diff options
55 files changed, 24504 insertions, 0 deletions
diff --git a/Documentation/filesystems/00-INDEX b/Documentation/filesystems/00-INDEX index d9b0a069186..2580ada100a 100644 --- a/Documentation/filesystems/00-INDEX +++ b/Documentation/filesystems/00-INDEX @@ -36,6 +36,8 @@ ntfs.txt - info and mount options for the NTFS filesystem (Windows NT). proc.txt - info on Linux's /proc filesystem. +ocfs2.txt + - info and mount options for the OCFS2 clustered filesystem. romfs.txt - Description of the ROMFS filesystem. smbfs.txt diff --git a/Documentation/filesystems/ocfs2.txt b/Documentation/filesystems/ocfs2.txt new file mode 100644 index 00000000000..f2595caf052 --- /dev/null +++ b/Documentation/filesystems/ocfs2.txt @@ -0,0 +1,55 @@ +OCFS2 filesystem +================== +OCFS2 is a general purpose extent based shared disk cluster file +system with many similarities to ext3. It supports 64 bit inode +numbers, and has automatically extending metadata groups which may +also make it attractive for non-clustered use. + +You'll want to install the ocfs2-tools package in order to at least +get "mount.ocfs2" and "ocfs2_hb_ctl". + +Project web page: http://oss.oracle.com/projects/ocfs2 +Tools web page: http://oss.oracle.com/projects/ocfs2-tools +OCFS2 mailing lists: http://oss.oracle.com/projects/ocfs2/mailman/ + +All code copyright 2005 Oracle except when otherwise noted. + +CREDITS: +Lots of code taken from ext3 and other projects. + +Authors in alphabetical order: +Joel Becker <joel.becker@oracle.com> +Zach Brown <zach.brown@oracle.com> +Mark Fasheh <mark.fasheh@oracle.com> +Kurt Hackel <kurt.hackel@oracle.com> +Sunil Mushran <sunil.mushran@oracle.com> +Manish Singh <manish.singh@oracle.com> + +Caveats +======= +Features which OCFS2 does not support yet: + - sparse files + - extended attributes + - shared writeable mmap + - loopback is supported, but data written will not + be cluster coherent. + - quotas + - cluster aware flock + - Directory change notification (F_NOTIFY) + - Distributed Caching (F_SETLEASE/F_GETLEASE/break_lease) + - POSIX ACLs + - readpages / writepages (not user visible) + +Mount options +============= + +OCFS2 supports the following mount options: +(*) == default + +barrier=1 This enables/disables barriers. barrier=0 disables it, + barrier=1 enables it. +errors=remount-ro(*) Remount the filesystem read-only on an error. +errors=panic Panic and halt the machine if an error occurs. +intr (*) Allow signals to interrupt cluster operations. +nointr Do not allow signals to interrupt cluster + operations. diff --git a/MAINTAINERS b/MAINTAINERS index 86ee06f4379..15888302025 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -1905,6 +1905,15 @@ M: ajoshi@shell.unixbox.com L: linux-nvidia@lists.surfsouth.com S: Maintained +ORACLE CLUSTER FILESYSTEM 2 (OCFS2) +P: Mark Fasheh +M: mark.fasheh@oracle.com +P: Kurt Hackel +M: kurt.hackel@oracle.com +L: ocfs2-devel@oss.oracle.com +W: http://oss.oracle.com/projects/ocfs2/ +S: Supported + OLYMPIC NETWORK DRIVER P: Peter De Shrijver M: p2@ace.ulyssis.student.kuleuven.ac.be diff --git a/fs/ocfs2/Makefile b/fs/ocfs2/Makefile new file mode 100644 index 00000000000..7d3be845a61 --- /dev/null +++ b/fs/ocfs2/Makefile @@ -0,0 +1,33 @@ +EXTRA_CFLAGS += -Ifs/ocfs2 + +EXTRA_CFLAGS += -DCATCH_BH_JBD_RACES + +obj-$(CONFIG_OCFS2_FS) += ocfs2.o + +ocfs2-objs := \ + alloc.o \ + aops.o \ + buffer_head_io.o \ + dcache.o \ + dir.o \ + dlmglue.o \ + export.o \ + extent_map.o \ + file.o \ + heartbeat.o \ + inode.o \ + journal.o \ + localalloc.o \ + mmap.o \ + namei.o \ + slot_map.o \ + suballoc.o \ + super.o \ + symlink.o \ + sysfile.o \ + uptodate.o \ + ver.o \ + vote.o + +obj-$(CONFIG_OCFS2_FS) += cluster/ +obj-$(CONFIG_OCFS2_FS) += dlm/ diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c new file mode 100644 index 00000000000..465f797451e --- /dev/null +++ b/fs/ocfs2/alloc.c @@ -0,0 +1,2040 @@ +/* -*- mode: c; c-basic-offset: 8; -*- + * vim: noexpandtab sw=8 ts=8 sts=0: + * + * alloc.c + * + * Extent allocs and frees + * + * Copyright (C) 2002, 2004 Oracle. All rights reserved. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public + * License as published by the Free Software Foundation; either + * version 2 of the License, or (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License for more details. + * + * You should have received a copy of the GNU General Public + * License along with this program; if not, write to the + * Free Software Foundation, Inc., 59 Temple Place - Suite 330, + * Boston, MA 021110-1307, USA. + */ + +#include <linux/fs.h> +#include <linux/types.h> +#include <linux/slab.h> +#include <linux/highmem.h> + +#define MLOG_MASK_PREFIX ML_DISK_ALLOC +#include <cluster/masklog.h> + +#include "ocfs2.h" + +#include "alloc.h" +#include "dlmglue.h" +#include "extent_map.h" +#include "inode.h" +#include "journal.h" +#include "localalloc.h" +#include "suballoc.h" +#include "sysfile.h" +#include "file.h" +#include "super.h" +#include "uptodate.h" + +#include "buffer_head_io.h" + +static int ocfs2_extent_contig(struct inode *inode, + struct ocfs2_extent_rec *ext, + u64 blkno); + +static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb, + struct ocfs2_journal_handle *handle, + struct inode *inode, + int wanted, + struct ocfs2_alloc_context *meta_ac, + struct buffer_head *bhs[]); + +static int ocfs2_add_branch(struct ocfs2_super *osb, + struct ocfs2_journal_handle *handle, + struct inode *inode, + struct buffer_head *fe_bh, + struct buffer_head *eb_bh, + struct buffer_head *last_eb_bh, + struct ocfs2_alloc_context *meta_ac); + +static int ocfs2_shift_tree_depth(struct ocfs2_super *osb, + struct ocfs2_journal_handle *handle, + struct inode *inode, + struct buffer_head *fe_bh, + struct ocfs2_alloc_context *meta_ac, + struct buffer_head **ret_new_eb_bh); + +static int ocfs2_do_insert_extent(struct ocfs2_super *osb, + struct ocfs2_journal_handle *handle, + struct inode *inode, + struct buffer_head *fe_bh, + u64 blkno, + u32 new_clusters); + +static int ocfs2_find_branch_target(struct ocfs2_super *osb, + struct inode *inode, + struct buffer_head *fe_bh, + struct buffer_head **target_bh); + +static int ocfs2_find_new_last_ext_blk(struct ocfs2_super *osb, + struct inode *inode, + struct ocfs2_dinode *fe, + unsigned int new_i_clusters, + struct buffer_head *old_last_eb, + struct buffer_head **new_last_eb); + +static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); + +static int ocfs2_extent_contig(struct inode *inode, + struct ocfs2_extent_rec *ext, + u64 blkno) +{ + return blkno == (le64_to_cpu(ext->e_blkno) + + ocfs2_clusters_to_blocks(inode->i_sb, + le32_to_cpu(ext->e_clusters))); +} + +/* + * How many free extents have we got before we need more meta data? + */ +int ocfs2_num_free_extents(struct ocfs2_super *osb, + struct inode *inode, + struct ocfs2_dinode *fe) +{ + int retval; + struct ocfs2_extent_list *el; + struct ocfs2_extent_block *eb; + struct buffer_head *eb_bh = NULL; + + mlog_entry_void(); + + if (!OCFS2_IS_VALID_DINODE(fe)) { + OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); + retval = -EIO; + goto bail; + } + + if (fe->i_last_eb_blk) { + retval = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk), + &eb_bh, OCFS2_BH_CACHED, inode); + if (retval < 0) { + mlog_errno(retval); + goto bail; + } + eb = (struct ocfs2_extent_block *) eb_bh->b_data; + el = &eb->h_list; + } else + el = &fe->id2.i_list; + + BUG_ON(el->l_tree_depth != 0); + + retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); +bail: + if (eb_bh) + brelse(eb_bh); + + mlog_exit(retval); + return retval; +} + +/* expects array to already be allocated + * + * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and + * l_count for you + */ +static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb, + struct ocfs2_journal_handle *handle, + struct inode *inode, + int wanted, + struct ocfs2_alloc_context *meta_ac, + struct buffer_head *bhs[]) +{ + int count, status, i; + u16 suballoc_bit_start; + u32 num_got; + u64 first_blkno; + struct ocfs2_extent_block *eb; + + mlog_entry_void(); + + count = 0; + while (count < wanted) { + status = ocfs2_claim_metadata(osb, + handle, + meta_ac, + wanted - count, + &suballoc_bit_start, + &num_got, + &first_blkno); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + for(i = count; i < (num_got + count); i++) { + bhs[i] = sb_getblk(osb->sb, first_blkno); + if (bhs[i] == NULL) { + status = -EIO; + mlog_errno(status); + goto bail; + } + ocfs2_set_new_buffer_uptodate(inode, bhs[i]); + + status = ocfs2_journal_access(handle, inode, bhs[i], + OCFS2_JOURNAL_ACCESS_CREATE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); + eb = (struct ocfs2_extent_block *) bhs[i]->b_data; + /* Ok, setup the minimal stuff here. */ + strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); + eb->h_blkno = cpu_to_le64(first_blkno); + eb->h_fs_generation = cpu_to_le32(osb->fs_generation); + +#ifndef OCFS2_USE_ALL_METADATA_SUBALLOCATORS + /* we always use slot zero's suballocator */ + eb->h_suballoc_slot = 0; +#else + eb->h_suballoc_slot = cpu_to_le16(osb->slot_num); +#endif + eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); + eb->h_list.l_count = + cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); + + suballoc_bit_start++; + first_blkno++; + + /* We'll also be dirtied by the caller, so + * this isn't absolutely necessary. */ + status = ocfs2_journal_dirty(handle, bhs[i]); + if (status < 0) { + mlog_errno(status); + goto bail; + } + } + + count += num_got; + } + + status = 0; +bail: + if (status < 0) { + for(i = 0; i < wanted; i++) { + if (bhs[i]) + brelse(bhs[i]); + bhs[i] = NULL; + } + } + mlog_exit(status); + return status; +} + +/* + * Add an entire tree branch to our inode. eb_bh is the extent block + * to start at, if we don't want to start the branch at the dinode + * structure. + * + * last_eb_bh is required as we have to update it's next_leaf pointer + * for the new last extent block. + * + * the new branch will be 'empty' in the sense that every block will + * contain a single record with e_clusters == 0. + */ +static int ocfs2_add_branch(struct ocfs2_super *osb, + struct ocfs2_journal_handle *handle, + struct inode *inode, + struct buffer_head *fe_bh, + struct buffer_head *eb_bh, + struct buffer_head *last_eb_bh, + struct ocfs2_alloc_context *meta_ac) +{ + int status, new_blocks, i; + u64 next_blkno, new_last_eb_blk; + struct buffer_head *bh; + struct buffer_head **new_eb_bhs = NULL; + struct ocfs2_dinode *fe; + struct ocfs2_extent_block *eb; + struct ocfs2_extent_list *eb_el; + struct ocfs2_extent_list *el; + + mlog_entry_void(); + + BUG_ON(!last_eb_bh); + + fe = (struct ocfs2_dinode *) fe_bh->b_data; + + if (eb_bh) { + eb = (struct ocfs2_extent_block *) eb_bh->b_data; + el = &eb->h_list; + } else + el = &fe->id2.i_list; + + /* we never add a branch to a leaf. */ + BUG_ON(!el->l_tree_depth); + + new_blocks = le16_to_cpu(el->l_tree_depth); + + /* allocate the number of new eb blocks we need */ + new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), + GFP_KERNEL); + if (!new_eb_bhs) { + status = -ENOMEM; + mlog_errno(status); + goto bail; + } + + status = ocfs2_create_new_meta_bhs(osb, handle, inode, new_blocks, + meta_ac, new_eb_bhs); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be + * linked with the rest of the tree. + * conversly, new_eb_bhs[0] is the new bottommost leaf. + * + * when we leave the loop, new_last_eb_blk will point to the + * newest leaf, and next_blkno will point to the topmost extent + * block. */ + next_blkno = new_last_eb_blk = 0; + for(i = 0; i < new_blocks; i++) { + bh = new_eb_bhs[i]; + eb = (struct ocfs2_extent_block *) bh->b_data; + if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { + OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); + status = -EIO; + goto bail; + } + eb_el = &eb->h_list; + + status = ocfs2_journal_access(handle, inode, bh, + OCFS2_JOURNAL_ACCESS_CREATE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + eb->h_next_leaf_blk = 0; + eb_el->l_tree_depth = cpu_to_le16(i); + eb_el->l_next_free_rec = cpu_to_le16(1); + eb_el->l_recs[0].e_cpos = fe->i_clusters; + eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); + eb_el->l_recs[0].e_clusters = cpu_to_le32(0); + if (!eb_el->l_tree_depth) + new_last_eb_blk = le64_to_cpu(eb->h_blkno); + + status = ocfs2_journal_dirty(handle, bh); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + next_blkno = le64_to_cpu(eb->h_blkno); + } + + /* This is a bit hairy. We want to update up to three blocks + * here without leaving any of them in an inconsistent state + * in case of error. We don't have to worry about + * journal_dirty erroring as it won't unless we've aborted the + * handle (in which case we would never be here) so reserving + * the write with journal_access is all we need to do. */ + status = ocfs2_journal_access(handle, inode, last_eb_bh, + OCFS2_JOURNAL_ACCESS_WRITE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + status = ocfs2_journal_access(handle, inode, fe_bh, + OCFS2_JOURNAL_ACCESS_WRITE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + if (eb_bh) { + status = ocfs2_journal_access(handle, inode, eb_bh, + OCFS2_JOURNAL_ACCESS_WRITE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + } + + /* Link the new branch into the rest of the tree (el will + * either be on the fe, or the extent block passed in. */ + i = le16_to_cpu(el->l_next_free_rec); + el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); + el->l_recs[i].e_cpos = fe->i_clusters; + el->l_recs[i].e_clusters = 0; + le16_add_cpu(&el->l_next_free_rec, 1); + + /* fe needs a new last extent block pointer, as does the + * next_leaf on the previously last-extent-block. */ + fe->i_last_eb_blk = cpu_to_le64(new_last_eb_blk); + + eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; + eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); + + status = ocfs2_journal_dirty(handle, last_eb_bh); + if (status < 0) + mlog_errno(status); + status = ocfs2_journal_dirty(handle, fe_bh); + if (status < 0) + mlog_errno(status); + if (eb_bh) { + status = ocfs2_journal_dirty(handle, eb_bh); + if (status < 0) + mlog_errno(status); + } + + status = 0; +bail: + if (new_eb_bhs) { + for (i = 0; i < new_blocks; i++) + if (new_eb_bhs[i]) + brelse(new_eb_bhs[i]); + kfree(new_eb_bhs); + } + + mlog_exit(status); + return status; +} + +/* + * adds another level to the allocation tree. + * returns back the new extent block so you can add a branch to it + * after this call. + */ +static int ocfs2_shift_tree_depth(struct ocfs2_super *osb, + struct ocfs2_journal_handle *handle, + struct inode *inode, + struct buffer_head *fe_bh, + struct ocfs2_alloc_context *meta_ac, + struct buffer_head **ret_new_eb_bh) +{ + int status, i; + struct buffer_head *new_eb_bh = NULL; + struct ocfs2_dinode *fe; + struct ocfs2_extent_block *eb; + struct ocfs2_extent_list *fe_el; + struct ocfs2_extent_list *eb_el; + + mlog_entry_void(); + + status = ocfs2_create_new_meta_bhs(osb, handle, inode, 1, meta_ac, + &new_eb_bh); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; + if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { + OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); + status = -EIO; + goto bail; + } + + eb_el = &eb->h_list; + fe = (struct ocfs2_dinode *) fe_bh->b_data; + fe_el = &fe->id2.i_list; + + status = ocfs2_journal_access(handle, inode, new_eb_bh, + OCFS2_JOURNAL_ACCESS_CREATE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + /* copy the fe data into the new extent block */ + eb_el->l_tree_depth = fe_el->l_tree_depth; + eb_el->l_next_free_rec = fe_el->l_next_free_rec; + for(i = 0; i < le16_to_cpu(fe_el->l_next_free_rec); i++) { + eb_el->l_recs[i].e_cpos = fe_el->l_recs[i].e_cpos; + eb_el->l_recs[i].e_clusters = fe_el->l_recs[i].e_clusters; + eb_el->l_recs[i].e_blkno = fe_el->l_recs[i].e_blkno; + } + + status = ocfs2_journal_dirty(handle, new_eb_bh); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + status = ocfs2_journal_access(handle, inode, fe_bh, + OCFS2_JOURNAL_ACCESS_WRITE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + /* update fe now */ + le16_add_cpu(&fe_el->l_tree_depth, 1); + fe_el->l_recs[0].e_cpos = 0; + fe_el->l_recs[0].e_blkno = eb->h_blkno; + fe_el->l_recs[0].e_clusters = fe->i_clusters; + for(i = 1; i < le16_to_cpu(fe_el->l_next_free_rec); i++) { + fe_el->l_recs[i].e_cpos = 0; + fe_el->l_recs[i].e_clusters = 0; + fe_el->l_recs[i].e_blkno = 0; + } + fe_el->l_next_free_rec = cpu_to_le16(1); + + /* If this is our 1st tree depth shift, then last_eb_blk + * becomes the allocated extent block */ + if (fe_el->l_tree_depth == cpu_to_le16(1)) + fe->i_last_eb_blk = eb->h_blkno; + + status = ocfs2_journal_dirty(handle, fe_bh); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + *ret_new_eb_bh = new_eb_bh; + new_eb_bh = NULL; + status = 0; +bail: + if (new_eb_bh) + brelse(new_eb_bh); + + mlog_exit(status); + return status; +} + +/* + * Expects the tree to already have room in the rightmost leaf for the + * extent. Updates all the extent blocks (and the dinode) on the way + * down. + */ +static int ocfs2_do_insert_extent(struct ocfs2_super *osb, + struct ocfs2_journal_handle *handle, + struct inode *inode, + struct buffer_head *fe_bh, + u64 start_blk, + u32 new_clusters) +{ + int status, i, num_bhs = 0; + u64 next_blkno; + u16 next_free; + struct buffer_head **eb_bhs = NULL; + struct ocfs2_dinode *fe; + struct ocfs2_extent_block *eb; + struct ocfs2_extent_list *el; + + mlog_entry_void(); + + status = ocfs2_journal_access(handle, inode, fe_bh, + OCFS2_JOURNAL_ACCESS_WRITE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + fe = (struct ocfs2_dinode *) fe_bh->b_data; + el = &fe->id2.i_list; + if (el->l_tree_depth) { + /* This is another operation where we want to be + * careful about our tree updates. An error here means + * none of the previous changes we made should roll + * forward. As a result, we have to record the buffers + * for this part of the tree in an array and reserve a + * journal write to them before making any changes. */ + num_bhs = le16_to_cpu(fe->id2.i_list.l_tree_depth); + eb_bhs = kcalloc(num_bhs, sizeof(struct buffer_head *), + GFP_KERNEL); + if (!eb_bhs) { + status = -ENOMEM; + mlog_errno(status); + goto bail; + } + + i = 0; + while(el->l_tree_depth) { + next_free = le16_to_cpu(el->l_next_free_rec); + if (next_free == 0) { + ocfs2_error(inode->i_sb, + "Dinode %"MLFu64" has a bad " + "extent list", + OCFS2_I(inode)->ip_blkno); + status = -EIO; + goto bail; + } + next_blkno = le64_to_cpu(el->l_recs[next_free - 1].e_blkno); + + BUG_ON(i >= num_bhs); + status = ocfs2_read_block(osb, next_blkno, &eb_bhs[i], + OCFS2_BH_CACHED, inode); + if (status < 0) { + mlog_errno(status); + goto bail; + } + eb = (struct ocfs2_extent_block *) eb_bhs[i]->b_data; + if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { + OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, + eb); + status = -EIO; + goto bail; + } + + status = ocfs2_journal_access(handle, inode, eb_bhs[i], + OCFS2_JOURNAL_ACCESS_WRITE); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + el = &eb->h_list; + i++; + /* When we leave this loop, eb_bhs[num_bhs - 1] will + * hold the bottom-most leaf extent block. */ + } + BUG_ON(el->l_tree_depth); + + el = &fe->id2.i_list; + /* If we have tree depth, then the fe update is + * trivial, and we want to switch el out for the + * bottom-most leaf in order to update it with the + * actual extent data below. */ + next_free = le16_to_cpu(el->l_next_free_rec); + if (next_free == 0) { + ocfs2_error(inode->i_sb, + "Dinode %"MLFu64" has a bad " + "extent list", + OCFS2_I(inode)->ip_blkno); + status = -EIO; + goto bail; + } + le32_add_cpu(&el->l_recs[next_free - 1].e_clusters, + new_cluste |