1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
5 * Author: Peter Braam <braam@clusterfs.com>
6 * Author: Phil Schwan <phil@clusterfs.com>
7 * Author: Andreas Dilger <adilger@clusterfs.com>
9 * This file is part of Lustre, http://www.lustre.org.
11 * Lustre is free software; you can redistribute it and/or
12 * modify it under the terms of version 2 of the GNU General Public
13 * License as published by the Free Software Foundation.
15 * Lustre 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 the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with Lustre; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <linux/lustre_dlm.h>
27 #include <linux/lustre_lite.h>
28 #include <linux/pagemap.h>
29 #include <linux/file.h>
30 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
31 #include <linux/lustre_compat25.h>
33 #include "llite_internal.h"
34 #include <linux/obd_lov.h>
36 int ll_mdc_close(struct obd_export *lmv_exp, struct inode *inode,
39 struct ll_file_data *fd = file->private_data;
40 struct ptlrpc_request *req = NULL;
41 struct obd_client_handle *och = &fd->fd_mds_och;
46 /* clear group lock, if present */
47 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
48 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
49 fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK);
50 rc = ll_extent_unlock(fd, inode, lsm, LCK_GROUP,
54 obdo.o_id = inode->i_ino;
55 obdo.o_valid = OBD_MD_FLID;
56 obdo_from_inode(&obdo, inode, (OBD_MD_FLTYPE | OBD_MD_FLMODE |
57 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
58 OBD_MD_FLATIME | OBD_MD_FLMTIME |
60 if (0 /* ll_is_inode_dirty(inode) */) {
61 obdo.o_flags = MDS_BFLAG_UNCOMMITTED_WRITES;
62 obdo.o_valid |= OBD_MD_FLFLAGS;
64 obdo.o_mds = id_group(&ll_i2info(inode)->lli_id);
65 rc = md_close(lmv_exp, &obdo, och, &req);
68 /* We are the last writer, so the MDS has instructed us to get
69 * the file size and any write cookies, then close again. */
70 //ll_queue_done_writing(inode);
73 CERROR("inode %lu mdc close failed: rc = %d\n",
77 rc = ll_objects_destroy(req, file->f_dentry->d_inode, 1);
79 CERROR("inode %lu ll_objects destroy: rc = %d\n",
83 mdc_clear_open_replay_data(lmv_exp, och);
84 ptlrpc_req_finished(req);
85 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
86 file->private_data = NULL;
87 OBD_SLAB_FREE(fd, ll_file_data_slab, sizeof *fd);
92 /* While this returns an error code, fput() the caller does not, so we need
93 * to make every effort to clean up all of our state here. Also, applications
94 * rarely check close errors and even if an error is returned they will not
95 * re-try the close call.
97 int ll_file_release(struct inode *inode, struct file *file)
99 struct ll_file_data *fd;
100 struct ll_sb_info *sbi = ll_i2sbi(inode);
104 CDEBUG(D_VFSTRACE, "VFS Op:inode="DLID4"(%p)\n",
105 OLID4(&ll_i2info(inode)->lli_id), inode);
107 /* don't do anything for / */
108 if (inode->i_sb->s_root == file->f_dentry)
111 lprocfs_counter_incr(sbi->ll_stats, LPROC_LL_RELEASE);
112 fd = (struct ll_file_data *)file->private_data;
115 rc = ll_mdc_close(sbi->ll_lmv_exp, inode, file);
119 static int ll_intent_file_open(struct file *file, void *lmm,
120 int lmmsize, struct lookup_intent *itp)
122 struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
123 struct dentry *parent = file->f_dentry->d_parent;
124 const char *name = file->f_dentry->d_name.name;
125 const int len = file->f_dentry->d_name.len;
126 struct lustre_handle lockh;
127 struct mdc_op_data data;
133 ll_prepare_mdc_data(&data, parent->d_inode, NULL, name, len, O_RDWR);
135 rc = md_enqueue(sbi->ll_lmv_exp, LDLM_IBITS, itp, LCK_PR, &data,
136 &lockh, lmm, lmmsize, ldlm_completion_ast,
137 ll_mdc_blocking_ast, NULL);
139 if (itp->d.lustre.it_lock_mode)
140 memcpy(&itp->d.lustre.it_lock_handle,
141 &lockh, sizeof(lockh));
143 CERROR("lock enqueue: err: %d\n", rc);
149 int ll_local_open(struct file *file, struct lookup_intent *it)
151 struct ptlrpc_request *req = it->d.lustre.it_data;
152 struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode);
153 struct obd_export *lmv_exp = ll_i2lmvexp(file->f_dentry->d_inode);
154 struct ll_file_data *fd;
155 struct mds_body *body;
158 body = lustre_msg_buf (req->rq_repmsg, 1, sizeof (*body));
159 LASSERT (body != NULL); /* reply already checked out */
160 LASSERT_REPSWABBED (req, 1); /* and swabbed down */
162 LASSERTF(file->private_data == NULL, "file %*s/%*s ino %lu/%u (%o)\n",
163 file->f_dentry->d_name.len, file->f_dentry->d_name.name,
164 file->f_dentry->d_parent->d_name.len,
165 file->f_dentry->d_parent->d_name.name,
166 file->f_dentry->d_inode->i_ino,
167 file->f_dentry->d_inode->i_generation,
168 file->f_dentry->d_inode->i_mode);
171 OBD_SLAB_ALLOC(fd, ll_file_data_slab, SLAB_KERNEL, sizeof *fd);
173 /* We can't handle this well without reorganizing ll_file_open and
174 * ll_mdc_close, so don't even try right now. */
177 memcpy(&fd->fd_mds_och.och_fh, &body->handle, sizeof(body->handle));
178 fd->fd_mds_och.och_magic = OBD_CLIENT_HANDLE_MAGIC;
179 file->private_data = fd;
180 ll_readahead_init(file->f_dentry->d_inode, &fd->fd_ras);
182 lli->lli_io_epoch = body->io_epoch;
184 mdc_set_open_replay_data(lmv_exp, &fd->fd_mds_och, it->d.lustre.it_data);
189 /* Open a file, and (for the very first open) create objects on the OSTs at
190 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
191 * creation or open until ll_lov_setstripe() ioctl is called. We grab
192 * lli_open_sem to ensure no other process will create objects, send the
193 * stripe MD to the MDS, or try to destroy the objects if that fails.
195 * If we already have the stripe MD locally then we don't request it in
196 * mdc_open(), by passing a lmm_size = 0.
198 * It is up to the application to ensure no other processes open this file
199 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
200 * used. We might be able to avoid races of that sort by getting lli_open_sem
201 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
202 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
204 int ll_file_open(struct inode *inode, struct file *file)
206 struct ll_inode_info *lli = ll_i2info(inode);
207 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
208 .it_flags = file->f_flags };
209 struct lov_stripe_md *lsm;
210 struct ptlrpc_request *req;
214 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), flags %o\n", inode->i_ino,
215 inode->i_generation, inode, file->f_flags);
217 /* don't do anything for / */
218 if (inode->i_sb->s_root == file->f_dentry)
223 if (!it || !it->d.lustre.it_disposition) {
225 rc = ll_intent_file_open(file, NULL, 0, it);
230 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_OPEN);
231 /* mdc_intent_lock() didn't get a request ref if there was an open
232 * error, so don't do cleanup on the request here (bug 3430) */
233 rc = it_open_error(DISP_OPEN_OPEN, it);
237 rc = ll_local_open(file, it);
239 LASSERTF(rc == 0, "rc = %d\n", rc);
241 if (!S_ISREG(inode->i_mode))
246 if (file->f_flags & O_LOV_DELAY_CREATE ||
247 !(file->f_mode & FMODE_WRITE)) {
248 CDEBUG(D_INODE, "object creation was delayed\n");
252 file->f_flags &= ~O_LOV_DELAY_CREATE;
255 req = it->d.lustre.it_data;
256 ptlrpc_req_finished(req);
258 ll_open_complete(inode);
262 /* Fills the obdo with the attributes for the inode defined by lsm */
263 int ll_lsm_getattr(struct obd_export *exp, struct lov_stripe_md *lsm,
266 struct ptlrpc_request_set *set;
270 LASSERT(lsm != NULL);
272 memset(oa, 0, sizeof *oa);
273 oa->o_id = lsm->lsm_object_id;
274 oa->o_gr = lsm->lsm_object_gr;
275 oa->o_mode = S_IFREG;
276 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE |
277 OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
278 OBD_MD_FLCTIME | OBD_MD_FLGROUP;
280 set = ptlrpc_prep_set();
284 rc = obd_getattr_async(exp, oa, lsm, set);
286 rc = ptlrpc_set_wait(set);
287 ptlrpc_set_destroy(set);
292 oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
293 OBD_MD_FLCTIME | OBD_MD_FLSIZE);
297 static inline void ll_remove_suid(struct inode *inode)
301 /* set S_IGID if S_IXGRP is set, and always set S_ISUID */
302 mode = (inode->i_mode & S_IXGRP)*(S_ISGID/S_IXGRP) | S_ISUID;
304 /* was any of the uid bits set? */
305 mode &= inode->i_mode;
306 if (mode && !capable(CAP_FSETID)) {
307 inode->i_mode &= ~mode;
308 // XXX careful here - we cannot change the size
312 static int ll_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock)
314 struct ll_inode_info *lli = ll_i2info(inode);
315 struct lov_stripe_md *lsm = lli->lli_smd;
316 struct obd_export *exp = ll_i2obdexp(inode);
319 struct ldlm_lock *lock;
320 struct lov_stripe_md *lsm;
321 } key = { .name = "lock_to_stripe", .lock = lock, .lsm = lsm };
322 __u32 stripe, vallen = sizeof(stripe);
326 if (lsm->lsm_stripe_count == 1)
327 GOTO(check, stripe = 0);
329 /* get our offset in the lov */
330 rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe);
332 CERROR("obd_get_info: rc = %d\n", rc);
335 LASSERT(stripe < lsm->lsm_stripe_count);
338 if (lsm->lsm_oinfo[stripe].loi_id != lock->l_resource->lr_name.name[0]||
339 lsm->lsm_oinfo[stripe].loi_gr != lock->l_resource->lr_name.name[2]){
340 LDLM_ERROR(lock, "resource doesn't match object "LPU64"/"LPU64
341 " inode=%lu/%u (%p)\n",
342 lsm->lsm_oinfo[stripe].loi_id,
343 lsm->lsm_oinfo[stripe].loi_gr,
344 inode->i_ino, inode->i_generation, inode);
345 RETURN(-ELDLM_NO_LOCK_DATA);
351 /* Flush the page cache for an extent as its canceled. When we're on an LOV,
352 * we get a lock cancellation for each stripe, so we have to map the obd's
353 * region back onto the stripes in the file that it held.
355 * No one can dirty the extent until we've finished our work and they can
356 * enqueue another lock. The DLM protects us from ll_file_read/write here,
357 * but other kernel actors could have pages locked.
359 * Called with the DLM lock held. */
360 void ll_pgcache_remove_extent(struct inode *inode, struct lov_stripe_md *lsm,
361 struct ldlm_lock *lock, __u32 stripe)
363 ldlm_policy_data_t tmpex;
364 unsigned long start, end, count, skip, i, j;
366 int rc, rc2, discard = lock->l_flags & LDLM_FL_DISCARD_DATA;
367 struct lustre_handle lockh;
370 memcpy(&tmpex, &lock->l_policy_data, sizeof(tmpex));
371 CDEBUG(D_INODE|D_PAGE, "inode %lu(%p) ["LPU64"->"LPU64"] size: %llu\n",
372 inode->i_ino, inode, tmpex.l_extent.start, tmpex.l_extent.end,
375 /* our locks are page granular thanks to osc_enqueue, we invalidate the
377 LASSERT((tmpex.l_extent.start & ~PAGE_CACHE_MASK) == 0);
378 LASSERT(((tmpex.l_extent.end + 1) & ~PAGE_CACHE_MASK) == 0);
382 start = tmpex.l_extent.start >> PAGE_CACHE_SHIFT;
383 end = tmpex.l_extent.end >> PAGE_CACHE_SHIFT;
384 if (lsm->lsm_stripe_count > 1) {
385 count = lsm->lsm_stripe_size >> PAGE_CACHE_SHIFT;
386 skip = (lsm->lsm_stripe_count - 1) * count;
387 start += start/count * skip + stripe * count;
389 end += end/count * skip + stripe * count;
391 if (end < tmpex.l_extent.end >> PAGE_CACHE_SHIFT)
394 i = inode->i_size ? (inode->i_size - 1) >> PAGE_CACHE_SHIFT : 0;
398 CDEBUG(D_INODE|D_PAGE, "walking page indices start: %lu j: %lu "
399 "count: %lu skip: %lu end: %lu%s\n", start, start % count,
400 count, skip, end, discard ? " (DISCARDING)" : "");
402 /* walk through the vmas on the inode and tear down mmaped pages that
403 * intersect with the lock. this stops immediately if there are no
404 * mmap()ed regions of the file. This is not efficient at all and
405 * should be short lived. We'll associate mmap()ed pages with the lock
406 * and will be able to find them directly */
408 for (i = start; i <= end; i += (j + skip)) {
409 j = min(count - (i % count), end - i + 1);
410 LASSERT(inode->i_mapping);
411 if (ll_teardown_mmaps(inode->i_mapping, i << PAGE_CACHE_SHIFT,
412 ((i+j) << PAGE_CACHE_SHIFT) - 1) )
416 /* this is the simplistic implementation of page eviction at
417 * cancelation. It is careful to get races with other page
418 * lockers handled correctly. fixes from bug 20 will make it
419 * more efficient by associating locks with pages and with
420 * batching writeback under the lock explicitly. */
421 for (i = start, j = start % count; i <= end;
422 j++, i++, tmpex.l_extent.start += PAGE_CACHE_SIZE) {
424 CDEBUG(D_PAGE, "skip index %lu to %lu\n", i, i + skip);
430 LASSERTF(tmpex.l_extent.start< lock->l_policy_data.l_extent.end,
431 LPU64" >= "LPU64" start %lu i %lu end %lu\n",
432 tmpex.l_extent.start, lock->l_policy_data.l_extent.end,
435 if (!mapping_has_pages(inode->i_mapping)) {
436 CDEBUG(D_INODE|D_PAGE, "nothing left\n");
442 page = find_get_page(inode->i_mapping, i);
445 LL_CDEBUG_PAGE(D_PAGE, page, "lock page idx %lu ext "LPU64"\n",
446 i, tmpex.l_extent.start);
449 /* page->mapping to check with racing against teardown */
450 if (!discard && clear_page_dirty_for_io(page)) {
451 rc = ll_call_writepage(inode, page);
453 CERROR("writepage of page %p failed: %d\n",
455 /* either waiting for io to complete or reacquiring
456 * the lock that the failed writepage released */
460 tmpex.l_extent.end = tmpex.l_extent.start + PAGE_CACHE_SIZE - 1;
461 /* check to see if another DLM lock covers this page */
462 rc2 = ldlm_lock_match(lock->l_resource->lr_namespace,
463 LDLM_FL_BLOCK_GRANTED|LDLM_FL_CBPENDING |
465 &lock->l_resource->lr_name, LDLM_EXTENT,
466 &tmpex, LCK_PR | LCK_PW, &lockh);
467 if (rc2 == 0 && page->mapping != NULL) {
468 // checking again to account for writeback's lock_page()
469 LL_CDEBUG_PAGE(D_PAGE, page, "truncating\n");
470 ll_ra_accounting(page, inode->i_mapping);
471 ll_truncate_complete_page(page);
474 page_cache_release(page);
476 LASSERTF(tmpex.l_extent.start <=
477 (lock->l_policy_data.l_extent.end == ~0ULL ? ~0ULL :
478 lock->l_policy_data.l_extent.end + 1),
479 "loop too long "LPU64" > "LPU64" start %lu i %lu end %lu\n",
480 tmpex.l_extent.start, lock->l_policy_data.l_extent.end,
485 static int ll_extent_lock_callback(struct ldlm_lock *lock,
486 struct ldlm_lock_desc *new, void *data,
489 struct lustre_handle lockh = { 0 };
493 if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
494 LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
499 case LDLM_CB_BLOCKING:
500 ldlm_lock2handle(lock, &lockh);
501 rc = ldlm_cli_cancel(&lockh);
503 CERROR("ldlm_cli_cancel failed: %d\n", rc);
505 case LDLM_CB_CANCELING: {
507 struct ll_inode_info *lli;
508 struct lov_stripe_md *lsm;
512 /* This lock wasn't granted, don't try to evict pages */
513 if (lock->l_req_mode != lock->l_granted_mode)
516 inode = ll_inode_from_lock(lock);
519 lli = ll_i2info(inode);
522 if (lli->lli_smd == NULL)
526 stripe = ll_lock_to_stripe_offset(inode, lock);
529 ll_pgcache_remove_extent(inode, lsm, lock, stripe);
531 /* grabbing the i_sem will wait for write() to complete. ns
532 * lock hold times should be very short as ast processing
533 * requires them and has a short timeout. so, i_sem before ns
537 l_lock(&lock->l_resource->lr_namespace->ns_lock);
538 kms = ldlm_extent_shift_kms(lock,
539 lsm->lsm_oinfo[stripe].loi_kms);
541 if (lsm->lsm_oinfo[stripe].loi_kms != kms)
542 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
543 lsm->lsm_oinfo[stripe].loi_kms, kms);
544 lsm->lsm_oinfo[stripe].loi_kms = kms;
545 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
547 //ll_try_done_writing(inode);
560 int ll_async_completion_ast(struct ldlm_lock *lock, int flags, void *data)
562 /* XXX ALLOCATE - 160 bytes */
563 struct inode *inode = ll_inode_from_lock(lock);
564 struct ll_inode_info *lli = ll_i2info(inode);
565 struct lustre_handle lockh = { 0 };
570 if (flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
571 LDLM_FL_BLOCK_CONV)) {
572 LBUG(); /* not expecting any blocked async locks yet */
573 LDLM_DEBUG(lock, "client-side async enqueue returned a blocked "
575 ldlm_lock_dump(D_OTHER, lock, 0);
576 ldlm_reprocess_all(lock->l_resource);
580 LDLM_DEBUG(lock, "client-side async enqueue: granted/glimpsed");
582 stripe = ll_lock_to_stripe_offset(inode, lock);
586 if (lock->l_lvb_len) {
587 struct lov_stripe_md *lsm = lli->lli_smd;
589 lvb = lock->l_lvb_data;
590 lsm->lsm_oinfo[stripe].loi_rss = lvb->lvb_size;
592 l_lock(&lock->l_resource->lr_namespace->ns_lock);
594 kms = MAX(lsm->lsm_oinfo[stripe].loi_kms, lvb->lvb_size);
595 kms = ldlm_extent_shift_kms(NULL, kms);
596 if (lsm->lsm_oinfo[stripe].loi_kms != kms)
597 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
598 lsm->lsm_oinfo[stripe].loi_kms, kms);
599 lsm->lsm_oinfo[stripe].loi_kms = kms;
601 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
606 wake_up(&lock->l_waitq);
608 ldlm_lock2handle(lock, &lockh);
609 ldlm_lock_decref(&lockh, LCK_PR);
614 static int ll_glimpse_callback(struct ldlm_lock *lock, void *reqp)
616 struct ptlrpc_request *req = reqp;
617 struct inode *inode = ll_inode_from_lock(lock);
618 struct ll_inode_info *lli;
620 struct lov_stripe_md *lsm;
621 int rc, size = sizeof(*lvb), stripe;
625 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
626 lli = ll_i2info(inode);
628 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
632 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
634 /* First, find out which stripe index this lock corresponds to. */
635 stripe = ll_lock_to_stripe_offset(inode, lock);
637 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
639 rc = lustre_pack_reply(req, 1, &size, NULL);
641 CERROR("lustre_pack_reply: %d\n", rc);
645 lvb = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*lvb));
646 lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe].loi_kms;
648 LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64,
649 inode->i_size, stripe, lvb->lvb_size);
655 /* These errors are normal races, so we don't want to fill the console
656 * with messages by calling ptlrpc_error() */
657 if (rc == -ELDLM_NO_LOCK_DATA)
658 lustre_pack_reply(req, 0, NULL, NULL);
664 __u64 lov_merge_size(struct lov_stripe_md *lsm, int kms);
665 __u64 lov_merge_blocks(struct lov_stripe_md *lsm);
666 __u64 lov_merge_mtime(struct lov_stripe_md *lsm, __u64 current_time);
668 /* NB: lov_merge_size will prefer locally cached writes if they extend the
669 * file (because it prefers KMS over RSS when larger) */
670 int ll_glimpse_size(struct inode *inode)
672 struct ll_inode_info *lli = ll_i2info(inode);
673 struct ll_sb_info *sbi = ll_i2sbi(inode);
674 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
675 struct lustre_handle lockh = { 0 };
676 int rc, flags = LDLM_FL_HAS_INTENT;
679 CDEBUG(D_DLMTRACE, "Glimpsing inode %lu\n", inode->i_ino);
681 rc = obd_enqueue(sbi->ll_lov_exp, lli->lli_smd, LDLM_EXTENT, &policy,
682 LCK_PR, &flags, ll_extent_lock_callback,
683 ldlm_completion_ast, ll_glimpse_callback, inode,
684 sizeof(struct ost_lvb), lustre_swab_ost_lvb, &lockh);
689 CERROR("obd_enqueue returned rc %d, returning -EIO\n", rc);
690 RETURN(rc > 0 ? -EIO : rc);
693 inode->i_size = lov_merge_size(lli->lli_smd, 0);
694 inode->i_blocks = lov_merge_blocks(lli->lli_smd);
695 //inode->i_mtime = lov_merge_mtime(lli->lli_smd, inode->i_mtime);
697 CDEBUG(D_DLMTRACE, "glimpse: size: "LPU64", blocks: "LPU64"\n",
698 (__u64)inode->i_size, (__u64)inode->i_blocks);
699 obd_cancel(sbi->ll_lov_exp, lli->lli_smd, LCK_PR, &lockh);
703 void ll_stime_record(struct ll_sb_info *sbi, struct timeval *start,
704 struct obd_service_time *stime)
707 do_gettimeofday(&stop);
709 spin_lock(&sbi->ll_lock);
710 lprocfs_stime_record(stime, &stop, start);
711 spin_unlock(&sbi->ll_lock);
714 int ll_extent_lock(struct ll_file_data *fd, struct inode *inode,
715 struct lov_stripe_md *lsm, int mode,
716 ldlm_policy_data_t *policy, struct lustre_handle *lockh,
717 int ast_flags, struct obd_service_time *stime)
719 struct ll_sb_info *sbi = ll_i2sbi(inode);
720 struct timeval start;
724 LASSERT(lockh->cookie == 0);
726 /* XXX phil: can we do this? won't it screw the file size up? */
727 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
728 (sbi->ll_flags & LL_SBI_NOLCK))
731 CDEBUG(D_DLMTRACE, "Locking inode %lu, start "LPU64" end "LPU64"\n",
732 inode->i_ino, policy->l_extent.start, policy->l_extent.end);
734 do_gettimeofday(&start);
735 rc = obd_enqueue(sbi->ll_lov_exp, lsm, LDLM_EXTENT, policy, mode,
736 &ast_flags, ll_extent_lock_callback,
737 ldlm_completion_ast, ll_glimpse_callback, inode,
738 sizeof(struct ost_lvb), lustre_swab_ost_lvb, lockh);
742 ll_stime_record(sbi, &start, stime);
744 if (policy->l_extent.start == 0 &&
745 policy->l_extent.end == OBD_OBJECT_EOF) {
746 /* vmtruncate()->ll_truncate() first sets the i_size and then
747 * the kms under both a DLM lock and the i_sem. If we don't
748 * get the i_sem here we can match the DLM lock and reset
749 * i_size from the kms before the truncating path has updated
750 * the kms. generic_file_write can then trust the stale i_size
751 * when doing appending writes and effectively cancel the
752 * result of the truncate. Getting the i_sem after the enqueue
753 * maintains the DLM -> i_sem acquiry order. */
755 inode->i_size = lov_merge_size(lsm, 1);
758 //inode->i_mtime = lov_merge_mtime(lsm, inode->i_mtime);
763 int ll_extent_unlock(struct ll_file_data *fd, struct inode *inode,
764 struct lov_stripe_md *lsm, int mode,
765 struct lustre_handle *lockh)
767 struct ll_sb_info *sbi = ll_i2sbi(inode);
771 /* XXX phil: can we do this? won't it screw the file size up? */
772 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
773 (sbi->ll_flags & LL_SBI_NOLCK))
776 rc = obd_cancel(sbi->ll_lov_exp, lsm, mode, lockh);
781 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
784 struct inode *inode = file->f_dentry->d_inode;
785 struct ll_inode_info *lli = ll_i2info(inode);
786 struct lov_stripe_md *lsm = lli->lli_smd;
787 struct ll_lock_tree tree;
788 struct ll_lock_tree_node *node;
793 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),size="LPSZ",offset=%Ld\n",
794 inode->i_ino, inode->i_generation, inode, count, *ppos);
796 /* "If nbyte is 0, read() will return 0 and have no other results."
797 * -- Single Unix Spec */
801 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats, LPROC_LL_READ_BYTES,
807 node = ll_node_from_inode(inode, *ppos, *ppos + count - 1,
810 tree.lt_fd = file->private_data;
812 rc = ll_tree_lock(&tree, node, inode, buf, count,
813 file->f_flags & O_NONBLOCK ? LDLM_FL_BLOCK_NOWAIT :0);
817 kms = lov_merge_size(lsm, 1);
818 if (*ppos + count - 1 > kms) {
819 /* A glimpse is necessary to determine whether we return a short
820 * read or some zeroes at the end of the buffer */
821 retval = ll_glimpse_size(inode);
828 CDEBUG(D_INFO, "Read ino %lu, "LPSZ" bytes, offset %lld, i_size %llu\n",
829 inode->i_ino, count, *ppos, inode->i_size);
831 /* turn off the kernel's read-ahead */
832 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
835 file->f_ra.ra_pages = 0;
837 retval = generic_file_read(file, buf, count, ppos);
840 ll_tree_unlock(&tree, inode);
845 * Write to a file (through the page cache).
847 static ssize_t ll_file_write(struct file *file, const char *buf,
848 size_t count, loff_t *ppos)
850 struct inode *inode = file->f_dentry->d_inode;
851 loff_t maxbytes = ll_file_maxbytes(inode);
852 struct ll_lock_tree tree;
853 struct ll_lock_tree_node *node;
858 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),size="LPSZ",offset=%Ld\n",
859 inode->i_ino, inode->i_generation, inode, count, *ppos);
861 SIGNAL_MASK_ASSERT(); /* XXX BUG 1511 */
863 /* POSIX, but surprised the VFS doesn't check this already */
867 /* If file was opened for LL_IOC_LOV_SETSTRIPE but the ioctl wasn't
868 * called on the file, don't fail the below assertion (bug 2388). */
869 if (file->f_flags & O_LOV_DELAY_CREATE &&
870 ll_i2info(inode)->lli_smd == NULL)
873 LASSERT(ll_i2info(inode)->lli_smd != NULL);
875 if (file->f_flags & O_APPEND)
876 node = ll_node_from_inode(inode, 0, OBD_OBJECT_EOF, LCK_PW);
878 node = ll_node_from_inode(inode, *ppos, *ppos + count - 1,
882 RETURN(PTR_ERR(node));
884 tree.lt_fd = file->private_data;
886 rc = ll_tree_lock(&tree, node, inode, buf, count,
887 file->f_flags & O_NONBLOCK ? LDLM_FL_BLOCK_NOWAIT :0);
891 /* this is ok, g_f_w will overwrite this under i_sem if it races
892 * with a local truncate, it just makes our maxbyte checking easier */
893 if (file->f_flags & O_APPEND)
894 *ppos = inode->i_size;
896 if (*ppos >= maxbytes) {
897 if (count || *ppos > maxbytes) {
898 send_sig(SIGXFSZ, current, 0);
899 GOTO(out, retval = -EFBIG);
902 if (*ppos + count > maxbytes)
903 count = maxbytes - *ppos;
905 CDEBUG(D_INFO, "Writing inode %lu, "LPSZ" bytes, offset %Lu\n",
906 inode->i_ino, count, *ppos);
908 /* generic_file_write handles O_APPEND after getting i_sem */
909 retval = generic_file_write(file, buf, count, ppos);
912 ll_tree_unlock(&tree, inode);
913 /* serialize with mmap/munmap/mremap */
914 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats, LPROC_LL_WRITE_BYTES,
915 retval > 0 ? retval : 0);
919 static int ll_lov_recreate_obj(struct inode *inode, struct file *file,
922 struct ll_inode_info *lli = ll_i2info(inode);
923 struct obd_export *exp = ll_i2obdexp(inode);
924 struct ll_recreate_obj ucreatp;
925 struct obd_trans_info oti = { 0 };
926 struct obdo *oa = NULL;
929 struct lov_stripe_md *lsm, *lsm2;
932 if (!capable (CAP_SYS_ADMIN))
935 rc = copy_from_user(&ucreatp, (struct ll_recreate_obj *)arg,
936 sizeof(struct ll_recreate_obj));
944 down(&lli->lli_open_sem);
947 GOTO(out, rc = -ENOENT);
948 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
949 (lsm->lsm_stripe_count));
951 OBD_ALLOC(lsm2, lsm_size);
953 GOTO(out, rc = -ENOMEM);
955 oa->o_id = ucreatp.lrc_id;
956 oa->o_nlink = ucreatp.lrc_ost_idx;
957 oa->o_gr = ucreatp.lrc_group;
958 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLFLAGS;
959 oa->o_flags |= OBD_FL_RECREATE_OBJS;
960 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
961 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
963 oti.oti_objid = NULL;
964 memcpy(lsm2, lsm, lsm_size);
965 rc = obd_create(exp, oa, &lsm2, &oti);
967 OBD_FREE(lsm2, lsm_size);
970 up(&lli->lli_open_sem);
975 static int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
976 int flags, struct lov_user_md *lum,
979 struct ll_inode_info *lli = ll_i2info(inode);
981 struct obd_export *exp = ll_i2obdexp(inode);
982 struct lov_stripe_md *lsm;
983 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
984 struct ptlrpc_request *req = NULL;
989 down(&lli->lli_open_sem);
992 up(&lli->lli_open_sem);
993 CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n",
998 f = get_empty_filp();
1002 f->f_dentry = file->f_dentry;
1003 f->f_vfsmnt = file->f_vfsmnt;
1005 rc = ll_intent_file_open(f, lum, lum_size, &oit);
1008 if (it_disposition(&oit, DISP_LOOKUP_NEG))
1010 req = oit.d.lustre.it_data;
1011 rc = oit.d.lustre.it_status;
1016 rc = mdc_req2lustre_md(ll_i2lmvexp(inode), req, 1, exp, &md);
1019 ll_update_inode(f->f_dentry->d_inode, &md);
1021 rc = ll_local_open(f, &oit);
1024 ll_intent_release(&oit);
1026 rc = ll_file_release(f->f_dentry->d_inode, f);
1031 up(&lli->lli_open_sem);
1033 ptlrpc_req_finished(req);
1037 static int ll_lov_setea(struct inode *inode, struct file *file,
1040 int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1041 struct lov_user_md *lump;
1042 int lum_size = sizeof(struct lov_user_md) +
1043 sizeof(struct lov_user_ost_data);
1047 if (!capable (CAP_SYS_ADMIN))
1050 OBD_ALLOC(lump, lum_size);
1054 rc = copy_from_user(lump, (struct lov_user_md *)arg, lum_size);
1056 OBD_FREE(lump, lum_size);
1060 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1062 OBD_FREE(lump, lum_size);
1066 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1069 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1071 int flags = FMODE_WRITE;
1074 /* Bug 1152: copy properly when this is no longer true */
1075 LASSERT(sizeof(lum) == sizeof(*lump));
1076 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1077 rc = copy_from_user(&lum, lump, sizeof(lum));
1081 rc = ll_lov_setstripe_ea_info(inode, file, flags, &lum, sizeof(lum));
1083 put_user(0, &lump->lmm_stripe_count);
1084 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2obdexp(inode),
1085 0, ll_i2info(inode)->lli_smd, lump);
1090 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1092 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1097 return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2obdexp(inode), 0, lsm,
1101 static int ll_get_grouplock(struct inode *inode, struct file *file,
1104 struct ll_file_data *fd = file->private_data;
1105 ldlm_policy_data_t policy = { .l_extent = { .start = 0,
1106 .end = OBD_OBJECT_EOF}};
1107 struct lustre_handle lockh = { 0 };
1108 struct ll_inode_info *lli = ll_i2info(inode);
1109 struct lov_stripe_md *lsm = lli->lli_smd;
1113 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1117 policy.l_extent.gid = arg;
1118 if (file->f_flags & O_NONBLOCK)
1119 flags = LDLM_FL_BLOCK_NOWAIT;
1121 rc = ll_extent_lock(fd, inode, lsm, LCK_GROUP, &policy, &lockh, flags,
1122 &ll_i2sbi(inode)->ll_grouplock_stime);
1126 fd->fd_flags |= LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK;
1128 memcpy(&fd->fd_cwlockh, &lockh, sizeof(lockh));
1133 static int ll_put_grouplock(struct inode *inode, struct file *file,
1136 struct ll_file_data *fd = file->private_data;
1137 struct ll_inode_info *lli = ll_i2info(inode);
1138 struct lov_stripe_md *lsm = lli->lli_smd;
1142 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1143 /* Ugh, it's already unlocked. */
1147 if (fd->fd_gid != arg) /* Ugh? Unlocking with different gid? */
1150 fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK);
1152 rc = ll_extent_unlock(fd, inode, lsm, LCK_GROUP, &fd->fd_cwlockh);
1157 memset(&fd->fd_cwlockh, 0, sizeof(fd->fd_cwlockh));
1162 int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
1165 struct ll_file_data *fd = file->private_data;
1166 struct ll_sb_info *sbi = ll_i2sbi(inode);
1170 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino,
1171 inode->i_generation, inode, cmd);
1173 if (_IOC_TYPE(cmd) == 'T') /* tty ioctls */
1176 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
1178 case LL_IOC_GETFLAGS:
1179 /* Get the current value of the file flags */
1180 return put_user(fd->fd_flags, (int *)arg);
1181 case LL_IOC_SETFLAGS:
1182 case LL_IOC_CLRFLAGS:
1183 /* Set or clear specific file flags */
1184 /* XXX This probably needs checks to ensure the flags are
1185 * not abused, and to handle any flag side effects.
1187 if (get_user(flags, (int *) arg))
1190 if (cmd == LL_IOC_SETFLAGS)
1191 fd->fd_flags |= flags;
1193 fd->fd_flags &= ~flags;
1195 case LL_IOC_LOV_SETSTRIPE:
1196 RETURN(ll_lov_setstripe(inode, file, arg));
1197 case LL_IOC_LOV_SETEA:
1198 RETURN(ll_lov_setea(inode, file, arg));
1199 case IOC_MDC_SHOWFID: {
1200 struct lustre_id *idp = (struct lustre_id *)arg;
1201 struct lustre_id id;
1205 filename = getname((const char *)arg);
1206 if (IS_ERR(filename))
1207 RETURN(PTR_ERR(filename));
1209 ll_inode2id(&id, inode);
1211 rc = ll_get_fid(sbi->ll_lmv_exp, &id, filename, &id);
1213 GOTO(out_filename, rc);
1215 rc = copy_to_user(idp, &id, sizeof(*idp));
1217 GOTO(out_filename, rc = -EFAULT);
1224 case LL_IOC_LOV_GETSTRIPE:
1225 RETURN(ll_lov_getstripe(inode, arg));
1226 case LL_IOC_RECREATE_OBJ:
1227 RETURN(ll_lov_recreate_obj(inode, file, arg));
1228 case EXT3_IOC_GETFLAGS:
1229 case EXT3_IOC_SETFLAGS:
1230 RETURN( ll_iocontrol(inode, file, cmd, arg) );
1231 case LL_IOC_GROUP_LOCK:
1232 RETURN(ll_get_grouplock(inode, file, arg));
1233 case LL_IOC_GROUP_UNLOCK:
1234 RETURN(ll_put_grouplock(inode, file, arg));
1235 case EXT3_IOC_GETVERSION_OLD:
1236 case EXT3_IOC_GETVERSION:
1237 return put_user(inode->i_generation, (int *) arg);
1238 /* We need to special case any other ioctls we want to handle,
1239 * to send them to the MDS/OST as appropriate and to properly
1240 * network encode the arg field.
1241 case EXT2_IOC_GETVERSION_OLD:
1242 case EXT2_IOC_GETVERSION_NEW:
1243 case EXT2_IOC_SETVERSION_OLD:
1244 case EXT2_IOC_SETVERSION_NEW:
1245 case EXT3_IOC_SETVERSION_OLD:
1246 case EXT3_IOC_SETVERSION:
1249 RETURN( obd_iocontrol(cmd, ll_i2obdexp(inode), 0, NULL,
1254 loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
1256 struct inode *inode = file->f_dentry->d_inode;
1257 struct ll_file_data *fd = file->private_data;
1258 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1259 struct lustre_handle lockh = {0};
1262 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),to=%llu\n", inode->i_ino,
1263 inode->i_generation, inode,
1264 offset + ((origin==2) ? inode->i_size : file->f_pos));
1266 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_LLSEEK);
1267 if (origin == 2) { /* SEEK_END */
1268 ldlm_policy_data_t policy = { .l_extent = {0, OBD_OBJECT_EOF }};
1269 int nonblock = 0, rc;
1271 if (file->f_flags & O_NONBLOCK)
1272 nonblock = LDLM_FL_BLOCK_NOWAIT;
1274 rc = ll_extent_lock(fd, inode, lsm, LCK_PR, &policy, &lockh,
1275 nonblock, &ll_i2sbi(inode)->ll_seek_stime);
1279 offset += inode->i_size;
1280 } else if (origin == 1) { /* SEEK_CUR */
1281 offset += file->f_pos;
1285 if (offset >= 0 && offset <= ll_file_maxbytes(inode)) {
1286 if (offset != file->f_pos) {
1287 file->f_pos = offset;
1288 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1290 file->f_version = ++event;
1297 ll_extent_unlock(fd, inode, lsm, LCK_PR, &lockh);
1301 int ll_fsync(struct file *file, struct dentry *dentry, int data)
1303 struct inode *inode = dentry->d_inode;
1304 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1305 struct lustre_id id;
1306 struct ptlrpc_request *req;
1309 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1310 inode->i_generation, inode);
1312 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_FSYNC);
1314 /* fsync's caller has already called _fdata{sync,write}, we want
1315 * that IO to finish before calling the osc and mdc sync methods */
1316 rc = filemap_fdatawait(inode->i_mapping);
1318 ll_inode2id(&id, inode);
1319 err = md_sync(ll_i2sbi(inode)->ll_lmv_exp, &id, &req);
1323 ptlrpc_req_finished(req);
1326 struct obdo *oa = obdo_alloc();
1329 RETURN(rc ? rc : -ENOMEM);
1331 oa->o_id = lsm->lsm_object_id;
1332 oa->o_gr = lsm->lsm_object_gr;
1333 oa->o_valid = OBD_MD_FLID;
1334 obdo_from_inode(oa, inode, (OBD_MD_FLTYPE | OBD_MD_FLATIME |
1335 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1338 err = obd_sync(ll_i2sbi(inode)->ll_lov_exp, oa, lsm,
1348 int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
1350 struct inode *inode = file->f_dentry->d_inode;
1351 struct ll_inode_info *li = ll_i2info(inode);
1352 struct ll_sb_info *sbi = ll_i2sbi(inode);
1353 struct obd_device *obddev;
1354 struct ldlm_res_id res_id =
1355 { .name = {id_fid(&li->lli_id), id_group(&li->lli_id), LDLM_FLOCK} };
1356 struct lustre_handle lockh = {0};
1357 ldlm_policy_data_t flock;
1358 ldlm_mode_t mode = 0;
1363 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu file_lock=%p\n",
1364 inode->i_ino, file_lock);
1366 flock.l_flock.pid = file_lock->fl_pid;
1367 flock.l_flock.start = file_lock->fl_start;
1368 flock.l_flock.end = file_lock->fl_end;
1370 switch (file_lock->fl_type) {
1375 /* An unlock request may or may not have any relation to
1376 * existing locks so we may not be able to pass a lock handle
1377 * via a normal ldlm_lock_cancel() request. The request may even
1378 * unlock a byte range in the middle of an existing lock. In
1379 * order to process an unlock request we need all of the same
1380 * information that is given with a normal read or write record
1381 * lock request. To avoid creating another ldlm unlock (cancel)
1382 * message we'll treat a LCK_NL flock request as an unlock. */
1389 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1404 flags = LDLM_FL_BLOCK_NOWAIT;
1410 flags = LDLM_FL_TEST_LOCK;
1411 /* Save the old mode so that if the mode in the lock changes we
1412 * can decrement the appropriate reader or writer refcount. */
1413 file_lock->fl_type = mode;
1416 CERROR("unknown fcntl lock command: %d\n", cmd);
1420 CDEBUG(D_DLMTRACE, "inode=%lu, pid="LPU64", flags=%#x, mode=%u, "
1421 "start="LPU64", end="LPU64"\n", inode->i_ino, flock.l_flock.pid,
1422 flags, mode, flock.l_flock.start, flock.l_flock.end);
1424 obddev = md_get_real_obd(sbi->ll_lmv_exp, NULL, 0);
1425 rc = ldlm_cli_enqueue(obddev->obd_self_export, NULL,
1426 obddev->obd_namespace,
1427 res_id, LDLM_FLOCK, &flock, mode, &flags,
1428 NULL, ldlm_flock_completion_ast, NULL, file_lock,
1429 NULL, 0, NULL, &lockh);
1433 int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it)
1435 struct lookup_intent oit = { .it_op = IT_GETATTR };
1436 struct inode *inode = dentry->d_inode;
1437 struct ptlrpc_request *req = NULL;
1438 struct ll_inode_info *lli;
1439 struct lov_stripe_md *lsm;
1440 struct ll_sb_info *sbi;
1441 struct lustre_id id;
1447 CERROR("REPORT THIS LINE TO PETER\n");
1451 sbi = ll_i2sbi(inode);
1453 ll_inode2id(&id, inode);
1454 lli = ll_i2info(inode);
1455 LASSERT(id_fid(&id) != 0);
1457 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), name=%s, intent=%s\n",
1458 inode->i_ino, inode->i_generation, inode, dentry->d_name.name,
1461 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,5,0))
1462 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_REVALIDATE);
1465 rc = md_intent_lock(sbi->ll_lmv_exp, &id, NULL, 0, NULL, 0, &id,
1466 &oit, 0, &req, ll_mdc_blocking_ast);
1470 rc = revalidate_it_finish(req, 1, &oit, dentry);
1472 ll_intent_release(&oit);
1476 if (dentry->d_iname[0] != '/') { /* Do not rehash root of the tree
1478 spin_lock(&dcache_lock);
1479 hlist_del_init(&dentry->d_hash);
1480 __d_rehash(dentry, 0);
1481 spin_unlock(&dcache_lock);
1483 ll_lookup_finish_locks(&oit, dentry);
1484 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
1488 if (lsm == NULL) /* object not yet allocated, don't validate size */
1492 * ll_glimpse_size() will prefer locally cached writes if they extend
1495 rc = ll_glimpse_size(inode);
1498 ptlrpc_req_finished(req);
1502 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1503 int ll_getattr(struct vfsmount *mnt, struct dentry *de,
1504 struct lookup_intent *it, struct kstat *stat)
1507 struct inode *inode = de->d_inode;
1509 res = ll_inode_revalidate_it(de, it);
1510 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_GETATTR);
1515 stat->ino = inode->i_ino;
1516 stat->mode = inode->i_mode;
1517 stat->nlink = inode->i_nlink;
1518 stat->uid = inode->i_uid;
1519 stat->gid = inode->i_gid;
1520 stat->atime = inode->i_atime;
1521 stat->mtime = inode->i_mtime;
1522 stat->ctime = inode->i_ctime;
1523 stat->size = inode->i_size;
1524 stat->blksize = inode->i_blksize;
1525 stat->blocks = inode->i_blocks;
1526 stat->rdev = kdev_t_to_nr(inode->i_rdev);
1527 stat->dev = id_group(&ll_i2info(inode)->lli_id);
1532 struct file_operations ll_file_operations = {
1533 .read = ll_file_read,
1534 .write = ll_file_write,
1535 .ioctl = ll_file_ioctl,
1536 .open = ll_file_open,
1537 .release = ll_file_release,
1538 .mmap = ll_file_mmap,
1539 .llseek = ll_file_seek,
1540 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1541 .sendfile = generic_file_sendfile,
1544 .lock = ll_file_flock
1547 struct inode_operations ll_file_inode_operations = {
1548 .setattr_raw = ll_setattr_raw,
1549 .setattr = ll_setattr,
1550 .truncate = ll_truncate,
1551 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1552 .getattr_it = ll_getattr,
1554 .revalidate_it = ll_inode_revalidate_it,