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" |
1b146fcf | 40 | #include "nfs4_fs.h" |
85e174ba RL |
41 | |
42 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
25c75333 | 43 | #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) |
85e174ba | 44 | |
02c35fca FI |
45 | /* Locking: |
46 | * | |
47 | * pnfs_spinlock: | |
48 | * protects pnfs_modules_tbl. | |
49 | */ | |
50 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
51 | ||
52 | /* | |
53 | * pnfs_modules_tbl holds all pnfs modules | |
54 | */ | |
55 | static LIST_HEAD(pnfs_modules_tbl); | |
56 | ||
13c13a6a | 57 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo); |
68f74479 TM |
58 | static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo, |
59 | struct list_head *free_me, | |
60 | const struct pnfs_layout_range *range, | |
61 | u32 seq); | |
fe1cf946 TM |
62 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
63 | struct list_head *tmp_list); | |
aa1e0e3a | 64 | |
02c35fca FI |
65 | /* Return the registered pnfs layout driver module matching given id */ |
66 | static struct pnfs_layoutdriver_type * | |
67 | find_pnfs_driver_locked(u32 id) | |
68 | { | |
69 | struct pnfs_layoutdriver_type *local; | |
70 | ||
71 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
72 | if (local->id == id) | |
73 | goto out; | |
74 | local = NULL; | |
75 | out: | |
76 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
77 | return local; | |
78 | } | |
79 | ||
85e174ba RL |
80 | static struct pnfs_layoutdriver_type * |
81 | find_pnfs_driver(u32 id) | |
82 | { | |
02c35fca FI |
83 | struct pnfs_layoutdriver_type *local; |
84 | ||
85 | spin_lock(&pnfs_spinlock); | |
86 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
87 | if (local != NULL && !try_module_get(local->owner)) { |
88 | dprintk("%s: Could not grab reference on module\n", __func__); | |
89 | local = NULL; | |
90 | } | |
02c35fca FI |
91 | spin_unlock(&pnfs_spinlock); |
92 | return local; | |
85e174ba RL |
93 | } |
94 | ||
95 | void | |
96 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
97 | { | |
738fd0f3 BH |
98 | if (nfss->pnfs_curr_ld) { |
99 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
100 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
101 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
102 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
103 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 104 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 105 | } |
85e174ba RL |
106 | nfss->pnfs_curr_ld = NULL; |
107 | } | |
108 | ||
ca440c38 JL |
109 | /* |
110 | * When the server sends a list of layout types, we choose one in the order | |
111 | * given in the list below. | |
112 | * | |
113 | * FIXME: should this list be configurable in some fashion? module param? | |
114 | * mount option? something else? | |
115 | */ | |
116 | static const u32 ld_prefs[] = { | |
117 | LAYOUT_SCSI, | |
118 | LAYOUT_BLOCK_VOLUME, | |
119 | LAYOUT_OSD2_OBJECTS, | |
120 | LAYOUT_FLEX_FILES, | |
121 | LAYOUT_NFSV4_1_FILES, | |
122 | 0 | |
123 | }; | |
124 | ||
125 | static int | |
126 | ld_cmp(const void *e1, const void *e2) | |
127 | { | |
128 | u32 ld1 = *((u32 *)e1); | |
129 | u32 ld2 = *((u32 *)e2); | |
130 | int i; | |
131 | ||
132 | for (i = 0; ld_prefs[i] != 0; i++) { | |
133 | if (ld1 == ld_prefs[i]) | |
134 | return -1; | |
135 | ||
136 | if (ld2 == ld_prefs[i]) | |
137 | return 1; | |
138 | } | |
139 | return 0; | |
140 | } | |
141 | ||
85e174ba RL |
142 | /* |
143 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
144 | * Currently only one pNFS layout driver per filesystem is supported. | |
145 | * | |
3132e49e | 146 | * @ids array of layout types supported by MDS. |
85e174ba RL |
147 | */ |
148 | void | |
738fd0f3 | 149 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
ca440c38 | 150 | struct nfs_fsinfo *fsinfo) |
85e174ba RL |
151 | { |
152 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
3132e49e | 153 | u32 id; |
ca440c38 | 154 | int i; |
85e174ba | 155 | |
19274716 AS |
156 | if (fsinfo->nlayouttypes == 0) |
157 | goto out_no_driver; | |
85e174ba RL |
158 | if (!(server->nfs_client->cl_exchange_flags & |
159 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
3132e49e JL |
160 | printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n", |
161 | __func__, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
162 | goto out_no_driver; |
163 | } | |
3132e49e | 164 | |
ca440c38 JL |
165 | sort(fsinfo->layouttype, fsinfo->nlayouttypes, |
166 | sizeof(*fsinfo->layouttype), ld_cmp, NULL); | |
3132e49e | 167 | |
ca440c38 JL |
168 | for (i = 0; i < fsinfo->nlayouttypes; i++) { |
169 | id = fsinfo->layouttype[i]; | |
85e174ba | 170 | ld_type = find_pnfs_driver(id); |
ca440c38 JL |
171 | if (!ld_type) { |
172 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, | |
173 | id); | |
174 | ld_type = find_pnfs_driver(id); | |
175 | } | |
176 | if (ld_type) | |
177 | break; | |
85e174ba | 178 | } |
3132e49e JL |
179 | |
180 | if (!ld_type) { | |
ca440c38 | 181 | dprintk("%s: No pNFS module found!\n", __func__); |
3132e49e JL |
182 | goto out_no_driver; |
183 | } | |
184 | ||
85e174ba | 185 | server->pnfs_curr_ld = ld_type; |
738fd0f3 BH |
186 | if (ld_type->set_layoutdriver |
187 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
188 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
189 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
190 | module_put(ld_type->owner); |
191 | goto out_no_driver; | |
192 | } | |
2a4c8994 TM |
193 | /* Bump the MDS count */ |
194 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 195 | |
85e174ba RL |
196 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
197 | return; | |
198 | ||
199 | out_no_driver: | |
200 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
201 | server->pnfs_curr_ld = NULL; | |
202 | } | |
02c35fca FI |
203 | |
204 | int | |
205 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
206 | { | |
207 | int status = -EINVAL; | |
208 | struct pnfs_layoutdriver_type *tmp; | |
209 | ||
210 | if (ld_type->id == 0) { | |
a030889a | 211 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
212 | return status; |
213 | } | |
b1f69b75 | 214 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 215 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
216 | "alloc_lseg and free_lseg.\n", __func__); |
217 | return status; | |
218 | } | |
02c35fca FI |
219 | |
220 | spin_lock(&pnfs_spinlock); | |
221 | tmp = find_pnfs_driver_locked(ld_type->id); | |
222 | if (!tmp) { | |
223 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
224 | status = 0; | |
225 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
226 | ld_type->name); | |
227 | } else { | |
a030889a | 228 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
229 | __func__, ld_type->id); |
230 | } | |
231 | spin_unlock(&pnfs_spinlock); | |
232 | ||
233 | return status; | |
234 | } | |
235 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
236 | ||
237 | void | |
238 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
239 | { | |
240 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
241 | spin_lock(&pnfs_spinlock); | |
242 | list_del(&ld_type->pnfs_tblid); | |
243 | spin_unlock(&pnfs_spinlock); | |
244 | } | |
245 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 246 | |
b1f69b75 AA |
247 | /* |
248 | * pNFS client layout cache | |
249 | */ | |
250 | ||
cc6e5340 | 251 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 252 | void |
70c3bd2b | 253 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 254 | { |
2b28a7be | 255 | refcount_inc(&lo->plh_refcount); |
e5e94017 BH |
256 | } |
257 | ||
636fb9c8 BH |
258 | static struct pnfs_layout_hdr * |
259 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
260 | { | |
261 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
57934278 | 262 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
263 | } |
264 | ||
265 | static void | |
266 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
267 | { | |
9c626381 TM |
268 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
269 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
270 | ||
cf6605d1 | 271 | if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) { |
9c626381 TM |
272 | struct nfs_client *clp = server->nfs_client; |
273 | ||
274 | spin_lock(&clp->cl_lock); | |
cf6605d1 | 275 | list_del_rcu(&lo->plh_layouts); |
9c626381 TM |
276 | spin_unlock(&clp->cl_lock); |
277 | } | |
a52458b4 | 278 | put_cred(lo->plh_lc_cred); |
57934278 | 279 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
280 | } |
281 | ||
e5e94017 | 282 | static void |
6622c3ea | 283 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 284 | { |
bb346f63 | 285 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 286 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
287 | nfsi->layout = NULL; |
288 | /* Reset MDS Threshold I/O counters */ | |
289 | nfsi->write_io = 0; | |
290 | nfsi->read_io = 0; | |
e5e94017 BH |
291 | } |
292 | ||
b1f69b75 | 293 | void |
70c3bd2b | 294 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 295 | { |
9c6376eb | 296 | struct inode *inode; |
cc6e5340 | 297 | |
9c6376eb TM |
298 | if (!lo) |
299 | return; | |
300 | inode = lo->plh_inode; | |
13c13a6a TM |
301 | pnfs_layoutreturn_before_put_layout_hdr(lo); |
302 | ||
2b28a7be | 303 | if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { |
566f8737 PT |
304 | if (!list_empty(&lo->plh_segs)) |
305 | WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n"); | |
6622c3ea | 306 | pnfs_detach_layout_hdr(lo); |
cc6e5340 | 307 | spin_unlock(&inode->i_lock); |
6622c3ea | 308 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 309 | } |
974cec8c AA |
310 | } |
311 | ||
b5fdf841 TM |
312 | static struct inode * |
313 | pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo) | |
314 | { | |
315 | struct inode *inode = igrab(lo->plh_inode); | |
316 | if (inode) | |
317 | return inode; | |
318 | set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags); | |
319 | return NULL; | |
320 | } | |
321 | ||
4aab9732 TM |
322 | static void |
323 | pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode, | |
324 | u32 seq) | |
325 | { | |
326 | if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode) | |
327 | iomode = IOMODE_ANY; | |
328 | lo->plh_return_iomode = iomode; | |
329 | set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); | |
330 | if (seq != 0) { | |
331 | WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq); | |
332 | lo->plh_return_seq = seq; | |
333 | } | |
334 | } | |
335 | ||
ae5a459d TM |
336 | static void |
337 | pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo) | |
338 | { | |
5466d214 | 339 | struct pnfs_layout_segment *lseg; |
ae5a459d TM |
340 | lo->plh_return_iomode = 0; |
341 | lo->plh_return_seq = 0; | |
342 | clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); | |
5466d214 TM |
343 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
344 | if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
345 | continue; | |
346 | pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0); | |
347 | } | |
ae5a459d TM |
348 | } |
349 | ||
362fb578 TM |
350 | static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo) |
351 | { | |
352 | clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
353 | clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags); | |
354 | smp_mb__after_atomic(); | |
355 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN); | |
356 | rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq); | |
357 | } | |
358 | ||
fe1cf946 TM |
359 | static void |
360 | pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg, | |
361 | struct list_head *free_me) | |
362 | { | |
363 | clear_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
364 | clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
365 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) | |
366 | pnfs_lseg_dec_and_remove_zero(lseg, free_me); | |
367 | if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
368 | pnfs_lseg_dec_and_remove_zero(lseg, free_me); | |
369 | } | |
370 | ||
7380020e | 371 | /* |
30cb3ee2 TM |
372 | * Update the seqid of a layout stateid after receiving |
373 | * NFS4ERR_OLD_STATEID | |
7380020e | 374 | */ |
30cb3ee2 | 375 | bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst, |
ecf84026 TM |
376 | struct pnfs_layout_range *dst_range, |
377 | struct inode *inode) | |
7380020e TM |
378 | { |
379 | struct pnfs_layout_hdr *lo; | |
c16467dc TM |
380 | struct pnfs_layout_range range = { |
381 | .iomode = IOMODE_ANY, | |
382 | .offset = 0, | |
383 | .length = NFS4_MAX_UINT64, | |
384 | }; | |
7380020e | 385 | bool ret = false; |
c16467dc TM |
386 | LIST_HEAD(head); |
387 | int err; | |
7380020e TM |
388 | |
389 | spin_lock(&inode->i_lock); | |
390 | lo = NFS_I(inode)->layout; | |
30cb3ee2 TM |
391 | if (lo && pnfs_layout_is_valid(lo) && |
392 | nfs4_stateid_match_other(dst, &lo->plh_stateid)) { | |
393 | /* Is our call using the most recent seqid? If so, bump it */ | |
394 | if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) { | |
395 | nfs4_stateid_seqid_inc(dst); | |
396 | ret = true; | |
397 | goto out; | |
398 | } | |
399 | /* Try to update the seqid to the most recent */ | |
c16467dc TM |
400 | err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0); |
401 | if (err != -EBUSY) { | |
402 | dst->seqid = lo->plh_stateid.seqid; | |
ecf84026 | 403 | *dst_range = range; |
c16467dc TM |
404 | ret = true; |
405 | } | |
7380020e | 406 | } |
30cb3ee2 | 407 | out: |
7380020e | 408 | spin_unlock(&inode->i_lock); |
c16467dc | 409 | pnfs_free_lseg_list(&head); |
7380020e TM |
410 | return ret; |
411 | } | |
412 | ||
2454dfea TM |
413 | /* |
414 | * Mark a pnfs_layout_hdr and all associated layout segments as invalid | |
415 | * | |
416 | * In order to continue using the pnfs_layout_hdr, a full recovery | |
417 | * is required. | |
418 | * Note that caller must hold inode->i_lock. | |
419 | */ | |
5f46be04 | 420 | int |
2454dfea TM |
421 | pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo, |
422 | struct list_head *lseg_list) | |
423 | { | |
424 | struct pnfs_layout_range range = { | |
425 | .iomode = IOMODE_ANY, | |
426 | .offset = 0, | |
427 | .length = NFS4_MAX_UINT64, | |
428 | }; | |
fe1cf946 | 429 | struct pnfs_layout_segment *lseg, *next; |
2454dfea TM |
430 | |
431 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
fe1cf946 TM |
432 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
433 | pnfs_clear_lseg_state(lseg, lseg_list); | |
5466d214 | 434 | pnfs_clear_layoutreturn_info(lo); |
68f74479 | 435 | pnfs_free_returned_lsegs(lo, lseg_list, &range, 0); |
362fb578 TM |
436 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) && |
437 | !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) | |
438 | pnfs_clear_layoutreturn_waitbit(lo); | |
fe1cf946 | 439 | return !list_empty(&lo->plh_segs); |
2454dfea TM |
440 | } |
441 | ||
b9e028fd TM |
442 | static int |
443 | pnfs_iomode_to_fail_bit(u32 iomode) | |
444 | { | |
445 | return iomode == IOMODE_RW ? | |
446 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
447 | } | |
448 | ||
449 | static void | |
3e621214 | 450 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 451 | { |
25c75333 | 452 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 453 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
2b28a7be | 454 | refcount_inc(&lo->plh_refcount); |
3e621214 TM |
455 | } |
456 | ||
457 | static void | |
458 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
459 | { | |
460 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
2b28a7be | 461 | refcount_dec(&lo->plh_refcount); |
3e621214 TM |
462 | } |
463 | ||
464 | static void | |
465 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
466 | { | |
467 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
468 | struct pnfs_layout_range range = { |
469 | .iomode = iomode, | |
470 | .offset = 0, | |
471 | .length = NFS4_MAX_UINT64, | |
472 | }; | |
473 | LIST_HEAD(head); | |
3e621214 TM |
474 | |
475 | spin_lock(&inode->i_lock); | |
476 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
6d597e17 | 477 | pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0); |
3e621214 | 478 | spin_unlock(&inode->i_lock); |
115ce575 | 479 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
480 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
481 | iomode == IOMODE_RW ? "RW" : "READ"); | |
482 | } | |
483 | ||
484 | static bool | |
485 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
486 | { | |
25c75333 | 487 | unsigned long start, end; |
3e621214 TM |
488 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
489 | ||
490 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
491 | return false; |
492 | end = jiffies; | |
493 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
494 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
495 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 496 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
497 | return false; |
498 | } | |
499 | return true; | |
b9e028fd TM |
500 | } |
501 | ||
974cec8c | 502 | static void |
119cef97 TM |
503 | pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg, |
504 | const struct pnfs_layout_range *range, | |
505 | const nfs4_stateid *stateid) | |
974cec8c | 506 | { |
566052c5 | 507 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 508 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
a9901899 | 509 | INIT_LIST_HEAD(&lseg->pls_commits); |
eba6dd69 | 510 | refcount_set(&lseg->pls_refcount, 1); |
4541d16c | 511 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); |
566052c5 | 512 | lseg->pls_layout = lo; |
119cef97 TM |
513 | lseg->pls_range = *range; |
514 | lseg->pls_seq = be32_to_cpu(stateid->seqid); | |
974cec8c AA |
515 | } |
516 | ||
905ca191 | 517 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 518 | { |
68f74479 TM |
519 | if (lseg != NULL) { |
520 | struct inode *inode = lseg->pls_layout->plh_inode; | |
521 | NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg); | |
522 | } | |
974cec8c AA |
523 | } |
524 | ||
d684d2ae | 525 | static void |
57036a37 TM |
526 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
527 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 528 | { |
d20581aa | 529 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 530 | list_del_init(&lseg->pls_list); |
8f0d27dc | 531 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
2b28a7be | 532 | refcount_dec(&lo->plh_refcount); |
abb3e1c8 TM |
533 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) |
534 | return; | |
7b650994 TM |
535 | if (list_empty(&lo->plh_segs) && |
536 | !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) && | |
537 | !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { | |
334a8f37 TM |
538 | if (atomic_read(&lo->plh_outstanding) == 0) |
539 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
173f77e9 | 540 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
2d148c7e | 541 | } |
d684d2ae FI |
542 | } |
543 | ||
68f74479 TM |
544 | static bool |
545 | pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo, | |
546 | struct pnfs_layout_segment *lseg) | |
547 | { | |
548 | if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && | |
549 | pnfs_layout_is_valid(lo)) { | |
4aab9732 | 550 | pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0); |
68f74479 TM |
551 | list_move_tail(&lseg->pls_list, &lo->plh_return_segs); |
552 | return true; | |
553 | } | |
554 | return false; | |
555 | } | |
556 | ||
bae724ef | 557 | void |
9369a431 | 558 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 559 | { |
57036a37 | 560 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
561 | struct inode *inode; |
562 | ||
563 | if (!lseg) | |
564 | return; | |
565 | ||
4541d16c | 566 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
eba6dd69 | 567 | refcount_read(&lseg->pls_refcount), |
4541d16c | 568 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
4ef2e4f8 | 569 | |
57036a37 TM |
570 | lo = lseg->pls_layout; |
571 | inode = lo->plh_inode; | |
4ef2e4f8 | 572 | |
eba6dd69 | 573 | if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
faa4a54f TM |
574 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { |
575 | spin_unlock(&inode->i_lock); | |
576 | return; | |
577 | } | |
8f0d27dc | 578 | pnfs_get_layout_hdr(lo); |
4ef2e4f8 | 579 | pnfs_layout_remove_lseg(lo, lseg); |
68f74479 TM |
580 | if (pnfs_cache_lseg_for_layoutreturn(lo, lseg)) |
581 | lseg = NULL; | |
4ef2e4f8 TM |
582 | spin_unlock(&inode->i_lock); |
583 | pnfs_free_lseg(lseg); | |
584 | pnfs_put_layout_hdr(lo); | |
4541d16c | 585 | } |
4541d16c | 586 | } |
9369a431 | 587 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 588 | |
fb3296eb BH |
589 | /* |
590 | * is l2 fully contained in l1? | |
591 | * start1 end1 | |
592 | * [----------------------------------) | |
593 | * start2 end2 | |
594 | * [----------------) | |
595 | */ | |
3cb2df17 | 596 | static bool |
7dc0ac70 | 597 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 598 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
599 | { |
600 | u64 start1 = l1->offset; | |
17822b20 | 601 | u64 end1 = pnfs_end_offset(start1, l1->length); |
fb3296eb | 602 | u64 start2 = l2->offset; |
17822b20 | 603 | u64 end2 = pnfs_end_offset(start2, l2->length); |
fb3296eb BH |
604 | |
605 | return (start1 <= start2) && (end1 >= end2); | |
606 | } | |
607 | ||
24956804 TM |
608 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
609 | struct list_head *tmp_list) | |
610 | { | |
eba6dd69 | 611 | if (!refcount_dec_and_test(&lseg->pls_refcount)) |
24956804 TM |
612 | return false; |
613 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
614 | list_add(&lseg->pls_list, tmp_list); | |
615 | return true; | |
616 | } | |
617 | ||
4541d16c FI |
618 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
619 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
620 | struct list_head *tmp_list) | |
621 | { | |
622 | int rv = 0; | |
623 | ||
624 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
625 | /* Remove the reference keeping the lseg in the | |
626 | * list. It will now be removed when all | |
627 | * outstanding io is finished. | |
628 | */ | |
d684d2ae | 629 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
eba6dd69 | 630 | refcount_read(&lseg->pls_refcount)); |
24956804 | 631 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 632 | rv = 1; |
4541d16c FI |
633 | } |
634 | return rv; | |
635 | } | |
636 | ||
6d597e17 JL |
637 | /* |
638 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
639 | * taking into account wraparound issues. | |
640 | */ | |
641 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
642 | { | |
643 | return (s32)(s1 - s2) > 0; | |
644 | } | |
645 | ||
e036f464 TM |
646 | static bool |
647 | pnfs_should_free_range(const struct pnfs_layout_range *lseg_range, | |
648 | const struct pnfs_layout_range *recall_range) | |
649 | { | |
650 | return (recall_range->iomode == IOMODE_ANY || | |
651 | lseg_range->iomode == recall_range->iomode) && | |
652 | pnfs_lseg_range_intersecting(lseg_range, recall_range); | |
653 | } | |
654 | ||
655 | static bool | |
656 | pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg, | |
657 | const struct pnfs_layout_range *recall_range, | |
658 | u32 seq) | |
659 | { | |
660 | if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq)) | |
661 | return false; | |
662 | if (recall_range == NULL) | |
663 | return true; | |
664 | return pnfs_should_free_range(&lseg->pls_range, recall_range); | |
665 | } | |
666 | ||
6d597e17 JL |
667 | /** |
668 | * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later | |
669 | * @lo: layout header containing the lsegs | |
670 | * @tmp_list: list head where doomed lsegs should go | |
671 | * @recall_range: optional recall range argument to match (may be NULL) | |
672 | * @seq: only invalidate lsegs obtained prior to this sequence (may be 0) | |
673 | * | |
674 | * Walk the list of lsegs in the layout header, and tear down any that should | |
675 | * be destroyed. If "recall_range" is specified then the segment must match | |
676 | * that range. If "seq" is non-zero, then only match segments that were handed | |
677 | * out at or before that sequence. | |
678 | * | |
679 | * Returns number of matching invalid lsegs remaining in list after scanning | |
680 | * it and purging them. | |
4541d16c | 681 | */ |
43f1b3da | 682 | int |
49a85061 | 683 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 684 | struct list_head *tmp_list, |
6d597e17 JL |
685 | const struct pnfs_layout_range *recall_range, |
686 | u32 seq) | |
974cec8c AA |
687 | { |
688 | struct pnfs_layout_segment *lseg, *next; | |
71b39854 | 689 | int remaining = 0; |
974cec8c AA |
690 | |
691 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
692 | ||
8006bfba | 693 | if (list_empty(&lo->plh_segs)) |
38511722 | 694 | return 0; |
4541d16c | 695 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
e036f464 | 696 | if (pnfs_match_lseg_recall(lseg, recall_range, seq)) { |
b3dce6a2 | 697 | dprintk("%s: freeing lseg %p iomode %d seq %u " |
4541d16c | 698 | "offset %llu length %llu\n", __func__, |
6d597e17 JL |
699 | lseg, lseg->pls_range.iomode, lseg->pls_seq, |
700 | lseg->pls_range.offset, lseg->pls_range.length); | |
71b39854 TM |
701 | if (!mark_lseg_invalid(lseg, tmp_list)) |
702 | remaining++; | |
4541d16c | 703 | } |
71b39854 TM |
704 | dprintk("%s:Return %i\n", __func__, remaining); |
705 | return remaining; | |
974cec8c AA |
706 | } |
707 | ||
68f74479 TM |
708 | static void |
709 | pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo, | |
710 | struct list_head *free_me, | |
711 | const struct pnfs_layout_range *range, | |
712 | u32 seq) | |
713 | { | |
714 | struct pnfs_layout_segment *lseg, *next; | |
715 | ||
716 | list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) { | |
717 | if (pnfs_match_lseg_recall(lseg, range, seq)) | |
718 | list_move_tail(&lseg->pls_list, free_me); | |
719 | } | |
720 | } | |
721 | ||
f49f9baa | 722 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 723 | void |
4541d16c | 724 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 725 | { |
4541d16c | 726 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
727 | |
728 | if (list_empty(free_me)) | |
729 | return; | |
730 | ||
4541d16c | 731 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 732 | list_del(&lseg->pls_list); |
905ca191 | 733 | pnfs_free_lseg(lseg); |
974cec8c AA |
734 | } |
735 | } | |
736 | ||
e5e94017 BH |
737 | void |
738 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
739 | { | |
740 | struct pnfs_layout_hdr *lo; | |
974cec8c | 741 | LIST_HEAD(tmp_list); |
e5e94017 BH |
742 | |
743 | spin_lock(&nfsi->vfs_inode.i_lock); | |
744 | lo = nfsi->layout; | |
745 | if (lo) { | |
3e621214 | 746 | pnfs_get_layout_hdr(lo); |
2454dfea | 747 | pnfs_mark_layout_stateid_invalid(lo, &tmp_list); |
3e621214 TM |
748 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); |
749 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
750 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
751 | pnfs_free_lseg_list(&tmp_list); | |
1f18b82c | 752 | nfs_commit_inode(&nfsi->vfs_inode, 0); |
3e621214 TM |
753 | pnfs_put_layout_hdr(lo); |
754 | } else | |
755 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c | 756 | } |
041245c8 | 757 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 758 | |
fd9a8d71 TM |
759 | static bool |
760 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
761 | struct list_head *layout_list) | |
974cec8c AA |
762 | { |
763 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 764 | bool ret = false; |
974cec8c | 765 | |
fd9a8d71 TM |
766 | spin_lock(&inode->i_lock); |
767 | lo = NFS_I(inode)->layout; | |
768 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
769 | pnfs_get_layout_hdr(lo); | |
770 | list_add(&lo->plh_bulk_destroy, layout_list); | |
771 | ret = true; | |
772 | } | |
773 | spin_unlock(&inode->i_lock); | |
774 | return ret; | |
775 | } | |
776 | ||
777 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
778 | static int | |
779 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
780 | struct nfs_server *server, | |
781 | struct list_head *layout_list) | |
5085607d TM |
782 | __must_hold(&clp->cl_lock) |
783 | __must_hold(RCU) | |
fd9a8d71 TM |
784 | { |
785 | struct pnfs_layout_hdr *lo, *next; | |
786 | struct inode *inode; | |
787 | ||
788 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
5085607d TM |
789 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) || |
790 | test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) || | |
791 | !list_empty(&lo->plh_bulk_destroy)) | |
b85f5620 | 792 | continue; |
5085607d TM |
793 | /* If the sb is being destroyed, just bail */ |
794 | if (!nfs_sb_active(server->super)) | |
795 | break; | |
b5fdf841 | 796 | inode = pnfs_grab_inode_layout_hdr(lo); |
5085607d | 797 | if (inode != NULL) { |
cf6605d1 TM |
798 | if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) |
799 | list_del_rcu(&lo->plh_layouts); | |
5085607d TM |
800 | if (pnfs_layout_add_bulk_destroy_list(inode, |
801 | layout_list)) | |
802 | continue; | |
803 | rcu_read_unlock(); | |
804 | spin_unlock(&clp->cl_lock); | |
805 | iput(inode); | |
806 | } else { | |
807 | rcu_read_unlock(); | |
808 | spin_unlock(&clp->cl_lock); | |
5085607d TM |
809 | } |
810 | nfs_sb_deactive(server->super); | |
fd9a8d71 TM |
811 | spin_lock(&clp->cl_lock); |
812 | rcu_read_lock(); | |
813 | return -EAGAIN; | |
814 | } | |
815 | return 0; | |
816 | } | |
817 | ||
818 | static int | |
819 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
820 | bool is_bulk_recall) | |
821 | { | |
822 | struct pnfs_layout_hdr *lo; | |
823 | struct inode *inode; | |
fd9a8d71 TM |
824 | LIST_HEAD(lseg_list); |
825 | int ret = 0; | |
826 | ||
827 | while (!list_empty(layout_list)) { | |
828 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
829 | plh_bulk_destroy); | |
830 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
831 | lo->plh_inode->i_ino); | |
832 | inode = lo->plh_inode; | |
7c5d1875 CH |
833 | |
834 | pnfs_layoutcommit_inode(inode, false); | |
835 | ||
fd9a8d71 TM |
836 | spin_lock(&inode->i_lock); |
837 | list_del_init(&lo->plh_bulk_destroy); | |
9fd4b9fc TM |
838 | if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) { |
839 | if (is_bulk_recall) | |
840 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
fd9a8d71 | 841 | ret = -EAGAIN; |
9fd4b9fc | 842 | } |
fd9a8d71 TM |
843 | spin_unlock(&inode->i_lock); |
844 | pnfs_free_lseg_list(&lseg_list); | |
b20135d0 TM |
845 | /* Free all lsegs that are attached to commit buckets */ |
846 | nfs_commit_inode(inode, 0); | |
fd9a8d71 | 847 | pnfs_put_layout_hdr(lo); |
5085607d | 848 | nfs_iput_and_deactive(inode); |
fd9a8d71 TM |
849 | } |
850 | return ret; | |
851 | } | |
852 | ||
853 | int | |
854 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
855 | struct nfs_fsid *fsid, | |
856 | bool is_recall) | |
857 | { | |
858 | struct nfs_server *server; | |
859 | LIST_HEAD(layout_list); | |
c47abcf8 | 860 | |
974cec8c | 861 | spin_lock(&clp->cl_lock); |
6382a441 | 862 | rcu_read_lock(); |
fd9a8d71 | 863 | restart: |
6382a441 | 864 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
865 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
866 | continue; | |
867 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
868 | server, | |
869 | &layout_list) != 0) | |
870 | goto restart; | |
6382a441 WAA |
871 | } |
872 | rcu_read_unlock(); | |
974cec8c AA |
873 | spin_unlock(&clp->cl_lock); |
874 | ||
fd9a8d71 TM |
875 | if (list_empty(&layout_list)) |
876 | return 0; | |
877 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
878 | } | |
879 | ||
880 | int | |
881 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
882 | bool is_recall) | |
883 | { | |
884 | struct nfs_server *server; | |
885 | LIST_HEAD(layout_list); | |
886 | ||
887 | spin_lock(&clp->cl_lock); | |
888 | rcu_read_lock(); | |
889 | restart: | |
890 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
891 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
892 | server, | |
893 | &layout_list) != 0) | |
894 | goto restart; | |
974cec8c | 895 | } |
fd9a8d71 TM |
896 | rcu_read_unlock(); |
897 | spin_unlock(&clp->cl_lock); | |
898 | ||
899 | if (list_empty(&layout_list)) | |
900 | return 0; | |
901 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
902 | } | |
903 | ||
904 | /* | |
905 | * Called by the state manger to remove all layouts established under an | |
906 | * expired lease. | |
907 | */ | |
908 | void | |
909 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
910 | { | |
911 | nfs4_deviceid_mark_client_invalid(clp); | |
912 | nfs4_deviceid_purge_client(clp); | |
913 | ||
914 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
915 | } |
916 | ||
59b56394 TM |
917 | static void |
918 | pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred) | |
919 | { | |
920 | const struct cred *old; | |
921 | ||
922 | if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) { | |
923 | old = xchg(&lo->plh_lc_cred, get_cred(cred)); | |
924 | put_cred(old); | |
925 | } | |
926 | } | |
927 | ||
fd6002e9 | 928 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
929 | void |
930 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
59b56394 | 931 | const struct cred *cred, bool update_barrier) |
b1f69b75 | 932 | { |
ecebb80b | 933 | u32 oldseq, newseq, new_barrier = 0; |
b1f69b75 | 934 | |
2d2f24ad TM |
935 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
936 | newseq = be32_to_cpu(new->seqid); | |
2a59a041 TM |
937 | |
938 | if (!pnfs_layout_is_valid(lo)) { | |
59b56394 | 939 | pnfs_set_layout_cred(lo, cred); |
2a59a041 TM |
940 | nfs4_stateid_copy(&lo->plh_stateid, new); |
941 | lo->plh_barrier = newseq; | |
942 | pnfs_clear_layoutreturn_info(lo); | |
943 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
944 | return; | |
945 | } | |
946 | if (pnfs_seqid_is_newer(newseq, oldseq)) { | |
f597c537 | 947 | nfs4_stateid_copy(&lo->plh_stateid, new); |
ecebb80b TM |
948 | /* |
949 | * Because of wraparound, we want to keep the barrier | |
950 | * "close" to the current seqids. | |
951 | */ | |
952 | new_barrier = newseq - atomic_read(&lo->plh_outstanding); | |
43f1b3da | 953 | } |
ecebb80b TM |
954 | if (update_barrier) |
955 | new_barrier = be32_to_cpu(new->seqid); | |
956 | else if (new_barrier == 0) | |
957 | return; | |
2a59a041 | 958 | if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) |
ecebb80b | 959 | lo->plh_barrier = new_barrier; |
b1f69b75 AA |
960 | } |
961 | ||
cf7d63f1 | 962 | static bool |
19c54aba TM |
963 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
964 | const nfs4_stateid *stateid) | |
43f1b3da | 965 | { |
19c54aba | 966 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 967 | |
19c54aba TM |
968 | return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); |
969 | } | |
970 | ||
971 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
972 | static bool | |
e1c06f80 | 973 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo) |
19c54aba | 974 | { |
f7e8917a | 975 | return lo->plh_block_lgets || |
e1c06f80 | 976 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
cf7d63f1 FI |
977 | } |
978 | ||
5e36e2a9 FI |
979 | static struct nfs_server * |
980 | pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx) | |
981 | { | |
982 | struct nfs_server *server; | |
983 | ||
78746a38 | 984 | if (inode) { |
5e36e2a9 | 985 | server = NFS_SERVER(inode); |
78746a38 | 986 | } else { |
5e36e2a9 FI |
987 | struct dentry *parent_dir = dget_parent(ctx->dentry); |
988 | server = NFS_SERVER(parent_dir->d_inode); | |
989 | dput(parent_dir); | |
990 | } | |
991 | return server; | |
992 | } | |
993 | ||
29a8bfe5 TM |
994 | static void nfs4_free_pages(struct page **pages, size_t size) |
995 | { | |
996 | int i; | |
997 | ||
998 | if (!pages) | |
999 | return; | |
1000 | ||
1001 | for (i = 0; i < size; i++) { | |
1002 | if (!pages[i]) | |
1003 | break; | |
1004 | __free_page(pages[i]); | |
1005 | } | |
1006 | kfree(pages); | |
1007 | } | |
1008 | ||
1009 | static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags) | |
1010 | { | |
1011 | struct page **pages; | |
1012 | int i; | |
1013 | ||
a2791d3a | 1014 | pages = kmalloc_array(size, sizeof(struct page *), gfp_flags); |
29a8bfe5 TM |
1015 | if (!pages) { |
1016 | dprintk("%s: can't alloc array of %zu pages\n", __func__, size); | |
1017 | return NULL; | |
1018 | } | |
1019 | ||
1020 | for (i = 0; i < size; i++) { | |
1021 | pages[i] = alloc_page(gfp_flags); | |
1022 | if (!pages[i]) { | |
1023 | dprintk("%s: failed to allocate page\n", __func__); | |
a2791d3a | 1024 | nfs4_free_pages(pages, i); |
29a8bfe5 TM |
1025 | return NULL; |
1026 | } | |
1027 | } | |
1028 | ||
1029 | return pages; | |
1030 | } | |
1031 | ||
587f03de | 1032 | static struct nfs4_layoutget * |
5e36e2a9 | 1033 | pnfs_alloc_init_layoutget_args(struct inode *ino, |
e5e94017 | 1034 | struct nfs_open_context *ctx, |
2409a976 | 1035 | const nfs4_stateid *stateid, |
e144e539 | 1036 | const struct pnfs_layout_range *range, |
587f03de | 1037 | gfp_t gfp_flags) |
e5e94017 | 1038 | { |
5e36e2a9 | 1039 | struct nfs_server *server = pnfs_find_server(ino, ctx); |
28ced9a8 | 1040 | size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response; |
dacb452d | 1041 | size_t max_pages = max_response_pages(server); |
b1f69b75 | 1042 | struct nfs4_layoutget *lgp; |
b1f69b75 AA |
1043 | |
1044 | dprintk("--> %s\n", __func__); | |
e5e94017 | 1045 | |
83026d80 JL |
1046 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
1047 | if (lgp == NULL) | |
587f03de | 1048 | return NULL; |
83026d80 | 1049 | |
28ced9a8 TM |
1050 | if (max_reply_sz) { |
1051 | size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1052 | if (npages < max_pages) | |
1053 | max_pages = npages; | |
1054 | } | |
1055 | ||
dacb452d FI |
1056 | lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags); |
1057 | if (!lgp->args.layout.pages) { | |
1058 | kfree(lgp); | |
1059 | return NULL; | |
1060 | } | |
1061 | lgp->args.layout.pglen = max_pages * PAGE_SIZE; | |
1062 | lgp->res.layoutp = &lgp->args.layout; | |
1063 | ||
d49e0d5b TM |
1064 | /* Don't confuse uninitialised result and success */ |
1065 | lgp->res.status = -NFS4ERR_DELAY; | |
83026d80 JL |
1066 | |
1067 | lgp->args.minlength = PAGE_SIZE; | |
1068 | if (lgp->args.minlength > range->length) | |
1069 | lgp->args.minlength = range->length; | |
5e36e2a9 FI |
1070 | if (ino) { |
1071 | loff_t i_size = i_size_read(ino); | |
1072 | ||
1073 | if (range->iomode == IOMODE_READ) { | |
1074 | if (range->offset >= i_size) | |
1075 | lgp->args.minlength = 0; | |
1076 | else if (i_size - range->offset < lgp->args.minlength) | |
1077 | lgp->args.minlength = i_size - range->offset; | |
1078 | } | |
d03ab29d | 1079 | } |
83026d80 JL |
1080 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
1081 | pnfs_copy_range(&lgp->args.range, range); | |
1082 | lgp->args.type = server->pnfs_curr_ld->id; | |
1083 | lgp->args.inode = ino; | |
1084 | lgp->args.ctx = get_nfs_open_context(ctx); | |
183d9e7b | 1085 | nfs4_stateid_copy(&lgp->args.stateid, stateid); |
83026d80 | 1086 | lgp->gfp_flags = gfp_flags; |
63ec2b69 | 1087 | lgp->cred = ctx->cred; |
587f03de | 1088 | return lgp; |
974cec8c AA |
1089 | } |
1090 | ||
29a8bfe5 TM |
1091 | void pnfs_layoutget_free(struct nfs4_layoutget *lgp) |
1092 | { | |
1093 | size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE; | |
1094 | ||
1095 | nfs4_free_pages(lgp->args.layout.pages, max_pages); | |
1096 | if (lgp->args.inode) | |
1097 | pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout); | |
29a8bfe5 TM |
1098 | put_nfs_open_context(lgp->args.ctx); |
1099 | kfree(lgp); | |
1100 | } | |
1101 | ||
24956804 TM |
1102 | static void pnfs_clear_layoutcommit(struct inode *inode, |
1103 | struct list_head *head) | |
1104 | { | |
1105 | struct nfs_inode *nfsi = NFS_I(inode); | |
1106 | struct pnfs_layout_segment *lseg, *tmp; | |
1107 | ||
1108 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
1109 | return; | |
1110 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
1111 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
1112 | continue; | |
1113 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
1114 | } | |
1115 | } | |
1116 | ||
68f74479 | 1117 | void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo, |
2a974425 | 1118 | const nfs4_stateid *arg_stateid, |
68f74479 | 1119 | const struct pnfs_layout_range *range, |
68f74479 TM |
1120 | const nfs4_stateid *stateid) |
1121 | { | |
1122 | struct inode *inode = lo->plh_inode; | |
1123 | LIST_HEAD(freeme); | |
1124 | ||
1125 | spin_lock(&inode->i_lock); | |
2a974425 TM |
1126 | if (!pnfs_layout_is_valid(lo) || !arg_stateid || |
1127 | !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid)) | |
1128 | goto out_unlock; | |
68f74479 | 1129 | if (stateid) { |
2a974425 TM |
1130 | u32 seq = be32_to_cpu(arg_stateid->seqid); |
1131 | ||
68f74479 TM |
1132 | pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq); |
1133 | pnfs_free_returned_lsegs(lo, &freeme, range, seq); | |
59b56394 | 1134 | pnfs_set_layout_stateid(lo, stateid, NULL, true); |
68f74479 TM |
1135 | } else |
1136 | pnfs_mark_layout_stateid_invalid(lo, &freeme); | |
2a974425 | 1137 | out_unlock: |
68f74479 TM |
1138 | pnfs_clear_layoutreturn_waitbit(lo); |
1139 | spin_unlock(&inode->i_lock); | |
1140 | pnfs_free_lseg_list(&freeme); | |
1141 | ||
1142 | } | |
1143 | ||
13c13a6a | 1144 | static bool |
e5fd1904 TM |
1145 | pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo, |
1146 | nfs4_stateid *stateid, | |
44ea8dfc | 1147 | const struct cred **cred, |
e5fd1904 | 1148 | enum pnfs_iomode *iomode) |
13c13a6a | 1149 | { |
bf0291dd TM |
1150 | /* Serialise LAYOUTGET/LAYOUTRETURN */ |
1151 | if (atomic_read(&lo->plh_outstanding) != 0) | |
1152 | return false; | |
6604b203 | 1153 | if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) |
13c13a6a | 1154 | return false; |
6604b203 | 1155 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); |
13c13a6a | 1156 | pnfs_get_layout_hdr(lo); |
e5fd1904 | 1157 | if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) { |
44ea8dfc TM |
1158 | nfs4_stateid_copy(stateid, &lo->plh_stateid); |
1159 | *cred = get_cred(lo->plh_lc_cred); | |
1160 | if (lo->plh_return_seq != 0) | |
1161 | stateid->seqid = cpu_to_be32(lo->plh_return_seq); | |
e5fd1904 TM |
1162 | if (iomode != NULL) |
1163 | *iomode = lo->plh_return_iomode; | |
1164 | pnfs_clear_layoutreturn_info(lo); | |
1165 | return true; | |
1166 | } | |
44ea8dfc TM |
1167 | nfs4_stateid_copy(stateid, &lo->plh_stateid); |
1168 | *cred = get_cred(lo->plh_lc_cred); | |
e5fd1904 TM |
1169 | if (iomode != NULL) |
1170 | *iomode = IOMODE_ANY; | |
13c13a6a TM |
1171 | return true; |
1172 | } | |
1173 | ||
828ed9ec TM |
1174 | static void |
1175 | pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args, | |
1176 | struct pnfs_layout_hdr *lo, | |
1177 | const nfs4_stateid *stateid, | |
1178 | enum pnfs_iomode iomode) | |
1179 | { | |
1180 | struct inode *inode = lo->plh_inode; | |
1181 | ||
1182 | args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id; | |
1183 | args->inode = inode; | |
1184 | args->range.iomode = iomode; | |
1185 | args->range.offset = 0; | |
1186 | args->range.length = NFS4_MAX_UINT64; | |
1187 | args->layout = lo; | |
1188 | nfs4_stateid_copy(&args->stateid, stateid); | |
1189 | } | |
1190 | ||
f40eb5d0 | 1191 | static int |
44ea8dfc TM |
1192 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, |
1193 | const nfs4_stateid *stateid, | |
1194 | const struct cred **pcred, | |
1195 | enum pnfs_iomode iomode, | |
1196 | bool sync) | |
f40eb5d0 PT |
1197 | { |
1198 | struct inode *ino = lo->plh_inode; | |
287bd3e9 | 1199 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; |
f40eb5d0 | 1200 | struct nfs4_layoutreturn *lrp; |
44ea8dfc | 1201 | const struct cred *cred = *pcred; |
f40eb5d0 PT |
1202 | int status = 0; |
1203 | ||
44ea8dfc | 1204 | *pcred = NULL; |
e4af440a | 1205 | lrp = kzalloc(sizeof(*lrp), GFP_NOFS); |
f40eb5d0 PT |
1206 | if (unlikely(lrp == NULL)) { |
1207 | status = -ENOMEM; | |
1208 | spin_lock(&ino->i_lock); | |
d67ae825 | 1209 | pnfs_clear_layoutreturn_waitbit(lo); |
f40eb5d0 | 1210 | spin_unlock(&ino->i_lock); |
44ea8dfc | 1211 | put_cred(cred); |
f40eb5d0 PT |
1212 | pnfs_put_layout_hdr(lo); |
1213 | goto out; | |
1214 | } | |
1215 | ||
828ed9ec | 1216 | pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode); |
4d796d75 | 1217 | lrp->args.ld_private = &lrp->ld_private; |
f40eb5d0 | 1218 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
44ea8dfc | 1219 | lrp->cred = cred; |
287bd3e9 TM |
1220 | if (ld->prepare_layoutreturn) |
1221 | ld->prepare_layoutreturn(&lrp->args); | |
f40eb5d0 | 1222 | |
6c16605d | 1223 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
1224 | out: |
1225 | dprintk("<-- %s status: %d\n", __func__, status); | |
1226 | return status; | |
1227 | } | |
1228 | ||
13c13a6a TM |
1229 | /* Return true if layoutreturn is needed */ |
1230 | static bool | |
1231 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo) | |
1232 | { | |
1233 | struct pnfs_layout_segment *s; | |
f0b42981 TM |
1234 | enum pnfs_iomode iomode; |
1235 | u32 seq; | |
13c13a6a | 1236 | |
2370abda | 1237 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
1238 | return false; |
1239 | ||
f0b42981 TM |
1240 | seq = lo->plh_return_seq; |
1241 | iomode = lo->plh_return_iomode; | |
1242 | ||
1243 | /* Defer layoutreturn until all recalled lsegs are done */ | |
13c13a6a | 1244 | list_for_each_entry(s, &lo->plh_segs, pls_list) { |
f0b42981 TM |
1245 | if (seq && pnfs_seqid_is_newer(s->pls_seq, seq)) |
1246 | continue; | |
1247 | if (iomode != IOMODE_ANY && s->pls_range.iomode != iomode) | |
1248 | continue; | |
13c13a6a TM |
1249 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags)) |
1250 | return false; | |
1251 | } | |
1252 | ||
1253 | return true; | |
1254 | } | |
1255 | ||
1256 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo) | |
1257 | { | |
1258 | struct inode *inode= lo->plh_inode; | |
1259 | ||
2370abda | 1260 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
1261 | return; |
1262 | spin_lock(&inode->i_lock); | |
1263 | if (pnfs_layout_need_return(lo)) { | |
44ea8dfc | 1264 | const struct cred *cred; |
13c13a6a TM |
1265 | nfs4_stateid stateid; |
1266 | enum pnfs_iomode iomode; | |
1267 | bool send; | |
1268 | ||
44ea8dfc | 1269 | send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode); |
13c13a6a TM |
1270 | spin_unlock(&inode->i_lock); |
1271 | if (send) { | |
1272 | /* Send an async layoutreturn so we dont deadlock */ | |
44ea8dfc | 1273 | pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false); |
13c13a6a TM |
1274 | } |
1275 | } else | |
1276 | spin_unlock(&inode->i_lock); | |
1277 | } | |
1278 | ||
293b3b06 AA |
1279 | /* |
1280 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
1281 | * when the layout segment list is empty. | |
1282 | * | |
1283 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
1284 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
1285 | * deviceid is marked invalid. | |
1286 | */ | |
cbe82603 BH |
1287 | int |
1288 | _pnfs_return_layout(struct inode *ino) | |
1289 | { | |
1290 | struct pnfs_layout_hdr *lo = NULL; | |
1291 | struct nfs_inode *nfsi = NFS_I(ino); | |
1292 | LIST_HEAD(tmp_list); | |
44ea8dfc | 1293 | const struct cred *cred; |
cbe82603 | 1294 | nfs4_stateid stateid; |
24408f52 | 1295 | int status = 0; |
93b7f7ad | 1296 | bool send, valid_layout; |
cbe82603 | 1297 | |
366d5052 | 1298 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
1299 | |
1300 | spin_lock(&ino->i_lock); | |
1301 | lo = nfsi->layout; | |
e5929f3c | 1302 | if (!lo) { |
cbe82603 | 1303 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
1304 | dprintk("NFS: %s no layout to return\n", __func__); |
1305 | goto out; | |
cbe82603 | 1306 | } |
cbe82603 | 1307 | /* Reference matched in nfs4_layoutreturn_release */ |
70c3bd2b | 1308 | pnfs_get_layout_hdr(lo); |
24408f52 TM |
1309 | /* Is there an outstanding layoutreturn ? */ |
1310 | if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) { | |
1311 | spin_unlock(&ino->i_lock); | |
1312 | if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
1313 | TASK_UNINTERRUPTIBLE)) | |
1314 | goto out_put_layout_hdr; | |
1315 | spin_lock(&ino->i_lock); | |
1316 | } | |
93b7f7ad | 1317 | valid_layout = pnfs_layout_is_valid(lo); |
24956804 | 1318 | pnfs_clear_layoutcommit(ino, &tmp_list); |
6d597e17 | 1319 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0); |
c88953d8 CH |
1320 | |
1321 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) { | |
1322 | struct pnfs_layout_range range = { | |
1323 | .iomode = IOMODE_ANY, | |
1324 | .offset = 0, | |
1325 | .length = NFS4_MAX_UINT64, | |
1326 | }; | |
1327 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); | |
1328 | } | |
1329 | ||
293b3b06 | 1330 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
93b7f7ad OK |
1331 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) || |
1332 | !valid_layout) { | |
293b3b06 | 1333 | spin_unlock(&ino->i_lock); |
293b3b06 | 1334 | dprintk("NFS: %s no layout segments to return\n", __func__); |
7bcc1058 | 1335 | goto out_wait_layoutreturn; |
293b3b06 | 1336 | } |
47abadef | 1337 | |
44ea8dfc | 1338 | send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL); |
cbe82603 | 1339 | spin_unlock(&ino->i_lock); |
7f27392c | 1340 | if (send) |
44ea8dfc | 1341 | status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true); |
7bcc1058 TM |
1342 | out_wait_layoutreturn: |
1343 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE); | |
7f27392c | 1344 | out_put_layout_hdr: |
ee6625a9 | 1345 | pnfs_free_lseg_list(&tmp_list); |
7f27392c | 1346 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
1347 | out: |
1348 | dprintk("<-- %s status: %d\n", __func__, status); | |
1349 | return status; | |
1350 | } | |
1351 | ||
24028672 TM |
1352 | int |
1353 | pnfs_commit_and_return_layout(struct inode *inode) | |
1354 | { | |
1355 | struct pnfs_layout_hdr *lo; | |
1356 | int ret; | |
1357 | ||
1358 | spin_lock(&inode->i_lock); | |
1359 | lo = NFS_I(inode)->layout; | |
1360 | if (lo == NULL) { | |
1361 | spin_unlock(&inode->i_lock); | |
1362 | return 0; | |
1363 | } | |
1364 | pnfs_get_layout_hdr(lo); | |
1365 | /* Block new layoutgets and read/write to ds */ | |
1366 | lo->plh_block_lgets++; | |
1367 | spin_unlock(&inode->i_lock); | |
1368 | filemap_fdatawait(inode->i_mapping); | |
1369 | ret = pnfs_layoutcommit_inode(inode, true); | |
1370 | if (ret == 0) | |
1371 | ret = _pnfs_return_layout(inode); | |
1372 | spin_lock(&inode->i_lock); | |
1373 | lo->plh_block_lgets--; | |
1374 | spin_unlock(&inode->i_lock); | |
1375 | pnfs_put_layout_hdr(lo); | |
1376 | return ret; | |
1377 | } | |
1378 | ||
1c5bd76d TM |
1379 | bool pnfs_roc(struct inode *ino, |
1380 | struct nfs4_layoutreturn_args *args, | |
1381 | struct nfs4_layoutreturn_res *res, | |
a52458b4 | 1382 | const struct cred *cred) |
f7e8917a | 1383 | { |
40dd4b7a TM |
1384 | struct nfs_inode *nfsi = NFS_I(ino); |
1385 | struct nfs_open_context *ctx; | |
1386 | struct nfs4_state *state; | |
f7e8917a | 1387 | struct pnfs_layout_hdr *lo; |
1c5bd76d | 1388 | struct pnfs_layout_segment *lseg, *next; |
44ea8dfc | 1389 | const struct cred *lc_cred; |
193e3aa2 | 1390 | nfs4_stateid stateid; |
1c5bd76d TM |
1391 | enum pnfs_iomode iomode = 0; |
1392 | bool layoutreturn = false, roc = false; | |
e71708d4 | 1393 | bool skip_read = false; |
f7e8917a | 1394 | |
1c5bd76d TM |
1395 | if (!nfs_have_layout(ino)) |
1396 | return false; | |
29ade5db | 1397 | retry: |
0de43976 | 1398 | rcu_read_lock(); |
f7e8917a | 1399 | spin_lock(&ino->i_lock); |
40dd4b7a | 1400 | lo = nfsi->layout; |
0cdc329e | 1401 | if (!lo || !pnfs_layout_is_valid(lo) || |
9c6376eb TM |
1402 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
1403 | lo = NULL; | |
40dd4b7a | 1404 | goto out_noroc; |
9c6376eb TM |
1405 | } |
1406 | pnfs_get_layout_hdr(lo); | |
29ade5db | 1407 | if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) { |
29ade5db | 1408 | spin_unlock(&ino->i_lock); |
0de43976 | 1409 | rcu_read_unlock(); |
29ade5db TM |
1410 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, |
1411 | TASK_UNINTERRUPTIBLE); | |
1412 | pnfs_put_layout_hdr(lo); | |
1413 | goto retry; | |
1414 | } | |
40dd4b7a | 1415 | |
e755d638 | 1416 | /* no roc if we hold a delegation */ |
e71708d4 TM |
1417 | if (nfs4_check_delegation(ino, FMODE_READ)) { |
1418 | if (nfs4_check_delegation(ino, FMODE_WRITE)) | |
1419 | goto out_noroc; | |
1420 | skip_read = true; | |
1421 | } | |
40dd4b7a | 1422 | |
0de43976 | 1423 | list_for_each_entry_rcu(ctx, &nfsi->open_files, list) { |
40dd4b7a | 1424 | state = ctx->state; |
e71708d4 TM |
1425 | if (state == NULL) |
1426 | continue; | |
40dd4b7a | 1427 | /* Don't return layout if there is open file state */ |
e71708d4 | 1428 | if (state->state & FMODE_WRITE) |
40dd4b7a | 1429 | goto out_noroc; |
e71708d4 TM |
1430 | if (state->state & FMODE_READ) |
1431 | skip_read = true; | |
40dd4b7a TM |
1432 | } |
1433 | ||
e755d638 | 1434 | |
1c5bd76d | 1435 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) { |
e71708d4 TM |
1436 | if (skip_read && lseg->pls_range.iomode == IOMODE_READ) |
1437 | continue; | |
e755d638 | 1438 | /* If we are sending layoutreturn, invalidate all valid lsegs */ |
1c5bd76d TM |
1439 | if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags)) |
1440 | continue; | |
1441 | /* | |
1442 | * Note: mark lseg for return so pnfs_layout_remove_lseg | |
1443 | * doesn't invalidate the layout for us. | |
1444 | */ | |
1445 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
1446 | if (!mark_lseg_invalid(lseg, &lo->plh_return_segs)) | |
1447 | continue; | |
1448 | pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0); | |
69820d22 TM |
1449 | } |
1450 | ||
1c5bd76d TM |
1451 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
1452 | goto out_noroc; | |
69820d22 | 1453 | |
500d701f | 1454 | /* ROC in two conditions: |
e755d638 PT |
1455 | * 1. there are ROC lsegs |
1456 | * 2. we don't send layoutreturn | |
e755d638 | 1457 | */ |
1c5bd76d | 1458 | /* lo ref dropped in pnfs_roc_release() */ |
44ea8dfc | 1459 | layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode); |
1c5bd76d | 1460 | /* If the creds don't match, we can't compound the layoutreturn */ |
4d8948c7 | 1461 | if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0) |
1c5bd76d TM |
1462 | goto out_noroc; |
1463 | ||
1464 | roc = layoutreturn; | |
1465 | pnfs_init_layoutreturn_args(args, lo, &stateid, iomode); | |
1466 | res->lrs_present = 0; | |
1467 | layoutreturn = false; | |
44ea8dfc | 1468 | put_cred(lc_cred); |
4d8948c7 | 1469 | |
40dd4b7a | 1470 | out_noroc: |
f7e8917a | 1471 | spin_unlock(&ino->i_lock); |
0de43976 | 1472 | rcu_read_unlock(); |
e755d638 | 1473 | pnfs_layoutcommit_inode(ino, true); |
287bd3e9 TM |
1474 | if (roc) { |
1475 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
1476 | if (ld->prepare_layoutreturn) | |
1477 | ld->prepare_layoutreturn(args); | |
9c6376eb | 1478 | pnfs_put_layout_hdr(lo); |
287bd3e9 TM |
1479 | return true; |
1480 | } | |
e755d638 | 1481 | if (layoutreturn) |
44ea8dfc | 1482 | pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true); |
9c6376eb | 1483 | pnfs_put_layout_hdr(lo); |
287bd3e9 | 1484 | return false; |
f7e8917a FI |
1485 | } |
1486 | ||
287a9c55 TM |
1487 | int pnfs_roc_done(struct rpc_task *task, struct inode *inode, |
1488 | struct nfs4_layoutreturn_args **argpp, | |
1489 | struct nfs4_layoutreturn_res **respp, | |
1490 | int *ret) | |
1491 | { | |
1492 | struct nfs4_layoutreturn_args *arg = *argpp; | |
1493 | int retval = -EAGAIN; | |
1494 | ||
1495 | if (!arg) | |
1496 | return 0; | |
1497 | /* Handle Layoutreturn errors */ | |
1498 | switch (*ret) { | |
1499 | case 0: | |
1500 | retval = 0; | |
1501 | break; | |
6109bcf7 TM |
1502 | case -NFS4ERR_NOMATCHING_LAYOUT: |
1503 | /* Was there an RPC level error? If not, retry */ | |
1504 | if (task->tk_rpc_status == 0) | |
1505 | break; | |
1506 | /* If the call was not sent, let caller handle it */ | |
1507 | if (!RPC_WAS_SENT(task)) | |
1508 | return 0; | |
1509 | /* | |
1510 | * Otherwise, assume the call succeeded and | |
1511 | * that we need to release the layout | |
1512 | */ | |
1513 | *ret = 0; | |
1514 | (*respp)->lrs_present = 0; | |
1515 | retval = 0; | |
1516 | break; | |
078a432d TM |
1517 | case -NFS4ERR_DELAY: |
1518 | /* Let the caller handle the retry */ | |
1519 | *ret = -NFS4ERR_NOMATCHING_LAYOUT; | |
1520 | return 0; | |
287a9c55 | 1521 | case -NFS4ERR_OLD_STATEID: |
30cb3ee2 | 1522 | if (!nfs4_layout_refresh_old_stateid(&arg->stateid, |
287a9c55 TM |
1523 | &arg->range, inode)) |
1524 | break; | |
1525 | *ret = -NFS4ERR_NOMATCHING_LAYOUT; | |
1526 | return -EAGAIN; | |
1527 | } | |
1528 | *argpp = NULL; | |
1529 | *respp = NULL; | |
1530 | return retval; | |
1531 | } | |
1532 | ||
1c5bd76d TM |
1533 | void pnfs_roc_release(struct nfs4_layoutreturn_args *args, |
1534 | struct nfs4_layoutreturn_res *res, | |
1535 | int ret) | |
f7e8917a | 1536 | { |
1c5bd76d TM |
1537 | struct pnfs_layout_hdr *lo = args->layout; |
1538 | const nfs4_stateid *arg_stateid = NULL; | |
1539 | const nfs4_stateid *res_stateid = NULL; | |
287bd3e9 | 1540 | struct nfs4_xdr_opaque_data *ld_private = args->ld_private; |
f7e8917a | 1541 | |
9c47b18c TM |
1542 | switch (ret) { |
1543 | case -NFS4ERR_NOMATCHING_LAYOUT: | |
1544 | break; | |
1545 | case 0: | |
1c5bd76d TM |
1546 | if (res->lrs_present) |
1547 | res_stateid = &res->stateid; | |
9c47b18c TM |
1548 | /* Fallthrough */ |
1549 | default: | |
1550 | arg_stateid = &args->stateid; | |
1c5bd76d | 1551 | } |
a19b4785 | 1552 | trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret); |
1c5bd76d TM |
1553 | pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range, |
1554 | res_stateid); | |
287bd3e9 TM |
1555 | if (ld_private && ld_private->ops && ld_private->ops->free) |
1556 | ld_private->ops->free(ld_private); | |
1c5bd76d | 1557 | pnfs_put_layout_hdr(lo); |
f7e8917a FI |
1558 | } |
1559 | ||
500d701f PT |
1560 | bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task) |
1561 | { | |
1562 | struct nfs_inode *nfsi = NFS_I(ino); | |
1563 | struct pnfs_layout_hdr *lo; | |
1564 | bool sleep = false; | |
1565 | ||
1566 | /* we might not have grabbed lo reference. so need to check under | |
1567 | * i_lock */ | |
1568 | spin_lock(&ino->i_lock); | |
1569 | lo = nfsi->layout; | |
ee284e35 TM |
1570 | if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
1571 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
500d701f | 1572 | sleep = true; |
ee284e35 | 1573 | } |
500d701f | 1574 | spin_unlock(&ino->i_lock); |
500d701f PT |
1575 | return sleep; |
1576 | } | |
1577 | ||
b1f69b75 AA |
1578 | /* |
1579 | * Compare two layout segments for sorting into layout cache. | |
1580 | * We want to preferentially return RW over RO layouts, so ensure those | |
1581 | * are seen first. | |
1582 | */ | |
1583 | static s64 | |
7dc0ac70 | 1584 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1585 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1586 | { |
fb3296eb BH |
1587 | s64 d; |
1588 | ||
1589 | /* high offset > low offset */ | |
1590 | d = l1->offset - l2->offset; | |
1591 | if (d) | |
1592 | return d; | |
1593 | ||
1594 | /* short length > long length */ | |
1595 | d = l2->length - l1->length; | |
1596 | if (d) | |
1597 | return d; | |
1598 | ||
b1f69b75 | 1599 | /* read > read/write */ |
fb3296eb | 1600 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1601 | } |
1602 | ||
03772d2f TM |
1603 | static bool |
1604 | pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1, | |
1605 | const struct pnfs_layout_range *l2) | |
1606 | { | |
1607 | return pnfs_lseg_range_cmp(l1, l2) > 0; | |
1608 | } | |
1609 | ||
1610 | static bool | |
1611 | pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg, | |
1612 | struct pnfs_layout_segment *old) | |
1613 | { | |
1614 | return false; | |
1615 | } | |
1616 | ||
1617 | void | |
1618 | pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1619 | struct pnfs_layout_segment *lseg, | |
1620 | bool (*is_after)(const struct pnfs_layout_range *, | |
1621 | const struct pnfs_layout_range *), | |
1622 | bool (*do_merge)(struct pnfs_layout_segment *, | |
1623 | struct pnfs_layout_segment *), | |
1624 | struct list_head *free_me) | |
974cec8c | 1625 | { |
03772d2f | 1626 | struct pnfs_layout_segment *lp, *tmp; |
b1f69b75 | 1627 | |
974cec8c AA |
1628 | dprintk("%s:Begin\n", __func__); |
1629 | ||
03772d2f TM |
1630 | list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) { |
1631 | if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0) | |
1632 | continue; | |
1633 | if (do_merge(lseg, lp)) { | |
1634 | mark_lseg_invalid(lp, free_me); | |
1635 | continue; | |
1636 | } | |
1637 | if (is_after(&lseg->pls_range, &lp->pls_range)) | |
b1f69b75 | 1638 | continue; |
566052c5 | 1639 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1640 | dprintk("%s: inserted lseg %p " |
1641 | "iomode %d offset %llu length %llu before " | |
1642 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1643 | __func__, lseg, lseg->pls_range.iomode, |
1644 | lseg->pls_range.offset, lseg->pls_range.length, | |
1645 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1646 | lp->pls_range.length); | |
fb3296eb | 1647 | goto out; |
974cec8c | 1648 | } |
fb3296eb BH |
1649 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1650 | dprintk("%s: inserted lseg %p " | |
1651 | "iomode %d offset %llu length %llu at tail\n", | |
1652 | __func__, lseg, lseg->pls_range.iomode, | |
1653 | lseg->pls_range.offset, lseg->pls_range.length); | |
1654 | out: | |
70c3bd2b | 1655 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1656 | |
1657 | dprintk("%s:Return\n", __func__); | |
e5e94017 | 1658 | } |
03772d2f TM |
1659 | EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg); |
1660 | ||
1661 | static void | |
1662 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1663 | struct pnfs_layout_segment *lseg, | |
1664 | struct list_head *free_me) | |
1665 | { | |
1666 | struct inode *inode = lo->plh_inode; | |
1667 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
1668 | ||
1669 | if (ld->add_lseg != NULL) | |
1670 | ld->add_lseg(lo, lseg, free_me); | |
1671 | else | |
1672 | pnfs_generic_layout_insert_lseg(lo, lseg, | |
1673 | pnfs_lseg_range_is_after, | |
1674 | pnfs_lseg_no_merge, | |
1675 | free_me); | |
1676 | } | |
e5e94017 BH |
1677 | |
1678 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1679 | alloc_init_layout_hdr(struct inode *ino, |
1680 | struct nfs_open_context *ctx, | |
1681 | gfp_t gfp_flags) | |
e5e94017 BH |
1682 | { |
1683 | struct pnfs_layout_hdr *lo; | |
1684 | ||
636fb9c8 | 1685 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1686 | if (!lo) |
1687 | return NULL; | |
2b28a7be | 1688 | refcount_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1689 | INIT_LIST_HEAD(&lo->plh_layouts); |
1690 | INIT_LIST_HEAD(&lo->plh_segs); | |
68f74479 | 1691 | INIT_LIST_HEAD(&lo->plh_return_segs); |
fd9a8d71 | 1692 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1693 | lo->plh_inode = ino; |
a52458b4 | 1694 | lo->plh_lc_cred = get_cred(ctx->cred); |
67a3b721 | 1695 | lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID; |
e5e94017 BH |
1696 | return lo; |
1697 | } | |
1698 | ||
1699 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1700 | pnfs_find_alloc_layout(struct inode *ino, |
1701 | struct nfs_open_context *ctx, | |
1702 | gfp_t gfp_flags) | |
e5241e43 TM |
1703 | __releases(&ino->i_lock) |
1704 | __acquires(&ino->i_lock) | |
e5e94017 BH |
1705 | { |
1706 | struct nfs_inode *nfsi = NFS_I(ino); | |
1707 | struct pnfs_layout_hdr *new = NULL; | |
1708 | ||
1709 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1710 | ||
251ec410 TM |
1711 | if (nfsi->layout != NULL) |
1712 | goto out_existing; | |
e5e94017 | 1713 | spin_unlock(&ino->i_lock); |
9fa40758 | 1714 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1715 | spin_lock(&ino->i_lock); |
1716 | ||
251ec410 | 1717 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1718 | nfsi->layout = new; |
251ec410 | 1719 | return new; |
7175fe90 YN |
1720 | } else if (new != NULL) |
1721 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1722 | out_existing: |
1723 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1724 | return nfsi->layout; |
1725 | } | |
1726 | ||
b1f69b75 AA |
1727 | /* |
1728 | * iomode matching rules: | |
c7d73af2 TH |
1729 | * iomode lseg strict match |
1730 | * iomode | |
1731 | * ----- ----- ------ ----- | |
1732 | * ANY READ N/A true | |
1733 | * ANY RW N/A true | |
1734 | * RW READ N/A false | |
1735 | * RW RW N/A true | |
1736 | * READ READ N/A true | |
1737 | * READ RW true false | |
1738 | * READ RW false true | |
b1f69b75 | 1739 | */ |
3cb2df17 | 1740 | static bool |
7dc0ac70 | 1741 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
c7d73af2 TH |
1742 | const struct pnfs_layout_range *range, |
1743 | bool strict_iomode) | |
b1f69b75 | 1744 | { |
fb3296eb BH |
1745 | struct pnfs_layout_range range1; |
1746 | ||
1747 | if ((range->iomode == IOMODE_RW && | |
1748 | ls_range->iomode != IOMODE_RW) || | |
c7d73af2 | 1749 | (range->iomode != ls_range->iomode && |
6089dd0d | 1750 | strict_iomode) || |
7dc0ac70 | 1751 | !pnfs_lseg_range_intersecting(ls_range, range)) |
10db5b7a | 1752 | return false; |
fb3296eb BH |
1753 | |
1754 | /* range1 covers only the first byte in the range */ | |
1755 | range1 = *range; | |
1756 | range1.length = 1; | |
7dc0ac70 | 1757 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1758 | } |
1759 | ||
1760 | /* | |
1761 | * lookup range in layout | |
1762 | */ | |
e5e94017 | 1763 | static struct pnfs_layout_segment * |
fb3296eb | 1764 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
c7d73af2 TH |
1765 | struct pnfs_layout_range *range, |
1766 | bool strict_iomode) | |
e5e94017 | 1767 | { |
b1f69b75 AA |
1768 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1769 | ||
1770 | dprintk("%s:Begin\n", __func__); | |
1771 | ||
b7edfaa1 | 1772 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1773 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
c7d73af2 TH |
1774 | pnfs_lseg_range_match(&lseg->pls_range, range, |
1775 | strict_iomode)) { | |
9369a431 | 1776 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1777 | break; |
1778 | } | |
b1f69b75 AA |
1779 | } |
1780 | ||
1781 | dprintk("%s:Return lseg %p ref %d\n", | |
eba6dd69 | 1782 | __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1783 | return ret; |
e5e94017 BH |
1784 | } |
1785 | ||
d23d61c8 AA |
1786 | /* |
1787 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1788 | * to the MDS or over pNFS | |
1789 | * | |
1790 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1791 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1792 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1793 | * WRITE request. | |
1794 | * | |
1795 | * A return of true means use MDS I/O. | |
1796 | * | |
1797 | * From rfc 5661: | |
1798 | * If a file's size is smaller than the file size threshold, data accesses | |
1799 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1800 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1801 | * server. If both file size and I/O size are provided, the client SHOULD | |
1802 | * reach or exceed both thresholds before sending its read or write | |
1803 | * requests to the data server. | |
1804 | */ | |
1805 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1806 | struct inode *ino, int iomode) | |
1807 | { | |
1808 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1809 | struct nfs_inode *nfsi = NFS_I(ino); | |
1810 | loff_t fsize = i_size_read(ino); | |
1811 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1812 | ||
1813 | if (t == NULL) | |
1814 | return ret; | |
1815 | ||
1816 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1817 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1818 | ||
1819 | switch (iomode) { | |
1820 | case IOMODE_READ: | |
1821 | if (t->bm & THRESHOLD_RD) { | |
1822 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1823 | size_set = true; | |
1824 | if (fsize < t->rd_sz) | |
1825 | size = true; | |
1826 | } | |
1827 | if (t->bm & THRESHOLD_RD_IO) { | |
1828 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1829 | nfsi->read_io); | |
1830 | io_set = true; | |
1831 | if (nfsi->read_io < t->rd_io_sz) | |
1832 | io = true; | |
1833 | } | |
1834 | break; | |
1835 | case IOMODE_RW: | |
1836 | if (t->bm & THRESHOLD_WR) { | |
1837 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1838 | size_set = true; | |
1839 | if (fsize < t->wr_sz) | |
1840 | size = true; | |
1841 | } | |
1842 | if (t->bm & THRESHOLD_WR_IO) { | |
1843 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1844 | nfsi->write_io); | |
1845 | io_set = true; | |
1846 | if (nfsi->write_io < t->wr_io_sz) | |
1847 | io = true; | |
1848 | } | |
1849 | break; | |
1850 | } | |
1851 | if (size_set && io_set) { | |
1852 | if (size && io) | |
1853 | ret = true; | |
1854 | } else if (size || io) | |
1855 | ret = true; | |
1856 | ||
1857 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1858 | return ret; | |
1859 | } | |
1860 | ||
d03360aa | 1861 | static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo) |
aa8a45ee PT |
1862 | { |
1863 | /* | |
1864 | * send layoutcommit as it can hold up layoutreturn due to lseg | |
1865 | * reference | |
1866 | */ | |
1867 | pnfs_layoutcommit_inode(lo->plh_inode, false); | |
d03360aa | 1868 | return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN, |
2e5b29f0 | 1869 | nfs_wait_bit_killable, |
d03360aa | 1870 | TASK_KILLABLE); |
aa8a45ee PT |
1871 | } |
1872 | ||
411ae722 TM |
1873 | static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo) |
1874 | { | |
1875 | atomic_inc(&lo->plh_outstanding); | |
1876 | } | |
1877 | ||
1878 | static void nfs_layoutget_end(struct pnfs_layout_hdr *lo) | |
1879 | { | |
1880 | if (atomic_dec_and_test(&lo->plh_outstanding)) | |
1881 | wake_up_var(&lo->plh_outstanding); | |
1882 | } | |
1883 | ||
d67ae825 TH |
1884 | static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo) |
1885 | { | |
1886 | unsigned long *bitlock = &lo->plh_flags; | |
1887 | ||
1888 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1889 | smp_mb__after_atomic(); | |
1890 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1891 | } | |
1892 | ||
78746a38 FI |
1893 | static void _add_to_server_list(struct pnfs_layout_hdr *lo, |
1894 | struct nfs_server *server) | |
1895 | { | |
cf6605d1 | 1896 | if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) { |
78746a38 FI |
1897 | struct nfs_client *clp = server->nfs_client; |
1898 | ||
1899 | /* The lo must be on the clp list if there is any | |
1900 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1901 | */ | |
1902 | spin_lock(&clp->cl_lock); | |
cf6605d1 | 1903 | list_add_tail_rcu(&lo->plh_layouts, &server->layouts); |
78746a38 FI |
1904 | spin_unlock(&clp->cl_lock); |
1905 | } | |
1906 | } | |
1907 | ||
e5e94017 BH |
1908 | /* |
1909 | * Layout segment is retreived from the server if not cached. | |
1910 | * The appropriate layout segment is referenced and returned to the caller. | |
1911 | */ | |
7c24d948 | 1912 | struct pnfs_layout_segment * |
e5e94017 BH |
1913 | pnfs_update_layout(struct inode *ino, |
1914 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1915 | loff_t pos, |
1916 | u64 count, | |
a75b9df9 | 1917 | enum pnfs_iomode iomode, |
c7d73af2 | 1918 | bool strict_iomode, |
a75b9df9 | 1919 | gfp_t gfp_flags) |
e5e94017 | 1920 | { |
fb3296eb BH |
1921 | struct pnfs_layout_range arg = { |
1922 | .iomode = iomode, | |
1923 | .offset = pos, | |
1924 | .length = count, | |
1925 | }; | |
70d2f7b1 | 1926 | unsigned pg_offset; |
6382a441 WAA |
1927 | struct nfs_server *server = NFS_SERVER(ino); |
1928 | struct nfs_client *clp = server->nfs_client; | |
183d9e7b | 1929 | struct pnfs_layout_hdr *lo = NULL; |
e5e94017 | 1930 | struct pnfs_layout_segment *lseg = NULL; |
587f03de | 1931 | struct nfs4_layoutget *lgp; |
183d9e7b JL |
1932 | nfs4_stateid stateid; |
1933 | long timeout = 0; | |
66b53f32 | 1934 | unsigned long giveup = jiffies + (clp->cl_lease_time << 1); |
30005121 | 1935 | bool first; |
e5e94017 | 1936 | |
9a4bf31d | 1937 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) { |
183d9e7b | 1938 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1939 | PNFS_UPDATE_LAYOUT_NO_PNFS); |
f86bbcf8 | 1940 | goto out; |
9a4bf31d | 1941 | } |
d23d61c8 | 1942 | |
9a4bf31d | 1943 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) { |
183d9e7b | 1944 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1945 | PNFS_UPDATE_LAYOUT_MDSTHRESH); |
f86bbcf8 | 1946 | goto out; |
9a4bf31d | 1947 | } |
d23d61c8 | 1948 | |
9bf87482 | 1949 | lookup_again: |
d03360aa TM |
1950 | lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp)); |
1951 | if (IS_ERR(lseg)) | |
1952 | goto out; | |
9bf87482 | 1953 | first = false; |
e5e94017 | 1954 | spin_lock(&ino->i_lock); |
9fa40758 | 1955 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
1956 | if (lo == NULL) { |
1957 | spin_unlock(&ino->i_lock); | |
183d9e7b | 1958 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1959 | PNFS_UPDATE_LAYOUT_NOMEM); |
830ffb56 TM |
1960 | goto out; |
1961 | } | |
e5e94017 | 1962 | |
43f1b3da | 1963 | /* Do we even need to bother with this? */ |
a59c30ac | 1964 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
183d9e7b | 1965 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1966 | PNFS_UPDATE_LAYOUT_BULK_RECALL); |
43f1b3da | 1967 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1968 | goto out_unlock; |
1969 | } | |
1970 | ||
1971 | /* if LAYOUTGET already failed once we don't try again */ | |
2e5b29f0 | 1972 | if (pnfs_layout_io_test_failed(lo, iomode)) { |
183d9e7b | 1973 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1974 | PNFS_UPDATE_LAYOUT_IO_TEST_FAIL); |
e5e94017 | 1975 | goto out_unlock; |
9a4bf31d | 1976 | } |
e5e94017 | 1977 | |
411ae722 TM |
1978 | /* |
1979 | * If the layout segment list is empty, but there are outstanding | |
1980 | * layoutget calls, then they might be subject to a layoutrecall. | |
1981 | */ | |
1982 | if (list_empty(&lo->plh_segs) && | |
1983 | atomic_read(&lo->plh_outstanding) != 0) { | |
1984 | spin_unlock(&ino->i_lock); | |
d03360aa | 1985 | lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding, |
400417b0 | 1986 | !atomic_read(&lo->plh_outstanding))); |
58bbeab4 | 1987 | if (IS_ERR(lseg)) |
411ae722 TM |
1988 | goto out_put_layout_hdr; |
1989 | pnfs_put_layout_hdr(lo); | |
1990 | goto lookup_again; | |
1991 | } | |
1992 | ||
c7d73af2 | 1993 | lseg = pnfs_find_lseg(lo, &arg, strict_iomode); |
183d9e7b JL |
1994 | if (lseg) { |
1995 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
1996 | PNFS_UPDATE_LAYOUT_FOUND_CACHED); | |
1997 | goto out_unlock; | |
1998 | } | |
1999 | ||
183d9e7b JL |
2000 | /* |
2001 | * Choose a stateid for the LAYOUTGET. If we don't have a layout | |
2002 | * stateid, or it has been invalidated, then we must use the open | |
2003 | * stateid. | |
2004 | */ | |
67a3b721 | 2005 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { |
d9aba2b4 | 2006 | int status; |
183d9e7b JL |
2007 | |
2008 | /* | |
2009 | * The first layoutget for the file. Need to serialize per | |
9bf87482 PT |
2010 | * RFC 5661 Errata 3208. |
2011 | */ | |
2012 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
2013 | &lo->plh_flags)) { | |
2014 | spin_unlock(&ino->i_lock); | |
d03360aa TM |
2015 | lseg = ERR_PTR(wait_on_bit(&lo->plh_flags, |
2016 | NFS_LAYOUT_FIRST_LAYOUTGET, | |
2017 | TASK_KILLABLE)); | |
2018 | if (IS_ERR(lseg)) | |
2019 | goto out_put_layout_hdr; | |
9bf87482 | 2020 | pnfs_put_layout_hdr(lo); |
183d9e7b | 2021 | dprintk("%s retrying\n", __func__); |
9bf87482 PT |
2022 | goto lookup_again; |
2023 | } | |
183d9e7b | 2024 | |
fbf4bcc9 | 2025 | spin_unlock(&ino->i_lock); |
183d9e7b | 2026 | first = true; |
d9aba2b4 | 2027 | status = nfs4_select_rw_stateid(ctx->state, |
70d2f7b1 | 2028 | iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ, |
d9aba2b4 TM |
2029 | NULL, &stateid, NULL); |
2030 | if (status != 0) { | |
731c74dd | 2031 | lseg = ERR_PTR(status); |
70d2f7b1 TM |
2032 | trace_pnfs_update_layout(ino, pos, count, |
2033 | iomode, lo, lseg, | |
2034 | PNFS_UPDATE_LAYOUT_INVALID_OPEN); | |
d9aba2b4 TM |
2035 | nfs4_schedule_stateid_recovery(server, ctx->state); |
2036 | pnfs_clear_first_layoutget(lo); | |
2037 | pnfs_put_layout_hdr(lo); | |
2038 | goto lookup_again; | |
70d2f7b1 | 2039 | } |
fbf4bcc9 | 2040 | spin_lock(&ino->i_lock); |
9bf87482 | 2041 | } else { |
183d9e7b | 2042 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
9bf87482 | 2043 | } |
568e8c49 | 2044 | |
aa8a45ee PT |
2045 | /* |
2046 | * Because we free lsegs before sending LAYOUTRETURN, we need to wait | |
2047 | * for LAYOUTRETURN even if first is true. | |
2048 | */ | |
8f70f53a | 2049 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
aa8a45ee PT |
2050 | spin_unlock(&ino->i_lock); |
2051 | dprintk("%s wait for layoutreturn\n", __func__); | |
d03360aa TM |
2052 | lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo)); |
2053 | if (!IS_ERR(lseg)) { | |
d67ae825 TH |
2054 | if (first) |
2055 | pnfs_clear_first_layoutget(lo); | |
aa8a45ee PT |
2056 | pnfs_put_layout_hdr(lo); |
2057 | dprintk("%s retrying\n", __func__); | |
183d9e7b JL |
2058 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, |
2059 | lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
aa8a45ee PT |
2060 | goto lookup_again; |
2061 | } | |
183d9e7b | 2062 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 2063 | PNFS_UPDATE_LAYOUT_RETURN); |
aa8a45ee PT |
2064 | goto out_put_layout_hdr; |
2065 | } | |
2066 | ||
9a4bf31d | 2067 | if (pnfs_layoutgets_blocked(lo)) { |
183d9e7b | 2068 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 2069 | PNFS_UPDATE_LAYOUT_BLOCKED); |
cf7d63f1 | 2070 | goto out_unlock; |
9a4bf31d | 2071 | } |
411ae722 | 2072 | nfs_layoutget_begin(lo); |
f49f9baa | 2073 | spin_unlock(&ino->i_lock); |
30005121 | 2074 | |
78746a38 | 2075 | _add_to_server_list(lo, server); |
e5e94017 | 2076 | |
09cbfeaf | 2077 | pg_offset = arg.offset & ~PAGE_MASK; |
707ed5fd BH |
2078 | if (pg_offset) { |
2079 | arg.offset -= pg_offset; | |
2080 | arg.length += pg_offset; | |
2081 | } | |
7c24d948 | 2082 | if (arg.length != NFS4_MAX_UINT64) |
09cbfeaf | 2083 | arg.length = PAGE_ALIGN(arg.length); |
707ed5fd | 2084 | |
5e36e2a9 | 2085 | lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags); |
587f03de FI |
2086 | if (!lgp) { |
2087 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL, | |
2088 | PNFS_UPDATE_LAYOUT_NOMEM); | |
411ae722 | 2089 | nfs_layoutget_end(lo); |
587f03de FI |
2090 | goto out_put_layout_hdr; |
2091 | } | |
2092 | ||
dacb452d | 2093 | lseg = nfs4_proc_layoutget(lgp, &timeout); |
183d9e7b JL |
2094 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
2095 | PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET); | |
411ae722 | 2096 | nfs_layoutget_end(lo); |
83026d80 | 2097 | if (IS_ERR(lseg)) { |
183d9e7b | 2098 | switch(PTR_ERR(lseg)) { |
e85d7ee4 | 2099 | case -EBUSY: |
183d9e7b JL |
2100 | if (time_after(jiffies, giveup)) |
2101 | lseg = NULL; | |
66b53f32 TM |
2102 | break; |
2103 | case -ERECALLCONFLICT: | |
183d9e7b | 2104 | case -EAGAIN: |
56b38a1f | 2105 | break; |
183d9e7b JL |
2106 | default: |
2107 | if (!nfs_error_is_fatal(PTR_ERR(lseg))) { | |
2108 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
2109 | lseg = NULL; | |
2110 | } | |
56b38a1f TM |
2111 | goto out_put_layout_hdr; |
2112 | } | |
2113 | if (lseg) { | |
2114 | if (first) | |
2115 | pnfs_clear_first_layoutget(lo); | |
2116 | trace_pnfs_update_layout(ino, pos, count, | |
2117 | iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
2118 | pnfs_put_layout_hdr(lo); | |
2119 | goto lookup_again; | |
83026d80 JL |
2120 | } |
2121 | } else { | |
2122 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
2123 | } | |
2124 | ||
830ffb56 | 2125 | out_put_layout_hdr: |
d67ae825 TH |
2126 | if (first) |
2127 | pnfs_clear_first_layoutget(lo); | |
d5b9216f TM |
2128 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
2129 | PNFS_UPDATE_LAYOUT_EXIT); | |
70c3bd2b | 2130 | pnfs_put_layout_hdr(lo); |
e5e94017 | 2131 | out: |
f86bbcf8 TM |
2132 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
2133 | "(%s, offset: %llu, length: %llu)\n", | |
2134 | __func__, ino->i_sb->s_id, | |
2135 | (unsigned long long)NFS_FILEID(ino), | |
d600ad1f | 2136 | IS_ERR_OR_NULL(lseg) ? "not found" : "found", |
f86bbcf8 TM |
2137 | iomode==IOMODE_RW ? "read/write" : "read-only", |
2138 | (unsigned long long)pos, | |
2139 | (unsigned long long)count); | |
e5e94017 BH |
2140 | return lseg; |
2141 | out_unlock: | |
2142 | spin_unlock(&ino->i_lock); | |
830ffb56 | 2143 | goto out_put_layout_hdr; |
e5e94017 | 2144 | } |
7c24d948 | 2145 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 2146 | |
540d9864 TM |
2147 | static bool |
2148 | pnfs_sanity_check_layout_range(struct pnfs_layout_range *range) | |
2149 | { | |
2150 | switch (range->iomode) { | |
2151 | case IOMODE_READ: | |
2152 | case IOMODE_RW: | |
2153 | break; | |
2154 | default: | |
2155 | return false; | |
2156 | } | |
2157 | if (range->offset == NFS4_MAX_UINT64) | |
2158 | return false; | |
2159 | if (range->length == 0) | |
2160 | return false; | |
2161 | if (range->length != NFS4_MAX_UINT64 && | |
2162 | range->length > NFS4_MAX_UINT64 - range->offset) | |
2163 | return false; | |
2164 | return true; | |
2165 | } | |
2166 | ||
78746a38 FI |
2167 | static struct pnfs_layout_hdr * |
2168 | _pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx) | |
2169 | { | |
2170 | struct pnfs_layout_hdr *lo; | |
2171 | ||
2172 | spin_lock(&ino->i_lock); | |
2173 | lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL); | |
2174 | if (!lo) | |
2175 | goto out_unlock; | |
2176 | if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) | |
2177 | goto out_unlock; | |
2178 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) | |
2179 | goto out_unlock; | |
2180 | if (pnfs_layoutgets_blocked(lo)) | |
2181 | goto out_unlock; | |
2182 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags)) | |
2183 | goto out_unlock; | |
411ae722 | 2184 | nfs_layoutget_begin(lo); |
78746a38 FI |
2185 | spin_unlock(&ino->i_lock); |
2186 | _add_to_server_list(lo, NFS_SERVER(ino)); | |
2187 | return lo; | |
2188 | ||
2189 | out_unlock: | |
2190 | spin_unlock(&ino->i_lock); | |
2191 | pnfs_put_layout_hdr(lo); | |
2192 | return NULL; | |
2193 | } | |
2194 | ||
2409a976 FI |
2195 | static void _lgopen_prepare_attached(struct nfs4_opendata *data, |
2196 | struct nfs_open_context *ctx) | |
2197 | { | |
78746a38 FI |
2198 | struct inode *ino = data->dentry->d_inode; |
2199 | struct pnfs_layout_range rng = { | |
2200 | .iomode = (data->o_arg.fmode & FMODE_WRITE) ? | |
2201 | IOMODE_RW: IOMODE_READ, | |
2202 | .offset = 0, | |
2203 | .length = NFS4_MAX_UINT64, | |
2204 | }; | |
2205 | struct nfs4_layoutget *lgp; | |
2206 | struct pnfs_layout_hdr *lo; | |
2207 | ||
64294b08 TM |
2208 | /* Heuristic: don't send layoutget if we have cached data */ |
2209 | if (rng.iomode == IOMODE_READ && | |
2210 | (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0)) | |
2211 | return; | |
2212 | ||
78746a38 FI |
2213 | lo = _pnfs_grab_empty_layout(ino, ctx); |
2214 | if (!lo) | |
2215 | return; | |
2216 | lgp = pnfs_alloc_init_layoutget_args(ino, ctx, ¤t_stateid, | |
2217 | &rng, GFP_KERNEL); | |
2218 | if (!lgp) { | |
2219 | pnfs_clear_first_layoutget(lo); | |
2220 | pnfs_put_layout_hdr(lo); | |
2221 | return; | |
2222 | } | |
2223 | data->lgp = lgp; | |
2224 | data->o_arg.lg_args = &lgp->args; | |
2225 | data->o_res.lg_res = &lgp->res; | |
2409a976 FI |
2226 | } |
2227 | ||
2228 | static void _lgopen_prepare_floating(struct nfs4_opendata *data, | |
2229 | struct nfs_open_context *ctx) | |
2230 | { | |
2231 | struct pnfs_layout_range rng = { | |
2232 | .iomode = (data->o_arg.fmode & FMODE_WRITE) ? | |
2233 | IOMODE_RW: IOMODE_READ, | |
2234 | .offset = 0, | |
2235 | .length = NFS4_MAX_UINT64, | |
2236 | }; | |
2237 | struct nfs4_layoutget *lgp; | |
2238 | ||
2239 | lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, ¤t_stateid, | |
2240 | &rng, GFP_KERNEL); | |
2241 | if (!lgp) | |
2242 | return; | |
2243 | data->lgp = lgp; | |
2244 | data->o_arg.lg_args = &lgp->args; | |
2245 | data->o_res.lg_res = &lgp->res; | |
2246 | } | |
2247 | ||
2248 | void pnfs_lgopen_prepare(struct nfs4_opendata *data, | |
2249 | struct nfs_open_context *ctx) | |
2250 | { | |
2251 | struct nfs_server *server = NFS_SERVER(data->dir->d_inode); | |
2252 | ||
2253 | if (!(pnfs_enabled_sb(server) && | |
2254 | server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN)) | |
2255 | return; | |
2256 | /* Could check on max_ops, but currently hardcoded high enough */ | |
6e01260c FI |
2257 | if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN)) |
2258 | return; | |
2409a976 FI |
2259 | if (data->state) |
2260 | _lgopen_prepare_attached(data, ctx); | |
2261 | else | |
2262 | _lgopen_prepare_floating(data, ctx); | |
2263 | } | |
2264 | ||
2265 | void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp, | |
2266 | struct nfs_open_context *ctx) | |
2267 | { | |
2268 | struct pnfs_layout_hdr *lo; | |
2269 | struct pnfs_layout_segment *lseg; | |
c49b5209 | 2270 | struct nfs_server *srv = NFS_SERVER(ino); |
2409a976 FI |
2271 | u32 iomode; |
2272 | ||
6e01260c | 2273 | if (!lgp) |
2409a976 | 2274 | return; |
6e01260c FI |
2275 | dprintk("%s: entered with status %i\n", __func__, lgp->res.status); |
2276 | if (lgp->res.status) { | |
2277 | switch (lgp->res.status) { | |
6e01260c | 2278 | default: |
8dc96566 TM |
2279 | break; |
2280 | /* | |
2281 | * Halt lgopen attempts if the server doesn't recognise | |
2282 | * the "current stateid" value, the layout type, or the | |
2283 | * layoutget operation as being valid. | |
2284 | * Also if it complains about too many ops in the compound | |
2285 | * or of the request/reply being too big. | |
2286 | */ | |
2287 | case -NFS4ERR_BAD_STATEID: | |
2288 | case -NFS4ERR_NOTSUPP: | |
2289 | case -NFS4ERR_REP_TOO_BIG: | |
2290 | case -NFS4ERR_REP_TOO_BIG_TO_CACHE: | |
2291 | case -NFS4ERR_REQ_TOO_BIG: | |
2292 | case -NFS4ERR_TOO_MANY_OPS: | |
2293 | case -NFS4ERR_UNKNOWN_LAYOUTTYPE: | |
c49b5209 | 2294 | srv->caps &= ~NFS_CAP_LGOPEN; |
6e01260c FI |
2295 | } |
2296 | return; | |
2297 | } | |
2409a976 | 2298 | if (!lgp->args.inode) { |
78746a38 FI |
2299 | lo = _pnfs_grab_empty_layout(ino, ctx); |
2300 | if (!lo) | |
2301 | return; | |
2409a976 FI |
2302 | lgp->args.inode = ino; |
2303 | } else | |
2304 | lo = NFS_I(lgp->args.inode)->layout; | |
2409a976 FI |
2305 | |
2306 | lseg = pnfs_layout_process(lgp); | |
32f1c28f | 2307 | if (!IS_ERR(lseg)) { |
2409a976 FI |
2308 | iomode = lgp->args.range.iomode; |
2309 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
2310 | pnfs_put_lseg(lseg); | |
2311 | } | |
30ae2412 FI |
2312 | } |
2313 | ||
2314 | void nfs4_lgopen_release(struct nfs4_layoutget *lgp) | |
2315 | { | |
2316 | if (lgp != NULL) { | |
2317 | struct inode *inode = lgp->args.inode; | |
2318 | if (inode) { | |
2319 | struct pnfs_layout_hdr *lo = NFS_I(inode)->layout; | |
30ae2412 | 2320 | pnfs_clear_first_layoutget(lo); |
411ae722 | 2321 | nfs_layoutget_end(lo); |
30ae2412 FI |
2322 | } |
2323 | pnfs_layoutget_free(lgp); | |
2324 | } | |
2409a976 FI |
2325 | } |
2326 | ||
a0b0a6e3 | 2327 | struct pnfs_layout_segment * |
b1f69b75 AA |
2328 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
2329 | { | |
2330 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
2331 | struct nfs4_layoutget_res *res = &lgp->res; | |
2332 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 2333 | struct inode *ino = lo->plh_inode; |
78096cca | 2334 | LIST_HEAD(free_me); |
540d9864 TM |
2335 | |
2336 | if (!pnfs_sanity_check_layout_range(&res->range)) | |
1b3c6d07 | 2337 | return ERR_PTR(-EINVAL); |
b1f69b75 AA |
2338 | |
2339 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 2340 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
1b3c6d07 | 2341 | if (IS_ERR_OR_NULL(lseg)) { |
b1f69b75 | 2342 | if (!lseg) |
1b3c6d07 JL |
2343 | lseg = ERR_PTR(-ENOMEM); |
2344 | ||
2345 | dprintk("%s: Could not allocate layout: error %ld\n", | |
2346 | __func__, PTR_ERR(lseg)); | |
2347 | return lseg; | |
b1f69b75 AA |
2348 | } |
2349 | ||
119cef97 | 2350 | pnfs_init_lseg(lo, lseg, &res->range, &res->stateid); |
1013df61 | 2351 | |
b1f69b75 | 2352 | spin_lock(&ino->i_lock); |
e1c06f80 | 2353 | if (pnfs_layoutgets_blocked(lo)) { |
43f1b3da | 2354 | dprintk("%s forget reply due to state\n", __func__); |
1b3c6d07 | 2355 | goto out_forget; |
43f1b3da | 2356 | } |
038d6493 | 2357 | |
9888d837 TM |
2358 | if (!pnfs_layout_is_valid(lo)) { |
2359 | /* We have a completely new layout */ | |
59b56394 | 2360 | pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true); |
9888d837 | 2361 | } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
362f7474 CH |
2362 | /* existing state ID, make sure the sequence number matches. */ |
2363 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
2364 | dprintk("%s forget reply due to sequence\n", __func__); | |
1b3c6d07 | 2365 | goto out_forget; |
362f7474 | 2366 | } |
59b56394 | 2367 | pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false); |
362f7474 CH |
2368 | } else { |
2369 | /* | |
2370 | * We got an entirely new state ID. Mark all segments for the | |
9888d837 | 2371 | * inode invalid, and retry the layoutget |
362f7474 | 2372 | */ |
d9b61708 | 2373 | pnfs_mark_layout_stateid_invalid(lo, &free_me); |
9888d837 | 2374 | goto out_forget; |
362f7474 | 2375 | } |
038d6493 | 2376 | |
9369a431 | 2377 | pnfs_get_lseg(lseg); |
03772d2f | 2378 | pnfs_layout_insert_lseg(lo, lseg, &free_me); |
8e0acf90 | 2379 | |
b1f69b75 | 2380 | |
3976143b | 2381 | if (res->return_on_close) |
f7e8917a | 2382 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); |
f7e8917a | 2383 | |
b1f69b75 | 2384 | spin_unlock(&ino->i_lock); |
78096cca | 2385 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 2386 | return lseg; |
43f1b3da | 2387 | |
1b3c6d07 | 2388 | out_forget: |
43f1b3da FI |
2389 | spin_unlock(&ino->i_lock); |
2390 | lseg->pls_layout = lo; | |
2391 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1b3c6d07 | 2392 | return ERR_PTR(-EAGAIN); |
b1f69b75 AA |
2393 | } |
2394 | ||
2f215968 TM |
2395 | /** |
2396 | * pnfs_mark_matching_lsegs_return - Free or return matching layout segments | |
2397 | * @lo: pointer to layout header | |
2398 | * @tmp_list: list header to be used with pnfs_free_lseg_list() | |
2399 | * @return_range: describe layout segment ranges to be returned | |
e0b7d420 | 2400 | * @seq: stateid seqid to match |
2f215968 TM |
2401 | * |
2402 | * This function is mainly intended for use by layoutrecall. It attempts | |
2403 | * to free the layout segment immediately, or else to mark it for return | |
2404 | * as soon as its reference count drops to zero. | |
e0b7d420 TM |
2405 | * |
2406 | * Returns | |
2407 | * - 0: a layoutreturn needs to be scheduled. | |
2408 | * - EBUSY: there are layout segment that are still in use. | |
2409 | * - ENOENT: there are no layout segments that need to be returned. | |
2f215968 | 2410 | */ |
10335556 | 2411 | int |
016256df PT |
2412 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, |
2413 | struct list_head *tmp_list, | |
6d597e17 JL |
2414 | const struct pnfs_layout_range *return_range, |
2415 | u32 seq) | |
016256df PT |
2416 | { |
2417 | struct pnfs_layout_segment *lseg, *next; | |
10335556 | 2418 | int remaining = 0; |
016256df PT |
2419 | |
2420 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
2421 | ||
fc7ff367 | 2422 | assert_spin_locked(&lo->plh_inode->i_lock); |
016256df PT |
2423 | |
2424 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) | |
e036f464 | 2425 | if (pnfs_match_lseg_recall(lseg, return_range, seq)) { |
016256df PT |
2426 | dprintk("%s: marking lseg %p iomode %d " |
2427 | "offset %llu length %llu\n", __func__, | |
2428 | lseg, lseg->pls_range.iomode, | |
2429 | lseg->pls_range.offset, | |
2430 | lseg->pls_range.length); | |
ff041727 | 2431 | if (mark_lseg_invalid(lseg, tmp_list)) |
2f215968 TM |
2432 | continue; |
2433 | remaining++; | |
016256df | 2434 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); |
016256df | 2435 | } |
6d597e17 | 2436 | |
e0b7d420 | 2437 | if (remaining) { |
6d597e17 | 2438 | pnfs_set_plh_return_info(lo, return_range->iomode, seq); |
e0b7d420 TM |
2439 | return -EBUSY; |
2440 | } | |
6d597e17 | 2441 | |
e0b7d420 TM |
2442 | if (!list_empty(&lo->plh_return_segs)) { |
2443 | pnfs_set_plh_return_info(lo, return_range->iomode, seq); | |
2444 | return 0; | |
2445 | } | |
2446 | ||
2447 | return -ENOENT; | |
016256df PT |
2448 | } |
2449 | ||
b5fdf841 TM |
2450 | static void |
2451 | pnfs_mark_layout_for_return(struct inode *inode, | |
2452 | const struct pnfs_layout_range *range) | |
016256df | 2453 | { |
b5fdf841 | 2454 | struct pnfs_layout_hdr *lo; |
10335556 | 2455 | bool return_now = false; |
016256df PT |
2456 | |
2457 | spin_lock(&inode->i_lock); | |
b5fdf841 | 2458 | lo = NFS_I(inode)->layout; |
bdebfccd TM |
2459 | if (!pnfs_layout_is_valid(lo)) { |
2460 | spin_unlock(&inode->i_lock); | |
2461 | return; | |
2462 | } | |
b5fdf841 | 2463 | pnfs_set_plh_return_info(lo, range->iomode, 0); |
016256df PT |
2464 | /* |
2465 | * mark all matching lsegs so that we are sure to have no live | |
2466 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
2467 | * for how it works. | |
2468 | */ | |
b5fdf841 | 2469 | if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) { |
44ea8dfc | 2470 | const struct cred *cred; |
10335556 | 2471 | nfs4_stateid stateid; |
e5fd1904 | 2472 | enum pnfs_iomode iomode; |
10335556 | 2473 | |
44ea8dfc | 2474 | return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode); |
10335556 TM |
2475 | spin_unlock(&inode->i_lock); |
2476 | if (return_now) | |
44ea8dfc | 2477 | pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false); |
10335556 TM |
2478 | } else { |
2479 | spin_unlock(&inode->i_lock); | |
2480 | nfs_commit_inode(inode, 0); | |
2481 | } | |
016256df | 2482 | } |
b5fdf841 TM |
2483 | |
2484 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
2485 | struct pnfs_layout_segment *lseg) | |
2486 | { | |
2487 | struct pnfs_layout_range range = { | |
2488 | .iomode = lseg->pls_range.iomode, | |
2489 | .offset = 0, | |
2490 | .length = NFS4_MAX_UINT64, | |
2491 | }; | |
2492 | ||
2493 | pnfs_mark_layout_for_return(inode, &range); | |
2494 | } | |
016256df PT |
2495 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); |
2496 | ||
b5fdf841 TM |
2497 | static bool |
2498 | pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo) | |
2499 | { | |
2500 | return pnfs_layout_is_valid(lo) && | |
2501 | !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) && | |
2502 | !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
2503 | } | |
2504 | ||
2505 | static struct pnfs_layout_segment * | |
2506 | pnfs_find_first_lseg(struct pnfs_layout_hdr *lo, | |
2507 | const struct pnfs_layout_range *range, | |
2508 | enum pnfs_iomode iomode) | |
2509 | { | |
2510 | struct pnfs_layout_segment *lseg; | |
2511 | ||
2512 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
2513 | if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) | |
2514 | continue; | |
2515 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
2516 | continue; | |
2517 | if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY) | |
2518 | continue; | |
2519 | if (pnfs_lseg_range_intersecting(&lseg->pls_range, range)) | |
2520 | return lseg; | |
2521 | } | |
2522 | return NULL; | |
2523 | } | |
2524 | ||
2525 | /* Find open file states whose mode matches that of the range */ | |
2526 | static bool | |
2527 | pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo, | |
2528 | const struct pnfs_layout_range *range) | |
2529 | { | |
2530 | struct list_head *head; | |
2531 | struct nfs_open_context *ctx; | |
2532 | fmode_t mode = 0; | |
2533 | ||
2534 | if (!pnfs_layout_can_be_returned(lo) || | |
2535 | !pnfs_find_first_lseg(lo, range, range->iomode)) | |
2536 | return false; | |
2537 | ||
2538 | head = &NFS_I(lo->plh_inode)->open_files; | |
2539 | list_for_each_entry_rcu(ctx, head, list) { | |
2540 | if (ctx->state) | |
2541 | mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE); | |
2542 | } | |
2543 | ||
2544 | switch (range->iomode) { | |
2545 | default: | |
2546 | break; | |
2547 | case IOMODE_READ: | |
2548 | mode &= ~FMODE_WRITE; | |
2549 | break; | |
2550 | case IOMODE_RW: | |
2551 | if (pnfs_find_first_lseg(lo, range, IOMODE_READ)) | |
2552 | mode &= ~FMODE_READ; | |
2553 | } | |
2554 | return mode == 0; | |
2555 | } | |
2556 | ||
2557 | static int | |
2558 | pnfs_layout_return_unused_byserver(struct nfs_server *server, void *data) | |
2559 | { | |
2560 | const struct pnfs_layout_range *range = data; | |
2561 | struct pnfs_layout_hdr *lo; | |
2562 | struct inode *inode; | |
2563 | restart: | |
2564 | rcu_read_lock(); | |
2565 | list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) { | |
2566 | if (!pnfs_layout_can_be_returned(lo) || | |
2567 | test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) | |
2568 | continue; | |
2569 | inode = lo->plh_inode; | |
2570 | spin_lock(&inode->i_lock); | |
2571 | if (!pnfs_should_return_unused_layout(lo, range)) { | |
2572 | spin_unlock(&inode->i_lock); | |
2573 | continue; | |
2574 | } | |
2575 | spin_unlock(&inode->i_lock); | |
2576 | inode = pnfs_grab_inode_layout_hdr(lo); | |
2577 | if (!inode) | |
2578 | continue; | |
2579 | rcu_read_unlock(); | |
2580 | pnfs_mark_layout_for_return(inode, range); | |
2581 | iput(inode); | |
2582 | cond_resched(); | |
2583 | goto restart; | |
2584 | } | |
2585 | rcu_read_unlock(); | |
2586 | return 0; | |
2587 | } | |
2588 | ||
2589 | void | |
2590 | pnfs_layout_return_unused_byclid(struct nfs_client *clp, | |
2591 | enum pnfs_iomode iomode) | |
2592 | { | |
2593 | struct pnfs_layout_range range = { | |
2594 | .iomode = iomode, | |
2595 | .offset = 0, | |
2596 | .length = NFS4_MAX_UINT64, | |
2597 | }; | |
2598 | ||
2599 | nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver, | |
2600 | &range); | |
2601 | } | |
2602 | ||
b3230e80 TM |
2603 | void |
2604 | pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio) | |
2605 | { | |
2606 | if (pgio->pg_lseg == NULL || | |
2607 | test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags)) | |
2608 | return; | |
2609 | pnfs_put_lseg(pgio->pg_lseg); | |
2610 | pgio->pg_lseg = NULL; | |
2611 | } | |
2612 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout); | |
2613 | ||
08cb5b0f BC |
2614 | /* |
2615 | * Check for any intersection between the request and the pgio->pg_lseg, | |
2616 | * and if none, put this pgio->pg_lseg away. | |
2617 | */ | |
e1e54ab7 | 2618 | void |
08cb5b0f BC |
2619 | pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) |
2620 | { | |
2621 | if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) { | |
2622 | pnfs_put_lseg(pgio->pg_lseg); | |
2623 | pgio->pg_lseg = NULL; | |
2624 | } | |
2625 | } | |
e1e54ab7 | 2626 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range); |
08cb5b0f | 2627 | |
d8007d4d TM |
2628 | void |
2629 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
2630 | { | |
1fd937bd PT |
2631 | u64 rd_size = req->wb_bytes; |
2632 | ||
b3230e80 | 2633 | pnfs_generic_pg_check_layout(pgio); |
08cb5b0f | 2634 | pnfs_generic_pg_check_range(pgio, req); |
cb5d04bc PT |
2635 | if (pgio->pg_lseg == NULL) { |
2636 | if (pgio->pg_dreq == NULL) | |
2637 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
2638 | else | |
2639 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
2640 | ||
2641 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
9fcd5960 | 2642 | nfs_req_openctx(req), |
cb5d04bc PT |
2643 | req_offset(req), |
2644 | rd_size, | |
2645 | IOMODE_READ, | |
c7d73af2 | 2646 | false, |
cb5d04bc | 2647 | GFP_KERNEL); |
d600ad1f PT |
2648 | if (IS_ERR(pgio->pg_lseg)) { |
2649 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
2650 | pgio->pg_lseg = NULL; | |
2651 | return; | |
2652 | } | |
cb5d04bc | 2653 | } |
e885de1a TM |
2654 | /* If no lseg, fall back to read through mds */ |
2655 | if (pgio->pg_lseg == NULL) | |
1f945357 | 2656 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 2657 | |
d8007d4d TM |
2658 | } |
2659 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
2660 | ||
2661 | void | |
6296556f PT |
2662 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
2663 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 2664 | { |
b3230e80 | 2665 | pnfs_generic_pg_check_layout(pgio); |
08cb5b0f | 2666 | pnfs_generic_pg_check_range(pgio, req); |
d600ad1f | 2667 | if (pgio->pg_lseg == NULL) { |
cb5d04bc | 2668 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
9fcd5960 | 2669 | nfs_req_openctx(req), |
cb5d04bc PT |
2670 | req_offset(req), |
2671 | wb_size, | |
2672 | IOMODE_RW, | |
c7d73af2 | 2673 | false, |
2b17d725 | 2674 | GFP_KERNEL); |
d600ad1f PT |
2675 | if (IS_ERR(pgio->pg_lseg)) { |
2676 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
2677 | pgio->pg_lseg = NULL; | |
2678 | return; | |
2679 | } | |
2680 | } | |
e885de1a TM |
2681 | /* If no lseg, fall back to write through mds */ |
2682 | if (pgio->pg_lseg == NULL) | |
1f945357 | 2683 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
2684 | } |
2685 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
2686 | ||
180bb5ec WAA |
2687 | void |
2688 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
2689 | { | |
2690 | if (desc->pg_lseg) { | |
2691 | pnfs_put_lseg(desc->pg_lseg); | |
2692 | desc->pg_lseg = NULL; | |
2693 | } | |
2694 | } | |
2695 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
2696 | ||
b4fdac1a WAA |
2697 | /* |
2698 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
2699 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
2700 | */ | |
2701 | size_t | |
a7d42ddb WAA |
2702 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
2703 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 2704 | { |
0f9c429e | 2705 | unsigned int size; |
c5e20cb7 | 2706 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
2707 | |
2708 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
2709 | if (!size) |
2710 | return 0; | |
94ad1c80 | 2711 | |
19982ba8 | 2712 | /* |
c5e20cb7 WAA |
2713 | * 'size' contains the number of bytes left in the current page (up |
2714 | * to the original size asked for in @req->wb_bytes). | |
2715 | * | |
2716 | * Calculate how many bytes are left in the layout segment | |
2717 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
2718 | * |
2719 | * Please also note that 'end_offset' is actually the offset of the | |
2720 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
2721 | * | |
2722 | */ | |
19b54848 | 2723 | if (pgio->pg_lseg) { |
17822b20 | 2724 | seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset, |
c5e20cb7 WAA |
2725 | pgio->pg_lseg->pls_range.length); |
2726 | req_start = req_offset(req); | |
08cb5b0f | 2727 | |
c5e20cb7 | 2728 | /* start of request is past the last byte of this segment */ |
08cb5b0f | 2729 | if (req_start >= seg_end) |
19b54848 | 2730 | return 0; |
c5e20cb7 WAA |
2731 | |
2732 | /* adjust 'size' iff there are fewer bytes left in the | |
2733 | * segment than what nfs_generic_pg_test returned */ | |
2734 | seg_left = seg_end - req_start; | |
2735 | if (seg_left < size) | |
2736 | size = (unsigned int)seg_left; | |
19b54848 | 2737 | } |
0f9c429e | 2738 | |
19b54848 | 2739 | return size; |
94ad1c80 | 2740 | } |
89a58e32 | 2741 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 2742 | |
53113ad3 | 2743 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
2744 | { |
2745 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
2746 | |
2747 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
2748 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
2749 | hdr->completion_ops); | |
c7070113 | 2750 | set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags); |
53113ad3 | 2751 | return nfs_pageio_resend(&pgio, hdr); |
e2fecb21 | 2752 | } |
e7dd79af | 2753 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 2754 | |
d45f60c6 | 2755 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2756 | { |
cd841605 FI |
2757 | |
2758 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
2759 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2760 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2761 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2762 | } |
6c75dc0d | 2763 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2764 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
2765 | } |
2766 | ||
d20581aa BH |
2767 | /* |
2768 | * Called by non rpc-based layout drivers | |
2769 | */ | |
d45f60c6 | 2770 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 2771 | { |
f8417b48 | 2772 | if (likely(!hdr->pnfs_error)) { |
67af7611 TM |
2773 | pnfs_set_layoutcommit(hdr->inode, hdr->lseg, |
2774 | hdr->mds_offset + hdr->res.count); | |
d45f60c6 | 2775 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2776 | } |
2777 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); | |
2778 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2779 | pnfs_ld_handle_write_error(hdr); |
2780 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 2781 | } |
d20581aa | 2782 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 2783 | |
dce81290 TM |
2784 | static void |
2785 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2786 | struct nfs_pgio_header *hdr) |
dce81290 | 2787 | { |
48d635f1 | 2788 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2789 | |
6c75dc0d | 2790 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2791 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 2792 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 2793 | mirror->pg_recoalesce = 1; |
6c75dc0d | 2794 | } |
ba4a76f7 | 2795 | hdr->completion_ops->completion(hdr); |
dce81290 TM |
2796 | } |
2797 | ||
2798 | static enum pnfs_try_status | |
d45f60c6 | 2799 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
2800 | const struct rpc_call_ops *call_ops, |
2801 | struct pnfs_layout_segment *lseg, | |
2802 | int how) | |
0382b744 | 2803 | { |
cd841605 | 2804 | struct inode *inode = hdr->inode; |
0382b744 AA |
2805 | enum pnfs_try_status trypnfs; |
2806 | struct nfs_server *nfss = NFS_SERVER(inode); | |
2807 | ||
cd841605 | 2808 | hdr->mds_ops = call_ops; |
0382b744 AA |
2809 | |
2810 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
2811 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
2812 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 2813 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 2814 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
2815 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2816 | return trypnfs; | |
2817 | } | |
2818 | ||
dce81290 | 2819 | static void |
7f714720 WAA |
2820 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
2821 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 2822 | { |
dce81290 TM |
2823 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2824 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2825 | enum pnfs_try_status trypnfs; |
dce81290 | 2826 | |
d45f60c6 | 2827 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
37f8aa16 TM |
2828 | switch (trypnfs) { |
2829 | case PNFS_NOT_ATTEMPTED: | |
d45f60c6 | 2830 | pnfs_write_through_mds(desc, hdr); |
37f8aa16 TM |
2831 | case PNFS_ATTEMPTED: |
2832 | break; | |
2833 | case PNFS_TRY_AGAIN: | |
2834 | /* cleanup hdr and prepare to redo pnfs */ | |
2835 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { | |
2836 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); | |
2837 | list_splice_init(&hdr->pages, &mirror->pg_list); | |
2838 | mirror->pg_recoalesce = 1; | |
2839 | } | |
2840 | hdr->mds_ops->rpc_release(hdr); | |
2841 | } | |
dce81290 TM |
2842 | } |
2843 | ||
6c75dc0d FI |
2844 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
2845 | { | |
9369a431 | 2846 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2847 | nfs_pgio_header_free(hdr); |
6c75dc0d FI |
2848 | } |
2849 | ||
dce81290 TM |
2850 | int |
2851 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
2852 | { | |
6c75dc0d | 2853 | struct nfs_pgio_header *hdr; |
dce81290 TM |
2854 | int ret; |
2855 | ||
1e7f3a48 WAA |
2856 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2857 | if (!hdr) { | |
2bff2288 PT |
2858 | desc->pg_error = -ENOMEM; |
2859 | return desc->pg_error; | |
dce81290 | 2860 | } |
6c75dc0d | 2861 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 2862 | |
9369a431 | 2863 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2864 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2865 | if (!ret) |
7f714720 | 2866 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 2867 | |
6c75dc0d | 2868 | return ret; |
dce81290 TM |
2869 | } |
2870 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
2871 | ||
53113ad3 | 2872 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
2873 | { |
2874 | struct nfs_pageio_descriptor pgio; | |
2875 | ||
1acbbb4e | 2876 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
2877 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
2878 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 2879 | } |
e7dd79af | 2880 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 2881 | |
d45f60c6 | 2882 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2883 | { |
cd841605 FI |
2884 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
2885 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2886 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2887 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2888 | } |
4db6e0b7 | 2889 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2890 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
2891 | } |
2892 | ||
d20581aa BH |
2893 | /* |
2894 | * Called by non rpc-based layout drivers | |
2895 | */ | |
d45f60c6 | 2896 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 2897 | { |
bfc505de | 2898 | if (likely(!hdr->pnfs_error)) |
d45f60c6 | 2899 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2900 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
2901 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2902 | pnfs_ld_handle_read_error(hdr); |
2903 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
2904 | } |
2905 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
2906 | ||
493292dd TM |
2907 | static void |
2908 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2909 | struct nfs_pgio_header *hdr) |
493292dd | 2910 | { |
48d635f1 | 2911 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2912 | |
4db6e0b7 | 2913 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2914 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 2915 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 2916 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 2917 | } |
ba4a76f7 | 2918 | hdr->completion_ops->completion(hdr); |
493292dd TM |
2919 | } |
2920 | ||
64419a9b AA |
2921 | /* |
2922 | * Call the appropriate parallel I/O subsystem read function. | |
2923 | */ | |
493292dd | 2924 | static enum pnfs_try_status |
d45f60c6 | 2925 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
2926 | const struct rpc_call_ops *call_ops, |
2927 | struct pnfs_layout_segment *lseg) | |
64419a9b | 2928 | { |
cd841605 | 2929 | struct inode *inode = hdr->inode; |
64419a9b AA |
2930 | struct nfs_server *nfss = NFS_SERVER(inode); |
2931 | enum pnfs_try_status trypnfs; | |
2932 | ||
cd841605 | 2933 | hdr->mds_ops = call_ops; |
64419a9b AA |
2934 | |
2935 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 2936 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 2937 | |
d45f60c6 | 2938 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 2939 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 2940 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
2941 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2942 | return trypnfs; | |
2943 | } | |
863a3c6c | 2944 | |
ceb11e13 | 2945 | /* Resend all requests through pnfs. */ |
1b1bc66b | 2946 | void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr) |
ceb11e13 PT |
2947 | { |
2948 | struct nfs_pageio_descriptor pgio; | |
2949 | ||
1b1bc66b | 2950 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
54e4a0df TM |
2951 | /* Prevent deadlocks with layoutreturn! */ |
2952 | pnfs_put_lseg(hdr->lseg); | |
2953 | hdr->lseg = NULL; | |
2954 | ||
1b1bc66b WAA |
2955 | nfs_pageio_init_read(&pgio, hdr->inode, false, |
2956 | hdr->completion_ops); | |
2957 | hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr); | |
2958 | } | |
ceb11e13 PT |
2959 | } |
2960 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
2961 | ||
493292dd | 2962 | static void |
7f714720 | 2963 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 2964 | { |
493292dd TM |
2965 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2966 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2967 | enum pnfs_try_status trypnfs; |
493292dd | 2968 | |
d45f60c6 | 2969 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
6aeafd05 TM |
2970 | switch (trypnfs) { |
2971 | case PNFS_NOT_ATTEMPTED: | |
d45f60c6 | 2972 | pnfs_read_through_mds(desc, hdr); |
6aeafd05 TM |
2973 | case PNFS_ATTEMPTED: |
2974 | break; | |
2975 | case PNFS_TRY_AGAIN: | |
2976 | /* cleanup hdr and prepare to redo pnfs */ | |
2977 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { | |
2978 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); | |
2979 | list_splice_init(&hdr->pages, &mirror->pg_list); | |
2980 | mirror->pg_recoalesce = 1; | |
2981 | } | |
2982 | hdr->mds_ops->rpc_release(hdr); | |
2983 | } | |
493292dd TM |
2984 | } |
2985 | ||
4db6e0b7 FI |
2986 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
2987 | { | |
9369a431 | 2988 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2989 | nfs_pgio_header_free(hdr); |
4db6e0b7 FI |
2990 | } |
2991 | ||
493292dd TM |
2992 | int |
2993 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
2994 | { | |
4db6e0b7 | 2995 | struct nfs_pgio_header *hdr; |
493292dd TM |
2996 | int ret; |
2997 | ||
1e7f3a48 WAA |
2998 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2999 | if (!hdr) { | |
2bff2288 PT |
3000 | desc->pg_error = -ENOMEM; |
3001 | return desc->pg_error; | |
493292dd | 3002 | } |
4db6e0b7 | 3003 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 3004 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 3005 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 3006 | if (!ret) |
7f714720 | 3007 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 3008 | return ret; |
493292dd TM |
3009 | } |
3010 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
3011 | ||
71244d9b TM |
3012 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
3013 | { | |
3014 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
3015 | ||
3016 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 3017 | smp_mb__after_atomic(); |
71244d9b TM |
3018 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
3019 | } | |
3020 | ||
863a3c6c | 3021 | /* |
a9bae566 | 3022 | * There can be multiple RW segments. |
863a3c6c | 3023 | */ |
a9bae566 | 3024 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 3025 | { |
a9bae566 | 3026 | struct pnfs_layout_segment *lseg; |
863a3c6c | 3027 | |
a9bae566 PT |
3028 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
3029 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 3030 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
3031 | list_add(&lseg->pls_lc_list, listp); |
3032 | } | |
863a3c6c AA |
3033 | } |
3034 | ||
a073dbff TM |
3035 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
3036 | { | |
3037 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
3038 | |
3039 | /* Matched by references in pnfs_set_layoutcommit */ | |
3040 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
3041 | list_del_init(&lseg->pls_lc_list); | |
3042 | pnfs_put_lseg(lseg); | |
3043 | } | |
3044 | ||
71244d9b | 3045 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
3046 | } |
3047 | ||
1b0ae068 PT |
3048 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
3049 | { | |
b9e028fd | 3050 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
3051 | } |
3052 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
3053 | ||
863a3c6c | 3054 | void |
67af7611 TM |
3055 | pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, |
3056 | loff_t end_pos) | |
863a3c6c | 3057 | { |
cd841605 | 3058 | struct nfs_inode *nfsi = NFS_I(inode); |
79a48a1f | 3059 | bool mark_as_dirty = false; |
863a3c6c | 3060 | |
cd841605 | 3061 | spin_lock(&inode->i_lock); |
863a3c6c | 3062 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
29559b11 | 3063 | nfsi->layout->plh_lwb = end_pos; |
79a48a1f | 3064 | mark_as_dirty = true; |
863a3c6c | 3065 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 3066 | __func__, inode->i_ino); |
29559b11 TM |
3067 | } else if (end_pos > nfsi->layout->plh_lwb) |
3068 | nfsi->layout->plh_lwb = end_pos; | |
67af7611 | 3069 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) { |
a9bae566 | 3070 | /* references matched in nfs4_layoutcommit_release */ |
67af7611 | 3071 | pnfs_get_lseg(lseg); |
a9bae566 | 3072 | } |
cd841605 | 3073 | spin_unlock(&inode->i_lock); |
acff5880 | 3074 | dprintk("%s: lseg %p end_pos %llu\n", |
67af7611 | 3075 | __func__, lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
3076 | |
3077 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
3078 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
3079 | if (mark_as_dirty) | |
cd841605 | 3080 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
3081 | } |
3082 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
3083 | ||
db29c089 AA |
3084 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
3085 | { | |
3086 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
3087 | ||
3088 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
3089 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 3090 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
3091 | } |
3092 | ||
de4b15c7 AA |
3093 | /* |
3094 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
3095 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
3096 | * data to disk to allow the server to recover the data if it crashes. | |
3097 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
3098 | * is off, and a COMMIT is sent to a data server, or | |
3099 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
3100 | */ | |
863a3c6c | 3101 | int |
ef311537 | 3102 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 3103 | { |
5f919c9f | 3104 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
3105 | struct nfs4_layoutcommit_data *data; |
3106 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 3107 | loff_t end_pos; |
71244d9b | 3108 | int status; |
863a3c6c | 3109 | |
71244d9b | 3110 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
3111 | return 0; |
3112 | ||
71244d9b | 3113 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 3114 | |
71244d9b | 3115 | status = -EAGAIN; |
92407e75 | 3116 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
3117 | if (!sync) |
3118 | goto out; | |
74316201 | 3119 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
3120 | NFS_INO_LAYOUTCOMMITTING, |
3121 | nfs_wait_bit_killable, | |
3122 | TASK_KILLABLE); | |
92407e75 | 3123 | if (status) |
71244d9b | 3124 | goto out; |
92407e75 PT |
3125 | } |
3126 | ||
71244d9b TM |
3127 | status = -ENOMEM; |
3128 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
3129 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
3130 | if (!data) | |
3131 | goto clear_layoutcommitting; | |
3132 | ||
3133 | status = 0; | |
de4b15c7 | 3134 | spin_lock(&inode->i_lock); |
71244d9b TM |
3135 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
3136 | goto out_unlock; | |
a9bae566 | 3137 | |
71244d9b | 3138 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 3139 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 3140 | |
acff5880 | 3141 | end_pos = nfsi->layout->plh_lwb; |
863a3c6c | 3142 | |
f597c537 | 3143 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
97a728f5 | 3144 | data->cred = get_cred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
3145 | spin_unlock(&inode->i_lock); |
3146 | ||
3147 | data->args.inode = inode; | |
863a3c6c AA |
3148 | nfs_fattr_init(&data->fattr); |
3149 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
3150 | data->res.fattr = &data->fattr; | |
2e18d4d8 TM |
3151 | if (end_pos != 0) |
3152 | data->args.lastbytewritten = end_pos - 1; | |
3153 | else | |
3154 | data->args.lastbytewritten = U64_MAX; | |
863a3c6c AA |
3155 | data->res.server = NFS_SERVER(inode); |
3156 | ||
5f919c9f CH |
3157 | if (ld->prepare_layoutcommit) { |
3158 | status = ld->prepare_layoutcommit(&data->args); | |
3159 | if (status) { | |
a52458b4 | 3160 | put_cred(data->cred); |
5f919c9f | 3161 | spin_lock(&inode->i_lock); |
29559b11 TM |
3162 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); |
3163 | if (end_pos > nfsi->layout->plh_lwb) | |
5f919c9f | 3164 | nfsi->layout->plh_lwb = end_pos; |
3471648a | 3165 | goto out_unlock; |
5f919c9f CH |
3166 | } |
3167 | } | |
3168 | ||
3169 | ||
863a3c6c AA |
3170 | status = nfs4_proc_layoutcommit(data, sync); |
3171 | out: | |
92407e75 PT |
3172 | if (status) |
3173 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
3174 | dprintk("<-- %s status %d\n", __func__, status); |
3175 | return status; | |
71244d9b TM |
3176 | out_unlock: |
3177 | spin_unlock(&inode->i_lock); | |
92407e75 | 3178 | kfree(data); |
71244d9b TM |
3179 | clear_layoutcommitting: |
3180 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 3181 | goto out; |
863a3c6c | 3182 | } |
72cff449 | 3183 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a | 3184 | |
5bb89b47 TM |
3185 | int |
3186 | pnfs_generic_sync(struct inode *inode, bool datasync) | |
3187 | { | |
3188 | return pnfs_layoutcommit_inode(inode, true); | |
3189 | } | |
3190 | EXPORT_SYMBOL_GPL(pnfs_generic_sync); | |
3191 | ||
82be417a AA |
3192 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) |
3193 | { | |
3194 | struct nfs4_threshold *thp; | |
3195 | ||
3196 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
3197 | if (!thp) { | |
3198 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
3199 | return NULL; | |
3200 | } | |
3201 | return thp; | |
3202 | } | |
8733408d | 3203 | |
865a7ecb | 3204 | #if IS_ENABLED(CONFIG_NFS_V4_2) |
8733408d | 3205 | int |
c8ad8894 | 3206 | pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags) |
8733408d PT |
3207 | { |
3208 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
3209 | struct nfs_server *server = NFS_SERVER(inode); | |
1bfe3b25 | 3210 | struct nfs_inode *nfsi = NFS_I(inode); |
8733408d PT |
3211 | struct nfs42_layoutstat_data *data; |
3212 | struct pnfs_layout_hdr *hdr; | |
3213 | int status = 0; | |
3214 | ||
3215 | if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats) | |
3216 | goto out; | |
3217 | ||
6c5a0d89 TM |
3218 | if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS)) |
3219 | goto out; | |
3220 | ||
1bfe3b25 PT |
3221 | if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags)) |
3222 | goto out; | |
3223 | ||
8733408d PT |
3224 | spin_lock(&inode->i_lock); |
3225 | if (!NFS_I(inode)->layout) { | |
3226 | spin_unlock(&inode->i_lock); | |
f538d0ba | 3227 | goto out_clear_layoutstats; |
8733408d PT |
3228 | } |
3229 | hdr = NFS_I(inode)->layout; | |
3230 | pnfs_get_layout_hdr(hdr); | |
3231 | spin_unlock(&inode->i_lock); | |
3232 | ||
c8ad8894 | 3233 | data = kzalloc(sizeof(*data), gfp_flags); |
8733408d PT |
3234 | if (!data) { |
3235 | status = -ENOMEM; | |
3236 | goto out_put; | |
3237 | } | |
3238 | ||
3239 | data->args.fh = NFS_FH(inode); | |
3240 | data->args.inode = inode; | |
8733408d PT |
3241 | status = ld->prepare_layoutstats(&data->args); |
3242 | if (status) | |
3243 | goto out_free; | |
3244 | ||
3245 | status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data); | |
3246 | ||
3247 | out: | |
3248 | dprintk("%s returns %d\n", __func__, status); | |
3249 | return status; | |
3250 | ||
3251 | out_free: | |
3252 | kfree(data); | |
3253 | out_put: | |
3254 | pnfs_put_layout_hdr(hdr); | |
f538d0ba | 3255 | out_clear_layoutstats: |
1bfe3b25 PT |
3256 | smp_mb__before_atomic(); |
3257 | clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags); | |
3258 | smp_mb__after_atomic(); | |
8733408d PT |
3259 | goto out; |
3260 | } | |
3261 | EXPORT_SYMBOL_GPL(pnfs_report_layoutstat); | |
865a7ecb | 3262 | #endif |
bbf58bf3 TM |
3263 | |
3264 | unsigned int layoutstats_timer; | |
3265 | module_param(layoutstats_timer, uint, 0644); | |
3266 | EXPORT_SYMBOL_GPL(layoutstats_timer); |