Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
6a5b99a4 | 18 | * http://www.gnu.org/licenses/gpl-2.0.html |
d7e09d03 | 19 | * |
d7e09d03 PT |
20 | * GPL HEADER END |
21 | */ | |
22 | /* | |
23 | * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Use is subject to license terms. | |
25 | * | |
1dc563a6 | 26 | * Copyright (c) 2011, 2015, Intel Corporation. |
d7e09d03 PT |
27 | */ |
28 | /* | |
29 | * This file is part of Lustre, http://www.lustre.org/ | |
30 | * Lustre is a trademark of Sun Microsystems, Inc. | |
31 | * | |
32 | * lustre/llite/file.c | |
33 | * | |
34 | * Author: Peter Braam <braam@clusterfs.com> | |
35 | * Author: Phil Schwan <phil@clusterfs.com> | |
36 | * Author: Andreas Dilger <adilger@clusterfs.com> | |
37 | */ | |
38 | ||
39 | #define DEBUG_SUBSYSTEM S_LLITE | |
67a235f5 GKH |
40 | #include "../include/lustre_dlm.h" |
41 | #include "../include/lustre_lite.h" | |
d7e09d03 PT |
42 | #include <linux/pagemap.h> |
43 | #include <linux/file.h> | |
bb41292b | 44 | #include <linux/mount.h> |
d7e09d03 | 45 | #include "llite_internal.h" |
67a235f5 | 46 | #include "../include/lustre/ll_fiemap.h" |
d7e09d03 | 47 | |
67a235f5 | 48 | #include "../include/cl_object.h" |
d7e09d03 | 49 | |
2d95f10e JH |
50 | static int |
51 | ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg); | |
52 | ||
53 | static int ll_lease_close(struct obd_client_handle *och, struct inode *inode, | |
54 | bool *lease_broken); | |
55 | ||
56 | static enum llioc_iter | |
57 | ll_iocontrol_call(struct inode *inode, struct file *file, | |
58 | unsigned int cmd, unsigned long arg, int *rcp); | |
59 | ||
60 | static struct ll_file_data *ll_file_data_get(void) | |
d7e09d03 PT |
61 | { |
62 | struct ll_file_data *fd; | |
63 | ||
21068c46 | 64 | fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS); |
6e16818b | 65 | if (!fd) |
73863d83 | 66 | return NULL; |
d7e09d03 PT |
67 | fd->fd_write_failed = false; |
68 | return fd; | |
69 | } | |
70 | ||
71 | static void ll_file_data_put(struct ll_file_data *fd) | |
72 | { | |
6e16818b | 73 | if (fd) |
50d30362 | 74 | kmem_cache_free(ll_file_data_slab, fd); |
d7e09d03 PT |
75 | } |
76 | ||
77 | void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data, | |
78 | struct lustre_handle *fh) | |
79 | { | |
80 | op_data->op_fid1 = ll_i2info(inode)->lli_fid; | |
81 | op_data->op_attr.ia_mode = inode->i_mode; | |
82 | op_data->op_attr.ia_atime = inode->i_atime; | |
83 | op_data->op_attr.ia_mtime = inode->i_mtime; | |
84 | op_data->op_attr.ia_ctime = inode->i_ctime; | |
85 | op_data->op_attr.ia_size = i_size_read(inode); | |
86 | op_data->op_attr_blocks = inode->i_blocks; | |
bb41292b | 87 | op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags); |
d7e09d03 PT |
88 | op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch; |
89 | if (fh) | |
90 | op_data->op_handle = *fh; | |
d7e09d03 | 91 | |
1f6eaf83 | 92 | if (ll_i2info(inode)->lli_flags & LLIF_DATA_MODIFIED) |
d7e09d03 PT |
93 | op_data->op_bias |= MDS_DATA_MODIFIED; |
94 | } | |
95 | ||
96 | /** | |
97 | * Closes the IO epoch and packs all the attributes into @op_data for | |
98 | * the CLOSE rpc. | |
99 | */ | |
100 | static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data, | |
101 | struct obd_client_handle *och) | |
102 | { | |
f57d9a72 EL |
103 | op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET | |
104 | ATTR_MTIME | ATTR_MTIME_SET | | |
105 | ATTR_CTIME | ATTR_CTIME_SET; | |
d7e09d03 PT |
106 | |
107 | if (!(och->och_flags & FMODE_WRITE)) | |
108 | goto out; | |
109 | ||
110 | if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode)) | |
111 | op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS; | |
112 | else | |
113 | ll_ioepoch_close(inode, op_data, &och, 0); | |
114 | ||
115 | out: | |
116 | ll_pack_inode2opdata(inode, op_data, &och->och_fh); | |
117 | ll_prep_md_op_data(op_data, inode, NULL, NULL, | |
118 | 0, 0, LUSTRE_OPC_ANY, NULL); | |
d7e09d03 PT |
119 | } |
120 | ||
121 | static int ll_close_inode_openhandle(struct obd_export *md_exp, | |
122 | struct inode *inode, | |
48d23e61 JX |
123 | struct obd_client_handle *och, |
124 | const __u64 *data_version) | |
d7e09d03 PT |
125 | { |
126 | struct obd_export *exp = ll_i2mdexp(inode); | |
127 | struct md_op_data *op_data; | |
128 | struct ptlrpc_request *req = NULL; | |
129 | struct obd_device *obd = class_exp2obd(exp); | |
130 | int epoch_close = 1; | |
131 | int rc; | |
d7e09d03 | 132 | |
6e16818b | 133 | if (!obd) { |
d7e09d03 PT |
134 | /* |
135 | * XXX: in case of LMV, is this correct to access | |
136 | * ->exp_handle? | |
137 | */ | |
55f5a824 | 138 | CERROR("Invalid MDC connection handle %#llx\n", |
d7e09d03 | 139 | ll_i2mdexp(inode)->exp_handle.h_cookie); |
34e1f2bb JL |
140 | rc = 0; |
141 | goto out; | |
d7e09d03 PT |
142 | } |
143 | ||
496a51bd JL |
144 | op_data = kzalloc(sizeof(*op_data), GFP_NOFS); |
145 | if (!op_data) { | |
34e1f2bb JL |
146 | /* XXX We leak openhandle and request here. */ |
147 | rc = -ENOMEM; | |
148 | goto out; | |
149 | } | |
d7e09d03 PT |
150 | |
151 | ll_prepare_close(inode, op_data, och); | |
6e16818b | 152 | if (data_version) { |
48d23e61 JX |
153 | /* Pass in data_version implies release. */ |
154 | op_data->op_bias |= MDS_HSM_RELEASE; | |
155 | op_data->op_data_version = *data_version; | |
156 | op_data->op_lease_handle = och->och_lease_handle; | |
157 | op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS; | |
158 | } | |
b6ee3824 | 159 | epoch_close = op_data->op_flags & MF_EPOCH_CLOSE; |
d7e09d03 PT |
160 | rc = md_close(md_exp, op_data, och->och_mod, &req); |
161 | if (rc == -EAGAIN) { | |
162 | /* This close must have the epoch closed. */ | |
163 | LASSERT(epoch_close); | |
164 | /* MDS has instructed us to obtain Size-on-MDS attribute from | |
c0894c6c OD |
165 | * OSTs and send setattr to back to MDS. |
166 | */ | |
d7e09d03 PT |
167 | rc = ll_som_update(inode, op_data); |
168 | if (rc) { | |
97a075cd JN |
169 | CERROR("%s: inode "DFID" mdc Size-on-MDS update failed: rc = %d\n", |
170 | ll_i2mdexp(inode)->exp_obd->obd_name, | |
171 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
172 | rc = 0; |
173 | } | |
174 | } else if (rc) { | |
97a075cd JN |
175 | CERROR("%s: inode "DFID" mdc close failed: rc = %d\n", |
176 | ll_i2mdexp(inode)->exp_obd->obd_name, | |
177 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
178 | } |
179 | ||
180 | /* DATA_MODIFIED flag was successfully sent on close, cancel data | |
c0894c6c OD |
181 | * modification flag. |
182 | */ | |
d7e09d03 PT |
183 | if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) { |
184 | struct ll_inode_info *lli = ll_i2info(inode); | |
185 | ||
186 | spin_lock(&lli->lli_lock); | |
187 | lli->lli_flags &= ~LLIF_DATA_MODIFIED; | |
188 | spin_unlock(&lli->lli_lock); | |
189 | } | |
190 | ||
d7e09d03 PT |
191 | if (rc == 0) { |
192 | rc = ll_objects_destroy(req, inode); | |
193 | if (rc) | |
194 | CERROR("inode %lu ll_objects destroy: rc = %d\n", | |
195 | inode->i_ino, rc); | |
196 | } | |
48d23e61 JX |
197 | if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) { |
198 | struct mdt_body *body; | |
cea812cd | 199 | |
48d23e61 JX |
200 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); |
201 | if (!(body->valid & OBD_MD_FLRELEASED)) | |
202 | rc = -EBUSY; | |
203 | } | |
204 | ||
205 | ll_finish_md_op_data(op_data); | |
d7e09d03 | 206 | |
d7e09d03 | 207 | out: |
d7e09d03 PT |
208 | if (exp_connect_som(exp) && !epoch_close && |
209 | S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) { | |
210 | ll_queue_done_writing(inode, LLIF_DONE_WRITING); | |
211 | } else { | |
212 | md_clear_open_replay_data(md_exp, och); | |
213 | /* Free @och if it is not waiting for DONE_WRITING. */ | |
214 | och->och_fh.cookie = DEAD_HANDLE_MAGIC; | |
97903a26 | 215 | kfree(och); |
d7e09d03 PT |
216 | } |
217 | if (req) /* This is close request */ | |
218 | ptlrpc_req_finished(req); | |
219 | return rc; | |
220 | } | |
221 | ||
45b2a010 | 222 | int ll_md_real_close(struct inode *inode, fmode_t fmode) |
d7e09d03 PT |
223 | { |
224 | struct ll_inode_info *lli = ll_i2info(inode); | |
225 | struct obd_client_handle **och_p; | |
226 | struct obd_client_handle *och; | |
227 | __u64 *och_usecount; | |
228 | int rc = 0; | |
d7e09d03 | 229 | |
45b2a010 | 230 | if (fmode & FMODE_WRITE) { |
d7e09d03 PT |
231 | och_p = &lli->lli_mds_write_och; |
232 | och_usecount = &lli->lli_open_fd_write_count; | |
45b2a010 | 233 | } else if (fmode & FMODE_EXEC) { |
d7e09d03 PT |
234 | och_p = &lli->lli_mds_exec_och; |
235 | och_usecount = &lli->lli_open_fd_exec_count; | |
236 | } else { | |
45b2a010 | 237 | LASSERT(fmode & FMODE_READ); |
d7e09d03 PT |
238 | och_p = &lli->lli_mds_read_och; |
239 | och_usecount = &lli->lli_open_fd_read_count; | |
240 | } | |
241 | ||
242 | mutex_lock(&lli->lli_och_mutex); | |
45b2a010 JH |
243 | if (*och_usecount > 0) { |
244 | /* There are still users of this handle, so skip | |
c0894c6c OD |
245 | * freeing it. |
246 | */ | |
d7e09d03 | 247 | mutex_unlock(&lli->lli_och_mutex); |
0a3bdb00 | 248 | return 0; |
d7e09d03 | 249 | } |
45b2a010 | 250 | |
57303e76 | 251 | och = *och_p; |
d7e09d03 PT |
252 | *och_p = NULL; |
253 | mutex_unlock(&lli->lli_och_mutex); | |
254 | ||
6e16818b | 255 | if (och) { |
45b2a010 | 256 | /* There might be a race and this handle may already |
c0894c6c OD |
257 | * be closed. |
258 | */ | |
d7e09d03 | 259 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, |
48d23e61 | 260 | inode, och, NULL); |
d7e09d03 PT |
261 | } |
262 | ||
0a3bdb00 | 263 | return rc; |
d7e09d03 PT |
264 | } |
265 | ||
2d95f10e JH |
266 | static int ll_md_close(struct obd_export *md_exp, struct inode *inode, |
267 | struct file *file) | |
d7e09d03 PT |
268 | { |
269 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
270 | struct ll_inode_info *lli = ll_i2info(inode); | |
74d01958 AV |
271 | int lockmode; |
272 | __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK; | |
273 | struct lustre_handle lockh; | |
8369cfff | 274 | ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_OPEN} }; |
d7e09d03 | 275 | int rc = 0; |
d7e09d03 PT |
276 | |
277 | /* clear group lock, if present */ | |
278 | if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED)) | |
98eae5e7 | 279 | ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid); |
d7e09d03 | 280 | |
6e16818b | 281 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
282 | bool lease_broken; |
283 | ||
284 | /* Usually the lease is not released when the | |
c0894c6c OD |
285 | * application crashed, we need to release here. |
286 | */ | |
d3a8a4e2 | 287 | rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken); |
e15ba45d OD |
288 | CDEBUG(rc ? D_ERROR : D_INODE, |
289 | "Clean up lease " DFID " %d/%d\n", | |
290 | PFID(&lli->lli_fid), rc, lease_broken); | |
d3a8a4e2 JX |
291 | |
292 | fd->fd_lease_och = NULL; | |
293 | } | |
294 | ||
6e16818b | 295 | if (fd->fd_och) { |
48d23e61 | 296 | rc = ll_close_inode_openhandle(md_exp, inode, fd->fd_och, NULL); |
d3a8a4e2 | 297 | fd->fd_och = NULL; |
34e1f2bb | 298 | goto out; |
d3a8a4e2 JX |
299 | } |
300 | ||
d7e09d03 | 301 | /* Let's see if we have good enough OPEN lock on the file and if |
c0894c6c OD |
302 | * we can skip talking to MDS |
303 | */ | |
d7e09d03 | 304 | |
74d01958 AV |
305 | mutex_lock(&lli->lli_och_mutex); |
306 | if (fd->fd_omode & FMODE_WRITE) { | |
307 | lockmode = LCK_CW; | |
308 | LASSERT(lli->lli_open_fd_write_count); | |
309 | lli->lli_open_fd_write_count--; | |
310 | } else if (fd->fd_omode & FMODE_EXEC) { | |
311 | lockmode = LCK_PR; | |
312 | LASSERT(lli->lli_open_fd_exec_count); | |
313 | lli->lli_open_fd_exec_count--; | |
d7e09d03 | 314 | } else { |
74d01958 AV |
315 | lockmode = LCK_CR; |
316 | LASSERT(lli->lli_open_fd_read_count); | |
317 | lli->lli_open_fd_read_count--; | |
d7e09d03 | 318 | } |
74d01958 AV |
319 | mutex_unlock(&lli->lli_och_mutex); |
320 | ||
321 | if (!md_lock_match(md_exp, flags, ll_inode2fid(inode), | |
322 | LDLM_IBITS, &policy, lockmode, &lockh)) | |
323 | rc = ll_md_real_close(inode, fd->fd_omode); | |
d7e09d03 | 324 | |
d3a8a4e2 | 325 | out: |
d7e09d03 PT |
326 | LUSTRE_FPRIVATE(file) = NULL; |
327 | ll_file_data_put(fd); | |
d7e09d03 | 328 | |
0a3bdb00 | 329 | return rc; |
d7e09d03 PT |
330 | } |
331 | ||
332 | /* While this returns an error code, fput() the caller does not, so we need | |
333 | * to make every effort to clean up all of our state here. Also, applications | |
334 | * rarely check close errors and even if an error is returned they will not | |
335 | * re-try the close call. | |
336 | */ | |
337 | int ll_file_release(struct inode *inode, struct file *file) | |
338 | { | |
339 | struct ll_file_data *fd; | |
340 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
341 | struct ll_inode_info *lli = ll_i2info(inode); | |
342 | int rc; | |
d7e09d03 | 343 | |
97a075cd JN |
344 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n", |
345 | PFID(ll_inode2fid(inode)), inode); | |
d7e09d03 | 346 | |
f76c23da | 347 | if (!is_root_inode(inode)) |
d7e09d03 PT |
348 | ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1); |
349 | fd = LUSTRE_FPRIVATE(file); | |
6e16818b | 350 | LASSERT(fd); |
d7e09d03 | 351 | |
c0894c6c | 352 | /* The last ref on @file, maybe not be the owner pid of statahead. |
d7e09d03 | 353 | * Different processes can open the same dir, "ll_opendir_key" means: |
c0894c6c OD |
354 | * it is me that should stop the statahead thread. |
355 | */ | |
d7e09d03 PT |
356 | if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd && |
357 | lli->lli_opendir_pid != 0) | |
358 | ll_stop_statahead(inode, lli->lli_opendir_key); | |
359 | ||
f76c23da | 360 | if (is_root_inode(inode)) { |
d7e09d03 PT |
361 | LUSTRE_FPRIVATE(file) = NULL; |
362 | ll_file_data_put(fd); | |
0a3bdb00 | 363 | return 0; |
d7e09d03 PT |
364 | } |
365 | ||
366 | if (!S_ISDIR(inode->i_mode)) { | |
367 | lov_read_and_clear_async_rc(lli->lli_clob); | |
368 | lli->lli_async_rc = 0; | |
369 | } | |
370 | ||
371 | rc = ll_md_close(sbi->ll_md_exp, inode, file); | |
372 | ||
373 | if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val)) | |
374 | libcfs_debug_dumplog(); | |
375 | ||
0a3bdb00 | 376 | return rc; |
d7e09d03 PT |
377 | } |
378 | ||
48eddfd5 | 379 | static int ll_intent_file_open(struct dentry *dentry, void *lmm, |
d7e09d03 PT |
380 | int lmmsize, struct lookup_intent *itp) |
381 | { | |
2b0143b5 | 382 | struct inode *inode = d_inode(dentry); |
48eddfd5 AV |
383 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
384 | struct dentry *parent = dentry->d_parent; | |
385 | const char *name = dentry->d_name.name; | |
386 | const int len = dentry->d_name.len; | |
d7e09d03 PT |
387 | struct md_op_data *op_data; |
388 | struct ptlrpc_request *req; | |
389 | __u32 opc = LUSTRE_OPC_ANY; | |
390 | int rc; | |
d7e09d03 | 391 | |
c0894c6c | 392 | /* Usually we come here only for NFSD, and we want open lock. */ |
d7e09d03 PT |
393 | /* We can also get here if there was cached open handle in revalidate_it |
394 | * but it disappeared while we were getting from there to ll_file_open. | |
bef31c78 | 395 | * But this means this file was closed and immediately opened which |
c0894c6c OD |
396 | * makes a good candidate for using OPEN lock |
397 | */ | |
d7e09d03 | 398 | /* If lmmsize & lmm are not 0, we are just setting stripe info |
c0894c6c OD |
399 | * parameters. No need for the open lock |
400 | */ | |
6e16818b | 401 | if (!lmm && lmmsize == 0) { |
6dad4d89 OD |
402 | struct ll_dentry_data *ldd = ll_d2d(dentry); |
403 | /* | |
404 | * If we came via ll_iget_for_nfs, then we need to request | |
405 | * struct ll_dentry_data *ldd = ll_d2d(file->f_dentry); | |
406 | * | |
407 | * NB: when ldd is NULL, it must have come via normal | |
408 | * lookup path only, since ll_iget_for_nfs always calls | |
409 | * ll_d_init(). | |
410 | */ | |
411 | if (ldd && ldd->lld_nfs_dentry) { | |
412 | ldd->lld_nfs_dentry = 0; | |
413 | itp->it_flags |= MDS_OPEN_LOCK; | |
414 | } | |
d7e09d03 PT |
415 | if (itp->it_flags & FMODE_WRITE) |
416 | opc = LUSTRE_OPC_CREATE; | |
417 | } | |
418 | ||
2b0143b5 | 419 | op_data = ll_prep_md_op_data(NULL, d_inode(parent), |
48eddfd5 | 420 | inode, name, len, |
d7e09d03 PT |
421 | O_RDWR, opc, NULL); |
422 | if (IS_ERR(op_data)) | |
0a3bdb00 | 423 | return PTR_ERR(op_data); |
d7e09d03 PT |
424 | |
425 | itp->it_flags |= MDS_OPEN_BY_FID; | |
426 | rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp, | |
427 | 0 /*unused */, &req, ll_md_blocking_ast, 0); | |
428 | ll_finish_md_op_data(op_data); | |
429 | if (rc == -ESTALE) { | |
430 | /* reason for keep own exit path - don`t flood log | |
431 | * with messages with -ESTALE errors. | |
432 | */ | |
433 | if (!it_disposition(itp, DISP_OPEN_OPEN) || | |
e15ba45d | 434 | it_open_error(DISP_OPEN_OPEN, itp)) |
34e1f2bb | 435 | goto out; |
e22fdcc8 | 436 | ll_release_openhandle(inode, itp); |
34e1f2bb | 437 | goto out; |
d7e09d03 PT |
438 | } |
439 | ||
34e1f2bb JL |
440 | if (it_disposition(itp, DISP_LOOKUP_NEG)) { |
441 | rc = -ENOENT; | |
442 | goto out; | |
443 | } | |
d7e09d03 PT |
444 | |
445 | if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) { | |
446 | rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp); | |
447 | CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc); | |
34e1f2bb | 448 | goto out; |
d7e09d03 PT |
449 | } |
450 | ||
48eddfd5 | 451 | rc = ll_prep_inode(&inode, req, NULL, itp); |
e476f2e5 | 452 | if (!rc && itp->it_lock_mode) |
48eddfd5 | 453 | ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL); |
d7e09d03 PT |
454 | |
455 | out: | |
f236f69b | 456 | ptlrpc_req_finished(req); |
d7e09d03 PT |
457 | ll_intent_drop_lock(itp); |
458 | ||
0a3bdb00 | 459 | return rc; |
d7e09d03 PT |
460 | } |
461 | ||
462 | /** | |
463 | * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does | |
464 | * not believe attributes if a few ioepoch holders exist. Attributes for | |
465 | * previous ioepoch if new one is opened are also skipped by MDS. | |
466 | */ | |
467 | void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch) | |
468 | { | |
469 | if (ioepoch && lli->lli_ioepoch != ioepoch) { | |
470 | lli->lli_ioepoch = ioepoch; | |
b0f5aad5 | 471 | CDEBUG(D_INODE, "Epoch %llu opened on "DFID"\n", |
d7e09d03 PT |
472 | ioepoch, PFID(&lli->lli_fid)); |
473 | } | |
474 | } | |
475 | ||
ea1db081 JH |
476 | static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it, |
477 | struct obd_client_handle *och) | |
d7e09d03 | 478 | { |
d7e09d03 PT |
479 | struct mdt_body *body; |
480 | ||
8bf86fd9 | 481 | body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY); |
ea1db081 JH |
482 | och->och_fh = body->handle; |
483 | och->och_fid = body->fid1; | |
e476f2e5 | 484 | och->och_lease_handle.cookie = it->it_lock_handle; |
d7e09d03 | 485 | och->och_magic = OBD_CLIENT_HANDLE_MAGIC; |
d7e09d03 | 486 | och->och_flags = it->it_flags; |
d7e09d03 | 487 | |
63d42578 | 488 | return md_set_open_replay_data(md_exp, och, it); |
d7e09d03 PT |
489 | } |
490 | ||
2d95f10e JH |
491 | static int ll_local_open(struct file *file, struct lookup_intent *it, |
492 | struct ll_file_data *fd, struct obd_client_handle *och) | |
d7e09d03 | 493 | { |
2a8a3597 | 494 | struct inode *inode = file_inode(file); |
d7e09d03 | 495 | struct ll_inode_info *lli = ll_i2info(inode); |
d7e09d03 PT |
496 | |
497 | LASSERT(!LUSTRE_FPRIVATE(file)); | |
498 | ||
6e16818b | 499 | LASSERT(fd); |
d7e09d03 PT |
500 | |
501 | if (och) { | |
d7e09d03 PT |
502 | struct mdt_body *body; |
503 | int rc; | |
504 | ||
ea1db081 JH |
505 | rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och); |
506 | if (rc != 0) | |
0a3bdb00 | 507 | return rc; |
d7e09d03 | 508 | |
8bf86fd9 JH |
509 | body = req_capsule_server_get(&it->it_request->rq_pill, |
510 | &RMF_MDT_BODY); | |
ea1db081 | 511 | ll_ioepoch_open(lli, body->ioepoch); |
d7e09d03 PT |
512 | } |
513 | ||
514 | LUSTRE_FPRIVATE(file) = fd; | |
515 | ll_readahead_init(inode, &fd->fd_ras); | |
d3a8a4e2 | 516 | fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC); |
966c4a8f JX |
517 | |
518 | /* ll_cl_context initialize */ | |
519 | rwlock_init(&fd->fd_lock); | |
520 | INIT_LIST_HEAD(&fd->fd_lccs); | |
521 | ||
0a3bdb00 | 522 | return 0; |
d7e09d03 PT |
523 | } |
524 | ||
525 | /* Open a file, and (for the very first open) create objects on the OSTs at | |
526 | * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object | |
527 | * creation or open until ll_lov_setstripe() ioctl is called. | |
528 | * | |
529 | * If we already have the stripe MD locally then we don't request it in | |
530 | * md_open(), by passing a lmm_size = 0. | |
531 | * | |
532 | * It is up to the application to ensure no other processes open this file | |
533 | * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be | |
534 | * used. We might be able to avoid races of that sort by getting lli_open_sem | |
535 | * before returning in the O_LOV_DELAY_CREATE case and dropping it here | |
536 | * or in ll_file_release(), but I'm not sure that is desirable/necessary. | |
537 | */ | |
538 | int ll_file_open(struct inode *inode, struct file *file) | |
539 | { | |
540 | struct ll_inode_info *lli = ll_i2info(inode); | |
541 | struct lookup_intent *it, oit = { .it_op = IT_OPEN, | |
542 | .it_flags = file->f_flags }; | |
543 | struct obd_client_handle **och_p = NULL; | |
544 | __u64 *och_usecount = NULL; | |
545 | struct ll_file_data *fd; | |
546 | int rc = 0, opendir_set = 0; | |
d7e09d03 | 547 | |
97a075cd JN |
548 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n", |
549 | PFID(ll_inode2fid(inode)), inode, file->f_flags); | |
d7e09d03 PT |
550 | |
551 | it = file->private_data; /* XXX: compat macro */ | |
552 | file->private_data = NULL; /* prevent ll_local_open assertion */ | |
553 | ||
554 | fd = ll_file_data_get(); | |
6e16818b | 555 | if (!fd) { |
34e1f2bb JL |
556 | rc = -ENOMEM; |
557 | goto out_openerr; | |
558 | } | |
d7e09d03 PT |
559 | |
560 | fd->fd_file = file; | |
561 | if (S_ISDIR(inode->i_mode)) { | |
562 | spin_lock(&lli->lli_sa_lock); | |
6e16818b | 563 | if (!lli->lli_opendir_key && !lli->lli_sai && |
d7e09d03 PT |
564 | lli->lli_opendir_pid == 0) { |
565 | lli->lli_opendir_key = fd; | |
566 | lli->lli_opendir_pid = current_pid(); | |
567 | opendir_set = 1; | |
568 | } | |
569 | spin_unlock(&lli->lli_sa_lock); | |
570 | } | |
571 | ||
f76c23da | 572 | if (is_root_inode(inode)) { |
d7e09d03 | 573 | LUSTRE_FPRIVATE(file) = fd; |
0a3bdb00 | 574 | return 0; |
d7e09d03 PT |
575 | } |
576 | ||
e476f2e5 | 577 | if (!it || !it->it_disposition) { |
d7e09d03 PT |
578 | /* Convert f_flags into access mode. We cannot use file->f_mode, |
579 | * because everything but O_ACCMODE mask was stripped from | |
c0894c6c OD |
580 | * there |
581 | */ | |
d7e09d03 PT |
582 | if ((oit.it_flags + 1) & O_ACCMODE) |
583 | oit.it_flags++; | |
584 | if (file->f_flags & O_TRUNC) | |
585 | oit.it_flags |= FMODE_WRITE; | |
586 | ||
587 | /* kernel only call f_op->open in dentry_open. filp_open calls | |
588 | * dentry_open after call to open_namei that checks permissions. | |
589 | * Only nfsd_open call dentry_open directly without checking | |
c0894c6c OD |
590 | * permissions and because of that this code below is safe. |
591 | */ | |
d7e09d03 PT |
592 | if (oit.it_flags & (FMODE_WRITE | FMODE_READ)) |
593 | oit.it_flags |= MDS_OPEN_OWNEROVERRIDE; | |
594 | ||
595 | /* We do not want O_EXCL here, presumably we opened the file | |
c0894c6c OD |
596 | * already? XXX - NFS implications? |
597 | */ | |
d7e09d03 PT |
598 | oit.it_flags &= ~O_EXCL; |
599 | ||
600 | /* bug20584, if "it_flags" contains O_CREAT, the file will be | |
601 | * created if necessary, then "IT_CREAT" should be set to keep | |
c0894c6c OD |
602 | * consistent with it |
603 | */ | |
d7e09d03 PT |
604 | if (oit.it_flags & O_CREAT) |
605 | oit.it_op |= IT_CREAT; | |
606 | ||
607 | it = &oit; | |
608 | } | |
609 | ||
610 | restart: | |
611 | /* Let's see if we have file open on MDS already. */ | |
612 | if (it->it_flags & FMODE_WRITE) { | |
613 | och_p = &lli->lli_mds_write_och; | |
614 | och_usecount = &lli->lli_open_fd_write_count; | |
615 | } else if (it->it_flags & FMODE_EXEC) { | |
616 | och_p = &lli->lli_mds_exec_och; | |
617 | och_usecount = &lli->lli_open_fd_exec_count; | |
618 | } else { | |
619 | och_p = &lli->lli_mds_read_och; | |
620 | och_usecount = &lli->lli_open_fd_read_count; | |
621 | } | |
622 | ||
623 | mutex_lock(&lli->lli_och_mutex); | |
624 | if (*och_p) { /* Open handle is present */ | |
625 | if (it_disposition(it, DISP_OPEN_OPEN)) { | |
626 | /* Well, there's extra open request that we do not need, | |
c0894c6c OD |
627 | * let's close it somehow. This will decref request. |
628 | */ | |
d7e09d03 PT |
629 | rc = it_open_error(DISP_OPEN_OPEN, it); |
630 | if (rc) { | |
631 | mutex_unlock(&lli->lli_och_mutex); | |
34e1f2bb | 632 | goto out_openerr; |
d7e09d03 PT |
633 | } |
634 | ||
e22fdcc8 | 635 | ll_release_openhandle(inode, it); |
d7e09d03 PT |
636 | } |
637 | (*och_usecount)++; | |
638 | ||
639 | rc = ll_local_open(file, it, fd, NULL); | |
640 | if (rc) { | |
641 | (*och_usecount)--; | |
642 | mutex_unlock(&lli->lli_och_mutex); | |
34e1f2bb | 643 | goto out_openerr; |
d7e09d03 PT |
644 | } |
645 | } else { | |
646 | LASSERT(*och_usecount == 0); | |
e476f2e5 | 647 | if (!it->it_disposition) { |
d7e09d03 | 648 | /* We cannot just request lock handle now, new ELC code |
c0894c6c OD |
649 | * means that one of other OPEN locks for this file |
650 | * could be cancelled, and since blocking ast handler | |
651 | * would attempt to grab och_mutex as well, that would | |
652 | * result in a deadlock | |
653 | */ | |
d7e09d03 PT |
654 | mutex_unlock(&lli->lli_och_mutex); |
655 | it->it_create_mode |= M_CHECK_STALE; | |
48eddfd5 | 656 | rc = ll_intent_file_open(file->f_path.dentry, NULL, 0, it); |
d7e09d03 PT |
657 | it->it_create_mode &= ~M_CHECK_STALE; |
658 | if (rc) | |
34e1f2bb | 659 | goto out_openerr; |
d7e09d03 PT |
660 | |
661 | goto restart; | |
662 | } | |
496a51bd | 663 | *och_p = kzalloc(sizeof(struct obd_client_handle), GFP_NOFS); |
34e1f2bb JL |
664 | if (!*och_p) { |
665 | rc = -ENOMEM; | |
666 | goto out_och_free; | |
667 | } | |
d7e09d03 PT |
668 | |
669 | (*och_usecount)++; | |
670 | ||
671 | /* md_intent_lock() didn't get a request ref if there was an | |
672 | * open error, so don't do cleanup on the request here | |
c0894c6c OD |
673 | * (bug 3430) |
674 | */ | |
d7e09d03 | 675 | /* XXX (green): Should not we bail out on any error here, not |
c0894c6c OD |
676 | * just open error? |
677 | */ | |
d7e09d03 PT |
678 | rc = it_open_error(DISP_OPEN_OPEN, it); |
679 | if (rc) | |
34e1f2bb | 680 | goto out_och_free; |
d7e09d03 | 681 | |
5787be94 AD |
682 | LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF), |
683 | "inode %p: disposition %x, status %d\n", inode, | |
e476f2e5 | 684 | it_disposition(it, ~0), it->it_status); |
d7e09d03 PT |
685 | |
686 | rc = ll_local_open(file, it, fd, *och_p); | |
687 | if (rc) | |
34e1f2bb | 688 | goto out_och_free; |
d7e09d03 PT |
689 | } |
690 | mutex_unlock(&lli->lli_och_mutex); | |
691 | fd = NULL; | |
692 | ||
693 | /* Must do this outside lli_och_mutex lock to prevent deadlock where | |
c0894c6c OD |
694 | * different kind of OPEN lock for this same inode gets cancelled |
695 | * by ldlm_cancel_lru | |
696 | */ | |
d7e09d03 | 697 | if (!S_ISREG(inode->i_mode)) |
34e1f2bb | 698 | goto out_och_free; |
d7e09d03 | 699 | |
38585ccc AD |
700 | if (!lli->lli_has_smd && |
701 | (cl_is_lov_delay_create(file->f_flags) || | |
702 | (file->f_mode & FMODE_WRITE) == 0)) { | |
703 | CDEBUG(D_INODE, "object creation was delayed\n"); | |
34e1f2bb | 704 | goto out_och_free; |
d7e09d03 | 705 | } |
38585ccc | 706 | cl_lov_delay_create_clear(&file->f_flags); |
34e1f2bb | 707 | goto out_och_free; |
d7e09d03 PT |
708 | |
709 | out_och_free: | |
710 | if (rc) { | |
711 | if (och_p && *och_p) { | |
97903a26 | 712 | kfree(*och_p); |
c0a2472f | 713 | *och_p = NULL; |
d7e09d03 PT |
714 | (*och_usecount)--; |
715 | } | |
716 | mutex_unlock(&lli->lli_och_mutex); | |
717 | ||
718 | out_openerr: | |
719 | if (opendir_set != 0) | |
720 | ll_stop_statahead(inode, lli->lli_opendir_key); | |
a5cb8880 | 721 | ll_file_data_put(fd); |
d7e09d03 PT |
722 | } else { |
723 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); | |
724 | } | |
725 | ||
726 | if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) { | |
8bf86fd9 | 727 | ptlrpc_req_finished(it->it_request); |
d7e09d03 PT |
728 | it_clear_disposition(it, DISP_ENQ_OPEN_REF); |
729 | } | |
730 | ||
731 | return rc; | |
732 | } | |
733 | ||
d3a8a4e2 | 734 | static int ll_md_blocking_lease_ast(struct ldlm_lock *lock, |
e15ba45d OD |
735 | struct ldlm_lock_desc *desc, |
736 | void *data, int flag) | |
d3a8a4e2 JX |
737 | { |
738 | int rc; | |
739 | struct lustre_handle lockh; | |
740 | ||
741 | switch (flag) { | |
742 | case LDLM_CB_BLOCKING: | |
743 | ldlm_lock2handle(lock, &lockh); | |
744 | rc = ldlm_cli_cancel(&lockh, LCF_ASYNC); | |
745 | if (rc < 0) { | |
746 | CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc); | |
747 | return rc; | |
748 | } | |
749 | break; | |
750 | case LDLM_CB_CANCELING: | |
751 | /* do nothing */ | |
752 | break; | |
753 | } | |
754 | return 0; | |
755 | } | |
756 | ||
757 | /** | |
758 | * Acquire a lease and open the file. | |
759 | */ | |
2d95f10e JH |
760 | static struct obd_client_handle * |
761 | ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode, | |
762 | __u64 open_flags) | |
d3a8a4e2 JX |
763 | { |
764 | struct lookup_intent it = { .it_op = IT_OPEN }; | |
765 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
766 | struct md_op_data *op_data; | |
767 | struct ptlrpc_request *req; | |
768 | struct lustre_handle old_handle = { 0 }; | |
769 | struct obd_client_handle *och = NULL; | |
770 | int rc; | |
771 | int rc2; | |
772 | ||
773 | if (fmode != FMODE_WRITE && fmode != FMODE_READ) | |
774 | return ERR_PTR(-EINVAL); | |
775 | ||
6e16818b | 776 | if (file) { |
d3a8a4e2 JX |
777 | struct ll_inode_info *lli = ll_i2info(inode); |
778 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
779 | struct obd_client_handle **och_p; | |
780 | __u64 *och_usecount; | |
781 | ||
782 | if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC)) | |
783 | return ERR_PTR(-EPERM); | |
784 | ||
785 | /* Get the openhandle of the file */ | |
786 | rc = -EBUSY; | |
787 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 788 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
789 | mutex_unlock(&lli->lli_och_mutex); |
790 | return ERR_PTR(rc); | |
791 | } | |
792 | ||
6e16818b | 793 | if (!fd->fd_och) { |
d3a8a4e2 | 794 | if (file->f_mode & FMODE_WRITE) { |
6e16818b | 795 | LASSERT(lli->lli_mds_write_och); |
d3a8a4e2 JX |
796 | och_p = &lli->lli_mds_write_och; |
797 | och_usecount = &lli->lli_open_fd_write_count; | |
798 | } else { | |
6e16818b | 799 | LASSERT(lli->lli_mds_read_och); |
d3a8a4e2 JX |
800 | och_p = &lli->lli_mds_read_och; |
801 | och_usecount = &lli->lli_open_fd_read_count; | |
802 | } | |
803 | if (*och_usecount == 1) { | |
804 | fd->fd_och = *och_p; | |
805 | *och_p = NULL; | |
806 | *och_usecount = 0; | |
807 | rc = 0; | |
808 | } | |
809 | } | |
810 | mutex_unlock(&lli->lli_och_mutex); | |
811 | if (rc < 0) /* more than 1 opener */ | |
812 | return ERR_PTR(rc); | |
813 | ||
6e16818b | 814 | LASSERT(fd->fd_och); |
d3a8a4e2 JX |
815 | old_handle = fd->fd_och->och_fh; |
816 | } | |
817 | ||
496a51bd JL |
818 | och = kzalloc(sizeof(*och), GFP_NOFS); |
819 | if (!och) | |
d3a8a4e2 JX |
820 | return ERR_PTR(-ENOMEM); |
821 | ||
822 | op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0, | |
e15ba45d | 823 | LUSTRE_OPC_ANY, NULL); |
34e1f2bb JL |
824 | if (IS_ERR(op_data)) { |
825 | rc = PTR_ERR(op_data); | |
826 | goto out; | |
827 | } | |
d3a8a4e2 JX |
828 | |
829 | /* To tell the MDT this openhandle is from the same owner */ | |
830 | op_data->op_handle = old_handle; | |
831 | ||
48d23e61 JX |
832 | it.it_flags = fmode | open_flags; |
833 | it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE; | |
d3a8a4e2 | 834 | rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, &req, |
e15ba45d | 835 | ll_md_blocking_lease_ast, |
d3a8a4e2 JX |
836 | /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise |
837 | * it can be cancelled which may mislead applications that the lease is | |
838 | * broken; | |
839 | * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal | |
840 | * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast | |
c0894c6c OD |
841 | * doesn't deal with openhandle, so normal openhandle will be leaked. |
842 | */ | |
d3a8a4e2 JX |
843 | LDLM_FL_NO_LRU | LDLM_FL_EXCL); |
844 | ll_finish_md_op_data(op_data); | |
f236f69b | 845 | ptlrpc_req_finished(req); |
d3a8a4e2 | 846 | if (rc < 0) |
34e1f2bb | 847 | goto out_release_it; |
d3a8a4e2 | 848 | |
34e1f2bb JL |
849 | if (it_disposition(&it, DISP_LOOKUP_NEG)) { |
850 | rc = -ENOENT; | |
851 | goto out_release_it; | |
852 | } | |
d3a8a4e2 JX |
853 | |
854 | rc = it_open_error(DISP_OPEN_OPEN, &it); | |
855 | if (rc) | |
34e1f2bb | 856 | goto out_release_it; |
d3a8a4e2 JX |
857 | |
858 | LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF)); | |
859 | ll_och_fill(sbi->ll_md_exp, &it, och); | |
860 | ||
34e1f2bb JL |
861 | if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */ { |
862 | rc = -EOPNOTSUPP; | |
863 | goto out_close; | |
864 | } | |
d3a8a4e2 JX |
865 | |
866 | /* already get lease, handle lease lock */ | |
867 | ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); | |
e476f2e5 JH |
868 | if (it.it_lock_mode == 0 || |
869 | it.it_lock_bits != MDS_INODELOCK_OPEN) { | |
d3a8a4e2 JX |
870 | /* open lock must return for lease */ |
871 | CERROR(DFID "lease granted but no open lock, %d/%llu.\n", | |
e476f2e5 JH |
872 | PFID(ll_inode2fid(inode)), it.it_lock_mode, |
873 | it.it_lock_bits); | |
34e1f2bb JL |
874 | rc = -EPROTO; |
875 | goto out_close; | |
d3a8a4e2 JX |
876 | } |
877 | ||
878 | ll_intent_release(&it); | |
879 | return och; | |
880 | ||
881 | out_close: | |
e55a68b6 | 882 | /* Cancel open lock */ |
e476f2e5 | 883 | if (it.it_lock_mode != 0) { |
d3a8a4e2 | 884 | ldlm_lock_decref_and_cancel(&och->och_lease_handle, |
e476f2e5 JH |
885 | it.it_lock_mode); |
886 | it.it_lock_mode = 0; | |
e55a68b6 | 887 | och->och_lease_handle.cookie = 0ULL; |
d3a8a4e2 | 888 | } |
e55a68b6 JX |
889 | rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, inode, och, NULL); |
890 | if (rc2 < 0) | |
891 | CERROR("%s: error closing file "DFID": %d\n", | |
892 | ll_get_fsname(inode->i_sb, NULL, 0), | |
893 | PFID(&ll_i2info(inode)->lli_fid), rc2); | |
894 | och = NULL; /* och has been freed in ll_close_inode_openhandle() */ | |
d3a8a4e2 JX |
895 | out_release_it: |
896 | ll_intent_release(&it); | |
897 | out: | |
97903a26 | 898 | kfree(och); |
d3a8a4e2 JX |
899 | return ERR_PTR(rc); |
900 | } | |
d3a8a4e2 JX |
901 | |
902 | /** | |
903 | * Release lease and close the file. | |
904 | * It will check if the lease has ever broken. | |
905 | */ | |
2d95f10e JH |
906 | static int ll_lease_close(struct obd_client_handle *och, struct inode *inode, |
907 | bool *lease_broken) | |
d3a8a4e2 JX |
908 | { |
909 | struct ldlm_lock *lock; | |
910 | bool cancelled = true; | |
911 | int rc; | |
912 | ||
913 | lock = ldlm_handle2lock(&och->och_lease_handle); | |
6e16818b | 914 | if (lock) { |
d3a8a4e2 JX |
915 | lock_res_and_lock(lock); |
916 | cancelled = ldlm_is_cancel(lock); | |
917 | unlock_res_and_lock(lock); | |
ead02808 | 918 | LDLM_LOCK_PUT(lock); |
d3a8a4e2 JX |
919 | } |
920 | ||
e15ba45d OD |
921 | CDEBUG(D_INODE, "lease for " DFID " broken? %d\n", |
922 | PFID(&ll_i2info(inode)->lli_fid), cancelled); | |
d3a8a4e2 JX |
923 | |
924 | if (!cancelled) | |
925 | ldlm_cli_cancel(&och->och_lease_handle, 0); | |
6e16818b | 926 | if (lease_broken) |
d3a8a4e2 JX |
927 | *lease_broken = cancelled; |
928 | ||
48d23e61 JX |
929 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och, |
930 | NULL); | |
d3a8a4e2 JX |
931 | return rc; |
932 | } | |
d3a8a4e2 | 933 | |
d7e09d03 PT |
934 | /* Fills the obdo with the attributes for the lsm */ |
935 | static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, | |
e1798006 | 936 | struct obdo *obdo, __u64 ioepoch, int dv_flags) |
d7e09d03 PT |
937 | { |
938 | struct ptlrpc_request_set *set; | |
45efd655 | 939 | struct obd_info oinfo = { }; |
d7e09d03 PT |
940 | int rc; |
941 | ||
6e16818b | 942 | LASSERT(lsm); |
d7e09d03 PT |
943 | |
944 | oinfo.oi_md = lsm; | |
945 | oinfo.oi_oa = obdo; | |
946 | oinfo.oi_oa->o_oi = lsm->lsm_oi; | |
947 | oinfo.oi_oa->o_mode = S_IFREG; | |
948 | oinfo.oi_oa->o_ioepoch = ioepoch; | |
949 | oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | | |
950 | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | | |
951 | OBD_MD_FLBLKSZ | OBD_MD_FLATIME | | |
952 | OBD_MD_FLMTIME | OBD_MD_FLCTIME | | |
953 | OBD_MD_FLGROUP | OBD_MD_FLEPOCH | | |
954 | OBD_MD_FLDATAVERSION; | |
e1798006 | 955 | if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) { |
d7e09d03 PT |
956 | oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS; |
957 | oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK; | |
e1798006 JX |
958 | if (dv_flags & LL_DV_WR_FLUSH) |
959 | oinfo.oi_oa->o_flags |= OBD_FL_FLUSH; | |
d7e09d03 PT |
960 | } |
961 | ||
962 | set = ptlrpc_prep_set(); | |
6e16818b | 963 | if (!set) { |
19b2056f | 964 | CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM); |
d7e09d03 PT |
965 | rc = -ENOMEM; |
966 | } else { | |
967 | rc = obd_getattr_async(exp, &oinfo, set); | |
968 | if (rc == 0) | |
969 | rc = ptlrpc_set_wait(set); | |
970 | ptlrpc_set_destroy(set); | |
971 | } | |
e1798006 | 972 | if (rc == 0) { |
d7e09d03 PT |
973 | oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | |
974 | OBD_MD_FLATIME | OBD_MD_FLMTIME | | |
975 | OBD_MD_FLCTIME | OBD_MD_FLSIZE | | |
e1798006 JX |
976 | OBD_MD_FLDATAVERSION | OBD_MD_FLFLAGS); |
977 | if (dv_flags & LL_DV_WR_FLUSH && | |
978 | !(oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS && | |
979 | oinfo.oi_oa->o_flags & OBD_FL_FLUSH)) | |
980 | return -ENOTSUPP; | |
981 | } | |
0a3bdb00 | 982 | return rc; |
d7e09d03 PT |
983 | } |
984 | ||
985 | /** | |
986 | * Performs the getattr on the inode and updates its fields. | |
987 | * If @sync != 0, perform the getattr under the server-side lock. | |
988 | */ | |
989 | int ll_inode_getattr(struct inode *inode, struct obdo *obdo, | |
990 | __u64 ioepoch, int sync) | |
991 | { | |
d7e09d03 PT |
992 | struct lov_stripe_md *lsm; |
993 | int rc; | |
d7e09d03 PT |
994 | |
995 | lsm = ccc_inode_lsm_get(inode); | |
996 | rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode), | |
e1798006 | 997 | obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0); |
d7e09d03 PT |
998 | if (rc == 0) { |
999 | struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi; | |
1000 | ||
1001 | obdo_refresh_inode(inode, obdo, obdo->o_valid); | |
2d00bd17 JP |
1002 | CDEBUG(D_INODE, "objid " DOSTID " size %llu, blocks %llu, blksize %lu\n", |
1003 | POSTID(oi), i_size_read(inode), | |
d7e09d03 | 1004 | (unsigned long long)inode->i_blocks, |
16e0631d | 1005 | 1UL << inode->i_blkbits); |
d7e09d03 PT |
1006 | } |
1007 | ccc_inode_lsm_put(inode, lsm); | |
0a3bdb00 | 1008 | return rc; |
d7e09d03 PT |
1009 | } |
1010 | ||
d2995737 | 1011 | int ll_merge_attr(const struct lu_env *env, struct inode *inode) |
d7e09d03 PT |
1012 | { |
1013 | struct ll_inode_info *lli = ll_i2info(inode); | |
1014 | struct cl_object *obj = lli->lli_clob; | |
9acc4500 | 1015 | struct cl_attr *attr = vvp_env_thread_attr(env); |
d2995737 JH |
1016 | s64 atime; |
1017 | s64 mtime; | |
1018 | s64 ctime; | |
d7e09d03 PT |
1019 | int rc = 0; |
1020 | ||
d7e09d03 | 1021 | ll_inode_size_lock(inode); |
d2995737 | 1022 | |
d7e09d03 | 1023 | /* merge timestamps the most recently obtained from mds with |
c0894c6c OD |
1024 | * timestamps obtained from osts |
1025 | */ | |
d2995737 JH |
1026 | LTIME_S(inode->i_atime) = lli->lli_atime; |
1027 | LTIME_S(inode->i_mtime) = lli->lli_mtime; | |
1028 | LTIME_S(inode->i_ctime) = lli->lli_ctime; | |
376ef86b | 1029 | |
d2995737 JH |
1030 | mtime = LTIME_S(inode->i_mtime); |
1031 | atime = LTIME_S(inode->i_atime); | |
1032 | ctime = LTIME_S(inode->i_ctime); | |
d7e09d03 PT |
1033 | |
1034 | cl_object_attr_lock(obj); | |
1035 | rc = cl_object_attr_get(env, obj, attr); | |
1036 | cl_object_attr_unlock(obj); | |
1037 | ||
d2995737 JH |
1038 | if (rc != 0) |
1039 | goto out_size_unlock; | |
d7e09d03 | 1040 | |
d2995737 JH |
1041 | if (atime < attr->cat_atime) |
1042 | atime = attr->cat_atime; | |
d7e09d03 | 1043 | |
d2995737 JH |
1044 | if (ctime < attr->cat_ctime) |
1045 | ctime = attr->cat_ctime; | |
d7e09d03 | 1046 | |
d2995737 JH |
1047 | if (mtime < attr->cat_mtime) |
1048 | mtime = attr->cat_mtime; | |
1049 | ||
1050 | CDEBUG(D_VFSTRACE, DFID " updating i_size %llu\n", | |
1051 | PFID(&lli->lli_fid), attr->cat_size); | |
1052 | ||
1929c433 | 1053 | i_size_write(inode, attr->cat_size); |
d2995737 JH |
1054 | |
1055 | inode->i_blocks = attr->cat_blocks; | |
1056 | ||
1057 | LTIME_S(inode->i_mtime) = mtime; | |
1058 | LTIME_S(inode->i_atime) = atime; | |
1059 | LTIME_S(inode->i_ctime) = ctime; | |
1060 | ||
1061 | out_size_unlock: | |
d7e09d03 PT |
1062 | ll_inode_size_unlock(inode); |
1063 | ||
0a3bdb00 | 1064 | return rc; |
d7e09d03 PT |
1065 | } |
1066 | ||
1067 | int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm, | |
1068 | lstat_t *st) | |
1069 | { | |
1070 | struct obdo obdo = { 0 }; | |
1071 | int rc; | |
1072 | ||
ef2e0f55 | 1073 | rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, &obdo, 0, 0); |
d7e09d03 PT |
1074 | if (rc == 0) { |
1075 | st->st_size = obdo.o_size; | |
1076 | st->st_blocks = obdo.o_blocks; | |
1077 | st->st_mtime = obdo.o_mtime; | |
1078 | st->st_atime = obdo.o_atime; | |
1079 | st->st_ctime = obdo.o_ctime; | |
1080 | } | |
1081 | return rc; | |
1082 | } | |
1083 | ||
ec9bca9c JH |
1084 | static bool file_is_noatime(const struct file *file) |
1085 | { | |
1086 | const struct vfsmount *mnt = file->f_path.mnt; | |
2a8a3597 | 1087 | const struct inode *inode = file_inode(file); |
ec9bca9c JH |
1088 | |
1089 | /* Adapted from file_accessed() and touch_atime().*/ | |
1090 | if (file->f_flags & O_NOATIME) | |
1091 | return true; | |
1092 | ||
1093 | if (inode->i_flags & S_NOATIME) | |
1094 | return true; | |
1095 | ||
1096 | if (IS_NOATIME(inode)) | |
1097 | return true; | |
1098 | ||
1099 | if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY)) | |
1100 | return true; | |
1101 | ||
1102 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1103 | return true; | |
1104 | ||
1105 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1106 | return true; | |
1107 | ||
1108 | return false; | |
1109 | } | |
1110 | ||
d7e09d03 PT |
1111 | void ll_io_init(struct cl_io *io, const struct file *file, int write) |
1112 | { | |
2a8a3597 | 1113 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
1114 | |
1115 | io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK; | |
1116 | if (write) { | |
1117 | io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); | |
1118 | io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || | |
1119 | file->f_flags & O_DIRECT || | |
1120 | IS_SYNC(inode); | |
1121 | } | |
1122 | io->ci_obj = ll_i2info(inode)->lli_clob; | |
1123 | io->ci_lockreq = CILR_MAYBE; | |
1124 | if (ll_file_nolock(file)) { | |
1125 | io->ci_lockreq = CILR_NEVER; | |
1126 | io->ci_no_srvlock = 1; | |
1127 | } else if (file->f_flags & O_APPEND) { | |
1128 | io->ci_lockreq = CILR_MANDATORY; | |
1129 | } | |
ec9bca9c JH |
1130 | |
1131 | io->ci_noatime = file_is_noatime(file); | |
d7e09d03 PT |
1132 | } |
1133 | ||
1134 | static ssize_t | |
1135 | ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, | |
1136 | struct file *file, enum cl_io_type iot, | |
1137 | loff_t *ppos, size_t count) | |
1138 | { | |
2a8a3597 | 1139 | struct ll_inode_info *lli = ll_i2info(file_inode(file)); |
d7e09d03 PT |
1140 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
1141 | struct cl_io *io; | |
1142 | ssize_t result; | |
d7e09d03 | 1143 | |
77605e41 JX |
1144 | CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: %llu, count: %zd\n", |
1145 | file->f_path.dentry->d_name.name, iot, *ppos, count); | |
1146 | ||
d7e09d03 | 1147 | restart: |
9acc4500 | 1148 | io = vvp_env_thread_io(env); |
d7e09d03 PT |
1149 | ll_io_init(io, file, iot == CIT_WRITE); |
1150 | ||
1151 | if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) { | |
e0a8144b | 1152 | struct vvp_io *vio = vvp_env_io(env); |
d7e09d03 PT |
1153 | int write_mutex_locked = 0; |
1154 | ||
e0a8144b | 1155 | vio->vui_fd = LUSTRE_FPRIVATE(file); |
82c156f8 AV |
1156 | vio->vui_iter = args->u.normal.via_iter; |
1157 | vio->vui_iocb = args->u.normal.via_iocb; | |
1158 | if ((iot == CIT_WRITE) && | |
1159 | !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { | |
1160 | if (mutex_lock_interruptible(&lli->lli_write_mutex)) { | |
1161 | result = -ERESTARTSYS; | |
1162 | goto out; | |
d7e09d03 | 1163 | } |
82c156f8 | 1164 | write_mutex_locked = 1; |
d7e09d03 | 1165 | } |
82c156f8 | 1166 | down_read(&lli->lli_trunc_sem); |
966c4a8f | 1167 | ll_cl_add(file, env, io); |
d7e09d03 | 1168 | result = cl_io_loop(env, io); |
966c4a8f | 1169 | ll_cl_remove(file, env); |
82c156f8 | 1170 | up_read(&lli->lli_trunc_sem); |
d7e09d03 PT |
1171 | if (write_mutex_locked) |
1172 | mutex_unlock(&lli->lli_write_mutex); | |
d7e09d03 PT |
1173 | } else { |
1174 | /* cl_io_rw_init() handled IO */ | |
1175 | result = io->ci_result; | |
1176 | } | |
1177 | ||
1178 | if (io->ci_nob > 0) { | |
1179 | result = io->ci_nob; | |
1180 | *ppos = io->u.ci_wr.wr.crw_pos; | |
1181 | } | |
34e1f2bb | 1182 | goto out; |
d7e09d03 PT |
1183 | out: |
1184 | cl_io_fini(env, io); | |
1185 | /* If any bit been read/written (result != 0), we just return | |
c0894c6c OD |
1186 | * short read/write instead of restart io. |
1187 | */ | |
5ea17d6c | 1188 | if ((result == 0 || result == -ENODATA) && io->ci_need_restart) { |
09561a53 | 1189 | CDEBUG(D_VFSTRACE, "Restart %s on %pD from %lld, count:%zd\n", |
d7e09d03 | 1190 | iot == CIT_READ ? "read" : "write", |
09561a53 | 1191 | file, *ppos, count); |
19b2056f | 1192 | LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob); |
d7e09d03 PT |
1193 | goto restart; |
1194 | } | |
1195 | ||
1196 | if (iot == CIT_READ) { | |
1197 | if (result >= 0) | |
2a8a3597 | 1198 | ll_stats_ops_tally(ll_i2sbi(file_inode(file)), |
d7e09d03 PT |
1199 | LPROC_LL_READ_BYTES, result); |
1200 | } else if (iot == CIT_WRITE) { | |
1201 | if (result >= 0) { | |
2a8a3597 | 1202 | ll_stats_ops_tally(ll_i2sbi(file_inode(file)), |
d7e09d03 PT |
1203 | LPROC_LL_WRITE_BYTES, result); |
1204 | fd->fd_write_failed = false; | |
1205 | } else if (result != -ERESTARTSYS) { | |
1206 | fd->fd_write_failed = true; | |
1207 | } | |
1208 | } | |
77605e41 | 1209 | CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result); |
d7e09d03 PT |
1210 | |
1211 | return result; | |
1212 | } | |
1213 | ||
b42b15fd | 1214 | static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
d7e09d03 PT |
1215 | { |
1216 | struct lu_env *env; | |
1217 | struct vvp_io_args *args; | |
d7e09d03 PT |
1218 | ssize_t result; |
1219 | int refcheck; | |
d7e09d03 | 1220 | |
d7e09d03 PT |
1221 | env = cl_env_get(&refcheck); |
1222 | if (IS_ERR(env)) | |
0a3bdb00 | 1223 | return PTR_ERR(env); |
d7e09d03 | 1224 | |
82c156f8 | 1225 | args = ll_env_args(env); |
b42b15fd | 1226 | args->u.normal.via_iter = to; |
d7e09d03 PT |
1227 | args->u.normal.via_iocb = iocb; |
1228 | ||
1229 | result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ, | |
b42b15fd | 1230 | &iocb->ki_pos, iov_iter_count(to)); |
d7e09d03 | 1231 | cl_env_put(env, &refcheck); |
0a3bdb00 | 1232 | return result; |
d7e09d03 PT |
1233 | } |
1234 | ||
1235 | /* | |
1236 | * Write to a file (through the page cache). | |
1237 | */ | |
b42b15fd | 1238 | static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
d7e09d03 PT |
1239 | { |
1240 | struct lu_env *env; | |
1241 | struct vvp_io_args *args; | |
d7e09d03 PT |
1242 | ssize_t result; |
1243 | int refcheck; | |
d7e09d03 | 1244 | |
d7e09d03 PT |
1245 | env = cl_env_get(&refcheck); |
1246 | if (IS_ERR(env)) | |
0a3bdb00 | 1247 | return PTR_ERR(env); |
d7e09d03 | 1248 | |
82c156f8 | 1249 | args = ll_env_args(env); |
b42b15fd | 1250 | args->u.normal.via_iter = from; |
d7e09d03 PT |
1251 | args->u.normal.via_iocb = iocb; |
1252 | ||
1253 | result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE, | |
e15ba45d | 1254 | &iocb->ki_pos, iov_iter_count(from)); |
d7e09d03 | 1255 | cl_env_put(env, &refcheck); |
0a3bdb00 | 1256 | return result; |
d7e09d03 PT |
1257 | } |
1258 | ||
21aef7d9 | 1259 | static int ll_lov_recreate(struct inode *inode, struct ost_id *oi, u32 ost_idx) |
d7e09d03 PT |
1260 | { |
1261 | struct obd_export *exp = ll_i2dtexp(inode); | |
1262 | struct obd_trans_info oti = { 0 }; | |
1263 | struct obdo *oa = NULL; | |
1264 | int lsm_size; | |
1265 | int rc = 0; | |
1266 | struct lov_stripe_md *lsm = NULL, *lsm2; | |
d7e09d03 | 1267 | |
21068c46 | 1268 | oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS); |
6e16818b | 1269 | if (!oa) |
0a3bdb00 | 1270 | return -ENOMEM; |
d7e09d03 PT |
1271 | |
1272 | lsm = ccc_inode_lsm_get(inode); | |
34e1f2bb JL |
1273 | if (!lsm_has_objects(lsm)) { |
1274 | rc = -ENOENT; | |
1275 | goto out; | |
1276 | } | |
d7e09d03 PT |
1277 | |
1278 | lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) * | |
1279 | (lsm->lsm_stripe_count)); | |
1280 | ||
e958f49b | 1281 | lsm2 = libcfs_kvzalloc(lsm_size, GFP_NOFS); |
6e16818b | 1282 | if (!lsm2) { |
34e1f2bb JL |
1283 | rc = -ENOMEM; |
1284 | goto out; | |
1285 | } | |
d7e09d03 PT |
1286 | |
1287 | oa->o_oi = *oi; | |
1288 | oa->o_nlink = ost_idx; | |
1289 | oa->o_flags |= OBD_FL_RECREATE_OBJS; | |
1290 | oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP; | |
1291 | obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | | |
1292 | OBD_MD_FLMTIME | OBD_MD_FLCTIME); | |
1293 | obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid); | |
1294 | memcpy(lsm2, lsm, lsm_size); | |
1295 | ll_inode_size_lock(inode); | |
1296 | rc = obd_create(NULL, exp, oa, &lsm2, &oti); | |
1297 | ll_inode_size_unlock(inode); | |
1298 | ||
e958f49b | 1299 | kvfree(lsm2); |
34e1f2bb | 1300 | goto out; |
d7e09d03 PT |
1301 | out: |
1302 | ccc_inode_lsm_put(inode, lsm); | |
2ba262fb | 1303 | kmem_cache_free(obdo_cachep, oa); |
d7e09d03 PT |
1304 | return rc; |
1305 | } | |
1306 | ||
1307 | static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg) | |
1308 | { | |
1309 | struct ll_recreate_obj ucreat; | |
1310 | struct ost_id oi; | |
d7e09d03 | 1311 | |
2eb90a75 | 1312 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1313 | return -EPERM; |
d7e09d03 | 1314 | |
02f9c12e | 1315 | if (copy_from_user(&ucreat, (struct ll_recreate_obj __user *)arg, |
d7e09d03 | 1316 | sizeof(ucreat))) |
0a3bdb00 | 1317 | return -EFAULT; |
d7e09d03 PT |
1318 | |
1319 | ostid_set_seq_mdt0(&oi); | |
1320 | ostid_set_id(&oi, ucreat.lrc_id); | |
0a3bdb00 | 1321 | return ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx); |
d7e09d03 PT |
1322 | } |
1323 | ||
1324 | static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg) | |
1325 | { | |
1326 | struct lu_fid fid; | |
1327 | struct ost_id oi; | |
21aef7d9 | 1328 | u32 ost_idx; |
d7e09d03 | 1329 | |
2eb90a75 | 1330 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1331 | return -EPERM; |
d7e09d03 | 1332 | |
02f9c12e | 1333 | if (copy_from_user(&fid, (struct lu_fid __user *)arg, sizeof(fid))) |
0a3bdb00 | 1334 | return -EFAULT; |
d7e09d03 PT |
1335 | |
1336 | fid_to_ostid(&fid, &oi); | |
1337 | ost_idx = (fid_seq(&fid) >> 16) & 0xffff; | |
0a3bdb00 | 1338 | return ll_lov_recreate(inode, &oi, ost_idx); |
d7e09d03 PT |
1339 | } |
1340 | ||
c139f3ce | 1341 | int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry, |
d467220e NY |
1342 | __u64 flags, struct lov_user_md *lum, |
1343 | int lum_size) | |
d7e09d03 PT |
1344 | { |
1345 | struct lov_stripe_md *lsm = NULL; | |
1346 | struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; | |
1347 | int rc = 0; | |
d7e09d03 PT |
1348 | |
1349 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1350 | if (lsm) { |
d7e09d03 | 1351 | ccc_inode_lsm_put(inode, lsm); |
97a075cd JN |
1352 | CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n", |
1353 | PFID(ll_inode2fid(inode))); | |
34e1f2bb JL |
1354 | rc = -EEXIST; |
1355 | goto out; | |
d7e09d03 PT |
1356 | } |
1357 | ||
1358 | ll_inode_size_lock(inode); | |
c139f3ce | 1359 | rc = ll_intent_file_open(dentry, lum, lum_size, &oit); |
d7e09d03 | 1360 | if (rc) |
34e1f2bb | 1361 | goto out_unlock; |
e476f2e5 | 1362 | rc = oit.it_status; |
d7e09d03 | 1363 | if (rc < 0) |
34e1f2bb | 1364 | goto out_req_free; |
d7e09d03 | 1365 | |
e22fdcc8 | 1366 | ll_release_openhandle(inode, &oit); |
d7e09d03 | 1367 | |
38585ccc | 1368 | out_unlock: |
d7e09d03 PT |
1369 | ll_inode_size_unlock(inode); |
1370 | ll_intent_release(&oit); | |
1371 | ccc_inode_lsm_put(inode, lsm); | |
38585ccc | 1372 | out: |
0a3bdb00 | 1373 | return rc; |
d7e09d03 | 1374 | out_req_free: |
8bf86fd9 | 1375 | ptlrpc_req_finished((struct ptlrpc_request *)oit.it_request); |
d7e09d03 PT |
1376 | goto out; |
1377 | } | |
1378 | ||
1379 | int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, | |
1380 | struct lov_mds_md **lmmp, int *lmm_size, | |
1381 | struct ptlrpc_request **request) | |
1382 | { | |
1383 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1384 | struct mdt_body *body; | |
1385 | struct lov_mds_md *lmm = NULL; | |
1386 | struct ptlrpc_request *req = NULL; | |
1387 | struct md_op_data *op_data; | |
1388 | int rc, lmmsize; | |
1389 | ||
44779340 | 1390 | rc = ll_get_default_mdsize(sbi, &lmmsize); |
d7e09d03 | 1391 | if (rc) |
0a3bdb00 | 1392 | return rc; |
d7e09d03 PT |
1393 | |
1394 | op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, | |
1395 | strlen(filename), lmmsize, | |
1396 | LUSTRE_OPC_ANY, NULL); | |
1397 | if (IS_ERR(op_data)) | |
0a3bdb00 | 1398 | return PTR_ERR(op_data); |
d7e09d03 PT |
1399 | |
1400 | op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; | |
1401 | rc = md_getattr_name(sbi->ll_md_exp, op_data, &req); | |
1402 | ll_finish_md_op_data(op_data); | |
1403 | if (rc < 0) { | |
2d00bd17 JP |
1404 | CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n", |
1405 | filename, rc); | |
34e1f2bb | 1406 | goto out; |
d7e09d03 PT |
1407 | } |
1408 | ||
1409 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
d7e09d03 PT |
1410 | |
1411 | lmmsize = body->eadatasize; | |
1412 | ||
1413 | if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) || | |
e15ba45d | 1414 | lmmsize == 0) { |
34e1f2bb JL |
1415 | rc = -ENODATA; |
1416 | goto out; | |
d7e09d03 PT |
1417 | } |
1418 | ||
1419 | lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize); | |
d7e09d03 PT |
1420 | |
1421 | if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) && | |
1422 | (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) { | |
34e1f2bb JL |
1423 | rc = -EPROTO; |
1424 | goto out; | |
d7e09d03 PT |
1425 | } |
1426 | ||
1427 | /* | |
1428 | * This is coming from the MDS, so is probably in | |
1429 | * little endian. We convert it to host endian before | |
1430 | * passing it to userspace. | |
1431 | */ | |
1f6eaf83 | 1432 | if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) { |
5dd16419 JX |
1433 | int stripe_count; |
1434 | ||
1435 | stripe_count = le16_to_cpu(lmm->lmm_stripe_count); | |
1436 | if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED) | |
1437 | stripe_count = 0; | |
1438 | ||
d7e09d03 | 1439 | /* if function called for directory - we should |
c0894c6c OD |
1440 | * avoid swab not existent lsm objects |
1441 | */ | |
d7e09d03 PT |
1442 | if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) { |
1443 | lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); | |
1444 | if (S_ISREG(body->mode)) | |
1445 | lustre_swab_lov_user_md_objects( | |
1446 | ((struct lov_user_md_v1 *)lmm)->lmm_objects, | |
5dd16419 | 1447 | stripe_count); |
d7e09d03 PT |
1448 | } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) { |
1449 | lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); | |
1450 | if (S_ISREG(body->mode)) | |
1451 | lustre_swab_lov_user_md_objects( | |
1452 | ((struct lov_user_md_v3 *)lmm)->lmm_objects, | |
5dd16419 | 1453 | stripe_count); |
d7e09d03 PT |
1454 | } |
1455 | } | |
1456 | ||
1457 | out: | |
1458 | *lmmp = lmm; | |
1459 | *lmm_size = lmmsize; | |
1460 | *request = req; | |
1461 | return rc; | |
1462 | } | |
1463 | ||
1464 | static int ll_lov_setea(struct inode *inode, struct file *file, | |
e15ba45d | 1465 | unsigned long arg) |
d7e09d03 | 1466 | { |
d467220e | 1467 | __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE; |
d7e09d03 PT |
1468 | struct lov_user_md *lump; |
1469 | int lum_size = sizeof(struct lov_user_md) + | |
1470 | sizeof(struct lov_user_ost_data); | |
1471 | int rc; | |
d7e09d03 | 1472 | |
2eb90a75 | 1473 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1474 | return -EPERM; |
d7e09d03 | 1475 | |
e958f49b | 1476 | lump = libcfs_kvzalloc(lum_size, GFP_NOFS); |
6e16818b | 1477 | if (!lump) |
0a3bdb00 | 1478 | return -ENOMEM; |
d7e09d03 | 1479 | |
02f9c12e | 1480 | if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) { |
e958f49b | 1481 | kvfree(lump); |
0a3bdb00 | 1482 | return -EFAULT; |
d7e09d03 PT |
1483 | } |
1484 | ||
c139f3ce | 1485 | rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump, |
e15ba45d | 1486 | lum_size); |
c139f3ce | 1487 | cl_lov_delay_create_clear(&file->f_flags); |
d7e09d03 | 1488 | |
e958f49b | 1489 | kvfree(lump); |
0a3bdb00 | 1490 | return rc; |
d7e09d03 PT |
1491 | } |
1492 | ||
1493 | static int ll_lov_setstripe(struct inode *inode, struct file *file, | |
1494 | unsigned long arg) | |
1495 | { | |
02f9c12e OD |
1496 | struct lov_user_md_v3 lumv3; |
1497 | struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; | |
1498 | struct lov_user_md_v1 __user *lumv1p = (void __user *)arg; | |
1499 | struct lov_user_md_v3 __user *lumv3p = (void __user *)arg; | |
1500 | int lum_size, rc; | |
d467220e | 1501 | __u64 flags = FMODE_WRITE; |
d7e09d03 PT |
1502 | |
1503 | /* first try with v1 which is smaller than v3 */ | |
1504 | lum_size = sizeof(struct lov_user_md_v1); | |
1505 | if (copy_from_user(lumv1, lumv1p, lum_size)) | |
0a3bdb00 | 1506 | return -EFAULT; |
d7e09d03 PT |
1507 | |
1508 | if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { | |
1509 | lum_size = sizeof(struct lov_user_md_v3); | |
1510 | if (copy_from_user(&lumv3, lumv3p, lum_size)) | |
0a3bdb00 | 1511 | return -EFAULT; |
d7e09d03 PT |
1512 | } |
1513 | ||
c139f3ce AV |
1514 | rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lumv1, |
1515 | lum_size); | |
1516 | cl_lov_delay_create_clear(&file->f_flags); | |
d7e09d03 PT |
1517 | if (rc == 0) { |
1518 | struct lov_stripe_md *lsm; | |
1519 | __u32 gen; | |
1520 | ||
1521 | put_user(0, &lumv1p->lmm_stripe_count); | |
1522 | ||
1523 | ll_layout_refresh(inode, &gen); | |
1524 | lsm = ccc_inode_lsm_get(inode); | |
1525 | rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), | |
e09bee34 | 1526 | 0, lsm, (void __user *)arg); |
d7e09d03 PT |
1527 | ccc_inode_lsm_put(inode, lsm); |
1528 | } | |
0a3bdb00 | 1529 | return rc; |
d7e09d03 PT |
1530 | } |
1531 | ||
1532 | static int ll_lov_getstripe(struct inode *inode, unsigned long arg) | |
1533 | { | |
1534 | struct lov_stripe_md *lsm; | |
1535 | int rc = -ENODATA; | |
d7e09d03 PT |
1536 | |
1537 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1538 | if (lsm) |
d7e09d03 | 1539 | rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, |
e09bee34 | 1540 | lsm, (void __user *)arg); |
d7e09d03 | 1541 | ccc_inode_lsm_put(inode, lsm); |
0a3bdb00 | 1542 | return rc; |
d7e09d03 PT |
1543 | } |
1544 | ||
2d95f10e JH |
1545 | static int |
1546 | ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) | |
d7e09d03 PT |
1547 | { |
1548 | struct ll_inode_info *lli = ll_i2info(inode); | |
1549 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
98eae5e7 | 1550 | struct ll_grouplock grouplock; |
d7e09d03 | 1551 | int rc; |
d7e09d03 | 1552 | |
431b5678 PF |
1553 | if (arg == 0) { |
1554 | CWARN("group id for group lock must not be 0\n"); | |
1555 | return -EINVAL; | |
1556 | } | |
1557 | ||
d7e09d03 | 1558 | if (ll_file_nolock(file)) |
0a3bdb00 | 1559 | return -EOPNOTSUPP; |
d7e09d03 PT |
1560 | |
1561 | spin_lock(&lli->lli_lock); | |
1562 | if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { | |
1563 | CWARN("group lock already existed with gid %lu\n", | |
98eae5e7 | 1564 | fd->fd_grouplock.lg_gid); |
d7e09d03 | 1565 | spin_unlock(&lli->lli_lock); |
0a3bdb00 | 1566 | return -EINVAL; |
d7e09d03 | 1567 | } |
98eae5e7 | 1568 | LASSERT(!fd->fd_grouplock.lg_lock); |
d7e09d03 PT |
1569 | spin_unlock(&lli->lli_lock); |
1570 | ||
1929c433 | 1571 | rc = cl_get_grouplock(ll_i2info(inode)->lli_clob, |
d7e09d03 PT |
1572 | arg, (file->f_flags & O_NONBLOCK), &grouplock); |
1573 | if (rc) | |
0a3bdb00 | 1574 | return rc; |
d7e09d03 PT |
1575 | |
1576 | spin_lock(&lli->lli_lock); | |
1577 | if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { | |
1578 | spin_unlock(&lli->lli_lock); | |
1579 | CERROR("another thread just won the race\n"); | |
1580 | cl_put_grouplock(&grouplock); | |
0a3bdb00 | 1581 | return -EINVAL; |
d7e09d03 PT |
1582 | } |
1583 | ||
1584 | fd->fd_flags |= LL_FILE_GROUP_LOCKED; | |
1585 | fd->fd_grouplock = grouplock; | |
1586 | spin_unlock(&lli->lli_lock); | |
1587 | ||
1588 | CDEBUG(D_INFO, "group lock %lu obtained\n", arg); | |
0a3bdb00 | 1589 | return 0; |
d7e09d03 PT |
1590 | } |
1591 | ||
920b4f2e LC |
1592 | static int ll_put_grouplock(struct inode *inode, struct file *file, |
1593 | unsigned long arg) | |
d7e09d03 PT |
1594 | { |
1595 | struct ll_inode_info *lli = ll_i2info(inode); | |
1596 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
98eae5e7 | 1597 | struct ll_grouplock grouplock; |
d7e09d03 PT |
1598 | |
1599 | spin_lock(&lli->lli_lock); | |
1600 | if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { | |
1601 | spin_unlock(&lli->lli_lock); | |
1602 | CWARN("no group lock held\n"); | |
0a3bdb00 | 1603 | return -EINVAL; |
d7e09d03 | 1604 | } |
98eae5e7 | 1605 | LASSERT(fd->fd_grouplock.lg_lock); |
d7e09d03 | 1606 | |
98eae5e7 | 1607 | if (fd->fd_grouplock.lg_gid != arg) { |
d7e09d03 | 1608 | CWARN("group lock %lu doesn't match current id %lu\n", |
98eae5e7 | 1609 | arg, fd->fd_grouplock.lg_gid); |
d7e09d03 | 1610 | spin_unlock(&lli->lli_lock); |
0a3bdb00 | 1611 | return -EINVAL; |
d7e09d03 PT |
1612 | } |
1613 | ||
1614 | grouplock = fd->fd_grouplock; | |
1615 | memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); | |
1616 | fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; | |
1617 | spin_unlock(&lli->lli_lock); | |
1618 | ||
1619 | cl_put_grouplock(&grouplock); | |
1620 | CDEBUG(D_INFO, "group lock %lu released\n", arg); | |
0a3bdb00 | 1621 | return 0; |
d7e09d03 PT |
1622 | } |
1623 | ||
1624 | /** | |
1625 | * Close inode open handle | |
1626 | * | |
e22fdcc8 | 1627 | * \param inode [in] inode in question |
d7e09d03 PT |
1628 | * \param it [in,out] intent which contains open info and result |
1629 | * | |
1630 | * \retval 0 success | |
1631 | * \retval <0 failure | |
1632 | */ | |
e22fdcc8 | 1633 | int ll_release_openhandle(struct inode *inode, struct lookup_intent *it) |
d7e09d03 | 1634 | { |
d7e09d03 PT |
1635 | struct obd_client_handle *och; |
1636 | int rc; | |
d7e09d03 PT |
1637 | |
1638 | LASSERT(inode); | |
1639 | ||
1640 | /* Root ? Do nothing. */ | |
f76c23da | 1641 | if (is_root_inode(inode)) |
0a3bdb00 | 1642 | return 0; |
d7e09d03 PT |
1643 | |
1644 | /* No open handle to close? Move away */ | |
1645 | if (!it_disposition(it, DISP_OPEN_OPEN)) | |
0a3bdb00 | 1646 | return 0; |
d7e09d03 PT |
1647 | |
1648 | LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0); | |
1649 | ||
496a51bd | 1650 | och = kzalloc(sizeof(*och), GFP_NOFS); |
34e1f2bb JL |
1651 | if (!och) { |
1652 | rc = -ENOMEM; | |
1653 | goto out; | |
1654 | } | |
d7e09d03 | 1655 | |
ea1db081 | 1656 | ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och); |
d7e09d03 PT |
1657 | |
1658 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, | |
48d23e61 JX |
1659 | inode, och, NULL); |
1660 | out: | |
d7e09d03 PT |
1661 | /* this one is in place of ll_file_open */ |
1662 | if (it_disposition(it, DISP_ENQ_OPEN_REF)) { | |
8bf86fd9 | 1663 | ptlrpc_req_finished(it->it_request); |
d7e09d03 PT |
1664 | it_clear_disposition(it, DISP_ENQ_OPEN_REF); |
1665 | } | |
0a3bdb00 | 1666 | return rc; |
d7e09d03 PT |
1667 | } |
1668 | ||
1669 | /** | |
1670 | * Get size for inode for which FIEMAP mapping is requested. | |
1671 | * Make the FIEMAP get_info call and returns the result. | |
1672 | */ | |
2d95f10e | 1673 | static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, |
ebdc4fc5 | 1674 | size_t num_bytes) |
d7e09d03 PT |
1675 | { |
1676 | struct obd_export *exp = ll_i2dtexp(inode); | |
1677 | struct lov_stripe_md *lsm = NULL; | |
1678 | struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, }; | |
ebdc4fc5 | 1679 | __u32 vallen = num_bytes; |
d7e09d03 | 1680 | int rc; |
d7e09d03 PT |
1681 | |
1682 | /* Checks for fiemap flags */ | |
1683 | if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) { | |
1684 | fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT; | |
1685 | return -EBADR; | |
1686 | } | |
1687 | ||
1688 | /* Check for FIEMAP_FLAG_SYNC */ | |
1689 | if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) { | |
1690 | rc = filemap_fdatawrite(inode->i_mapping); | |
1691 | if (rc) | |
1692 | return rc; | |
1693 | } | |
1694 | ||
1695 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1696 | if (!lsm) |
d7e09d03 PT |
1697 | return -ENOENT; |
1698 | ||
1699 | /* If the stripe_count > 1 and the application does not understand | |
1700 | * DEVICE_ORDER flag, then it cannot interpret the extents correctly. | |
1701 | */ | |
1702 | if (lsm->lsm_stripe_count > 1 && | |
34e1f2bb JL |
1703 | !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) { |
1704 | rc = -EOPNOTSUPP; | |
1705 | goto out; | |
1706 | } | |
d7e09d03 PT |
1707 | |
1708 | fm_key.oa.o_oi = lsm->lsm_oi; | |
1709 | fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; | |
1710 | ||
a915ffda LD |
1711 | if (i_size_read(inode) == 0) { |
1712 | rc = ll_glimpse_size(inode); | |
1713 | if (rc) | |
1714 | goto out; | |
1715 | } | |
1716 | ||
d7e09d03 PT |
1717 | obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE); |
1718 | obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid); | |
1719 | /* If filesize is 0, then there would be no objects for mapping */ | |
1720 | if (fm_key.oa.o_size == 0) { | |
1721 | fiemap->fm_mapped_extents = 0; | |
34e1f2bb JL |
1722 | rc = 0; |
1723 | goto out; | |
d7e09d03 PT |
1724 | } |
1725 | ||
1726 | memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap)); | |
1727 | ||
1728 | rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen, | |
1729 | fiemap, lsm); | |
1730 | if (rc) | |
1731 | CERROR("obd_get_info failed: rc = %d\n", rc); | |
1732 | ||
1733 | out: | |
1734 | ccc_inode_lsm_put(inode, lsm); | |
0a3bdb00 | 1735 | return rc; |
d7e09d03 PT |
1736 | } |
1737 | ||
2b358b4e | 1738 | int ll_fid2path(struct inode *inode, void __user *arg) |
d7e09d03 | 1739 | { |
2b358b4e FZ |
1740 | struct obd_export *exp = ll_i2mdexp(inode); |
1741 | const struct getinfo_fid2path __user *gfin = arg; | |
1742 | struct getinfo_fid2path *gfout; | |
1743 | u32 pathlen; | |
1744 | size_t outsize; | |
1745 | int rc; | |
d7e09d03 | 1746 | |
2eb90a75 | 1747 | if (!capable(CFS_CAP_DAC_READ_SEARCH) && |
d7e09d03 | 1748 | !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH)) |
0a3bdb00 | 1749 | return -EPERM; |
d7e09d03 | 1750 | |
2b358b4e FZ |
1751 | /* Only need to get the buflen */ |
1752 | if (get_user(pathlen, &gfin->gf_pathlen)) | |
0a3bdb00 | 1753 | return -EFAULT; |
d7e09d03 | 1754 | |
c7b09efa OD |
1755 | if (pathlen > PATH_MAX) |
1756 | return -EINVAL; | |
1757 | ||
2b358b4e FZ |
1758 | outsize = sizeof(*gfout) + pathlen; |
1759 | ||
496a51bd JL |
1760 | gfout = kzalloc(outsize, GFP_NOFS); |
1761 | if (!gfout) | |
0a3bdb00 | 1762 | return -ENOMEM; |
2b358b4e | 1763 | |
34e1f2bb JL |
1764 | if (copy_from_user(gfout, arg, sizeof(*gfout))) { |
1765 | rc = -EFAULT; | |
1766 | goto gf_free; | |
1767 | } | |
d7e09d03 PT |
1768 | |
1769 | /* Call mdc_iocontrol */ | |
1770 | rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); | |
2b358b4e | 1771 | if (rc != 0) |
34e1f2bb | 1772 | goto gf_free; |
d7e09d03 PT |
1773 | |
1774 | if (copy_to_user(arg, gfout, outsize)) | |
1775 | rc = -EFAULT; | |
1776 | ||
1777 | gf_free: | |
97903a26 | 1778 | kfree(gfout); |
0a3bdb00 | 1779 | return rc; |
d7e09d03 PT |
1780 | } |
1781 | ||
1782 | static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) | |
1783 | { | |
1784 | struct ll_user_fiemap *fiemap_s; | |
1785 | size_t num_bytes, ret_bytes; | |
1786 | unsigned int extent_count; | |
1787 | int rc = 0; | |
1788 | ||
1789 | /* Get the extent count so we can calculate the size of | |
c0894c6c OD |
1790 | * required fiemap buffer |
1791 | */ | |
d7e09d03 | 1792 | if (get_user(extent_count, |
e15ba45d | 1793 | &((struct ll_user_fiemap __user *)arg)->fm_extent_count)) |
0a3bdb00 | 1794 | return -EFAULT; |
7bc3dfa3 VO |
1795 | |
1796 | if (extent_count >= | |
1797 | (SIZE_MAX - sizeof(*fiemap_s)) / sizeof(struct ll_fiemap_extent)) | |
1798 | return -EINVAL; | |
d7e09d03 PT |
1799 | num_bytes = sizeof(*fiemap_s) + (extent_count * |
1800 | sizeof(struct ll_fiemap_extent)); | |
1801 | ||
e958f49b | 1802 | fiemap_s = libcfs_kvzalloc(num_bytes, GFP_NOFS); |
6e16818b | 1803 | if (!fiemap_s) |
0a3bdb00 | 1804 | return -ENOMEM; |
d7e09d03 PT |
1805 | |
1806 | /* get the fiemap value */ | |
1807 | if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg, | |
34e1f2bb JL |
1808 | sizeof(*fiemap_s))) { |
1809 | rc = -EFAULT; | |
1810 | goto error; | |
1811 | } | |
d7e09d03 PT |
1812 | |
1813 | /* If fm_extent_count is non-zero, read the first extent since | |
1814 | * it is used to calculate end_offset and device from previous | |
c0894c6c OD |
1815 | * fiemap call. |
1816 | */ | |
d7e09d03 PT |
1817 | if (extent_count) { |
1818 | if (copy_from_user(&fiemap_s->fm_extents[0], | |
e15ba45d OD |
1819 | (char __user *)arg + sizeof(*fiemap_s), |
1820 | sizeof(struct ll_fiemap_extent))) { | |
34e1f2bb JL |
1821 | rc = -EFAULT; |
1822 | goto error; | |
1823 | } | |
d7e09d03 PT |
1824 | } |
1825 | ||
1826 | rc = ll_do_fiemap(inode, fiemap_s, num_bytes); | |
1827 | if (rc) | |
34e1f2bb | 1828 | goto error; |
d7e09d03 PT |
1829 | |
1830 | ret_bytes = sizeof(struct ll_user_fiemap); | |
1831 | ||
1832 | if (extent_count != 0) | |
1833 | ret_bytes += (fiemap_s->fm_mapped_extents * | |
1834 | sizeof(struct ll_fiemap_extent)); | |
1835 | ||
02f9c12e | 1836 | if (copy_to_user((void __user *)arg, fiemap_s, ret_bytes)) |
d7e09d03 PT |
1837 | rc = -EFAULT; |
1838 | ||
1839 | error: | |
e958f49b | 1840 | kvfree(fiemap_s); |
0a3bdb00 | 1841 | return rc; |
d7e09d03 PT |
1842 | } |
1843 | ||
1844 | /* | |
1845 | * Read the data_version for inode. | |
1846 | * | |
1847 | * This value is computed using stripe object version on OST. | |
1848 | * Version is computed using server side locking. | |
1849 | * | |
e1798006 JX |
1850 | * @param sync if do sync on the OST side; |
1851 | * 0: no sync | |
1852 | * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs | |
1853 | * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs | |
d7e09d03 | 1854 | */ |
e1798006 | 1855 | int ll_data_version(struct inode *inode, __u64 *data_version, int flags) |
d7e09d03 PT |
1856 | { |
1857 | struct lov_stripe_md *lsm = NULL; | |
1858 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1859 | struct obdo *obdo = NULL; | |
1860 | int rc; | |
d7e09d03 PT |
1861 | |
1862 | /* If no stripe, we consider version is 0. */ | |
1863 | lsm = ccc_inode_lsm_get(inode); | |
5dd16419 | 1864 | if (!lsm_has_objects(lsm)) { |
d7e09d03 PT |
1865 | *data_version = 0; |
1866 | CDEBUG(D_INODE, "No object for inode\n"); | |
34e1f2bb JL |
1867 | rc = 0; |
1868 | goto out; | |
d7e09d03 PT |
1869 | } |
1870 | ||
496a51bd JL |
1871 | obdo = kzalloc(sizeof(*obdo), GFP_NOFS); |
1872 | if (!obdo) { | |
34e1f2bb JL |
1873 | rc = -ENOMEM; |
1874 | goto out; | |
1875 | } | |
d7e09d03 | 1876 | |
e1798006 | 1877 | rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, obdo, 0, flags); |
5dd16419 | 1878 | if (rc == 0) { |
d7e09d03 PT |
1879 | if (!(obdo->o_valid & OBD_MD_FLDATAVERSION)) |
1880 | rc = -EOPNOTSUPP; | |
1881 | else | |
1882 | *data_version = obdo->o_data_version; | |
1883 | } | |
1884 | ||
97903a26 | 1885 | kfree(obdo); |
5dd16419 | 1886 | out: |
d7e09d03 | 1887 | ccc_inode_lsm_put(inode, lsm); |
0a3bdb00 | 1888 | return rc; |
d7e09d03 PT |
1889 | } |
1890 | ||
48d23e61 JX |
1891 | /* |
1892 | * Trigger a HSM release request for the provided inode. | |
1893 | */ | |
1894 | int ll_hsm_release(struct inode *inode) | |
1895 | { | |
1896 | struct cl_env_nest nest; | |
1897 | struct lu_env *env; | |
1898 | struct obd_client_handle *och = NULL; | |
1899 | __u64 data_version = 0; | |
1900 | int rc; | |
1901 | ||
48d23e61 JX |
1902 | CDEBUG(D_INODE, "%s: Releasing file "DFID".\n", |
1903 | ll_get_fsname(inode->i_sb, NULL, 0), | |
1904 | PFID(&ll_i2info(inode)->lli_fid)); | |
1905 | ||
1906 | och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE); | |
34e1f2bb JL |
1907 | if (IS_ERR(och)) { |
1908 | rc = PTR_ERR(och); | |
1909 | goto out; | |
1910 | } | |
48d23e61 JX |
1911 | |
1912 | /* Grab latest data_version and [am]time values */ | |
e1798006 | 1913 | rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH); |
48d23e61 | 1914 | if (rc != 0) |
34e1f2bb | 1915 | goto out; |
48d23e61 JX |
1916 | |
1917 | env = cl_env_nested_get(&nest); | |
34e1f2bb JL |
1918 | if (IS_ERR(env)) { |
1919 | rc = PTR_ERR(env); | |
1920 | goto out; | |
1921 | } | |
48d23e61 | 1922 | |
d2995737 | 1923 | ll_merge_attr(env, inode); |
48d23e61 JX |
1924 | cl_env_nested_put(&nest, env); |
1925 | ||
1926 | /* Release the file. | |
1927 | * NB: lease lock handle is released in mdc_hsm_release_pack() because | |
c0894c6c OD |
1928 | * we still need it to pack l_remote_handle to MDT. |
1929 | */ | |
48d23e61 JX |
1930 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och, |
1931 | &data_version); | |
1932 | och = NULL; | |
1933 | ||
48d23e61 | 1934 | out: |
6e16818b | 1935 | if (och && !IS_ERR(och)) /* close the file */ |
48d23e61 JX |
1936 | ll_lease_close(och, inode, NULL); |
1937 | ||
1938 | return rc; | |
1939 | } | |
1940 | ||
d7e09d03 PT |
1941 | struct ll_swap_stack { |
1942 | struct iattr ia1, ia2; | |
1943 | __u64 dv1, dv2; | |
1944 | struct inode *inode1, *inode2; | |
1945 | bool check_dv1, check_dv2; | |
1946 | }; | |
1947 | ||
1948 | static int ll_swap_layouts(struct file *file1, struct file *file2, | |
1949 | struct lustre_swap_layouts *lsl) | |
1950 | { | |
1951 | struct mdc_swap_layouts msl; | |
1952 | struct md_op_data *op_data; | |
1953 | __u32 gid; | |
1954 | __u64 dv; | |
1955 | struct ll_swap_stack *llss = NULL; | |
1956 | int rc; | |
1957 | ||
496a51bd JL |
1958 | llss = kzalloc(sizeof(*llss), GFP_NOFS); |
1959 | if (!llss) | |
0a3bdb00 | 1960 | return -ENOMEM; |
d7e09d03 | 1961 | |
2a8a3597 AV |
1962 | llss->inode1 = file_inode(file1); |
1963 | llss->inode2 = file_inode(file2); | |
d7e09d03 | 1964 | |
34e1f2bb JL |
1965 | if (!S_ISREG(llss->inode2->i_mode)) { |
1966 | rc = -EINVAL; | |
1967 | goto free; | |
1968 | } | |
d7e09d03 | 1969 | |
9c5fb72c | 1970 | if (inode_permission(llss->inode1, MAY_WRITE) || |
34e1f2bb JL |
1971 | inode_permission(llss->inode2, MAY_WRITE)) { |
1972 | rc = -EPERM; | |
1973 | goto free; | |
1974 | } | |
d7e09d03 | 1975 | |
34e1f2bb JL |
1976 | if (llss->inode2->i_sb != llss->inode1->i_sb) { |
1977 | rc = -EXDEV; | |
1978 | goto free; | |
1979 | } | |
d7e09d03 PT |
1980 | |
1981 | /* we use 2 bool because it is easier to swap than 2 bits */ | |
1982 | if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1) | |
1983 | llss->check_dv1 = true; | |
1984 | ||
1985 | if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2) | |
1986 | llss->check_dv2 = true; | |
1987 | ||
1988 | /* we cannot use lsl->sl_dvX directly because we may swap them */ | |
1989 | llss->dv1 = lsl->sl_dv1; | |
1990 | llss->dv2 = lsl->sl_dv2; | |
1991 | ||
1992 | rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2)); | |
34e1f2bb JL |
1993 | if (rc == 0) /* same file, done! */ { |
1994 | rc = 0; | |
1995 | goto free; | |
1996 | } | |
d7e09d03 PT |
1997 | |
1998 | if (rc < 0) { /* sequentialize it */ | |
1999 | swap(llss->inode1, llss->inode2); | |
2000 | swap(file1, file2); | |
2001 | swap(llss->dv1, llss->dv2); | |
2002 | swap(llss->check_dv1, llss->check_dv2); | |
2003 | } | |
2004 | ||
2005 | gid = lsl->sl_gid; | |
2006 | if (gid != 0) { /* application asks to flush dirty cache */ | |
2007 | rc = ll_get_grouplock(llss->inode1, file1, gid); | |
2008 | if (rc < 0) | |
34e1f2bb | 2009 | goto free; |
d7e09d03 PT |
2010 | |
2011 | rc = ll_get_grouplock(llss->inode2, file2, gid); | |
2012 | if (rc < 0) { | |
2013 | ll_put_grouplock(llss->inode1, file1, gid); | |
34e1f2bb | 2014 | goto free; |
d7e09d03 PT |
2015 | } |
2016 | } | |
2017 | ||
2018 | /* to be able to restore mtime and atime after swap | |
c0894c6c OD |
2019 | * we need to first save them |
2020 | */ | |
d7e09d03 PT |
2021 | if (lsl->sl_flags & |
2022 | (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) { | |
2023 | llss->ia1.ia_mtime = llss->inode1->i_mtime; | |
2024 | llss->ia1.ia_atime = llss->inode1->i_atime; | |
2025 | llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME; | |
2026 | llss->ia2.ia_mtime = llss->inode2->i_mtime; | |
2027 | llss->ia2.ia_atime = llss->inode2->i_atime; | |
2028 | llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME; | |
2029 | } | |
2030 | ||
d0a0acc3 | 2031 | /* ultimate check, before swapping the layouts we check if |
c0894c6c OD |
2032 | * dataversion has changed (if requested) |
2033 | */ | |
d7e09d03 PT |
2034 | if (llss->check_dv1) { |
2035 | rc = ll_data_version(llss->inode1, &dv, 0); | |
2036 | if (rc) | |
34e1f2bb JL |
2037 | goto putgl; |
2038 | if (dv != llss->dv1) { | |
2039 | rc = -EAGAIN; | |
2040 | goto putgl; | |
2041 | } | |
d7e09d03 PT |
2042 | } |
2043 | ||
2044 | if (llss->check_dv2) { | |
2045 | rc = ll_data_version(llss->inode2, &dv, 0); | |
2046 | if (rc) | |
34e1f2bb JL |
2047 | goto putgl; |
2048 | if (dv != llss->dv2) { | |
2049 | rc = -EAGAIN; | |
2050 | goto putgl; | |
2051 | } | |
d7e09d03 PT |
2052 | } |
2053 | ||
2054 | /* struct md_op_data is used to send the swap args to the mdt | |
2055 | * only flags is missing, so we use struct mdc_swap_layouts | |
c0894c6c OD |
2056 | * through the md_op_data->op_data |
2057 | */ | |
d7e09d03 | 2058 | /* flags from user space have to be converted before they are send to |
c0894c6c OD |
2059 | * server, no flag is sent today, they are only used on the client |
2060 | */ | |
d7e09d03 PT |
2061 | msl.msl_flags = 0; |
2062 | rc = -ENOMEM; | |
2063 | op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0, | |
2064 | 0, LUSTRE_OPC_ANY, &msl); | |
34e1f2bb JL |
2065 | if (IS_ERR(op_data)) { |
2066 | rc = PTR_ERR(op_data); | |
2067 | goto free; | |
2068 | } | |
79a8726a JH |
2069 | |
2070 | rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1), | |
2071 | sizeof(*op_data), op_data, NULL); | |
2072 | ll_finish_md_op_data(op_data); | |
d7e09d03 PT |
2073 | |
2074 | putgl: | |
2075 | if (gid != 0) { | |
2076 | ll_put_grouplock(llss->inode2, file2, gid); | |
2077 | ll_put_grouplock(llss->inode1, file1, gid); | |
2078 | } | |
2079 | ||
2080 | /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */ | |
2081 | if (rc != 0) | |
34e1f2bb | 2082 | goto free; |
d7e09d03 PT |
2083 | |
2084 | /* clear useless flags */ | |
2085 | if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) { | |
2086 | llss->ia1.ia_valid &= ~ATTR_MTIME; | |
2087 | llss->ia2.ia_valid &= ~ATTR_MTIME; | |
2088 | } | |
2089 | ||
2090 | if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) { | |
2091 | llss->ia1.ia_valid &= ~ATTR_ATIME; | |
2092 | llss->ia2.ia_valid &= ~ATTR_ATIME; | |
2093 | } | |
2094 | ||
2095 | /* update time if requested */ | |
2096 | rc = 0; | |
2097 | if (llss->ia2.ia_valid != 0) { | |
5955102c | 2098 | inode_lock(llss->inode1); |
b583043e | 2099 | rc = ll_setattr(file1->f_path.dentry, &llss->ia2); |
5955102c | 2100 | inode_unlock(llss->inode1); |
d7e09d03 PT |
2101 | } |
2102 | ||
2103 | if (llss->ia1.ia_valid != 0) { | |
2104 | int rc1; | |
2105 | ||
5955102c | 2106 | inode_lock(llss->inode2); |
b583043e | 2107 | rc1 = ll_setattr(file2->f_path.dentry, &llss->ia1); |
5955102c | 2108 | inode_unlock(llss->inode2); |
d7e09d03 PT |
2109 | if (rc == 0) |
2110 | rc = rc1; | |
2111 | } | |
2112 | ||
2113 | free: | |
e6b9a3b2 | 2114 | kfree(llss); |
d7e09d03 | 2115 | |
0a3bdb00 | 2116 | return rc; |
d7e09d03 PT |
2117 | } |
2118 | ||
a720b790 JL |
2119 | static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss) |
2120 | { | |
2121 | struct md_op_data *op_data; | |
2122 | int rc; | |
2123 | ||
2742c75e BF |
2124 | /* Detect out-of range masks */ |
2125 | if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK) | |
2126 | return -EINVAL; | |
2127 | ||
a720b790 | 2128 | /* Non-root users are forbidden to set or clear flags which are |
c0894c6c OD |
2129 | * NOT defined in HSM_USER_MASK. |
2130 | */ | |
a720b790 | 2131 | if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) && |
2eb90a75 | 2132 | !capable(CFS_CAP_SYS_ADMIN)) |
a720b790 JL |
2133 | return -EPERM; |
2134 | ||
2742c75e BF |
2135 | /* Detect out-of range archive id */ |
2136 | if ((hss->hss_valid & HSS_ARCHIVE_ID) && | |
2137 | (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE)) | |
2138 | return -EINVAL; | |
2139 | ||
a720b790 JL |
2140 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, |
2141 | LUSTRE_OPC_ANY, hss); | |
2142 | if (IS_ERR(op_data)) | |
2143 | return PTR_ERR(op_data); | |
2144 | ||
2145 | rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode), | |
2146 | sizeof(*op_data), op_data, NULL); | |
2147 | ||
2148 | ll_finish_md_op_data(op_data); | |
2149 | ||
2150 | return rc; | |
2151 | } | |
2152 | ||
2153 | static int ll_hsm_import(struct inode *inode, struct file *file, | |
2154 | struct hsm_user_import *hui) | |
2155 | { | |
2156 | struct hsm_state_set *hss = NULL; | |
2157 | struct iattr *attr = NULL; | |
2158 | int rc; | |
2159 | ||
a720b790 JL |
2160 | if (!S_ISREG(inode->i_mode)) |
2161 | return -EINVAL; | |
2162 | ||
2163 | /* set HSM flags */ | |
496a51bd | 2164 | hss = kzalloc(sizeof(*hss), GFP_NOFS); |
e6b9a3b2 JL |
2165 | if (!hss) |
2166 | return -ENOMEM; | |
a720b790 JL |
2167 | |
2168 | hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID; | |
2169 | hss->hss_archive_id = hui->hui_archive_id; | |
2170 | hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED; | |
2171 | rc = ll_hsm_state_set(inode, hss); | |
2172 | if (rc != 0) | |
e6b9a3b2 | 2173 | goto free_hss; |
a720b790 | 2174 | |
496a51bd JL |
2175 | attr = kzalloc(sizeof(*attr), GFP_NOFS); |
2176 | if (!attr) { | |
34e1f2bb | 2177 | rc = -ENOMEM; |
e6b9a3b2 | 2178 | goto free_hss; |
34e1f2bb | 2179 | } |
a720b790 JL |
2180 | |
2181 | attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO); | |
2182 | attr->ia_mode |= S_IFREG; | |
2183 | attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid); | |
2184 | attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid); | |
2185 | attr->ia_size = hui->hui_size; | |
2186 | attr->ia_mtime.tv_sec = hui->hui_mtime; | |
2187 | attr->ia_mtime.tv_nsec = hui->hui_mtime_ns; | |
2188 | attr->ia_atime.tv_sec = hui->hui_atime; | |
2189 | attr->ia_atime.tv_nsec = hui->hui_atime_ns; | |
2190 | ||
2191 | attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE | | |
2192 | ATTR_UID | ATTR_GID | | |
2193 | ATTR_MTIME | ATTR_MTIME_SET | | |
2194 | ATTR_ATIME | ATTR_ATIME_SET; | |
2195 | ||
5955102c | 2196 | inode_lock(inode); |
b6ee56fe | 2197 | |
b583043e | 2198 | rc = ll_setattr_raw(file->f_path.dentry, attr, true); |
a720b790 JL |
2199 | if (rc == -ENODATA) |
2200 | rc = 0; | |
2201 | ||
5955102c | 2202 | inode_unlock(inode); |
b6ee56fe | 2203 | |
e6b9a3b2 JL |
2204 | kfree(attr); |
2205 | free_hss: | |
2206 | kfree(hss); | |
a720b790 JL |
2207 | return rc; |
2208 | } | |
2209 | ||
2d95f10e JH |
2210 | static long |
2211 | ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
d7e09d03 | 2212 | { |
2a8a3597 | 2213 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2214 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
2215 | int flags, rc; | |
d7e09d03 | 2216 | |
97a075cd JN |
2217 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),cmd=%x\n", |
2218 | PFID(ll_inode2fid(inode)), inode, cmd); | |
d7e09d03 PT |
2219 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1); |
2220 | ||
2221 | /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */ | |
2222 | if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */ | |
0a3bdb00 | 2223 | return -ENOTTY; |
d7e09d03 | 2224 | |
a58a38ac | 2225 | switch (cmd) { |
d7e09d03 PT |
2226 | case LL_IOC_GETFLAGS: |
2227 | /* Get the current value of the file flags */ | |
02f9c12e | 2228 | return put_user(fd->fd_flags, (int __user *)arg); |
d7e09d03 PT |
2229 | case LL_IOC_SETFLAGS: |
2230 | case LL_IOC_CLRFLAGS: | |
2231 | /* Set or clear specific file flags */ | |
2232 | /* XXX This probably needs checks to ensure the flags are | |
2233 | * not abused, and to handle any flag side effects. | |
2234 | */ | |
02f9c12e | 2235 | if (get_user(flags, (int __user *)arg)) |
0a3bdb00 | 2236 | return -EFAULT; |
d7e09d03 PT |
2237 | |
2238 | if (cmd == LL_IOC_SETFLAGS) { | |
2239 | if ((flags & LL_FILE_IGNORE_LOCK) && | |
2240 | !(file->f_flags & O_DIRECT)) { | |
2d00bd17 JP |
2241 | CERROR("%s: unable to disable locking on non-O_DIRECT file\n", |
2242 | current->comm); | |
0a3bdb00 | 2243 | return -EINVAL; |
d7e09d03 PT |
2244 | } |
2245 | ||
2246 | fd->fd_flags |= flags; | |
2247 | } else { | |
2248 | fd->fd_flags &= ~flags; | |
2249 | } | |
0a3bdb00 | 2250 | return 0; |
d7e09d03 | 2251 | case LL_IOC_LOV_SETSTRIPE: |
0a3bdb00 | 2252 | return ll_lov_setstripe(inode, file, arg); |
d7e09d03 | 2253 | case LL_IOC_LOV_SETEA: |
0a3bdb00 | 2254 | return ll_lov_setea(inode, file, arg); |
d7e09d03 PT |
2255 | case LL_IOC_LOV_SWAP_LAYOUTS: { |
2256 | struct file *file2; | |
2257 | struct lustre_swap_layouts lsl; | |
2258 | ||
02f9c12e OD |
2259 | if (copy_from_user(&lsl, (char __user *)arg, |
2260 | sizeof(struct lustre_swap_layouts))) | |
0a3bdb00 | 2261 | return -EFAULT; |
d7e09d03 PT |
2262 | |
2263 | if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */ | |
0a3bdb00 | 2264 | return -EPERM; |
d7e09d03 PT |
2265 | |
2266 | file2 = fget(lsl.sl_fd); | |
6e16818b | 2267 | if (!file2) |
0a3bdb00 | 2268 | return -EBADF; |
d7e09d03 PT |
2269 | |
2270 | rc = -EPERM; | |
2271 | if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */ | |
2272 | rc = ll_swap_layouts(file, file2, &lsl); | |
2273 | fput(file2); | |
0a3bdb00 | 2274 | return rc; |
d7e09d03 PT |
2275 | } |
2276 | case LL_IOC_LOV_GETSTRIPE: | |
0a3bdb00 | 2277 | return ll_lov_getstripe(inode, arg); |
d7e09d03 | 2278 | case LL_IOC_RECREATE_OBJ: |
0a3bdb00 | 2279 | return ll_lov_recreate_obj(inode, arg); |
d7e09d03 | 2280 | case LL_IOC_RECREATE_FID: |
0a3bdb00 | 2281 | return ll_lov_recreate_fid(inode, arg); |
d7e09d03 | 2282 | case FSFILT_IOC_FIEMAP: |
0a3bdb00 | 2283 | return ll_ioctl_fiemap(inode, arg); |
d7e09d03 PT |
2284 | case FSFILT_IOC_GETFLAGS: |
2285 | case FSFILT_IOC_SETFLAGS: | |
0a3bdb00 | 2286 | return ll_iocontrol(inode, file, cmd, arg); |
d7e09d03 PT |
2287 | case FSFILT_IOC_GETVERSION_OLD: |
2288 | case FSFILT_IOC_GETVERSION: | |
02f9c12e | 2289 | return put_user(inode->i_generation, (int __user *)arg); |
d7e09d03 | 2290 | case LL_IOC_GROUP_LOCK: |
0a3bdb00 | 2291 | return ll_get_grouplock(inode, file, arg); |
d7e09d03 | 2292 | case LL_IOC_GROUP_UNLOCK: |
0a3bdb00 | 2293 | return ll_put_grouplock(inode, file, arg); |
d7e09d03 | 2294 | case IOC_OBD_STATFS: |
4c6243ec | 2295 | return ll_obd_statfs(inode, (void __user *)arg); |
d7e09d03 PT |
2296 | |
2297 | /* We need to special case any other ioctls we want to handle, | |
2298 | * to send them to the MDS/OST as appropriate and to properly | |
2299 | * network encode the arg field. | |
2300 | case FSFILT_IOC_SETVERSION_OLD: | |
2301 | case FSFILT_IOC_SETVERSION: | |
2302 | */ | |
2303 | case LL_IOC_FLUSHCTX: | |
0a3bdb00 | 2304 | return ll_flush_ctx(inode); |
d7e09d03 | 2305 | case LL_IOC_PATH2FID: { |
02f9c12e | 2306 | if (copy_to_user((void __user *)arg, ll_inode2fid(inode), |
d7e09d03 | 2307 | sizeof(struct lu_fid))) |
0a3bdb00 | 2308 | return -EFAULT; |
d7e09d03 | 2309 | |
0a3bdb00 | 2310 | return 0; |
d7e09d03 PT |
2311 | } |
2312 | case OBD_IOC_FID2PATH: | |
61dad0ba | 2313 | return ll_fid2path(inode, (void __user *)arg); |
d7e09d03 PT |
2314 | case LL_IOC_DATA_VERSION: { |
2315 | struct ioc_data_version idv; | |
2316 | int rc; | |
2317 | ||
02f9c12e | 2318 | if (copy_from_user(&idv, (char __user *)arg, sizeof(idv))) |
0a3bdb00 | 2319 | return -EFAULT; |
d7e09d03 | 2320 | |
e1798006 JX |
2321 | idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH; |
2322 | rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags); | |
02f9c12e OD |
2323 | if (rc == 0 && copy_to_user((char __user *)arg, &idv, |
2324 | sizeof(idv))) | |
0a3bdb00 | 2325 | return -EFAULT; |
d7e09d03 | 2326 | |
0a3bdb00 | 2327 | return rc; |
d7e09d03 PT |
2328 | } |
2329 | ||
2330 | case LL_IOC_GET_MDTIDX: { | |
2331 | int mdtidx; | |
2332 | ||
2333 | mdtidx = ll_get_mdt_idx(inode); | |
2334 | if (mdtidx < 0) | |
0a3bdb00 | 2335 | return mdtidx; |
d7e09d03 | 2336 | |
02f9c12e | 2337 | if (put_user(mdtidx, (int __user *)arg)) |
0a3bdb00 | 2338 | return -EFAULT; |
d7e09d03 | 2339 | |
0a3bdb00 | 2340 | return 0; |
d7e09d03 PT |
2341 | } |
2342 | case OBD_IOC_GETDTNAME: | |
2343 | case OBD_IOC_GETMDNAME: | |
0a3bdb00 | 2344 | return ll_get_obd_name(inode, cmd, arg); |
d7e09d03 PT |
2345 | case LL_IOC_HSM_STATE_GET: { |
2346 | struct md_op_data *op_data; | |
2347 | struct hsm_user_state *hus; | |
2348 | int rc; | |
2349 | ||
496a51bd JL |
2350 | hus = kzalloc(sizeof(*hus), GFP_NOFS); |
2351 | if (!hus) | |
0a3bdb00 | 2352 | return -ENOMEM; |
d7e09d03 PT |
2353 | |
2354 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2355 | LUSTRE_OPC_ANY, hus); | |
79a8726a | 2356 | if (IS_ERR(op_data)) { |
97903a26 | 2357 | kfree(hus); |
0a3bdb00 | 2358 | return PTR_ERR(op_data); |
d7e09d03 PT |
2359 | } |
2360 | ||
2361 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2362 | op_data, NULL); | |
2363 | ||
02f9c12e | 2364 | if (copy_to_user((void __user *)arg, hus, sizeof(*hus))) |
d7e09d03 PT |
2365 | rc = -EFAULT; |
2366 | ||
2367 | ll_finish_md_op_data(op_data); | |
97903a26 | 2368 | kfree(hus); |
0a3bdb00 | 2369 | return rc; |
d7e09d03 PT |
2370 | } |
2371 | case LL_IOC_HSM_STATE_SET: { | |
d7e09d03 PT |
2372 | struct hsm_state_set *hss; |
2373 | int rc; | |
2374 | ||
02f9c12e | 2375 | hss = memdup_user((char __user *)arg, sizeof(*hss)); |
0c027bc3 AH |
2376 | if (IS_ERR(hss)) |
2377 | return PTR_ERR(hss); | |
d7e09d03 | 2378 | |
a720b790 | 2379 | rc = ll_hsm_state_set(inode, hss); |
d7e09d03 | 2380 | |
97903a26 | 2381 | kfree(hss); |
0a3bdb00 | 2382 | return rc; |
d7e09d03 PT |
2383 | } |
2384 | case LL_IOC_HSM_ACTION: { | |
2385 | struct md_op_data *op_data; | |
2386 | struct hsm_current_action *hca; | |
2387 | int rc; | |
2388 | ||
496a51bd JL |
2389 | hca = kzalloc(sizeof(*hca), GFP_NOFS); |
2390 | if (!hca) | |
0a3bdb00 | 2391 | return -ENOMEM; |
d7e09d03 PT |
2392 | |
2393 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2394 | LUSTRE_OPC_ANY, hca); | |
79a8726a | 2395 | if (IS_ERR(op_data)) { |
97903a26 | 2396 | kfree(hca); |
0a3bdb00 | 2397 | return PTR_ERR(op_data); |
d7e09d03 PT |
2398 | } |
2399 | ||
2400 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2401 | op_data, NULL); | |
2402 | ||
02f9c12e | 2403 | if (copy_to_user((char __user *)arg, hca, sizeof(*hca))) |
d7e09d03 PT |
2404 | rc = -EFAULT; |
2405 | ||
2406 | ll_finish_md_op_data(op_data); | |
97903a26 | 2407 | kfree(hca); |
0a3bdb00 | 2408 | return rc; |
d7e09d03 | 2409 | } |
d3a8a4e2 JX |
2410 | case LL_IOC_SET_LEASE: { |
2411 | struct ll_inode_info *lli = ll_i2info(inode); | |
2412 | struct obd_client_handle *och = NULL; | |
2413 | bool lease_broken; | |
2414 | fmode_t mode = 0; | |
2415 | ||
2416 | switch (arg) { | |
2417 | case F_WRLCK: | |
2418 | if (!(file->f_mode & FMODE_WRITE)) | |
2419 | return -EPERM; | |
2420 | mode = FMODE_WRITE; | |
2421 | break; | |
2422 | case F_RDLCK: | |
2423 | if (!(file->f_mode & FMODE_READ)) | |
2424 | return -EPERM; | |
2425 | mode = FMODE_READ; | |
2426 | break; | |
2427 | case F_UNLCK: | |
2428 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2429 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
2430 | och = fd->fd_lease_och; |
2431 | fd->fd_lease_och = NULL; | |
2432 | } | |
2433 | mutex_unlock(&lli->lli_och_mutex); | |
2434 | ||
6e16818b | 2435 | if (och) { |
d3a8a4e2 JX |
2436 | mode = och->och_flags & |
2437 | (FMODE_READ|FMODE_WRITE); | |
2438 | rc = ll_lease_close(och, inode, &lease_broken); | |
2439 | if (rc == 0 && lease_broken) | |
2440 | mode = 0; | |
2441 | } else { | |
2442 | rc = -ENOLCK; | |
2443 | } | |
2444 | ||
2445 | /* return the type of lease or error */ | |
2446 | return rc < 0 ? rc : (int)mode; | |
2447 | default: | |
2448 | return -EINVAL; | |
2449 | } | |
2450 | ||
2451 | CDEBUG(D_INODE, "Set lease with mode %d\n", mode); | |
2452 | ||
2453 | /* apply for lease */ | |
48d23e61 | 2454 | och = ll_lease_open(inode, file, mode, 0); |
d3a8a4e2 JX |
2455 | if (IS_ERR(och)) |
2456 | return PTR_ERR(och); | |
2457 | ||
2458 | rc = 0; | |
2459 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2460 | if (!fd->fd_lease_och) { |
d3a8a4e2 JX |
2461 | fd->fd_lease_och = och; |
2462 | och = NULL; | |
2463 | } | |
2464 | mutex_unlock(&lli->lli_och_mutex); | |
6e16818b | 2465 | if (och) { |
d3a8a4e2 JX |
2466 | /* impossible now that only excl is supported for now */ |
2467 | ll_lease_close(och, inode, &lease_broken); | |
2468 | rc = -EBUSY; | |
2469 | } | |
2470 | return rc; | |
2471 | } | |
2472 | case LL_IOC_GET_LEASE: { | |
2473 | struct ll_inode_info *lli = ll_i2info(inode); | |
2474 | struct ldlm_lock *lock = NULL; | |
2475 | ||
2476 | rc = 0; | |
2477 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2478 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
2479 | struct obd_client_handle *och = fd->fd_lease_och; |
2480 | ||
2481 | lock = ldlm_handle2lock(&och->och_lease_handle); | |
6e16818b | 2482 | if (lock) { |
d3a8a4e2 JX |
2483 | lock_res_and_lock(lock); |
2484 | if (!ldlm_is_cancel(lock)) | |
2485 | rc = och->och_flags & | |
2486 | (FMODE_READ | FMODE_WRITE); | |
2487 | unlock_res_and_lock(lock); | |
ead02808 | 2488 | LDLM_LOCK_PUT(lock); |
d3a8a4e2 JX |
2489 | } |
2490 | } | |
2491 | mutex_unlock(&lli->lli_och_mutex); | |
a720b790 JL |
2492 | return rc; |
2493 | } | |
2494 | case LL_IOC_HSM_IMPORT: { | |
2495 | struct hsm_user_import *hui; | |
2496 | ||
02f9c12e | 2497 | hui = memdup_user((void __user *)arg, sizeof(*hui)); |
0c027bc3 AH |
2498 | if (IS_ERR(hui)) |
2499 | return PTR_ERR(hui); | |
a720b790 JL |
2500 | |
2501 | rc = ll_hsm_import(inode, file, hui); | |
d3a8a4e2 | 2502 | |
97903a26 | 2503 | kfree(hui); |
d3a8a4e2 JX |
2504 | return rc; |
2505 | } | |
d7e09d03 PT |
2506 | default: { |
2507 | int err; | |
2508 | ||
1f6eaf83 JL |
2509 | if (ll_iocontrol_call(inode, file, cmd, arg, &err) == |
2510 | LLIOC_STOP) | |
0a3bdb00 | 2511 | return err; |
d7e09d03 | 2512 | |
0a3bdb00 | 2513 | return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, |
e09bee34 | 2514 | (void __user *)arg); |
d7e09d03 PT |
2515 | } |
2516 | } | |
2517 | } | |
2518 | ||
2d95f10e | 2519 | static loff_t ll_file_seek(struct file *file, loff_t offset, int origin) |
d7e09d03 | 2520 | { |
2a8a3597 | 2521 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2522 | loff_t retval, eof = 0; |
2523 | ||
d7e09d03 PT |
2524 | retval = offset + ((origin == SEEK_END) ? i_size_read(inode) : |
2525 | (origin == SEEK_CUR) ? file->f_pos : 0); | |
97a075cd JN |
2526 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n", |
2527 | PFID(ll_inode2fid(inode)), inode, retval, retval, origin); | |
d7e09d03 PT |
2528 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); |
2529 | ||
2530 | if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) { | |
2531 | retval = ll_glimpse_size(inode); | |
2532 | if (retval != 0) | |
0a3bdb00 | 2533 | return retval; |
d7e09d03 PT |
2534 | eof = i_size_read(inode); |
2535 | } | |
2536 | ||
6f014339 | 2537 | retval = generic_file_llseek_size(file, offset, origin, |
d7e09d03 | 2538 | ll_file_maxbytes(inode), eof); |
0a3bdb00 | 2539 | return retval; |
d7e09d03 PT |
2540 | } |
2541 | ||
2d95f10e | 2542 | static int ll_flush(struct file *file, fl_owner_t id) |
d7e09d03 | 2543 | { |
2a8a3597 | 2544 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2545 | struct ll_inode_info *lli = ll_i2info(inode); |
2546 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
2547 | int rc, err; | |
2548 | ||
2549 | LASSERT(!S_ISDIR(inode->i_mode)); | |
2550 | ||
2551 | /* catch async errors that were recorded back when async writeback | |
c0894c6c OD |
2552 | * failed for pages in this mapping. |
2553 | */ | |
d7e09d03 PT |
2554 | rc = lli->lli_async_rc; |
2555 | lli->lli_async_rc = 0; | |
2556 | err = lov_read_and_clear_async_rc(lli->lli_clob); | |
2557 | if (rc == 0) | |
2558 | rc = err; | |
2559 | ||
c0894c6c OD |
2560 | /* The application has been told about write failure already. |
2561 | * Do not report failure again. | |
2562 | */ | |
d7e09d03 PT |
2563 | if (fd->fd_write_failed) |
2564 | return 0; | |
2565 | return rc ? -EIO : 0; | |
2566 | } | |
2567 | ||
2568 | /** | |
2569 | * Called to make sure a portion of file has been written out. | |
05289927 | 2570 | * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST. |
d7e09d03 PT |
2571 | * |
2572 | * Return how many pages have been written. | |
2573 | */ | |
2574 | int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end, | |
65fb55d1 | 2575 | enum cl_fsync_mode mode, int ignore_layout) |
d7e09d03 PT |
2576 | { |
2577 | struct cl_env_nest nest; | |
2578 | struct lu_env *env; | |
2579 | struct cl_io *io; | |
d7e09d03 PT |
2580 | struct cl_fsync_io *fio; |
2581 | int result; | |
d7e09d03 PT |
2582 | |
2583 | if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL && | |
2584 | mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL) | |
0a3bdb00 | 2585 | return -EINVAL; |
d7e09d03 PT |
2586 | |
2587 | env = cl_env_nested_get(&nest); | |
2588 | if (IS_ERR(env)) | |
0a3bdb00 | 2589 | return PTR_ERR(env); |
d7e09d03 | 2590 | |
9acc4500 | 2591 | io = vvp_env_thread_io(env); |
1929c433 | 2592 | io->ci_obj = ll_i2info(inode)->lli_clob; |
65fb55d1 | 2593 | io->ci_ignore_layout = ignore_layout; |
d7e09d03 PT |
2594 | |
2595 | /* initialize parameters for sync */ | |
2596 | fio = &io->u.ci_fsync; | |
d7e09d03 PT |
2597 | fio->fi_start = start; |
2598 | fio->fi_end = end; | |
2599 | fio->fi_fid = ll_inode2fid(inode); | |
2600 | fio->fi_mode = mode; | |
2601 | fio->fi_nr_written = 0; | |
2602 | ||
2603 | if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0) | |
2604 | result = cl_io_loop(env, io); | |
2605 | else | |
2606 | result = io->ci_result; | |
2607 | if (result == 0) | |
2608 | result = fio->fi_nr_written; | |
2609 | cl_io_fini(env, io); | |
2610 | cl_env_nested_put(&nest, env); | |
2611 | ||
0a3bdb00 | 2612 | return result; |
d7e09d03 PT |
2613 | } |
2614 | ||
d7e09d03 PT |
2615 | int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
2616 | { | |
2a8a3597 | 2617 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2618 | struct ll_inode_info *lli = ll_i2info(inode); |
2619 | struct ptlrpc_request *req; | |
d7e09d03 | 2620 | int rc, err; |
d7e09d03 | 2621 | |
97a075cd JN |
2622 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n", |
2623 | PFID(ll_inode2fid(inode)), inode); | |
d7e09d03 PT |
2624 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1); |
2625 | ||
2626 | rc = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
5955102c | 2627 | inode_lock(inode); |
d7e09d03 PT |
2628 | |
2629 | /* catch async errors that were recorded back when async writeback | |
c0894c6c OD |
2630 | * failed for pages in this mapping. |
2631 | */ | |
d7e09d03 PT |
2632 | if (!S_ISDIR(inode->i_mode)) { |
2633 | err = lli->lli_async_rc; | |
2634 | lli->lli_async_rc = 0; | |
2635 | if (rc == 0) | |
2636 | rc = err; | |
2637 | err = lov_read_and_clear_async_rc(lli->lli_clob); | |
2638 | if (rc == 0) | |
2639 | rc = err; | |
2640 | } | |
2641 | ||
ef2e0f55 | 2642 | err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req); |
d7e09d03 PT |
2643 | if (!rc) |
2644 | rc = err; | |
2645 | if (!err) | |
2646 | ptlrpc_req_finished(req); | |
2647 | ||
8d97deb9 | 2648 | if (S_ISREG(inode->i_mode)) { |
d7e09d03 PT |
2649 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
2650 | ||
05289927 | 2651 | err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0); |
d7e09d03 PT |
2652 | if (rc == 0 && err < 0) |
2653 | rc = err; | |
2654 | if (rc < 0) | |
2655 | fd->fd_write_failed = true; | |
2656 | else | |
2657 | fd->fd_write_failed = false; | |
2658 | } | |
2659 | ||
5955102c | 2660 | inode_unlock(inode); |
0a3bdb00 | 2661 | return rc; |
d7e09d03 PT |
2662 | } |
2663 | ||
2d95f10e JH |
2664 | static int |
2665 | ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) | |
d7e09d03 | 2666 | { |
2a8a3597 | 2667 | struct inode *inode = file_inode(file); |
d7e09d03 | 2668 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
f2145eae BK |
2669 | struct ldlm_enqueue_info einfo = { |
2670 | .ei_type = LDLM_FLOCK, | |
2671 | .ei_cb_cp = ldlm_flock_completion_ast, | |
2672 | .ei_cbdata = file_lock, | |
2673 | }; | |
d7e09d03 PT |
2674 | struct md_op_data *op_data; |
2675 | struct lustre_handle lockh = {0}; | |
8369cfff | 2676 | ldlm_policy_data_t flock = { {0} }; |
875332d4 | 2677 | __u64 flags = 0; |
d7e09d03 PT |
2678 | int rc; |
2679 | int rc2 = 0; | |
d7e09d03 | 2680 | |
97a075cd JN |
2681 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n", |
2682 | PFID(ll_inode2fid(inode)), file_lock); | |
d7e09d03 PT |
2683 | |
2684 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1); | |
2685 | ||
130d1f95 | 2686 | if (file_lock->fl_flags & FL_FLOCK) |
d7e09d03 | 2687 | LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK)); |
130d1f95 | 2688 | else if (!(file_lock->fl_flags & FL_POSIX)) |
0a3bdb00 | 2689 | return -EINVAL; |
130d1f95 JL |
2690 | |
2691 | flock.l_flock.owner = (unsigned long)file_lock->fl_owner; | |
d7e09d03 | 2692 | flock.l_flock.pid = file_lock->fl_pid; |
130d1f95 JL |
2693 | flock.l_flock.start = file_lock->fl_start; |
2694 | flock.l_flock.end = file_lock->fl_end; | |
d7e09d03 PT |
2695 | |
2696 | /* Somewhat ugly workaround for svc lockd. | |
2697 | * lockd installs custom fl_lmops->lm_compare_owner that checks | |
2698 | * for the fl_owner to be the same (which it always is on local node | |
2699 | * I guess between lockd processes) and then compares pid. | |
2700 | * As such we assign pid to the owner field to make it all work, | |
2701 | * conflict with normal locks is unlikely since pid space and | |
c0894c6c OD |
2702 | * pointer space for current->files are not intersecting |
2703 | */ | |
d7e09d03 PT |
2704 | if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner) |
2705 | flock.l_flock.owner = (unsigned long)file_lock->fl_pid; | |
2706 | ||
2707 | switch (file_lock->fl_type) { | |
2708 | case F_RDLCK: | |
2709 | einfo.ei_mode = LCK_PR; | |
2710 | break; | |
2711 | case F_UNLCK: | |
2712 | /* An unlock request may or may not have any relation to | |
2713 | * existing locks so we may not be able to pass a lock handle | |
2714 | * via a normal ldlm_lock_cancel() request. The request may even | |
2715 | * unlock a byte range in the middle of an existing lock. In | |
2716 | * order to process an unlock request we need all of the same | |
2717 | * information that is given with a normal read or write record | |
2718 | * lock request. To avoid creating another ldlm unlock (cancel) | |
c0894c6c OD |
2719 | * message we'll treat a LCK_NL flock request as an unlock. |
2720 | */ | |
d7e09d03 PT |
2721 | einfo.ei_mode = LCK_NL; |
2722 | break; | |
2723 | case F_WRLCK: | |
2724 | einfo.ei_mode = LCK_PW; | |
2725 | break; | |
2726 | default: | |
2727 | CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", | |
e15ba45d | 2728 | file_lock->fl_type); |
0a3bdb00 | 2729 | return -ENOTSUPP; |
d7e09d03 PT |
2730 | } |
2731 | ||
2732 | switch (cmd) { | |
2733 | case F_SETLKW: | |
2faedcd5 | 2734 | #ifdef F_SETLKW64 |
d7e09d03 | 2735 | case F_SETLKW64: |
2faedcd5 | 2736 | #endif |
d7e09d03 PT |
2737 | flags = 0; |
2738 | break; | |
2739 | case F_SETLK: | |
2faedcd5 | 2740 | #ifdef F_SETLK64 |
d7e09d03 | 2741 | case F_SETLK64: |
2faedcd5 | 2742 | #endif |
d7e09d03 PT |
2743 | flags = LDLM_FL_BLOCK_NOWAIT; |
2744 | break; | |
2745 | case F_GETLK: | |
2faedcd5 | 2746 | #ifdef F_GETLK64 |
d7e09d03 | 2747 | case F_GETLK64: |
2faedcd5 | 2748 | #endif |
d7e09d03 PT |
2749 | flags = LDLM_FL_TEST_LOCK; |
2750 | /* Save the old mode so that if the mode in the lock changes we | |
c0894c6c OD |
2751 | * can decrement the appropriate reader or writer refcount. |
2752 | */ | |
d7e09d03 PT |
2753 | file_lock->fl_type = einfo.ei_mode; |
2754 | break; | |
2755 | default: | |
2756 | CERROR("unknown fcntl lock command: %d\n", cmd); | |
0a3bdb00 | 2757 | return -EINVAL; |
d7e09d03 PT |
2758 | } |
2759 | ||
2760 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2761 | LUSTRE_OPC_ANY, NULL); | |
2762 | if (IS_ERR(op_data)) | |
0a3bdb00 | 2763 | return PTR_ERR(op_data); |
d7e09d03 | 2764 | |
97a075cd JN |
2765 | CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n", |
2766 | PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags, | |
2767 | einfo.ei_mode, flock.l_flock.start, flock.l_flock.end); | |
d7e09d03 PT |
2768 | |
2769 | rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, | |
2770 | op_data, &lockh, &flock, 0, NULL /* req */, flags); | |
2771 | ||
4f656367 | 2772 | if ((rc == 0 || file_lock->fl_type == F_UNLCK) && |
d7e09d03 | 2773 | !(flags & LDLM_FL_TEST_LOCK)) |
4f656367 | 2774 | rc2 = locks_lock_file_wait(file, file_lock); |
d7e09d03 PT |
2775 | |
2776 | if (rc2 && file_lock->fl_type != F_UNLCK) { | |
2777 | einfo.ei_mode = LCK_NL; | |
2778 | md_enqueue(sbi->ll_md_exp, &einfo, NULL, | |
e15ba45d | 2779 | op_data, &lockh, &flock, 0, NULL /* req */, flags); |
d7e09d03 PT |
2780 | rc = rc2; |
2781 | } | |
2782 | ||
2783 | ll_finish_md_op_data(op_data); | |
2784 | ||
0a3bdb00 | 2785 | return rc; |
d7e09d03 PT |
2786 | } |
2787 | ||
2d95f10e JH |
2788 | static int |
2789 | ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock) | |
d7e09d03 | 2790 | { |
0a3bdb00 | 2791 | return -ENOSYS; |
d7e09d03 PT |
2792 | } |
2793 | ||
2794 | /** | |
2795 | * test if some locks matching bits and l_req_mode are acquired | |
2796 | * - bits can be in different locks | |
2797 | * - if found clear the common lock bits in *bits | |
2798 | * - the bits not found, are kept in *bits | |
2799 | * \param inode [IN] | |
2800 | * \param bits [IN] searched lock bits [IN] | |
2801 | * \param l_req_mode [IN] searched lock mode | |
2802 | * \retval boolean, true iff all bits are found | |
2803 | */ | |
52ee0d20 OD |
2804 | int ll_have_md_lock(struct inode *inode, __u64 *bits, |
2805 | enum ldlm_mode l_req_mode) | |
d7e09d03 PT |
2806 | { |
2807 | struct lustre_handle lockh; | |
2808 | ldlm_policy_data_t policy; | |
52ee0d20 | 2809 | enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ? |
d7e09d03 PT |
2810 | (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode; |
2811 | struct lu_fid *fid; | |
2812 | __u64 flags; | |
2813 | int i; | |
d7e09d03 PT |
2814 | |
2815 | if (!inode) | |
ef075edc | 2816 | return 0; |
d7e09d03 PT |
2817 | |
2818 | fid = &ll_i2info(inode)->lli_fid; | |
2819 | CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid), | |
2820 | ldlm_lockname[mode]); | |
2821 | ||
2822 | flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK; | |
1253b2e8 | 2823 | for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) { |
d7e09d03 PT |
2824 | policy.l_inodebits.bits = *bits & (1 << i); |
2825 | if (policy.l_inodebits.bits == 0) | |
2826 | continue; | |
2827 | ||
2828 | if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, | |
2829 | &policy, mode, &lockh)) { | |
2830 | struct ldlm_lock *lock; | |
2831 | ||
2832 | lock = ldlm_handle2lock(&lockh); | |
2833 | if (lock) { | |
2834 | *bits &= | |
2835 | ~(lock->l_policy_data.l_inodebits.bits); | |
2836 | LDLM_LOCK_PUT(lock); | |
2837 | } else { | |
2838 | *bits &= ~policy.l_inodebits.bits; | |
2839 | } | |
2840 | } | |
2841 | } | |
0a3bdb00 | 2842 | return *bits == 0; |
d7e09d03 PT |
2843 | } |
2844 | ||
52ee0d20 OD |
2845 | enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits, |
2846 | struct lustre_handle *lockh, __u64 flags, | |
2847 | enum ldlm_mode mode) | |
d7e09d03 | 2848 | { |
57303e76 | 2849 | ldlm_policy_data_t policy = { .l_inodebits = {bits} }; |
d7e09d03 | 2850 | struct lu_fid *fid; |
52ee0d20 | 2851 | enum ldlm_mode rc; |
d7e09d03 PT |
2852 | |
2853 | fid = &ll_i2info(inode)->lli_fid; | |
2854 | CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid)); | |
2855 | ||
1f6eaf83 | 2856 | rc = md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED, |
7fc1f831 AP |
2857 | fid, LDLM_IBITS, &policy, mode, lockh); |
2858 | ||
0a3bdb00 | 2859 | return rc; |
d7e09d03 PT |
2860 | } |
2861 | ||
2862 | static int ll_inode_revalidate_fini(struct inode *inode, int rc) | |
2863 | { | |
2864 | /* Already unlinked. Just update nlink and return success */ | |
2865 | if (rc == -ENOENT) { | |
2866 | clear_nlink(inode); | |
2867 | /* This path cannot be hit for regular files unless in | |
bef31c78 MI |
2868 | * case of obscure races, so no need to validate size. |
2869 | */ | |
d7e09d03 PT |
2870 | if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode)) |
2871 | return 0; | |
2872 | } else if (rc != 0) { | |
e49634bb AD |
2873 | CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR, |
2874 | "%s: revalidate FID "DFID" error: rc = %d\n", | |
2875 | ll_get_fsname(inode->i_sb, NULL, 0), | |
2876 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
2877 | } |
2878 | ||
2879 | return rc; | |
2880 | } | |
2881 | ||
2d95f10e | 2882 | static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits) |
d7e09d03 | 2883 | { |
2b0143b5 | 2884 | struct inode *inode = d_inode(dentry); |
d7e09d03 PT |
2885 | struct ptlrpc_request *req = NULL; |
2886 | struct obd_export *exp; | |
2887 | int rc = 0; | |
d7e09d03 | 2888 | |
97a075cd JN |
2889 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%pd\n", |
2890 | PFID(ll_inode2fid(inode)), inode, dentry); | |
d7e09d03 PT |
2891 | |
2892 | exp = ll_i2mdexp(inode); | |
2893 | ||
2894 | /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC. | |
2895 | * But under CMD case, it caused some lock issues, should be fixed | |
c0894c6c OD |
2896 | * with new CMD ibits lock. See bug 12718 |
2897 | */ | |
d7e09d03 PT |
2898 | if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) { |
2899 | struct lookup_intent oit = { .it_op = IT_GETATTR }; | |
2900 | struct md_op_data *op_data; | |
2901 | ||
2902 | if (ibits == MDS_INODELOCK_LOOKUP) | |
2903 | oit.it_op = IT_LOOKUP; | |
2904 | ||
2905 | /* Call getattr by fid, so do not provide name at all. */ | |
dbca51dd AV |
2906 | op_data = ll_prep_md_op_data(NULL, inode, |
2907 | inode, NULL, 0, 0, | |
d7e09d03 PT |
2908 | LUSTRE_OPC_ANY, NULL); |
2909 | if (IS_ERR(op_data)) | |
0a3bdb00 | 2910 | return PTR_ERR(op_data); |
d7e09d03 PT |
2911 | |
2912 | oit.it_create_mode |= M_CHECK_STALE; | |
2913 | rc = md_intent_lock(exp, op_data, NULL, 0, | |
2914 | /* we are not interested in name | |
c0894c6c OD |
2915 | * based lookup |
2916 | */ | |
d7e09d03 PT |
2917 | &oit, 0, &req, |
2918 | ll_md_blocking_ast, 0); | |
2919 | ll_finish_md_op_data(op_data); | |
2920 | oit.it_create_mode &= ~M_CHECK_STALE; | |
2921 | if (rc < 0) { | |
2922 | rc = ll_inode_revalidate_fini(inode, rc); | |
34e1f2bb | 2923 | goto out; |
d7e09d03 PT |
2924 | } |
2925 | ||
dbca51dd | 2926 | rc = ll_revalidate_it_finish(req, &oit, inode); |
d7e09d03 PT |
2927 | if (rc != 0) { |
2928 | ll_intent_release(&oit); | |
34e1f2bb | 2929 | goto out; |
d7e09d03 PT |
2930 | } |
2931 | ||
2932 | /* Unlinked? Unhash dentry, so it is not picked up later by | |
c0894c6c OD |
2933 | * do_lookup() -> ll_revalidate_it(). We cannot use d_drop |
2934 | * here to preserve get_cwd functionality on 2.6. | |
2935 | * Bug 10503 | |
2936 | */ | |
c9cc8d0f BF |
2937 | if (!d_inode(dentry)->i_nlink) { |
2938 | spin_lock(&inode->i_lock); | |
b1d2a127 | 2939 | d_lustre_invalidate(dentry, 0); |
c9cc8d0f BF |
2940 | spin_unlock(&inode->i_lock); |
2941 | } | |
d7e09d03 | 2942 | |
dbca51dd | 2943 | ll_lookup_finish_locks(&oit, inode); |
2b0143b5 DH |
2944 | } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) { |
2945 | struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry)); | |
21aef7d9 | 2946 | u64 valid = OBD_MD_FLGETATTR; |
d7e09d03 PT |
2947 | struct md_op_data *op_data; |
2948 | int ealen = 0; | |
2949 | ||
2950 | if (S_ISREG(inode->i_mode)) { | |
44779340 | 2951 | rc = ll_get_default_mdsize(sbi, &ealen); |
d7e09d03 | 2952 | if (rc) |
0a3bdb00 | 2953 | return rc; |
d7e09d03 PT |
2954 | valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE; |
2955 | } | |
2956 | ||
2957 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, | |
2958 | 0, ealen, LUSTRE_OPC_ANY, | |
2959 | NULL); | |
2960 | if (IS_ERR(op_data)) | |
0a3bdb00 | 2961 | return PTR_ERR(op_data); |
d7e09d03 PT |
2962 | |
2963 | op_data->op_valid = valid; | |
d7e09d03 PT |
2964 | rc = md_getattr(sbi->ll_md_exp, op_data, &req); |
2965 | ll_finish_md_op_data(op_data); | |
2966 | if (rc) { | |
2967 | rc = ll_inode_revalidate_fini(inode, rc); | |
0a3bdb00 | 2968 | return rc; |
d7e09d03 PT |
2969 | } |
2970 | ||
2971 | rc = ll_prep_inode(&inode, req, NULL, NULL); | |
2972 | } | |
2973 | out: | |
2974 | ptlrpc_req_finished(req); | |
2975 | return rc; | |
2976 | } | |
2977 | ||
2d95f10e | 2978 | static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits) |
d7e09d03 | 2979 | { |
2b0143b5 | 2980 | struct inode *inode = d_inode(dentry); |
d7e09d03 | 2981 | int rc; |
d7e09d03 | 2982 | |
2d95f10e | 2983 | rc = __ll_inode_revalidate(dentry, ibits); |
d7e09d03 | 2984 | if (rc != 0) |
0a3bdb00 | 2985 | return rc; |
d7e09d03 PT |
2986 | |
2987 | /* if object isn't regular file, don't validate size */ | |
2988 | if (!S_ISREG(inode->i_mode)) { | |
d2995737 JH |
2989 | LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime; |
2990 | LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime; | |
2991 | LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime; | |
d7e09d03 | 2992 | } else { |
5ea17d6c JL |
2993 | /* In case of restore, the MDT has the right size and has |
2994 | * already send it back without granting the layout lock, | |
2995 | * inode is up-to-date so glimpse is useless. | |
2996 | * Also to glimpse we need the layout, in case of a running | |
2997 | * restore the MDT holds the layout lock so the glimpse will | |
2998 | * block up to the end of restore (getattr will block) | |
2999 | */ | |
3000 | if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING)) | |
3001 | rc = ll_glimpse_size(inode); | |
d7e09d03 | 3002 | } |
0a3bdb00 | 3003 | return rc; |
d7e09d03 PT |
3004 | } |
3005 | ||
2d95f10e | 3006 | int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat) |
d7e09d03 | 3007 | { |
2b0143b5 | 3008 | struct inode *inode = d_inode(de); |
d7e09d03 PT |
3009 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
3010 | struct ll_inode_info *lli = ll_i2info(inode); | |
f82ced5d | 3011 | int res; |
d7e09d03 | 3012 | |
2d95f10e JH |
3013 | res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE | |
3014 | MDS_INODELOCK_LOOKUP); | |
d7e09d03 PT |
3015 | ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1); |
3016 | ||
3017 | if (res) | |
3018 | return res; | |
3019 | ||
3020 | stat->dev = inode->i_sb->s_dev; | |
3021 | if (ll_need_32bit_api(sbi)) | |
3022 | stat->ino = cl_fid_build_ino(&lli->lli_fid, 1); | |
3023 | else | |
3024 | stat->ino = inode->i_ino; | |
3025 | stat->mode = inode->i_mode; | |
3026 | stat->nlink = inode->i_nlink; | |
3027 | stat->uid = inode->i_uid; | |
3028 | stat->gid = inode->i_gid; | |
3029 | stat->rdev = inode->i_rdev; | |
3030 | stat->atime = inode->i_atime; | |
3031 | stat->mtime = inode->i_mtime; | |
3032 | stat->ctime = inode->i_ctime; | |
3033 | stat->blksize = 1 << inode->i_blkbits; | |
3034 | ||
3035 | stat->size = i_size_read(inode); | |
3036 | stat->blocks = inode->i_blocks; | |
3037 | ||
3038 | return 0; | |
3039 | } | |
d7e09d03 | 3040 | |
2d95f10e JH |
3041 | static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3042 | __u64 start, __u64 len) | |
89580e37 PT |
3043 | { |
3044 | int rc; | |
3045 | size_t num_bytes; | |
3046 | struct ll_user_fiemap *fiemap; | |
3047 | unsigned int extent_count = fieinfo->fi_extents_max; | |
3048 | ||
3049 | num_bytes = sizeof(*fiemap) + (extent_count * | |
3050 | sizeof(struct ll_fiemap_extent)); | |
e958f49b | 3051 | fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS); |
89580e37 | 3052 | |
6e16818b | 3053 | if (!fiemap) |
89580e37 PT |
3054 | return -ENOMEM; |
3055 | ||
3056 | fiemap->fm_flags = fieinfo->fi_flags; | |
3057 | fiemap->fm_extent_count = fieinfo->fi_extents_max; | |
3058 | fiemap->fm_start = start; | |
3059 | fiemap->fm_length = len; | |
97514241 OD |
3060 | if (extent_count > 0 && |
3061 | copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start, | |
3062 | sizeof(struct ll_fiemap_extent)) != 0) { | |
3063 | rc = -EFAULT; | |
3064 | goto out; | |
3065 | } | |
89580e37 PT |
3066 | |
3067 | rc = ll_do_fiemap(inode, fiemap, num_bytes); | |
3068 | ||
3069 | fieinfo->fi_flags = fiemap->fm_flags; | |
3070 | fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents; | |
97514241 OD |
3071 | if (extent_count > 0 && |
3072 | copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0], | |
3073 | fiemap->fm_mapped_extents * | |
3074 | sizeof(struct ll_fiemap_extent)) != 0) { | |
3075 | rc = -EFAULT; | |
3076 | goto out; | |
3077 | } | |
89580e37 | 3078 | |
97514241 | 3079 | out: |
e958f49b | 3080 | kvfree(fiemap); |
89580e37 PT |
3081 | return rc; |
3082 | } | |
d7e09d03 | 3083 | |
2d95f10e | 3084 | struct posix_acl *ll_get_acl(struct inode *inode, int type) |
d7e09d03 PT |
3085 | { |
3086 | struct ll_inode_info *lli = ll_i2info(inode); | |
3087 | struct posix_acl *acl = NULL; | |
d7e09d03 PT |
3088 | |
3089 | spin_lock(&lli->lli_lock); | |
3090 | /* VFS' acl_permission_check->check_acl will release the refcount */ | |
3091 | acl = posix_acl_dup(lli->lli_posix_acl); | |
ed7bdf5c | 3092 | #ifdef CONFIG_FS_POSIX_ACL |
b788dc51 | 3093 | forget_cached_acl(inode, type); |
ed7bdf5c | 3094 | #endif |
d7e09d03 PT |
3095 | spin_unlock(&lli->lli_lock); |
3096 | ||
0a3bdb00 | 3097 | return acl; |
d7e09d03 PT |
3098 | } |
3099 | ||
d7e09d03 PT |
3100 | int ll_inode_permission(struct inode *inode, int mask) |
3101 | { | |
3102 | int rc = 0; | |
d7e09d03 | 3103 | |
d7e09d03 PT |
3104 | if (mask & MAY_NOT_BLOCK) |
3105 | return -ECHILD; | |
d7e09d03 PT |
3106 | |
3107 | /* as root inode are NOT getting validated in lookup operation, | |
c0894c6c OD |
3108 | * need to do it before permission check. |
3109 | */ | |
d7e09d03 | 3110 | |
f76c23da | 3111 | if (is_root_inode(inode)) { |
2d95f10e JH |
3112 | rc = __ll_inode_revalidate(inode->i_sb->s_root, |
3113 | MDS_INODELOCK_LOOKUP); | |
d7e09d03 | 3114 | if (rc) |
0a3bdb00 | 3115 | return rc; |
d7e09d03 PT |
3116 | } |
3117 | ||
97a075cd JN |
3118 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n", |
3119 | PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask); | |
d7e09d03 | 3120 | |
d7e09d03 | 3121 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); |
8707c96e | 3122 | rc = generic_permission(inode, mask); |
d7e09d03 | 3123 | |
0a3bdb00 | 3124 | return rc; |
d7e09d03 PT |
3125 | } |
3126 | ||
d7e09d03 PT |
3127 | /* -o localflock - only provides locally consistent flock locks */ |
3128 | struct file_operations ll_file_operations = { | |
b42b15fd | 3129 | .read_iter = ll_file_read_iter, |
b42b15fd | 3130 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3131 | .unlocked_ioctl = ll_file_ioctl, |
3132 | .open = ll_file_open, | |
3133 | .release = ll_file_release, | |
3134 | .mmap = ll_file_mmap, | |
3135 | .llseek = ll_file_seek, | |
82c156f8 | 3136 | .splice_read = generic_file_splice_read, |
d7e09d03 PT |
3137 | .fsync = ll_fsync, |
3138 | .flush = ll_flush | |
3139 | }; | |
3140 | ||
3141 | struct file_operations ll_file_operations_flock = { | |
b42b15fd | 3142 | .read_iter = ll_file_read_iter, |
b42b15fd | 3143 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3144 | .unlocked_ioctl = ll_file_ioctl, |
3145 | .open = ll_file_open, | |
3146 | .release = ll_file_release, | |
3147 | .mmap = ll_file_mmap, | |
3148 | .llseek = ll_file_seek, | |
82c156f8 | 3149 | .splice_read = generic_file_splice_read, |
d7e09d03 PT |
3150 | .fsync = ll_fsync, |
3151 | .flush = ll_flush, | |
3152 | .flock = ll_file_flock, | |
3153 | .lock = ll_file_flock | |
3154 | }; | |
3155 | ||
3156 | /* These are for -o noflock - to return ENOSYS on flock calls */ | |
3157 | struct file_operations ll_file_operations_noflock = { | |
b42b15fd | 3158 | .read_iter = ll_file_read_iter, |
b42b15fd | 3159 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3160 | .unlocked_ioctl = ll_file_ioctl, |
3161 | .open = ll_file_open, | |
3162 | .release = ll_file_release, | |
3163 | .mmap = ll_file_mmap, | |
3164 | .llseek = ll_file_seek, | |
82c156f8 | 3165 | .splice_read = generic_file_splice_read, |
d7e09d03 PT |
3166 | .fsync = ll_fsync, |
3167 | .flush = ll_flush, | |
3168 | .flock = ll_file_noflock, | |
3169 | .lock = ll_file_noflock | |
3170 | }; | |
3171 | ||
d2d32738 | 3172 | const struct inode_operations ll_file_inode_operations = { |
d7e09d03 PT |
3173 | .setattr = ll_setattr, |
3174 | .getattr = ll_getattr, | |
3175 | .permission = ll_inode_permission, | |
3176 | .setxattr = ll_setxattr, | |
3177 | .getxattr = ll_getxattr, | |
3178 | .listxattr = ll_listxattr, | |
3179 | .removexattr = ll_removexattr, | |
89580e37 | 3180 | .fiemap = ll_fiemap, |
d7e09d03 PT |
3181 | .get_acl = ll_get_acl, |
3182 | }; | |
3183 | ||
d0a0acc3 | 3184 | /* dynamic ioctl number support routines */ |
d7e09d03 PT |
3185 | static struct llioc_ctl_data { |
3186 | struct rw_semaphore ioc_sem; | |
3187 | struct list_head ioc_head; | |
3188 | } llioc = { | |
3189 | __RWSEM_INITIALIZER(llioc.ioc_sem), | |
3190 | LIST_HEAD_INIT(llioc.ioc_head) | |
3191 | }; | |
3192 | ||
d7e09d03 PT |
3193 | struct llioc_data { |
3194 | struct list_head iocd_list; | |
3195 | unsigned int iocd_size; | |
3196 | llioc_callback_t iocd_cb; | |
3197 | unsigned int iocd_count; | |
3198 | unsigned int iocd_cmd[0]; | |
3199 | }; | |
3200 | ||
3201 | void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd) | |
3202 | { | |
3203 | unsigned int size; | |
3204 | struct llioc_data *in_data = NULL; | |
d7e09d03 | 3205 | |
6e16818b | 3206 | if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0) |
0a3bdb00 | 3207 | return NULL; |
d7e09d03 PT |
3208 | |
3209 | size = sizeof(*in_data) + count * sizeof(unsigned int); | |
496a51bd JL |
3210 | in_data = kzalloc(size, GFP_NOFS); |
3211 | if (!in_data) | |
0a3bdb00 | 3212 | return NULL; |
d7e09d03 PT |
3213 | |
3214 | memset(in_data, 0, sizeof(*in_data)); | |
3215 | in_data->iocd_size = size; | |
3216 | in_data->iocd_cb = cb; | |
3217 | in_data->iocd_count = count; | |
3218 | memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count); | |
3219 | ||
3220 | down_write(&llioc.ioc_sem); | |
3221 | list_add_tail(&in_data->iocd_list, &llioc.ioc_head); | |
3222 | up_write(&llioc.ioc_sem); | |
3223 | ||
0a3bdb00 | 3224 | return in_data; |
d7e09d03 | 3225 | } |
93133eb4 | 3226 | EXPORT_SYMBOL(ll_iocontrol_register); |
d7e09d03 PT |
3227 | |
3228 | void ll_iocontrol_unregister(void *magic) | |
3229 | { | |
3230 | struct llioc_data *tmp; | |
3231 | ||
6e16818b | 3232 | if (!magic) |
d7e09d03 PT |
3233 | return; |
3234 | ||
3235 | down_write(&llioc.ioc_sem); | |
3236 | list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) { | |
3237 | if (tmp == magic) { | |
d7e09d03 PT |
3238 | list_del(&tmp->iocd_list); |
3239 | up_write(&llioc.ioc_sem); | |
3240 | ||
97903a26 | 3241 | kfree(tmp); |
d7e09d03 PT |
3242 | return; |
3243 | } | |
3244 | } | |
3245 | up_write(&llioc.ioc_sem); | |
3246 | ||
3247 | CWARN("didn't find iocontrol register block with magic: %p\n", magic); | |
3248 | } | |
d7e09d03 PT |
3249 | EXPORT_SYMBOL(ll_iocontrol_unregister); |
3250 | ||
2d95f10e JH |
3251 | static enum llioc_iter |
3252 | ll_iocontrol_call(struct inode *inode, struct file *file, | |
3253 | unsigned int cmd, unsigned long arg, int *rcp) | |
d7e09d03 PT |
3254 | { |
3255 | enum llioc_iter ret = LLIOC_CONT; | |
3256 | struct llioc_data *data; | |
3257 | int rc = -EINVAL, i; | |
3258 | ||
3259 | down_read(&llioc.ioc_sem); | |
3260 | list_for_each_entry(data, &llioc.ioc_head, iocd_list) { | |
3261 | for (i = 0; i < data->iocd_count; i++) { | |
3262 | if (cmd != data->iocd_cmd[i]) | |
3263 | continue; | |
3264 | ||
3265 | ret = data->iocd_cb(inode, file, cmd, arg, data, &rc); | |
3266 | break; | |
3267 | } | |
3268 | ||
3269 | if (ret == LLIOC_STOP) | |
3270 | break; | |
3271 | } | |
3272 | up_read(&llioc.ioc_sem); | |
3273 | ||
3274 | if (rcp) | |
3275 | *rcp = rc; | |
3276 | return ret; | |
3277 | } | |
3278 | ||
3279 | int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf) | |
3280 | { | |
3281 | struct ll_inode_info *lli = ll_i2info(inode); | |
3282 | struct cl_env_nest nest; | |
3283 | struct lu_env *env; | |
3284 | int result; | |
d7e09d03 | 3285 | |
6e16818b | 3286 | if (!lli->lli_clob) |
0a3bdb00 | 3287 | return 0; |
d7e09d03 PT |
3288 | |
3289 | env = cl_env_nested_get(&nest); | |
3290 | if (IS_ERR(env)) | |
0a3bdb00 | 3291 | return PTR_ERR(env); |
d7e09d03 PT |
3292 | |
3293 | result = cl_conf_set(env, lli->lli_clob, conf); | |
3294 | cl_env_nested_put(&nest, env); | |
3295 | ||
3296 | if (conf->coc_opc == OBJECT_CONF_SET) { | |
3297 | struct ldlm_lock *lock = conf->coc_lock; | |
3298 | ||
6e16818b | 3299 | LASSERT(lock); |
d7e09d03 PT |
3300 | LASSERT(ldlm_has_layout(lock)); |
3301 | if (result == 0) { | |
3302 | /* it can only be allowed to match after layout is | |
3303 | * applied to inode otherwise false layout would be | |
d0a0acc3 | 3304 | * seen. Applying layout should happen before dropping |
c0894c6c OD |
3305 | * the intent lock. |
3306 | */ | |
d7e09d03 PT |
3307 | ldlm_lock_allow_match(lock); |
3308 | } | |
3309 | } | |
0a3bdb00 | 3310 | return result; |
d7e09d03 PT |
3311 | } |
3312 | ||
3313 | /* Fetch layout from MDT with getxattr request, if it's not ready yet */ | |
3314 | static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock) | |
3315 | ||
3316 | { | |
3317 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
d7e09d03 PT |
3318 | struct ptlrpc_request *req; |
3319 | struct mdt_body *body; | |
3320 | void *lvbdata; | |
3321 | void *lmm; | |
3322 | int lmmsize; | |
3323 | int rc; | |
d7e09d03 | 3324 | |
e2335e5d | 3325 | CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n", |
5a9a80ba | 3326 | PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock), |
e2335e5d | 3327 | lock->l_lvb_data, lock->l_lvb_len); |
3328 | ||
5a9a80ba | 3329 | if (lock->l_lvb_data && ldlm_is_lvb_ready(lock)) |
0a3bdb00 | 3330 | return 0; |
d7e09d03 PT |
3331 | |
3332 | /* if layout lock was granted right away, the layout is returned | |
3333 | * within DLM_LVB of dlm reply; otherwise if the lock was ever | |
3334 | * blocked and then granted via completion ast, we have to fetch | |
3335 | * layout here. Please note that we can't use the LVB buffer in | |
c0894c6c OD |
3336 | * completion AST because it doesn't have a large enough buffer |
3337 | */ | |
44779340 | 3338 | rc = ll_get_default_mdsize(sbi, &lmmsize); |
d7e09d03 | 3339 | if (rc == 0) |
ef2e0f55 OD |
3340 | rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), |
3341 | OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0, | |
3342 | lmmsize, 0, &req); | |
d7e09d03 | 3343 | if (rc < 0) |
0a3bdb00 | 3344 | return rc; |
d7e09d03 PT |
3345 | |
3346 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
6e16818b | 3347 | if (!body) { |
34e1f2bb JL |
3348 | rc = -EPROTO; |
3349 | goto out; | |
3350 | } | |
d7e09d03 PT |
3351 | |
3352 | lmmsize = body->eadatasize; | |
34e1f2bb JL |
3353 | if (lmmsize == 0) /* empty layout */ { |
3354 | rc = 0; | |
3355 | goto out; | |
3356 | } | |
d7e09d03 PT |
3357 | |
3358 | lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize); | |
6e16818b | 3359 | if (!lmm) { |
34e1f2bb JL |
3360 | rc = -EFAULT; |
3361 | goto out; | |
3362 | } | |
d7e09d03 | 3363 | |
e958f49b | 3364 | lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS); |
6e16818b | 3365 | if (!lvbdata) { |
34e1f2bb JL |
3366 | rc = -ENOMEM; |
3367 | goto out; | |
3368 | } | |
d7e09d03 PT |
3369 | |
3370 | memcpy(lvbdata, lmm, lmmsize); | |
3371 | lock_res_and_lock(lock); | |
6e16818b | 3372 | if (lock->l_lvb_data) |
e958f49b | 3373 | kvfree(lock->l_lvb_data); |
e2335e5d | 3374 | |
3375 | lock->l_lvb_data = lvbdata; | |
3376 | lock->l_lvb_len = lmmsize; | |
d7e09d03 PT |
3377 | unlock_res_and_lock(lock); |
3378 | ||
d7e09d03 PT |
3379 | out: |
3380 | ptlrpc_req_finished(req); | |
3381 | return rc; | |
3382 | } | |
3383 | ||
3384 | /** | |
3385 | * Apply the layout to the inode. Layout lock is held and will be released | |
3386 | * in this function. | |
3387 | */ | |
52ee0d20 OD |
3388 | static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode, |
3389 | struct inode *inode, __u32 *gen, bool reconf) | |
d7e09d03 PT |
3390 | { |
3391 | struct ll_inode_info *lli = ll_i2info(inode); | |
3392 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
3393 | struct ldlm_lock *lock; | |
3394 | struct lustre_md md = { NULL }; | |
3395 | struct cl_object_conf conf; | |
3396 | int rc = 0; | |
3397 | bool lvb_ready; | |
3398 | bool wait_layout = false; | |
d7e09d03 PT |
3399 | |
3400 | LASSERT(lustre_handle_is_used(lockh)); | |
3401 | ||
3402 | lock = ldlm_handle2lock(lockh); | |
6e16818b | 3403 | LASSERT(lock); |
d7e09d03 PT |
3404 | LASSERT(ldlm_has_layout(lock)); |
3405 | ||
97a075cd JN |
3406 | LDLM_DEBUG(lock, "File "DFID"(%p) being reconfigured: %d", |
3407 | PFID(&lli->lli_fid), inode, reconf); | |
d7e09d03 | 3408 | |
bc969176 JL |
3409 | /* in case this is a caching lock and reinstate with new inode */ |
3410 | md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL); | |
3411 | ||
d7e09d03 | 3412 | lock_res_and_lock(lock); |
5a9a80ba | 3413 | lvb_ready = ldlm_is_lvb_ready(lock); |
d7e09d03 PT |
3414 | unlock_res_and_lock(lock); |
3415 | /* checking lvb_ready is racy but this is okay. The worst case is | |
c0894c6c OD |
3416 | * that multi processes may configure the file on the same time. |
3417 | */ | |
d7e09d03 PT |
3418 | if (lvb_ready || !reconf) { |
3419 | rc = -ENODATA; | |
3420 | if (lvb_ready) { | |
3421 | /* layout_gen must be valid if layout lock is not | |
c0894c6c OD |
3422 | * cancelled and stripe has already set |
3423 | */ | |
09aed8a5 | 3424 | *gen = ll_layout_version_get(lli); |
d7e09d03 PT |
3425 | rc = 0; |
3426 | } | |
34e1f2bb | 3427 | goto out; |
d7e09d03 PT |
3428 | } |
3429 | ||
3430 | rc = ll_layout_fetch(inode, lock); | |
3431 | if (rc < 0) | |
34e1f2bb | 3432 | goto out; |
d7e09d03 PT |
3433 | |
3434 | /* for layout lock, lmm is returned in lock's lvb. | |
3435 | * lvb_data is immutable if the lock is held so it's safe to access it | |
3436 | * without res lock. See the description in ldlm_lock_decref_internal() | |
c0894c6c OD |
3437 | * for the condition to free lvb_data of layout lock |
3438 | */ | |
6e16818b | 3439 | if (lock->l_lvb_data) { |
d7e09d03 PT |
3440 | rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm, |
3441 | lock->l_lvb_data, lock->l_lvb_len); | |
3442 | if (rc >= 0) { | |
3443 | *gen = LL_LAYOUT_GEN_EMPTY; | |
6e16818b | 3444 | if (md.lsm) |
d7e09d03 PT |
3445 | *gen = md.lsm->lsm_layout_gen; |
3446 | rc = 0; | |
3447 | } else { | |
e15ba45d OD |
3448 | CERROR("%s: file " DFID " unpackmd error: %d\n", |
3449 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3450 | PFID(&lli->lli_fid), rc); | |
d7e09d03 PT |
3451 | } |
3452 | } | |
3453 | if (rc < 0) | |
34e1f2bb | 3454 | goto out; |
d7e09d03 PT |
3455 | |
3456 | /* set layout to file. Unlikely this will fail as old layout was | |
c0894c6c OD |
3457 | * surely eliminated |
3458 | */ | |
ec83e611 | 3459 | memset(&conf, 0, sizeof(conf)); |
d7e09d03 PT |
3460 | conf.coc_opc = OBJECT_CONF_SET; |
3461 | conf.coc_inode = inode; | |
3462 | conf.coc_lock = lock; | |
3463 | conf.u.coc_md = &md; | |
3464 | rc = ll_layout_conf(inode, &conf); | |
3465 | ||
6e16818b | 3466 | if (md.lsm) |
d7e09d03 PT |
3467 | obd_free_memmd(sbi->ll_dt_exp, &md.lsm); |
3468 | ||
3469 | /* refresh layout failed, need to wait */ | |
3470 | wait_layout = rc == -EBUSY; | |
d7e09d03 PT |
3471 | |
3472 | out: | |
3473 | LDLM_LOCK_PUT(lock); | |
3474 | ldlm_lock_decref(lockh, mode); | |
3475 | ||
3476 | /* wait for IO to complete if it's still being used. */ | |
3477 | if (wait_layout) { | |
97a075cd | 3478 | CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n", |
e15ba45d | 3479 | ll_get_fsname(inode->i_sb, NULL, 0), |
97a075cd | 3480 | PFID(&lli->lli_fid), inode); |
d7e09d03 | 3481 | |
ec83e611 | 3482 | memset(&conf, 0, sizeof(conf)); |
d7e09d03 PT |
3483 | conf.coc_opc = OBJECT_CONF_WAIT; |
3484 | conf.coc_inode = inode; | |
3485 | rc = ll_layout_conf(inode, &conf); | |
3486 | if (rc == 0) | |
3487 | rc = -EAGAIN; | |
3488 | ||
97a075cd JN |
3489 | CDEBUG(D_INODE, "%s: file="DFID" waiting layout return: %d.\n", |
3490 | ll_get_fsname(inode->i_sb, NULL, 0), | |
e15ba45d | 3491 | PFID(&lli->lli_fid), rc); |
d7e09d03 | 3492 | } |
0a3bdb00 | 3493 | return rc; |
d7e09d03 PT |
3494 | } |
3495 | ||
3496 | /** | |
3497 | * This function checks if there exists a LAYOUT lock on the client side, | |
3498 | * or enqueues it if it doesn't have one in cache. | |
3499 | * | |
3500 | * This function will not hold layout lock so it may be revoked any time after | |
3501 | * this function returns. Any operations depend on layout should be redone | |
3502 | * in that case. | |
3503 | * | |
3504 | * This function should be called before lov_io_init() to get an uptodate | |
3505 | * layout version, the caller should save the version number and after IO | |
3506 | * is finished, this function should be called again to verify that layout | |
3507 | * is not changed during IO time. | |
3508 | */ | |
3509 | int ll_layout_refresh(struct inode *inode, __u32 *gen) | |
3510 | { | |
3511 | struct ll_inode_info *lli = ll_i2info(inode); | |
3512 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
3513 | struct md_op_data *op_data; | |
3514 | struct lookup_intent it; | |
3515 | struct lustre_handle lockh; | |
52ee0d20 | 3516 | enum ldlm_mode mode; |
f2145eae BK |
3517 | struct ldlm_enqueue_info einfo = { |
3518 | .ei_type = LDLM_IBITS, | |
3519 | .ei_mode = LCK_CR, | |
3520 | .ei_cb_bl = ll_md_blocking_ast, | |
3521 | .ei_cb_cp = ldlm_completion_ast, | |
3522 | }; | |
d7e09d03 | 3523 | int rc; |
d7e09d03 | 3524 | |
09aed8a5 JX |
3525 | *gen = ll_layout_version_get(lli); |
3526 | if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE) | |
0a3bdb00 | 3527 | return 0; |
d7e09d03 PT |
3528 | |
3529 | /* sanity checks */ | |
3530 | LASSERT(fid_is_sane(ll_inode2fid(inode))); | |
3531 | LASSERT(S_ISREG(inode->i_mode)); | |
3532 | ||
d7e09d03 PT |
3533 | /* take layout lock mutex to enqueue layout lock exclusively. */ |
3534 | mutex_lock(&lli->lli_layout_mutex); | |
3535 | ||
3536 | again: | |
09aed8a5 | 3537 | /* mostly layout lock is caching on the local side, so try to match |
c0894c6c OD |
3538 | * it before grabbing layout lock mutex. |
3539 | */ | |
7fc1f831 AP |
3540 | mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0, |
3541 | LCK_CR | LCK_CW | LCK_PR | LCK_PW); | |
d7e09d03 PT |
3542 | if (mode != 0) { /* hit cached lock */ |
3543 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); | |
3544 | if (rc == -EAGAIN) | |
3545 | goto again; | |
3546 | ||
3547 | mutex_unlock(&lli->lli_layout_mutex); | |
0a3bdb00 | 3548 | return rc; |
d7e09d03 PT |
3549 | } |
3550 | ||
3551 | op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, | |
e15ba45d | 3552 | 0, 0, LUSTRE_OPC_ANY, NULL); |
d7e09d03 PT |
3553 | if (IS_ERR(op_data)) { |
3554 | mutex_unlock(&lli->lli_layout_mutex); | |
0a3bdb00 | 3555 | return PTR_ERR(op_data); |
d7e09d03 PT |
3556 | } |
3557 | ||
3558 | /* have to enqueue one */ | |
3559 | memset(&it, 0, sizeof(it)); | |
3560 | it.it_op = IT_LAYOUT; | |
3561 | lockh.cookie = 0ULL; | |
3562 | ||
97a075cd JN |
3563 | LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)", |
3564 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3565 | PFID(&lli->lli_fid), inode); | |
d7e09d03 PT |
3566 | |
3567 | rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh, | |
3568 | NULL, 0, NULL, 0); | |
8bf86fd9 JH |
3569 | ptlrpc_req_finished(it.it_request); |
3570 | it.it_request = NULL; | |
d7e09d03 PT |
3571 | |
3572 | ll_finish_md_op_data(op_data); | |
3573 | ||
e476f2e5 JH |
3574 | mode = it.it_lock_mode; |
3575 | it.it_lock_mode = 0; | |
d7e09d03 PT |
3576 | ll_intent_drop_lock(&it); |
3577 | ||
3578 | if (rc == 0) { | |
3579 | /* set lock data in case this is a new lock */ | |
3580 | ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); | |
3581 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); | |
3582 | if (rc == -EAGAIN) | |
3583 | goto again; | |
3584 | } | |
3585 | mutex_unlock(&lli->lli_layout_mutex); | |
3586 | ||
0a3bdb00 | 3587 | return rc; |
d7e09d03 | 3588 | } |
5ea17d6c JL |
3589 | |
3590 | /** | |
3591 | * This function send a restore request to the MDT | |
3592 | */ | |
1b1594da | 3593 | int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length) |
5ea17d6c JL |
3594 | { |
3595 | struct hsm_user_request *hur; | |
3596 | int len, rc; | |
3597 | ||
3598 | len = sizeof(struct hsm_user_request) + | |
3599 | sizeof(struct hsm_user_item); | |
496a51bd JL |
3600 | hur = kzalloc(len, GFP_NOFS); |
3601 | if (!hur) | |
5ea17d6c JL |
3602 | return -ENOMEM; |
3603 | ||
3604 | hur->hur_request.hr_action = HUA_RESTORE; | |
3605 | hur->hur_request.hr_archive_id = 0; | |
3606 | hur->hur_request.hr_flags = 0; | |
3607 | memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid, | |
3608 | sizeof(hur->hur_user_item[0].hui_fid)); | |
1b1594da JX |
3609 | hur->hur_user_item[0].hui_extent.offset = offset; |
3610 | hur->hur_user_item[0].hui_extent.length = length; | |
5ea17d6c | 3611 | hur->hur_request.hr_itemcount = 1; |
1929c433 | 3612 | rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp, |
5ea17d6c | 3613 | len, hur, NULL); |
97903a26 | 3614 | kfree(hur); |
5ea17d6c JL |
3615 | return rc; |
3616 | } |