]>
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> |
ca440c38 | 33 | #include <linux/sort.h> |
974cec8c | 34 | #include "internal.h" |
85e174ba | 35 | #include "pnfs.h" |
64419a9b | 36 | #include "iostat.h" |
cc668ab3 | 37 | #include "nfs4trace.h" |
40dd4b7a | 38 | #include "delegation.h" |
8733408d | 39 | #include "nfs42.h" |
85e174ba RL |
40 | |
41 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
25c75333 | 42 | #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) |
85e174ba | 43 | |
02c35fca FI |
44 | /* Locking: |
45 | * | |
46 | * pnfs_spinlock: | |
47 | * protects pnfs_modules_tbl. | |
48 | */ | |
49 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
50 | ||
51 | /* | |
52 | * pnfs_modules_tbl holds all pnfs modules | |
53 | */ | |
54 | static LIST_HEAD(pnfs_modules_tbl); | |
55 | ||
13c13a6a | 56 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo); |
68f74479 TM |
57 | static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo, |
58 | struct list_head *free_me, | |
59 | const struct pnfs_layout_range *range, | |
60 | u32 seq); | |
aa1e0e3a | 61 | |
02c35fca FI |
62 | /* Return the registered pnfs layout driver module matching given id */ |
63 | static struct pnfs_layoutdriver_type * | |
64 | find_pnfs_driver_locked(u32 id) | |
65 | { | |
66 | struct pnfs_layoutdriver_type *local; | |
67 | ||
68 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
69 | if (local->id == id) | |
70 | goto out; | |
71 | local = NULL; | |
72 | out: | |
73 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
74 | return local; | |
75 | } | |
76 | ||
85e174ba RL |
77 | static struct pnfs_layoutdriver_type * |
78 | find_pnfs_driver(u32 id) | |
79 | { | |
02c35fca FI |
80 | struct pnfs_layoutdriver_type *local; |
81 | ||
82 | spin_lock(&pnfs_spinlock); | |
83 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
84 | if (local != NULL && !try_module_get(local->owner)) { |
85 | dprintk("%s: Could not grab reference on module\n", __func__); | |
86 | local = NULL; | |
87 | } | |
02c35fca FI |
88 | spin_unlock(&pnfs_spinlock); |
89 | return local; | |
85e174ba RL |
90 | } |
91 | ||
92 | void | |
93 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
94 | { | |
738fd0f3 BH |
95 | if (nfss->pnfs_curr_ld) { |
96 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
97 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
98 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
99 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
100 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 101 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 102 | } |
85e174ba RL |
103 | nfss->pnfs_curr_ld = NULL; |
104 | } | |
105 | ||
ca440c38 JL |
106 | /* |
107 | * When the server sends a list of layout types, we choose one in the order | |
108 | * given in the list below. | |
109 | * | |
110 | * FIXME: should this list be configurable in some fashion? module param? | |
111 | * mount option? something else? | |
112 | */ | |
113 | static const u32 ld_prefs[] = { | |
114 | LAYOUT_SCSI, | |
115 | LAYOUT_BLOCK_VOLUME, | |
116 | LAYOUT_OSD2_OBJECTS, | |
117 | LAYOUT_FLEX_FILES, | |
118 | LAYOUT_NFSV4_1_FILES, | |
119 | 0 | |
120 | }; | |
121 | ||
122 | static int | |
123 | ld_cmp(const void *e1, const void *e2) | |
124 | { | |
125 | u32 ld1 = *((u32 *)e1); | |
126 | u32 ld2 = *((u32 *)e2); | |
127 | int i; | |
128 | ||
129 | for (i = 0; ld_prefs[i] != 0; i++) { | |
130 | if (ld1 == ld_prefs[i]) | |
131 | return -1; | |
132 | ||
133 | if (ld2 == ld_prefs[i]) | |
134 | return 1; | |
135 | } | |
136 | return 0; | |
137 | } | |
138 | ||
85e174ba RL |
139 | /* |
140 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
141 | * Currently only one pNFS layout driver per filesystem is supported. | |
142 | * | |
3132e49e | 143 | * @ids array of layout types supported by MDS. |
85e174ba RL |
144 | */ |
145 | void | |
738fd0f3 | 146 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
ca440c38 | 147 | struct nfs_fsinfo *fsinfo) |
85e174ba RL |
148 | { |
149 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
3132e49e | 150 | u32 id; |
ca440c38 | 151 | int i; |
85e174ba | 152 | |
19274716 AS |
153 | if (fsinfo->nlayouttypes == 0) |
154 | goto out_no_driver; | |
85e174ba RL |
155 | if (!(server->nfs_client->cl_exchange_flags & |
156 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
3132e49e JL |
157 | printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n", |
158 | __func__, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
159 | goto out_no_driver; |
160 | } | |
3132e49e | 161 | |
ca440c38 JL |
162 | sort(fsinfo->layouttype, fsinfo->nlayouttypes, |
163 | sizeof(*fsinfo->layouttype), ld_cmp, NULL); | |
3132e49e | 164 | |
ca440c38 JL |
165 | for (i = 0; i < fsinfo->nlayouttypes; i++) { |
166 | id = fsinfo->layouttype[i]; | |
85e174ba | 167 | ld_type = find_pnfs_driver(id); |
ca440c38 JL |
168 | if (!ld_type) { |
169 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, | |
170 | id); | |
171 | ld_type = find_pnfs_driver(id); | |
172 | } | |
173 | if (ld_type) | |
174 | break; | |
85e174ba | 175 | } |
3132e49e JL |
176 | |
177 | if (!ld_type) { | |
ca440c38 | 178 | dprintk("%s: No pNFS module found!\n", __func__); |
3132e49e JL |
179 | goto out_no_driver; |
180 | } | |
181 | ||
85e174ba | 182 | server->pnfs_curr_ld = ld_type; |
738fd0f3 BH |
183 | if (ld_type->set_layoutdriver |
184 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
185 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
186 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
187 | module_put(ld_type->owner); |
188 | goto out_no_driver; | |
189 | } | |
2a4c8994 TM |
190 | /* Bump the MDS count */ |
191 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 192 | |
85e174ba RL |
193 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
194 | return; | |
195 | ||
196 | out_no_driver: | |
197 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
198 | server->pnfs_curr_ld = NULL; | |
199 | } | |
02c35fca FI |
200 | |
201 | int | |
202 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
203 | { | |
204 | int status = -EINVAL; | |
205 | struct pnfs_layoutdriver_type *tmp; | |
206 | ||
207 | if (ld_type->id == 0) { | |
a030889a | 208 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
209 | return status; |
210 | } | |
b1f69b75 | 211 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 212 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
213 | "alloc_lseg and free_lseg.\n", __func__); |
214 | return status; | |
215 | } | |
02c35fca FI |
216 | |
217 | spin_lock(&pnfs_spinlock); | |
218 | tmp = find_pnfs_driver_locked(ld_type->id); | |
219 | if (!tmp) { | |
220 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
221 | status = 0; | |
222 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
223 | ld_type->name); | |
224 | } else { | |
a030889a | 225 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
226 | __func__, ld_type->id); |
227 | } | |
228 | spin_unlock(&pnfs_spinlock); | |
229 | ||
230 | return status; | |
231 | } | |
232 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
233 | ||
234 | void | |
235 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
236 | { | |
237 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
238 | spin_lock(&pnfs_spinlock); | |
239 | list_del(&ld_type->pnfs_tblid); | |
240 | spin_unlock(&pnfs_spinlock); | |
241 | } | |
242 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 243 | |
b1f69b75 AA |
244 | /* |
245 | * pNFS client layout cache | |
246 | */ | |
247 | ||
cc6e5340 | 248 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 249 | void |
70c3bd2b | 250 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 251 | { |
cc6e5340 | 252 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
253 | } |
254 | ||
636fb9c8 BH |
255 | static struct pnfs_layout_hdr * |
256 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
257 | { | |
258 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
57934278 | 259 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
260 | } |
261 | ||
262 | static void | |
263 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
264 | { | |
9c626381 TM |
265 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
266 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
267 | ||
268 | if (!list_empty(&lo->plh_layouts)) { | |
269 | struct nfs_client *clp = server->nfs_client; | |
270 | ||
271 | spin_lock(&clp->cl_lock); | |
272 | list_del_init(&lo->plh_layouts); | |
273 | spin_unlock(&clp->cl_lock); | |
274 | } | |
9fa40758 | 275 | put_rpccred(lo->plh_lc_cred); |
57934278 | 276 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
277 | } |
278 | ||
e5e94017 | 279 | static void |
6622c3ea | 280 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 281 | { |
bb346f63 | 282 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 283 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
284 | nfsi->layout = NULL; |
285 | /* Reset MDS Threshold I/O counters */ | |
286 | nfsi->write_io = 0; | |
287 | nfsi->read_io = 0; | |
e5e94017 BH |
288 | } |
289 | ||
b1f69b75 | 290 | void |
70c3bd2b | 291 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 292 | { |
cc6e5340 FI |
293 | struct inode *inode = lo->plh_inode; |
294 | ||
13c13a6a TM |
295 | pnfs_layoutreturn_before_put_layout_hdr(lo); |
296 | ||
cc6e5340 | 297 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { |
566f8737 PT |
298 | if (!list_empty(&lo->plh_segs)) |
299 | WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n"); | |
6622c3ea | 300 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 301 | spin_unlock(&inode->i_lock); |
6622c3ea | 302 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 303 | } |
974cec8c AA |
304 | } |
305 | ||
ae5a459d TM |
306 | static void |
307 | pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo) | |
308 | { | |
309 | lo->plh_return_iomode = 0; | |
310 | lo->plh_return_seq = 0; | |
311 | clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); | |
312 | } | |
313 | ||
2454dfea TM |
314 | /* |
315 | * Mark a pnfs_layout_hdr and all associated layout segments as invalid | |
316 | * | |
317 | * In order to continue using the pnfs_layout_hdr, a full recovery | |
318 | * is required. | |
319 | * Note that caller must hold inode->i_lock. | |
320 | */ | |
5f46be04 | 321 | int |
2454dfea TM |
322 | pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo, |
323 | struct list_head *lseg_list) | |
324 | { | |
325 | struct pnfs_layout_range range = { | |
326 | .iomode = IOMODE_ANY, | |
327 | .offset = 0, | |
328 | .length = NFS4_MAX_UINT64, | |
329 | }; | |
330 | ||
331 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
ae5a459d | 332 | pnfs_clear_layoutreturn_info(lo); |
68f74479 | 333 | pnfs_free_returned_lsegs(lo, lseg_list, &range, 0); |
6d597e17 | 334 | return pnfs_mark_matching_lsegs_invalid(lo, lseg_list, &range, 0); |
2454dfea TM |
335 | } |
336 | ||
b9e028fd TM |
337 | static int |
338 | pnfs_iomode_to_fail_bit(u32 iomode) | |
339 | { | |
340 | return iomode == IOMODE_RW ? | |
341 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
342 | } | |
343 | ||
344 | static void | |
3e621214 | 345 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 346 | { |
25c75333 | 347 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 348 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
3e621214 TM |
349 | atomic_inc(&lo->plh_refcount); |
350 | } | |
351 | ||
352 | static void | |
353 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
354 | { | |
355 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
356 | atomic_dec(&lo->plh_refcount); | |
357 | } | |
358 | ||
359 | static void | |
360 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
361 | { | |
362 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
363 | struct pnfs_layout_range range = { |
364 | .iomode = iomode, | |
365 | .offset = 0, | |
366 | .length = NFS4_MAX_UINT64, | |
367 | }; | |
368 | LIST_HEAD(head); | |
3e621214 TM |
369 | |
370 | spin_lock(&inode->i_lock); | |
371 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
6d597e17 | 372 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0); |
3e621214 | 373 | spin_unlock(&inode->i_lock); |
115ce575 | 374 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
375 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
376 | iomode == IOMODE_RW ? "RW" : "READ"); | |
377 | } | |
378 | ||
379 | static bool | |
380 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
381 | { | |
25c75333 | 382 | unsigned long start, end; |
3e621214 TM |
383 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
384 | ||
385 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
386 | return false; |
387 | end = jiffies; | |
388 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
389 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
390 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 391 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
392 | return false; |
393 | } | |
394 | return true; | |
b9e028fd TM |
395 | } |
396 | ||
974cec8c | 397 | static void |
119cef97 TM |
398 | pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg, |
399 | const struct pnfs_layout_range *range, | |
400 | const nfs4_stateid *stateid) | |
974cec8c | 401 | { |
566052c5 | 402 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 403 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c | 404 | atomic_set(&lseg->pls_refcount, 1); |
4541d16c | 405 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); |
566052c5 | 406 | lseg->pls_layout = lo; |
119cef97 TM |
407 | lseg->pls_range = *range; |
408 | lseg->pls_seq = be32_to_cpu(stateid->seqid); | |
974cec8c AA |
409 | } |
410 | ||
905ca191 | 411 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 412 | { |
68f74479 TM |
413 | if (lseg != NULL) { |
414 | struct inode *inode = lseg->pls_layout->plh_inode; | |
415 | NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg); | |
416 | } | |
974cec8c AA |
417 | } |
418 | ||
d684d2ae | 419 | static void |
57036a37 TM |
420 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
421 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 422 | { |
d20581aa | 423 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 424 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
425 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
426 | atomic_dec(&lo->plh_refcount); | |
7b650994 TM |
427 | if (list_empty(&lo->plh_segs) && |
428 | !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) && | |
429 | !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { | |
334a8f37 TM |
430 | if (atomic_read(&lo->plh_outstanding) == 0) |
431 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
173f77e9 | 432 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
2d148c7e | 433 | } |
d684d2ae FI |
434 | } |
435 | ||
68f74479 TM |
436 | static bool |
437 | pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo, | |
438 | struct pnfs_layout_segment *lseg) | |
439 | { | |
440 | if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && | |
441 | pnfs_layout_is_valid(lo)) { | |
442 | list_move_tail(&lseg->pls_list, &lo->plh_return_segs); | |
443 | return true; | |
444 | } | |
445 | return false; | |
446 | } | |
447 | ||
bae724ef | 448 | void |
9369a431 | 449 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 450 | { |
57036a37 | 451 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
452 | struct inode *inode; |
453 | ||
454 | if (!lseg) | |
455 | return; | |
456 | ||
4541d16c FI |
457 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
458 | atomic_read(&lseg->pls_refcount), | |
459 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
4ef2e4f8 | 460 | |
57036a37 TM |
461 | lo = lseg->pls_layout; |
462 | inode = lo->plh_inode; | |
4ef2e4f8 | 463 | |
d684d2ae | 464 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
faa4a54f TM |
465 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { |
466 | spin_unlock(&inode->i_lock); | |
467 | return; | |
468 | } | |
8f0d27dc | 469 | pnfs_get_layout_hdr(lo); |
4ef2e4f8 | 470 | pnfs_layout_remove_lseg(lo, lseg); |
68f74479 TM |
471 | if (pnfs_cache_lseg_for_layoutreturn(lo, lseg)) |
472 | lseg = NULL; | |
4ef2e4f8 TM |
473 | spin_unlock(&inode->i_lock); |
474 | pnfs_free_lseg(lseg); | |
475 | pnfs_put_layout_hdr(lo); | |
4541d16c | 476 | } |
4541d16c | 477 | } |
9369a431 | 478 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 479 | |
6543f803 | 480 | static void pnfs_free_lseg_async_work(struct work_struct *work) |
e6cf82d1 WAA |
481 | { |
482 | struct pnfs_layout_segment *lseg; | |
6543f803 | 483 | struct pnfs_layout_hdr *lo; |
e6cf82d1 WAA |
484 | |
485 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
6543f803 | 486 | lo = lseg->pls_layout; |
e6cf82d1 | 487 | |
6543f803 TM |
488 | pnfs_free_lseg(lseg); |
489 | pnfs_put_layout_hdr(lo); | |
e6cf82d1 WAA |
490 | } |
491 | ||
6543f803 | 492 | static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg) |
e6cf82d1 | 493 | { |
6543f803 | 494 | INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work); |
e6cf82d1 WAA |
495 | schedule_work(&lseg->pls_work); |
496 | } | |
6543f803 TM |
497 | |
498 | void | |
499 | pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg) | |
500 | { | |
501 | if (!lseg) | |
502 | return; | |
503 | ||
504 | assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock); | |
505 | ||
506 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, | |
507 | atomic_read(&lseg->pls_refcount), | |
508 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
509 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
510 | struct pnfs_layout_hdr *lo = lseg->pls_layout; | |
faa4a54f TM |
511 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) |
512 | return; | |
6543f803 | 513 | pnfs_layout_remove_lseg(lo, lseg); |
68f74479 TM |
514 | if (!pnfs_cache_lseg_for_layoutreturn(lo, lseg)) { |
515 | pnfs_get_layout_hdr(lo); | |
516 | pnfs_free_lseg_async(lseg); | |
517 | } | |
6543f803 TM |
518 | } |
519 | } | |
520 | EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked); | |
e6cf82d1 | 521 | |
fb3296eb BH |
522 | /* |
523 | * is l2 fully contained in l1? | |
524 | * start1 end1 | |
525 | * [----------------------------------) | |
526 | * start2 end2 | |
527 | * [----------------) | |
528 | */ | |
3cb2df17 | 529 | static bool |
7dc0ac70 | 530 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 531 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
532 | { |
533 | u64 start1 = l1->offset; | |
17822b20 | 534 | u64 end1 = pnfs_end_offset(start1, l1->length); |
fb3296eb | 535 | u64 start2 = l2->offset; |
17822b20 | 536 | u64 end2 = pnfs_end_offset(start2, l2->length); |
fb3296eb BH |
537 | |
538 | return (start1 <= start2) && (end1 >= end2); | |
539 | } | |
540 | ||
24956804 TM |
541 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
542 | struct list_head *tmp_list) | |
543 | { | |
544 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
545 | return false; | |
546 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
547 | list_add(&lseg->pls_list, tmp_list); | |
548 | return true; | |
549 | } | |
550 | ||
4541d16c FI |
551 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
552 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
553 | struct list_head *tmp_list) | |
554 | { | |
555 | int rv = 0; | |
556 | ||
557 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
558 | /* Remove the reference keeping the lseg in the | |
559 | * list. It will now be removed when all | |
560 | * outstanding io is finished. | |
561 | */ | |
d684d2ae FI |
562 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
563 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 564 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 565 | rv = 1; |
4541d16c FI |
566 | } |
567 | return rv; | |
568 | } | |
569 | ||
6d597e17 JL |
570 | /* |
571 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
572 | * taking into account wraparound issues. | |
573 | */ | |
574 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
575 | { | |
576 | return (s32)(s1 - s2) > 0; | |
577 | } | |
578 | ||
e036f464 TM |
579 | static bool |
580 | pnfs_should_free_range(const struct pnfs_layout_range *lseg_range, | |
581 | const struct pnfs_layout_range *recall_range) | |
582 | { | |
583 | return (recall_range->iomode == IOMODE_ANY || | |
584 | lseg_range->iomode == recall_range->iomode) && | |
585 | pnfs_lseg_range_intersecting(lseg_range, recall_range); | |
586 | } | |
587 | ||
588 | static bool | |
589 | pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg, | |
590 | const struct pnfs_layout_range *recall_range, | |
591 | u32 seq) | |
592 | { | |
593 | if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq)) | |
594 | return false; | |
595 | if (recall_range == NULL) | |
596 | return true; | |
597 | return pnfs_should_free_range(&lseg->pls_range, recall_range); | |
598 | } | |
599 | ||
6d597e17 JL |
600 | /** |
601 | * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later | |
602 | * @lo: layout header containing the lsegs | |
603 | * @tmp_list: list head where doomed lsegs should go | |
604 | * @recall_range: optional recall range argument to match (may be NULL) | |
605 | * @seq: only invalidate lsegs obtained prior to this sequence (may be 0) | |
606 | * | |
607 | * Walk the list of lsegs in the layout header, and tear down any that should | |
608 | * be destroyed. If "recall_range" is specified then the segment must match | |
609 | * that range. If "seq" is non-zero, then only match segments that were handed | |
610 | * out at or before that sequence. | |
611 | * | |
612 | * Returns number of matching invalid lsegs remaining in list after scanning | |
613 | * it and purging them. | |
4541d16c | 614 | */ |
43f1b3da | 615 | int |
49a85061 | 616 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 617 | struct list_head *tmp_list, |
6d597e17 JL |
618 | const struct pnfs_layout_range *recall_range, |
619 | u32 seq) | |
974cec8c AA |
620 | { |
621 | struct pnfs_layout_segment *lseg, *next; | |
71b39854 | 622 | int remaining = 0; |
974cec8c AA |
623 | |
624 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
625 | ||
8006bfba | 626 | if (list_empty(&lo->plh_segs)) |
38511722 | 627 | return 0; |
4541d16c | 628 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
e036f464 | 629 | if (pnfs_match_lseg_recall(lseg, recall_range, seq)) { |
6d597e17 | 630 | dprintk("%s: freeing lseg %p iomode %d seq %u" |
4541d16c | 631 | "offset %llu length %llu\n", __func__, |
6d597e17 JL |
632 | lseg, lseg->pls_range.iomode, lseg->pls_seq, |
633 | lseg->pls_range.offset, lseg->pls_range.length); | |
71b39854 TM |
634 | if (!mark_lseg_invalid(lseg, tmp_list)) |
635 | remaining++; | |
4541d16c | 636 | } |
71b39854 TM |
637 | dprintk("%s:Return %i\n", __func__, remaining); |
638 | return remaining; | |
974cec8c AA |
639 | } |
640 | ||
68f74479 TM |
641 | static void |
642 | pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo, | |
643 | struct list_head *free_me, | |
644 | const struct pnfs_layout_range *range, | |
645 | u32 seq) | |
646 | { | |
647 | struct pnfs_layout_segment *lseg, *next; | |
648 | ||
649 | list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) { | |
650 | if (pnfs_match_lseg_recall(lseg, range, seq)) | |
651 | list_move_tail(&lseg->pls_list, free_me); | |
652 | } | |
653 | } | |
654 | ||
f49f9baa | 655 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 656 | void |
4541d16c | 657 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 658 | { |
4541d16c | 659 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
660 | |
661 | if (list_empty(free_me)) | |
662 | return; | |
663 | ||
4541d16c | 664 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 665 | list_del(&lseg->pls_list); |
905ca191 | 666 | pnfs_free_lseg(lseg); |
974cec8c AA |
667 | } |
668 | } | |
669 | ||
e5e94017 BH |
670 | void |
671 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
672 | { | |
673 | struct pnfs_layout_hdr *lo; | |
974cec8c | 674 | LIST_HEAD(tmp_list); |
e5e94017 BH |
675 | |
676 | spin_lock(&nfsi->vfs_inode.i_lock); | |
677 | lo = nfsi->layout; | |
678 | if (lo) { | |
3e621214 | 679 | pnfs_get_layout_hdr(lo); |
2454dfea | 680 | pnfs_mark_layout_stateid_invalid(lo, &tmp_list); |
3e621214 TM |
681 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); |
682 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
683 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
684 | pnfs_free_lseg_list(&tmp_list); | |
685 | pnfs_put_layout_hdr(lo); | |
686 | } else | |
687 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 688 | } |
041245c8 | 689 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 690 | |
fd9a8d71 TM |
691 | static bool |
692 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
693 | struct list_head *layout_list) | |
974cec8c AA |
694 | { |
695 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 696 | bool ret = false; |
974cec8c | 697 | |
fd9a8d71 TM |
698 | spin_lock(&inode->i_lock); |
699 | lo = NFS_I(inode)->layout; | |
700 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
701 | pnfs_get_layout_hdr(lo); | |
702 | list_add(&lo->plh_bulk_destroy, layout_list); | |
703 | ret = true; | |
704 | } | |
705 | spin_unlock(&inode->i_lock); | |
706 | return ret; | |
707 | } | |
708 | ||
709 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
710 | static int | |
711 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
712 | struct nfs_server *server, | |
713 | struct list_head *layout_list) | |
714 | { | |
715 | struct pnfs_layout_hdr *lo, *next; | |
716 | struct inode *inode; | |
717 | ||
718 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
719 | inode = igrab(lo->plh_inode); | |
720 | if (inode == NULL) | |
721 | continue; | |
722 | list_del_init(&lo->plh_layouts); | |
723 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
724 | continue; | |
725 | rcu_read_unlock(); | |
726 | spin_unlock(&clp->cl_lock); | |
727 | iput(inode); | |
728 | spin_lock(&clp->cl_lock); | |
729 | rcu_read_lock(); | |
730 | return -EAGAIN; | |
731 | } | |
732 | return 0; | |
733 | } | |
734 | ||
735 | static int | |
736 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
737 | bool is_bulk_recall) | |
738 | { | |
739 | struct pnfs_layout_hdr *lo; | |
740 | struct inode *inode; | |
fd9a8d71 TM |
741 | LIST_HEAD(lseg_list); |
742 | int ret = 0; | |
743 | ||
744 | while (!list_empty(layout_list)) { | |
745 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
746 | plh_bulk_destroy); | |
747 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
748 | lo->plh_inode->i_ino); | |
749 | inode = lo->plh_inode; | |
7c5d1875 CH |
750 | |
751 | pnfs_layoutcommit_inode(inode, false); | |
752 | ||
fd9a8d71 TM |
753 | spin_lock(&inode->i_lock); |
754 | list_del_init(&lo->plh_bulk_destroy); | |
9fd4b9fc TM |
755 | if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) { |
756 | if (is_bulk_recall) | |
757 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
fd9a8d71 | 758 | ret = -EAGAIN; |
9fd4b9fc | 759 | } |
fd9a8d71 TM |
760 | spin_unlock(&inode->i_lock); |
761 | pnfs_free_lseg_list(&lseg_list); | |
b20135d0 TM |
762 | /* Free all lsegs that are attached to commit buckets */ |
763 | nfs_commit_inode(inode, 0); | |
fd9a8d71 TM |
764 | pnfs_put_layout_hdr(lo); |
765 | iput(inode); | |
766 | } | |
767 | return ret; | |
768 | } | |
769 | ||
770 | int | |
771 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
772 | struct nfs_fsid *fsid, | |
773 | bool is_recall) | |
774 | { | |
775 | struct nfs_server *server; | |
776 | LIST_HEAD(layout_list); | |
c47abcf8 | 777 | |
974cec8c | 778 | spin_lock(&clp->cl_lock); |
6382a441 | 779 | rcu_read_lock(); |
fd9a8d71 | 780 | restart: |
6382a441 | 781 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
782 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
783 | continue; | |
784 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
785 | server, | |
786 | &layout_list) != 0) | |
787 | goto restart; | |
6382a441 WAA |
788 | } |
789 | rcu_read_unlock(); | |
974cec8c AA |
790 | spin_unlock(&clp->cl_lock); |
791 | ||
fd9a8d71 TM |
792 | if (list_empty(&layout_list)) |
793 | return 0; | |
794 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
795 | } | |
796 | ||
797 | int | |
798 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
799 | bool is_recall) | |
800 | { | |
801 | struct nfs_server *server; | |
802 | LIST_HEAD(layout_list); | |
803 | ||
804 | spin_lock(&clp->cl_lock); | |
805 | rcu_read_lock(); | |
806 | restart: | |
807 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
808 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
809 | server, | |
810 | &layout_list) != 0) | |
811 | goto restart; | |
974cec8c | 812 | } |
fd9a8d71 TM |
813 | rcu_read_unlock(); |
814 | spin_unlock(&clp->cl_lock); | |
815 | ||
816 | if (list_empty(&layout_list)) | |
817 | return 0; | |
818 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
819 | } | |
820 | ||
821 | /* | |
822 | * Called by the state manger to remove all layouts established under an | |
823 | * expired lease. | |
824 | */ | |
825 | void | |
826 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
827 | { | |
828 | nfs4_deviceid_mark_client_invalid(clp); | |
829 | nfs4_deviceid_purge_client(clp); | |
830 | ||
831 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
832 | } |
833 | ||
fd6002e9 | 834 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
835 | void |
836 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
837 | bool update_barrier) | |
b1f69b75 | 838 | { |
ecebb80b | 839 | u32 oldseq, newseq, new_barrier = 0; |
b1f69b75 | 840 | |
2d2f24ad TM |
841 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
842 | newseq = be32_to_cpu(new->seqid); | |
2a59a041 TM |
843 | |
844 | if (!pnfs_layout_is_valid(lo)) { | |
845 | nfs4_stateid_copy(&lo->plh_stateid, new); | |
846 | lo->plh_barrier = newseq; | |
847 | pnfs_clear_layoutreturn_info(lo); | |
848 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
849 | return; | |
850 | } | |
851 | if (pnfs_seqid_is_newer(newseq, oldseq)) { | |
f597c537 | 852 | nfs4_stateid_copy(&lo->plh_stateid, new); |
ecebb80b TM |
853 | /* |
854 | * Because of wraparound, we want to keep the barrier | |
855 | * "close" to the current seqids. | |
856 | */ | |
857 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); | |
43f1b3da | 858 | } |
ecebb80b TM |
859 | if (update_barrier) |
860 | new_barrier = be32_to_cpu(new->seqid); | |
861 | else if (new_barrier == 0) | |
862 | return; | |
2a59a041 | 863 | if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
ecebb80b | 864 | lo->plh_barrier = new_barrier; |
b1f69b75 AA |
865 | } |
866 | ||
cf7d63f1 | 867 | static bool |
19c54aba TM |
868 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
869 | const nfs4_stateid *stateid) | |
43f1b3da | 870 | { |
19c54aba | 871 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 872 | |
19c54aba TM |
873 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
874 | } | |
875 | ||
876 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
877 | static bool | |
e1c06f80 | 878 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo) |
19c54aba | 879 | { |
f7e8917a | 880 | return lo->plh_block_lgets || |
e1c06f80 | 881 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
cf7d63f1 FI |
882 | } |
883 | ||
b1f69b75 | 884 | /* |
183d9e7b JL |
885 | * Get layout from server. |
886 | * for now, assume that whole file layouts are requested. | |
887 | * arg->offset: 0 | |
888 | * arg->length: all ones | |
889 | */ | |
e5e94017 BH |
890 | static struct pnfs_layout_segment * |
891 | send_layoutget(struct pnfs_layout_hdr *lo, | |
892 | struct nfs_open_context *ctx, | |
183d9e7b | 893 | nfs4_stateid *stateid, |
e144e539 | 894 | const struct pnfs_layout_range *range, |
183d9e7b | 895 | long *timeout, gfp_t gfp_flags) |
e5e94017 | 896 | { |
b7edfaa1 | 897 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
898 | struct nfs_server *server = NFS_SERVER(ino); |
899 | struct nfs4_layoutget *lgp; | |
2d89a1d3 | 900 | loff_t i_size; |
b1f69b75 AA |
901 | |
902 | dprintk("--> %s\n", __func__); | |
e5e94017 | 903 | |
4f2e9dce JL |
904 | /* |
905 | * Synchronously retrieve layout information from server and | |
906 | * store in lseg. If we race with a concurrent seqid morphing | |
907 | * op, then re-send the LAYOUTGET. | |
908 | */ | |
83026d80 JL |
909 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
910 | if (lgp == NULL) | |
911 | return ERR_PTR(-ENOMEM); | |
912 | ||
913 | i_size = i_size_read(ino); | |
914 | ||
915 | lgp->args.minlength = PAGE_SIZE; | |
916 | if (lgp->args.minlength > range->length) | |
917 | lgp->args.minlength = range->length; | |
918 | if (range->iomode == IOMODE_READ) { | |
919 | if (range->offset >= i_size) | |
920 | lgp->args.minlength = 0; | |
921 | else if (i_size - range->offset < lgp->args.minlength) | |
922 | lgp->args.minlength = i_size - range->offset; | |
d03ab29d | 923 | } |
83026d80 JL |
924 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
925 | pnfs_copy_range(&lgp->args.range, range); | |
926 | lgp->args.type = server->pnfs_curr_ld->id; | |
927 | lgp->args.inode = ino; | |
928 | lgp->args.ctx = get_nfs_open_context(ctx); | |
183d9e7b | 929 | nfs4_stateid_copy(&lgp->args.stateid, stateid); |
83026d80 JL |
930 | lgp->gfp_flags = gfp_flags; |
931 | lgp->cred = lo->plh_lc_cred; | |
35124a09 | 932 | |
183d9e7b | 933 | return nfs4_proc_layoutget(lgp, timeout, gfp_flags); |
974cec8c AA |
934 | } |
935 | ||
24956804 TM |
936 | static void pnfs_clear_layoutcommit(struct inode *inode, |
937 | struct list_head *head) | |
938 | { | |
939 | struct nfs_inode *nfsi = NFS_I(inode); | |
940 | struct pnfs_layout_segment *lseg, *tmp; | |
941 | ||
942 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
943 | return; | |
944 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
945 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
946 | continue; | |
947 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
948 | } | |
949 | } | |
950 | ||
68f74479 | 951 | static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo) |
d67ae825 TH |
952 | { |
953 | clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
6604b203 | 954 | clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags); |
d67ae825 TH |
955 | smp_mb__after_atomic(); |
956 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN); | |
7f27392c | 957 | rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq); |
d67ae825 TH |
958 | } |
959 | ||
68f74479 | 960 | void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo, |
2a974425 | 961 | const nfs4_stateid *arg_stateid, |
68f74479 | 962 | const struct pnfs_layout_range *range, |
68f74479 TM |
963 | const nfs4_stateid *stateid) |
964 | { | |
965 | struct inode *inode = lo->plh_inode; | |
966 | LIST_HEAD(freeme); | |
967 | ||
968 | spin_lock(&inode->i_lock); | |
2a974425 TM |
969 | if (!pnfs_layout_is_valid(lo) || !arg_stateid || |
970 | !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid)) | |
971 | goto out_unlock; | |
68f74479 | 972 | if (stateid) { |
2a974425 TM |
973 | u32 seq = be32_to_cpu(arg_stateid->seqid); |
974 | ||
68f74479 TM |
975 | pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq); |
976 | pnfs_free_returned_lsegs(lo, &freeme, range, seq); | |
977 | pnfs_set_layout_stateid(lo, stateid, true); | |
978 | } else | |
979 | pnfs_mark_layout_stateid_invalid(lo, &freeme); | |
2a974425 | 980 | out_unlock: |
68f74479 TM |
981 | pnfs_clear_layoutreturn_waitbit(lo); |
982 | spin_unlock(&inode->i_lock); | |
983 | pnfs_free_lseg_list(&freeme); | |
984 | ||
985 | } | |
986 | ||
13c13a6a | 987 | static bool |
e5fd1904 TM |
988 | pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo, |
989 | nfs4_stateid *stateid, | |
990 | enum pnfs_iomode *iomode) | |
13c13a6a | 991 | { |
bf0291dd TM |
992 | /* Serialise LAYOUTGET/LAYOUTRETURN */ |
993 | if (atomic_read(&lo->plh_outstanding) != 0) | |
994 | return false; | |
6604b203 | 995 | if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) |
13c13a6a | 996 | return false; |
6604b203 | 997 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); |
13c13a6a | 998 | pnfs_get_layout_hdr(lo); |
e5fd1904 TM |
999 | if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) { |
1000 | if (stateid != NULL) { | |
1001 | nfs4_stateid_copy(stateid, &lo->plh_stateid); | |
1002 | if (lo->plh_return_seq != 0) | |
1003 | stateid->seqid = cpu_to_be32(lo->plh_return_seq); | |
1004 | } | |
1005 | if (iomode != NULL) | |
1006 | *iomode = lo->plh_return_iomode; | |
1007 | pnfs_clear_layoutreturn_info(lo); | |
1008 | return true; | |
1009 | } | |
1010 | if (stateid != NULL) | |
1011 | nfs4_stateid_copy(stateid, &lo->plh_stateid); | |
1012 | if (iomode != NULL) | |
1013 | *iomode = IOMODE_ANY; | |
13c13a6a TM |
1014 | return true; |
1015 | } | |
1016 | ||
828ed9ec TM |
1017 | static void |
1018 | pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args, | |
1019 | struct pnfs_layout_hdr *lo, | |
1020 | const nfs4_stateid *stateid, | |
1021 | enum pnfs_iomode iomode) | |
1022 | { | |
1023 | struct inode *inode = lo->plh_inode; | |
1024 | ||
1025 | args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id; | |
1026 | args->inode = inode; | |
1027 | args->range.iomode = iomode; | |
1028 | args->range.offset = 0; | |
1029 | args->range.length = NFS4_MAX_UINT64; | |
1030 | args->layout = lo; | |
1031 | nfs4_stateid_copy(&args->stateid, stateid); | |
1032 | } | |
1033 | ||
f40eb5d0 | 1034 | static int |
ed429d6b | 1035 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid, |
6c16605d | 1036 | enum pnfs_iomode iomode, bool sync) |
f40eb5d0 PT |
1037 | { |
1038 | struct inode *ino = lo->plh_inode; | |
1039 | struct nfs4_layoutreturn *lrp; | |
1040 | int status = 0; | |
1041 | ||
e4af440a | 1042 | lrp = kzalloc(sizeof(*lrp), GFP_NOFS); |
f40eb5d0 PT |
1043 | if (unlikely(lrp == NULL)) { |
1044 | status = -ENOMEM; | |
1045 | spin_lock(&ino->i_lock); | |
d67ae825 | 1046 | pnfs_clear_layoutreturn_waitbit(lo); |
f40eb5d0 PT |
1047 | spin_unlock(&ino->i_lock); |
1048 | pnfs_put_layout_hdr(lo); | |
1049 | goto out; | |
1050 | } | |
1051 | ||
828ed9ec | 1052 | pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode); |
f40eb5d0 PT |
1053 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
1054 | lrp->cred = lo->plh_lc_cred; | |
1055 | ||
6c16605d | 1056 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
1057 | out: |
1058 | dprintk("<-- %s status: %d\n", __func__, status); | |
1059 | return status; | |
1060 | } | |
1061 | ||
13c13a6a TM |
1062 | /* Return true if layoutreturn is needed */ |
1063 | static bool | |
1064 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo) | |
1065 | { | |
1066 | struct pnfs_layout_segment *s; | |
1067 | ||
2370abda | 1068 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
1069 | return false; |
1070 | ||
1071 | /* Defer layoutreturn until all lsegs are done */ | |
1072 | list_for_each_entry(s, &lo->plh_segs, pls_list) { | |
1073 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags)) | |
1074 | return false; | |
1075 | } | |
1076 | ||
1077 | return true; | |
1078 | } | |
1079 | ||
1080 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo) | |
1081 | { | |
1082 | struct inode *inode= lo->plh_inode; | |
1083 | ||
2370abda | 1084 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
1085 | return; |
1086 | spin_lock(&inode->i_lock); | |
1087 | if (pnfs_layout_need_return(lo)) { | |
1088 | nfs4_stateid stateid; | |
1089 | enum pnfs_iomode iomode; | |
1090 | bool send; | |
1091 | ||
e5fd1904 | 1092 | send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode); |
13c13a6a TM |
1093 | spin_unlock(&inode->i_lock); |
1094 | if (send) { | |
1095 | /* Send an async layoutreturn so we dont deadlock */ | |
1096 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); | |
1097 | } | |
1098 | } else | |
1099 | spin_unlock(&inode->i_lock); | |
1100 | } | |
1101 | ||
293b3b06 AA |
1102 | /* |
1103 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
1104 | * when the layout segment list is empty. | |
1105 | * | |
1106 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
1107 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
1108 | * deviceid is marked invalid. | |
1109 | */ | |
cbe82603 BH |
1110 | int |
1111 | _pnfs_return_layout(struct inode *ino) | |
1112 | { | |
1113 | struct pnfs_layout_hdr *lo = NULL; | |
1114 | struct nfs_inode *nfsi = NFS_I(ino); | |
1115 | LIST_HEAD(tmp_list); | |
cbe82603 | 1116 | nfs4_stateid stateid; |
293b3b06 | 1117 | int status = 0, empty; |
7f27392c | 1118 | bool send; |
cbe82603 | 1119 | |
366d5052 | 1120 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
1121 | |
1122 | spin_lock(&ino->i_lock); | |
1123 | lo = nfsi->layout; | |
e5929f3c | 1124 | if (!lo) { |
cbe82603 | 1125 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
1126 | dprintk("NFS: %s no layout to return\n", __func__); |
1127 | goto out; | |
cbe82603 | 1128 | } |
cbe82603 | 1129 | /* Reference matched in nfs4_layoutreturn_release */ |
70c3bd2b | 1130 | pnfs_get_layout_hdr(lo); |
293b3b06 | 1131 | empty = list_empty(&lo->plh_segs); |
24956804 | 1132 | pnfs_clear_layoutcommit(ino, &tmp_list); |
6d597e17 | 1133 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0); |
c88953d8 CH |
1134 | |
1135 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
1136 | struct pnfs_layout_range range = { | |
1137 | .iomode = IOMODE_ANY, | |
1138 | .offset = 0, | |
1139 | .length = NFS4_MAX_UINT64, | |
1140 | }; | |
1141 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
1142 | } | |
1143 | ||
293b3b06 AA |
1144 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
1145 | if (empty) { | |
1146 | spin_unlock(&ino->i_lock); | |
293b3b06 | 1147 | dprintk("NFS: %s no layout segments to return\n", __func__); |
7f27392c | 1148 | goto out_put_layout_hdr; |
293b3b06 | 1149 | } |
47abadef | 1150 | |
e5fd1904 | 1151 | send = pnfs_prepare_layoutreturn(lo, &stateid, NULL); |
cbe82603 BH |
1152 | spin_unlock(&ino->i_lock); |
1153 | pnfs_free_lseg_list(&tmp_list); | |
7f27392c | 1154 | if (send) |
ed429d6b | 1155 | status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
7f27392c TM |
1156 | out_put_layout_hdr: |
1157 | pnfs_put_layout_hdr(lo); | |
cbe82603 BH |
1158 | out: |
1159 | dprintk("<-- %s status: %d\n", __func__, status); | |
1160 | return status; | |
1161 | } | |
0a57cdac | 1162 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 1163 | |
24028672 TM |
1164 | int |
1165 | pnfs_commit_and_return_layout(struct inode *inode) | |
1166 | { | |
1167 | struct pnfs_layout_hdr *lo; | |
1168 | int ret; | |
1169 | ||
1170 | spin_lock(&inode->i_lock); | |
1171 | lo = NFS_I(inode)->layout; | |
1172 | if (lo == NULL) { | |
1173 | spin_unlock(&inode->i_lock); | |
1174 | return 0; | |
1175 | } | |
1176 | pnfs_get_layout_hdr(lo); | |
1177 | /* Block new layoutgets and read/write to ds */ | |
1178 | lo->plh_block_lgets++; | |
1179 | spin_unlock(&inode->i_lock); | |
1180 | filemap_fdatawait(inode->i_mapping); | |
1181 | ret = pnfs_layoutcommit_inode(inode, true); | |
1182 | if (ret == 0) | |
1183 | ret = _pnfs_return_layout(inode); | |
1184 | spin_lock(&inode->i_lock); | |
1185 | lo->plh_block_lgets--; | |
1186 | spin_unlock(&inode->i_lock); | |
1187 | pnfs_put_layout_hdr(lo); | |
1188 | return ret; | |
1189 | } | |
1190 | ||
f7e8917a FI |
1191 | bool pnfs_roc(struct inode *ino) |
1192 | { | |
40dd4b7a TM |
1193 | struct nfs_inode *nfsi = NFS_I(ino); |
1194 | struct nfs_open_context *ctx; | |
1195 | struct nfs4_state *state; | |
f7e8917a FI |
1196 | struct pnfs_layout_hdr *lo; |
1197 | struct pnfs_layout_segment *lseg, *tmp; | |
193e3aa2 | 1198 | nfs4_stateid stateid; |
f7e8917a | 1199 | LIST_HEAD(tmp_list); |
e755d638 | 1200 | bool found = false, layoutreturn = false, roc = false; |
f7e8917a FI |
1201 | |
1202 | spin_lock(&ino->i_lock); | |
40dd4b7a | 1203 | lo = nfsi->layout; |
0cdc329e TM |
1204 | if (!lo || !pnfs_layout_is_valid(lo) || |
1205 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
40dd4b7a TM |
1206 | goto out_noroc; |
1207 | ||
e755d638 | 1208 | /* no roc if we hold a delegation */ |
40dd4b7a TM |
1209 | if (nfs4_check_delegation(ino, FMODE_READ)) |
1210 | goto out_noroc; | |
1211 | ||
1212 | list_for_each_entry(ctx, &nfsi->open_files, list) { | |
1213 | state = ctx->state; | |
1214 | /* Don't return layout if there is open file state */ | |
1215 | if (state != NULL && state->state != 0) | |
1216 | goto out_noroc; | |
1217 | } | |
1218 | ||
e755d638 | 1219 | |
69820d22 | 1220 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) { |
e755d638 | 1221 | /* If we are sending layoutreturn, invalidate all valid lsegs */ |
69820d22 | 1222 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { |
f7e8917a FI |
1223 | mark_lseg_invalid(lseg, &tmp_list); |
1224 | found = true; | |
1225 | } | |
69820d22 TM |
1226 | } |
1227 | ||
1228 | /* always send layoutreturn if being marked so */ | |
1229 | if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) { | |
1230 | layoutreturn = pnfs_prepare_layoutreturn(lo, | |
1231 | &stateid, NULL); | |
1232 | if (layoutreturn) | |
1233 | goto out_noroc; | |
1234 | } | |
1235 | ||
500d701f | 1236 | /* ROC in two conditions: |
e755d638 PT |
1237 | * 1. there are ROC lsegs |
1238 | * 2. we don't send layoutreturn | |
e755d638 | 1239 | */ |
69820d22 | 1240 | if (found) { |
500d701f PT |
1241 | /* lo ref dropped in pnfs_roc_release() */ |
1242 | pnfs_get_layout_hdr(lo); | |
e755d638 | 1243 | roc = true; |
500d701f | 1244 | } |
f7e8917a | 1245 | |
40dd4b7a | 1246 | out_noroc: |
f7e8917a | 1247 | spin_unlock(&ino->i_lock); |
e755d638 PT |
1248 | pnfs_free_lseg_list(&tmp_list); |
1249 | pnfs_layoutcommit_inode(ino, true); | |
1250 | if (layoutreturn) | |
ed429d6b | 1251 | pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
e755d638 | 1252 | return roc; |
f7e8917a FI |
1253 | } |
1254 | ||
1255 | void pnfs_roc_release(struct inode *ino) | |
1256 | { | |
1257 | struct pnfs_layout_hdr *lo; | |
1258 | ||
1259 | spin_lock(&ino->i_lock); | |
1260 | lo = NFS_I(ino)->layout; | |
5c4a79fb | 1261 | pnfs_clear_layoutreturn_waitbit(lo); |
6622c3ea TM |
1262 | if (atomic_dec_and_test(&lo->plh_refcount)) { |
1263 | pnfs_detach_layout_hdr(lo); | |
1264 | spin_unlock(&ino->i_lock); | |
1265 | pnfs_free_layout_hdr(lo); | |
1266 | } else | |
1267 | spin_unlock(&ino->i_lock); | |
f7e8917a FI |
1268 | } |
1269 | ||
1270 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
1271 | { | |
1272 | struct pnfs_layout_hdr *lo; | |
1273 | ||
1274 | spin_lock(&ino->i_lock); | |
1275 | lo = NFS_I(ino)->layout; | |
0f35ad6f | 1276 | if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) |
f7e8917a FI |
1277 | lo->plh_barrier = barrier; |
1278 | spin_unlock(&ino->i_lock); | |
6a463beb | 1279 | trace_nfs4_layoutreturn_on_close(ino, 0); |
f7e8917a FI |
1280 | } |
1281 | ||
4ff376fe | 1282 | void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier) |
f7e8917a FI |
1283 | { |
1284 | struct nfs_inode *nfsi = NFS_I(ino); | |
7fdab069 | 1285 | struct pnfs_layout_hdr *lo; |
7fdab069 | 1286 | u32 current_seqid; |
f7e8917a FI |
1287 | |
1288 | spin_lock(&ino->i_lock); | |
7fdab069 TM |
1289 | lo = nfsi->layout; |
1290 | current_seqid = be32_to_cpu(lo->plh_stateid.seqid); | |
f7e8917a | 1291 | |
7fdab069 TM |
1292 | /* Since close does not return a layout stateid for use as |
1293 | * a barrier, we choose the worst-case barrier. | |
1294 | */ | |
1295 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
f7e8917a | 1296 | spin_unlock(&ino->i_lock); |
f7e8917a FI |
1297 | } |
1298 | ||
500d701f PT |
1299 | bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task) |
1300 | { | |
1301 | struct nfs_inode *nfsi = NFS_I(ino); | |
1302 | struct pnfs_layout_hdr *lo; | |
1303 | bool sleep = false; | |
1304 | ||
1305 | /* we might not have grabbed lo reference. so need to check under | |
1306 | * i_lock */ | |
1307 | spin_lock(&ino->i_lock); | |
1308 | lo = nfsi->layout; | |
ee284e35 TM |
1309 | if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
1310 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
500d701f | 1311 | sleep = true; |
ee284e35 | 1312 | } |
500d701f | 1313 | spin_unlock(&ino->i_lock); |
500d701f PT |
1314 | return sleep; |
1315 | } | |
1316 | ||
b1f69b75 AA |
1317 | /* |
1318 | * Compare two layout segments for sorting into layout cache. | |
1319 | * We want to preferentially return RW over RO layouts, so ensure those | |
1320 | * are seen first. | |
1321 | */ | |
1322 | static s64 | |
7dc0ac70 | 1323 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1324 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1325 | { |
fb3296eb BH |
1326 | s64 d; |
1327 | ||
1328 | /* high offset > low offset */ | |
1329 | d = l1->offset - l2->offset; | |
1330 | if (d) | |
1331 | return d; | |
1332 | ||
1333 | /* short length > long length */ | |
1334 | d = l2->length - l1->length; | |
1335 | if (d) | |
1336 | return d; | |
1337 | ||
b1f69b75 | 1338 | /* read > read/write */ |
fb3296eb | 1339 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1340 | } |
1341 | ||
03772d2f TM |
1342 | static bool |
1343 | pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1, | |
1344 | const struct pnfs_layout_range *l2) | |
1345 | { | |
1346 | return pnfs_lseg_range_cmp(l1, l2) > 0; | |
1347 | } | |
1348 | ||
1349 | static bool | |
1350 | pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg, | |
1351 | struct pnfs_layout_segment *old) | |
1352 | { | |
1353 | return false; | |
1354 | } | |
1355 | ||
1356 | void | |
1357 | pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1358 | struct pnfs_layout_segment *lseg, | |
1359 | bool (*is_after)(const struct pnfs_layout_range *, | |
1360 | const struct pnfs_layout_range *), | |
1361 | bool (*do_merge)(struct pnfs_layout_segment *, | |
1362 | struct pnfs_layout_segment *), | |
1363 | struct list_head *free_me) | |
974cec8c | 1364 | { |
03772d2f | 1365 | struct pnfs_layout_segment *lp, *tmp; |
b1f69b75 | 1366 | |
974cec8c AA |
1367 | dprintk("%s:Begin\n", __func__); |
1368 | ||
03772d2f TM |
1369 | list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) { |
1370 | if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0) | |
1371 | continue; | |
1372 | if (do_merge(lseg, lp)) { | |
1373 | mark_lseg_invalid(lp, free_me); | |
1374 | continue; | |
1375 | } | |
1376 | if (is_after(&lseg->pls_range, &lp->pls_range)) | |
b1f69b75 | 1377 | continue; |
566052c5 | 1378 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1379 | dprintk("%s: inserted lseg %p " |
1380 | "iomode %d offset %llu length %llu before " | |
1381 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1382 | __func__, lseg, lseg->pls_range.iomode, |
1383 | lseg->pls_range.offset, lseg->pls_range.length, | |
1384 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1385 | lp->pls_range.length); | |
fb3296eb | 1386 | goto out; |
974cec8c | 1387 | } |
fb3296eb BH |
1388 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1389 | dprintk("%s: inserted lseg %p " | |
1390 | "iomode %d offset %llu length %llu at tail\n", | |
1391 | __func__, lseg, lseg->pls_range.iomode, | |
1392 | lseg->pls_range.offset, lseg->pls_range.length); | |
1393 | out: | |
70c3bd2b | 1394 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1395 | |
1396 | dprintk("%s:Return\n", __func__); | |
e5e94017 | 1397 | } |
03772d2f TM |
1398 | EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg); |
1399 | ||
1400 | static void | |
1401 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1402 | struct pnfs_layout_segment *lseg, | |
1403 | struct list_head *free_me) | |
1404 | { | |
1405 | struct inode *inode = lo->plh_inode; | |
1406 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
1407 | ||
1408 | if (ld->add_lseg != NULL) | |
1409 | ld->add_lseg(lo, lseg, free_me); | |
1410 | else | |
1411 | pnfs_generic_layout_insert_lseg(lo, lseg, | |
1412 | pnfs_lseg_range_is_after, | |
1413 | pnfs_lseg_no_merge, | |
1414 | free_me); | |
1415 | } | |
e5e94017 BH |
1416 | |
1417 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1418 | alloc_init_layout_hdr(struct inode *ino, |
1419 | struct nfs_open_context *ctx, | |
1420 | gfp_t gfp_flags) | |
e5e94017 BH |
1421 | { |
1422 | struct pnfs_layout_hdr *lo; | |
1423 | ||
636fb9c8 | 1424 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1425 | if (!lo) |
1426 | return NULL; | |
cc6e5340 | 1427 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1428 | INIT_LIST_HEAD(&lo->plh_layouts); |
1429 | INIT_LIST_HEAD(&lo->plh_segs); | |
68f74479 | 1430 | INIT_LIST_HEAD(&lo->plh_return_segs); |
fd9a8d71 | 1431 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1432 | lo->plh_inode = ino; |
5cc2216d | 1433 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
67a3b721 | 1434 | lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID; |
e5e94017 BH |
1435 | return lo; |
1436 | } | |
1437 | ||
1438 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1439 | pnfs_find_alloc_layout(struct inode *ino, |
1440 | struct nfs_open_context *ctx, | |
1441 | gfp_t gfp_flags) | |
e5241e43 TM |
1442 | __releases(&ino->i_lock) |
1443 | __acquires(&ino->i_lock) | |
e5e94017 BH |
1444 | { |
1445 | struct nfs_inode *nfsi = NFS_I(ino); | |
1446 | struct pnfs_layout_hdr *new = NULL; | |
1447 | ||
1448 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1449 | ||
251ec410 TM |
1450 | if (nfsi->layout != NULL) |
1451 | goto out_existing; | |
e5e94017 | 1452 | spin_unlock(&ino->i_lock); |
9fa40758 | 1453 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1454 | spin_lock(&ino->i_lock); |
1455 | ||
251ec410 | 1456 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1457 | nfsi->layout = new; |
251ec410 | 1458 | return new; |
7175fe90 YN |
1459 | } else if (new != NULL) |
1460 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1461 | out_existing: |
1462 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1463 | return nfsi->layout; |
1464 | } | |
1465 | ||
b1f69b75 AA |
1466 | /* |
1467 | * iomode matching rules: | |
c7d73af2 TH |
1468 | * iomode lseg strict match |
1469 | * iomode | |
1470 | * ----- ----- ------ ----- | |
1471 | * ANY READ N/A true | |
1472 | * ANY RW N/A true | |
1473 | * RW READ N/A false | |
1474 | * RW RW N/A true | |
1475 | * READ READ N/A true | |
1476 | * READ RW true false | |
1477 | * READ RW false true | |
b1f69b75 | 1478 | */ |
3cb2df17 | 1479 | static bool |
7dc0ac70 | 1480 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
c7d73af2 TH |
1481 | const struct pnfs_layout_range *range, |
1482 | bool strict_iomode) | |
b1f69b75 | 1483 | { |
fb3296eb BH |
1484 | struct pnfs_layout_range range1; |
1485 | ||
1486 | if ((range->iomode == IOMODE_RW && | |
1487 | ls_range->iomode != IOMODE_RW) || | |
c7d73af2 TH |
1488 | (range->iomode != ls_range->iomode && |
1489 | strict_iomode == true) || | |
7dc0ac70 | 1490 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1491 | return 0; |
1492 | ||
1493 | /* range1 covers only the first byte in the range */ | |
1494 | range1 = *range; | |
1495 | range1.length = 1; | |
7dc0ac70 | 1496 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1497 | } |
1498 | ||
1499 | /* | |
1500 | * lookup range in layout | |
1501 | */ | |
e5e94017 | 1502 | static struct pnfs_layout_segment * |
fb3296eb | 1503 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
c7d73af2 TH |
1504 | struct pnfs_layout_range *range, |
1505 | bool strict_iomode) | |
e5e94017 | 1506 | { |
b1f69b75 AA |
1507 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1508 | ||
1509 | dprintk("%s:Begin\n", __func__); | |
1510 | ||
b7edfaa1 | 1511 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1512 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1513 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
c7d73af2 TH |
1514 | pnfs_lseg_range_match(&lseg->pls_range, range, |
1515 | strict_iomode)) { | |
9369a431 | 1516 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1517 | break; |
1518 | } | |
b1f69b75 AA |
1519 | } |
1520 | ||
1521 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1522 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1523 | return ret; |
e5e94017 BH |
1524 | } |
1525 | ||
d23d61c8 AA |
1526 | /* |
1527 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1528 | * to the MDS or over pNFS | |
1529 | * | |
1530 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1531 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1532 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1533 | * WRITE request. | |
1534 | * | |
1535 | * A return of true means use MDS I/O. | |
1536 | * | |
1537 | * From rfc 5661: | |
1538 | * If a file's size is smaller than the file size threshold, data accesses | |
1539 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1540 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1541 | * server. If both file size and I/O size are provided, the client SHOULD | |
1542 | * reach or exceed both thresholds before sending its read or write | |
1543 | * requests to the data server. | |
1544 | */ | |
1545 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1546 | struct inode *ino, int iomode) | |
1547 | { | |
1548 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1549 | struct nfs_inode *nfsi = NFS_I(ino); | |
1550 | loff_t fsize = i_size_read(ino); | |
1551 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1552 | ||
1553 | if (t == NULL) | |
1554 | return ret; | |
1555 | ||
1556 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1557 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1558 | ||
1559 | switch (iomode) { | |
1560 | case IOMODE_READ: | |
1561 | if (t->bm & THRESHOLD_RD) { | |
1562 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1563 | size_set = true; | |
1564 | if (fsize < t->rd_sz) | |
1565 | size = true; | |
1566 | } | |
1567 | if (t->bm & THRESHOLD_RD_IO) { | |
1568 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1569 | nfsi->read_io); | |
1570 | io_set = true; | |
1571 | if (nfsi->read_io < t->rd_io_sz) | |
1572 | io = true; | |
1573 | } | |
1574 | break; | |
1575 | case IOMODE_RW: | |
1576 | if (t->bm & THRESHOLD_WR) { | |
1577 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1578 | size_set = true; | |
1579 | if (fsize < t->wr_sz) | |
1580 | size = true; | |
1581 | } | |
1582 | if (t->bm & THRESHOLD_WR_IO) { | |
1583 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1584 | nfsi->write_io); | |
1585 | io_set = true; | |
1586 | if (nfsi->write_io < t->wr_io_sz) | |
1587 | io = true; | |
1588 | } | |
1589 | break; | |
1590 | } | |
1591 | if (size_set && io_set) { | |
1592 | if (size && io) | |
1593 | ret = true; | |
1594 | } else if (size || io) | |
1595 | ret = true; | |
1596 | ||
1597 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1598 | return ret; | |
1599 | } | |
1600 | ||
aa8a45ee PT |
1601 | static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo) |
1602 | { | |
1603 | /* | |
1604 | * send layoutcommit as it can hold up layoutreturn due to lseg | |
1605 | * reference | |
1606 | */ | |
1607 | pnfs_layoutcommit_inode(lo->plh_inode, false); | |
1608 | return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
2e5b29f0 | 1609 | nfs_wait_bit_killable, |
aa8a45ee PT |
1610 | TASK_UNINTERRUPTIBLE); |
1611 | } | |
1612 | ||
d67ae825 TH |
1613 | static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo) |
1614 | { | |
1615 | unsigned long *bitlock = &lo->plh_flags; | |
1616 | ||
1617 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1618 | smp_mb__after_atomic(); | |
1619 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1620 | } | |
1621 | ||
e5e94017 BH |
1622 | /* |
1623 | * Layout segment is retreived from the server if not cached. | |
1624 | * The appropriate layout segment is referenced and returned to the caller. | |
1625 | */ | |
7c24d948 | 1626 | struct pnfs_layout_segment * |
e5e94017 BH |
1627 | pnfs_update_layout(struct inode *ino, |
1628 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1629 | loff_t pos, |
1630 | u64 count, | |
a75b9df9 | 1631 | enum pnfs_iomode iomode, |
c7d73af2 | 1632 | bool strict_iomode, |
a75b9df9 | 1633 | gfp_t gfp_flags) |
e5e94017 | 1634 | { |
fb3296eb BH |
1635 | struct pnfs_layout_range arg = { |
1636 | .iomode = iomode, | |
1637 | .offset = pos, | |
1638 | .length = count, | |
1639 | }; | |
183d9e7b | 1640 | unsigned pg_offset, seq; |
6382a441 WAA |
1641 | struct nfs_server *server = NFS_SERVER(ino); |
1642 | struct nfs_client *clp = server->nfs_client; | |
183d9e7b | 1643 | struct pnfs_layout_hdr *lo = NULL; |
e5e94017 | 1644 | struct pnfs_layout_segment *lseg = NULL; |
183d9e7b JL |
1645 | nfs4_stateid stateid; |
1646 | long timeout = 0; | |
66b53f32 | 1647 | unsigned long giveup = jiffies + (clp->cl_lease_time << 1); |
30005121 | 1648 | bool first; |
e5e94017 | 1649 | |
9a4bf31d | 1650 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) { |
183d9e7b | 1651 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1652 | PNFS_UPDATE_LAYOUT_NO_PNFS); |
f86bbcf8 | 1653 | goto out; |
9a4bf31d | 1654 | } |
d23d61c8 | 1655 | |
9a4bf31d | 1656 | if (iomode == IOMODE_READ && i_size_read(ino) == 0) { |
183d9e7b | 1657 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1658 | PNFS_UPDATE_LAYOUT_RD_ZEROLEN); |
4ae93560 | 1659 | goto out; |
9a4bf31d | 1660 | } |
4ae93560 | 1661 | |
9a4bf31d | 1662 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) { |
183d9e7b | 1663 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1664 | PNFS_UPDATE_LAYOUT_MDSTHRESH); |
f86bbcf8 | 1665 | goto out; |
9a4bf31d | 1666 | } |
d23d61c8 | 1667 | |
9bf87482 | 1668 | lookup_again: |
b88fa69e | 1669 | nfs4_client_recover_expired_lease(clp); |
9bf87482 | 1670 | first = false; |
e5e94017 | 1671 | spin_lock(&ino->i_lock); |
9fa40758 | 1672 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1673 | if (lo == NULL) { |
1674 | spin_unlock(&ino->i_lock); | |
183d9e7b | 1675 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1676 | PNFS_UPDATE_LAYOUT_NOMEM); |
830ffb56 TM |
1677 | goto out; |
1678 | } | |
e5e94017 | 1679 | |
43f1b3da | 1680 | /* Do we even need to bother with this? */ |
a59c30ac | 1681 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
183d9e7b | 1682 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1683 | PNFS_UPDATE_LAYOUT_BULK_RECALL); |
43f1b3da | 1684 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1685 | goto out_unlock; |
1686 | } | |
1687 | ||
1688 | /* if LAYOUTGET already failed once we don't try again */ | |
2e5b29f0 | 1689 | if (pnfs_layout_io_test_failed(lo, iomode)) { |
183d9e7b | 1690 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1691 | PNFS_UPDATE_LAYOUT_IO_TEST_FAIL); |
e5e94017 | 1692 | goto out_unlock; |
9a4bf31d | 1693 | } |
e5e94017 | 1694 | |
c7d73af2 | 1695 | lseg = pnfs_find_lseg(lo, &arg, strict_iomode); |
183d9e7b JL |
1696 | if (lseg) { |
1697 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1698 | PNFS_UPDATE_LAYOUT_FOUND_CACHED); | |
1699 | goto out_unlock; | |
1700 | } | |
1701 | ||
1702 | if (!nfs4_valid_open_stateid(ctx->state)) { | |
1703 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1704 | PNFS_UPDATE_LAYOUT_INVALID_OPEN); | |
1705 | goto out_unlock; | |
1706 | } | |
1707 | ||
1708 | /* | |
1709 | * Choose a stateid for the LAYOUTGET. If we don't have a layout | |
1710 | * stateid, or it has been invalidated, then we must use the open | |
1711 | * stateid. | |
1712 | */ | |
67a3b721 | 1713 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { |
183d9e7b JL |
1714 | |
1715 | /* | |
1716 | * The first layoutget for the file. Need to serialize per | |
9bf87482 PT |
1717 | * RFC 5661 Errata 3208. |
1718 | */ | |
1719 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1720 | &lo->plh_flags)) { | |
1721 | spin_unlock(&ino->i_lock); | |
1722 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1723 | TASK_UNINTERRUPTIBLE); | |
1724 | pnfs_put_layout_hdr(lo); | |
183d9e7b | 1725 | dprintk("%s retrying\n", __func__); |
9bf87482 PT |
1726 | goto lookup_again; |
1727 | } | |
183d9e7b JL |
1728 | |
1729 | first = true; | |
1730 | do { | |
1731 | seq = read_seqbegin(&ctx->state->seqlock); | |
1732 | nfs4_stateid_copy(&stateid, &ctx->state->stateid); | |
1733 | } while (read_seqretry(&ctx->state->seqlock, seq)); | |
9bf87482 | 1734 | } else { |
183d9e7b | 1735 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
9bf87482 | 1736 | } |
568e8c49 | 1737 | |
aa8a45ee PT |
1738 | /* |
1739 | * Because we free lsegs before sending LAYOUTRETURN, we need to wait | |
1740 | * for LAYOUTRETURN even if first is true. | |
1741 | */ | |
8f70f53a | 1742 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
aa8a45ee PT |
1743 | spin_unlock(&ino->i_lock); |
1744 | dprintk("%s wait for layoutreturn\n", __func__); | |
1745 | if (pnfs_prepare_to_retry_layoutget(lo)) { | |
d67ae825 TH |
1746 | if (first) |
1747 | pnfs_clear_first_layoutget(lo); | |
aa8a45ee PT |
1748 | pnfs_put_layout_hdr(lo); |
1749 | dprintk("%s retrying\n", __func__); | |
183d9e7b JL |
1750 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
1751 | lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
aa8a45ee PT |
1752 | goto lookup_again; |
1753 | } | |
183d9e7b | 1754 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1755 | PNFS_UPDATE_LAYOUT_RETURN); |
aa8a45ee PT |
1756 | goto out_put_layout_hdr; |
1757 | } | |
1758 | ||
9a4bf31d | 1759 | if (pnfs_layoutgets_blocked(lo)) { |
183d9e7b | 1760 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1761 | PNFS_UPDATE_LAYOUT_BLOCKED); |
cf7d63f1 | 1762 | goto out_unlock; |
9a4bf31d | 1763 | } |
cf7d63f1 | 1764 | atomic_inc(&lo->plh_outstanding); |
f49f9baa | 1765 | spin_unlock(&ino->i_lock); |
30005121 | 1766 | |
abb9a007 | 1767 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1768 | /* The lo must be on the clp list if there is any |
1769 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1770 | */ | |
1771 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1772 | if (list_empty(&lo->plh_layouts)) |
1773 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1774 | spin_unlock(&clp->cl_lock); |
1775 | } | |
e5e94017 | 1776 | |
09cbfeaf | 1777 | pg_offset = arg.offset & ~PAGE_MASK; |
707ed5fd BH |
1778 | if (pg_offset) { |
1779 | arg.offset -= pg_offset; | |
1780 | arg.length += pg_offset; | |
1781 | } | |
7c24d948 | 1782 | if (arg.length != NFS4_MAX_UINT64) |
09cbfeaf | 1783 | arg.length = PAGE_ALIGN(arg.length); |
707ed5fd | 1784 | |
183d9e7b JL |
1785 | lseg = send_layoutget(lo, ctx, &stateid, &arg, &timeout, gfp_flags); |
1786 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1787 | PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET); | |
56b38a1f | 1788 | atomic_dec(&lo->plh_outstanding); |
83026d80 | 1789 | if (IS_ERR(lseg)) { |
183d9e7b | 1790 | switch(PTR_ERR(lseg)) { |
e85d7ee4 | 1791 | case -EBUSY: |
183d9e7b JL |
1792 | if (time_after(jiffies, giveup)) |
1793 | lseg = NULL; | |
66b53f32 TM |
1794 | break; |
1795 | case -ERECALLCONFLICT: | |
1796 | /* Huh? We hold no layouts, how is there a recall? */ | |
1797 | if (first) { | |
1798 | lseg = NULL; | |
1799 | break; | |
1800 | } | |
1801 | /* Destroy the existing layout and start over */ | |
1802 | if (time_after(jiffies, giveup)) | |
1803 | pnfs_destroy_layout(NFS_I(ino)); | |
183d9e7b JL |
1804 | /* Fallthrough */ |
1805 | case -EAGAIN: | |
56b38a1f | 1806 | break; |
183d9e7b JL |
1807 | default: |
1808 | if (!nfs_error_is_fatal(PTR_ERR(lseg))) { | |
1809 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
1810 | lseg = NULL; | |
1811 | } | |
56b38a1f TM |
1812 | goto out_put_layout_hdr; |
1813 | } | |
1814 | if (lseg) { | |
1815 | if (first) | |
1816 | pnfs_clear_first_layoutget(lo); | |
1817 | trace_pnfs_update_layout(ino, pos, count, | |
1818 | iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
1819 | pnfs_put_layout_hdr(lo); | |
1820 | goto lookup_again; | |
83026d80 JL |
1821 | } |
1822 | } else { | |
1823 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
1824 | } | |
1825 | ||
830ffb56 | 1826 | out_put_layout_hdr: |
d67ae825 TH |
1827 | if (first) |
1828 | pnfs_clear_first_layoutget(lo); | |
70c3bd2b | 1829 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1830 | out: |
f86bbcf8 TM |
1831 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1832 | "(%s, offset: %llu, length: %llu)\n", | |
1833 | __func__, ino->i_sb->s_id, | |
1834 | (unsigned long long)NFS_FILEID(ino), | |
d600ad1f | 1835 | IS_ERR_OR_NULL(lseg) ? "not found" : "found", |
f86bbcf8 TM |
1836 | iomode==IOMODE_RW ? "read/write" : "read-only", |
1837 | (unsigned long long)pos, | |
1838 | (unsigned long long)count); | |
e5e94017 BH |
1839 | return lseg; |
1840 | out_unlock: | |
1841 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1842 | goto out_put_layout_hdr; |
e5e94017 | 1843 | } |
7c24d948 | 1844 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1845 | |
540d9864 TM |
1846 | static bool |
1847 | pnfs_sanity_check_layout_range(struct pnfs_layout_range *range) | |
1848 | { | |
1849 | switch (range->iomode) { | |
1850 | case IOMODE_READ: | |
1851 | case IOMODE_RW: | |
1852 | break; | |
1853 | default: | |
1854 | return false; | |
1855 | } | |
1856 | if (range->offset == NFS4_MAX_UINT64) | |
1857 | return false; | |
1858 | if (range->length == 0) | |
1859 | return false; | |
1860 | if (range->length != NFS4_MAX_UINT64 && | |
1861 | range->length > NFS4_MAX_UINT64 - range->offset) | |
1862 | return false; | |
1863 | return true; | |
1864 | } | |
1865 | ||
a0b0a6e3 | 1866 | struct pnfs_layout_segment * |
b1f69b75 AA |
1867 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1868 | { | |
1869 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1870 | struct nfs4_layoutget_res *res = &lgp->res; | |
1871 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1872 | struct inode *ino = lo->plh_inode; |
78096cca | 1873 | LIST_HEAD(free_me); |
540d9864 TM |
1874 | |
1875 | if (!pnfs_sanity_check_layout_range(&res->range)) | |
1b3c6d07 | 1876 | return ERR_PTR(-EINVAL); |
b1f69b75 AA |
1877 | |
1878 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1879 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
1b3c6d07 | 1880 | if (IS_ERR_OR_NULL(lseg)) { |
b1f69b75 | 1881 | if (!lseg) |
1b3c6d07 JL |
1882 | lseg = ERR_PTR(-ENOMEM); |
1883 | ||
1884 | dprintk("%s: Could not allocate layout: error %ld\n", | |
1885 | __func__, PTR_ERR(lseg)); | |
1886 | return lseg; | |
b1f69b75 AA |
1887 | } |
1888 | ||
119cef97 | 1889 | pnfs_init_lseg(lo, lseg, &res->range, &res->stateid); |
1013df61 | 1890 | |
b1f69b75 | 1891 | spin_lock(&ino->i_lock); |
e1c06f80 | 1892 | if (pnfs_layoutgets_blocked(lo)) { |
43f1b3da | 1893 | dprintk("%s forget reply due to state\n", __func__); |
1b3c6d07 | 1894 | goto out_forget; |
43f1b3da | 1895 | } |
038d6493 | 1896 | |
9888d837 TM |
1897 | if (!pnfs_layout_is_valid(lo)) { |
1898 | /* We have a completely new layout */ | |
1899 | pnfs_set_layout_stateid(lo, &res->stateid, true); | |
1900 | } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { | |
362f7474 CH |
1901 | /* existing state ID, make sure the sequence number matches. */ |
1902 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1903 | dprintk("%s forget reply due to sequence\n", __func__); | |
1b3c6d07 | 1904 | goto out_forget; |
362f7474 CH |
1905 | } |
1906 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1907 | } else { | |
1908 | /* | |
1909 | * We got an entirely new state ID. Mark all segments for the | |
9888d837 | 1910 | * inode invalid, and retry the layoutget |
362f7474 | 1911 | */ |
d9b61708 | 1912 | pnfs_mark_layout_stateid_invalid(lo, &free_me); |
9888d837 | 1913 | goto out_forget; |
362f7474 | 1914 | } |
038d6493 | 1915 | |
9369a431 | 1916 | pnfs_get_lseg(lseg); |
03772d2f | 1917 | pnfs_layout_insert_lseg(lo, lseg, &free_me); |
8e0acf90 | 1918 | |
b1f69b75 | 1919 | |
3976143b | 1920 | if (res->return_on_close) |
f7e8917a | 1921 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); |
f7e8917a | 1922 | |
b1f69b75 | 1923 | spin_unlock(&ino->i_lock); |
78096cca | 1924 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1925 | return lseg; |
43f1b3da | 1926 | |
1b3c6d07 | 1927 | out_forget: |
43f1b3da FI |
1928 | spin_unlock(&ino->i_lock); |
1929 | lseg->pls_layout = lo; | |
1930 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1b3c6d07 | 1931 | return ERR_PTR(-EAGAIN); |
b1f69b75 AA |
1932 | } |
1933 | ||
016256df | 1934 | static void |
3982a6a2 JL |
1935 | pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode, |
1936 | u32 seq) | |
5c97f5de | 1937 | { |
2d6cf5ab | 1938 | if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode) |
5c97f5de TM |
1939 | iomode = IOMODE_ANY; |
1940 | lo->plh_return_iomode = iomode; | |
e0fa0d01 | 1941 | set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); |
2d6cf5ab TM |
1942 | if (seq != 0) { |
1943 | WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq); | |
3982a6a2 | 1944 | lo->plh_return_seq = seq; |
2d6cf5ab | 1945 | } |
5c97f5de TM |
1946 | } |
1947 | ||
2f215968 TM |
1948 | /** |
1949 | * pnfs_mark_matching_lsegs_return - Free or return matching layout segments | |
1950 | * @lo: pointer to layout header | |
1951 | * @tmp_list: list header to be used with pnfs_free_lseg_list() | |
1952 | * @return_range: describe layout segment ranges to be returned | |
1953 | * | |
1954 | * This function is mainly intended for use by layoutrecall. It attempts | |
1955 | * to free the layout segment immediately, or else to mark it for return | |
1956 | * as soon as its reference count drops to zero. | |
1957 | */ | |
10335556 | 1958 | int |
016256df PT |
1959 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, |
1960 | struct list_head *tmp_list, | |
6d597e17 JL |
1961 | const struct pnfs_layout_range *return_range, |
1962 | u32 seq) | |
016256df PT |
1963 | { |
1964 | struct pnfs_layout_segment *lseg, *next; | |
10335556 | 1965 | int remaining = 0; |
016256df PT |
1966 | |
1967 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
1968 | ||
1969 | if (list_empty(&lo->plh_segs)) | |
10335556 | 1970 | return 0; |
016256df | 1971 | |
fc7ff367 | 1972 | assert_spin_locked(&lo->plh_inode->i_lock); |
016256df PT |
1973 | |
1974 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
e036f464 | 1975 | if (pnfs_match_lseg_recall(lseg, return_range, seq)) { |
016256df PT |
1976 | dprintk("%s: marking lseg %p iomode %d " |
1977 | "offset %llu length %llu\n", __func__, | |
1978 | lseg, lseg->pls_range.iomode, | |
1979 | lseg->pls_range.offset, | |
1980 | lseg->pls_range.length); | |
2f215968 TM |
1981 | if (mark_lseg_invalid(lseg, tmp_list)) |
1982 | continue; | |
1983 | remaining++; | |
016256df | 1984 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); |
016256df | 1985 | } |
6d597e17 JL |
1986 | |
1987 | if (remaining) | |
1988 | pnfs_set_plh_return_info(lo, return_range->iomode, seq); | |
1989 | ||
10335556 | 1990 | return remaining; |
016256df PT |
1991 | } |
1992 | ||
1993 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
1994 | struct pnfs_layout_segment *lseg) | |
1995 | { | |
1996 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
016256df PT |
1997 | struct pnfs_layout_range range = { |
1998 | .iomode = lseg->pls_range.iomode, | |
1999 | .offset = 0, | |
2000 | .length = NFS4_MAX_UINT64, | |
2001 | }; | |
10335556 | 2002 | bool return_now = false; |
016256df PT |
2003 | |
2004 | spin_lock(&inode->i_lock); | |
2d6cf5ab | 2005 | pnfs_set_plh_return_info(lo, range.iomode, 0); |
6604b203 TM |
2006 | /* Block LAYOUTGET */ |
2007 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
016256df PT |
2008 | /* |
2009 | * mark all matching lsegs so that we are sure to have no live | |
2010 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
2011 | * for how it works. | |
2012 | */ | |
68f74479 | 2013 | if (!pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0)) { |
10335556 | 2014 | nfs4_stateid stateid; |
e5fd1904 | 2015 | enum pnfs_iomode iomode; |
10335556 | 2016 | |
e5fd1904 | 2017 | return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode); |
10335556 TM |
2018 | spin_unlock(&inode->i_lock); |
2019 | if (return_now) | |
2020 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); | |
2021 | } else { | |
2022 | spin_unlock(&inode->i_lock); | |
2023 | nfs_commit_inode(inode, 0); | |
2024 | } | |
016256df PT |
2025 | } |
2026 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
2027 | ||
d8007d4d TM |
2028 | void |
2029 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
2030 | { | |
1fd937bd PT |
2031 | u64 rd_size = req->wb_bytes; |
2032 | ||
cb5d04bc PT |
2033 | if (pgio->pg_lseg == NULL) { |
2034 | if (pgio->pg_dreq == NULL) | |
2035 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
2036 | else | |
2037 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
2038 | ||
2039 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
2040 | req->wb_context, | |
2041 | req_offset(req), | |
2042 | rd_size, | |
2043 | IOMODE_READ, | |
c7d73af2 | 2044 | false, |
cb5d04bc | 2045 | GFP_KERNEL); |
d600ad1f PT |
2046 | if (IS_ERR(pgio->pg_lseg)) { |
2047 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
2048 | pgio->pg_lseg = NULL; | |
2049 | return; | |
2050 | } | |
cb5d04bc | 2051 | } |
e885de1a TM |
2052 | /* If no lseg, fall back to read through mds */ |
2053 | if (pgio->pg_lseg == NULL) | |
1f945357 | 2054 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 2055 | |
d8007d4d TM |
2056 | } |
2057 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
2058 | ||
2059 | void | |
6296556f PT |
2060 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
2061 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 2062 | { |
d600ad1f | 2063 | if (pgio->pg_lseg == NULL) { |
cb5d04bc PT |
2064 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
2065 | req->wb_context, | |
2066 | req_offset(req), | |
2067 | wb_size, | |
2068 | IOMODE_RW, | |
c7d73af2 | 2069 | false, |
cb5d04bc | 2070 | GFP_NOFS); |
d600ad1f PT |
2071 | if (IS_ERR(pgio->pg_lseg)) { |
2072 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
2073 | pgio->pg_lseg = NULL; | |
2074 | return; | |
2075 | } | |
2076 | } | |
e885de1a TM |
2077 | /* If no lseg, fall back to write through mds */ |
2078 | if (pgio->pg_lseg == NULL) | |
1f945357 | 2079 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
2080 | } |
2081 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
2082 | ||
180bb5ec WAA |
2083 | void |
2084 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
2085 | { | |
2086 | if (desc->pg_lseg) { | |
2087 | pnfs_put_lseg(desc->pg_lseg); | |
2088 | desc->pg_lseg = NULL; | |
2089 | } | |
2090 | } | |
2091 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
2092 | ||
b4fdac1a WAA |
2093 | /* |
2094 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
2095 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
2096 | */ | |
2097 | size_t | |
a7d42ddb WAA |
2098 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
2099 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 2100 | { |
0f9c429e | 2101 | unsigned int size; |
c5e20cb7 | 2102 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
2103 | |
2104 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
2105 | if (!size) |
2106 | return 0; | |
94ad1c80 | 2107 | |
19982ba8 | 2108 | /* |
c5e20cb7 WAA |
2109 | * 'size' contains the number of bytes left in the current page (up |
2110 | * to the original size asked for in @req->wb_bytes). | |
2111 | * | |
2112 | * Calculate how many bytes are left in the layout segment | |
2113 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
2114 | * |
2115 | * Please also note that 'end_offset' is actually the offset of the | |
2116 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
2117 | * | |
2118 | */ | |
19b54848 | 2119 | if (pgio->pg_lseg) { |
17822b20 | 2120 | seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset, |
c5e20cb7 WAA |
2121 | pgio->pg_lseg->pls_range.length); |
2122 | req_start = req_offset(req); | |
7c13789e | 2123 | WARN_ON_ONCE(req_start >= seg_end); |
c5e20cb7 | 2124 | /* start of request is past the last byte of this segment */ |
7c13789e WAA |
2125 | if (req_start >= seg_end) { |
2126 | /* reference the new lseg */ | |
2127 | if (pgio->pg_ops->pg_cleanup) | |
2128 | pgio->pg_ops->pg_cleanup(pgio); | |
2129 | if (pgio->pg_ops->pg_init) | |
2130 | pgio->pg_ops->pg_init(pgio, req); | |
19b54848 | 2131 | return 0; |
7c13789e | 2132 | } |
c5e20cb7 WAA |
2133 | |
2134 | /* adjust 'size' iff there are fewer bytes left in the | |
2135 | * segment than what nfs_generic_pg_test returned */ | |
2136 | seg_left = seg_end - req_start; | |
2137 | if (seg_left < size) | |
2138 | size = (unsigned int)seg_left; | |
19b54848 | 2139 | } |
0f9c429e | 2140 | |
19b54848 | 2141 | return size; |
94ad1c80 | 2142 | } |
89a58e32 | 2143 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 2144 | |
53113ad3 | 2145 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
2146 | { |
2147 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
2148 | |
2149 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
2150 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
2151 | hdr->completion_ops); | |
c7070113 | 2152 | set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags); |
53113ad3 | 2153 | return nfs_pageio_resend(&pgio, hdr); |
e2fecb21 | 2154 | } |
e7dd79af | 2155 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 2156 | |
d45f60c6 | 2157 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2158 | { |
cd841605 FI |
2159 | |
2160 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
2161 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2162 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2163 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2164 | } |
6c75dc0d | 2165 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2166 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
2167 | } |
2168 | ||
d20581aa BH |
2169 | /* |
2170 | * Called by non rpc-based layout drivers | |
2171 | */ | |
d45f60c6 | 2172 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 2173 | { |
f8417b48 | 2174 | if (likely(!hdr->pnfs_error)) { |
67af7611 TM |
2175 | pnfs_set_layoutcommit(hdr->inode, hdr->lseg, |
2176 | hdr->mds_offset + hdr->res.count); | |
d45f60c6 | 2177 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2178 | } |
2179 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); | |
2180 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2181 | pnfs_ld_handle_write_error(hdr); |
2182 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 2183 | } |
d20581aa | 2184 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 2185 | |
dce81290 TM |
2186 | static void |
2187 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2188 | struct nfs_pgio_header *hdr) |
dce81290 | 2189 | { |
48d635f1 | 2190 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2191 | |
6c75dc0d | 2192 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2193 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 2194 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 2195 | mirror->pg_recoalesce = 1; |
6c75dc0d | 2196 | } |
d45f60c6 | 2197 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2198 | hdr->release(hdr); |
dce81290 TM |
2199 | } |
2200 | ||
2201 | static enum pnfs_try_status | |
d45f60c6 | 2202 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
2203 | const struct rpc_call_ops *call_ops, |
2204 | struct pnfs_layout_segment *lseg, | |
2205 | int how) | |
0382b744 | 2206 | { |
cd841605 | 2207 | struct inode *inode = hdr->inode; |
0382b744 AA |
2208 | enum pnfs_try_status trypnfs; |
2209 | struct nfs_server *nfss = NFS_SERVER(inode); | |
2210 | ||
cd841605 | 2211 | hdr->mds_ops = call_ops; |
0382b744 AA |
2212 | |
2213 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
2214 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
2215 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 2216 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 2217 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
2218 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2219 | return trypnfs; | |
2220 | } | |
2221 | ||
dce81290 | 2222 | static void |
7f714720 WAA |
2223 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
2224 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 2225 | { |
dce81290 TM |
2226 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2227 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2228 | enum pnfs_try_status trypnfs; |
dce81290 | 2229 | |
d45f60c6 | 2230 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
7f714720 | 2231 | if (trypnfs == PNFS_NOT_ATTEMPTED) |
d45f60c6 | 2232 | pnfs_write_through_mds(desc, hdr); |
dce81290 TM |
2233 | } |
2234 | ||
6c75dc0d FI |
2235 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
2236 | { | |
9369a431 | 2237 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2238 | nfs_pgio_header_free(hdr); |
6c75dc0d FI |
2239 | } |
2240 | ||
dce81290 TM |
2241 | int |
2242 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
2243 | { | |
6c75dc0d | 2244 | struct nfs_pgio_header *hdr; |
dce81290 TM |
2245 | int ret; |
2246 | ||
1e7f3a48 WAA |
2247 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2248 | if (!hdr) { | |
2bff2288 PT |
2249 | desc->pg_error = -ENOMEM; |
2250 | return desc->pg_error; | |
dce81290 | 2251 | } |
6c75dc0d | 2252 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 2253 | |
9369a431 | 2254 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2255 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2256 | if (!ret) |
7f714720 | 2257 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 2258 | |
6c75dc0d | 2259 | return ret; |
dce81290 TM |
2260 | } |
2261 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
2262 | ||
53113ad3 | 2263 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
2264 | { |
2265 | struct nfs_pageio_descriptor pgio; | |
2266 | ||
1acbbb4e | 2267 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
2268 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
2269 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 2270 | } |
e7dd79af | 2271 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 2272 | |
d45f60c6 | 2273 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2274 | { |
cd841605 FI |
2275 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
2276 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2277 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2278 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2279 | } |
4db6e0b7 | 2280 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2281 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
2282 | } |
2283 | ||
d20581aa BH |
2284 | /* |
2285 | * Called by non rpc-based layout drivers | |
2286 | */ | |
d45f60c6 | 2287 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 2288 | { |
bfc505de | 2289 | if (likely(!hdr->pnfs_error)) |
d45f60c6 | 2290 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2291 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
2292 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2293 | pnfs_ld_handle_read_error(hdr); |
2294 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
2295 | } |
2296 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
2297 | ||
493292dd TM |
2298 | static void |
2299 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2300 | struct nfs_pgio_header *hdr) |
493292dd | 2301 | { |
48d635f1 | 2302 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2303 | |
4db6e0b7 | 2304 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2305 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 2306 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 2307 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 2308 | } |
d45f60c6 | 2309 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2310 | hdr->release(hdr); |
493292dd TM |
2311 | } |
2312 | ||
64419a9b AA |
2313 | /* |
2314 | * Call the appropriate parallel I/O subsystem read function. | |
2315 | */ | |
493292dd | 2316 | static enum pnfs_try_status |
d45f60c6 | 2317 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
2318 | const struct rpc_call_ops *call_ops, |
2319 | struct pnfs_layout_segment *lseg) | |
64419a9b | 2320 | { |
cd841605 | 2321 | struct inode *inode = hdr->inode; |
64419a9b AA |
2322 | struct nfs_server *nfss = NFS_SERVER(inode); |
2323 | enum pnfs_try_status trypnfs; | |
2324 | ||
cd841605 | 2325 | hdr->mds_ops = call_ops; |
64419a9b AA |
2326 | |
2327 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 2328 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 2329 | |
d45f60c6 | 2330 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 2331 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 2332 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
2333 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2334 | return trypnfs; | |
2335 | } | |
863a3c6c | 2336 | |
ceb11e13 | 2337 | /* Resend all requests through pnfs. */ |
1b1bc66b | 2338 | void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) |
ceb11e13 PT |
2339 | { |
2340 | struct nfs_pageio_descriptor pgio; | |
2341 | ||
1b1bc66b | 2342 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
54e4a0df TM |
2343 | /* Prevent deadlocks with layoutreturn! */ |
2344 | pnfs_put_lseg(hdr->lseg); | |
2345 | hdr->lseg = NULL; | |
2346 | ||
1b1bc66b WAA |
2347 | nfs_pageio_init_read(&pgio, hdr->inode, false, |
2348 | hdr->completion_ops); | |
2349 | hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr); | |
2350 | } | |
ceb11e13 PT |
2351 | } |
2352 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
2353 | ||
493292dd | 2354 | static void |
7f714720 | 2355 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 2356 | { |
493292dd TM |
2357 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2358 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2359 | enum pnfs_try_status trypnfs; |
493292dd | 2360 | |
d45f60c6 | 2361 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
ceb11e13 | 2362 | if (trypnfs == PNFS_TRY_AGAIN) |
1b1bc66b WAA |
2363 | pnfs_read_resend_pnfs(hdr); |
2364 | if (trypnfs == PNFS_NOT_ATTEMPTED || hdr->task.tk_status) | |
d45f60c6 | 2365 | pnfs_read_through_mds(desc, hdr); |
493292dd TM |
2366 | } |
2367 | ||
4db6e0b7 FI |
2368 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
2369 | { | |
9369a431 | 2370 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2371 | nfs_pgio_header_free(hdr); |
4db6e0b7 FI |
2372 | } |
2373 | ||
493292dd TM |
2374 | int |
2375 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
2376 | { | |
4db6e0b7 | 2377 | struct nfs_pgio_header *hdr; |
493292dd TM |
2378 | int ret; |
2379 | ||
1e7f3a48 WAA |
2380 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2381 | if (!hdr) { | |
2bff2288 PT |
2382 | desc->pg_error = -ENOMEM; |
2383 | return desc->pg_error; | |
493292dd | 2384 | } |
4db6e0b7 | 2385 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 2386 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2387 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2388 | if (!ret) |
7f714720 | 2389 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 2390 | return ret; |
493292dd TM |
2391 | } |
2392 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
2393 | ||
71244d9b TM |
2394 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
2395 | { | |
2396 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
2397 | ||
2398 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 2399 | smp_mb__after_atomic(); |
71244d9b TM |
2400 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
2401 | } | |
2402 | ||
863a3c6c | 2403 | /* |
a9bae566 | 2404 | * There can be multiple RW segments. |
863a3c6c | 2405 | */ |
a9bae566 | 2406 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 2407 | { |
a9bae566 | 2408 | struct pnfs_layout_segment *lseg; |
863a3c6c | 2409 | |
a9bae566 PT |
2410 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
2411 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 2412 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
2413 | list_add(&lseg->pls_lc_list, listp); |
2414 | } | |
863a3c6c AA |
2415 | } |
2416 | ||
a073dbff TM |
2417 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
2418 | { | |
2419 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
2420 | |
2421 | /* Matched by references in pnfs_set_layoutcommit */ | |
2422 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
2423 | list_del_init(&lseg->pls_lc_list); | |
2424 | pnfs_put_lseg(lseg); | |
2425 | } | |
2426 | ||
71244d9b | 2427 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
2428 | } |
2429 | ||
1b0ae068 PT |
2430 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
2431 | { | |
b9e028fd | 2432 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
2433 | } |
2434 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
2435 | ||
863a3c6c | 2436 | void |
67af7611 TM |
2437 | pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, |
2438 | loff_t end_pos) | |
863a3c6c | 2439 | { |
cd841605 | 2440 | struct nfs_inode *nfsi = NFS_I(inode); |
79a48a1f | 2441 | bool mark_as_dirty = false; |
863a3c6c | 2442 | |
cd841605 | 2443 | spin_lock(&inode->i_lock); |
863a3c6c | 2444 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
29559b11 | 2445 | nfsi->layout->plh_lwb = end_pos; |
79a48a1f | 2446 | mark_as_dirty = true; |
863a3c6c | 2447 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 2448 | __func__, inode->i_ino); |
29559b11 TM |
2449 | } else if (end_pos > nfsi->layout->plh_lwb) |
2450 | nfsi->layout->plh_lwb = end_pos; | |
67af7611 | 2451 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) { |
a9bae566 | 2452 | /* references matched in nfs4_layoutcommit_release */ |
67af7611 | 2453 | pnfs_get_lseg(lseg); |
a9bae566 | 2454 | } |
cd841605 | 2455 | spin_unlock(&inode->i_lock); |
acff5880 | 2456 | dprintk("%s: lseg %p end_pos %llu\n", |
67af7611 | 2457 | __func__, lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
2458 | |
2459 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2460 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2461 | if (mark_as_dirty) | |
cd841605 | 2462 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
2463 | } |
2464 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
2465 | ||
db29c089 AA |
2466 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2467 | { | |
2468 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2469 | ||
2470 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2471 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2472 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2473 | } |
2474 | ||
de4b15c7 AA |
2475 | /* |
2476 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2477 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2478 | * data to disk to allow the server to recover the data if it crashes. | |
2479 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2480 | * is off, and a COMMIT is sent to a data server, or | |
2481 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2482 | */ | |
863a3c6c | 2483 | int |
ef311537 | 2484 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2485 | { |
5f919c9f | 2486 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2487 | struct nfs4_layoutcommit_data *data; |
2488 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2489 | loff_t end_pos; |
71244d9b | 2490 | int status; |
863a3c6c | 2491 | |
71244d9b | 2492 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2493 | return 0; |
2494 | ||
71244d9b | 2495 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2496 | |
71244d9b | 2497 | status = -EAGAIN; |
92407e75 | 2498 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2499 | if (!sync) |
2500 | goto out; | |
74316201 | 2501 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2502 | NFS_INO_LAYOUTCOMMITTING, |
2503 | nfs_wait_bit_killable, | |
2504 | TASK_KILLABLE); | |
92407e75 | 2505 | if (status) |
71244d9b | 2506 | goto out; |
92407e75 PT |
2507 | } |
2508 | ||
71244d9b TM |
2509 | status = -ENOMEM; |
2510 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2511 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2512 | if (!data) | |
2513 | goto clear_layoutcommitting; | |
2514 | ||
2515 | status = 0; | |
de4b15c7 | 2516 | spin_lock(&inode->i_lock); |
71244d9b TM |
2517 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2518 | goto out_unlock; | |
a9bae566 | 2519 | |
71244d9b | 2520 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2521 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2522 | |
acff5880 | 2523 | end_pos = nfsi->layout->plh_lwb; |
863a3c6c | 2524 | |
f597c537 | 2525 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2526 | spin_unlock(&inode->i_lock); |
2527 | ||
2528 | data->args.inode = inode; | |
9fa40758 | 2529 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2530 | nfs_fattr_init(&data->fattr); |
2531 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2532 | data->res.fattr = &data->fattr; | |
2e18d4d8 TM |
2533 | if (end_pos != 0) |
2534 | data->args.lastbytewritten = end_pos - 1; | |
2535 | else | |
2536 | data->args.lastbytewritten = U64_MAX; | |
863a3c6c AA |
2537 | data->res.server = NFS_SERVER(inode); |
2538 | ||
5f919c9f CH |
2539 | if (ld->prepare_layoutcommit) { |
2540 | status = ld->prepare_layoutcommit(&data->args); | |
2541 | if (status) { | |
3471648a | 2542 | put_rpccred(data->cred); |
5f919c9f | 2543 | spin_lock(&inode->i_lock); |
29559b11 TM |
2544 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); |
2545 | if (end_pos > nfsi->layout->plh_lwb) | |
5f919c9f | 2546 | nfsi->layout->plh_lwb = end_pos; |
3471648a | 2547 | goto out_unlock; |
5f919c9f CH |
2548 | } |
2549 | } | |
2550 | ||
2551 | ||
863a3c6c AA |
2552 | status = nfs4_proc_layoutcommit(data, sync); |
2553 | out: | |
92407e75 PT |
2554 | if (status) |
2555 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2556 | dprintk("<-- %s status %d\n", __func__, status); |
2557 | return status; | |
71244d9b TM |
2558 | out_unlock: |
2559 | spin_unlock(&inode->i_lock); | |
92407e75 | 2560 | kfree(data); |
71244d9b TM |
2561 | clear_layoutcommitting: |
2562 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2563 | goto out; |
863a3c6c | 2564 | } |
72cff449 | 2565 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a | 2566 | |
5bb89b47 TM |
2567 | int |
2568 | pnfs_generic_sync(struct inode *inode, bool datasync) | |
2569 | { | |
2570 | return pnfs_layoutcommit_inode(inode, true); | |
2571 | } | |
2572 | EXPORT_SYMBOL_GPL(pnfs_generic_sync); | |
2573 | ||
82be417a AA |
2574 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) |
2575 | { | |
2576 | struct nfs4_threshold *thp; | |
2577 | ||
2578 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2579 | if (!thp) { | |
2580 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2581 | return NULL; | |
2582 | } | |
2583 | return thp; | |
2584 | } | |
8733408d | 2585 | |
865a7ecb | 2586 | #if IS_ENABLED(CONFIG_NFS_V4_2) |
8733408d | 2587 | int |
c8ad8894 | 2588 | pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags) |
8733408d PT |
2589 | { |
2590 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
2591 | struct nfs_server *server = NFS_SERVER(inode); | |
1bfe3b25 | 2592 | struct nfs_inode *nfsi = NFS_I(inode); |
8733408d PT |
2593 | struct nfs42_layoutstat_data *data; |
2594 | struct pnfs_layout_hdr *hdr; | |
2595 | int status = 0; | |
2596 | ||
2597 | if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats) | |
2598 | goto out; | |
2599 | ||
6c5a0d89 TM |
2600 | if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS)) |
2601 | goto out; | |
2602 | ||
1bfe3b25 PT |
2603 | if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags)) |
2604 | goto out; | |
2605 | ||
8733408d PT |
2606 | spin_lock(&inode->i_lock); |
2607 | if (!NFS_I(inode)->layout) { | |
2608 | spin_unlock(&inode->i_lock); | |
f538d0ba | 2609 | goto out_clear_layoutstats; |
8733408d PT |
2610 | } |
2611 | hdr = NFS_I(inode)->layout; | |
2612 | pnfs_get_layout_hdr(hdr); | |
2613 | spin_unlock(&inode->i_lock); | |
2614 | ||
c8ad8894 | 2615 | data = kzalloc(sizeof(*data), gfp_flags); |
8733408d PT |
2616 | if (!data) { |
2617 | status = -ENOMEM; | |
2618 | goto out_put; | |
2619 | } | |
2620 | ||
2621 | data->args.fh = NFS_FH(inode); | |
2622 | data->args.inode = inode; | |
8733408d PT |
2623 | status = ld->prepare_layoutstats(&data->args); |
2624 | if (status) | |
2625 | goto out_free; | |
2626 | ||
2627 | status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data); | |
2628 | ||
2629 | out: | |
2630 | dprintk("%s returns %d\n", __func__, status); | |
2631 | return status; | |
2632 | ||
2633 | out_free: | |
2634 | kfree(data); | |
2635 | out_put: | |
2636 | pnfs_put_layout_hdr(hdr); | |
f538d0ba | 2637 | out_clear_layoutstats: |
1bfe3b25 PT |
2638 | smp_mb__before_atomic(); |
2639 | clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags); | |
2640 | smp_mb__after_atomic(); | |
8733408d PT |
2641 | goto out; |
2642 | } | |
2643 | EXPORT_SYMBOL_GPL(pnfs_report_layoutstat); | |
865a7ecb | 2644 | #endif |
bbf58bf3 TM |
2645 | |
2646 | unsigned int layoutstats_timer; | |
2647 | module_param(layoutstats_timer, uint, 0644); | |
2648 | EXPORT_SYMBOL_GPL(layoutstats_timer); |