projects / deliverable / linux.git / blob
? search:


00b466e667ab9735c60322b082a71d8ef3fd427c
[deliverable/linux.git] / fs / cifs / file.c
1 /*
2 * fs/cifs/file.c
3 *
4 * vfs operations that deal with files
5 *
6 * Copyright (C) International Business Machines Corp., 2002,2010
7 * Author(s): Steve French (sfrench@us.ibm.com)
8 * Jeremy Allison (jra@samba.org)
9 *
10 * This library is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU Lesser General Public License as published
12 * by the Free Software Foundation; either version 2.1 of the License, or
13 * (at your option) any later version.
14 *
15 * This library is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
18 * the GNU Lesser General Public License for more details.
19 *
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this library; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 */
24 #include <linux/fs.h>
25 #include <linux/backing-dev.h>
26 #include <linux/stat.h>
27 #include <linux/fcntl.h>
28 #include <linux/pagemap.h>
29 #include <linux/pagevec.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/delay.h>
33 #include <linux/mount.h>
34 #include <linux/slab.h>
35 #include <asm/div64.h>
36 #include "cifsfs.h"
37 #include "cifspdu.h"
38 #include "cifsglob.h"
39 #include "cifsproto.h"
40 #include "cifs_unicode.h"
41 #include "cifs_debug.h"
42 #include "cifs_fs_sb.h"
43 #include "fscache.h"
44
45 static inline int cifs_convert_flags(unsigned int flags)
46 {
47 if ((flags & O_ACCMODE) == O_RDONLY)
48 return GENERIC_READ;
49 else if ((flags & O_ACCMODE) == O_WRONLY)
50 return GENERIC_WRITE;
51 else if ((flags & O_ACCMODE) == O_RDWR) {
52 /* GENERIC_ALL is too much permission to request
53 can cause unnecessary access denied on create */
54 /* return GENERIC_ALL; */
55 return (GENERIC_READ | GENERIC_WRITE);
56 }
57
58 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
59 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
60 FILE_READ_DATA);
61 }
62
63 static u32 cifs_posix_convert_flags(unsigned int flags)
64 {
65 u32 posix_flags = 0;
66
67 if ((flags & O_ACCMODE) == O_RDONLY)
68 posix_flags = SMB_O_RDONLY;
69 else if ((flags & O_ACCMODE) == O_WRONLY)
70 posix_flags = SMB_O_WRONLY;
71 else if ((flags & O_ACCMODE) == O_RDWR)
72 posix_flags = SMB_O_RDWR;
73
74 if (flags & O_CREAT)
75 posix_flags |= SMB_O_CREAT;
76 if (flags & O_EXCL)
77 posix_flags |= SMB_O_EXCL;
78 if (flags & O_TRUNC)
79 posix_flags |= SMB_O_TRUNC;
80 /* be safe and imply O_SYNC for O_DSYNC */
81 if (flags & O_DSYNC)
82 posix_flags |= SMB_O_SYNC;
83 if (flags & O_DIRECTORY)
84 posix_flags |= SMB_O_DIRECTORY;
85 if (flags & O_NOFOLLOW)
86 posix_flags |= SMB_O_NOFOLLOW;
87 if (flags & O_DIRECT)
88 posix_flags |= SMB_O_DIRECT;
89
90 return posix_flags;
91 }
92
93 static inline int cifs_get_disposition(unsigned int flags)
94 {
95 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
96 return FILE_CREATE;
97 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
98 return FILE_OVERWRITE_IF;
99 else if ((flags & O_CREAT) == O_CREAT)
100 return FILE_OPEN_IF;
101 else if ((flags & O_TRUNC) == O_TRUNC)
102 return FILE_OVERWRITE;
103 else
104 return FILE_OPEN;
105 }
106
107 int cifs_posix_open(char *full_path, struct inode **pinode,
108 struct super_block *sb, int mode, unsigned int f_flags,
109 __u32 *poplock, __u16 *pnetfid, int xid)
110 {
111 int rc;
112 FILE_UNIX_BASIC_INFO *presp_data;
113 __u32 posix_flags = 0;
114 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
115 struct cifs_fattr fattr;
116 struct tcon_link *tlink;
117 struct cifsTconInfo *tcon;
118
119 cFYI(1, "posix open %s", full_path);
120
121 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
122 if (presp_data == NULL)
123 return -ENOMEM;
124
125 tlink = cifs_sb_tlink(cifs_sb);
126 if (IS_ERR(tlink)) {
127 rc = PTR_ERR(tlink);
128 goto posix_open_ret;
129 }
130
131 tcon = tlink_tcon(tlink);
132 mode &= ~current_umask();
133
134 posix_flags = cifs_posix_convert_flags(f_flags);
135 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
136 poplock, full_path, cifs_sb->local_nls,
137 cifs_sb->mnt_cifs_flags &
138 CIFS_MOUNT_MAP_SPECIAL_CHR);
139 cifs_put_tlink(tlink);
140
141 if (rc)
142 goto posix_open_ret;
143
144 if (presp_data->Type == cpu_to_le32(-1))
145 goto posix_open_ret; /* open ok, caller does qpathinfo */
146
147 if (!pinode)
148 goto posix_open_ret; /* caller does not need info */
149
150 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
151
152 /* get new inode and set it up */
153 if (*pinode == NULL) {
154 cifs_fill_uniqueid(sb, &fattr);
155 *pinode = cifs_iget(sb, &fattr);
156 if (!*pinode) {
157 rc = -ENOMEM;
158 goto posix_open_ret;
159 }
160 } else {
161 cifs_fattr_to_inode(*pinode, &fattr);
162 }
163
164 posix_open_ret:
165 kfree(presp_data);
166 return rc;
167 }
168
169 static int
170 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
171 struct cifsTconInfo *tcon, unsigned int f_flags, __u32 *poplock,
172 __u16 *pnetfid, int xid)
173 {
174 int rc;
175 int desiredAccess;
176 int disposition;
177 FILE_ALL_INFO *buf;
178
179 desiredAccess = cifs_convert_flags(f_flags);
180
181 /*********************************************************************
182 * open flag mapping table:
183 *
184 * POSIX Flag CIFS Disposition
185 * ---------- ----------------
186 * O_CREAT FILE_OPEN_IF
187 * O_CREAT | O_EXCL FILE_CREATE
188 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
189 * O_TRUNC FILE_OVERWRITE
190 * none of the above FILE_OPEN
191 *
192 * Note that there is not a direct match between disposition
193 * FILE_SUPERSEDE (ie create whether or not file exists although
194 * O_CREAT | O_TRUNC is similar but truncates the existing
195 * file rather than creating a new file as FILE_SUPERSEDE does
196 * (which uses the attributes / metadata passed in on open call)
197 *?
198 *? O_SYNC is a reasonable match to CIFS writethrough flag
199 *? and the read write flags match reasonably. O_LARGEFILE
200 *? is irrelevant because largefile support is always used
201 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
202 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
203 *********************************************************************/
204
205 disposition = cifs_get_disposition(f_flags);
206
207 /* BB pass O_SYNC flag through on file attributes .. BB */
208
209 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
210 if (!buf)
211 return -ENOMEM;
212
213 if (tcon->ses->capabilities & CAP_NT_SMBS)
214 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
215 desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
216 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
217 & CIFS_MOUNT_MAP_SPECIAL_CHR);
218 else
219 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
220 desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
221 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
222 & CIFS_MOUNT_MAP_SPECIAL_CHR);
223
224 if (rc)
225 goto out;
226
227 if (tcon->unix_ext)
228 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
229 xid);
230 else
231 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
232 xid, pnetfid);
233
234 out:
235 kfree(buf);
236 return rc;
237 }
238
239 struct cifsFileInfo *
240 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
241 struct tcon_link *tlink, __u32 oplock)
242 {
243 struct dentry *dentry = file->f_path.dentry;
244 struct inode *inode = dentry->d_inode;
245 struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
246 struct cifsFileInfo *pCifsFile;
247
248 pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
249 if (pCifsFile == NULL)
250 return pCifsFile;
251
252 pCifsFile->count = 1;
253 pCifsFile->netfid = fileHandle;
254 pCifsFile->pid = current->tgid;
255 pCifsFile->uid = current_fsuid();
256 pCifsFile->dentry = dget(dentry);
257 pCifsFile->f_flags = file->f_flags;
258 pCifsFile->invalidHandle = false;
259 pCifsFile->tlink = cifs_get_tlink(tlink);
260 mutex_init(&pCifsFile->fh_mutex);
261 mutex_init(&pCifsFile->lock_mutex);
262 INIT_LIST_HEAD(&pCifsFile->llist);
263 INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
264
265 spin_lock(&cifs_file_list_lock);
266 list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
267 /* if readable file instance put first in list*/
268 if (file->f_mode & FMODE_READ)
269 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
270 else
271 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
272 spin_unlock(&cifs_file_list_lock);
273
274 cifs_set_oplock_level(pCifsInode, oplock);
275
276 file->private_data = pCifsFile;
277 return pCifsFile;
278 }
279
280 /*
281 * Release a reference on the file private data. This may involve closing
282 * the filehandle out on the server. Must be called without holding
283 * cifs_file_list_lock.
284 */
285 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
286 {
287 struct inode *inode = cifs_file->dentry->d_inode;
288 struct cifsTconInfo *tcon = tlink_tcon(cifs_file->tlink);
289 struct cifsInodeInfo *cifsi = CIFS_I(inode);
290 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
291 struct cifsLockInfo *li, *tmp;
292
293 spin_lock(&cifs_file_list_lock);
294 if (--cifs_file->count > 0) {
295 spin_unlock(&cifs_file_list_lock);
296 return;
297 }
298
299 /* remove it from the lists */
300 list_del(&cifs_file->flist);
301 list_del(&cifs_file->tlist);
302
303 if (list_empty(&cifsi->openFileList)) {
304 cFYI(1, "closing last open instance for inode %p",
305 cifs_file->dentry->d_inode);
306
307 /* in strict cache mode we need invalidate mapping on the last
308 close because it may cause a error when we open this file
309 again and get at least level II oplock */
310 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
311 CIFS_I(inode)->invalid_mapping = true;
312
313 cifs_set_oplock_level(cifsi, 0);
314 }
315 spin_unlock(&cifs_file_list_lock);
316
317 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
318 int xid, rc;
319
320 xid = GetXid();
321 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
322 FreeXid(xid);
323 }
324
325 /* Delete any outstanding lock records. We'll lose them when the file
326 * is closed anyway.
327 */
328 mutex_lock(&cifs_file->lock_mutex);
329 list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
330 list_del(&li->llist);
331 kfree(li);
332 }
333 mutex_unlock(&cifs_file->lock_mutex);
334
335 cifs_put_tlink(cifs_file->tlink);
336 dput(cifs_file->dentry);
337 kfree(cifs_file);
338 }
339
340 int cifs_open(struct inode *inode, struct file *file)
341 {
342 int rc = -EACCES;
343 int xid;
344 __u32 oplock;
345 struct cifs_sb_info *cifs_sb;
346 struct cifsTconInfo *tcon;
347 struct tcon_link *tlink;
348 struct cifsFileInfo *pCifsFile = NULL;
349 char *full_path = NULL;
350 bool posix_open_ok = false;
351 __u16 netfid;
352
353 xid = GetXid();
354
355 cifs_sb = CIFS_SB(inode->i_sb);
356 tlink = cifs_sb_tlink(cifs_sb);
357 if (IS_ERR(tlink)) {
358 FreeXid(xid);
359 return PTR_ERR(tlink);
360 }
361 tcon = tlink_tcon(tlink);
362
363 full_path = build_path_from_dentry(file->f_path.dentry);
364 if (full_path == NULL) {
365 rc = -ENOMEM;
366 goto out;
367 }
368
369 cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
370 inode, file->f_flags, full_path);
371
372 if (oplockEnabled)
373 oplock = REQ_OPLOCK;
374 else
375 oplock = 0;
376
377 if (!tcon->broken_posix_open && tcon->unix_ext &&
378 (tcon->ses->capabilities & CAP_UNIX) &&
379 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
380 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
381 /* can not refresh inode info since size could be stale */
382 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
383 cifs_sb->mnt_file_mode /* ignored */,
384 file->f_flags, &oplock, &netfid, xid);
385 if (rc == 0) {
386 cFYI(1, "posix open succeeded");
387 posix_open_ok = true;
388 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
389 if (tcon->ses->serverNOS)
390 cERROR(1, "server %s of type %s returned"
391 " unexpected error on SMB posix open"
392 ", disabling posix open support."
393 " Check if server update available.",
394 tcon->ses->serverName,
395 tcon->ses->serverNOS);
396 tcon->broken_posix_open = true;
397 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
398 (rc != -EOPNOTSUPP)) /* path not found or net err */
399 goto out;
400 /* else fallthrough to retry open the old way on network i/o
401 or DFS errors */
402 }
403
404 if (!posix_open_ok) {
405 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
406 file->f_flags, &oplock, &netfid, xid);
407 if (rc)
408 goto out;
409 }
410
411 pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
412 if (pCifsFile == NULL) {
413 CIFSSMBClose(xid, tcon, netfid);
414 rc = -ENOMEM;
415 goto out;
416 }
417
418 cifs_fscache_set_inode_cookie(inode, file);
419
420 if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
421 /* time to set mode which we can not set earlier due to
422 problems creating new read-only files */
423 struct cifs_unix_set_info_args args = {
424 .mode = inode->i_mode,
425 .uid = NO_CHANGE_64,
426 .gid = NO_CHANGE_64,
427 .ctime = NO_CHANGE_64,
428 .atime = NO_CHANGE_64,
429 .mtime = NO_CHANGE_64,
430 .device = 0,
431 };
432 CIFSSMBUnixSetFileInfo(xid, tcon, &args, netfid,
433 pCifsFile->pid);
434 }
435
436 out:
437 kfree(full_path);
438 FreeXid(xid);
439 cifs_put_tlink(tlink);
440 return rc;
441 }
442
443 /* Try to reacquire byte range locks that were released when session */
444 /* to server was lost */
445 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
446 {
447 int rc = 0;
448
449 /* BB list all locks open on this file and relock */
450
451 return rc;
452 }
453
454 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
455 {
456 int rc = -EACCES;
457 int xid;
458 __u32 oplock;
459 struct cifs_sb_info *cifs_sb;
460 struct cifsTconInfo *tcon;
461 struct cifsInodeInfo *pCifsInode;
462 struct inode *inode;
463 char *full_path = NULL;
464 int desiredAccess;
465 int disposition = FILE_OPEN;
466 __u16 netfid;
467
468 xid = GetXid();
469 mutex_lock(&pCifsFile->fh_mutex);
470 if (!pCifsFile->invalidHandle) {
471 mutex_unlock(&pCifsFile->fh_mutex);
472 rc = 0;
473 FreeXid(xid);
474 return rc;
475 }
476
477 inode = pCifsFile->dentry->d_inode;
478 cifs_sb = CIFS_SB(inode->i_sb);
479 tcon = tlink_tcon(pCifsFile->tlink);
480
481 /* can not grab rename sem here because various ops, including
482 those that already have the rename sem can end up causing writepage
483 to get called and if the server was down that means we end up here,
484 and we can never tell if the caller already has the rename_sem */
485 full_path = build_path_from_dentry(pCifsFile->dentry);
486 if (full_path == NULL) {
487 rc = -ENOMEM;
488 mutex_unlock(&pCifsFile->fh_mutex);
489 FreeXid(xid);
490 return rc;
491 }
492
493 cFYI(1, "inode = 0x%p file flags 0x%x for %s",
494 inode, pCifsFile->f_flags, full_path);
495
496 if (oplockEnabled)
497 oplock = REQ_OPLOCK;
498 else
499 oplock = 0;
500
501 if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
502 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
503 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
504
505 /*
506 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
507 * original open. Must mask them off for a reopen.
508 */
509 unsigned int oflags = pCifsFile->f_flags &
510 ~(O_CREAT | O_EXCL | O_TRUNC);
511
512 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
513 cifs_sb->mnt_file_mode /* ignored */,
514 oflags, &oplock, &netfid, xid);
515 if (rc == 0) {
516 cFYI(1, "posix reopen succeeded");
517 goto reopen_success;
518 }
519 /* fallthrough to retry open the old way on errors, especially
520 in the reconnect path it is important to retry hard */
521 }
522
523 desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
524
525 /* Can not refresh inode by passing in file_info buf to be returned
526 by SMBOpen and then calling get_inode_info with returned buf
527 since file might have write behind data that needs to be flushed
528 and server version of file size can be stale. If we knew for sure
529 that inode was not dirty locally we could do this */
530
531 rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
532 CREATE_NOT_DIR, &netfid, &oplock, NULL,
533 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
534 CIFS_MOUNT_MAP_SPECIAL_CHR);
535 if (rc) {
536 mutex_unlock(&pCifsFile->fh_mutex);
537 cFYI(1, "cifs_open returned 0x%x", rc);
538 cFYI(1, "oplock: %d", oplock);
539 goto reopen_error_exit;
540 }
541
542 reopen_success:
543 pCifsFile->netfid = netfid;
544 pCifsFile->invalidHandle = false;
545 mutex_unlock(&pCifsFile->fh_mutex);
546 pCifsInode = CIFS_I(inode);
547
548 if (can_flush) {
549 rc = filemap_write_and_wait(inode->i_mapping);
550 mapping_set_error(inode->i_mapping, rc);
551
552 if (tcon->unix_ext)
553 rc = cifs_get_inode_info_unix(&inode,
554 full_path, inode->i_sb, xid);
555 else
556 rc = cifs_get_inode_info(&inode,
557 full_path, NULL, inode->i_sb,
558 xid, NULL);
559 } /* else we are writing out data to server already
560 and could deadlock if we tried to flush data, and
561 since we do not know if we have data that would
562 invalidate the current end of file on the server
563 we can not go to the server to get the new inod
564 info */
565
566 cifs_set_oplock_level(pCifsInode, oplock);
567
568 cifs_relock_file(pCifsFile);
569
570 reopen_error_exit:
571 kfree(full_path);
572 FreeXid(xid);
573 return rc;
574 }
575
576 int cifs_close(struct inode *inode, struct file *file)
577 {
578 if (file->private_data != NULL) {
579 cifsFileInfo_put(file->private_data);
580 file->private_data = NULL;
581 }
582
583 /* return code from the ->release op is always ignored */
584 return 0;
585 }
586
587 int cifs_closedir(struct inode *inode, struct file *file)
588 {
589 int rc = 0;
590 int xid;
591 struct cifsFileInfo *pCFileStruct = file->private_data;
592 char *ptmp;
593
594 cFYI(1, "Closedir inode = 0x%p", inode);
595
596 xid = GetXid();
597
598 if (pCFileStruct) {
599 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
600
601 cFYI(1, "Freeing private data in close dir");
602 spin_lock(&cifs_file_list_lock);
603 if (!pCFileStruct->srch_inf.endOfSearch &&
604 !pCFileStruct->invalidHandle) {
605 pCFileStruct->invalidHandle = true;
606 spin_unlock(&cifs_file_list_lock);
607 rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
608 cFYI(1, "Closing uncompleted readdir with rc %d",
609 rc);
610 /* not much we can do if it fails anyway, ignore rc */
611 rc = 0;
612 } else
613 spin_unlock(&cifs_file_list_lock);
614 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
615 if (ptmp) {
616 cFYI(1, "closedir free smb buf in srch struct");
617 pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
618 if (pCFileStruct->srch_inf.smallBuf)
619 cifs_small_buf_release(ptmp);
620 else
621 cifs_buf_release(ptmp);
622 }
623 cifs_put_tlink(pCFileStruct->tlink);
624 kfree(file->private_data);
625 file->private_data = NULL;
626 }
627 /* BB can we lock the filestruct while this is going on? */
628 FreeXid(xid);
629 return rc;
630 }
631
632 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
633 __u64 offset, __u8 lockType)
634 {
635 struct cifsLockInfo *li =
636 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
637 if (li == NULL)
638 return -ENOMEM;
639 li->offset = offset;
640 li->length = len;
641 li->type = lockType;
642 mutex_lock(&fid->lock_mutex);
643 list_add(&li->llist, &fid->llist);
644 mutex_unlock(&fid->lock_mutex);
645 return 0;
646 }
647
648 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
649 {
650 int rc, xid;
651 __u32 numLock = 0;
652 __u32 numUnlock = 0;
653 __u64 length;
654 bool wait_flag = false;
655 struct cifs_sb_info *cifs_sb;
656 struct cifsTconInfo *tcon;
657 __u16 netfid;
658 __u8 lockType = LOCKING_ANDX_LARGE_FILES;
659 bool posix_locking = 0;
660
661 length = 1 + pfLock->fl_end - pfLock->fl_start;
662 rc = -EACCES;
663 xid = GetXid();
664
665 cFYI(1, "Lock parm: 0x%x flockflags: "
666 "0x%x flocktype: 0x%x start: %lld end: %lld",
667 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
668 pfLock->fl_end);
669
670 if (pfLock->fl_flags & FL_POSIX)
671 cFYI(1, "Posix");
672 if (pfLock->fl_flags & FL_FLOCK)
673 cFYI(1, "Flock");
674 if (pfLock->fl_flags & FL_SLEEP) {
675 cFYI(1, "Blocking lock");
676 wait_flag = true;
677 }
678 if (pfLock->fl_flags & FL_ACCESS)
679 cFYI(1, "Process suspended by mandatory locking - "
680 "not implemented yet");
681 if (pfLock->fl_flags & FL_LEASE)
682 cFYI(1, "Lease on file - not implemented yet");
683 if (pfLock->fl_flags &
684 (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
685 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
686
687 if (pfLock->fl_type == F_WRLCK) {
688 cFYI(1, "F_WRLCK ");
689 numLock = 1;
690 } else if (pfLock->fl_type == F_UNLCK) {
691 cFYI(1, "F_UNLCK");
692 numUnlock = 1;
693 /* Check if unlock includes more than
694 one lock range */
695 } else if (pfLock->fl_type == F_RDLCK) {
696 cFYI(1, "F_RDLCK");
697 lockType |= LOCKING_ANDX_SHARED_LOCK;
698 numLock = 1;
699 } else if (pfLock->fl_type == F_EXLCK) {
700 cFYI(1, "F_EXLCK");
701 numLock = 1;
702 } else if (pfLock->fl_type == F_SHLCK) {
703 cFYI(1, "F_SHLCK");
704 lockType |= LOCKING_ANDX_SHARED_LOCK;
705 numLock = 1;
706 } else
707 cFYI(1, "Unknown type of lock");
708
709 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
710 tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
711 netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
712
713 if ((tcon->ses->capabilities & CAP_UNIX) &&
714 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
715 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
716 posix_locking = 1;
717 /* BB add code here to normalize offset and length to
718 account for negative length which we can not accept over the
719 wire */
720 if (IS_GETLK(cmd)) {
721 if (posix_locking) {
722 int posix_lock_type;
723 if (lockType & LOCKING_ANDX_SHARED_LOCK)
724 posix_lock_type = CIFS_RDLCK;
725 else
726 posix_lock_type = CIFS_WRLCK;
727 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
728 length, pfLock,
729 posix_lock_type, wait_flag);
730 FreeXid(xid);
731 return rc;
732 }
733
734 /* BB we could chain these into one lock request BB */
735 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
736 0, 1, lockType, 0 /* wait flag */, 0);
737 if (rc == 0) {
738 rc = CIFSSMBLock(xid, tcon, netfid, length,
739 pfLock->fl_start, 1 /* numUnlock */ ,
740 0 /* numLock */ , lockType,
741 0 /* wait flag */, 0);
742 pfLock->fl_type = F_UNLCK;
743 if (rc != 0)
744 cERROR(1, "Error unlocking previously locked "
745 "range %d during test of lock", rc);
746 rc = 0;
747
748 } else {
749 /* if rc == ERR_SHARING_VIOLATION ? */
750 rc = 0;
751
752 if (lockType & LOCKING_ANDX_SHARED_LOCK) {
753 pfLock->fl_type = F_WRLCK;
754 } else {
755 rc = CIFSSMBLock(xid, tcon, netfid, length,
756 pfLock->fl_start, 0, 1,
757 lockType | LOCKING_ANDX_SHARED_LOCK,
758 0 /* wait flag */, 0);
759 if (rc == 0) {
760 rc = CIFSSMBLock(xid, tcon, netfid,
761 length, pfLock->fl_start, 1, 0,
762 lockType |
763 LOCKING_ANDX_SHARED_LOCK,
764 0 /* wait flag */, 0);
765 pfLock->fl_type = F_RDLCK;
766 if (rc != 0)
767 cERROR(1, "Error unlocking "
768 "previously locked range %d "
769 "during test of lock", rc);
770 rc = 0;
771 } else {
772 pfLock->fl_type = F_WRLCK;
773 rc = 0;
774 }
775 }
776 }
777
778 FreeXid(xid);
779 return rc;
780 }
781
782 if (!numLock && !numUnlock) {
783 /* if no lock or unlock then nothing
784 to do since we do not know what it is */
785 FreeXid(xid);
786 return -EOPNOTSUPP;
787 }
788
789 if (posix_locking) {
790 int posix_lock_type;
791 if (lockType & LOCKING_ANDX_SHARED_LOCK)
792 posix_lock_type = CIFS_RDLCK;
793 else
794 posix_lock_type = CIFS_WRLCK;
795
796 if (numUnlock == 1)
797 posix_lock_type = CIFS_UNLCK;
798
799 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
800 length, pfLock,
801 posix_lock_type, wait_flag);
802 } else {
803 struct cifsFileInfo *fid = file->private_data;
804
805 if (numLock) {
806 rc = CIFSSMBLock(xid, tcon, netfid, length,
807 pfLock->fl_start, 0, numLock, lockType,
808 wait_flag, 0);
809
810 if (rc == 0) {
811 /* For Windows locks we must store them. */
812 rc = store_file_lock(fid, length,
813 pfLock->fl_start, lockType);
814 }
815 } else if (numUnlock) {
816 /* For each stored lock that this unlock overlaps
817 completely, unlock it. */
818 int stored_rc = 0;
819 struct cifsLockInfo *li, *tmp;
820
821 rc = 0;
822 mutex_lock(&fid->lock_mutex);
823 list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
824 if (pfLock->fl_start <= li->offset &&
825 (pfLock->fl_start + length) >=
826 (li->offset + li->length)) {
827 stored_rc = CIFSSMBLock(xid, tcon,
828 netfid, li->length,
829 li->offset, 1, 0,
830 li->type, false, 0);
831 if (stored_rc)
832 rc = stored_rc;
833 else {
834 list_del(&li->llist);
835 kfree(li);
836 }
837 }
838 }
839 mutex_unlock(&fid->lock_mutex);
840 }
841 }
842
843 if (pfLock->fl_flags & FL_POSIX)
844 posix_lock_file_wait(file, pfLock);
845 FreeXid(xid);
846 return rc;
847 }
848
849 /* update the file size (if needed) after a write */
850 void
851 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
852 unsigned int bytes_written)
853 {
854 loff_t end_of_write = offset + bytes_written;
855
856 if (end_of_write > cifsi->server_eof)
857 cifsi->server_eof = end_of_write;
858 }
859
860 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
861 size_t write_size, loff_t *poffset)
862 {
863 struct inode *inode = file->f_path.dentry->d_inode;
864 int rc = 0;
865 unsigned int bytes_written = 0;
866 unsigned int total_written;
867 struct cifs_sb_info *cifs_sb;
868 struct cifsTconInfo *pTcon;
869 int xid;
870 struct cifsFileInfo *open_file;
871 struct cifsInodeInfo *cifsi = CIFS_I(inode);
872
873 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
874
875 /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
876 *poffset, file->f_path.dentry->d_name.name); */
877
878 if (file->private_data == NULL)
879 return -EBADF;
880
881 open_file = file->private_data;
882 pTcon = tlink_tcon(open_file->tlink);
883
884 rc = generic_write_checks(file, poffset, &write_size, 0);
885 if (rc)
886 return rc;
887
888 xid = GetXid();
889
890 for (total_written = 0; write_size > total_written;
891 total_written += bytes_written) {
892 rc = -EAGAIN;
893 while (rc == -EAGAIN) {
894 if (file->private_data == NULL) {
895 /* file has been closed on us */
896 FreeXid(xid);
897 /* if we have gotten here we have written some data
898 and blocked, and the file has been freed on us while
899 we blocked so return what we managed to write */
900 return total_written;
901 }
902 if (open_file->invalidHandle) {
903 /* we could deadlock if we called
904 filemap_fdatawait from here so tell
905 reopen_file not to flush data to server
906 now */
907 rc = cifs_reopen_file(open_file, false);
908 if (rc != 0)
909 break;
910 }
911
912 rc = CIFSSMBWrite(xid, pTcon,
913 open_file->netfid,
914 min_t(const int, cifs_sb->wsize,
915 write_size - total_written),
916 *poffset, &bytes_written,
917 NULL, write_data + total_written, 0);
918 }
919 if (rc || (bytes_written == 0)) {
920 if (total_written)
921 break;
922 else {
923 FreeXid(xid);
924 return rc;
925 }
926 } else {
927 cifs_update_eof(cifsi, *poffset, bytes_written);
928 *poffset += bytes_written;
929 }
930 }
931
932 cifs_stats_bytes_written(pTcon, total_written);
933
934 /* Do not update local mtime - server will set its actual value on write
935 * inode->i_ctime = inode->i_mtime =
936 * current_fs_time(inode->i_sb);*/
937 if (total_written > 0) {
938 spin_lock(&inode->i_lock);
939 if (*poffset > inode->i_size)
940 i_size_write(inode, *poffset);
941 spin_unlock(&inode->i_lock);
942 }
943 mark_inode_dirty_sync(inode);
944
945 FreeXid(xid);
946 return total_written;
947 }
948
949 static ssize_t cifs_write(struct cifsFileInfo *open_file,
950 const char *write_data, size_t write_size,
951 loff_t *poffset)
952 {
953 int rc = 0;
954 unsigned int bytes_written = 0;
955 unsigned int total_written;
956 struct cifs_sb_info *cifs_sb;
957 struct cifsTconInfo *pTcon;
958 int xid;
959 struct dentry *dentry = open_file->dentry;
960 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
961
962 cifs_sb = CIFS_SB(dentry->d_sb);
963
964 cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
965 *poffset, dentry->d_name.name);
966
967 pTcon = tlink_tcon(open_file->tlink);
968
969 xid = GetXid();
970
971 for (total_written = 0; write_size > total_written;
972 total_written += bytes_written) {
973 rc = -EAGAIN;
974 while (rc == -EAGAIN) {
975 struct kvec iov[2];
976 unsigned int len;
977
978 if (open_file->invalidHandle) {
979 /* we could deadlock if we called
980 filemap_fdatawait from here so tell
981 reopen_file not to flush data to
982 server now */
983 rc = cifs_reopen_file(open_file, false);
984 if (rc != 0)
985 break;
986 }
987
988 len = min((size_t)cifs_sb->wsize,
989 write_size - total_written);
990 /* iov[0] is reserved for smb header */
991 iov[1].iov_base = (char *)write_data + total_written;
992 iov[1].iov_len = len;
993 rc = CIFSSMBWrite2(xid, pTcon, open_file->netfid, len,
994 *poffset, &bytes_written, iov, 1, 0);
995 }
996 if (rc || (bytes_written == 0)) {
997 if (total_written)
998 break;
999 else {
1000 FreeXid(xid);
1001 return rc;
1002 }
1003 } else {
1004 cifs_update_eof(cifsi, *poffset, bytes_written);
1005 *poffset += bytes_written;
1006 }
1007 }
1008
1009 cifs_stats_bytes_written(pTcon, total_written);
1010
1011 if (total_written > 0) {
1012 spin_lock(&dentry->d_inode->i_lock);
1013 if (*poffset > dentry->d_inode->i_size)
1014 i_size_write(dentry->d_inode, *poffset);
1015 spin_unlock(&dentry->d_inode->i_lock);
1016 }
1017 mark_inode_dirty_sync(dentry->d_inode);
1018 FreeXid(xid);
1019 return total_written;
1020 }
1021
1022 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1023 bool fsuid_only)
1024 {
1025 struct cifsFileInfo *open_file = NULL;
1026 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1027
1028 /* only filter by fsuid on multiuser mounts */
1029 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1030 fsuid_only = false;
1031
1032 spin_lock(&cifs_file_list_lock);
1033 /* we could simply get the first_list_entry since write-only entries
1034 are always at the end of the list but since the first entry might
1035 have a close pending, we go through the whole list */
1036 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1037 if (fsuid_only && open_file->uid != current_fsuid())
1038 continue;
1039 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1040 if (!open_file->invalidHandle) {
1041 /* found a good file */
1042 /* lock it so it will not be closed on us */
1043 cifsFileInfo_get(open_file);
1044 spin_unlock(&cifs_file_list_lock);
1045 return open_file;
1046 } /* else might as well continue, and look for
1047 another, or simply have the caller reopen it
1048 again rather than trying to fix this handle */
1049 } else /* write only file */
1050 break; /* write only files are last so must be done */
1051 }
1052 spin_unlock(&cifs_file_list_lock);
1053 return NULL;
1054 }
1055
1056 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1057 bool fsuid_only)
1058 {
1059 struct cifsFileInfo *open_file;
1060 struct cifs_sb_info *cifs_sb;
1061 bool any_available = false;
1062 int rc;
1063
1064 /* Having a null inode here (because mapping->host was set to zero by
1065 the VFS or MM) should not happen but we had reports of on oops (due to
1066 it being zero) during stress testcases so we need to check for it */
1067
1068 if (cifs_inode == NULL) {
1069 cERROR(1, "Null inode passed to cifs_writeable_file");
1070 dump_stack();
1071 return NULL;
1072 }
1073
1074 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1075
1076 /* only filter by fsuid on multiuser mounts */
1077 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1078 fsuid_only = false;
1079
1080 spin_lock(&cifs_file_list_lock);
1081 refind_writable:
1082 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1083 if (!any_available && open_file->pid != current->tgid)
1084 continue;
1085 if (fsuid_only && open_file->uid != current_fsuid())
1086 continue;
1087 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1088 cifsFileInfo_get(open_file);
1089
1090 if (!open_file->invalidHandle) {
1091 /* found a good writable file */
1092 spin_unlock(&cifs_file_list_lock);
1093 return open_file;
1094 }
1095
1096 spin_unlock(&cifs_file_list_lock);
1097
1098 /* Had to unlock since following call can block */
1099 rc = cifs_reopen_file(open_file, false);
1100 if (!rc)
1101 return open_file;
1102
1103 /* if it fails, try another handle if possible */
1104 cFYI(1, "wp failed on reopen file");
1105 cifsFileInfo_put(open_file);
1106
1107 spin_lock(&cifs_file_list_lock);
1108
1109 /* else we simply continue to the next entry. Thus
1110 we do not loop on reopen errors. If we
1111 can not reopen the file, for example if we
1112 reconnected to a server with another client
1113 racing to delete or lock the file we would not
1114 make progress if we restarted before the beginning
1115 of the loop here. */
1116 }
1117 }
1118 /* couldn't find useable FH with same pid, try any available */
1119 if (!any_available) {
1120 any_available = true;
1121 goto refind_writable;
1122 }
1123 spin_unlock(&cifs_file_list_lock);
1124 return NULL;
1125 }
1126
1127 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1128 {
1129 struct address_space *mapping = page->mapping;
1130 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1131 char *write_data;
1132 int rc = -EFAULT;
1133 int bytes_written = 0;
1134 struct inode *inode;
1135 struct cifsFileInfo *open_file;
1136
1137 if (!mapping || !mapping->host)
1138 return -EFAULT;
1139
1140 inode = page->mapping->host;
1141
1142 offset += (loff_t)from;
1143 write_data = kmap(page);
1144 write_data += from;
1145
1146 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1147 kunmap(page);
1148 return -EIO;
1149 }
1150
1151 /* racing with truncate? */
1152 if (offset > mapping->host->i_size) {
1153 kunmap(page);
1154 return 0; /* don't care */
1155 }
1156
1157 /* check to make sure that we are not extending the file */
1158 if (mapping->host->i_size - offset < (loff_t)to)
1159 to = (unsigned)(mapping->host->i_size - offset);
1160
1161 open_file = find_writable_file(CIFS_I(mapping->host), false);
1162 if (open_file) {
1163 bytes_written = cifs_write(open_file, write_data,
1164 to - from, &offset);
1165 cifsFileInfo_put(open_file);
1166 /* Does mm or vfs already set times? */
1167 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1168 if ((bytes_written > 0) && (offset))
1169 rc = 0;
1170 else if (bytes_written < 0)
1171 rc = bytes_written;
1172 } else {
1173 cFYI(1, "No writeable filehandles for inode");
1174 rc = -EIO;
1175 }
1176
1177 kunmap(page);
1178 return rc;
1179 }
1180
1181 static int cifs_writepages(struct address_space *mapping,
1182 struct writeback_control *wbc)
1183 {
1184 unsigned int bytes_to_write;
1185 unsigned int bytes_written;
1186 struct cifs_sb_info *cifs_sb;
1187 int done = 0;
1188 pgoff_t end;
1189 pgoff_t index;
1190 int range_whole = 0;
1191 struct kvec *iov;
1192 int len;
1193 int n_iov = 0;
1194 pgoff_t next;
1195 int nr_pages;
1196 __u64 offset = 0;
1197 struct cifsFileInfo *open_file;
1198 struct cifsTconInfo *tcon;
1199 struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1200 struct page *page;
1201 struct pagevec pvec;
1202 int rc = 0;
1203 int scanned = 0;
1204 int xid;
1205
1206 cifs_sb = CIFS_SB(mapping->host->i_sb);
1207
1208 /*
1209 * If wsize is smaller that the page cache size, default to writing
1210 * one page at a time via cifs_writepage
1211 */
1212 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1213 return generic_writepages(mapping, wbc);
1214
1215 iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1216 if (iov == NULL)
1217 return generic_writepages(mapping, wbc);
1218
1219 /*
1220 * if there's no open file, then this is likely to fail too,
1221 * but it'll at least handle the return. Maybe it should be
1222 * a BUG() instead?
1223 */
1224 open_file = find_writable_file(CIFS_I(mapping->host), false);
1225 if (!open_file) {
1226 kfree(iov);
1227 return generic_writepages(mapping, wbc);
1228 }
1229
1230 tcon = tlink_tcon(open_file->tlink);
1231 cifsFileInfo_put(open_file);
1232
1233 xid = GetXid();
1234
1235 pagevec_init(&pvec, 0);
1236 if (wbc->range_cyclic) {
1237 index = mapping->writeback_index; /* Start from prev offset */
1238 end = -1;
1239 } else {
1240 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1241 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1242 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1243 range_whole = 1;
1244 scanned = 1;
1245 }
1246 retry:
1247 while (!done && (index <= end) &&
1248 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1249 PAGECACHE_TAG_DIRTY,
1250 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1251 int first;
1252 unsigned int i;
1253
1254 first = -1;
1255 next = 0;
1256 n_iov = 0;
1257 bytes_to_write = 0;
1258
1259 for (i = 0; i < nr_pages; i++) {
1260 page = pvec.pages[i];
1261 /*
1262 * At this point we hold neither mapping->tree_lock nor
1263 * lock on the page itself: the page may be truncated or
1264 * invalidated (changing page->mapping to NULL), or even
1265 * swizzled back from swapper_space to tmpfs file
1266 * mapping
1267 */
1268
1269 if (first < 0)
1270 lock_page(page);
1271 else if (!trylock_page(page))
1272 break;
1273
1274 if (unlikely(page->mapping != mapping)) {
1275 unlock_page(page);
1276 break;
1277 }
1278
1279 if (!wbc->range_cyclic && page->index > end) {
1280 done = 1;
1281 unlock_page(page);
1282 break;
1283 }
1284
1285 if (next && (page->index != next)) {
1286 /* Not next consecutive page */
1287 unlock_page(page);
1288 break;
1289 }
1290
1291 if (wbc->sync_mode != WB_SYNC_NONE)
1292 wait_on_page_writeback(page);
1293
1294 if (PageWriteback(page) ||
1295 !clear_page_dirty_for_io(page)) {
1296 unlock_page(page);
1297 break;
1298 }
1299
1300 /*
1301 * This actually clears the dirty bit in the radix tree.
1302 * See cifs_writepage() for more commentary.
1303 */
1304 set_page_writeback(page);
1305
1306 if (page_offset(page) >= mapping->host->i_size) {
1307 done = 1;
1308 unlock_page(page);
1309 end_page_writeback(page);
1310 break;
1311 }
1312
1313 /*
1314 * BB can we get rid of this? pages are held by pvec
1315 */
1316 page_cache_get(page);
1317
1318 len = min(mapping->host->i_size - page_offset(page),
1319 (loff_t)PAGE_CACHE_SIZE);
1320
1321 /* reserve iov[0] for the smb header */
1322 n_iov++;
1323 iov[n_iov].iov_base = kmap(page);
1324 iov[n_iov].iov_len = len;
1325 bytes_to_write += len;
1326
1327 if (first < 0) {
1328 first = i;
1329 offset = page_offset(page);
1330 }
1331 next = page->index + 1;
1332 if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1333 break;
1334 }
1335 if (n_iov) {
1336 retry_write:
1337 open_file = find_writable_file(CIFS_I(mapping->host),
1338 false);
1339 if (!open_file) {
1340 cERROR(1, "No writable handles for inode");
1341 rc = -EBADF;
1342 } else {
1343 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1344 bytes_to_write, offset,
1345 &bytes_written, iov, n_iov,
1346 0);
1347 cifsFileInfo_put(open_file);
1348 }
1349
1350 cFYI(1, "Write2 rc=%d, wrote=%u", rc, bytes_written);
1351
1352 /*
1353 * For now, treat a short write as if nothing got
1354 * written. A zero length write however indicates
1355 * ENOSPC or EFBIG. We have no way to know which
1356 * though, so call it ENOSPC for now. EFBIG would
1357 * get translated to AS_EIO anyway.
1358 *
1359 * FIXME: make it take into account the data that did
1360 * get written
1361 */
1362 if (rc == 0) {
1363 if (bytes_written == 0)
1364 rc = -ENOSPC;
1365 else if (bytes_written < bytes_to_write)
1366 rc = -EAGAIN;
1367 }
1368
1369 /* retry on data-integrity flush */
1370 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN)
1371 goto retry_write;
1372
1373 /* fix the stats and EOF */
1374 if (bytes_written > 0) {
1375 cifs_stats_bytes_written(tcon, bytes_written);
1376 cifs_update_eof(cifsi, offset, bytes_written);
1377 }
1378
1379 for (i = 0; i < n_iov; i++) {
1380 page = pvec.pages[first + i];
1381 /* on retryable write error, redirty page */
1382 if (rc == -EAGAIN)
1383 redirty_page_for_writepage(wbc, page);
1384 else if (rc != 0)
1385 SetPageError(page);
1386 kunmap(page);
1387 unlock_page(page);
1388 end_page_writeback(page);
1389 page_cache_release(page);
1390 }
1391
1392 if (rc != -EAGAIN)
1393 mapping_set_error(mapping, rc);
1394 else
1395 rc = 0;
1396
1397 if ((wbc->nr_to_write -= n_iov) <= 0)
1398 done = 1;
1399 index = next;
1400 } else
1401 /* Need to re-find the pages we skipped */
1402 index = pvec.pages[0]->index + 1;
1403
1404 pagevec_release(&pvec);
1405 }
1406 if (!scanned && !done) {
1407 /*
1408 * We hit the last page and there is more work to be done: wrap
1409 * back to the start of the file
1410 */
1411 scanned = 1;
1412 index = 0;
1413 goto retry;
1414 }
1415 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1416 mapping->writeback_index = index;
1417
1418 FreeXid(xid);
1419 kfree(iov);
1420 return rc;
1421 }
1422
1423 static int
1424 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
1425 {
1426 int rc;
1427 int xid;
1428
1429 xid = GetXid();
1430 /* BB add check for wbc flags */
1431 page_cache_get(page);
1432 if (!PageUptodate(page))
1433 cFYI(1, "ppw - page not up to date");
1434
1435 /*
1436 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1437 *
1438 * A writepage() implementation always needs to do either this,
1439 * or re-dirty the page with "redirty_page_for_writepage()" in
1440 * the case of a failure.
1441 *
1442 * Just unlocking the page will cause the radix tree tag-bits
1443 * to fail to update with the state of the page correctly.
1444 */
1445 set_page_writeback(page);
1446 retry_write:
1447 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1448 if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
1449 goto retry_write;
1450 else if (rc == -EAGAIN)
1451 redirty_page_for_writepage(wbc, page);
1452 else if (rc != 0)
1453 SetPageError(page);
1454 else
1455 SetPageUptodate(page);
1456 end_page_writeback(page);
1457 page_cache_release(page);
1458 FreeXid(xid);
1459 return rc;
1460 }
1461
1462 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1463 {
1464 int rc = cifs_writepage_locked(page, wbc);
1465 unlock_page(page);
1466 return rc;
1467 }
1468
1469 static int cifs_write_end(struct file *file, struct address_space *mapping,
1470 loff_t pos, unsigned len, unsigned copied,
1471 struct page *page, void *fsdata)
1472 {
1473 int rc;
1474 struct inode *inode = mapping->host;
1475
1476 cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1477 page, pos, copied);
1478
1479 if (PageChecked(page)) {
1480 if (copied == len)
1481 SetPageUptodate(page);
1482 ClearPageChecked(page);
1483 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1484 SetPageUptodate(page);
1485
1486 if (!PageUptodate(page)) {
1487 char *page_data;
1488 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1489 int xid;
1490
1491 xid = GetXid();
1492 /* this is probably better than directly calling
1493 partialpage_write since in this function the file handle is
1494 known which we might as well leverage */
1495 /* BB check if anything else missing out of ppw
1496 such as updating last write time */
1497 page_data = kmap(page);
1498 rc = cifs_write(file->private_data, page_data + offset,
1499 copied, &pos);
1500 /* if (rc < 0) should we set writebehind rc? */
1501 kunmap(page);
1502
1503 FreeXid(xid);
1504 } else {
1505 rc = copied;
1506 pos += copied;
1507 set_page_dirty(page);
1508 }
1509
1510 if (rc > 0) {
1511 spin_lock(&inode->i_lock);
1512 if (pos > inode->i_size)
1513 i_size_write(inode, pos);
1514 spin_unlock(&inode->i_lock);
1515 }
1516
1517 unlock_page(page);
1518 page_cache_release(page);
1519
1520 return rc;
1521 }
1522
1523 int cifs_strict_fsync(struct file *file, int datasync)
1524 {
1525 int xid;
1526 int rc = 0;
1527 struct cifsTconInfo *tcon;
1528 struct cifsFileInfo *smbfile = file->private_data;
1529 struct inode *inode = file->f_path.dentry->d_inode;
1530 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1531
1532 xid = GetXid();
1533
1534 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1535 file->f_path.dentry->d_name.name, datasync);
1536
1537 if (!CIFS_I(inode)->clientCanCacheRead)
1538 cifs_invalidate_mapping(inode);
1539
1540 tcon = tlink_tcon(smbfile->tlink);
1541 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1542 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1543
1544 FreeXid(xid);
1545 return rc;
1546 }
1547
1548 int cifs_fsync(struct file *file, int datasync)
1549 {
1550 int xid;
1551 int rc = 0;
1552 struct cifsTconInfo *tcon;
1553 struct cifsFileInfo *smbfile = file->private_data;
1554 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1555
1556 xid = GetXid();
1557
1558 cFYI(1, "Sync file - name: %s datasync: 0x%x",
1559 file->f_path.dentry->d_name.name, datasync);
1560
1561 tcon = tlink_tcon(smbfile->tlink);
1562 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1563 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1564
1565 FreeXid(xid);
1566 return rc;
1567 }
1568
1569 /*
1570 * As file closes, flush all cached write data for this inode checking
1571 * for write behind errors.
1572 */
1573 int cifs_flush(struct file *file, fl_owner_t id)
1574 {
1575 struct inode *inode = file->f_path.dentry->d_inode;
1576 int rc = 0;
1577
1578 if (file->f_mode & FMODE_WRITE)
1579 rc = filemap_write_and_wait(inode->i_mapping);
1580
1581 cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1582
1583 return rc;
1584 }
1585
1586 static int
1587 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
1588 {
1589 int rc = 0;
1590 unsigned long i;
1591
1592 for (i = 0; i < num_pages; i++) {
1593 pages[i] = alloc_page(__GFP_HIGHMEM);
1594 if (!pages[i]) {
1595 /*
1596 * save number of pages we have already allocated and
1597 * return with ENOMEM error
1598 */
1599 num_pages = i;
1600 rc = -ENOMEM;
1601 goto error;
1602 }
1603 }
1604
1605 return rc;
1606
1607 error:
1608 for (i = 0; i < num_pages; i++)
1609 put_page(pages[i]);
1610 return rc;
1611 }
1612
1613 static inline
1614 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
1615 {
1616 size_t num_pages;
1617 size_t clen;
1618
1619 clen = min_t(const size_t, len, wsize);
1620 num_pages = clen / PAGE_CACHE_SIZE;
1621 if (clen % PAGE_CACHE_SIZE)
1622 num_pages++;
1623
1624 if (cur_len)
1625 *cur_len = clen;
1626
1627 return num_pages;
1628 }
1629
1630 static ssize_t
1631 cifs_iovec_write(struct file *file, const struct iovec *iov,
1632 unsigned long nr_segs, loff_t *poffset)
1633 {
1634 unsigned int written;
1635 unsigned long num_pages, npages, i;
1636 size_t copied, len, cur_len;
1637 ssize_t total_written = 0;
1638 struct kvec *to_send;
1639 struct page **pages;
1640 struct iov_iter it;
1641 struct inode *inode;
1642 struct cifsFileInfo *open_file;
1643 struct cifsTconInfo *pTcon;
1644 struct cifs_sb_info *cifs_sb;
1645 int xid, rc;
1646
1647 len = iov_length(iov, nr_segs);
1648 if (!len)
1649 return 0;
1650
1651 rc = generic_write_checks(file, poffset, &len, 0);
1652 if (rc)
1653 return rc;
1654
1655 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1656 num_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
1657
1658 pages = kmalloc(sizeof(struct pages *)*num_pages, GFP_KERNEL);
1659 if (!pages)
1660 return -ENOMEM;
1661
1662 to_send = kmalloc(sizeof(struct kvec)*(num_pages + 1), GFP_KERNEL);
1663 if (!to_send) {
1664 kfree(pages);
1665 return -ENOMEM;
1666 }
1667
1668 rc = cifs_write_allocate_pages(pages, num_pages);
1669 if (rc) {
1670 kfree(pages);
1671 kfree(to_send);
1672 return rc;
1673 }
1674
1675 xid = GetXid();
1676 open_file = file->private_data;
1677 pTcon = tlink_tcon(open_file->tlink);
1678 inode = file->f_path.dentry->d_inode;
1679
1680 iov_iter_init(&it, iov, nr_segs, len, 0);
1681 npages = num_pages;
1682
1683 do {
1684 size_t save_len = cur_len;
1685 for (i = 0; i < npages; i++) {
1686 copied = min_t(const size_t, cur_len, PAGE_CACHE_SIZE);
1687 copied = iov_iter_copy_from_user(pages[i], &it, 0,
1688 copied);
1689 cur_len -= copied;
1690 iov_iter_advance(&it, copied);
1691 to_send[i+1].iov_base = kmap(pages[i]);
1692 to_send[i+1].iov_len = copied;
1693 }
1694
1695 cur_len = save_len - cur_len;
1696
1697 do {
1698 if (open_file->invalidHandle) {
1699 rc = cifs_reopen_file(open_file, false);
1700 if (rc != 0)
1701 break;
1702 }
1703 rc = CIFSSMBWrite2(xid, pTcon, open_file->netfid,
1704 cur_len, *poffset, &written,
1705 to_send, npages, 0);
1706 } while (rc == -EAGAIN);
1707
1708 for (i = 0; i < npages; i++)
1709 kunmap(pages[i]);
1710
1711 if (written) {
1712 len -= written;
1713 total_written += written;
1714 cifs_update_eof(CIFS_I(inode), *poffset, written);
1715 *poffset += written;
1716 } else if (rc < 0) {
1717 if (!total_written)
1718 total_written = rc;
1719 break;
1720 }
1721
1722 /* get length and number of kvecs of the next write */
1723 npages = get_numpages(cifs_sb->wsize, len, &cur_len);
1724 } while (len > 0);
1725
1726 if (total_written > 0) {
1727 spin_lock(&inode->i_lock);
1728 if (*poffset > inode->i_size)
1729 i_size_write(inode, *poffset);
1730 spin_unlock(&inode->i_lock);
1731 }
1732
1733 cifs_stats_bytes_written(pTcon, total_written);
1734 mark_inode_dirty_sync(inode);
1735
1736 for (i = 0; i < num_pages; i++)
1737 put_page(pages[i]);
1738 kfree(to_send);
1739 kfree(pages);
1740 FreeXid(xid);
1741 return total_written;
1742 }
1743
1744 static ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
1745 unsigned long nr_segs, loff_t pos)
1746 {
1747 ssize_t written;
1748 struct inode *inode;
1749
1750 inode = iocb->ki_filp->f_path.dentry->d_inode;
1751
1752 /*
1753 * BB - optimize the way when signing is disabled. We can drop this
1754 * extra memory-to-memory copying and use iovec buffers for constructing
1755 * write request.
1756 */
1757
1758 written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
1759 if (written > 0) {
1760 CIFS_I(inode)->invalid_mapping = true;
1761 iocb->ki_pos = pos;
1762 }
1763
1764 return written;
1765 }
1766
1767 ssize_t cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
1768 unsigned long nr_segs, loff_t pos)
1769 {
1770 struct inode *inode;
1771
1772 inode = iocb->ki_filp->f_path.dentry->d_inode;
1773
1774 if (CIFS_I(inode)->clientCanCacheAll)
1775 return generic_file_aio_write(iocb, iov, nr_segs, pos);
1776
1777 /*
1778 * In strict cache mode we need to write the data to the server exactly
1779 * from the pos to pos+len-1 rather than flush all affected pages
1780 * because it may cause a error with mandatory locks on these pages but
1781 * not on the region from pos to ppos+len-1.
1782 */
1783
1784 return cifs_user_writev(iocb, iov, nr_segs, pos);
1785 }
1786
1787 static ssize_t
1788 cifs_iovec_read(struct file *file, const struct iovec *iov,
1789 unsigned long nr_segs, loff_t *poffset)
1790 {
1791 int rc;
1792 int xid;
1793 ssize_t total_read;
1794 unsigned int bytes_read = 0;
1795 size_t len, cur_len;
1796 int iov_offset = 0;
1797 struct cifs_sb_info *cifs_sb;
1798 struct cifsTconInfo *pTcon;
1799 struct cifsFileInfo *open_file;
1800 struct smb_com_read_rsp *pSMBr;
1801 char *read_data;
1802
1803 if (!nr_segs)
1804 return 0;
1805
1806 len = iov_length(iov, nr_segs);
1807 if (!len)
1808 return 0;
1809
1810 xid = GetXid();
1811 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1812
1813 open_file = file->private_data;
1814 pTcon = tlink_tcon(open_file->tlink);
1815
1816 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1817 cFYI(1, "attempting read on write only file instance");
1818
1819 for (total_read = 0; total_read < len; total_read += bytes_read) {
1820 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
1821 rc = -EAGAIN;
1822 read_data = NULL;
1823
1824 while (rc == -EAGAIN) {
1825 int buf_type = CIFS_NO_BUFFER;
1826 if (open_file->invalidHandle) {
1827 rc = cifs_reopen_file(open_file, true);
1828 if (rc != 0)
1829 break;
1830 }
1831 rc = CIFSSMBRead(xid, pTcon, open_file->netfid,
1832 cur_len, *poffset, &bytes_read,
1833 &read_data, &buf_type);
1834 pSMBr = (struct smb_com_read_rsp *)read_data;
1835 if (read_data) {
1836 char *data_offset = read_data + 4 +
1837 le16_to_cpu(pSMBr->DataOffset);
1838 if (memcpy_toiovecend(iov, data_offset,
1839 iov_offset, bytes_read))
1840 rc = -EFAULT;
1841 if (buf_type == CIFS_SMALL_BUFFER)
1842 cifs_small_buf_release(read_data);
1843 else if (buf_type == CIFS_LARGE_BUFFER)
1844 cifs_buf_release(read_data);
1845 read_data = NULL;
1846 iov_offset += bytes_read;
1847 }
1848 }
1849
1850 if (rc || (bytes_read == 0)) {
1851 if (total_read) {
1852 break;
1853 } else {
1854 FreeXid(xid);
1855 return rc;
1856 }
1857 } else {
1858 cifs_stats_bytes_read(pTcon, bytes_read);
1859 *poffset += bytes_read;
1860 }
1861 }
1862
1863 FreeXid(xid);
1864 return total_read;
1865 }
1866
1867 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1868 size_t read_size, loff_t *poffset)
1869 {
1870 struct iovec iov;
1871 iov.iov_base = read_data;
1872 iov.iov_len = read_size;
1873
1874 return cifs_iovec_read(file, &iov, 1, poffset);
1875 }
1876
1877 static ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
1878 unsigned long nr_segs, loff_t pos)
1879 {
1880 ssize_t read;
1881
1882 read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
1883 if (read > 0)
1884 iocb->ki_pos = pos;
1885
1886 return read;
1887 }
1888
1889 ssize_t cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
1890 unsigned long nr_segs, loff_t pos)
1891 {
1892 struct inode *inode;
1893
1894 inode = iocb->ki_filp->f_path.dentry->d_inode;
1895
1896 if (CIFS_I(inode)->clientCanCacheRead)
1897 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1898
1899 /*
1900 * In strict cache mode we need to read from the server all the time
1901 * if we don't have level II oplock because the server can delay mtime
1902 * change - so we can't make a decision about inode invalidating.
1903 * And we can also fail with pagereading if there are mandatory locks
1904 * on pages affected by this read but not on the region from pos to
1905 * pos+len-1.
1906 */
1907
1908 return cifs_user_readv(iocb, iov, nr_segs, pos);
1909 }
1910
1911 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1912 loff_t *poffset)
1913 {
1914 int rc = -EACCES;
1915 unsigned int bytes_read = 0;
1916 unsigned int total_read;
1917 unsigned int current_read_size;
1918 struct cifs_sb_info *cifs_sb;
1919 struct cifsTconInfo *pTcon;
1920 int xid;
1921 char *current_offset;
1922 struct cifsFileInfo *open_file;
1923 int buf_type = CIFS_NO_BUFFER;
1924
1925 xid = GetXid();
1926 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1927
1928 if (file->private_data == NULL) {
1929 rc = -EBADF;
1930 FreeXid(xid);
1931 return rc;
1932 }
1933 open_file = file->private_data;
1934 pTcon = tlink_tcon(open_file->tlink);
1935
1936 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1937 cFYI(1, "attempting read on write only file instance");
1938
1939 for (total_read = 0, current_offset = read_data;
1940 read_size > total_read;
1941 total_read += bytes_read, current_offset += bytes_read) {
1942 current_read_size = min_t(const int, read_size - total_read,
1943 cifs_sb->rsize);
1944 /* For windows me and 9x we do not want to request more
1945 than it negotiated since it will refuse the read then */
1946 if ((pTcon->ses) &&
1947 !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1948 current_read_size = min_t(const int, current_read_size,
1949 pTcon->ses->server->maxBuf - 128);
1950 }
1951 rc = -EAGAIN;
1952 while (rc == -EAGAIN) {
1953 if (open_file->invalidHandle) {
1954 rc = cifs_reopen_file(open_file, true);
1955 if (rc != 0)
1956 break;
1957 }
1958 rc = CIFSSMBRead(xid, pTcon,
1959 open_file->netfid,
1960 current_read_size, *poffset,
1961 &bytes_read, &current_offset,
1962 &buf_type);
1963 }
1964 if (rc || (bytes_read == 0)) {
1965 if (total_read) {
1966 break;
1967 } else {
1968 FreeXid(xid);
1969 return rc;
1970 }
1971 } else {
1972 cifs_stats_bytes_read(pTcon, total_read);
1973 *poffset += bytes_read;
1974 }
1975 }
1976 FreeXid(xid);
1977 return total_read;
1978 }
1979
1980 /*
1981 * If the page is mmap'ed into a process' page tables, then we need to make
1982 * sure that it doesn't change while being written back.
1983 */
1984 static int
1985 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1986 {
1987 struct page *page = vmf->page;
1988
1989 lock_page(page);
1990 return VM_FAULT_LOCKED;
1991 }
1992
1993 static struct vm_operations_struct cifs_file_vm_ops = {
1994 .fault = filemap_fault,
1995 .page_mkwrite = cifs_page_mkwrite,
1996 };
1997
1998 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
1999 {
2000 int rc, xid;
2001 struct inode *inode = file->f_path.dentry->d_inode;
2002
2003 xid = GetXid();
2004
2005 if (!CIFS_I(inode)->clientCanCacheRead)
2006 cifs_invalidate_mapping(inode);
2007
2008 rc = generic_file_mmap(file, vma);
2009 if (rc == 0)
2010 vma->vm_ops = &cifs_file_vm_ops;
2011 FreeXid(xid);
2012 return rc;
2013 }
2014
2015 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
2016 {
2017 int rc, xid;
2018
2019 xid = GetXid();
2020 rc = cifs_revalidate_file(file);
2021 if (rc) {
2022 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
2023 FreeXid(xid);
2024 return rc;
2025 }
2026 rc = generic_file_mmap(file, vma);
2027 if (rc == 0)
2028 vma->vm_ops = &cifs_file_vm_ops;
2029 FreeXid(xid);
2030 return rc;
2031 }
2032
2033
2034 static void cifs_copy_cache_pages(struct address_space *mapping,
2035 struct list_head *pages, int bytes_read, char *data)
2036 {
2037 struct page *page;
2038 char *target;
2039
2040 while (bytes_read > 0) {
2041 if (list_empty(pages))
2042 break;
2043
2044 page = list_entry(pages->prev, struct page, lru);
2045 list_del(&page->lru);
2046
2047 if (add_to_page_cache_lru(page, mapping, page->index,
2048 GFP_KERNEL)) {
2049 page_cache_release(page);
2050 cFYI(1, "Add page cache failed");
2051 data += PAGE_CACHE_SIZE;
2052 bytes_read -= PAGE_CACHE_SIZE;
2053 continue;
2054 }
2055 page_cache_release(page);
2056
2057 target = kmap_atomic(page, KM_USER0);
2058
2059 if (PAGE_CACHE_SIZE > bytes_read) {
2060 memcpy(target, data, bytes_read);
2061 /* zero the tail end of this partial page */
2062 memset(target + bytes_read, 0,
2063 PAGE_CACHE_SIZE - bytes_read);
2064 bytes_read = 0;
2065 } else {
2066 memcpy(target, data, PAGE_CACHE_SIZE);
2067 bytes_read -= PAGE_CACHE_SIZE;
2068 }
2069 kunmap_atomic(target, KM_USER0);
2070
2071 flush_dcache_page(page);
2072 SetPageUptodate(page);
2073 unlock_page(page);
2074 data += PAGE_CACHE_SIZE;
2075
2076 /* add page to FS-Cache */
2077 cifs_readpage_to_fscache(mapping->host, page);
2078 }
2079 return;
2080 }
2081
2082 static int cifs_readpages(struct file *file, struct address_space *mapping,
2083 struct list_head *page_list, unsigned num_pages)
2084 {
2085 int rc = -EACCES;
2086 int xid;
2087 loff_t offset;
2088 struct page *page;
2089 struct cifs_sb_info *cifs_sb;
2090 struct cifsTconInfo *pTcon;
2091 unsigned int bytes_read = 0;
2092 unsigned int read_size, i;
2093 char *smb_read_data = NULL;
2094 struct smb_com_read_rsp *pSMBr;
2095 struct cifsFileInfo *open_file;
2096 int buf_type = CIFS_NO_BUFFER;
2097
2098 xid = GetXid();
2099 if (file->private_data == NULL) {
2100 rc = -EBADF;
2101 FreeXid(xid);
2102 return rc;
2103 }
2104 open_file = file->private_data;
2105 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2106 pTcon = tlink_tcon(open_file->tlink);
2107
2108 /*
2109 * Reads as many pages as possible from fscache. Returns -ENOBUFS
2110 * immediately if the cookie is negative
2111 */
2112 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2113 &num_pages);
2114 if (rc == 0)
2115 goto read_complete;
2116
2117 cFYI(DBG2, "rpages: num pages %d", num_pages);
2118 for (i = 0; i < num_pages; ) {
2119 unsigned contig_pages;
2120 struct page *tmp_page;
2121 unsigned long expected_index;
2122
2123 if (list_empty(page_list))
2124 break;
2125
2126 page = list_entry(page_list->prev, struct page, lru);
2127 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2128
2129 /* count adjacent pages that we will read into */
2130 contig_pages = 0;
2131 expected_index =
2132 list_entry(page_list->prev, struct page, lru)->index;
2133 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2134 if (tmp_page->index == expected_index) {
2135 contig_pages++;
2136 expected_index++;
2137 } else
2138 break;
2139 }
2140 if (contig_pages + i > num_pages)
2141 contig_pages = num_pages - i;
2142
2143 /* for reads over a certain size could initiate async
2144 read ahead */
2145
2146 read_size = contig_pages * PAGE_CACHE_SIZE;
2147 /* Read size needs to be in multiples of one page */
2148 read_size = min_t(const unsigned int, read_size,
2149 cifs_sb->rsize & PAGE_CACHE_MASK);
2150 cFYI(DBG2, "rpages: read size 0x%x contiguous pages %d",
2151 read_size, contig_pages);
2152 rc = -EAGAIN;
2153 while (rc == -EAGAIN) {
2154 if (open_file->invalidHandle) {
2155 rc = cifs_reopen_file(open_file, true);
2156 if (rc != 0)
2157 break;
2158 }
2159
2160 rc = CIFSSMBRead(xid, pTcon,
2161 open_file->netfid,
2162 read_size, offset,
2163 &bytes_read, &smb_read_data,
2164 &buf_type);
2165 /* BB more RC checks ? */
2166 if (rc == -EAGAIN) {
2167 if (smb_read_data) {
2168 if (buf_type == CIFS_SMALL_BUFFER)
2169 cifs_small_buf_release(smb_read_data);
2170 else if (buf_type == CIFS_LARGE_BUFFER)
2171 cifs_buf_release(smb_read_data);
2172 smb_read_data = NULL;
2173 }
2174 }
2175 }
2176 if ((rc < 0) || (smb_read_data == NULL)) {
2177 cFYI(1, "Read error in readpages: %d", rc);
2178 break;
2179 } else if (bytes_read > 0) {
2180 task_io_account_read(bytes_read);
2181 pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2182 cifs_copy_cache_pages(mapping, page_list, bytes_read,
2183 smb_read_data + 4 /* RFC1001 hdr */ +
2184 le16_to_cpu(pSMBr->DataOffset));
2185
2186 i += bytes_read >> PAGE_CACHE_SHIFT;
2187 cifs_stats_bytes_read(pTcon, bytes_read);
2188 if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2189 i++; /* account for partial page */
2190
2191 /* server copy of file can have smaller size
2192 than client */
2193 /* BB do we need to verify this common case ?
2194 this case is ok - if we are at server EOF
2195 we will hit it on next read */
2196
2197 /* break; */
2198 }
2199 } else {
2200 cFYI(1, "No bytes read (%d) at offset %lld . "
2201 "Cleaning remaining pages from readahead list",
2202 bytes_read, offset);
2203 /* BB turn off caching and do new lookup on
2204 file size at server? */
2205 break;
2206 }
2207 if (smb_read_data) {
2208 if (buf_type == CIFS_SMALL_BUFFER)
2209 cifs_small_buf_release(smb_read_data);
2210 else if (buf_type == CIFS_LARGE_BUFFER)
2211 cifs_buf_release(smb_read_data);
2212 smb_read_data = NULL;
2213 }
2214 bytes_read = 0;
2215 }
2216
2217 /* need to free smb_read_data buf before exit */
2218 if (smb_read_data) {
2219 if (buf_type == CIFS_SMALL_BUFFER)
2220 cifs_small_buf_release(smb_read_data);
2221 else if (buf_type == CIFS_LARGE_BUFFER)
2222 cifs_buf_release(smb_read_data);
2223 smb_read_data = NULL;
2224 }
2225
2226 read_complete:
2227 FreeXid(xid);
2228 return rc;
2229 }
2230
2231 static int cifs_readpage_worker(struct file *file, struct page *page,
2232 loff_t *poffset)
2233 {
2234 char *read_data;
2235 int rc;
2236
2237 /* Is the page cached? */
2238 rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2239 if (rc == 0)
2240 goto read_complete;
2241
2242 page_cache_get(page);
2243 read_data = kmap(page);
2244 /* for reads over a certain size could initiate async read ahead */
2245
2246 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2247
2248 if (rc < 0)
2249 goto io_error;
2250 else
2251 cFYI(1, "Bytes read %d", rc);
2252
2253 file->f_path.dentry->d_inode->i_atime =
2254 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2255
2256 if (PAGE_CACHE_SIZE > rc)
2257 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2258
2259 flush_dcache_page(page);
2260 SetPageUptodate(page);
2261
2262 /* send this page to the cache */
2263 cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2264
2265 rc = 0;
2266
2267 io_error:
2268 kunmap(page);
2269 page_cache_release(page);
2270
2271 read_complete:
2272 return rc;
2273 }
2274
2275 static int cifs_readpage(struct file *file, struct page *page)
2276 {
2277 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2278 int rc = -EACCES;
2279 int xid;
2280
2281 xid = GetXid();
2282
2283 if (file->private_data == NULL) {
2284 rc = -EBADF;
2285 FreeXid(xid);
2286 return rc;
2287 }
2288
2289 cFYI(1, "readpage %p at offset %d 0x%x\n",
2290 page, (int)offset, (int)offset);
2291
2292 rc = cifs_readpage_worker(file, page, &offset);
2293
2294 unlock_page(page);
2295
2296 FreeXid(xid);
2297 return rc;
2298 }
2299
2300 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2301 {
2302 struct cifsFileInfo *open_file;
2303
2304 spin_lock(&cifs_file_list_lock);
2305 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2306 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2307 spin_unlock(&cifs_file_list_lock);
2308 return 1;
2309 }
2310 }
2311 spin_unlock(&cifs_file_list_lock);
2312 return 0;
2313 }
2314
2315 /* We do not want to update the file size from server for inodes
2316 open for write - to avoid races with writepage extending
2317 the file - in the future we could consider allowing
2318 refreshing the inode only on increases in the file size
2319 but this is tricky to do without racing with writebehind
2320 page caching in the current Linux kernel design */
2321 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2322 {
2323 if (!cifsInode)
2324 return true;
2325
2326 if (is_inode_writable(cifsInode)) {
2327 /* This inode is open for write at least once */
2328 struct cifs_sb_info *cifs_sb;
2329
2330 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2331 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2332 /* since no page cache to corrupt on directio
2333 we can change size safely */
2334 return true;
2335 }
2336
2337 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2338 return true;
2339
2340 return false;
2341 } else
2342 return true;
2343 }
2344
2345 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2346 loff_t pos, unsigned len, unsigned flags,
2347 struct page **pagep, void **fsdata)
2348 {
2349 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2350 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2351 loff_t page_start = pos & PAGE_MASK;
2352 loff_t i_size;
2353 struct page *page;
2354 int rc = 0;
2355
2356 cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2357
2358 page = grab_cache_page_write_begin(mapping, index, flags);
2359 if (!page) {
2360 rc = -ENOMEM;
2361 goto out;
2362 }
2363
2364 if (PageUptodate(page))
2365 goto out;
2366
2367 /*
2368 * If we write a full page it will be up to date, no need to read from
2369 * the server. If the write is short, we'll end up doing a sync write
2370 * instead.
2371 */
2372 if (len == PAGE_CACHE_SIZE)
2373 goto out;
2374
2375 /*
2376 * optimize away the read when we have an oplock, and we're not
2377 * expecting to use any of the data we'd be reading in. That
2378 * is, when the page lies beyond the EOF, or straddles the EOF
2379 * and the write will cover all of the existing data.
2380 */
2381 if (CIFS_I(mapping->host)->clientCanCacheRead) {
2382 i_size = i_size_read(mapping->host);
2383 if (page_start >= i_size ||
2384 (offset == 0 && (pos + len) >= i_size)) {
2385 zero_user_segments(page, 0, offset,
2386 offset + len,
2387 PAGE_CACHE_SIZE);
2388 /*
2389 * PageChecked means that the parts of the page
2390 * to which we're not writing are considered up
2391 * to date. Once the data is copied to the
2392 * page, it can be set uptodate.
2393 */
2394 SetPageChecked(page);
2395 goto out;
2396 }
2397 }
2398
2399 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2400 /*
2401 * might as well read a page, it is fast enough. If we get
2402 * an error, we don't need to return it. cifs_write_end will
2403 * do a sync write instead since PG_uptodate isn't set.
2404 */
2405 cifs_readpage_worker(file, page, &page_start);
2406 } else {
2407 /* we could try using another file handle if there is one -
2408 but how would we lock it to prevent close of that handle
2409 racing with this read? In any case
2410 this will be written out by write_end so is fine */
2411 }
2412 out:
2413 *pagep = page;
2414 return rc;
2415 }
2416
2417 static int cifs_release_page(struct page *page, gfp_t gfp)
2418 {
2419 if (PagePrivate(page))
2420 return 0;
2421
2422 return cifs_fscache_release_page(page, gfp);
2423 }
2424
2425 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2426 {
2427 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2428
2429 if (offset == 0)
2430 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2431 }
2432
2433 static int cifs_launder_page(struct page *page)
2434 {
2435 int rc = 0;
2436 loff_t range_start = page_offset(page);
2437 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
2438 struct writeback_control wbc = {
2439 .sync_mode = WB_SYNC_ALL,
2440 .nr_to_write = 0,
2441 .range_start = range_start,
2442 .range_end = range_end,
2443 };
2444
2445 cFYI(1, "Launder page: %p", page);
2446
2447 if (clear_page_dirty_for_io(page))
2448 rc = cifs_writepage_locked(page, &wbc);
2449
2450 cifs_fscache_invalidate_page(page, page->mapping->host);
2451 return rc;
2452 }
2453
2454 void cifs_oplock_break(struct work_struct *work)
2455 {
2456 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2457 oplock_break);
2458 struct inode *inode = cfile->dentry->d_inode;
2459 struct cifsInodeInfo *cinode = CIFS_I(inode);
2460 int rc = 0;
2461
2462 if (inode && S_ISREG(inode->i_mode)) {
2463 if (cinode->clientCanCacheRead)
2464 break_lease(inode, O_RDONLY);
2465 else
2466 break_lease(inode, O_WRONLY);
2467 rc = filemap_fdatawrite(inode->i_mapping);
2468 if (cinode->clientCanCacheRead == 0) {
2469 rc = filemap_fdatawait(inode->i_mapping);
2470 mapping_set_error(inode->i_mapping, rc);
2471 invalidate_remote_inode(inode);
2472 }
2473 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2474 }
2475
2476 /*
2477 * releasing stale oplock after recent reconnect of smb session using
2478 * a now incorrect file handle is not a data integrity issue but do
2479 * not bother sending an oplock release if session to server still is
2480 * disconnected since oplock already released by the server
2481 */
2482 if (!cfile->oplock_break_cancelled) {
2483 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2484 0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false,
2485 cinode->clientCanCacheRead ? 1 : 0);
2486 cFYI(1, "Oplock release rc = %d", rc);
2487 }
2488
2489 /*
2490 * We might have kicked in before is_valid_oplock_break()
2491 * finished grabbing reference for us. Make sure it's done by
2492 * waiting for cifs_file_list_lock.
2493 */
2494 spin_lock(&cifs_file_list_lock);
2495 spin_unlock(&cifs_file_list_lock);
2496
2497 cifs_oplock_break_put(cfile);
2498 }
2499
2500 /* must be called while holding cifs_file_list_lock */
2501 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2502 {
2503 cifs_sb_active(cfile->dentry->d_sb);
2504 cifsFileInfo_get(cfile);
2505 }
2506
2507 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2508 {
2509 struct super_block *sb = cfile->dentry->d_sb;
2510
2511 cifsFileInfo_put(cfile);
2512 cifs_sb_deactive(sb);
2513 }
2514
2515 const struct address_space_operations cifs_addr_ops = {
2516 .readpage = cifs_readpage,
2517 .readpages = cifs_readpages,
2518 .writepage = cifs_writepage,
2519 .writepages = cifs_writepages,
2520 .write_begin = cifs_write_begin,
2521 .write_end = cifs_write_end,
2522 .set_page_dirty = __set_page_dirty_nobuffers,
2523 .releasepage = cifs_release_page,
2524 .invalidatepage = cifs_invalidate_page,
2525 .launder_page = cifs_launder_page,
2526 };
2527
2528 /*
2529 * cifs_readpages requires the server to support a buffer large enough to
2530 * contain the header plus one complete page of data. Otherwise, we need
2531 * to leave cifs_readpages out of the address space operations.
2532 */
2533 const struct address_space_operations cifs_addr_ops_smallbuf = {
2534 .readpage = cifs_readpage,
2535 .writepage = cifs_writepage,
2536 .writepages = cifs_writepages,
2537 .write_begin = cifs_write_begin,
2538 .write_end = cifs_write_end,
2539 .set_page_dirty = __set_page_dirty_nobuffers,
2540 .releasepage = cifs_release_page,
2541 .invalidatepage = cifs_invalidate_page,
2542 .launder_page = cifs_launder_page,
2543 };
Linux kernel - Deliverables
This page took 0.122036 seconds and 4 git commands to generate.