X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Flvfs%2Ffsfilt_ext3.c;h=e65ce42a9e657da1b3263de9556ce3788d778164;hb=b5bf04fb27d8e3e455aec024d8b908a10434519f;hp=2c04a3775a0025ba3db92ec4d6e1730f66b43fc3;hpb=2a481e4f72a409938043049ed33d47bd0399e770;p=fs%2Flustre-release.git diff --git a/lustre/lvfs/fsfilt_ext3.c b/lustre/lvfs/fsfilt_ext3.c index 2c04a37..e65ce42 100644 --- a/lustre/lvfs/fsfilt_ext3.c +++ b/lustre/lvfs/fsfilt_ext3.c @@ -34,26 +34,43 @@ #include #include #include -#include #include -#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)) -#include -#else +#include +#include +#include +#include #include + +#include +#include +#include +#include +#include +#include + +#ifdef EXT3_MULTIBLOCK_ALLOCATOR +#include #endif -#include -#include -#include -#include -#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)) -#include -#include +#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15) +#define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS +#define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS +#else +#define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS(sb) +#define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS(sb) #endif +#ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB +/* for kernels 2.6.18 and later */ +#define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb) +#else +#define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS +#endif -static kmem_cache_t *fcb_cache; -static atomic_t fcb_cache_count = ATOMIC_INIT(0); +#define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks) +#define fsfilt_ext3_journal_stop(handle) ext3_journal_stop(handle) + +static cfs_mem_cache_t *fcb_cache; struct fsfilt_cb_data { struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */ @@ -66,7 +83,67 @@ struct fsfilt_cb_data { #ifndef EXT3_XATTR_INDEX_TRUSTED /* temporary until we hit l28 kernel */ #define EXT3_XATTR_INDEX_TRUSTED 4 #endif -#define XATTR_LUSTRE_MDS_LOV_EA "lov" + +static char *fsfilt_ext3_get_label(struct super_block *sb) +{ + return EXT3_SB(sb)->s_es->s_volume_name; +} + +static int fsfilt_ext3_set_label(struct super_block *sb, char *label) +{ + /* see e.g. fsfilt_ext3_write_record() */ + journal_t *journal; + handle_t *handle; + int err; + + journal = EXT3_SB(sb)->s_journal; + handle = journal_start(journal, 1); + if (IS_ERR(handle)) { + CERROR("can't start transaction\n"); + return(PTR_ERR(handle)); + } + + err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh); + if (err) + goto out; + + memcpy(EXT3_SB(sb)->s_es->s_volume_name, label, + sizeof(EXT3_SB(sb)->s_es->s_volume_name)); + + err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh); + +out: + journal_stop(handle); + + return(err); +} + +static char *fsfilt_ext3_uuid(struct super_block *sb) +{ + return EXT3_SB(sb)->s_es->s_uuid; +} + +#ifdef HAVE_DISK_INODE_VERSION +/* + * Get the 64-bit version for an inode. + */ +static __u64 fsfilt_ext3_get_version(struct inode *inode) +{ + return EXT3_I(inode)->i_fs_version; +} + +/* + * Set the 64-bit version and return the old version. + */ +static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version) +{ + __u64 old_version = EXT3_I(inode)->i_fs_version; + + (EXT3_I(inode))->i_fs_version = new_version; + return old_version; +} + +#endif /* * We don't currently need any additional blocks for rmdir and @@ -77,8 +154,8 @@ struct fsfilt_cb_data { static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private, int logs) { - /* For updates to the last recieved file */ - int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS; + /* For updates to the last received file */ + int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb); journal_t *journal; void *handle; @@ -88,40 +165,39 @@ static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private, goto journal_start; } - /* XXX BUG 3188 -- must return to one set of opcodes */ - /* FIXME - cache hook */ - if (op & 0x20) { - nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS; - op = op & ~0x20; - } - - /* FIXME - kml */ - if (op & 0x10) { - nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS; - op = op & ~0x10; - } - switch(op) { case FSFILT_OP_RMDIR: case FSFILT_OP_UNLINK: /* delete one file + create/update logs for each stripe */ - nblocks += EXT3_DELETE_TRANS_BLOCKS; + nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb); nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS + - EXT3_SINGLEDATA_TRANS_BLOCKS) * logs; + FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs; break; case FSFILT_OP_RENAME: /* modify additional directory */ - nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS; + nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb); /* no break */ case FSFILT_OP_SYMLINK: /* additional block + block bitmap + GDT for long symlink */ nblocks += 3; /* no break */ - case FSFILT_OP_CREATE: - /* create/update logs for each stripe */ - nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS + - EXT3_SINGLEDATA_TRANS_BLOCKS) * logs; + case FSFILT_OP_CREATE: { +#if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) + static int warned; + if (!warned) { + if (!test_opt(inode->i_sb, EXTENTS)) { + warned = 1; + } else if (((EXT3_I(inode)->i_flags & + cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) == + cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) { + CWARN("extent-mapped directory found - contact " + "CFS: support@clusterfs.com\n"); + warned = 1; + } + } +#endif /* no break */ + } case FSFILT_OP_MKDIR: case FSFILT_OP_MKNOD: /* modify one inode + block bitmap + GDT */ @@ -130,20 +206,38 @@ static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private, case FSFILT_OP_LINK: /* modify parent directory */ nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS + - EXT3_DATA_TRANS_BLOCKS; + FSFILT_DATA_TRANS_BLOCKS(inode->i_sb); + /* create/update logs for each stripe */ + nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS + + FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs; break; case FSFILT_OP_SETATTR: /* Setattr on inode */ nblocks += 1; + nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS + + FSFILT_DATA_TRANS_BLOCKS(inode->i_sb); + /* quota chown log for each stripe */ + nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS + + FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs; break; case FSFILT_OP_CANCEL_UNLINK: /* blocks for log header bitmap update OR * blocks for catalog header bitmap update + unlink of logs */ nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) + - EXT3_DELETE_TRANS_BLOCKS * logs; + FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs; break; - case FSFILT_OP_NOOP: - nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS; + case FSFILT_OP_JOIN: + /* delete 2 file(file + array id) + create 1 file (array id) + * create/update logs for each stripe */ + nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb); + + /*create array log for head file*/ + nblocks += 3; + nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS + + FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)); + /*update head file array */ + nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS + + FSFILT_DATA_TRANS_BLOCKS(inode->i_sb); break; default: CERROR("unknown transaction start op %d\n", op); LBUG(); @@ -152,15 +246,14 @@ static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private, LASSERT(current->journal_info == desc_private); journal = EXT3_SB(inode->i_sb)->s_journal; if (nblocks > journal->j_max_transaction_buffers) { - CERROR("too many credits %d for op %ux%u using %d instead\n", + CWARN("too many credits %d for op %ux%u using %d instead\n", nblocks, op, logs, journal->j_max_transaction_buffers); nblocks = journal->j_max_transaction_buffers; } journal_start: - lock_kernel(); - handle = journal_start(EXT3_JOURNAL(inode), nblocks); - unlock_kernel(); + LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks); + handle = fsfilt_ext3_journal_start(inode, nblocks); if (!IS_ERR(handle)) LASSERT(current->journal_info == handle); @@ -199,7 +292,7 @@ static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso, { struct super_block *sb = fso->fso_dentry->d_inode->i_sb; __u64 next_indir; - const int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits); + const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits); int nbitmaps = 0, ngdblocks; int needed = objcount + 1; /* inodes + superblock */ int i, j; @@ -240,14 +333,14 @@ static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso, needed += nbitmaps + ngdblocks; /* last_rcvd update */ - needed += EXT3_DATA_TRANS_BLOCKS; + needed += FSFILT_DATA_TRANS_BLOCKS(sb); -#if defined(CONFIG_QUOTA) && !defined(__x86_64__) /* XXX */ +#if defined(CONFIG_QUOTA) /* We assume that there will be 1 bit set in s_dquot.flags for each * quota file that is active. This is at least true for now. */ needed += hweight32(sb_any_quota_enabled(sb)) * - EXT3_SINGLEDATA_TRANS_BLOCKS; + FSFILT_SINGLEDATA_TRANS_BLOCKS(sb); #endif return needed; @@ -296,9 +389,8 @@ static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso, needed = journal->j_max_transaction_buffers; } - lock_kernel(); - handle = journal_start(journal, needed); - unlock_kernel(); + LASSERTF(needed > 0, "can't start %d credit transaction\n", needed); + handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed); if (IS_ERR(handle)) { CERROR("can't get handle for %d credits: rc = %ld\n", needed, PTR_ERR(handle)); @@ -310,8 +402,26 @@ static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso, RETURN(handle); } -static int fsfilt_ext3_commit(struct super_block *sb, struct inode *inode, - void *h, int force_sync) +static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h) +{ + handle_t *handle = h; + + /* fsfilt_extend called with nblocks = 0 for testing in special cases */ + if (nblocks == 0) { + handle->h_buffer_credits = 0; + CWARN("setting credits of handle %p to zero by request\n", h); + } + + if (handle->h_buffer_credits > nblocks) + return 0; + if (journal_extend(handle, nblocks) == 0) + return 0; + + ext3_mark_inode_dirty(handle, inode); + return journal_restart(handle, nblocks); +} + +static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync) { int rc; handle_t *handle = h; @@ -320,9 +430,7 @@ static int fsfilt_ext3_commit(struct super_block *sb, struct inode *inode, if (force_sync) handle->h_sync = 1; /* recovery likes this */ - lock_kernel(); - rc = journal_stop(handle); - unlock_kernel(); + rc = fsfilt_ext3_journal_stop(handle); return rc; } @@ -332,36 +440,23 @@ static int fsfilt_ext3_commit_async(struct inode *inode, void *h, { unsigned long tid; transaction_t *transaction; -#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0) - unsigned long rtid; -#endif handle_t *handle = h; journal_t *journal; int rc; LASSERT(current->journal_info == handle); - lock_kernel(); transaction = handle->h_transaction; journal = transaction->t_journal; tid = transaction->t_tid; /* we don't want to be blocked */ handle->h_sync = 0; - rc = journal_stop(handle); + rc = fsfilt_ext3_journal_stop(handle); if (rc) { CERROR("error while stopping transaction: %d\n", rc); - unlock_kernel(); return rc; } -#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0) - rtid = log_start_commit(journal, transaction); - if (rtid != tid) - CERROR("strange race: %lu != %lu\n", - (unsigned long) tid, (unsigned long) rtid); -#else - log_start_commit(journal, transaction->t_tid); -#endif - unlock_kernel(); + log_start_commit(journal, tid); *wait_handle = (void *) tid; CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid); @@ -370,14 +465,17 @@ static int fsfilt_ext3_commit_async(struct inode *inode, void *h, static int fsfilt_ext3_commit_wait(struct inode *inode, void *h) { + journal_t *journal = EXT3_JOURNAL(inode); tid_t tid = (tid_t)(long)h; CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid); - if (is_journal_aborted(EXT3_JOURNAL(inode))) + if (unlikely(is_journal_aborted(journal))) return -EIO; log_wait_commit(EXT3_JOURNAL(inode), tid); + if (unlikely(is_journal_aborted(journal))) + return -EIO; return 0; } @@ -385,30 +483,44 @@ static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle, struct iattr *iattr, int do_trunc) { struct inode *inode = dentry->d_inode; - int rc; - - lock_kernel(); + int rc = 0; - /* A _really_ horrible hack to avoid removing the data stored - * in the block pointers; this is really the "small" stripe MD data. - * We can avoid further hackery by virtue of the MDS file size being - * zero all the time (which doesn't invoke block truncate at unlink - * time), so we assert we never change the MDS file size from zero. */ + /* Avoid marking the inode dirty on the superblock list unnecessarily. + * We are already writing the inode to disk as part of this + * transaction and want to avoid a lot of extra inode writeout + * later on. b=9828 */ if (iattr->ia_valid & ATTR_SIZE && !do_trunc) { /* ATTR_SIZE would invoke truncate: clear it */ iattr->ia_valid &= ~ATTR_SIZE; - EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size; - - /* make sure _something_ gets set - so new inode - * goes to disk (probably won't work over XFS */ - if (!(iattr->ia_valid & (ATTR_MODE | ATTR_MTIME | ATTR_CTIME))){ - iattr->ia_valid |= ATTR_MODE; - iattr->ia_mode = inode->i_mode; + EXT3_I(inode)->i_disksize = iattr->ia_size; + i_size_write(inode, iattr->ia_size); + + if (iattr->ia_valid & ATTR_UID) + inode->i_uid = iattr->ia_uid; + if (iattr->ia_valid & ATTR_GID) + inode->i_gid = iattr->ia_gid; + if (iattr->ia_valid & ATTR_ATIME) + inode->i_atime = iattr->ia_atime; + if (iattr->ia_valid & ATTR_MTIME) + inode->i_mtime = iattr->ia_mtime; + if (iattr->ia_valid & ATTR_CTIME) + inode->i_ctime = iattr->ia_ctime; + if (iattr->ia_valid & ATTR_MODE) { + inode->i_mode = iattr->ia_mode; + + if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) + inode->i_mode &= ~S_ISGID; } + + inode->i_sb->s_op->dirty_inode(inode); + + goto out; } /* Don't allow setattr to change file type */ - iattr->ia_mode = (inode->i_mode & S_IFMT)|(iattr->ia_mode & ~S_IFMT); + if (iattr->ia_valid & ATTR_MODE) + iattr->ia_mode = (inode->i_mode & S_IFMT) | + (iattr->ia_mode & ~S_IFMT); /* We set these flags on the client, but have already checked perms * so don't confuse inode_change_ok. */ @@ -422,9 +534,8 @@ static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle, rc = inode_setattr(inode, iattr); } - unlock_kernel(); - - return rc; + out: + RETURN(rc); } static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file, @@ -433,6 +544,12 @@ static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file, int rc = 0; ENTRY; + /* FIXME: Can't do this because of nested transaction deadlock */ + if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) { + CERROR("can't set data journal flag on file\n"); + RETURN(-EPERM); + } + if (inode->i_fop->ioctl) rc = inode->i_fop->ioctl(inode, file, cmd, arg); else @@ -442,38 +559,32 @@ static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file, } static int fsfilt_ext3_set_md(struct inode *inode, void *handle, - void *lmm, int lmm_size) + void *lmm, int lmm_size, const char *name) { int rc; - /* keep this when we get rid of OLD_EA (too noisy during conversion) */ - if (EXT3_I(inode)->i_file_acl /* || large inode EA flag */) - CWARN("setting EA on %lu/%u again... interesting\n", - inode->i_ino, inode->i_generation); + LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0); - lock_kernel(); rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED, - XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size, 0); + name, lmm, lmm_size, 0); - unlock_kernel(); - if (rc) + if (rc && rc != -EROFS) CERROR("error adding MD data to inode %lu: rc = %d\n", inode->i_ino, rc); return rc; } -/* Must be called with i_sem held */ -static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size) +/* Must be called with i_mutex held */ +static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size, + const char *name) { int rc; - LASSERT(down_trylock(&inode->i_sem) != 0); - lock_kernel(); + LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0); rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED, - XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size); - unlock_kernel(); + name, lmm, lmm_size); /* This gives us the MD size */ if (lmm == NULL) @@ -481,7 +592,7 @@ static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size) if (rc < 0) { CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n", - EXT3_XATTR_INDEX_TRUSTED, XATTR_LUSTRE_MDS_LOV_EA, + EXT3_XATTR_INDEX_TRUSTED, name, inode->i_ino, rc); memset(lmm, 0, lmm_size); return (rc == -ENODATA) ? 0 : rc; @@ -490,56 +601,11 @@ static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size) return rc; } -#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0)) -static int fsfilt_ext3_send_bio(struct inode *inode, struct bio *bio) +static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio) { - submit_bio(WRITE, bio); + submit_bio(rw, bio); return 0; } -#else -static int fsfilt_ext3_send_bio(struct inode *inode, struct kiobuf *bio) -{ - int rc, blocks_per_page; - - rc = brw_kiovec(WRITE, 1, &bio, inode->i_dev, - bio->blocks, 1 << inode->i_blkbits); - - blocks_per_page = PAGE_SIZE >> inode->i_blkbits; - - if (rc != (1 << inode->i_blkbits) * bio->nr_pages * - blocks_per_page) { - CERROR("short write? expected %d, wrote %d\n", - (1 << inode->i_blkbits) * bio->nr_pages * - blocks_per_page, rc); - } - - return rc; -} -#endif - -/* FIXME-UMKA: This should be used in 2.6.x io code later. */ -static struct page *fsfilt_ext3_getpage(struct inode *inode, long int index) -{ - int rc; - struct page *page; - - page = grab_cache_page(inode->i_mapping, index); - if (page == NULL) - return ERR_PTR(-ENOMEM); - - if (PageUptodate(page)) { - unlock_page(page); - return page; - } - - rc = inode->i_mapping->a_ops->readpage(NULL, page); - if (rc < 0) { - page_cache_release(page); - return ERR_PTR(rc); - } - - return page; -} static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count, loff_t *off) @@ -559,7 +625,7 @@ static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count, struct buffer_head *bh; bh = NULL; - if (*off < inode->i_size) { + if (*off < i_size_read(inode)) { int err = 0; bh = ext3_bread(NULL, inode, *off >> blkbits, @@ -585,7 +651,7 @@ static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count, CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off); memset(fake, 0, sizeof(*fake)); - fake->rec_len = cpu_to_le32(blksize); + fake->rec_len = cpu_to_le16(blksize); } count -= blksize; buf += blksize; @@ -604,32 +670,26 @@ static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error) fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error); OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb); - atomic_dec(&fcb_cache_count); } -static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, - struct super_block *sb, - __u64 last_rcvd, +static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd, void *handle, fsfilt_cb_t cb_func, void *cb_data) { struct fsfilt_cb_data *fcb; - OBD_SLAB_ALLOC(fcb, fcb_cache, GFP_NOFS, sizeof *fcb); + OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb); if (fcb == NULL) RETURN(-ENOMEM); - atomic_inc(&fcb_cache_count); fcb->cb_func = cb_func; fcb->cb_obd = obd; fcb->cb_last_rcvd = last_rcvd; fcb->cb_data = cb_data; CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd); - lock_kernel(); journal_callback_set(handle, fsfilt_ext3_cb_func, (struct journal_callback *)fcb); - unlock_kernel(); return 0; } @@ -648,7 +708,7 @@ static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs) memset(&sfs, 0, sizeof(sfs)); - rc = sb->s_op->statfs(sb, &sfs); + rc = ll_do_statfs(sb, &sfs); if (!rc && sfs.f_bfree < sfs.f_ffree) { sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree; @@ -664,7 +724,53 @@ static int fsfilt_ext3_sync(struct super_block *sb) return ext3_force_commit(sb); } +#if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK)) +#warning "kernel code has old extents/mballoc patch, disabling" +#undef EXT3_MULTIBLOCK_ALLOCATOR +#endif +#ifndef EXT3_EXTENTS_FL +#define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */ +#endif + #ifdef EXT3_MULTIBLOCK_ALLOCATOR +#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)) +#define ext3_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem); +#define ext3_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem); +#else +#define ext3_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex); +#define ext3_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex); +#endif + +#ifndef EXT_ASSERT +#define EXT_ASSERT(cond) BUG_ON(!(cond)) +#endif + +#ifdef EXT3_EXT_HAS_NO_TREE +/* for kernels 2.6.18 and later */ +#define ext3_ext_base inode +#define ext3_ext_base2inode(inode) (inode) +#define EXT_DEPTH(inode) ext_depth(inode) +#define EXT_GENERATION(inode) ext_generation(inode) +#define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \ + ext3_ext_walk_space(inode, block, num, cb, cbdata); +#else +#define ext3_ext_base ext3_extents_tree +#define ext3_ext_base2inode(tree) (tree->inode) +#define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \ + ext3_ext_walk_space(tree, block, num, cb); +#endif + +#include +#if EXT3_EXT_MAGIC == 0xf301 +#define ee_start e_start +#define ee_block e_block +#define ee_len e_num +#endif +#ifndef EXT3_BB_MAX_BLOCKS +#define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \ + ext3_new_blocks(handle, inode, count, goal, err) +#endif + struct bpointers { unsigned long *blocks; int *created; @@ -675,65 +781,135 @@ struct bpointers { }; static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path, - unsigned long block) + unsigned long block, int *aflags) +{ + struct ext3_inode_info *ei = EXT3_I(inode); + unsigned long bg_start; + unsigned long colour; + int depth; + + if (path) { + struct ext3_extent *ex; + depth = path->p_depth; + + /* try to predict block placement */ + if ((ex = path[depth].p_ext)) { +#if 0 + /* This prefers to eat into a contiguous extent + * rather than find an extent that the whole + * request will fit into. This can fragment data + * block allocation and prevents our lovely 1M I/Os + * from reaching the disk intact. */ + if (ex->ee_block + ex->ee_len == block) + *aflags |= 1; +#endif + return ex->ee_start + (block - ex->ee_block); + } + + /* it looks index is empty + * try to find starting from index itself */ + if (path[depth].p_bh) + return path[depth].p_bh->b_blocknr; + } + + /* OK. use inode's group */ + bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) + + le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block); + colour = (current->pid % 16) * + (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16); + return bg_start + colour + block; +} + +#define ll_unmap_underlying_metadata(sb, blocknr) \ + unmap_underlying_metadata((sb)->s_bdev, blocknr) + +#ifndef EXT3_MB_HINT_GROUP_ALLOC +static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base, + struct ext3_ext_path *path, unsigned long block, + unsigned long *count, int *err) +{ + unsigned long pblock, goal; + int aflags = 0; + struct inode *inode = ext3_ext_base2inode(base); + + goal = ext3_ext_find_goal(inode, path, block, &aflags); + aflags |= 2; /* block have been already reserved */ + pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err); + return pblock; + +} +#else +static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base, + struct ext3_ext_path *path, unsigned long block, + unsigned long *count, int *err) { - struct ext3_inode_info *ei = EXT3_I(inode); - unsigned long bg_start; - unsigned long colour; - int depth; - - if (path) { - struct ext3_extent *ex; - depth = path->p_depth; - - /* try to predict block placement */ - if ((ex = path[depth].p_ext)) - return ex->ee_start + (block - ex->ee_block); - - /* it looks index is empty - * try to find starting from index itself */ - if (path[depth].p_bh) - return path[depth].p_bh->b_blocknr; - } - - /* OK. use inode's group */ - bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) + - le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block); - colour = (current->pid % 16) * - (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16); - return bg_start + colour + block; + struct inode *inode = ext3_ext_base2inode(base); + struct ext3_allocation_request ar; + unsigned long pblock; + int aflags; + + /* find neighbour allocated blocks */ + ar.lleft = block; + *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft); + if (*err) + return 0; + ar.lright = block; + *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright); + if (*err) + return 0; + + /* allocate new block */ + ar.goal = ext3_ext_find_goal(inode, path, block, &aflags); + ar.inode = inode; + ar.logical = block; + ar.len = *count; + ar.flags = EXT3_MB_HINT_DATA; + pblock = ext3_mb_new_blocks(handle, &ar, err); + *count = ar.len; + return pblock; + } +#endif -static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree, +#ifdef EXT3_EXT_HAS_NO_TREE +static int ext3_ext_new_extent_cb(struct ext3_ext_base *base, + struct ext3_ext_path *path, + struct ext3_ext_cache *cex, + void *cbdata) +{ + struct bpointers *bp = cbdata; +#else +static int ext3_ext_new_extent_cb(struct ext3_ext_base *base, struct ext3_ext_path *path, - struct ext3_extent *newex, int exist) + struct ext3_ext_cache *cex) { - struct inode *inode = tree->inode; - struct bpointers *bp = tree->private; - int count, err, goal; + struct bpointers *bp = base->private; +#endif + struct inode *inode = ext3_ext_base2inode(base); + struct ext3_extent nex; unsigned long pblock; unsigned long tgen; - loff_t new_i_size; + int err, i; + unsigned long count; handle_t *handle; - int i; - i = EXT_DEPTH(tree); + i = EXT_DEPTH(base); EXT_ASSERT(i == path->p_depth); EXT_ASSERT(path[i].p_hdr); - if (exist) { + if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) { err = EXT_CONTINUE; goto map; } if (bp->create == 0) { i = 0; - if (newex->ee_block < bp->start) - i = bp->start - newex->ee_block; - if (i >= newex->ee_len) + if (cex->ec_block < bp->start) + i = bp->start - cex->ec_block; + if (i >= cex->ec_len) CERROR("nothing to do?! i = %d, e_num = %u\n", - i, newex->ee_len); - for (; i < newex->ee_len && bp->num; i++) { + i, cex->ec_len); + for (; i < cex->ec_len && bp->num; i++) { *(bp->created) = 0; bp->created++; *(bp->blocks) = 0; @@ -745,56 +921,53 @@ static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree, return EXT_CONTINUE; } - tgen = EXT_GENERATION(tree); - count = ext3_ext_calc_credits_for_insert(tree, path); - up_write(&EXT3_I(inode)->truncate_sem); + tgen = EXT_GENERATION(base); + count = ext3_ext_calc_credits_for_insert(base, path); + ext3_up_truncate_sem(inode); - lock_kernel(); - handle = journal_start(EXT3_JOURNAL(inode), count + EXT3_ALLOC_NEEDED + 1); - unlock_kernel(); + handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1); if (IS_ERR(handle)) { - down_write(&EXT3_I(inode)->truncate_sem); + ext3_down_truncate_sem(inode); return PTR_ERR(handle); } - if (tgen != EXT_GENERATION(tree)) { + ext3_down_truncate_sem(inode); + if (tgen != EXT_GENERATION(base)) { /* the tree has changed. so path can be invalid at moment */ - lock_kernel(); - journal_stop(handle); - unlock_kernel(); - down_write(&EXT3_I(inode)->truncate_sem); + fsfilt_ext3_journal_stop(handle); return EXT_REPEAT; } - down_write(&EXT3_I(inode)->truncate_sem); - goal = ext3_ext_find_goal(inode, path, newex->ee_block); - count = newex->ee_len; - pblock = ext3_new_blocks(handle, inode, &count, goal, &err); + count = cex->ec_len; + pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err); if (!pblock) goto out; - EXT_ASSERT(count <= newex->ee_len); + EXT_ASSERT(count <= cex->ec_len); /* insert new extent */ - newex->ee_start = pblock; - newex->ee_len = count; - err = ext3_ext_insert_extent(handle, tree, path, newex); - if (err) + nex.ee_block = cex->ec_block; + nex.ee_start = pblock; + nex.ee_len = count; + err = ext3_ext_insert_extent(handle, base, path, &nex); + if (err) { + CERROR("can't insert extent: %d\n", err); + /* XXX: export ext3_free_blocks() */ + /*ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);*/ goto out; - - /* correct on-disk inode size */ - if (newex->ee_len > 0) { - new_i_size = (loff_t) newex->ee_block + newex->ee_len; - new_i_size = new_i_size << inode->i_blkbits; - if (new_i_size > EXT3_I(inode)->i_disksize) { - EXT3_I(inode)->i_disksize = new_i_size; - err = ext3_mark_inode_dirty(handle, inode); - } } + /* + * Putting len of the actual extent we just inserted, + * we are asking ext3_ext_walk_space() to continue + * scaning after that block + */ + cex->ec_len = nex.ee_len; + cex->ec_start = nex.ee_start; + BUG_ON(nex.ee_len == 0); + BUG_ON(nex.ee_block != cex->ec_block); + out: - lock_kernel(); - journal_stop(handle); - unlock_kernel(); + fsfilt_ext3_journal_stop(handle); map: if (err >= 0) { /* map blocks */ @@ -803,21 +976,30 @@ map: CERROR("initial space: %lu:%u\n", bp->start, bp->init_num); CERROR("current extent: %u/%u/%u %d\n", - newex->ee_block, newex->ee_len, - newex->ee_start, exist); + cex->ec_block, cex->ec_len, + cex->ec_start, cex->ec_type); } i = 0; - if (newex->ee_block < bp->start) - i = bp->start - newex->ee_block; - if (i >= newex->ee_len) + if (cex->ec_block < bp->start) + i = bp->start - cex->ec_block; + if (i >= cex->ec_len) CERROR("nothing to do?! i = %d, e_num = %u\n", - i, newex->ee_len); - for (; i < newex->ee_len && bp->num; i++) { - *(bp->created) = (exist == 0 ? 1 : 0); + i, cex->ec_len); + for (; i < cex->ec_len && bp->num; i++) { + *(bp->blocks) = cex->ec_start + i; + if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) { + *(bp->created) = 0; + } else { + *(bp->created) = 1; + /* unmap any possible underlying metadata from + * the block device mapping. bug 6998. */ + ll_unmap_underlying_metadata(inode->i_sb, + *(bp->blocks)); + } bp->created++; - *(bp->blocks) = newex->ee_start + i; bp->blocks++; bp->num--; + bp->start++; } } return err; @@ -827,25 +1009,33 @@ int fsfilt_map_nblocks(struct inode *inode, unsigned long block, unsigned long num, unsigned long *blocks, int *created, int create) { +#ifdef EXT3_EXT_HAS_NO_TREE + struct ext3_ext_base *base = inode; +#else struct ext3_extents_tree tree; + struct ext3_ext_base *base = &tree; +#endif struct bpointers bp; int err; CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n", - block, block + num, (unsigned) inode->i_ino); + block, block + num - 1, (unsigned) inode->i_ino); - ext3_init_tree_desc(&tree, inode); +#ifndef EXT3_EXT_HAS_NO_TREE + ext3_init_tree_desc(base, inode); tree.private = &bp; +#endif bp.blocks = blocks; bp.created = created; bp.start = block; bp.init_num = bp.num = num; bp.create = create; - down_write(&EXT3_I(inode)->truncate_sem); - err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb); - ext3_ext_invalidate_cache(&tree); - up_write(&EXT3_I(inode)->truncate_sem); + ext3_down_truncate_sem(inode); + err = fsfilt_ext3_ext_walk_space(base, block, num, + ext3_ext_new_extent_cb, &bp); + ext3_ext_invalidate_cache(base); + ext3_up_truncate_sem(inode); return err; } @@ -854,7 +1044,7 @@ int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page, int pages, unsigned long *blocks, int *created, int create) { - int blocks_per_page = PAGE_SIZE >> inode->i_blkbits; + int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits; int rc = 0, i = 0; struct page *fp = NULL; int clen = 0; @@ -899,7 +1089,7 @@ int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page, cleanup: return rc; } -#endif +#endif /* EXT3_MULTIBLOCK_ALLOCATOR */ extern int ext3_map_inode_page(struct inode *inode, struct page *page, unsigned long *blocks, int *created, int create); @@ -907,7 +1097,7 @@ int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page, int pages, unsigned long *blocks, int *created, int create) { - int blocks_per_page = PAGE_SIZE >> inode->i_blkbits; + int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits; unsigned long *b; int rc = 0, i, *cr; @@ -948,30 +1138,20 @@ int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page, return rc; } -extern int ext3_prep_san_write(struct inode *inode, long *blocks, - int nblocks, loff_t newsize); -static int fsfilt_ext3_prep_san_write(struct inode *inode, long *blocks, - int nblocks, loff_t newsize) -{ - return ext3_prep_san_write(inode, blocks, nblocks, newsize); -} - -static int fsfilt_ext3_read_record(struct file * file, void *buf, - int size, loff_t *offs) +int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs) { - struct inode *inode = file->f_dentry->d_inode; unsigned long block; struct buffer_head *bh; - int err, blocksize, csize, boffs; + int err, blocksize, csize, boffs, osize = size; /* prevent reading after eof */ lock_kernel(); - if (inode->i_size < *offs + size) { - size = inode->i_size - *offs; + if (i_size_read(inode) < *offs + size) { + size = i_size_read(inode) - *offs; unlock_kernel(); if (size < 0) { CERROR("size %llu is too short for read %u@%llu\n", - inode->i_size, size, *offs); + i_size_read(inode), size, *offs); return -EIO; } else if (size == 0) { return 0; @@ -999,35 +1179,28 @@ static int fsfilt_ext3_read_record(struct file * file, void *buf, buf += csize; size -= csize; } - return 0; + return osize; } +EXPORT_SYMBOL(fsfilt_ext3_read); -static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize, - loff_t *offs, int force_sync) +static int fsfilt_ext3_read_record(struct file * file, void *buf, + int size, loff_t *offs) +{ + int rc; + rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs); + if (rc > 0) + rc = 0; + return rc; +} + +int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize, + loff_t *offs, handle_t *handle) { struct buffer_head *bh = NULL; + loff_t old_size = i_size_read(inode), offset = *offs; + loff_t new_size = i_size_read(inode); unsigned long block; - struct inode *inode = file->f_dentry->d_inode; - loff_t old_size = inode->i_size, offset = *offs; - loff_t new_size = inode->i_size; - journal_t *journal; - handle_t *handle; - int err, block_count = 0, blocksize, size, boffs; - - /* Determine how many transaction credits are needed */ - blocksize = 1 << inode->i_blkbits; - block_count = (*offs & (blocksize - 1)) + bufsize; - block_count = (block_count + blocksize - 1) >> inode->i_blkbits; - - journal = EXT3_SB(inode->i_sb)->s_journal; - lock_kernel(); - handle = journal_start(journal, - block_count * EXT3_DATA_TRANS_BLOCKS + 2); - unlock_kernel(); - if (IS_ERR(handle)) { - CERROR("can't start transaction\n"); - return PTR_ERR(handle); - } + int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs; while (bufsize > 0) { if (bh != NULL) @@ -1039,14 +1212,14 @@ static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize, bh = ext3_bread(handle, inode, block, 1, &err); if (!bh) { CERROR("can't read/create block: %d\n", err); - goto out; + break; } err = ext3_journal_get_write_access(handle, bh); if (err) { CERROR("journal_get_write_access() returned error %d\n", err); - goto out; + break; } LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size); memcpy(bh->b_data + boffs, buf, size); @@ -1054,7 +1227,7 @@ static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize, if (err) { CERROR("journal_dirty_metadata() returned error %d\n", err); - goto out; + break; } if (offset + size > new_size) new_size = offset + size; @@ -1062,39 +1235,59 @@ static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize, bufsize -= size; buf += size; } - - if (force_sync) - handle->h_sync = 1; /* recovery likes this */ -out: if (bh) brelse(bh); /* correct in-core and on-disk sizes */ - if (new_size > inode->i_size) { + if (new_size > i_size_read(inode)) { lock_kernel(); - if (new_size > inode->i_size) - inode->i_size = new_size; - if (inode->i_size > EXT3_I(inode)->i_disksize) - EXT3_I(inode)->i_disksize = inode->i_size; - if (inode->i_size > old_size) + if (new_size > i_size_read(inode)) + i_size_write(inode, new_size); + if (i_size_read(inode) > EXT3_I(inode)->i_disksize) + EXT3_I(inode)->i_disksize = i_size_read(inode); + if (i_size_read(inode) > old_size) mark_inode_dirty(inode); unlock_kernel(); } - lock_kernel(); - journal_stop(handle); - unlock_kernel(); - if (err == 0) *offs = offset; return err; } +EXPORT_SYMBOL(fsfilt_ext3_write_handle); -static int fsfilt_ext3_setup(struct obd_device *obd, struct super_block *sb) +static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize, + loff_t *offs, int force_sync) +{ + struct inode *inode = file->f_dentry->d_inode; + handle_t *handle; + int err, block_count = 0, blocksize; + + /* Determine how many transaction credits are needed */ + blocksize = 1 << inode->i_blkbits; + block_count = (*offs & (blocksize - 1)) + bufsize; + block_count = (block_count + blocksize - 1) >> inode->i_blkbits; + + handle = fsfilt_ext3_journal_start(inode, + block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2); + if (IS_ERR(handle)) { + CERROR("can't start transaction for %d blocks (%d bytes)\n", + block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize); + return PTR_ERR(handle); + } + + err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle); + + if (!err && force_sync) + handle->h_sync = 1; /* recovery likes this */ + + fsfilt_ext3_journal_stop(handle); + + return err; +} + +static int fsfilt_ext3_setup(struct super_block *sb) { -#ifdef EXT3_FEATURE_INCOMPAT_MDSNUM - struct mds_obd *mds = &obd->u.mds; -#endif #if 0 EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock; EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock; @@ -1104,152 +1297,17 @@ static int fsfilt_ext3_setup(struct obd_device *obd, struct super_block *sb) set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS); sb->s_flags |= S_PDIROPS; #endif - /* setup mdsnum in underlying fs */ -#ifdef EXT3_FEATURE_INCOMPAT_MDSNUM - if (mds->mds_lmv_obd) { - struct ext3_sb_info *sbi = EXT3_SB(sb); - struct ext3_super_block *es = sbi->s_es; - handle_t *handle; - int err; - - if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MDSNUM)) { - CWARN("%s: set mdsnum %d in ext3fs\n", - obd->obd_name, mds->mds_num); - lock_kernel(); - handle = journal_start(sbi->s_journal, 1); - unlock_kernel(); - LASSERT(!IS_ERR(handle)); - err = ext3_journal_get_write_access(handle, sbi->s_sbh); - LASSERT(err == 0); - EXT3_SET_INCOMPAT_FEATURE(sb, - EXT3_FEATURE_INCOMPAT_MDSNUM); - es->s_mdsnum = mds->mds_num; - err = ext3_journal_dirty_metadata(handle, sbi->s_sbh); - LASSERT(err == 0); - lock_kernel(); - journal_stop(handle); - unlock_kernel(); - } else { - CWARN("%s: mdsnum initialized to %u in ext3fs\n", - obd->obd_name, es->s_mdsnum); - } - sbi->s_mdsnum = es->s_mdsnum; - } + if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX)) + CWARN("filesystem doesn't have dir_index feature enabled\n"); +#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13)) && HAVE_QUOTA_SUPPORT + set_opt(EXT3_SB(sb)->s_mount_opt, QUOTA); #endif return 0; } -extern int ext3_add_dir_entry(struct dentry *dentry); -extern int ext3_del_dir_entry(struct dentry *dentry); - -static int fsfilt_ext3_add_dir_entry(struct obd_device *obd, - struct dentry *parent, - char *name, int namelen, - unsigned long ino, - unsigned long generation, - unsigned mds) -{ -#ifdef EXT3_FEATURE_INCOMPAT_MDSNUM - struct dentry *dentry; - int err; - LASSERT(ino != 0); - LASSERT(namelen != 0); - dentry = ll_lookup_one_len(name, parent, namelen); - if (IS_ERR(dentry)) { - CERROR("can't lookup %*s in %lu/%lu: %d\n", dentry->d_name.len, - dentry->d_name.name, dentry->d_inode->i_ino, - (unsigned long) dentry->d_inode->i_generation, - (int) PTR_ERR(dentry)); - RETURN(PTR_ERR(dentry)); - } - if (dentry->d_inode != NULL) { - CERROR("dentry %*s(0x%p) found\n", dentry->d_name.len, - dentry->d_name.name, dentry); - l_dput(dentry); - RETURN(-EEXIST); - } - - /* mds_reint_rename() may use this method to add dir entry - * that points onto local inode. and we don't want to find - * it cross-ref by subsequent lookups */ - d_drop(dentry); - - dentry->d_flags |= DCACHE_CROSS_REF; - dentry->d_inum = ino; - dentry->d_mdsnum = mds; - dentry->d_generation = generation; - lock_kernel(); - err = ext3_add_dir_entry(dentry); - unlock_kernel(); - - l_dput(dentry); - - RETURN(err); -#else -#error "rebuild kernel and lustre with ext3-mds-num patch!" - LASSERT(0); -#endif -} - -static int fsfilt_ext3_del_dir_entry(struct obd_device *obd, - struct dentry *dentry) -{ -#ifdef EXT3_FEATURE_INCOMPAT_MDSNUM - int err; - lock_kernel(); - err = ext3_del_dir_entry(dentry); - unlock_kernel(); - if (err == 0) - d_drop(dentry); - return err; -#else -#error "rebuild kernel and lustre with ext3-mds-num patch!" - LASSERT(0); -#endif -} - -static int fsfilt_ext3_set_xattr(struct inode * inode, void *handle, char *name, - void *buffer, int buffer_size) -{ - int rc = 0; - - lock_kernel(); - - rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED, - name, buffer, buffer_size, 0); - unlock_kernel(); - if (rc) - CERROR("set xattr %s from inode %lu: rc %d\n", - name, inode->i_ino, rc); - return rc; -} - -static int fsfilt_ext3_get_xattr(struct inode *inode, char *name, - void *buffer, int buffer_size) -{ - int rc = 0; - - lock_kernel(); - - rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED, - name, buffer, buffer_size); - unlock_kernel(); - - if (buffer == NULL) - return (rc == -ENODATA) ? 0 : rc; - if (rc < 0) { - CDEBUG(D_INFO, "error getting EA %s from inode %lu: rc %d\n", - name, inode->i_ino, rc); - memset(buffer, 0, buffer_size); - return (rc == -ENODATA) ? 0 : rc; - } - - return rc; -} - -/* If fso is NULL, op is FSFILT operation, otherwise op is number of fso - objects. Logs is number of logfiles to update */ -static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs) +/* If fso is NULL, op is FSFILT operation, otherwise op is number of fso + objects. Logs is number of logfiles to update */ +static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs) { if ( !fso ) { switch(op) { @@ -1263,7 +1321,7 @@ static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs) int i; int needed = 0; struct super_block *sb = fso->fso_dentry->d_inode->i_sb; - int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits); + int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits); int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp; for (i = 0; i < op; i++, fso++) { int nblocks = fso->fso_bufcnt * blockpp; @@ -1278,73 +1336,702 @@ static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs) return 0; } +static const char *op_quotafile[] = { "lquota.user", "lquota.group" }; + +#define DQINFO_COPY(out, in) \ +do { \ + Q_COPY(out, in, dqi_bgrace); \ + Q_COPY(out, in, dqi_igrace); \ + Q_COPY(out, in, dqi_flags); \ + Q_COPY(out, in, dqi_valid); \ +} while (0) + +#define DQBLK_COPY(out, in) \ +do { \ + Q_COPY(out, in, dqb_bhardlimit); \ + Q_COPY(out, in, dqb_bsoftlimit); \ + Q_COPY(out, in, dqb_curspace); \ + Q_COPY(out, in, dqb_ihardlimit); \ + Q_COPY(out, in, dqb_isoftlimit); \ + Q_COPY(out, in, dqb_curinodes); \ + Q_COPY(out, in, dqb_btime); \ + Q_COPY(out, in, dqb_itime); \ + Q_COPY(out, in, dqb_valid); \ +} while (0) + + + +static int fsfilt_ext3_quotactl(struct super_block *sb, + struct obd_quotactl *oqc) +{ + int i, rc = 0, error = 0; + struct quotactl_ops *qcop; + struct if_dqinfo *info; + struct if_dqblk *dqblk; + ENTRY; + + if (!sb->s_qcop) + RETURN(-ENOSYS); + + OBD_ALLOC_PTR(info); + if (!info) + RETURN(-ENOMEM); + OBD_ALLOC_PTR(dqblk); + if (!dqblk) { + OBD_FREE_PTR(info); + RETURN(-ENOMEM); + } + + DQINFO_COPY(info, &oqc->qc_dqinfo); + DQBLK_COPY(dqblk, &oqc->qc_dqblk); + + qcop = sb->s_qcop; + if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) { + for (i = 0; i < MAXQUOTAS; i++) { + if (!Q_TYPESET(oqc, i)) + continue; + + if (oqc->qc_cmd == Q_QUOTAON) { + if (!qcop->quota_on) + GOTO(out, rc = -ENOSYS); + rc = qcop->quota_on(sb, i, oqc->qc_id, + (char *)op_quotafile[i]); + } else if (oqc->qc_cmd == Q_QUOTAOFF) { + if (!qcop->quota_off) + GOTO(out, rc = -ENOSYS); + rc = qcop->quota_off(sb, i); + } + + if (rc == -EBUSY) + error = rc; + else if (rc) + GOTO(out, rc); + } + GOTO(out, rc ?: error); + } + + switch (oqc->qc_cmd) { + case Q_GETOINFO: + case Q_GETINFO: + if (!qcop->get_info) + GOTO(out, rc = -ENOSYS); + rc = qcop->get_info(sb, oqc->qc_type, info); + break; + case Q_SETQUOTA: + case Q_INITQUOTA: + if (!qcop->set_dqblk) + GOTO(out, rc = -ENOSYS); + rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk); + break; + case Q_GETOQUOTA: + case Q_GETQUOTA: + if (!qcop->get_dqblk) + GOTO(out, rc = -ENOSYS); + rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk); + break; + case Q_SYNC: + if (!sb->s_qcop->quota_sync) + GOTO(out, rc = -ENOSYS); + qcop->quota_sync(sb, oqc->qc_type); + break; + default: + CERROR("unsupported quotactl command: %d", oqc->qc_cmd); + LBUG(); + } +out: + DQINFO_COPY(&oqc->qc_dqinfo, info); + DQBLK_COPY(&oqc->qc_dqblk, dqblk); + + OBD_FREE_PTR(info); + OBD_FREE_PTR(dqblk); + + if (rc) + CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d " + "failed: %d\n", + oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc); + RETURN(rc); +} + +struct chk_dqblk{ + struct hlist_node dqb_hash; /* quotacheck hash */ + struct list_head dqb_list; /* in list also */ + qid_t dqb_id; /* uid/gid */ + short dqb_type; /* USRQUOTA/GRPQUOTA */ + __u32 dqb_bhardlimit; /* block hard limit */ + __u32 dqb_bsoftlimit; /* block soft limit */ + qsize_t dqb_curspace; /* current space */ + __u32 dqb_ihardlimit; /* inode hard limit */ + __u32 dqb_isoftlimit; /* inode soft limit */ + __u32 dqb_curinodes; /* current inodes */ + __u64 dqb_btime; /* block grace time */ + __u64 dqb_itime; /* inode grace time */ + __u32 dqb_valid; /* flag for above fields */ +}; -#define EXTENTS_EA "write_extents" -#define EXTENTS_EA_SIZE 64 +static inline unsigned int chkquot_hash(qid_t id, int type) + __attribute__((__const__)); -int ext3_ext_in_ea_alloc_space(struct inode *, int, const char *, unsigned long, unsigned long); -int ext3_ext_in_ea_remove_space(struct inode *, int, const char *, unsigned long, unsigned long); -int ext3_ext_in_ea_get_extents(struct inode *, int, const char *, char **, int *); -int ext3_ext_in_ea_get_extents_num(struct inode *, int, const char *, int *); +static inline unsigned int chkquot_hash(qid_t id, int type) +{ + return (id * (MAXQUOTAS - type)) % NR_DQHASH; +} -static int fsfilt_ext3_insert_extents_ea(struct inode *inode, - unsigned long from, - unsigned long num) +static inline struct chk_dqblk * +find_chkquot(struct hlist_head *head, qid_t id, int type) { - int rc = 0; + struct hlist_node *node; + struct chk_dqblk *cdqb; + + hlist_for_each(node, head) { + cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash); + if (cdqb->dqb_id == id && cdqb->dqb_type == type) + return cdqb; + } - rc = ext3_ext_in_ea_alloc_space(inode, EXT3_XATTR_INDEX_TRUSTED, - EXTENTS_EA, from, num); + return NULL; +} + +static struct chk_dqblk *alloc_chkquot(qid_t id, int type) +{ + struct chk_dqblk *cdqb; + + OBD_ALLOC_PTR(cdqb); + if (cdqb) { + INIT_HLIST_NODE(&cdqb->dqb_hash); + INIT_LIST_HEAD(&cdqb->dqb_list); + cdqb->dqb_id = id; + cdqb->dqb_type = type; + } + + return cdqb; +} + +static struct chk_dqblk * +cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list, + qid_t id, int type, int first_check) +{ + struct hlist_head *head = hash + chkquot_hash(id, type); + struct if_dqblk dqb; + struct chk_dqblk *cdqb; + int rc; + + cdqb = find_chkquot(head, id, type); + if (cdqb) + return cdqb; + + cdqb = alloc_chkquot(id, type); + if (!cdqb) + return NULL; + + if (!first_check) { + rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb); + if (rc) { + CERROR("get_dqblk of id %u, type %d failed: %d\n", + id, type, rc); + } else { + DQBLK_COPY(cdqb, &dqb); + cdqb->dqb_curspace = 0; + cdqb->dqb_curinodes = 0; + } + } + + hlist_add_head(&cdqb->dqb_hash, head); + list_add_tail(&cdqb->dqb_list, list); + + return cdqb; +} + +static inline int quota_onoff(struct super_block *sb, int cmd, int type) +{ + struct obd_quotactl *oqctl; + int rc; + + OBD_ALLOC_PTR(oqctl); + if (!oqctl) + RETURN(-ENOMEM); + + oqctl->qc_cmd = cmd; + oqctl->qc_id = QFMT_LDISKFS; + oqctl->qc_type = type; + rc = fsfilt_ext3_quotactl(sb, oqctl); + + OBD_FREE_PTR(oqctl); return rc; } -static int fsfilt_ext3_remove_extents_ea(struct inode *inode, - unsigned long from, - unsigned long num) +static inline int read_old_dqinfo(struct super_block *sb, int type, + struct if_dqinfo *dqinfo) { - int rc = 0; + struct obd_quotactl *oqctl; + int rc; + ENTRY; + + OBD_ALLOC_PTR(oqctl); + if (!oqctl) + RETURN(-ENOMEM); + + oqctl->qc_cmd = Q_GETINFO; + oqctl->qc_type = type; + rc = fsfilt_ext3_quotactl(sb, oqctl); + if (!rc) + ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo; + + OBD_FREE_PTR(oqctl); + RETURN(rc); +} + +static inline struct ext3_group_desc * +get_group_desc(struct super_block *sb, int group) +{ + unsigned long desc_block, desc; + struct ext3_group_desc *gdp; + + desc_block = group / EXT3_DESC_PER_BLOCK(sb); + desc = group % EXT3_DESC_PER_BLOCK(sb); + gdp = (struct ext3_group_desc *) + EXT3_SB(sb)->s_group_desc[desc_block]->b_data; + + return gdp + desc; +} + +static inline struct buffer_head * +read_inode_bitmap(struct super_block *sb, unsigned long group) +{ + struct ext3_group_desc *desc; + struct buffer_head *bh; + + desc = get_group_desc(sb, group); + bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap)); + + return bh; +} + +static inline struct inode *ext3_iget_inuse(struct super_block *sb, + struct buffer_head *bitmap_bh, + int index, unsigned long ino) +{ + struct inode *inode = NULL; + + if (ext3_test_bit(index, bitmap_bh->b_data)) + inode = iget(sb, ino); + + return inode; +} + +struct qchk_ctxt { + struct hlist_head qckt_hash[NR_DQHASH]; /* quotacheck hash */ + struct list_head qckt_list; /* quotacheck list */ + int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */ + struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */ +}; + +static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt, + struct obd_quotactl *oqc) +{ + struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, }; + loff_t size = 0; + qid_t qid[MAXQUOTAS]; + int cnt, i, rc = 0; + + if (!inode) + return 0; + + qid[USRQUOTA] = inode->i_uid; + qid[GRPQUOTA] = inode->i_gid; + + if (S_ISDIR(inode->i_mode) || + S_ISREG(inode->i_mode) || + S_ISLNK(inode->i_mode)) + size = inode_get_bytes(inode); + + for (cnt = 0; cnt < MAXQUOTAS; cnt++) { + if (!Q_TYPESET(oqc, cnt)) + continue; + + cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash, + &qctxt->qckt_list, qid[cnt], cnt, + qctxt->qckt_first_check[cnt]); + if (!cdqb[cnt]) { + rc = -ENOMEM; + break; + } + + cdqb[cnt]->dqb_curspace += size; + cdqb[cnt]->dqb_curinodes++; + } + + if (rc) { + for (i = 0; i < cnt; i++) { + if (!Q_TYPESET(oqc, i)) + continue; + LASSERT(cdqb[i]); + cdqb[i]->dqb_curspace -= size; + cdqb[i]->dqb_curinodes--; + } + } - rc = ext3_ext_in_ea_remove_space(inode, EXT3_XATTR_INDEX_TRUSTED, - EXTENTS_EA, from, num); return rc; } -extern int ext3_init_tree_in_ea(struct inode *inode, int name_index, - const char *eaname, int size); - -static int fsfilt_ext3_init_extents_ea(struct inode *inode) +static int v2_write_dqheader(struct file *f, int type) +{ + static const __u32 quota_magics[] = V2_INITQMAGICS; + static const __u32 quota_versions[] = V2_INITQVERSIONS; + struct v2_disk_dqheader dqhead; + loff_t offset = 0; + + CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions)); + LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics)); + + dqhead.dqh_magic = cpu_to_le32(quota_magics[type]); + dqhead.dqh_version = cpu_to_le32(quota_versions[type]); + + return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset); +} + +/* write dqinfo struct in a new quota file */ +static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info) +{ + struct v2_disk_dqinfo dqinfo; + __u32 blocks = V2_DQTREEOFF + 1; + loff_t offset = V2_DQINFOOFF; + + if (info) { + dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace); + dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace); + dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK & + ~DQF_INFO_DIRTY); + } else { + dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME); + dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME); + dqinfo.dqi_flags = 0; + } + + dqinfo.dqi_blocks = cpu_to_le32(blocks); + dqinfo.dqi_free_blk = 0; + dqinfo.dqi_free_entry = 0; + + return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset); +} + +static int create_new_quota_files(struct qchk_ctxt *qctxt, + struct obd_quotactl *oqc) +{ + int i, rc = 0; + ENTRY; + + for (i = 0; i < MAXQUOTAS; i++) { + struct if_dqinfo *info = qctxt->qckt_first_check[i]? + NULL : &qctxt->qckt_dqinfo[i]; + struct file *file; + + if (!Q_TYPESET(oqc, i)) + continue; + + file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC, + 0644); + if (IS_ERR(file)) { + rc = PTR_ERR(file); + CERROR("can't create %s file: rc = %d\n", + op_quotafile[i], rc); + GOTO(out, rc); + } + + if (!S_ISREG(file->f_dentry->d_inode->i_mode)) { + CERROR("file %s is not regular", op_quotafile[i]); + filp_close(file, 0); + GOTO(out, rc = -EINVAL); + } + + rc = v2_write_dqheader(file, i); + if (rc) { + filp_close(file, 0); + GOTO(out, rc); + } + + rc = v2_write_dqinfo(file, i, info); + filp_close(file, 0); + if (rc) + GOTO(out, rc); + } + +out: + RETURN(rc); +} + + +static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt, + struct chk_dqblk *cdqb) { + struct obd_quotactl *oqc; + long now; + int rc; + ENTRY; + + OBD_ALLOC_PTR(oqc); + if (!oqc) + RETURN(-ENOMEM); + + now = cfs_time_current_sec(); + + if (cdqb->dqb_bsoftlimit && + toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit && + !cdqb->dqb_btime) + cdqb->dqb_btime = + now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace; + + if (cdqb->dqb_isoftlimit && + cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit && + !cdqb->dqb_itime) + cdqb->dqb_itime = + now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace; + + cdqb->dqb_valid = QIF_ALL; + + oqc->qc_cmd = Q_SETQUOTA; + oqc->qc_type = cdqb->dqb_type; + oqc->qc_id = cdqb->dqb_id; + DQBLK_COPY(&oqc->qc_dqblk, cdqb); + + rc = fsfilt_ext3_quotactl(sb, oqc); + OBD_FREE_PTR(oqc); + RETURN(rc); +} + +static int prune_chkquots(struct super_block *sb, + struct qchk_ctxt *qctxt, int error) +{ + struct chk_dqblk *cdqb, *tmp; + int rc; + + list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) { + if (!error) { + rc = commit_chkquot(sb, qctxt, cdqb); + if (rc) + error = rc; + } + hlist_del_init(&cdqb->dqb_hash); + list_del(&cdqb->dqb_list); + OBD_FREE_PTR(cdqb); + } + + return error; +} + +static int fsfilt_ext3_quotacheck(struct super_block *sb, + struct obd_quotactl *oqc) +{ + struct ext3_sb_info *sbi = EXT3_SB(sb); + int i, group; + struct qchk_ctxt *qctxt; + struct buffer_head *bitmap_bh = NULL; + unsigned long ino; + struct inode *inode; int rc = 0; + ENTRY; - rc = ext3_init_tree_in_ea(inode, EXT3_XATTR_INDEX_TRUSTED, - EXTENTS_EA, 64); - return rc; + /* turn on quota and read dqinfo if existed */ + OBD_ALLOC_PTR(qctxt); + if (!qctxt) { + oqc->qc_stat = -ENOMEM; + RETURN(-ENOMEM); + } + + for (i = 0; i < NR_DQHASH; i++) + INIT_HLIST_HEAD(&qctxt->qckt_hash[i]); + INIT_LIST_HEAD(&qctxt->qckt_list); + + for (i = 0; i < MAXQUOTAS; i++) { + if (!Q_TYPESET(oqc, i)) + continue; + + rc = quota_onoff(sb, Q_QUOTAON, i); + if (!rc || rc == -EBUSY) { + rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo); + if (rc) + GOTO(out, rc); + } else if (rc == -ENOENT) { + qctxt->qckt_first_check[i] = 1; + } else if (rc) { + GOTO(out, rc); + } + } + + /* check quota and update in hash */ + for (group = 0; group < sbi->s_groups_count; group++) { + ino = group * sbi->s_inodes_per_group + 1; + bitmap_bh = read_inode_bitmap(sb, group); + if (!bitmap_bh) { + CERROR("read_inode_bitmap group %d failed", group); + GOTO(out, rc = -EIO); + } + + for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) { + if (ino < sbi->s_first_ino) + continue; + + inode = ext3_iget_inuse(sb, bitmap_bh, i, ino); + rc = add_inode_quota(inode, qctxt, oqc); + iput(inode); + if (rc) { + brelse(bitmap_bh); + GOTO(out, rc); + } + } + + brelse(bitmap_bh); + } + + /* read old quota limits from old quota file. (only for the user + * has limits but hasn't file) */ +#ifdef HAVE_QUOTA_SUPPORT + for (i = 0; i < MAXQUOTAS; i++) { + struct list_head id_list; + struct dquot_id *dqid, *tmp; + + if (!Q_TYPESET(oqc, i)) + continue; + + if (qctxt->qckt_first_check[i]) + continue; + + + LASSERT(sb_dqopt(sb)->files[i] != NULL); + INIT_LIST_HEAD(&id_list); +#ifndef KERNEL_SUPPORTS_QUOTA_READ + rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list); +#else + rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list); +#endif + if (rc) + CERROR("read old limits failed. (rc:%d)\n", rc); + + list_for_each_entry_safe(dqid, tmp, &id_list, di_link) { + list_del_init(&dqid->di_link); + + if (!rc) + cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list, + dqid->di_id, i, + qctxt->qckt_first_check[i]); + kfree(dqid); + } + } +#endif + /* turn off quota cause we are to dump chk_dqblk to files */ + quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type); + + rc = create_new_quota_files(qctxt, oqc); + if (rc) + GOTO(out, rc); + + /* we use vfs functions to set dqblk, so turn quota on */ + rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type); +out: + /* dump and free chk_dqblk */ + rc = prune_chkquots(sb, qctxt, rc); + OBD_FREE_PTR(qctxt); + + /* turn off quota, `lfs quotacheck` will turn on when all + * nodes quotacheck finish. */ + quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type); + + oqc->qc_stat = rc; + if (rc) + CERROR("quotacheck failed: rc = %d\n", rc); + + RETURN(rc); } -static int fsfilt_ext3_get_inode_write_extents(struct inode *inode, - char **pbuf, int *size) +#ifdef HAVE_QUOTA_SUPPORT +static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type, + int cmd) { int rc = 0; - - rc = ext3_ext_in_ea_get_extents(inode, EXT3_XATTR_INDEX_TRUSTED, - EXTENTS_EA, pbuf, size); - return rc; -} + ENTRY; + + if (lqi->qi_files[type] == NULL) { + CERROR("operate qinfo before it's enabled!\n"); + RETURN(-EIO); + } + + switch (cmd) { + case QFILE_CHK: + rc = lustre_check_quota_file(lqi, type); + break; + case QFILE_RD_INFO: + rc = lustre_read_quota_info(lqi, type); + break; + case QFILE_WR_INFO: + rc = lustre_write_quota_info(lqi, type); + break; + case QFILE_INIT_INFO: + rc = lustre_init_quota_info(lqi, type); + break; + default: + CERROR("Unsupported admin quota file cmd %d\n", cmd); + LBUG(); + break; + } + RETURN(rc); +} -static int fsfilt_ext3_get_write_extents_num(struct inode *inode, int *size) +static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type, + struct list_head *list) +{ + return lustre_get_qids(file, inode, type, list); +} + +static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd) { int rc = 0; - - rc = ext3_ext_in_ea_get_extents_num(inode, EXT3_XATTR_INDEX_TRUSTED, - EXTENTS_EA, size); - return rc; -} + ENTRY; + + if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) { + CERROR("operate dquot before it's enabled!\n"); + RETURN(-EIO); + } + + switch (cmd) { + case QFILE_RD_DQUOT: + rc = lustre_read_dquot(dquot); + break; + case QFILE_WR_DQUOT: + if (dquot->dq_dqb.dqb_ihardlimit || + dquot->dq_dqb.dqb_isoftlimit || + dquot->dq_dqb.dqb_bhardlimit || + dquot->dq_dqb.dqb_bsoftlimit) + clear_bit(DQ_FAKE_B, &dquot->dq_flags); + else + set_bit(DQ_FAKE_B, &dquot->dq_flags); + + rc = lustre_commit_dquot(dquot); + if (rc >= 0) + rc = 0; + break; + default: + CERROR("Unsupported admin quota file cmd %d\n", cmd); + LBUG(); + break; + } + RETURN(rc); +} +#endif + +lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb) +{ + return (EXT3_SB(sb)->journal_bdev); +} +EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev); static struct fsfilt_operations fsfilt_ext3_ops = { .fs_type = "ext3", .fs_owner = THIS_MODULE, + .fs_getlabel = fsfilt_ext3_get_label, + .fs_setlabel = fsfilt_ext3_set_label, + .fs_uuid = fsfilt_ext3_uuid, .fs_start = fsfilt_ext3_start, .fs_brw_start = fsfilt_ext3_brw_start, + .fs_extend = fsfilt_ext3_extend, .fs_commit = fsfilt_ext3_commit, .fs_commit_async = fsfilt_ext3_commit_async, .fs_commit_wait = fsfilt_ext3_commit_wait, @@ -1357,31 +2044,31 @@ static struct fsfilt_operations fsfilt_ext3_ops = { .fs_statfs = fsfilt_ext3_statfs, .fs_sync = fsfilt_ext3_sync, .fs_map_inode_pages = fsfilt_ext3_map_inode_pages, - .fs_prep_san_write = fsfilt_ext3_prep_san_write, .fs_write_record = fsfilt_ext3_write_record, .fs_read_record = fsfilt_ext3_read_record, .fs_setup = fsfilt_ext3_setup, - .fs_getpage = fsfilt_ext3_getpage, .fs_send_bio = fsfilt_ext3_send_bio, - .fs_set_xattr = fsfilt_ext3_set_xattr, - .fs_get_xattr = fsfilt_ext3_get_xattr, .fs_get_op_len = fsfilt_ext3_get_op_len, - .fs_add_dir_entry = fsfilt_ext3_add_dir_entry, - .fs_del_dir_entry = fsfilt_ext3_del_dir_entry, - .fs_init_extents_ea = fsfilt_ext3_init_extents_ea, - .fs_insert_extents_ea = fsfilt_ext3_insert_extents_ea, - .fs_remove_extents_ea = fsfilt_ext3_remove_extents_ea, - .fs_get_inode_write_extents = fsfilt_ext3_get_inode_write_extents, - .fs_get_write_extents_num = fsfilt_ext3_get_write_extents_num, + .fs_quotactl = fsfilt_ext3_quotactl, + .fs_quotacheck = fsfilt_ext3_quotacheck, +#ifdef HAVE_DISK_INODE_VERSION + .fs_get_version = fsfilt_ext3_get_version, + .fs_set_version = fsfilt_ext3_set_version, +#endif +#ifdef HAVE_QUOTA_SUPPORT + .fs_quotainfo = fsfilt_ext3_quotainfo, + .fs_qids = fsfilt_ext3_qids, + .fs_dquot = fsfilt_ext3_dquot, +#endif + .fs_journal_sbdev = fsfilt_ext3_journal_sbdev, }; static int __init fsfilt_ext3_init(void) { int rc; - fcb_cache = kmem_cache_create("fsfilt_ext3_fcb", - sizeof(struct fsfilt_cb_data), 0, - 0, NULL, NULL); + fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb", + sizeof(struct fsfilt_cb_data), 0, 0); if (!fcb_cache) { CERROR("error allocating fsfilt journal callback cache\n"); GOTO(out, rc = -ENOMEM); @@ -1389,8 +2076,10 @@ static int __init fsfilt_ext3_init(void) rc = fsfilt_register_ops(&fsfilt_ext3_ops); - if (rc) - kmem_cache_destroy(fcb_cache); + if (rc) { + int err = cfs_mem_cache_destroy(fcb_cache); + LASSERTF(err == 0, "error destroying new cache: rc %d\n", err); + } out: return rc; } @@ -1400,12 +2089,8 @@ static void __exit fsfilt_ext3_exit(void) int rc; fsfilt_unregister_ops(&fsfilt_ext3_ops); - rc = kmem_cache_destroy(fcb_cache); - - if (rc || atomic_read(&fcb_cache_count)) { - CERROR("can't free fsfilt callback cache: count %d, rc = %d\n", - atomic_read(&fcb_cache_count), rc); - } + rc = cfs_mem_cache_destroy(fcb_cache); + LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n"); } module_init(fsfilt_ext3_init);