4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lvfs/lvfs_linux.c
38 * Author: Andreas Dilger <adilger@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_FILTER
43 #include <linux/version.h>
45 #include <asm/unistd.h>
46 #include <linux/slab.h>
47 #include <linux/pagemap.h>
48 #include <linux/quotaops.h>
49 #include <linux/version.h>
50 #include <libcfs/libcfs.h>
51 #include <lustre_fsfilt.h>
53 #include <linux/module.h>
54 #include <linux/init.h>
55 #include <linux/lustre_compat25.h>
57 #include "lvfs_internal.h"
60 #include <lustre_lib.h>
61 #include <lustre_quota.h>
63 __u64 obd_max_pages = 0;
64 __u64 obd_max_alloc = 0;
65 struct lprocfs_stats *obd_memory = NULL;
66 EXPORT_SYMBOL(obd_memory);
67 cfs_spinlock_t obd_updatemax_lock = CFS_SPIN_LOCK_UNLOCKED;
68 /* refine later and change to seqlock or simlar from libcfs */
70 /* Debugging check only needed during development */
72 # define ASSERT_CTXT_MAGIC(magic) LASSERT((magic) == OBD_RUN_CTXT_MAGIC)
73 # define ASSERT_NOT_KERNEL_CTXT(msg) LASSERTF(!segment_eq(get_fs(), get_ds()),\
75 # define ASSERT_KERNEL_CTXT(msg) LASSERTF(segment_eq(get_fs(), get_ds()), msg)
77 # define ASSERT_CTXT_MAGIC(magic) do {} while(0)
78 # define ASSERT_NOT_KERNEL_CTXT(msg) do {} while(0)
79 # define ASSERT_KERNEL_CTXT(msg) do {} while(0)
82 static void push_group_info(struct lvfs_run_ctxt *save,
83 struct group_info *ginfo)
86 save->ngroups = current_ngroups;
91 save->group_info = current_cred()->group_info;
92 if ((cred = prepare_creds())) {
93 cred->group_info = ginfo;
100 static void pop_group_info(struct lvfs_run_ctxt *save,
101 struct group_info *ginfo)
104 current_ngroups = save->ngroups;
108 if ((cred = prepare_creds())) {
109 cred->group_info = save->group_info;
112 task_unlock(current);
116 /* push / pop to root of obd store */
117 void push_ctxt(struct lvfs_run_ctxt *save, struct lvfs_run_ctxt *new_ctx,
118 struct lvfs_ucred *uc)
120 //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
121 ASSERT_CTXT_MAGIC(new_ctx->magic);
122 OBD_SET_CTXT_MAGIC(save);
125 LASSERT(cfs_atomic_read(&cfs_fs_pwd(current->fs)->d_count));
126 LASSERT(cfs_atomic_read(&new_ctx->pwd->d_count));
127 save->pwd = dget(cfs_fs_pwd(current->fs));
128 save->pwdmnt = mntget(cfs_fs_mnt(current->fs));
129 save->luc.luc_umask = cfs_curproc_umask();
130 save->ngroups = current_cred()->group_info->ngroups;
133 LASSERT(save->pwdmnt);
134 LASSERT(new_ctx->pwd);
135 LASSERT(new_ctx->pwdmnt);
139 save->luc.luc_uid = current_uid();
140 save->luc.luc_gid = current_gid();
141 save->luc.luc_fsuid = current_fsuid();
142 save->luc.luc_fsgid = current_fsgid();
143 save->luc.luc_cap = current_cap();
145 if ((cred = prepare_creds())) {
146 cred->uid = uc->luc_uid;
147 cred->gid = uc->luc_gid;
148 cred->fsuid = uc->luc_fsuid;
149 cred->fsgid = uc->luc_fsgid;
150 cred->cap_effective = uc->luc_cap;
154 push_group_info(save,
156 uc->luc_identity ? uc->luc_identity->mi_ginfo :
159 current->fs->umask = 0; /* umask already applied on client */
161 ll_set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
163 EXPORT_SYMBOL(push_ctxt);
165 void pop_ctxt(struct lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
166 struct lvfs_ucred *uc)
168 ASSERT_CTXT_MAGIC(saved->magic);
169 ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
171 LASSERTF(cfs_fs_pwd(current->fs) == new_ctx->pwd, "%p != %p\n",
172 cfs_fs_pwd(current->fs), new_ctx->pwd);
173 LASSERTF(cfs_fs_mnt(current->fs) == new_ctx->pwdmnt, "%p != %p\n",
174 cfs_fs_mnt(current->fs), new_ctx->pwdmnt);
177 ll_set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
180 mntput(saved->pwdmnt);
181 current->fs->umask = saved->luc.luc_umask;
184 if ((cred = prepare_creds())) {
185 cred->uid = saved->luc.luc_uid;
186 cred->gid = saved->luc.luc_gid;
187 cred->fsuid = saved->luc.luc_fsuid;
188 cred->fsgid = saved->luc.luc_fsgid;
189 cred->cap_effective = saved->luc.luc_cap;
193 pop_group_info(saved,
195 uc->luc_identity ? uc->luc_identity->mi_ginfo :
199 EXPORT_SYMBOL(pop_ctxt);
201 /* utility to make a file */
202 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
204 struct dentry *dchild;
208 // ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
209 CDEBUG(D_INODE, "creating file %.*s\n", (int)strlen(name), name);
211 dchild = ll_lookup_one_len(name, dir, strlen(name));
213 GOTO(out_up, dchild);
215 if (dchild->d_inode) {
216 int old_mode = dchild->d_inode->i_mode;
217 if (!S_ISREG(old_mode))
218 GOTO(out_err, err = -EEXIST);
220 /* Fixup file permissions if necessary */
221 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
222 CWARN("fixing permissions on %s from %o to %o\n",
223 name, old_mode, mode);
224 dchild->d_inode->i_mode = (mode & S_IALLUGO) |
225 (old_mode & ~S_IALLUGO);
226 mark_inode_dirty(dchild->d_inode);
228 GOTO(out_up, dchild);
231 err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
240 dchild = ERR_PTR(err);
244 EXPORT_SYMBOL(simple_mknod);
246 /* utility to make a directory */
247 struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
248 const char *name, int mode, int fix)
250 struct dentry *dchild;
254 // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
255 CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
256 dchild = ll_lookup_one_len(name, dir, strlen(name));
258 GOTO(out_up, dchild);
260 if (dchild->d_inode) {
261 int old_mode = dchild->d_inode->i_mode;
262 if (!S_ISDIR(old_mode)) {
263 CERROR("found %s (%lu/%u) is mode %o\n", name,
264 dchild->d_inode->i_ino,
265 dchild->d_inode->i_generation, old_mode);
266 GOTO(out_err, err = -ENOTDIR);
269 /* Fixup directory permissions if necessary */
270 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
272 "fixing permissions on %s from %o to %o\n",
273 name, old_mode, mode);
274 dchild->d_inode->i_mode = (mode & S_IALLUGO) |
275 (old_mode & ~S_IALLUGO);
276 mark_inode_dirty(dchild->d_inode);
278 GOTO(out_up, dchild);
281 err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode);
289 dchild = ERR_PTR(err);
293 EXPORT_SYMBOL(simple_mkdir);
295 /* utility to rename a file */
296 int lustre_rename(struct dentry *dir, struct vfsmount *mnt,
297 char *oldname, char *newname)
299 struct dentry *dchild_old, *dchild_new;
303 ASSERT_KERNEL_CTXT("kernel doing rename outside kernel context\n");
304 CDEBUG(D_INODE, "renaming file %.*s to %.*s\n",
305 (int)strlen(oldname), oldname, (int)strlen(newname), newname);
307 dchild_old = ll_lookup_one_len(oldname, dir, strlen(oldname));
308 if (IS_ERR(dchild_old))
309 RETURN(PTR_ERR(dchild_old));
311 if (!dchild_old->d_inode)
312 GOTO(put_old, err = -ENOENT);
314 dchild_new = ll_lookup_one_len(newname, dir, strlen(newname));
315 if (IS_ERR(dchild_new))
316 GOTO(put_old, err = PTR_ERR(dchild_new));
318 err = ll_vfs_rename(dir->d_inode, dchild_old, mnt,
319 dir->d_inode, dchild_new, mnt);
326 EXPORT_SYMBOL(lustre_rename);
329 * Read a file from within kernel context. Prior to calling this
330 * function we should already have done a push_ctxt().
332 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
334 ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
335 if (!file || !file->f_op || !file->f_op->read || !off)
338 return file->f_op->read(file, buf, len, off);
340 EXPORT_SYMBOL(lustre_fread);
343 * Write a file from within kernel context. Prior to calling this
344 * function we should already have done a push_ctxt().
346 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
349 ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
357 if (!file->f_op->write)
360 RETURN(file->f_op->write(file, buf, len, off));
362 EXPORT_SYMBOL(lustre_fwrite);
365 * Sync a file from within kernel context. Prior to calling this
366 * function we should already have done a push_ctxt().
368 int lustre_fsync(struct file *file)
371 ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
372 if (!file || !file->f_op || !file->f_op->fsync)
375 RETURN(cfs_do_fsync(file, 0));
377 EXPORT_SYMBOL(lustre_fsync);
379 /* Note: dput(dchild) will be called if there is an error */
380 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
383 mntget(ctxt->pwdmnt);
384 return ll_dentry_open(de, ctxt->pwdmnt, flags, current_cred());
386 EXPORT_SYMBOL(l_dentry_open);
388 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
389 u64 ino, unsigned int d_type)
391 struct l_linux_dirent *dirent;
392 struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
394 dirent = buf->lrc_dirent;
396 dirent->lld_off = offset;
398 OBD_ALLOC(dirent, sizeof(*dirent));
403 cfs_list_add_tail(&dirent->lld_list, buf->lrc_list);
405 buf->lrc_dirent = dirent;
406 dirent->lld_ino = ino;
407 LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
408 memcpy(dirent->lld_name, name, namlen);
413 long l_readdir(struct file *file, cfs_list_t *dentry_list)
415 struct l_linux_dirent *lastdirent;
416 struct l_readdir_callback buf;
419 buf.lrc_dirent = NULL;
420 buf.lrc_list = dentry_list;
422 error = vfs_readdir(file, l_filldir, &buf);
426 lastdirent = buf.lrc_dirent;
428 lastdirent->lld_off = file->f_pos;
432 EXPORT_SYMBOL(l_readdir);
434 int l_notify_change(struct vfsmount *mnt, struct dentry *dchild,
435 struct iattr *newattrs)
439 LOCK_INODE_MUTEX(dchild->d_inode);
440 #ifdef HAVE_SECURITY_PLUG
441 rc = notify_change(dchild, mnt, newattrs);
443 rc = notify_change(dchild, newattrs);
445 UNLOCK_INODE_MUTEX(dchild->d_inode);
448 EXPORT_SYMBOL(l_notify_change);
450 /* utility to truncate a file */
451 int simple_truncate(struct dentry *dir, struct vfsmount *mnt,
452 char *name, loff_t length)
454 struct dentry *dchild;
455 struct iattr newattrs;
459 CDEBUG(D_INODE, "truncating file %.*s to %lld\n", (int)strlen(name),
460 name, (long long)length);
461 dchild = ll_lookup_one_len(name, dir, strlen(name));
463 GOTO(out, err = PTR_ERR(dchild));
465 if (dchild->d_inode) {
466 int old_mode = dchild->d_inode->i_mode;
467 if (S_ISDIR(old_mode)) {
468 CERROR("found %s (%lu/%u) is mode %o\n", name,
469 dchild->d_inode->i_ino,
470 dchild->d_inode->i_generation, old_mode);
471 GOTO(out_dput, err = -EISDIR);
474 newattrs.ia_size = length;
475 newattrs.ia_valid = ATTR_SIZE;
476 err = l_notify_change(mnt, dchild, &newattrs);
484 EXPORT_SYMBOL(simple_truncate);
486 int __lvfs_set_rdonly(lvfs_sbdev_type dev, lvfs_sbdev_type jdev)
488 #ifdef HAVE_DEV_SET_RDONLY
489 if (jdev && (jdev != dev)) {
490 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
492 dev_set_rdonly(jdev);
494 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
499 CERROR("DEV %lx CANNOT BE SET READONLY\n", (long)dev);
504 EXPORT_SYMBOL(__lvfs_set_rdonly);
506 int lvfs_check_rdonly(lvfs_sbdev_type dev)
508 #ifdef HAVE_DEV_SET_RDONLY
509 return dev_check_rdonly(dev);
514 EXPORT_SYMBOL(lvfs_check_rdonly);
516 int lvfs_check_io_health(struct obd_device *obd, struct file *file)
518 char *write_page = NULL;
523 OBD_ALLOC(write_page, CFS_PAGE_SIZE);
527 rc = fsfilt_write_record(obd, file, write_page, CFS_PAGE_SIZE, &offset, 1);
529 OBD_FREE(write_page, CFS_PAGE_SIZE);
531 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
534 EXPORT_SYMBOL(lvfs_check_io_health);
536 void obd_update_maxusage()
540 max1 = obd_pages_sum();
541 max2 = obd_memory_sum();
543 cfs_spin_lock(&obd_updatemax_lock);
544 if (max1 > obd_max_pages)
545 obd_max_pages = max1;
546 if (max2 > obd_max_alloc)
547 obd_max_alloc = max2;
548 cfs_spin_unlock(&obd_updatemax_lock);
551 EXPORT_SYMBOL(obd_update_maxusage);
553 __u64 obd_memory_max(void)
557 cfs_spin_lock(&obd_updatemax_lock);
559 cfs_spin_unlock(&obd_updatemax_lock);
563 EXPORT_SYMBOL(obd_memory_max);
565 __u64 obd_pages_max(void)
569 cfs_spin_lock(&obd_updatemax_lock);
571 cfs_spin_unlock(&obd_updatemax_lock);
575 EXPORT_SYMBOL(obd_pages_max);
578 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
579 enum lprocfs_fields_flags field)
587 centry = cfs_atomic_read(&lc->lc_cntl.la_entry);
590 case LPROCFS_FIELDS_FLAGS_CONFIG:
593 case LPROCFS_FIELDS_FLAGS_SUM:
594 ret = lc->lc_sum + lc->lc_sum_irq;
596 case LPROCFS_FIELDS_FLAGS_MIN:
599 case LPROCFS_FIELDS_FLAGS_MAX:
602 case LPROCFS_FIELDS_FLAGS_AVG:
603 ret = (lc->lc_max - lc->lc_min)/2;
605 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
606 ret = lc->lc_sumsquare;
608 case LPROCFS_FIELDS_FLAGS_COUNT:
614 } while (centry != cfs_atomic_read(&lc->lc_cntl.la_entry) &&
615 centry != cfs_atomic_read(&lc->lc_cntl.la_exit));
619 EXPORT_SYMBOL(lprocfs_read_helper);
622 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
623 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
624 MODULE_LICENSE("GPL");