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); | |
fe1cf946 TM |
61 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
62 | struct list_head *tmp_list); | |
aa1e0e3a | 63 | |
02c35fca FI |
64 | /* Return the registered pnfs layout driver module matching given id */ |
65 | static struct pnfs_layoutdriver_type * | |
66 | find_pnfs_driver_locked(u32 id) | |
67 | { | |
68 | struct pnfs_layoutdriver_type *local; | |
69 | ||
70 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
71 | if (local->id == id) | |
72 | goto out; | |
73 | local = NULL; | |
74 | out: | |
75 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
76 | return local; | |
77 | } | |
78 | ||
85e174ba RL |
79 | static struct pnfs_layoutdriver_type * |
80 | find_pnfs_driver(u32 id) | |
81 | { | |
02c35fca FI |
82 | struct pnfs_layoutdriver_type *local; |
83 | ||
84 | spin_lock(&pnfs_spinlock); | |
85 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
86 | if (local != NULL && !try_module_get(local->owner)) { |
87 | dprintk("%s: Could not grab reference on module\n", __func__); | |
88 | local = NULL; | |
89 | } | |
02c35fca FI |
90 | spin_unlock(&pnfs_spinlock); |
91 | return local; | |
85e174ba RL |
92 | } |
93 | ||
94 | void | |
95 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
96 | { | |
738fd0f3 BH |
97 | if (nfss->pnfs_curr_ld) { |
98 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
99 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
100 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
101 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
102 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 103 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 104 | } |
85e174ba RL |
105 | nfss->pnfs_curr_ld = NULL; |
106 | } | |
107 | ||
ca440c38 JL |
108 | /* |
109 | * When the server sends a list of layout types, we choose one in the order | |
110 | * given in the list below. | |
111 | * | |
112 | * FIXME: should this list be configurable in some fashion? module param? | |
113 | * mount option? something else? | |
114 | */ | |
115 | static const u32 ld_prefs[] = { | |
116 | LAYOUT_SCSI, | |
117 | LAYOUT_BLOCK_VOLUME, | |
118 | LAYOUT_OSD2_OBJECTS, | |
119 | LAYOUT_FLEX_FILES, | |
120 | LAYOUT_NFSV4_1_FILES, | |
121 | 0 | |
122 | }; | |
123 | ||
124 | static int | |
125 | ld_cmp(const void *e1, const void *e2) | |
126 | { | |
127 | u32 ld1 = *((u32 *)e1); | |
128 | u32 ld2 = *((u32 *)e2); | |
129 | int i; | |
130 | ||
131 | for (i = 0; ld_prefs[i] != 0; i++) { | |
132 | if (ld1 == ld_prefs[i]) | |
133 | return -1; | |
134 | ||
135 | if (ld2 == ld_prefs[i]) | |
136 | return 1; | |
137 | } | |
138 | return 0; | |
139 | } | |
140 | ||
85e174ba RL |
141 | /* |
142 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
143 | * Currently only one pNFS layout driver per filesystem is supported. | |
144 | * | |
3132e49e | 145 | * @ids array of layout types supported by MDS. |
85e174ba RL |
146 | */ |
147 | void | |
738fd0f3 | 148 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
ca440c38 | 149 | struct nfs_fsinfo *fsinfo) |
85e174ba RL |
150 | { |
151 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
3132e49e | 152 | u32 id; |
ca440c38 | 153 | int i; |
85e174ba | 154 | |
19274716 AS |
155 | if (fsinfo->nlayouttypes == 0) |
156 | goto out_no_driver; | |
85e174ba RL |
157 | if (!(server->nfs_client->cl_exchange_flags & |
158 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
3132e49e JL |
159 | printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n", |
160 | __func__, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
161 | goto out_no_driver; |
162 | } | |
3132e49e | 163 | |
ca440c38 JL |
164 | sort(fsinfo->layouttype, fsinfo->nlayouttypes, |
165 | sizeof(*fsinfo->layouttype), ld_cmp, NULL); | |
3132e49e | 166 | |
ca440c38 JL |
167 | for (i = 0; i < fsinfo->nlayouttypes; i++) { |
168 | id = fsinfo->layouttype[i]; | |
85e174ba | 169 | ld_type = find_pnfs_driver(id); |
ca440c38 JL |
170 | if (!ld_type) { |
171 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, | |
172 | id); | |
173 | ld_type = find_pnfs_driver(id); | |
174 | } | |
175 | if (ld_type) | |
176 | break; | |
85e174ba | 177 | } |
3132e49e JL |
178 | |
179 | if (!ld_type) { | |
ca440c38 | 180 | dprintk("%s: No pNFS module found!\n", __func__); |
3132e49e JL |
181 | goto out_no_driver; |
182 | } | |
183 | ||
85e174ba | 184 | server->pnfs_curr_ld = ld_type; |
738fd0f3 BH |
185 | if (ld_type->set_layoutdriver |
186 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
187 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
188 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
189 | module_put(ld_type->owner); |
190 | goto out_no_driver; | |
191 | } | |
2a4c8994 TM |
192 | /* Bump the MDS count */ |
193 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 194 | |
85e174ba RL |
195 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
196 | return; | |
197 | ||
198 | out_no_driver: | |
199 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
200 | server->pnfs_curr_ld = NULL; | |
201 | } | |
02c35fca FI |
202 | |
203 | int | |
204 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
205 | { | |
206 | int status = -EINVAL; | |
207 | struct pnfs_layoutdriver_type *tmp; | |
208 | ||
209 | if (ld_type->id == 0) { | |
a030889a | 210 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
211 | return status; |
212 | } | |
b1f69b75 | 213 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 214 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
215 | "alloc_lseg and free_lseg.\n", __func__); |
216 | return status; | |
217 | } | |
02c35fca FI |
218 | |
219 | spin_lock(&pnfs_spinlock); | |
220 | tmp = find_pnfs_driver_locked(ld_type->id); | |
221 | if (!tmp) { | |
222 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
223 | status = 0; | |
224 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
225 | ld_type->name); | |
226 | } else { | |
a030889a | 227 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
228 | __func__, ld_type->id); |
229 | } | |
230 | spin_unlock(&pnfs_spinlock); | |
231 | ||
232 | return status; | |
233 | } | |
234 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
235 | ||
236 | void | |
237 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
238 | { | |
239 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
240 | spin_lock(&pnfs_spinlock); | |
241 | list_del(&ld_type->pnfs_tblid); | |
242 | spin_unlock(&pnfs_spinlock); | |
243 | } | |
244 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 245 | |
b1f69b75 AA |
246 | /* |
247 | * pNFS client layout cache | |
248 | */ | |
249 | ||
cc6e5340 | 250 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 251 | void |
70c3bd2b | 252 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 253 | { |
cc6e5340 | 254 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
255 | } |
256 | ||
636fb9c8 BH |
257 | static struct pnfs_layout_hdr * |
258 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
259 | { | |
260 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
57934278 | 261 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
262 | } |
263 | ||
264 | static void | |
265 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
266 | { | |
9c626381 TM |
267 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
268 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
269 | ||
270 | if (!list_empty(&lo->plh_layouts)) { | |
271 | struct nfs_client *clp = server->nfs_client; | |
272 | ||
273 | spin_lock(&clp->cl_lock); | |
274 | list_del_init(&lo->plh_layouts); | |
275 | spin_unlock(&clp->cl_lock); | |
276 | } | |
9fa40758 | 277 | put_rpccred(lo->plh_lc_cred); |
57934278 | 278 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
279 | } |
280 | ||
e5e94017 | 281 | static void |
6622c3ea | 282 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 283 | { |
bb346f63 | 284 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 285 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
286 | nfsi->layout = NULL; |
287 | /* Reset MDS Threshold I/O counters */ | |
288 | nfsi->write_io = 0; | |
289 | nfsi->read_io = 0; | |
e5e94017 BH |
290 | } |
291 | ||
b1f69b75 | 292 | void |
70c3bd2b | 293 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 294 | { |
cc6e5340 FI |
295 | struct inode *inode = lo->plh_inode; |
296 | ||
13c13a6a TM |
297 | pnfs_layoutreturn_before_put_layout_hdr(lo); |
298 | ||
cc6e5340 | 299 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { |
566f8737 PT |
300 | if (!list_empty(&lo->plh_segs)) |
301 | WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n"); | |
6622c3ea | 302 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 303 | spin_unlock(&inode->i_lock); |
6622c3ea | 304 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 305 | } |
974cec8c AA |
306 | } |
307 | ||
4aab9732 TM |
308 | static void |
309 | pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode, | |
310 | u32 seq) | |
311 | { | |
312 | if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode) | |
313 | iomode = IOMODE_ANY; | |
314 | lo->plh_return_iomode = iomode; | |
315 | set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); | |
316 | if (seq != 0) { | |
317 | WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq); | |
318 | lo->plh_return_seq = seq; | |
319 | } | |
320 | } | |
321 | ||
ae5a459d TM |
322 | static void |
323 | pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo) | |
324 | { | |
325 | lo->plh_return_iomode = 0; | |
326 | lo->plh_return_seq = 0; | |
327 | clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); | |
328 | } | |
329 | ||
362fb578 TM |
330 | static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo) |
331 | { | |
332 | clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
333 | clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags); | |
334 | smp_mb__after_atomic(); | |
335 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN); | |
336 | rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq); | |
337 | } | |
338 | ||
fe1cf946 TM |
339 | static void |
340 | pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg, | |
341 | struct list_head *free_me) | |
342 | { | |
343 | clear_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
344 | clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
345 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) | |
346 | pnfs_lseg_dec_and_remove_zero(lseg, free_me); | |
347 | if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
348 | pnfs_lseg_dec_and_remove_zero(lseg, free_me); | |
349 | } | |
350 | ||
2454dfea TM |
351 | /* |
352 | * Mark a pnfs_layout_hdr and all associated layout segments as invalid | |
353 | * | |
354 | * In order to continue using the pnfs_layout_hdr, a full recovery | |
355 | * is required. | |
356 | * Note that caller must hold inode->i_lock. | |
357 | */ | |
5f46be04 | 358 | int |
2454dfea TM |
359 | pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo, |
360 | struct list_head *lseg_list) | |
361 | { | |
362 | struct pnfs_layout_range range = { | |
363 | .iomode = IOMODE_ANY, | |
364 | .offset = 0, | |
365 | .length = NFS4_MAX_UINT64, | |
366 | }; | |
fe1cf946 | 367 | struct pnfs_layout_segment *lseg, *next; |
2454dfea TM |
368 | |
369 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
ae5a459d | 370 | pnfs_clear_layoutreturn_info(lo); |
fe1cf946 TM |
371 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
372 | pnfs_clear_lseg_state(lseg, lseg_list); | |
68f74479 | 373 | pnfs_free_returned_lsegs(lo, lseg_list, &range, 0); |
362fb578 TM |
374 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) && |
375 | !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) | |
376 | pnfs_clear_layoutreturn_waitbit(lo); | |
fe1cf946 | 377 | return !list_empty(&lo->plh_segs); |
2454dfea TM |
378 | } |
379 | ||
b9e028fd TM |
380 | static int |
381 | pnfs_iomode_to_fail_bit(u32 iomode) | |
382 | { | |
383 | return iomode == IOMODE_RW ? | |
384 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
385 | } | |
386 | ||
387 | static void | |
3e621214 | 388 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 389 | { |
25c75333 | 390 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 391 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
3e621214 TM |
392 | atomic_inc(&lo->plh_refcount); |
393 | } | |
394 | ||
395 | static void | |
396 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
397 | { | |
398 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
399 | atomic_dec(&lo->plh_refcount); | |
400 | } | |
401 | ||
402 | static void | |
403 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
404 | { | |
405 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
406 | struct pnfs_layout_range range = { |
407 | .iomode = iomode, | |
408 | .offset = 0, | |
409 | .length = NFS4_MAX_UINT64, | |
410 | }; | |
411 | LIST_HEAD(head); | |
3e621214 TM |
412 | |
413 | spin_lock(&inode->i_lock); | |
414 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
6d597e17 | 415 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0); |
3e621214 | 416 | spin_unlock(&inode->i_lock); |
115ce575 | 417 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
418 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
419 | iomode == IOMODE_RW ? "RW" : "READ"); | |
420 | } | |
421 | ||
422 | static bool | |
423 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
424 | { | |
25c75333 | 425 | unsigned long start, end; |
3e621214 TM |
426 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
427 | ||
428 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
429 | return false; |
430 | end = jiffies; | |
431 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
432 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
433 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 434 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
435 | return false; |
436 | } | |
437 | return true; | |
b9e028fd TM |
438 | } |
439 | ||
974cec8c | 440 | static void |
119cef97 TM |
441 | pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg, |
442 | const struct pnfs_layout_range *range, | |
443 | const nfs4_stateid *stateid) | |
974cec8c | 444 | { |
566052c5 | 445 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 446 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c | 447 | atomic_set(&lseg->pls_refcount, 1); |
4541d16c | 448 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); |
566052c5 | 449 | lseg->pls_layout = lo; |
119cef97 TM |
450 | lseg->pls_range = *range; |
451 | lseg->pls_seq = be32_to_cpu(stateid->seqid); | |
974cec8c AA |
452 | } |
453 | ||
905ca191 | 454 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 455 | { |
68f74479 TM |
456 | if (lseg != NULL) { |
457 | struct inode *inode = lseg->pls_layout->plh_inode; | |
458 | NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg); | |
459 | } | |
974cec8c AA |
460 | } |
461 | ||
d684d2ae | 462 | static void |
57036a37 TM |
463 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
464 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 465 | { |
d20581aa | 466 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 467 | list_del_init(&lseg->pls_list); |
8f0d27dc TM |
468 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
469 | atomic_dec(&lo->plh_refcount); | |
abb3e1c8 TM |
470 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) |
471 | return; | |
7b650994 TM |
472 | if (list_empty(&lo->plh_segs) && |
473 | !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) && | |
474 | !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { | |
334a8f37 TM |
475 | if (atomic_read(&lo->plh_outstanding) == 0) |
476 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
173f77e9 | 477 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
2d148c7e | 478 | } |
d684d2ae FI |
479 | } |
480 | ||
68f74479 TM |
481 | static bool |
482 | pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo, | |
483 | struct pnfs_layout_segment *lseg) | |
484 | { | |
485 | if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && | |
486 | pnfs_layout_is_valid(lo)) { | |
4aab9732 | 487 | pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0); |
68f74479 TM |
488 | list_move_tail(&lseg->pls_list, &lo->plh_return_segs); |
489 | return true; | |
490 | } | |
491 | return false; | |
492 | } | |
493 | ||
bae724ef | 494 | void |
9369a431 | 495 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 496 | { |
57036a37 | 497 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
498 | struct inode *inode; |
499 | ||
500 | if (!lseg) | |
501 | return; | |
502 | ||
4541d16c FI |
503 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
504 | atomic_read(&lseg->pls_refcount), | |
505 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
4ef2e4f8 | 506 | |
57036a37 TM |
507 | lo = lseg->pls_layout; |
508 | inode = lo->plh_inode; | |
4ef2e4f8 | 509 | |
d684d2ae | 510 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
faa4a54f TM |
511 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { |
512 | spin_unlock(&inode->i_lock); | |
513 | return; | |
514 | } | |
8f0d27dc | 515 | pnfs_get_layout_hdr(lo); |
4ef2e4f8 | 516 | pnfs_layout_remove_lseg(lo, lseg); |
68f74479 TM |
517 | if (pnfs_cache_lseg_for_layoutreturn(lo, lseg)) |
518 | lseg = NULL; | |
4ef2e4f8 TM |
519 | spin_unlock(&inode->i_lock); |
520 | pnfs_free_lseg(lseg); | |
521 | pnfs_put_layout_hdr(lo); | |
4541d16c | 522 | } |
4541d16c | 523 | } |
9369a431 | 524 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 525 | |
6543f803 | 526 | static void pnfs_free_lseg_async_work(struct work_struct *work) |
e6cf82d1 WAA |
527 | { |
528 | struct pnfs_layout_segment *lseg; | |
6543f803 | 529 | struct pnfs_layout_hdr *lo; |
e6cf82d1 WAA |
530 | |
531 | lseg = container_of(work, struct pnfs_layout_segment, pls_work); | |
6543f803 | 532 | lo = lseg->pls_layout; |
e6cf82d1 | 533 | |
6543f803 TM |
534 | pnfs_free_lseg(lseg); |
535 | pnfs_put_layout_hdr(lo); | |
e6cf82d1 WAA |
536 | } |
537 | ||
6543f803 | 538 | static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg) |
e6cf82d1 | 539 | { |
6543f803 | 540 | INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work); |
e6cf82d1 WAA |
541 | schedule_work(&lseg->pls_work); |
542 | } | |
6543f803 TM |
543 | |
544 | void | |
545 | pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg) | |
546 | { | |
547 | if (!lseg) | |
548 | return; | |
549 | ||
550 | assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock); | |
551 | ||
552 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, | |
553 | atomic_read(&lseg->pls_refcount), | |
554 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
555 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
556 | struct pnfs_layout_hdr *lo = lseg->pls_layout; | |
faa4a54f TM |
557 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) |
558 | return; | |
6543f803 | 559 | pnfs_layout_remove_lseg(lo, lseg); |
68f74479 TM |
560 | if (!pnfs_cache_lseg_for_layoutreturn(lo, lseg)) { |
561 | pnfs_get_layout_hdr(lo); | |
562 | pnfs_free_lseg_async(lseg); | |
563 | } | |
6543f803 TM |
564 | } |
565 | } | |
e6cf82d1 | 566 | |
fb3296eb BH |
567 | /* |
568 | * is l2 fully contained in l1? | |
569 | * start1 end1 | |
570 | * [----------------------------------) | |
571 | * start2 end2 | |
572 | * [----------------) | |
573 | */ | |
3cb2df17 | 574 | static bool |
7dc0ac70 | 575 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 576 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
577 | { |
578 | u64 start1 = l1->offset; | |
17822b20 | 579 | u64 end1 = pnfs_end_offset(start1, l1->length); |
fb3296eb | 580 | u64 start2 = l2->offset; |
17822b20 | 581 | u64 end2 = pnfs_end_offset(start2, l2->length); |
fb3296eb BH |
582 | |
583 | return (start1 <= start2) && (end1 >= end2); | |
584 | } | |
585 | ||
24956804 TM |
586 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
587 | struct list_head *tmp_list) | |
588 | { | |
589 | if (!atomic_dec_and_test(&lseg->pls_refcount)) | |
590 | return false; | |
591 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
592 | list_add(&lseg->pls_list, tmp_list); | |
593 | return true; | |
594 | } | |
595 | ||
4541d16c FI |
596 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
597 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
598 | struct list_head *tmp_list) | |
599 | { | |
600 | int rv = 0; | |
601 | ||
602 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
603 | /* Remove the reference keeping the lseg in the | |
604 | * list. It will now be removed when all | |
605 | * outstanding io is finished. | |
606 | */ | |
d684d2ae FI |
607 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
608 | atomic_read(&lseg->pls_refcount)); | |
24956804 | 609 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 610 | rv = 1; |
4541d16c FI |
611 | } |
612 | return rv; | |
613 | } | |
614 | ||
6d597e17 JL |
615 | /* |
616 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
617 | * taking into account wraparound issues. | |
618 | */ | |
619 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
620 | { | |
621 | return (s32)(s1 - s2) > 0; | |
622 | } | |
623 | ||
e036f464 TM |
624 | static bool |
625 | pnfs_should_free_range(const struct pnfs_layout_range *lseg_range, | |
626 | const struct pnfs_layout_range *recall_range) | |
627 | { | |
628 | return (recall_range->iomode == IOMODE_ANY || | |
629 | lseg_range->iomode == recall_range->iomode) && | |
630 | pnfs_lseg_range_intersecting(lseg_range, recall_range); | |
631 | } | |
632 | ||
633 | static bool | |
634 | pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg, | |
635 | const struct pnfs_layout_range *recall_range, | |
636 | u32 seq) | |
637 | { | |
638 | if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq)) | |
639 | return false; | |
640 | if (recall_range == NULL) | |
641 | return true; | |
642 | return pnfs_should_free_range(&lseg->pls_range, recall_range); | |
643 | } | |
644 | ||
6d597e17 JL |
645 | /** |
646 | * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later | |
647 | * @lo: layout header containing the lsegs | |
648 | * @tmp_list: list head where doomed lsegs should go | |
649 | * @recall_range: optional recall range argument to match (may be NULL) | |
650 | * @seq: only invalidate lsegs obtained prior to this sequence (may be 0) | |
651 | * | |
652 | * Walk the list of lsegs in the layout header, and tear down any that should | |
653 | * be destroyed. If "recall_range" is specified then the segment must match | |
654 | * that range. If "seq" is non-zero, then only match segments that were handed | |
655 | * out at or before that sequence. | |
656 | * | |
657 | * Returns number of matching invalid lsegs remaining in list after scanning | |
658 | * it and purging them. | |
4541d16c | 659 | */ |
43f1b3da | 660 | int |
49a85061 | 661 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 662 | struct list_head *tmp_list, |
6d597e17 JL |
663 | const struct pnfs_layout_range *recall_range, |
664 | u32 seq) | |
974cec8c AA |
665 | { |
666 | struct pnfs_layout_segment *lseg, *next; | |
71b39854 | 667 | int remaining = 0; |
974cec8c AA |
668 | |
669 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
670 | ||
8006bfba | 671 | if (list_empty(&lo->plh_segs)) |
38511722 | 672 | return 0; |
4541d16c | 673 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
e036f464 | 674 | if (pnfs_match_lseg_recall(lseg, recall_range, seq)) { |
6d597e17 | 675 | dprintk("%s: freeing lseg %p iomode %d seq %u" |
4541d16c | 676 | "offset %llu length %llu\n", __func__, |
6d597e17 JL |
677 | lseg, lseg->pls_range.iomode, lseg->pls_seq, |
678 | lseg->pls_range.offset, lseg->pls_range.length); | |
71b39854 TM |
679 | if (!mark_lseg_invalid(lseg, tmp_list)) |
680 | remaining++; | |
4541d16c | 681 | } |
71b39854 TM |
682 | dprintk("%s:Return %i\n", __func__, remaining); |
683 | return remaining; | |
974cec8c AA |
684 | } |
685 | ||
68f74479 TM |
686 | static void |
687 | pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo, | |
688 | struct list_head *free_me, | |
689 | const struct pnfs_layout_range *range, | |
690 | u32 seq) | |
691 | { | |
692 | struct pnfs_layout_segment *lseg, *next; | |
693 | ||
694 | list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) { | |
695 | if (pnfs_match_lseg_recall(lseg, range, seq)) | |
696 | list_move_tail(&lseg->pls_list, free_me); | |
697 | } | |
698 | } | |
699 | ||
f49f9baa | 700 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 701 | void |
4541d16c | 702 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 703 | { |
4541d16c | 704 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
705 | |
706 | if (list_empty(free_me)) | |
707 | return; | |
708 | ||
4541d16c | 709 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 710 | list_del(&lseg->pls_list); |
905ca191 | 711 | pnfs_free_lseg(lseg); |
974cec8c AA |
712 | } |
713 | } | |
714 | ||
e5e94017 BH |
715 | void |
716 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
717 | { | |
718 | struct pnfs_layout_hdr *lo; | |
974cec8c | 719 | LIST_HEAD(tmp_list); |
e5e94017 BH |
720 | |
721 | spin_lock(&nfsi->vfs_inode.i_lock); | |
722 | lo = nfsi->layout; | |
723 | if (lo) { | |
3e621214 | 724 | pnfs_get_layout_hdr(lo); |
2454dfea | 725 | pnfs_mark_layout_stateid_invalid(lo, &tmp_list); |
3e621214 TM |
726 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); |
727 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
728 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
729 | pnfs_free_lseg_list(&tmp_list); | |
730 | pnfs_put_layout_hdr(lo); | |
731 | } else | |
732 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 733 | } |
041245c8 | 734 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 735 | |
fd9a8d71 TM |
736 | static bool |
737 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
738 | struct list_head *layout_list) | |
974cec8c AA |
739 | { |
740 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 741 | bool ret = false; |
974cec8c | 742 | |
fd9a8d71 TM |
743 | spin_lock(&inode->i_lock); |
744 | lo = NFS_I(inode)->layout; | |
745 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
746 | pnfs_get_layout_hdr(lo); | |
747 | list_add(&lo->plh_bulk_destroy, layout_list); | |
748 | ret = true; | |
749 | } | |
750 | spin_unlock(&inode->i_lock); | |
751 | return ret; | |
752 | } | |
753 | ||
754 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
755 | static int | |
756 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
757 | struct nfs_server *server, | |
758 | struct list_head *layout_list) | |
759 | { | |
760 | struct pnfs_layout_hdr *lo, *next; | |
761 | struct inode *inode; | |
762 | ||
763 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
b85f5620 TM |
764 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) |
765 | continue; | |
fd9a8d71 TM |
766 | inode = igrab(lo->plh_inode); |
767 | if (inode == NULL) | |
768 | continue; | |
769 | list_del_init(&lo->plh_layouts); | |
770 | if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) | |
771 | continue; | |
772 | rcu_read_unlock(); | |
773 | spin_unlock(&clp->cl_lock); | |
774 | iput(inode); | |
775 | spin_lock(&clp->cl_lock); | |
776 | rcu_read_lock(); | |
777 | return -EAGAIN; | |
778 | } | |
779 | return 0; | |
780 | } | |
781 | ||
782 | static int | |
783 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
784 | bool is_bulk_recall) | |
785 | { | |
786 | struct pnfs_layout_hdr *lo; | |
787 | struct inode *inode; | |
fd9a8d71 TM |
788 | LIST_HEAD(lseg_list); |
789 | int ret = 0; | |
790 | ||
791 | while (!list_empty(layout_list)) { | |
792 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
793 | plh_bulk_destroy); | |
794 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
795 | lo->plh_inode->i_ino); | |
796 | inode = lo->plh_inode; | |
7c5d1875 CH |
797 | |
798 | pnfs_layoutcommit_inode(inode, false); | |
799 | ||
fd9a8d71 TM |
800 | spin_lock(&inode->i_lock); |
801 | list_del_init(&lo->plh_bulk_destroy); | |
9fd4b9fc TM |
802 | if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) { |
803 | if (is_bulk_recall) | |
804 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
fd9a8d71 | 805 | ret = -EAGAIN; |
9fd4b9fc | 806 | } |
fd9a8d71 TM |
807 | spin_unlock(&inode->i_lock); |
808 | pnfs_free_lseg_list(&lseg_list); | |
b20135d0 TM |
809 | /* Free all lsegs that are attached to commit buckets */ |
810 | nfs_commit_inode(inode, 0); | |
fd9a8d71 TM |
811 | pnfs_put_layout_hdr(lo); |
812 | iput(inode); | |
813 | } | |
814 | return ret; | |
815 | } | |
816 | ||
817 | int | |
818 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
819 | struct nfs_fsid *fsid, | |
820 | bool is_recall) | |
821 | { | |
822 | struct nfs_server *server; | |
823 | LIST_HEAD(layout_list); | |
c47abcf8 | 824 | |
974cec8c | 825 | spin_lock(&clp->cl_lock); |
6382a441 | 826 | rcu_read_lock(); |
fd9a8d71 | 827 | restart: |
6382a441 | 828 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
829 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
830 | continue; | |
831 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
832 | server, | |
833 | &layout_list) != 0) | |
834 | goto restart; | |
6382a441 WAA |
835 | } |
836 | rcu_read_unlock(); | |
974cec8c AA |
837 | spin_unlock(&clp->cl_lock); |
838 | ||
fd9a8d71 TM |
839 | if (list_empty(&layout_list)) |
840 | return 0; | |
841 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
842 | } | |
843 | ||
844 | int | |
845 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
846 | bool is_recall) | |
847 | { | |
848 | struct nfs_server *server; | |
849 | LIST_HEAD(layout_list); | |
850 | ||
851 | spin_lock(&clp->cl_lock); | |
852 | rcu_read_lock(); | |
853 | restart: | |
854 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
855 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
856 | server, | |
857 | &layout_list) != 0) | |
858 | goto restart; | |
974cec8c | 859 | } |
fd9a8d71 TM |
860 | rcu_read_unlock(); |
861 | spin_unlock(&clp->cl_lock); | |
862 | ||
863 | if (list_empty(&layout_list)) | |
864 | return 0; | |
865 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
866 | } | |
867 | ||
868 | /* | |
869 | * Called by the state manger to remove all layouts established under an | |
870 | * expired lease. | |
871 | */ | |
872 | void | |
873 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
874 | { | |
875 | nfs4_deviceid_mark_client_invalid(clp); | |
876 | nfs4_deviceid_purge_client(clp); | |
877 | ||
878 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
879 | } |
880 | ||
fd6002e9 | 881 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
882 | void |
883 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
884 | bool update_barrier) | |
b1f69b75 | 885 | { |
ecebb80b | 886 | u32 oldseq, newseq, new_barrier = 0; |
b1f69b75 | 887 | |
2d2f24ad TM |
888 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
889 | newseq = be32_to_cpu(new->seqid); | |
2a59a041 TM |
890 | |
891 | if (!pnfs_layout_is_valid(lo)) { | |
892 | nfs4_stateid_copy(&lo->plh_stateid, new); | |
893 | lo->plh_barrier = newseq; | |
894 | pnfs_clear_layoutreturn_info(lo); | |
895 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
896 | return; | |
897 | } | |
898 | if (pnfs_seqid_is_newer(newseq, oldseq)) { | |
f597c537 | 899 | nfs4_stateid_copy(&lo->plh_stateid, new); |
ecebb80b TM |
900 | /* |
901 | * Because of wraparound, we want to keep the barrier | |
902 | * "close" to the current seqids. | |
903 | */ | |
904 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); | |
43f1b3da | 905 | } |
ecebb80b TM |
906 | if (update_barrier) |
907 | new_barrier = be32_to_cpu(new->seqid); | |
908 | else if (new_barrier == 0) | |
909 | return; | |
2a59a041 | 910 | if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
ecebb80b | 911 | lo->plh_barrier = new_barrier; |
b1f69b75 AA |
912 | } |
913 | ||
cf7d63f1 | 914 | static bool |
19c54aba TM |
915 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
916 | const nfs4_stateid *stateid) | |
43f1b3da | 917 | { |
19c54aba | 918 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 919 | |
19c54aba TM |
920 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
921 | } | |
922 | ||
923 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
924 | static bool | |
e1c06f80 | 925 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo) |
19c54aba | 926 | { |
f7e8917a | 927 | return lo->plh_block_lgets || |
e1c06f80 | 928 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
cf7d63f1 FI |
929 | } |
930 | ||
b1f69b75 | 931 | /* |
183d9e7b JL |
932 | * Get layout from server. |
933 | * for now, assume that whole file layouts are requested. | |
934 | * arg->offset: 0 | |
935 | * arg->length: all ones | |
936 | */ | |
e5e94017 BH |
937 | static struct pnfs_layout_segment * |
938 | send_layoutget(struct pnfs_layout_hdr *lo, | |
939 | struct nfs_open_context *ctx, | |
183d9e7b | 940 | nfs4_stateid *stateid, |
e144e539 | 941 | const struct pnfs_layout_range *range, |
183d9e7b | 942 | long *timeout, gfp_t gfp_flags) |
e5e94017 | 943 | { |
b7edfaa1 | 944 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
945 | struct nfs_server *server = NFS_SERVER(ino); |
946 | struct nfs4_layoutget *lgp; | |
2d89a1d3 | 947 | loff_t i_size; |
b1f69b75 AA |
948 | |
949 | dprintk("--> %s\n", __func__); | |
e5e94017 | 950 | |
4f2e9dce JL |
951 | /* |
952 | * Synchronously retrieve layout information from server and | |
953 | * store in lseg. If we race with a concurrent seqid morphing | |
954 | * op, then re-send the LAYOUTGET. | |
955 | */ | |
83026d80 JL |
956 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
957 | if (lgp == NULL) | |
958 | return ERR_PTR(-ENOMEM); | |
959 | ||
960 | i_size = i_size_read(ino); | |
961 | ||
962 | lgp->args.minlength = PAGE_SIZE; | |
963 | if (lgp->args.minlength > range->length) | |
964 | lgp->args.minlength = range->length; | |
965 | if (range->iomode == IOMODE_READ) { | |
966 | if (range->offset >= i_size) | |
967 | lgp->args.minlength = 0; | |
968 | else if (i_size - range->offset < lgp->args.minlength) | |
969 | lgp->args.minlength = i_size - range->offset; | |
d03ab29d | 970 | } |
83026d80 JL |
971 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
972 | pnfs_copy_range(&lgp->args.range, range); | |
973 | lgp->args.type = server->pnfs_curr_ld->id; | |
974 | lgp->args.inode = ino; | |
975 | lgp->args.ctx = get_nfs_open_context(ctx); | |
183d9e7b | 976 | nfs4_stateid_copy(&lgp->args.stateid, stateid); |
83026d80 JL |
977 | lgp->gfp_flags = gfp_flags; |
978 | lgp->cred = lo->plh_lc_cred; | |
35124a09 | 979 | |
183d9e7b | 980 | return nfs4_proc_layoutget(lgp, timeout, gfp_flags); |
974cec8c AA |
981 | } |
982 | ||
24956804 TM |
983 | static void pnfs_clear_layoutcommit(struct inode *inode, |
984 | struct list_head *head) | |
985 | { | |
986 | struct nfs_inode *nfsi = NFS_I(inode); | |
987 | struct pnfs_layout_segment *lseg, *tmp; | |
988 | ||
989 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
990 | return; | |
991 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
992 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
993 | continue; | |
994 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
995 | } | |
996 | } | |
997 | ||
68f74479 | 998 | void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo, |
2a974425 | 999 | const nfs4_stateid *arg_stateid, |
68f74479 | 1000 | const struct pnfs_layout_range *range, |
68f74479 TM |
1001 | const nfs4_stateid *stateid) |
1002 | { | |
1003 | struct inode *inode = lo->plh_inode; | |
1004 | LIST_HEAD(freeme); | |
1005 | ||
1006 | spin_lock(&inode->i_lock); | |
2a974425 TM |
1007 | if (!pnfs_layout_is_valid(lo) || !arg_stateid || |
1008 | !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid)) | |
1009 | goto out_unlock; | |
68f74479 | 1010 | if (stateid) { |
2a974425 TM |
1011 | u32 seq = be32_to_cpu(arg_stateid->seqid); |
1012 | ||
68f74479 TM |
1013 | pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq); |
1014 | pnfs_free_returned_lsegs(lo, &freeme, range, seq); | |
1015 | pnfs_set_layout_stateid(lo, stateid, true); | |
1016 | } else | |
1017 | pnfs_mark_layout_stateid_invalid(lo, &freeme); | |
2a974425 | 1018 | out_unlock: |
68f74479 TM |
1019 | pnfs_clear_layoutreturn_waitbit(lo); |
1020 | spin_unlock(&inode->i_lock); | |
1021 | pnfs_free_lseg_list(&freeme); | |
1022 | ||
1023 | } | |
1024 | ||
13c13a6a | 1025 | static bool |
e5fd1904 TM |
1026 | pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo, |
1027 | nfs4_stateid *stateid, | |
1028 | enum pnfs_iomode *iomode) | |
13c13a6a | 1029 | { |
bf0291dd TM |
1030 | /* Serialise LAYOUTGET/LAYOUTRETURN */ |
1031 | if (atomic_read(&lo->plh_outstanding) != 0) | |
1032 | return false; | |
6604b203 | 1033 | if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) |
13c13a6a | 1034 | return false; |
6604b203 | 1035 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); |
13c13a6a | 1036 | pnfs_get_layout_hdr(lo); |
e5fd1904 TM |
1037 | if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) { |
1038 | if (stateid != NULL) { | |
1039 | nfs4_stateid_copy(stateid, &lo->plh_stateid); | |
1040 | if (lo->plh_return_seq != 0) | |
1041 | stateid->seqid = cpu_to_be32(lo->plh_return_seq); | |
1042 | } | |
1043 | if (iomode != NULL) | |
1044 | *iomode = lo->plh_return_iomode; | |
1045 | pnfs_clear_layoutreturn_info(lo); | |
1046 | return true; | |
1047 | } | |
1048 | if (stateid != NULL) | |
1049 | nfs4_stateid_copy(stateid, &lo->plh_stateid); | |
1050 | if (iomode != NULL) | |
1051 | *iomode = IOMODE_ANY; | |
13c13a6a TM |
1052 | return true; |
1053 | } | |
1054 | ||
828ed9ec TM |
1055 | static void |
1056 | pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args, | |
1057 | struct pnfs_layout_hdr *lo, | |
1058 | const nfs4_stateid *stateid, | |
1059 | enum pnfs_iomode iomode) | |
1060 | { | |
1061 | struct inode *inode = lo->plh_inode; | |
1062 | ||
1063 | args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id; | |
1064 | args->inode = inode; | |
1065 | args->range.iomode = iomode; | |
1066 | args->range.offset = 0; | |
1067 | args->range.length = NFS4_MAX_UINT64; | |
1068 | args->layout = lo; | |
1069 | nfs4_stateid_copy(&args->stateid, stateid); | |
1070 | } | |
1071 | ||
f40eb5d0 | 1072 | static int |
ed429d6b | 1073 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid, |
6c16605d | 1074 | enum pnfs_iomode iomode, bool sync) |
f40eb5d0 PT |
1075 | { |
1076 | struct inode *ino = lo->plh_inode; | |
287bd3e9 | 1077 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; |
f40eb5d0 PT |
1078 | struct nfs4_layoutreturn *lrp; |
1079 | int status = 0; | |
1080 | ||
e4af440a | 1081 | lrp = kzalloc(sizeof(*lrp), GFP_NOFS); |
f40eb5d0 PT |
1082 | if (unlikely(lrp == NULL)) { |
1083 | status = -ENOMEM; | |
1084 | spin_lock(&ino->i_lock); | |
d67ae825 | 1085 | pnfs_clear_layoutreturn_waitbit(lo); |
f40eb5d0 PT |
1086 | spin_unlock(&ino->i_lock); |
1087 | pnfs_put_layout_hdr(lo); | |
1088 | goto out; | |
1089 | } | |
1090 | ||
828ed9ec | 1091 | pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode); |
4d796d75 | 1092 | lrp->args.ld_private = &lrp->ld_private; |
f40eb5d0 PT |
1093 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
1094 | lrp->cred = lo->plh_lc_cred; | |
287bd3e9 TM |
1095 | if (ld->prepare_layoutreturn) |
1096 | ld->prepare_layoutreturn(&lrp->args); | |
f40eb5d0 | 1097 | |
6c16605d | 1098 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
1099 | out: |
1100 | dprintk("<-- %s status: %d\n", __func__, status); | |
1101 | return status; | |
1102 | } | |
1103 | ||
13c13a6a TM |
1104 | /* Return true if layoutreturn is needed */ |
1105 | static bool | |
1106 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo) | |
1107 | { | |
1108 | struct pnfs_layout_segment *s; | |
1109 | ||
2370abda | 1110 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
1111 | return false; |
1112 | ||
1113 | /* Defer layoutreturn until all lsegs are done */ | |
1114 | list_for_each_entry(s, &lo->plh_segs, pls_list) { | |
1115 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags)) | |
1116 | return false; | |
1117 | } | |
1118 | ||
1119 | return true; | |
1120 | } | |
1121 | ||
1122 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo) | |
1123 | { | |
1124 | struct inode *inode= lo->plh_inode; | |
1125 | ||
2370abda | 1126 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
1127 | return; |
1128 | spin_lock(&inode->i_lock); | |
1129 | if (pnfs_layout_need_return(lo)) { | |
1130 | nfs4_stateid stateid; | |
1131 | enum pnfs_iomode iomode; | |
1132 | bool send; | |
1133 | ||
e5fd1904 | 1134 | send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode); |
13c13a6a TM |
1135 | spin_unlock(&inode->i_lock); |
1136 | if (send) { | |
1137 | /* Send an async layoutreturn so we dont deadlock */ | |
1138 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); | |
1139 | } | |
1140 | } else | |
1141 | spin_unlock(&inode->i_lock); | |
1142 | } | |
1143 | ||
293b3b06 AA |
1144 | /* |
1145 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
1146 | * when the layout segment list is empty. | |
1147 | * | |
1148 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
1149 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
1150 | * deviceid is marked invalid. | |
1151 | */ | |
cbe82603 BH |
1152 | int |
1153 | _pnfs_return_layout(struct inode *ino) | |
1154 | { | |
1155 | struct pnfs_layout_hdr *lo = NULL; | |
1156 | struct nfs_inode *nfsi = NFS_I(ino); | |
1157 | LIST_HEAD(tmp_list); | |
cbe82603 | 1158 | nfs4_stateid stateid; |
24408f52 | 1159 | int status = 0; |
7f27392c | 1160 | bool send; |
cbe82603 | 1161 | |
366d5052 | 1162 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
1163 | |
1164 | spin_lock(&ino->i_lock); | |
1165 | lo = nfsi->layout; | |
e5929f3c | 1166 | if (!lo) { |
cbe82603 | 1167 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
1168 | dprintk("NFS: %s no layout to return\n", __func__); |
1169 | goto out; | |
cbe82603 | 1170 | } |
cbe82603 | 1171 | /* Reference matched in nfs4_layoutreturn_release */ |
70c3bd2b | 1172 | pnfs_get_layout_hdr(lo); |
24408f52 TM |
1173 | /* Is there an outstanding layoutreturn ? */ |
1174 | if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) { | |
1175 | spin_unlock(&ino->i_lock); | |
1176 | if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
1177 | TASK_UNINTERRUPTIBLE)) | |
1178 | goto out_put_layout_hdr; | |
1179 | spin_lock(&ino->i_lock); | |
1180 | } | |
24956804 | 1181 | pnfs_clear_layoutcommit(ino, &tmp_list); |
6d597e17 | 1182 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0); |
c88953d8 CH |
1183 | |
1184 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
1185 | struct pnfs_layout_range range = { | |
1186 | .iomode = IOMODE_ANY, | |
1187 | .offset = 0, | |
1188 | .length = NFS4_MAX_UINT64, | |
1189 | }; | |
1190 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
1191 | } | |
1192 | ||
293b3b06 | 1193 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
24408f52 | 1194 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) { |
293b3b06 | 1195 | spin_unlock(&ino->i_lock); |
293b3b06 | 1196 | dprintk("NFS: %s no layout segments to return\n", __func__); |
7f27392c | 1197 | goto out_put_layout_hdr; |
293b3b06 | 1198 | } |
47abadef | 1199 | |
e5fd1904 | 1200 | send = pnfs_prepare_layoutreturn(lo, &stateid, NULL); |
cbe82603 | 1201 | spin_unlock(&ino->i_lock); |
7f27392c | 1202 | if (send) |
ed429d6b | 1203 | status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true); |
7f27392c | 1204 | out_put_layout_hdr: |
ee6625a9 | 1205 | pnfs_free_lseg_list(&tmp_list); |
7f27392c | 1206 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
1207 | out: |
1208 | dprintk("<-- %s status: %d\n", __func__, status); | |
1209 | return status; | |
1210 | } | |
1211 | ||
24028672 TM |
1212 | int |
1213 | pnfs_commit_and_return_layout(struct inode *inode) | |
1214 | { | |
1215 | struct pnfs_layout_hdr *lo; | |
1216 | int ret; | |
1217 | ||
1218 | spin_lock(&inode->i_lock); | |
1219 | lo = NFS_I(inode)->layout; | |
1220 | if (lo == NULL) { | |
1221 | spin_unlock(&inode->i_lock); | |
1222 | return 0; | |
1223 | } | |
1224 | pnfs_get_layout_hdr(lo); | |
1225 | /* Block new layoutgets and read/write to ds */ | |
1226 | lo->plh_block_lgets++; | |
1227 | spin_unlock(&inode->i_lock); | |
1228 | filemap_fdatawait(inode->i_mapping); | |
1229 | ret = pnfs_layoutcommit_inode(inode, true); | |
1230 | if (ret == 0) | |
1231 | ret = _pnfs_return_layout(inode); | |
1232 | spin_lock(&inode->i_lock); | |
1233 | lo->plh_block_lgets--; | |
1234 | spin_unlock(&inode->i_lock); | |
1235 | pnfs_put_layout_hdr(lo); | |
1236 | return ret; | |
1237 | } | |
1238 | ||
1c5bd76d TM |
1239 | bool pnfs_roc(struct inode *ino, |
1240 | struct nfs4_layoutreturn_args *args, | |
1241 | struct nfs4_layoutreturn_res *res, | |
1242 | const struct rpc_cred *cred) | |
f7e8917a | 1243 | { |
40dd4b7a TM |
1244 | struct nfs_inode *nfsi = NFS_I(ino); |
1245 | struct nfs_open_context *ctx; | |
1246 | struct nfs4_state *state; | |
f7e8917a | 1247 | struct pnfs_layout_hdr *lo; |
1c5bd76d | 1248 | struct pnfs_layout_segment *lseg, *next; |
193e3aa2 | 1249 | nfs4_stateid stateid; |
1c5bd76d TM |
1250 | enum pnfs_iomode iomode = 0; |
1251 | bool layoutreturn = false, roc = false; | |
e71708d4 | 1252 | bool skip_read = false; |
f7e8917a | 1253 | |
1c5bd76d TM |
1254 | if (!nfs_have_layout(ino)) |
1255 | return false; | |
29ade5db | 1256 | retry: |
f7e8917a | 1257 | spin_lock(&ino->i_lock); |
40dd4b7a | 1258 | lo = nfsi->layout; |
0cdc329e TM |
1259 | if (!lo || !pnfs_layout_is_valid(lo) || |
1260 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
40dd4b7a | 1261 | goto out_noroc; |
29ade5db TM |
1262 | if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) { |
1263 | pnfs_get_layout_hdr(lo); | |
1264 | spin_unlock(&ino->i_lock); | |
1265 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
1266 | TASK_UNINTERRUPTIBLE); | |
1267 | pnfs_put_layout_hdr(lo); | |
1268 | goto retry; | |
1269 | } | |
40dd4b7a | 1270 | |
e755d638 | 1271 | /* no roc if we hold a delegation */ |
e71708d4 TM |
1272 | if (nfs4_check_delegation(ino, FMODE_READ)) { |
1273 | if (nfs4_check_delegation(ino, FMODE_WRITE)) | |
1274 | goto out_noroc; | |
1275 | skip_read = true; | |
1276 | } | |
40dd4b7a TM |
1277 | |
1278 | list_for_each_entry(ctx, &nfsi->open_files, list) { | |
1279 | state = ctx->state; | |
e71708d4 TM |
1280 | if (state == NULL) |
1281 | continue; | |
40dd4b7a | 1282 | /* Don't return layout if there is open file state */ |
e71708d4 | 1283 | if (state->state & FMODE_WRITE) |
40dd4b7a | 1284 | goto out_noroc; |
e71708d4 TM |
1285 | if (state->state & FMODE_READ) |
1286 | skip_read = true; | |
40dd4b7a TM |
1287 | } |
1288 | ||
e755d638 | 1289 | |
1c5bd76d | 1290 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) { |
e71708d4 TM |
1291 | if (skip_read && lseg->pls_range.iomode == IOMODE_READ) |
1292 | continue; | |
e755d638 | 1293 | /* If we are sending layoutreturn, invalidate all valid lsegs */ |
1c5bd76d TM |
1294 | if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags)) |
1295 | continue; | |
1296 | /* | |
1297 | * Note: mark lseg for return so pnfs_layout_remove_lseg | |
1298 | * doesn't invalidate the layout for us. | |
1299 | */ | |
1300 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
1301 | if (!mark_lseg_invalid(lseg, &lo->plh_return_segs)) | |
1302 | continue; | |
1303 | pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0); | |
69820d22 TM |
1304 | } |
1305 | ||
1c5bd76d TM |
1306 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
1307 | goto out_noroc; | |
69820d22 | 1308 | |
500d701f | 1309 | /* ROC in two conditions: |
e755d638 PT |
1310 | * 1. there are ROC lsegs |
1311 | * 2. we don't send layoutreturn | |
e755d638 | 1312 | */ |
1c5bd76d TM |
1313 | /* lo ref dropped in pnfs_roc_release() */ |
1314 | layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode); | |
1315 | /* If the creds don't match, we can't compound the layoutreturn */ | |
1316 | if (!layoutreturn || cred != lo->plh_lc_cred) | |
1317 | goto out_noroc; | |
1318 | ||
1319 | roc = layoutreturn; | |
1320 | pnfs_init_layoutreturn_args(args, lo, &stateid, iomode); | |
1321 | res->lrs_present = 0; | |
1322 | layoutreturn = false; | |
f7e8917a | 1323 | |
40dd4b7a | 1324 | out_noroc: |
f7e8917a | 1325 | spin_unlock(&ino->i_lock); |
e755d638 | 1326 | pnfs_layoutcommit_inode(ino, true); |
287bd3e9 TM |
1327 | if (roc) { |
1328 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
1329 | if (ld->prepare_layoutreturn) | |
1330 | ld->prepare_layoutreturn(args); | |
1331 | return true; | |
1332 | } | |
e755d638 | 1333 | if (layoutreturn) |
1c5bd76d | 1334 | pnfs_send_layoutreturn(lo, &stateid, iomode, true); |
287bd3e9 | 1335 | return false; |
f7e8917a FI |
1336 | } |
1337 | ||
1c5bd76d TM |
1338 | void pnfs_roc_release(struct nfs4_layoutreturn_args *args, |
1339 | struct nfs4_layoutreturn_res *res, | |
1340 | int ret) | |
f7e8917a | 1341 | { |
1c5bd76d TM |
1342 | struct pnfs_layout_hdr *lo = args->layout; |
1343 | const nfs4_stateid *arg_stateid = NULL; | |
1344 | const nfs4_stateid *res_stateid = NULL; | |
287bd3e9 | 1345 | struct nfs4_xdr_opaque_data *ld_private = args->ld_private; |
f7e8917a | 1346 | |
1c5bd76d TM |
1347 | if (ret == 0) { |
1348 | arg_stateid = &args->stateid; | |
1349 | if (res->lrs_present) | |
1350 | res_stateid = &res->stateid; | |
1351 | } | |
1352 | pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range, | |
1353 | res_stateid); | |
287bd3e9 TM |
1354 | if (ld_private && ld_private->ops && ld_private->ops->free) |
1355 | ld_private->ops->free(ld_private); | |
1c5bd76d TM |
1356 | pnfs_put_layout_hdr(lo); |
1357 | trace_nfs4_layoutreturn_on_close(args->inode, 0); | |
f7e8917a FI |
1358 | } |
1359 | ||
500d701f PT |
1360 | bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task) |
1361 | { | |
1362 | struct nfs_inode *nfsi = NFS_I(ino); | |
1363 | struct pnfs_layout_hdr *lo; | |
1364 | bool sleep = false; | |
1365 | ||
1366 | /* we might not have grabbed lo reference. so need to check under | |
1367 | * i_lock */ | |
1368 | spin_lock(&ino->i_lock); | |
1369 | lo = nfsi->layout; | |
ee284e35 TM |
1370 | if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
1371 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
500d701f | 1372 | sleep = true; |
ee284e35 | 1373 | } |
500d701f | 1374 | spin_unlock(&ino->i_lock); |
500d701f PT |
1375 | return sleep; |
1376 | } | |
1377 | ||
b1f69b75 AA |
1378 | /* |
1379 | * Compare two layout segments for sorting into layout cache. | |
1380 | * We want to preferentially return RW over RO layouts, so ensure those | |
1381 | * are seen first. | |
1382 | */ | |
1383 | static s64 | |
7dc0ac70 | 1384 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1385 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1386 | { |
fb3296eb BH |
1387 | s64 d; |
1388 | ||
1389 | /* high offset > low offset */ | |
1390 | d = l1->offset - l2->offset; | |
1391 | if (d) | |
1392 | return d; | |
1393 | ||
1394 | /* short length > long length */ | |
1395 | d = l2->length - l1->length; | |
1396 | if (d) | |
1397 | return d; | |
1398 | ||
b1f69b75 | 1399 | /* read > read/write */ |
fb3296eb | 1400 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1401 | } |
1402 | ||
03772d2f TM |
1403 | static bool |
1404 | pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1, | |
1405 | const struct pnfs_layout_range *l2) | |
1406 | { | |
1407 | return pnfs_lseg_range_cmp(l1, l2) > 0; | |
1408 | } | |
1409 | ||
1410 | static bool | |
1411 | pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg, | |
1412 | struct pnfs_layout_segment *old) | |
1413 | { | |
1414 | return false; | |
1415 | } | |
1416 | ||
1417 | void | |
1418 | pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1419 | struct pnfs_layout_segment *lseg, | |
1420 | bool (*is_after)(const struct pnfs_layout_range *, | |
1421 | const struct pnfs_layout_range *), | |
1422 | bool (*do_merge)(struct pnfs_layout_segment *, | |
1423 | struct pnfs_layout_segment *), | |
1424 | struct list_head *free_me) | |
974cec8c | 1425 | { |
03772d2f | 1426 | struct pnfs_layout_segment *lp, *tmp; |
b1f69b75 | 1427 | |
974cec8c AA |
1428 | dprintk("%s:Begin\n", __func__); |
1429 | ||
03772d2f TM |
1430 | list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) { |
1431 | if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0) | |
1432 | continue; | |
1433 | if (do_merge(lseg, lp)) { | |
1434 | mark_lseg_invalid(lp, free_me); | |
1435 | continue; | |
1436 | } | |
1437 | if (is_after(&lseg->pls_range, &lp->pls_range)) | |
b1f69b75 | 1438 | continue; |
566052c5 | 1439 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1440 | dprintk("%s: inserted lseg %p " |
1441 | "iomode %d offset %llu length %llu before " | |
1442 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1443 | __func__, lseg, lseg->pls_range.iomode, |
1444 | lseg->pls_range.offset, lseg->pls_range.length, | |
1445 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1446 | lp->pls_range.length); | |
fb3296eb | 1447 | goto out; |
974cec8c | 1448 | } |
fb3296eb BH |
1449 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1450 | dprintk("%s: inserted lseg %p " | |
1451 | "iomode %d offset %llu length %llu at tail\n", | |
1452 | __func__, lseg, lseg->pls_range.iomode, | |
1453 | lseg->pls_range.offset, lseg->pls_range.length); | |
1454 | out: | |
70c3bd2b | 1455 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1456 | |
1457 | dprintk("%s:Return\n", __func__); | |
e5e94017 | 1458 | } |
03772d2f TM |
1459 | EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg); |
1460 | ||
1461 | static void | |
1462 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1463 | struct pnfs_layout_segment *lseg, | |
1464 | struct list_head *free_me) | |
1465 | { | |
1466 | struct inode *inode = lo->plh_inode; | |
1467 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
1468 | ||
1469 | if (ld->add_lseg != NULL) | |
1470 | ld->add_lseg(lo, lseg, free_me); | |
1471 | else | |
1472 | pnfs_generic_layout_insert_lseg(lo, lseg, | |
1473 | pnfs_lseg_range_is_after, | |
1474 | pnfs_lseg_no_merge, | |
1475 | free_me); | |
1476 | } | |
e5e94017 BH |
1477 | |
1478 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1479 | alloc_init_layout_hdr(struct inode *ino, |
1480 | struct nfs_open_context *ctx, | |
1481 | gfp_t gfp_flags) | |
e5e94017 BH |
1482 | { |
1483 | struct pnfs_layout_hdr *lo; | |
1484 | ||
636fb9c8 | 1485 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1486 | if (!lo) |
1487 | return NULL; | |
cc6e5340 | 1488 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1489 | INIT_LIST_HEAD(&lo->plh_layouts); |
1490 | INIT_LIST_HEAD(&lo->plh_segs); | |
68f74479 | 1491 | INIT_LIST_HEAD(&lo->plh_return_segs); |
fd9a8d71 | 1492 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1493 | lo->plh_inode = ino; |
5cc2216d | 1494 | lo->plh_lc_cred = get_rpccred(ctx->cred); |
67a3b721 | 1495 | lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID; |
e5e94017 BH |
1496 | return lo; |
1497 | } | |
1498 | ||
1499 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1500 | pnfs_find_alloc_layout(struct inode *ino, |
1501 | struct nfs_open_context *ctx, | |
1502 | gfp_t gfp_flags) | |
e5241e43 TM |
1503 | __releases(&ino->i_lock) |
1504 | __acquires(&ino->i_lock) | |
e5e94017 BH |
1505 | { |
1506 | struct nfs_inode *nfsi = NFS_I(ino); | |
1507 | struct pnfs_layout_hdr *new = NULL; | |
1508 | ||
1509 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1510 | ||
251ec410 TM |
1511 | if (nfsi->layout != NULL) |
1512 | goto out_existing; | |
e5e94017 | 1513 | spin_unlock(&ino->i_lock); |
9fa40758 | 1514 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1515 | spin_lock(&ino->i_lock); |
1516 | ||
251ec410 | 1517 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1518 | nfsi->layout = new; |
251ec410 | 1519 | return new; |
7175fe90 YN |
1520 | } else if (new != NULL) |
1521 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1522 | out_existing: |
1523 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1524 | return nfsi->layout; |
1525 | } | |
1526 | ||
b1f69b75 AA |
1527 | /* |
1528 | * iomode matching rules: | |
c7d73af2 TH |
1529 | * iomode lseg strict match |
1530 | * iomode | |
1531 | * ----- ----- ------ ----- | |
1532 | * ANY READ N/A true | |
1533 | * ANY RW N/A true | |
1534 | * RW READ N/A false | |
1535 | * RW RW N/A true | |
1536 | * READ READ N/A true | |
1537 | * READ RW true false | |
1538 | * READ RW false true | |
b1f69b75 | 1539 | */ |
3cb2df17 | 1540 | static bool |
7dc0ac70 | 1541 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
c7d73af2 TH |
1542 | const struct pnfs_layout_range *range, |
1543 | bool strict_iomode) | |
b1f69b75 | 1544 | { |
fb3296eb BH |
1545 | struct pnfs_layout_range range1; |
1546 | ||
1547 | if ((range->iomode == IOMODE_RW && | |
1548 | ls_range->iomode != IOMODE_RW) || | |
c7d73af2 TH |
1549 | (range->iomode != ls_range->iomode && |
1550 | strict_iomode == true) || | |
7dc0ac70 | 1551 | !pnfs_lseg_range_intersecting(ls_range, range)) |
fb3296eb BH |
1552 | return 0; |
1553 | ||
1554 | /* range1 covers only the first byte in the range */ | |
1555 | range1 = *range; | |
1556 | range1.length = 1; | |
7dc0ac70 | 1557 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1558 | } |
1559 | ||
1560 | /* | |
1561 | * lookup range in layout | |
1562 | */ | |
e5e94017 | 1563 | static struct pnfs_layout_segment * |
fb3296eb | 1564 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
c7d73af2 TH |
1565 | struct pnfs_layout_range *range, |
1566 | bool strict_iomode) | |
e5e94017 | 1567 | { |
b1f69b75 AA |
1568 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1569 | ||
1570 | dprintk("%s:Begin\n", __func__); | |
1571 | ||
b7edfaa1 | 1572 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1573 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
ce6ab4f2 | 1574 | !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && |
c7d73af2 TH |
1575 | pnfs_lseg_range_match(&lseg->pls_range, range, |
1576 | strict_iomode)) { | |
9369a431 | 1577 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1578 | break; |
1579 | } | |
b1f69b75 AA |
1580 | } |
1581 | ||
1582 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 1583 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1584 | return ret; |
e5e94017 BH |
1585 | } |
1586 | ||
d23d61c8 AA |
1587 | /* |
1588 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1589 | * to the MDS or over pNFS | |
1590 | * | |
1591 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1592 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1593 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1594 | * WRITE request. | |
1595 | * | |
1596 | * A return of true means use MDS I/O. | |
1597 | * | |
1598 | * From rfc 5661: | |
1599 | * If a file's size is smaller than the file size threshold, data accesses | |
1600 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1601 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1602 | * server. If both file size and I/O size are provided, the client SHOULD | |
1603 | * reach or exceed both thresholds before sending its read or write | |
1604 | * requests to the data server. | |
1605 | */ | |
1606 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1607 | struct inode *ino, int iomode) | |
1608 | { | |
1609 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1610 | struct nfs_inode *nfsi = NFS_I(ino); | |
1611 | loff_t fsize = i_size_read(ino); | |
1612 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1613 | ||
1614 | if (t == NULL) | |
1615 | return ret; | |
1616 | ||
1617 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1618 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1619 | ||
1620 | switch (iomode) { | |
1621 | case IOMODE_READ: | |
1622 | if (t->bm & THRESHOLD_RD) { | |
1623 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1624 | size_set = true; | |
1625 | if (fsize < t->rd_sz) | |
1626 | size = true; | |
1627 | } | |
1628 | if (t->bm & THRESHOLD_RD_IO) { | |
1629 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1630 | nfsi->read_io); | |
1631 | io_set = true; | |
1632 | if (nfsi->read_io < t->rd_io_sz) | |
1633 | io = true; | |
1634 | } | |
1635 | break; | |
1636 | case IOMODE_RW: | |
1637 | if (t->bm & THRESHOLD_WR) { | |
1638 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1639 | size_set = true; | |
1640 | if (fsize < t->wr_sz) | |
1641 | size = true; | |
1642 | } | |
1643 | if (t->bm & THRESHOLD_WR_IO) { | |
1644 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1645 | nfsi->write_io); | |
1646 | io_set = true; | |
1647 | if (nfsi->write_io < t->wr_io_sz) | |
1648 | io = true; | |
1649 | } | |
1650 | break; | |
1651 | } | |
1652 | if (size_set && io_set) { | |
1653 | if (size && io) | |
1654 | ret = true; | |
1655 | } else if (size || io) | |
1656 | ret = true; | |
1657 | ||
1658 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1659 | return ret; | |
1660 | } | |
1661 | ||
aa8a45ee PT |
1662 | static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo) |
1663 | { | |
1664 | /* | |
1665 | * send layoutcommit as it can hold up layoutreturn due to lseg | |
1666 | * reference | |
1667 | */ | |
1668 | pnfs_layoutcommit_inode(lo->plh_inode, false); | |
1669 | return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
2e5b29f0 | 1670 | nfs_wait_bit_killable, |
aa8a45ee PT |
1671 | TASK_UNINTERRUPTIBLE); |
1672 | } | |
1673 | ||
d67ae825 TH |
1674 | static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo) |
1675 | { | |
1676 | unsigned long *bitlock = &lo->plh_flags; | |
1677 | ||
1678 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1679 | smp_mb__after_atomic(); | |
1680 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1681 | } | |
1682 | ||
e5e94017 BH |
1683 | /* |
1684 | * Layout segment is retreived from the server if not cached. | |
1685 | * The appropriate layout segment is referenced and returned to the caller. | |
1686 | */ | |
7c24d948 | 1687 | struct pnfs_layout_segment * |
e5e94017 BH |
1688 | pnfs_update_layout(struct inode *ino, |
1689 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1690 | loff_t pos, |
1691 | u64 count, | |
a75b9df9 | 1692 | enum pnfs_iomode iomode, |
c7d73af2 | 1693 | bool strict_iomode, |
a75b9df9 | 1694 | gfp_t gfp_flags) |
e5e94017 | 1695 | { |
fb3296eb BH |
1696 | struct pnfs_layout_range arg = { |
1697 | .iomode = iomode, | |
1698 | .offset = pos, | |
1699 | .length = count, | |
1700 | }; | |
183d9e7b | 1701 | unsigned pg_offset, seq; |
6382a441 WAA |
1702 | struct nfs_server *server = NFS_SERVER(ino); |
1703 | struct nfs_client *clp = server->nfs_client; | |
183d9e7b | 1704 | struct pnfs_layout_hdr *lo = NULL; |
e5e94017 | 1705 | struct pnfs_layout_segment *lseg = NULL; |
183d9e7b JL |
1706 | nfs4_stateid stateid; |
1707 | long timeout = 0; | |
66b53f32 | 1708 | unsigned long giveup = jiffies + (clp->cl_lease_time << 1); |
30005121 | 1709 | bool first; |
e5e94017 | 1710 | |
9a4bf31d | 1711 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) { |
183d9e7b | 1712 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1713 | PNFS_UPDATE_LAYOUT_NO_PNFS); |
f86bbcf8 | 1714 | goto out; |
9a4bf31d | 1715 | } |
d23d61c8 | 1716 | |
9a4bf31d | 1717 | if (iomode == IOMODE_READ && i_size_read(ino) == 0) { |
183d9e7b | 1718 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1719 | PNFS_UPDATE_LAYOUT_RD_ZEROLEN); |
4ae93560 | 1720 | goto out; |
9a4bf31d | 1721 | } |
4ae93560 | 1722 | |
9a4bf31d | 1723 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) { |
183d9e7b | 1724 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1725 | PNFS_UPDATE_LAYOUT_MDSTHRESH); |
f86bbcf8 | 1726 | goto out; |
9a4bf31d | 1727 | } |
d23d61c8 | 1728 | |
9bf87482 | 1729 | lookup_again: |
b88fa69e | 1730 | nfs4_client_recover_expired_lease(clp); |
9bf87482 | 1731 | first = false; |
e5e94017 | 1732 | spin_lock(&ino->i_lock); |
9fa40758 | 1733 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1734 | if (lo == NULL) { |
1735 | spin_unlock(&ino->i_lock); | |
183d9e7b | 1736 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1737 | PNFS_UPDATE_LAYOUT_NOMEM); |
830ffb56 TM |
1738 | goto out; |
1739 | } | |
e5e94017 | 1740 | |
43f1b3da | 1741 | /* Do we even need to bother with this? */ |
a59c30ac | 1742 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
183d9e7b | 1743 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1744 | PNFS_UPDATE_LAYOUT_BULK_RECALL); |
43f1b3da | 1745 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1746 | goto out_unlock; |
1747 | } | |
1748 | ||
1749 | /* if LAYOUTGET already failed once we don't try again */ | |
2e5b29f0 | 1750 | if (pnfs_layout_io_test_failed(lo, iomode)) { |
183d9e7b | 1751 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1752 | PNFS_UPDATE_LAYOUT_IO_TEST_FAIL); |
e5e94017 | 1753 | goto out_unlock; |
9a4bf31d | 1754 | } |
e5e94017 | 1755 | |
c7d73af2 | 1756 | lseg = pnfs_find_lseg(lo, &arg, strict_iomode); |
183d9e7b JL |
1757 | if (lseg) { |
1758 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1759 | PNFS_UPDATE_LAYOUT_FOUND_CACHED); | |
1760 | goto out_unlock; | |
1761 | } | |
1762 | ||
1763 | if (!nfs4_valid_open_stateid(ctx->state)) { | |
1764 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1765 | PNFS_UPDATE_LAYOUT_INVALID_OPEN); | |
1766 | goto out_unlock; | |
1767 | } | |
1768 | ||
1769 | /* | |
1770 | * Choose a stateid for the LAYOUTGET. If we don't have a layout | |
1771 | * stateid, or it has been invalidated, then we must use the open | |
1772 | * stateid. | |
1773 | */ | |
67a3b721 | 1774 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { |
183d9e7b JL |
1775 | |
1776 | /* | |
1777 | * The first layoutget for the file. Need to serialize per | |
9bf87482 PT |
1778 | * RFC 5661 Errata 3208. |
1779 | */ | |
1780 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
1781 | &lo->plh_flags)) { | |
1782 | spin_unlock(&ino->i_lock); | |
1783 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET, | |
1784 | TASK_UNINTERRUPTIBLE); | |
1785 | pnfs_put_layout_hdr(lo); | |
183d9e7b | 1786 | dprintk("%s retrying\n", __func__); |
9bf87482 PT |
1787 | goto lookup_again; |
1788 | } | |
183d9e7b JL |
1789 | |
1790 | first = true; | |
1791 | do { | |
1792 | seq = read_seqbegin(&ctx->state->seqlock); | |
1793 | nfs4_stateid_copy(&stateid, &ctx->state->stateid); | |
1794 | } while (read_seqretry(&ctx->state->seqlock, seq)); | |
9bf87482 | 1795 | } else { |
183d9e7b | 1796 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
9bf87482 | 1797 | } |
568e8c49 | 1798 | |
aa8a45ee PT |
1799 | /* |
1800 | * Because we free lsegs before sending LAYOUTRETURN, we need to wait | |
1801 | * for LAYOUTRETURN even if first is true. | |
1802 | */ | |
8f70f53a | 1803 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
aa8a45ee PT |
1804 | spin_unlock(&ino->i_lock); |
1805 | dprintk("%s wait for layoutreturn\n", __func__); | |
1806 | if (pnfs_prepare_to_retry_layoutget(lo)) { | |
d67ae825 TH |
1807 | if (first) |
1808 | pnfs_clear_first_layoutget(lo); | |
aa8a45ee PT |
1809 | pnfs_put_layout_hdr(lo); |
1810 | dprintk("%s retrying\n", __func__); | |
183d9e7b JL |
1811 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
1812 | lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
aa8a45ee PT |
1813 | goto lookup_again; |
1814 | } | |
183d9e7b | 1815 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1816 | PNFS_UPDATE_LAYOUT_RETURN); |
aa8a45ee PT |
1817 | goto out_put_layout_hdr; |
1818 | } | |
1819 | ||
9a4bf31d | 1820 | if (pnfs_layoutgets_blocked(lo)) { |
183d9e7b | 1821 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1822 | PNFS_UPDATE_LAYOUT_BLOCKED); |
cf7d63f1 | 1823 | goto out_unlock; |
9a4bf31d | 1824 | } |
cf7d63f1 | 1825 | atomic_inc(&lo->plh_outstanding); |
f49f9baa | 1826 | spin_unlock(&ino->i_lock); |
30005121 | 1827 | |
abb9a007 | 1828 | if (list_empty(&lo->plh_layouts)) { |
2130ff66 FI |
1829 | /* The lo must be on the clp list if there is any |
1830 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1831 | */ | |
1832 | spin_lock(&clp->cl_lock); | |
abb9a007 PT |
1833 | if (list_empty(&lo->plh_layouts)) |
1834 | list_add_tail(&lo->plh_layouts, &server->layouts); | |
2130ff66 FI |
1835 | spin_unlock(&clp->cl_lock); |
1836 | } | |
e5e94017 | 1837 | |
09cbfeaf | 1838 | pg_offset = arg.offset & ~PAGE_MASK; |
707ed5fd BH |
1839 | if (pg_offset) { |
1840 | arg.offset -= pg_offset; | |
1841 | arg.length += pg_offset; | |
1842 | } | |
7c24d948 | 1843 | if (arg.length != NFS4_MAX_UINT64) |
09cbfeaf | 1844 | arg.length = PAGE_ALIGN(arg.length); |
707ed5fd | 1845 | |
183d9e7b JL |
1846 | lseg = send_layoutget(lo, ctx, &stateid, &arg, &timeout, gfp_flags); |
1847 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1848 | PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET); | |
56b38a1f | 1849 | atomic_dec(&lo->plh_outstanding); |
83026d80 | 1850 | if (IS_ERR(lseg)) { |
183d9e7b | 1851 | switch(PTR_ERR(lseg)) { |
e85d7ee4 | 1852 | case -EBUSY: |
183d9e7b JL |
1853 | if (time_after(jiffies, giveup)) |
1854 | lseg = NULL; | |
66b53f32 TM |
1855 | break; |
1856 | case -ERECALLCONFLICT: | |
1857 | /* Huh? We hold no layouts, how is there a recall? */ | |
1858 | if (first) { | |
1859 | lseg = NULL; | |
1860 | break; | |
1861 | } | |
1862 | /* Destroy the existing layout and start over */ | |
1863 | if (time_after(jiffies, giveup)) | |
1864 | pnfs_destroy_layout(NFS_I(ino)); | |
183d9e7b JL |
1865 | /* Fallthrough */ |
1866 | case -EAGAIN: | |
56b38a1f | 1867 | break; |
183d9e7b JL |
1868 | default: |
1869 | if (!nfs_error_is_fatal(PTR_ERR(lseg))) { | |
1870 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
1871 | lseg = NULL; | |
1872 | } | |
56b38a1f TM |
1873 | goto out_put_layout_hdr; |
1874 | } | |
1875 | if (lseg) { | |
1876 | if (first) | |
1877 | pnfs_clear_first_layoutget(lo); | |
1878 | trace_pnfs_update_layout(ino, pos, count, | |
1879 | iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
1880 | pnfs_put_layout_hdr(lo); | |
1881 | goto lookup_again; | |
83026d80 JL |
1882 | } |
1883 | } else { | |
1884 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
1885 | } | |
1886 | ||
830ffb56 | 1887 | out_put_layout_hdr: |
d67ae825 TH |
1888 | if (first) |
1889 | pnfs_clear_first_layoutget(lo); | |
70c3bd2b | 1890 | pnfs_put_layout_hdr(lo); |
e5e94017 | 1891 | out: |
f86bbcf8 TM |
1892 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
1893 | "(%s, offset: %llu, length: %llu)\n", | |
1894 | __func__, ino->i_sb->s_id, | |
1895 | (unsigned long long)NFS_FILEID(ino), | |
d600ad1f | 1896 | IS_ERR_OR_NULL(lseg) ? "not found" : "found", |
f86bbcf8 TM |
1897 | iomode==IOMODE_RW ? "read/write" : "read-only", |
1898 | (unsigned long long)pos, | |
1899 | (unsigned long long)count); | |
e5e94017 BH |
1900 | return lseg; |
1901 | out_unlock: | |
1902 | spin_unlock(&ino->i_lock); | |
830ffb56 | 1903 | goto out_put_layout_hdr; |
e5e94017 | 1904 | } |
7c24d948 | 1905 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1906 | |
540d9864 TM |
1907 | static bool |
1908 | pnfs_sanity_check_layout_range(struct pnfs_layout_range *range) | |
1909 | { | |
1910 | switch (range->iomode) { | |
1911 | case IOMODE_READ: | |
1912 | case IOMODE_RW: | |
1913 | break; | |
1914 | default: | |
1915 | return false; | |
1916 | } | |
1917 | if (range->offset == NFS4_MAX_UINT64) | |
1918 | return false; | |
1919 | if (range->length == 0) | |
1920 | return false; | |
1921 | if (range->length != NFS4_MAX_UINT64 && | |
1922 | range->length > NFS4_MAX_UINT64 - range->offset) | |
1923 | return false; | |
1924 | return true; | |
1925 | } | |
1926 | ||
a0b0a6e3 | 1927 | struct pnfs_layout_segment * |
b1f69b75 AA |
1928 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1929 | { | |
1930 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1931 | struct nfs4_layoutget_res *res = &lgp->res; | |
1932 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1933 | struct inode *ino = lo->plh_inode; |
78096cca | 1934 | LIST_HEAD(free_me); |
540d9864 TM |
1935 | |
1936 | if (!pnfs_sanity_check_layout_range(&res->range)) | |
1b3c6d07 | 1937 | return ERR_PTR(-EINVAL); |
b1f69b75 AA |
1938 | |
1939 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1940 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
1b3c6d07 | 1941 | if (IS_ERR_OR_NULL(lseg)) { |
b1f69b75 | 1942 | if (!lseg) |
1b3c6d07 JL |
1943 | lseg = ERR_PTR(-ENOMEM); |
1944 | ||
1945 | dprintk("%s: Could not allocate layout: error %ld\n", | |
1946 | __func__, PTR_ERR(lseg)); | |
1947 | return lseg; | |
b1f69b75 AA |
1948 | } |
1949 | ||
119cef97 | 1950 | pnfs_init_lseg(lo, lseg, &res->range, &res->stateid); |
1013df61 | 1951 | |
b1f69b75 | 1952 | spin_lock(&ino->i_lock); |
e1c06f80 | 1953 | if (pnfs_layoutgets_blocked(lo)) { |
43f1b3da | 1954 | dprintk("%s forget reply due to state\n", __func__); |
1b3c6d07 | 1955 | goto out_forget; |
43f1b3da | 1956 | } |
038d6493 | 1957 | |
9888d837 TM |
1958 | if (!pnfs_layout_is_valid(lo)) { |
1959 | /* We have a completely new layout */ | |
1960 | pnfs_set_layout_stateid(lo, &res->stateid, true); | |
1961 | } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { | |
362f7474 CH |
1962 | /* existing state ID, make sure the sequence number matches. */ |
1963 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
1964 | dprintk("%s forget reply due to sequence\n", __func__); | |
1b3c6d07 | 1965 | goto out_forget; |
362f7474 CH |
1966 | } |
1967 | pnfs_set_layout_stateid(lo, &res->stateid, false); | |
1968 | } else { | |
1969 | /* | |
1970 | * We got an entirely new state ID. Mark all segments for the | |
9888d837 | 1971 | * inode invalid, and retry the layoutget |
362f7474 | 1972 | */ |
d9b61708 | 1973 | pnfs_mark_layout_stateid_invalid(lo, &free_me); |
9888d837 | 1974 | goto out_forget; |
362f7474 | 1975 | } |
038d6493 | 1976 | |
9369a431 | 1977 | pnfs_get_lseg(lseg); |
03772d2f | 1978 | pnfs_layout_insert_lseg(lo, lseg, &free_me); |
8e0acf90 | 1979 | |
b1f69b75 | 1980 | |
3976143b | 1981 | if (res->return_on_close) |
f7e8917a | 1982 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); |
f7e8917a | 1983 | |
b1f69b75 | 1984 | spin_unlock(&ino->i_lock); |
78096cca | 1985 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 1986 | return lseg; |
43f1b3da | 1987 | |
1b3c6d07 | 1988 | out_forget: |
43f1b3da FI |
1989 | spin_unlock(&ino->i_lock); |
1990 | lseg->pls_layout = lo; | |
1991 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1b3c6d07 | 1992 | return ERR_PTR(-EAGAIN); |
b1f69b75 AA |
1993 | } |
1994 | ||
2f215968 TM |
1995 | /** |
1996 | * pnfs_mark_matching_lsegs_return - Free or return matching layout segments | |
1997 | * @lo: pointer to layout header | |
1998 | * @tmp_list: list header to be used with pnfs_free_lseg_list() | |
1999 | * @return_range: describe layout segment ranges to be returned | |
2000 | * | |
2001 | * This function is mainly intended for use by layoutrecall. It attempts | |
2002 | * to free the layout segment immediately, or else to mark it for return | |
2003 | * as soon as its reference count drops to zero. | |
2004 | */ | |
10335556 | 2005 | int |
016256df PT |
2006 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, |
2007 | struct list_head *tmp_list, | |
6d597e17 JL |
2008 | const struct pnfs_layout_range *return_range, |
2009 | u32 seq) | |
016256df PT |
2010 | { |
2011 | struct pnfs_layout_segment *lseg, *next; | |
10335556 | 2012 | int remaining = 0; |
016256df PT |
2013 | |
2014 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
2015 | ||
2016 | if (list_empty(&lo->plh_segs)) | |
10335556 | 2017 | return 0; |
016256df | 2018 | |
fc7ff367 | 2019 | assert_spin_locked(&lo->plh_inode->i_lock); |
016256df PT |
2020 | |
2021 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
e036f464 | 2022 | if (pnfs_match_lseg_recall(lseg, return_range, seq)) { |
016256df PT |
2023 | dprintk("%s: marking lseg %p iomode %d " |
2024 | "offset %llu length %llu\n", __func__, | |
2025 | lseg, lseg->pls_range.iomode, | |
2026 | lseg->pls_range.offset, | |
2027 | lseg->pls_range.length); | |
2f215968 TM |
2028 | if (mark_lseg_invalid(lseg, tmp_list)) |
2029 | continue; | |
2030 | remaining++; | |
016256df | 2031 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); |
016256df | 2032 | } |
6d597e17 JL |
2033 | |
2034 | if (remaining) | |
2035 | pnfs_set_plh_return_info(lo, return_range->iomode, seq); | |
2036 | ||
10335556 | 2037 | return remaining; |
016256df PT |
2038 | } |
2039 | ||
2040 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
2041 | struct pnfs_layout_segment *lseg) | |
2042 | { | |
2043 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
016256df PT |
2044 | struct pnfs_layout_range range = { |
2045 | .iomode = lseg->pls_range.iomode, | |
2046 | .offset = 0, | |
2047 | .length = NFS4_MAX_UINT64, | |
2048 | }; | |
10335556 | 2049 | bool return_now = false; |
016256df PT |
2050 | |
2051 | spin_lock(&inode->i_lock); | |
bdebfccd TM |
2052 | if (!pnfs_layout_is_valid(lo)) { |
2053 | spin_unlock(&inode->i_lock); | |
2054 | return; | |
2055 | } | |
2d6cf5ab | 2056 | pnfs_set_plh_return_info(lo, range.iomode, 0); |
6604b203 TM |
2057 | /* Block LAYOUTGET */ |
2058 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
016256df PT |
2059 | /* |
2060 | * mark all matching lsegs so that we are sure to have no live | |
2061 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
2062 | * for how it works. | |
2063 | */ | |
68f74479 | 2064 | if (!pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0)) { |
10335556 | 2065 | nfs4_stateid stateid; |
e5fd1904 | 2066 | enum pnfs_iomode iomode; |
10335556 | 2067 | |
e5fd1904 | 2068 | return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode); |
10335556 TM |
2069 | spin_unlock(&inode->i_lock); |
2070 | if (return_now) | |
2071 | pnfs_send_layoutreturn(lo, &stateid, iomode, false); | |
2072 | } else { | |
2073 | spin_unlock(&inode->i_lock); | |
2074 | nfs_commit_inode(inode, 0); | |
2075 | } | |
016256df PT |
2076 | } |
2077 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); | |
2078 | ||
b3230e80 TM |
2079 | void |
2080 | pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio) | |
2081 | { | |
2082 | if (pgio->pg_lseg == NULL || | |
2083 | test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags)) | |
2084 | return; | |
2085 | pnfs_put_lseg(pgio->pg_lseg); | |
2086 | pgio->pg_lseg = NULL; | |
2087 | } | |
2088 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout); | |
2089 | ||
d8007d4d TM |
2090 | void |
2091 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
2092 | { | |
1fd937bd PT |
2093 | u64 rd_size = req->wb_bytes; |
2094 | ||
b3230e80 | 2095 | pnfs_generic_pg_check_layout(pgio); |
cb5d04bc PT |
2096 | if (pgio->pg_lseg == NULL) { |
2097 | if (pgio->pg_dreq == NULL) | |
2098 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
2099 | else | |
2100 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
2101 | ||
2102 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
2103 | req->wb_context, | |
2104 | req_offset(req), | |
2105 | rd_size, | |
2106 | IOMODE_READ, | |
c7d73af2 | 2107 | false, |
cb5d04bc | 2108 | GFP_KERNEL); |
d600ad1f PT |
2109 | if (IS_ERR(pgio->pg_lseg)) { |
2110 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
2111 | pgio->pg_lseg = NULL; | |
2112 | return; | |
2113 | } | |
cb5d04bc | 2114 | } |
e885de1a TM |
2115 | /* If no lseg, fall back to read through mds */ |
2116 | if (pgio->pg_lseg == NULL) | |
1f945357 | 2117 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 2118 | |
d8007d4d TM |
2119 | } |
2120 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
2121 | ||
2122 | void | |
6296556f PT |
2123 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
2124 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 2125 | { |
b3230e80 | 2126 | pnfs_generic_pg_check_layout(pgio); |
d600ad1f | 2127 | if (pgio->pg_lseg == NULL) { |
cb5d04bc PT |
2128 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
2129 | req->wb_context, | |
2130 | req_offset(req), | |
2131 | wb_size, | |
2132 | IOMODE_RW, | |
c7d73af2 | 2133 | false, |
cb5d04bc | 2134 | GFP_NOFS); |
d600ad1f PT |
2135 | if (IS_ERR(pgio->pg_lseg)) { |
2136 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
2137 | pgio->pg_lseg = NULL; | |
2138 | return; | |
2139 | } | |
2140 | } | |
e885de1a TM |
2141 | /* If no lseg, fall back to write through mds */ |
2142 | if (pgio->pg_lseg == NULL) | |
1f945357 | 2143 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
2144 | } |
2145 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
2146 | ||
180bb5ec WAA |
2147 | void |
2148 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
2149 | { | |
2150 | if (desc->pg_lseg) { | |
2151 | pnfs_put_lseg(desc->pg_lseg); | |
2152 | desc->pg_lseg = NULL; | |
2153 | } | |
2154 | } | |
2155 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
2156 | ||
b4fdac1a WAA |
2157 | /* |
2158 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
2159 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
2160 | */ | |
2161 | size_t | |
a7d42ddb WAA |
2162 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
2163 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 2164 | { |
0f9c429e | 2165 | unsigned int size; |
c5e20cb7 | 2166 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
2167 | |
2168 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
2169 | if (!size) |
2170 | return 0; | |
94ad1c80 | 2171 | |
19982ba8 | 2172 | /* |
c5e20cb7 WAA |
2173 | * 'size' contains the number of bytes left in the current page (up |
2174 | * to the original size asked for in @req->wb_bytes). | |
2175 | * | |
2176 | * Calculate how many bytes are left in the layout segment | |
2177 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
2178 | * |
2179 | * Please also note that 'end_offset' is actually the offset of the | |
2180 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
2181 | * | |
2182 | */ | |
19b54848 | 2183 | if (pgio->pg_lseg) { |
17822b20 | 2184 | seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset, |
c5e20cb7 WAA |
2185 | pgio->pg_lseg->pls_range.length); |
2186 | req_start = req_offset(req); | |
7c13789e | 2187 | WARN_ON_ONCE(req_start >= seg_end); |
c5e20cb7 | 2188 | /* start of request is past the last byte of this segment */ |
7c13789e WAA |
2189 | if (req_start >= seg_end) { |
2190 | /* reference the new lseg */ | |
2191 | if (pgio->pg_ops->pg_cleanup) | |
2192 | pgio->pg_ops->pg_cleanup(pgio); | |
2193 | if (pgio->pg_ops->pg_init) | |
2194 | pgio->pg_ops->pg_init(pgio, req); | |
19b54848 | 2195 | return 0; |
7c13789e | 2196 | } |
c5e20cb7 WAA |
2197 | |
2198 | /* adjust 'size' iff there are fewer bytes left in the | |
2199 | * segment than what nfs_generic_pg_test returned */ | |
2200 | seg_left = seg_end - req_start; | |
2201 | if (seg_left < size) | |
2202 | size = (unsigned int)seg_left; | |
19b54848 | 2203 | } |
0f9c429e | 2204 | |
19b54848 | 2205 | return size; |
94ad1c80 | 2206 | } |
89a58e32 | 2207 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 2208 | |
53113ad3 | 2209 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
2210 | { |
2211 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
2212 | |
2213 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
2214 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
2215 | hdr->completion_ops); | |
c7070113 | 2216 | set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags); |
53113ad3 | 2217 | return nfs_pageio_resend(&pgio, hdr); |
e2fecb21 | 2218 | } |
e7dd79af | 2219 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 2220 | |
d45f60c6 | 2221 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2222 | { |
cd841605 FI |
2223 | |
2224 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
2225 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2226 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2227 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2228 | } |
6c75dc0d | 2229 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2230 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
2231 | } |
2232 | ||
d20581aa BH |
2233 | /* |
2234 | * Called by non rpc-based layout drivers | |
2235 | */ | |
d45f60c6 | 2236 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 2237 | { |
f8417b48 | 2238 | if (likely(!hdr->pnfs_error)) { |
67af7611 TM |
2239 | pnfs_set_layoutcommit(hdr->inode, hdr->lseg, |
2240 | hdr->mds_offset + hdr->res.count); | |
d45f60c6 | 2241 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2242 | } |
2243 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); | |
2244 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2245 | pnfs_ld_handle_write_error(hdr); |
2246 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 2247 | } |
d20581aa | 2248 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 2249 | |
dce81290 TM |
2250 | static void |
2251 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2252 | struct nfs_pgio_header *hdr) |
dce81290 | 2253 | { |
48d635f1 | 2254 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2255 | |
6c75dc0d | 2256 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2257 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 2258 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 2259 | mirror->pg_recoalesce = 1; |
6c75dc0d | 2260 | } |
d45f60c6 | 2261 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2262 | hdr->release(hdr); |
dce81290 TM |
2263 | } |
2264 | ||
2265 | static enum pnfs_try_status | |
d45f60c6 | 2266 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
2267 | const struct rpc_call_ops *call_ops, |
2268 | struct pnfs_layout_segment *lseg, | |
2269 | int how) | |
0382b744 | 2270 | { |
cd841605 | 2271 | struct inode *inode = hdr->inode; |
0382b744 AA |
2272 | enum pnfs_try_status trypnfs; |
2273 | struct nfs_server *nfss = NFS_SERVER(inode); | |
2274 | ||
cd841605 | 2275 | hdr->mds_ops = call_ops; |
0382b744 AA |
2276 | |
2277 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
2278 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
2279 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 2280 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 2281 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
2282 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2283 | return trypnfs; | |
2284 | } | |
2285 | ||
dce81290 | 2286 | static void |
7f714720 WAA |
2287 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
2288 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 2289 | { |
dce81290 TM |
2290 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2291 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2292 | enum pnfs_try_status trypnfs; |
dce81290 | 2293 | |
d45f60c6 | 2294 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
37f8aa16 TM |
2295 | switch (trypnfs) { |
2296 | case PNFS_NOT_ATTEMPTED: | |
d45f60c6 | 2297 | pnfs_write_through_mds(desc, hdr); |
37f8aa16 TM |
2298 | case PNFS_ATTEMPTED: |
2299 | break; | |
2300 | case PNFS_TRY_AGAIN: | |
2301 | /* cleanup hdr and prepare to redo pnfs */ | |
2302 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { | |
2303 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); | |
2304 | list_splice_init(&hdr->pages, &mirror->pg_list); | |
2305 | mirror->pg_recoalesce = 1; | |
2306 | } | |
2307 | hdr->mds_ops->rpc_release(hdr); | |
2308 | } | |
dce81290 TM |
2309 | } |
2310 | ||
6c75dc0d FI |
2311 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
2312 | { | |
9369a431 | 2313 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2314 | nfs_pgio_header_free(hdr); |
6c75dc0d FI |
2315 | } |
2316 | ||
dce81290 TM |
2317 | int |
2318 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
2319 | { | |
6c75dc0d | 2320 | struct nfs_pgio_header *hdr; |
dce81290 TM |
2321 | int ret; |
2322 | ||
1e7f3a48 WAA |
2323 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2324 | if (!hdr) { | |
2bff2288 PT |
2325 | desc->pg_error = -ENOMEM; |
2326 | return desc->pg_error; | |
dce81290 | 2327 | } |
6c75dc0d | 2328 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 2329 | |
9369a431 | 2330 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2331 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2332 | if (!ret) |
7f714720 | 2333 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 2334 | |
6c75dc0d | 2335 | return ret; |
dce81290 TM |
2336 | } |
2337 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
2338 | ||
53113ad3 | 2339 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
2340 | { |
2341 | struct nfs_pageio_descriptor pgio; | |
2342 | ||
1acbbb4e | 2343 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
2344 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
2345 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 2346 | } |
e7dd79af | 2347 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 2348 | |
d45f60c6 | 2349 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2350 | { |
cd841605 FI |
2351 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
2352 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2353 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2354 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2355 | } |
4db6e0b7 | 2356 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2357 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
2358 | } |
2359 | ||
d20581aa BH |
2360 | /* |
2361 | * Called by non rpc-based layout drivers | |
2362 | */ | |
d45f60c6 | 2363 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 2364 | { |
bfc505de | 2365 | if (likely(!hdr->pnfs_error)) |
d45f60c6 | 2366 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2367 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
2368 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2369 | pnfs_ld_handle_read_error(hdr); |
2370 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
2371 | } |
2372 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
2373 | ||
493292dd TM |
2374 | static void |
2375 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2376 | struct nfs_pgio_header *hdr) |
493292dd | 2377 | { |
48d635f1 | 2378 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2379 | |
4db6e0b7 | 2380 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2381 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 2382 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 2383 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 2384 | } |
d45f60c6 | 2385 | nfs_pgio_data_destroy(hdr); |
1ca018d2 | 2386 | hdr->release(hdr); |
493292dd TM |
2387 | } |
2388 | ||
64419a9b AA |
2389 | /* |
2390 | * Call the appropriate parallel I/O subsystem read function. | |
2391 | */ | |
493292dd | 2392 | static enum pnfs_try_status |
d45f60c6 | 2393 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
2394 | const struct rpc_call_ops *call_ops, |
2395 | struct pnfs_layout_segment *lseg) | |
64419a9b | 2396 | { |
cd841605 | 2397 | struct inode *inode = hdr->inode; |
64419a9b AA |
2398 | struct nfs_server *nfss = NFS_SERVER(inode); |
2399 | enum pnfs_try_status trypnfs; | |
2400 | ||
cd841605 | 2401 | hdr->mds_ops = call_ops; |
64419a9b AA |
2402 | |
2403 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 2404 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 2405 | |
d45f60c6 | 2406 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 2407 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 2408 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
2409 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2410 | return trypnfs; | |
2411 | } | |
863a3c6c | 2412 | |
ceb11e13 | 2413 | /* Resend all requests through pnfs. */ |
1b1bc66b | 2414 | void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) |
ceb11e13 PT |
2415 | { |
2416 | struct nfs_pageio_descriptor pgio; | |
2417 | ||
1b1bc66b | 2418 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
54e4a0df TM |
2419 | /* Prevent deadlocks with layoutreturn! */ |
2420 | pnfs_put_lseg(hdr->lseg); | |
2421 | hdr->lseg = NULL; | |
2422 | ||
1b1bc66b WAA |
2423 | nfs_pageio_init_read(&pgio, hdr->inode, false, |
2424 | hdr->completion_ops); | |
2425 | hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr); | |
2426 | } | |
ceb11e13 PT |
2427 | } |
2428 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
2429 | ||
493292dd | 2430 | static void |
7f714720 | 2431 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 2432 | { |
493292dd TM |
2433 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2434 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2435 | enum pnfs_try_status trypnfs; |
493292dd | 2436 | |
d45f60c6 | 2437 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
6aeafd05 TM |
2438 | switch (trypnfs) { |
2439 | case PNFS_NOT_ATTEMPTED: | |
d45f60c6 | 2440 | pnfs_read_through_mds(desc, hdr); |
6aeafd05 TM |
2441 | case PNFS_ATTEMPTED: |
2442 | break; | |
2443 | case PNFS_TRY_AGAIN: | |
2444 | /* cleanup hdr and prepare to redo pnfs */ | |
2445 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { | |
2446 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); | |
2447 | list_splice_init(&hdr->pages, &mirror->pg_list); | |
2448 | mirror->pg_recoalesce = 1; | |
2449 | } | |
2450 | hdr->mds_ops->rpc_release(hdr); | |
2451 | } | |
493292dd TM |
2452 | } |
2453 | ||
4db6e0b7 FI |
2454 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
2455 | { | |
9369a431 | 2456 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2457 | nfs_pgio_header_free(hdr); |
4db6e0b7 FI |
2458 | } |
2459 | ||
493292dd TM |
2460 | int |
2461 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
2462 | { | |
4db6e0b7 | 2463 | struct nfs_pgio_header *hdr; |
493292dd TM |
2464 | int ret; |
2465 | ||
1e7f3a48 WAA |
2466 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2467 | if (!hdr) { | |
2bff2288 PT |
2468 | desc->pg_error = -ENOMEM; |
2469 | return desc->pg_error; | |
493292dd | 2470 | } |
4db6e0b7 | 2471 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 2472 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2473 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2474 | if (!ret) |
7f714720 | 2475 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 2476 | return ret; |
493292dd TM |
2477 | } |
2478 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
2479 | ||
71244d9b TM |
2480 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
2481 | { | |
2482 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
2483 | ||
2484 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 2485 | smp_mb__after_atomic(); |
71244d9b TM |
2486 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
2487 | } | |
2488 | ||
863a3c6c | 2489 | /* |
a9bae566 | 2490 | * There can be multiple RW segments. |
863a3c6c | 2491 | */ |
a9bae566 | 2492 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 2493 | { |
a9bae566 | 2494 | struct pnfs_layout_segment *lseg; |
863a3c6c | 2495 | |
a9bae566 PT |
2496 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
2497 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 2498 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
2499 | list_add(&lseg->pls_lc_list, listp); |
2500 | } | |
863a3c6c AA |
2501 | } |
2502 | ||
a073dbff TM |
2503 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
2504 | { | |
2505 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
2506 | |
2507 | /* Matched by references in pnfs_set_layoutcommit */ | |
2508 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
2509 | list_del_init(&lseg->pls_lc_list); | |
2510 | pnfs_put_lseg(lseg); | |
2511 | } | |
2512 | ||
71244d9b | 2513 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
2514 | } |
2515 | ||
1b0ae068 PT |
2516 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
2517 | { | |
b9e028fd | 2518 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
2519 | } |
2520 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
2521 | ||
863a3c6c | 2522 | void |
67af7611 TM |
2523 | pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, |
2524 | loff_t end_pos) | |
863a3c6c | 2525 | { |
cd841605 | 2526 | struct nfs_inode *nfsi = NFS_I(inode); |
79a48a1f | 2527 | bool mark_as_dirty = false; |
863a3c6c | 2528 | |
cd841605 | 2529 | spin_lock(&inode->i_lock); |
863a3c6c | 2530 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
29559b11 | 2531 | nfsi->layout->plh_lwb = end_pos; |
79a48a1f | 2532 | mark_as_dirty = true; |
863a3c6c | 2533 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 2534 | __func__, inode->i_ino); |
29559b11 TM |
2535 | } else if (end_pos > nfsi->layout->plh_lwb) |
2536 | nfsi->layout->plh_lwb = end_pos; | |
67af7611 | 2537 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) { |
a9bae566 | 2538 | /* references matched in nfs4_layoutcommit_release */ |
67af7611 | 2539 | pnfs_get_lseg(lseg); |
a9bae566 | 2540 | } |
cd841605 | 2541 | spin_unlock(&inode->i_lock); |
acff5880 | 2542 | dprintk("%s: lseg %p end_pos %llu\n", |
67af7611 | 2543 | __func__, lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
2544 | |
2545 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
2546 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
2547 | if (mark_as_dirty) | |
cd841605 | 2548 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
2549 | } |
2550 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
2551 | ||
db29c089 AA |
2552 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
2553 | { | |
2554 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
2555 | ||
2556 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
2557 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 2558 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
2559 | } |
2560 | ||
de4b15c7 AA |
2561 | /* |
2562 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
2563 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
2564 | * data to disk to allow the server to recover the data if it crashes. | |
2565 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
2566 | * is off, and a COMMIT is sent to a data server, or | |
2567 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
2568 | */ | |
863a3c6c | 2569 | int |
ef311537 | 2570 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 2571 | { |
5f919c9f | 2572 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
2573 | struct nfs4_layoutcommit_data *data; |
2574 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 2575 | loff_t end_pos; |
71244d9b | 2576 | int status; |
863a3c6c | 2577 | |
71244d9b | 2578 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
2579 | return 0; |
2580 | ||
71244d9b | 2581 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 2582 | |
71244d9b | 2583 | status = -EAGAIN; |
92407e75 | 2584 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
2585 | if (!sync) |
2586 | goto out; | |
74316201 | 2587 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
2588 | NFS_INO_LAYOUTCOMMITTING, |
2589 | nfs_wait_bit_killable, | |
2590 | TASK_KILLABLE); | |
92407e75 | 2591 | if (status) |
71244d9b | 2592 | goto out; |
92407e75 PT |
2593 | } |
2594 | ||
71244d9b TM |
2595 | status = -ENOMEM; |
2596 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
2597 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
2598 | if (!data) | |
2599 | goto clear_layoutcommitting; | |
2600 | ||
2601 | status = 0; | |
de4b15c7 | 2602 | spin_lock(&inode->i_lock); |
71244d9b TM |
2603 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
2604 | goto out_unlock; | |
a9bae566 | 2605 | |
71244d9b | 2606 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 2607 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 2608 | |
acff5880 | 2609 | end_pos = nfsi->layout->plh_lwb; |
863a3c6c | 2610 | |
f597c537 | 2611 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
2612 | spin_unlock(&inode->i_lock); |
2613 | ||
2614 | data->args.inode = inode; | |
9fa40758 | 2615 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
2616 | nfs_fattr_init(&data->fattr); |
2617 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
2618 | data->res.fattr = &data->fattr; | |
2e18d4d8 TM |
2619 | if (end_pos != 0) |
2620 | data->args.lastbytewritten = end_pos - 1; | |
2621 | else | |
2622 | data->args.lastbytewritten = U64_MAX; | |
863a3c6c AA |
2623 | data->res.server = NFS_SERVER(inode); |
2624 | ||
5f919c9f CH |
2625 | if (ld->prepare_layoutcommit) { |
2626 | status = ld->prepare_layoutcommit(&data->args); | |
2627 | if (status) { | |
3471648a | 2628 | put_rpccred(data->cred); |
5f919c9f | 2629 | spin_lock(&inode->i_lock); |
29559b11 TM |
2630 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); |
2631 | if (end_pos > nfsi->layout->plh_lwb) | |
5f919c9f | 2632 | nfsi->layout->plh_lwb = end_pos; |
3471648a | 2633 | goto out_unlock; |
5f919c9f CH |
2634 | } |
2635 | } | |
2636 | ||
2637 | ||
863a3c6c AA |
2638 | status = nfs4_proc_layoutcommit(data, sync); |
2639 | out: | |
92407e75 PT |
2640 | if (status) |
2641 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
2642 | dprintk("<-- %s status %d\n", __func__, status); |
2643 | return status; | |
71244d9b TM |
2644 | out_unlock: |
2645 | spin_unlock(&inode->i_lock); | |
92407e75 | 2646 | kfree(data); |
71244d9b TM |
2647 | clear_layoutcommitting: |
2648 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 2649 | goto out; |
863a3c6c | 2650 | } |
72cff449 | 2651 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a | 2652 | |
5bb89b47 TM |
2653 | int |
2654 | pnfs_generic_sync(struct inode *inode, bool datasync) | |
2655 | { | |
2656 | return pnfs_layoutcommit_inode(inode, true); | |
2657 | } | |
2658 | EXPORT_SYMBOL_GPL(pnfs_generic_sync); | |
2659 | ||
82be417a AA |
2660 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) |
2661 | { | |
2662 | struct nfs4_threshold *thp; | |
2663 | ||
2664 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
2665 | if (!thp) { | |
2666 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
2667 | return NULL; | |
2668 | } | |
2669 | return thp; | |
2670 | } | |
8733408d | 2671 | |
865a7ecb | 2672 | #if IS_ENABLED(CONFIG_NFS_V4_2) |
8733408d | 2673 | int |
c8ad8894 | 2674 | pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags) |
8733408d PT |
2675 | { |
2676 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
2677 | struct nfs_server *server = NFS_SERVER(inode); | |
1bfe3b25 | 2678 | struct nfs_inode *nfsi = NFS_I(inode); |
8733408d PT |
2679 | struct nfs42_layoutstat_data *data; |
2680 | struct pnfs_layout_hdr *hdr; | |
2681 | int status = 0; | |
2682 | ||
2683 | if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats) | |
2684 | goto out; | |
2685 | ||
6c5a0d89 TM |
2686 | if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS)) |
2687 | goto out; | |
2688 | ||
1bfe3b25 PT |
2689 | if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags)) |
2690 | goto out; | |
2691 | ||
8733408d PT |
2692 | spin_lock(&inode->i_lock); |
2693 | if (!NFS_I(inode)->layout) { | |
2694 | spin_unlock(&inode->i_lock); | |
f538d0ba | 2695 | goto out_clear_layoutstats; |
8733408d PT |
2696 | } |
2697 | hdr = NFS_I(inode)->layout; | |
2698 | pnfs_get_layout_hdr(hdr); | |
2699 | spin_unlock(&inode->i_lock); | |
2700 | ||
c8ad8894 | 2701 | data = kzalloc(sizeof(*data), gfp_flags); |
8733408d PT |
2702 | if (!data) { |
2703 | status = -ENOMEM; | |
2704 | goto out_put; | |
2705 | } | |
2706 | ||
2707 | data->args.fh = NFS_FH(inode); | |
2708 | data->args.inode = inode; | |
8733408d PT |
2709 | status = ld->prepare_layoutstats(&data->args); |
2710 | if (status) | |
2711 | goto out_free; | |
2712 | ||
2713 | status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data); | |
2714 | ||
2715 | out: | |
2716 | dprintk("%s returns %d\n", __func__, status); | |
2717 | return status; | |
2718 | ||
2719 | out_free: | |
2720 | kfree(data); | |
2721 | out_put: | |
2722 | pnfs_put_layout_hdr(hdr); | |
f538d0ba | 2723 | out_clear_layoutstats: |
1bfe3b25 PT |
2724 | smp_mb__before_atomic(); |
2725 | clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags); | |
2726 | smp_mb__after_atomic(); | |
8733408d PT |
2727 | goto out; |
2728 | } | |
2729 | EXPORT_SYMBOL_GPL(pnfs_report_layoutstat); | |
865a7ecb | 2730 | #endif |
bbf58bf3 TM |
2731 | |
2732 | unsigned int layoutstats_timer; | |
2733 | module_param(layoutstats_timer, uint, 0644); | |
2734 | EXPORT_SYMBOL_GPL(layoutstats_timer); |