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 | */ | |
9fd4b9fc | 262 | static 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 AA |
336 | static void |
337 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
338 | { | |
566052c5 | 339 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 340 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c FI |
341 | atomic_set(&lseg->pls_refcount, 1); |
342 | smp_mb(); | |
343 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 344 | lseg->pls_layout = lo; |
974cec8c AA |
345 | } |
346 | ||
905ca191 | 347 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 348 | { |
b7edfaa1 | 349 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 350 | |
b1f69b75 | 351 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
974cec8c AA |
352 | } |
353 | ||
d684d2ae | 354 | static void |
57036a37 TM |
355 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
356 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 357 | { |
57036a37 | 358 | struct inode *inode = lo->plh_inode; |
d684d2ae | 359 | |
d20581aa | 360 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 361 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
362 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
363 | atomic_dec(&lo->plh_refcount); | |
2d148c7e TM |
364 | if (list_empty(&lo->plh_segs)) { |
365 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
173f77e9 | 366 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
2d148c7e | 367 | } |
d684d2ae FI |
368 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); |
369 | } | |
370 | ||
bae724ef | 371 | void |
9369a431 | 372 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 373 | { |
57036a37 | 374 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
375 | struct inode *inode; |
376 | ||
377 | if (!lseg) | |
378 | return; | |
379 | ||
4541d16c FI |
380 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
381 | atomic_read(&lseg->pls_refcount), | |
382 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
4ef2e4f8 | 383 | |
57036a37 TM |
384 | lo = lseg->pls_layout; |
385 | inode = lo->plh_inode; | |
4ef2e4f8 | 386 | |
d684d2ae | 387 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
faa4a54f TM |
388 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { |
389 | spin_unlock(&inode->i_lock); | |
390 | return; | |
391 | } | |
8f0d27dc | 392 | pnfs_get_layout_hdr(lo); |
4ef2e4f8 TM |
393 | pnfs_layout_remove_lseg(lo, lseg); |
394 | spin_unlock(&inode->i_lock); | |
395 | pnfs_free_lseg(lseg); | |
396 | pnfs_put_layout_hdr(lo); | |
4541d16c | 397 | } |
4541d16c | 398 | } |
9369a431 | 399 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 400 | |
6543f803 | 401 | static void pnfs_free_lseg_async_work(struct work_struct *work) |
e6cf82d1 WAA |
402 | { |
403 | struct pnfs_layout_segment *lseg; | |
6543f803 | 404 | struct pnfs_layout_hdr *lo; |
e6cf82d1 WAA |
405 | |
406 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
6543f803 | 407 | lo = lseg->pls_layout; |
e6cf82d1 | 408 | |
6543f803 TM |
409 | pnfs_free_lseg(lseg); |
410 | pnfs_put_layout_hdr(lo); | |
e6cf82d1 WAA |
411 | } |
412 | ||
6543f803 | 413 | static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg) |
e6cf82d1 | 414 | { |
6543f803 | 415 | INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work); |
e6cf82d1 WAA |
416 | schedule_work(&lseg->pls_work); |
417 | } | |
6543f803 TM |
418 | |
419 | void | |
420 | pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg) | |
421 | { | |
422 | if (!lseg) | |
423 | return; | |
424 | ||
425 | assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock); | |
426 | ||
427 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, | |
428 | atomic_read(&lseg->pls_refcount), | |
429 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
430 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
431 | struct pnfs_layout_hdr *lo = lseg->pls_layout; | |
faa4a54f TM |
432 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) |
433 | return; | |
6543f803 TM |
434 | pnfs_get_layout_hdr(lo); |
435 | pnfs_layout_remove_lseg(lo, lseg); | |
436 | pnfs_free_lseg_async(lseg); | |
437 | } | |
438 | } | |
439 | EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked); | |
e6cf82d1 | 440 | |
3cb2df17 | 441 | static u64 |
fb3296eb BH |
442 | end_offset(u64 start, u64 len) |
443 | { | |
444 | u64 end; | |
445 | ||
446 | end = start + len; | |
447 | return end >= start ? end : NFS4_MAX_UINT64; | |
448 | } | |
449 | ||
fb3296eb BH |
450 | /* |
451 | * is l2 fully contained in l1? | |
452 | * start1 end1 | |
453 | * [----------------------------------) | |
454 | * start2 end2 | |
455 | * [----------------) | |
456 | */ | |
3cb2df17 | 457 | static bool |
7dc0ac70 | 458 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 459 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
460 | { |
461 | u64 start1 = l1->offset; | |
462 | u64 end1 = end_offset(start1, l1->length); | |
463 | u64 start2 = l2->offset; | |
464 | u64 end2 = end_offset(start2, l2->length); | |
465 | ||
466 | return (start1 <= start2) && (end1 >= end2); | |
467 | } | |
468 | ||
469 | /* | |
470 | * is l1 and l2 intersecting? | |
471 | * start1 end1 | |
472 | * [----------------------------------) | |
473 | * start2 end2 | |
474 | * [----------------) | |
475 | */ | |
3cb2df17 | 476 | static bool |
7dc0ac70 | 477 | pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1, |
3cb2df17 | 478 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
479 | { |
480 | u64 start1 = l1->offset; | |
481 | u64 end1 = end_offset(start1, l1->length); | |
482 | u64 start2 = l2->offset; | |
483 | u64 end2 = end_offset(start2, l2->length); | |
484 | ||
485 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
486 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
487 | } | |
488 | ||
4541d16c | 489 | static bool |
3cb2df17 TM |
490 | should_free_lseg(const struct pnfs_layout_range *lseg_range, |
491 | const struct pnfs_layout_range *recall_range) | |
4541d16c | 492 | { |
778b5502 BH |
493 | return (recall_range->iomode == IOMODE_ANY || |
494 | lseg_range->iomode == recall_range->iomode) && | |
7dc0ac70 | 495 | pnfs_lseg_range_intersecting(lseg_range, recall_range); |
974cec8c AA |
496 | } |
497 | ||
24956804 TM |
498 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
499 | struct list_head *tmp_list) | |
500 | { | |
501 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
502 | return false; | |
503 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
504 | list_add(&lseg->pls_list, tmp_list); | |
505 | return true; | |
506 | } | |
507 | ||
4541d16c FI |
508 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
509 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
510 | struct list_head *tmp_list) | |
511 | { | |
512 | int rv = 0; | |
513 | ||
514 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
515 | /* Remove the reference keeping the lseg in the | |
516 | * list. It will now be removed when all | |
517 | * outstanding io is finished. | |
518 | */ | |
d684d2ae FI |
519 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
520 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 521 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 522 | rv = 1; |
4541d16c FI |
523 | } |
524 | return rv; | |
525 | } | |
526 | ||
6d597e17 JL |
527 | /* |
528 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
529 | * taking into account wraparound issues. | |
530 | */ | |
531 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
532 | { | |
533 | return (s32)(s1 - s2) > 0; | |
534 | } | |
535 | ||
536 | /** | |
537 | * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later | |
538 | * @lo: layout header containing the lsegs | |
539 | * @tmp_list: list head where doomed lsegs should go | |
540 | * @recall_range: optional recall range argument to match (may be NULL) | |
541 | * @seq: only invalidate lsegs obtained prior to this sequence (may be 0) | |
542 | * | |
543 | * Walk the list of lsegs in the layout header, and tear down any that should | |
544 | * be destroyed. If "recall_range" is specified then the segment must match | |
545 | * that range. If "seq" is non-zero, then only match segments that were handed | |
546 | * out at or before that sequence. | |
547 | * | |
548 | * Returns number of matching invalid lsegs remaining in list after scanning | |
549 | * it and purging them. | |
4541d16c | 550 | */ |
43f1b3da | 551 | int |
49a85061 | 552 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 553 | struct list_head *tmp_list, |
6d597e17 JL |
554 | const struct pnfs_layout_range *recall_range, |
555 | u32 seq) | |
974cec8c AA |
556 | { |
557 | struct pnfs_layout_segment *lseg, *next; | |
71b39854 | 558 | int remaining = 0; |
974cec8c AA |
559 | |
560 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
561 | ||
8006bfba | 562 | if (list_empty(&lo->plh_segs)) |
38511722 | 563 | return 0; |
4541d16c | 564 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
565 | if (!recall_range || |
566 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
6d597e17 JL |
567 | if (seq && pnfs_seqid_is_newer(lseg->pls_seq, seq)) |
568 | continue; | |
569 | dprintk("%s: freeing lseg %p iomode %d seq %u" | |
4541d16c | 570 | "offset %llu length %llu\n", __func__, |
6d597e17 JL |
571 | lseg, lseg->pls_range.iomode, lseg->pls_seq, |
572 | lseg->pls_range.offset, lseg->pls_range.length); | |
71b39854 TM |
573 | if (!mark_lseg_invalid(lseg, tmp_list)) |
574 | remaining++; | |
4541d16c | 575 | } |
71b39854 TM |
576 | dprintk("%s:Return %i\n", __func__, remaining); |
577 | return remaining; | |
974cec8c AA |
578 | } |
579 | ||
f49f9baa | 580 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 581 | void |
4541d16c | 582 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 583 | { |
4541d16c | 584 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
585 | |
586 | if (list_empty(free_me)) | |
587 | return; | |
588 | ||
4541d16c | 589 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 590 | list_del(&lseg->pls_list); |
905ca191 | 591 | pnfs_free_lseg(lseg); |
974cec8c AA |
592 | } |
593 | } | |
594 | ||
e5e94017 BH |
595 | void |
596 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
597 | { | |
598 | struct pnfs_layout_hdr *lo; | |
974cec8c | 599 | LIST_HEAD(tmp_list); |
e5e94017 BH |
600 | |
601 | spin_lock(&nfsi->vfs_inode.i_lock); | |
602 | lo = nfsi->layout; | |
603 | if (lo) { | |
3e621214 | 604 | pnfs_get_layout_hdr(lo); |
2454dfea | 605 | pnfs_mark_layout_stateid_invalid(lo, &tmp_list); |
3e621214 TM |
606 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); |
607 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
608 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
609 | pnfs_free_lseg_list(&tmp_list); | |
610 | pnfs_put_layout_hdr(lo); | |
611 | } else | |
612 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 613 | } |
041245c8 | 614 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 615 | |
fd9a8d71 TM |
616 | static bool |
617 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
618 | struct list_head *layout_list) | |
974cec8c AA |
619 | { |
620 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 621 | bool ret = false; |
974cec8c | 622 | |
fd9a8d71 TM |
623 | spin_lock(&inode->i_lock); |
624 | lo = NFS_I(inode)->layout; | |
625 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
626 | pnfs_get_layout_hdr(lo); | |
627 | list_add(&lo->plh_bulk_destroy, layout_list); | |
628 | ret = true; | |
629 | } | |
630 | spin_unlock(&inode->i_lock); | |
631 | return ret; | |
632 | } | |
633 | ||
634 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
635 | static int | |
636 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
637 | struct nfs_server *server, | |
638 | struct list_head *layout_list) | |
639 | { | |
640 | struct pnfs_layout_hdr *lo, *next; | |
641 | struct inode *inode; | |
642 | ||
643 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
644 | inode = igrab(lo->plh_inode); | |
645 | if (inode == NULL) | |
646 | continue; | |
647 | list_del_init(&lo->plh_layouts); | |
648 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
649 | continue; | |
650 | rcu_read_unlock(); | |
651 | spin_unlock(&clp->cl_lock); | |
652 | iput(inode); | |
653 | spin_lock(&clp->cl_lock); | |
654 | rcu_read_lock(); | |
655 | return -EAGAIN; | |
656 | } | |
657 | return 0; | |
658 | } | |
659 | ||
660 | static int | |
661 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
662 | bool is_bulk_recall) | |
663 | { | |
664 | struct pnfs_layout_hdr *lo; | |
665 | struct inode *inode; | |
fd9a8d71 TM |
666 | LIST_HEAD(lseg_list); |
667 | int ret = 0; | |
668 | ||
669 | while (!list_empty(layout_list)) { | |
670 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
671 | plh_bulk_destroy); | |
672 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
673 | lo->plh_inode->i_ino); | |
674 | inode = lo->plh_inode; | |
7c5d1875 CH |
675 | |
676 | pnfs_layoutcommit_inode(inode, false); | |
677 | ||
fd9a8d71 TM |
678 | spin_lock(&inode->i_lock); |
679 | list_del_init(&lo->plh_bulk_destroy); | |
9fd4b9fc TM |
680 | if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) { |
681 | if (is_bulk_recall) | |
682 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
fd9a8d71 | 683 | ret = -EAGAIN; |
9fd4b9fc | 684 | } |
fd9a8d71 TM |
685 | spin_unlock(&inode->i_lock); |
686 | pnfs_free_lseg_list(&lseg_list); | |
b20135d0 TM |
687 | /* Free all lsegs that are attached to commit buckets */ |
688 | nfs_commit_inode(inode, 0); | |
fd9a8d71 TM |
689 | pnfs_put_layout_hdr(lo); |
690 | iput(inode); | |
691 | } | |
692 | return ret; | |
693 | } | |
694 | ||
695 | int | |
696 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
697 | struct nfs_fsid *fsid, | |
698 | bool is_recall) | |
699 | { | |
700 | struct nfs_server *server; | |
701 | LIST_HEAD(layout_list); | |
c47abcf8 | 702 | |
974cec8c | 703 | spin_lock(&clp->cl_lock); |
6382a441 | 704 | rcu_read_lock(); |
fd9a8d71 | 705 | restart: |
6382a441 | 706 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
707 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
708 | continue; | |
709 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
710 | server, | |
711 | &layout_list) != 0) | |
712 | goto restart; | |
6382a441 WAA |
713 | } |
714 | rcu_read_unlock(); | |
974cec8c AA |
715 | spin_unlock(&clp->cl_lock); |
716 | ||
fd9a8d71 TM |
717 | if (list_empty(&layout_list)) |
718 | return 0; | |
719 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
720 | } | |
721 | ||
722 | int | |
723 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
724 | bool is_recall) | |
725 | { | |
726 | struct nfs_server *server; | |
727 | LIST_HEAD(layout_list); | |
728 | ||
729 | spin_lock(&clp->cl_lock); | |
730 | rcu_read_lock(); | |
731 | restart: | |
732 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
733 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
734 | server, | |
735 | &layout_list) != 0) | |
736 | goto restart; | |
974cec8c | 737 | } |
fd9a8d71 TM |
738 | rcu_read_unlock(); |
739 | spin_unlock(&clp->cl_lock); | |
740 | ||
741 | if (list_empty(&layout_list)) | |
742 | return 0; | |
743 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
744 | } | |
745 | ||
746 | /* | |
747 | * Called by the state manger to remove all layouts established under an | |
748 | * expired lease. | |
749 | */ | |
750 | void | |
751 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
752 | { | |
753 | nfs4_deviceid_mark_client_invalid(clp); | |
754 | nfs4_deviceid_purge_client(clp); | |
755 | ||
756 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
757 | } |
758 | ||
fd6002e9 | 759 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
760 | void |
761 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
762 | bool update_barrier) | |
b1f69b75 | 763 | { |
22aaf714 TM |
764 | u32 oldseq, newseq, new_barrier; |
765 | int empty = list_empty(&lo->plh_segs); | |
b1f69b75 | 766 | |
2d2f24ad TM |
767 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
768 | newseq = be32_to_cpu(new->seqid); | |
22aaf714 | 769 | if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { |
f597c537 | 770 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 771 | if (update_barrier) { |
22aaf714 | 772 | new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
773 | } else { |
774 | /* Because of wraparound, we want to keep the barrier | |
22aaf714 | 775 | * "close" to the current seqids. |
43f1b3da | 776 | */ |
22aaf714 | 777 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); |
43f1b3da | 778 | } |
22aaf714 TM |
779 | if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
780 | lo->plh_barrier = new_barrier; | |
43f1b3da | 781 | } |
b1f69b75 AA |
782 | } |
783 | ||
cf7d63f1 | 784 | static bool |
19c54aba TM |
785 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
786 | const nfs4_stateid *stateid) | |
43f1b3da | 787 | { |
19c54aba | 788 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 789 | |
19c54aba TM |
790 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
791 | } | |
792 | ||
793 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
794 | static bool | |
e1c06f80 | 795 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo) |
19c54aba | 796 | { |
f7e8917a | 797 | return lo->plh_block_lgets || |
e1c06f80 | 798 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
cf7d63f1 FI |
799 | } |
800 | ||
b1f69b75 | 801 | /* |
183d9e7b JL |
802 | * Get layout from server. |
803 | * for now, assume that whole file layouts are requested. | |
804 | * arg->offset: 0 | |
805 | * arg->length: all ones | |
806 | */ | |
e5e94017 BH |
807 | static struct pnfs_layout_segment * |
808 | send_layoutget(struct pnfs_layout_hdr *lo, | |
809 | struct nfs_open_context *ctx, | |
183d9e7b | 810 | nfs4_stateid *stateid, |
e144e539 | 811 | const struct pnfs_layout_range *range, |
183d9e7b | 812 | long *timeout, gfp_t gfp_flags) |
e5e94017 | 813 | { |
b7edfaa1 | 814 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
815 | struct nfs_server *server = NFS_SERVER(ino); |
816 | struct nfs4_layoutget *lgp; | |
2d89a1d3 | 817 | loff_t i_size; |
b1f69b75 AA |
818 | |
819 | dprintk("--> %s\n", __func__); | |
e5e94017 | 820 | |
4f2e9dce JL |
821 | /* |
822 | * Synchronously retrieve layout information from server and | |
823 | * store in lseg. If we race with a concurrent seqid morphing | |
824 | * op, then re-send the LAYOUTGET. | |
825 | */ | |
83026d80 JL |
826 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
827 | if (lgp == NULL) | |
828 | return ERR_PTR(-ENOMEM); | |
829 | ||
830 | i_size = i_size_read(ino); | |
831 | ||
832 | lgp->args.minlength = PAGE_SIZE; | |
833 | if (lgp->args.minlength > range->length) | |
834 | lgp->args.minlength = range->length; | |
835 | if (range->iomode == IOMODE_READ) { | |
836 | if (range->offset >= i_size) | |
837 | lgp->args.minlength = 0; | |
838 | else if (i_size - range->offset < lgp->args.minlength) | |
839 | lgp->args.minlength = i_size - range->offset; | |
d03ab29d | 840 | } |
83026d80 JL |
841 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
842 | pnfs_copy_range(&lgp->args.range, range); | |
843 | lgp->args.type = server->pnfs_curr_ld->id; | |
844 | lgp->args.inode = ino; | |
845 | lgp->args.ctx = get_nfs_open_context(ctx); | |
183d9e7b | 846 | nfs4_stateid_copy(&lgp->args.stateid, stateid); |
83026d80 JL |
847 | lgp->gfp_flags = gfp_flags; |
848 | lgp->cred = lo->plh_lc_cred; | |
35124a09 | 849 | |
183d9e7b | 850 | return nfs4_proc_layoutget(lgp, timeout, gfp_flags); |
974cec8c AA |
851 | } |
852 | ||
24956804 TM |
853 | static void pnfs_clear_layoutcommit(struct inode *inode, |
854 | struct list_head *head) | |
855 | { | |
856 | struct nfs_inode *nfsi = NFS_I(inode); | |
857 | struct pnfs_layout_segment *lseg, *tmp; | |
858 | ||
859 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
860 | return; | |
861 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
862 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
863 | continue; | |
864 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
865 | } | |
866 | } | |
867 | ||
d67ae825 TH |
868 | void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo) |
869 | { | |
870 | clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
871 | smp_mb__after_atomic(); | |
872 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN); | |
7f27392c | 873 | rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq); |
d67ae825 TH |
874 | } |
875 | ||
13c13a6a TM |
876 | static bool |
877 | pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo) | |
878 | { | |
879 | if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) | |
880 | return false; | |
881 | lo->plh_return_iomode = 0; | |
3982a6a2 | 882 | lo->plh_return_seq = 0; |
13c13a6a | 883 | pnfs_get_layout_hdr(lo); |
2370abda | 884 | clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); |
13c13a6a TM |
885 | return true; |
886 | } | |
887 | ||
f40eb5d0 | 888 | static int |
ed429d6b | 889 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid, |
6c16605d | 890 | enum pnfs_iomode iomode, bool sync) |
f40eb5d0 PT |
891 | { |
892 | struct inode *ino = lo->plh_inode; | |
893 | struct nfs4_layoutreturn *lrp; | |
894 | int status = 0; | |
895 | ||
e4af440a | 896 | lrp = kzalloc(sizeof(*lrp), GFP_NOFS); |
f40eb5d0 PT |
897 | if (unlikely(lrp == NULL)) { |
898 | status = -ENOMEM; | |
899 | spin_lock(&ino->i_lock); | |
d67ae825 | 900 | pnfs_clear_layoutreturn_waitbit(lo); |
f40eb5d0 PT |
901 | spin_unlock(&ino->i_lock); |
902 | pnfs_put_layout_hdr(lo); | |
903 | goto out; | |
904 | } | |
905 | ||
ed429d6b | 906 | nfs4_stateid_copy(&lrp->args.stateid, stateid); |
f40eb5d0 PT |
907 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; |
908 | lrp->args.inode = ino; | |
15eb67c1 PT |
909 | lrp->args.range.iomode = iomode; |
910 | lrp->args.range.offset = 0; | |
911 | lrp->args.range.length = NFS4_MAX_UINT64; | |
f40eb5d0 PT |
912 | lrp->args.layout = lo; |
913 | lrp->clp = NFS_SERVER(ino)->nfs_client; | |
914 | lrp->cred = lo->plh_lc_cred; | |
915 | ||
6c16605d | 916 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
917 | out: |
918 | dprintk("<-- %s status: %d\n", __func__, status); | |
919 | return status; | |
920 | } | |
921 | ||
13c13a6a TM |
922 | /* Return true if layoutreturn is needed */ |
923 | static bool | |
924 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo) | |
925 | { | |
926 | struct pnfs_layout_segment *s; | |
927 | ||
2370abda | 928 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
929 | return false; |
930 | ||
931 | /* Defer layoutreturn until all lsegs are done */ | |
932 | list_for_each_entry(s, &lo->plh_segs, pls_list) { | |
933 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags)) | |
934 | return false; | |
935 | } | |
936 | ||
937 | return true; | |
938 | } | |
939 | ||
940 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo) | |
941 | { | |
942 | struct inode *inode= lo->plh_inode; | |
943 | ||
2370abda | 944 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
945 | return; |
946 | spin_lock(&inode->i_lock); | |
947 | if (pnfs_layout_need_return(lo)) { | |
948 | nfs4_stateid stateid; | |
949 | enum pnfs_iomode iomode; | |
950 | bool send; | |
951 | ||
952 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); | |
3982a6a2 | 953 | stateid.seqid = cpu_to_be32(lo->plh_return_seq); |
13c13a6a TM |
954 | iomode = lo->plh_return_iomode; |
955 | send = pnfs_prepare_layoutreturn(lo); | |
956 | spin_unlock(&inode->i_lock); | |
957 | if (send) { | |
958 | /* Send an async layoutreturn so we dont deadlock */ | |
959 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); | |
960 | } | |
961 | } else | |
962 | spin_unlock(&inode->i_lock); | |
963 | } | |
964 | ||
293b3b06 AA |
965 | /* |
966 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
967 | * when the layout segment list is empty. | |
968 | * | |
969 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
970 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
971 | * deviceid is marked invalid. | |
972 | */ | |
cbe82603 BH |
973 | int |
974 | _pnfs_return_layout(struct inode *ino) | |
975 | { | |
976 | struct pnfs_layout_hdr *lo = NULL; | |
977 | struct nfs_inode *nfsi = NFS_I(ino); | |
978 | LIST_HEAD(tmp_list); | |
cbe82603 | 979 | nfs4_stateid stateid; |
293b3b06 | 980 | int status = 0, empty; |
7f27392c | 981 | bool send; |
cbe82603 | 982 | |
366d5052 | 983 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
984 | |
985 | spin_lock(&ino->i_lock); | |
986 | lo = nfsi->layout; | |
e5929f3c | 987 | if (!lo) { |
cbe82603 | 988 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
989 | dprintk("NFS: %s no layout to return\n", __func__); |
990 | goto out; | |
cbe82603 | 991 | } |
50f563ef | 992 | nfs4_stateid_copy(&stateid, &nfsi->layout->plh_stateid); |
cbe82603 | 993 | /* Reference matched in nfs4_layoutreturn_release */ |
70c3bd2b | 994 | pnfs_get_layout_hdr(lo); |
293b3b06 | 995 | empty = list_empty(&lo->plh_segs); |
24956804 | 996 | pnfs_clear_layoutcommit(ino, &tmp_list); |
6d597e17 | 997 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0); |
c88953d8 CH |
998 | |
999 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
1000 | struct pnfs_layout_range range = { | |
1001 | .iomode = IOMODE_ANY, | |
1002 | .offset = 0, | |
1003 | .length = NFS4_MAX_UINT64, | |
1004 | }; | |
1005 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
1006 | } | |
1007 | ||
293b3b06 AA |
1008 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
1009 | if (empty) { | |
1010 | spin_unlock(&ino->i_lock); | |
293b3b06 | 1011 | dprintk("NFS: %s no layout segments to return\n", __func__); |
7f27392c | 1012 | goto out_put_layout_hdr; |
293b3b06 | 1013 | } |
47abadef CH |
1014 | |
1015 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
7f27392c | 1016 | send = pnfs_prepare_layoutreturn(lo); |
cbe82603 BH |
1017 | spin_unlock(&ino->i_lock); |
1018 | pnfs_free_lseg_list(&tmp_list); | |
7f27392c | 1019 | if (send) |
ed429d6b | 1020 | status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
7f27392c TM |
1021 | out_put_layout_hdr: |
1022 | pnfs_put_layout_hdr(lo); | |
cbe82603 BH |
1023 | out: |
1024 | dprintk("<-- %s status: %d\n", __func__, status); | |
1025 | return status; | |
1026 | } | |
0a57cdac | 1027 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 1028 | |
24028672 TM |
1029 | int |
1030 | pnfs_commit_and_return_layout(struct inode *inode) | |
1031 | { | |
1032 | struct pnfs_layout_hdr *lo; | |
1033 | int ret; | |
1034 | ||
1035 | spin_lock(&inode->i_lock); | |
1036 | lo = NFS_I(inode)->layout; | |
1037 | if (lo == NULL) { | |
1038 | spin_unlock(&inode->i_lock); | |
1039 | return 0; | |
1040 | } | |
1041 | pnfs_get_layout_hdr(lo); | |
1042 | /* Block new layoutgets and read/write to ds */ | |
1043 | lo->plh_block_lgets++; | |
1044 | spin_unlock(&inode->i_lock); | |
1045 | filemap_fdatawait(inode->i_mapping); | |
1046 | ret = pnfs_layoutcommit_inode(inode, true); | |
1047 | if (ret == 0) | |
1048 | ret = _pnfs_return_layout(inode); | |
1049 | spin_lock(&inode->i_lock); | |
1050 | lo->plh_block_lgets--; | |
1051 | spin_unlock(&inode->i_lock); | |
1052 | pnfs_put_layout_hdr(lo); | |
1053 | return ret; | |
1054 | } | |
1055 | ||
f7e8917a FI |
1056 | bool pnfs_roc(struct inode *ino) |
1057 | { | |
40dd4b7a TM |
1058 | struct nfs_inode *nfsi = NFS_I(ino); |
1059 | struct nfs_open_context *ctx; | |
1060 | struct nfs4_state *state; | |
f7e8917a FI |
1061 | struct pnfs_layout_hdr *lo; |
1062 | struct pnfs_layout_segment *lseg, *tmp; | |
193e3aa2 | 1063 | nfs4_stateid stateid; |
f7e8917a | 1064 | LIST_HEAD(tmp_list); |
e755d638 | 1065 | bool found = false, layoutreturn = false, roc = false; |
f7e8917a FI |
1066 | |
1067 | spin_lock(&ino->i_lock); | |
40dd4b7a | 1068 | lo = nfsi->layout; |
3976143b | 1069 | if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) |
40dd4b7a TM |
1070 | goto out_noroc; |
1071 | ||
e755d638 | 1072 | /* no roc if we hold a delegation */ |
40dd4b7a TM |
1073 | if (nfs4_check_delegation(ino, FMODE_READ)) |
1074 | goto out_noroc; | |
1075 | ||
1076 | list_for_each_entry(ctx, &nfsi->open_files, list) { | |
1077 | state = ctx->state; | |
1078 | /* Don't return layout if there is open file state */ | |
1079 | if (state != NULL && state->state != 0) | |
1080 | goto out_noroc; | |
1081 | } | |
1082 | ||
50f563ef | 1083 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
e755d638 | 1084 | /* always send layoutreturn if being marked so */ |
2370abda | 1085 | if (test_and_clear_bit(NFS_LAYOUT_RETURN_REQUESTED, |
e755d638 PT |
1086 | &lo->plh_flags)) |
1087 | layoutreturn = pnfs_prepare_layoutreturn(lo); | |
1088 | ||
f7e8917a | 1089 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) |
e755d638 PT |
1090 | /* If we are sending layoutreturn, invalidate all valid lsegs */ |
1091 | if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
f7e8917a FI |
1092 | mark_lseg_invalid(lseg, &tmp_list); |
1093 | found = true; | |
1094 | } | |
500d701f | 1095 | /* ROC in two conditions: |
e755d638 PT |
1096 | * 1. there are ROC lsegs |
1097 | * 2. we don't send layoutreturn | |
e755d638 | 1098 | */ |
500d701f PT |
1099 | if (found && !layoutreturn) { |
1100 | /* lo ref dropped in pnfs_roc_release() */ | |
1101 | pnfs_get_layout_hdr(lo); | |
e755d638 | 1102 | roc = true; |
500d701f | 1103 | } |
f7e8917a | 1104 | |
40dd4b7a | 1105 | out_noroc: |
f7e8917a | 1106 | spin_unlock(&ino->i_lock); |
e755d638 PT |
1107 | pnfs_free_lseg_list(&tmp_list); |
1108 | pnfs_layoutcommit_inode(ino, true); | |
1109 | if (layoutreturn) | |
ed429d6b | 1110 | pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
e755d638 | 1111 | return roc; |
f7e8917a FI |
1112 | } |
1113 | ||
1114 | void pnfs_roc_release(struct inode *ino) | |
1115 | { | |
1116 | struct pnfs_layout_hdr *lo; | |
1117 | ||
1118 | spin_lock(&ino->i_lock); | |
1119 | lo = NFS_I(ino)->layout; | |
5c4a79fb | 1120 | pnfs_clear_layoutreturn_waitbit(lo); |
6622c3ea TM |
1121 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
1122 | pnfs_detach_layout_hdr(lo); | |
1123 | spin_unlock(&ino->i_lock); | |
1124 | pnfs_free_layout_hdr(lo); | |
1125 | } else | |
1126 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
1127 | } |
1128 | ||
1129 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
1130 | { | |
1131 | struct pnfs_layout_hdr *lo; | |
1132 | ||
1133 | spin_lock(&ino->i_lock); | |
1134 | lo = NFS_I(ino)->layout; | |
0654cc72 | 1135 | pnfs_mark_layout_returned_if_empty(lo); |
0f35ad6f | 1136 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
1137 | lo->plh_barrier = barrier; |
1138 | spin_unlock(&ino->i_lock); | |
6a463beb | 1139 | trace_nfs4_layoutreturn_on_close(ino, 0); |
f7e8917a FI |
1140 | } |
1141 | ||
4ff376fe | 1142 | void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier) |
f7e8917a FI |
1143 | { |
1144 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 1145 | struct pnfs_layout_hdr *lo; |
7fdab069 | 1146 | u32 current_seqid; |
f7e8917a FI |
1147 | |
1148 | spin_lock(&ino->i_lock); | |
7fdab069 TM |
1149 | lo = nfsi->layout; |
1150 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1151 | |
7fdab069 TM |
1152 | /* Since close does not return a layout stateid for use as |
1153 | * a barrier, we choose the worst-case barrier. | |
1154 | */ | |
1155 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
f7e8917a | 1156 | spin_unlock(&ino->i_lock); |
f7e8917a FI |
1157 | } |
1158 | ||
500d701f PT |
1159 | bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task) |
1160 | { | |
1161 | struct nfs_inode *nfsi = NFS_I(ino); | |
1162 | struct pnfs_layout_hdr *lo; | |
1163 | bool sleep = false; | |
1164 | ||
1165 | /* we might not have grabbed lo reference. so need to check under | |
1166 | * i_lock */ | |
1167 | spin_lock(&ino->i_lock); | |
1168 | lo = nfsi->layout; | |
1169 | if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) | |
1170 | sleep = true; | |
1171 | spin_unlock(&ino->i_lock); | |
1172 | ||
1173 | if (sleep) | |
1174 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
1175 | ||
1176 | return sleep; | |
1177 | } | |
1178 | ||
b1f69b75 AA |
1179 | /* |
1180 | * Compare two layout segments for sorting into layout cache. | |
1181 | * We want to preferentially return RW over RO layouts, so ensure those | |
1182 | * are seen first. | |
1183 | */ | |
1184 | static s64 | |
7dc0ac70 | 1185 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1186 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1187 | { |
fb3296eb BH |
1188 | s64 d; |
1189 | ||
1190 | /* high offset > low offset */ | |
1191 | d = l1->offset - l2->offset; | |
1192 | if (d) | |
1193 | return d; | |
1194 | ||
1195 | /* short length > long length */ | |
1196 | d = l2->length - l1->length; | |
1197 | if (d) | |
1198 | return d; | |
1199 | ||
b1f69b75 | 1200 | /* read > read/write */ |
fb3296eb | 1201 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1202 | } |
1203 | ||
03772d2f TM |
1204 | static bool |
1205 | pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1, | |
1206 | const struct pnfs_layout_range *l2) | |
1207 | { | |
1208 | return pnfs_lseg_range_cmp(l1, l2) > 0; | |
1209 | } | |
1210 | ||
1211 | static bool | |
1212 | pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg, | |
1213 | struct pnfs_layout_segment *old) | |
1214 | { | |
1215 | return false; | |
1216 | } | |
1217 | ||
1218 | void | |
1219 | pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1220 | struct pnfs_layout_segment *lseg, | |
1221 | bool (*is_after)(const struct pnfs_layout_range *, | |
1222 | const struct pnfs_layout_range *), | |
1223 | bool (*do_merge)(struct pnfs_layout_segment *, | |
1224 | struct pnfs_layout_segment *), | |
1225 | struct list_head *free_me) | |
974cec8c | 1226 | { |
03772d2f | 1227 | struct pnfs_layout_segment *lp, *tmp; |
b1f69b75 | 1228 | |
974cec8c AA |
1229 | dprintk("%s:Begin\n", __func__); |
1230 | ||
03772d2f TM |
1231 | list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) { |
1232 | if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0) | |
1233 | continue; | |
1234 | if (do_merge(lseg, lp)) { | |
1235 | mark_lseg_invalid(lp, free_me); | |
1236 | continue; | |
1237 | } | |
1238 | if (is_after(&lseg->pls_range, &lp->pls_range)) | |
b1f69b75 | 1239 | continue; |
566052c5 | 1240 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1241 | dprintk("%s: inserted lseg %p " |
1242 | "iomode %d offset %llu length %llu before " | |
1243 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1244 | __func__, lseg, lseg->pls_range.iomode, |
1245 | lseg->pls_range.offset, lseg->pls_range.length, | |
1246 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1247 | lp->pls_range.length); | |
fb3296eb | 1248 | goto out; |
974cec8c | 1249 | } |
fb3296eb BH |
1250 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1251 | dprintk("%s: inserted lseg %p " | |
1252 | "iomode %d offset %llu length %llu at tail\n", | |
1253 | __func__, lseg, lseg->pls_range.iomode, | |
1254 | lseg->pls_range.offset, lseg->pls_range.length); | |
1255 | out: | |
70c3bd2b | 1256 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1257 | |
1258 | dprintk("%s:Return\n", __func__); | |
e5e94017 | 1259 | } |
03772d2f TM |
1260 | EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg); |
1261 | ||
1262 | static void | |
1263 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1264 | struct pnfs_layout_segment *lseg, | |
1265 | struct list_head *free_me) | |
1266 | { | |
1267 | struct inode *inode = lo->plh_inode; | |
1268 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
1269 | ||
1270 | if (ld->add_lseg != NULL) | |
1271 | ld->add_lseg(lo, lseg, free_me); | |
1272 | else | |
1273 | pnfs_generic_layout_insert_lseg(lo, lseg, | |
1274 | pnfs_lseg_range_is_after, | |
1275 | pnfs_lseg_no_merge, | |
1276 | free_me); | |
1277 | } | |
e5e94017 BH |
1278 | |
1279 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1280 | alloc_init_layout_hdr(struct inode *ino, |
1281 | struct nfs_open_context *ctx, | |
1282 | gfp_t gfp_flags) | |
e5e94017 BH |
1283 | { |
1284 | struct pnfs_layout_hdr *lo; | |
1285 | ||
636fb9c8 | 1286 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1287 | if (!lo) |
1288 | return NULL; | |
cc6e5340 | 1289 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1290 | INIT_LIST_HEAD(&lo->plh_layouts); |
1291 | INIT_LIST_HEAD(&lo->plh_segs); | |
fd9a8d71 | 1292 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1293 | lo->plh_inode = ino; |
5cc2216d | 1294 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
67a3b721 | 1295 | lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID; |
e5e94017 BH |
1296 | return lo; |
1297 | } | |
1298 | ||
1299 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1300 | pnfs_find_alloc_layout(struct inode *ino, |
1301 | struct nfs_open_context *ctx, | |
1302 | gfp_t gfp_flags) | |
e5241e43 TM |
1303 | __releases(&ino->i_lock) |
1304 | __acquires(&ino->i_lock) | |
e5e94017 BH |
1305 | { |
1306 | struct nfs_inode *nfsi = NFS_I(ino); | |
1307 | struct pnfs_layout_hdr *new = NULL; | |
1308 | ||
1309 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1310 | ||
251ec410 TM |
1311 | if (nfsi->layout != NULL) |
1312 | goto out_existing; | |
e5e94017 | 1313 | spin_unlock(&ino->i_lock); |
9fa40758 | 1314 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1315 | spin_lock(&ino->i_lock); |
1316 | ||
251ec410 | 1317 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1318 | nfsi->layout = new; |
251ec410 | 1319 | return new; |
7175fe90 YN |
1320 | } else if (new != NULL) |
1321 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1322 | out_existing: |
1323 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1324 | return nfsi->layout; |
1325 | } | |
1326 | ||
b1f69b75 AA |
1327 | /* |
1328 | * iomode matching rules: | |
c7d73af2 TH |
1329 | * iomode lseg strict match |
1330 | * iomode | |
1331 | * ----- ----- ------ ----- | |
1332 | * ANY READ N/A true | |
1333 | * ANY RW N/A true | |
1334 | * RW READ N/A false | |
1335 | * RW RW N/A true | |
1336 | * READ READ N/A true | |
1337 | * READ RW true false | |
1338 | * READ RW false true | |
b1f69b75 | 1339 | */ |
3cb2df17 | 1340 | static bool |
7dc0ac70 | 1341 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
c7d73af2 TH |
1342 | const struct pnfs_layout_range *range, |
1343 | bool strict_iomode) | |
b1f69b75 | 1344 | { |
fb3296eb BH |
1345 | struct pnfs_layout_range range1; |
1346 | ||
1347 | if ((range->iomode == IOMODE_RW && | |
1348 | ls_range->iomode != IOMODE_RW) || | |
c7d73af2 TH |
1349 | (range->iomode != ls_range->iomode && |
1350 | strict_iomode == true) || | |
7dc0ac70 | 1351 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1352 | return 0; |
1353 | ||
1354 | /* range1 covers only the first byte in the range */ | |
1355 | range1 = *range; | |
1356 | range1.length = 1; | |
7dc0ac70 | 1357 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1358 | } |
1359 | ||
1360 | /* | |
1361 | * lookup range in layout | |
1362 | */ | |
e5e94017 | 1363 | static struct pnfs_layout_segment * |
fb3296eb | 1364 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
c7d73af2 TH |
1365 | struct pnfs_layout_range *range, |
1366 | bool strict_iomode) | |
e5e94017 | 1367 | { |
b1f69b75 AA |
1368 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1369 | ||
1370 | dprintk("%s:Begin\n", __func__); | |
1371 | ||
b7edfaa1 | 1372 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1373 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1374 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
c7d73af2 TH |
1375 | pnfs_lseg_range_match(&lseg->pls_range, range, |
1376 | strict_iomode)) { | |
9369a431 | 1377 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1378 | break; |
1379 | } | |
b1f69b75 AA |
1380 | } |
1381 | ||
1382 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1383 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1384 | return ret; |
e5e94017 BH |
1385 | } |
1386 | ||
d23d61c8 AA |
1387 | /* |
1388 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1389 | * to the MDS or over pNFS | |
1390 | * | |
1391 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1392 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1393 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1394 | * WRITE request. | |
1395 | * | |
1396 | * A return of true means use MDS I/O. | |
1397 | * | |
1398 | * From rfc 5661: | |
1399 | * If a file's size is smaller than the file size threshold, data accesses | |
1400 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1401 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1402 | * server. If both file size and I/O size are provided, the client SHOULD | |
1403 | * reach or exceed both thresholds before sending its read or write | |
1404 | * requests to the data server. | |
1405 | */ | |
1406 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1407 | struct inode *ino, int iomode) | |
1408 | { | |
1409 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1410 | struct nfs_inode *nfsi = NFS_I(ino); | |
1411 | loff_t fsize = i_size_read(ino); | |
1412 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1413 | ||
1414 | if (t == NULL) | |
1415 | return ret; | |
1416 | ||
1417 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1418 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1419 | ||
1420 | switch (iomode) { | |
1421 | case IOMODE_READ: | |
1422 | if (t->bm & THRESHOLD_RD) { | |
1423 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1424 | size_set = true; | |
1425 | if (fsize < t->rd_sz) | |
1426 | size = true; | |
1427 | } | |
1428 | if (t->bm & THRESHOLD_RD_IO) { | |
1429 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1430 | nfsi->read_io); | |
1431 | io_set = true; | |
1432 | if (nfsi->read_io < t->rd_io_sz) | |
1433 | io = true; | |
1434 | } | |
1435 | break; | |
1436 | case IOMODE_RW: | |
1437 | if (t->bm & THRESHOLD_WR) { | |
1438 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1439 | size_set = true; | |
1440 | if (fsize < t->wr_sz) | |
1441 | size = true; | |
1442 | } | |
1443 | if (t->bm & THRESHOLD_WR_IO) { | |
1444 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1445 | nfsi->write_io); | |
1446 | io_set = true; | |
1447 | if (nfsi->write_io < t->wr_io_sz) | |
1448 | io = true; | |
1449 | } | |
1450 | break; | |
1451 | } | |
1452 | if (size_set && io_set) { | |
1453 | if (size && io) | |
1454 | ret = true; | |
1455 | } else if (size || io) | |
1456 | ret = true; | |
1457 | ||
1458 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1459 | return ret; | |
1460 | } | |
1461 | ||
aa8a45ee PT |
1462 | static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo) |
1463 | { | |
1464 | /* | |
1465 | * send layoutcommit as it can hold up layoutreturn due to lseg | |
1466 | * reference | |
1467 | */ | |
1468 | pnfs_layoutcommit_inode(lo->plh_inode, false); | |
1469 | return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
2e5b29f0 | 1470 | nfs_wait_bit_killable, |
aa8a45ee PT |
1471 | TASK_UNINTERRUPTIBLE); |
1472 | } | |
1473 | ||
d67ae825 TH |
1474 | static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo) |
1475 | { | |
1476 | unsigned long *bitlock = &lo->plh_flags; | |
1477 | ||
1478 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1479 | smp_mb__after_atomic(); | |
1480 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1481 | } | |
1482 | ||
e5e94017 BH |
1483 | /* |
1484 | * Layout segment is retreived from the server if not cached. | |
1485 | * The appropriate layout segment is referenced and returned to the caller. | |
1486 | */ | |
7c24d948 | 1487 | struct pnfs_layout_segment * |
e5e94017 BH |
1488 | pnfs_update_layout(struct inode *ino, |
1489 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1490 | loff_t pos, |
1491 | u64 count, | |
a75b9df9 | 1492 | enum pnfs_iomode iomode, |
c7d73af2 | 1493 | bool strict_iomode, |
a75b9df9 | 1494 | gfp_t gfp_flags) |
e5e94017 | 1495 | { |
fb3296eb BH |
1496 | struct pnfs_layout_range arg = { |
1497 | .iomode = iomode, | |
1498 | .offset = pos, | |
1499 | .length = count, | |
1500 | }; | |
183d9e7b | 1501 | unsigned pg_offset, seq; |
6382a441 WAA |
1502 | struct nfs_server *server = NFS_SERVER(ino); |
1503 | struct nfs_client *clp = server->nfs_client; | |
183d9e7b | 1504 | struct pnfs_layout_hdr *lo = NULL; |
e5e94017 | 1505 | struct pnfs_layout_segment *lseg = NULL; |
183d9e7b JL |
1506 | nfs4_stateid stateid; |
1507 | long timeout = 0; | |
1508 | unsigned long giveup = jiffies + rpc_get_timeout(server->client); | |
30005121 | 1509 | bool first; |
e5e94017 | 1510 | |
9a4bf31d | 1511 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) { |
183d9e7b | 1512 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1513 | PNFS_UPDATE_LAYOUT_NO_PNFS); |
f86bbcf8 | 1514 | goto out; |
9a4bf31d | 1515 | } |
d23d61c8 | 1516 | |
9a4bf31d | 1517 | if (iomode == IOMODE_READ && i_size_read(ino) == 0) { |
183d9e7b | 1518 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1519 | PNFS_UPDATE_LAYOUT_RD_ZEROLEN); |
4ae93560 | 1520 | goto out; |
9a4bf31d | 1521 | } |
4ae93560 | 1522 | |
9a4bf31d | 1523 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) { |
183d9e7b | 1524 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1525 | PNFS_UPDATE_LAYOUT_MDSTHRESH); |
f86bbcf8 | 1526 | goto out; |
9a4bf31d | 1527 | } |
d23d61c8 | 1528 | |
9bf87482 PT |
1529 | lookup_again: |
1530 | first = false; | |
e5e94017 | 1531 | spin_lock(&ino->i_lock); |
9fa40758 | 1532 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1533 | if (lo == NULL) { |
1534 | spin_unlock(&ino->i_lock); | |
183d9e7b | 1535 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1536 | PNFS_UPDATE_LAYOUT_NOMEM); |
830ffb56 TM |
1537 | goto out; |
1538 | } | |
e5e94017 | 1539 | |
43f1b3da | 1540 | /* Do we even need to bother with this? */ |
a59c30ac | 1541 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
183d9e7b | 1542 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1543 | PNFS_UPDATE_LAYOUT_BULK_RECALL); |
43f1b3da | 1544 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1545 | goto out_unlock; |
1546 | } | |
1547 | ||
1548 | /* if LAYOUTGET already failed once we don't try again */ | |
2e5b29f0 | 1549 | if (pnfs_layout_io_test_failed(lo, iomode)) { |
183d9e7b | 1550 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1551 | PNFS_UPDATE_LAYOUT_IO_TEST_FAIL); |
e5e94017 | 1552 | goto out_unlock; |
9a4bf31d | 1553 | } |
e5e94017 | 1554 | |
c7d73af2 | 1555 | lseg = pnfs_find_lseg(lo, &arg, strict_iomode); |
183d9e7b JL |
1556 | if (lseg) { |
1557 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1558 | PNFS_UPDATE_LAYOUT_FOUND_CACHED); | |
1559 | goto out_unlock; | |
1560 | } | |
1561 | ||
1562 | if (!nfs4_valid_open_stateid(ctx->state)) { | |
1563 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1564 | PNFS_UPDATE_LAYOUT_INVALID_OPEN); | |
1565 | goto out_unlock; | |
1566 | } | |
1567 | ||
1568 | /* | |
1569 | * Choose a stateid for the LAYOUTGET. If we don't have a layout | |
1570 | * stateid, or it has been invalidated, then we must use the open | |
1571 | * stateid. | |
1572 | */ | |
67a3b721 | 1573 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { |
183d9e7b JL |
1574 | |
1575 | /* | |
1576 | * The first layoutget for the file. Need to serialize per | |
9bf87482 PT |
1577 | * RFC 5661 Errata 3208. |
1578 | */ | |
1579 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1580 | &lo->plh_flags)) { | |
1581 | spin_unlock(&ino->i_lock); | |
1582 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1583 | TASK_UNINTERRUPTIBLE); | |
1584 | pnfs_put_layout_hdr(lo); | |
183d9e7b | 1585 | dprintk("%s retrying\n", __func__); |
9bf87482 PT |
1586 | goto lookup_again; |
1587 | } | |
183d9e7b JL |
1588 | |
1589 | first = true; | |
1590 | do { | |
1591 | seq = read_seqbegin(&ctx->state->seqlock); | |
1592 | nfs4_stateid_copy(&stateid, &ctx->state->stateid); | |
1593 | } while (read_seqretry(&ctx->state->seqlock, seq)); | |
9bf87482 | 1594 | } else { |
183d9e7b | 1595 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
9bf87482 | 1596 | } |
568e8c49 | 1597 | |
aa8a45ee PT |
1598 | /* |
1599 | * Because we free lsegs before sending LAYOUTRETURN, we need to wait | |
1600 | * for LAYOUTRETURN even if first is true. | |
1601 | */ | |
8f70f53a | 1602 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
aa8a45ee PT |
1603 | spin_unlock(&ino->i_lock); |
1604 | dprintk("%s wait for layoutreturn\n", __func__); | |
1605 | if (pnfs_prepare_to_retry_layoutget(lo)) { | |
d67ae825 TH |
1606 | if (first) |
1607 | pnfs_clear_first_layoutget(lo); | |
aa8a45ee PT |
1608 | pnfs_put_layout_hdr(lo); |
1609 | dprintk("%s retrying\n", __func__); | |
183d9e7b JL |
1610 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
1611 | lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
aa8a45ee PT |
1612 | goto lookup_again; |
1613 | } | |
183d9e7b | 1614 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1615 | PNFS_UPDATE_LAYOUT_RETURN); |
aa8a45ee PT |
1616 | goto out_put_layout_hdr; |
1617 | } | |
1618 | ||
9a4bf31d | 1619 | if (pnfs_layoutgets_blocked(lo)) { |
183d9e7b | 1620 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1621 | PNFS_UPDATE_LAYOUT_BLOCKED); |
cf7d63f1 | 1622 | goto out_unlock; |
9a4bf31d | 1623 | } |
cf7d63f1 | 1624 | atomic_inc(&lo->plh_outstanding); |
f49f9baa | 1625 | spin_unlock(&ino->i_lock); |
30005121 | 1626 | |
abb9a007 | 1627 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1628 | /* The lo must be on the clp list if there is any |
1629 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1630 | */ | |
1631 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1632 | if (list_empty(&lo->plh_layouts)) |
1633 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1634 | spin_unlock(&clp->cl_lock); |
1635 | } | |
e5e94017 | 1636 | |
09cbfeaf | 1637 | pg_offset = arg.offset & ~PAGE_MASK; |
707ed5fd BH |
1638 | if (pg_offset) { |
1639 | arg.offset -= pg_offset; | |
1640 | arg.length += pg_offset; | |
1641 | } | |
7c24d948 | 1642 | if (arg.length != NFS4_MAX_UINT64) |
09cbfeaf | 1643 | arg.length = PAGE_ALIGN(arg.length); |
707ed5fd | 1644 | |
183d9e7b JL |
1645 | lseg = send_layoutget(lo, ctx, &stateid, &arg, &timeout, gfp_flags); |
1646 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1647 | PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET); | |
83026d80 | 1648 | if (IS_ERR(lseg)) { |
183d9e7b JL |
1649 | switch(PTR_ERR(lseg)) { |
1650 | case -ERECALLCONFLICT: | |
1651 | if (time_after(jiffies, giveup)) | |
1652 | lseg = NULL; | |
1653 | /* Fallthrough */ | |
1654 | case -EAGAIN: | |
1655 | pnfs_put_layout_hdr(lo); | |
83026d80 JL |
1656 | if (first) |
1657 | pnfs_clear_first_layoutget(lo); | |
183d9e7b JL |
1658 | if (lseg) { |
1659 | trace_pnfs_update_layout(ino, pos, count, | |
1660 | iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
1661 | goto lookup_again; | |
1662 | } | |
1663 | /* Fallthrough */ | |
1664 | default: | |
1665 | if (!nfs_error_is_fatal(PTR_ERR(lseg))) { | |
1666 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
1667 | lseg = NULL; | |
1668 | } | |
83026d80 JL |
1669 | } |
1670 | } else { | |
1671 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
1672 | } | |
1673 | ||
cf7d63f1 | 1674 | atomic_dec(&lo->plh_outstanding); |
830ffb56 | 1675 | out_put_layout_hdr: |
d67ae825 TH |
1676 | if (first) |
1677 | pnfs_clear_first_layoutget(lo); | |
70c3bd2b | 1678 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1679 | out: |
f86bbcf8 TM |
1680 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1681 | "(%s, offset: %llu, length: %llu)\n", | |
1682 | __func__, ino->i_sb->s_id, | |
1683 | (unsigned long long)NFS_FILEID(ino), | |
d600ad1f | 1684 | IS_ERR_OR_NULL(lseg) ? "not found" : "found", |
f86bbcf8 TM |
1685 | iomode==IOMODE_RW ? "read/write" : "read-only", |
1686 | (unsigned long long)pos, | |
1687 | (unsigned long long)count); | |
e5e94017 BH |
1688 | return lseg; |
1689 | out_unlock: | |
1690 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1691 | goto out_put_layout_hdr; |
e5e94017 | 1692 | } |
7c24d948 | 1693 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1694 | |
540d9864 TM |
1695 | static bool |
1696 | pnfs_sanity_check_layout_range(struct pnfs_layout_range *range) | |
1697 | { | |
1698 | switch (range->iomode) { | |
1699 | case IOMODE_READ: | |
1700 | case IOMODE_RW: | |
1701 | break; | |
1702 | default: | |
1703 | return false; | |
1704 | } | |
1705 | if (range->offset == NFS4_MAX_UINT64) | |
1706 | return false; | |
1707 | if (range->length == 0) | |
1708 | return false; | |
1709 | if (range->length != NFS4_MAX_UINT64 && | |
1710 | range->length > NFS4_MAX_UINT64 - range->offset) | |
1711 | return false; | |
1712 | return true; | |
1713 | } | |
1714 | ||
a0b0a6e3 | 1715 | struct pnfs_layout_segment * |
b1f69b75 AA |
1716 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1717 | { | |
1718 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1719 | struct nfs4_layoutget_res *res = &lgp->res; | |
1720 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1721 | struct inode *ino = lo->plh_inode; |
78096cca | 1722 | LIST_HEAD(free_me); |
540d9864 TM |
1723 | |
1724 | if (!pnfs_sanity_check_layout_range(&res->range)) | |
1b3c6d07 | 1725 | return ERR_PTR(-EINVAL); |
b1f69b75 AA |
1726 | |
1727 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1728 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
1b3c6d07 | 1729 | if (IS_ERR_OR_NULL(lseg)) { |
b1f69b75 | 1730 | if (!lseg) |
1b3c6d07 JL |
1731 | lseg = ERR_PTR(-ENOMEM); |
1732 | ||
1733 | dprintk("%s: Could not allocate layout: error %ld\n", | |
1734 | __func__, PTR_ERR(lseg)); | |
1735 | return lseg; | |
b1f69b75 AA |
1736 | } |
1737 | ||
1013df61 CH |
1738 | init_lseg(lo, lseg); |
1739 | lseg->pls_range = res->range; | |
66755283 | 1740 | lseg->pls_seq = be32_to_cpu(res->stateid.seqid); |
1013df61 | 1741 | |
b1f69b75 | 1742 | spin_lock(&ino->i_lock); |
e1c06f80 | 1743 | if (pnfs_layoutgets_blocked(lo)) { |
43f1b3da | 1744 | dprintk("%s forget reply due to state\n", __func__); |
1b3c6d07 | 1745 | goto out_forget; |
43f1b3da | 1746 | } |
038d6493 | 1747 | |
362f7474 CH |
1748 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
1749 | /* existing state ID, make sure the sequence number matches. */ | |
1750 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1751 | dprintk("%s forget reply due to sequence\n", __func__); | |
1b3c6d07 | 1752 | goto out_forget; |
362f7474 CH |
1753 | } |
1754 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1755 | } else { | |
1756 | /* | |
1757 | * We got an entirely new state ID. Mark all segments for the | |
1758 | * inode invalid, and don't bother validating the stateid | |
1759 | * sequence number. | |
1760 | */ | |
6d597e17 | 1761 | pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL, 0); |
362f7474 CH |
1762 | |
1763 | nfs4_stateid_copy(&lo->plh_stateid, &res->stateid); | |
1764 | lo->plh_barrier = be32_to_cpu(res->stateid.seqid); | |
1765 | } | |
038d6493 | 1766 | |
47abadef CH |
1767 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); |
1768 | ||
9369a431 | 1769 | pnfs_get_lseg(lseg); |
03772d2f | 1770 | pnfs_layout_insert_lseg(lo, lseg, &free_me); |
b1f69b75 | 1771 | |
3976143b | 1772 | if (res->return_on_close) |
f7e8917a | 1773 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); |
f7e8917a | 1774 | |
b1f69b75 | 1775 | spin_unlock(&ino->i_lock); |
78096cca | 1776 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1777 | return lseg; |
43f1b3da | 1778 | |
1b3c6d07 | 1779 | out_forget: |
43f1b3da FI |
1780 | spin_unlock(&ino->i_lock); |
1781 | lseg->pls_layout = lo; | |
1782 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1b3c6d07 | 1783 | return ERR_PTR(-EAGAIN); |
b1f69b75 AA |
1784 | } |
1785 | ||
016256df | 1786 | static void |
3982a6a2 JL |
1787 | pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode, |
1788 | u32 seq) | |
5c97f5de TM |
1789 | { |
1790 | if (lo->plh_return_iomode == iomode) | |
1791 | return; | |
1792 | if (lo->plh_return_iomode != 0) | |
1793 | iomode = IOMODE_ANY; | |
1794 | lo->plh_return_iomode = iomode; | |
e0fa0d01 | 1795 | set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); |
3982a6a2 JL |
1796 | if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq)) |
1797 | lo->plh_return_seq = seq; | |
5c97f5de TM |
1798 | } |
1799 | ||
2f215968 TM |
1800 | /** |
1801 | * pnfs_mark_matching_lsegs_return - Free or return matching layout segments | |
1802 | * @lo: pointer to layout header | |
1803 | * @tmp_list: list header to be used with pnfs_free_lseg_list() | |
1804 | * @return_range: describe layout segment ranges to be returned | |
1805 | * | |
1806 | * This function is mainly intended for use by layoutrecall. It attempts | |
1807 | * to free the layout segment immediately, or else to mark it for return | |
1808 | * as soon as its reference count drops to zero. | |
1809 | */ | |
10335556 | 1810 | int |
016256df PT |
1811 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, |
1812 | struct list_head *tmp_list, | |
6d597e17 JL |
1813 | const struct pnfs_layout_range *return_range, |
1814 | u32 seq) | |
016256df PT |
1815 | { |
1816 | struct pnfs_layout_segment *lseg, *next; | |
10335556 | 1817 | int remaining = 0; |
016256df PT |
1818 | |
1819 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
1820 | ||
1821 | if (list_empty(&lo->plh_segs)) | |
10335556 | 1822 | return 0; |
016256df | 1823 | |
fc7ff367 | 1824 | assert_spin_locked(&lo->plh_inode->i_lock); |
016256df PT |
1825 | |
1826 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
1827 | if (should_free_lseg(&lseg->pls_range, return_range)) { | |
1828 | dprintk("%s: marking lseg %p iomode %d " | |
1829 | "offset %llu length %llu\n", __func__, | |
1830 | lseg, lseg->pls_range.iomode, | |
1831 | lseg->pls_range.offset, | |
1832 | lseg->pls_range.length); | |
2f215968 TM |
1833 | if (mark_lseg_invalid(lseg, tmp_list)) |
1834 | continue; | |
1835 | remaining++; | |
016256df | 1836 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); |
016256df | 1837 | } |
6d597e17 JL |
1838 | |
1839 | if (remaining) | |
1840 | pnfs_set_plh_return_info(lo, return_range->iomode, seq); | |
1841 | ||
10335556 | 1842 | return remaining; |
016256df PT |
1843 | } |
1844 | ||
1845 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
1846 | struct pnfs_layout_segment *lseg) | |
1847 | { | |
1848 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
016256df PT |
1849 | struct pnfs_layout_range range = { |
1850 | .iomode = lseg->pls_range.iomode, | |
1851 | .offset = 0, | |
1852 | .length = NFS4_MAX_UINT64, | |
1853 | }; | |
1854 | LIST_HEAD(free_me); | |
10335556 | 1855 | bool return_now = false; |
016256df PT |
1856 | |
1857 | spin_lock(&inode->i_lock); | |
3982a6a2 | 1858 | pnfs_set_plh_return_info(lo, range.iomode, lseg->pls_seq); |
016256df PT |
1859 | /* |
1860 | * mark all matching lsegs so that we are sure to have no live | |
1861 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
1862 | * for how it works. | |
1863 | */ | |
6d597e17 JL |
1864 | if (!pnfs_mark_matching_lsegs_return(lo, &free_me, |
1865 | &range, lseg->pls_seq)) { | |
10335556 TM |
1866 | nfs4_stateid stateid; |
1867 | enum pnfs_iomode iomode = lo->plh_return_iomode; | |
1868 | ||
1869 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); | |
1870 | return_now = pnfs_prepare_layoutreturn(lo); | |
1871 | spin_unlock(&inode->i_lock); | |
1872 | if (return_now) | |
1873 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); | |
1874 | } else { | |
1875 | spin_unlock(&inode->i_lock); | |
1876 | nfs_commit_inode(inode, 0); | |
1877 | } | |
016256df PT |
1878 | pnfs_free_lseg_list(&free_me); |
1879 | } | |
1880 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
1881 | ||
d8007d4d TM |
1882 | void |
1883 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1884 | { | |
1fd937bd PT |
1885 | u64 rd_size = req->wb_bytes; |
1886 | ||
cb5d04bc PT |
1887 | if (pgio->pg_lseg == NULL) { |
1888 | if (pgio->pg_dreq == NULL) | |
1889 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
1890 | else | |
1891 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
1892 | ||
1893 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
1894 | req->wb_context, | |
1895 | req_offset(req), | |
1896 | rd_size, | |
1897 | IOMODE_READ, | |
c7d73af2 | 1898 | false, |
cb5d04bc | 1899 | GFP_KERNEL); |
d600ad1f PT |
1900 | if (IS_ERR(pgio->pg_lseg)) { |
1901 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
1902 | pgio->pg_lseg = NULL; | |
1903 | return; | |
1904 | } | |
cb5d04bc | 1905 | } |
e885de1a TM |
1906 | /* If no lseg, fall back to read through mds */ |
1907 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1908 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1909 | |
d8007d4d TM |
1910 | } |
1911 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1912 | ||
1913 | void | |
6296556f PT |
1914 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
1915 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 1916 | { |
d600ad1f | 1917 | if (pgio->pg_lseg == NULL) { |
cb5d04bc PT |
1918 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1919 | req->wb_context, | |
1920 | req_offset(req), | |
1921 | wb_size, | |
1922 | IOMODE_RW, | |
c7d73af2 | 1923 | false, |
cb5d04bc | 1924 | GFP_NOFS); |
d600ad1f PT |
1925 | if (IS_ERR(pgio->pg_lseg)) { |
1926 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
1927 | pgio->pg_lseg = NULL; | |
1928 | return; | |
1929 | } | |
1930 | } | |
e885de1a TM |
1931 | /* If no lseg, fall back to write through mds */ |
1932 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1933 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1934 | } |
1935 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1936 | ||
180bb5ec WAA |
1937 | void |
1938 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
1939 | { | |
1940 | if (desc->pg_lseg) { | |
1941 | pnfs_put_lseg(desc->pg_lseg); | |
1942 | desc->pg_lseg = NULL; | |
1943 | } | |
1944 | } | |
1945 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
1946 | ||
b4fdac1a WAA |
1947 | /* |
1948 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
1949 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
1950 | */ | |
1951 | size_t | |
a7d42ddb WAA |
1952 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
1953 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 1954 | { |
0f9c429e | 1955 | unsigned int size; |
c5e20cb7 | 1956 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
1957 | |
1958 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
1959 | if (!size) |
1960 | return 0; | |
94ad1c80 | 1961 | |
19982ba8 | 1962 | /* |
c5e20cb7 WAA |
1963 | * 'size' contains the number of bytes left in the current page (up |
1964 | * to the original size asked for in @req->wb_bytes). | |
1965 | * | |
1966 | * Calculate how many bytes are left in the layout segment | |
1967 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
1968 | * |
1969 | * Please also note that 'end_offset' is actually the offset of the | |
1970 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1971 | * | |
1972 | */ | |
19b54848 | 1973 | if (pgio->pg_lseg) { |
c5e20cb7 WAA |
1974 | seg_end = end_offset(pgio->pg_lseg->pls_range.offset, |
1975 | pgio->pg_lseg->pls_range.length); | |
1976 | req_start = req_offset(req); | |
7c13789e | 1977 | WARN_ON_ONCE(req_start >= seg_end); |
c5e20cb7 | 1978 | /* start of request is past the last byte of this segment */ |
7c13789e WAA |
1979 | if (req_start >= seg_end) { |
1980 | /* reference the new lseg */ | |
1981 | if (pgio->pg_ops->pg_cleanup) | |
1982 | pgio->pg_ops->pg_cleanup(pgio); | |
1983 | if (pgio->pg_ops->pg_init) | |
1984 | pgio->pg_ops->pg_init(pgio, req); | |
19b54848 | 1985 | return 0; |
7c13789e | 1986 | } |
c5e20cb7 WAA |
1987 | |
1988 | /* adjust 'size' iff there are fewer bytes left in the | |
1989 | * segment than what nfs_generic_pg_test returned */ | |
1990 | seg_left = seg_end - req_start; | |
1991 | if (seg_left < size) | |
1992 | size = (unsigned int)seg_left; | |
19b54848 | 1993 | } |
0f9c429e | 1994 | |
19b54848 | 1995 | return size; |
94ad1c80 | 1996 | } |
89a58e32 | 1997 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1998 | |
53113ad3 | 1999 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
2000 | { |
2001 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
2002 | |
2003 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
2004 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
2005 | hdr->completion_ops); | |
c7070113 | 2006 | set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags); |
53113ad3 | 2007 | return nfs_pageio_resend(&pgio, hdr); |
e2fecb21 | 2008 | } |
e7dd79af | 2009 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 2010 | |
d45f60c6 | 2011 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2012 | { |
cd841605 FI |
2013 | |
2014 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
2015 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2016 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2017 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2018 | } |
6c75dc0d | 2019 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2020 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
2021 | } |
2022 | ||
d20581aa BH |
2023 | /* |
2024 | * Called by non rpc-based layout drivers | |
2025 | */ | |
d45f60c6 | 2026 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 2027 | { |
f8417b48 | 2028 | if (likely(!hdr->pnfs_error)) { |
67af7611 TM |
2029 | pnfs_set_layoutcommit(hdr->inode, hdr->lseg, |
2030 | hdr->mds_offset + hdr->res.count); | |
d45f60c6 | 2031 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2032 | } |
2033 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); | |
2034 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2035 | pnfs_ld_handle_write_error(hdr); |
2036 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 2037 | } |
d20581aa | 2038 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 2039 | |
dce81290 TM |
2040 | static void |
2041 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2042 | struct nfs_pgio_header *hdr) |
dce81290 | 2043 | { |
48d635f1 | 2044 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2045 | |
6c75dc0d | 2046 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2047 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 2048 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 2049 | mirror->pg_recoalesce = 1; |
6c75dc0d | 2050 | } |
d45f60c6 | 2051 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2052 | hdr->release(hdr); |
dce81290 TM |
2053 | } |
2054 | ||
2055 | static enum pnfs_try_status | |
d45f60c6 | 2056 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
2057 | const struct rpc_call_ops *call_ops, |
2058 | struct pnfs_layout_segment *lseg, | |
2059 | int how) | |
0382b744 | 2060 | { |
cd841605 | 2061 | struct inode *inode = hdr->inode; |
0382b744 AA |
2062 | enum pnfs_try_status trypnfs; |
2063 | struct nfs_server *nfss = NFS_SERVER(inode); | |
2064 | ||
cd841605 | 2065 | hdr->mds_ops = call_ops; |
0382b744 AA |
2066 | |
2067 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
2068 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
2069 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 2070 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 2071 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
2072 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2073 | return trypnfs; | |
2074 | } | |
2075 | ||
dce81290 | 2076 | static void |
7f714720 WAA |
2077 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
2078 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 2079 | { |
dce81290 TM |
2080 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2081 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2082 | enum pnfs_try_status trypnfs; |
dce81290 | 2083 | |
d45f60c6 | 2084 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 2085 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 2086 | pnfs_write_through_mds(desc, hdr); |
dce81290 TM |
2087 | } |
2088 | ||
6c75dc0d FI |
2089 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
2090 | { | |
9369a431 | 2091 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2092 | nfs_pgio_header_free(hdr); |
6c75dc0d FI |
2093 | } |
2094 | ||
dce81290 TM |
2095 | int |
2096 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
2097 | { | |
6c75dc0d | 2098 | struct nfs_pgio_header *hdr; |
dce81290 TM |
2099 | int ret; |
2100 | ||
1e7f3a48 WAA |
2101 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2102 | if (!hdr) { | |
2bff2288 PT |
2103 | desc->pg_error = -ENOMEM; |
2104 | return desc->pg_error; | |
dce81290 | 2105 | } |
6c75dc0d | 2106 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 2107 | |
9369a431 | 2108 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2109 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2110 | if (!ret) |
7f714720 | 2111 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 2112 | |
6c75dc0d | 2113 | return ret; |
dce81290 TM |
2114 | } |
2115 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
2116 | ||
53113ad3 | 2117 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
2118 | { |
2119 | struct nfs_pageio_descriptor pgio; | |
2120 | ||
1acbbb4e | 2121 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
2122 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
2123 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 2124 | } |
e7dd79af | 2125 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 2126 | |
d45f60c6 | 2127 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2128 | { |
cd841605 FI |
2129 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
2130 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2131 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2132 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2133 | } |
4db6e0b7 | 2134 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2135 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
2136 | } |
2137 | ||
d20581aa BH |
2138 | /* |
2139 | * Called by non rpc-based layout drivers | |
2140 | */ | |
d45f60c6 | 2141 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 2142 | { |
cd841605 | 2143 | if (likely(!hdr->pnfs_error)) { |
d45f60c6 WAA |
2144 | __nfs4_read_done_cb(hdr); |
2145 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); | |
f8417b48 KM |
2146 | } |
2147 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); | |
2148 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2149 | pnfs_ld_handle_read_error(hdr); |
2150 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
2151 | } |
2152 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
2153 | ||
493292dd TM |
2154 | static void |
2155 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2156 | struct nfs_pgio_header *hdr) |
493292dd | 2157 | { |
48d635f1 | 2158 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2159 | |
4db6e0b7 | 2160 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2161 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 2162 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 2163 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 2164 | } |
d45f60c6 | 2165 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2166 | hdr->release(hdr); |
493292dd TM |
2167 | } |
2168 | ||
64419a9b AA |
2169 | /* |
2170 | * Call the appropriate parallel I/O subsystem read function. | |
2171 | */ | |
493292dd | 2172 | static enum pnfs_try_status |
d45f60c6 | 2173 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
2174 | const struct rpc_call_ops *call_ops, |
2175 | struct pnfs_layout_segment *lseg) | |
64419a9b | 2176 | { |
cd841605 | 2177 | struct inode *inode = hdr->inode; |
64419a9b AA |
2178 | struct nfs_server *nfss = NFS_SERVER(inode); |
2179 | enum pnfs_try_status trypnfs; | |
2180 | ||
cd841605 | 2181 | hdr->mds_ops = call_ops; |
64419a9b AA |
2182 | |
2183 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 2184 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 2185 | |
d45f60c6 | 2186 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 2187 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 2188 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
2189 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2190 | return trypnfs; | |
2191 | } | |
863a3c6c | 2192 | |
ceb11e13 | 2193 | /* Resend all requests through pnfs. */ |
1b1bc66b | 2194 | void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) |
ceb11e13 PT |
2195 | { |
2196 | struct nfs_pageio_descriptor pgio; | |
2197 | ||
1b1bc66b WAA |
2198 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
2199 | nfs_pageio_init_read(&pgio, hdr->inode, false, | |
2200 | hdr->completion_ops); | |
2201 | hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr); | |
2202 | } | |
ceb11e13 PT |
2203 | } |
2204 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
2205 | ||
493292dd | 2206 | static void |
7f714720 | 2207 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 2208 | { |
493292dd TM |
2209 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2210 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2211 | enum pnfs_try_status trypnfs; |
493292dd | 2212 | |
d45f60c6 | 2213 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
ceb11e13 | 2214 | if (trypnfs == PNFS_TRY_AGAIN) |
1b1bc66b WAA |
2215 | pnfs_read_resend_pnfs(hdr); |
2216 | if (trypnfs == PNFS_NOT_ATTEMPTED || hdr->task.tk_status) | |
d45f60c6 | 2217 | pnfs_read_through_mds(desc, hdr); |
493292dd TM |
2218 | } |
2219 | ||
4db6e0b7 FI |
2220 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
2221 | { | |
9369a431 | 2222 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2223 | nfs_pgio_header_free(hdr); |
4db6e0b7 FI |
2224 | } |
2225 | ||
493292dd TM |
2226 | int |
2227 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
2228 | { | |
4db6e0b7 | 2229 | struct nfs_pgio_header *hdr; |
493292dd TM |
2230 | int ret; |
2231 | ||
1e7f3a48 WAA |
2232 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2233 | if (!hdr) { | |
2bff2288 PT |
2234 | desc->pg_error = -ENOMEM; |
2235 | return desc->pg_error; | |
493292dd | 2236 | } |
4db6e0b7 | 2237 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 2238 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2239 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2240 | if (!ret) |
7f714720 | 2241 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 2242 | return ret; |
493292dd TM |
2243 | } |
2244 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
2245 | ||
71244d9b TM |
2246 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
2247 | { | |
2248 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
2249 | ||
2250 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 2251 | smp_mb__after_atomic(); |
71244d9b TM |
2252 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
2253 | } | |
2254 | ||
863a3c6c | 2255 | /* |
a9bae566 | 2256 | * There can be multiple RW segments. |
863a3c6c | 2257 | */ |
a9bae566 | 2258 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 2259 | { |
a9bae566 | 2260 | struct pnfs_layout_segment *lseg; |
863a3c6c | 2261 | |
a9bae566 PT |
2262 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
2263 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 2264 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
2265 | list_add(&lseg->pls_lc_list, listp); |
2266 | } | |
863a3c6c AA |
2267 | } |
2268 | ||
a073dbff TM |
2269 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
2270 | { | |
2271 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
2272 | |
2273 | /* Matched by references in pnfs_set_layoutcommit */ | |
2274 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
2275 | list_del_init(&lseg->pls_lc_list); | |
2276 | pnfs_put_lseg(lseg); | |
2277 | } | |
2278 | ||
71244d9b | 2279 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
2280 | } |
2281 | ||
1b0ae068 PT |
2282 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
2283 | { | |
b9e028fd | 2284 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
2285 | } |
2286 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
2287 | ||
863a3c6c | 2288 | void |
67af7611 TM |
2289 | pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, |
2290 | loff_t end_pos) | |
863a3c6c | 2291 | { |
cd841605 | 2292 | struct nfs_inode *nfsi = NFS_I(inode); |
79a48a1f | 2293 | bool mark_as_dirty = false; |
863a3c6c | 2294 | |
cd841605 | 2295 | spin_lock(&inode->i_lock); |
863a3c6c | 2296 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
29559b11 | 2297 | nfsi->layout->plh_lwb = end_pos; |
79a48a1f | 2298 | mark_as_dirty = true; |
863a3c6c | 2299 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 2300 | __func__, inode->i_ino); |
29559b11 TM |
2301 | } else if (end_pos > nfsi->layout->plh_lwb) |
2302 | nfsi->layout->plh_lwb = end_pos; | |
67af7611 | 2303 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) { |
a9bae566 | 2304 | /* references matched in nfs4_layoutcommit_release */ |
67af7611 | 2305 | pnfs_get_lseg(lseg); |
a9bae566 | 2306 | } |
cd841605 | 2307 | spin_unlock(&inode->i_lock); |
acff5880 | 2308 | dprintk("%s: lseg %p end_pos %llu\n", |
67af7611 | 2309 | __func__, lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
2310 | |
2311 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2312 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2313 | if (mark_as_dirty) | |
cd841605 | 2314 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
2315 | } |
2316 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
2317 | ||
db29c089 AA |
2318 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2319 | { | |
2320 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2321 | ||
2322 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2323 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2324 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2325 | } |
2326 | ||
de4b15c7 AA |
2327 | /* |
2328 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2329 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2330 | * data to disk to allow the server to recover the data if it crashes. | |
2331 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2332 | * is off, and a COMMIT is sent to a data server, or | |
2333 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2334 | */ | |
863a3c6c | 2335 | int |
ef311537 | 2336 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2337 | { |
5f919c9f | 2338 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2339 | struct nfs4_layoutcommit_data *data; |
2340 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2341 | loff_t end_pos; |
71244d9b | 2342 | int status; |
863a3c6c | 2343 | |
71244d9b | 2344 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2345 | return 0; |
2346 | ||
71244d9b | 2347 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2348 | |
71244d9b | 2349 | status = -EAGAIN; |
92407e75 | 2350 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2351 | if (!sync) |
2352 | goto out; | |
74316201 | 2353 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2354 | NFS_INO_LAYOUTCOMMITTING, |
2355 | nfs_wait_bit_killable, | |
2356 | TASK_KILLABLE); | |
92407e75 | 2357 | if (status) |
71244d9b | 2358 | goto out; |
92407e75 PT |
2359 | } |
2360 | ||
71244d9b TM |
2361 | status = -ENOMEM; |
2362 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2363 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2364 | if (!data) | |
2365 | goto clear_layoutcommitting; | |
2366 | ||
2367 | status = 0; | |
de4b15c7 | 2368 | spin_lock(&inode->i_lock); |
71244d9b TM |
2369 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2370 | goto out_unlock; | |
a9bae566 | 2371 | |
71244d9b | 2372 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2373 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2374 | |
acff5880 | 2375 | end_pos = nfsi->layout->plh_lwb; |
863a3c6c | 2376 | |
f597c537 | 2377 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2378 | spin_unlock(&inode->i_lock); |
2379 | ||
2380 | data->args.inode = inode; | |
9fa40758 | 2381 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2382 | nfs_fattr_init(&data->fattr); |
2383 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2384 | data->res.fattr = &data->fattr; | |
2385 | data->args.lastbytewritten = end_pos - 1; | |
2386 | data->res.server = NFS_SERVER(inode); | |
2387 | ||
5f919c9f CH |
2388 | if (ld->prepare_layoutcommit) { |
2389 | status = ld->prepare_layoutcommit(&data->args); | |
2390 | if (status) { | |
3471648a | 2391 | put_rpccred(data->cred); |
5f919c9f | 2392 | spin_lock(&inode->i_lock); |
29559b11 TM |
2393 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); |
2394 | if (end_pos > nfsi->layout->plh_lwb) | |
5f919c9f | 2395 | nfsi->layout->plh_lwb = end_pos; |
3471648a | 2396 | goto out_unlock; |
5f919c9f CH |
2397 | } |
2398 | } | |
2399 | ||
2400 | ||
863a3c6c AA |
2401 | status = nfs4_proc_layoutcommit(data, sync); |
2402 | out: | |
92407e75 PT |
2403 | if (status) |
2404 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2405 | dprintk("<-- %s status %d\n", __func__, status); |
2406 | return status; | |
71244d9b TM |
2407 | out_unlock: |
2408 | spin_unlock(&inode->i_lock); | |
92407e75 | 2409 | kfree(data); |
71244d9b TM |
2410 | clear_layoutcommitting: |
2411 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2412 | goto out; |
863a3c6c | 2413 | } |
72cff449 | 2414 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a | 2415 | |
5bb89b47 TM |
2416 | int |
2417 | pnfs_generic_sync(struct inode *inode, bool datasync) | |
2418 | { | |
2419 | return pnfs_layoutcommit_inode(inode, true); | |
2420 | } | |
2421 | EXPORT_SYMBOL_GPL(pnfs_generic_sync); | |
2422 | ||
82be417a AA |
2423 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) |
2424 | { | |
2425 | struct nfs4_threshold *thp; | |
2426 | ||
2427 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2428 | if (!thp) { | |
2429 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2430 | return NULL; | |
2431 | } | |
2432 | return thp; | |
2433 | } | |
8733408d | 2434 | |
865a7ecb | 2435 | #if IS_ENABLED(CONFIG_NFS_V4_2) |
8733408d | 2436 | int |
c8ad8894 | 2437 | pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags) |
8733408d PT |
2438 | { |
2439 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
2440 | struct nfs_server *server = NFS_SERVER(inode); | |
1bfe3b25 | 2441 | struct nfs_inode *nfsi = NFS_I(inode); |
8733408d PT |
2442 | struct nfs42_layoutstat_data *data; |
2443 | struct pnfs_layout_hdr *hdr; | |
2444 | int status = 0; | |
2445 | ||
2446 | if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats) | |
2447 | goto out; | |
2448 | ||
6c5a0d89 TM |
2449 | if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS)) |
2450 | goto out; | |
2451 | ||
1bfe3b25 PT |
2452 | if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags)) |
2453 | goto out; | |
2454 | ||
8733408d PT |
2455 | spin_lock(&inode->i_lock); |
2456 | if (!NFS_I(inode)->layout) { | |
2457 | spin_unlock(&inode->i_lock); | |
f538d0ba | 2458 | goto out_clear_layoutstats; |
8733408d PT |
2459 | } |
2460 | hdr = NFS_I(inode)->layout; | |
2461 | pnfs_get_layout_hdr(hdr); | |
2462 | spin_unlock(&inode->i_lock); | |
2463 | ||
c8ad8894 | 2464 | data = kzalloc(sizeof(*data), gfp_flags); |
8733408d PT |
2465 | if (!data) { |
2466 | status = -ENOMEM; | |
2467 | goto out_put; | |
2468 | } | |
2469 | ||
2470 | data->args.fh = NFS_FH(inode); | |
2471 | data->args.inode = inode; | |
2472 | nfs4_stateid_copy(&data->args.stateid, &hdr->plh_stateid); | |
2473 | status = ld->prepare_layoutstats(&data->args); | |
2474 | if (status) | |
2475 | goto out_free; | |
2476 | ||
2477 | status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data); | |
2478 | ||
2479 | out: | |
2480 | dprintk("%s returns %d\n", __func__, status); | |
2481 | return status; | |
2482 | ||
2483 | out_free: | |
2484 | kfree(data); | |
2485 | out_put: | |
2486 | pnfs_put_layout_hdr(hdr); | |
f538d0ba | 2487 | out_clear_layoutstats: |
1bfe3b25 PT |
2488 | smp_mb__before_atomic(); |
2489 | clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags); | |
2490 | smp_mb__after_atomic(); | |
8733408d PT |
2491 | goto out; |
2492 | } | |
2493 | EXPORT_SYMBOL_GPL(pnfs_report_layoutstat); | |
865a7ecb | 2494 | #endif |
bbf58bf3 TM |
2495 | |
2496 | unsigned int layoutstats_timer; | |
2497 | module_param(layoutstats_timer, uint, 0644); | |
2498 | EXPORT_SYMBOL_GPL(layoutstats_timer); |