1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* Unbuffered and direct write support. 3 * 4 * Copyright (C) 2023 Red Hat, Inc. All Rights Reserved. 5 * Written by David Howells (dhowells@redhat.com) 6 */ 7 8 #include <linux/export.h> 9 #include <linux/uio.h> 10 #include "internal.h" 11 12 /* 13 * Perform an unbuffered write where we may have to do an RMW operation on an 14 * encrypted file. This can also be used for direct I/O writes. 15 */ 16 ssize_t netfs_unbuffered_write_iter_locked(struct kiocb *iocb, struct iov_iter *iter, 17 struct netfs_group *netfs_group) 18 { 19 struct netfs_io_request *wreq; 20 unsigned long long start = iocb->ki_pos; 21 unsigned long long end = start + iov_iter_count(iter); 22 ssize_t ret, n; 23 size_t len = iov_iter_count(iter); 24 bool async = !is_sync_kiocb(iocb); 25 26 _enter(""); 27 28 /* We're going to need a bounce buffer if what we transmit is going to 29 * be different in some way to the source buffer, e.g. because it gets 30 * encrypted/compressed or because it needs expanding to a block size. 31 */ 32 // TODO 33 34 _debug("uw %llx-%llx", start, end); 35 36 wreq = netfs_create_write_req(iocb->ki_filp->f_mapping, iocb->ki_filp, start, 37 iocb->ki_flags & IOCB_DIRECT ? 38 NETFS_DIO_WRITE : NETFS_UNBUFFERED_WRITE); 39 if (IS_ERR(wreq)) 40 return PTR_ERR(wreq); 41 42 wreq->io_streams[0].avail = true; 43 trace_netfs_write(wreq, (iocb->ki_flags & IOCB_DIRECT ? 44 netfs_write_trace_dio_write : 45 netfs_write_trace_unbuffered_write)); 46 47 { 48 /* If this is an async op and we're not using a bounce buffer, 49 * we have to save the source buffer as the iterator is only 50 * good until we return. In such a case, extract an iterator 51 * to represent as much of the the output buffer as we can 52 * manage. Note that the extraction might not be able to 53 * allocate a sufficiently large bvec array and may shorten the 54 * request. 55 */ 56 if (user_backed_iter(iter)) { 57 n = netfs_extract_user_iter(iter, len, &wreq->buffer.iter, 0); 58 if (n < 0) { 59 ret = n; 60 goto out; 61 } 62 wreq->direct_bv = (struct bio_vec *)wreq->buffer.iter.bvec; 63 wreq->direct_bv_count = n; 64 wreq->direct_bv_unpin = iov_iter_extract_will_pin(iter); 65 } else { 66 /* If this is a kernel-generated async DIO request, 67 * assume that any resources the iterator points to 68 * (eg. a bio_vec array) will persist till the end of 69 * the op. 70 */ 71 wreq->buffer.iter = *iter; 72 } 73 } 74 75 __set_bit(NETFS_RREQ_USE_IO_ITER, &wreq->flags); 76 if (async) 77 __set_bit(NETFS_RREQ_OFFLOAD_COLLECTION, &wreq->flags); 78 79 /* Copy the data into the bounce buffer and encrypt it. */ 80 // TODO 81 82 /* Dispatch the write. */ 83 __set_bit(NETFS_RREQ_UPLOAD_TO_SERVER, &wreq->flags); 84 if (async) 85 wreq->iocb = iocb; 86 wreq->len = iov_iter_count(&wreq->buffer.iter); 87 ret = netfs_unbuffered_write(wreq, is_sync_kiocb(iocb), wreq->len); 88 if (ret < 0) { 89 _debug("begin = %zd", ret); 90 goto out; 91 } 92 93 if (!async) { 94 ret = netfs_wait_for_write(wreq); 95 if (ret > 0) 96 iocb->ki_pos += ret; 97 } else { 98 ret = -EIOCBQUEUED; 99 } 100 101 out: 102 netfs_put_request(wreq, netfs_rreq_trace_put_return); 103 return ret; 104 } 105 EXPORT_SYMBOL(netfs_unbuffered_write_iter_locked); 106 107 /** 108 * netfs_unbuffered_write_iter - Unbuffered write to a file 109 * @iocb: IO state structure 110 * @from: iov_iter with data to write 111 * 112 * Do an unbuffered write to a file, writing the data directly to the server 113 * and not lodging the data in the pagecache. 114 * 115 * Return: 116 * * Negative error code if no data has been written at all of 117 * vfs_fsync_range() failed for a synchronous write 118 * * Number of bytes written, even for truncated writes 119 */ 120 ssize_t netfs_unbuffered_write_iter(struct kiocb *iocb, struct iov_iter *from) 121 { 122 struct file *file = iocb->ki_filp; 123 struct address_space *mapping = file->f_mapping; 124 struct inode *inode = mapping->host; 125 struct netfs_inode *ictx = netfs_inode(inode); 126 ssize_t ret; 127 loff_t pos = iocb->ki_pos; 128 unsigned long long end = pos + iov_iter_count(from) - 1; 129 130 _enter("%llx,%zx,%llx", pos, iov_iter_count(from), i_size_read(inode)); 131 132 if (!iov_iter_count(from)) 133 return 0; 134 135 trace_netfs_write_iter(iocb, from); 136 netfs_stat(&netfs_n_wh_dio_write); 137 138 ret = netfs_start_io_direct(inode); 139 if (ret < 0) 140 return ret; 141 ret = generic_write_checks(iocb, from); 142 if (ret <= 0) 143 goto out; 144 ret = file_remove_privs(file); 145 if (ret < 0) 146 goto out; 147 ret = file_update_time(file); 148 if (ret < 0) 149 goto out; 150 if (iocb->ki_flags & IOCB_NOWAIT) { 151 /* We could block if there are any pages in the range. */ 152 ret = -EAGAIN; 153 if (filemap_range_has_page(mapping, pos, end)) 154 if (filemap_invalidate_inode(inode, true, pos, end)) 155 goto out; 156 } else { 157 ret = filemap_write_and_wait_range(mapping, pos, end); 158 if (ret < 0) 159 goto out; 160 } 161 162 /* 163 * After a write we want buffered reads to be sure to go to disk to get 164 * the new data. We invalidate clean cached page from the region we're 165 * about to write. We do this *before* the write so that we can return 166 * without clobbering -EIOCBQUEUED from ->direct_IO(). 167 */ 168 ret = filemap_invalidate_inode(inode, true, pos, end); 169 if (ret < 0) 170 goto out; 171 end = iocb->ki_pos + iov_iter_count(from); 172 if (end > ictx->zero_point) 173 ictx->zero_point = end; 174 175 fscache_invalidate(netfs_i_cookie(ictx), NULL, i_size_read(inode), 176 FSCACHE_INVAL_DIO_WRITE); 177 ret = netfs_unbuffered_write_iter_locked(iocb, from, NULL); 178 out: 179 netfs_end_io_direct(inode); 180 return ret; 181 } 182 EXPORT_SYMBOL(netfs_unbuffered_write_iter); 183