret = pnfs_layoutcommit_inode(inode, true); | |
1014 | if (ret == 0) | |
1015 | ret = _pnfs_return_layout(inode); | |
1016 | spin_lock(&inode->i_lock); | |
1017 | lo->plh_block_lgets--; | |
1018 | spin_unlock(&inode->i_lock); | |
1019 | pnfs_put_layout_hdr(lo); | |
1020 | return ret; | |
1021 | } | |
1022 | ||
f7e8917a FI |
1023 | bool pnfs_roc(struct inode *ino) |
1024 | { | |
1025 | struct pnfs_layout_hdr *lo; | |
1026 | struct pnfs_layout_segment *lseg, *tmp; | |
193e3aa2 | 1027 | nfs4_stateid stateid; |
f7e8917a | 1028 | LIST_HEAD(tmp_list); |
193e3aa2 | 1029 | bool found = false, layoutreturn = false; |
f7e8917a FI |
1030 | |
1031 | spin_lock(&ino->i_lock); | |
1032 | lo = NFS_I(ino)->layout; | |
1033 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
1034 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
1035 | goto out_nolayout; | |
1036 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
1037 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
1038 | mark_lseg_invalid(lseg, &tmp_list); | |
1039 | found = true; | |
1040 | } | |
1041 | if (!found) | |
1042 | goto out_nolayout; | |
1043 | lo->plh_block_lgets++; | |
70c3bd2b | 1044 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
1045 | spin_unlock(&ino->i_lock); |
1046 | pnfs_free_lseg_list(&tmp_list); | |
1047 | return true; | |
1048 | ||
1049 | out_nolayout: | |
193e3aa2 PT |
1050 | if (lo) { |
1051 | stateid = lo->plh_stateid; | |
1052 | layoutreturn = | |
1053 | test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, | |
1054 | &lo->plh_flags); | |
1055 | if (layoutreturn) { | |
1056 | lo->plh_block_lgets++; | |
1057 | pnfs_get_layout_hdr(lo); | |
1058 | } | |
1059 | } | |
f7e8917a | 1060 | spin_unlock(&ino->i_lock); |
193e3aa2 PT |
1061 | if (layoutreturn) |
1062 | pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, 0, | |
1063 | NFS4_MAX_UINT64, true); | |
f7e8917a FI |
1064 | return false; |
1065 | } | |
1066 | ||
1067 | void pnfs_roc_release(struct inode *ino) | |
1068 | { | |
1069 | struct pnfs_layout_hdr *lo; | |
1070 | ||
1071 | spin_lock(&ino->i_lock); | |
1072 | lo = NFS_I(ino)->layout; | |
1073 | lo->plh_block_lgets--; | |
6622c3ea TM |
1074 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
1075 | pnfs_detach_layout_hdr(lo); | |
1076 | spin_unlock(&ino->i_lock); | |
1077 | pnfs_free_layout_hdr(lo); | |
1078 | } else | |
1079 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
1080 | } |
1081 | ||
1082 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
1083 | { | |
1084 | struct pnfs_layout_hdr *lo; | |
1085 | ||
1086 | spin_lock(&ino->i_lock); | |
1087 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 1088 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
1089 | lo->plh_barrier = barrier; |
1090 | spin_unlock(&ino->i_lock); | |
1091 | } | |
1092 | ||
7fdab069 | 1093 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
1094 | { |
1095 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 1096 | struct pnfs_layout_hdr *lo; |
f7e8917a | 1097 | struct pnfs_layout_segment *lseg; |
193e3aa2 | 1098 | nfs4_stateid stateid; |
7fdab069 | 1099 | u32 current_seqid; |
193e3aa2 | 1100 | bool found = false, layoutreturn = false; |
f7e8917a FI |
1101 | |
1102 | spin_lock(&ino->i_lock); | |
1103 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
1104 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 1105 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 1106 | found = true; |
7fdab069 | 1107 | goto out; |
f7e8917a | 1108 | } |
7fdab069 TM |
1109 | lo = nfsi->layout; |
1110 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1111 | |
7fdab069 TM |
1112 | /* Since close does not return a layout stateid for use as |
1113 | * a barrier, we choose the worst-case barrier. | |
1114 | */ | |
1115 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
1116 | out: | |
193e3aa2 PT |
1117 | if (!found) { |
1118 | stateid = lo->plh_stateid; | |
1119 | layoutreturn = | |
1120 | test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, | |
1121 | &lo->plh_flags); | |
1122 | if (layoutreturn) { | |
1123 | lo->plh_block_lgets++; | |
1124 | pnfs_get_layout_hdr(lo); | |
1125 | } | |
1126 | } | |
f7e8917a | 1127 | spin_unlock(&ino->i_lock); |
193e3aa2 PT |
1128 | if (layoutreturn) { |
1129 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
1130 | pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, 0, | |
1131 | NFS4_MAX_UINT64, false); | |
1132 | } | |
f7e8917a FI |
1133 | return found; |
1134 | } | |
1135 | ||
b1f69b75 AA |
1136 | /* |
1137 | * Compare two layout segments for sorting into layout cache. | |
1138 | * We want to preferentially return RW over RO layouts, so ensure those | |
1139 | * are seen first. | |
1140 | */ | |
1141 | static s64 | |
7dc0ac70 | 1142 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1143 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1144 | { |
fb3296eb BH |
1145 | s64 d; |
1146 | ||
1147 | /* high offset > low offset */ | |
1148 | d = l1->offset - l2->offset; | |
1149 | if (d) | |
1150 | return d; | |
1151 | ||
1152 | /* short length > long length */ | |
1153 | d = l2->length - l1->length; | |
1154 | if (d) | |
1155 | return d; | |
1156 | ||
b1f69b75 | 1157 | /* read > read/write */ |
fb3296eb | 1158 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1159 | } |
1160 | ||
974cec8c | 1161 | static void |
57036a37 | 1162 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
1163 | struct pnfs_layout_segment *lseg) |
1164 | { | |
b1f69b75 | 1165 | struct pnfs_layout_segment *lp; |
b1f69b75 | 1166 | |
974cec8c AA |
1167 | dprintk("%s:Begin\n", __func__); |
1168 | ||
b7edfaa1 | 1169 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
7dc0ac70 | 1170 | if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 1171 | continue; |
566052c5 | 1172 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1173 | dprintk("%s: inserted lseg %p " |
1174 | "iomode %d offset %llu length %llu before " | |
1175 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1176 | __func__, lseg, lseg->pls_range.iomode, |
1177 | lseg->pls_range.offset, lseg->pls_range.length, | |
1178 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1179 | lp->pls_range.length); | |
fb3296eb | 1180 | goto out; |
974cec8c | 1181 | } |
fb3296eb BH |
1182 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1183 | dprintk("%s: inserted lseg %p " | |
1184 | "iomode %d offset %llu length %llu at tail\n", | |
1185 | __func__, lseg, lseg->pls_range.iomode, | |
1186 | lseg->pls_range.offset, lseg->pls_range.length); | |
1187 | out: | |
70c3bd2b | 1188 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1189 | |
1190 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
1191 | } |
1192 | ||
1193 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1194 | alloc_init_layout_hdr(struct inode *ino, |
1195 | struct nfs_open_context *ctx, | |
1196 | gfp_t gfp_flags) | |
e5e94017 BH |
1197 | { |
1198 | struct pnfs_layout_hdr *lo; | |
1199 | ||
636fb9c8 | 1200 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1201 | if (!lo) |
1202 | return NULL; | |
cc6e5340 | 1203 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1204 | INIT_LIST_HEAD(&lo->plh_layouts); |
1205 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1206 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1207 | lo->plh_inode = ino; |
5cc2216d | 1208 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
e5e94017 BH |
1209 | return lo; |
1210 | } | |
1211 | ||
1212 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1213 | pnfs_find_alloc_layout(struct inode *ino, |
1214 | struct nfs_open_context *ctx, | |
1215 | gfp_t gfp_flags) | |
e5e94017 BH |
1216 | { |
1217 | struct nfs_inode *nfsi = NFS_I(ino); | |
1218 | struct pnfs_layout_hdr *new = NULL; | |
1219 | ||
1220 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1221 | ||
251ec410 TM |
1222 | if (nfsi->layout != NULL) |
1223 | goto out_existing; | |
e5e94017 | 1224 | spin_unlock(&ino->i_lock); |
9fa40758 | 1225 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1226 | spin_lock(&ino->i_lock); |
1227 | ||
251ec410 | 1228 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1229 | nfsi->layout = new; |
251ec410 | 1230 | return new; |
7175fe90 YN |
1231 | } else if (new != NULL) |
1232 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1233 | out_existing: |
1234 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1235 | return nfsi->layout; |
1236 | } | |
1237 | ||
b1f69b75 AA |
1238 | /* |
1239 | * iomode matching rules: | |
1240 | * iomode lseg match | |
1241 | * ----- ----- ----- | |
1242 | * ANY READ true | |
1243 | * ANY RW true | |
1244 | * RW READ false | |
1245 | * RW RW true | |
1246 | * READ READ true | |
1247 | * READ RW true | |
1248 | */ | |
3cb2df17 | 1249 | static bool |
7dc0ac70 | 1250 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
3cb2df17 | 1251 | const struct pnfs_layout_range *range) |
b1f69b75 | 1252 | { |
fb3296eb BH |
1253 | struct pnfs_layout_range range1; |
1254 | ||
1255 | if ((range->iomode == IOMODE_RW && | |
1256 | ls_range->iomode != IOMODE_RW) || | |
7dc0ac70 | 1257 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1258 | return 0; |
1259 | ||
1260 | /* range1 covers only the first byte in the range */ | |
1261 | range1 = *range; | |
1262 | range1.length = 1; | |
7dc0ac70 | 1263 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1264 | } |
1265 | ||
1266 | /* | |
1267 | * lookup range in layout | |
1268 | */ | |
e5e94017 | 1269 | static struct pnfs_layout_segment * |
fb3296eb BH |
1270 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
1271 | struct pnfs_layout_range *range) | |
e5e94017 | 1272 | { |
b1f69b75 AA |
1273 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1274 | ||
1275 | dprintk("%s:Begin\n", __func__); | |
1276 | ||
b7edfaa1 | 1277 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1278 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1279 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
7dc0ac70 | 1280 | pnfs_lseg_range_match(&lseg->pls_range, range)) { |
9369a431 | 1281 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1282 | break; |
1283 | } | |
d771e3a4 | 1284 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
1285 | break; |
1286 | } | |
1287 | ||
1288 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1289 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1290 | return ret; |
e5e94017 BH |
1291 | } |
1292 | ||
d23d61c8 AA |
1293 | /* |
1294 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1295 | * to the MDS or over pNFS | |
1296 | * | |
1297 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1298 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1299 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1300 | * WRITE request. | |
1301 | * | |
1302 | * A return of true means use MDS I/O. | |
1303 | * | |
1304 | * From rfc 5661: | |
1305 | * If a file's size is smaller than the file size threshold, data accesses | |
1306 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1307 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1308 | * server. If both file size and I/O size are provided, the client SHOULD | |
1309 | * reach or exceed both thresholds before sending its read or write | |
1310 | * requests to the data server. | |
1311 | */ | |
1312 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1313 | struct inode *ino, int iomode) | |
1314 | { | |
1315 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1316 | struct nfs_inode *nfsi = NFS_I(ino); | |
1317 | loff_t fsize = i_size_read(ino); | |
1318 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1319 | ||
1320 | if (t == NULL) | |
1321 | return ret; | |
1322 | ||
1323 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1324 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1325 | ||
1326 | switch (iomode) { | |
1327 | case IOMODE_READ: | |
1328 | if (t->bm & THRESHOLD_RD) { | |
1329 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1330 | size_set = true; | |
1331 | if (fsize < t->rd_sz) | |
1332 | size = true; | |
1333 | } | |
1334 | if (t->bm & THRESHOLD_RD_IO) { | |
1335 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1336 | nfsi->read_io); | |
1337 | io_set = true; | |
1338 | if (nfsi->read_io < t->rd_io_sz) | |
1339 | io = true; | |
1340 | } | |
1341 | break; | |
1342 | case IOMODE_RW: | |
1343 | if (t->bm & THRESHOLD_WR) { | |
1344 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1345 | size_set = true; | |
1346 | if (fsize < t->wr_sz) | |
1347 | size = true; | |
1348 | } | |
1349 | if (t->bm & THRESHOLD_WR_IO) { | |
1350 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1351 | nfsi->write_io); | |
1352 | io_set = true; | |
1353 | if (nfsi->write_io < t->wr_io_sz) | |
1354 | io = true; | |
1355 | } | |
1356 | break; | |
1357 | } | |
1358 | if (size_set && io_set) { | |
1359 | if (size && io) | |
1360 | ret = true; | |
1361 | } else if (size || io) | |
1362 | ret = true; | |
1363 | ||
1364 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1365 | return ret; | |
1366 | } | |
1367 | ||
e5e94017 BH |
1368 | /* |
1369 | * Layout segment is retreived from the server if not cached. | |
1370 | * The appropriate layout segment is referenced and returned to the caller. | |
1371 | */ | |
7c24d948 | 1372 | struct pnfs_layout_segment * |
e5e94017 BH |
1373 | pnfs_update_layout(struct inode *ino, |
1374 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1375 | loff_t pos, |
1376 | u64 count, | |
a75b9df9 TM |
1377 | enum pnfs_iomode iomode, |
1378 | gfp_t gfp_flags) | |
e5e94017 | 1379 | { |
fb3296eb BH |
1380 | struct pnfs_layout_range arg = { |
1381 | .iomode = iomode, | |
1382 | .offset = pos, | |
1383 | .length = count, | |
1384 | }; | |
707ed5fd | 1385 | unsigned pg_offset; |
6382a441 WAA |
1386 | struct nfs_server *server = NFS_SERVER(ino); |
1387 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1388 | struct pnfs_layout_hdr *lo; |
1389 | struct pnfs_layout_segment *lseg = NULL; | |
9bf87482 | 1390 | bool first; |
e5e94017 BH |
1391 | |
1392 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1393 | goto out; |
d23d61c8 AA |
1394 | |
1395 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1396 | goto out; |
d23d61c8 | 1397 | |
9bf87482 PT |
1398 | lookup_again: |
1399 | first = false; | |
e5e94017 | 1400 | spin_lock(&ino->i_lock); |
9fa40758 | 1401 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1402 | if (lo == NULL) { |
1403 | spin_unlock(&ino->i_lock); | |
1404 | goto out; | |
1405 | } | |
e5e94017 | 1406 | |
43f1b3da | 1407 | /* Do we even need to bother with this? */ |
a59c30ac | 1408 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1409 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1410 | goto out_unlock; |
1411 | } | |
1412 | ||
1413 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1414 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1415 | goto out_unlock; |
1416 | ||
9bf87482 PT |
1417 | first = list_empty(&lo->plh_segs); |
1418 | if (first) { | |
1419 | /* The first layoutget for the file. Need to serialize per | |
1420 | * RFC 5661 Errata 3208. | |
1421 | */ | |
1422 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1423 | &lo->plh_flags)) { | |
1424 | spin_unlock(&ino->i_lock); | |
1425 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1426 | TASK_UNINTERRUPTIBLE); | |
1427 | pnfs_put_layout_hdr(lo); | |
1428 | goto lookup_again; | |
1429 | } | |
1430 | } else { | |
1431 | /* Check to see if the layout for the given range | |
1432 | * already exists | |
1433 | */ | |
1434 | lseg = pnfs_find_lseg(lo, &arg); | |
1435 | if (lseg) | |
1436 | goto out_unlock; | |
1437 | } | |
568e8c49 | 1438 | |
ce6ab4f2 | 1439 | if (pnfs_layoutgets_blocked(lo, &arg, 0)) |
cf7d63f1 FI |
1440 | goto out_unlock; |
1441 | atomic_inc(&lo->plh_outstanding); | |
f49f9baa | 1442 | spin_unlock(&ino->i_lock); |
30005121 | 1443 | |
abb9a007 | 1444 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1445 | /* The lo must be on the clp list if there is any |
1446 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1447 | */ | |
1448 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1449 | if (list_empty(&lo->plh_layouts)) |
1450 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1451 | spin_unlock(&clp->cl_lock); |
1452 | } | |
e5e94017 | 1453 | |
707ed5fd BH |
1454 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1455 | if (pg_offset) { | |
1456 | arg.offset -= pg_offset; | |
1457 | arg.length += pg_offset; | |
1458 | } | |
7c24d948 AA |
1459 | if (arg.length != NFS4_MAX_UINT64) |
1460 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1461 | |
fb3296eb | 1462 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
cf7d63f1 | 1463 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1464 | out_put_layout_hdr: |
9bf87482 PT |
1465 | if (first) { |
1466 | unsigned long *bitlock = &lo->plh_flags; | |
1467 | ||
1468 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1469 | smp_mb__after_atomic(); | |
1470 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1471 | } | |
70c3bd2b | 1472 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1473 | out: |
f86bbcf8 TM |
1474 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1475 | "(%s, offset: %llu, length: %llu)\n", | |
1476 | __func__, ino->i_sb->s_id, | |
1477 | (unsigned long long)NFS_FILEID(ino), | |
1478 | lseg == NULL ? "not found" : "found", | |
1479 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1480 | (unsigned long long)pos, | |
1481 | (unsigned long long)count); | |
e5e94017 BH |
1482 | return lseg; |
1483 | out_unlock: | |
1484 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1485 | goto out_put_layout_hdr; |
e5e94017 | 1486 | } |
7c24d948 | 1487 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1488 | |
a0b0a6e3 | 1489 | struct pnfs_layout_segment * |
b1f69b75 AA |
1490 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1491 | { | |
1492 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1493 | struct nfs4_layoutget_res *res = &lgp->res; | |
1494 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1495 | struct inode *ino = lo->plh_inode; |
78096cca | 1496 | LIST_HEAD(free_me); |
b1f69b75 AA |
1497 | int status = 0; |
1498 | ||
1499 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1500 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1501 | if (!lseg || IS_ERR(lseg)) { |
1502 | if (!lseg) | |
1503 | status = -ENOMEM; | |
1504 | else | |
1505 | status = PTR_ERR(lseg); | |
1506 | dprintk("%s: Could not allocate layout: error %d\n", | |
1507 | __func__, status); | |
1508 | goto out; | |
1509 | } | |
1510 | ||
1013df61 CH |
1511 | init_lseg(lo, lseg); |
1512 | lseg->pls_range = res->range; | |
1513 | ||
b1f69b75 | 1514 | spin_lock(&ino->i_lock); |
a59c30ac | 1515 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1516 | dprintk("%s forget reply due to recall\n", __func__); |
1517 | goto out_forget_reply; | |
1518 | } | |
1519 | ||
ce6ab4f2 | 1520 | if (pnfs_layoutgets_blocked(lo, &lgp->args.range, 1)) { |
43f1b3da FI |
1521 | dprintk("%s forget reply due to state\n", __func__); |
1522 | goto out_forget_reply; | |
1523 | } | |
038d6493 | 1524 | |
362f7474 CH |
1525 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1526 | /* existing state ID, make sure the sequence number matches. */ | |
1527 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1528 | dprintk("%s forget reply due to sequence\n", __func__); | |
1529 | goto out_forget_reply; | |
1530 | } | |
1531 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1532 | } else { | |
1533 | /* | |
1534 | * We got an entirely new state ID. Mark all segments for the | |
1535 | * inode invalid, and don't bother validating the stateid | |
1536 | * sequence number. | |
1537 | */ | |
1538 | pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL); | |
1539 | ||
1540 | nfs4_stateid_copy(&lo->plh_stateid, &res->stateid); | |
1541 | lo->plh_barrier = be32_to_cpu(res->stateid.seqid); | |
1542 | } | |
038d6493 | 1543 | |
47abadef CH |
1544 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); |
1545 | ||
9369a431 | 1546 | pnfs_get_lseg(lseg); |
57036a37 | 1547 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1548 | |
f7e8917a FI |
1549 | if (res->return_on_close) { |
1550 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1551 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1552 | } | |
1553 | ||
b1f69b75 | 1554 | spin_unlock(&ino->i_lock); |
78096cca | 1555 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1556 | return lseg; |
b1f69b75 | 1557 | out: |
a0b0a6e3 | 1558 | return ERR_PTR(status); |
43f1b3da FI |
1559 | |
1560 | out_forget_reply: | |
1561 | spin_unlock(&ino->i_lock); | |
1562 | lseg->pls_layout = lo; | |
1563 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1564 | goto out; | |
b1f69b75 AA |
1565 | } |
1566 | ||
016256df PT |
1567 | static void |
1568 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, | |
1569 | struct list_head *tmp_list, | |
1570 | struct pnfs_layout_range *return_range) | |
1571 | { | |
1572 | struct pnfs_layout_segment *lseg, *next; | |
1573 | ||
1574 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
1575 | ||
1576 | if (list_empty(&lo->plh_segs)) | |
1577 | return; | |
1578 | ||
1579 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
1580 | if (should_free_lseg(&lseg->pls_range, return_range)) { | |
1581 | dprintk("%s: marking lseg %p iomode %d " | |
1582 | "offset %llu length %llu\n", __func__, | |
1583 | lseg, lseg->pls_range.iomode, | |
1584 | lseg->pls_range.offset, | |
1585 | lseg->pls_range.length); | |
1586 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
1587 | mark_lseg_invalid(lseg, tmp_list); | |
1588 | } | |
1589 | } | |
1590 | ||
1591 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
1592 | struct pnfs_layout_segment *lseg) | |
1593 | { | |
1594 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
1595 | int iomode = pnfs_iomode_to_fail_bit(lseg->pls_range.iomode); | |
1596 | struct pnfs_layout_range range = { | |
1597 | .iomode = lseg->pls_range.iomode, | |
1598 | .offset = 0, | |
1599 | .length = NFS4_MAX_UINT64, | |
1600 | }; | |
1601 | LIST_HEAD(free_me); | |
1602 | ||
1603 | spin_lock(&inode->i_lock); | |
1604 | /* set failure bit so that pnfs path will be retried later */ | |
1605 | pnfs_layout_set_fail_bit(lo, iomode); | |
1606 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
1607 | if (lo->plh_return_iomode == 0) | |
1608 | lo->plh_return_iomode = range.iomode; | |
1609 | else if (lo->plh_return_iomode != range.iomode) | |
1610 | lo->plh_return_iomode = IOMODE_ANY; | |
1611 | /* | |
1612 | * mark all matching lsegs so that we are sure to have no live | |
1613 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
1614 | * for how it works. | |
1615 | */ | |
1616 | pnfs_mark_matching_lsegs_return(lo, &free_me, &range); | |
1617 | spin_unlock(&inode->i_lock); | |
1618 | pnfs_free_lseg_list(&free_me); | |
1619 | } | |
1620 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
1621 | ||
d8007d4d TM |
1622 | void |
1623 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1624 | { | |
1fd937bd PT |
1625 | u64 rd_size = req->wb_bytes; |
1626 | ||
bc5a89b3 | 1627 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d | 1628 | |
1fd937bd PT |
1629 | if (pgio->pg_dreq == NULL) |
1630 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1631 | else | |
1632 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1633 | ||
d8007d4d TM |
1634 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1635 | req->wb_context, | |
1636 | req_offset(req), | |
1fd937bd | 1637 | rd_size, |
d8007d4d TM |
1638 | IOMODE_READ, |
1639 | GFP_KERNEL); | |
e885de1a TM |
1640 | /* If no lseg, fall back to read through mds */ |
1641 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1642 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1643 | |
d8007d4d TM |
1644 | } |
1645 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1646 | ||
1647 | void | |
6296556f PT |
1648 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1649 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1650 | { |
bc5a89b3 | 1651 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d TM |
1652 | |
1653 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1654 | req->wb_context, | |
1655 | req_offset(req), | |
6296556f | 1656 | wb_size, |
d8007d4d TM |
1657 | IOMODE_RW, |
1658 | GFP_NOFS); | |
e885de1a TM |
1659 | /* If no lseg, fall back to write through mds */ |
1660 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1661 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1662 | } |
1663 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1664 | ||
180bb5ec WAA |
1665 | void |
1666 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
1667 | { | |
1668 | if (desc->pg_lseg) { | |
1669 | pnfs_put_lseg(desc->pg_lseg); | |
1670 | desc->pg_lseg = NULL; | |
1671 | } | |
1672 | } | |
1673 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
1674 | ||
b4fdac1a WAA |
1675 | /* |
1676 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1677 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1678 | */ | |
1679 | size_t | |
a7d42ddb WAA |
1680 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
1681 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 1682 | { |
0f9c429e | 1683 | unsigned int size; |
c5e20cb7 | 1684 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1685 | |
1686 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1687 | if (!size) |
1688 | return 0; | |
94ad1c80 | 1689 | |
19982ba8 | 1690 | /* |
c5e20cb7 WAA |
1691 | * 'size' contains the number of bytes left in the current page (up |
1692 | * to the original size asked for in @req->wb_bytes). | |
1693 | * | |
1694 | * Calculate how many bytes are left in the layout segment | |
1695 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1696 | * |
1697 | * Please also note that 'end_offset' is actually the offset of the | |
1698 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1699 | * | |
1700 | */ | |
19b54848 | 1701 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1702 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1703 | pgio->pg_lseg->pls_range.length); | |
1704 | req_start = req_offset(req); | |
1705 | WARN_ON_ONCE(req_start > seg_end); | |
1706 | /* start of request is past the last byte of this segment */ | |
1707 | if (req_start >= seg_end) | |
19b54848 | 1708 | return 0; |
c5e20cb7 WAA |
1709 | |
1710 | /* adjust 'size' iff there are fewer bytes left in the | |
1711 | * segment than what nfs_generic_pg_test returned */ | |
1712 | seg_left = seg_end - req_start; | |
1713 | if (seg_left < size) | |
1714 | size = (unsigned int)seg_left; | |
19b54848 | 1715 | } |
0f9c429e | 1716 | |
19b54848 | 1717 | return size; |
94ad1c80 | 1718 | } |
89a58e32 | 1719 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1720 | |
53113ad3 | 1721 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
1722 | { |
1723 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
1724 | |
1725 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
1726 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
1727 | hdr->completion_ops); | |
1728 | return nfs_pageio_resend(&pgio, hdr); | |
e2fecb21 | 1729 | } |
e7dd79af | 1730 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1731 | |
d45f60c6 | 1732 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1733 | { |
cd841605 FI |
1734 | |
1735 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1736 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1737 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1738 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1739 | } |
6c75dc0d | 1740 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1741 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
1742 | } |
1743 | ||
d20581aa BH |
1744 | /* |
1745 | * Called by non rpc-based layout drivers | |
1746 | */ | |
d45f60c6 | 1747 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 1748 | { |
d45f60c6 | 1749 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); |
cd841605 | 1750 | if (!hdr->pnfs_error) { |
d45f60c6 WAA |
1751 | pnfs_set_layoutcommit(hdr); |
1752 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
1acbbb4e | 1753 | } else |
d45f60c6 WAA |
1754 | pnfs_ld_handle_write_error(hdr); |
1755 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 1756 | } |
d20581aa | 1757 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1758 | |
dce81290 TM |
1759 | static void |
1760 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1761 | struct nfs_pgio_header *hdr) |
dce81290 | 1762 | { |
48d635f1 | 1763 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1764 | |
6c75dc0d | 1765 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 1766 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 1767 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 1768 | mirror->pg_recoalesce = 1; |
6c75dc0d | 1769 | } |
d45f60c6 | 1770 | nfs_pgio_data_destroy(hdr); |
dce81290 TM |
1771 | } |
1772 | ||
1773 | static enum pnfs_try_status | |
d45f60c6 | 1774 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
1775 | const struct rpc_call_ops *call_ops, |
1776 | struct pnfs_layout_segment *lseg, | |
1777 | int how) | |
0382b744 | 1778 | { |
cd841605 | 1779 | struct inode *inode = hdr->inode; |
0382b744 AA |
1780 | enum pnfs_try_status trypnfs; |
1781 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1782 | ||
cd841605 | 1783 | hdr->mds_ops = call_ops; |
0382b744 AA |
1784 | |
1785 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
1786 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
1787 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 1788 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1789 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1790 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1791 | return trypnfs; | |
1792 | } | |
1793 | ||
dce81290 | 1794 | static void |
7f714720 WAA |
1795 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
1796 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 1797 | { |
dce81290 TM |
1798 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1799 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1800 | enum pnfs_try_status trypnfs; |
dce81290 | 1801 | |
d45f60c6 | 1802 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 1803 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1804 | pnfs_write_through_mds(desc, hdr); |
dce81290 TM |
1805 | } |
1806 | ||
6c75dc0d FI |
1807 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1808 | { | |
9369a431 | 1809 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1810 | nfs_pgio_header_free(hdr); |
6c75dc0d | 1811 | } |
89d77c8f | 1812 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1813 | |
dce81290 TM |
1814 | int |
1815 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1816 | { | |
48d635f1 | 1817 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1818 | |
6c75dc0d | 1819 | struct nfs_pgio_header *hdr; |
dce81290 TM |
1820 | int ret; |
1821 | ||
1e7f3a48 WAA |
1822 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1823 | if (!hdr) { | |
a7d42ddb | 1824 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); |
6c75dc0d | 1825 | return -ENOMEM; |
dce81290 | 1826 | } |
6c75dc0d | 1827 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 1828 | |
9369a431 | 1829 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1830 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 1831 | if (!ret) |
7f714720 | 1832 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 1833 | |
6c75dc0d | 1834 | return ret; |
dce81290 TM |
1835 | } |
1836 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1837 | ||
53113ad3 | 1838 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
1839 | { |
1840 | struct nfs_pageio_descriptor pgio; | |
1841 | ||
1acbbb4e | 1842 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
1843 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
1844 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 1845 | } |
e7dd79af | 1846 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 1847 | |
d45f60c6 | 1848 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1849 | { |
cd841605 FI |
1850 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
1851 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1852 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1853 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1854 | } |
4db6e0b7 | 1855 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1856 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
1857 | } |
1858 | ||
d20581aa BH |
1859 | /* |
1860 | * Called by non rpc-based layout drivers | |
1861 | */ | |
d45f60c6 | 1862 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 1863 | { |
d45f60c6 | 1864 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
cd841605 | 1865 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
1866 | __nfs4_read_done_cb(hdr); |
1867 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
62e4a769 | 1868 | } else |
d45f60c6 WAA |
1869 | pnfs_ld_handle_read_error(hdr); |
1870 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
1871 | } |
1872 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1873 | ||
493292dd TM |
1874 | static void |
1875 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1876 | struct nfs_pgio_header *hdr) |
493292dd | 1877 | { |
48d635f1 | 1878 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1879 | |
4db6e0b7 | 1880 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 1881 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 1882 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 1883 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 1884 | } |
d45f60c6 | 1885 | nfs_pgio_data_destroy(hdr); |
493292dd TM |
1886 | } |
1887 | ||
64419a9b AA |
1888 | /* |
1889 | * Call the appropriate parallel I/O subsystem read function. | |
1890 | */ | |
493292dd | 1891 | static enum pnfs_try_status |
d45f60c6 | 1892 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
1893 | const struct rpc_call_ops *call_ops, |
1894 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1895 | { |
cd841605 | 1896 | struct inode *inode = hdr->inode; |
64419a9b AA |
1897 | struct nfs_server *nfss = NFS_SERVER(inode); |
1898 | enum pnfs_try_status trypnfs; | |
1899 | ||
cd841605 | 1900 | hdr->mds_ops = call_ops; |
64419a9b AA |
1901 | |
1902 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 1903 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 1904 | |
d45f60c6 | 1905 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 1906 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1907 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1908 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1909 | return trypnfs; | |
1910 | } | |
863a3c6c | 1911 | |
ceb11e13 PT |
1912 | /* Resend all requests through pnfs. */ |
1913 | int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) | |
1914 | { | |
1915 | struct nfs_pageio_descriptor pgio; | |
1916 | ||
1917 | nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops); | |
1918 | return nfs_pageio_resend(&pgio, hdr); | |
1919 | } | |
1920 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
1921 | ||
493292dd | 1922 | static void |
7f714720 | 1923 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 1924 | { |
493292dd TM |
1925 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1926 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1927 | enum pnfs_try_status trypnfs; |
ceb11e13 | 1928 | int err = 0; |
493292dd | 1929 | |
d45f60c6 | 1930 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
ceb11e13 PT |
1931 | if (trypnfs == PNFS_TRY_AGAIN) |
1932 | err = pnfs_read_resend_pnfs(hdr); | |
1933 | if (trypnfs == PNFS_NOT_ATTEMPTED || err) | |
d45f60c6 | 1934 | pnfs_read_through_mds(desc, hdr); |
493292dd TM |
1935 | } |
1936 | ||
4db6e0b7 FI |
1937 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1938 | { | |
9369a431 | 1939 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1940 | nfs_pgio_header_free(hdr); |
4db6e0b7 | 1941 | } |
89d77c8f | 1942 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1943 | |
493292dd TM |
1944 | int |
1945 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1946 | { | |
48d635f1 | 1947 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1948 | |
4db6e0b7 | 1949 | struct nfs_pgio_header *hdr; |
493292dd TM |
1950 | int ret; |
1951 | ||
1e7f3a48 WAA |
1952 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1953 | if (!hdr) { | |
a7d42ddb | 1954 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); |
180bb5ec | 1955 | return -ENOMEM; |
493292dd | 1956 | } |
4db6e0b7 | 1957 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 1958 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1959 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 1960 | if (!ret) |
7f714720 | 1961 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 1962 | return ret; |
493292dd TM |
1963 | } |
1964 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1965 | ||
71244d9b TM |
1966 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
1967 | { | |
1968 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
1969 | ||
1970 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 1971 | smp_mb__after_atomic(); |
71244d9b TM |
1972 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
1973 | } | |
1974 | ||
863a3c6c | 1975 | /* |
a9bae566 | 1976 | * There can be multiple RW segments. |
863a3c6c | 1977 | */ |
a9bae566 | 1978 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1979 | { |
a9bae566 | 1980 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1981 | |
a9bae566 PT |
1982 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1983 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 1984 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
1985 | list_add(&lseg->pls_lc_list, listp); |
1986 | } | |
863a3c6c AA |
1987 | } |
1988 | ||
a073dbff TM |
1989 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
1990 | { | |
1991 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
1992 | |
1993 | /* Matched by references in pnfs_set_layoutcommit */ | |
1994 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
1995 | list_del_init(&lseg->pls_lc_list); | |
1996 | pnfs_put_lseg(lseg); | |
1997 | } | |
1998 | ||
71244d9b | 1999 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
2000 | } |
2001 | ||
1b0ae068 PT |
2002 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
2003 | { | |
b9e028fd | 2004 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
2005 | } |
2006 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
2007 | ||
863a3c6c | 2008 | void |
d45f60c6 | 2009 | pnfs_set_layoutcommit(struct nfs_pgio_header *hdr) |
863a3c6c | 2010 | { |
cd841605 FI |
2011 | struct inode *inode = hdr->inode; |
2012 | struct nfs_inode *nfsi = NFS_I(inode); | |
d45f60c6 | 2013 | loff_t end_pos = hdr->mds_offset + hdr->res.count; |
79a48a1f | 2014 | bool mark_as_dirty = false; |
863a3c6c | 2015 | |
cd841605 | 2016 | spin_lock(&inode->i_lock); |
863a3c6c | 2017 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 2018 | mark_as_dirty = true; |
863a3c6c | 2019 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 2020 | __func__, inode->i_ino); |
863a3c6c | 2021 | } |
cd841605 | 2022 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 2023 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 2024 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 2025 | } |
acff5880 PT |
2026 | if (end_pos > nfsi->layout->plh_lwb) |
2027 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 2028 | spin_unlock(&inode->i_lock); |
acff5880 | 2029 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 2030 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
2031 | |
2032 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2033 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2034 | if (mark_as_dirty) | |
cd841605 | 2035 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
2036 | } |
2037 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
2038 | ||
378520b8 PT |
2039 | void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data) |
2040 | { | |
2041 | struct inode *inode = data->inode; | |
2042 | struct nfs_inode *nfsi = NFS_I(inode); | |
2043 | bool mark_as_dirty = false; | |
2044 | ||
2045 | spin_lock(&inode->i_lock); | |
2046 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { | |
2047 | mark_as_dirty = true; | |
2048 | dprintk("%s: Set layoutcommit for inode %lu ", | |
2049 | __func__, inode->i_ino); | |
2050 | } | |
2051 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) { | |
2052 | /* references matched in nfs4_layoutcommit_release */ | |
2053 | pnfs_get_lseg(data->lseg); | |
2054 | } | |
2055 | if (data->lwb > nfsi->layout->plh_lwb) | |
2056 | nfsi->layout->plh_lwb = data->lwb; | |
2057 | spin_unlock(&inode->i_lock); | |
2058 | dprintk("%s: lseg %p end_pos %llu\n", | |
2059 | __func__, data->lseg, nfsi->layout->plh_lwb); | |
2060 | ||
2061 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2062 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2063 | if (mark_as_dirty) | |
2064 | mark_inode_dirty_sync(inode); | |
2065 | } | |
2066 | EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit); | |
2067 | ||
db29c089 AA |
2068 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2069 | { | |
2070 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2071 | ||
2072 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2073 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2074 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2075 | } |
2076 | ||
de4b15c7 AA |
2077 | /* |
2078 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2079 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2080 | * data to disk to allow the server to recover the data if it crashes. | |
2081 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2082 | * is off, and a COMMIT is sent to a data server, or | |
2083 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2084 | */ | |
863a3c6c | 2085 | int |
ef311537 | 2086 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2087 | { |
5f919c9f | 2088 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2089 | struct nfs4_layoutcommit_data *data; |
2090 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2091 | loff_t end_pos; |
71244d9b | 2092 | int status; |
863a3c6c | 2093 | |
71244d9b | 2094 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2095 | return 0; |
2096 | ||
71244d9b | 2097 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2098 | |
71244d9b | 2099 | status = -EAGAIN; |
92407e75 | 2100 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2101 | if (!sync) |
2102 | goto out; | |
74316201 | 2103 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2104 | NFS_INO_LAYOUTCOMMITTING, |
2105 | nfs_wait_bit_killable, | |
2106 | TASK_KILLABLE); | |
92407e75 | 2107 | if (status) |
71244d9b | 2108 | goto out; |
92407e75 PT |
2109 | } |
2110 | ||
71244d9b TM |
2111 | status = -ENOMEM; |
2112 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2113 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2114 | if (!data) | |
2115 | goto clear_layoutcommitting; | |
2116 | ||
2117 | status = 0; | |
de4b15c7 | 2118 | spin_lock(&inode->i_lock); |
71244d9b TM |
2119 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2120 | goto out_unlock; | |
a9bae566 | 2121 | |
71244d9b | 2122 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2123 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2124 | |
acff5880 | 2125 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 2126 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 2127 | |
f597c537 | 2128 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2129 | spin_unlock(&inode->i_lock); |
2130 | ||
2131 | data->args.inode = inode; | |
9fa40758 | 2132 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2133 | nfs_fattr_init(&data->fattr); |
2134 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2135 | data->res.fattr = &data->fattr; | |
2136 | data->args.lastbytewritten = end_pos - 1; | |
2137 | data->res.server = NFS_SERVER(inode); | |
2138 | ||
5f919c9f CH |
2139 | if (ld->prepare_layoutcommit) { |
2140 | status = ld->prepare_layoutcommit(&data->args); | |
2141 | if (status) { | |
2142 | spin_lock(&inode->i_lock); | |
2143 | if (end_pos < nfsi->layout->plh_lwb) | |
2144 | nfsi->layout->plh_lwb = end_pos; | |
2145 | spin_unlock(&inode->i_lock); | |
2146 | put_rpccred(data->cred); | |
2147 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); | |
2148 | goto clear_layoutcommitting; | |
2149 | } | |
2150 | } | |
2151 | ||
2152 | ||
863a3c6c AA |
2153 | status = nfs4_proc_layoutcommit(data, sync); |
2154 | out: | |
92407e75 PT |
2155 | if (status) |
2156 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2157 | dprintk("<-- %s status %d\n", __func__, status); |
2158 | return status; | |
71244d9b TM |
2159 | out_unlock: |
2160 | spin_unlock(&inode->i_lock); | |
92407e75 | 2161 | kfree(data); |
71244d9b TM |
2162 | clear_layoutcommitting: |
2163 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2164 | goto out; |
863a3c6c | 2165 | } |
72cff449 | 2166 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a AA |
2167 | |
2168 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
2169 | { | |
2170 | struct nfs4_threshold *thp; | |
2171 | ||
2172 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2173 | if (!thp) { | |
2174 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2175 | return NULL; | |
2176 | } | |
2177 | return thp; | |
2178 | } |