X-Git-Url: https://git.camperquake.de/gitweb.cgi?a=blobdiff_plain;f=module%2Fzfs%2Fzfs_vnops.c;h=3b671b4acd42b58fa5cc7df54ccf0c9cbbd1ad2a;hb=389cf730cedd42dd1ef653e9358635c114e458d5;hp=cb66741c2adf2d4d1be31f6c07fd75570c4ebed0;hpb=126400a1ca656d41dea9d2ad88afbec3ed32d391;p=zfs.git diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c index cb66741..3b671b4 100644 --- a/module/zfs/zfs_vnops.c +++ b/module/zfs/zfs_vnops.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012 by Delphix. All rights reserved. */ /* Portions Copyright 2007 Jeremy Teo */ @@ -63,6 +64,7 @@ #include #include #include "fs/fs_subr.h" +#include #include #include #include @@ -72,6 +74,7 @@ #include #include #include +#include /* * Programming rules. @@ -220,7 +223,11 @@ zfs_close(struct inode *ip, int flag, cred_t *cr) ZFS_ENTER(zsb); ZFS_VERIFY_ZP(zp); - /* Decrement the synchronous opens in the znode */ + /* + * Zero the synchronous opens in the znode. Under Linux the + * zfs_close() hook is not symmetric with zfs_open(), it is + * only called once when the last reference is dropped. + */ if (flag & O_SYNC) zp->z_sync_cnt = 0; @@ -335,7 +342,7 @@ mappedread(struct inode *ip, int nbytes, uio_t *uio) } #endif /* _KERNEL */ -offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ +unsigned long zfs_read_chunk_size = 1024 * 1024; /* Tunable */ /* * Read bytes from specified file into supplied buffer. @@ -394,18 +401,14 @@ zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr) return (0); } -#ifdef HAVE_MANDLOCKS /* * Check for mandatory locks */ - if (MANDMODE(zp->z_mode)) { - if (error = chklock(ip, FREAD, - uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { - ZFS_EXIT(zsb); - return (error); - } + if (mandatory_lock(ip) && + !lock_may_read(ip, uio->uio_loffset, uio->uio_resid)) { + ZFS_EXIT(zsb); + return (EAGAIN); } -#endif /* HAVE_MANDLOCK */ /* * If we're in FRSYNC mode, sync out this znode before reading it. @@ -576,17 +579,14 @@ zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr) return (EINVAL); } -#ifdef HAVE_MANDLOCKS /* * Check for mandatory locks before calling zfs_range_lock() * in order to prevent a deadlock with locks set via fcntl(). */ - if (MANDMODE((mode_t)zp->z_mode) && - (error = chklock(ip, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { + if (mandatory_lock(ip) && !lock_may_write(ip, woff, n)) { ZFS_EXIT(zsb); - return (error); + return (EAGAIN); } -#endif /* HAVE_MANDLOCKS */ #ifdef HAVE_UIO_ZEROCOPY /* @@ -869,7 +869,7 @@ iput_async(struct inode *ip, taskq_t *taskq) { ASSERT(atomic_read(&ip->i_count) > 0); if (atomic_read(&ip->i_count) == 1) - taskq_dispatch(taskq, (task_func_t *)iput, ip, TQ_SLEEP); + taskq_dispatch(taskq, (task_func_t *)iput, ip, TQ_PUSHPAGE); else iput(ip); } @@ -935,7 +935,7 @@ zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) return (ENOENT); } - zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); + zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_PUSHPAGE); zgd->zgd_zilog = zsb->z_log; zgd->zgd_private = zp; @@ -1115,14 +1115,6 @@ zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags, if (flags & LOOKUP_XATTR) { /* - * If the xattr property is off, refuse the lookup request. - */ - if (!(zsb->z_flags & ZSB_XATTR_USER)) { - ZFS_EXIT(zsb); - return (EINVAL); - } - - /* * We don't allow recursive attributes.. * Maybe someday we will. */ @@ -1532,7 +1524,7 @@ top: &xattr_obj, sizeof (xattr_obj)); if (error == 0 && xattr_obj) { error = zfs_zget(zsb, xattr_obj, &xzp); - ASSERT3U(error, ==, 0); + ASSERT0(error); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); } @@ -1909,13 +1901,13 @@ top: out: zfs_dirent_unlock(dl); + zfs_inode_update(dzp); + zfs_inode_update(zp); iput(ip); if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zilog, 0); - zfs_inode_update(dzp); - zfs_inode_update(zp); ZFS_EXIT(zsb); return (error); } @@ -2022,12 +2014,22 @@ zfs_readdir(struct inode *ip, void *dirent, filldir_t filldir, goto update; } + /* + * Allow multiple entries provided the first entry is + * the object id. Non-zpl consumers may safely make + * use of the additional space. + * + * XXX: This should be a feature flag for compatibility + */ if (zap.za_integer_length != 8 || - zap.za_num_integers != 1) { + zap.za_num_integers == 0) { cmn_err(CE_WARN, "zap_readdir: bad directory " - "entry, obj = %lld, offset = %lld\n", + "entry, obj = %lld, offset = %lld, " + "length = %d, num = %lld\n", (u_longlong_t)zp->z_id, - (u_longlong_t)*pos); + (u_longlong_t)*pos, + zap.za_integer_length, + (u_longlong_t)zap.za_num_integers); error = ENXIO; goto update; } @@ -2045,7 +2047,7 @@ zfs_readdir(struct inode *ip, void *dirent, filldir_t filldir, dmu_prefetch(os, objnum, 0, 0); } - if (*pos >= 2) { + if (*pos > 2 || (*pos == 2 && !zfs_show_ctldir(zp))) { zap_cursor_advance(&zc); *pos = zap_cursor_serialize(&zc); } else { @@ -2293,6 +2295,49 @@ zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr) EXPORT_SYMBOL(zfs_getattr); /* + * Get the basic file attributes and place them in the provided kstat + * structure. The inode is assumed to be the authoritative source + * for most of the attributes. However, the znode currently has the + * authoritative atime, blksize, and block count. + * + * IN: ip - inode of file. + * + * OUT: sp - kstat values. + * + * RETURN: 0 (always succeeds) + */ +/* ARGSUSED */ +int +zfs_getattr_fast(struct inode *ip, struct kstat *sp) +{ + znode_t *zp = ITOZ(ip); + zfs_sb_t *zsb = ITOZSB(ip); + + ZFS_ENTER(zsb); + ZFS_VERIFY_ZP(zp); + + mutex_enter(&zp->z_lock); + + generic_fillattr(ip, sp); + ZFS_TIME_DECODE(&sp->atime, zp->z_atime); + + sa_object_size(zp->z_sa_hdl, (uint32_t *)&sp->blksize, &sp->blocks); + if (unlikely(zp->z_blksz == 0)) { + /* + * Block size hasn't been set; suggest maximal I/O transfers. + */ + sp->blksize = zsb->z_max_blksz; + } + + mutex_exit(&zp->z_lock); + + ZFS_EXIT(zsb); + + return (0); +} +EXPORT_SYMBOL(zfs_getattr_fast); + +/* * Set the file attributes to the values contained in the * vattr structure. * @@ -2335,7 +2380,7 @@ zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr) zfs_acl_t *aclp; boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; boolean_t fuid_dirtied = B_FALSE; - sa_bulk_attr_t bulk[7], xattr_bulk[7]; + sa_bulk_attr_t *bulk, *xattr_bulk; int count = 0, xattr_count = 0; if (mask == 0) @@ -2378,6 +2423,9 @@ zfs_setattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr) tmpxvattr = kmem_alloc(sizeof(xvattr_t), KM_SLEEP); xva_init(tmpxvattr); + bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 7, KM_SLEEP); + xattr_bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 7, KM_SLEEP); + /* * Immutable files can only alter immutable bit and atime */ @@ -2412,7 +2460,7 @@ top: aclp = NULL; /* Can this be moved to before the top label? */ - if (zsb->z_vfs->mnt_flags & MNT_READONLY) { + if (zfs_is_readonly(zsb)) { err = EROFS; goto out3; } @@ -2426,6 +2474,8 @@ top: if (err) goto out3; + truncate_setsize(ip, vap->va_size); + /* * XXX - Note, we are not providing any open * mode flags here (like FNDELAY), so we may @@ -2436,11 +2486,6 @@ top: err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); if (err) goto out3; - - /* Careful negative Linux return code here */ - err = -vmtruncate(ip, vap->va_size); - if (err) - goto out3; } if (mask & (ATTR_ATIME|ATTR_MTIME) || @@ -2793,7 +2838,7 @@ top: zp->z_mode = new_mode; ASSERT3P(aclp, !=, NULL); err = zfs_aclset_common(zp, aclp, cr, tx); - ASSERT3U(err, ==, 0); + ASSERT0(err); if (zp->z_acl_cached) zfs_acl_free(zp->z_acl_cached); zp->z_acl_cached = aclp; @@ -2918,6 +2963,8 @@ out2: zil_commit(zilog, 0); out3: + kmem_free(xattr_bulk, sizeof(sa_bulk_attr_t) * 7); + kmem_free(bulk, sizeof(sa_bulk_attr_t) * 7); kmem_free(tmpxvattr, sizeof(xvattr_t)); ZFS_EXIT(zsb); return (err); @@ -3300,7 +3347,7 @@ top: error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb), (void *)&szp->z_pflags, sizeof (uint64_t), tx); - ASSERT3U(error, ==, 0); + ASSERT0(error); error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); if (error == 0) { @@ -3705,242 +3752,203 @@ top: } EXPORT_SYMBOL(zfs_link); -#ifdef HAVE_MMAP -/* - * zfs_null_putapage() is used when the file system has been force - * unmounted. It just drops the pages. - */ -/* ARGSUSED */ -static int -zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, - size_t *lenp, int flags, cred_t *cr) +static void +zfs_putpage_commit_cb(void *arg, int error) { - pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); - return (0); + struct page *pp = arg; + + if (error) { + __set_page_dirty_nobuffers(pp); + + if (error != ECANCELED) + SetPageError(pp); + } else { + ClearPageError(pp); + } + + end_page_writeback(pp); } /* - * Push a page out to disk, klustering if possible. - * - * IN: vp - file to push page to. - * pp - page to push. - * flags - additional flags. - * cr - credentials of caller. + * Push a page out to disk, once the page is on stable storage the + * registered commit callback will be run as notification of completion. * - * OUT: offp - start of range pushed. - * lenp - len of range pushed. + * IN: ip - page mapped for inode. + * pp - page to push (page is locked) + * wbc - writeback control data * * RETURN: 0 if success * error code if failure * - * NOTE: callers must have locked the page to be pushed. On - * exit, the page (and all other pages in the kluster) must be - * unlocked. + * Timestamps: + * ip - ctime|mtime updated */ /* ARGSUSED */ -static int -zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, - size_t *lenp, int flags, cred_t *cr) +int +zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc) { - znode_t *zp = VTOZ(vp); - zfsvfs_t *zfsvfs = zp->z_zfsvfs; + znode_t *zp = ITOZ(ip); + zfs_sb_t *zsb = ITOZSB(ip); + loff_t offset; + loff_t pgoff; + unsigned int pglen; + rl_t *rl; dmu_tx_t *tx; - u_offset_t off, koff; - size_t len, klen; - int err; + caddr_t va; + int err = 0; + uint64_t mtime[2], ctime[2]; + sa_bulk_attr_t bulk[3]; + int cnt = 0; + int sync; - off = pp->p_offset; - len = PAGESIZE; - /* - * If our blocksize is bigger than the page size, try to kluster - * multiple pages so that we write a full block (thus avoiding - * a read-modify-write). - */ - if (off < zp->z_size && zp->z_blksz > PAGESIZE) { - klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); - koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; - ASSERT(koff <= zp->z_size); - if (koff + klen > zp->z_size) - klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); - pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); - } - ASSERT3U(btop(len), ==, btopr(len)); + ZFS_ENTER(zsb); + ZFS_VERIFY_ZP(zp); - /* - * Can't push pages past end-of-file. - */ - if (off >= zp->z_size) { - /* ignore all pages */ - err = 0; - goto out; - } else if (off + len > zp->z_size) { - int npages = btopr(zp->z_size - off); - page_t *trunc; + ASSERT(PageLocked(pp)); + + pgoff = page_offset(pp); /* Page byte-offset in file */ + offset = i_size_read(ip); /* File length in bytes */ + pglen = MIN(PAGE_CACHE_SIZE, /* Page length in bytes */ + P2ROUNDUP(offset, PAGE_CACHE_SIZE)-pgoff); - page_list_break(&pp, &trunc, npages); - /* ignore pages past end of file */ - if (trunc) - pvn_write_done(trunc, flags); - len = zp->z_size - off; + /* Page is beyond end of file */ + if (pgoff >= offset) { + unlock_page(pp); + ZFS_EXIT(zsb); + return (0); } - if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || - zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { + /* Truncate page length to end of file */ + if (pgoff + pglen > offset) + pglen = offset - pgoff; + +#if 0 + /* + * FIXME: Allow mmap writes past its quota. The correct fix + * is to register a page_mkwrite() handler to count the page + * against its quota when it is about to be dirtied. + */ + if (zfs_owner_overquota(zsb, zp, B_FALSE) || + zfs_owner_overquota(zsb, zp, B_TRUE)) { err = EDQUOT; - goto out; } -top: - tx = dmu_tx_create(zfsvfs->z_os); - dmu_tx_hold_write(tx, zp->z_id, off, len); +#endif + + set_page_writeback(pp); + unlock_page(pp); + + rl = zfs_range_lock(zp, pgoff, pglen, RL_WRITER); + tx = dmu_tx_create(zsb->z_os); + + sync = ((zsb->z_os->os_sync == ZFS_SYNC_ALWAYS) || + (wbc->sync_mode == WB_SYNC_ALL)); + if (!sync) + dmu_tx_callback_register(tx, zfs_putpage_commit_cb, pp); + + dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen); dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); zfs_sa_upgrade_txholds(tx, zp); err = dmu_tx_assign(tx, TXG_NOWAIT); if (err != 0) { - if (err == ERESTART) { + if (err == ERESTART) dmu_tx_wait(tx); - dmu_tx_abort(tx); - goto top; - } + + /* Will call all registered commit callbacks */ dmu_tx_abort(tx); - goto out; - } - if (zp->z_blksz <= PAGESIZE) { - caddr_t va = zfs_map_page(pp, S_READ); - ASSERT3U(len, <=, PAGESIZE); - dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); - zfs_unmap_page(pp, va); - } else { - err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); + /* + * For the synchronous case the commit callback must be + * explicitly called because there is no registered callback. + */ + if (sync) + zfs_putpage_commit_cb(pp, ECANCELED); + + zfs_range_unlock(rl); + ZFS_EXIT(zsb); + return (err); } - if (err == 0) { - uint64_t mtime[2], ctime[2]; - sa_bulk_attr_t bulk[3]; - int count = 0; + va = kmap(pp); + ASSERT3U(pglen, <=, PAGE_CACHE_SIZE); + dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx); + kunmap(pp); - SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, - &mtime, 16); - SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, - &ctime, 16); - SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, - &zp->z_pflags, 8); - zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, - B_TRUE); - zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); - } + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8); + + /* Preserve the mtime and ctime provided by the inode */ + ZFS_TIME_ENCODE(&ip->i_mtime, mtime); + ZFS_TIME_ENCODE(&ip->i_ctime, ctime); + zp->z_atime_dirty = 0; + zp->z_seq++; + + err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx); + + zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0); dmu_tx_commit(tx); -out: - pvn_write_done(pp, (err ? B_ERROR : 0) | flags); - if (offp) - *offp = off; - if (lenp) - *lenp = len; + zfs_range_unlock(rl); + if (sync) { + zil_commit(zsb->z_log, zp->z_id); + zfs_putpage_commit_cb(pp, err); + } + + ZFS_EXIT(zsb); return (err); } /* - * Copy the portion of the file indicated from pages into the file. - * The pages are stored in a page list attached to the files vnode. - * - * IN: vp - vnode of file to push page data to. - * off - position in file to put data. - * len - amount of data to write. - * flags - flags to control the operation. - * cr - credentials of caller. - * ct - caller context. - * - * RETURN: 0 if success - * error code if failure - * - * Timestamps: - * vp - ctime|mtime updated + * Update the system attributes when the inode has been dirtied. For the + * moment we're conservative and only update the atime, mtime, and ctime. */ -/*ARGSUSED*/ -static int -zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr) +int +zfs_dirty_inode(struct inode *ip, int flags) { - znode_t *zp = VTOZ(vp); - zfsvfs_t *zfsvfs = zp->z_zfsvfs; - page_t *pp; - size_t io_len; - u_offset_t io_off; - uint_t blksz; - rl_t *rl; - int error = 0; + znode_t *zp = ITOZ(ip); + zfs_sb_t *zsb = ITOZSB(ip); + dmu_tx_t *tx; + uint64_t atime[2], mtime[2], ctime[2]; + sa_bulk_attr_t bulk[3]; + int error; + int cnt = 0; - ZFS_ENTER(zfsvfs); + ZFS_ENTER(zsb); ZFS_VERIFY_ZP(zp); - /* - * Align this request to the file block size in case we kluster. - * XXX - this can result in pretty aggresive locking, which can - * impact simultanious read/write access. One option might be - * to break up long requests (len == 0) into block-by-block - * operations to get narrower locking. - */ - blksz = zp->z_blksz; - if (ISP2(blksz)) - io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); - else - io_off = 0; - if (len > 0 && ISP2(blksz)) - io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); - else - io_len = 0; + tx = dmu_tx_create(zsb->z_os); - if (io_len == 0) { - /* - * Search the entire vp list for pages >= io_off. - */ - rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); - error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); - goto out; - } - rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); + dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); + zfs_sa_upgrade_txholds(tx, zp); - if (off > zp->z_size) { - /* past end of file */ - zfs_range_unlock(rl); - ZFS_EXIT(zfsvfs); - return (0); + error = dmu_tx_assign(tx, TXG_WAIT); + if (error) { + dmu_tx_abort(tx); + goto out; } - len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); + mutex_enter(&zp->z_lock); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16); - for (off = io_off; io_off < off + len; io_off += io_len) { - if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { - pp = page_lookup(vp, io_off, - (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); - } else { - pp = page_lookup_nowait(vp, io_off, - (flags & B_FREE) ? SE_EXCL : SE_SHARED); - } + /* Preserve the mtime and ctime provided by the inode */ + ZFS_TIME_ENCODE(&ip->i_atime, atime); + ZFS_TIME_ENCODE(&ip->i_mtime, mtime); + ZFS_TIME_ENCODE(&ip->i_ctime, ctime); + zp->z_atime_dirty = 0; - if (pp != NULL && pvn_getdirty(pp, flags)) { - int err; + error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx); + mutex_exit(&zp->z_lock); - /* - * Found a dirty page to push - */ - err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); - if (err) - error = err; - } else { - io_len = PAGESIZE; - } - } + dmu_tx_commit(tx); out: - zfs_range_unlock(rl); - if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) - zil_commit(zfsvfs->z_log, zp->z_id); - ZFS_EXIT(zfsvfs); + ZFS_EXIT(zsb); return (error); } -#endif /* HAVE_MMAP */ +EXPORT_SYMBOL(zfs_dirty_inode); /*ARGSUSED*/ void @@ -3950,9 +3958,10 @@ zfs_inactive(struct inode *ip) zfs_sb_t *zsb = ITOZSB(ip); int error; -#ifdef HAVE_SNAPSHOT - /* Early return for snapshot inode? */ -#endif /* HAVE_SNAPSHOT */ + if (zfsctl_is_node(ip)) { + zfsctl_inode_inactive(ip); + return; + } rw_enter(&zsb->z_teardown_inactive_lock, RW_READER); if (zp->z_sa_hdl == NULL) { @@ -4004,130 +4013,60 @@ zfs_seek(struct inode *ip, offset_t ooff, offset_t *noffp) } EXPORT_SYMBOL(zfs_seek); -#ifdef HAVE_MMAP /* - * Pre-filter the generic locking function to trap attempts to place - * a mandatory lock on a memory mapped file. + * Fill pages with data from the disk. */ static int -zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, - flk_callback_t *flk_cbp, cred_t *cr) +zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages) { - znode_t *zp = VTOZ(vp); - zfsvfs_t *zfsvfs = zp->z_zfsvfs; - - ZFS_ENTER(zfsvfs); - ZFS_VERIFY_ZP(zp); + znode_t *zp = ITOZ(ip); + zfs_sb_t *zsb = ITOZSB(ip); + objset_t *os; + struct page *cur_pp; + u_offset_t io_off, total; + size_t io_len; + loff_t i_size; + unsigned page_idx; + int err; + + os = zsb->z_os; + io_len = nr_pages << PAGE_CACHE_SHIFT; + i_size = i_size_read(ip); + io_off = page_offset(pl[0]); + + if (io_off + io_len > i_size) + io_len = i_size - io_off; /* - * We are following the UFS semantics with respect to mapcnt - * here: If we see that the file is mapped already, then we will - * return an error, but we don't worry about races between this - * function and zfs_map(). + * Iterate over list of pages and read each page individually. */ - if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { - ZFS_EXIT(zfsvfs); - return (EAGAIN); - } - ZFS_EXIT(zfsvfs); - return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); -} - -/* - * If we can't find a page in the cache, we will create a new page - * and fill it with file data. For efficiency, we may try to fill - * multiple pages at once (klustering) to fill up the supplied page - * list. Note that the pages to be filled are held with an exclusive - * lock to prevent access by other threads while they are being filled. - */ -static int -zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, - caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) -{ - znode_t *zp = VTOZ(vp); - page_t *pp, *cur_pp; - objset_t *os = zp->z_zfsvfs->z_os; - u_offset_t io_off, total; - size_t io_len; - int err; - - if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { - /* - * We only have a single page, don't bother klustering - */ - io_off = off; - io_len = PAGESIZE; - pp = page_create_va(vp, io_off, io_len, - PG_EXCL | PG_WAIT, seg, addr); - } else { - /* - * Try to find enough pages to fill the page list - */ - pp = pvn_read_kluster(vp, off, seg, addr, &io_off, - &io_len, off, plsz, 0); - } - if (pp == NULL) { - /* - * The page already exists, nothing to do here. - */ - *pl = NULL; - return (0); - } - - /* - * Fill the pages in the kluster. - */ - cur_pp = pp; + page_idx = 0; + cur_pp = pl[0]; for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { caddr_t va; - ASSERT3U(io_off, ==, cur_pp->p_offset); - va = zfs_map_page(cur_pp, S_WRITE); + va = kmap(cur_pp); err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, DMU_READ_PREFETCH); - zfs_unmap_page(cur_pp, va); + kunmap(cur_pp); if (err) { - /* On error, toss the entire kluster */ - pvn_read_done(pp, B_ERROR); /* convert checksum errors into IO errors */ if (err == ECKSUM) err = EIO; return (err); } - cur_pp = cur_pp->p_next; + cur_pp = pl[++page_idx]; } - /* - * Fill in the page list array from the kluster starting - * from the desired offset `off'. - * NOTE: the page list will always be null terminated. - */ - pvn_plist_init(pp, pl, plsz, off, io_len, rw); - ASSERT(pl == NULL || (*pl)->p_offset == off); - return (0); } /* - * Return pointers to the pages for the file region [off, off + len] - * in the pl array. If plsz is greater than len, this function may - * also return page pointers from after the specified region - * (i.e. the region [off, off + plsz]). These additional pages are - * only returned if they are already in the cache, or were created as - * part of a klustered read. + * Uses zfs_fillpage to read data from the file and fill the pages. * - * IN: vp - vnode of file to get data from. - * off - position in file to get data from. - * len - amount of data to retrieve. - * plsz - length of provided page list. - * seg - segment to obtain pages for. - * addr - virtual address of fault. - * rw - mode of created pages. - * cr - credentials of caller. - * ct - caller context. - * - * OUT: protp - protection mode of created pages. - * pl - list of pages created. + * IN: ip - inode of file to get data from. + * pl - list of pages to read + * nr_pages - number of pages to read * * RETURN: 0 if success * error code if failure @@ -4136,217 +4075,73 @@ zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, * vp - atime updated */ /* ARGSUSED */ -static int -zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, - page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, - enum seg_rw rw, cred_t *cr) +int +zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages) { - znode_t *zp = VTOZ(vp); - zfsvfs_t *zfsvfs = zp->z_zfsvfs; - page_t **pl0 = pl; - int err = 0; + znode_t *zp = ITOZ(ip); + zfs_sb_t *zsb = ITOZSB(ip); + int err; - /* we do our own caching, faultahead is unnecessary */ if (pl == NULL) return (0); - else if (len > plsz) - len = plsz; - else - len = P2ROUNDUP(len, PAGESIZE); - ASSERT(plsz >= len); - ZFS_ENTER(zfsvfs); + ZFS_ENTER(zsb); ZFS_VERIFY_ZP(zp); - if (protp) - *protp = PROT_ALL; + err = zfs_fillpage(ip, pl, nr_pages); - /* - * Loop through the requested range [off, off + len) looking - * for pages. If we don't find a page, we will need to create - * a new page and fill it with data from the file. - */ - while (len > 0) { - if (*pl = page_lookup(vp, off, SE_SHARED)) - *(pl+1) = NULL; - else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) - goto out; - while (*pl) { - ASSERT3U((*pl)->p_offset, ==, off); - off += PAGESIZE; - addr += PAGESIZE; - if (len > 0) { - ASSERT3U(len, >=, PAGESIZE); - len -= PAGESIZE; - } - ASSERT3U(plsz, >=, PAGESIZE); - plsz -= PAGESIZE; - pl++; - } - } - - /* - * Fill out the page array with any pages already in the cache. - */ - while (plsz > 0 && - (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { - off += PAGESIZE; - plsz -= PAGESIZE; - } -out: - if (err) { - /* - * Release any pages we have previously locked. - */ - while (pl > pl0) - page_unlock(*--pl); - } else { - ZFS_ACCESSTIME_STAMP(zfsvfs, zp); - } + if (!err) + ZFS_ACCESSTIME_STAMP(zsb, zp); - *pl = NULL; - - ZFS_EXIT(zfsvfs); + ZFS_EXIT(zsb); return (err); } +EXPORT_SYMBOL(zfs_getpage); /* - * Request a memory map for a section of a file. This code interacts - * with common code and the VM system as follows: + * Check ZFS specific permissions to memory map a section of a file. * - * common code calls mmap(), which ends up in smmap_common() + * IN: ip - inode of the file to mmap + * off - file offset + * addrp - start address in memory region + * len - length of memory region + * vm_flags- address flags * - * this calls VOP_MAP(), which takes you into (say) zfs - * - * zfs_map() calls as_map(), passing segvn_create() as the callback - * - * segvn_create() creates the new segment and calls VOP_ADDMAP() - * - * zfs_addmap() updates z_mapcnt + * RETURN: 0 if success + * error code if failure */ /*ARGSUSED*/ -static int -zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, - size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) +int +zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len, + unsigned long vm_flags) { - znode_t *zp = VTOZ(vp); - zfsvfs_t *zfsvfs = zp->z_zfsvfs; - segvn_crargs_t vn_a; - int error; + znode_t *zp = ITOZ(ip); + zfs_sb_t *zsb = ITOZSB(ip); - ZFS_ENTER(zfsvfs); + ZFS_ENTER(zsb); ZFS_VERIFY_ZP(zp); - if ((prot & PROT_WRITE) && (zp->z_pflags & + if ((vm_flags & VM_WRITE) && (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { - ZFS_EXIT(zfsvfs); + ZFS_EXIT(zsb); return (EPERM); } - if ((prot & (PROT_READ | PROT_EXEC)) && + if ((vm_flags & (VM_READ | VM_EXEC)) && (zp->z_pflags & ZFS_AV_QUARANTINED)) { - ZFS_EXIT(zfsvfs); + ZFS_EXIT(zsb); return (EACCES); } - if (vp->v_flag & VNOMAP) { - ZFS_EXIT(zfsvfs); - return (ENOSYS); - } - if (off < 0 || len > MAXOFFSET_T - off) { - ZFS_EXIT(zfsvfs); + ZFS_EXIT(zsb); return (ENXIO); } - if (vp->v_type != VREG) { - ZFS_EXIT(zfsvfs); - return (ENODEV); - } - - /* - * If file is locked, disallow mapping. - */ - if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { - ZFS_EXIT(zfsvfs); - return (EAGAIN); - } - - as_rangelock(as); - error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); - if (error != 0) { - as_rangeunlock(as); - ZFS_EXIT(zfsvfs); - return (error); - } - - vn_a.vp = vp; - vn_a.offset = (u_offset_t)off; - vn_a.type = flags & MAP_TYPE; - vn_a.prot = prot; - vn_a.maxprot = maxprot; - vn_a.cred = cr; - vn_a.amp = NULL; - vn_a.flags = flags & ~MAP_TYPE; - vn_a.szc = 0; - vn_a.lgrp_mem_policy_flags = 0; - - error = as_map(as, *addrp, len, segvn_create, &vn_a); - - as_rangeunlock(as); - ZFS_EXIT(zfsvfs); - return (error); -} - -/* ARGSUSED */ -static int -zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, - size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) -{ - uint64_t pages = btopr(len); - - atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); - return (0); -} - -/* - * The reason we push dirty pages as part of zfs_delmap() is so that we get a - * more accurate mtime for the associated file. Since we don't have a way of - * detecting when the data was actually modified, we have to resort to - * heuristics. If an explicit msync() is done, then we mark the mtime when the - * last page is pushed. The problem occurs when the msync() call is omitted, - * which by far the most common case: - * - * open() - * mmap() - * - * munmap() - * close() - *