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> | |
c948390f | 44 | #include <linux/sched.h> |
bb41292b | 45 | #include <linux/mount.h> |
d7e09d03 | 46 | #include "llite_internal.h" |
67a235f5 | 47 | #include "../include/lustre/ll_fiemap.h" |
8877d3bf | 48 | #include "../include/lustre/lustre_ioctl.h" |
d7e09d03 | 49 | |
67a235f5 | 50 | #include "../include/cl_object.h" |
d7e09d03 | 51 | |
2d95f10e JH |
52 | static int |
53 | ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg); | |
54 | ||
55 | static int ll_lease_close(struct obd_client_handle *och, struct inode *inode, | |
56 | bool *lease_broken); | |
57 | ||
58 | static enum llioc_iter | |
59 | ll_iocontrol_call(struct inode *inode, struct file *file, | |
60 | unsigned int cmd, unsigned long arg, int *rcp); | |
61 | ||
62 | static struct ll_file_data *ll_file_data_get(void) | |
d7e09d03 PT |
63 | { |
64 | struct ll_file_data *fd; | |
65 | ||
21068c46 | 66 | fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS); |
6e16818b | 67 | if (!fd) |
73863d83 | 68 | return NULL; |
d7e09d03 PT |
69 | fd->fd_write_failed = false; |
70 | return fd; | |
71 | } | |
72 | ||
73 | static void ll_file_data_put(struct ll_file_data *fd) | |
74 | { | |
6e16818b | 75 | if (fd) |
50d30362 | 76 | kmem_cache_free(ll_file_data_slab, fd); |
d7e09d03 PT |
77 | } |
78 | ||
79 | void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data, | |
80 | struct lustre_handle *fh) | |
81 | { | |
82 | op_data->op_fid1 = ll_i2info(inode)->lli_fid; | |
83 | op_data->op_attr.ia_mode = inode->i_mode; | |
84 | op_data->op_attr.ia_atime = inode->i_atime; | |
85 | op_data->op_attr.ia_mtime = inode->i_mtime; | |
86 | op_data->op_attr.ia_ctime = inode->i_ctime; | |
87 | op_data->op_attr.ia_size = i_size_read(inode); | |
88 | op_data->op_attr_blocks = inode->i_blocks; | |
bb41292b | 89 | op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags); |
d7e09d03 PT |
90 | op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch; |
91 | if (fh) | |
92 | op_data->op_handle = *fh; | |
d7e09d03 | 93 | |
1f6eaf83 | 94 | if (ll_i2info(inode)->lli_flags & LLIF_DATA_MODIFIED) |
d7e09d03 PT |
95 | op_data->op_bias |= MDS_DATA_MODIFIED; |
96 | } | |
97 | ||
98 | /** | |
99 | * Closes the IO epoch and packs all the attributes into @op_data for | |
100 | * the CLOSE rpc. | |
101 | */ | |
102 | static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data, | |
103 | struct obd_client_handle *och) | |
104 | { | |
f57d9a72 EL |
105 | op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET | |
106 | ATTR_MTIME | ATTR_MTIME_SET | | |
107 | ATTR_CTIME | ATTR_CTIME_SET; | |
d7e09d03 PT |
108 | |
109 | if (!(och->och_flags & FMODE_WRITE)) | |
110 | goto out; | |
111 | ||
112 | if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode)) | |
113 | op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS; | |
114 | else | |
115 | ll_ioepoch_close(inode, op_data, &och, 0); | |
116 | ||
117 | out: | |
118 | ll_pack_inode2opdata(inode, op_data, &och->och_fh); | |
119 | ll_prep_md_op_data(op_data, inode, NULL, NULL, | |
120 | 0, 0, LUSTRE_OPC_ANY, NULL); | |
d7e09d03 PT |
121 | } |
122 | ||
123 | static int ll_close_inode_openhandle(struct obd_export *md_exp, | |
124 | struct inode *inode, | |
48d23e61 JX |
125 | struct obd_client_handle *och, |
126 | const __u64 *data_version) | |
d7e09d03 PT |
127 | { |
128 | struct obd_export *exp = ll_i2mdexp(inode); | |
129 | struct md_op_data *op_data; | |
130 | struct ptlrpc_request *req = NULL; | |
131 | struct obd_device *obd = class_exp2obd(exp); | |
132 | int epoch_close = 1; | |
133 | int rc; | |
d7e09d03 | 134 | |
6e16818b | 135 | if (!obd) { |
d7e09d03 PT |
136 | /* |
137 | * XXX: in case of LMV, is this correct to access | |
138 | * ->exp_handle? | |
139 | */ | |
55f5a824 | 140 | CERROR("Invalid MDC connection handle %#llx\n", |
d7e09d03 | 141 | ll_i2mdexp(inode)->exp_handle.h_cookie); |
34e1f2bb JL |
142 | rc = 0; |
143 | goto out; | |
d7e09d03 PT |
144 | } |
145 | ||
496a51bd JL |
146 | op_data = kzalloc(sizeof(*op_data), GFP_NOFS); |
147 | if (!op_data) { | |
34e1f2bb JL |
148 | /* XXX We leak openhandle and request here. */ |
149 | rc = -ENOMEM; | |
150 | goto out; | |
151 | } | |
d7e09d03 PT |
152 | |
153 | ll_prepare_close(inode, op_data, och); | |
6e16818b | 154 | if (data_version) { |
48d23e61 JX |
155 | /* Pass in data_version implies release. */ |
156 | op_data->op_bias |= MDS_HSM_RELEASE; | |
157 | op_data->op_data_version = *data_version; | |
158 | op_data->op_lease_handle = och->och_lease_handle; | |
159 | op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS; | |
160 | } | |
b6ee3824 | 161 | epoch_close = op_data->op_flags & MF_EPOCH_CLOSE; |
d7e09d03 PT |
162 | rc = md_close(md_exp, op_data, och->och_mod, &req); |
163 | if (rc == -EAGAIN) { | |
164 | /* This close must have the epoch closed. */ | |
165 | LASSERT(epoch_close); | |
166 | /* MDS has instructed us to obtain Size-on-MDS attribute from | |
c0894c6c OD |
167 | * OSTs and send setattr to back to MDS. |
168 | */ | |
d7e09d03 PT |
169 | rc = ll_som_update(inode, op_data); |
170 | if (rc) { | |
97a075cd JN |
171 | CERROR("%s: inode "DFID" mdc Size-on-MDS update failed: rc = %d\n", |
172 | ll_i2mdexp(inode)->exp_obd->obd_name, | |
173 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
174 | rc = 0; |
175 | } | |
176 | } else if (rc) { | |
97a075cd JN |
177 | CERROR("%s: inode "DFID" mdc close failed: rc = %d\n", |
178 | ll_i2mdexp(inode)->exp_obd->obd_name, | |
179 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
180 | } |
181 | ||
182 | /* DATA_MODIFIED flag was successfully sent on close, cancel data | |
c0894c6c OD |
183 | * modification flag. |
184 | */ | |
d7e09d03 PT |
185 | if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) { |
186 | struct ll_inode_info *lli = ll_i2info(inode); | |
187 | ||
188 | spin_lock(&lli->lli_lock); | |
189 | lli->lli_flags &= ~LLIF_DATA_MODIFIED; | |
190 | spin_unlock(&lli->lli_lock); | |
191 | } | |
192 | ||
d7e09d03 PT |
193 | if (rc == 0) { |
194 | rc = ll_objects_destroy(req, inode); | |
195 | if (rc) | |
196 | CERROR("inode %lu ll_objects destroy: rc = %d\n", | |
197 | inode->i_ino, rc); | |
198 | } | |
48d23e61 JX |
199 | if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) { |
200 | struct mdt_body *body; | |
cea812cd | 201 | |
48d23e61 | 202 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); |
2e1b5b8b | 203 | if (!(body->mbo_valid & OBD_MD_FLRELEASED)) |
48d23e61 JX |
204 | rc = -EBUSY; |
205 | } | |
206 | ||
207 | ll_finish_md_op_data(op_data); | |
d7e09d03 | 208 | |
d7e09d03 | 209 | out: |
d7e09d03 PT |
210 | if (exp_connect_som(exp) && !epoch_close && |
211 | S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) { | |
212 | ll_queue_done_writing(inode, LLIF_DONE_WRITING); | |
213 | } else { | |
214 | md_clear_open_replay_data(md_exp, och); | |
215 | /* Free @och if it is not waiting for DONE_WRITING. */ | |
216 | och->och_fh.cookie = DEAD_HANDLE_MAGIC; | |
97903a26 | 217 | kfree(och); |
d7e09d03 PT |
218 | } |
219 | if (req) /* This is close request */ | |
220 | ptlrpc_req_finished(req); | |
221 | return rc; | |
222 | } | |
223 | ||
45b2a010 | 224 | int ll_md_real_close(struct inode *inode, fmode_t fmode) |
d7e09d03 PT |
225 | { |
226 | struct ll_inode_info *lli = ll_i2info(inode); | |
227 | struct obd_client_handle **och_p; | |
228 | struct obd_client_handle *och; | |
229 | __u64 *och_usecount; | |
230 | int rc = 0; | |
d7e09d03 | 231 | |
45b2a010 | 232 | if (fmode & FMODE_WRITE) { |
d7e09d03 PT |
233 | och_p = &lli->lli_mds_write_och; |
234 | och_usecount = &lli->lli_open_fd_write_count; | |
45b2a010 | 235 | } else if (fmode & FMODE_EXEC) { |
d7e09d03 PT |
236 | och_p = &lli->lli_mds_exec_och; |
237 | och_usecount = &lli->lli_open_fd_exec_count; | |
238 | } else { | |
45b2a010 | 239 | LASSERT(fmode & FMODE_READ); |
d7e09d03 PT |
240 | och_p = &lli->lli_mds_read_och; |
241 | och_usecount = &lli->lli_open_fd_read_count; | |
242 | } | |
243 | ||
244 | mutex_lock(&lli->lli_och_mutex); | |
45b2a010 JH |
245 | if (*och_usecount > 0) { |
246 | /* There are still users of this handle, so skip | |
c0894c6c OD |
247 | * freeing it. |
248 | */ | |
d7e09d03 | 249 | mutex_unlock(&lli->lli_och_mutex); |
0a3bdb00 | 250 | return 0; |
d7e09d03 | 251 | } |
45b2a010 | 252 | |
57303e76 | 253 | och = *och_p; |
d7e09d03 PT |
254 | *och_p = NULL; |
255 | mutex_unlock(&lli->lli_och_mutex); | |
256 | ||
6e16818b | 257 | if (och) { |
45b2a010 | 258 | /* There might be a race and this handle may already |
c0894c6c OD |
259 | * be closed. |
260 | */ | |
d7e09d03 | 261 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, |
48d23e61 | 262 | inode, och, NULL); |
d7e09d03 PT |
263 | } |
264 | ||
0a3bdb00 | 265 | return rc; |
d7e09d03 PT |
266 | } |
267 | ||
2d95f10e JH |
268 | static int ll_md_close(struct obd_export *md_exp, struct inode *inode, |
269 | struct file *file) | |
d7e09d03 PT |
270 | { |
271 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
272 | struct ll_inode_info *lli = ll_i2info(inode); | |
74d01958 AV |
273 | int lockmode; |
274 | __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK; | |
275 | struct lustre_handle lockh; | |
8369cfff | 276 | ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_OPEN} }; |
d7e09d03 | 277 | int rc = 0; |
d7e09d03 PT |
278 | |
279 | /* clear group lock, if present */ | |
280 | if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED)) | |
98eae5e7 | 281 | ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid); |
d7e09d03 | 282 | |
6e16818b | 283 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
284 | bool lease_broken; |
285 | ||
286 | /* Usually the lease is not released when the | |
c0894c6c OD |
287 | * application crashed, we need to release here. |
288 | */ | |
d3a8a4e2 | 289 | rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken); |
e15ba45d OD |
290 | CDEBUG(rc ? D_ERROR : D_INODE, |
291 | "Clean up lease " DFID " %d/%d\n", | |
292 | PFID(&lli->lli_fid), rc, lease_broken); | |
d3a8a4e2 JX |
293 | |
294 | fd->fd_lease_och = NULL; | |
295 | } | |
296 | ||
6e16818b | 297 | if (fd->fd_och) { |
48d23e61 | 298 | rc = ll_close_inode_openhandle(md_exp, inode, fd->fd_och, NULL); |
d3a8a4e2 | 299 | fd->fd_och = NULL; |
34e1f2bb | 300 | goto out; |
d3a8a4e2 JX |
301 | } |
302 | ||
d7e09d03 | 303 | /* Let's see if we have good enough OPEN lock on the file and if |
c0894c6c OD |
304 | * we can skip talking to MDS |
305 | */ | |
d7e09d03 | 306 | |
74d01958 AV |
307 | mutex_lock(&lli->lli_och_mutex); |
308 | if (fd->fd_omode & FMODE_WRITE) { | |
309 | lockmode = LCK_CW; | |
310 | LASSERT(lli->lli_open_fd_write_count); | |
311 | lli->lli_open_fd_write_count--; | |
312 | } else if (fd->fd_omode & FMODE_EXEC) { | |
313 | lockmode = LCK_PR; | |
314 | LASSERT(lli->lli_open_fd_exec_count); | |
315 | lli->lli_open_fd_exec_count--; | |
d7e09d03 | 316 | } else { |
74d01958 AV |
317 | lockmode = LCK_CR; |
318 | LASSERT(lli->lli_open_fd_read_count); | |
319 | lli->lli_open_fd_read_count--; | |
d7e09d03 | 320 | } |
74d01958 AV |
321 | mutex_unlock(&lli->lli_och_mutex); |
322 | ||
323 | if (!md_lock_match(md_exp, flags, ll_inode2fid(inode), | |
324 | LDLM_IBITS, &policy, lockmode, &lockh)) | |
325 | rc = ll_md_real_close(inode, fd->fd_omode); | |
d7e09d03 | 326 | |
d3a8a4e2 | 327 | out: |
d7e09d03 PT |
328 | LUSTRE_FPRIVATE(file) = NULL; |
329 | ll_file_data_put(fd); | |
d7e09d03 | 330 | |
0a3bdb00 | 331 | return rc; |
d7e09d03 PT |
332 | } |
333 | ||
334 | /* While this returns an error code, fput() the caller does not, so we need | |
335 | * to make every effort to clean up all of our state here. Also, applications | |
336 | * rarely check close errors and even if an error is returned they will not | |
337 | * re-try the close call. | |
338 | */ | |
339 | int ll_file_release(struct inode *inode, struct file *file) | |
340 | { | |
341 | struct ll_file_data *fd; | |
342 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
343 | struct ll_inode_info *lli = ll_i2info(inode); | |
344 | int rc; | |
d7e09d03 | 345 | |
97a075cd JN |
346 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n", |
347 | PFID(ll_inode2fid(inode)), inode); | |
d7e09d03 | 348 | |
f76c23da | 349 | if (!is_root_inode(inode)) |
d7e09d03 PT |
350 | ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1); |
351 | fd = LUSTRE_FPRIVATE(file); | |
6e16818b | 352 | LASSERT(fd); |
d7e09d03 | 353 | |
c0894c6c | 354 | /* The last ref on @file, maybe not be the owner pid of statahead. |
d7e09d03 | 355 | * Different processes can open the same dir, "ll_opendir_key" means: |
c0894c6c OD |
356 | * it is me that should stop the statahead thread. |
357 | */ | |
d7e09d03 PT |
358 | if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd && |
359 | lli->lli_opendir_pid != 0) | |
360 | ll_stop_statahead(inode, lli->lli_opendir_key); | |
361 | ||
f76c23da | 362 | if (is_root_inode(inode)) { |
d7e09d03 PT |
363 | LUSTRE_FPRIVATE(file) = NULL; |
364 | ll_file_data_put(fd); | |
0a3bdb00 | 365 | return 0; |
d7e09d03 PT |
366 | } |
367 | ||
368 | if (!S_ISDIR(inode->i_mode)) { | |
79496845 | 369 | if (lli->lli_clob) |
370 | lov_read_and_clear_async_rc(lli->lli_clob); | |
d7e09d03 PT |
371 | lli->lli_async_rc = 0; |
372 | } | |
373 | ||
374 | rc = ll_md_close(sbi->ll_md_exp, inode, file); | |
375 | ||
376 | if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val)) | |
377 | libcfs_debug_dumplog(); | |
378 | ||
0a3bdb00 | 379 | return rc; |
d7e09d03 PT |
380 | } |
381 | ||
48eddfd5 | 382 | static int ll_intent_file_open(struct dentry *dentry, void *lmm, |
d7e09d03 PT |
383 | int lmmsize, struct lookup_intent *itp) |
384 | { | |
2b0143b5 | 385 | struct inode *inode = d_inode(dentry); |
48eddfd5 AV |
386 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
387 | struct dentry *parent = dentry->d_parent; | |
388 | const char *name = dentry->d_name.name; | |
389 | const int len = dentry->d_name.len; | |
d7e09d03 PT |
390 | struct md_op_data *op_data; |
391 | struct ptlrpc_request *req; | |
392 | __u32 opc = LUSTRE_OPC_ANY; | |
393 | int rc; | |
d7e09d03 | 394 | |
c0894c6c | 395 | /* Usually we come here only for NFSD, and we want open lock. */ |
d7e09d03 PT |
396 | /* We can also get here if there was cached open handle in revalidate_it |
397 | * but it disappeared while we were getting from there to ll_file_open. | |
bef31c78 | 398 | * But this means this file was closed and immediately opened which |
c0894c6c OD |
399 | * makes a good candidate for using OPEN lock |
400 | */ | |
d7e09d03 | 401 | /* If lmmsize & lmm are not 0, we are just setting stripe info |
c0894c6c OD |
402 | * parameters. No need for the open lock |
403 | */ | |
6e16818b | 404 | if (!lmm && lmmsize == 0) { |
6dad4d89 OD |
405 | struct ll_dentry_data *ldd = ll_d2d(dentry); |
406 | /* | |
407 | * If we came via ll_iget_for_nfs, then we need to request | |
408 | * struct ll_dentry_data *ldd = ll_d2d(file->f_dentry); | |
409 | * | |
410 | * NB: when ldd is NULL, it must have come via normal | |
411 | * lookup path only, since ll_iget_for_nfs always calls | |
412 | * ll_d_init(). | |
413 | */ | |
414 | if (ldd && ldd->lld_nfs_dentry) { | |
415 | ldd->lld_nfs_dentry = 0; | |
416 | itp->it_flags |= MDS_OPEN_LOCK; | |
417 | } | |
d7e09d03 PT |
418 | if (itp->it_flags & FMODE_WRITE) |
419 | opc = LUSTRE_OPC_CREATE; | |
420 | } | |
421 | ||
2b0143b5 | 422 | op_data = ll_prep_md_op_data(NULL, d_inode(parent), |
48eddfd5 | 423 | inode, name, len, |
d7e09d03 PT |
424 | O_RDWR, opc, NULL); |
425 | if (IS_ERR(op_data)) | |
0a3bdb00 | 426 | return PTR_ERR(op_data); |
d7e09d03 PT |
427 | |
428 | itp->it_flags |= MDS_OPEN_BY_FID; | |
429 | rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp, | |
430 | 0 /*unused */, &req, ll_md_blocking_ast, 0); | |
431 | ll_finish_md_op_data(op_data); | |
432 | if (rc == -ESTALE) { | |
433 | /* reason for keep own exit path - don`t flood log | |
434 | * with messages with -ESTALE errors. | |
435 | */ | |
436 | if (!it_disposition(itp, DISP_OPEN_OPEN) || | |
e15ba45d | 437 | it_open_error(DISP_OPEN_OPEN, itp)) |
34e1f2bb | 438 | goto out; |
e22fdcc8 | 439 | ll_release_openhandle(inode, itp); |
34e1f2bb | 440 | goto out; |
d7e09d03 PT |
441 | } |
442 | ||
34e1f2bb JL |
443 | if (it_disposition(itp, DISP_LOOKUP_NEG)) { |
444 | rc = -ENOENT; | |
445 | goto out; | |
446 | } | |
d7e09d03 PT |
447 | |
448 | if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) { | |
449 | rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp); | |
450 | CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc); | |
34e1f2bb | 451 | goto out; |
d7e09d03 PT |
452 | } |
453 | ||
48eddfd5 | 454 | rc = ll_prep_inode(&inode, req, NULL, itp); |
e476f2e5 | 455 | if (!rc && itp->it_lock_mode) |
48eddfd5 | 456 | ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL); |
d7e09d03 PT |
457 | |
458 | out: | |
f236f69b | 459 | ptlrpc_req_finished(req); |
d7e09d03 PT |
460 | ll_intent_drop_lock(itp); |
461 | ||
0a3bdb00 | 462 | return rc; |
d7e09d03 PT |
463 | } |
464 | ||
465 | /** | |
466 | * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does | |
467 | * not believe attributes if a few ioepoch holders exist. Attributes for | |
468 | * previous ioepoch if new one is opened are also skipped by MDS. | |
469 | */ | |
470 | void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch) | |
471 | { | |
472 | if (ioepoch && lli->lli_ioepoch != ioepoch) { | |
473 | lli->lli_ioepoch = ioepoch; | |
b0f5aad5 | 474 | CDEBUG(D_INODE, "Epoch %llu opened on "DFID"\n", |
d7e09d03 PT |
475 | ioepoch, PFID(&lli->lli_fid)); |
476 | } | |
477 | } | |
478 | ||
ea1db081 JH |
479 | static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it, |
480 | struct obd_client_handle *och) | |
d7e09d03 | 481 | { |
d7e09d03 PT |
482 | struct mdt_body *body; |
483 | ||
8bf86fd9 | 484 | body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY); |
2e1b5b8b JH |
485 | och->och_fh = body->mbo_handle; |
486 | och->och_fid = body->mbo_fid1; | |
e476f2e5 | 487 | och->och_lease_handle.cookie = it->it_lock_handle; |
d7e09d03 | 488 | och->och_magic = OBD_CLIENT_HANDLE_MAGIC; |
d7e09d03 | 489 | och->och_flags = it->it_flags; |
d7e09d03 | 490 | |
63d42578 | 491 | return md_set_open_replay_data(md_exp, och, it); |
d7e09d03 PT |
492 | } |
493 | ||
2d95f10e JH |
494 | static int ll_local_open(struct file *file, struct lookup_intent *it, |
495 | struct ll_file_data *fd, struct obd_client_handle *och) | |
d7e09d03 | 496 | { |
2a8a3597 | 497 | struct inode *inode = file_inode(file); |
d7e09d03 | 498 | struct ll_inode_info *lli = ll_i2info(inode); |
d7e09d03 PT |
499 | |
500 | LASSERT(!LUSTRE_FPRIVATE(file)); | |
501 | ||
6e16818b | 502 | LASSERT(fd); |
d7e09d03 PT |
503 | |
504 | if (och) { | |
d7e09d03 PT |
505 | struct mdt_body *body; |
506 | int rc; | |
507 | ||
ea1db081 JH |
508 | rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och); |
509 | if (rc != 0) | |
0a3bdb00 | 510 | return rc; |
d7e09d03 | 511 | |
8bf86fd9 JH |
512 | body = req_capsule_server_get(&it->it_request->rq_pill, |
513 | &RMF_MDT_BODY); | |
2e1b5b8b | 514 | ll_ioepoch_open(lli, body->mbo_ioepoch); |
d7e09d03 PT |
515 | } |
516 | ||
517 | LUSTRE_FPRIVATE(file) = fd; | |
518 | ll_readahead_init(inode, &fd->fd_ras); | |
d3a8a4e2 | 519 | fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC); |
966c4a8f JX |
520 | |
521 | /* ll_cl_context initialize */ | |
522 | rwlock_init(&fd->fd_lock); | |
523 | INIT_LIST_HEAD(&fd->fd_lccs); | |
524 | ||
0a3bdb00 | 525 | return 0; |
d7e09d03 PT |
526 | } |
527 | ||
528 | /* Open a file, and (for the very first open) create objects on the OSTs at | |
529 | * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object | |
530 | * creation or open until ll_lov_setstripe() ioctl is called. | |
531 | * | |
532 | * If we already have the stripe MD locally then we don't request it in | |
533 | * md_open(), by passing a lmm_size = 0. | |
534 | * | |
535 | * It is up to the application to ensure no other processes open this file | |
536 | * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be | |
537 | * used. We might be able to avoid races of that sort by getting lli_open_sem | |
538 | * before returning in the O_LOV_DELAY_CREATE case and dropping it here | |
539 | * or in ll_file_release(), but I'm not sure that is desirable/necessary. | |
540 | */ | |
541 | int ll_file_open(struct inode *inode, struct file *file) | |
542 | { | |
543 | struct ll_inode_info *lli = ll_i2info(inode); | |
544 | struct lookup_intent *it, oit = { .it_op = IT_OPEN, | |
545 | .it_flags = file->f_flags }; | |
546 | struct obd_client_handle **och_p = NULL; | |
547 | __u64 *och_usecount = NULL; | |
548 | struct ll_file_data *fd; | |
549 | int rc = 0, opendir_set = 0; | |
d7e09d03 | 550 | |
97a075cd JN |
551 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n", |
552 | PFID(ll_inode2fid(inode)), inode, file->f_flags); | |
d7e09d03 PT |
553 | |
554 | it = file->private_data; /* XXX: compat macro */ | |
555 | file->private_data = NULL; /* prevent ll_local_open assertion */ | |
556 | ||
557 | fd = ll_file_data_get(); | |
6e16818b | 558 | if (!fd) { |
34e1f2bb JL |
559 | rc = -ENOMEM; |
560 | goto out_openerr; | |
561 | } | |
d7e09d03 PT |
562 | |
563 | fd->fd_file = file; | |
564 | if (S_ISDIR(inode->i_mode)) { | |
565 | spin_lock(&lli->lli_sa_lock); | |
6e16818b | 566 | if (!lli->lli_opendir_key && !lli->lli_sai && |
d7e09d03 PT |
567 | lli->lli_opendir_pid == 0) { |
568 | lli->lli_opendir_key = fd; | |
569 | lli->lli_opendir_pid = current_pid(); | |
570 | opendir_set = 1; | |
571 | } | |
572 | spin_unlock(&lli->lli_sa_lock); | |
573 | } | |
574 | ||
f76c23da | 575 | if (is_root_inode(inode)) { |
d7e09d03 | 576 | LUSTRE_FPRIVATE(file) = fd; |
0a3bdb00 | 577 | return 0; |
d7e09d03 PT |
578 | } |
579 | ||
e476f2e5 | 580 | if (!it || !it->it_disposition) { |
d7e09d03 PT |
581 | /* Convert f_flags into access mode. We cannot use file->f_mode, |
582 | * because everything but O_ACCMODE mask was stripped from | |
c0894c6c OD |
583 | * there |
584 | */ | |
d7e09d03 PT |
585 | if ((oit.it_flags + 1) & O_ACCMODE) |
586 | oit.it_flags++; | |
587 | if (file->f_flags & O_TRUNC) | |
588 | oit.it_flags |= FMODE_WRITE; | |
589 | ||
590 | /* kernel only call f_op->open in dentry_open. filp_open calls | |
591 | * dentry_open after call to open_namei that checks permissions. | |
592 | * Only nfsd_open call dentry_open directly without checking | |
c0894c6c OD |
593 | * permissions and because of that this code below is safe. |
594 | */ | |
d7e09d03 PT |
595 | if (oit.it_flags & (FMODE_WRITE | FMODE_READ)) |
596 | oit.it_flags |= MDS_OPEN_OWNEROVERRIDE; | |
597 | ||
598 | /* We do not want O_EXCL here, presumably we opened the file | |
c0894c6c OD |
599 | * already? XXX - NFS implications? |
600 | */ | |
d7e09d03 PT |
601 | oit.it_flags &= ~O_EXCL; |
602 | ||
603 | /* bug20584, if "it_flags" contains O_CREAT, the file will be | |
604 | * created if necessary, then "IT_CREAT" should be set to keep | |
c0894c6c OD |
605 | * consistent with it |
606 | */ | |
d7e09d03 PT |
607 | if (oit.it_flags & O_CREAT) |
608 | oit.it_op |= IT_CREAT; | |
609 | ||
610 | it = &oit; | |
611 | } | |
612 | ||
613 | restart: | |
614 | /* Let's see if we have file open on MDS already. */ | |
615 | if (it->it_flags & FMODE_WRITE) { | |
616 | och_p = &lli->lli_mds_write_och; | |
617 | och_usecount = &lli->lli_open_fd_write_count; | |
618 | } else if (it->it_flags & FMODE_EXEC) { | |
619 | och_p = &lli->lli_mds_exec_och; | |
620 | och_usecount = &lli->lli_open_fd_exec_count; | |
621 | } else { | |
622 | och_p = &lli->lli_mds_read_och; | |
623 | och_usecount = &lli->lli_open_fd_read_count; | |
624 | } | |
625 | ||
626 | mutex_lock(&lli->lli_och_mutex); | |
627 | if (*och_p) { /* Open handle is present */ | |
628 | if (it_disposition(it, DISP_OPEN_OPEN)) { | |
629 | /* Well, there's extra open request that we do not need, | |
c0894c6c OD |
630 | * let's close it somehow. This will decref request. |
631 | */ | |
d7e09d03 PT |
632 | rc = it_open_error(DISP_OPEN_OPEN, it); |
633 | if (rc) { | |
634 | mutex_unlock(&lli->lli_och_mutex); | |
34e1f2bb | 635 | goto out_openerr; |
d7e09d03 PT |
636 | } |
637 | ||
e22fdcc8 | 638 | ll_release_openhandle(inode, it); |
d7e09d03 PT |
639 | } |
640 | (*och_usecount)++; | |
641 | ||
642 | rc = ll_local_open(file, it, fd, NULL); | |
643 | if (rc) { | |
644 | (*och_usecount)--; | |
645 | mutex_unlock(&lli->lli_och_mutex); | |
34e1f2bb | 646 | goto out_openerr; |
d7e09d03 PT |
647 | } |
648 | } else { | |
649 | LASSERT(*och_usecount == 0); | |
e476f2e5 | 650 | if (!it->it_disposition) { |
d7e09d03 | 651 | /* We cannot just request lock handle now, new ELC code |
c0894c6c OD |
652 | * means that one of other OPEN locks for this file |
653 | * could be cancelled, and since blocking ast handler | |
654 | * would attempt to grab och_mutex as well, that would | |
655 | * result in a deadlock | |
656 | */ | |
d7e09d03 PT |
657 | mutex_unlock(&lli->lli_och_mutex); |
658 | it->it_create_mode |= M_CHECK_STALE; | |
48eddfd5 | 659 | rc = ll_intent_file_open(file->f_path.dentry, NULL, 0, it); |
d7e09d03 PT |
660 | it->it_create_mode &= ~M_CHECK_STALE; |
661 | if (rc) | |
34e1f2bb | 662 | goto out_openerr; |
d7e09d03 PT |
663 | |
664 | goto restart; | |
665 | } | |
496a51bd | 666 | *och_p = kzalloc(sizeof(struct obd_client_handle), GFP_NOFS); |
34e1f2bb JL |
667 | if (!*och_p) { |
668 | rc = -ENOMEM; | |
669 | goto out_och_free; | |
670 | } | |
d7e09d03 PT |
671 | |
672 | (*och_usecount)++; | |
673 | ||
674 | /* md_intent_lock() didn't get a request ref if there was an | |
675 | * open error, so don't do cleanup on the request here | |
c0894c6c OD |
676 | * (bug 3430) |
677 | */ | |
d7e09d03 | 678 | /* XXX (green): Should not we bail out on any error here, not |
c0894c6c OD |
679 | * just open error? |
680 | */ | |
d7e09d03 PT |
681 | rc = it_open_error(DISP_OPEN_OPEN, it); |
682 | if (rc) | |
34e1f2bb | 683 | goto out_och_free; |
d7e09d03 | 684 | |
5787be94 AD |
685 | LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF), |
686 | "inode %p: disposition %x, status %d\n", inode, | |
e476f2e5 | 687 | it_disposition(it, ~0), it->it_status); |
d7e09d03 PT |
688 | |
689 | rc = ll_local_open(file, it, fd, *och_p); | |
690 | if (rc) | |
34e1f2bb | 691 | goto out_och_free; |
d7e09d03 PT |
692 | } |
693 | mutex_unlock(&lli->lli_och_mutex); | |
694 | fd = NULL; | |
695 | ||
696 | /* Must do this outside lli_och_mutex lock to prevent deadlock where | |
c0894c6c OD |
697 | * different kind of OPEN lock for this same inode gets cancelled |
698 | * by ldlm_cancel_lru | |
699 | */ | |
d7e09d03 | 700 | if (!S_ISREG(inode->i_mode)) |
34e1f2bb | 701 | goto out_och_free; |
d7e09d03 | 702 | |
38585ccc AD |
703 | if (!lli->lli_has_smd && |
704 | (cl_is_lov_delay_create(file->f_flags) || | |
705 | (file->f_mode & FMODE_WRITE) == 0)) { | |
706 | CDEBUG(D_INODE, "object creation was delayed\n"); | |
34e1f2bb | 707 | goto out_och_free; |
d7e09d03 | 708 | } |
38585ccc | 709 | cl_lov_delay_create_clear(&file->f_flags); |
34e1f2bb | 710 | goto out_och_free; |
d7e09d03 PT |
711 | |
712 | out_och_free: | |
713 | if (rc) { | |
714 | if (och_p && *och_p) { | |
97903a26 | 715 | kfree(*och_p); |
c0a2472f | 716 | *och_p = NULL; |
d7e09d03 PT |
717 | (*och_usecount)--; |
718 | } | |
719 | mutex_unlock(&lli->lli_och_mutex); | |
720 | ||
721 | out_openerr: | |
722 | if (opendir_set != 0) | |
723 | ll_stop_statahead(inode, lli->lli_opendir_key); | |
a5cb8880 | 724 | ll_file_data_put(fd); |
d7e09d03 PT |
725 | } else { |
726 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); | |
727 | } | |
728 | ||
729 | if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) { | |
8bf86fd9 | 730 | ptlrpc_req_finished(it->it_request); |
d7e09d03 PT |
731 | it_clear_disposition(it, DISP_ENQ_OPEN_REF); |
732 | } | |
733 | ||
734 | return rc; | |
735 | } | |
736 | ||
d3a8a4e2 | 737 | static int ll_md_blocking_lease_ast(struct ldlm_lock *lock, |
e15ba45d OD |
738 | struct ldlm_lock_desc *desc, |
739 | void *data, int flag) | |
d3a8a4e2 JX |
740 | { |
741 | int rc; | |
742 | struct lustre_handle lockh; | |
743 | ||
744 | switch (flag) { | |
745 | case LDLM_CB_BLOCKING: | |
746 | ldlm_lock2handle(lock, &lockh); | |
747 | rc = ldlm_cli_cancel(&lockh, LCF_ASYNC); | |
748 | if (rc < 0) { | |
749 | CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc); | |
750 | return rc; | |
751 | } | |
752 | break; | |
753 | case LDLM_CB_CANCELING: | |
754 | /* do nothing */ | |
755 | break; | |
756 | } | |
757 | return 0; | |
758 | } | |
759 | ||
760 | /** | |
761 | * Acquire a lease and open the file. | |
762 | */ | |
2d95f10e JH |
763 | static struct obd_client_handle * |
764 | ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode, | |
765 | __u64 open_flags) | |
d3a8a4e2 JX |
766 | { |
767 | struct lookup_intent it = { .it_op = IT_OPEN }; | |
768 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
769 | struct md_op_data *op_data; | |
770 | struct ptlrpc_request *req; | |
771 | struct lustre_handle old_handle = { 0 }; | |
772 | struct obd_client_handle *och = NULL; | |
773 | int rc; | |
774 | int rc2; | |
775 | ||
776 | if (fmode != FMODE_WRITE && fmode != FMODE_READ) | |
777 | return ERR_PTR(-EINVAL); | |
778 | ||
6e16818b | 779 | if (file) { |
d3a8a4e2 JX |
780 | struct ll_inode_info *lli = ll_i2info(inode); |
781 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
782 | struct obd_client_handle **och_p; | |
783 | __u64 *och_usecount; | |
784 | ||
785 | if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC)) | |
786 | return ERR_PTR(-EPERM); | |
787 | ||
788 | /* Get the openhandle of the file */ | |
789 | rc = -EBUSY; | |
790 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 791 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
792 | mutex_unlock(&lli->lli_och_mutex); |
793 | return ERR_PTR(rc); | |
794 | } | |
795 | ||
6e16818b | 796 | if (!fd->fd_och) { |
d3a8a4e2 | 797 | if (file->f_mode & FMODE_WRITE) { |
6e16818b | 798 | LASSERT(lli->lli_mds_write_och); |
d3a8a4e2 JX |
799 | och_p = &lli->lli_mds_write_och; |
800 | och_usecount = &lli->lli_open_fd_write_count; | |
801 | } else { | |
6e16818b | 802 | LASSERT(lli->lli_mds_read_och); |
d3a8a4e2 JX |
803 | och_p = &lli->lli_mds_read_och; |
804 | och_usecount = &lli->lli_open_fd_read_count; | |
805 | } | |
806 | if (*och_usecount == 1) { | |
807 | fd->fd_och = *och_p; | |
808 | *och_p = NULL; | |
809 | *och_usecount = 0; | |
810 | rc = 0; | |
811 | } | |
812 | } | |
813 | mutex_unlock(&lli->lli_och_mutex); | |
814 | if (rc < 0) /* more than 1 opener */ | |
815 | return ERR_PTR(rc); | |
816 | ||
6e16818b | 817 | LASSERT(fd->fd_och); |
d3a8a4e2 JX |
818 | old_handle = fd->fd_och->och_fh; |
819 | } | |
820 | ||
496a51bd JL |
821 | och = kzalloc(sizeof(*och), GFP_NOFS); |
822 | if (!och) | |
d3a8a4e2 JX |
823 | return ERR_PTR(-ENOMEM); |
824 | ||
825 | op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0, | |
e15ba45d | 826 | LUSTRE_OPC_ANY, NULL); |
34e1f2bb JL |
827 | if (IS_ERR(op_data)) { |
828 | rc = PTR_ERR(op_data); | |
829 | goto out; | |
830 | } | |
d3a8a4e2 JX |
831 | |
832 | /* To tell the MDT this openhandle is from the same owner */ | |
833 | op_data->op_handle = old_handle; | |
834 | ||
48d23e61 JX |
835 | it.it_flags = fmode | open_flags; |
836 | it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE; | |
d3a8a4e2 | 837 | rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, &req, |
e15ba45d | 838 | ll_md_blocking_lease_ast, |
d3a8a4e2 JX |
839 | /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise |
840 | * it can be cancelled which may mislead applications that the lease is | |
841 | * broken; | |
842 | * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal | |
843 | * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast | |
c0894c6c OD |
844 | * doesn't deal with openhandle, so normal openhandle will be leaked. |
845 | */ | |
d3a8a4e2 JX |
846 | LDLM_FL_NO_LRU | LDLM_FL_EXCL); |
847 | ll_finish_md_op_data(op_data); | |
f236f69b | 848 | ptlrpc_req_finished(req); |
d3a8a4e2 | 849 | if (rc < 0) |
34e1f2bb | 850 | goto out_release_it; |
d3a8a4e2 | 851 | |
34e1f2bb JL |
852 | if (it_disposition(&it, DISP_LOOKUP_NEG)) { |
853 | rc = -ENOENT; | |
854 | goto out_release_it; | |
855 | } | |
d3a8a4e2 JX |
856 | |
857 | rc = it_open_error(DISP_OPEN_OPEN, &it); | |
858 | if (rc) | |
34e1f2bb | 859 | goto out_release_it; |
d3a8a4e2 JX |
860 | |
861 | LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF)); | |
862 | ll_och_fill(sbi->ll_md_exp, &it, och); | |
863 | ||
34e1f2bb JL |
864 | if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */ { |
865 | rc = -EOPNOTSUPP; | |
866 | goto out_close; | |
867 | } | |
d3a8a4e2 JX |
868 | |
869 | /* already get lease, handle lease lock */ | |
870 | ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); | |
e476f2e5 JH |
871 | if (it.it_lock_mode == 0 || |
872 | it.it_lock_bits != MDS_INODELOCK_OPEN) { | |
d3a8a4e2 JX |
873 | /* open lock must return for lease */ |
874 | CERROR(DFID "lease granted but no open lock, %d/%llu.\n", | |
e476f2e5 JH |
875 | PFID(ll_inode2fid(inode)), it.it_lock_mode, |
876 | it.it_lock_bits); | |
34e1f2bb JL |
877 | rc = -EPROTO; |
878 | goto out_close; | |
d3a8a4e2 JX |
879 | } |
880 | ||
881 | ll_intent_release(&it); | |
882 | return och; | |
883 | ||
884 | out_close: | |
e55a68b6 | 885 | /* Cancel open lock */ |
e476f2e5 | 886 | if (it.it_lock_mode != 0) { |
d3a8a4e2 | 887 | ldlm_lock_decref_and_cancel(&och->och_lease_handle, |
e476f2e5 JH |
888 | it.it_lock_mode); |
889 | it.it_lock_mode = 0; | |
e55a68b6 | 890 | och->och_lease_handle.cookie = 0ULL; |
d3a8a4e2 | 891 | } |
e55a68b6 JX |
892 | rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, inode, och, NULL); |
893 | if (rc2 < 0) | |
894 | CERROR("%s: error closing file "DFID": %d\n", | |
895 | ll_get_fsname(inode->i_sb, NULL, 0), | |
896 | PFID(&ll_i2info(inode)->lli_fid), rc2); | |
897 | och = NULL; /* och has been freed in ll_close_inode_openhandle() */ | |
d3a8a4e2 JX |
898 | out_release_it: |
899 | ll_intent_release(&it); | |
900 | out: | |
97903a26 | 901 | kfree(och); |
d3a8a4e2 JX |
902 | return ERR_PTR(rc); |
903 | } | |
d3a8a4e2 JX |
904 | |
905 | /** | |
906 | * Release lease and close the file. | |
907 | * It will check if the lease has ever broken. | |
908 | */ | |
2d95f10e JH |
909 | static int ll_lease_close(struct obd_client_handle *och, struct inode *inode, |
910 | bool *lease_broken) | |
d3a8a4e2 JX |
911 | { |
912 | struct ldlm_lock *lock; | |
913 | bool cancelled = true; | |
914 | int rc; | |
915 | ||
916 | lock = ldlm_handle2lock(&och->och_lease_handle); | |
6e16818b | 917 | if (lock) { |
d3a8a4e2 JX |
918 | lock_res_and_lock(lock); |
919 | cancelled = ldlm_is_cancel(lock); | |
920 | unlock_res_and_lock(lock); | |
ead02808 | 921 | LDLM_LOCK_PUT(lock); |
d3a8a4e2 JX |
922 | } |
923 | ||
e15ba45d OD |
924 | CDEBUG(D_INODE, "lease for " DFID " broken? %d\n", |
925 | PFID(&ll_i2info(inode)->lli_fid), cancelled); | |
d3a8a4e2 JX |
926 | |
927 | if (!cancelled) | |
928 | ldlm_cli_cancel(&och->och_lease_handle, 0); | |
6e16818b | 929 | if (lease_broken) |
d3a8a4e2 JX |
930 | *lease_broken = cancelled; |
931 | ||
48d23e61 JX |
932 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och, |
933 | NULL); | |
d3a8a4e2 JX |
934 | return rc; |
935 | } | |
d3a8a4e2 | 936 | |
d7e09d03 PT |
937 | /* Fills the obdo with the attributes for the lsm */ |
938 | static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, | |
e1798006 | 939 | struct obdo *obdo, __u64 ioepoch, int dv_flags) |
d7e09d03 PT |
940 | { |
941 | struct ptlrpc_request_set *set; | |
45efd655 | 942 | struct obd_info oinfo = { }; |
d7e09d03 PT |
943 | int rc; |
944 | ||
6e16818b | 945 | LASSERT(lsm); |
d7e09d03 PT |
946 | |
947 | oinfo.oi_md = lsm; | |
948 | oinfo.oi_oa = obdo; | |
949 | oinfo.oi_oa->o_oi = lsm->lsm_oi; | |
950 | oinfo.oi_oa->o_mode = S_IFREG; | |
951 | oinfo.oi_oa->o_ioepoch = ioepoch; | |
952 | oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | | |
953 | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | | |
954 | OBD_MD_FLBLKSZ | OBD_MD_FLATIME | | |
955 | OBD_MD_FLMTIME | OBD_MD_FLCTIME | | |
956 | OBD_MD_FLGROUP | OBD_MD_FLEPOCH | | |
957 | OBD_MD_FLDATAVERSION; | |
e1798006 | 958 | if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) { |
d7e09d03 PT |
959 | oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS; |
960 | oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK; | |
e1798006 JX |
961 | if (dv_flags & LL_DV_WR_FLUSH) |
962 | oinfo.oi_oa->o_flags |= OBD_FL_FLUSH; | |
d7e09d03 PT |
963 | } |
964 | ||
965 | set = ptlrpc_prep_set(); | |
6e16818b | 966 | if (!set) { |
19b2056f | 967 | CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM); |
d7e09d03 PT |
968 | rc = -ENOMEM; |
969 | } else { | |
970 | rc = obd_getattr_async(exp, &oinfo, set); | |
971 | if (rc == 0) | |
972 | rc = ptlrpc_set_wait(set); | |
973 | ptlrpc_set_destroy(set); | |
974 | } | |
e1798006 | 975 | if (rc == 0) { |
d7e09d03 PT |
976 | oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | |
977 | OBD_MD_FLATIME | OBD_MD_FLMTIME | | |
978 | OBD_MD_FLCTIME | OBD_MD_FLSIZE | | |
e1798006 JX |
979 | OBD_MD_FLDATAVERSION | OBD_MD_FLFLAGS); |
980 | if (dv_flags & LL_DV_WR_FLUSH && | |
981 | !(oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS && | |
982 | oinfo.oi_oa->o_flags & OBD_FL_FLUSH)) | |
983 | return -ENOTSUPP; | |
984 | } | |
0a3bdb00 | 985 | return rc; |
d7e09d03 PT |
986 | } |
987 | ||
988 | /** | |
989 | * Performs the getattr on the inode and updates its fields. | |
990 | * If @sync != 0, perform the getattr under the server-side lock. | |
991 | */ | |
992 | int ll_inode_getattr(struct inode *inode, struct obdo *obdo, | |
993 | __u64 ioepoch, int sync) | |
994 | { | |
d7e09d03 PT |
995 | struct lov_stripe_md *lsm; |
996 | int rc; | |
d7e09d03 PT |
997 | |
998 | lsm = ccc_inode_lsm_get(inode); | |
999 | rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode), | |
e1798006 | 1000 | obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0); |
d7e09d03 PT |
1001 | if (rc == 0) { |
1002 | struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi; | |
1003 | ||
1004 | obdo_refresh_inode(inode, obdo, obdo->o_valid); | |
2d00bd17 JP |
1005 | CDEBUG(D_INODE, "objid " DOSTID " size %llu, blocks %llu, blksize %lu\n", |
1006 | POSTID(oi), i_size_read(inode), | |
d7e09d03 | 1007 | (unsigned long long)inode->i_blocks, |
16e0631d | 1008 | 1UL << inode->i_blkbits); |
d7e09d03 PT |
1009 | } |
1010 | ccc_inode_lsm_put(inode, lsm); | |
0a3bdb00 | 1011 | return rc; |
d7e09d03 PT |
1012 | } |
1013 | ||
d2995737 | 1014 | int ll_merge_attr(const struct lu_env *env, struct inode *inode) |
d7e09d03 PT |
1015 | { |
1016 | struct ll_inode_info *lli = ll_i2info(inode); | |
1017 | struct cl_object *obj = lli->lli_clob; | |
9acc4500 | 1018 | struct cl_attr *attr = vvp_env_thread_attr(env); |
d2995737 JH |
1019 | s64 atime; |
1020 | s64 mtime; | |
1021 | s64 ctime; | |
d7e09d03 PT |
1022 | int rc = 0; |
1023 | ||
d7e09d03 | 1024 | ll_inode_size_lock(inode); |
d2995737 | 1025 | |
d7e09d03 | 1026 | /* merge timestamps the most recently obtained from mds with |
c0894c6c OD |
1027 | * timestamps obtained from osts |
1028 | */ | |
d2995737 JH |
1029 | LTIME_S(inode->i_atime) = lli->lli_atime; |
1030 | LTIME_S(inode->i_mtime) = lli->lli_mtime; | |
1031 | LTIME_S(inode->i_ctime) = lli->lli_ctime; | |
376ef86b | 1032 | |
d2995737 JH |
1033 | mtime = LTIME_S(inode->i_mtime); |
1034 | atime = LTIME_S(inode->i_atime); | |
1035 | ctime = LTIME_S(inode->i_ctime); | |
d7e09d03 PT |
1036 | |
1037 | cl_object_attr_lock(obj); | |
1038 | rc = cl_object_attr_get(env, obj, attr); | |
1039 | cl_object_attr_unlock(obj); | |
1040 | ||
d2995737 JH |
1041 | if (rc != 0) |
1042 | goto out_size_unlock; | |
d7e09d03 | 1043 | |
d2995737 JH |
1044 | if (atime < attr->cat_atime) |
1045 | atime = attr->cat_atime; | |
d7e09d03 | 1046 | |
d2995737 JH |
1047 | if (ctime < attr->cat_ctime) |
1048 | ctime = attr->cat_ctime; | |
d7e09d03 | 1049 | |
d2995737 JH |
1050 | if (mtime < attr->cat_mtime) |
1051 | mtime = attr->cat_mtime; | |
1052 | ||
1053 | CDEBUG(D_VFSTRACE, DFID " updating i_size %llu\n", | |
1054 | PFID(&lli->lli_fid), attr->cat_size); | |
1055 | ||
1929c433 | 1056 | i_size_write(inode, attr->cat_size); |
d2995737 JH |
1057 | |
1058 | inode->i_blocks = attr->cat_blocks; | |
1059 | ||
1060 | LTIME_S(inode->i_mtime) = mtime; | |
1061 | LTIME_S(inode->i_atime) = atime; | |
1062 | LTIME_S(inode->i_ctime) = ctime; | |
1063 | ||
1064 | out_size_unlock: | |
d7e09d03 PT |
1065 | ll_inode_size_unlock(inode); |
1066 | ||
0a3bdb00 | 1067 | return rc; |
d7e09d03 PT |
1068 | } |
1069 | ||
1070 | int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm, | |
1071 | lstat_t *st) | |
1072 | { | |
1073 | struct obdo obdo = { 0 }; | |
1074 | int rc; | |
1075 | ||
ef2e0f55 | 1076 | rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, &obdo, 0, 0); |
d7e09d03 PT |
1077 | if (rc == 0) { |
1078 | st->st_size = obdo.o_size; | |
1079 | st->st_blocks = obdo.o_blocks; | |
1080 | st->st_mtime = obdo.o_mtime; | |
1081 | st->st_atime = obdo.o_atime; | |
1082 | st->st_ctime = obdo.o_ctime; | |
1083 | } | |
1084 | return rc; | |
1085 | } | |
1086 | ||
ec9bca9c JH |
1087 | static bool file_is_noatime(const struct file *file) |
1088 | { | |
1089 | const struct vfsmount *mnt = file->f_path.mnt; | |
2a8a3597 | 1090 | const struct inode *inode = file_inode(file); |
ec9bca9c JH |
1091 | |
1092 | /* Adapted from file_accessed() and touch_atime().*/ | |
1093 | if (file->f_flags & O_NOATIME) | |
1094 | return true; | |
1095 | ||
1096 | if (inode->i_flags & S_NOATIME) | |
1097 | return true; | |
1098 | ||
1099 | if (IS_NOATIME(inode)) | |
1100 | return true; | |
1101 | ||
1102 | if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY)) | |
1103 | return true; | |
1104 | ||
1105 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1106 | return true; | |
1107 | ||
1108 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1109 | return true; | |
1110 | ||
1111 | return false; | |
1112 | } | |
1113 | ||
d7e09d03 PT |
1114 | void ll_io_init(struct cl_io *io, const struct file *file, int write) |
1115 | { | |
2a8a3597 | 1116 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
1117 | |
1118 | io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK; | |
1119 | if (write) { | |
1120 | io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); | |
1121 | io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || | |
1122 | file->f_flags & O_DIRECT || | |
1123 | IS_SYNC(inode); | |
1124 | } | |
1125 | io->ci_obj = ll_i2info(inode)->lli_clob; | |
1126 | io->ci_lockreq = CILR_MAYBE; | |
1127 | if (ll_file_nolock(file)) { | |
1128 | io->ci_lockreq = CILR_NEVER; | |
1129 | io->ci_no_srvlock = 1; | |
1130 | } else if (file->f_flags & O_APPEND) { | |
1131 | io->ci_lockreq = CILR_MANDATORY; | |
1132 | } | |
ec9bca9c JH |
1133 | |
1134 | io->ci_noatime = file_is_noatime(file); | |
d7e09d03 PT |
1135 | } |
1136 | ||
1137 | static ssize_t | |
1138 | ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, | |
1139 | struct file *file, enum cl_io_type iot, | |
1140 | loff_t *ppos, size_t count) | |
1141 | { | |
2a8a3597 | 1142 | struct ll_inode_info *lli = ll_i2info(file_inode(file)); |
d7e09d03 PT |
1143 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
1144 | struct cl_io *io; | |
1145 | ssize_t result; | |
d7e09d03 | 1146 | |
77605e41 JX |
1147 | CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: %llu, count: %zd\n", |
1148 | file->f_path.dentry->d_name.name, iot, *ppos, count); | |
1149 | ||
d7e09d03 | 1150 | restart: |
9acc4500 | 1151 | io = vvp_env_thread_io(env); |
d7e09d03 PT |
1152 | ll_io_init(io, file, iot == CIT_WRITE); |
1153 | ||
1154 | if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) { | |
e0a8144b | 1155 | struct vvp_io *vio = vvp_env_io(env); |
d7e09d03 PT |
1156 | int write_mutex_locked = 0; |
1157 | ||
e0a8144b JH |
1158 | vio->vui_fd = LUSTRE_FPRIVATE(file); |
1159 | vio->vui_io_subtype = args->via_io_subtype; | |
d7e09d03 | 1160 | |
e0a8144b | 1161 | switch (vio->vui_io_subtype) { |
d7e09d03 | 1162 | case IO_NORMAL: |
e0a8144b JH |
1163 | vio->vui_iter = args->u.normal.via_iter; |
1164 | vio->vui_iocb = args->u.normal.via_iocb; | |
d7e09d03 | 1165 | if ((iot == CIT_WRITE) && |
e0a8144b | 1166 | !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { |
d7e09d03 | 1167 | if (mutex_lock_interruptible(&lli-> |
34e1f2bb JL |
1168 | lli_write_mutex)) { |
1169 | result = -ERESTARTSYS; | |
1170 | goto out; | |
1171 | } | |
d7e09d03 | 1172 | write_mutex_locked = 1; |
d7e09d03 | 1173 | } |
77605e41 | 1174 | down_read(&lli->lli_trunc_sem); |
d7e09d03 | 1175 | break; |
d7e09d03 | 1176 | case IO_SPLICE: |
e0a8144b JH |
1177 | vio->u.splice.vui_pipe = args->u.splice.via_pipe; |
1178 | vio->u.splice.vui_flags = args->u.splice.via_flags; | |
d7e09d03 PT |
1179 | break; |
1180 | default: | |
e0a8144b | 1181 | CERROR("Unknown IO type - %u\n", vio->vui_io_subtype); |
d7e09d03 PT |
1182 | LBUG(); |
1183 | } | |
966c4a8f | 1184 | ll_cl_add(file, env, io); |
d7e09d03 | 1185 | result = cl_io_loop(env, io); |
966c4a8f | 1186 | ll_cl_remove(file, env); |
77605e41 JX |
1187 | if (args->via_io_subtype == IO_NORMAL) |
1188 | up_read(&lli->lli_trunc_sem); | |
d7e09d03 PT |
1189 | if (write_mutex_locked) |
1190 | mutex_unlock(&lli->lli_write_mutex); | |
d7e09d03 PT |
1191 | } else { |
1192 | /* cl_io_rw_init() handled IO */ | |
1193 | result = io->ci_result; | |
1194 | } | |
1195 | ||
1196 | if (io->ci_nob > 0) { | |
1197 | result = io->ci_nob; | |
1198 | *ppos = io->u.ci_wr.wr.crw_pos; | |
1199 | } | |
34e1f2bb | 1200 | goto out; |
d7e09d03 PT |
1201 | out: |
1202 | cl_io_fini(env, io); | |
1203 | /* If any bit been read/written (result != 0), we just return | |
c0894c6c OD |
1204 | * short read/write instead of restart io. |
1205 | */ | |
5ea17d6c | 1206 | if ((result == 0 || result == -ENODATA) && io->ci_need_restart) { |
09561a53 | 1207 | CDEBUG(D_VFSTRACE, "Restart %s on %pD from %lld, count:%zd\n", |
d7e09d03 | 1208 | iot == CIT_READ ? "read" : "write", |
09561a53 | 1209 | file, *ppos, count); |
19b2056f | 1210 | LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob); |
d7e09d03 PT |
1211 | goto restart; |
1212 | } | |
1213 | ||
1214 | if (iot == CIT_READ) { | |
1215 | if (result >= 0) | |
2a8a3597 | 1216 | ll_stats_ops_tally(ll_i2sbi(file_inode(file)), |
d7e09d03 PT |
1217 | LPROC_LL_READ_BYTES, result); |
1218 | } else if (iot == CIT_WRITE) { | |
1219 | if (result >= 0) { | |
2a8a3597 | 1220 | ll_stats_ops_tally(ll_i2sbi(file_inode(file)), |
d7e09d03 PT |
1221 | LPROC_LL_WRITE_BYTES, result); |
1222 | fd->fd_write_failed = false; | |
1223 | } else if (result != -ERESTARTSYS) { | |
1224 | fd->fd_write_failed = true; | |
1225 | } | |
1226 | } | |
77605e41 | 1227 | CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result); |
d7e09d03 PT |
1228 | |
1229 | return result; | |
1230 | } | |
1231 | ||
b42b15fd | 1232 | static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
d7e09d03 PT |
1233 | { |
1234 | struct lu_env *env; | |
1235 | struct vvp_io_args *args; | |
d7e09d03 PT |
1236 | ssize_t result; |
1237 | int refcheck; | |
d7e09d03 | 1238 | |
d7e09d03 PT |
1239 | env = cl_env_get(&refcheck); |
1240 | if (IS_ERR(env)) | |
0a3bdb00 | 1241 | return PTR_ERR(env); |
d7e09d03 | 1242 | |
9989a58e | 1243 | args = ll_env_args(env, IO_NORMAL); |
b42b15fd | 1244 | args->u.normal.via_iter = to; |
d7e09d03 PT |
1245 | args->u.normal.via_iocb = iocb; |
1246 | ||
1247 | result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ, | |
b42b15fd | 1248 | &iocb->ki_pos, iov_iter_count(to)); |
d7e09d03 | 1249 | cl_env_put(env, &refcheck); |
0a3bdb00 | 1250 | return result; |
d7e09d03 PT |
1251 | } |
1252 | ||
1253 | /* | |
1254 | * Write to a file (through the page cache). | |
1255 | */ | |
b42b15fd | 1256 | static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
d7e09d03 PT |
1257 | { |
1258 | struct lu_env *env; | |
1259 | struct vvp_io_args *args; | |
d7e09d03 PT |
1260 | ssize_t result; |
1261 | int refcheck; | |
d7e09d03 | 1262 | |
d7e09d03 PT |
1263 | env = cl_env_get(&refcheck); |
1264 | if (IS_ERR(env)) | |
0a3bdb00 | 1265 | return PTR_ERR(env); |
d7e09d03 | 1266 | |
9989a58e | 1267 | args = ll_env_args(env, IO_NORMAL); |
b42b15fd | 1268 | args->u.normal.via_iter = from; |
d7e09d03 PT |
1269 | args->u.normal.via_iocb = iocb; |
1270 | ||
1271 | result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE, | |
e15ba45d | 1272 | &iocb->ki_pos, iov_iter_count(from)); |
d7e09d03 | 1273 | cl_env_put(env, &refcheck); |
0a3bdb00 | 1274 | return result; |
d7e09d03 PT |
1275 | } |
1276 | ||
d7e09d03 PT |
1277 | /* |
1278 | * Send file content (through pagecache) somewhere with helper | |
1279 | */ | |
1280 | static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos, | |
1281 | struct pipe_inode_info *pipe, size_t count, | |
1282 | unsigned int flags) | |
1283 | { | |
1284 | struct lu_env *env; | |
1285 | struct vvp_io_args *args; | |
1286 | ssize_t result; | |
1287 | int refcheck; | |
d7e09d03 PT |
1288 | |
1289 | env = cl_env_get(&refcheck); | |
1290 | if (IS_ERR(env)) | |
0a3bdb00 | 1291 | return PTR_ERR(env); |
d7e09d03 | 1292 | |
9989a58e | 1293 | args = ll_env_args(env, IO_SPLICE); |
d7e09d03 PT |
1294 | args->u.splice.via_pipe = pipe; |
1295 | args->u.splice.via_flags = flags; | |
1296 | ||
1297 | result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count); | |
1298 | cl_env_put(env, &refcheck); | |
0a3bdb00 | 1299 | return result; |
d7e09d03 PT |
1300 | } |
1301 | ||
21aef7d9 | 1302 | static int ll_lov_recreate(struct inode *inode, struct ost_id *oi, u32 ost_idx) |
d7e09d03 PT |
1303 | { |
1304 | struct obd_export *exp = ll_i2dtexp(inode); | |
1305 | struct obd_trans_info oti = { 0 }; | |
1306 | struct obdo *oa = NULL; | |
1307 | int lsm_size; | |
1308 | int rc = 0; | |
1309 | struct lov_stripe_md *lsm = NULL, *lsm2; | |
d7e09d03 | 1310 | |
21068c46 | 1311 | oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS); |
6e16818b | 1312 | if (!oa) |
0a3bdb00 | 1313 | return -ENOMEM; |
d7e09d03 PT |
1314 | |
1315 | lsm = ccc_inode_lsm_get(inode); | |
34e1f2bb JL |
1316 | if (!lsm_has_objects(lsm)) { |
1317 | rc = -ENOENT; | |
1318 | goto out; | |
1319 | } | |
d7e09d03 PT |
1320 | |
1321 | lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) * | |
1322 | (lsm->lsm_stripe_count)); | |
1323 | ||
e958f49b | 1324 | lsm2 = libcfs_kvzalloc(lsm_size, GFP_NOFS); |
6e16818b | 1325 | if (!lsm2) { |
34e1f2bb JL |
1326 | rc = -ENOMEM; |
1327 | goto out; | |
1328 | } | |
d7e09d03 PT |
1329 | |
1330 | oa->o_oi = *oi; | |
1331 | oa->o_nlink = ost_idx; | |
1332 | oa->o_flags |= OBD_FL_RECREATE_OBJS; | |
1333 | oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP; | |
1334 | obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | | |
1335 | OBD_MD_FLMTIME | OBD_MD_FLCTIME); | |
1336 | obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid); | |
1337 | memcpy(lsm2, lsm, lsm_size); | |
1338 | ll_inode_size_lock(inode); | |
1339 | rc = obd_create(NULL, exp, oa, &lsm2, &oti); | |
1340 | ll_inode_size_unlock(inode); | |
1341 | ||
e958f49b | 1342 | kvfree(lsm2); |
34e1f2bb | 1343 | goto out; |
d7e09d03 PT |
1344 | out: |
1345 | ccc_inode_lsm_put(inode, lsm); | |
2ba262fb | 1346 | kmem_cache_free(obdo_cachep, oa); |
d7e09d03 PT |
1347 | return rc; |
1348 | } | |
1349 | ||
1350 | static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg) | |
1351 | { | |
1352 | struct ll_recreate_obj ucreat; | |
1353 | struct ost_id oi; | |
d7e09d03 | 1354 | |
2eb90a75 | 1355 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1356 | return -EPERM; |
d7e09d03 | 1357 | |
02f9c12e | 1358 | if (copy_from_user(&ucreat, (struct ll_recreate_obj __user *)arg, |
d7e09d03 | 1359 | sizeof(ucreat))) |
0a3bdb00 | 1360 | return -EFAULT; |
d7e09d03 PT |
1361 | |
1362 | ostid_set_seq_mdt0(&oi); | |
1363 | ostid_set_id(&oi, ucreat.lrc_id); | |
0a3bdb00 | 1364 | return ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx); |
d7e09d03 PT |
1365 | } |
1366 | ||
1367 | static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg) | |
1368 | { | |
1369 | struct lu_fid fid; | |
1370 | struct ost_id oi; | |
21aef7d9 | 1371 | u32 ost_idx; |
d7e09d03 | 1372 | |
2eb90a75 | 1373 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1374 | return -EPERM; |
d7e09d03 | 1375 | |
02f9c12e | 1376 | if (copy_from_user(&fid, (struct lu_fid __user *)arg, sizeof(fid))) |
0a3bdb00 | 1377 | return -EFAULT; |
d7e09d03 PT |
1378 | |
1379 | fid_to_ostid(&fid, &oi); | |
1380 | ost_idx = (fid_seq(&fid) >> 16) & 0xffff; | |
0a3bdb00 | 1381 | return ll_lov_recreate(inode, &oi, ost_idx); |
d7e09d03 PT |
1382 | } |
1383 | ||
c139f3ce | 1384 | int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry, |
d467220e NY |
1385 | __u64 flags, struct lov_user_md *lum, |
1386 | int lum_size) | |
d7e09d03 PT |
1387 | { |
1388 | struct lov_stripe_md *lsm = NULL; | |
1389 | struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; | |
1390 | int rc = 0; | |
d7e09d03 PT |
1391 | |
1392 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1393 | if (lsm) { |
d7e09d03 | 1394 | ccc_inode_lsm_put(inode, lsm); |
97a075cd JN |
1395 | CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n", |
1396 | PFID(ll_inode2fid(inode))); | |
34e1f2bb JL |
1397 | rc = -EEXIST; |
1398 | goto out; | |
d7e09d03 PT |
1399 | } |
1400 | ||
1401 | ll_inode_size_lock(inode); | |
c139f3ce | 1402 | rc = ll_intent_file_open(dentry, lum, lum_size, &oit); |
d7e09d03 | 1403 | if (rc) |
34e1f2bb | 1404 | goto out_unlock; |
e476f2e5 | 1405 | rc = oit.it_status; |
d7e09d03 | 1406 | if (rc < 0) |
34e1f2bb | 1407 | goto out_req_free; |
d7e09d03 | 1408 | |
e22fdcc8 | 1409 | ll_release_openhandle(inode, &oit); |
d7e09d03 | 1410 | |
38585ccc | 1411 | out_unlock: |
d7e09d03 PT |
1412 | ll_inode_size_unlock(inode); |
1413 | ll_intent_release(&oit); | |
1414 | ccc_inode_lsm_put(inode, lsm); | |
38585ccc | 1415 | out: |
0a3bdb00 | 1416 | return rc; |
d7e09d03 | 1417 | out_req_free: |
8bf86fd9 | 1418 | ptlrpc_req_finished((struct ptlrpc_request *)oit.it_request); |
d7e09d03 PT |
1419 | goto out; |
1420 | } | |
1421 | ||
1422 | int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, | |
1423 | struct lov_mds_md **lmmp, int *lmm_size, | |
1424 | struct ptlrpc_request **request) | |
1425 | { | |
1426 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1427 | struct mdt_body *body; | |
1428 | struct lov_mds_md *lmm = NULL; | |
1429 | struct ptlrpc_request *req = NULL; | |
1430 | struct md_op_data *op_data; | |
1431 | int rc, lmmsize; | |
1432 | ||
44779340 | 1433 | rc = ll_get_default_mdsize(sbi, &lmmsize); |
d7e09d03 | 1434 | if (rc) |
0a3bdb00 | 1435 | return rc; |
d7e09d03 PT |
1436 | |
1437 | op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, | |
1438 | strlen(filename), lmmsize, | |
1439 | LUSTRE_OPC_ANY, NULL); | |
1440 | if (IS_ERR(op_data)) | |
0a3bdb00 | 1441 | return PTR_ERR(op_data); |
d7e09d03 PT |
1442 | |
1443 | op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; | |
1444 | rc = md_getattr_name(sbi->ll_md_exp, op_data, &req); | |
1445 | ll_finish_md_op_data(op_data); | |
1446 | if (rc < 0) { | |
2d00bd17 JP |
1447 | CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n", |
1448 | filename, rc); | |
34e1f2bb | 1449 | goto out; |
d7e09d03 PT |
1450 | } |
1451 | ||
1452 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
d7e09d03 | 1453 | |
2e1b5b8b | 1454 | lmmsize = body->mbo_eadatasize; |
d7e09d03 | 1455 | |
2e1b5b8b | 1456 | if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) || |
e15ba45d | 1457 | lmmsize == 0) { |
34e1f2bb JL |
1458 | rc = -ENODATA; |
1459 | goto out; | |
d7e09d03 PT |
1460 | } |
1461 | ||
1462 | lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize); | |
d7e09d03 PT |
1463 | |
1464 | if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) && | |
1465 | (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) { | |
34e1f2bb JL |
1466 | rc = -EPROTO; |
1467 | goto out; | |
d7e09d03 PT |
1468 | } |
1469 | ||
1470 | /* | |
1471 | * This is coming from the MDS, so is probably in | |
1472 | * little endian. We convert it to host endian before | |
1473 | * passing it to userspace. | |
1474 | */ | |
1f6eaf83 | 1475 | if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) { |
5dd16419 JX |
1476 | int stripe_count; |
1477 | ||
1478 | stripe_count = le16_to_cpu(lmm->lmm_stripe_count); | |
1479 | if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED) | |
1480 | stripe_count = 0; | |
1481 | ||
d7e09d03 | 1482 | /* if function called for directory - we should |
c0894c6c OD |
1483 | * avoid swab not existent lsm objects |
1484 | */ | |
d7e09d03 PT |
1485 | if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) { |
1486 | lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); | |
2e1b5b8b | 1487 | if (S_ISREG(body->mbo_mode)) |
d7e09d03 PT |
1488 | lustre_swab_lov_user_md_objects( |
1489 | ((struct lov_user_md_v1 *)lmm)->lmm_objects, | |
5dd16419 | 1490 | stripe_count); |
d7e09d03 PT |
1491 | } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) { |
1492 | lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); | |
2e1b5b8b | 1493 | if (S_ISREG(body->mbo_mode)) |
d7e09d03 PT |
1494 | lustre_swab_lov_user_md_objects( |
1495 | ((struct lov_user_md_v3 *)lmm)->lmm_objects, | |
5dd16419 | 1496 | stripe_count); |
d7e09d03 PT |
1497 | } |
1498 | } | |
1499 | ||
1500 | out: | |
1501 | *lmmp = lmm; | |
1502 | *lmm_size = lmmsize; | |
1503 | *request = req; | |
1504 | return rc; | |
1505 | } | |
1506 | ||
1507 | static int ll_lov_setea(struct inode *inode, struct file *file, | |
e15ba45d | 1508 | unsigned long arg) |
d7e09d03 | 1509 | { |
d467220e | 1510 | __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE; |
d7e09d03 PT |
1511 | struct lov_user_md *lump; |
1512 | int lum_size = sizeof(struct lov_user_md) + | |
1513 | sizeof(struct lov_user_ost_data); | |
1514 | int rc; | |
d7e09d03 | 1515 | |
2eb90a75 | 1516 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1517 | return -EPERM; |
d7e09d03 | 1518 | |
e958f49b | 1519 | lump = libcfs_kvzalloc(lum_size, GFP_NOFS); |
6e16818b | 1520 | if (!lump) |
0a3bdb00 | 1521 | return -ENOMEM; |
d7e09d03 | 1522 | |
02f9c12e | 1523 | if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) { |
e958f49b | 1524 | kvfree(lump); |
0a3bdb00 | 1525 | return -EFAULT; |
d7e09d03 PT |
1526 | } |
1527 | ||
c139f3ce | 1528 | rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump, |
e15ba45d | 1529 | lum_size); |
c139f3ce | 1530 | cl_lov_delay_create_clear(&file->f_flags); |
d7e09d03 | 1531 | |
e958f49b | 1532 | kvfree(lump); |
0a3bdb00 | 1533 | return rc; |
d7e09d03 PT |
1534 | } |
1535 | ||
1536 | static int ll_lov_setstripe(struct inode *inode, struct file *file, | |
1537 | unsigned long arg) | |
1538 | { | |
02f9c12e OD |
1539 | struct lov_user_md_v3 lumv3; |
1540 | struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; | |
1541 | struct lov_user_md_v1 __user *lumv1p = (void __user *)arg; | |
1542 | struct lov_user_md_v3 __user *lumv3p = (void __user *)arg; | |
1543 | int lum_size, rc; | |
d467220e | 1544 | __u64 flags = FMODE_WRITE; |
d7e09d03 PT |
1545 | |
1546 | /* first try with v1 which is smaller than v3 */ | |
1547 | lum_size = sizeof(struct lov_user_md_v1); | |
1548 | if (copy_from_user(lumv1, lumv1p, lum_size)) | |
0a3bdb00 | 1549 | return -EFAULT; |
d7e09d03 PT |
1550 | |
1551 | if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { | |
1552 | lum_size = sizeof(struct lov_user_md_v3); | |
1553 | if (copy_from_user(&lumv3, lumv3p, lum_size)) | |
0a3bdb00 | 1554 | return -EFAULT; |
d7e09d03 PT |
1555 | } |
1556 | ||
c139f3ce AV |
1557 | rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lumv1, |
1558 | lum_size); | |
1559 | cl_lov_delay_create_clear(&file->f_flags); | |
d7e09d03 PT |
1560 | if (rc == 0) { |
1561 | struct lov_stripe_md *lsm; | |
1562 | __u32 gen; | |
1563 | ||
1564 | put_user(0, &lumv1p->lmm_stripe_count); | |
1565 | ||
1566 | ll_layout_refresh(inode, &gen); | |
1567 | lsm = ccc_inode_lsm_get(inode); | |
1568 | rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), | |
e09bee34 | 1569 | 0, lsm, (void __user *)arg); |
d7e09d03 PT |
1570 | ccc_inode_lsm_put(inode, lsm); |
1571 | } | |
0a3bdb00 | 1572 | return rc; |
d7e09d03 PT |
1573 | } |
1574 | ||
1575 | static int ll_lov_getstripe(struct inode *inode, unsigned long arg) | |
1576 | { | |
1577 | struct lov_stripe_md *lsm; | |
1578 | int rc = -ENODATA; | |
d7e09d03 PT |
1579 | |
1580 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1581 | if (lsm) |
d7e09d03 | 1582 | rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, |
e09bee34 | 1583 | lsm, (void __user *)arg); |
d7e09d03 | 1584 | ccc_inode_lsm_put(inode, lsm); |
0a3bdb00 | 1585 | return rc; |
d7e09d03 PT |
1586 | } |
1587 | ||
2d95f10e JH |
1588 | static int |
1589 | ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) | |
d7e09d03 PT |
1590 | { |
1591 | struct ll_inode_info *lli = ll_i2info(inode); | |
1592 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
98eae5e7 | 1593 | struct ll_grouplock grouplock; |
d7e09d03 | 1594 | int rc; |
d7e09d03 | 1595 | |
431b5678 PF |
1596 | if (arg == 0) { |
1597 | CWARN("group id for group lock must not be 0\n"); | |
1598 | return -EINVAL; | |
1599 | } | |
1600 | ||
d7e09d03 | 1601 | if (ll_file_nolock(file)) |
0a3bdb00 | 1602 | return -EOPNOTSUPP; |
d7e09d03 PT |
1603 | |
1604 | spin_lock(&lli->lli_lock); | |
1605 | if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { | |
1606 | CWARN("group lock already existed with gid %lu\n", | |
98eae5e7 | 1607 | fd->fd_grouplock.lg_gid); |
d7e09d03 | 1608 | spin_unlock(&lli->lli_lock); |
0a3bdb00 | 1609 | return -EINVAL; |
d7e09d03 | 1610 | } |
98eae5e7 | 1611 | LASSERT(!fd->fd_grouplock.lg_lock); |
d7e09d03 PT |
1612 | spin_unlock(&lli->lli_lock); |
1613 | ||
1929c433 | 1614 | rc = cl_get_grouplock(ll_i2info(inode)->lli_clob, |
d7e09d03 PT |
1615 | arg, (file->f_flags & O_NONBLOCK), &grouplock); |
1616 | if (rc) | |
0a3bdb00 | 1617 | return rc; |
d7e09d03 PT |
1618 | |
1619 | spin_lock(&lli->lli_lock); | |
1620 | if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { | |
1621 | spin_unlock(&lli->lli_lock); | |
1622 | CERROR("another thread just won the race\n"); | |
1623 | cl_put_grouplock(&grouplock); | |
0a3bdb00 | 1624 | return -EINVAL; |
d7e09d03 PT |
1625 | } |
1626 | ||
1627 | fd->fd_flags |= LL_FILE_GROUP_LOCKED; | |
1628 | fd->fd_grouplock = grouplock; | |
1629 | spin_unlock(&lli->lli_lock); | |
1630 | ||
1631 | CDEBUG(D_INFO, "group lock %lu obtained\n", arg); | |
0a3bdb00 | 1632 | return 0; |
d7e09d03 PT |
1633 | } |
1634 | ||
920b4f2e LC |
1635 | static int ll_put_grouplock(struct inode *inode, struct file *file, |
1636 | unsigned long arg) | |
d7e09d03 PT |
1637 | { |
1638 | struct ll_inode_info *lli = ll_i2info(inode); | |
1639 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
98eae5e7 | 1640 | struct ll_grouplock grouplock; |
d7e09d03 PT |
1641 | |
1642 | spin_lock(&lli->lli_lock); | |
1643 | if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { | |
1644 | spin_unlock(&lli->lli_lock); | |
1645 | CWARN("no group lock held\n"); | |
0a3bdb00 | 1646 | return -EINVAL; |
d7e09d03 | 1647 | } |
98eae5e7 | 1648 | LASSERT(fd->fd_grouplock.lg_lock); |
d7e09d03 | 1649 | |
98eae5e7 | 1650 | if (fd->fd_grouplock.lg_gid != arg) { |
d7e09d03 | 1651 | CWARN("group lock %lu doesn't match current id %lu\n", |
98eae5e7 | 1652 | arg, fd->fd_grouplock.lg_gid); |
d7e09d03 | 1653 | spin_unlock(&lli->lli_lock); |
0a3bdb00 | 1654 | return -EINVAL; |
d7e09d03 PT |
1655 | } |
1656 | ||
1657 | grouplock = fd->fd_grouplock; | |
1658 | memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); | |
1659 | fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; | |
1660 | spin_unlock(&lli->lli_lock); | |
1661 | ||
1662 | cl_put_grouplock(&grouplock); | |
1663 | CDEBUG(D_INFO, "group lock %lu released\n", arg); | |
0a3bdb00 | 1664 | return 0; |
d7e09d03 PT |
1665 | } |
1666 | ||
1667 | /** | |
1668 | * Close inode open handle | |
1669 | * | |
e22fdcc8 | 1670 | * \param inode [in] inode in question |
d7e09d03 PT |
1671 | * \param it [in,out] intent which contains open info and result |
1672 | * | |
1673 | * \retval 0 success | |
1674 | * \retval <0 failure | |
1675 | */ | |
e22fdcc8 | 1676 | int ll_release_openhandle(struct inode *inode, struct lookup_intent *it) |
d7e09d03 | 1677 | { |
d7e09d03 PT |
1678 | struct obd_client_handle *och; |
1679 | int rc; | |
d7e09d03 PT |
1680 | |
1681 | LASSERT(inode); | |
1682 | ||
1683 | /* Root ? Do nothing. */ | |
f76c23da | 1684 | if (is_root_inode(inode)) |
0a3bdb00 | 1685 | return 0; |
d7e09d03 PT |
1686 | |
1687 | /* No open handle to close? Move away */ | |
1688 | if (!it_disposition(it, DISP_OPEN_OPEN)) | |
0a3bdb00 | 1689 | return 0; |
d7e09d03 PT |
1690 | |
1691 | LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0); | |
1692 | ||
496a51bd | 1693 | och = kzalloc(sizeof(*och), GFP_NOFS); |
34e1f2bb JL |
1694 | if (!och) { |
1695 | rc = -ENOMEM; | |
1696 | goto out; | |
1697 | } | |
d7e09d03 | 1698 | |
ea1db081 | 1699 | ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och); |
d7e09d03 PT |
1700 | |
1701 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, | |
48d23e61 JX |
1702 | inode, och, NULL); |
1703 | out: | |
d7e09d03 PT |
1704 | /* this one is in place of ll_file_open */ |
1705 | if (it_disposition(it, DISP_ENQ_OPEN_REF)) { | |
8bf86fd9 | 1706 | ptlrpc_req_finished(it->it_request); |
d7e09d03 PT |
1707 | it_clear_disposition(it, DISP_ENQ_OPEN_REF); |
1708 | } | |
0a3bdb00 | 1709 | return rc; |
d7e09d03 PT |
1710 | } |
1711 | ||
1712 | /** | |
1713 | * Get size for inode for which FIEMAP mapping is requested. | |
1714 | * Make the FIEMAP get_info call and returns the result. | |
1715 | */ | |
2d95f10e | 1716 | static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, |
ebdc4fc5 | 1717 | size_t num_bytes) |
d7e09d03 PT |
1718 | { |
1719 | struct obd_export *exp = ll_i2dtexp(inode); | |
1720 | struct lov_stripe_md *lsm = NULL; | |
1721 | struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, }; | |
ebdc4fc5 | 1722 | __u32 vallen = num_bytes; |
d7e09d03 | 1723 | int rc; |
d7e09d03 PT |
1724 | |
1725 | /* Checks for fiemap flags */ | |
1726 | if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) { | |
1727 | fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT; | |
1728 | return -EBADR; | |
1729 | } | |
1730 | ||
1731 | /* Check for FIEMAP_FLAG_SYNC */ | |
1732 | if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) { | |
1733 | rc = filemap_fdatawrite(inode->i_mapping); | |
1734 | if (rc) | |
1735 | return rc; | |
1736 | } | |
1737 | ||
1738 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1739 | if (!lsm) |
d7e09d03 PT |
1740 | return -ENOENT; |
1741 | ||
1742 | /* If the stripe_count > 1 and the application does not understand | |
1743 | * DEVICE_ORDER flag, then it cannot interpret the extents correctly. | |
1744 | */ | |
1745 | if (lsm->lsm_stripe_count > 1 && | |
34e1f2bb JL |
1746 | !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) { |
1747 | rc = -EOPNOTSUPP; | |
1748 | goto out; | |
1749 | } | |
d7e09d03 PT |
1750 | |
1751 | fm_key.oa.o_oi = lsm->lsm_oi; | |
1752 | fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; | |
1753 | ||
a915ffda LD |
1754 | if (i_size_read(inode) == 0) { |
1755 | rc = ll_glimpse_size(inode); | |
1756 | if (rc) | |
1757 | goto out; | |
1758 | } | |
1759 | ||
d7e09d03 PT |
1760 | obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE); |
1761 | obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid); | |
1762 | /* If filesize is 0, then there would be no objects for mapping */ | |
1763 | if (fm_key.oa.o_size == 0) { | |
1764 | fiemap->fm_mapped_extents = 0; | |
34e1f2bb JL |
1765 | rc = 0; |
1766 | goto out; | |
d7e09d03 PT |
1767 | } |
1768 | ||
1769 | memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap)); | |
1770 | ||
1771 | rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen, | |
1772 | fiemap, lsm); | |
1773 | if (rc) | |
1774 | CERROR("obd_get_info failed: rc = %d\n", rc); | |
1775 | ||
1776 | out: | |
1777 | ccc_inode_lsm_put(inode, lsm); | |
0a3bdb00 | 1778 | return rc; |
d7e09d03 PT |
1779 | } |
1780 | ||
2b358b4e | 1781 | int ll_fid2path(struct inode *inode, void __user *arg) |
d7e09d03 | 1782 | { |
2b358b4e FZ |
1783 | struct obd_export *exp = ll_i2mdexp(inode); |
1784 | const struct getinfo_fid2path __user *gfin = arg; | |
1785 | struct getinfo_fid2path *gfout; | |
1786 | u32 pathlen; | |
1787 | size_t outsize; | |
1788 | int rc; | |
d7e09d03 | 1789 | |
2eb90a75 | 1790 | if (!capable(CFS_CAP_DAC_READ_SEARCH) && |
d7e09d03 | 1791 | !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH)) |
0a3bdb00 | 1792 | return -EPERM; |
d7e09d03 | 1793 | |
2b358b4e FZ |
1794 | /* Only need to get the buflen */ |
1795 | if (get_user(pathlen, &gfin->gf_pathlen)) | |
0a3bdb00 | 1796 | return -EFAULT; |
d7e09d03 | 1797 | |
c7b09efa OD |
1798 | if (pathlen > PATH_MAX) |
1799 | return -EINVAL; | |
1800 | ||
2b358b4e FZ |
1801 | outsize = sizeof(*gfout) + pathlen; |
1802 | ||
496a51bd JL |
1803 | gfout = kzalloc(outsize, GFP_NOFS); |
1804 | if (!gfout) | |
0a3bdb00 | 1805 | return -ENOMEM; |
2b358b4e | 1806 | |
34e1f2bb JL |
1807 | if (copy_from_user(gfout, arg, sizeof(*gfout))) { |
1808 | rc = -EFAULT; | |
1809 | goto gf_free; | |
1810 | } | |
d7e09d03 PT |
1811 | |
1812 | /* Call mdc_iocontrol */ | |
1813 | rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); | |
2b358b4e | 1814 | if (rc != 0) |
34e1f2bb | 1815 | goto gf_free; |
d7e09d03 PT |
1816 | |
1817 | if (copy_to_user(arg, gfout, outsize)) | |
1818 | rc = -EFAULT; | |
1819 | ||
1820 | gf_free: | |
97903a26 | 1821 | kfree(gfout); |
0a3bdb00 | 1822 | return rc; |
d7e09d03 PT |
1823 | } |
1824 | ||
1825 | static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) | |
1826 | { | |
1827 | struct ll_user_fiemap *fiemap_s; | |
1828 | size_t num_bytes, ret_bytes; | |
1829 | unsigned int extent_count; | |
1830 | int rc = 0; | |
1831 | ||
1832 | /* Get the extent count so we can calculate the size of | |
c0894c6c OD |
1833 | * required fiemap buffer |
1834 | */ | |
d7e09d03 | 1835 | if (get_user(extent_count, |
e15ba45d | 1836 | &((struct ll_user_fiemap __user *)arg)->fm_extent_count)) |
0a3bdb00 | 1837 | return -EFAULT; |
7bc3dfa3 VO |
1838 | |
1839 | if (extent_count >= | |
1840 | (SIZE_MAX - sizeof(*fiemap_s)) / sizeof(struct ll_fiemap_extent)) | |
1841 | return -EINVAL; | |
d7e09d03 PT |
1842 | num_bytes = sizeof(*fiemap_s) + (extent_count * |
1843 | sizeof(struct ll_fiemap_extent)); | |
1844 | ||
e958f49b | 1845 | fiemap_s = libcfs_kvzalloc(num_bytes, GFP_NOFS); |
6e16818b | 1846 | if (!fiemap_s) |
0a3bdb00 | 1847 | return -ENOMEM; |
d7e09d03 PT |
1848 | |
1849 | /* get the fiemap value */ | |
1850 | if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg, | |
34e1f2bb JL |
1851 | sizeof(*fiemap_s))) { |
1852 | rc = -EFAULT; | |
1853 | goto error; | |
1854 | } | |
d7e09d03 PT |
1855 | |
1856 | /* If fm_extent_count is non-zero, read the first extent since | |
1857 | * it is used to calculate end_offset and device from previous | |
c0894c6c OD |
1858 | * fiemap call. |
1859 | */ | |
d7e09d03 PT |
1860 | if (extent_count) { |
1861 | if (copy_from_user(&fiemap_s->fm_extents[0], | |
e15ba45d OD |
1862 | (char __user *)arg + sizeof(*fiemap_s), |
1863 | sizeof(struct ll_fiemap_extent))) { | |
34e1f2bb JL |
1864 | rc = -EFAULT; |
1865 | goto error; | |
1866 | } | |
d7e09d03 PT |
1867 | } |
1868 | ||
1869 | rc = ll_do_fiemap(inode, fiemap_s, num_bytes); | |
1870 | if (rc) | |
34e1f2bb | 1871 | goto error; |
d7e09d03 PT |
1872 | |
1873 | ret_bytes = sizeof(struct ll_user_fiemap); | |
1874 | ||
1875 | if (extent_count != 0) | |
1876 | ret_bytes += (fiemap_s->fm_mapped_extents * | |
1877 | sizeof(struct ll_fiemap_extent)); | |
1878 | ||
02f9c12e | 1879 | if (copy_to_user((void __user *)arg, fiemap_s, ret_bytes)) |
d7e09d03 PT |
1880 | rc = -EFAULT; |
1881 | ||
1882 | error: | |
e958f49b | 1883 | kvfree(fiemap_s); |
0a3bdb00 | 1884 | return rc; |
d7e09d03 PT |
1885 | } |
1886 | ||
1887 | /* | |
1888 | * Read the data_version for inode. | |
1889 | * | |
1890 | * This value is computed using stripe object version on OST. | |
1891 | * Version is computed using server side locking. | |
1892 | * | |
e1798006 JX |
1893 | * @param sync if do sync on the OST side; |
1894 | * 0: no sync | |
1895 | * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs | |
1896 | * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs | |
d7e09d03 | 1897 | */ |
e1798006 | 1898 | int ll_data_version(struct inode *inode, __u64 *data_version, int flags) |
d7e09d03 PT |
1899 | { |
1900 | struct lov_stripe_md *lsm = NULL; | |
1901 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1902 | struct obdo *obdo = NULL; | |
1903 | int rc; | |
d7e09d03 PT |
1904 | |
1905 | /* If no stripe, we consider version is 0. */ | |
1906 | lsm = ccc_inode_lsm_get(inode); | |
5dd16419 | 1907 | if (!lsm_has_objects(lsm)) { |
d7e09d03 PT |
1908 | *data_version = 0; |
1909 | CDEBUG(D_INODE, "No object for inode\n"); | |
34e1f2bb JL |
1910 | rc = 0; |
1911 | goto out; | |
d7e09d03 PT |
1912 | } |
1913 | ||
496a51bd JL |
1914 | obdo = kzalloc(sizeof(*obdo), GFP_NOFS); |
1915 | if (!obdo) { | |
34e1f2bb JL |
1916 | rc = -ENOMEM; |
1917 | goto out; | |
1918 | } | |
d7e09d03 | 1919 | |
e1798006 | 1920 | rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, obdo, 0, flags); |
5dd16419 | 1921 | if (rc == 0) { |
d7e09d03 PT |
1922 | if (!(obdo->o_valid & OBD_MD_FLDATAVERSION)) |
1923 | rc = -EOPNOTSUPP; | |
1924 | else | |
1925 | *data_version = obdo->o_data_version; | |
1926 | } | |
1927 | ||
97903a26 | 1928 | kfree(obdo); |
5dd16419 | 1929 | out: |
d7e09d03 | 1930 | ccc_inode_lsm_put(inode, lsm); |
0a3bdb00 | 1931 | return rc; |
d7e09d03 PT |
1932 | } |
1933 | ||
48d23e61 JX |
1934 | /* |
1935 | * Trigger a HSM release request for the provided inode. | |
1936 | */ | |
1937 | int ll_hsm_release(struct inode *inode) | |
1938 | { | |
1939 | struct cl_env_nest nest; | |
1940 | struct lu_env *env; | |
1941 | struct obd_client_handle *och = NULL; | |
1942 | __u64 data_version = 0; | |
1943 | int rc; | |
1944 | ||
48d23e61 JX |
1945 | CDEBUG(D_INODE, "%s: Releasing file "DFID".\n", |
1946 | ll_get_fsname(inode->i_sb, NULL, 0), | |
1947 | PFID(&ll_i2info(inode)->lli_fid)); | |
1948 | ||
1949 | och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE); | |
34e1f2bb JL |
1950 | if (IS_ERR(och)) { |
1951 | rc = PTR_ERR(och); | |
1952 | goto out; | |
1953 | } | |
48d23e61 JX |
1954 | |
1955 | /* Grab latest data_version and [am]time values */ | |
e1798006 | 1956 | rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH); |
48d23e61 | 1957 | if (rc != 0) |
34e1f2bb | 1958 | goto out; |
48d23e61 JX |
1959 | |
1960 | env = cl_env_nested_get(&nest); | |
34e1f2bb JL |
1961 | if (IS_ERR(env)) { |
1962 | rc = PTR_ERR(env); | |
1963 | goto out; | |
1964 | } | |
48d23e61 | 1965 | |
d2995737 | 1966 | ll_merge_attr(env, inode); |
48d23e61 JX |
1967 | cl_env_nested_put(&nest, env); |
1968 | ||
1969 | /* Release the file. | |
1970 | * NB: lease lock handle is released in mdc_hsm_release_pack() because | |
c0894c6c OD |
1971 | * we still need it to pack l_remote_handle to MDT. |
1972 | */ | |
48d23e61 JX |
1973 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och, |
1974 | &data_version); | |
1975 | och = NULL; | |
1976 | ||
48d23e61 | 1977 | out: |
6e16818b | 1978 | if (och && !IS_ERR(och)) /* close the file */ |
48d23e61 JX |
1979 | ll_lease_close(och, inode, NULL); |
1980 | ||
1981 | return rc; | |
1982 | } | |
1983 | ||
d7e09d03 PT |
1984 | struct ll_swap_stack { |
1985 | struct iattr ia1, ia2; | |
1986 | __u64 dv1, dv2; | |
1987 | struct inode *inode1, *inode2; | |
1988 | bool check_dv1, check_dv2; | |
1989 | }; | |
1990 | ||
1991 | static int ll_swap_layouts(struct file *file1, struct file *file2, | |
1992 | struct lustre_swap_layouts *lsl) | |
1993 | { | |
1994 | struct mdc_swap_layouts msl; | |
1995 | struct md_op_data *op_data; | |
1996 | __u32 gid; | |
1997 | __u64 dv; | |
1998 | struct ll_swap_stack *llss = NULL; | |
1999 | int rc; | |
2000 | ||
496a51bd JL |
2001 | llss = kzalloc(sizeof(*llss), GFP_NOFS); |
2002 | if (!llss) | |
0a3bdb00 | 2003 | return -ENOMEM; |
d7e09d03 | 2004 | |
2a8a3597 AV |
2005 | llss->inode1 = file_inode(file1); |
2006 | llss->inode2 = file_inode(file2); | |
d7e09d03 | 2007 | |
34e1f2bb JL |
2008 | if (!S_ISREG(llss->inode2->i_mode)) { |
2009 | rc = -EINVAL; | |
2010 | goto free; | |
2011 | } | |
d7e09d03 | 2012 | |
9c5fb72c | 2013 | if (inode_permission(llss->inode1, MAY_WRITE) || |
34e1f2bb JL |
2014 | inode_permission(llss->inode2, MAY_WRITE)) { |
2015 | rc = -EPERM; | |
2016 | goto free; | |
2017 | } | |
d7e09d03 | 2018 | |
34e1f2bb JL |
2019 | if (llss->inode2->i_sb != llss->inode1->i_sb) { |
2020 | rc = -EXDEV; | |
2021 | goto free; | |
2022 | } | |
d7e09d03 PT |
2023 | |
2024 | /* we use 2 bool because it is easier to swap than 2 bits */ | |
2025 | if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1) | |
2026 | llss->check_dv1 = true; | |
2027 | ||
2028 | if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2) | |
2029 | llss->check_dv2 = true; | |
2030 | ||
2031 | /* we cannot use lsl->sl_dvX directly because we may swap them */ | |
2032 | llss->dv1 = lsl->sl_dv1; | |
2033 | llss->dv2 = lsl->sl_dv2; | |
2034 | ||
2035 | rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2)); | |
34e1f2bb JL |
2036 | if (rc == 0) /* same file, done! */ { |
2037 | rc = 0; | |
2038 | goto free; | |
2039 | } | |
d7e09d03 PT |
2040 | |
2041 | if (rc < 0) { /* sequentialize it */ | |
2042 | swap(llss->inode1, llss->inode2); | |
2043 | swap(file1, file2); | |
2044 | swap(llss->dv1, llss->dv2); | |
2045 | swap(llss->check_dv1, llss->check_dv2); | |
2046 | } | |
2047 | ||
2048 | gid = lsl->sl_gid; | |
2049 | if (gid != 0) { /* application asks to flush dirty cache */ | |
2050 | rc = ll_get_grouplock(llss->inode1, file1, gid); | |
2051 | if (rc < 0) | |
34e1f2bb | 2052 | goto free; |
d7e09d03 PT |
2053 | |
2054 | rc = ll_get_grouplock(llss->inode2, file2, gid); | |
2055 | if (rc < 0) { | |
2056 | ll_put_grouplock(llss->inode1, file1, gid); | |
34e1f2bb | 2057 | goto free; |
d7e09d03 PT |
2058 | } |
2059 | } | |
2060 | ||
2061 | /* to be able to restore mtime and atime after swap | |
c0894c6c OD |
2062 | * we need to first save them |
2063 | */ | |
d7e09d03 PT |
2064 | if (lsl->sl_flags & |
2065 | (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) { | |
2066 | llss->ia1.ia_mtime = llss->inode1->i_mtime; | |
2067 | llss->ia1.ia_atime = llss->inode1->i_atime; | |
2068 | llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME; | |
2069 | llss->ia2.ia_mtime = llss->inode2->i_mtime; | |
2070 | llss->ia2.ia_atime = llss->inode2->i_atime; | |
2071 | llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME; | |
2072 | } | |
2073 | ||
d0a0acc3 | 2074 | /* ultimate check, before swapping the layouts we check if |
c0894c6c OD |
2075 | * dataversion has changed (if requested) |
2076 | */ | |
d7e09d03 PT |
2077 | if (llss->check_dv1) { |
2078 | rc = ll_data_version(llss->inode1, &dv, 0); | |
2079 | if (rc) | |
34e1f2bb JL |
2080 | goto putgl; |
2081 | if (dv != llss->dv1) { | |
2082 | rc = -EAGAIN; | |
2083 | goto putgl; | |
2084 | } | |
d7e09d03 PT |
2085 | } |
2086 | ||
2087 | if (llss->check_dv2) { | |
2088 | rc = ll_data_version(llss->inode2, &dv, 0); | |
2089 | if (rc) | |
34e1f2bb JL |
2090 | goto putgl; |
2091 | if (dv != llss->dv2) { | |
2092 | rc = -EAGAIN; | |
2093 | goto putgl; | |
2094 | } | |
d7e09d03 PT |
2095 | } |
2096 | ||
2097 | /* struct md_op_data is used to send the swap args to the mdt | |
2098 | * only flags is missing, so we use struct mdc_swap_layouts | |
c0894c6c OD |
2099 | * through the md_op_data->op_data |
2100 | */ | |
d7e09d03 | 2101 | /* flags from user space have to be converted before they are send to |
c0894c6c OD |
2102 | * server, no flag is sent today, they are only used on the client |
2103 | */ | |
d7e09d03 PT |
2104 | msl.msl_flags = 0; |
2105 | rc = -ENOMEM; | |
2106 | op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0, | |
2107 | 0, LUSTRE_OPC_ANY, &msl); | |
34e1f2bb JL |
2108 | if (IS_ERR(op_data)) { |
2109 | rc = PTR_ERR(op_data); | |
2110 | goto free; | |
2111 | } | |
79a8726a JH |
2112 | |
2113 | rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1), | |
2114 | sizeof(*op_data), op_data, NULL); | |
2115 | ll_finish_md_op_data(op_data); | |
d7e09d03 PT |
2116 | |
2117 | putgl: | |
2118 | if (gid != 0) { | |
2119 | ll_put_grouplock(llss->inode2, file2, gid); | |
2120 | ll_put_grouplock(llss->inode1, file1, gid); | |
2121 | } | |
2122 | ||
2123 | /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */ | |
2124 | if (rc != 0) | |
34e1f2bb | 2125 | goto free; |
d7e09d03 PT |
2126 | |
2127 | /* clear useless flags */ | |
2128 | if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) { | |
2129 | llss->ia1.ia_valid &= ~ATTR_MTIME; | |
2130 | llss->ia2.ia_valid &= ~ATTR_MTIME; | |
2131 | } | |
2132 | ||
2133 | if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) { | |
2134 | llss->ia1.ia_valid &= ~ATTR_ATIME; | |
2135 | llss->ia2.ia_valid &= ~ATTR_ATIME; | |
2136 | } | |
2137 | ||
2138 | /* update time if requested */ | |
2139 | rc = 0; | |
2140 | if (llss->ia2.ia_valid != 0) { | |
5955102c | 2141 | inode_lock(llss->inode1); |
b583043e | 2142 | rc = ll_setattr(file1->f_path.dentry, &llss->ia2); |
5955102c | 2143 | inode_unlock(llss->inode1); |
d7e09d03 PT |
2144 | } |
2145 | ||
2146 | if (llss->ia1.ia_valid != 0) { | |
2147 | int rc1; | |
2148 | ||
5955102c | 2149 | inode_lock(llss->inode2); |
b583043e | 2150 | rc1 = ll_setattr(file2->f_path.dentry, &llss->ia1); |
5955102c | 2151 | inode_unlock(llss->inode2); |
d7e09d03 PT |
2152 | if (rc == 0) |
2153 | rc = rc1; | |
2154 | } | |
2155 | ||
2156 | free: | |
e6b9a3b2 | 2157 | kfree(llss); |
d7e09d03 | 2158 | |
0a3bdb00 | 2159 | return rc; |
d7e09d03 PT |
2160 | } |
2161 | ||
a720b790 JL |
2162 | static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss) |
2163 | { | |
2164 | struct md_op_data *op_data; | |
2165 | int rc; | |
2166 | ||
2742c75e BF |
2167 | /* Detect out-of range masks */ |
2168 | if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK) | |
2169 | return -EINVAL; | |
2170 | ||
a720b790 | 2171 | /* Non-root users are forbidden to set or clear flags which are |
c0894c6c OD |
2172 | * NOT defined in HSM_USER_MASK. |
2173 | */ | |
a720b790 | 2174 | if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) && |
2eb90a75 | 2175 | !capable(CFS_CAP_SYS_ADMIN)) |
a720b790 JL |
2176 | return -EPERM; |
2177 | ||
2742c75e BF |
2178 | /* Detect out-of range archive id */ |
2179 | if ((hss->hss_valid & HSS_ARCHIVE_ID) && | |
2180 | (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE)) | |
2181 | return -EINVAL; | |
2182 | ||
a720b790 JL |
2183 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, |
2184 | LUSTRE_OPC_ANY, hss); | |
2185 | if (IS_ERR(op_data)) | |
2186 | return PTR_ERR(op_data); | |
2187 | ||
2188 | rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode), | |
2189 | sizeof(*op_data), op_data, NULL); | |
2190 | ||
2191 | ll_finish_md_op_data(op_data); | |
2192 | ||
2193 | return rc; | |
2194 | } | |
2195 | ||
2196 | static int ll_hsm_import(struct inode *inode, struct file *file, | |
2197 | struct hsm_user_import *hui) | |
2198 | { | |
2199 | struct hsm_state_set *hss = NULL; | |
2200 | struct iattr *attr = NULL; | |
2201 | int rc; | |
2202 | ||
a720b790 JL |
2203 | if (!S_ISREG(inode->i_mode)) |
2204 | return -EINVAL; | |
2205 | ||
2206 | /* set HSM flags */ | |
496a51bd | 2207 | hss = kzalloc(sizeof(*hss), GFP_NOFS); |
e6b9a3b2 JL |
2208 | if (!hss) |
2209 | return -ENOMEM; | |
a720b790 JL |
2210 | |
2211 | hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID; | |
2212 | hss->hss_archive_id = hui->hui_archive_id; | |
2213 | hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED; | |
2214 | rc = ll_hsm_state_set(inode, hss); | |
2215 | if (rc != 0) | |
e6b9a3b2 | 2216 | goto free_hss; |
a720b790 | 2217 | |
496a51bd JL |
2218 | attr = kzalloc(sizeof(*attr), GFP_NOFS); |
2219 | if (!attr) { | |
34e1f2bb | 2220 | rc = -ENOMEM; |
e6b9a3b2 | 2221 | goto free_hss; |
34e1f2bb | 2222 | } |
a720b790 JL |
2223 | |
2224 | attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO); | |
2225 | attr->ia_mode |= S_IFREG; | |
2226 | attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid); | |
2227 | attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid); | |
2228 | attr->ia_size = hui->hui_size; | |
2229 | attr->ia_mtime.tv_sec = hui->hui_mtime; | |
2230 | attr->ia_mtime.tv_nsec = hui->hui_mtime_ns; | |
2231 | attr->ia_atime.tv_sec = hui->hui_atime; | |
2232 | attr->ia_atime.tv_nsec = hui->hui_atime_ns; | |
2233 | ||
2234 | attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE | | |
2235 | ATTR_UID | ATTR_GID | | |
2236 | ATTR_MTIME | ATTR_MTIME_SET | | |
2237 | ATTR_ATIME | ATTR_ATIME_SET; | |
2238 | ||
5955102c | 2239 | inode_lock(inode); |
b6ee56fe | 2240 | |
b583043e | 2241 | rc = ll_setattr_raw(file->f_path.dentry, attr, true); |
a720b790 JL |
2242 | if (rc == -ENODATA) |
2243 | rc = 0; | |
2244 | ||
5955102c | 2245 | inode_unlock(inode); |
b6ee56fe | 2246 | |
e6b9a3b2 JL |
2247 | kfree(attr); |
2248 | free_hss: | |
2249 | kfree(hss); | |
a720b790 JL |
2250 | return rc; |
2251 | } | |
2252 | ||
2d95f10e JH |
2253 | static long |
2254 | ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
d7e09d03 | 2255 | { |
2a8a3597 | 2256 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2257 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
2258 | int flags, rc; | |
d7e09d03 | 2259 | |
97a075cd JN |
2260 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),cmd=%x\n", |
2261 | PFID(ll_inode2fid(inode)), inode, cmd); | |
d7e09d03 PT |
2262 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1); |
2263 | ||
2264 | /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */ | |
2265 | if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */ | |
0a3bdb00 | 2266 | return -ENOTTY; |
d7e09d03 | 2267 | |
a58a38ac | 2268 | switch (cmd) { |
d7e09d03 PT |
2269 | case LL_IOC_GETFLAGS: |
2270 | /* Get the current value of the file flags */ | |
02f9c12e | 2271 | return put_user(fd->fd_flags, (int __user *)arg); |
d7e09d03 PT |
2272 | case LL_IOC_SETFLAGS: |
2273 | case LL_IOC_CLRFLAGS: | |
2274 | /* Set or clear specific file flags */ | |
2275 | /* XXX This probably needs checks to ensure the flags are | |
2276 | * not abused, and to handle any flag side effects. | |
2277 | */ | |
02f9c12e | 2278 | if (get_user(flags, (int __user *)arg)) |
0a3bdb00 | 2279 | return -EFAULT; |
d7e09d03 PT |
2280 | |
2281 | if (cmd == LL_IOC_SETFLAGS) { | |
2282 | if ((flags & LL_FILE_IGNORE_LOCK) && | |
2283 | !(file->f_flags & O_DIRECT)) { | |
2d00bd17 JP |
2284 | CERROR("%s: unable to disable locking on non-O_DIRECT file\n", |
2285 | current->comm); | |
0a3bdb00 | 2286 | return -EINVAL; |
d7e09d03 PT |
2287 | } |
2288 | ||
2289 | fd->fd_flags |= flags; | |
2290 | } else { | |
2291 | fd->fd_flags &= ~flags; | |
2292 | } | |
0a3bdb00 | 2293 | return 0; |
d7e09d03 | 2294 | case LL_IOC_LOV_SETSTRIPE: |
0a3bdb00 | 2295 | return ll_lov_setstripe(inode, file, arg); |
d7e09d03 | 2296 | case LL_IOC_LOV_SETEA: |
0a3bdb00 | 2297 | return ll_lov_setea(inode, file, arg); |
d7e09d03 PT |
2298 | case LL_IOC_LOV_SWAP_LAYOUTS: { |
2299 | struct file *file2; | |
2300 | struct lustre_swap_layouts lsl; | |
2301 | ||
02f9c12e OD |
2302 | if (copy_from_user(&lsl, (char __user *)arg, |
2303 | sizeof(struct lustre_swap_layouts))) | |
0a3bdb00 | 2304 | return -EFAULT; |
d7e09d03 PT |
2305 | |
2306 | if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */ | |
0a3bdb00 | 2307 | return -EPERM; |
d7e09d03 PT |
2308 | |
2309 | file2 = fget(lsl.sl_fd); | |
6e16818b | 2310 | if (!file2) |
0a3bdb00 | 2311 | return -EBADF; |
d7e09d03 PT |
2312 | |
2313 | rc = -EPERM; | |
2314 | if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */ | |
2315 | rc = ll_swap_layouts(file, file2, &lsl); | |
2316 | fput(file2); | |
0a3bdb00 | 2317 | return rc; |
d7e09d03 PT |
2318 | } |
2319 | case LL_IOC_LOV_GETSTRIPE: | |
0a3bdb00 | 2320 | return ll_lov_getstripe(inode, arg); |
d7e09d03 | 2321 | case LL_IOC_RECREATE_OBJ: |
0a3bdb00 | 2322 | return ll_lov_recreate_obj(inode, arg); |
d7e09d03 | 2323 | case LL_IOC_RECREATE_FID: |
0a3bdb00 | 2324 | return ll_lov_recreate_fid(inode, arg); |
d7e09d03 | 2325 | case FSFILT_IOC_FIEMAP: |
0a3bdb00 | 2326 | return ll_ioctl_fiemap(inode, arg); |
d7e09d03 PT |
2327 | case FSFILT_IOC_GETFLAGS: |
2328 | case FSFILT_IOC_SETFLAGS: | |
0a3bdb00 | 2329 | return ll_iocontrol(inode, file, cmd, arg); |
d7e09d03 PT |
2330 | case FSFILT_IOC_GETVERSION_OLD: |
2331 | case FSFILT_IOC_GETVERSION: | |
02f9c12e | 2332 | return put_user(inode->i_generation, (int __user *)arg); |
d7e09d03 | 2333 | case LL_IOC_GROUP_LOCK: |
0a3bdb00 | 2334 | return ll_get_grouplock(inode, file, arg); |
d7e09d03 | 2335 | case LL_IOC_GROUP_UNLOCK: |
0a3bdb00 | 2336 | return ll_put_grouplock(inode, file, arg); |
d7e09d03 | 2337 | case IOC_OBD_STATFS: |
4c6243ec | 2338 | return ll_obd_statfs(inode, (void __user *)arg); |
d7e09d03 PT |
2339 | |
2340 | /* We need to special case any other ioctls we want to handle, | |
2341 | * to send them to the MDS/OST as appropriate and to properly | |
2342 | * network encode the arg field. | |
2343 | case FSFILT_IOC_SETVERSION_OLD: | |
2344 | case FSFILT_IOC_SETVERSION: | |
2345 | */ | |
2346 | case LL_IOC_FLUSHCTX: | |
0a3bdb00 | 2347 | return ll_flush_ctx(inode); |
d7e09d03 | 2348 | case LL_IOC_PATH2FID: { |
02f9c12e | 2349 | if (copy_to_user((void __user *)arg, ll_inode2fid(inode), |
d7e09d03 | 2350 | sizeof(struct lu_fid))) |
0a3bdb00 | 2351 | return -EFAULT; |
d7e09d03 | 2352 | |
0a3bdb00 | 2353 | return 0; |
d7e09d03 PT |
2354 | } |
2355 | case OBD_IOC_FID2PATH: | |
61dad0ba | 2356 | return ll_fid2path(inode, (void __user *)arg); |
d7e09d03 PT |
2357 | case LL_IOC_DATA_VERSION: { |
2358 | struct ioc_data_version idv; | |
2359 | int rc; | |
2360 | ||
02f9c12e | 2361 | if (copy_from_user(&idv, (char __user *)arg, sizeof(idv))) |
0a3bdb00 | 2362 | return -EFAULT; |
d7e09d03 | 2363 | |
e1798006 JX |
2364 | idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH; |
2365 | rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags); | |
02f9c12e OD |
2366 | if (rc == 0 && copy_to_user((char __user *)arg, &idv, |
2367 | sizeof(idv))) | |
0a3bdb00 | 2368 | return -EFAULT; |
d7e09d03 | 2369 | |
0a3bdb00 | 2370 | return rc; |
d7e09d03 PT |
2371 | } |
2372 | ||
2373 | case LL_IOC_GET_MDTIDX: { | |
2374 | int mdtidx; | |
2375 | ||
2376 | mdtidx = ll_get_mdt_idx(inode); | |
2377 | if (mdtidx < 0) | |
0a3bdb00 | 2378 | return mdtidx; |
d7e09d03 | 2379 | |
02f9c12e | 2380 | if (put_user(mdtidx, (int __user *)arg)) |
0a3bdb00 | 2381 | return -EFAULT; |
d7e09d03 | 2382 | |
0a3bdb00 | 2383 | return 0; |
d7e09d03 PT |
2384 | } |
2385 | case OBD_IOC_GETDTNAME: | |
2386 | case OBD_IOC_GETMDNAME: | |
0a3bdb00 | 2387 | return ll_get_obd_name(inode, cmd, arg); |
d7e09d03 PT |
2388 | case LL_IOC_HSM_STATE_GET: { |
2389 | struct md_op_data *op_data; | |
2390 | struct hsm_user_state *hus; | |
2391 | int rc; | |
2392 | ||
496a51bd JL |
2393 | hus = kzalloc(sizeof(*hus), GFP_NOFS); |
2394 | if (!hus) | |
0a3bdb00 | 2395 | return -ENOMEM; |
d7e09d03 PT |
2396 | |
2397 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2398 | LUSTRE_OPC_ANY, hus); | |
79a8726a | 2399 | if (IS_ERR(op_data)) { |
97903a26 | 2400 | kfree(hus); |
0a3bdb00 | 2401 | return PTR_ERR(op_data); |
d7e09d03 PT |
2402 | } |
2403 | ||
2404 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2405 | op_data, NULL); | |
2406 | ||
02f9c12e | 2407 | if (copy_to_user((void __user *)arg, hus, sizeof(*hus))) |
d7e09d03 PT |
2408 | rc = -EFAULT; |
2409 | ||
2410 | ll_finish_md_op_data(op_data); | |
97903a26 | 2411 | kfree(hus); |
0a3bdb00 | 2412 | return rc; |
d7e09d03 PT |
2413 | } |
2414 | case LL_IOC_HSM_STATE_SET: { | |
d7e09d03 PT |
2415 | struct hsm_state_set *hss; |
2416 | int rc; | |
2417 | ||
02f9c12e | 2418 | hss = memdup_user((char __user *)arg, sizeof(*hss)); |
0c027bc3 AH |
2419 | if (IS_ERR(hss)) |
2420 | return PTR_ERR(hss); | |
d7e09d03 | 2421 | |
a720b790 | 2422 | rc = ll_hsm_state_set(inode, hss); |
d7e09d03 | 2423 | |
97903a26 | 2424 | kfree(hss); |
0a3bdb00 | 2425 | return rc; |
d7e09d03 PT |
2426 | } |
2427 | case LL_IOC_HSM_ACTION: { | |
2428 | struct md_op_data *op_data; | |
2429 | struct hsm_current_action *hca; | |
2430 | int rc; | |
2431 | ||
496a51bd JL |
2432 | hca = kzalloc(sizeof(*hca), GFP_NOFS); |
2433 | if (!hca) | |
0a3bdb00 | 2434 | return -ENOMEM; |
d7e09d03 PT |
2435 | |
2436 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2437 | LUSTRE_OPC_ANY, hca); | |
79a8726a | 2438 | if (IS_ERR(op_data)) { |
97903a26 | 2439 | kfree(hca); |
0a3bdb00 | 2440 | return PTR_ERR(op_data); |
d7e09d03 PT |
2441 | } |
2442 | ||
2443 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2444 | op_data, NULL); | |
2445 | ||
02f9c12e | 2446 | if (copy_to_user((char __user *)arg, hca, sizeof(*hca))) |
d7e09d03 PT |
2447 | rc = -EFAULT; |
2448 | ||
2449 | ll_finish_md_op_data(op_data); | |
97903a26 | 2450 | kfree(hca); |
0a3bdb00 | 2451 | return rc; |
d7e09d03 | 2452 | } |
d3a8a4e2 JX |
2453 | case LL_IOC_SET_LEASE: { |
2454 | struct ll_inode_info *lli = ll_i2info(inode); | |
2455 | struct obd_client_handle *och = NULL; | |
2456 | bool lease_broken; | |
2457 | fmode_t mode = 0; | |
2458 | ||
2459 | switch (arg) { | |
2460 | case F_WRLCK: | |
2461 | if (!(file->f_mode & FMODE_WRITE)) | |
2462 | return -EPERM; | |
2463 | mode = FMODE_WRITE; | |
2464 | break; | |
2465 | case F_RDLCK: | |
2466 | if (!(file->f_mode & FMODE_READ)) | |
2467 | return -EPERM; | |
2468 | mode = FMODE_READ; | |
2469 | break; | |
2470 | case F_UNLCK: | |
2471 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2472 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
2473 | och = fd->fd_lease_och; |
2474 | fd->fd_lease_och = NULL; | |
2475 | } | |
2476 | mutex_unlock(&lli->lli_och_mutex); | |
2477 | ||
6e16818b | 2478 | if (och) { |
d3a8a4e2 JX |
2479 | mode = och->och_flags & |
2480 | (FMODE_READ|FMODE_WRITE); | |
2481 | rc = ll_lease_close(och, inode, &lease_broken); | |
2482 | if (rc == 0 && lease_broken) | |
2483 | mode = 0; | |
2484 | } else { | |
2485 | rc = -ENOLCK; | |
2486 | } | |
2487 | ||
2488 | /* return the type of lease or error */ | |
2489 | return rc < 0 ? rc : (int)mode; | |
2490 | default: | |
2491 | return -EINVAL; | |
2492 | } | |
2493 | ||
2494 | CDEBUG(D_INODE, "Set lease with mode %d\n", mode); | |
2495 | ||
2496 | /* apply for lease */ | |
48d23e61 | 2497 | och = ll_lease_open(inode, file, mode, 0); |
d3a8a4e2 JX |
2498 | if (IS_ERR(och)) |
2499 | return PTR_ERR(och); | |
2500 | ||
2501 | rc = 0; | |
2502 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2503 | if (!fd->fd_lease_och) { |
d3a8a4e2 JX |
2504 | fd->fd_lease_och = och; |
2505 | och = NULL; | |
2506 | } | |
2507 | mutex_unlock(&lli->lli_och_mutex); | |
6e16818b | 2508 | if (och) { |
d3a8a4e2 JX |
2509 | /* impossible now that only excl is supported for now */ |
2510 | ll_lease_close(och, inode, &lease_broken); | |
2511 | rc = -EBUSY; | |
2512 | } | |
2513 | return rc; | |
2514 | } | |
2515 | case LL_IOC_GET_LEASE: { | |
2516 | struct ll_inode_info *lli = ll_i2info(inode); | |
2517 | struct ldlm_lock *lock = NULL; | |
2518 | ||
2519 | rc = 0; | |
2520 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2521 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
2522 | struct obd_client_handle *och = fd->fd_lease_och; |
2523 | ||
2524 | lock = ldlm_handle2lock(&och->och_lease_handle); | |
6e16818b | 2525 | if (lock) { |
d3a8a4e2 JX |
2526 | lock_res_and_lock(lock); |
2527 | if (!ldlm_is_cancel(lock)) | |
2528 | rc = och->och_flags & | |
2529 | (FMODE_READ | FMODE_WRITE); | |
2530 | unlock_res_and_lock(lock); | |
ead02808 | 2531 | LDLM_LOCK_PUT(lock); |
d3a8a4e2 JX |
2532 | } |
2533 | } | |
2534 | mutex_unlock(&lli->lli_och_mutex); | |
a720b790 JL |
2535 | return rc; |
2536 | } | |
2537 | case LL_IOC_HSM_IMPORT: { | |
2538 | struct hsm_user_import *hui; | |
2539 | ||
02f9c12e | 2540 | hui = memdup_user((void __user *)arg, sizeof(*hui)); |
0c027bc3 AH |
2541 | if (IS_ERR(hui)) |
2542 | return PTR_ERR(hui); | |
a720b790 JL |
2543 | |
2544 | rc = ll_hsm_import(inode, file, hui); | |
d3a8a4e2 | 2545 | |
97903a26 | 2546 | kfree(hui); |
d3a8a4e2 JX |
2547 | return rc; |
2548 | } | |
d7e09d03 PT |
2549 | default: { |
2550 | int err; | |
2551 | ||
1f6eaf83 JL |
2552 | if (ll_iocontrol_call(inode, file, cmd, arg, &err) == |
2553 | LLIOC_STOP) | |
0a3bdb00 | 2554 | return err; |
d7e09d03 | 2555 | |
0a3bdb00 | 2556 | return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, |
e09bee34 | 2557 | (void __user *)arg); |
d7e09d03 PT |
2558 | } |
2559 | } | |
2560 | } | |
2561 | ||
2d95f10e | 2562 | static loff_t ll_file_seek(struct file *file, loff_t offset, int origin) |
d7e09d03 | 2563 | { |
2a8a3597 | 2564 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2565 | loff_t retval, eof = 0; |
2566 | ||
d7e09d03 PT |
2567 | retval = offset + ((origin == SEEK_END) ? i_size_read(inode) : |
2568 | (origin == SEEK_CUR) ? file->f_pos : 0); | |
97a075cd JN |
2569 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n", |
2570 | PFID(ll_inode2fid(inode)), inode, retval, retval, origin); | |
d7e09d03 PT |
2571 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); |
2572 | ||
2573 | if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) { | |
2574 | retval = ll_glimpse_size(inode); | |
2575 | if (retval != 0) | |
0a3bdb00 | 2576 | return retval; |
d7e09d03 PT |
2577 | eof = i_size_read(inode); |
2578 | } | |
2579 | ||
6f014339 | 2580 | retval = generic_file_llseek_size(file, offset, origin, |
d7e09d03 | 2581 | ll_file_maxbytes(inode), eof); |
0a3bdb00 | 2582 | return retval; |
d7e09d03 PT |
2583 | } |
2584 | ||
2d95f10e | 2585 | static int ll_flush(struct file *file, fl_owner_t id) |
d7e09d03 | 2586 | { |
2a8a3597 | 2587 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2588 | struct ll_inode_info *lli = ll_i2info(inode); |
2589 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
2590 | int rc, err; | |
2591 | ||
2592 | LASSERT(!S_ISDIR(inode->i_mode)); | |
2593 | ||
2594 | /* catch async errors that were recorded back when async writeback | |
c0894c6c OD |
2595 | * failed for pages in this mapping. |
2596 | */ | |
d7e09d03 PT |
2597 | rc = lli->lli_async_rc; |
2598 | lli->lli_async_rc = 0; | |
79496845 | 2599 | if (lli->lli_clob) { |
2600 | err = lov_read_and_clear_async_rc(lli->lli_clob); | |
2601 | if (!rc) | |
2602 | rc = err; | |
2603 | } | |
d7e09d03 | 2604 | |
c0894c6c OD |
2605 | /* The application has been told about write failure already. |
2606 | * Do not report failure again. | |
2607 | */ | |
d7e09d03 PT |
2608 | if (fd->fd_write_failed) |
2609 | return 0; | |
2610 | return rc ? -EIO : 0; | |
2611 | } | |
2612 | ||
2613 | /** | |
2614 | * Called to make sure a portion of file has been written out. | |
05289927 | 2615 | * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST. |
d7e09d03 PT |
2616 | * |
2617 | * Return how many pages have been written. | |
2618 | */ | |
2619 | int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end, | |
65fb55d1 | 2620 | enum cl_fsync_mode mode, int ignore_layout) |
d7e09d03 PT |
2621 | { |
2622 | struct cl_env_nest nest; | |
2623 | struct lu_env *env; | |
2624 | struct cl_io *io; | |
d7e09d03 PT |
2625 | struct cl_fsync_io *fio; |
2626 | int result; | |
d7e09d03 PT |
2627 | |
2628 | if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL && | |
2629 | mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL) | |
0a3bdb00 | 2630 | return -EINVAL; |
d7e09d03 PT |
2631 | |
2632 | env = cl_env_nested_get(&nest); | |
2633 | if (IS_ERR(env)) | |
0a3bdb00 | 2634 | return PTR_ERR(env); |
d7e09d03 | 2635 | |
9acc4500 | 2636 | io = vvp_env_thread_io(env); |
1929c433 | 2637 | io->ci_obj = ll_i2info(inode)->lli_clob; |
65fb55d1 | 2638 | io->ci_ignore_layout = ignore_layout; |
d7e09d03 PT |
2639 | |
2640 | /* initialize parameters for sync */ | |
2641 | fio = &io->u.ci_fsync; | |
d7e09d03 PT |
2642 | fio->fi_start = start; |
2643 | fio->fi_end = end; | |
2644 | fio->fi_fid = ll_inode2fid(inode); | |
2645 | fio->fi_mode = mode; | |
2646 | fio->fi_nr_written = 0; | |
2647 | ||
2648 | if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0) | |
2649 | result = cl_io_loop(env, io); | |
2650 | else | |
2651 | result = io->ci_result; | |
2652 | if (result == 0) | |
2653 | result = fio->fi_nr_written; | |
2654 | cl_io_fini(env, io); | |
2655 | cl_env_nested_put(&nest, env); | |
2656 | ||
0a3bdb00 | 2657 | return result; |
d7e09d03 PT |
2658 | } |
2659 | ||
d7e09d03 PT |
2660 | int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
2661 | { | |
2a8a3597 | 2662 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2663 | struct ll_inode_info *lli = ll_i2info(inode); |
2664 | struct ptlrpc_request *req; | |
d7e09d03 | 2665 | int rc, err; |
d7e09d03 | 2666 | |
97a075cd JN |
2667 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n", |
2668 | PFID(ll_inode2fid(inode)), inode); | |
d7e09d03 PT |
2669 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1); |
2670 | ||
2671 | rc = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
5955102c | 2672 | inode_lock(inode); |
d7e09d03 PT |
2673 | |
2674 | /* catch async errors that were recorded back when async writeback | |
c0894c6c OD |
2675 | * failed for pages in this mapping. |
2676 | */ | |
d7e09d03 PT |
2677 | if (!S_ISDIR(inode->i_mode)) { |
2678 | err = lli->lli_async_rc; | |
2679 | lli->lli_async_rc = 0; | |
2680 | if (rc == 0) | |
2681 | rc = err; | |
2682 | err = lov_read_and_clear_async_rc(lli->lli_clob); | |
2683 | if (rc == 0) | |
2684 | rc = err; | |
2685 | } | |
2686 | ||
ef2e0f55 | 2687 | err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req); |
d7e09d03 PT |
2688 | if (!rc) |
2689 | rc = err; | |
2690 | if (!err) | |
2691 | ptlrpc_req_finished(req); | |
2692 | ||
8d97deb9 | 2693 | if (S_ISREG(inode->i_mode)) { |
d7e09d03 PT |
2694 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
2695 | ||
05289927 | 2696 | err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0); |
d7e09d03 PT |
2697 | if (rc == 0 && err < 0) |
2698 | rc = err; | |
2699 | if (rc < 0) | |
2700 | fd->fd_write_failed = true; | |
2701 | else | |
2702 | fd->fd_write_failed = false; | |
2703 | } | |
2704 | ||
5955102c | 2705 | inode_unlock(inode); |
0a3bdb00 | 2706 | return rc; |
d7e09d03 PT |
2707 | } |
2708 | ||
2d95f10e JH |
2709 | static int |
2710 | ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) | |
d7e09d03 | 2711 | { |
2a8a3597 | 2712 | struct inode *inode = file_inode(file); |
d7e09d03 | 2713 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
f2145eae BK |
2714 | struct ldlm_enqueue_info einfo = { |
2715 | .ei_type = LDLM_FLOCK, | |
2716 | .ei_cb_cp = ldlm_flock_completion_ast, | |
2717 | .ei_cbdata = file_lock, | |
2718 | }; | |
d7e09d03 PT |
2719 | struct md_op_data *op_data; |
2720 | struct lustre_handle lockh = {0}; | |
8369cfff | 2721 | ldlm_policy_data_t flock = { {0} }; |
c68c3fa4 | 2722 | int fl_type = file_lock->fl_type; |
875332d4 | 2723 | __u64 flags = 0; |
d7e09d03 PT |
2724 | int rc; |
2725 | int rc2 = 0; | |
d7e09d03 | 2726 | |
97a075cd JN |
2727 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n", |
2728 | PFID(ll_inode2fid(inode)), file_lock); | |
d7e09d03 PT |
2729 | |
2730 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1); | |
2731 | ||
130d1f95 | 2732 | if (file_lock->fl_flags & FL_FLOCK) |
d7e09d03 | 2733 | LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK)); |
130d1f95 | 2734 | else if (!(file_lock->fl_flags & FL_POSIX)) |
0a3bdb00 | 2735 | return -EINVAL; |
130d1f95 JL |
2736 | |
2737 | flock.l_flock.owner = (unsigned long)file_lock->fl_owner; | |
d7e09d03 | 2738 | flock.l_flock.pid = file_lock->fl_pid; |
130d1f95 JL |
2739 | flock.l_flock.start = file_lock->fl_start; |
2740 | flock.l_flock.end = file_lock->fl_end; | |
d7e09d03 PT |
2741 | |
2742 | /* Somewhat ugly workaround for svc lockd. | |
2743 | * lockd installs custom fl_lmops->lm_compare_owner that checks | |
2744 | * for the fl_owner to be the same (which it always is on local node | |
2745 | * I guess between lockd processes) and then compares pid. | |
2746 | * As such we assign pid to the owner field to make it all work, | |
2747 | * conflict with normal locks is unlikely since pid space and | |
c0894c6c OD |
2748 | * pointer space for current->files are not intersecting |
2749 | */ | |
d7e09d03 PT |
2750 | if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner) |
2751 | flock.l_flock.owner = (unsigned long)file_lock->fl_pid; | |
2752 | ||
c68c3fa4 | 2753 | switch (fl_type) { |
d7e09d03 PT |
2754 | case F_RDLCK: |
2755 | einfo.ei_mode = LCK_PR; | |
2756 | break; | |
2757 | case F_UNLCK: | |
2758 | /* An unlock request may or may not have any relation to | |
2759 | * existing locks so we may not be able to pass a lock handle | |
2760 | * via a normal ldlm_lock_cancel() request. The request may even | |
2761 | * unlock a byte range in the middle of an existing lock. In | |
2762 | * order to process an unlock request we need all of the same | |
2763 | * information that is given with a normal read or write record | |
2764 | * lock request. To avoid creating another ldlm unlock (cancel) | |
c0894c6c OD |
2765 | * message we'll treat a LCK_NL flock request as an unlock. |
2766 | */ | |
d7e09d03 PT |
2767 | einfo.ei_mode = LCK_NL; |
2768 | break; | |
2769 | case F_WRLCK: | |
2770 | einfo.ei_mode = LCK_PW; | |
2771 | break; | |
2772 | default: | |
c68c3fa4 | 2773 | CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type); |
0a3bdb00 | 2774 | return -ENOTSUPP; |
d7e09d03 PT |
2775 | } |
2776 | ||
2777 | switch (cmd) { | |
2778 | case F_SETLKW: | |
2faedcd5 | 2779 | #ifdef F_SETLKW64 |
d7e09d03 | 2780 | case F_SETLKW64: |
2faedcd5 | 2781 | #endif |
d7e09d03 PT |
2782 | flags = 0; |
2783 | break; | |
2784 | case F_SETLK: | |
2faedcd5 | 2785 | #ifdef F_SETLK64 |
d7e09d03 | 2786 | case F_SETLK64: |
2faedcd5 | 2787 | #endif |
d7e09d03 PT |
2788 | flags = LDLM_FL_BLOCK_NOWAIT; |
2789 | break; | |
2790 | case F_GETLK: | |
2faedcd5 | 2791 | #ifdef F_GETLK64 |
d7e09d03 | 2792 | case F_GETLK64: |
2faedcd5 | 2793 | #endif |
d7e09d03 | 2794 | flags = LDLM_FL_TEST_LOCK; |
d7e09d03 PT |
2795 | break; |
2796 | default: | |
2797 | CERROR("unknown fcntl lock command: %d\n", cmd); | |
0a3bdb00 | 2798 | return -EINVAL; |
d7e09d03 PT |
2799 | } |
2800 | ||
c68c3fa4 VF |
2801 | /* |
2802 | * Save the old mode so that if the mode in the lock changes we | |
2803 | * can decrement the appropriate reader or writer refcount. | |
2804 | */ | |
2805 | file_lock->fl_type = einfo.ei_mode; | |
2806 | ||
d7e09d03 PT |
2807 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, |
2808 | LUSTRE_OPC_ANY, NULL); | |
2809 | if (IS_ERR(op_data)) | |
0a3bdb00 | 2810 | return PTR_ERR(op_data); |
d7e09d03 | 2811 | |
97a075cd JN |
2812 | CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n", |
2813 | PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags, | |
2814 | einfo.ei_mode, flock.l_flock.start, flock.l_flock.end); | |
d7e09d03 PT |
2815 | |
2816 | rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, | |
2817 | op_data, &lockh, &flock, 0, NULL /* req */, flags); | |
2818 | ||
c68c3fa4 VF |
2819 | /* Restore the file lock type if not TEST lock. */ |
2820 | if (!(flags & LDLM_FL_TEST_LOCK)) | |
2821 | file_lock->fl_type = fl_type; | |
2822 | ||
4f656367 | 2823 | if ((rc == 0 || file_lock->fl_type == F_UNLCK) && |
d7e09d03 | 2824 | !(flags & LDLM_FL_TEST_LOCK)) |
4f656367 | 2825 | rc2 = locks_lock_file_wait(file, file_lock); |
d7e09d03 PT |
2826 | |
2827 | if (rc2 && file_lock->fl_type != F_UNLCK) { | |
2828 | einfo.ei_mode = LCK_NL; | |
2829 | md_enqueue(sbi->ll_md_exp, &einfo, NULL, | |
e15ba45d | 2830 | op_data, &lockh, &flock, 0, NULL /* req */, flags); |
d7e09d03 PT |
2831 | rc = rc2; |
2832 | } | |
2833 | ||
2834 | ll_finish_md_op_data(op_data); | |
2835 | ||
0a3bdb00 | 2836 | return rc; |
d7e09d03 PT |
2837 | } |
2838 | ||
1d82425f | 2839 | int ll_get_fid_by_name(struct inode *parent, const char *name, |
2840 | int namelen, struct lu_fid *fid) | |
79496845 | 2841 | { |
2842 | struct md_op_data *op_data = NULL; | |
2843 | struct ptlrpc_request *req; | |
2844 | struct mdt_body *body; | |
2845 | int rc; | |
2846 | ||
2847 | op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0, | |
2848 | LUSTRE_OPC_ANY, NULL); | |
2849 | if (IS_ERR(op_data)) | |
2850 | return PTR_ERR(op_data); | |
2851 | ||
2852 | op_data->op_valid = OBD_MD_FLID; | |
2853 | rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req); | |
1d82425f | 2854 | ll_finish_md_op_data(op_data); |
79496845 | 2855 | if (rc < 0) |
1d82425f | 2856 | return rc; |
79496845 | 2857 | |
2858 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
2859 | if (!body) { | |
2860 | rc = -EFAULT; | |
2861 | goto out_req; | |
2862 | } | |
1d82425f | 2863 | if (fid) |
2e1b5b8b | 2864 | *fid = body->mbo_fid1; |
79496845 | 2865 | out_req: |
2866 | ptlrpc_req_finished(req); | |
79496845 | 2867 | return rc; |
2868 | } | |
2869 | ||
2870 | int ll_migrate(struct inode *parent, struct file *file, int mdtidx, | |
2871 | const char *name, int namelen) | |
2872 | { | |
2873 | struct ptlrpc_request *request = NULL; | |
1d82425f | 2874 | struct inode *child_inode = NULL; |
79496845 | 2875 | struct dentry *dchild = NULL; |
2876 | struct md_op_data *op_data; | |
2877 | struct qstr qstr; | |
2878 | int rc; | |
2879 | ||
1d82425f | 2880 | CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%d\n", |
79496845 | 2881 | name, PFID(ll_inode2fid(parent)), mdtidx); |
2882 | ||
2883 | op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, | |
2884 | 0, LUSTRE_OPC_ANY, NULL); | |
2885 | if (IS_ERR(op_data)) | |
2886 | return PTR_ERR(op_data); | |
2887 | ||
2888 | /* Get child FID first */ | |
2889 | qstr.hash = full_name_hash(parent, name, namelen); | |
2890 | qstr.name = name; | |
2891 | qstr.len = namelen; | |
2892 | dchild = d_lookup(file_dentry(file), &qstr); | |
2893 | if (dchild && dchild->d_inode) { | |
2894 | op_data->op_fid3 = *ll_inode2fid(dchild->d_inode); | |
1d82425f | 2895 | if (dchild->d_inode) { |
2896 | child_inode = igrab(dchild->d_inode); | |
2897 | ll_invalidate_aliases(child_inode); | |
2898 | } | |
2899 | dput(dchild); | |
79496845 | 2900 | } else { |
1d82425f | 2901 | rc = ll_get_fid_by_name(parent, name, namelen, |
79496845 | 2902 | &op_data->op_fid3); |
2903 | if (rc) | |
2904 | goto out_free; | |
2905 | } | |
2906 | ||
2907 | if (!fid_is_sane(&op_data->op_fid3)) { | |
2908 | CERROR("%s: migrate %s, but fid "DFID" is insane\n", | |
2909 | ll_get_fsname(parent->i_sb, NULL, 0), name, | |
2910 | PFID(&op_data->op_fid3)); | |
1d82425f | 2911 | rc = -EINVAL; |
79496845 | 2912 | goto out_free; |
2913 | } | |
2914 | ||
2915 | rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3); | |
2916 | if (rc < 0) | |
2917 | goto out_free; | |
2918 | ||
2919 | if (rc == mdtidx) { | |
2920 | CDEBUG(D_INFO, "%s:"DFID" is already on MDT%d.\n", name, | |
2921 | PFID(&op_data->op_fid3), mdtidx); | |
2922 | rc = 0; | |
2923 | goto out_free; | |
2924 | } | |
2925 | ||
2926 | op_data->op_mds = mdtidx; | |
2927 | op_data->op_cli_flags = CLI_MIGRATE; | |
2928 | rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name, | |
1d82425f | 2929 | namelen, name, namelen, &request); |
79496845 | 2930 | if (!rc) |
2931 | ll_update_times(request, parent); | |
2932 | ||
2933 | ptlrpc_req_finished(request); | |
2934 | ||
2935 | out_free: | |
1d82425f | 2936 | if (child_inode) { |
2937 | clear_nlink(child_inode); | |
2938 | iput(child_inode); | |
79496845 | 2939 | } |
2940 | ||
2941 | ll_finish_md_op_data(op_data); | |
2942 | return rc; | |
2943 | } | |
2944 | ||
2d95f10e JH |
2945 | static int |
2946 | ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock) | |
d7e09d03 | 2947 | { |
0a3bdb00 | 2948 | return -ENOSYS; |
d7e09d03 PT |
2949 | } |
2950 | ||
2951 | /** | |
2952 | * test if some locks matching bits and l_req_mode are acquired | |
2953 | * - bits can be in different locks | |
2954 | * - if found clear the common lock bits in *bits | |
2955 | * - the bits not found, are kept in *bits | |
2956 | * \param inode [IN] | |
2957 | * \param bits [IN] searched lock bits [IN] | |
2958 | * \param l_req_mode [IN] searched lock mode | |
2959 | * \retval boolean, true iff all bits are found | |
2960 | */ | |
52ee0d20 OD |
2961 | int ll_have_md_lock(struct inode *inode, __u64 *bits, |
2962 | enum ldlm_mode l_req_mode) | |
d7e09d03 PT |
2963 | { |
2964 | struct lustre_handle lockh; | |
2965 | ldlm_policy_data_t policy; | |
52ee0d20 | 2966 | enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ? |
d7e09d03 PT |
2967 | (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode; |
2968 | struct lu_fid *fid; | |
2969 | __u64 flags; | |
2970 | int i; | |
d7e09d03 PT |
2971 | |
2972 | if (!inode) | |
ef075edc | 2973 | return 0; |
d7e09d03 PT |
2974 | |
2975 | fid = &ll_i2info(inode)->lli_fid; | |
2976 | CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid), | |
2977 | ldlm_lockname[mode]); | |
2978 | ||
2979 | flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK; | |
1253b2e8 | 2980 | for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) { |
d7e09d03 PT |
2981 | policy.l_inodebits.bits = *bits & (1 << i); |
2982 | if (policy.l_inodebits.bits == 0) | |
2983 | continue; | |
2984 | ||
2985 | if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, | |
2986 | &policy, mode, &lockh)) { | |
2987 | struct ldlm_lock *lock; | |
2988 | ||
2989 | lock = ldlm_handle2lock(&lockh); | |
2990 | if (lock) { | |
2991 | *bits &= | |
2992 | ~(lock->l_policy_data.l_inodebits.bits); | |
2993 | LDLM_LOCK_PUT(lock); | |
2994 | } else { | |
2995 | *bits &= ~policy.l_inodebits.bits; | |
2996 | } | |
2997 | } | |
2998 | } | |
0a3bdb00 | 2999 | return *bits == 0; |
d7e09d03 PT |
3000 | } |
3001 | ||
52ee0d20 OD |
3002 | enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits, |
3003 | struct lustre_handle *lockh, __u64 flags, | |
3004 | enum ldlm_mode mode) | |
d7e09d03 | 3005 | { |
57303e76 | 3006 | ldlm_policy_data_t policy = { .l_inodebits = {bits} }; |
d7e09d03 | 3007 | struct lu_fid *fid; |
52ee0d20 | 3008 | enum ldlm_mode rc; |
d7e09d03 PT |
3009 | |
3010 | fid = &ll_i2info(inode)->lli_fid; | |
3011 | CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid)); | |
3012 | ||
1f6eaf83 | 3013 | rc = md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED, |
7fc1f831 AP |
3014 | fid, LDLM_IBITS, &policy, mode, lockh); |
3015 | ||
0a3bdb00 | 3016 | return rc; |
d7e09d03 PT |
3017 | } |
3018 | ||
3019 | static int ll_inode_revalidate_fini(struct inode *inode, int rc) | |
3020 | { | |
3021 | /* Already unlinked. Just update nlink and return success */ | |
3022 | if (rc == -ENOENT) { | |
3023 | clear_nlink(inode); | |
3024 | /* This path cannot be hit for regular files unless in | |
bef31c78 MI |
3025 | * case of obscure races, so no need to validate size. |
3026 | */ | |
d7e09d03 PT |
3027 | if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode)) |
3028 | return 0; | |
3029 | } else if (rc != 0) { | |
e49634bb AD |
3030 | CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR, |
3031 | "%s: revalidate FID "DFID" error: rc = %d\n", | |
3032 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3033 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
3034 | } |
3035 | ||
3036 | return rc; | |
3037 | } | |
3038 | ||
2d95f10e | 3039 | static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits) |
d7e09d03 | 3040 | { |
2b0143b5 | 3041 | struct inode *inode = d_inode(dentry); |
d7e09d03 PT |
3042 | struct ptlrpc_request *req = NULL; |
3043 | struct obd_export *exp; | |
3044 | int rc = 0; | |
d7e09d03 | 3045 | |
97a075cd JN |
3046 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%pd\n", |
3047 | PFID(ll_inode2fid(inode)), inode, dentry); | |
d7e09d03 PT |
3048 | |
3049 | exp = ll_i2mdexp(inode); | |
3050 | ||
3051 | /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC. | |
3052 | * But under CMD case, it caused some lock issues, should be fixed | |
c0894c6c OD |
3053 | * with new CMD ibits lock. See bug 12718 |
3054 | */ | |
d7e09d03 PT |
3055 | if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) { |
3056 | struct lookup_intent oit = { .it_op = IT_GETATTR }; | |
3057 | struct md_op_data *op_data; | |
3058 | ||
3059 | if (ibits == MDS_INODELOCK_LOOKUP) | |
3060 | oit.it_op = IT_LOOKUP; | |
3061 | ||
3062 | /* Call getattr by fid, so do not provide name at all. */ | |
dbca51dd AV |
3063 | op_data = ll_prep_md_op_data(NULL, inode, |
3064 | inode, NULL, 0, 0, | |
d7e09d03 PT |
3065 | LUSTRE_OPC_ANY, NULL); |
3066 | if (IS_ERR(op_data)) | |
0a3bdb00 | 3067 | return PTR_ERR(op_data); |
d7e09d03 PT |
3068 | |
3069 | oit.it_create_mode |= M_CHECK_STALE; | |
3070 | rc = md_intent_lock(exp, op_data, NULL, 0, | |
3071 | /* we are not interested in name | |
c0894c6c OD |
3072 | * based lookup |
3073 | */ | |
d7e09d03 PT |
3074 | &oit, 0, &req, |
3075 | ll_md_blocking_ast, 0); | |
3076 | ll_finish_md_op_data(op_data); | |
3077 | oit.it_create_mode &= ~M_CHECK_STALE; | |
3078 | if (rc < 0) { | |
3079 | rc = ll_inode_revalidate_fini(inode, rc); | |
34e1f2bb | 3080 | goto out; |
d7e09d03 PT |
3081 | } |
3082 | ||
dbca51dd | 3083 | rc = ll_revalidate_it_finish(req, &oit, inode); |
d7e09d03 PT |
3084 | if (rc != 0) { |
3085 | ll_intent_release(&oit); | |
34e1f2bb | 3086 | goto out; |
d7e09d03 PT |
3087 | } |
3088 | ||
3089 | /* Unlinked? Unhash dentry, so it is not picked up later by | |
c0894c6c OD |
3090 | * do_lookup() -> ll_revalidate_it(). We cannot use d_drop |
3091 | * here to preserve get_cwd functionality on 2.6. | |
3092 | * Bug 10503 | |
3093 | */ | |
c9cc8d0f BF |
3094 | if (!d_inode(dentry)->i_nlink) { |
3095 | spin_lock(&inode->i_lock); | |
b1d2a127 | 3096 | d_lustre_invalidate(dentry, 0); |
c9cc8d0f BF |
3097 | spin_unlock(&inode->i_lock); |
3098 | } | |
d7e09d03 | 3099 | |
dbca51dd | 3100 | ll_lookup_finish_locks(&oit, inode); |
2b0143b5 DH |
3101 | } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) { |
3102 | struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry)); | |
21aef7d9 | 3103 | u64 valid = OBD_MD_FLGETATTR; |
d7e09d03 PT |
3104 | struct md_op_data *op_data; |
3105 | int ealen = 0; | |
3106 | ||
3107 | if (S_ISREG(inode->i_mode)) { | |
44779340 | 3108 | rc = ll_get_default_mdsize(sbi, &ealen); |
d7e09d03 | 3109 | if (rc) |
0a3bdb00 | 3110 | return rc; |
d7e09d03 PT |
3111 | valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE; |
3112 | } | |
3113 | ||
3114 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, | |
3115 | 0, ealen, LUSTRE_OPC_ANY, | |
3116 | NULL); | |
3117 | if (IS_ERR(op_data)) | |
0a3bdb00 | 3118 | return PTR_ERR(op_data); |
d7e09d03 PT |
3119 | |
3120 | op_data->op_valid = valid; | |
d7e09d03 PT |
3121 | rc = md_getattr(sbi->ll_md_exp, op_data, &req); |
3122 | ll_finish_md_op_data(op_data); | |
3123 | if (rc) { | |
3124 | rc = ll_inode_revalidate_fini(inode, rc); | |
0a3bdb00 | 3125 | return rc; |
d7e09d03 PT |
3126 | } |
3127 | ||
3128 | rc = ll_prep_inode(&inode, req, NULL, NULL); | |
3129 | } | |
3130 | out: | |
3131 | ptlrpc_req_finished(req); | |
3132 | return rc; | |
3133 | } | |
3134 | ||
2de35386 | 3135 | static int ll_merge_md_attr(struct inode *inode) |
3136 | { | |
3137 | struct cl_attr attr = { 0 }; | |
3138 | int rc; | |
3139 | ||
3140 | LASSERT(ll_i2info(inode)->lli_lsm_md); | |
3141 | rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md, | |
3142 | &attr); | |
3143 | if (rc) | |
3144 | return rc; | |
3145 | ||
3146 | ll_i2info(inode)->lli_stripe_dir_size = attr.cat_size; | |
3147 | ll_i2info(inode)->lli_stripe_dir_nlink = attr.cat_nlink; | |
3148 | ||
3149 | ll_i2info(inode)->lli_atime = attr.cat_atime; | |
3150 | ll_i2info(inode)->lli_mtime = attr.cat_mtime; | |
3151 | ll_i2info(inode)->lli_ctime = attr.cat_ctime; | |
3152 | ||
3153 | return 0; | |
3154 | } | |
3155 | ||
2d95f10e | 3156 | static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits) |
d7e09d03 | 3157 | { |
2b0143b5 | 3158 | struct inode *inode = d_inode(dentry); |
d7e09d03 | 3159 | int rc; |
d7e09d03 | 3160 | |
2d95f10e | 3161 | rc = __ll_inode_revalidate(dentry, ibits); |
d7e09d03 | 3162 | if (rc != 0) |
0a3bdb00 | 3163 | return rc; |
d7e09d03 PT |
3164 | |
3165 | /* if object isn't regular file, don't validate size */ | |
3166 | if (!S_ISREG(inode->i_mode)) { | |
2de35386 | 3167 | if (S_ISDIR(inode->i_mode) && |
3168 | ll_i2info(inode)->lli_lsm_md) { | |
3169 | rc = ll_merge_md_attr(inode); | |
3170 | if (rc) | |
3171 | return rc; | |
3172 | } | |
3173 | ||
d2995737 JH |
3174 | LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime; |
3175 | LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime; | |
3176 | LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime; | |
d7e09d03 | 3177 | } else { |
5ea17d6c JL |
3178 | /* In case of restore, the MDT has the right size and has |
3179 | * already send it back without granting the layout lock, | |
3180 | * inode is up-to-date so glimpse is useless. | |
3181 | * Also to glimpse we need the layout, in case of a running | |
3182 | * restore the MDT holds the layout lock so the glimpse will | |
3183 | * block up to the end of restore (getattr will block) | |
3184 | */ | |
3185 | if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING)) | |
3186 | rc = ll_glimpse_size(inode); | |
d7e09d03 | 3187 | } |
0a3bdb00 | 3188 | return rc; |
d7e09d03 PT |
3189 | } |
3190 | ||
2d95f10e | 3191 | int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat) |
d7e09d03 | 3192 | { |
2b0143b5 | 3193 | struct inode *inode = d_inode(de); |
d7e09d03 PT |
3194 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
3195 | struct ll_inode_info *lli = ll_i2info(inode); | |
f82ced5d | 3196 | int res; |
d7e09d03 | 3197 | |
2d95f10e JH |
3198 | res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE | |
3199 | MDS_INODELOCK_LOOKUP); | |
d7e09d03 PT |
3200 | ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1); |
3201 | ||
3202 | if (res) | |
3203 | return res; | |
3204 | ||
3205 | stat->dev = inode->i_sb->s_dev; | |
3206 | if (ll_need_32bit_api(sbi)) | |
3207 | stat->ino = cl_fid_build_ino(&lli->lli_fid, 1); | |
3208 | else | |
3209 | stat->ino = inode->i_ino; | |
3210 | stat->mode = inode->i_mode; | |
d7e09d03 PT |
3211 | stat->uid = inode->i_uid; |
3212 | stat->gid = inode->i_gid; | |
3213 | stat->rdev = inode->i_rdev; | |
3214 | stat->atime = inode->i_atime; | |
3215 | stat->mtime = inode->i_mtime; | |
3216 | stat->ctime = inode->i_ctime; | |
3217 | stat->blksize = 1 << inode->i_blkbits; | |
d7e09d03 PT |
3218 | stat->blocks = inode->i_blocks; |
3219 | ||
2de35386 | 3220 | if (S_ISDIR(inode->i_mode) && |
3221 | ll_i2info(inode)->lli_lsm_md) { | |
3222 | stat->nlink = lli->lli_stripe_dir_nlink; | |
3223 | stat->size = lli->lli_stripe_dir_size; | |
3224 | } else { | |
3225 | stat->nlink = inode->i_nlink; | |
3226 | stat->size = i_size_read(inode); | |
3227 | } | |
3228 | ||
d7e09d03 PT |
3229 | return 0; |
3230 | } | |
d7e09d03 | 3231 | |
2d95f10e JH |
3232 | static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3233 | __u64 start, __u64 len) | |
89580e37 PT |
3234 | { |
3235 | int rc; | |
3236 | size_t num_bytes; | |
3237 | struct ll_user_fiemap *fiemap; | |
3238 | unsigned int extent_count = fieinfo->fi_extents_max; | |
3239 | ||
3240 | num_bytes = sizeof(*fiemap) + (extent_count * | |
3241 | sizeof(struct ll_fiemap_extent)); | |
e958f49b | 3242 | fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS); |
89580e37 | 3243 | |
6e16818b | 3244 | if (!fiemap) |
89580e37 PT |
3245 | return -ENOMEM; |
3246 | ||
3247 | fiemap->fm_flags = fieinfo->fi_flags; | |
3248 | fiemap->fm_extent_count = fieinfo->fi_extents_max; | |
3249 | fiemap->fm_start = start; | |
3250 | fiemap->fm_length = len; | |
97514241 OD |
3251 | if (extent_count > 0 && |
3252 | copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start, | |
3253 | sizeof(struct ll_fiemap_extent)) != 0) { | |
3254 | rc = -EFAULT; | |
3255 | goto out; | |
3256 | } | |
89580e37 PT |
3257 | |
3258 | rc = ll_do_fiemap(inode, fiemap, num_bytes); | |
3259 | ||
3260 | fieinfo->fi_flags = fiemap->fm_flags; | |
3261 | fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents; | |
97514241 OD |
3262 | if (extent_count > 0 && |
3263 | copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0], | |
3264 | fiemap->fm_mapped_extents * | |
3265 | sizeof(struct ll_fiemap_extent)) != 0) { | |
3266 | rc = -EFAULT; | |
3267 | goto out; | |
3268 | } | |
89580e37 | 3269 | |
97514241 | 3270 | out: |
e958f49b | 3271 | kvfree(fiemap); |
89580e37 PT |
3272 | return rc; |
3273 | } | |
d7e09d03 | 3274 | |
2d95f10e | 3275 | struct posix_acl *ll_get_acl(struct inode *inode, int type) |
d7e09d03 PT |
3276 | { |
3277 | struct ll_inode_info *lli = ll_i2info(inode); | |
3278 | struct posix_acl *acl = NULL; | |
d7e09d03 PT |
3279 | |
3280 | spin_lock(&lli->lli_lock); | |
3281 | /* VFS' acl_permission_check->check_acl will release the refcount */ | |
3282 | acl = posix_acl_dup(lli->lli_posix_acl); | |
ed7bdf5c | 3283 | #ifdef CONFIG_FS_POSIX_ACL |
b788dc51 | 3284 | forget_cached_acl(inode, type); |
ed7bdf5c | 3285 | #endif |
d7e09d03 PT |
3286 | spin_unlock(&lli->lli_lock); |
3287 | ||
0a3bdb00 | 3288 | return acl; |
d7e09d03 PT |
3289 | } |
3290 | ||
d7e09d03 PT |
3291 | int ll_inode_permission(struct inode *inode, int mask) |
3292 | { | |
c948390f GP |
3293 | struct ll_sb_info *sbi; |
3294 | struct root_squash_info *squash; | |
3295 | const struct cred *old_cred = NULL; | |
3296 | struct cred *cred = NULL; | |
3297 | bool squash_id = false; | |
3298 | cfs_cap_t cap; | |
d7e09d03 | 3299 | int rc = 0; |
d7e09d03 | 3300 | |
d7e09d03 PT |
3301 | if (mask & MAY_NOT_BLOCK) |
3302 | return -ECHILD; | |
d7e09d03 PT |
3303 | |
3304 | /* as root inode are NOT getting validated in lookup operation, | |
c0894c6c OD |
3305 | * need to do it before permission check. |
3306 | */ | |
d7e09d03 | 3307 | |
f76c23da | 3308 | if (is_root_inode(inode)) { |
2d95f10e JH |
3309 | rc = __ll_inode_revalidate(inode->i_sb->s_root, |
3310 | MDS_INODELOCK_LOOKUP); | |
d7e09d03 | 3311 | if (rc) |
0a3bdb00 | 3312 | return rc; |
d7e09d03 PT |
3313 | } |
3314 | ||
97a075cd JN |
3315 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n", |
3316 | PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask); | |
d7e09d03 | 3317 | |
c948390f GP |
3318 | /* squash fsuid/fsgid if needed */ |
3319 | sbi = ll_i2sbi(inode); | |
3320 | squash = &sbi->ll_squash; | |
3321 | if (unlikely(squash->rsi_uid && | |
3322 | uid_eq(current_fsuid(), GLOBAL_ROOT_UID) && | |
3323 | !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) { | |
3324 | squash_id = true; | |
3325 | } | |
3326 | ||
3327 | if (squash_id) { | |
3328 | CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n", | |
3329 | __kuid_val(current_fsuid()), __kgid_val(current_fsgid()), | |
3330 | squash->rsi_uid, squash->rsi_gid); | |
3331 | ||
3332 | /* | |
3333 | * update current process's credentials | |
3334 | * and FS capability | |
3335 | */ | |
3336 | cred = prepare_creds(); | |
3337 | if (!cred) | |
3338 | return -ENOMEM; | |
3339 | ||
3340 | cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid); | |
3341 | cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid); | |
3342 | for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) { | |
3343 | if ((1 << cap) & CFS_CAP_FS_MASK) | |
3344 | cap_lower(cred->cap_effective, cap); | |
3345 | } | |
3346 | old_cred = override_creds(cred); | |
3347 | } | |
3348 | ||
d7e09d03 | 3349 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); |
8707c96e | 3350 | rc = generic_permission(inode, mask); |
d7e09d03 | 3351 | |
c948390f GP |
3352 | /* restore current process's credentials and FS capability */ |
3353 | if (squash_id) { | |
3354 | revert_creds(old_cred); | |
3355 | put_cred(cred); | |
3356 | } | |
3357 | ||
0a3bdb00 | 3358 | return rc; |
d7e09d03 PT |
3359 | } |
3360 | ||
d7e09d03 PT |
3361 | /* -o localflock - only provides locally consistent flock locks */ |
3362 | struct file_operations ll_file_operations = { | |
b42b15fd | 3363 | .read_iter = ll_file_read_iter, |
b42b15fd | 3364 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3365 | .unlocked_ioctl = ll_file_ioctl, |
3366 | .open = ll_file_open, | |
3367 | .release = ll_file_release, | |
3368 | .mmap = ll_file_mmap, | |
3369 | .llseek = ll_file_seek, | |
3370 | .splice_read = ll_file_splice_read, | |
3371 | .fsync = ll_fsync, | |
3372 | .flush = ll_flush | |
3373 | }; | |
3374 | ||
3375 | struct file_operations ll_file_operations_flock = { | |
b42b15fd | 3376 | .read_iter = ll_file_read_iter, |
b42b15fd | 3377 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3378 | .unlocked_ioctl = ll_file_ioctl, |
3379 | .open = ll_file_open, | |
3380 | .release = ll_file_release, | |
3381 | .mmap = ll_file_mmap, | |
3382 | .llseek = ll_file_seek, | |
3383 | .splice_read = ll_file_splice_read, | |
3384 | .fsync = ll_fsync, | |
3385 | .flush = ll_flush, | |
3386 | .flock = ll_file_flock, | |
3387 | .lock = ll_file_flock | |
3388 | }; | |
3389 | ||
3390 | /* These are for -o noflock - to return ENOSYS on flock calls */ | |
3391 | struct file_operations ll_file_operations_noflock = { | |
b42b15fd | 3392 | .read_iter = ll_file_read_iter, |
b42b15fd | 3393 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3394 | .unlocked_ioctl = ll_file_ioctl, |
3395 | .open = ll_file_open, | |
3396 | .release = ll_file_release, | |
3397 | .mmap = ll_file_mmap, | |
3398 | .llseek = ll_file_seek, | |
3399 | .splice_read = ll_file_splice_read, | |
3400 | .fsync = ll_fsync, | |
3401 | .flush = ll_flush, | |
3402 | .flock = ll_file_noflock, | |
3403 | .lock = ll_file_noflock | |
3404 | }; | |
3405 | ||
d2d32738 | 3406 | const struct inode_operations ll_file_inode_operations = { |
d7e09d03 PT |
3407 | .setattr = ll_setattr, |
3408 | .getattr = ll_getattr, | |
3409 | .permission = ll_inode_permission, | |
2c563880 JS |
3410 | .setxattr = generic_setxattr, |
3411 | .getxattr = generic_getxattr, | |
d7e09d03 | 3412 | .listxattr = ll_listxattr, |
2c563880 | 3413 | .removexattr = generic_removexattr, |
89580e37 | 3414 | .fiemap = ll_fiemap, |
d7e09d03 PT |
3415 | .get_acl = ll_get_acl, |
3416 | }; | |
3417 | ||
d0a0acc3 | 3418 | /* dynamic ioctl number support routines */ |
d7e09d03 PT |
3419 | static struct llioc_ctl_data { |
3420 | struct rw_semaphore ioc_sem; | |
3421 | struct list_head ioc_head; | |
3422 | } llioc = { | |
3423 | __RWSEM_INITIALIZER(llioc.ioc_sem), | |
3424 | LIST_HEAD_INIT(llioc.ioc_head) | |
3425 | }; | |
3426 | ||
d7e09d03 PT |
3427 | struct llioc_data { |
3428 | struct list_head iocd_list; | |
3429 | unsigned int iocd_size; | |
3430 | llioc_callback_t iocd_cb; | |
3431 | unsigned int iocd_count; | |
3432 | unsigned int iocd_cmd[0]; | |
3433 | }; | |
3434 | ||
3435 | void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd) | |
3436 | { | |
3437 | unsigned int size; | |
3438 | struct llioc_data *in_data = NULL; | |
d7e09d03 | 3439 | |
6e16818b | 3440 | if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0) |
0a3bdb00 | 3441 | return NULL; |
d7e09d03 PT |
3442 | |
3443 | size = sizeof(*in_data) + count * sizeof(unsigned int); | |
496a51bd JL |
3444 | in_data = kzalloc(size, GFP_NOFS); |
3445 | if (!in_data) | |
0a3bdb00 | 3446 | return NULL; |
d7e09d03 PT |
3447 | |
3448 | memset(in_data, 0, sizeof(*in_data)); | |
3449 | in_data->iocd_size = size; | |
3450 | in_data->iocd_cb = cb; | |
3451 | in_data->iocd_count = count; | |
3452 | memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count); | |
3453 | ||
3454 | down_write(&llioc.ioc_sem); | |
3455 | list_add_tail(&in_data->iocd_list, &llioc.ioc_head); | |
3456 | up_write(&llioc.ioc_sem); | |
3457 | ||
0a3bdb00 | 3458 | return in_data; |
d7e09d03 | 3459 | } |
93133eb4 | 3460 | EXPORT_SYMBOL(ll_iocontrol_register); |
d7e09d03 PT |
3461 | |
3462 | void ll_iocontrol_unregister(void *magic) | |
3463 | { | |
3464 | struct llioc_data *tmp; | |
3465 | ||
6e16818b | 3466 | if (!magic) |
d7e09d03 PT |
3467 | return; |
3468 | ||
3469 | down_write(&llioc.ioc_sem); | |
3470 | list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) { | |
3471 | if (tmp == magic) { | |
d7e09d03 PT |
3472 | list_del(&tmp->iocd_list); |
3473 | up_write(&llioc.ioc_sem); | |
3474 | ||
97903a26 | 3475 | kfree(tmp); |
d7e09d03 PT |
3476 | return; |
3477 | } | |
3478 | } | |
3479 | up_write(&llioc.ioc_sem); | |
3480 | ||
3481 | CWARN("didn't find iocontrol register block with magic: %p\n", magic); | |
3482 | } | |
d7e09d03 PT |
3483 | EXPORT_SYMBOL(ll_iocontrol_unregister); |
3484 | ||
2d95f10e JH |
3485 | static enum llioc_iter |
3486 | ll_iocontrol_call(struct inode *inode, struct file *file, | |
3487 | unsigned int cmd, unsigned long arg, int *rcp) | |
d7e09d03 PT |
3488 | { |
3489 | enum llioc_iter ret = LLIOC_CONT; | |
3490 | struct llioc_data *data; | |
3491 | int rc = -EINVAL, i; | |
3492 | ||
3493 | down_read(&llioc.ioc_sem); | |
3494 | list_for_each_entry(data, &llioc.ioc_head, iocd_list) { | |
3495 | for (i = 0; i < data->iocd_count; i++) { | |
3496 | if (cmd != data->iocd_cmd[i]) | |
3497 | continue; | |
3498 | ||
3499 | ret = data->iocd_cb(inode, file, cmd, arg, data, &rc); | |
3500 | break; | |
3501 | } | |
3502 | ||
3503 | if (ret == LLIOC_STOP) | |
3504 | break; | |
3505 | } | |
3506 | up_read(&llioc.ioc_sem); | |
3507 | ||
3508 | if (rcp) | |
3509 | *rcp = rc; | |
3510 | return ret; | |
3511 | } | |
3512 | ||
3513 | int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf) | |
3514 | { | |
3515 | struct ll_inode_info *lli = ll_i2info(inode); | |
3516 | struct cl_env_nest nest; | |
3517 | struct lu_env *env; | |
3518 | int result; | |
d7e09d03 | 3519 | |
6e16818b | 3520 | if (!lli->lli_clob) |
0a3bdb00 | 3521 | return 0; |
d7e09d03 PT |
3522 | |
3523 | env = cl_env_nested_get(&nest); | |
3524 | if (IS_ERR(env)) | |
0a3bdb00 | 3525 | return PTR_ERR(env); |
d7e09d03 PT |
3526 | |
3527 | result = cl_conf_set(env, lli->lli_clob, conf); | |
3528 | cl_env_nested_put(&nest, env); | |
3529 | ||
3530 | if (conf->coc_opc == OBJECT_CONF_SET) { | |
3531 | struct ldlm_lock *lock = conf->coc_lock; | |
3532 | ||
6e16818b | 3533 | LASSERT(lock); |
d7e09d03 PT |
3534 | LASSERT(ldlm_has_layout(lock)); |
3535 | if (result == 0) { | |
3536 | /* it can only be allowed to match after layout is | |
3537 | * applied to inode otherwise false layout would be | |
d0a0acc3 | 3538 | * seen. Applying layout should happen before dropping |
c0894c6c OD |
3539 | * the intent lock. |
3540 | */ | |
d7e09d03 PT |
3541 | ldlm_lock_allow_match(lock); |
3542 | } | |
3543 | } | |
0a3bdb00 | 3544 | return result; |
d7e09d03 PT |
3545 | } |
3546 | ||
3547 | /* Fetch layout from MDT with getxattr request, if it's not ready yet */ | |
3548 | static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock) | |
3549 | ||
3550 | { | |
3551 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
d7e09d03 PT |
3552 | struct ptlrpc_request *req; |
3553 | struct mdt_body *body; | |
3554 | void *lvbdata; | |
3555 | void *lmm; | |
3556 | int lmmsize; | |
3557 | int rc; | |
d7e09d03 | 3558 | |
e2335e5d | 3559 | CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n", |
5a9a80ba | 3560 | PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock), |
e2335e5d | 3561 | lock->l_lvb_data, lock->l_lvb_len); |
3562 | ||
5a9a80ba | 3563 | if (lock->l_lvb_data && ldlm_is_lvb_ready(lock)) |
0a3bdb00 | 3564 | return 0; |
d7e09d03 PT |
3565 | |
3566 | /* if layout lock was granted right away, the layout is returned | |
3567 | * within DLM_LVB of dlm reply; otherwise if the lock was ever | |
3568 | * blocked and then granted via completion ast, we have to fetch | |
3569 | * layout here. Please note that we can't use the LVB buffer in | |
c0894c6c OD |
3570 | * completion AST because it doesn't have a large enough buffer |
3571 | */ | |
44779340 | 3572 | rc = ll_get_default_mdsize(sbi, &lmmsize); |
d7e09d03 | 3573 | if (rc == 0) |
ef2e0f55 OD |
3574 | rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), |
3575 | OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0, | |
3576 | lmmsize, 0, &req); | |
d7e09d03 | 3577 | if (rc < 0) |
0a3bdb00 | 3578 | return rc; |
d7e09d03 PT |
3579 | |
3580 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
6e16818b | 3581 | if (!body) { |
34e1f2bb JL |
3582 | rc = -EPROTO; |
3583 | goto out; | |
3584 | } | |
d7e09d03 | 3585 | |
2e1b5b8b | 3586 | lmmsize = body->mbo_eadatasize; |
34e1f2bb JL |
3587 | if (lmmsize == 0) /* empty layout */ { |
3588 | rc = 0; | |
3589 | goto out; | |
3590 | } | |
d7e09d03 PT |
3591 | |
3592 | lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize); | |
6e16818b | 3593 | if (!lmm) { |
34e1f2bb JL |
3594 | rc = -EFAULT; |
3595 | goto out; | |
3596 | } | |
d7e09d03 | 3597 | |
e958f49b | 3598 | lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS); |
6e16818b | 3599 | if (!lvbdata) { |
34e1f2bb JL |
3600 | rc = -ENOMEM; |
3601 | goto out; | |
3602 | } | |
d7e09d03 PT |
3603 | |
3604 | memcpy(lvbdata, lmm, lmmsize); | |
3605 | lock_res_and_lock(lock); | |
6e16818b | 3606 | if (lock->l_lvb_data) |
e958f49b | 3607 | kvfree(lock->l_lvb_data); |
e2335e5d | 3608 | |
3609 | lock->l_lvb_data = lvbdata; | |
3610 | lock->l_lvb_len = lmmsize; | |
d7e09d03 PT |
3611 | unlock_res_and_lock(lock); |
3612 | ||
d7e09d03 PT |
3613 | out: |
3614 | ptlrpc_req_finished(req); | |
3615 | return rc; | |
3616 | } | |
3617 | ||
3618 | /** | |
3619 | * Apply the layout to the inode. Layout lock is held and will be released | |
3620 | * in this function. | |
3621 | */ | |
52ee0d20 OD |
3622 | static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode, |
3623 | struct inode *inode, __u32 *gen, bool reconf) | |
d7e09d03 PT |
3624 | { |
3625 | struct ll_inode_info *lli = ll_i2info(inode); | |
3626 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
3627 | struct ldlm_lock *lock; | |
3628 | struct lustre_md md = { NULL }; | |
3629 | struct cl_object_conf conf; | |
3630 | int rc = 0; | |
3631 | bool lvb_ready; | |
3632 | bool wait_layout = false; | |
d7e09d03 PT |
3633 | |
3634 | LASSERT(lustre_handle_is_used(lockh)); | |
3635 | ||
3636 | lock = ldlm_handle2lock(lockh); | |
6e16818b | 3637 | LASSERT(lock); |
d7e09d03 PT |
3638 | LASSERT(ldlm_has_layout(lock)); |
3639 | ||
97a075cd JN |
3640 | LDLM_DEBUG(lock, "File "DFID"(%p) being reconfigured: %d", |
3641 | PFID(&lli->lli_fid), inode, reconf); | |
d7e09d03 | 3642 | |
bc969176 JL |
3643 | /* in case this is a caching lock and reinstate with new inode */ |
3644 | md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL); | |
3645 | ||
d7e09d03 | 3646 | lock_res_and_lock(lock); |
5a9a80ba | 3647 | lvb_ready = ldlm_is_lvb_ready(lock); |
d7e09d03 PT |
3648 | unlock_res_and_lock(lock); |
3649 | /* checking lvb_ready is racy but this is okay. The worst case is | |
c0894c6c OD |
3650 | * that multi processes may configure the file on the same time. |
3651 | */ | |
d7e09d03 PT |
3652 | if (lvb_ready || !reconf) { |
3653 | rc = -ENODATA; | |
3654 | if (lvb_ready) { | |
3655 | /* layout_gen must be valid if layout lock is not | |
c0894c6c OD |
3656 | * cancelled and stripe has already set |
3657 | */ | |
09aed8a5 | 3658 | *gen = ll_layout_version_get(lli); |
d7e09d03 PT |
3659 | rc = 0; |
3660 | } | |
34e1f2bb | 3661 | goto out; |
d7e09d03 PT |
3662 | } |
3663 | ||
3664 | rc = ll_layout_fetch(inode, lock); | |
3665 | if (rc < 0) | |
34e1f2bb | 3666 | goto out; |
d7e09d03 PT |
3667 | |
3668 | /* for layout lock, lmm is returned in lock's lvb. | |
3669 | * lvb_data is immutable if the lock is held so it's safe to access it | |
3670 | * without res lock. See the description in ldlm_lock_decref_internal() | |
c0894c6c OD |
3671 | * for the condition to free lvb_data of layout lock |
3672 | */ | |
6e16818b | 3673 | if (lock->l_lvb_data) { |
d7e09d03 PT |
3674 | rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm, |
3675 | lock->l_lvb_data, lock->l_lvb_len); | |
3676 | if (rc >= 0) { | |
3677 | *gen = LL_LAYOUT_GEN_EMPTY; | |
6e16818b | 3678 | if (md.lsm) |
d7e09d03 PT |
3679 | *gen = md.lsm->lsm_layout_gen; |
3680 | rc = 0; | |
3681 | } else { | |
e15ba45d OD |
3682 | CERROR("%s: file " DFID " unpackmd error: %d\n", |
3683 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3684 | PFID(&lli->lli_fid), rc); | |
d7e09d03 PT |
3685 | } |
3686 | } | |
3687 | if (rc < 0) | |
34e1f2bb | 3688 | goto out; |
d7e09d03 PT |
3689 | |
3690 | /* set layout to file. Unlikely this will fail as old layout was | |
c0894c6c OD |
3691 | * surely eliminated |
3692 | */ | |
ec83e611 | 3693 | memset(&conf, 0, sizeof(conf)); |
d7e09d03 PT |
3694 | conf.coc_opc = OBJECT_CONF_SET; |
3695 | conf.coc_inode = inode; | |
3696 | conf.coc_lock = lock; | |
3697 | conf.u.coc_md = &md; | |
3698 | rc = ll_layout_conf(inode, &conf); | |
3699 | ||
6e16818b | 3700 | if (md.lsm) |
d7e09d03 PT |
3701 | obd_free_memmd(sbi->ll_dt_exp, &md.lsm); |
3702 | ||
3703 | /* refresh layout failed, need to wait */ | |
3704 | wait_layout = rc == -EBUSY; | |
d7e09d03 PT |
3705 | |
3706 | out: | |
3707 | LDLM_LOCK_PUT(lock); | |
3708 | ldlm_lock_decref(lockh, mode); | |
3709 | ||
3710 | /* wait for IO to complete if it's still being used. */ | |
3711 | if (wait_layout) { | |
97a075cd | 3712 | CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n", |
e15ba45d | 3713 | ll_get_fsname(inode->i_sb, NULL, 0), |
97a075cd | 3714 | PFID(&lli->lli_fid), inode); |
d7e09d03 | 3715 | |
ec83e611 | 3716 | memset(&conf, 0, sizeof(conf)); |
d7e09d03 PT |
3717 | conf.coc_opc = OBJECT_CONF_WAIT; |
3718 | conf.coc_inode = inode; | |
3719 | rc = ll_layout_conf(inode, &conf); | |
3720 | if (rc == 0) | |
3721 | rc = -EAGAIN; | |
3722 | ||
97a075cd JN |
3723 | CDEBUG(D_INODE, "%s: file="DFID" waiting layout return: %d.\n", |
3724 | ll_get_fsname(inode->i_sb, NULL, 0), | |
e15ba45d | 3725 | PFID(&lli->lli_fid), rc); |
d7e09d03 | 3726 | } |
0a3bdb00 | 3727 | return rc; |
d7e09d03 PT |
3728 | } |
3729 | ||
3730 | /** | |
3731 | * This function checks if there exists a LAYOUT lock on the client side, | |
3732 | * or enqueues it if it doesn't have one in cache. | |
3733 | * | |
3734 | * This function will not hold layout lock so it may be revoked any time after | |
3735 | * this function returns. Any operations depend on layout should be redone | |
3736 | * in that case. | |
3737 | * | |
3738 | * This function should be called before lov_io_init() to get an uptodate | |
3739 | * layout version, the caller should save the version number and after IO | |
3740 | * is finished, this function should be called again to verify that layout | |
3741 | * is not changed during IO time. | |
3742 | */ | |
3743 | int ll_layout_refresh(struct inode *inode, __u32 *gen) | |
3744 | { | |
3745 | struct ll_inode_info *lli = ll_i2info(inode); | |
3746 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
3747 | struct md_op_data *op_data; | |
3748 | struct lookup_intent it; | |
3749 | struct lustre_handle lockh; | |
52ee0d20 | 3750 | enum ldlm_mode mode; |
f2145eae BK |
3751 | struct ldlm_enqueue_info einfo = { |
3752 | .ei_type = LDLM_IBITS, | |
3753 | .ei_mode = LCK_CR, | |
3754 | .ei_cb_bl = ll_md_blocking_ast, | |
3755 | .ei_cb_cp = ldlm_completion_ast, | |
3756 | }; | |
d7e09d03 | 3757 | int rc; |
d7e09d03 | 3758 | |
09aed8a5 JX |
3759 | *gen = ll_layout_version_get(lli); |
3760 | if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE) | |
0a3bdb00 | 3761 | return 0; |
d7e09d03 PT |
3762 | |
3763 | /* sanity checks */ | |
3764 | LASSERT(fid_is_sane(ll_inode2fid(inode))); | |
3765 | LASSERT(S_ISREG(inode->i_mode)); | |
3766 | ||
d7e09d03 PT |
3767 | /* take layout lock mutex to enqueue layout lock exclusively. */ |
3768 | mutex_lock(&lli->lli_layout_mutex); | |
3769 | ||
3770 | again: | |
09aed8a5 | 3771 | /* mostly layout lock is caching on the local side, so try to match |
c0894c6c OD |
3772 | * it before grabbing layout lock mutex. |
3773 | */ | |
7fc1f831 AP |
3774 | mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0, |
3775 | LCK_CR | LCK_CW | LCK_PR | LCK_PW); | |
d7e09d03 PT |
3776 | if (mode != 0) { /* hit cached lock */ |
3777 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); | |
3778 | if (rc == -EAGAIN) | |
3779 | goto again; | |
3780 | ||
3781 | mutex_unlock(&lli->lli_layout_mutex); | |
0a3bdb00 | 3782 | return rc; |
d7e09d03 PT |
3783 | } |
3784 | ||
3785 | op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, | |
e15ba45d | 3786 | 0, 0, LUSTRE_OPC_ANY, NULL); |
d7e09d03 PT |
3787 | if (IS_ERR(op_data)) { |
3788 | mutex_unlock(&lli->lli_layout_mutex); | |
0a3bdb00 | 3789 | return PTR_ERR(op_data); |
d7e09d03 PT |
3790 | } |
3791 | ||
3792 | /* have to enqueue one */ | |
3793 | memset(&it, 0, sizeof(it)); | |
3794 | it.it_op = IT_LAYOUT; | |
3795 | lockh.cookie = 0ULL; | |
3796 | ||
97a075cd JN |
3797 | LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)", |
3798 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3799 | PFID(&lli->lli_fid), inode); | |
d7e09d03 PT |
3800 | |
3801 | rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh, | |
3802 | NULL, 0, NULL, 0); | |
8bf86fd9 JH |
3803 | ptlrpc_req_finished(it.it_request); |
3804 | it.it_request = NULL; | |
d7e09d03 PT |
3805 | |
3806 | ll_finish_md_op_data(op_data); | |
3807 | ||
e476f2e5 JH |
3808 | mode = it.it_lock_mode; |
3809 | it.it_lock_mode = 0; | |
d7e09d03 PT |
3810 | ll_intent_drop_lock(&it); |
3811 | ||
3812 | if (rc == 0) { | |
3813 | /* set lock data in case this is a new lock */ | |
3814 | ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); | |
3815 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); | |
3816 | if (rc == -EAGAIN) | |
3817 | goto again; | |
3818 | } | |
3819 | mutex_unlock(&lli->lli_layout_mutex); | |
3820 | ||
0a3bdb00 | 3821 | return rc; |
d7e09d03 | 3822 | } |
5ea17d6c JL |
3823 | |
3824 | /** | |
3825 | * This function send a restore request to the MDT | |
3826 | */ | |
1b1594da | 3827 | int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length) |
5ea17d6c JL |
3828 | { |
3829 | struct hsm_user_request *hur; | |
3830 | int len, rc; | |
3831 | ||
3832 | len = sizeof(struct hsm_user_request) + | |
3833 | sizeof(struct hsm_user_item); | |
496a51bd JL |
3834 | hur = kzalloc(len, GFP_NOFS); |
3835 | if (!hur) | |
5ea17d6c JL |
3836 | return -ENOMEM; |
3837 | ||
3838 | hur->hur_request.hr_action = HUA_RESTORE; | |
3839 | hur->hur_request.hr_archive_id = 0; | |
3840 | hur->hur_request.hr_flags = 0; | |
3841 | memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid, | |
3842 | sizeof(hur->hur_user_item[0].hui_fid)); | |
1b1594da JX |
3843 | hur->hur_user_item[0].hui_extent.offset = offset; |
3844 | hur->hur_user_item[0].hui_extent.length = length; | |
5ea17d6c | 3845 | hur->hur_request.hr_itemcount = 1; |
1929c433 | 3846 | rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp, |
5ea17d6c | 3847 | len, hur, NULL); |
97903a26 | 3848 | kfree(hur); |
5ea17d6c JL |
3849 | return rc; |
3850 | } |