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, | |
55 | enum pnfs_iomode iomode); | |
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)) { | |
6622c3ea | 245 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 246 | spin_unlock(&inode->i_lock); |
6622c3ea | 247 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 248 | } |
974cec8c AA |
249 | } |
250 | ||
b9e028fd TM |
251 | static int |
252 | pnfs_iomode_to_fail_bit(u32 iomode) | |
253 | { | |
254 | return iomode == IOMODE_RW ? | |
255 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
256 | } | |
257 | ||
258 | static void | |
3e621214 | 259 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 260 | { |
25c75333 | 261 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 262 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
3e621214 TM |
263 | atomic_inc(&lo->plh_refcount); |
264 | } | |
265 | ||
266 | static void | |
267 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
268 | { | |
269 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
270 | atomic_dec(&lo->plh_refcount); | |
271 | } | |
272 | ||
273 | static void | |
274 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
275 | { | |
276 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
277 | struct pnfs_layout_range range = { |
278 | .iomode = iomode, | |
279 | .offset = 0, | |
280 | .length = NFS4_MAX_UINT64, | |
281 | }; | |
282 | LIST_HEAD(head); | |
3e621214 TM |
283 | |
284 | spin_lock(&inode->i_lock); | |
285 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
115ce575 | 286 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range); |
3e621214 | 287 | spin_unlock(&inode->i_lock); |
115ce575 | 288 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
289 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
290 | iomode == IOMODE_RW ? "RW" : "READ"); | |
291 | } | |
292 | ||
293 | static bool | |
294 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
295 | { | |
25c75333 | 296 | unsigned long start, end; |
3e621214 TM |
297 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
298 | ||
299 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
300 | return false; |
301 | end = jiffies; | |
302 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
303 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
304 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 305 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
306 | return false; |
307 | } | |
308 | return true; | |
b9e028fd TM |
309 | } |
310 | ||
974cec8c AA |
311 | static void |
312 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
313 | { | |
566052c5 | 314 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 315 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c FI |
316 | atomic_set(&lseg->pls_refcount, 1); |
317 | smp_mb(); | |
318 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 319 | lseg->pls_layout = lo; |
974cec8c AA |
320 | } |
321 | ||
905ca191 | 322 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 323 | { |
b7edfaa1 | 324 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 325 | |
b1f69b75 | 326 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
974cec8c AA |
327 | } |
328 | ||
d684d2ae | 329 | static void |
57036a37 TM |
330 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
331 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 332 | { |
57036a37 | 333 | struct inode *inode = lo->plh_inode; |
d684d2ae | 334 | |
d20581aa | 335 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 336 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
337 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
338 | atomic_dec(&lo->plh_refcount); | |
173f77e9 TM |
339 | if (list_empty(&lo->plh_segs)) |
340 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
d684d2ae FI |
341 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); |
342 | } | |
343 | ||
aa1e0e3a PT |
344 | /* Return true if layoutreturn is needed */ |
345 | static bool | |
346 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo, | |
347 | struct pnfs_layout_segment *lseg, | |
348 | nfs4_stateid *stateid, enum pnfs_iomode *iomode) | |
349 | { | |
350 | struct pnfs_layout_segment *s; | |
351 | ||
352 | if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
353 | return false; | |
354 | ||
355 | list_for_each_entry(s, &lo->plh_segs, pls_list) | |
356 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
357 | return false; | |
358 | ||
359 | *stateid = lo->plh_stateid; | |
360 | *iomode = lo->plh_return_iomode; | |
361 | /* decreased in pnfs_send_layoutreturn() */ | |
362 | lo->plh_block_lgets++; | |
363 | lo->plh_return_iomode = 0; | |
364 | return true; | |
365 | } | |
366 | ||
bae724ef | 367 | void |
9369a431 | 368 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 369 | { |
57036a37 | 370 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
371 | struct inode *inode; |
372 | ||
373 | if (!lseg) | |
374 | return; | |
375 | ||
4541d16c FI |
376 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
377 | atomic_read(&lseg->pls_refcount), | |
378 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
57036a37 TM |
379 | lo = lseg->pls_layout; |
380 | inode = lo->plh_inode; | |
d684d2ae | 381 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
aa1e0e3a PT |
382 | bool need_return; |
383 | nfs4_stateid stateid; | |
384 | enum pnfs_iomode iomode; | |
385 | ||
8f0d27dc | 386 | pnfs_get_layout_hdr(lo); |
57036a37 | 387 | pnfs_layout_remove_lseg(lo, lseg); |
aa1e0e3a PT |
388 | need_return = pnfs_layout_need_return(lo, lseg, |
389 | &stateid, &iomode); | |
d684d2ae | 390 | spin_unlock(&inode->i_lock); |
905ca191 | 391 | pnfs_free_lseg(lseg); |
aa1e0e3a PT |
392 | if (need_return) |
393 | pnfs_send_layoutreturn(lo, stateid, iomode); | |
394 | else | |
395 | pnfs_put_layout_hdr(lo); | |
4541d16c | 396 | } |
4541d16c | 397 | } |
9369a431 | 398 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 399 | |
6543f803 | 400 | static void pnfs_free_lseg_async_work(struct work_struct *work) |
e6cf82d1 WAA |
401 | { |
402 | struct pnfs_layout_segment *lseg; | |
6543f803 | 403 | struct pnfs_layout_hdr *lo; |
e6cf82d1 WAA |
404 | |
405 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
6543f803 | 406 | lo = lseg->pls_layout; |
e6cf82d1 | 407 | |
6543f803 TM |
408 | pnfs_free_lseg(lseg); |
409 | pnfs_put_layout_hdr(lo); | |
e6cf82d1 WAA |
410 | } |
411 | ||
6543f803 | 412 | static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg) |
e6cf82d1 | 413 | { |
6543f803 | 414 | INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work); |
e6cf82d1 WAA |
415 | schedule_work(&lseg->pls_work); |
416 | } | |
6543f803 TM |
417 | |
418 | void | |
419 | pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg) | |
420 | { | |
421 | if (!lseg) | |
422 | return; | |
423 | ||
424 | assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock); | |
425 | ||
426 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, | |
427 | atomic_read(&lseg->pls_refcount), | |
428 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
429 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
430 | struct pnfs_layout_hdr *lo = lseg->pls_layout; | |
431 | pnfs_get_layout_hdr(lo); | |
432 | pnfs_layout_remove_lseg(lo, lseg); | |
433 | pnfs_free_lseg_async(lseg); | |
434 | } | |
435 | } | |
436 | EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked); | |
e6cf82d1 | 437 | |
3cb2df17 | 438 | static u64 |
fb3296eb BH |
439 | end_offset(u64 start, u64 len) |
440 | { | |
441 | u64 end; | |
442 | ||
443 | end = start + len; | |
444 | return end >= start ? end : NFS4_MAX_UINT64; | |
445 | } | |
446 | ||
fb3296eb BH |
447 | /* |
448 | * is l2 fully contained in l1? | |
449 | * start1 end1 | |
450 | * [----------------------------------) | |
451 | * start2 end2 | |
452 | * [----------------) | |
453 | */ | |
3cb2df17 | 454 | static bool |
7dc0ac70 | 455 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 456 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
457 | { |
458 | u64 start1 = l1->offset; | |
459 | u64 end1 = end_offset(start1, l1->length); | |
460 | u64 start2 = l2->offset; | |
461 | u64 end2 = end_offset(start2, l2->length); | |
462 | ||
463 | return (start1 <= start2) && (end1 >= end2); | |
464 | } | |
465 | ||
466 | /* | |
467 | * is l1 and l2 intersecting? | |
468 | * start1 end1 | |
469 | * [----------------------------------) | |
470 | * start2 end2 | |
471 | * [----------------) | |
472 | */ | |
3cb2df17 | 473 | static bool |
7dc0ac70 | 474 | pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1, |
3cb2df17 | 475 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
476 | { |
477 | u64 start1 = l1->offset; | |
478 | u64 end1 = end_offset(start1, l1->length); | |
479 | u64 start2 = l2->offset; | |
480 | u64 end2 = end_offset(start2, l2->length); | |
481 | ||
482 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
483 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
484 | } | |
485 | ||
4541d16c | 486 | static bool |
3cb2df17 TM |
487 | should_free_lseg(const struct pnfs_layout_range *lseg_range, |
488 | const struct pnfs_layout_range *recall_range) | |
4541d16c | 489 | { |
778b5502 BH |
490 | return (recall_range->iomode == IOMODE_ANY || |
491 | lseg_range->iomode == recall_range->iomode) && | |
7dc0ac70 | 492 | pnfs_lseg_range_intersecting(lseg_range, recall_range); |
974cec8c AA |
493 | } |
494 | ||
24956804 TM |
495 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
496 | struct list_head *tmp_list) | |
497 | { | |
498 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
499 | return false; | |
500 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
501 | list_add(&lseg->pls_list, tmp_list); | |
502 | return true; | |
503 | } | |
504 | ||
4541d16c FI |
505 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
506 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
507 | struct list_head *tmp_list) | |
508 | { | |
509 | int rv = 0; | |
510 | ||
511 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
512 | /* Remove the reference keeping the lseg in the | |
513 | * list. It will now be removed when all | |
514 | * outstanding io is finished. | |
515 | */ | |
d684d2ae FI |
516 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
517 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 518 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 519 | rv = 1; |
4541d16c FI |
520 | } |
521 | return rv; | |
522 | } | |
523 | ||
524 | /* Returns count of number of matching invalid lsegs remaining in list | |
525 | * after call. | |
526 | */ | |
43f1b3da | 527 | int |
49a85061 | 528 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 529 | struct list_head *tmp_list, |
778b5502 | 530 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
531 | { |
532 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 533 | int invalid = 0, removed = 0; |
974cec8c AA |
534 | |
535 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
536 | ||
8006bfba | 537 | if (list_empty(&lo->plh_segs)) |
38511722 | 538 | return 0; |
4541d16c | 539 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
540 | if (!recall_range || |
541 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
542 | dprintk("%s: freeing lseg %p iomode %d " |
543 | "offset %llu length %llu\n", __func__, | |
544 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
545 | lseg->pls_range.length); | |
546 | invalid++; | |
547 | removed += mark_lseg_invalid(lseg, tmp_list); | |
548 | } | |
549 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
550 | return invalid - removed; | |
974cec8c AA |
551 | } |
552 | ||
f49f9baa | 553 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 554 | void |
4541d16c | 555 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 556 | { |
4541d16c | 557 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
558 | |
559 | if (list_empty(free_me)) | |
560 | return; | |
561 | ||
4541d16c | 562 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 563 | list_del(&lseg->pls_list); |
905ca191 | 564 | pnfs_free_lseg(lseg); |
974cec8c AA |
565 | } |
566 | } | |
567 | ||
e5e94017 BH |
568 | void |
569 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
570 | { | |
571 | struct pnfs_layout_hdr *lo; | |
974cec8c | 572 | LIST_HEAD(tmp_list); |
e5e94017 BH |
573 | |
574 | spin_lock(&nfsi->vfs_inode.i_lock); | |
575 | lo = nfsi->layout; | |
576 | if (lo) { | |
38511722 | 577 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 578 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
3e621214 TM |
579 | pnfs_get_layout_hdr(lo); |
580 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); | |
581 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
582 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
583 | pnfs_free_lseg_list(&tmp_list); | |
584 | pnfs_put_layout_hdr(lo); | |
585 | } else | |
586 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 587 | } |
041245c8 | 588 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 589 | |
fd9a8d71 TM |
590 | static bool |
591 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
592 | struct list_head *layout_list) | |
974cec8c AA |
593 | { |
594 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 595 | bool ret = false; |
974cec8c | 596 | |
fd9a8d71 TM |
597 | spin_lock(&inode->i_lock); |
598 | lo = NFS_I(inode)->layout; | |
599 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
600 | pnfs_get_layout_hdr(lo); | |
601 | list_add(&lo->plh_bulk_destroy, layout_list); | |
602 | ret = true; | |
603 | } | |
604 | spin_unlock(&inode->i_lock); | |
605 | return ret; | |
606 | } | |
607 | ||
608 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
609 | static int | |
610 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
611 | struct nfs_server *server, | |
612 | struct list_head *layout_list) | |
613 | { | |
614 | struct pnfs_layout_hdr *lo, *next; | |
615 | struct inode *inode; | |
616 | ||
617 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
618 | inode = igrab(lo->plh_inode); | |
619 | if (inode == NULL) | |
620 | continue; | |
621 | list_del_init(&lo->plh_layouts); | |
622 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
623 | continue; | |
624 | rcu_read_unlock(); | |
625 | spin_unlock(&clp->cl_lock); | |
626 | iput(inode); | |
627 | spin_lock(&clp->cl_lock); | |
628 | rcu_read_lock(); | |
629 | return -EAGAIN; | |
630 | } | |
631 | return 0; | |
632 | } | |
633 | ||
634 | static int | |
635 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
636 | bool is_bulk_recall) | |
637 | { | |
638 | struct pnfs_layout_hdr *lo; | |
639 | struct inode *inode; | |
640 | struct pnfs_layout_range range = { | |
641 | .iomode = IOMODE_ANY, | |
642 | .offset = 0, | |
643 | .length = NFS4_MAX_UINT64, | |
644 | }; | |
645 | LIST_HEAD(lseg_list); | |
646 | int ret = 0; | |
647 | ||
648 | while (!list_empty(layout_list)) { | |
649 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
650 | plh_bulk_destroy); | |
651 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
652 | lo->plh_inode->i_ino); | |
653 | inode = lo->plh_inode; | |
7c5d1875 CH |
654 | |
655 | pnfs_layoutcommit_inode(inode, false); | |
656 | ||
fd9a8d71 TM |
657 | spin_lock(&inode->i_lock); |
658 | list_del_init(&lo->plh_bulk_destroy); | |
659 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ | |
660 | if (is_bulk_recall) | |
661 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
662 | if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range)) | |
663 | ret = -EAGAIN; | |
664 | spin_unlock(&inode->i_lock); | |
665 | pnfs_free_lseg_list(&lseg_list); | |
666 | pnfs_put_layout_hdr(lo); | |
667 | iput(inode); | |
668 | } | |
669 | return ret; | |
670 | } | |
671 | ||
672 | int | |
673 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
674 | struct nfs_fsid *fsid, | |
675 | bool is_recall) | |
676 | { | |
677 | struct nfs_server *server; | |
678 | LIST_HEAD(layout_list); | |
c47abcf8 | 679 | |
974cec8c | 680 | spin_lock(&clp->cl_lock); |
6382a441 | 681 | rcu_read_lock(); |
fd9a8d71 | 682 | restart: |
6382a441 | 683 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
684 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
685 | continue; | |
686 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
687 | server, | |
688 | &layout_list) != 0) | |
689 | goto restart; | |
6382a441 WAA |
690 | } |
691 | rcu_read_unlock(); | |
974cec8c AA |
692 | spin_unlock(&clp->cl_lock); |
693 | ||
fd9a8d71 TM |
694 | if (list_empty(&layout_list)) |
695 | return 0; | |
696 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
697 | } | |
698 | ||
699 | int | |
700 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
701 | bool is_recall) | |
702 | { | |
703 | struct nfs_server *server; | |
704 | LIST_HEAD(layout_list); | |
705 | ||
706 | spin_lock(&clp->cl_lock); | |
707 | rcu_read_lock(); | |
708 | restart: | |
709 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
710 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
711 | server, | |
712 | &layout_list) != 0) | |
713 | goto restart; | |
974cec8c | 714 | } |
fd9a8d71 TM |
715 | rcu_read_unlock(); |
716 | spin_unlock(&clp->cl_lock); | |
717 | ||
718 | if (list_empty(&layout_list)) | |
719 | return 0; | |
720 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
721 | } | |
722 | ||
723 | /* | |
724 | * Called by the state manger to remove all layouts established under an | |
725 | * expired lease. | |
726 | */ | |
727 | void | |
728 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
729 | { | |
730 | nfs4_deviceid_mark_client_invalid(clp); | |
731 | nfs4_deviceid_purge_client(clp); | |
732 | ||
733 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
734 | } |
735 | ||
5a65503f TM |
736 | /* |
737 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
738 | * taking into account wraparound issues. | |
739 | */ | |
740 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
741 | { | |
2c64c57d | 742 | return (s32)(s1 - s2) > 0; |
5a65503f TM |
743 | } |
744 | ||
fd6002e9 | 745 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
746 | void |
747 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
748 | bool update_barrier) | |
b1f69b75 | 749 | { |
22aaf714 TM |
750 | u32 oldseq, newseq, new_barrier; |
751 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 752 | |
2d2f24ad TM |
753 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
754 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 755 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 756 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 757 | if (update_barrier) { |
22aaf714 | 758 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
759 | } else { |
760 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 761 | * "close" to the current seqids. |
43f1b3da | 762 | */ |
22aaf714 | 763 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 764 | } |
22aaf714 TM |
765 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
766 | lo->plh_barrier = new_barrier; | |
43f1b3da | 767 | } |
b1f69b75 AA |
768 | } |
769 | ||
cf7d63f1 | 770 | static bool |
19c54aba TM |
771 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
772 | const nfs4_stateid *stateid) | |
43f1b3da | 773 | { |
19c54aba | 774 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 775 | |
19c54aba TM |
776 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
777 | } | |
778 | ||
ce6ab4f2 PT |
779 | static bool |
780 | pnfs_layout_returning(const struct pnfs_layout_hdr *lo, | |
781 | struct pnfs_layout_range *range) | |
782 | { | |
783 | return test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) && | |
784 | (lo->plh_return_iomode == IOMODE_ANY || | |
785 | lo->plh_return_iomode == range->iomode); | |
786 | } | |
787 | ||
19c54aba TM |
788 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
789 | static bool | |
ce6ab4f2 PT |
790 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, |
791 | struct pnfs_layout_range *range, int lget) | |
19c54aba | 792 | { |
f7e8917a FI |
793 | return lo->plh_block_lgets || |
794 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || | |
43f1b3da | 795 | (list_empty(&lo->plh_segs) && |
ce6ab4f2 PT |
796 | (atomic_read(&lo->plh_outstanding) > lget)) || |
797 | pnfs_layout_returning(lo, range); | |
cf7d63f1 FI |
798 | } |
799 | ||
fd6002e9 FI |
800 | int |
801 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
ce6ab4f2 | 802 | struct pnfs_layout_range *range, |
fd6002e9 | 803 | struct nfs4_state *open_state) |
b1f69b75 | 804 | { |
fd6002e9 | 805 | int status = 0; |
974cec8c | 806 | |
b1f69b75 | 807 | dprintk("--> %s\n", __func__); |
fd6002e9 | 808 | spin_lock(&lo->plh_inode->i_lock); |
ce6ab4f2 | 809 | if (pnfs_layoutgets_blocked(lo, range, 1)) { |
cf7d63f1 | 810 | status = -EAGAIN; |
5d422301 TM |
811 | } else if (!nfs4_valid_open_stateid(open_state)) { |
812 | status = -EBADF; | |
47abadef CH |
813 | } else if (list_empty(&lo->plh_segs) || |
814 | test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { | |
fd6002e9 FI |
815 | int seq; |
816 | ||
817 | do { | |
818 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 819 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
820 | } while (read_seqretry(&open_state->seqlock, seq)); |
821 | } else | |
f597c537 | 822 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 823 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 824 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 825 | return status; |
b1f69b75 AA |
826 | } |
827 | ||
828 | /* | |
829 | * Get layout from server. | |
830 | * for now, assume that whole file layouts are requested. | |
831 | * arg->offset: 0 | |
832 | * arg->length: all ones | |
833 | */ | |
e5e94017 BH |
834 | static struct pnfs_layout_segment * |
835 | send_layoutget(struct pnfs_layout_hdr *lo, | |
836 | struct nfs_open_context *ctx, | |
fb3296eb | 837 | struct pnfs_layout_range *range, |
a75b9df9 | 838 | gfp_t gfp_flags) |
e5e94017 | 839 | { |
b7edfaa1 | 840 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
841 | struct nfs_server *server = NFS_SERVER(ino); |
842 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 843 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
844 | |
845 | dprintk("--> %s\n", __func__); | |
e5e94017 | 846 | |
a75b9df9 | 847 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 848 | if (lgp == NULL) |
b1f69b75 | 849 | return NULL; |
35124a09 | 850 | |
fb3296eb BH |
851 | lgp->args.minlength = PAGE_CACHE_SIZE; |
852 | if (lgp->args.minlength > range->length) | |
853 | lgp->args.minlength = range->length; | |
b1f69b75 | 854 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 855 | lgp->args.range = *range; |
b1f69b75 AA |
856 | lgp->args.type = server->pnfs_curr_ld->id; |
857 | lgp->args.inode = ino; | |
858 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 859 | lgp->gfp_flags = gfp_flags; |
6ab59344 | 860 | lgp->cred = lo->plh_lc_cred; |
b1f69b75 AA |
861 | |
862 | /* Synchronously retrieve layout information from server and | |
863 | * store in lseg. | |
864 | */ | |
a0b0a6e3 TM |
865 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
866 | if (IS_ERR(lseg)) { | |
867 | switch (PTR_ERR(lseg)) { | |
868 | case -ENOMEM: | |
869 | case -ERESTARTSYS: | |
870 | break; | |
871 | default: | |
872 | /* remember that LAYOUTGET failed and suspend trying */ | |
b9e028fd | 873 | pnfs_layout_io_set_failed(lo, range->iomode); |
a0b0a6e3 TM |
874 | } |
875 | return NULL; | |
974cec8c | 876 | } |
35124a09 | 877 | |
974cec8c AA |
878 | return lseg; |
879 | } | |
880 | ||
24956804 TM |
881 | static void pnfs_clear_layoutcommit(struct inode *inode, |
882 | struct list_head *head) | |
883 | { | |
884 | struct nfs_inode *nfsi = NFS_I(inode); | |
885 | struct pnfs_layout_segment *lseg, *tmp; | |
886 | ||
887 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
888 | return; | |
889 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
890 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
891 | continue; | |
892 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
893 | } | |
894 | } | |
895 | ||
f40eb5d0 PT |
896 | static int |
897 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid, | |
898 | enum pnfs_iomode iomode) | |
899 | { | |
900 | struct inode *ino = lo->plh_inode; | |
901 | struct nfs4_layoutreturn *lrp; | |
902 | int status = 0; | |
903 | ||
904 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); | |
905 | if (unlikely(lrp == NULL)) { | |
906 | status = -ENOMEM; | |
907 | spin_lock(&ino->i_lock); | |
908 | lo->plh_block_lgets--; | |
909 | spin_unlock(&ino->i_lock); | |
910 | pnfs_put_layout_hdr(lo); | |
911 | goto out; | |
912 | } | |
913 | ||
914 | lrp->args.stateid = stateid; | |
915 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
916 | lrp->args.inode = ino; | |
917 | lrp->args.iomode = iomode; | |
918 | lrp->args.layout = lo; | |
919 | lrp->clp = NFS_SERVER(ino)->nfs_client; | |
920 | lrp->cred = lo->plh_lc_cred; | |
921 | ||
922 | status = nfs4_proc_layoutreturn(lrp); | |
923 | out: | |
e736a5b9 PT |
924 | if (status) { |
925 | spin_lock(&ino->i_lock); | |
926 | clear_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
927 | spin_unlock(&ino->i_lock); | |
928 | } | |
f40eb5d0 PT |
929 | dprintk("<-- %s status: %d\n", __func__, status); |
930 | return status; | |
931 | } | |
932 | ||
293b3b06 AA |
933 | /* |
934 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
935 | * when the layout segment list is empty. | |
936 | * | |
937 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
938 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
939 | * deviceid is marked invalid. | |
940 | */ | |
cbe82603 BH |
941 | int |
942 | _pnfs_return_layout(struct inode *ino) | |
943 | { | |
944 | struct pnfs_layout_hdr *lo = NULL; | |
945 | struct nfs_inode *nfsi = NFS_I(ino); | |
946 | LIST_HEAD(tmp_list); | |
cbe82603 | 947 | nfs4_stateid stateid; |
293b3b06 | 948 | int status = 0, empty; |
cbe82603 | 949 | |
366d5052 | 950 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
951 | |
952 | spin_lock(&ino->i_lock); | |
953 | lo = nfsi->layout; | |
e5929f3c | 954 | if (!lo) { |
cbe82603 | 955 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
956 | dprintk("NFS: %s no layout to return\n", __func__); |
957 | goto out; | |
cbe82603 BH |
958 | } |
959 | stateid = nfsi->layout->plh_stateid; | |
960 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 961 | pnfs_get_layout_hdr(lo); |
293b3b06 | 962 | empty = list_empty(&lo->plh_segs); |
24956804 | 963 | pnfs_clear_layoutcommit(ino, &tmp_list); |
49a85061 | 964 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
c88953d8 CH |
965 | |
966 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
967 | struct pnfs_layout_range range = { | |
968 | .iomode = IOMODE_ANY, | |
969 | .offset = 0, | |
970 | .length = NFS4_MAX_UINT64, | |
971 | }; | |
972 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
973 | } | |
974 | ||
293b3b06 AA |
975 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
976 | if (empty) { | |
977 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 978 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
979 | dprintk("NFS: %s no layout segments to return\n", __func__); |
980 | goto out; | |
981 | } | |
47abadef CH |
982 | |
983 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
ea0ded74 | 984 | lo->plh_block_lgets++; |
cbe82603 BH |
985 | spin_unlock(&ino->i_lock); |
986 | pnfs_free_lseg_list(&tmp_list); | |
987 | ||
f40eb5d0 | 988 | status = pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY); |
cbe82603 BH |
989 | out: |
990 | dprintk("<-- %s status: %d\n", __func__, status); | |
991 | return status; | |
992 | } | |
0a57cdac | 993 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 994 | |
24028672 TM |
995 | int |
996 | pnfs_commit_and_return_layout(struct inode *inode) | |
997 | { | |
998 | struct pnfs_layout_hdr *lo; | |
999 | int ret; | |
1000 | ||
1001 | spin_lock(&inode->i_lock); | |
1002 | lo = NFS_I(inode)->layout; | |
1003 | if (lo == NULL) { | |
1004 | spin_unlock(&inode->i_lock); | |
1005 | return 0; | |
1006 | } | |
1007 | pnfs_get_layout_hdr(lo); | |
1008 | /* Block new layoutgets and read/write to ds */ | |
1009 | lo->plh_block_lgets++; | |
1010 | spin_unlock(&inode->i_lock); | |
1011 | filemap_fdatawait(inode->i_mapping); | |
1012 | ret = pnfs_layoutcommit_inode(inode, true); | |
1013 | if (ret == 0) | |
1014 | ret = _pnfs_return_layout(inode); | |
1015 | spin_lock(&inode->i_lock); | |
1016 | lo->plh_block_lgets--; | |
1017 | spin_unlock(&inode->i_lock); | |
1018 | pnfs_put_layout_hdr(lo); | |
1019 | return ret; | |
1020 | } | |
1021 | ||
f7e8917a FI |
1022 | bool pnfs_roc(struct inode *ino) |
1023 | { | |
1024 | struct pnfs_layout_hdr *lo; | |
1025 | struct pnfs_layout_segment *lseg, *tmp; | |
1026 | LIST_HEAD(tmp_list); | |
1027 | bool found = false; | |
1028 | ||
1029 | spin_lock(&ino->i_lock); | |
1030 | lo = NFS_I(ino)->layout; | |
1031 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
1032 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
1033 | goto out_nolayout; | |
1034 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
1035 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
1036 | mark_lseg_invalid(lseg, &tmp_list); | |
1037 | found = true; | |
1038 | } | |
1039 | if (!found) | |
1040 | goto out_nolayout; | |
1041 | lo->plh_block_lgets++; | |
70c3bd2b | 1042 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
1043 | spin_unlock(&ino->i_lock); |
1044 | pnfs_free_lseg_list(&tmp_list); | |
1045 | return true; | |
1046 | ||
1047 | out_nolayout: | |
1048 | spin_unlock(&ino->i_lock); | |
1049 | return false; | |
1050 | } | |
1051 | ||
1052 | void pnfs_roc_release(struct inode *ino) | |
1053 | { | |
1054 | struct pnfs_layout_hdr *lo; | |
1055 | ||
1056 | spin_lock(&ino->i_lock); | |
1057 | lo = NFS_I(ino)->layout; | |
1058 | lo->plh_block_lgets--; | |
6622c3ea TM |
1059 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
1060 | pnfs_detach_layout_hdr(lo); | |
1061 | spin_unlock(&ino->i_lock); | |
1062 | pnfs_free_layout_hdr(lo); | |
1063 | } else | |
1064 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
1065 | } |
1066 | ||
1067 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
1068 | { | |
1069 | struct pnfs_layout_hdr *lo; | |
1070 | ||
1071 | spin_lock(&ino->i_lock); | |
1072 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 1073 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
1074 | lo->plh_barrier = barrier; |
1075 | spin_unlock(&ino->i_lock); | |
1076 | } | |
1077 | ||
7fdab069 | 1078 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) |
f7e8917a FI |
1079 | { |
1080 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 1081 | struct pnfs_layout_hdr *lo; |
f7e8917a | 1082 | struct pnfs_layout_segment *lseg; |
7fdab069 | 1083 | u32 current_seqid; |
f7e8917a FI |
1084 | bool found = false; |
1085 | ||
1086 | spin_lock(&ino->i_lock); | |
1087 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
1088 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
7fdab069 | 1089 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); |
f7e8917a | 1090 | found = true; |
7fdab069 | 1091 | goto out; |
f7e8917a | 1092 | } |
7fdab069 TM |
1093 | lo = nfsi->layout; |
1094 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1095 | |
7fdab069 TM |
1096 | /* Since close does not return a layout stateid for use as |
1097 | * a barrier, we choose the worst-case barrier. | |
1098 | */ | |
1099 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
1100 | out: | |
f7e8917a FI |
1101 | spin_unlock(&ino->i_lock); |
1102 | return found; | |
1103 | } | |
1104 | ||
b1f69b75 AA |
1105 | /* |
1106 | * Compare two layout segments for sorting into layout cache. | |
1107 | * We want to preferentially return RW over RO layouts, so ensure those | |
1108 | * are seen first. | |
1109 | */ | |
1110 | static s64 | |
7dc0ac70 | 1111 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1112 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1113 | { |
fb3296eb BH |
1114 | s64 d; |
1115 | ||
1116 | /* high offset > low offset */ | |
1117 | d = l1->offset - l2->offset; | |
1118 | if (d) | |
1119 | return d; | |
1120 | ||
1121 | /* short length > long length */ | |
1122 | d = l2->length - l1->length; | |
1123 | if (d) | |
1124 | return d; | |
1125 | ||
b1f69b75 | 1126 | /* read > read/write */ |
fb3296eb | 1127 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1128 | } |
1129 | ||
974cec8c | 1130 | static void |
57036a37 | 1131 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, |
974cec8c AA |
1132 | struct pnfs_layout_segment *lseg) |
1133 | { | |
b1f69b75 | 1134 | struct pnfs_layout_segment *lp; |
b1f69b75 | 1135 | |
974cec8c AA |
1136 | dprintk("%s:Begin\n", __func__); |
1137 | ||
b7edfaa1 | 1138 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
7dc0ac70 | 1139 | if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 1140 | continue; |
566052c5 | 1141 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1142 | dprintk("%s: inserted lseg %p " |
1143 | "iomode %d offset %llu length %llu before " | |
1144 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1145 | __func__, lseg, lseg->pls_range.iomode, |
1146 | lseg->pls_range.offset, lseg->pls_range.length, | |
1147 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1148 | lp->pls_range.length); | |
fb3296eb | 1149 | goto out; |
974cec8c | 1150 | } |
fb3296eb BH |
1151 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1152 | dprintk("%s: inserted lseg %p " | |
1153 | "iomode %d offset %llu length %llu at tail\n", | |
1154 | __func__, lseg, lseg->pls_range.iomode, | |
1155 | lseg->pls_range.offset, lseg->pls_range.length); | |
1156 | out: | |
70c3bd2b | 1157 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1158 | |
1159 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
1160 | } |
1161 | ||
1162 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1163 | alloc_init_layout_hdr(struct inode *ino, |
1164 | struct nfs_open_context *ctx, | |
1165 | gfp_t gfp_flags) | |
e5e94017 BH |
1166 | { |
1167 | struct pnfs_layout_hdr *lo; | |
1168 | ||
636fb9c8 | 1169 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1170 | if (!lo) |
1171 | return NULL; | |
cc6e5340 | 1172 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1173 | INIT_LIST_HEAD(&lo->plh_layouts); |
1174 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1175 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1176 | lo->plh_inode = ino; |
5cc2216d | 1177 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
e5e94017 BH |
1178 | return lo; |
1179 | } | |
1180 | ||
1181 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1182 | pnfs_find_alloc_layout(struct inode *ino, |
1183 | struct nfs_open_context *ctx, | |
1184 | gfp_t gfp_flags) | |
e5e94017 BH |
1185 | { |
1186 | struct nfs_inode *nfsi = NFS_I(ino); | |
1187 | struct pnfs_layout_hdr *new = NULL; | |
1188 | ||
1189 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1190 | ||
251ec410 TM |
1191 | if (nfsi->layout != NULL) |
1192 | goto out_existing; | |
e5e94017 | 1193 | spin_unlock(&ino->i_lock); |
9fa40758 | 1194 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1195 | spin_lock(&ino->i_lock); |
1196 | ||
251ec410 | 1197 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1198 | nfsi->layout = new; |
251ec410 | 1199 | return new; |
7175fe90 YN |
1200 | } else if (new != NULL) |
1201 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1202 | out_existing: |
1203 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1204 | return nfsi->layout; |
1205 | } | |
1206 | ||
b1f69b75 AA |
1207 | /* |
1208 | * iomode matching rules: | |
1209 | * iomode lseg match | |
1210 | * ----- ----- ----- | |
1211 | * ANY READ true | |
1212 | * ANY RW true | |
1213 | * RW READ false | |
1214 | * RW RW true | |
1215 | * READ READ true | |
1216 | * READ RW true | |
1217 | */ | |
3cb2df17 | 1218 | static bool |
7dc0ac70 | 1219 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
3cb2df17 | 1220 | const struct pnfs_layout_range *range) |
b1f69b75 | 1221 | { |
fb3296eb BH |
1222 | struct pnfs_layout_range range1; |
1223 | ||
1224 | if ((range->iomode == IOMODE_RW && | |
1225 | ls_range->iomode != IOMODE_RW) || | |
7dc0ac70 | 1226 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1227 | return 0; |
1228 | ||
1229 | /* range1 covers only the first byte in the range */ | |
1230 | range1 = *range; | |
1231 | range1.length = 1; | |
7dc0ac70 | 1232 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1233 | } |
1234 | ||
1235 | /* | |
1236 | * lookup range in layout | |
1237 | */ | |
e5e94017 | 1238 | static struct pnfs_layout_segment * |
fb3296eb BH |
1239 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
1240 | struct pnfs_layout_range *range) | |
e5e94017 | 1241 | { |
b1f69b75 AA |
1242 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1243 | ||
1244 | dprintk("%s:Begin\n", __func__); | |
1245 | ||
b7edfaa1 | 1246 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1247 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1248 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
7dc0ac70 | 1249 | pnfs_lseg_range_match(&lseg->pls_range, range)) { |
9369a431 | 1250 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1251 | break; |
1252 | } | |
d771e3a4 | 1253 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
1254 | break; |
1255 | } | |
1256 | ||
1257 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1258 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1259 | return ret; |
e5e94017 BH |
1260 | } |
1261 | ||
d23d61c8 AA |
1262 | /* |
1263 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1264 | * to the MDS or over pNFS | |
1265 | * | |
1266 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1267 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1268 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1269 | * WRITE request. | |
1270 | * | |
1271 | * A return of true means use MDS I/O. | |
1272 | * | |
1273 | * From rfc 5661: | |
1274 | * If a file's size is smaller than the file size threshold, data accesses | |
1275 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1276 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1277 | * server. If both file size and I/O size are provided, the client SHOULD | |
1278 | * reach or exceed both thresholds before sending its read or write | |
1279 | * requests to the data server. | |
1280 | */ | |
1281 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1282 | struct inode *ino, int iomode) | |
1283 | { | |
1284 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1285 | struct nfs_inode *nfsi = NFS_I(ino); | |
1286 | loff_t fsize = i_size_read(ino); | |
1287 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1288 | ||
1289 | if (t == NULL) | |
1290 | return ret; | |
1291 | ||
1292 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1293 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1294 | ||
1295 | switch (iomode) { | |
1296 | case IOMODE_READ: | |
1297 | if (t->bm & THRESHOLD_RD) { | |
1298 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1299 | size_set = true; | |
1300 | if (fsize < t->rd_sz) | |
1301 | size = true; | |
1302 | } | |
1303 | if (t->bm & THRESHOLD_RD_IO) { | |
1304 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1305 | nfsi->read_io); | |
1306 | io_set = true; | |
1307 | if (nfsi->read_io < t->rd_io_sz) | |
1308 | io = true; | |
1309 | } | |
1310 | break; | |
1311 | case IOMODE_RW: | |
1312 | if (t->bm & THRESHOLD_WR) { | |
1313 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1314 | size_set = true; | |
1315 | if (fsize < t->wr_sz) | |
1316 | size = true; | |
1317 | } | |
1318 | if (t->bm & THRESHOLD_WR_IO) { | |
1319 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1320 | nfsi->write_io); | |
1321 | io_set = true; | |
1322 | if (nfsi->write_io < t->wr_io_sz) | |
1323 | io = true; | |
1324 | } | |
1325 | break; | |
1326 | } | |
1327 | if (size_set && io_set) { | |
1328 | if (size && io) | |
1329 | ret = true; | |
1330 | } else if (size || io) | |
1331 | ret = true; | |
1332 | ||
1333 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1334 | return ret; | |
1335 | } | |
1336 | ||
e5e94017 BH |
1337 | /* |
1338 | * Layout segment is retreived from the server if not cached. | |
1339 | * The appropriate layout segment is referenced and returned to the caller. | |
1340 | */ | |
7c24d948 | 1341 | struct pnfs_layout_segment * |
e5e94017 BH |
1342 | pnfs_update_layout(struct inode *ino, |
1343 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1344 | loff_t pos, |
1345 | u64 count, | |
a75b9df9 TM |
1346 | enum pnfs_iomode iomode, |
1347 | gfp_t gfp_flags) | |
e5e94017 | 1348 | { |
fb3296eb BH |
1349 | struct pnfs_layout_range arg = { |
1350 | .iomode = iomode, | |
1351 | .offset = pos, | |
1352 | .length = count, | |
1353 | }; | |
707ed5fd | 1354 | unsigned pg_offset; |
6382a441 WAA |
1355 | struct nfs_server *server = NFS_SERVER(ino); |
1356 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1357 | struct pnfs_layout_hdr *lo; |
1358 | struct pnfs_layout_segment *lseg = NULL; | |
9bf87482 | 1359 | bool first; |
e5e94017 BH |
1360 | |
1361 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
f86bbcf8 | 1362 | goto out; |
d23d61c8 AA |
1363 | |
1364 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
f86bbcf8 | 1365 | goto out; |
d23d61c8 | 1366 | |
9bf87482 PT |
1367 | lookup_again: |
1368 | first = false; | |
e5e94017 | 1369 | spin_lock(&ino->i_lock); |
9fa40758 | 1370 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1371 | if (lo == NULL) { |
1372 | spin_unlock(&ino->i_lock); | |
1373 | goto out; | |
1374 | } | |
e5e94017 | 1375 | |
43f1b3da | 1376 | /* Do we even need to bother with this? */ |
a59c30ac | 1377 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1378 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1379 | goto out_unlock; |
1380 | } | |
1381 | ||
1382 | /* if LAYOUTGET already failed once we don't try again */ | |
b9e028fd | 1383 | if (pnfs_layout_io_test_failed(lo, iomode)) |
e5e94017 BH |
1384 | goto out_unlock; |
1385 | ||
9bf87482 PT |
1386 | first = list_empty(&lo->plh_segs); |
1387 | if (first) { | |
1388 | /* The first layoutget for the file. Need to serialize per | |
1389 | * RFC 5661 Errata 3208. | |
1390 | */ | |
1391 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1392 | &lo->plh_flags)) { | |
1393 | spin_unlock(&ino->i_lock); | |
1394 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1395 | TASK_UNINTERRUPTIBLE); | |
1396 | pnfs_put_layout_hdr(lo); | |
1397 | goto lookup_again; | |
1398 | } | |
1399 | } else { | |
1400 | /* Check to see if the layout for the given range | |
1401 | * already exists | |
1402 | */ | |
1403 | lseg = pnfs_find_lseg(lo, &arg); | |
1404 | if (lseg) | |
1405 | goto out_unlock; | |
1406 | } | |
568e8c49 | 1407 | |
ce6ab4f2 | 1408 | if (pnfs_layoutgets_blocked(lo, &arg, 0)) |
cf7d63f1 FI |
1409 | goto out_unlock; |
1410 | atomic_inc(&lo->plh_outstanding); | |
f49f9baa | 1411 | spin_unlock(&ino->i_lock); |
30005121 | 1412 | |
abb9a007 | 1413 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1414 | /* The lo must be on the clp list if there is any |
1415 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1416 | */ | |
1417 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1418 | if (list_empty(&lo->plh_layouts)) |
1419 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1420 | spin_unlock(&clp->cl_lock); |
1421 | } | |
e5e94017 | 1422 | |
707ed5fd BH |
1423 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1424 | if (pg_offset) { | |
1425 | arg.offset -= pg_offset; | |
1426 | arg.length += pg_offset; | |
1427 | } | |
7c24d948 AA |
1428 | if (arg.length != NFS4_MAX_UINT64) |
1429 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1430 | |
fb3296eb | 1431 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
cf7d63f1 | 1432 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1433 | out_put_layout_hdr: |
9bf87482 PT |
1434 | if (first) { |
1435 | unsigned long *bitlock = &lo->plh_flags; | |
1436 | ||
1437 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1438 | smp_mb__after_atomic(); | |
1439 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1440 | } | |
70c3bd2b | 1441 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1442 | out: |
f86bbcf8 TM |
1443 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1444 | "(%s, offset: %llu, length: %llu)\n", | |
1445 | __func__, ino->i_sb->s_id, | |
1446 | (unsigned long long)NFS_FILEID(ino), | |
1447 | lseg == NULL ? "not found" : "found", | |
1448 | iomode==IOMODE_RW ? "read/write" : "read-only", | |
1449 | (unsigned long long)pos, | |
1450 | (unsigned long long)count); | |
e5e94017 BH |
1451 | return lseg; |
1452 | out_unlock: | |
1453 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1454 | goto out_put_layout_hdr; |
e5e94017 | 1455 | } |
7c24d948 | 1456 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1457 | |
a0b0a6e3 | 1458 | struct pnfs_layout_segment * |
b1f69b75 AA |
1459 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1460 | { | |
1461 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1462 | struct nfs4_layoutget_res *res = &lgp->res; | |
1463 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1464 | struct inode *ino = lo->plh_inode; |
78096cca | 1465 | LIST_HEAD(free_me); |
b1f69b75 AA |
1466 | int status = 0; |
1467 | ||
1468 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1469 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1470 | if (!lseg || IS_ERR(lseg)) { |
1471 | if (!lseg) | |
1472 | status = -ENOMEM; | |
1473 | else | |
1474 | status = PTR_ERR(lseg); | |
1475 | dprintk("%s: Could not allocate layout: error %d\n", | |
1476 | __func__, status); | |
1477 | goto out; | |
1478 | } | |
1479 | ||
1013df61 CH |
1480 | init_lseg(lo, lseg); |
1481 | lseg->pls_range = res->range; | |
1482 | ||
b1f69b75 | 1483 | spin_lock(&ino->i_lock); |
a59c30ac | 1484 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1485 | dprintk("%s forget reply due to recall\n", __func__); |
1486 | goto out_forget_reply; | |
1487 | } | |
1488 | ||
ce6ab4f2 | 1489 | if (pnfs_layoutgets_blocked(lo, &lgp->args.range, 1)) { |
43f1b3da FI |
1490 | dprintk("%s forget reply due to state\n", __func__); |
1491 | goto out_forget_reply; | |
1492 | } | |
038d6493 | 1493 | |
362f7474 CH |
1494 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1495 | /* existing state ID, make sure the sequence number matches. */ | |
1496 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1497 | dprintk("%s forget reply due to sequence\n", __func__); | |
1498 | goto out_forget_reply; | |
1499 | } | |
1500 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1501 | } else { | |
1502 | /* | |
1503 | * We got an entirely new state ID. Mark all segments for the | |
1504 | * inode invalid, and don't bother validating the stateid | |
1505 | * sequence number. | |
1506 | */ | |
1507 | pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL); | |
1508 | ||
1509 | nfs4_stateid_copy(&lo->plh_stateid, &res->stateid); | |
1510 | lo->plh_barrier = be32_to_cpu(res->stateid.seqid); | |
1511 | } | |
038d6493 | 1512 | |
47abadef CH |
1513 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); |
1514 | ||
9369a431 | 1515 | pnfs_get_lseg(lseg); |
57036a37 | 1516 | pnfs_layout_insert_lseg(lo, lseg); |
b1f69b75 | 1517 | |
f7e8917a FI |
1518 | if (res->return_on_close) { |
1519 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1520 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1521 | } | |
1522 | ||
b1f69b75 | 1523 | spin_unlock(&ino->i_lock); |
78096cca | 1524 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1525 | return lseg; |
b1f69b75 | 1526 | out: |
a0b0a6e3 | 1527 | return ERR_PTR(status); |
43f1b3da FI |
1528 | |
1529 | out_forget_reply: | |
1530 | spin_unlock(&ino->i_lock); | |
1531 | lseg->pls_layout = lo; | |
1532 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1533 | goto out; | |
b1f69b75 AA |
1534 | } |
1535 | ||
016256df PT |
1536 | static void |
1537 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, | |
1538 | struct list_head *tmp_list, | |
1539 | struct pnfs_layout_range *return_range) | |
1540 | { | |
1541 | struct pnfs_layout_segment *lseg, *next; | |
1542 | ||
1543 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
1544 | ||
1545 | if (list_empty(&lo->plh_segs)) | |
1546 | return; | |
1547 | ||
1548 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
1549 | if (should_free_lseg(&lseg->pls_range, return_range)) { | |
1550 | dprintk("%s: marking lseg %p iomode %d " | |
1551 | "offset %llu length %llu\n", __func__, | |
1552 | lseg, lseg->pls_range.iomode, | |
1553 | lseg->pls_range.offset, | |
1554 | lseg->pls_range.length); | |
1555 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
1556 | mark_lseg_invalid(lseg, tmp_list); | |
1557 | } | |
1558 | } | |
1559 | ||
1560 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
1561 | struct pnfs_layout_segment *lseg) | |
1562 | { | |
1563 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
1564 | int iomode = pnfs_iomode_to_fail_bit(lseg->pls_range.iomode); | |
1565 | struct pnfs_layout_range range = { | |
1566 | .iomode = lseg->pls_range.iomode, | |
1567 | .offset = 0, | |
1568 | .length = NFS4_MAX_UINT64, | |
1569 | }; | |
1570 | LIST_HEAD(free_me); | |
1571 | ||
1572 | spin_lock(&inode->i_lock); | |
1573 | /* set failure bit so that pnfs path will be retried later */ | |
1574 | pnfs_layout_set_fail_bit(lo, iomode); | |
1575 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
1576 | if (lo->plh_return_iomode == 0) | |
1577 | lo->plh_return_iomode = range.iomode; | |
1578 | else if (lo->plh_return_iomode != range.iomode) | |
1579 | lo->plh_return_iomode = IOMODE_ANY; | |
1580 | /* | |
1581 | * mark all matching lsegs so that we are sure to have no live | |
1582 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
1583 | * for how it works. | |
1584 | */ | |
1585 | pnfs_mark_matching_lsegs_return(lo, &free_me, &range); | |
1586 | spin_unlock(&inode->i_lock); | |
1587 | pnfs_free_lseg_list(&free_me); | |
1588 | } | |
1589 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
1590 | ||
d8007d4d TM |
1591 | void |
1592 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1593 | { | |
1fd937bd PT |
1594 | u64 rd_size = req->wb_bytes; |
1595 | ||
bc5a89b3 | 1596 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d | 1597 | |
1fd937bd PT |
1598 | if (pgio->pg_dreq == NULL) |
1599 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1600 | else | |
1601 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1602 | ||
d8007d4d TM |
1603 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1604 | req->wb_context, | |
1605 | req_offset(req), | |
1fd937bd | 1606 | rd_size, |
d8007d4d TM |
1607 | IOMODE_READ, |
1608 | GFP_KERNEL); | |
e885de1a TM |
1609 | /* If no lseg, fall back to read through mds */ |
1610 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1611 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1612 | |
d8007d4d TM |
1613 | } |
1614 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1615 | ||
1616 | void | |
6296556f PT |
1617 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1618 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1619 | { |
bc5a89b3 | 1620 | WARN_ON_ONCE(pgio->pg_lseg != NULL); |
d8007d4d TM |
1621 | |
1622 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1623 | req->wb_context, | |
1624 | req_offset(req), | |
6296556f | 1625 | wb_size, |
d8007d4d TM |
1626 | IOMODE_RW, |
1627 | GFP_NOFS); | |
e885de1a TM |
1628 | /* If no lseg, fall back to write through mds */ |
1629 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1630 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1631 | } |
1632 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1633 | ||
b4fdac1a WAA |
1634 | /* |
1635 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1636 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1637 | */ | |
1638 | size_t | |
dfed206b BH |
1639 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
1640 | struct nfs_page *req) | |
94ad1c80 | 1641 | { |
0f9c429e | 1642 | unsigned int size; |
c5e20cb7 | 1643 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1644 | |
1645 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1646 | if (!size) |
1647 | return 0; | |
94ad1c80 | 1648 | |
19982ba8 | 1649 | /* |
c5e20cb7 WAA |
1650 | * 'size' contains the number of bytes left in the current page (up |
1651 | * to the original size asked for in @req->wb_bytes). | |
1652 | * | |
1653 | * Calculate how many bytes are left in the layout segment | |
1654 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1655 | * |
1656 | * Please also note that 'end_offset' is actually the offset of the | |
1657 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1658 | * | |
1659 | */ | |
19b54848 | 1660 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1661 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1662 | pgio->pg_lseg->pls_range.length); | |
1663 | req_start = req_offset(req); | |
1664 | WARN_ON_ONCE(req_start > seg_end); | |
1665 | /* start of request is past the last byte of this segment */ | |
1666 | if (req_start >= seg_end) | |
19b54848 | 1667 | return 0; |
c5e20cb7 WAA |
1668 | |
1669 | /* adjust 'size' iff there are fewer bytes left in the | |
1670 | * segment than what nfs_generic_pg_test returned */ | |
1671 | seg_left = seg_end - req_start; | |
1672 | if (seg_left < size) | |
1673 | size = (unsigned int)seg_left; | |
19b54848 | 1674 | } |
0f9c429e | 1675 | |
19b54848 | 1676 | return size; |
94ad1c80 | 1677 | } |
89a58e32 | 1678 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1679 | |
53113ad3 | 1680 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
1681 | { |
1682 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
1683 | |
1684 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
1685 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
1686 | hdr->completion_ops); | |
1687 | return nfs_pageio_resend(&pgio, hdr); | |
e2fecb21 | 1688 | } |
e7dd79af | 1689 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1690 | |
d45f60c6 | 1691 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1692 | { |
cd841605 FI |
1693 | |
1694 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1695 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1696 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1697 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1698 | } |
6c75dc0d | 1699 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1700 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
1701 | } |
1702 | ||
d20581aa BH |
1703 | /* |
1704 | * Called by non rpc-based layout drivers | |
1705 | */ | |
d45f60c6 | 1706 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 1707 | { |
d45f60c6 | 1708 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); |
cd841605 | 1709 | if (!hdr->pnfs_error) { |
d45f60c6 WAA |
1710 | pnfs_set_layoutcommit(hdr); |
1711 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
1acbbb4e | 1712 | } else |
d45f60c6 WAA |
1713 | pnfs_ld_handle_write_error(hdr); |
1714 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 1715 | } |
d20581aa | 1716 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1717 | |
dce81290 TM |
1718 | static void |
1719 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1720 | struct nfs_pgio_header *hdr) |
dce81290 | 1721 | { |
6c75dc0d FI |
1722 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1723 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1724 | nfs_pageio_reset_write_mds(desc); | |
1725 | desc->pg_recoalesce = 1; | |
1726 | } | |
d45f60c6 | 1727 | nfs_pgio_data_destroy(hdr); |
dce81290 TM |
1728 | } |
1729 | ||
1730 | static enum pnfs_try_status | |
d45f60c6 | 1731 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
1732 | const struct rpc_call_ops *call_ops, |
1733 | struct pnfs_layout_segment *lseg, | |
1734 | int how) | |
0382b744 | 1735 | { |
cd841605 | 1736 | struct inode *inode = hdr->inode; |
0382b744 AA |
1737 | enum pnfs_try_status trypnfs; |
1738 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1739 | ||
cd841605 | 1740 | hdr->mds_ops = call_ops; |
0382b744 AA |
1741 | |
1742 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
1743 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
1744 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 1745 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1746 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1747 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1748 | return trypnfs; | |
1749 | } | |
1750 | ||
dce81290 | 1751 | static void |
7f714720 WAA |
1752 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
1753 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 1754 | { |
dce81290 TM |
1755 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1756 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1757 | enum pnfs_try_status trypnfs; |
dce81290 TM |
1758 | |
1759 | desc->pg_lseg = NULL; | |
d45f60c6 | 1760 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 1761 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1762 | pnfs_write_through_mds(desc, hdr); |
9369a431 | 1763 | pnfs_put_lseg(lseg); |
dce81290 TM |
1764 | } |
1765 | ||
6c75dc0d FI |
1766 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1767 | { | |
9369a431 | 1768 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1769 | nfs_pgio_header_free(hdr); |
6c75dc0d | 1770 | } |
89d77c8f | 1771 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1772 | |
dce81290 TM |
1773 | int |
1774 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1775 | { | |
6c75dc0d | 1776 | struct nfs_pgio_header *hdr; |
dce81290 TM |
1777 | int ret; |
1778 | ||
1e7f3a48 WAA |
1779 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1780 | if (!hdr) { | |
9b5415b5 | 1781 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
9369a431 | 1782 | pnfs_put_lseg(desc->pg_lseg); |
dce81290 | 1783 | desc->pg_lseg = NULL; |
6c75dc0d | 1784 | return -ENOMEM; |
dce81290 | 1785 | } |
6c75dc0d | 1786 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
9369a431 | 1787 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1788 | ret = nfs_generic_pgio(desc, hdr); |
6c75dc0d | 1789 | if (ret != 0) { |
9369a431 | 1790 | pnfs_put_lseg(desc->pg_lseg); |
6c75dc0d | 1791 | desc->pg_lseg = NULL; |
6c75dc0d | 1792 | } else |
7f714720 | 1793 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
6c75dc0d | 1794 | return ret; |
dce81290 TM |
1795 | } |
1796 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1797 | ||
53113ad3 | 1798 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
1799 | { |
1800 | struct nfs_pageio_descriptor pgio; | |
1801 | ||
1acbbb4e | 1802 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
1803 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
1804 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 1805 | } |
e7dd79af | 1806 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 1807 | |
d45f60c6 | 1808 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 1809 | { |
cd841605 FI |
1810 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
1811 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1812 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 1813 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 1814 | } |
4db6e0b7 | 1815 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 1816 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
1817 | } |
1818 | ||
d20581aa BH |
1819 | /* |
1820 | * Called by non rpc-based layout drivers | |
1821 | */ | |
d45f60c6 | 1822 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 1823 | { |
d45f60c6 | 1824 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
cd841605 | 1825 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
1826 | __nfs4_read_done_cb(hdr); |
1827 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
62e4a769 | 1828 | } else |
d45f60c6 WAA |
1829 | pnfs_ld_handle_read_error(hdr); |
1830 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
1831 | } |
1832 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1833 | ||
493292dd TM |
1834 | static void |
1835 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 1836 | struct nfs_pgio_header *hdr) |
493292dd | 1837 | { |
4db6e0b7 FI |
1838 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1839 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1840 | nfs_pageio_reset_read_mds(desc); | |
1841 | desc->pg_recoalesce = 1; | |
1842 | } | |
d45f60c6 | 1843 | nfs_pgio_data_destroy(hdr); |
493292dd TM |
1844 | } |
1845 | ||
64419a9b AA |
1846 | /* |
1847 | * Call the appropriate parallel I/O subsystem read function. | |
1848 | */ | |
493292dd | 1849 | static enum pnfs_try_status |
d45f60c6 | 1850 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
1851 | const struct rpc_call_ops *call_ops, |
1852 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1853 | { |
cd841605 | 1854 | struct inode *inode = hdr->inode; |
64419a9b AA |
1855 | struct nfs_server *nfss = NFS_SERVER(inode); |
1856 | enum pnfs_try_status trypnfs; | |
1857 | ||
cd841605 | 1858 | hdr->mds_ops = call_ops; |
64419a9b AA |
1859 | |
1860 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 1861 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 1862 | |
d45f60c6 | 1863 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 1864 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1865 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1866 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1867 | return trypnfs; | |
1868 | } | |
863a3c6c | 1869 | |
493292dd | 1870 | static void |
7f714720 | 1871 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 1872 | { |
493292dd TM |
1873 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
1874 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 1875 | enum pnfs_try_status trypnfs; |
493292dd TM |
1876 | |
1877 | desc->pg_lseg = NULL; | |
d45f60c6 | 1878 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
7f714720 | 1879 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 1880 | pnfs_read_through_mds(desc, hdr); |
9369a431 | 1881 | pnfs_put_lseg(lseg); |
493292dd TM |
1882 | } |
1883 | ||
4db6e0b7 FI |
1884 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1885 | { | |
9369a431 | 1886 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 1887 | nfs_pgio_header_free(hdr); |
4db6e0b7 | 1888 | } |
89d77c8f | 1889 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1890 | |
493292dd TM |
1891 | int |
1892 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1893 | { | |
4db6e0b7 | 1894 | struct nfs_pgio_header *hdr; |
493292dd TM |
1895 | int ret; |
1896 | ||
1e7f3a48 WAA |
1897 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
1898 | if (!hdr) { | |
061ae2ed | 1899 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
4db6e0b7 | 1900 | ret = -ENOMEM; |
9369a431 | 1901 | pnfs_put_lseg(desc->pg_lseg); |
493292dd TM |
1902 | desc->pg_lseg = NULL; |
1903 | return ret; | |
1904 | } | |
4db6e0b7 | 1905 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 1906 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 1907 | ret = nfs_generic_pgio(desc, hdr); |
4db6e0b7 | 1908 | if (ret != 0) { |
9369a431 | 1909 | pnfs_put_lseg(desc->pg_lseg); |
4db6e0b7 | 1910 | desc->pg_lseg = NULL; |
4db6e0b7 | 1911 | } else |
7f714720 | 1912 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 1913 | return ret; |
493292dd TM |
1914 | } |
1915 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1916 | ||
71244d9b TM |
1917 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
1918 | { | |
1919 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
1920 | ||
1921 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 1922 | smp_mb__after_atomic(); |
71244d9b TM |
1923 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
1924 | } | |
1925 | ||
863a3c6c | 1926 | /* |
a9bae566 | 1927 | * There can be multiple RW segments. |
863a3c6c | 1928 | */ |
a9bae566 | 1929 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1930 | { |
a9bae566 | 1931 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1932 | |
a9bae566 PT |
1933 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1934 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 1935 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
1936 | list_add(&lseg->pls_lc_list, listp); |
1937 | } | |
863a3c6c AA |
1938 | } |
1939 | ||
a073dbff TM |
1940 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
1941 | { | |
1942 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
1943 | |
1944 | /* Matched by references in pnfs_set_layoutcommit */ | |
1945 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
1946 | list_del_init(&lseg->pls_lc_list); | |
1947 | pnfs_put_lseg(lseg); | |
1948 | } | |
1949 | ||
71244d9b | 1950 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
1951 | } |
1952 | ||
1b0ae068 PT |
1953 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
1954 | { | |
b9e028fd | 1955 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
1956 | } |
1957 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
1958 | ||
863a3c6c | 1959 | void |
d45f60c6 | 1960 | pnfs_set_layoutcommit(struct nfs_pgio_header *hdr) |
863a3c6c | 1961 | { |
cd841605 FI |
1962 | struct inode *inode = hdr->inode; |
1963 | struct nfs_inode *nfsi = NFS_I(inode); | |
d45f60c6 | 1964 | loff_t end_pos = hdr->mds_offset + hdr->res.count; |
79a48a1f | 1965 | bool mark_as_dirty = false; |
863a3c6c | 1966 | |
cd841605 | 1967 | spin_lock(&inode->i_lock); |
863a3c6c | 1968 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 1969 | mark_as_dirty = true; |
863a3c6c | 1970 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 1971 | __func__, inode->i_ino); |
863a3c6c | 1972 | } |
cd841605 | 1973 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 1974 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 1975 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 1976 | } |
acff5880 PT |
1977 | if (end_pos > nfsi->layout->plh_lwb) |
1978 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 1979 | spin_unlock(&inode->i_lock); |
acff5880 | 1980 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 1981 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
1982 | |
1983 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1984 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1985 | if (mark_as_dirty) | |
cd841605 | 1986 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
1987 | } |
1988 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
1989 | ||
378520b8 PT |
1990 | void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data) |
1991 | { | |
1992 | struct inode *inode = data->inode; | |
1993 | struct nfs_inode *nfsi = NFS_I(inode); | |
1994 | bool mark_as_dirty = false; | |
1995 | ||
1996 | spin_lock(&inode->i_lock); | |
1997 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { | |
1998 | mark_as_dirty = true; | |
1999 | dprintk("%s: Set layoutcommit for inode %lu ", | |
2000 | __func__, inode->i_ino); | |
2001 | } | |
2002 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) { | |
2003 | /* references matched in nfs4_layoutcommit_release */ | |
2004 | pnfs_get_lseg(data->lseg); | |
2005 | } | |
2006 | if (data->lwb > nfsi->layout->plh_lwb) | |
2007 | nfsi->layout->plh_lwb = data->lwb; | |
2008 | spin_unlock(&inode->i_lock); | |
2009 | dprintk("%s: lseg %p end_pos %llu\n", | |
2010 | __func__, data->lseg, nfsi->layout->plh_lwb); | |
2011 | ||
2012 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2013 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2014 | if (mark_as_dirty) | |
2015 | mark_inode_dirty_sync(inode); | |
2016 | } | |
2017 | EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit); | |
2018 | ||
db29c089 AA |
2019 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2020 | { | |
2021 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2022 | ||
2023 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2024 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2025 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2026 | } |
2027 | ||
de4b15c7 AA |
2028 | /* |
2029 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2030 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2031 | * data to disk to allow the server to recover the data if it crashes. | |
2032 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2033 | * is off, and a COMMIT is sent to a data server, or | |
2034 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2035 | */ | |
863a3c6c | 2036 | int |
ef311537 | 2037 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2038 | { |
5f919c9f | 2039 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2040 | struct nfs4_layoutcommit_data *data; |
2041 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2042 | loff_t end_pos; |
71244d9b | 2043 | int status; |
863a3c6c | 2044 | |
71244d9b | 2045 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2046 | return 0; |
2047 | ||
71244d9b | 2048 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2049 | |
71244d9b | 2050 | status = -EAGAIN; |
92407e75 | 2051 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2052 | if (!sync) |
2053 | goto out; | |
74316201 | 2054 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2055 | NFS_INO_LAYOUTCOMMITTING, |
2056 | nfs_wait_bit_killable, | |
2057 | TASK_KILLABLE); | |
92407e75 | 2058 | if (status) |
71244d9b | 2059 | goto out; |
92407e75 PT |
2060 | } |
2061 | ||
71244d9b TM |
2062 | status = -ENOMEM; |
2063 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2064 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2065 | if (!data) | |
2066 | goto clear_layoutcommitting; | |
2067 | ||
2068 | status = 0; | |
de4b15c7 | 2069 | spin_lock(&inode->i_lock); |
71244d9b TM |
2070 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2071 | goto out_unlock; | |
a9bae566 | 2072 | |
71244d9b | 2073 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2074 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2075 | |
acff5880 | 2076 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 2077 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 2078 | |
f597c537 | 2079 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2080 | spin_unlock(&inode->i_lock); |
2081 | ||
2082 | data->args.inode = inode; | |
9fa40758 | 2083 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2084 | nfs_fattr_init(&data->fattr); |
2085 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2086 | data->res.fattr = &data->fattr; | |
2087 | data->args.lastbytewritten = end_pos - 1; | |
2088 | data->res.server = NFS_SERVER(inode); | |
2089 | ||
5f919c9f CH |
2090 | if (ld->prepare_layoutcommit) { |
2091 | status = ld->prepare_layoutcommit(&data->args); | |
2092 | if (status) { | |
2093 | spin_lock(&inode->i_lock); | |
2094 | if (end_pos < nfsi->layout->plh_lwb) | |
2095 | nfsi->layout->plh_lwb = end_pos; | |
2096 | spin_unlock(&inode->i_lock); | |
2097 | put_rpccred(data->cred); | |
2098 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); | |
2099 | goto clear_layoutcommitting; | |
2100 | } | |
2101 | } | |
2102 | ||
2103 | ||
863a3c6c AA |
2104 | status = nfs4_proc_layoutcommit(data, sync); |
2105 | out: | |
92407e75 PT |
2106 | if (status) |
2107 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2108 | dprintk("<-- %s status %d\n", __func__, status); |
2109 | return status; | |
71244d9b TM |
2110 | out_unlock: |
2111 | spin_unlock(&inode->i_lock); | |
92407e75 | 2112 | kfree(data); |
71244d9b TM |
2113 | clear_layoutcommitting: |
2114 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2115 | goto out; |
863a3c6c | 2116 | } |
72cff449 | 2117 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a AA |
2118 | |
2119 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
2120 | { | |
2121 | struct nfs4_threshold *thp; | |
2122 | ||
2123 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2124 | if (!thp) { | |
2125 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2126 | return NULL; | |
2127 | } | |
2128 | return thp; | |
2129 | } |