1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * This code is issued under the GNU General Public License.
7 * See the file COPYING in this distribution
9 * Copyright (C) 1992, 1993, 1994, 1995
10 * Remy Card (card@masi.ibp.fr)
11 * Laboratoire MASI - Institut Blaise Pascal
12 * Universite Pierre et Marie Curie (Paris VI)
16 * linux/fs/minix/file.c
18 * Copyright (C) 1991, 1992 Linus Torvalds
20 * ext2 fs regular file handling primitives
22 * 64-bit file support on 64-bit platforms by Jakub Jelinek
23 * (jj@sunsite.ms.mff.cuni.cz)
26 #define DEBUG_SUBSYSTEM S_LLITE
28 #include <linux/lustre_dlm.h>
29 #include <linux/lustre_lite.h>
31 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc);
32 extern int ll_setattr(struct dentry *de, struct iattr *attr);
34 static int ll_file_open(struct inode *inode, struct file *file)
36 struct ptlrpc_request *req = NULL;
37 struct ll_file_data *fd;
38 struct obdo *oa = NULL;
39 struct lov_stripe_md *lsm = NULL;
40 struct ll_sb_info *sbi = ll_i2sbi(inode);
41 struct ll_inode_info *lli = ll_i2info(inode);
45 if (file->private_data)
48 CHECK_MOUNT_EPOCH(inode);
52 /* delayed create of object (intent created inode) */
53 /* XXX object needs to be cleaned up if mdc_open fails */
54 /* XXX error handling appropriate here? */
56 struct inode *inode = file->f_dentry->d_inode;
58 down(&lli->lli_open_sem);
59 /* Check to see if we lost the race */
60 if (lli->lli_smd == NULL) {
63 up(&lli->lli_open_sem);
66 oa->o_mode = S_IFREG | 0600;
67 oa->o_easize = ll_mds_easize(inode->i_sb);
68 oa->o_id = inode->i_ino;
69 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
70 OBD_MD_FLMODE | OBD_MD_FLEASIZE;
71 rc = obd_create(ll_i2obdconn(inode), oa, &lli->lli_smd);
75 up(&lli->lli_open_sem);
79 if (lli->lli_smd->lsm_object_id == 0)
81 up(&lli->lli_open_sem);
85 fd = kmem_cache_alloc(ll_file_data_slab, SLAB_KERNEL);
87 GOTO(out, rc = -ENOMEM);
88 memset(fd, 0, sizeof(*fd));
90 rc = mdc_open(&sbi->ll_mdc_conn, inode->i_ino, S_IFREG | inode->i_mode,
91 file->f_flags, lsm, (__u64)(unsigned long)file,
92 &fd->fd_mdshandle, &req);
94 ptlrpc_req_finished(req);
96 GOTO(out_req, -abs(rc));
97 if (!fd->fd_mdshandle) {
98 CERROR("mdc_open didn't assign fd_mdshandle\n");
99 /* XXX handle this how, abort or is it non-fatal? */
105 GOTO(out_mdc, rc = -EINVAL);
107 oa->o_id = lsm->lsm_object_id;
108 oa->o_mode = S_IFREG;
109 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE;
110 rc = obd_open(ll_i2obdconn(inode), oa, lsm);
113 GOTO(out_mdc, rc = -abs(rc));
117 file->private_data = fd;
123 mdc_close(&sbi->ll_mdc_conn, inode->i_ino,
124 S_IFREG, fd->fd_mdshandle, &req);
126 ptlrpc_free_req(req);
129 kmem_cache_free(ll_file_data_slab, fd);
130 file->private_data = NULL;
135 int ll_size_lock(struct inode *inode, struct lov_stripe_md *lsm, __u64 start,
136 int mode, struct lustre_handle **lockhs_p)
138 struct ll_sb_info *sbi = ll_i2sbi(inode);
139 struct ldlm_extent extent;
140 struct lustre_handle *lockhs = NULL;
141 int rc, flags = 0, stripe_count;
144 if (sbi->ll_flags & LL_SBI_NOLCK) {
149 stripe_count = lsm->lsm_stripe_count;
153 OBD_ALLOC(lockhs, stripe_count * sizeof(*lockhs));
157 extent.start = start;
160 rc = obd_enqueue(&sbi->ll_osc_conn, lsm, NULL, LDLM_EXTENT, &extent,
161 sizeof(extent), mode, &flags, ll_lock_callback,
162 inode, sizeof(*inode), lockhs);
163 if (rc != ELDLM_OK) {
164 CERROR("lock enqueue: %d\n", rc);
165 OBD_FREE(lockhs, stripe_count * sizeof(*lockhs));
171 int ll_size_unlock(struct inode *inode, struct lov_stripe_md *lsm, int mode,
172 struct lustre_handle *lockhs)
174 struct ll_sb_info *sbi = ll_i2sbi(inode);
175 int rc, stripe_count;
178 if (sbi->ll_flags & LL_SBI_NOLCK)
181 if (lockhs == NULL) {
186 rc = obd_cancel(&sbi->ll_osc_conn, lsm, mode, lockhs);
187 if (rc != ELDLM_OK) {
188 CERROR("lock cancel: %d\n", rc);
192 stripe_count = lsm->lsm_stripe_count;
196 OBD_FREE(lockhs, stripe_count * sizeof(*lockhs));
200 int ll_file_size(struct inode *inode, struct lov_stripe_md *lsm)
202 struct ll_sb_info *sbi = ll_i2sbi(inode);
203 struct lustre_handle *lockhs;
211 rc = ll_size_lock(inode, lsm, 0, LCK_PR, &lockhs);
212 if (rc != ELDLM_OK) {
213 CERROR("lock enqueue: %d\n", rc);
217 oa.o_id = lsm->lsm_object_id;
219 oa.o_valid = OBD_MD_FLID|OBD_MD_FLTYPE|OBD_MD_FLSIZE|OBD_MD_FLBLOCKS;
220 rc = obd_getattr(&sbi->ll_osc_conn, &oa, lsm);
222 obdo_to_inode(inode, &oa,
223 oa.o_valid & ~(OBD_MD_FLTYPE | OBD_MD_FLMODE));
225 err = ll_size_unlock(inode, lsm, LCK_PR, lockhs);
226 if (err != ELDLM_OK) {
227 CERROR("lock cancel: %d\n", err);
233 static int ll_file_release(struct inode *inode, struct file *file)
236 struct ptlrpc_request *req = NULL;
237 struct ll_file_data *fd;
239 struct ll_sb_info *sbi = ll_i2sbi(inode);
240 struct ll_inode_info *lli = ll_i2info(inode);
244 CHECK_MOUNT_EPOCH(inode);
246 fd = (struct ll_file_data *)file->private_data;
247 if (!fd || !fd->fd_mdshandle) {
249 GOTO(out, rc = -EINVAL);
252 memset(&oa, 0, sizeof(oa));
253 oa.o_id = lli->lli_smd->lsm_object_id;
255 oa.o_valid = OBD_MD_FLTYPE | OBD_MD_FLID;
256 rc = obd_close(ll_i2obdconn(inode), &oa, lli->lli_smd);
258 GOTO(out_fd, abs(rc));
260 /* If this fails and we goto out_fd, the file size on the MDS is out of
261 * date. Is that a big deal? */
262 if (file->f_mode & FMODE_WRITE) {
263 struct lustre_handle *lockhs;
265 rc = ll_size_lock(inode, lli->lli_smd, 0, LCK_PR, &lockhs);
267 GOTO(out_fd, abs(rc));
269 oa.o_id = lli->lli_smd->lsm_object_id;
271 oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE |
273 rc = obd_getattr(&sbi->ll_osc_conn, &oa, lli->lli_smd);
276 attr.ia_valid = (ATTR_MTIME | ATTR_CTIME | ATTR_ATIME |
278 attr.ia_mtime = inode->i_mtime;
279 attr.ia_ctime = inode->i_ctime;
280 attr.ia_atime = inode->i_atime;
281 attr.ia_size = oa.o_size;
283 inode->i_blocks = oa.o_blocks;
285 /* XXX: this introduces a small race that we should
287 rc = ll_inode_setattr(inode, &attr, 0);
289 CERROR("failed - %d.\n", rc);
290 rc = -EIO; /* XXX - GOTO(out)? -phil */
293 rc = ll_size_unlock(inode, lli->lli_smd, LCK_PR, lockhs);
295 CERROR("lock cancel: %d\n", rc);
300 rc = mdc_close(&sbi->ll_mdc_conn, inode->i_ino,
301 S_IFREG, fd->fd_mdshandle, &req);
302 ptlrpc_req_finished(req);
308 ptlrpc_req_finished(fd->fd_req);
310 rc = obd_cancel_unused(ll_i2obdconn(inode), lli->lli_smd, 0);
312 CERROR("obd_cancel_unused: %d\n", rc);
317 kmem_cache_free(ll_file_data_slab, fd);
318 file->private_data = NULL;
324 static inline void ll_remove_suid(struct inode *inode)
328 /* set S_IGID if S_IXGRP is set, and always set S_ISUID */
329 mode = (inode->i_mode & S_IXGRP)*(S_ISGID/S_IXGRP) | S_ISUID;
331 /* was any of the uid bits set? */
332 mode &= inode->i_mode;
333 if (mode && !capable(CAP_FSETID)) {
334 inode->i_mode &= ~mode;
335 // XXX careful here - we cannot change the size
339 static void ll_update_atime(struct inode *inode)
343 attr.ia_atime = CURRENT_TIME;
344 attr.ia_valid = ATTR_ATIME;
346 if (inode->i_atime == attr.ia_atime) return;
347 if (IS_RDONLY(inode)) return;
348 if (IS_NOATIME(inode)) return;
350 /* ll_inode_setattr() sets inode->i_atime from attr.ia_atime */
351 ll_inode_setattr(inode, &attr, 0);
354 int ll_lock_callback(struct ldlm_lock *lock, struct ldlm_lock_desc *new,
355 void *data, __u32 data_len, int flag)
357 struct inode *inode = data;
358 struct lustre_handle lockh;
362 if (data_len != sizeof(struct inode))
369 case LDLM_CB_BLOCKING:
370 ldlm_lock2handle(lock, &lockh);
371 rc = ldlm_cli_cancel(&lockh);
373 CERROR("ldlm_cli_cancel failed: %d\n", rc);
375 case LDLM_CB_CANCELING:
376 CDEBUG(D_INODE, "invalidating obdo/inode %ld\n", inode->i_ino);
377 /* FIXME: do something better than throwing away everything */
378 //down(&inode->i_sem);
379 invalidate_inode_pages(inode);
389 static ssize_t ll_file_read(struct file *filp, char *buf, size_t count,
392 struct ll_file_data *fd = (struct ll_file_data *)filp->private_data;
393 struct inode *inode = filp->f_dentry->d_inode;
394 struct ll_sb_info *sbi = ll_i2sbi(inode);
395 struct ldlm_extent extent;
396 struct lustre_handle *lockhs = NULL;
397 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
403 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
404 sbi->ll_flags & LL_SBI_NOLCK) {
405 OBD_ALLOC(lockhs, lsm->lsm_stripe_count * sizeof(*lockhs));
409 extent.start = *ppos;
410 extent.end = *ppos + count;
411 CDEBUG(D_INFO, "Locking inode %ld, start "LPU64" end "LPU64"\n",
412 inode->i_ino, extent.start, extent.end);
414 err = obd_enqueue(&sbi->ll_osc_conn, lsm, NULL, LDLM_EXTENT,
415 &extent, sizeof(extent), LCK_PR, &flags,
416 ll_lock_callback, inode, sizeof(*inode),
418 if (err != ELDLM_OK) {
419 OBD_FREE(lockhs, lsm->lsm_stripe_count*sizeof(*lockhs));
420 CERROR("lock enqueue: err: %d\n", err);
425 CDEBUG(D_INFO, "Reading inode %ld, %d bytes, offset %Ld\n",
426 inode->i_ino, count, *ppos);
427 retval = generic_file_read(filp, buf, count, ppos);
430 ll_update_atime(inode);
432 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
433 sbi->ll_flags & LL_SBI_NOLCK) {
434 err = obd_cancel(&sbi->ll_osc_conn, lsm, LCK_PR, lockhs);
435 if (err != ELDLM_OK) {
436 CERROR("lock cancel: err: %d\n", err);
442 OBD_FREE(lockhs, lsm->lsm_stripe_count * sizeof(*lockhs));
447 * Write to a file (through the page cache).
450 ll_file_write(struct file *file, const char *buf, size_t count, loff_t *ppos)
452 struct ll_file_data *fd = (struct ll_file_data *)file->private_data;
453 struct inode *inode = file->f_dentry->d_inode;
454 struct ll_sb_info *sbi = ll_i2sbi(inode);
455 struct ldlm_extent extent;
456 struct lustre_handle *lockhs = NULL, *eof_lockhs = NULL;
457 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
463 if (!S_ISBLK(inode->i_mode) && file->f_flags & O_APPEND) {
470 err = ll_size_lock(inode, lsm, 0, LCK_PW, &eof_lockhs);
476 oa->o_id = lsm->lsm_object_id;
477 oa->o_mode = inode->i_mode;
478 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE |
480 retval = obd_getattr(&sbi->ll_osc_conn, oa, lsm);
483 GOTO(out_eof, retval);
487 obdo_to_inode(inode, oa, oa->o_valid);
491 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
492 sbi->ll_flags & LL_SBI_NOLCK) {
493 OBD_ALLOC(lockhs, lsm->lsm_stripe_count * sizeof(*lockhs));
495 GOTO(out_eof, retval = -ENOMEM);
496 extent.start = *ppos;
497 extent.end = *ppos + count;
498 CDEBUG(D_INFO, "Locking inode %ld, start "LPU64" end "LPU64"\n",
499 inode->i_ino, extent.start, extent.end);
501 err = obd_enqueue(&sbi->ll_osc_conn, lsm, NULL, LDLM_EXTENT,
502 &extent, sizeof(extent), LCK_PW, &flags,
503 ll_lock_callback, inode, sizeof(*inode),
505 if (err != ELDLM_OK) {
506 CERROR("lock enqueue: err: %d\n", err);
507 GOTO(out_free, retval = err);
511 CDEBUG(D_INFO, "Writing inode %ld, %ld bytes, offset "LPD64"\n",
512 inode->i_ino, (long)count, *ppos);
514 retval = generic_file_write(file, buf, count, ppos);
516 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
517 sbi->ll_flags & LL_SBI_NOLCK) {
518 err = obd_cancel(&sbi->ll_osc_conn, lsm, LCK_PW, lockhs);
519 if (err != ELDLM_OK) {
520 CERROR("lock cancel: err: %d\n", err);
521 GOTO(out_free, retval = err);
528 OBD_FREE(lockhs, lsm->lsm_stripe_count * sizeof(*lockhs));
531 if (!S_ISBLK(inode->i_mode) && file->f_flags & O_APPEND) {
532 err = ll_size_unlock(inode, lsm, LCK_PW, eof_lockhs);
540 int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
543 struct ll_file_data *fd = (struct ll_file_data *)file->private_data;
547 case LL_IOC_GETFLAGS:
548 /* Get the current value of the file flags */
549 return put_user(fd->fd_flags, (int *)arg);
550 case LL_IOC_SETFLAGS:
551 case LL_IOC_CLRFLAGS:
552 /* Set or clear specific file flags */
553 /* XXX This probably needs checks to ensure the flags are
554 * not abused, and to handle any flag side effects.
556 if (get_user(flags, (int *) arg))
559 if (cmd == LL_IOC_SETFLAGS)
560 fd->fd_flags |= flags;
562 fd->fd_flags &= ~flags;
565 /* We need to special case any other ioctls we want to handle,
566 * to send them to the MDS/OST as appropriate and to properly
567 * network encode the arg field.
568 case EXT2_IOC_GETFLAGS:
569 case EXT2_IOC_SETFLAGS:
570 case EXT2_IOC_GETVERSION_OLD:
571 case EXT2_IOC_GETVERSION_NEW:
572 case EXT2_IOC_SETVERSION_OLD:
573 case EXT2_IOC_SETVERSION_NEW:
580 loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
582 struct inode *inode = file->f_dentry->d_inode;
586 CHECK_MOUNT_EPOCH(inode);
590 struct ll_inode_info *lli = ll_i2info(inode);
592 retval = ll_file_size(inode, lli->lli_smd);
596 offset += inode->i_size;
600 offset += file->f_pos;
603 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
604 if (offset != file->f_pos) {
605 file->f_pos = offset;
607 file->f_version = ++event;
614 /* XXX this does not need to do anything for data, it _does_ need to
616 int ll_fsync(struct file *file, struct dentry *dentry, int data)
621 static int ll_inode_revalidate(struct dentry *dentry)
623 struct inode *inode = dentry->d_inode;
629 RETURN(ll_file_size(inode, ll_i2info(inode)->lli_smd));
632 struct file_operations ll_file_operations = {
634 write: ll_file_write,
635 ioctl: ll_file_ioctl,
637 release: ll_file_release,
638 mmap: generic_file_mmap,
639 llseek: ll_file_seek,
643 struct inode_operations ll_file_inode_operations = {
644 truncate: ll_truncate,
646 revalidate: ll_inode_revalidate