]> git.ipfire.org Git - thirdparty/kernel/linux.git/blob - fs/cifs/file.c
projects / thirdparty / kernel / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
CIFS: Separate page sending from writepages
[thirdparty/kernel/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 <linux/swap.h>
36 #include <asm/div64.h>
37 #include "cifsfs.h"
38 #include "cifspdu.h"
39 #include "cifsglob.h"
40 #include "cifsproto.h"
41 #include "cifs_unicode.h"
42 #include "cifs_debug.h"
43 #include "cifs_fs_sb.h"
44 #include "fscache.h"
45
46
47 static inline int cifs_convert_flags(unsigned int flags)
48 {
49 if ((flags & O_ACCMODE) == O_RDONLY)
50 return GENERIC_READ;
51 else if ((flags & O_ACCMODE) == O_WRONLY)
52 return GENERIC_WRITE;
53 else if ((flags & O_ACCMODE) == O_RDWR) {
54 /* GENERIC_ALL is too much permission to request
55 can cause unnecessary access denied on create */
56 /* return GENERIC_ALL; */
57 return (GENERIC_READ | GENERIC_WRITE);
58 }
59
60 return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
61 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
62 FILE_READ_DATA);
63 }
64
65 static u32 cifs_posix_convert_flags(unsigned int flags)
66 {
67 u32 posix_flags = 0;
68
69 if ((flags & O_ACCMODE) == O_RDONLY)
70 posix_flags = SMB_O_RDONLY;
71 else if ((flags & O_ACCMODE) == O_WRONLY)
72 posix_flags = SMB_O_WRONLY;
73 else if ((flags & O_ACCMODE) == O_RDWR)
74 posix_flags = SMB_O_RDWR;
75
76 if (flags & O_CREAT) {
77 posix_flags |= SMB_O_CREAT;
78 if (flags & O_EXCL)
79 posix_flags |= SMB_O_EXCL;
80 } else if (flags & O_EXCL)
81 cifs_dbg(FYI, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
82 current->comm, current->tgid);
83
84 if (flags & O_TRUNC)
85 posix_flags |= SMB_O_TRUNC;
86 /* be safe and imply O_SYNC for O_DSYNC */
87 if (flags & O_DSYNC)
88 posix_flags |= SMB_O_SYNC;
89 if (flags & O_DIRECTORY)
90 posix_flags |= SMB_O_DIRECTORY;
91 if (flags & O_NOFOLLOW)
92 posix_flags |= SMB_O_NOFOLLOW;
93 if (flags & O_DIRECT)
94 posix_flags |= SMB_O_DIRECT;
95
96 return posix_flags;
97 }
98
99 static inline int cifs_get_disposition(unsigned int flags)
100 {
101 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
102 return FILE_CREATE;
103 else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
104 return FILE_OVERWRITE_IF;
105 else if ((flags & O_CREAT) == O_CREAT)
106 return FILE_OPEN_IF;
107 else if ((flags & O_TRUNC) == O_TRUNC)
108 return FILE_OVERWRITE;
109 else
110 return FILE_OPEN;
111 }
112
113 int cifs_posix_open(char *full_path, struct inode **pinode,
114 struct super_block *sb, int mode, unsigned int f_flags,
115 __u32 *poplock, __u16 *pnetfid, unsigned int xid)
116 {
117 int rc;
118 FILE_UNIX_BASIC_INFO *presp_data;
119 __u32 posix_flags = 0;
120 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
121 struct cifs_fattr fattr;
122 struct tcon_link *tlink;
123 struct cifs_tcon *tcon;
124
125 cifs_dbg(FYI, "posix open %s\n", full_path);
126
127 presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
128 if (presp_data == NULL)
129 return -ENOMEM;
130
131 tlink = cifs_sb_tlink(cifs_sb);
132 if (IS_ERR(tlink)) {
133 rc = PTR_ERR(tlink);
134 goto posix_open_ret;
135 }
136
137 tcon = tlink_tcon(tlink);
138 mode &= ~current_umask();
139
140 posix_flags = cifs_posix_convert_flags(f_flags);
141 rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
142 poplock, full_path, cifs_sb->local_nls,
143 cifs_sb->mnt_cifs_flags &
144 CIFS_MOUNT_MAP_SPECIAL_CHR);
145 cifs_put_tlink(tlink);
146
147 if (rc)
148 goto posix_open_ret;
149
150 if (presp_data->Type == cpu_to_le32(-1))
151 goto posix_open_ret; /* open ok, caller does qpathinfo */
152
153 if (!pinode)
154 goto posix_open_ret; /* caller does not need info */
155
156 cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
157
158 /* get new inode and set it up */
159 if (*pinode == NULL) {
160 cifs_fill_uniqueid(sb, &fattr);
161 *pinode = cifs_iget(sb, &fattr);
162 if (!*pinode) {
163 rc = -ENOMEM;
164 goto posix_open_ret;
165 }
166 } else {
167 cifs_fattr_to_inode(*pinode, &fattr);
168 }
169
170 posix_open_ret:
171 kfree(presp_data);
172 return rc;
173 }
174
175 static int
176 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
177 struct cifs_tcon *tcon, unsigned int f_flags, __u32 *oplock,
178 struct cifs_fid *fid, unsigned int xid)
179 {
180 int rc;
181 int desired_access;
182 int disposition;
183 int create_options = CREATE_NOT_DIR;
184 FILE_ALL_INFO *buf;
185 struct TCP_Server_Info *server = tcon->ses->server;
186 struct cifs_open_parms oparms;
187
188 if (!server->ops->open)
189 return -ENOSYS;
190
191 desired_access = cifs_convert_flags(f_flags);
192
193 /*********************************************************************
194 * open flag mapping table:
195 *
196 * POSIX Flag CIFS Disposition
197 * ---------- ----------------
198 * O_CREAT FILE_OPEN_IF
199 * O_CREAT | O_EXCL FILE_CREATE
200 * O_CREAT | O_TRUNC FILE_OVERWRITE_IF
201 * O_TRUNC FILE_OVERWRITE
202 * none of the above FILE_OPEN
203 *
204 * Note that there is not a direct match between disposition
205 * FILE_SUPERSEDE (ie create whether or not file exists although
206 * O_CREAT | O_TRUNC is similar but truncates the existing
207 * file rather than creating a new file as FILE_SUPERSEDE does
208 * (which uses the attributes / metadata passed in on open call)
209 *?
210 *? O_SYNC is a reasonable match to CIFS writethrough flag
211 *? and the read write flags match reasonably. O_LARGEFILE
212 *? is irrelevant because largefile support is always used
213 *? by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
214 * O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
215 *********************************************************************/
216
217 disposition = cifs_get_disposition(f_flags);
218
219 /* BB pass O_SYNC flag through on file attributes .. BB */
220
221 buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
222 if (!buf)
223 return -ENOMEM;
224
225 if (backup_cred(cifs_sb))
226 create_options |= CREATE_OPEN_BACKUP_INTENT;
227
228 oparms.tcon = tcon;
229 oparms.cifs_sb = cifs_sb;
230 oparms.desired_access = desired_access;
231 oparms.create_options = create_options;
232 oparms.disposition = disposition;
233 oparms.path = full_path;
234 oparms.fid = fid;
235 oparms.reconnect = false;
236
237 rc = server->ops->open(xid, &oparms, oplock, buf);
238
239 if (rc)
240 goto out;
241
242 if (tcon->unix_ext)
243 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
244 xid);
245 else
246 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
247 xid, fid);
248
249 out:
250 kfree(buf);
251 return rc;
252 }
253
254 static bool
255 cifs_has_mand_locks(struct cifsInodeInfo *cinode)
256 {
257 struct cifs_fid_locks *cur;
258 bool has_locks = false;
259
260 down_read(&cinode->lock_sem);
261 list_for_each_entry(cur, &cinode->llist, llist) {
262 if (!list_empty(&cur->locks)) {
263 has_locks = true;
264 break;
265 }
266 }
267 up_read(&cinode->lock_sem);
268 return has_locks;
269 }
270
271 struct cifsFileInfo *
272 cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
273 struct tcon_link *tlink, __u32 oplock)
274 {
275 struct dentry *dentry = file->f_path.dentry;
276 struct inode *inode = dentry->d_inode;
277 struct cifsInodeInfo *cinode = CIFS_I(inode);
278 struct cifsFileInfo *cfile;
279 struct cifs_fid_locks *fdlocks;
280 struct cifs_tcon *tcon = tlink_tcon(tlink);
281 struct TCP_Server_Info *server = tcon->ses->server;
282
283 cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
284 if (cfile == NULL)
285 return cfile;
286
287 fdlocks = kzalloc(sizeof(struct cifs_fid_locks), GFP_KERNEL);
288 if (!fdlocks) {
289 kfree(cfile);
290 return NULL;
291 }
292
293 INIT_LIST_HEAD(&fdlocks->locks);
294 fdlocks->cfile = cfile;
295 cfile->llist = fdlocks;
296 down_write(&cinode->lock_sem);
297 list_add(&fdlocks->llist, &cinode->llist);
298 up_write(&cinode->lock_sem);
299
300 cfile->count = 1;
301 cfile->pid = current->tgid;
302 cfile->uid = current_fsuid();
303 cfile->dentry = dget(dentry);
304 cfile->f_flags = file->f_flags;
305 cfile->invalidHandle = false;
306 cfile->tlink = cifs_get_tlink(tlink);
307 INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
308 mutex_init(&cfile->fh_mutex);
309
310 cifs_sb_active(inode->i_sb);
311
312 /*
313 * If the server returned a read oplock and we have mandatory brlocks,
314 * set oplock level to None.
315 */
316 if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
317 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
318 oplock = 0;
319 }
320
321 spin_lock(&cifs_file_list_lock);
322 if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
323 oplock = fid->pending_open->oplock;
324 list_del(&fid->pending_open->olist);
325
326 fid->purge_cache = false;
327 server->ops->set_fid(cfile, fid, oplock);
328
329 list_add(&cfile->tlist, &tcon->openFileList);
330 /* if readable file instance put first in list*/
331 if (file->f_mode & FMODE_READ)
332 list_add(&cfile->flist, &cinode->openFileList);
333 else
334 list_add_tail(&cfile->flist, &cinode->openFileList);
335 spin_unlock(&cifs_file_list_lock);
336
337 if (fid->purge_cache)
338 cifs_zap_mapping(inode);
339
340 file->private_data = cfile;
341 return cfile;
342 }
343
344 struct cifsFileInfo *
345 cifsFileInfo_get(struct cifsFileInfo *cifs_file)
346 {
347 spin_lock(&cifs_file_list_lock);
348 cifsFileInfo_get_locked(cifs_file);
349 spin_unlock(&cifs_file_list_lock);
350 return cifs_file;
351 }
352
353 /*
354 * Release a reference on the file private data. This may involve closing
355 * the filehandle out on the server. Must be called without holding
356 * cifs_file_list_lock.
357 */
358 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
359 {
360 struct inode *inode = cifs_file->dentry->d_inode;
361 struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
362 struct TCP_Server_Info *server = tcon->ses->server;
363 struct cifsInodeInfo *cifsi = CIFS_I(inode);
364 struct super_block *sb = inode->i_sb;
365 struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
366 struct cifsLockInfo *li, *tmp;
367 struct cifs_fid fid;
368 struct cifs_pending_open open;
369
370 spin_lock(&cifs_file_list_lock);
371 if (--cifs_file->count > 0) {
372 spin_unlock(&cifs_file_list_lock);
373 return;
374 }
375
376 if (server->ops->get_lease_key)
377 server->ops->get_lease_key(inode, &fid);
378
379 /* store open in pending opens to make sure we don't miss lease break */
380 cifs_add_pending_open_locked(&fid, cifs_file->tlink, &open);
381
382 /* remove it from the lists */
383 list_del(&cifs_file->flist);
384 list_del(&cifs_file->tlist);
385
386 if (list_empty(&cifsi->openFileList)) {
387 cifs_dbg(FYI, "closing last open instance for inode %p\n",
388 cifs_file->dentry->d_inode);
389 /*
390 * In strict cache mode we need invalidate mapping on the last
391 * close because it may cause a error when we open this file
392 * again and get at least level II oplock.
393 */
394 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
395 set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
396 cifs_set_oplock_level(cifsi, 0);
397 }
398 spin_unlock(&cifs_file_list_lock);
399
400 cancel_work_sync(&cifs_file->oplock_break);
401
402 if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
403 struct TCP_Server_Info *server = tcon->ses->server;
404 unsigned int xid;
405
406 xid = get_xid();
407 if (server->ops->close)
408 server->ops->close(xid, tcon, &cifs_file->fid);
409 _free_xid(xid);
410 }
411
412 cifs_del_pending_open(&open);
413
414 /*
415 * Delete any outstanding lock records. We'll lose them when the file
416 * is closed anyway.
417 */
418 down_write(&cifsi->lock_sem);
419 list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
420 list_del(&li->llist);
421 cifs_del_lock_waiters(li);
422 kfree(li);
423 }
424 list_del(&cifs_file->llist->llist);
425 kfree(cifs_file->llist);
426 up_write(&cifsi->lock_sem);
427
428 cifs_put_tlink(cifs_file->tlink);
429 dput(cifs_file->dentry);
430 cifs_sb_deactive(sb);
431 kfree(cifs_file);
432 }
433
434 int cifs_open(struct inode *inode, struct file *file)
435
436 {
437 int rc = -EACCES;
438 unsigned int xid;
439 __u32 oplock;
440 struct cifs_sb_info *cifs_sb;
441 struct TCP_Server_Info *server;
442 struct cifs_tcon *tcon;
443 struct tcon_link *tlink;
444 struct cifsFileInfo *cfile = NULL;
445 char *full_path = NULL;
446 bool posix_open_ok = false;
447 struct cifs_fid fid;
448 struct cifs_pending_open open;
449
450 xid = get_xid();
451
452 cifs_sb = CIFS_SB(inode->i_sb);
453 tlink = cifs_sb_tlink(cifs_sb);
454 if (IS_ERR(tlink)) {
455 free_xid(xid);
456 return PTR_ERR(tlink);
457 }
458 tcon = tlink_tcon(tlink);
459 server = tcon->ses->server;
460
461 full_path = build_path_from_dentry(file->f_path.dentry);
462 if (full_path == NULL) {
463 rc = -ENOMEM;
464 goto out;
465 }
466
467 cifs_dbg(FYI, "inode = 0x%p file flags are 0x%x for %s\n",
468 inode, file->f_flags, full_path);
469
470 if (server->oplocks)
471 oplock = REQ_OPLOCK;
472 else
473 oplock = 0;
474
475 if (!tcon->broken_posix_open && tcon->unix_ext &&
476 cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
477 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
478 /* can not refresh inode info since size could be stale */
479 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
480 cifs_sb->mnt_file_mode /* ignored */,
481 file->f_flags, &oplock, &fid.netfid, xid);
482 if (rc == 0) {
483 cifs_dbg(FYI, "posix open succeeded\n");
484 posix_open_ok = true;
485 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
486 if (tcon->ses->serverNOS)
487 cifs_dbg(VFS, "server %s of type %s returned unexpected error on SMB posix open, disabling posix open support. Check if server update available.\n",
488 tcon->ses->serverName,
489 tcon->ses->serverNOS);
490 tcon->broken_posix_open = true;
491 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
492 (rc != -EOPNOTSUPP)) /* path not found or net err */
493 goto out;
494 /*
495 * Else fallthrough to retry open the old way on network i/o
496 * or DFS errors.
497 */
498 }
499
500 if (server->ops->get_lease_key)
501 server->ops->get_lease_key(inode, &fid);
502
503 cifs_add_pending_open(&fid, tlink, &open);
504
505 if (!posix_open_ok) {
506 if (server->ops->get_lease_key)
507 server->ops->get_lease_key(inode, &fid);
508
509 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
510 file->f_flags, &oplock, &fid, xid);
511 if (rc) {
512 cifs_del_pending_open(&open);
513 goto out;
514 }
515 }
516
517 cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
518 if (cfile == NULL) {
519 if (server->ops->close)
520 server->ops->close(xid, tcon, &fid);
521 cifs_del_pending_open(&open);
522 rc = -ENOMEM;
523 goto out;
524 }
525
526 cifs_fscache_set_inode_cookie(inode, file);
527
528 if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
529 /*
530 * Time to set mode which we can not set earlier due to
531 * problems creating new read-only files.
532 */
533 struct cifs_unix_set_info_args args = {
534 .mode = inode->i_mode,
535 .uid = INVALID_UID, /* no change */
536 .gid = INVALID_GID, /* no change */
537 .ctime = NO_CHANGE_64,
538 .atime = NO_CHANGE_64,
539 .mtime = NO_CHANGE_64,
540 .device = 0,
541 };
542 CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
543 cfile->pid);
544 }
545
546 out:
547 kfree(full_path);
548 free_xid(xid);
549 cifs_put_tlink(tlink);
550 return rc;
551 }
552
553 static int cifs_push_posix_locks(struct cifsFileInfo *cfile);
554
555 /*
556 * Try to reacquire byte range locks that were released when session
557 * to server was lost.
558 */
559 static int
560 cifs_relock_file(struct cifsFileInfo *cfile)
561 {
562 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
563 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
564 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
565 int rc = 0;
566
567 down_read(&cinode->lock_sem);
568 if (cinode->can_cache_brlcks) {
569 /* can cache locks - no need to relock */
570 up_read(&cinode->lock_sem);
571 return rc;
572 }
573
574 if (cap_unix(tcon->ses) &&
575 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
576 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
577 rc = cifs_push_posix_locks(cfile);
578 else
579 rc = tcon->ses->server->ops->push_mand_locks(cfile);
580
581 up_read(&cinode->lock_sem);
582 return rc;
583 }
584
585 static int
586 cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
587 {
588 int rc = -EACCES;
589 unsigned int xid;
590 __u32 oplock;
591 struct cifs_sb_info *cifs_sb;
592 struct cifs_tcon *tcon;
593 struct TCP_Server_Info *server;
594 struct cifsInodeInfo *cinode;
595 struct inode *inode;
596 char *full_path = NULL;
597 int desired_access;
598 int disposition = FILE_OPEN;
599 int create_options = CREATE_NOT_DIR;
600 struct cifs_open_parms oparms;
601
602 xid = get_xid();
603 mutex_lock(&cfile->fh_mutex);
604 if (!cfile->invalidHandle) {
605 mutex_unlock(&cfile->fh_mutex);
606 rc = 0;
607 free_xid(xid);
608 return rc;
609 }
610
611 inode = cfile->dentry->d_inode;
612 cifs_sb = CIFS_SB(inode->i_sb);
613 tcon = tlink_tcon(cfile->tlink);
614 server = tcon->ses->server;
615
616 /*
617 * Can not grab rename sem here because various ops, including those
618 * that already have the rename sem can end up causing writepage to get
619 * called and if the server was down that means we end up here, and we
620 * can never tell if the caller already has the rename_sem.
621 */
622 full_path = build_path_from_dentry(cfile->dentry);
623 if (full_path == NULL) {
624 rc = -ENOMEM;
625 mutex_unlock(&cfile->fh_mutex);
626 free_xid(xid);
627 return rc;
628 }
629
630 cifs_dbg(FYI, "inode = 0x%p file flags 0x%x for %s\n",
631 inode, cfile->f_flags, full_path);
632
633 if (tcon->ses->server->oplocks)
634 oplock = REQ_OPLOCK;
635 else
636 oplock = 0;
637
638 if (tcon->unix_ext && cap_unix(tcon->ses) &&
639 (CIFS_UNIX_POSIX_PATH_OPS_CAP &
640 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
641 /*
642 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
643 * original open. Must mask them off for a reopen.
644 */
645 unsigned int oflags = cfile->f_flags &
646 ~(O_CREAT | O_EXCL | O_TRUNC);
647
648 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
649 cifs_sb->mnt_file_mode /* ignored */,
650 oflags, &oplock, &cfile->fid.netfid, xid);
651 if (rc == 0) {
652 cifs_dbg(FYI, "posix reopen succeeded\n");
653 oparms.reconnect = true;
654 goto reopen_success;
655 }
656 /*
657 * fallthrough to retry open the old way on errors, especially
658 * in the reconnect path it is important to retry hard
659 */
660 }
661
662 desired_access = cifs_convert_flags(cfile->f_flags);
663
664 if (backup_cred(cifs_sb))
665 create_options |= CREATE_OPEN_BACKUP_INTENT;
666
667 if (server->ops->get_lease_key)
668 server->ops->get_lease_key(inode, &cfile->fid);
669
670 oparms.tcon = tcon;
671 oparms.cifs_sb = cifs_sb;
672 oparms.desired_access = desired_access;
673 oparms.create_options = create_options;
674 oparms.disposition = disposition;
675 oparms.path = full_path;
676 oparms.fid = &cfile->fid;
677 oparms.reconnect = true;
678
679 /*
680 * Can not refresh inode by passing in file_info buf to be returned by
681 * ops->open and then calling get_inode_info with returned buf since
682 * file might have write behind data that needs to be flushed and server
683 * version of file size can be stale. If we knew for sure that inode was
684 * not dirty locally we could do this.
685 */
686 rc = server->ops->open(xid, &oparms, &oplock, NULL);
687 if (rc == -ENOENT && oparms.reconnect == false) {
688 /* durable handle timeout is expired - open the file again */
689 rc = server->ops->open(xid, &oparms, &oplock, NULL);
690 /* indicate that we need to relock the file */
691 oparms.reconnect = true;
692 }
693
694 if (rc) {
695 mutex_unlock(&cfile->fh_mutex);
696 cifs_dbg(FYI, "cifs_reopen returned 0x%x\n", rc);
697 cifs_dbg(FYI, "oplock: %d\n", oplock);
698 goto reopen_error_exit;
699 }
700
701 reopen_success:
702 cfile->invalidHandle = false;
703 mutex_unlock(&cfile->fh_mutex);
704 cinode = CIFS_I(inode);
705
706 if (can_flush) {
707 rc = filemap_write_and_wait(inode->i_mapping);
708 mapping_set_error(inode->i_mapping, rc);
709
710 if (tcon->unix_ext)
711 rc = cifs_get_inode_info_unix(&inode, full_path,
712 inode->i_sb, xid);
713 else
714 rc = cifs_get_inode_info(&inode, full_path, NULL,
715 inode->i_sb, xid, NULL);
716 }
717 /*
718 * Else we are writing out data to server already and could deadlock if
719 * we tried to flush data, and since we do not know if we have data that
720 * would invalidate the current end of file on the server we can not go
721 * to the server to get the new inode info.
722 */
723
724 server->ops->set_fid(cfile, &cfile->fid, oplock);
725 if (oparms.reconnect)
726 cifs_relock_file(cfile);
727
728 reopen_error_exit:
729 kfree(full_path);
730 free_xid(xid);
731 return rc;
732 }
733
734 int cifs_close(struct inode *inode, struct file *file)
735 {
736 if (file->private_data != NULL) {
737 cifsFileInfo_put(file->private_data);
738 file->private_data = NULL;
739 }
740
741 /* return code from the ->release op is always ignored */
742 return 0;
743 }
744
745 int cifs_closedir(struct inode *inode, struct file *file)
746 {
747 int rc = 0;
748 unsigned int xid;
749 struct cifsFileInfo *cfile = file->private_data;
750 struct cifs_tcon *tcon;
751 struct TCP_Server_Info *server;
752 char *buf;
753
754 cifs_dbg(FYI, "Closedir inode = 0x%p\n", inode);
755
756 if (cfile == NULL)
757 return rc;
758
759 xid = get_xid();
760 tcon = tlink_tcon(cfile->tlink);
761 server = tcon->ses->server;
762
763 cifs_dbg(FYI, "Freeing private data in close dir\n");
764 spin_lock(&cifs_file_list_lock);
765 if (!cfile->srch_inf.endOfSearch && !cfile->invalidHandle) {
766 cfile->invalidHandle = true;
767 spin_unlock(&cifs_file_list_lock);
768 if (server->ops->close_dir)
769 rc = server->ops->close_dir(xid, tcon, &cfile->fid);
770 else
771 rc = -ENOSYS;
772 cifs_dbg(FYI, "Closing uncompleted readdir with rc %d\n", rc);
773 /* not much we can do if it fails anyway, ignore rc */
774 rc = 0;
775 } else
776 spin_unlock(&cifs_file_list_lock);
777
778 buf = cfile->srch_inf.ntwrk_buf_start;
779 if (buf) {
780 cifs_dbg(FYI, "closedir free smb buf in srch struct\n");
781 cfile->srch_inf.ntwrk_buf_start = NULL;
782 if (cfile->srch_inf.smallBuf)
783 cifs_small_buf_release(buf);
784 else
785 cifs_buf_release(buf);
786 }
787
788 cifs_put_tlink(cfile->tlink);
789 kfree(file->private_data);
790 file->private_data = NULL;
791 /* BB can we lock the filestruct while this is going on? */
792 free_xid(xid);
793 return rc;
794 }
795
796 static struct cifsLockInfo *
797 cifs_lock_init(__u64 offset, __u64 length, __u8 type)
798 {
799 struct cifsLockInfo *lock =
800 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
801 if (!lock)
802 return lock;
803 lock->offset = offset;
804 lock->length = length;
805 lock->type = type;
806 lock->pid = current->tgid;
807 INIT_LIST_HEAD(&lock->blist);
808 init_waitqueue_head(&lock->block_q);
809 return lock;
810 }
811
812 void
813 cifs_del_lock_waiters(struct cifsLockInfo *lock)
814 {
815 struct cifsLockInfo *li, *tmp;
816 list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
817 list_del_init(&li->blist);
818 wake_up(&li->block_q);
819 }
820 }
821
822 #define CIFS_LOCK_OP 0
823 #define CIFS_READ_OP 1
824 #define CIFS_WRITE_OP 2
825
826 /* @rw_check : 0 - no op, 1 - read, 2 - write */
827 static bool
828 cifs_find_fid_lock_conflict(struct cifs_fid_locks *fdlocks, __u64 offset,
829 __u64 length, __u8 type, struct cifsFileInfo *cfile,
830 struct cifsLockInfo **conf_lock, int rw_check)
831 {
832 struct cifsLockInfo *li;
833 struct cifsFileInfo *cur_cfile = fdlocks->cfile;
834 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
835
836 list_for_each_entry(li, &fdlocks->locks, llist) {
837 if (offset + length <= li->offset ||
838 offset >= li->offset + li->length)
839 continue;
840 if (rw_check != CIFS_LOCK_OP && current->tgid == li->pid &&
841 server->ops->compare_fids(cfile, cur_cfile)) {
842 /* shared lock prevents write op through the same fid */
843 if (!(li->type & server->vals->shared_lock_type) ||
844 rw_check != CIFS_WRITE_OP)
845 continue;
846 }
847 if ((type & server->vals->shared_lock_type) &&
848 ((server->ops->compare_fids(cfile, cur_cfile) &&
849 current->tgid == li->pid) || type == li->type))
850 continue;
851 if (conf_lock)
852 *conf_lock = li;
853 return true;
854 }
855 return false;
856 }
857
858 bool
859 cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
860 __u8 type, struct cifsLockInfo **conf_lock,
861 int rw_check)
862 {
863 bool rc = false;
864 struct cifs_fid_locks *cur;
865 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
866
867 list_for_each_entry(cur, &cinode->llist, llist) {
868 rc = cifs_find_fid_lock_conflict(cur, offset, length, type,
869 cfile, conf_lock, rw_check);
870 if (rc)
871 break;
872 }
873
874 return rc;
875 }
876
877 /*
878 * Check if there is another lock that prevents us to set the lock (mandatory
879 * style). If such a lock exists, update the flock structure with its
880 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
881 * or leave it the same if we can't. Returns 0 if we don't need to request to
882 * the server or 1 otherwise.
883 */
884 static int
885 cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
886 __u8 type, struct file_lock *flock)
887 {
888 int rc = 0;
889 struct cifsLockInfo *conf_lock;
890 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
891 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
892 bool exist;
893
894 down_read(&cinode->lock_sem);
895
896 exist = cifs_find_lock_conflict(cfile, offset, length, type,
897 &conf_lock, CIFS_LOCK_OP);
898 if (exist) {
899 flock->fl_start = conf_lock->offset;
900 flock->fl_end = conf_lock->offset + conf_lock->length - 1;
901 flock->fl_pid = conf_lock->pid;
902 if (conf_lock->type & server->vals->shared_lock_type)
903 flock->fl_type = F_RDLCK;
904 else
905 flock->fl_type = F_WRLCK;
906 } else if (!cinode->can_cache_brlcks)
907 rc = 1;
908 else
909 flock->fl_type = F_UNLCK;
910
911 up_read(&cinode->lock_sem);
912 return rc;
913 }
914
915 static void
916 cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
917 {
918 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
919 down_write(&cinode->lock_sem);
920 list_add_tail(&lock->llist, &cfile->llist->locks);
921 up_write(&cinode->lock_sem);
922 }
923
924 /*
925 * Set the byte-range lock (mandatory style). Returns:
926 * 1) 0, if we set the lock and don't need to request to the server;
927 * 2) 1, if no locks prevent us but we need to request to the server;
928 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
929 */
930 static int
931 cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
932 bool wait)
933 {
934 struct cifsLockInfo *conf_lock;
935 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
936 bool exist;
937 int rc = 0;
938
939 try_again:
940 exist = false;
941 down_write(&cinode->lock_sem);
942
943 exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
944 lock->type, &conf_lock, CIFS_LOCK_OP);
945 if (!exist && cinode->can_cache_brlcks) {
946 list_add_tail(&lock->llist, &cfile->llist->locks);
947 up_write(&cinode->lock_sem);
948 return rc;
949 }
950
951 if (!exist)
952 rc = 1;
953 else if (!wait)
954 rc = -EACCES;
955 else {
956 list_add_tail(&lock->blist, &conf_lock->blist);
957 up_write(&cinode->lock_sem);
958 rc = wait_event_interruptible(lock->block_q,
959 (lock->blist.prev == &lock->blist) &&
960 (lock->blist.next == &lock->blist));
961 if (!rc)
962 goto try_again;
963 down_write(&cinode->lock_sem);
964 list_del_init(&lock->blist);
965 }
966
967 up_write(&cinode->lock_sem);
968 return rc;
969 }
970
971 /*
972 * Check if there is another lock that prevents us to set the lock (posix
973 * style). If such a lock exists, update the flock structure with its
974 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
975 * or leave it the same if we can't. Returns 0 if we don't need to request to
976 * the server or 1 otherwise.
977 */
978 static int
979 cifs_posix_lock_test(struct file *file, struct file_lock *flock)
980 {
981 int rc = 0;
982 struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
983 unsigned char saved_type = flock->fl_type;
984
985 if ((flock->fl_flags & FL_POSIX) == 0)
986 return 1;
987
988 down_read(&cinode->lock_sem);
989 posix_test_lock(file, flock);
990
991 if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
992 flock->fl_type = saved_type;
993 rc = 1;
994 }
995
996 up_read(&cinode->lock_sem);
997 return rc;
998 }
999
1000 /*
1001 * Set the byte-range lock (posix style). Returns:
1002 * 1) 0, if we set the lock and don't need to request to the server;
1003 * 2) 1, if we need to request to the server;
1004 * 3) <0, if the error occurs while setting the lock.
1005 */
1006 static int
1007 cifs_posix_lock_set(struct file *file, struct file_lock *flock)
1008 {
1009 struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1010 int rc = 1;
1011
1012 if ((flock->fl_flags & FL_POSIX) == 0)
1013 return rc;
1014
1015 try_again:
1016 down_write(&cinode->lock_sem);
1017 if (!cinode->can_cache_brlcks) {
1018 up_write(&cinode->lock_sem);
1019 return rc;
1020 }
1021
1022 rc = posix_lock_file(file, flock, NULL);
1023 up_write(&cinode->lock_sem);
1024 if (rc == FILE_LOCK_DEFERRED) {
1025 rc = wait_event_interruptible(flock->fl_wait, !flock->fl_next);
1026 if (!rc)
1027 goto try_again;
1028 posix_unblock_lock(flock);
1029 }
1030 return rc;
1031 }
1032
1033 int
1034 cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
1035 {
1036 unsigned int xid;
1037 int rc = 0, stored_rc;
1038 struct cifsLockInfo *li, *tmp;
1039 struct cifs_tcon *tcon;
1040 unsigned int num, max_num, max_buf;
1041 LOCKING_ANDX_RANGE *buf, *cur;
1042 int types[] = {LOCKING_ANDX_LARGE_FILES,
1043 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1044 int i;
1045
1046 xid = get_xid();
1047 tcon = tlink_tcon(cfile->tlink);
1048
1049 /*
1050 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1051 * and check it for zero before using.
1052 */
1053 max_buf = tcon->ses->server->maxBuf;
1054 if (!max_buf) {
1055 free_xid(xid);
1056 return -EINVAL;
1057 }
1058
1059 max_num = (max_buf - sizeof(struct smb_hdr)) /
1060 sizeof(LOCKING_ANDX_RANGE);
1061 buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1062 if (!buf) {
1063 free_xid(xid);
1064 return -ENOMEM;
1065 }
1066
1067 for (i = 0; i < 2; i++) {
1068 cur = buf;
1069 num = 0;
1070 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1071 if (li->type != types[i])
1072 continue;
1073 cur->Pid = cpu_to_le16(li->pid);
1074 cur->LengthLow = cpu_to_le32((u32)li->length);
1075 cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1076 cur->OffsetLow = cpu_to_le32((u32)li->offset);
1077 cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1078 if (++num == max_num) {
1079 stored_rc = cifs_lockv(xid, tcon,
1080 cfile->fid.netfid,
1081 (__u8)li->type, 0, num,
1082 buf);
1083 if (stored_rc)
1084 rc = stored_rc;
1085 cur = buf;
1086 num = 0;
1087 } else
1088 cur++;
1089 }
1090
1091 if (num) {
1092 stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1093 (__u8)types[i], 0, num, buf);
1094 if (stored_rc)
1095 rc = stored_rc;
1096 }
1097 }
1098
1099 kfree(buf);
1100 free_xid(xid);
1101 return rc;
1102 }
1103
1104 /* copied from fs/locks.c with a name change */
1105 #define cifs_for_each_lock(inode, lockp) \
1106 for (lockp = &inode->i_flock; *lockp != NULL; \
1107 lockp = &(*lockp)->fl_next)
1108
1109 struct lock_to_push {
1110 struct list_head llist;
1111 __u64 offset;
1112 __u64 length;
1113 __u32 pid;
1114 __u16 netfid;
1115 __u8 type;
1116 };
1117
1118 static int
1119 cifs_push_posix_locks(struct cifsFileInfo *cfile)
1120 {
1121 struct inode *inode = cfile->dentry->d_inode;
1122 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1123 struct file_lock *flock, **before;
1124 unsigned int count = 0, i = 0;
1125 int rc = 0, xid, type;
1126 struct list_head locks_to_send, *el;
1127 struct lock_to_push *lck, *tmp;
1128 __u64 length;
1129
1130 xid = get_xid();
1131
1132 spin_lock(&inode->i_lock);
1133 cifs_for_each_lock(inode, before) {
1134 if ((*before)->fl_flags & FL_POSIX)
1135 count++;
1136 }
1137 spin_unlock(&inode->i_lock);
1138
1139 INIT_LIST_HEAD(&locks_to_send);
1140
1141 /*
1142 * Allocating count locks is enough because no FL_POSIX locks can be
1143 * added to the list while we are holding cinode->lock_sem that
1144 * protects locking operations of this inode.
1145 */
1146 for (; i < count; i++) {
1147 lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1148 if (!lck) {
1149 rc = -ENOMEM;
1150 goto err_out;
1151 }
1152 list_add_tail(&lck->llist, &locks_to_send);
1153 }
1154
1155 el = locks_to_send.next;
1156 spin_lock(&inode->i_lock);
1157 cifs_for_each_lock(inode, before) {
1158 flock = *before;
1159 if ((flock->fl_flags & FL_POSIX) == 0)
1160 continue;
1161 if (el == &locks_to_send) {
1162 /*
1163 * The list ended. We don't have enough allocated
1164 * structures - something is really wrong.
1165 */
1166 cifs_dbg(VFS, "Can't push all brlocks!\n");
1167 break;
1168 }
1169 length = 1 + flock->fl_end - flock->fl_start;
1170 if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1171 type = CIFS_RDLCK;
1172 else
1173 type = CIFS_WRLCK;
1174 lck = list_entry(el, struct lock_to_push, llist);
1175 lck->pid = flock->fl_pid;
1176 lck->netfid = cfile->fid.netfid;
1177 lck->length = length;
1178 lck->type = type;
1179 lck->offset = flock->fl_start;
1180 el = el->next;
1181 }
1182 spin_unlock(&inode->i_lock);
1183
1184 list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1185 int stored_rc;
1186
1187 stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1188 lck->offset, lck->length, NULL,
1189 lck->type, 0);
1190 if (stored_rc)
1191 rc = stored_rc;
1192 list_del(&lck->llist);
1193 kfree(lck);
1194 }
1195
1196 out:
1197 free_xid(xid);
1198 return rc;
1199 err_out:
1200 list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1201 list_del(&lck->llist);
1202 kfree(lck);
1203 }
1204 goto out;
1205 }
1206
1207 static int
1208 cifs_push_locks(struct cifsFileInfo *cfile)
1209 {
1210 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1211 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
1212 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1213 int rc = 0;
1214
1215 /* we are going to update can_cache_brlcks here - need a write access */
1216 down_write(&cinode->lock_sem);
1217 if (!cinode->can_cache_brlcks) {
1218 up_write(&cinode->lock_sem);
1219 return rc;
1220 }
1221
1222 if (cap_unix(tcon->ses) &&
1223 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1224 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1225 rc = cifs_push_posix_locks(cfile);
1226 else
1227 rc = tcon->ses->server->ops->push_mand_locks(cfile);
1228
1229 cinode->can_cache_brlcks = false;
1230 up_write(&cinode->lock_sem);
1231 return rc;
1232 }
1233
1234 static void
1235 cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1236 bool *wait_flag, struct TCP_Server_Info *server)
1237 {
1238 if (flock->fl_flags & FL_POSIX)
1239 cifs_dbg(FYI, "Posix\n");
1240 if (flock->fl_flags & FL_FLOCK)
1241 cifs_dbg(FYI, "Flock\n");
1242 if (flock->fl_flags & FL_SLEEP) {
1243 cifs_dbg(FYI, "Blocking lock\n");
1244 *wait_flag = true;
1245 }
1246 if (flock->fl_flags & FL_ACCESS)
1247 cifs_dbg(FYI, "Process suspended by mandatory locking - not implemented yet\n");
1248 if (flock->fl_flags & FL_LEASE)
1249 cifs_dbg(FYI, "Lease on file - not implemented yet\n");
1250 if (flock->fl_flags &
1251 (~(FL_POSIX | FL_FLOCK | FL_SLEEP |
1252 FL_ACCESS | FL_LEASE | FL_CLOSE)))
1253 cifs_dbg(FYI, "Unknown lock flags 0x%x\n", flock->fl_flags);
1254
1255 *type = server->vals->large_lock_type;
1256 if (flock->fl_type == F_WRLCK) {
1257 cifs_dbg(FYI, "F_WRLCK\n");
1258 *type |= server->vals->exclusive_lock_type;
1259 *lock = 1;
1260 } else if (flock->fl_type == F_UNLCK) {
1261 cifs_dbg(FYI, "F_UNLCK\n");
1262 *type |= server->vals->unlock_lock_type;
1263 *unlock = 1;
1264 /* Check if unlock includes more than one lock range */
1265 } else if (flock->fl_type == F_RDLCK) {
1266 cifs_dbg(FYI, "F_RDLCK\n");
1267 *type |= server->vals->shared_lock_type;
1268 *lock = 1;
1269 } else if (flock->fl_type == F_EXLCK) {
1270 cifs_dbg(FYI, "F_EXLCK\n");
1271 *type |= server->vals->exclusive_lock_type;
1272 *lock = 1;
1273 } else if (flock->fl_type == F_SHLCK) {
1274 cifs_dbg(FYI, "F_SHLCK\n");
1275 *type |= server->vals->shared_lock_type;
1276 *lock = 1;
1277 } else
1278 cifs_dbg(FYI, "Unknown type of lock\n");
1279 }
1280
1281 static int
1282 cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1283 bool wait_flag, bool posix_lck, unsigned int xid)
1284 {
1285 int rc = 0;
1286 __u64 length = 1 + flock->fl_end - flock->fl_start;
1287 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1288 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1289 struct TCP_Server_Info *server = tcon->ses->server;
1290 __u16 netfid = cfile->fid.netfid;
1291
1292 if (posix_lck) {
1293 int posix_lock_type;
1294
1295 rc = cifs_posix_lock_test(file, flock);
1296 if (!rc)
1297 return rc;
1298
1299 if (type & server->vals->shared_lock_type)
1300 posix_lock_type = CIFS_RDLCK;
1301 else
1302 posix_lock_type = CIFS_WRLCK;
1303 rc = CIFSSMBPosixLock(xid, tcon, netfid, current->tgid,
1304 flock->fl_start, length, flock,
1305 posix_lock_type, wait_flag);
1306 return rc;
1307 }
1308
1309 rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1310 if (!rc)
1311 return rc;
1312
1313 /* BB we could chain these into one lock request BB */
1314 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length, type,
1315 1, 0, false);
1316 if (rc == 0) {
1317 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1318 type, 0, 1, false);
1319 flock->fl_type = F_UNLCK;
1320 if (rc != 0)
1321 cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1322 rc);
1323 return 0;
1324 }
1325
1326 if (type & server->vals->shared_lock_type) {
1327 flock->fl_type = F_WRLCK;
1328 return 0;
1329 }
1330
1331 type &= ~server->vals->exclusive_lock_type;
1332
1333 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1334 type | server->vals->shared_lock_type,
1335 1, 0, false);
1336 if (rc == 0) {
1337 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1338 type | server->vals->shared_lock_type, 0, 1, false);
1339 flock->fl_type = F_RDLCK;
1340 if (rc != 0)
1341 cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1342 rc);
1343 } else
1344 flock->fl_type = F_WRLCK;
1345
1346 return 0;
1347 }
1348
1349 void
1350 cifs_move_llist(struct list_head *source, struct list_head *dest)
1351 {
1352 struct list_head *li, *tmp;
1353 list_for_each_safe(li, tmp, source)
1354 list_move(li, dest);
1355 }
1356
1357 void
1358 cifs_free_llist(struct list_head *llist)
1359 {
1360 struct cifsLockInfo *li, *tmp;
1361 list_for_each_entry_safe(li, tmp, llist, llist) {
1362 cifs_del_lock_waiters(li);
1363 list_del(&li->llist);
1364 kfree(li);
1365 }
1366 }
1367
1368 int
1369 cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1370 unsigned int xid)
1371 {
1372 int rc = 0, stored_rc;
1373 int types[] = {LOCKING_ANDX_LARGE_FILES,
1374 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1375 unsigned int i;
1376 unsigned int max_num, num, max_buf;
1377 LOCKING_ANDX_RANGE *buf, *cur;
1378 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1379 struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
1380 struct cifsLockInfo *li, *tmp;
1381 __u64 length = 1 + flock->fl_end - flock->fl_start;
1382 struct list_head tmp_llist;
1383
1384 INIT_LIST_HEAD(&tmp_llist);
1385
1386 /*
1387 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1388 * and check it for zero before using.
1389 */
1390 max_buf = tcon->ses->server->maxBuf;
1391 if (!max_buf)
1392 return -EINVAL;
1393
1394 max_num = (max_buf - sizeof(struct smb_hdr)) /
1395 sizeof(LOCKING_ANDX_RANGE);
1396 buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1397 if (!buf)
1398 return -ENOMEM;
1399
1400 down_write(&cinode->lock_sem);
1401 for (i = 0; i < 2; i++) {
1402 cur = buf;
1403 num = 0;
1404 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1405 if (flock->fl_start > li->offset ||
1406 (flock->fl_start + length) <
1407 (li->offset + li->length))
1408 continue;
1409 if (current->tgid != li->pid)
1410 continue;
1411 if (types[i] != li->type)
1412 continue;
1413 if (cinode->can_cache_brlcks) {
1414 /*
1415 * We can cache brlock requests - simply remove
1416 * a lock from the file's list.
1417 */
1418 list_del(&li->llist);
1419 cifs_del_lock_waiters(li);
1420 kfree(li);
1421 continue;
1422 }
1423 cur->Pid = cpu_to_le16(li->pid);
1424 cur->LengthLow = cpu_to_le32((u32)li->length);
1425 cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1426 cur->OffsetLow = cpu_to_le32((u32)li->offset);
1427 cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1428 /*
1429 * We need to save a lock here to let us add it again to
1430 * the file's list if the unlock range request fails on
1431 * the server.
1432 */
1433 list_move(&li->llist, &tmp_llist);
1434 if (++num == max_num) {
1435 stored_rc = cifs_lockv(xid, tcon,
1436 cfile->fid.netfid,
1437 li->type, num, 0, buf);
1438 if (stored_rc) {
1439 /*
1440 * We failed on the unlock range
1441 * request - add all locks from the tmp
1442 * list to the head of the file's list.
1443 */
1444 cifs_move_llist(&tmp_llist,
1445 &cfile->llist->locks);
1446 rc = stored_rc;
1447 } else
1448 /*
1449 * The unlock range request succeed -
1450 * free the tmp list.
1451 */
1452 cifs_free_llist(&tmp_llist);
1453 cur = buf;
1454 num = 0;
1455 } else
1456 cur++;
1457 }
1458 if (num) {
1459 stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1460 types[i], num, 0, buf);
1461 if (stored_rc) {
1462 cifs_move_llist(&tmp_llist,
1463 &cfile->llist->locks);
1464 rc = stored_rc;
1465 } else
1466 cifs_free_llist(&tmp_llist);
1467 }
1468 }
1469
1470 up_write(&cinode->lock_sem);
1471 kfree(buf);
1472 return rc;
1473 }
1474
1475 static int
1476 cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1477 bool wait_flag, bool posix_lck, int lock, int unlock,
1478 unsigned int xid)
1479 {
1480 int rc = 0;
1481 __u64 length = 1 + flock->fl_end - flock->fl_start;
1482 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1483 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1484 struct TCP_Server_Info *server = tcon->ses->server;
1485 struct inode *inode = cfile->dentry->d_inode;
1486
1487 if (posix_lck) {
1488 int posix_lock_type;
1489
1490 rc = cifs_posix_lock_set(file, flock);
1491 if (!rc || rc < 0)
1492 return rc;
1493
1494 if (type & server->vals->shared_lock_type)
1495 posix_lock_type = CIFS_RDLCK;
1496 else
1497 posix_lock_type = CIFS_WRLCK;
1498
1499 if (unlock == 1)
1500 posix_lock_type = CIFS_UNLCK;
1501
1502 rc = CIFSSMBPosixLock(xid, tcon, cfile->fid.netfid,
1503 current->tgid, flock->fl_start, length,
1504 NULL, posix_lock_type, wait_flag);
1505 goto out;
1506 }
1507
1508 if (lock) {
1509 struct cifsLockInfo *lock;
1510
1511 lock = cifs_lock_init(flock->fl_start, length, type);
1512 if (!lock)
1513 return -ENOMEM;
1514
1515 rc = cifs_lock_add_if(cfile, lock, wait_flag);
1516 if (rc < 0) {
1517 kfree(lock);
1518 return rc;
1519 }
1520 if (!rc)
1521 goto out;
1522
1523 /*
1524 * Windows 7 server can delay breaking lease from read to None
1525 * if we set a byte-range lock on a file - break it explicitly
1526 * before sending the lock to the server to be sure the next
1527 * read won't conflict with non-overlapted locks due to
1528 * pagereading.
1529 */
1530 if (!CIFS_CACHE_WRITE(CIFS_I(inode)) &&
1531 CIFS_CACHE_READ(CIFS_I(inode))) {
1532 cifs_zap_mapping(inode);
1533 cifs_dbg(FYI, "Set no oplock for inode=%p due to mand locks\n",
1534 inode);
1535 CIFS_I(inode)->oplock = 0;
1536 }
1537
1538 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1539 type, 1, 0, wait_flag);
1540 if (rc) {
1541 kfree(lock);
1542 return rc;
1543 }
1544
1545 cifs_lock_add(cfile, lock);
1546 } else if (unlock)
1547 rc = server->ops->mand_unlock_range(cfile, flock, xid);
1548
1549 out:
1550 if (flock->fl_flags & FL_POSIX)
1551 posix_lock_file_wait(file, flock);
1552 return rc;
1553 }
1554
1555 int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1556 {
1557 int rc, xid;
1558 int lock = 0, unlock = 0;
1559 bool wait_flag = false;
1560 bool posix_lck = false;
1561 struct cifs_sb_info *cifs_sb;
1562 struct cifs_tcon *tcon;
1563 struct cifsInodeInfo *cinode;
1564 struct cifsFileInfo *cfile;
1565 __u16 netfid;
1566 __u32 type;
1567
1568 rc = -EACCES;
1569 xid = get_xid();
1570
1571 cifs_dbg(FYI, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1572 cmd, flock->fl_flags, flock->fl_type,
1573 flock->fl_start, flock->fl_end);
1574
1575 cfile = (struct cifsFileInfo *)file->private_data;
1576 tcon = tlink_tcon(cfile->tlink);
1577
1578 cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1579 tcon->ses->server);
1580
1581 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1582 netfid = cfile->fid.netfid;
1583 cinode = CIFS_I(file_inode(file));
1584
1585 if (cap_unix(tcon->ses) &&
1586 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1587 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1588 posix_lck = true;
1589 /*
1590 * BB add code here to normalize offset and length to account for
1591 * negative length which we can not accept over the wire.
1592 */
1593 if (IS_GETLK(cmd)) {
1594 rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1595 free_xid(xid);
1596 return rc;
1597 }
1598
1599 if (!lock && !unlock) {
1600 /*
1601 * if no lock or unlock then nothing to do since we do not
1602 * know what it is
1603 */
1604 free_xid(xid);
1605 return -EOPNOTSUPP;
1606 }
1607
1608 rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1609 xid);
1610 free_xid(xid);
1611 return rc;
1612 }
1613
1614 /*
1615 * update the file size (if needed) after a write. Should be called with
1616 * the inode->i_lock held
1617 */
1618 void
1619 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1620 unsigned int bytes_written)
1621 {
1622 loff_t end_of_write = offset + bytes_written;
1623
1624 if (end_of_write > cifsi->server_eof)
1625 cifsi->server_eof = end_of_write;
1626 }
1627
1628 static ssize_t
1629 cifs_write(struct cifsFileInfo *open_file, __u32 pid, const char *write_data,
1630 size_t write_size, loff_t *offset)
1631 {
1632 int rc = 0;
1633 unsigned int bytes_written = 0;
1634 unsigned int total_written;
1635 struct cifs_sb_info *cifs_sb;
1636 struct cifs_tcon *tcon;
1637 struct TCP_Server_Info *server;
1638 unsigned int xid;
1639 struct dentry *dentry = open_file->dentry;
1640 struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1641 struct cifs_io_parms io_parms;
1642
1643 cifs_sb = CIFS_SB(dentry->d_sb);
1644
1645 cifs_dbg(FYI, "write %zd bytes to offset %lld of %s\n",
1646 write_size, *offset, dentry->d_name.name);
1647
1648 tcon = tlink_tcon(open_file->tlink);
1649 server = tcon->ses->server;
1650
1651 if (!server->ops->sync_write)
1652 return -ENOSYS;
1653
1654 xid = get_xid();
1655
1656 for (total_written = 0; write_size > total_written;
1657 total_written += bytes_written) {
1658 rc = -EAGAIN;
1659 while (rc == -EAGAIN) {
1660 struct kvec iov[2];
1661 unsigned int len;
1662
1663 if (open_file->invalidHandle) {
1664 /* we could deadlock if we called
1665 filemap_fdatawait from here so tell
1666 reopen_file not to flush data to
1667 server now */
1668 rc = cifs_reopen_file(open_file, false);
1669 if (rc != 0)
1670 break;
1671 }
1672
1673 len = min((size_t)cifs_sb->wsize,
1674 write_size - total_written);
1675 /* iov[0] is reserved for smb header */
1676 iov[1].iov_base = (char *)write_data + total_written;
1677 iov[1].iov_len = len;
1678 io_parms.pid = pid;
1679 io_parms.tcon = tcon;
1680 io_parms.offset = *offset;
1681 io_parms.length = len;
1682 rc = server->ops->sync_write(xid, open_file, &io_parms,
1683 &bytes_written, iov, 1);
1684 }
1685 if (rc || (bytes_written == 0)) {
1686 if (total_written)
1687 break;
1688 else {
1689 free_xid(xid);
1690 return rc;
1691 }
1692 } else {
1693 spin_lock(&dentry->d_inode->i_lock);
1694 cifs_update_eof(cifsi, *offset, bytes_written);
1695 spin_unlock(&dentry->d_inode->i_lock);
1696 *offset += bytes_written;
1697 }
1698 }
1699
1700 cifs_stats_bytes_written(tcon, total_written);
1701
1702 if (total_written > 0) {
1703 spin_lock(&dentry->d_inode->i_lock);
1704 if (*offset > dentry->d_inode->i_size)
1705 i_size_write(dentry->d_inode, *offset);
1706 spin_unlock(&dentry->d_inode->i_lock);
1707 }
1708 mark_inode_dirty_sync(dentry->d_inode);
1709 free_xid(xid);
1710 return total_written;
1711 }
1712
1713 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1714 bool fsuid_only)
1715 {
1716 struct cifsFileInfo *open_file = NULL;
1717 struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1718
1719 /* only filter by fsuid on multiuser mounts */
1720 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1721 fsuid_only = false;
1722
1723 spin_lock(&cifs_file_list_lock);
1724 /* we could simply get the first_list_entry since write-only entries
1725 are always at the end of the list but since the first entry might
1726 have a close pending, we go through the whole list */
1727 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1728 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1729 continue;
1730 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1731 if (!open_file->invalidHandle) {
1732 /* found a good file */
1733 /* lock it so it will not be closed on us */
1734 cifsFileInfo_get_locked(open_file);
1735 spin_unlock(&cifs_file_list_lock);
1736 return open_file;
1737 } /* else might as well continue, and look for
1738 another, or simply have the caller reopen it
1739 again rather than trying to fix this handle */
1740 } else /* write only file */
1741 break; /* write only files are last so must be done */
1742 }
1743 spin_unlock(&cifs_file_list_lock);
1744 return NULL;
1745 }
1746
1747 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1748 bool fsuid_only)
1749 {
1750 struct cifsFileInfo *open_file, *inv_file = NULL;
1751 struct cifs_sb_info *cifs_sb;
1752 bool any_available = false;
1753 int rc;
1754 unsigned int refind = 0;
1755
1756 /* Having a null inode here (because mapping->host was set to zero by
1757 the VFS or MM) should not happen but we had reports of on oops (due to
1758 it being zero) during stress testcases so we need to check for it */
1759
1760 if (cifs_inode == NULL) {
1761 cifs_dbg(VFS, "Null inode passed to cifs_writeable_file\n");
1762 dump_stack();
1763 return NULL;
1764 }
1765
1766 cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1767
1768 /* only filter by fsuid on multiuser mounts */
1769 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1770 fsuid_only = false;
1771
1772 spin_lock(&cifs_file_list_lock);
1773 refind_writable:
1774 if (refind > MAX_REOPEN_ATT) {
1775 spin_unlock(&cifs_file_list_lock);
1776 return NULL;
1777 }
1778 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1779 if (!any_available && open_file->pid != current->tgid)
1780 continue;
1781 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1782 continue;
1783 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1784 if (!open_file->invalidHandle) {
1785 /* found a good writable file */
1786 cifsFileInfo_get_locked(open_file);
1787 spin_unlock(&cifs_file_list_lock);
1788 return open_file;
1789 } else {
1790 if (!inv_file)
1791 inv_file = open_file;
1792 }
1793 }
1794 }
1795 /* couldn't find useable FH with same pid, try any available */
1796 if (!any_available) {
1797 any_available = true;
1798 goto refind_writable;
1799 }
1800
1801 if (inv_file) {
1802 any_available = false;
1803 cifsFileInfo_get_locked(inv_file);
1804 }
1805
1806 spin_unlock(&cifs_file_list_lock);
1807
1808 if (inv_file) {
1809 rc = cifs_reopen_file(inv_file, false);
1810 if (!rc)
1811 return inv_file;
1812 else {
1813 spin_lock(&cifs_file_list_lock);
1814 list_move_tail(&inv_file->flist,
1815 &cifs_inode->openFileList);
1816 spin_unlock(&cifs_file_list_lock);
1817 cifsFileInfo_put(inv_file);
1818 spin_lock(&cifs_file_list_lock);
1819 ++refind;
1820 goto refind_writable;
1821 }
1822 }
1823
1824 return NULL;
1825 }
1826
1827 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1828 {
1829 struct address_space *mapping = page->mapping;
1830 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1831 char *write_data;
1832 int rc = -EFAULT;
1833 int bytes_written = 0;
1834 struct inode *inode;
1835 struct cifsFileInfo *open_file;
1836
1837 if (!mapping || !mapping->host)
1838 return -EFAULT;
1839
1840 inode = page->mapping->host;
1841
1842 offset += (loff_t)from;
1843 write_data = kmap(page);
1844 write_data += from;
1845
1846 if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1847 kunmap(page);
1848 return -EIO;
1849 }
1850
1851 /* racing with truncate? */
1852 if (offset > mapping->host->i_size) {
1853 kunmap(page);
1854 return 0; /* don't care */
1855 }
1856
1857 /* check to make sure that we are not extending the file */
1858 if (mapping->host->i_size - offset < (loff_t)to)
1859 to = (unsigned)(mapping->host->i_size - offset);
1860
1861 open_file = find_writable_file(CIFS_I(mapping->host), false);
1862 if (open_file) {
1863 bytes_written = cifs_write(open_file, open_file->pid,
1864 write_data, to - from, &offset);
1865 cifsFileInfo_put(open_file);
1866 /* Does mm or vfs already set times? */
1867 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1868 if ((bytes_written > 0) && (offset))
1869 rc = 0;
1870 else if (bytes_written < 0)
1871 rc = bytes_written;
1872 } else {
1873 cifs_dbg(FYI, "No writeable filehandles for inode\n");
1874 rc = -EIO;
1875 }
1876
1877 kunmap(page);
1878 return rc;
1879 }
1880
1881 static unsigned int
1882 wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
1883 struct address_space *mapping,
1884 struct writeback_control *wbc,
1885 pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
1886 {
1887 unsigned int nr_pages = 0, i;
1888 struct page *page;
1889
1890 for (i = 0; i < found_pages; i++) {
1891 page = wdata->pages[i];
1892 /*
1893 * At this point we hold neither mapping->tree_lock nor
1894 * lock on the page itself: the page may be truncated or
1895 * invalidated (changing page->mapping to NULL), or even
1896 * swizzled back from swapper_space to tmpfs file
1897 * mapping
1898 */
1899
1900 if (nr_pages == 0)
1901 lock_page(page);
1902 else if (!trylock_page(page))
1903 break;
1904
1905 if (unlikely(page->mapping != mapping)) {
1906 unlock_page(page);
1907 break;
1908 }
1909
1910 if (!wbc->range_cyclic && page->index > end) {
1911 *done = true;
1912 unlock_page(page);
1913 break;
1914 }
1915
1916 if (*next && (page->index != *next)) {
1917 /* Not next consecutive page */
1918 unlock_page(page);
1919 break;
1920 }
1921
1922 if (wbc->sync_mode != WB_SYNC_NONE)
1923 wait_on_page_writeback(page);
1924
1925 if (PageWriteback(page) ||
1926 !clear_page_dirty_for_io(page)) {
1927 unlock_page(page);
1928 break;
1929 }
1930
1931 /*
1932 * This actually clears the dirty bit in the radix tree.
1933 * See cifs_writepage() for more commentary.
1934 */
1935 set_page_writeback(page);
1936 if (page_offset(page) >= i_size_read(mapping->host)) {
1937 *done = true;
1938 unlock_page(page);
1939 end_page_writeback(page);
1940 break;
1941 }
1942
1943 wdata->pages[i] = page;
1944 *next = page->index + 1;
1945 ++nr_pages;
1946 }
1947
1948 /* reset index to refind any pages skipped */
1949 if (nr_pages == 0)
1950 *index = wdata->pages[0]->index + 1;
1951
1952 /* put any pages we aren't going to use */
1953 for (i = nr_pages; i < found_pages; i++) {
1954 page_cache_release(wdata->pages[i]);
1955 wdata->pages[i] = NULL;
1956 }
1957
1958 return nr_pages;
1959 }
1960
1961 static int
1962 wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
1963 struct address_space *mapping, struct writeback_control *wbc)
1964 {
1965 int rc = 0;
1966 struct TCP_Server_Info *server;
1967 unsigned int i;
1968
1969 wdata->sync_mode = wbc->sync_mode;
1970 wdata->nr_pages = nr_pages;
1971 wdata->offset = page_offset(wdata->pages[0]);
1972 wdata->pagesz = PAGE_CACHE_SIZE;
1973 wdata->tailsz = min(i_size_read(mapping->host) -
1974 page_offset(wdata->pages[nr_pages - 1]),
1975 (loff_t)PAGE_CACHE_SIZE);
1976 wdata->bytes = ((nr_pages - 1) * PAGE_CACHE_SIZE) + wdata->tailsz;
1977
1978 do {
1979 if (wdata->cfile != NULL)
1980 cifsFileInfo_put(wdata->cfile);
1981 wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
1982 if (!wdata->cfile) {
1983 cifs_dbg(VFS, "No writable handles for inode\n");
1984 rc = -EBADF;
1985 break;
1986 }
1987 wdata->pid = wdata->cfile->pid;
1988 server = tlink_tcon(wdata->cfile->tlink)->ses->server;
1989 rc = server->ops->async_writev(wdata, cifs_writedata_release);
1990 } while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
1991
1992 for (i = 0; i < nr_pages; ++i)
1993 unlock_page(wdata->pages[i]);
1994
1995 return rc;
1996 }
1997
1998 static int cifs_writepages(struct address_space *mapping,
1999 struct writeback_control *wbc)
2000 {
2001 struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
2002 bool done = false, scanned = false, range_whole = false;
2003 pgoff_t end, index;
2004 struct cifs_writedata *wdata;
2005 int rc = 0;
2006
2007 /*
2008 * If wsize is smaller than the page cache size, default to writing
2009 * one page at a time via cifs_writepage
2010 */
2011 if (cifs_sb->wsize < PAGE_CACHE_SIZE)
2012 return generic_writepages(mapping, wbc);
2013
2014 if (wbc->range_cyclic) {
2015 index = mapping->writeback_index; /* Start from prev offset */
2016 end = -1;
2017 } else {
2018 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2019 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2020 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2021 range_whole = true;
2022 scanned = true;
2023 }
2024 retry:
2025 while (!done && index <= end) {
2026 unsigned int i, nr_pages, found_pages;
2027 pgoff_t next = 0, tofind;
2028 struct page **pages;
2029
2030 tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
2031 end - index) + 1;
2032
2033 wdata = cifs_writedata_alloc((unsigned int)tofind,
2034 cifs_writev_complete);
2035 if (!wdata) {
2036 rc = -ENOMEM;
2037 break;
2038 }
2039
2040 /*
2041 * find_get_pages_tag seems to return a max of 256 on each
2042 * iteration, so we must call it several times in order to
2043 * fill the array or the wsize is effectively limited to
2044 * 256 * PAGE_CACHE_SIZE.
2045 */
2046 found_pages = 0;
2047 pages = wdata->pages;
2048 do {
2049 nr_pages = find_get_pages_tag(mapping, &index,
2050 PAGECACHE_TAG_DIRTY,
2051 tofind, pages);
2052 found_pages += nr_pages;
2053 tofind -= nr_pages;
2054 pages += nr_pages;
2055 } while (nr_pages && tofind && index <= end);
2056
2057 if (found_pages == 0) {
2058 kref_put(&wdata->refcount, cifs_writedata_release);
2059 break;
2060 }
2061
2062 nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2063 end, &index, &next, &done);
2064
2065 /* nothing to write? */
2066 if (nr_pages == 0) {
2067 kref_put(&wdata->refcount, cifs_writedata_release);
2068 continue;
2069 }
2070
2071 rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
2072
2073 /* send failure -- clean up the mess */
2074 if (rc != 0) {
2075 for (i = 0; i < nr_pages; ++i) {
2076 if (rc == -EAGAIN)
2077 redirty_page_for_writepage(wbc,
2078 wdata->pages[i]);
2079 else
2080 SetPageError(wdata->pages[i]);
2081 end_page_writeback(wdata->pages[i]);
2082 page_cache_release(wdata->pages[i]);
2083 }
2084 if (rc != -EAGAIN)
2085 mapping_set_error(mapping, rc);
2086 }
2087 kref_put(&wdata->refcount, cifs_writedata_release);
2088
2089 wbc->nr_to_write -= nr_pages;
2090 if (wbc->nr_to_write <= 0)
2091 done = true;
2092
2093 index = next;
2094 }
2095
2096 if (!scanned && !done) {
2097 /*
2098 * We hit the last page and there is more work to be done: wrap
2099 * back to the start of the file
2100 */
2101 scanned = true;
2102 index = 0;
2103 goto retry;
2104 }
2105
2106 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2107 mapping->writeback_index = index;
2108
2109 return rc;
2110 }
2111
2112 static int
2113 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2114 {
2115 int rc;
2116 unsigned int xid;
2117
2118 xid = get_xid();
2119 /* BB add check for wbc flags */
2120 page_cache_get(page);
2121 if (!PageUptodate(page))
2122 cifs_dbg(FYI, "ppw - page not up to date\n");
2123
2124 /*
2125 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2126 *
2127 * A writepage() implementation always needs to do either this,
2128 * or re-dirty the page with "redirty_page_for_writepage()" in
2129 * the case of a failure.
2130 *
2131 * Just unlocking the page will cause the radix tree tag-bits
2132 * to fail to update with the state of the page correctly.
2133 */
2134 set_page_writeback(page);
2135 retry_write:
2136 rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
2137 if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
2138 goto retry_write;
2139 else if (rc == -EAGAIN)
2140 redirty_page_for_writepage(wbc, page);
2141 else if (rc != 0)
2142 SetPageError(page);
2143 else
2144 SetPageUptodate(page);
2145 end_page_writeback(page);
2146 page_cache_release(page);
2147 free_xid(xid);
2148 return rc;
2149 }
2150
2151 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2152 {
2153 int rc = cifs_writepage_locked(page, wbc);
2154 unlock_page(page);
2155 return rc;
2156 }
2157
2158 static int cifs_write_end(struct file *file, struct address_space *mapping,
2159 loff_t pos, unsigned len, unsigned copied,
2160 struct page *page, void *fsdata)
2161 {
2162 int rc;
2163 struct inode *inode = mapping->host;
2164 struct cifsFileInfo *cfile = file->private_data;
2165 struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2166 __u32 pid;
2167
2168 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2169 pid = cfile->pid;
2170 else
2171 pid = current->tgid;
2172
2173 cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2174 page, pos, copied);
2175
2176 if (PageChecked(page)) {
2177 if (copied == len)
2178 SetPageUptodate(page);
2179 ClearPageChecked(page);
2180 } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
2181 SetPageUptodate(page);
2182
2183 if (!PageUptodate(page)) {
2184 char *page_data;
2185 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2186 unsigned int xid;
2187
2188 xid = get_xid();
2189 /* this is probably better than directly calling
2190 partialpage_write since in this function the file handle is
2191 known which we might as well leverage */
2192 /* BB check if anything else missing out of ppw
2193 such as updating last write time */
2194 page_data = kmap(page);
2195 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2196 /* if (rc < 0) should we set writebehind rc? */
2197 kunmap(page);
2198
2199 free_xid(xid);
2200 } else {
2201 rc = copied;
2202 pos += copied;
2203 set_page_dirty(page);
2204 }
2205
2206 if (rc > 0) {
2207 spin_lock(&inode->i_lock);
2208 if (pos > inode->i_size)
2209 i_size_write(inode, pos);
2210 spin_unlock(&inode->i_lock);
2211 }
2212
2213 unlock_page(page);
2214 page_cache_release(page);
2215
2216 return rc;
2217 }
2218
2219 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2220 int datasync)
2221 {
2222 unsigned int xid;
2223 int rc = 0;
2224 struct cifs_tcon *tcon;
2225 struct TCP_Server_Info *server;
2226 struct cifsFileInfo *smbfile = file->private_data;
2227 struct inode *inode = file_inode(file);
2228 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2229
2230 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2231 if (rc)
2232 return rc;
2233 mutex_lock(&inode->i_mutex);
2234
2235 xid = get_xid();
2236
2237 cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2238 file->f_path.dentry->d_name.name, datasync);
2239
2240 if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2241 rc = cifs_zap_mapping(inode);
2242 if (rc) {
2243 cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2244 rc = 0; /* don't care about it in fsync */
2245 }
2246 }
2247
2248 tcon = tlink_tcon(smbfile->tlink);
2249 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2250 server = tcon->ses->server;
2251 if (server->ops->flush)
2252 rc = server->ops->flush(xid, tcon, &smbfile->fid);
2253 else
2254 rc = -ENOSYS;
2255 }
2256
2257 free_xid(xid);
2258 mutex_unlock(&inode->i_mutex);
2259 return rc;
2260 }
2261
2262 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2263 {
2264 unsigned int xid;
2265 int rc = 0;
2266 struct cifs_tcon *tcon;
2267 struct TCP_Server_Info *server;
2268 struct cifsFileInfo *smbfile = file->private_data;
2269 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2270 struct inode *inode = file->f_mapping->host;
2271
2272 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2273 if (rc)
2274 return rc;
2275 mutex_lock(&inode->i_mutex);
2276
2277 xid = get_xid();
2278
2279 cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2280 file->f_path.dentry->d_name.name, datasync);
2281
2282 tcon = tlink_tcon(smbfile->tlink);
2283 if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2284 server = tcon->ses->server;
2285 if (server->ops->flush)
2286 rc = server->ops->flush(xid, tcon, &smbfile->fid);
2287 else
2288 rc = -ENOSYS;
2289 }
2290
2291 free_xid(xid);
2292 mutex_unlock(&inode->i_mutex);
2293 return rc;
2294 }
2295
2296 /*
2297 * As file closes, flush all cached write data for this inode checking
2298 * for write behind errors.
2299 */
2300 int cifs_flush(struct file *file, fl_owner_t id)
2301 {
2302 struct inode *inode = file_inode(file);
2303 int rc = 0;
2304
2305 if (file->f_mode & FMODE_WRITE)
2306 rc = filemap_write_and_wait(inode->i_mapping);
2307
2308 cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2309
2310 return rc;
2311 }
2312
2313 static int
2314 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2315 {
2316 int rc = 0;
2317 unsigned long i;
2318
2319 for (i = 0; i < num_pages; i++) {
2320 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2321 if (!pages[i]) {
2322 /*
2323 * save number of pages we have already allocated and
2324 * return with ENOMEM error
2325 */
2326 num_pages = i;
2327 rc = -ENOMEM;
2328 break;
2329 }
2330 }
2331
2332 if (rc) {
2333 for (i = 0; i < num_pages; i++)
2334 put_page(pages[i]);
2335 }
2336 return rc;
2337 }
2338
2339 static inline
2340 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2341 {
2342 size_t num_pages;
2343 size_t clen;
2344
2345 clen = min_t(const size_t, len, wsize);
2346 num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2347
2348 if (cur_len)
2349 *cur_len = clen;
2350
2351 return num_pages;
2352 }
2353
2354 static void
2355 cifs_uncached_writedata_release(struct kref *refcount)
2356 {
2357 int i;
2358 struct cifs_writedata *wdata = container_of(refcount,
2359 struct cifs_writedata, refcount);
2360
2361 for (i = 0; i < wdata->nr_pages; i++)
2362 put_page(wdata->pages[i]);
2363 cifs_writedata_release(refcount);
2364 }
2365
2366 static void
2367 cifs_uncached_writev_complete(struct work_struct *work)
2368 {
2369 struct cifs_writedata *wdata = container_of(work,
2370 struct cifs_writedata, work);
2371 struct inode *inode = wdata->cfile->dentry->d_inode;
2372 struct cifsInodeInfo *cifsi = CIFS_I(inode);
2373
2374 spin_lock(&inode->i_lock);
2375 cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2376 if (cifsi->server_eof > inode->i_size)
2377 i_size_write(inode, cifsi->server_eof);
2378 spin_unlock(&inode->i_lock);
2379
2380 complete(&wdata->done);
2381
2382 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2383 }
2384
2385 /* attempt to send write to server, retry on any -EAGAIN errors */
2386 static int
2387 cifs_uncached_retry_writev(struct cifs_writedata *wdata)
2388 {
2389 int rc;
2390 struct TCP_Server_Info *server;
2391
2392 server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2393
2394 do {
2395 if (wdata->cfile->invalidHandle) {
2396 rc = cifs_reopen_file(wdata->cfile, false);
2397 if (rc != 0)
2398 continue;
2399 }
2400 rc = server->ops->async_writev(wdata,
2401 cifs_uncached_writedata_release);
2402 } while (rc == -EAGAIN);
2403
2404 return rc;
2405 }
2406
2407 static ssize_t
2408 cifs_iovec_write(struct file *file, struct iov_iter *from, loff_t *poffset)
2409 {
2410 unsigned long nr_pages, i;
2411 size_t bytes, copied, len, cur_len;
2412 ssize_t total_written = 0;
2413 loff_t offset;
2414 struct cifsFileInfo *open_file;
2415 struct cifs_tcon *tcon;
2416 struct cifs_sb_info *cifs_sb;
2417 struct cifs_writedata *wdata, *tmp;
2418 struct list_head wdata_list;
2419 int rc;
2420 pid_t pid;
2421
2422 len = iov_iter_count(from);
2423 rc = generic_write_checks(file, poffset, &len, 0);
2424 if (rc)
2425 return rc;
2426
2427 if (!len)
2428 return 0;
2429
2430 iov_iter_truncate(from, len);
2431
2432 INIT_LIST_HEAD(&wdata_list);
2433 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2434 open_file = file->private_data;
2435 tcon = tlink_tcon(open_file->tlink);
2436
2437 if (!tcon->ses->server->ops->async_writev)
2438 return -ENOSYS;
2439
2440 offset = *poffset;
2441
2442 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2443 pid = open_file->pid;
2444 else
2445 pid = current->tgid;
2446
2447 do {
2448 size_t save_len;
2449
2450 nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
2451 wdata = cifs_writedata_alloc(nr_pages,
2452 cifs_uncached_writev_complete);
2453 if (!wdata) {
2454 rc = -ENOMEM;
2455 break;
2456 }
2457
2458 rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2459 if (rc) {
2460 kfree(wdata);
2461 break;
2462 }
2463
2464 save_len = cur_len;
2465 for (i = 0; i < nr_pages; i++) {
2466 bytes = min_t(size_t, cur_len, PAGE_SIZE);
2467 copied = copy_page_from_iter(wdata->pages[i], 0, bytes,
2468 from);
2469 cur_len -= copied;
2470 /*
2471 * If we didn't copy as much as we expected, then that
2472 * may mean we trod into an unmapped area. Stop copying
2473 * at that point. On the next pass through the big
2474 * loop, we'll likely end up getting a zero-length
2475 * write and bailing out of it.
2476 */
2477 if (copied < bytes)
2478 break;
2479 }
2480 cur_len = save_len - cur_len;
2481
2482 /*
2483 * If we have no data to send, then that probably means that
2484 * the copy above failed altogether. That's most likely because
2485 * the address in the iovec was bogus. Set the rc to -EFAULT,
2486 * free anything we allocated and bail out.
2487 */
2488 if (!cur_len) {
2489 for (i = 0; i < nr_pages; i++)
2490 put_page(wdata->pages[i]);
2491 kfree(wdata);
2492 rc = -EFAULT;
2493 break;
2494 }
2495
2496 /*
2497 * i + 1 now represents the number of pages we actually used in
2498 * the copy phase above. Bring nr_pages down to that, and free
2499 * any pages that we didn't use.
2500 */
2501 for ( ; nr_pages > i + 1; nr_pages--)
2502 put_page(wdata->pages[nr_pages - 1]);
2503
2504 wdata->sync_mode = WB_SYNC_ALL;
2505 wdata->nr_pages = nr_pages;
2506 wdata->offset = (__u64)offset;
2507 wdata->cfile = cifsFileInfo_get(open_file);
2508 wdata->pid = pid;
2509 wdata->bytes = cur_len;
2510 wdata->pagesz = PAGE_SIZE;
2511 wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2512 rc = cifs_uncached_retry_writev(wdata);
2513 if (rc) {
2514 kref_put(&wdata->refcount,
2515 cifs_uncached_writedata_release);
2516 break;
2517 }
2518
2519 list_add_tail(&wdata->list, &wdata_list);
2520 offset += cur_len;
2521 len -= cur_len;
2522 } while (len > 0);
2523
2524 /*
2525 * If at least one write was successfully sent, then discard any rc
2526 * value from the later writes. If the other write succeeds, then
2527 * we'll end up returning whatever was written. If it fails, then
2528 * we'll get a new rc value from that.
2529 */
2530 if (!list_empty(&wdata_list))
2531 rc = 0;
2532
2533 /*
2534 * Wait for and collect replies for any successful sends in order of
2535 * increasing offset. Once an error is hit or we get a fatal signal
2536 * while waiting, then return without waiting for any more replies.
2537 */
2538 restart_loop:
2539 list_for_each_entry_safe(wdata, tmp, &wdata_list, list) {
2540 if (!rc) {
2541 /* FIXME: freezable too? */
2542 rc = wait_for_completion_killable(&wdata->done);
2543 if (rc)
2544 rc = -EINTR;
2545 else if (wdata->result)
2546 rc = wdata->result;
2547 else
2548 total_written += wdata->bytes;
2549
2550 /* resend call if it's a retryable error */
2551 if (rc == -EAGAIN) {
2552 rc = cifs_uncached_retry_writev(wdata);
2553 goto restart_loop;
2554 }
2555 }
2556 list_del_init(&wdata->list);
2557 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2558 }
2559
2560 if (total_written > 0)
2561 *poffset += total_written;
2562
2563 cifs_stats_bytes_written(tcon, total_written);
2564 return total_written ? total_written : (ssize_t)rc;
2565 }
2566
2567 ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
2568 {
2569 ssize_t written;
2570 struct inode *inode;
2571 loff_t pos = iocb->ki_pos;
2572
2573 inode = file_inode(iocb->ki_filp);
2574
2575 /*
2576 * BB - optimize the way when signing is disabled. We can drop this
2577 * extra memory-to-memory copying and use iovec buffers for constructing
2578 * write request.
2579 */
2580
2581 written = cifs_iovec_write(iocb->ki_filp, from, &pos);
2582 if (written > 0) {
2583 set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(inode)->flags);
2584 iocb->ki_pos = pos;
2585 }
2586
2587 return written;
2588 }
2589
2590 static ssize_t
2591 cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2592 {
2593 struct file *file = iocb->ki_filp;
2594 struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2595 struct inode *inode = file->f_mapping->host;
2596 struct cifsInodeInfo *cinode = CIFS_I(inode);
2597 struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2598 ssize_t rc = -EACCES;
2599 loff_t lock_pos = iocb->ki_pos;
2600
2601 /*
2602 * We need to hold the sem to be sure nobody modifies lock list
2603 * with a brlock that prevents writing.
2604 */
2605 down_read(&cinode->lock_sem);
2606 mutex_lock(&inode->i_mutex);
2607 if (file->f_flags & O_APPEND)
2608 lock_pos = i_size_read(inode);
2609 if (!cifs_find_lock_conflict(cfile, lock_pos, iov_iter_count(from),
2610 server->vals->exclusive_lock_type, NULL,
2611 CIFS_WRITE_OP)) {
2612 rc = __generic_file_write_iter(iocb, from);
2613 mutex_unlock(&inode->i_mutex);
2614
2615 if (rc > 0) {
2616 ssize_t err;
2617
2618 err = generic_write_sync(file, iocb->ki_pos - rc, rc);
2619 if (err < 0)
2620 rc = err;
2621 }
2622 } else {
2623 mutex_unlock(&inode->i_mutex);
2624 }
2625 up_read(&cinode->lock_sem);
2626 return rc;
2627 }
2628
2629 ssize_t
2630 cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2631 {
2632 struct inode *inode = file_inode(iocb->ki_filp);
2633 struct cifsInodeInfo *cinode = CIFS_I(inode);
2634 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2635 struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2636 iocb->ki_filp->private_data;
2637 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2638 ssize_t written;
2639
2640 written = cifs_get_writer(cinode);
2641 if (written)
2642 return written;
2643
2644 if (CIFS_CACHE_WRITE(cinode)) {
2645 if (cap_unix(tcon->ses) &&
2646 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2647 && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2648 written = generic_file_write_iter(iocb, from);
2649 goto out;
2650 }
2651 written = cifs_writev(iocb, from);
2652 goto out;
2653 }
2654 /*
2655 * For non-oplocked files in strict cache mode we need to write the data
2656 * to the server exactly from the pos to pos+len-1 rather than flush all
2657 * affected pages because it may cause a error with mandatory locks on
2658 * these pages but not on the region from pos to ppos+len-1.
2659 */
2660 written = cifs_user_writev(iocb, from);
2661 if (written > 0 && CIFS_CACHE_READ(cinode)) {
2662 /*
2663 * Windows 7 server can delay breaking level2 oplock if a write
2664 * request comes - break it on the client to prevent reading
2665 * an old data.
2666 */
2667 cifs_zap_mapping(inode);
2668 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
2669 inode);
2670 cinode->oplock = 0;
2671 }
2672 out:
2673 cifs_put_writer(cinode);
2674 return written;
2675 }
2676
2677 static struct cifs_readdata *
2678 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
2679 {
2680 struct cifs_readdata *rdata;
2681
2682 rdata = kzalloc(sizeof(*rdata) + (sizeof(struct page *) * nr_pages),
2683 GFP_KERNEL);
2684 if (rdata != NULL) {
2685 kref_init(&rdata->refcount);
2686 INIT_LIST_HEAD(&rdata->list);
2687 init_completion(&rdata->done);
2688 INIT_WORK(&rdata->work, complete);
2689 }
2690
2691 return rdata;
2692 }
2693
2694 void
2695 cifs_readdata_release(struct kref *refcount)
2696 {
2697 struct cifs_readdata *rdata = container_of(refcount,
2698 struct cifs_readdata, refcount);
2699
2700 if (rdata->cfile)
2701 cifsFileInfo_put(rdata->cfile);
2702
2703 kfree(rdata);
2704 }
2705
2706 static int
2707 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
2708 {
2709 int rc = 0;
2710 struct page *page;
2711 unsigned int i;
2712
2713 for (i = 0; i < nr_pages; i++) {
2714 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2715 if (!page) {
2716 rc = -ENOMEM;
2717 break;
2718 }
2719 rdata->pages[i] = page;
2720 }
2721
2722 if (rc) {
2723 for (i = 0; i < nr_pages; i++) {
2724 put_page(rdata->pages[i]);
2725 rdata->pages[i] = NULL;
2726 }
2727 }
2728 return rc;
2729 }
2730
2731 static void
2732 cifs_uncached_readdata_release(struct kref *refcount)
2733 {
2734 struct cifs_readdata *rdata = container_of(refcount,
2735 struct cifs_readdata, refcount);
2736 unsigned int i;
2737
2738 for (i = 0; i < rdata->nr_pages; i++) {
2739 put_page(rdata->pages[i]);
2740 rdata->pages[i] = NULL;
2741 }
2742 cifs_readdata_release(refcount);
2743 }
2744
2745 static int
2746 cifs_retry_async_readv(struct cifs_readdata *rdata)
2747 {
2748 int rc;
2749 struct TCP_Server_Info *server;
2750
2751 server = tlink_tcon(rdata->cfile->tlink)->ses->server;
2752
2753 do {
2754 if (rdata->cfile->invalidHandle) {
2755 rc = cifs_reopen_file(rdata->cfile, true);
2756 if (rc != 0)
2757 continue;
2758 }
2759 rc = server->ops->async_readv(rdata);
2760 } while (rc == -EAGAIN);
2761
2762 return rc;
2763 }
2764
2765 /**
2766 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2767 * @rdata: the readdata response with list of pages holding data
2768 * @iter: destination for our data
2769 *
2770 * This function copies data from a list of pages in a readdata response into
2771 * an array of iovecs. It will first calculate where the data should go
2772 * based on the info in the readdata and then copy the data into that spot.
2773 */
2774 static int
2775 cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
2776 {
2777 size_t remaining = rdata->bytes;
2778 unsigned int i;
2779
2780 for (i = 0; i < rdata->nr_pages; i++) {
2781 struct page *page = rdata->pages[i];
2782 size_t copy = min_t(size_t, remaining, PAGE_SIZE);
2783 size_t written = copy_page_to_iter(page, 0, copy, iter);
2784 remaining -= written;
2785 if (written < copy && iov_iter_count(iter) > 0)
2786 break;
2787 }
2788 return remaining ? -EFAULT : 0;
2789 }
2790
2791 static void
2792 cifs_uncached_readv_complete(struct work_struct *work)
2793 {
2794 struct cifs_readdata *rdata = container_of(work,
2795 struct cifs_readdata, work);
2796
2797 complete(&rdata->done);
2798 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2799 }
2800
2801 static int
2802 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
2803 struct cifs_readdata *rdata, unsigned int len)
2804 {
2805 int total_read = 0, result = 0;
2806 unsigned int i;
2807 unsigned int nr_pages = rdata->nr_pages;
2808 struct kvec iov;
2809
2810 rdata->tailsz = PAGE_SIZE;
2811 for (i = 0; i < nr_pages; i++) {
2812 struct page *page = rdata->pages[i];
2813
2814 if (len >= PAGE_SIZE) {
2815 /* enough data to fill the page */
2816 iov.iov_base = kmap(page);
2817 iov.iov_len = PAGE_SIZE;
2818 cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2819 i, iov.iov_base, iov.iov_len);
2820 len -= PAGE_SIZE;
2821 } else if (len > 0) {
2822 /* enough for partial page, fill and zero the rest */
2823 iov.iov_base = kmap(page);
2824 iov.iov_len = len;
2825 cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2826 i, iov.iov_base, iov.iov_len);
2827 memset(iov.iov_base + len, '\0', PAGE_SIZE - len);
2828 rdata->tailsz = len;
2829 len = 0;
2830 } else {
2831 /* no need to hold page hostage */
2832 rdata->pages[i] = NULL;
2833 rdata->nr_pages--;
2834 put_page(page);
2835 continue;
2836 }
2837
2838 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
2839 kunmap(page);
2840 if (result < 0)
2841 break;
2842
2843 total_read += result;
2844 }
2845
2846 return total_read > 0 && result != -EAGAIN ? total_read : result;
2847 }
2848
2849 ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
2850 {
2851 struct file *file = iocb->ki_filp;
2852 ssize_t rc;
2853 size_t len, cur_len;
2854 ssize_t total_read = 0;
2855 loff_t offset = iocb->ki_pos;
2856 unsigned int npages;
2857 struct cifs_sb_info *cifs_sb;
2858 struct cifs_tcon *tcon;
2859 struct cifsFileInfo *open_file;
2860 struct cifs_readdata *rdata, *tmp;
2861 struct list_head rdata_list;
2862 pid_t pid;
2863
2864 len = iov_iter_count(to);
2865 if (!len)
2866 return 0;
2867
2868 INIT_LIST_HEAD(&rdata_list);
2869 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2870 open_file = file->private_data;
2871 tcon = tlink_tcon(open_file->tlink);
2872
2873 if (!tcon->ses->server->ops->async_readv)
2874 return -ENOSYS;
2875
2876 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2877 pid = open_file->pid;
2878 else
2879 pid = current->tgid;
2880
2881 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2882 cifs_dbg(FYI, "attempting read on write only file instance\n");
2883
2884 do {
2885 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
2886 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
2887
2888 /* allocate a readdata struct */
2889 rdata = cifs_readdata_alloc(npages,
2890 cifs_uncached_readv_complete);
2891 if (!rdata) {
2892 rc = -ENOMEM;
2893 break;
2894 }
2895
2896 rc = cifs_read_allocate_pages(rdata, npages);
2897 if (rc)
2898 goto error;
2899
2900 rdata->cfile = cifsFileInfo_get(open_file);
2901 rdata->nr_pages = npages;
2902 rdata->offset = offset;
2903 rdata->bytes = cur_len;
2904 rdata->pid = pid;
2905 rdata->pagesz = PAGE_SIZE;
2906 rdata->read_into_pages = cifs_uncached_read_into_pages;
2907
2908 rc = cifs_retry_async_readv(rdata);
2909 error:
2910 if (rc) {
2911 kref_put(&rdata->refcount,
2912 cifs_uncached_readdata_release);
2913 break;
2914 }
2915
2916 list_add_tail(&rdata->list, &rdata_list);
2917 offset += cur_len;
2918 len -= cur_len;
2919 } while (len > 0);
2920
2921 /* if at least one read request send succeeded, then reset rc */
2922 if (!list_empty(&rdata_list))
2923 rc = 0;
2924
2925 len = iov_iter_count(to);
2926 /* the loop below should proceed in the order of increasing offsets */
2927 list_for_each_entry_safe(rdata, tmp, &rdata_list, list) {
2928 again:
2929 if (!rc) {
2930 /* FIXME: freezable sleep too? */
2931 rc = wait_for_completion_killable(&rdata->done);
2932 if (rc)
2933 rc = -EINTR;
2934 else if (rdata->result) {
2935 rc = rdata->result;
2936 /* resend call if it's a retryable error */
2937 if (rc == -EAGAIN) {
2938 rc = cifs_retry_async_readv(rdata);
2939 goto again;
2940 }
2941 } else {
2942 rc = cifs_readdata_to_iov(rdata, to);
2943 }
2944
2945 }
2946 list_del_init(&rdata->list);
2947 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2948 }
2949
2950 total_read = len - iov_iter_count(to);
2951
2952 cifs_stats_bytes_read(tcon, total_read);
2953
2954 /* mask nodata case */
2955 if (rc == -ENODATA)
2956 rc = 0;
2957
2958 if (total_read) {
2959 iocb->ki_pos += total_read;
2960 return total_read;
2961 }
2962 return rc;
2963 }
2964
2965 ssize_t
2966 cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
2967 {
2968 struct inode *inode = file_inode(iocb->ki_filp);
2969 struct cifsInodeInfo *cinode = CIFS_I(inode);
2970 struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2971 struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2972 iocb->ki_filp->private_data;
2973 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2974 int rc = -EACCES;
2975
2976 /*
2977 * In strict cache mode we need to read from the server all the time
2978 * if we don't have level II oplock because the server can delay mtime
2979 * change - so we can't make a decision about inode invalidating.
2980 * And we can also fail with pagereading if there are mandatory locks
2981 * on pages affected by this read but not on the region from pos to
2982 * pos+len-1.
2983 */
2984 if (!CIFS_CACHE_READ(cinode))
2985 return cifs_user_readv(iocb, to);
2986
2987 if (cap_unix(tcon->ses) &&
2988 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
2989 ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
2990 return generic_file_read_iter(iocb, to);
2991
2992 /*
2993 * We need to hold the sem to be sure nobody modifies lock list
2994 * with a brlock that prevents reading.
2995 */
2996 down_read(&cinode->lock_sem);
2997 if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
2998 tcon->ses->server->vals->shared_lock_type,
2999 NULL, CIFS_READ_OP))
3000 rc = generic_file_read_iter(iocb, to);
3001 up_read(&cinode->lock_sem);
3002 return rc;
3003 }
3004
3005 static ssize_t
3006 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3007 {
3008 int rc = -EACCES;
3009 unsigned int bytes_read = 0;
3010 unsigned int total_read;
3011 unsigned int current_read_size;
3012 unsigned int rsize;
3013 struct cifs_sb_info *cifs_sb;
3014 struct cifs_tcon *tcon;
3015 struct TCP_Server_Info *server;
3016 unsigned int xid;
3017 char *cur_offset;
3018 struct cifsFileInfo *open_file;
3019 struct cifs_io_parms io_parms;
3020 int buf_type = CIFS_NO_BUFFER;
3021 __u32 pid;
3022
3023 xid = get_xid();
3024 cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3025
3026 /* FIXME: set up handlers for larger reads and/or convert to async */
3027 rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3028
3029 if (file->private_data == NULL) {
3030 rc = -EBADF;
3031 free_xid(xid);
3032 return rc;
3033 }
3034 open_file = file->private_data;
3035 tcon = tlink_tcon(open_file->tlink);
3036 server = tcon->ses->server;
3037
3038 if (!server->ops->sync_read) {
3039 free_xid(xid);
3040 return -ENOSYS;
3041 }
3042
3043 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3044 pid = open_file->pid;
3045 else
3046 pid = current->tgid;
3047
3048 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3049 cifs_dbg(FYI, "attempting read on write only file instance\n");
3050
3051 for (total_read = 0, cur_offset = read_data; read_size > total_read;
3052 total_read += bytes_read, cur_offset += bytes_read) {
3053 current_read_size = min_t(uint, read_size - total_read, rsize);
3054 /*
3055 * For windows me and 9x we do not want to request more than it
3056 * negotiated since it will refuse the read then.
3057 */
3058 if ((tcon->ses) && !(tcon->ses->capabilities &
3059 tcon->ses->server->vals->cap_large_files)) {
3060 current_read_size = min_t(uint, current_read_size,
3061 CIFSMaxBufSize);
3062 }
3063 rc = -EAGAIN;
3064 while (rc == -EAGAIN) {
3065 if (open_file->invalidHandle) {
3066 rc = cifs_reopen_file(open_file, true);
3067 if (rc != 0)
3068 break;
3069 }
3070 io_parms.pid = pid;
3071 io_parms.tcon = tcon;
3072 io_parms.offset = *offset;
3073 io_parms.length = current_read_size;
3074 rc = server->ops->sync_read(xid, open_file, &io_parms,
3075 &bytes_read, &cur_offset,
3076 &buf_type);
3077 }
3078 if (rc || (bytes_read == 0)) {
3079 if (total_read) {
3080 break;
3081 } else {
3082 free_xid(xid);
3083 return rc;
3084 }
3085 } else {
3086 cifs_stats_bytes_read(tcon, total_read);
3087 *offset += bytes_read;
3088 }
3089 }
3090 free_xid(xid);
3091 return total_read;
3092 }
3093
3094 /*
3095 * If the page is mmap'ed into a process' page tables, then we need to make
3096 * sure that it doesn't change while being written back.
3097 */
3098 static int
3099 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3100 {
3101 struct page *page = vmf->page;
3102
3103 lock_page(page);
3104 return VM_FAULT_LOCKED;
3105 }
3106
3107 static struct vm_operations_struct cifs_file_vm_ops = {
3108 .fault = filemap_fault,
3109 .map_pages = filemap_map_pages,
3110 .page_mkwrite = cifs_page_mkwrite,
3111 .remap_pages = generic_file_remap_pages,
3112 };
3113
3114 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3115 {
3116 int rc, xid;
3117 struct inode *inode = file_inode(file);
3118
3119 xid = get_xid();
3120
3121 if (!CIFS_CACHE_READ(CIFS_I(inode))) {
3122 rc = cifs_zap_mapping(inode);
3123 if (rc)
3124 return rc;
3125 }
3126
3127 rc = generic_file_mmap(file, vma);
3128 if (rc == 0)
3129 vma->vm_ops = &cifs_file_vm_ops;
3130 free_xid(xid);
3131 return rc;
3132 }
3133
3134 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3135 {
3136 int rc, xid;
3137
3138 xid = get_xid();
3139 rc = cifs_revalidate_file(file);
3140 if (rc) {
3141 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3142 rc);
3143 free_xid(xid);
3144 return rc;
3145 }
3146 rc = generic_file_mmap(file, vma);
3147 if (rc == 0)
3148 vma->vm_ops = &cifs_file_vm_ops;
3149 free_xid(xid);
3150 return rc;
3151 }
3152
3153 static void
3154 cifs_readv_complete(struct work_struct *work)
3155 {
3156 unsigned int i;
3157 struct cifs_readdata *rdata = container_of(work,
3158 struct cifs_readdata, work);
3159
3160 for (i = 0; i < rdata->nr_pages; i++) {
3161 struct page *page = rdata->pages[i];
3162
3163 lru_cache_add_file(page);
3164
3165 if (rdata->result == 0) {
3166 flush_dcache_page(page);
3167 SetPageUptodate(page);
3168 }
3169
3170 unlock_page(page);
3171
3172 if (rdata->result == 0)
3173 cifs_readpage_to_fscache(rdata->mapping->host, page);
3174
3175 page_cache_release(page);
3176 rdata->pages[i] = NULL;
3177 }
3178 kref_put(&rdata->refcount, cifs_readdata_release);
3179 }
3180
3181 static int
3182 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3183 struct cifs_readdata *rdata, unsigned int len)
3184 {
3185 int total_read = 0, result = 0;
3186 unsigned int i;
3187 u64 eof;
3188 pgoff_t eof_index;
3189 unsigned int nr_pages = rdata->nr_pages;
3190 struct kvec iov;
3191
3192 /* determine the eof that the server (probably) has */
3193 eof = CIFS_I(rdata->mapping->host)->server_eof;
3194 eof_index = eof ? (eof - 1) >> PAGE_CACHE_SHIFT : 0;
3195 cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3196
3197 rdata->tailsz = PAGE_CACHE_SIZE;
3198 for (i = 0; i < nr_pages; i++) {
3199 struct page *page = rdata->pages[i];
3200
3201 if (len >= PAGE_CACHE_SIZE) {
3202 /* enough data to fill the page */
3203 iov.iov_base = kmap(page);
3204 iov.iov_len = PAGE_CACHE_SIZE;
3205 cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3206 i, page->index, iov.iov_base, iov.iov_len);
3207 len -= PAGE_CACHE_SIZE;
3208 } else if (len > 0) {
3209 /* enough for partial page, fill and zero the rest */
3210 iov.iov_base = kmap(page);
3211 iov.iov_len = len;
3212 cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3213 i, page->index, iov.iov_base, iov.iov_len);
3214 memset(iov.iov_base + len,
3215 '\0', PAGE_CACHE_SIZE - len);
3216 rdata->tailsz = len;
3217 len = 0;
3218 } else if (page->index > eof_index) {
3219 /*
3220 * The VFS will not try to do readahead past the
3221 * i_size, but it's possible that we have outstanding
3222 * writes with gaps in the middle and the i_size hasn't
3223 * caught up yet. Populate those with zeroed out pages
3224 * to prevent the VFS from repeatedly attempting to
3225 * fill them until the writes are flushed.
3226 */
3227 zero_user(page, 0, PAGE_CACHE_SIZE);
3228 lru_cache_add_file(page);
3229 flush_dcache_page(page);
3230 SetPageUptodate(page);
3231 unlock_page(page);
3232 page_cache_release(page);
3233 rdata->pages[i] = NULL;
3234 rdata->nr_pages--;
3235 continue;
3236 } else {
3237 /* no need to hold page hostage */
3238 lru_cache_add_file(page);
3239 unlock_page(page);
3240 page_cache_release(page);
3241 rdata->pages[i] = NULL;
3242 rdata->nr_pages--;
3243 continue;
3244 }
3245
3246 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
3247 kunmap(page);
3248 if (result < 0)
3249 break;
3250
3251 total_read += result;
3252 }
3253
3254 return total_read > 0 && result != -EAGAIN ? total_read : result;
3255 }
3256
3257 static int cifs_readpages(struct file *file, struct address_space *mapping,
3258 struct list_head *page_list, unsigned num_pages)
3259 {
3260 int rc;
3261 struct list_head tmplist;
3262 struct cifsFileInfo *open_file = file->private_data;
3263 struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3264 unsigned int rsize = cifs_sb->rsize;
3265 pid_t pid;
3266
3267 /*
3268 * Give up immediately if rsize is too small to read an entire page.
3269 * The VFS will fall back to readpage. We should never reach this
3270 * point however since we set ra_pages to 0 when the rsize is smaller
3271 * than a cache page.
3272 */
3273 if (unlikely(rsize < PAGE_CACHE_SIZE))
3274 return 0;
3275
3276 /*
3277 * Reads as many pages as possible from fscache. Returns -ENOBUFS
3278 * immediately if the cookie is negative
3279 *
3280 * After this point, every page in the list might have PG_fscache set,
3281 * so we will need to clean that up off of every page we don't use.
3282 */
3283 rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3284 &num_pages);
3285 if (rc == 0)
3286 return rc;
3287
3288 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3289 pid = open_file->pid;
3290 else
3291 pid = current->tgid;
3292
3293 rc = 0;
3294 INIT_LIST_HEAD(&tmplist);
3295
3296 cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
3297 __func__, file, mapping, num_pages);
3298
3299 /*
3300 * Start with the page at end of list and move it to private
3301 * list. Do the same with any following pages until we hit
3302 * the rsize limit, hit an index discontinuity, or run out of
3303 * pages. Issue the async read and then start the loop again
3304 * until the list is empty.
3305 *
3306 * Note that list order is important. The page_list is in
3307 * the order of declining indexes. When we put the pages in
3308 * the rdata->pages, then we want them in increasing order.
3309 */
3310 while (!list_empty(page_list)) {
3311 unsigned int i;
3312 unsigned int bytes = PAGE_CACHE_SIZE;
3313 unsigned int expected_index;
3314 unsigned int nr_pages = 1;
3315 loff_t offset;
3316 struct page *page, *tpage;
3317 struct cifs_readdata *rdata;
3318
3319 page = list_entry(page_list->prev, struct page, lru);
3320
3321 /*
3322 * Lock the page and put it in the cache. Since no one else
3323 * should have access to this page, we're safe to simply set
3324 * PG_locked without checking it first.
3325 */
3326 __set_page_locked(page);
3327 rc = add_to_page_cache_locked(page, mapping,
3328 page->index, GFP_KERNEL);
3329
3330 /* give up if we can't stick it in the cache */
3331 if (rc) {
3332 __clear_page_locked(page);
3333 break;
3334 }
3335
3336 /* move first page to the tmplist */
3337 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3338 list_move_tail(&page->lru, &tmplist);
3339
3340 /* now try and add more pages onto the request */
3341 expected_index = page->index + 1;
3342 list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3343 /* discontinuity ? */
3344 if (page->index != expected_index)
3345 break;
3346
3347 /* would this page push the read over the rsize? */
3348 if (bytes + PAGE_CACHE_SIZE > rsize)
3349 break;
3350
3351 __set_page_locked(page);
3352 if (add_to_page_cache_locked(page, mapping,
3353 page->index, GFP_KERNEL)) {
3354 __clear_page_locked(page);
3355 break;
3356 }
3357 list_move_tail(&page->lru, &tmplist);
3358 bytes += PAGE_CACHE_SIZE;
3359 expected_index++;
3360 nr_pages++;
3361 }
3362
3363 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3364 if (!rdata) {
3365 /* best to give up if we're out of mem */
3366 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3367 list_del(&page->lru);
3368 lru_cache_add_file(page);
3369 unlock_page(page);
3370 page_cache_release(page);
3371 }
3372 rc = -ENOMEM;
3373 break;
3374 }
3375
3376 rdata->cfile = cifsFileInfo_get(open_file);
3377 rdata->mapping = mapping;
3378 rdata->offset = offset;
3379 rdata->bytes = bytes;
3380 rdata->pid = pid;
3381 rdata->pagesz = PAGE_CACHE_SIZE;
3382 rdata->read_into_pages = cifs_readpages_read_into_pages;
3383
3384 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3385 list_del(&page->lru);
3386 rdata->pages[rdata->nr_pages++] = page;
3387 }
3388
3389 rc = cifs_retry_async_readv(rdata);
3390 if (rc != 0) {
3391 for (i = 0; i < rdata->nr_pages; i++) {
3392 page = rdata->pages[i];
3393 lru_cache_add_file(page);
3394 unlock_page(page);
3395 page_cache_release(page);
3396 }
3397 kref_put(&rdata->refcount, cifs_readdata_release);
3398 break;
3399 }
3400
3401 kref_put(&rdata->refcount, cifs_readdata_release);
3402 }
3403
3404 /* Any pages that have been shown to fscache but didn't get added to
3405 * the pagecache must be uncached before they get returned to the
3406 * allocator.
3407 */
3408 cifs_fscache_readpages_cancel(mapping->host, page_list);
3409 return rc;
3410 }
3411
3412 /*
3413 * cifs_readpage_worker must be called with the page pinned
3414 */
3415 static int cifs_readpage_worker(struct file *file, struct page *page,
3416 loff_t *poffset)
3417 {
3418 char *read_data;
3419 int rc;
3420
3421 /* Is the page cached? */
3422 rc = cifs_readpage_from_fscache(file_inode(file), page);
3423 if (rc == 0)
3424 goto read_complete;
3425
3426 read_data = kmap(page);
3427 /* for reads over a certain size could initiate async read ahead */
3428
3429 rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
3430
3431 if (rc < 0)
3432 goto io_error;
3433 else
3434 cifs_dbg(FYI, "Bytes read %d\n", rc);
3435
3436 file_inode(file)->i_atime =
3437 current_fs_time(file_inode(file)->i_sb);
3438
3439 if (PAGE_CACHE_SIZE > rc)
3440 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
3441
3442 flush_dcache_page(page);
3443 SetPageUptodate(page);
3444
3445 /* send this page to the cache */
3446 cifs_readpage_to_fscache(file_inode(file), page);
3447
3448 rc = 0;
3449
3450 io_error:
3451 kunmap(page);
3452 unlock_page(page);
3453
3454 read_complete:
3455 return rc;
3456 }
3457
3458 static int cifs_readpage(struct file *file, struct page *page)
3459 {
3460 loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3461 int rc = -EACCES;
3462 unsigned int xid;
3463
3464 xid = get_xid();
3465
3466 if (file->private_data == NULL) {
3467 rc = -EBADF;
3468 free_xid(xid);
3469 return rc;
3470 }
3471
3472 cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
3473 page, (int)offset, (int)offset);
3474
3475 rc = cifs_readpage_worker(file, page, &offset);
3476
3477 free_xid(xid);
3478 return rc;
3479 }
3480
3481 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3482 {
3483 struct cifsFileInfo *open_file;
3484
3485 spin_lock(&cifs_file_list_lock);
3486 list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3487 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3488 spin_unlock(&cifs_file_list_lock);
3489 return 1;
3490 }
3491 }
3492 spin_unlock(&cifs_file_list_lock);
3493 return 0;
3494 }
3495
3496 /* We do not want to update the file size from server for inodes
3497 open for write - to avoid races with writepage extending
3498 the file - in the future we could consider allowing
3499 refreshing the inode only on increases in the file size
3500 but this is tricky to do without racing with writebehind
3501 page caching in the current Linux kernel design */
3502 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3503 {
3504 if (!cifsInode)
3505 return true;
3506
3507 if (is_inode_writable(cifsInode)) {
3508 /* This inode is open for write at least once */
3509 struct cifs_sb_info *cifs_sb;
3510
3511 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3512 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3513 /* since no page cache to corrupt on directio
3514 we can change size safely */
3515 return true;
3516 }
3517
3518 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3519 return true;
3520
3521 return false;
3522 } else
3523 return true;
3524 }
3525
3526 static int cifs_write_begin(struct file *file, struct address_space *mapping,
3527 loff_t pos, unsigned len, unsigned flags,
3528 struct page **pagep, void **fsdata)
3529 {
3530 int oncethru = 0;
3531 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
3532 loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
3533 loff_t page_start = pos & PAGE_MASK;
3534 loff_t i_size;
3535 struct page *page;
3536 int rc = 0;
3537
3538 cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
3539
3540 start:
3541 page = grab_cache_page_write_begin(mapping, index, flags);
3542 if (!page) {
3543 rc = -ENOMEM;
3544 goto out;
3545 }
3546
3547 if (PageUptodate(page))
3548 goto out;
3549
3550 /*
3551 * If we write a full page it will be up to date, no need to read from
3552 * the server. If the write is short, we'll end up doing a sync write
3553 * instead.
3554 */
3555 if (len == PAGE_CACHE_SIZE)
3556 goto out;
3557
3558 /*
3559 * optimize away the read when we have an oplock, and we're not
3560 * expecting to use any of the data we'd be reading in. That
3561 * is, when the page lies beyond the EOF, or straddles the EOF
3562 * and the write will cover all of the existing data.
3563 */
3564 if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
3565 i_size = i_size_read(mapping->host);
3566 if (page_start >= i_size ||
3567 (offset == 0 && (pos + len) >= i_size)) {
3568 zero_user_segments(page, 0, offset,
3569 offset + len,
3570 PAGE_CACHE_SIZE);
3571 /*
3572 * PageChecked means that the parts of the page
3573 * to which we're not writing are considered up
3574 * to date. Once the data is copied to the
3575 * page, it can be set uptodate.
3576 */
3577 SetPageChecked(page);
3578 goto out;
3579 }
3580 }
3581
3582 if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
3583 /*
3584 * might as well read a page, it is fast enough. If we get
3585 * an error, we don't need to return it. cifs_write_end will
3586 * do a sync write instead since PG_uptodate isn't set.
3587 */
3588 cifs_readpage_worker(file, page, &page_start);
3589 page_cache_release(page);
3590 oncethru = 1;
3591 goto start;
3592 } else {
3593 /* we could try using another file handle if there is one -
3594 but how would we lock it to prevent close of that handle
3595 racing with this read? In any case
3596 this will be written out by write_end so is fine */
3597 }
3598 out:
3599 *pagep = page;
3600 return rc;
3601 }
3602
3603 static int cifs_release_page(struct page *page, gfp_t gfp)
3604 {
3605 if (PagePrivate(page))
3606 return 0;
3607
3608 return cifs_fscache_release_page(page, gfp);
3609 }
3610
3611 static void cifs_invalidate_page(struct page *page, unsigned int offset,
3612 unsigned int length)
3613 {
3614 struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
3615
3616 if (offset == 0 && length == PAGE_CACHE_SIZE)
3617 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
3618 }
3619
3620 static int cifs_launder_page(struct page *page)
3621 {
3622 int rc = 0;
3623 loff_t range_start = page_offset(page);
3624 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
3625 struct writeback_control wbc = {
3626 .sync_mode = WB_SYNC_ALL,
3627 .nr_to_write = 0,
3628 .range_start = range_start,
3629 .range_end = range_end,
3630 };
3631
3632 cifs_dbg(FYI, "Launder page: %p\n", page);
3633
3634 if (clear_page_dirty_for_io(page))
3635 rc = cifs_writepage_locked(page, &wbc);
3636
3637 cifs_fscache_invalidate_page(page, page->mapping->host);
3638 return rc;
3639 }
3640
3641 static int
3642 cifs_pending_writers_wait(void *unused)
3643 {
3644 schedule();
3645 return 0;
3646 }
3647
3648 void cifs_oplock_break(struct work_struct *work)
3649 {
3650 struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
3651 oplock_break);
3652 struct inode *inode = cfile->dentry->d_inode;
3653 struct cifsInodeInfo *cinode = CIFS_I(inode);
3654 struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3655 struct TCP_Server_Info *server = tcon->ses->server;
3656 int rc = 0;
3657
3658 wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
3659 cifs_pending_writers_wait, TASK_UNINTERRUPTIBLE);
3660
3661 server->ops->downgrade_oplock(server, cinode,
3662 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
3663
3664 if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
3665 cifs_has_mand_locks(cinode)) {
3666 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
3667 inode);
3668 cinode->oplock = 0;
3669 }
3670
3671 if (inode && S_ISREG(inode->i_mode)) {
3672 if (CIFS_CACHE_READ(cinode))
3673 break_lease(inode, O_RDONLY);
3674 else
3675 break_lease(inode, O_WRONLY);
3676 rc = filemap_fdatawrite(inode->i_mapping);
3677 if (!CIFS_CACHE_READ(cinode)) {
3678 rc = filemap_fdatawait(inode->i_mapping);
3679 mapping_set_error(inode->i_mapping, rc);
3680 cifs_zap_mapping(inode);
3681 }
3682 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
3683 }
3684
3685 rc = cifs_push_locks(cfile);
3686 if (rc)
3687 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
3688
3689 /*
3690 * releasing stale oplock after recent reconnect of smb session using
3691 * a now incorrect file handle is not a data integrity issue but do
3692 * not bother sending an oplock release if session to server still is
3693 * disconnected since oplock already released by the server
3694 */
3695 if (!cfile->oplock_break_cancelled) {
3696 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
3697 cinode);
3698 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
3699 }
3700 cifs_done_oplock_break(cinode);
3701 }
3702
3703 /*
3704 * The presence of cifs_direct_io() in the address space ops vector
3705 * allowes open() O_DIRECT flags which would have failed otherwise.
3706 *
3707 * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
3708 * so this method should never be called.
3709 *
3710 * Direct IO is not yet supported in the cached mode.
3711 */
3712 static ssize_t
3713 cifs_direct_io(int rw, struct kiocb *iocb, struct iov_iter *iter,
3714 loff_t pos)
3715 {
3716 /*
3717 * FIXME
3718 * Eventually need to support direct IO for non forcedirectio mounts
3719 */
3720 return -EINVAL;
3721 }
3722
3723
3724 const struct address_space_operations cifs_addr_ops = {
3725 .readpage = cifs_readpage,
3726 .readpages = cifs_readpages,
3727 .writepage = cifs_writepage,
3728 .writepages = cifs_writepages,
3729 .write_begin = cifs_write_begin,
3730 .write_end = cifs_write_end,
3731 .set_page_dirty = __set_page_dirty_nobuffers,
3732 .releasepage = cifs_release_page,
3733 .direct_IO = cifs_direct_io,
3734 .invalidatepage = cifs_invalidate_page,
3735 .launder_page = cifs_launder_page,
3736 };
3737
3738 /*
3739 * cifs_readpages requires the server to support a buffer large enough to
3740 * contain the header plus one complete page of data. Otherwise, we need
3741 * to leave cifs_readpages out of the address space operations.
3742 */
3743 const struct address_space_operations cifs_addr_ops_smallbuf = {
3744 .readpage = cifs_readpage,
3745 .writepage = cifs_writepage,
3746 .writepages = cifs_writepages,
3747 .write_begin = cifs_write_begin,
3748 .write_end = cifs_write_end,
3749 .set_page_dirty = __set_page_dirty_nobuffers,
3750 .releasepage = cifs_release_page,
3751 .invalidatepage = cifs_invalidate_page,
3752 .launder_page = cifs_launder_page,
3753 };
A mirror of Linus' kernel repository