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)
37 struct ptlrpc_request *req = NULL;
38 struct ll_file_data *fd;
39 struct obdo *oa = NULL;
40 struct lov_stripe_md *lsm = NULL;
41 struct ll_sb_info *sbi = ll_i2sbi(inode);
42 struct ll_inode_info *lli = ll_i2info(inode);
45 if (file->private_data)
50 /* delayed create of object (intent created inode) */
51 /* XXX object needs to be cleaned up if mdc_open fails */
52 /* XXX error handling appropriate here? */
54 struct inode *inode = file->f_dentry->d_inode;
56 down(&lli->lli_open_sem);
57 /* Check to see if we lost the race */
58 if (lli->lli_smd == NULL) {
61 up(&lli->lli_open_sem);
64 oa->o_mode = S_IFREG | 0600;
65 oa->o_easize = ll_mds_easize(inode->i_sb);
66 oa->o_id = inode->i_ino;
67 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
68 OBD_MD_FLMODE | OBD_MD_FLEASIZE;
69 rc = obd_create(ll_i2obdconn(inode), oa, &lli->lli_smd);
73 up(&lli->lli_open_sem);
77 if (lli->lli_smd->lsm_object_id == 0)
79 up(&lli->lli_open_sem);
83 fd = kmem_cache_alloc(ll_file_data_slab, SLAB_KERNEL);
85 GOTO(out, rc = -ENOMEM);
86 memset(fd, 0, sizeof(*fd));
88 rc = mdc_open(&sbi->ll_mdc_conn, inode->i_ino, S_IFREG | inode->i_mode,
89 file->f_flags, lsm, (__u64)(unsigned long)file,
90 &fd->fd_mdshandle, &req);
92 ptlrpc_req_finished(req);
94 GOTO(out_req, -abs(rc));
95 if (!fd->fd_mdshandle) {
96 CERROR("mdc_open didn't assign fd_mdshandle\n");
97 /* XXX handle this how, abort or is it non-fatal? */
100 if (oa == NULL && (oa = obdo_alloc()) == NULL)
101 GOTO(out_mdc, rc = -EINVAL);
103 oa->o_id = lsm->lsm_object_id;
104 oa->o_mode = S_IFREG;
105 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE;
106 rc = obd_open(ll_i2obdconn(inode), oa, lsm);
111 GOTO(out_mdc, rc = -abs(rc));
113 file->private_data = fd;
119 mdc_close(&sbi->ll_mdc_conn, inode->i_ino,
120 S_IFREG, fd->fd_mdshandle, &req);
122 ptlrpc_free_req(req);
125 kmem_cache_free(ll_file_data_slab, fd);
126 file->private_data = NULL;
131 int ll_size_lock(struct inode *inode, struct lov_stripe_md *lsm, __u64 start,
132 int mode, struct lustre_handle **lockhs_p)
134 struct ll_sb_info *sbi = ll_i2sbi(inode);
135 struct ldlm_extent extent;
136 struct lustre_handle *lockhs = NULL;
137 int rc, flags = 0, stripe_count;
140 if (sbi->ll_flags & LL_SBI_NOLCK) {
145 stripe_count = lsm->lsm_stripe_count;
149 OBD_ALLOC(lockhs, stripe_count * sizeof(*lockhs));
153 extent.start = start;
156 rc = obd_enqueue(&sbi->ll_osc_conn, lsm, NULL, LDLM_EXTENT, &extent,
157 sizeof(extent), mode, &flags, ll_lock_callback,
158 inode, sizeof(*inode), lockhs);
159 if (rc != ELDLM_OK) {
160 CERROR("lock enqueue: %d\n", rc);
161 OBD_FREE(lockhs, stripe_count * sizeof(*lockhs));
167 int ll_size_unlock(struct inode *inode, struct lov_stripe_md *lsm, int mode,
168 struct lustre_handle *lockhs)
170 struct ll_sb_info *sbi = ll_i2sbi(inode);
171 int rc, stripe_count;
174 if (sbi->ll_flags & LL_SBI_NOLCK)
177 if (lockhs == NULL) {
182 rc = obd_cancel(&sbi->ll_osc_conn, lsm, mode, lockhs);
183 if (rc != ELDLM_OK) {
184 CERROR("lock cancel: %d\n", rc);
188 stripe_count = lsm->lsm_stripe_count;
192 OBD_FREE(lockhs, stripe_count * sizeof(*lockhs));
196 int ll_file_size(struct inode *inode, struct lov_stripe_md *lsm)
198 struct ll_sb_info *sbi = ll_i2sbi(inode);
199 struct lustre_handle *lockhs;
207 rc = ll_size_lock(inode, lsm, 0, LCK_PR, &lockhs);
208 if (rc != ELDLM_OK) {
209 CERROR("lock enqueue: %d\n", rc);
213 oa.o_id = lsm->lsm_object_id;
215 oa.o_valid = OBD_MD_FLID|OBD_MD_FLTYPE|OBD_MD_FLSIZE|OBD_MD_FLBLOCKS;
216 rc = obd_getattr(&sbi->ll_osc_conn, &oa, lsm);
218 obdo_to_inode(inode, &oa,
219 oa.o_valid & ~(OBD_MD_FLTYPE | OBD_MD_FLMODE));
221 err = ll_size_unlock(inode, lsm, LCK_PR, lockhs);
222 if (err != ELDLM_OK) {
223 CERROR("lock cancel: %d\n", err);
229 static int ll_file_release(struct inode *inode, struct file *file)
232 struct ptlrpc_request *req = NULL;
233 struct ll_file_data *fd;
235 struct ll_sb_info *sbi = ll_i2sbi(inode);
236 struct ll_inode_info *lli = ll_i2info(inode);
240 fd = (struct ll_file_data *)file->private_data;
241 if (!fd || !fd->fd_mdshandle) {
243 GOTO(out, rc = -EINVAL);
246 memset(&oa, 0, sizeof(oa));
247 oa.o_id = lli->lli_smd->lsm_object_id;
249 oa.o_valid = OBD_MD_FLTYPE | OBD_MD_FLID;
250 rc = obd_close(ll_i2obdconn(inode), &oa, lli->lli_smd);
252 GOTO(out_fd, abs(rc));
254 /* If this fails and we goto out_fd, the file size on the MDS is out of
255 * date. Is that a big deal? */
256 if (file->f_mode & FMODE_WRITE) {
257 struct lustre_handle *lockhs;
259 rc = ll_size_lock(inode, lli->lli_smd, 0, LCK_PR, &lockhs);
261 GOTO(out_fd, abs(rc));
263 oa.o_id = lli->lli_smd->lsm_object_id;
265 oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE |
267 rc = obd_getattr(&sbi->ll_osc_conn, &oa, lli->lli_smd);
270 attr.ia_valid = (ATTR_MTIME | ATTR_CTIME | ATTR_ATIME |
272 attr.ia_mtime = inode->i_mtime;
273 attr.ia_ctime = inode->i_ctime;
274 attr.ia_atime = inode->i_atime;
275 attr.ia_size = oa.o_size;
277 inode->i_blocks = oa.o_blocks;
279 /* XXX: this introduces a small race that we should
281 rc = ll_inode_setattr(inode, &attr, 0);
283 CERROR("failed - %d.\n", rc);
284 rc = -EIO; /* XXX - GOTO(out)? -phil */
287 rc = ll_size_unlock(inode, lli->lli_smd, LCK_PR, lockhs);
289 CERROR("lock cancel: %d\n", rc);
294 rc = mdc_close(&sbi->ll_mdc_conn, inode->i_ino,
295 S_IFREG, fd->fd_mdshandle, &req);
296 ptlrpc_req_finished(req);
302 ptlrpc_req_finished(fd->fd_req);
304 rc = obd_cancel_unused(ll_i2obdconn(inode), lli->lli_smd, 0);
306 CERROR("obd_cancel_unused: %d\n", rc);
311 kmem_cache_free(ll_file_data_slab, fd);
312 file->private_data = NULL;
318 static inline void ll_remove_suid(struct inode *inode)
322 /* set S_IGID if S_IXGRP is set, and always set S_ISUID */
323 mode = (inode->i_mode & S_IXGRP)*(S_ISGID/S_IXGRP) | S_ISUID;
325 /* was any of the uid bits set? */
326 mode &= inode->i_mode;
327 if (mode && !capable(CAP_FSETID)) {
328 inode->i_mode &= ~mode;
329 // XXX careful here - we cannot change the size
333 static void ll_update_atime(struct inode *inode)
337 attr.ia_atime = CURRENT_TIME;
338 attr.ia_valid = ATTR_ATIME;
340 if (inode->i_atime == attr.ia_atime) return;
341 if (IS_RDONLY(inode)) return;
342 if (IS_NOATIME(inode)) return;
344 /* ll_inode_setattr() sets inode->i_atime from attr.ia_atime */
345 ll_inode_setattr(inode, &attr, 0);
348 int ll_lock_callback(struct ldlm_lock *lock, struct ldlm_lock_desc *new,
349 void *data, __u32 data_len, int flag)
351 struct inode *inode = data;
352 struct lustre_handle lockh;
356 if (data_len != sizeof(struct inode))
363 case LDLM_CB_BLOCKING:
364 ldlm_lock2handle(lock, &lockh);
365 rc = ldlm_cli_cancel(&lockh);
367 CERROR("ldlm_cli_cancel failed: %d\n", rc);
369 case LDLM_CB_CANCELING:
371 CDEBUG(D_INODE, "invalidating obdo/inode %ld\n", inode->i_ino);
372 /* FIXME: do something better than throwing away everything */
373 invalidate_inode_pages(inode);
383 static ssize_t ll_file_read(struct file *filp, char *buf, size_t count,
386 struct ll_file_data *fd = (struct ll_file_data *)filp->private_data;
387 struct inode *inode = filp->f_dentry->d_inode;
388 struct ll_sb_info *sbi = ll_i2sbi(inode);
389 struct ldlm_extent extent;
390 struct lustre_handle *lockhs = NULL;
391 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
397 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
398 sbi->ll_flags & LL_SBI_NOLCK) {
399 OBD_ALLOC(lockhs, lsm->lsm_stripe_count * sizeof(*lockhs));
403 extent.start = *ppos;
404 extent.end = *ppos + count;
405 CDEBUG(D_INFO, "Locking inode %ld, start "LPU64" end "LPU64"\n",
406 inode->i_ino, extent.start, extent.end);
408 err = obd_enqueue(&sbi->ll_osc_conn, lsm, NULL, LDLM_EXTENT,
409 &extent, sizeof(extent), LCK_PR, &flags,
410 ll_lock_callback, inode, sizeof(*inode),
412 if (err != ELDLM_OK) {
413 OBD_FREE(lockhs, lsm->lsm_stripe_count*sizeof(*lockhs));
414 CERROR("lock enqueue: err: %d\n", err);
419 CDEBUG(D_INFO, "Reading inode %ld, %d bytes, offset %Ld\n",
420 inode->i_ino, count, *ppos);
421 retval = generic_file_read(filp, buf, count, ppos);
424 ll_update_atime(inode);
426 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
427 sbi->ll_flags & LL_SBI_NOLCK) {
428 err = obd_cancel(&sbi->ll_osc_conn, lsm, LCK_PR, lockhs);
429 if (err != ELDLM_OK) {
430 CERROR("lock cancel: err: %d\n", err);
436 OBD_FREE(lockhs, lsm->lsm_stripe_count * sizeof(*lockhs));
441 * Write to a file (through the page cache).
444 ll_file_write(struct file *file, const char *buf, size_t count, loff_t *ppos)
446 struct ll_file_data *fd = (struct ll_file_data *)file->private_data;
447 struct inode *inode = file->f_dentry->d_inode;
448 struct ll_sb_info *sbi = ll_i2sbi(inode);
449 struct ldlm_extent extent;
450 struct lustre_handle *lockhs = NULL, *eof_lockhs = NULL;
451 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
457 if (!S_ISBLK(inode->i_mode) && file->f_flags & O_APPEND) {
459 err = ll_size_lock(inode, lsm, 0, LCK_PW, &eof_lockhs);
463 oa.o_id = lsm->lsm_object_id;
464 oa.o_mode = inode->i_mode;
465 oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE |
467 retval = obd_getattr(&sbi->ll_osc_conn, &oa, lsm);
469 GOTO(out_eof, retval);
472 obdo_to_inode(inode, &oa, oa.o_valid);
475 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
476 sbi->ll_flags & LL_SBI_NOLCK) {
477 OBD_ALLOC(lockhs, lsm->lsm_stripe_count * sizeof(*lockhs));
479 GOTO(out_eof, retval = -ENOMEM);
480 extent.start = *ppos;
481 extent.end = *ppos + count;
482 CDEBUG(D_INFO, "Locking inode %ld, start "LPU64" end "LPU64"\n",
483 inode->i_ino, extent.start, extent.end);
485 err = obd_enqueue(&sbi->ll_osc_conn, lsm, NULL, LDLM_EXTENT,
486 &extent, sizeof(extent), LCK_PW, &flags,
487 ll_lock_callback, inode, sizeof(*inode),
489 if (err != ELDLM_OK) {
490 CERROR("lock enqueue: err: %d\n", err);
491 GOTO(out_free, retval = err);
495 CDEBUG(D_INFO, "Writing inode %ld, %ld bytes, offset "LPD64"\n",
496 inode->i_ino, (long)count, *ppos);
498 retval = generic_file_write(file, buf, count, ppos);
500 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
501 sbi->ll_flags & LL_SBI_NOLCK) {
502 err = obd_cancel(&sbi->ll_osc_conn, lsm, LCK_PW, lockhs);
503 if (err != ELDLM_OK) {
504 CERROR("lock cancel: err: %d\n", err);
505 GOTO(out_free, retval = err);
512 OBD_FREE(lockhs, lsm->lsm_stripe_count * sizeof(*lockhs));
515 if (!S_ISBLK(inode->i_mode) && file->f_flags & O_APPEND) {
516 err = ll_size_unlock(inode, lsm, LCK_PW, eof_lockhs);
524 int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
527 struct ll_file_data *fd = (struct ll_file_data *)file->private_data;
531 case LL_IOC_GETFLAGS:
532 /* Get the current value of the file flags */
533 return put_user(fd->fd_flags, (int *)arg);
534 case LL_IOC_SETFLAGS:
535 case LL_IOC_CLRFLAGS:
536 /* Set or clear specific file flags */
537 /* XXX This probably needs checks to ensure the flags are
538 * not abused, and to handle any flag side effects.
540 if (get_user(flags, (int *) arg))
543 if (cmd == LL_IOC_SETFLAGS)
544 fd->fd_flags |= flags;
546 fd->fd_flags &= ~flags;
549 /* We need to special case any other ioctls we want to handle,
550 * to send them to the MDS/OST as appropriate and to properly
551 * network encode the arg field.
552 case EXT2_IOC_GETFLAGS:
553 case EXT2_IOC_SETFLAGS:
554 case EXT2_IOC_GETVERSION_OLD:
555 case EXT2_IOC_GETVERSION_NEW:
556 case EXT2_IOC_SETVERSION_OLD:
557 case EXT2_IOC_SETVERSION_NEW:
564 loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
566 struct inode *inode = file->f_dentry->d_inode;
572 struct ll_inode_info *lli = ll_i2info(inode);
574 retval = ll_file_size(inode, lli->lli_smd);
578 offset += inode->i_size;
582 offset += file->f_pos;
585 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
586 if (offset != file->f_pos) {
587 file->f_pos = offset;
589 file->f_version = ++event;
596 /* XXX this does not need to do anything for data, it _does_ need to
598 int ll_fsync(struct file *file, struct dentry *dentry, int data)
603 static int ll_inode_revalidate(struct dentry *dentry)
605 struct inode *inode = dentry->d_inode;
611 RETURN(ll_file_size(inode, ll_i2info(inode)->lli_smd));
614 struct file_operations ll_file_operations = {
616 write: ll_file_write,
617 ioctl: ll_file_ioctl,
619 release: ll_file_release,
620 mmap: generic_file_mmap,
621 llseek: ll_file_seek,
625 struct inode_operations ll_file_inode_operations = {
626 truncate: ll_truncate,
628 revalidate: ll_inode_revalidate