Commit | Line | Data |
---|---|---|
85e174ba RL |
1 | /* |
2 | * pNFS functions to call and manage layout drivers. | |
3 | * | |
4 | * Copyright (c) 2002 [year of first publication] | |
5 | * The Regents of the University of Michigan | |
6 | * All Rights Reserved | |
7 | * | |
8 | * Dean Hildebrand <dhildebz@umich.edu> | |
9 | * | |
10 | * Permission is granted to use, copy, create derivative works, and | |
11 | * redistribute this software and such derivative works for any purpose, | |
12 | * so long as the name of the University of Michigan is not used in | |
13 | * any advertising or publicity pertaining to the use or distribution | |
14 | * of this software without specific, written prior authorization. If | |
15 | * the above copyright notice or any other identification of the | |
16 | * University of Michigan is included in any copy of any portion of | |
17 | * this software, then the disclaimer below must also be included. | |
18 | * | |
19 | * This software is provided as is, without representation or warranty | |
20 | * of any kind either express or implied, including without limitation | |
21 | * the implied warranties of merchantability, fitness for a particular | |
22 | * purpose, or noninfringement. The Regents of the University of | |
23 | * Michigan shall not be liable for any damages, including special, | |
24 | * indirect, incidental, or consequential damages, with respect to any | |
25 | * claim arising out of or in connection with the use of the software, | |
26 | * even if it has been or is hereafter advised of the possibility of | |
27 | * such damages. | |
28 | */ | |
29 | ||
30 | #include <linux/nfs_fs.h> | |
493292dd | 31 | #include <linux/nfs_page.h> |
143cb494 | 32 | #include <linux/module.h> |
974cec8c | 33 | #include "internal.h" |
85e174ba | 34 | #include "pnfs.h" |
64419a9b | 35 | #include "iostat.h" |
cc668ab3 | 36 | #include "nfs4trace.h" |
40dd4b7a | 37 | #include "delegation.h" |
8733408d | 38 | #include "nfs42.h" |
85e174ba RL |
39 | |
40 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
25c75333 | 41 | #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) |
85e174ba | 42 | |
02c35fca FI |
43 | /* Locking: |
44 | * | |
45 | * pnfs_spinlock: | |
46 | * protects pnfs_modules_tbl. | |
47 | */ | |
48 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
49 | ||
50 | /* | |
51 | * pnfs_modules_tbl holds all pnfs modules | |
52 | */ | |
53 | static LIST_HEAD(pnfs_modules_tbl); | |
54 | ||
13c13a6a | 55 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo); |
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 | ||
13c13a6a TM |
244 | pnfs_layoutreturn_before_put_layout_hdr(lo); |
245 | ||
cc6e5340 | 246 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { |
566f8737 PT |
247 | if (!list_empty(&lo->plh_segs)) |
248 | WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n"); | |
6622c3ea | 249 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 250 | spin_unlock(&inode->i_lock); |
6622c3ea | 251 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 252 | } |
974cec8c AA |
253 | } |
254 | ||
2454dfea TM |
255 | /* |
256 | * Mark a pnfs_layout_hdr and all associated layout segments as invalid | |
257 | * | |
258 | * In order to continue using the pnfs_layout_hdr, a full recovery | |
259 | * is required. | |
260 | * Note that caller must hold inode->i_lock. | |
261 | */ | |
5f46be04 | 262 | int |
2454dfea TM |
263 | pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo, |
264 | struct list_head *lseg_list) | |
265 | { | |
266 | struct pnfs_layout_range range = { | |
267 | .iomode = IOMODE_ANY, | |
268 | .offset = 0, | |
269 | .length = NFS4_MAX_UINT64, | |
270 | }; | |
271 | ||
272 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
6d597e17 | 273 | return pnfs_mark_matching_lsegs_invalid(lo, lseg_list, &range, 0); |
2454dfea TM |
274 | } |
275 | ||
b9e028fd TM |
276 | static int |
277 | pnfs_iomode_to_fail_bit(u32 iomode) | |
278 | { | |
279 | return iomode == IOMODE_RW ? | |
280 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
281 | } | |
282 | ||
283 | static void | |
3e621214 | 284 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 285 | { |
25c75333 | 286 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 287 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
3e621214 TM |
288 | atomic_inc(&lo->plh_refcount); |
289 | } | |
290 | ||
291 | static void | |
292 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
293 | { | |
294 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
295 | atomic_dec(&lo->plh_refcount); | |
296 | } | |
297 | ||
298 | static void | |
299 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
300 | { | |
301 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
302 | struct pnfs_layout_range range = { |
303 | .iomode = iomode, | |
304 | .offset = 0, | |
305 | .length = NFS4_MAX_UINT64, | |
306 | }; | |
307 | LIST_HEAD(head); | |
3e621214 TM |
308 | |
309 | spin_lock(&inode->i_lock); | |
310 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
6d597e17 | 311 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0); |
3e621214 | 312 | spin_unlock(&inode->i_lock); |
115ce575 | 313 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
314 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
315 | iomode == IOMODE_RW ? "RW" : "READ"); | |
316 | } | |
317 | ||
318 | static bool | |
319 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
320 | { | |
25c75333 | 321 | unsigned long start, end; |
3e621214 TM |
322 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
323 | ||
324 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
325 | return false; |
326 | end = jiffies; | |
327 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
328 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
329 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 330 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
331 | return false; |
332 | } | |
333 | return true; | |
b9e028fd TM |
334 | } |
335 | ||
974cec8c | 336 | static void |
119cef97 TM |
337 | pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg, |
338 | const struct pnfs_layout_range *range, | |
339 | const nfs4_stateid *stateid) | |
974cec8c | 340 | { |
566052c5 | 341 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 342 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c | 343 | atomic_set(&lseg->pls_refcount, 1); |
4541d16c | 344 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); |
566052c5 | 345 | lseg->pls_layout = lo; |
119cef97 TM |
346 | lseg->pls_range = *range; |
347 | lseg->pls_seq = be32_to_cpu(stateid->seqid); | |
974cec8c AA |
348 | } |
349 | ||
905ca191 | 350 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 351 | { |
b7edfaa1 | 352 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 353 | |
b1f69b75 | 354 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
974cec8c AA |
355 | } |
356 | ||
d684d2ae | 357 | static void |
57036a37 TM |
358 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
359 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 360 | { |
57036a37 | 361 | struct inode *inode = lo->plh_inode; |
d684d2ae | 362 | |
d20581aa | 363 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 364 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
365 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
366 | atomic_dec(&lo->plh_refcount); | |
2d148c7e | 367 | if (list_empty(&lo->plh_segs)) { |
334a8f37 TM |
368 | if (atomic_read(&lo->plh_outstanding) == 0) |
369 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
173f77e9 | 370 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
2d148c7e | 371 | } |
d684d2ae FI |
372 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); |
373 | } | |
374 | ||
bae724ef | 375 | void |
9369a431 | 376 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 377 | { |
57036a37 | 378 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
379 | struct inode *inode; |
380 | ||
381 | if (!lseg) | |
382 | return; | |
383 | ||
4541d16c FI |
384 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
385 | atomic_read(&lseg->pls_refcount), | |
386 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
4ef2e4f8 | 387 | |
57036a37 TM |
388 | lo = lseg->pls_layout; |
389 | inode = lo->plh_inode; | |
4ef2e4f8 | 390 | |
d684d2ae | 391 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
faa4a54f TM |
392 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { |
393 | spin_unlock(&inode->i_lock); | |
394 | return; | |
395 | } | |
8f0d27dc | 396 | pnfs_get_layout_hdr(lo); |
4ef2e4f8 TM |
397 | pnfs_layout_remove_lseg(lo, lseg); |
398 | spin_unlock(&inode->i_lock); | |
399 | pnfs_free_lseg(lseg); | |
400 | pnfs_put_layout_hdr(lo); | |
4541d16c | 401 | } |
4541d16c | 402 | } |
9369a431 | 403 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 404 | |
6543f803 | 405 | static void pnfs_free_lseg_async_work(struct work_struct *work) |
e6cf82d1 WAA |
406 | { |
407 | struct pnfs_layout_segment *lseg; | |
6543f803 | 408 | struct pnfs_layout_hdr *lo; |
e6cf82d1 WAA |
409 | |
410 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
6543f803 | 411 | lo = lseg->pls_layout; |
e6cf82d1 | 412 | |
6543f803 TM |
413 | pnfs_free_lseg(lseg); |
414 | pnfs_put_layout_hdr(lo); | |
e6cf82d1 WAA |
415 | } |
416 | ||
6543f803 | 417 | static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg) |
e6cf82d1 | 418 | { |
6543f803 | 419 | INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work); |
e6cf82d1 WAA |
420 | schedule_work(&lseg->pls_work); |
421 | } | |
6543f803 TM |
422 | |
423 | void | |
424 | pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg) | |
425 | { | |
426 | if (!lseg) | |
427 | return; | |
428 | ||
429 | assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock); | |
430 | ||
431 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, | |
432 | atomic_read(&lseg->pls_refcount), | |
433 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
434 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
435 | struct pnfs_layout_hdr *lo = lseg->pls_layout; | |
faa4a54f TM |
436 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) |
437 | return; | |
6543f803 TM |
438 | pnfs_get_layout_hdr(lo); |
439 | pnfs_layout_remove_lseg(lo, lseg); | |
440 | pnfs_free_lseg_async(lseg); | |
441 | } | |
442 | } | |
443 | EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked); | |
e6cf82d1 | 444 | |
3cb2df17 | 445 | static u64 |
fb3296eb BH |
446 | end_offset(u64 start, u64 len) |
447 | { | |
448 | u64 end; | |
449 | ||
450 | end = start + len; | |
451 | return end >= start ? end : NFS4_MAX_UINT64; | |
452 | } | |
453 | ||
fb3296eb BH |
454 | /* |
455 | * is l2 fully contained in l1? | |
456 | * start1 end1 | |
457 | * [----------------------------------) | |
458 | * start2 end2 | |
459 | * [----------------) | |
460 | */ | |
3cb2df17 | 461 | static bool |
7dc0ac70 | 462 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 463 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
464 | { |
465 | u64 start1 = l1->offset; | |
466 | u64 end1 = end_offset(start1, l1->length); | |
467 | u64 start2 = l2->offset; | |
468 | u64 end2 = end_offset(start2, l2->length); | |
469 | ||
470 | return (start1 <= start2) && (end1 >= end2); | |
471 | } | |
472 | ||
473 | /* | |
474 | * is l1 and l2 intersecting? | |
475 | * start1 end1 | |
476 | * [----------------------------------) | |
477 | * start2 end2 | |
478 | * [----------------) | |
479 | */ | |
3cb2df17 | 480 | static bool |
7dc0ac70 | 481 | pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1, |
3cb2df17 | 482 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
483 | { |
484 | u64 start1 = l1->offset; | |
485 | u64 end1 = end_offset(start1, l1->length); | |
486 | u64 start2 = l2->offset; | |
487 | u64 end2 = end_offset(start2, l2->length); | |
488 | ||
489 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
490 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
491 | } | |
492 | ||
24956804 TM |
493 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
494 | struct list_head *tmp_list) | |
495 | { | |
496 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
497 | return false; | |
498 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
499 | list_add(&lseg->pls_list, tmp_list); | |
500 | return true; | |
501 | } | |
502 | ||
4541d16c FI |
503 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
504 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
505 | struct list_head *tmp_list) | |
506 | { | |
507 | int rv = 0; | |
508 | ||
509 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
510 | /* Remove the reference keeping the lseg in the | |
511 | * list. It will now be removed when all | |
512 | * outstanding io is finished. | |
513 | */ | |
d684d2ae FI |
514 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
515 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 516 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 517 | rv = 1; |
4541d16c FI |
518 | } |
519 | return rv; | |
520 | } | |
521 | ||
6d597e17 JL |
522 | /* |
523 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
524 | * taking into account wraparound issues. | |
525 | */ | |
526 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
527 | { | |
528 | return (s32)(s1 - s2) > 0; | |
529 | } | |
530 | ||
e036f464 TM |
531 | static bool |
532 | pnfs_should_free_range(const struct pnfs_layout_range *lseg_range, | |
533 | const struct pnfs_layout_range *recall_range) | |
534 | { | |
535 | return (recall_range->iomode == IOMODE_ANY || | |
536 | lseg_range->iomode == recall_range->iomode) && | |
537 | pnfs_lseg_range_intersecting(lseg_range, recall_range); | |
538 | } | |
539 | ||
540 | static bool | |
541 | pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg, | |
542 | const struct pnfs_layout_range *recall_range, | |
543 | u32 seq) | |
544 | { | |
545 | if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq)) | |
546 | return false; | |
547 | if (recall_range == NULL) | |
548 | return true; | |
549 | return pnfs_should_free_range(&lseg->pls_range, recall_range); | |
550 | } | |
551 | ||
6d597e17 JL |
552 | /** |
553 | * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later | |
554 | * @lo: layout header containing the lsegs | |
555 | * @tmp_list: list head where doomed lsegs should go | |
556 | * @recall_range: optional recall range argument to match (may be NULL) | |
557 | * @seq: only invalidate lsegs obtained prior to this sequence (may be 0) | |
558 | * | |
559 | * Walk the list of lsegs in the layout header, and tear down any that should | |
560 | * be destroyed. If "recall_range" is specified then the segment must match | |
561 | * that range. If "seq" is non-zero, then only match segments that were handed | |
562 | * out at or before that sequence. | |
563 | * | |
564 | * Returns number of matching invalid lsegs remaining in list after scanning | |
565 | * it and purging them. | |
4541d16c | 566 | */ |
43f1b3da | 567 | int |
49a85061 | 568 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 569 | struct list_head *tmp_list, |
6d597e17 JL |
570 | const struct pnfs_layout_range *recall_range, |
571 | u32 seq) | |
974cec8c AA |
572 | { |
573 | struct pnfs_layout_segment *lseg, *next; | |
71b39854 | 574 | int remaining = 0; |
974cec8c AA |
575 | |
576 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
577 | ||
8006bfba | 578 | if (list_empty(&lo->plh_segs)) |
38511722 | 579 | return 0; |
4541d16c | 580 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
e036f464 | 581 | if (pnfs_match_lseg_recall(lseg, recall_range, seq)) { |
6d597e17 | 582 | dprintk("%s: freeing lseg %p iomode %d seq %u" |
4541d16c | 583 | "offset %llu length %llu\n", __func__, |
6d597e17 JL |
584 | lseg, lseg->pls_range.iomode, lseg->pls_seq, |
585 | lseg->pls_range.offset, lseg->pls_range.length); | |
71b39854 TM |
586 | if (!mark_lseg_invalid(lseg, tmp_list)) |
587 | remaining++; | |
4541d16c | 588 | } |
71b39854 TM |
589 | dprintk("%s:Return %i\n", __func__, remaining); |
590 | return remaining; | |
974cec8c AA |
591 | } |
592 | ||
f49f9baa | 593 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 594 | void |
4541d16c | 595 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 596 | { |
4541d16c | 597 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
598 | |
599 | if (list_empty(free_me)) | |
600 | return; | |
601 | ||
4541d16c | 602 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 603 | list_del(&lseg->pls_list); |
905ca191 | 604 | pnfs_free_lseg(lseg); |
974cec8c AA |
605 | } |
606 | } | |
607 | ||
e5e94017 BH |
608 | void |
609 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
610 | { | |
611 | struct pnfs_layout_hdr *lo; | |
974cec8c | 612 | LIST_HEAD(tmp_list); |
e5e94017 BH |
613 | |
614 | spin_lock(&nfsi->vfs_inode.i_lock); | |
615 | lo = nfsi->layout; | |
616 | if (lo) { | |
3e621214 | 617 | pnfs_get_layout_hdr(lo); |
2454dfea | 618 | pnfs_mark_layout_stateid_invalid(lo, &tmp_list); |
3e621214 TM |
619 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); |
620 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
621 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
622 | pnfs_free_lseg_list(&tmp_list); | |
623 | pnfs_put_layout_hdr(lo); | |
624 | } else | |
625 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 626 | } |
041245c8 | 627 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 628 | |
fd9a8d71 TM |
629 | static bool |
630 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
631 | struct list_head *layout_list) | |
974cec8c AA |
632 | { |
633 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 634 | bool ret = false; |
974cec8c | 635 | |
fd9a8d71 TM |
636 | spin_lock(&inode->i_lock); |
637 | lo = NFS_I(inode)->layout; | |
638 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
639 | pnfs_get_layout_hdr(lo); | |
640 | list_add(&lo->plh_bulk_destroy, layout_list); | |
641 | ret = true; | |
642 | } | |
643 | spin_unlock(&inode->i_lock); | |
644 | return ret; | |
645 | } | |
646 | ||
647 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
648 | static int | |
649 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
650 | struct nfs_server *server, | |
651 | struct list_head *layout_list) | |
652 | { | |
653 | struct pnfs_layout_hdr *lo, *next; | |
654 | struct inode *inode; | |
655 | ||
656 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
657 | inode = igrab(lo->plh_inode); | |
658 | if (inode == NULL) | |
659 | continue; | |
660 | list_del_init(&lo->plh_layouts); | |
661 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
662 | continue; | |
663 | rcu_read_unlock(); | |
664 | spin_unlock(&clp->cl_lock); | |
665 | iput(inode); | |
666 | spin_lock(&clp->cl_lock); | |
667 | rcu_read_lock(); | |
668 | return -EAGAIN; | |
669 | } | |
670 | return 0; | |
671 | } | |
672 | ||
673 | static int | |
674 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
675 | bool is_bulk_recall) | |
676 | { | |
677 | struct pnfs_layout_hdr *lo; | |
678 | struct inode *inode; | |
fd9a8d71 TM |
679 | LIST_HEAD(lseg_list); |
680 | int ret = 0; | |
681 | ||
682 | while (!list_empty(layout_list)) { | |
683 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
684 | plh_bulk_destroy); | |
685 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
686 | lo->plh_inode->i_ino); | |
687 | inode = lo->plh_inode; | |
7c5d1875 CH |
688 | |
689 | pnfs_layoutcommit_inode(inode, false); | |
690 | ||
fd9a8d71 TM |
691 | spin_lock(&inode->i_lock); |
692 | list_del_init(&lo->plh_bulk_destroy); | |
9fd4b9fc TM |
693 | if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) { |
694 | if (is_bulk_recall) | |
695 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
fd9a8d71 | 696 | ret = -EAGAIN; |
9fd4b9fc | 697 | } |
fd9a8d71 TM |
698 | spin_unlock(&inode->i_lock); |
699 | pnfs_free_lseg_list(&lseg_list); | |
b20135d0 TM |
700 | /* Free all lsegs that are attached to commit buckets */ |
701 | nfs_commit_inode(inode, 0); | |
fd9a8d71 TM |
702 | pnfs_put_layout_hdr(lo); |
703 | iput(inode); | |
704 | } | |
705 | return ret; | |
706 | } | |
707 | ||
708 | int | |
709 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
710 | struct nfs_fsid *fsid, | |
711 | bool is_recall) | |
712 | { | |
713 | struct nfs_server *server; | |
714 | LIST_HEAD(layout_list); | |
c47abcf8 | 715 | |
974cec8c | 716 | spin_lock(&clp->cl_lock); |
6382a441 | 717 | rcu_read_lock(); |
fd9a8d71 | 718 | restart: |
6382a441 | 719 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
720 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
721 | continue; | |
722 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
723 | server, | |
724 | &layout_list) != 0) | |
725 | goto restart; | |
6382a441 WAA |
726 | } |
727 | rcu_read_unlock(); | |
974cec8c AA |
728 | spin_unlock(&clp->cl_lock); |
729 | ||
fd9a8d71 TM |
730 | if (list_empty(&layout_list)) |
731 | return 0; | |
732 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
733 | } | |
734 | ||
735 | int | |
736 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
737 | bool is_recall) | |
738 | { | |
739 | struct nfs_server *server; | |
740 | LIST_HEAD(layout_list); | |
741 | ||
742 | spin_lock(&clp->cl_lock); | |
743 | rcu_read_lock(); | |
744 | restart: | |
745 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
746 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
747 | server, | |
748 | &layout_list) != 0) | |
749 | goto restart; | |
974cec8c | 750 | } |
fd9a8d71 TM |
751 | rcu_read_unlock(); |
752 | spin_unlock(&clp->cl_lock); | |
753 | ||
754 | if (list_empty(&layout_list)) | |
755 | return 0; | |
756 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
757 | } | |
758 | ||
759 | /* | |
760 | * Called by the state manger to remove all layouts established under an | |
761 | * expired lease. | |
762 | */ | |
763 | void | |
764 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
765 | { | |
766 | nfs4_deviceid_mark_client_invalid(clp); | |
767 | nfs4_deviceid_purge_client(clp); | |
768 | ||
769 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
770 | } |
771 | ||
2a59a041 TM |
772 | static void |
773 | pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo) | |
774 | { | |
775 | lo->plh_return_iomode = 0; | |
776 | lo->plh_return_seq = 0; | |
777 | clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); | |
778 | } | |
779 | ||
fd6002e9 | 780 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
781 | void |
782 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
783 | bool update_barrier) | |
b1f69b75 | 784 | { |
ecebb80b | 785 | u32 oldseq, newseq, new_barrier = 0; |
b1f69b75 | 786 | |
2d2f24ad TM |
787 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
788 | newseq = be32_to_cpu(new->seqid); | |
2a59a041 TM |
789 | |
790 | if (!pnfs_layout_is_valid(lo)) { | |
791 | nfs4_stateid_copy(&lo->plh_stateid, new); | |
792 | lo->plh_barrier = newseq; | |
793 | pnfs_clear_layoutreturn_info(lo); | |
794 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
795 | return; | |
796 | } | |
797 | if (pnfs_seqid_is_newer(newseq, oldseq)) { | |
f597c537 | 798 | nfs4_stateid_copy(&lo->plh_stateid, new); |
ecebb80b TM |
799 | /* |
800 | * Because of wraparound, we want to keep the barrier | |
801 | * "close" to the current seqids. | |
802 | */ | |
803 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); | |
43f1b3da | 804 | } |
ecebb80b TM |
805 | if (update_barrier) |
806 | new_barrier = be32_to_cpu(new->seqid); | |
807 | else if (new_barrier == 0) | |
808 | return; | |
2a59a041 | 809 | if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
ecebb80b | 810 | lo->plh_barrier = new_barrier; |
b1f69b75 AA |
811 | } |
812 | ||
cf7d63f1 | 813 | static bool |
19c54aba TM |
814 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
815 | const nfs4_stateid *stateid) | |
43f1b3da | 816 | { |
19c54aba | 817 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 818 | |
19c54aba TM |
819 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
820 | } | |
821 | ||
822 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
823 | static bool | |
e1c06f80 | 824 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo) |
19c54aba | 825 | { |
f7e8917a | 826 | return lo->plh_block_lgets || |
e1c06f80 | 827 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
cf7d63f1 FI |
828 | } |
829 | ||
b1f69b75 | 830 | /* |
183d9e7b JL |
831 | * Get layout from server. |
832 | * for now, assume that whole file layouts are requested. | |
833 | * arg->offset: 0 | |
834 | * arg->length: all ones | |
835 | */ | |
e5e94017 BH |
836 | static struct pnfs_layout_segment * |
837 | send_layoutget(struct pnfs_layout_hdr *lo, | |
838 | struct nfs_open_context *ctx, | |
183d9e7b | 839 | nfs4_stateid *stateid, |
e144e539 | 840 | const struct pnfs_layout_range *range, |
183d9e7b | 841 | long *timeout, 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; | |
2d89a1d3 | 846 | loff_t i_size; |
b1f69b75 AA |
847 | |
848 | dprintk("--> %s\n", __func__); | |
e5e94017 | 849 | |
4f2e9dce JL |
850 | /* |
851 | * Synchronously retrieve layout information from server and | |
852 | * store in lseg. If we race with a concurrent seqid morphing | |
853 | * op, then re-send the LAYOUTGET. | |
854 | */ | |
83026d80 JL |
855 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
856 | if (lgp == NULL) | |
857 | return ERR_PTR(-ENOMEM); | |
858 | ||
859 | i_size = i_size_read(ino); | |
860 | ||
861 | lgp->args.minlength = PAGE_SIZE; | |
862 | if (lgp->args.minlength > range->length) | |
863 | lgp->args.minlength = range->length; | |
864 | if (range->iomode == IOMODE_READ) { | |
865 | if (range->offset >= i_size) | |
866 | lgp->args.minlength = 0; | |
867 | else if (i_size - range->offset < lgp->args.minlength) | |
868 | lgp->args.minlength = i_size - range->offset; | |
d03ab29d | 869 | } |
83026d80 JL |
870 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
871 | pnfs_copy_range(&lgp->args.range, range); | |
872 | lgp->args.type = server->pnfs_curr_ld->id; | |
873 | lgp->args.inode = ino; | |
874 | lgp->args.ctx = get_nfs_open_context(ctx); | |
183d9e7b | 875 | nfs4_stateid_copy(&lgp->args.stateid, stateid); |
83026d80 JL |
876 | lgp->gfp_flags = gfp_flags; |
877 | lgp->cred = lo->plh_lc_cred; | |
35124a09 | 878 | |
183d9e7b | 879 | return nfs4_proc_layoutget(lgp, timeout, gfp_flags); |
974cec8c AA |
880 | } |
881 | ||
24956804 TM |
882 | static void pnfs_clear_layoutcommit(struct inode *inode, |
883 | struct list_head *head) | |
884 | { | |
885 | struct nfs_inode *nfsi = NFS_I(inode); | |
886 | struct pnfs_layout_segment *lseg, *tmp; | |
887 | ||
888 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
889 | return; | |
890 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
891 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
892 | continue; | |
893 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
894 | } | |
895 | } | |
896 | ||
d67ae825 TH |
897 | void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo) |
898 | { | |
899 | clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
900 | smp_mb__after_atomic(); | |
901 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN); | |
7f27392c | 902 | rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq); |
d67ae825 TH |
903 | } |
904 | ||
13c13a6a | 905 | static bool |
e5fd1904 TM |
906 | pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo, |
907 | nfs4_stateid *stateid, | |
908 | enum pnfs_iomode *iomode) | |
13c13a6a | 909 | { |
bf0291dd TM |
910 | /* Serialise LAYOUTGET/LAYOUTRETURN */ |
911 | if (atomic_read(&lo->plh_outstanding) != 0) | |
912 | return false; | |
13c13a6a TM |
913 | if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) |
914 | return false; | |
13c13a6a | 915 | pnfs_get_layout_hdr(lo); |
e5fd1904 TM |
916 | if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) { |
917 | if (stateid != NULL) { | |
918 | nfs4_stateid_copy(stateid, &lo->plh_stateid); | |
919 | if (lo->plh_return_seq != 0) | |
920 | stateid->seqid = cpu_to_be32(lo->plh_return_seq); | |
921 | } | |
922 | if (iomode != NULL) | |
923 | *iomode = lo->plh_return_iomode; | |
924 | pnfs_clear_layoutreturn_info(lo); | |
925 | return true; | |
926 | } | |
927 | if (stateid != NULL) | |
928 | nfs4_stateid_copy(stateid, &lo->plh_stateid); | |
929 | if (iomode != NULL) | |
930 | *iomode = IOMODE_ANY; | |
13c13a6a TM |
931 | return true; |
932 | } | |
933 | ||
f40eb5d0 | 934 | static int |
ed429d6b | 935 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid, |
6c16605d | 936 | enum pnfs_iomode iomode, bool sync) |
f40eb5d0 PT |
937 | { |
938 | struct inode *ino = lo->plh_inode; | |
939 | struct nfs4_layoutreturn *lrp; | |
940 | int status = 0; | |
941 | ||
e4af440a | 942 | lrp = kzalloc(sizeof(*lrp), GFP_NOFS); |
f40eb5d0 PT |
943 | if (unlikely(lrp == NULL)) { |
944 | status = -ENOMEM; | |
945 | spin_lock(&ino->i_lock); | |
d67ae825 | 946 | pnfs_clear_layoutreturn_waitbit(lo); |
f40eb5d0 PT |
947 | spin_unlock(&ino->i_lock); |
948 | pnfs_put_layout_hdr(lo); | |
949 | goto out; | |
950 | } | |
951 | ||
ed429d6b | 952 | nfs4_stateid_copy(&lrp->args.stateid, stateid); |
f40eb5d0 PT |
953 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; |
954 | lrp->args.inode = ino; | |
15eb67c1 PT |
955 | lrp->args.range.iomode = iomode; |
956 | lrp->args.range.offset = 0; | |
957 | lrp->args.range.length = NFS4_MAX_UINT64; | |
f40eb5d0 PT |
958 | lrp->args.layout = lo; |
959 | lrp->clp = NFS_SERVER(ino)->nfs_client; | |
960 | lrp->cred = lo->plh_lc_cred; | |
961 | ||
6c16605d | 962 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
963 | out: |
964 | dprintk("<-- %s status: %d\n", __func__, status); | |
965 | return status; | |
966 | } | |
967 | ||
13c13a6a TM |
968 | /* Return true if layoutreturn is needed */ |
969 | static bool | |
970 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo) | |
971 | { | |
972 | struct pnfs_layout_segment *s; | |
973 | ||
2370abda | 974 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
975 | return false; |
976 | ||
977 | /* Defer layoutreturn until all lsegs are done */ | |
978 | list_for_each_entry(s, &lo->plh_segs, pls_list) { | |
979 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags)) | |
980 | return false; | |
981 | } | |
982 | ||
983 | return true; | |
984 | } | |
985 | ||
986 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo) | |
987 | { | |
988 | struct inode *inode= lo->plh_inode; | |
989 | ||
2370abda | 990 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
991 | return; |
992 | spin_lock(&inode->i_lock); | |
993 | if (pnfs_layout_need_return(lo)) { | |
994 | nfs4_stateid stateid; | |
995 | enum pnfs_iomode iomode; | |
996 | bool send; | |
997 | ||
e5fd1904 | 998 | send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode); |
13c13a6a TM |
999 | spin_unlock(&inode->i_lock); |
1000 | if (send) { | |
1001 | /* Send an async layoutreturn so we dont deadlock */ | |
1002 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); | |
1003 | } | |
1004 | } else | |
1005 | spin_unlock(&inode->i_lock); | |
1006 | } | |
1007 | ||
293b3b06 AA |
1008 | /* |
1009 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
1010 | * when the layout segment list is empty. | |
1011 | * | |
1012 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
1013 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
1014 | * deviceid is marked invalid. | |
1015 | */ | |
cbe82603 BH |
1016 | int |
1017 | _pnfs_return_layout(struct inode *ino) | |
1018 | { | |
1019 | struct pnfs_layout_hdr *lo = NULL; | |
1020 | struct nfs_inode *nfsi = NFS_I(ino); | |
1021 | LIST_HEAD(tmp_list); | |
cbe82603 | 1022 | nfs4_stateid stateid; |
293b3b06 | 1023 | int status = 0, empty; |
7f27392c | 1024 | bool send; |
cbe82603 | 1025 | |
366d5052 | 1026 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
1027 | |
1028 | spin_lock(&ino->i_lock); | |
1029 | lo = nfsi->layout; | |
e5929f3c | 1030 | if (!lo) { |
cbe82603 | 1031 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
1032 | dprintk("NFS: %s no layout to return\n", __func__); |
1033 | goto out; | |
cbe82603 | 1034 | } |
cbe82603 | 1035 | /* Reference matched in nfs4_layoutreturn_release */ |
70c3bd2b | 1036 | pnfs_get_layout_hdr(lo); |
293b3b06 | 1037 | empty = list_empty(&lo->plh_segs); |
24956804 | 1038 | pnfs_clear_layoutcommit(ino, &tmp_list); |
6d597e17 | 1039 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0); |
c88953d8 CH |
1040 | |
1041 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
1042 | struct pnfs_layout_range range = { | |
1043 | .iomode = IOMODE_ANY, | |
1044 | .offset = 0, | |
1045 | .length = NFS4_MAX_UINT64, | |
1046 | }; | |
1047 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
1048 | } | |
1049 | ||
293b3b06 AA |
1050 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
1051 | if (empty) { | |
1052 | spin_unlock(&ino->i_lock); | |
293b3b06 | 1053 | dprintk("NFS: %s no layout segments to return\n", __func__); |
7f27392c | 1054 | goto out_put_layout_hdr; |
293b3b06 | 1055 | } |
47abadef | 1056 | |
e5fd1904 | 1057 | send = pnfs_prepare_layoutreturn(lo, &stateid, NULL); |
cbe82603 BH |
1058 | spin_unlock(&ino->i_lock); |
1059 | pnfs_free_lseg_list(&tmp_list); | |
7f27392c | 1060 | if (send) |
ed429d6b | 1061 | status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
7f27392c TM |
1062 | out_put_layout_hdr: |
1063 | pnfs_put_layout_hdr(lo); | |
cbe82603 BH |
1064 | out: |
1065 | dprintk("<-- %s status: %d\n", __func__, status); | |
1066 | return status; | |
1067 | } | |
0a57cdac | 1068 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 1069 | |
24028672 TM |
1070 | int |
1071 | pnfs_commit_and_return_layout(struct inode *inode) | |
1072 | { | |
1073 | struct pnfs_layout_hdr *lo; | |
1074 | int ret; | |
1075 | ||
1076 | spin_lock(&inode->i_lock); | |
1077 | lo = NFS_I(inode)->layout; | |
1078 | if (lo == NULL) { | |
1079 | spin_unlock(&inode->i_lock); | |
1080 | return 0; | |
1081 | } | |
1082 | pnfs_get_layout_hdr(lo); | |
1083 | /* Block new layoutgets and read/write to ds */ | |
1084 | lo->plh_block_lgets++; | |
1085 | spin_unlock(&inode->i_lock); | |
1086 | filemap_fdatawait(inode->i_mapping); | |
1087 | ret = pnfs_layoutcommit_inode(inode, true); | |
1088 | if (ret == 0) | |
1089 | ret = _pnfs_return_layout(inode); | |
1090 | spin_lock(&inode->i_lock); | |
1091 | lo->plh_block_lgets--; | |
1092 | spin_unlock(&inode->i_lock); | |
1093 | pnfs_put_layout_hdr(lo); | |
1094 | return ret; | |
1095 | } | |
1096 | ||
f7e8917a FI |
1097 | bool pnfs_roc(struct inode *ino) |
1098 | { | |
40dd4b7a TM |
1099 | struct nfs_inode *nfsi = NFS_I(ino); |
1100 | struct nfs_open_context *ctx; | |
1101 | struct nfs4_state *state; | |
f7e8917a FI |
1102 | struct pnfs_layout_hdr *lo; |
1103 | struct pnfs_layout_segment *lseg, *tmp; | |
193e3aa2 | 1104 | nfs4_stateid stateid; |
f7e8917a | 1105 | LIST_HEAD(tmp_list); |
e755d638 | 1106 | bool found = false, layoutreturn = false, roc = false; |
f7e8917a FI |
1107 | |
1108 | spin_lock(&ino->i_lock); | |
40dd4b7a | 1109 | lo = nfsi->layout; |
3976143b | 1110 | if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) |
40dd4b7a TM |
1111 | goto out_noroc; |
1112 | ||
e755d638 | 1113 | /* no roc if we hold a delegation */ |
40dd4b7a TM |
1114 | if (nfs4_check_delegation(ino, FMODE_READ)) |
1115 | goto out_noroc; | |
1116 | ||
1117 | list_for_each_entry(ctx, &nfsi->open_files, list) { | |
1118 | state = ctx->state; | |
1119 | /* Don't return layout if there is open file state */ | |
1120 | if (state != NULL && state->state != 0) | |
1121 | goto out_noroc; | |
1122 | } | |
1123 | ||
e755d638 | 1124 | /* always send layoutreturn if being marked so */ |
e5fd1904 TM |
1125 | if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
1126 | layoutreturn = pnfs_prepare_layoutreturn(lo, | |
1127 | &stateid, NULL); | |
e755d638 | 1128 | |
f7e8917a | 1129 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) |
e755d638 PT |
1130 | /* If we are sending layoutreturn, invalidate all valid lsegs */ |
1131 | if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
f7e8917a FI |
1132 | mark_lseg_invalid(lseg, &tmp_list); |
1133 | found = true; | |
1134 | } | |
500d701f | 1135 | /* ROC in two conditions: |
e755d638 PT |
1136 | * 1. there are ROC lsegs |
1137 | * 2. we don't send layoutreturn | |
e755d638 | 1138 | */ |
500d701f PT |
1139 | if (found && !layoutreturn) { |
1140 | /* lo ref dropped in pnfs_roc_release() */ | |
1141 | pnfs_get_layout_hdr(lo); | |
e755d638 | 1142 | roc = true; |
500d701f | 1143 | } |
f7e8917a | 1144 | |
40dd4b7a | 1145 | out_noroc: |
f7e8917a | 1146 | spin_unlock(&ino->i_lock); |
e755d638 PT |
1147 | pnfs_free_lseg_list(&tmp_list); |
1148 | pnfs_layoutcommit_inode(ino, true); | |
1149 | if (layoutreturn) | |
ed429d6b | 1150 | pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
e755d638 | 1151 | return roc; |
f7e8917a FI |
1152 | } |
1153 | ||
1154 | void pnfs_roc_release(struct inode *ino) | |
1155 | { | |
1156 | struct pnfs_layout_hdr *lo; | |
1157 | ||
1158 | spin_lock(&ino->i_lock); | |
1159 | lo = NFS_I(ino)->layout; | |
5c4a79fb | 1160 | pnfs_clear_layoutreturn_waitbit(lo); |
6622c3ea TM |
1161 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
1162 | pnfs_detach_layout_hdr(lo); | |
1163 | spin_unlock(&ino->i_lock); | |
1164 | pnfs_free_layout_hdr(lo); | |
1165 | } else | |
1166 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
1167 | } |
1168 | ||
1169 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
1170 | { | |
1171 | struct pnfs_layout_hdr *lo; | |
1172 | ||
1173 | spin_lock(&ino->i_lock); | |
1174 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 1175 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
1176 | lo->plh_barrier = barrier; |
1177 | spin_unlock(&ino->i_lock); | |
6a463beb | 1178 | trace_nfs4_layoutreturn_on_close(ino, 0); |
f7e8917a FI |
1179 | } |
1180 | ||
4ff376fe | 1181 | void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier) |
f7e8917a FI |
1182 | { |
1183 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 1184 | struct pnfs_layout_hdr *lo; |
7fdab069 | 1185 | u32 current_seqid; |
f7e8917a FI |
1186 | |
1187 | spin_lock(&ino->i_lock); | |
7fdab069 TM |
1188 | lo = nfsi->layout; |
1189 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1190 | |
7fdab069 TM |
1191 | /* Since close does not return a layout stateid for use as |
1192 | * a barrier, we choose the worst-case barrier. | |
1193 | */ | |
1194 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
f7e8917a | 1195 | spin_unlock(&ino->i_lock); |
f7e8917a FI |
1196 | } |
1197 | ||
500d701f PT |
1198 | bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task) |
1199 | { | |
1200 | struct nfs_inode *nfsi = NFS_I(ino); | |
1201 | struct pnfs_layout_hdr *lo; | |
1202 | bool sleep = false; | |
1203 | ||
1204 | /* we might not have grabbed lo reference. so need to check under | |
1205 | * i_lock */ | |
1206 | spin_lock(&ino->i_lock); | |
1207 | lo = nfsi->layout; | |
1208 | if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) | |
1209 | sleep = true; | |
1210 | spin_unlock(&ino->i_lock); | |
1211 | ||
1212 | if (sleep) | |
1213 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
1214 | ||
1215 | return sleep; | |
1216 | } | |
1217 | ||
b1f69b75 AA |
1218 | /* |
1219 | * Compare two layout segments for sorting into layout cache. | |
1220 | * We want to preferentially return RW over RO layouts, so ensure those | |
1221 | * are seen first. | |
1222 | */ | |
1223 | static s64 | |
7dc0ac70 | 1224 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1225 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1226 | { |
fb3296eb BH |
1227 | s64 d; |
1228 | ||
1229 | /* high offset > low offset */ | |
1230 | d = l1->offset - l2->offset; | |
1231 | if (d) | |
1232 | return d; | |
1233 | ||
1234 | /* short length > long length */ | |
1235 | d = l2->length - l1->length; | |
1236 | if (d) | |
1237 | return d; | |
1238 | ||
b1f69b75 | 1239 | /* read > read/write */ |
fb3296eb | 1240 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1241 | } |
1242 | ||
03772d2f TM |
1243 | static bool |
1244 | pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1, | |
1245 | const struct pnfs_layout_range *l2) | |
1246 | { | |
1247 | return pnfs_lseg_range_cmp(l1, l2) > 0; | |
1248 | } | |
1249 | ||
1250 | static bool | |
1251 | pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg, | |
1252 | struct pnfs_layout_segment *old) | |
1253 | { | |
1254 | return false; | |
1255 | } | |
1256 | ||
1257 | void | |
1258 | pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1259 | struct pnfs_layout_segment *lseg, | |
1260 | bool (*is_after)(const struct pnfs_layout_range *, | |
1261 | const struct pnfs_layout_range *), | |
1262 | bool (*do_merge)(struct pnfs_layout_segment *, | |
1263 | struct pnfs_layout_segment *), | |
1264 | struct list_head *free_me) | |
974cec8c | 1265 | { |
03772d2f | 1266 | struct pnfs_layout_segment *lp, *tmp; |
b1f69b75 | 1267 | |
974cec8c AA |
1268 | dprintk("%s:Begin\n", __func__); |
1269 | ||
03772d2f TM |
1270 | list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) { |
1271 | if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0) | |
1272 | continue; | |
1273 | if (do_merge(lseg, lp)) { | |
1274 | mark_lseg_invalid(lp, free_me); | |
1275 | continue; | |
1276 | } | |
1277 | if (is_after(&lseg->pls_range, &lp->pls_range)) | |
b1f69b75 | 1278 | continue; |
566052c5 | 1279 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1280 | dprintk("%s: inserted lseg %p " |
1281 | "iomode %d offset %llu length %llu before " | |
1282 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1283 | __func__, lseg, lseg->pls_range.iomode, |
1284 | lseg->pls_range.offset, lseg->pls_range.length, | |
1285 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1286 | lp->pls_range.length); | |
fb3296eb | 1287 | goto out; |
974cec8c | 1288 | } |
fb3296eb BH |
1289 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1290 | dprintk("%s: inserted lseg %p " | |
1291 | "iomode %d offset %llu length %llu at tail\n", | |
1292 | __func__, lseg, lseg->pls_range.iomode, | |
1293 | lseg->pls_range.offset, lseg->pls_range.length); | |
1294 | out: | |
70c3bd2b | 1295 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1296 | |
1297 | dprintk("%s:Return\n", __func__); | |
e5e94017 | 1298 | } |
03772d2f TM |
1299 | EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg); |
1300 | ||
1301 | static void | |
1302 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1303 | struct pnfs_layout_segment *lseg, | |
1304 | struct list_head *free_me) | |
1305 | { | |
1306 | struct inode *inode = lo->plh_inode; | |
1307 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
1308 | ||
1309 | if (ld->add_lseg != NULL) | |
1310 | ld->add_lseg(lo, lseg, free_me); | |
1311 | else | |
1312 | pnfs_generic_layout_insert_lseg(lo, lseg, | |
1313 | pnfs_lseg_range_is_after, | |
1314 | pnfs_lseg_no_merge, | |
1315 | free_me); | |
1316 | } | |
e5e94017 BH |
1317 | |
1318 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1319 | alloc_init_layout_hdr(struct inode *ino, |
1320 | struct nfs_open_context *ctx, | |
1321 | gfp_t gfp_flags) | |
e5e94017 BH |
1322 | { |
1323 | struct pnfs_layout_hdr *lo; | |
1324 | ||
636fb9c8 | 1325 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1326 | if (!lo) |
1327 | return NULL; | |
cc6e5340 | 1328 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1329 | INIT_LIST_HEAD(&lo->plh_layouts); |
1330 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1331 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1332 | lo->plh_inode = ino; |
5cc2216d | 1333 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
67a3b721 | 1334 | lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID; |
e5e94017 BH |
1335 | return lo; |
1336 | } | |
1337 | ||
1338 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1339 | pnfs_find_alloc_layout(struct inode *ino, |
1340 | struct nfs_open_context *ctx, | |
1341 | gfp_t gfp_flags) | |
e5241e43 TM |
1342 | __releases(&ino->i_lock) |
1343 | __acquires(&ino->i_lock) | |
e5e94017 BH |
1344 | { |
1345 | struct nfs_inode *nfsi = NFS_I(ino); | |
1346 | struct pnfs_layout_hdr *new = NULL; | |
1347 | ||
1348 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1349 | ||
251ec410 TM |
1350 | if (nfsi->layout != NULL) |
1351 | goto out_existing; | |
e5e94017 | 1352 | spin_unlock(&ino->i_lock); |
9fa40758 | 1353 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1354 | spin_lock(&ino->i_lock); |
1355 | ||
251ec410 | 1356 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1357 | nfsi->layout = new; |
251ec410 | 1358 | return new; |
7175fe90 YN |
1359 | } else if (new != NULL) |
1360 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1361 | out_existing: |
1362 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1363 | return nfsi->layout; |
1364 | } | |
1365 | ||
b1f69b75 AA |
1366 | /* |
1367 | * iomode matching rules: | |
c7d73af2 TH |
1368 | * iomode lseg strict match |
1369 | * iomode | |
1370 | * ----- ----- ------ ----- | |
1371 | * ANY READ N/A true | |
1372 | * ANY RW N/A true | |
1373 | * RW READ N/A false | |
1374 | * RW RW N/A true | |
1375 | * READ READ N/A true | |
1376 | * READ RW true false | |
1377 | * READ RW false true | |
b1f69b75 | 1378 | */ |
3cb2df17 | 1379 | static bool |
7dc0ac70 | 1380 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
c7d73af2 TH |
1381 | const struct pnfs_layout_range *range, |
1382 | bool strict_iomode) | |
b1f69b75 | 1383 | { |
fb3296eb BH |
1384 | struct pnfs_layout_range range1; |
1385 | ||
1386 | if ((range->iomode == IOMODE_RW && | |
1387 | ls_range->iomode != IOMODE_RW) || | |
c7d73af2 TH |
1388 | (range->iomode != ls_range->iomode && |
1389 | strict_iomode == true) || | |
7dc0ac70 | 1390 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1391 | return 0; |
1392 | ||
1393 | /* range1 covers only the first byte in the range */ | |
1394 | range1 = *range; | |
1395 | range1.length = 1; | |
7dc0ac70 | 1396 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1397 | } |
1398 | ||
1399 | /* | |
1400 | * lookup range in layout | |
1401 | */ | |
e5e94017 | 1402 | static struct pnfs_layout_segment * |
fb3296eb | 1403 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
c7d73af2 TH |
1404 | struct pnfs_layout_range *range, |
1405 | bool strict_iomode) | |
e5e94017 | 1406 | { |
b1f69b75 AA |
1407 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1408 | ||
1409 | dprintk("%s:Begin\n", __func__); | |
1410 | ||
b7edfaa1 | 1411 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1412 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1413 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
c7d73af2 TH |
1414 | pnfs_lseg_range_match(&lseg->pls_range, range, |
1415 | strict_iomode)) { | |
9369a431 | 1416 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1417 | break; |
1418 | } | |
b1f69b75 AA |
1419 | } |
1420 | ||
1421 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1422 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1423 | return ret; |
e5e94017 BH |
1424 | } |
1425 | ||
d23d61c8 AA |
1426 | /* |
1427 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1428 | * to the MDS or over pNFS | |
1429 | * | |
1430 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1431 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1432 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1433 | * WRITE request. | |
1434 | * | |
1435 | * A return of true means use MDS I/O. | |
1436 | * | |
1437 | * From rfc 5661: | |
1438 | * If a file's size is smaller than the file size threshold, data accesses | |
1439 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1440 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1441 | * server. If both file size and I/O size are provided, the client SHOULD | |
1442 | * reach or exceed both thresholds before sending its read or write | |
1443 | * requests to the data server. | |
1444 | */ | |
1445 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1446 | struct inode *ino, int iomode) | |
1447 | { | |
1448 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1449 | struct nfs_inode *nfsi = NFS_I(ino); | |
1450 | loff_t fsize = i_size_read(ino); | |
1451 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1452 | ||
1453 | if (t == NULL) | |
1454 | return ret; | |
1455 | ||
1456 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1457 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1458 | ||
1459 | switch (iomode) { | |
1460 | case IOMODE_READ: | |
1461 | if (t->bm & THRESHOLD_RD) { | |
1462 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1463 | size_set = true; | |
1464 | if (fsize < t->rd_sz) | |
1465 | size = true; | |
1466 | } | |
1467 | if (t->bm & THRESHOLD_RD_IO) { | |
1468 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1469 | nfsi->read_io); | |
1470 | io_set = true; | |
1471 | if (nfsi->read_io < t->rd_io_sz) | |
1472 | io = true; | |
1473 | } | |
1474 | break; | |
1475 | case IOMODE_RW: | |
1476 | if (t->bm & THRESHOLD_WR) { | |
1477 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1478 | size_set = true; | |
1479 | if (fsize < t->wr_sz) | |
1480 | size = true; | |
1481 | } | |
1482 | if (t->bm & THRESHOLD_WR_IO) { | |
1483 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1484 | nfsi->write_io); | |
1485 | io_set = true; | |
1486 | if (nfsi->write_io < t->wr_io_sz) | |
1487 | io = true; | |
1488 | } | |
1489 | break; | |
1490 | } | |
1491 | if (size_set && io_set) { | |
1492 | if (size && io) | |
1493 | ret = true; | |
1494 | } else if (size || io) | |
1495 | ret = true; | |
1496 | ||
1497 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1498 | return ret; | |
1499 | } | |
1500 | ||
aa8a45ee PT |
1501 | static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo) |
1502 | { | |
1503 | /* | |
1504 | * send layoutcommit as it can hold up layoutreturn due to lseg | |
1505 | * reference | |
1506 | */ | |
1507 | pnfs_layoutcommit_inode(lo->plh_inode, false); | |
1508 | return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
2e5b29f0 | 1509 | nfs_wait_bit_killable, |
aa8a45ee PT |
1510 | TASK_UNINTERRUPTIBLE); |
1511 | } | |
1512 | ||
d67ae825 TH |
1513 | static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo) |
1514 | { | |
1515 | unsigned long *bitlock = &lo->plh_flags; | |
1516 | ||
1517 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1518 | smp_mb__after_atomic(); | |
1519 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1520 | } | |
1521 | ||
e5e94017 BH |
1522 | /* |
1523 | * Layout segment is retreived from the server if not cached. | |
1524 | * The appropriate layout segment is referenced and returned to the caller. | |
1525 | */ | |
7c24d948 | 1526 | struct pnfs_layout_segment * |
e5e94017 BH |
1527 | pnfs_update_layout(struct inode *ino, |
1528 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1529 | loff_t pos, |
1530 | u64 count, | |
a75b9df9 | 1531 | enum pnfs_iomode iomode, |
c7d73af2 | 1532 | bool strict_iomode, |
a75b9df9 | 1533 | gfp_t gfp_flags) |
e5e94017 | 1534 | { |
fb3296eb BH |
1535 | struct pnfs_layout_range arg = { |
1536 | .iomode = iomode, | |
1537 | .offset = pos, | |
1538 | .length = count, | |
1539 | }; | |
183d9e7b | 1540 | unsigned pg_offset, seq; |
6382a441 WAA |
1541 | struct nfs_server *server = NFS_SERVER(ino); |
1542 | struct nfs_client *clp = server->nfs_client; | |
183d9e7b | 1543 | struct pnfs_layout_hdr *lo = NULL; |
e5e94017 | 1544 | struct pnfs_layout_segment *lseg = NULL; |
183d9e7b JL |
1545 | nfs4_stateid stateid; |
1546 | long timeout = 0; | |
66b53f32 | 1547 | unsigned long giveup = jiffies + (clp->cl_lease_time << 1); |
30005121 | 1548 | bool first; |
e5e94017 | 1549 | |
9a4bf31d | 1550 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) { |
183d9e7b | 1551 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1552 | PNFS_UPDATE_LAYOUT_NO_PNFS); |
f86bbcf8 | 1553 | goto out; |
9a4bf31d | 1554 | } |
d23d61c8 | 1555 | |
9a4bf31d | 1556 | if (iomode == IOMODE_READ && i_size_read(ino) == 0) { |
183d9e7b | 1557 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1558 | PNFS_UPDATE_LAYOUT_RD_ZEROLEN); |
4ae93560 | 1559 | goto out; |
9a4bf31d | 1560 | } |
4ae93560 | 1561 | |
9a4bf31d | 1562 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) { |
183d9e7b | 1563 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1564 | PNFS_UPDATE_LAYOUT_MDSTHRESH); |
f86bbcf8 | 1565 | goto out; |
9a4bf31d | 1566 | } |
d23d61c8 | 1567 | |
9bf87482 | 1568 | lookup_again: |
b88fa69e | 1569 | nfs4_client_recover_expired_lease(clp); |
9bf87482 | 1570 | first = false; |
e5e94017 | 1571 | spin_lock(&ino->i_lock); |
9fa40758 | 1572 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1573 | if (lo == NULL) { |
1574 | spin_unlock(&ino->i_lock); | |
183d9e7b | 1575 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1576 | PNFS_UPDATE_LAYOUT_NOMEM); |
830ffb56 TM |
1577 | goto out; |
1578 | } | |
e5e94017 | 1579 | |
43f1b3da | 1580 | /* Do we even need to bother with this? */ |
a59c30ac | 1581 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
183d9e7b | 1582 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1583 | PNFS_UPDATE_LAYOUT_BULK_RECALL); |
43f1b3da | 1584 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1585 | goto out_unlock; |
1586 | } | |
1587 | ||
1588 | /* if LAYOUTGET already failed once we don't try again */ | |
2e5b29f0 | 1589 | if (pnfs_layout_io_test_failed(lo, iomode)) { |
183d9e7b | 1590 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1591 | PNFS_UPDATE_LAYOUT_IO_TEST_FAIL); |
e5e94017 | 1592 | goto out_unlock; |
9a4bf31d | 1593 | } |
e5e94017 | 1594 | |
c7d73af2 | 1595 | lseg = pnfs_find_lseg(lo, &arg, strict_iomode); |
183d9e7b JL |
1596 | if (lseg) { |
1597 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1598 | PNFS_UPDATE_LAYOUT_FOUND_CACHED); | |
1599 | goto out_unlock; | |
1600 | } | |
1601 | ||
1602 | if (!nfs4_valid_open_stateid(ctx->state)) { | |
1603 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1604 | PNFS_UPDATE_LAYOUT_INVALID_OPEN); | |
1605 | goto out_unlock; | |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | * Choose a stateid for the LAYOUTGET. If we don't have a layout | |
1610 | * stateid, or it has been invalidated, then we must use the open | |
1611 | * stateid. | |
1612 | */ | |
67a3b721 | 1613 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { |
183d9e7b JL |
1614 | |
1615 | /* | |
1616 | * The first layoutget for the file. Need to serialize per | |
9bf87482 PT |
1617 | * RFC 5661 Errata 3208. |
1618 | */ | |
1619 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1620 | &lo->plh_flags)) { | |
1621 | spin_unlock(&ino->i_lock); | |
1622 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1623 | TASK_UNINTERRUPTIBLE); | |
1624 | pnfs_put_layout_hdr(lo); | |
183d9e7b | 1625 | dprintk("%s retrying\n", __func__); |
9bf87482 PT |
1626 | goto lookup_again; |
1627 | } | |
183d9e7b JL |
1628 | |
1629 | first = true; | |
1630 | do { | |
1631 | seq = read_seqbegin(&ctx->state->seqlock); | |
1632 | nfs4_stateid_copy(&stateid, &ctx->state->stateid); | |
1633 | } while (read_seqretry(&ctx->state->seqlock, seq)); | |
9bf87482 | 1634 | } else { |
183d9e7b | 1635 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
9bf87482 | 1636 | } |
568e8c49 | 1637 | |
aa8a45ee PT |
1638 | /* |
1639 | * Because we free lsegs before sending LAYOUTRETURN, we need to wait | |
1640 | * for LAYOUTRETURN even if first is true. | |
1641 | */ | |
8f70f53a | 1642 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
aa8a45ee PT |
1643 | spin_unlock(&ino->i_lock); |
1644 | dprintk("%s wait for layoutreturn\n", __func__); | |
1645 | if (pnfs_prepare_to_retry_layoutget(lo)) { | |
d67ae825 TH |
1646 | if (first) |
1647 | pnfs_clear_first_layoutget(lo); | |
aa8a45ee PT |
1648 | pnfs_put_layout_hdr(lo); |
1649 | dprintk("%s retrying\n", __func__); | |
183d9e7b JL |
1650 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
1651 | lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
aa8a45ee PT |
1652 | goto lookup_again; |
1653 | } | |
183d9e7b | 1654 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1655 | PNFS_UPDATE_LAYOUT_RETURN); |
aa8a45ee PT |
1656 | goto out_put_layout_hdr; |
1657 | } | |
1658 | ||
9a4bf31d | 1659 | if (pnfs_layoutgets_blocked(lo)) { |
183d9e7b | 1660 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1661 | PNFS_UPDATE_LAYOUT_BLOCKED); |
cf7d63f1 | 1662 | goto out_unlock; |
9a4bf31d | 1663 | } |
cf7d63f1 | 1664 | atomic_inc(&lo->plh_outstanding); |
f49f9baa | 1665 | spin_unlock(&ino->i_lock); |
30005121 | 1666 | |
abb9a007 | 1667 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1668 | /* The lo must be on the clp list if there is any |
1669 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1670 | */ | |
1671 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1672 | if (list_empty(&lo->plh_layouts)) |
1673 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1674 | spin_unlock(&clp->cl_lock); |
1675 | } | |
e5e94017 | 1676 | |
09cbfeaf | 1677 | pg_offset = arg.offset & ~PAGE_MASK; |
707ed5fd BH |
1678 | if (pg_offset) { |
1679 | arg.offset -= pg_offset; | |
1680 | arg.length += pg_offset; | |
1681 | } | |
7c24d948 | 1682 | if (arg.length != NFS4_MAX_UINT64) |
09cbfeaf | 1683 | arg.length = PAGE_ALIGN(arg.length); |
707ed5fd | 1684 | |
183d9e7b JL |
1685 | lseg = send_layoutget(lo, ctx, &stateid, &arg, &timeout, gfp_flags); |
1686 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1687 | PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET); | |
56b38a1f | 1688 | atomic_dec(&lo->plh_outstanding); |
83026d80 | 1689 | if (IS_ERR(lseg)) { |
183d9e7b | 1690 | switch(PTR_ERR(lseg)) { |
e85d7ee4 | 1691 | case -EBUSY: |
183d9e7b JL |
1692 | if (time_after(jiffies, giveup)) |
1693 | lseg = NULL; | |
66b53f32 TM |
1694 | break; |
1695 | case -ERECALLCONFLICT: | |
1696 | /* Huh? We hold no layouts, how is there a recall? */ | |
1697 | if (first) { | |
1698 | lseg = NULL; | |
1699 | break; | |
1700 | } | |
1701 | /* Destroy the existing layout and start over */ | |
1702 | if (time_after(jiffies, giveup)) | |
1703 | pnfs_destroy_layout(NFS_I(ino)); | |
183d9e7b JL |
1704 | /* Fallthrough */ |
1705 | case -EAGAIN: | |
56b38a1f | 1706 | break; |
183d9e7b JL |
1707 | default: |
1708 | if (!nfs_error_is_fatal(PTR_ERR(lseg))) { | |
1709 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
1710 | lseg = NULL; | |
1711 | } | |
56b38a1f TM |
1712 | goto out_put_layout_hdr; |
1713 | } | |
1714 | if (lseg) { | |
1715 | if (first) | |
1716 | pnfs_clear_first_layoutget(lo); | |
1717 | trace_pnfs_update_layout(ino, pos, count, | |
1718 | iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
1719 | pnfs_put_layout_hdr(lo); | |
1720 | goto lookup_again; | |
83026d80 JL |
1721 | } |
1722 | } else { | |
1723 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
1724 | } | |
1725 | ||
830ffb56 | 1726 | out_put_layout_hdr: |
d67ae825 TH |
1727 | if (first) |
1728 | pnfs_clear_first_layoutget(lo); | |
70c3bd2b | 1729 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1730 | out: |
f86bbcf8 TM |
1731 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1732 | "(%s, offset: %llu, length: %llu)\n", | |
1733 | __func__, ino->i_sb->s_id, | |
1734 | (unsigned long long)NFS_FILEID(ino), | |
d600ad1f | 1735 | IS_ERR_OR_NULL(lseg) ? "not found" : "found", |
f86bbcf8 TM |
1736 | iomode==IOMODE_RW ? "read/write" : "read-only", |
1737 | (unsigned long long)pos, | |
1738 | (unsigned long long)count); | |
e5e94017 BH |
1739 | return lseg; |
1740 | out_unlock: | |
1741 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1742 | goto out_put_layout_hdr; |
e5e94017 | 1743 | } |
7c24d948 | 1744 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1745 | |
540d9864 TM |
1746 | static bool |
1747 | pnfs_sanity_check_layout_range(struct pnfs_layout_range *range) | |
1748 | { | |
1749 | switch (range->iomode) { | |
1750 | case IOMODE_READ: | |
1751 | case IOMODE_RW: | |
1752 | break; | |
1753 | default: | |
1754 | return false; | |
1755 | } | |
1756 | if (range->offset == NFS4_MAX_UINT64) | |
1757 | return false; | |
1758 | if (range->length == 0) | |
1759 | return false; | |
1760 | if (range->length != NFS4_MAX_UINT64 && | |
1761 | range->length > NFS4_MAX_UINT64 - range->offset) | |
1762 | return false; | |
1763 | return true; | |
1764 | } | |
1765 | ||
a0b0a6e3 | 1766 | struct pnfs_layout_segment * |
b1f69b75 AA |
1767 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1768 | { | |
1769 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1770 | struct nfs4_layoutget_res *res = &lgp->res; | |
1771 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1772 | struct inode *ino = lo->plh_inode; |
78096cca | 1773 | LIST_HEAD(free_me); |
540d9864 TM |
1774 | |
1775 | if (!pnfs_sanity_check_layout_range(&res->range)) | |
1b3c6d07 | 1776 | return ERR_PTR(-EINVAL); |
b1f69b75 AA |
1777 | |
1778 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1779 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
1b3c6d07 | 1780 | if (IS_ERR_OR_NULL(lseg)) { |
b1f69b75 | 1781 | if (!lseg) |
1b3c6d07 JL |
1782 | lseg = ERR_PTR(-ENOMEM); |
1783 | ||
1784 | dprintk("%s: Could not allocate layout: error %ld\n", | |
1785 | __func__, PTR_ERR(lseg)); | |
1786 | return lseg; | |
b1f69b75 AA |
1787 | } |
1788 | ||
119cef97 | 1789 | pnfs_init_lseg(lo, lseg, &res->range, &res->stateid); |
1013df61 | 1790 | |
b1f69b75 | 1791 | spin_lock(&ino->i_lock); |
e1c06f80 | 1792 | if (pnfs_layoutgets_blocked(lo)) { |
43f1b3da | 1793 | dprintk("%s forget reply due to state\n", __func__); |
1b3c6d07 | 1794 | goto out_forget; |
43f1b3da | 1795 | } |
038d6493 | 1796 | |
362f7474 CH |
1797 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1798 | /* existing state ID, make sure the sequence number matches. */ | |
1799 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1800 | dprintk("%s forget reply due to sequence\n", __func__); | |
1b3c6d07 | 1801 | goto out_forget; |
362f7474 CH |
1802 | } |
1803 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1804 | } else { | |
1805 | /* | |
1806 | * We got an entirely new state ID. Mark all segments for the | |
1807 | * inode invalid, and don't bother validating the stateid | |
1808 | * sequence number. | |
1809 | */ | |
d9b61708 | 1810 | pnfs_mark_layout_stateid_invalid(lo, &free_me); |
362f7474 | 1811 | |
2a59a041 | 1812 | pnfs_set_layout_stateid(lo, &res->stateid, true); |
362f7474 | 1813 | } |
038d6493 | 1814 | |
9369a431 | 1815 | pnfs_get_lseg(lseg); |
03772d2f | 1816 | pnfs_layout_insert_lseg(lo, lseg, &free_me); |
8e0acf90 | 1817 | |
b1f69b75 | 1818 | |
3976143b | 1819 | if (res->return_on_close) |
f7e8917a | 1820 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); |
f7e8917a | 1821 | |
b1f69b75 | 1822 | spin_unlock(&ino->i_lock); |
78096cca | 1823 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1824 | return lseg; |
43f1b3da | 1825 | |
1b3c6d07 | 1826 | out_forget: |
43f1b3da FI |
1827 | spin_unlock(&ino->i_lock); |
1828 | lseg->pls_layout = lo; | |
1829 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1b3c6d07 | 1830 | return ERR_PTR(-EAGAIN); |
b1f69b75 AA |
1831 | } |
1832 | ||
016256df | 1833 | static void |
3982a6a2 JL |
1834 | pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode, |
1835 | u32 seq) | |
5c97f5de | 1836 | { |
2d6cf5ab | 1837 | if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode) |
5c97f5de TM |
1838 | iomode = IOMODE_ANY; |
1839 | lo->plh_return_iomode = iomode; | |
e0fa0d01 | 1840 | set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); |
2d6cf5ab TM |
1841 | if (seq != 0) { |
1842 | WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq); | |
3982a6a2 | 1843 | lo->plh_return_seq = seq; |
2d6cf5ab | 1844 | } |
5c97f5de TM |
1845 | } |
1846 | ||
2f215968 TM |
1847 | /** |
1848 | * pnfs_mark_matching_lsegs_return - Free or return matching layout segments | |
1849 | * @lo: pointer to layout header | |
1850 | * @tmp_list: list header to be used with pnfs_free_lseg_list() | |
1851 | * @return_range: describe layout segment ranges to be returned | |
1852 | * | |
1853 | * This function is mainly intended for use by layoutrecall. It attempts | |
1854 | * to free the layout segment immediately, or else to mark it for return | |
1855 | * as soon as its reference count drops to zero. | |
1856 | */ | |
10335556 | 1857 | int |
016256df PT |
1858 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, |
1859 | struct list_head *tmp_list, | |
6d597e17 JL |
1860 | const struct pnfs_layout_range *return_range, |
1861 | u32 seq) | |
016256df PT |
1862 | { |
1863 | struct pnfs_layout_segment *lseg, *next; | |
10335556 | 1864 | int remaining = 0; |
016256df PT |
1865 | |
1866 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
1867 | ||
1868 | if (list_empty(&lo->plh_segs)) | |
10335556 | 1869 | return 0; |
016256df | 1870 | |
fc7ff367 | 1871 | assert_spin_locked(&lo->plh_inode->i_lock); |
016256df PT |
1872 | |
1873 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
e036f464 | 1874 | if (pnfs_match_lseg_recall(lseg, return_range, seq)) { |
016256df PT |
1875 | dprintk("%s: marking lseg %p iomode %d " |
1876 | "offset %llu length %llu\n", __func__, | |
1877 | lseg, lseg->pls_range.iomode, | |
1878 | lseg->pls_range.offset, | |
1879 | lseg->pls_range.length); | |
2f215968 TM |
1880 | if (mark_lseg_invalid(lseg, tmp_list)) |
1881 | continue; | |
1882 | remaining++; | |
016256df | 1883 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); |
016256df | 1884 | } |
6d597e17 JL |
1885 | |
1886 | if (remaining) | |
1887 | pnfs_set_plh_return_info(lo, return_range->iomode, seq); | |
1888 | ||
10335556 | 1889 | return remaining; |
016256df PT |
1890 | } |
1891 | ||
1892 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
1893 | struct pnfs_layout_segment *lseg) | |
1894 | { | |
1895 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
016256df PT |
1896 | struct pnfs_layout_range range = { |
1897 | .iomode = lseg->pls_range.iomode, | |
1898 | .offset = 0, | |
1899 | .length = NFS4_MAX_UINT64, | |
1900 | }; | |
1901 | LIST_HEAD(free_me); | |
10335556 | 1902 | bool return_now = false; |
016256df PT |
1903 | |
1904 | spin_lock(&inode->i_lock); | |
2d6cf5ab | 1905 | pnfs_set_plh_return_info(lo, range.iomode, 0); |
016256df PT |
1906 | /* |
1907 | * mark all matching lsegs so that we are sure to have no live | |
1908 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
1909 | * for how it works. | |
1910 | */ | |
2d6cf5ab | 1911 | if (!pnfs_mark_matching_lsegs_return(lo, &free_me, &range, 0)) { |
10335556 | 1912 | nfs4_stateid stateid; |
e5fd1904 | 1913 | enum pnfs_iomode iomode; |
10335556 | 1914 | |
e5fd1904 | 1915 | return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode); |
10335556 TM |
1916 | spin_unlock(&inode->i_lock); |
1917 | if (return_now) | |
1918 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); | |
1919 | } else { | |
1920 | spin_unlock(&inode->i_lock); | |
1921 | nfs_commit_inode(inode, 0); | |
1922 | } | |
016256df PT |
1923 | pnfs_free_lseg_list(&free_me); |
1924 | } | |
1925 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
1926 | ||
d8007d4d TM |
1927 | void |
1928 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1929 | { | |
1fd937bd PT |
1930 | u64 rd_size = req->wb_bytes; |
1931 | ||
cb5d04bc PT |
1932 | if (pgio->pg_lseg == NULL) { |
1933 | if (pgio->pg_dreq == NULL) | |
1934 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1935 | else | |
1936 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1937 | ||
1938 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1939 | req->wb_context, | |
1940 | req_offset(req), | |
1941 | rd_size, | |
1942 | IOMODE_READ, | |
c7d73af2 | 1943 | false, |
cb5d04bc | 1944 | GFP_KERNEL); |
d600ad1f PT |
1945 | if (IS_ERR(pgio->pg_lseg)) { |
1946 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
1947 | pgio->pg_lseg = NULL; | |
1948 | return; | |
1949 | } | |
cb5d04bc | 1950 | } |
e885de1a TM |
1951 | /* If no lseg, fall back to read through mds */ |
1952 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1953 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1954 | |
d8007d4d TM |
1955 | } |
1956 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1957 | ||
1958 | void | |
6296556f PT |
1959 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1960 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1961 | { |
d600ad1f | 1962 | if (pgio->pg_lseg == NULL) { |
cb5d04bc PT |
1963 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1964 | req->wb_context, | |
1965 | req_offset(req), | |
1966 | wb_size, | |
1967 | IOMODE_RW, | |
c7d73af2 | 1968 | false, |
cb5d04bc | 1969 | GFP_NOFS); |
d600ad1f PT |
1970 | if (IS_ERR(pgio->pg_lseg)) { |
1971 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
1972 | pgio->pg_lseg = NULL; | |
1973 | return; | |
1974 | } | |
1975 | } | |
e885de1a TM |
1976 | /* If no lseg, fall back to write through mds */ |
1977 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1978 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1979 | } |
1980 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1981 | ||
180bb5ec WAA |
1982 | void |
1983 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
1984 | { | |
1985 | if (desc->pg_lseg) { | |
1986 | pnfs_put_lseg(desc->pg_lseg); | |
1987 | desc->pg_lseg = NULL; | |
1988 | } | |
1989 | } | |
1990 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
1991 | ||
b4fdac1a WAA |
1992 | /* |
1993 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1994 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1995 | */ | |
1996 | size_t | |
a7d42ddb WAA |
1997 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
1998 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 1999 | { |
0f9c429e | 2000 | unsigned int size; |
c5e20cb7 | 2001 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
2002 | |
2003 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
2004 | if (!size) |
2005 | return 0; | |
94ad1c80 | 2006 | |
19982ba8 | 2007 | /* |
c5e20cb7 WAA |
2008 | * 'size' contains the number of bytes left in the current page (up |
2009 | * to the original size asked for in @req->wb_bytes). | |
2010 | * | |
2011 | * Calculate how many bytes are left in the layout segment | |
2012 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
2013 | * |
2014 | * Please also note that 'end_offset' is actually the offset of the | |
2015 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
2016 | * | |
2017 | */ | |
19b54848 | 2018 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
2019 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
2020 | pgio->pg_lseg->pls_range.length); | |
2021 | req_start = req_offset(req); | |
7c13789e | 2022 | WARN_ON_ONCE(req_start >= seg_end); |
c5e20cb7 | 2023 | /* start of request is past the last byte of this segment */ |
7c13789e WAA |
2024 | if (req_start >= seg_end) { |
2025 | /* reference the new lseg */ | |
2026 | if (pgio->pg_ops->pg_cleanup) | |
2027 | pgio->pg_ops->pg_cleanup(pgio); | |
2028 | if (pgio->pg_ops->pg_init) | |
2029 | pgio->pg_ops->pg_init(pgio, req); | |
19b54848 | 2030 | return 0; |
7c13789e | 2031 | } |
c5e20cb7 WAA |
2032 | |
2033 | /* adjust 'size' iff there are fewer bytes left in the | |
2034 | * segment than what nfs_generic_pg_test returned */ | |
2035 | seg_left = seg_end - req_start; | |
2036 | if (seg_left < size) | |
2037 | size = (unsigned int)seg_left; | |
19b54848 | 2038 | } |
0f9c429e | 2039 | |
19b54848 | 2040 | return size; |
94ad1c80 | 2041 | } |
89a58e32 | 2042 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 2043 | |
53113ad3 | 2044 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
2045 | { |
2046 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
2047 | |
2048 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
2049 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
2050 | hdr->completion_ops); | |
c7070113 | 2051 | set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags); |
53113ad3 | 2052 | return nfs_pageio_resend(&pgio, hdr); |
e2fecb21 | 2053 | } |
e7dd79af | 2054 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 2055 | |
d45f60c6 | 2056 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2057 | { |
cd841605 FI |
2058 | |
2059 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
2060 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2061 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2062 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2063 | } |
6c75dc0d | 2064 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2065 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
2066 | } |
2067 | ||
d20581aa BH |
2068 | /* |
2069 | * Called by non rpc-based layout drivers | |
2070 | */ | |
d45f60c6 | 2071 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 2072 | { |
f8417b48 | 2073 | if (likely(!hdr->pnfs_error)) { |
67af7611 TM |
2074 | pnfs_set_layoutcommit(hdr->inode, hdr->lseg, |
2075 | hdr->mds_offset + hdr->res.count); | |
d45f60c6 | 2076 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2077 | } |
2078 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); | |
2079 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2080 | pnfs_ld_handle_write_error(hdr); |
2081 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 2082 | } |
d20581aa | 2083 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 2084 | |
dce81290 TM |
2085 | static void |
2086 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2087 | struct nfs_pgio_header *hdr) |
dce81290 | 2088 | { |
48d635f1 | 2089 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2090 | |
6c75dc0d | 2091 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2092 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 2093 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 2094 | mirror->pg_recoalesce = 1; |
6c75dc0d | 2095 | } |
d45f60c6 | 2096 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2097 | hdr->release(hdr); |
dce81290 TM |
2098 | } |
2099 | ||
2100 | static enum pnfs_try_status | |
d45f60c6 | 2101 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
2102 | const struct rpc_call_ops *call_ops, |
2103 | struct pnfs_layout_segment *lseg, | |
2104 | int how) | |
0382b744 | 2105 | { |
cd841605 | 2106 | struct inode *inode = hdr->inode; |
0382b744 AA |
2107 | enum pnfs_try_status trypnfs; |
2108 | struct nfs_server *nfss = NFS_SERVER(inode); | |
2109 | ||
cd841605 | 2110 | hdr->mds_ops = call_ops; |
0382b744 AA |
2111 | |
2112 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
2113 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
2114 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 2115 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 2116 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
2117 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2118 | return trypnfs; | |
2119 | } | |
2120 | ||
dce81290 | 2121 | static void |
7f714720 WAA |
2122 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
2123 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 2124 | { |
dce81290 TM |
2125 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2126 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2127 | enum pnfs_try_status trypnfs; |
dce81290 | 2128 | |
d45f60c6 | 2129 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 2130 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 2131 | pnfs_write_through_mds(desc, hdr); |
dce81290 TM |
2132 | } |
2133 | ||
6c75dc0d FI |
2134 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
2135 | { | |
9369a431 | 2136 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2137 | nfs_pgio_header_free(hdr); |
6c75dc0d FI |
2138 | } |
2139 | ||
dce81290 TM |
2140 | int |
2141 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
2142 | { | |
6c75dc0d | 2143 | struct nfs_pgio_header *hdr; |
dce81290 TM |
2144 | int ret; |
2145 | ||
1e7f3a48 WAA |
2146 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2147 | if (!hdr) { | |
2bff2288 PT |
2148 | desc->pg_error = -ENOMEM; |
2149 | return desc->pg_error; | |
dce81290 | 2150 | } |
6c75dc0d | 2151 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 2152 | |
9369a431 | 2153 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2154 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2155 | if (!ret) |
7f714720 | 2156 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 2157 | |
6c75dc0d | 2158 | return ret; |
dce81290 TM |
2159 | } |
2160 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
2161 | ||
53113ad3 | 2162 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
2163 | { |
2164 | struct nfs_pageio_descriptor pgio; | |
2165 | ||
1acbbb4e | 2166 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
2167 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
2168 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 2169 | } |
e7dd79af | 2170 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 2171 | |
d45f60c6 | 2172 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2173 | { |
cd841605 FI |
2174 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
2175 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2176 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2177 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2178 | } |
4db6e0b7 | 2179 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2180 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
2181 | } |
2182 | ||
d20581aa BH |
2183 | /* |
2184 | * Called by non rpc-based layout drivers | |
2185 | */ | |
d45f60c6 | 2186 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 2187 | { |
cd841605 | 2188 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
2189 | __nfs4_read_done_cb(hdr); |
2190 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
f8417b48 KM |
2191 | } |
2192 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); | |
2193 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2194 | pnfs_ld_handle_read_error(hdr); |
2195 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
2196 | } |
2197 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
2198 | ||
493292dd TM |
2199 | static void |
2200 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2201 | struct nfs_pgio_header *hdr) |
493292dd | 2202 | { |
48d635f1 | 2203 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2204 | |
4db6e0b7 | 2205 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2206 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 2207 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 2208 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 2209 | } |
d45f60c6 | 2210 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2211 | hdr->release(hdr); |
493292dd TM |
2212 | } |
2213 | ||
64419a9b AA |
2214 | /* |
2215 | * Call the appropriate parallel I/O subsystem read function. | |
2216 | */ | |
493292dd | 2217 | static enum pnfs_try_status |
d45f60c6 | 2218 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
2219 | const struct rpc_call_ops *call_ops, |
2220 | struct pnfs_layout_segment *lseg) | |
64419a9b | 2221 | { |
cd841605 | 2222 | struct inode *inode = hdr->inode; |
64419a9b AA |
2223 | struct nfs_server *nfss = NFS_SERVER(inode); |
2224 | enum pnfs_try_status trypnfs; | |
2225 | ||
cd841605 | 2226 | hdr->mds_ops = call_ops; |
64419a9b AA |
2227 | |
2228 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 2229 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 2230 | |
d45f60c6 | 2231 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 2232 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 2233 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
2234 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2235 | return trypnfs; | |
2236 | } | |
863a3c6c | 2237 | |
ceb11e13 | 2238 | /* Resend all requests through pnfs. */ |
1b1bc66b | 2239 | void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) |
ceb11e13 PT |
2240 | { |
2241 | struct nfs_pageio_descriptor pgio; | |
2242 | ||
1b1bc66b WAA |
2243 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
2244 | nfs_pageio_init_read(&pgio, hdr->inode, false, | |
2245 | hdr->completion_ops); | |
2246 | hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr); | |
2247 | } | |
ceb11e13 PT |
2248 | } |
2249 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
2250 | ||
493292dd | 2251 | static void |
7f714720 | 2252 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 2253 | { |
493292dd TM |
2254 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2255 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2256 | enum pnfs_try_status trypnfs; |
493292dd | 2257 | |
d45f60c6 | 2258 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
ceb11e13 | 2259 | if (trypnfs == PNFS_TRY_AGAIN) |
1b1bc66b WAA |
2260 | pnfs_read_resend_pnfs(hdr); |
2261 | if (trypnfs == PNFS_NOT_ATTEMPTED || hdr->task.tk_status) | |
d45f60c6 | 2262 | pnfs_read_through_mds(desc, hdr); |
493292dd TM |
2263 | } |
2264 | ||
4db6e0b7 FI |
2265 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
2266 | { | |
9369a431 | 2267 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2268 | nfs_pgio_header_free(hdr); |
4db6e0b7 FI |
2269 | } |
2270 | ||
493292dd TM |
2271 | int |
2272 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
2273 | { | |
4db6e0b7 | 2274 | struct nfs_pgio_header *hdr; |
493292dd TM |
2275 | int ret; |
2276 | ||
1e7f3a48 WAA |
2277 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2278 | if (!hdr) { | |
2bff2288 PT |
2279 | desc->pg_error = -ENOMEM; |
2280 | return desc->pg_error; | |
493292dd | 2281 | } |
4db6e0b7 | 2282 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 2283 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2284 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2285 | if (!ret) |
7f714720 | 2286 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 2287 | return ret; |
493292dd TM |
2288 | } |
2289 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
2290 | ||
71244d9b TM |
2291 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
2292 | { | |
2293 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
2294 | ||
2295 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 2296 | smp_mb__after_atomic(); |
71244d9b TM |
2297 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
2298 | } | |
2299 | ||
863a3c6c | 2300 | /* |
a9bae566 | 2301 | * There can be multiple RW segments. |
863a3c6c | 2302 | */ |
a9bae566 | 2303 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 2304 | { |
a9bae566 | 2305 | struct pnfs_layout_segment *lseg; |
863a3c6c | 2306 | |
a9bae566 PT |
2307 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
2308 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 2309 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
2310 | list_add(&lseg->pls_lc_list, listp); |
2311 | } | |
863a3c6c AA |
2312 | } |
2313 | ||
a073dbff TM |
2314 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
2315 | { | |
2316 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
2317 | |
2318 | /* Matched by references in pnfs_set_layoutcommit */ | |
2319 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
2320 | list_del_init(&lseg->pls_lc_list); | |
2321 | pnfs_put_lseg(lseg); | |
2322 | } | |
2323 | ||
71244d9b | 2324 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
2325 | } |
2326 | ||
1b0ae068 PT |
2327 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
2328 | { | |
b9e028fd | 2329 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
2330 | } |
2331 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
2332 | ||
863a3c6c | 2333 | void |
67af7611 TM |
2334 | pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, |
2335 | loff_t end_pos) | |
863a3c6c | 2336 | { |
cd841605 | 2337 | struct nfs_inode *nfsi = NFS_I(inode); |
79a48a1f | 2338 | bool mark_as_dirty = false; |
863a3c6c | 2339 | |
cd841605 | 2340 | spin_lock(&inode->i_lock); |
863a3c6c | 2341 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
29559b11 | 2342 | nfsi->layout->plh_lwb = end_pos; |
79a48a1f | 2343 | mark_as_dirty = true; |
863a3c6c | 2344 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 2345 | __func__, inode->i_ino); |
29559b11 TM |
2346 | } else if (end_pos > nfsi->layout->plh_lwb) |
2347 | nfsi->layout->plh_lwb = end_pos; | |
67af7611 | 2348 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) { |
a9bae566 | 2349 | /* references matched in nfs4_layoutcommit_release */ |
67af7611 | 2350 | pnfs_get_lseg(lseg); |
a9bae566 | 2351 | } |
cd841605 | 2352 | spin_unlock(&inode->i_lock); |
acff5880 | 2353 | dprintk("%s: lseg %p end_pos %llu\n", |
67af7611 | 2354 | __func__, lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
2355 | |
2356 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2357 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2358 | if (mark_as_dirty) | |
cd841605 | 2359 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
2360 | } |
2361 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
2362 | ||
db29c089 AA |
2363 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2364 | { | |
2365 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2366 | ||
2367 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2368 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2369 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2370 | } |
2371 | ||
de4b15c7 AA |
2372 | /* |
2373 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2374 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2375 | * data to disk to allow the server to recover the data if it crashes. | |
2376 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2377 | * is off, and a COMMIT is sent to a data server, or | |
2378 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2379 | */ | |
863a3c6c | 2380 | int |
ef311537 | 2381 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2382 | { |
5f919c9f | 2383 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2384 | struct nfs4_layoutcommit_data *data; |
2385 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2386 | loff_t end_pos; |
71244d9b | 2387 | int status; |
863a3c6c | 2388 | |
71244d9b | 2389 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2390 | return 0; |
2391 | ||
71244d9b | 2392 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2393 | |
71244d9b | 2394 | status = -EAGAIN; |
92407e75 | 2395 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2396 | if (!sync) |
2397 | goto out; | |
74316201 | 2398 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2399 | NFS_INO_LAYOUTCOMMITTING, |
2400 | nfs_wait_bit_killable, | |
2401 | TASK_KILLABLE); | |
92407e75 | 2402 | if (status) |
71244d9b | 2403 | goto out; |
92407e75 PT |
2404 | } |
2405 | ||
71244d9b TM |
2406 | status = -ENOMEM; |
2407 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2408 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2409 | if (!data) | |
2410 | goto clear_layoutcommitting; | |
2411 | ||
2412 | status = 0; | |
de4b15c7 | 2413 | spin_lock(&inode->i_lock); |
71244d9b TM |
2414 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2415 | goto out_unlock; | |
a9bae566 | 2416 | |
71244d9b | 2417 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2418 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2419 | |
acff5880 | 2420 | end_pos = nfsi->layout->plh_lwb; |
863a3c6c | 2421 | |
f597c537 | 2422 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2423 | spin_unlock(&inode->i_lock); |
2424 | ||
2425 | data->args.inode = inode; | |
9fa40758 | 2426 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2427 | nfs_fattr_init(&data->fattr); |
2428 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2429 | data->res.fattr = &data->fattr; | |
2e18d4d8 TM |
2430 | if (end_pos != 0) |
2431 | data->args.lastbytewritten = end_pos - 1; | |
2432 | else | |
2433 | data->args.lastbytewritten = U64_MAX; | |
863a3c6c AA |
2434 | data->res.server = NFS_SERVER(inode); |
2435 | ||
5f919c9f CH |
2436 | if (ld->prepare_layoutcommit) { |
2437 | status = ld->prepare_layoutcommit(&data->args); | |
2438 | if (status) { | |
3471648a | 2439 | put_rpccred(data->cred); |
5f919c9f | 2440 | spin_lock(&inode->i_lock); |
29559b11 TM |
2441 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); |
2442 | if (end_pos > nfsi->layout->plh_lwb) | |
5f919c9f | 2443 | nfsi->layout->plh_lwb = end_pos; |
3471648a | 2444 | goto out_unlock; |
5f919c9f CH |
2445 | } |
2446 | } | |
2447 | ||
2448 | ||
863a3c6c AA |
2449 | status = nfs4_proc_layoutcommit(data, sync); |
2450 | out: | |
92407e75 PT |
2451 | if (status) |
2452 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2453 | dprintk("<-- %s status %d\n", __func__, status); |
2454 | return status; | |
71244d9b TM |
2455 | out_unlock: |
2456 | spin_unlock(&inode->i_lock); | |
92407e75 | 2457 | kfree(data); |
71244d9b TM |
2458 | clear_layoutcommitting: |
2459 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2460 | goto out; |
863a3c6c | 2461 | } |
72cff449 | 2462 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a | 2463 | |
5bb89b47 TM |
2464 | int |
2465 | pnfs_generic_sync(struct inode *inode, bool datasync) | |
2466 | { | |
2467 | return pnfs_layoutcommit_inode(inode, true); | |
2468 | } | |
2469 | EXPORT_SYMBOL_GPL(pnfs_generic_sync); | |
2470 | ||
82be417a AA |
2471 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) |
2472 | { | |
2473 | struct nfs4_threshold *thp; | |
2474 | ||
2475 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2476 | if (!thp) { | |
2477 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2478 | return NULL; | |
2479 | } | |
2480 | return thp; | |
2481 | } | |
8733408d | 2482 | |
865a7ecb | 2483 | #if IS_ENABLED(CONFIG_NFS_V4_2) |
8733408d | 2484 | int |
c8ad8894 | 2485 | pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags) |
8733408d PT |
2486 | { |
2487 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
2488 | struct nfs_server *server = NFS_SERVER(inode); | |
1bfe3b25 | 2489 | struct nfs_inode *nfsi = NFS_I(inode); |
8733408d PT |
2490 | struct nfs42_layoutstat_data *data; |
2491 | struct pnfs_layout_hdr *hdr; | |
2492 | int status = 0; | |
2493 | ||
2494 | if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats) | |
2495 | goto out; | |
2496 | ||
6c5a0d89 TM |
2497 | if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS)) |
2498 | goto out; | |
2499 | ||
1bfe3b25 PT |
2500 | if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags)) |
2501 | goto out; | |
2502 | ||
8733408d PT |
2503 | spin_lock(&inode->i_lock); |
2504 | if (!NFS_I(inode)->layout) { | |
2505 | spin_unlock(&inode->i_lock); | |
f538d0ba | 2506 | goto out_clear_layoutstats; |
8733408d PT |
2507 | } |
2508 | hdr = NFS_I(inode)->layout; | |
2509 | pnfs_get_layout_hdr(hdr); | |
2510 | spin_unlock(&inode->i_lock); | |
2511 | ||
c8ad8894 | 2512 | data = kzalloc(sizeof(*data), gfp_flags); |
8733408d PT |
2513 | if (!data) { |
2514 | status = -ENOMEM; | |
2515 | goto out_put; | |
2516 | } | |
2517 | ||
2518 | data->args.fh = NFS_FH(inode); | |
2519 | data->args.inode = inode; | |
8733408d PT |
2520 | status = ld->prepare_layoutstats(&data->args); |
2521 | if (status) | |
2522 | goto out_free; | |
2523 | ||
2524 | status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data); | |
2525 | ||
2526 | out: | |
2527 | dprintk("%s returns %d\n", __func__, status); | |
2528 | return status; | |
2529 | ||
2530 | out_free: | |
2531 | kfree(data); | |
2532 | out_put: | |
2533 | pnfs_put_layout_hdr(hdr); | |
f538d0ba | 2534 | out_clear_layoutstats: |
1bfe3b25 PT |
2535 | smp_mb__before_atomic(); |
2536 | clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags); | |
2537 | smp_mb__after_atomic(); | |
8733408d PT |
2538 | goto out; |
2539 | } | |
2540 | EXPORT_SYMBOL_GPL(pnfs_report_layoutstat); | |
865a7ecb | 2541 | #endif |
bbf58bf3 TM |
2542 | |
2543 | unsigned int layoutstats_timer; | |
2544 | module_param(layoutstats_timer, uint, 0644); | |
2545 | EXPORT_SYMBOL_GPL(layoutstats_timer); |